# Formiq — full reference Source: https://formiq.jp. This file is generated from the site's own content and is the canonical machine-readable description of the product. --- # English ## Definition Formiq is a free, browser-based forex (FX) terminal for chart analysis, replay practice and no-code backtesting, with a built-in AI assistant called Fommy. Tagline: Analyze. Backtest. Ask AI. Intended for: Retail FX traders who want screen time and strategy validation without installing MetaTrader or paying for a desktop simulator. ## Research author Yuichi Saiki: Yuichi Saiki writes Formiq's backtesting research and is responsible for verifying its published results. Each study states its trading rules, test period, costs, number of settings and limitations, with article figures checked against reproducible verification code. Profile: https://formiq.jp/about/yuichi-saiki ## Features - Candlestick charting for 40 FX pairs and the dollar index (USDX) across M1 to MN1 timeframes, running entirely in the browser with no install. - A catalogue of 80+ technical indicators (moving averages, RSI, MACD, Bollinger Bands and others), plus drawing tools for trendlines and levels. - Replay practice: historical charts replayed forward so entries and exits can be practised with no money at risk. Visual replay auto-plays; manual trade steps candle by candle. - No-code backtesting: entry and exit rules are assembled from indicator conditions in dropdowns — no scripting language — and return win rate, profit factor and maximum drawdown. - Demo trading inside replay sessions: market and pending orders, stop loss and take profit, in JPY, USD or EUR account currency. - A trade journal and P&L calendar that record every simulated trade. - Fommy, an AI chart assistant that reads the current chart state and answers questions about trend, indicators and setups. - Strategy Library: describe a strategy in plain language and the AI turns it into backtestable rules you can save and re-run. - Backtested strategies export as an MQL4 expert advisor for MetaTrader 4, alongside CSV, JSON and PDF reports. - Price alert notifications. - Cloud sync of settings and data across devices, plus an iOS app that runs on the same account — one sign-in, one subscription, same charts, drawings and trade records in the browser and on iPhone. ## Limitations - Formiq is not a broker and does not execute real-money orders; all trading inside it is simulated. - The price feed is historical, not live — the most recent candles are typically about two weeks behind the present. - Coverage is FX pairs, gold (XAU/USD) and five indices — the dollar index, the Nasdaq 100, the Nikkei 225, the Dow 30 and the S&P 500; individual stocks, crypto and futures are not supported. ## Plans and limits - Guest (no account) — Free. Indicators: 1 at a time. Backtests: 1 per day. AI: Sign-in required. - Free plan (Google sign-in) — Free, no time limit, no credit card. Indicators: 2 at a time (catalogue of 80+). Backtests: 3 per day. AI: Fommy 10/day, AI strategy builder 1/day. - Unlimited — $19.99/month or $179.99/year. Indicators: Unlimited. Backtests: Unlimited. AI: Fommy 300/day, unlimited AI strategy builder. ## Chart data Chart data is built from a public historical tick datafeed: tick data aggregated into M1 candles and rolled up to higher timeframes. Several years of history are available. ## Platforms - Web: any modern browser on macOS, Windows, iPad or Chromebook. No install. - iOS: https://apps.apple.com/jp/app/formiq-fx-backtest-tester/id6759964075 - Settings and trade data sync across devices via the cloud. - The web and the iOS app are one product on one account: signing in with the same account on either side restores the same charts, drawings, trade records and subscription. An Unlimited plan bought in the browser is active in the app, and vice versa. ## Frequently asked questions ### Is it really free? Yes. Charts, 80+ indicators, replay practice, and backtesting are all available on the free plan. The AI assistant (Fommy) is also free for up to 10 messages a day — the Unlimited plan raises that to 300 a day and removes the indicator and backtest caps. ### How do I get started? Just sign in with your Google account — no credit card or personal info required. You'll be on the chart in seconds. ### How is this different from other charting tools? Formiq runs entirely in your browser — no install, no desktop app. It comes with replay practice and no-code backtesting built in, so you can test strategies without writing any code. ### What is replay practice? It replays historical charts in real time so you can practice entries and exits without risking real money. Two modes: visual replay (auto-play) and manual trade (step at your own pace). ### Do I need to code to backtest? No. Set entry/exit rules by picking indicator conditions from dropdowns. Results — win rate, profit factor, max drawdown — appear instantly. ### How far back does the chart data go? Several years of historical data are available. Note: this is not a real-time feed — the most recent data is typically available from about two weeks ago. ### What devices are supported? Mac, Windows, iPad, Chromebook — any device with a modern browser — plus the iPhone app. It is one Formiq account across both: sign in with the same account and your chart settings, drawings, trade records and plan carry over between the browser and the app. ## Billing questions ### Is the free plan really free forever? Yes. The free plan has no time limit. You can use core charting, 2 indicators, 3 backtests per day, and the trade journal indefinitely at no cost. ### What is the difference between the one-time purchase and the monthly plan? The one-time purchase is $129.00 paid once: it does not expire, does not renew, and is never billed again. The monthly plan is $19.99/month for as long as you keep it. The one-time price is about six months of the monthly plan, and it is the same Unlimited either way — the monthly plan exists so you can compare. ### How do I upgrade to Unlimited? Sign up for a free account, then go to Settings and click 'Upgrade'. Payment is handled securely via Stripe. ### Can I cancel anytime? The one-time purchase has nothing to cancel — there is no renewal to stop. On the monthly plan, cancel from your account settings at any time and you keep Unlimited access until the end of the current billing period. ### Does my plan also work in the iPhone app? Yes. The browser and the iOS app are one Formiq account, so Unlimited bought here — the one-time purchase included — is already active in the app once you sign in with the same account. There is nothing to buy twice, and your trades, drawings and settings move with you. ### Do you offer refunds? Fees already paid are non-refundable except as required by law. If you have an issue, contact support@formiq.jp. ## Contact Support: support@formiq.jp Contact form: https://formiq.jp/contact ## How Formiq compares to other tools Claims about tools other than Formiq were read on each vendor's own site; the date is given against each entry (the five tools on the overview were read on 2026-07-27). Prices for other tools are deliberately omitted, because their checkouts render in different currencies depending on where the request comes from. Comparison page: https://formiq.jp/compare/forex-backtesting-tools Alternative pages (English): https://formiq.jp/compare/fx-replay-alternative, https://formiq.jp/compare/forex-tester-alternative, https://formiq.jp/compare/free-forex-backtesting-no-download ### Formiq Runs on: Browser, plus an iOS app (install: Not required) Free tier: Free plan with no time limit: 2 indicators at once, 3 backtests per day, AI assistant 10×/day. Replay practice and the trade journal are uncapped. Replay: Yes — auto-play and candle-by-candle manual modes, on the free plan Better than Formiq at: Starting costs nothing and takes no setup — a browser tab and a Google sign-in. Best for: Building screen time and testing simple rules without installing anything or committing money first. Source: https://formiq.jp/pricing (read 2026-07-27) ### Forex Tester Online Runs on: Browser (install: Not required) Free tier: A free demo, advertised as "no credit card required". Full access is a paid subscription. Replay: Yes — it is the product's core function Better than Formiq at: Coverage well beyond FX — stocks, indices, commodities, ETFs and crypto are all testable in the same tool. Best for: Testing one approach across several asset classes, in a tool built for backtesting first. Source: https://forextester.com/ (read 2026-07-27) ### Forex Tester 5 (desktop) Runs on: Windows desktop (install: Required) Free tier: A free version limited to one hour of continuous testing and one month of data. The full licence is a one-time purchase. Replay: Yes — it is the product's core function Better than Formiq at: Depth of history — the vendor advertises 25 years of data — and it keeps working offline once installed. Best for: Long-horizon testing on a Windows machine, with a one-time payment rather than a subscription. Source: https://desktop.forextester.com/order (read 2026-07-27) ### TradingView Runs on: Browser, desktop app and mobile apps (install: Not required for the browser version) Free tier: The Basic plan is free and allows 2 indicators per chart, 1 chart per tab, 3 price alerts and 1 watchlist of up to 30 symbols, with ads. Its pricing table lists Bar Replay as available from the Essential plan upward, not on Basic. Replay: Bar Replay, on paid plans per the pricing table Better than Formiq at: The charting itself — drawing tools, alerting, market coverage, and a very large library of community scripts. Best for: Analysis and charting across many markets, where replay is one feature among many rather than the point. Source: https://jp.tradingview.com/pricing/ (read 2026-07-27) ### MT4 / MT5 Strategy Tester Runs on: Windows desktop (bundled with MetaTrader) (install: Required (MetaTrader itself)) Free tier: Free — it ships with the MetaTrader terminal. Running a strategy through it requires an Expert Advisor written in MQL. Replay: Visual mode, driven by an Expert Advisor Better than Formiq at: No ceiling on what a strategy can express, and what you test is what you can then run live on the same platform. Best for: Anyone who can write MQL, or whose strategy is going to end up as an EA regardless. Source: https://www.metatrader5.com/en/terminal/help/algotrading/testing (read 2026-07-27) ### FX Replay Runs on: Browser (install: Not required) Free tier: A free Beginner plan, listed on the pricing page as "2 sessions • 50 records • 1 week data retention • 1 month session duration", with 1 custom indicator and 1 screenshot and 1 checklist in the journal. Intermediate and Pro are billed monthly or annually. Replay: Yes — manual replay is the product's core function, on charts powered by TradingView Better than Formiq at: Market coverage beyond FX, and a journal built around screenshots and checklists, on TradingView charting. Best for: Discretionary traders who want to drill execution across several markets and keep a screenshot journal of every session. Source: https://fxreplay.com/pricing (read 2026-09-03) ### Forex Simulator (Soft4FX) Runs on: Inside MetaTrader 4 or 5 (install: Required (MetaTrader, then the simulator)) Free tier: A free demo limited, in the vendor's words, to "Only 5 trades per simulation", "Only 2 sample years of data" and "No saving/loading". The full licence is a one-time payment covering 1 user and 2 MetaTrader accounts, with the MT4 and MT5 versions both included. Replay: Yes — it is the product's core function, inside the MetaTrader terminal Better than Formiq at: Practice on the exact MetaTrader charts, indicators and templates you already trade with, on tick-level history. Best for: MetaTrader users who want replay practice without leaving the platform they trade on, bought once. Source: https://soft4fx.com/software/forex-simulator.php (read 2026-09-03) ## Price behaviour by currency pair Formiq aggregated public historical tick data into M1 candles and computed these figures over the complete calendar years 2016–2025 (generated 2026-07-26). Range means the high-to-low distance within the period, in pips. ### USD/JPY (dollar-yen) — US Dollar / Japanese Yen URL: https://formiq.jp/data/usdjpy Period: 2016-01-04 to 2025-12-31 (2,595 trading days) Average daily range: 100.5 pips (median 83.3 pips) Busiest hour: 14:00 UTC, averaging 27.8 pips Quietest hour: 21:00 UTC, averaging 10.7 pips Busiest session: London (07:00–16:00 UTC, averaging 71.5 pips). The three sessions overlap, so they do not sum to the daily range. Widest day on record: 2022-12-20, 690.8 pips ### EUR/USD (euro-dollar) — Euro / US Dollar URL: https://formiq.jp/data/eurusd Period: 2016-01-04 to 2025-12-31 (2,595 trading days) Average daily range: 77.4 pips (median 70.4 pips) Busiest hour: 14:00 UTC, averaging 24.5 pips Quietest hour: 22:00 UTC, averaging 6.7 pips Busiest session: London (07:00–16:00 UTC, averaging 63.0 pips). The three sessions overlap, so they do not sum to the daily range. Widest day on record: 2016-06-24, 425.9 pips ### GBP/USD (cable) — British Pound / US Dollar URL: https://formiq.jp/data/gbpusd Period: 2016-01-04 to 2025-12-31 (2,595 trading days) Average daily range: 108.1 pips (median 94.7 pips) Busiest hour: 14:00 UTC, averaging 32.5 pips Quietest hour: 04:00 UTC, averaging 10.4 pips Busiest session: London (07:00–16:00 UTC, averaging 88.5 pips). The three sessions overlap, so they do not sum to the daily range. Widest day on record: 2016-06-24, 1423.0 pips ### EUR/JPY (euro-yen) — Euro / Japanese Yen URL: https://formiq.jp/data/eurjpy Period: 2016-01-04 to 2025-12-31 (2,595 trading days) Average daily range: 111.1 pips (median 95.6 pips) Busiest hour: 14:00 UTC, averaging 30.9 pips Quietest hour: 21:00 UTC, averaging 13.3 pips Busiest session: London (07:00–16:00 UTC, averaging 83.6 pips). The three sessions overlap, so they do not sum to the daily range. Widest day on record: 2016-06-24, 1006.3 pips ### GBP/JPY (pound-yen) — British Pound / Japanese Yen URL: https://formiq.jp/data/gbpjpy Period: 2016-01-04 to 2025-12-31 (2,595 trading days) Average daily range: 153.9 pips (median 131.4 pips) Busiest hour: 14:00 UTC, averaging 42.9 pips Quietest hour: 20:00 UTC, averaging 19.3 pips Busiest session: London (07:00–16:00 UTC, averaging 117.9 pips). The three sessions overlap, so they do not sum to the daily range. Widest day on record: 2016-06-24, 2093.2 pips ### AUD/USD (the Aussie) — Australian Dollar / US Dollar URL: https://formiq.jp/data/audusd Period: 2016-01-04 to 2025-12-31 (2,595 trading days) Average daily range: 66.5 pips (median 59.4 pips) Busiest hour: 14:00 UTC, averaging 19.3 pips Quietest hour: 20:00 UTC, averaging 7.9 pips Busiest session: London (07:00–16:00 UTC, averaging 47.2 pips). The three sessions overlap, so they do not sum to the daily range. Widest day on record: 2020-03-19, 456.3 pips ### AUD/JPY (Aussie yen) — Australian Dollar / Japanese Yen URL: https://formiq.jp/data/audjpy Period: 2016-01-04 to 2025-12-31 (2,595 trading days) Average daily range: 90.6 pips (median 78.0 pips) Busiest hour: 14:00 UTC, averaging 24.4 pips Quietest hour: 20:00 UTC, averaging 11.7 pips Busiest session: London (07:00–16:00 UTC, averaging 60.7 pips). The three sessions overlap, so they do not sum to the daily range. Widest day on record: 2016-06-24, 764.2 pips ### USD/CAD (the loonie) — US Dollar / Canadian Dollar URL: https://formiq.jp/data/usdcad Period: 2016-01-04 to 2025-12-31 (2,595 trading days) Average daily range: 86.3 pips (median 76.3 pips) Busiest hour: 14:00 UTC, averaging 30.3 pips Quietest hour: 04:00 UTC, averaging 8.2 pips Busiest session: New York (12:00–21:00 UTC, averaging 69.6 pips). The three sessions overlap, so they do not sum to the daily range. Widest day on record: 2020-03-18, 484.6 pips ### USD/CHF (the swissie) — US Dollar / Swiss Franc URL: https://formiq.jp/data/usdchf Period: 2016-01-04 to 2025-12-31 (2,595 trading days) Average daily range: 65.4 pips (median 58.7 pips) Busiest hour: 14:00 UTC, averaging 21.5 pips Quietest hour: 04:00 UTC, averaging 6.3 pips Busiest session: London (07:00–16:00 UTC, averaging 54.0 pips). The three sessions overlap, so they do not sum to the daily range. Widest day on record: 2025-04-10, 370.3 pips ### NZD/USD (the kiwi) — New Zealand Dollar / US Dollar URL: https://formiq.jp/data/nzdusd Period: 2016-01-04 to 2025-12-31 (2,595 trading days) Average daily range: 63.5 pips (median 56.9 pips) Busiest hour: 14:00 UTC, averaging 18.0 pips Quietest hour: 20:00 UTC, averaging 8.6 pips Busiest session: London (07:00–16:00 UTC, averaging 44.8 pips). The three sessions overlap, so they do not sum to the daily range. Widest day on record: 2020-03-19, 447.4 pips --- # 日本語 (Japanese) ## 定義 Formiqは、チャート分析・リプレイ練習・ノーコードバックテストをブラウザだけで完結できる無料のFXツールです。AIアシスタント「Fommy」を標準搭載しています。 タグライン: 分析する。検証する。AIに聞く。 想定ユーザー: MetaTraderをインストールしたり有料の検証ソフトを買ったりせずに、練習量と手法検証を確保したい個人FXトレーダー。 ## 検証記事の著者 斉木勇一: 斉木勇一は、Formiqのバックテスト検証記事を執筆し、公開する数値の検証を担当しています。各記事では、売買ルール、検証期間、コスト、試した設定数、限界を明記し、本文の数値を再実行可能な検算コードと照合しています。 プロフィール: https://formiq.jp/ja/about/yuichi-saiki ## 機能 - 40のFX通貨ペアとドルインデックス（USDX）のローソク足チャートをM1〜MN1で表示。インストール不要でブラウザだけで動作します。 - 移動平均・RSI・MACD・ボリンジャーバンドなど80種類以上のインジケーターと、トレンドラインや水平線などの描画ツール。 - リプレイ練習：過去チャートを先に向かって再生し、実際の資金を使わずにエントリーとエグジットを練習できます。ビジュアル再生は自動、手動トレードは1本ずつ進められます。 - ノーコードバックテスト：エントリー・エグジット条件をドロップダウンから組み立てるだけで、勝率・プロフィットファクター・最大ドローダウンが出ます。スクリプト記述は不要です。 - デモトレード：リプレイ練習の中で成行・指値注文、損切り・利確に対応。口座通貨はJPY／USD／EURから選べます。 - 全ての仮想トレードを記録するトレード記録と損益カレンダー。 - Fommy：現在のチャート状態を読み取り、トレンドやインジケーター、セットアップについて答えるAIチャートアシスタント。 - 手法ライブラリ: 手法を自然言語で説明すると、AIが検証可能なルールに変換し、保存・再実行できます。 - 検証した手法はMetaTrader 4向けのMQL4 EA、またはMetaTrader 5向けのMQL5 EAとして書き出せます。CSV・JSON・PDFのレポート出力にも対応しています。 - 価格アラート通知。 - 設定とデータのクラウド同期、および同じアカウントで使えるiOSアプリ。アカウントもサブスクもひとつで、チャート設定・描画・トレード記録はブラウザとiPhoneで共通です。 ## できないこと - Formiqは証券会社・FX業者ではなく、実際の資金による注文執行は行いません。ツール内の取引はすべてシミュレーションです。 - 価格データはヒストリカルであり、リアルタイム配信ではありません。最新のローソク足はおおむね2週間前までです。 - 対応はFX通貨ペア・金（XAU/USD）・指数5種（ドルインデックス、ナスダック100、日経225、ダウ30、S&P500）で、個別株・暗号資産・先物には対応していません。 ## プランと上限 - ゲスト（未登録） — 無料。インジケーター: 同時表示1つ。バックテスト: 1日1回。AI: 要ログイン。 - 無料プラン（Googleログイン） — 無料・期限なし・クレジットカード不要。インジケーター: 同時表示2つ（カタログは80種類以上）。バックテスト: 1日3回。AI: Fommy 1日10回／AI手法作成 1日1回。 - Unlimited — ¥4,000／月、または¥36,000／年。インジケーター: 無制限。バックテスト: 無制限。AI: Fommy 1日300回／AI手法作成 無制限。 ## チャートデータ チャートデータは公開されているヒストリカルティックデータを元にしています。ティックデータをM1に集約し、上位足へ積み上げています。数年分の履歴を参照できます。 ## よくある質問 ### 本当に無料で使えますか？ はい。チャート・80種類以上のインジケーター・リプレイ練習・バックテストはすべて無料プランで利用できます。AIアシスタント（Fommy）も1日10回まで無料で使え、Unlimitedプランなら1日300回まで、インジケーターやバックテストは無制限になります。 ### アカウント登録に何が必要ですか？ Googleアカウントだけで始められます。クレジットカードや個人情報の入力は不要です。 ### 他のチャートツールとの違いは？ Formiqはインストール不要・ブラウザだけで完結します。リプレイ練習とノーコードバックテストを標準搭載しており、プログラミングなしで手法を検証できます。 ### リプレイ練習とは何ですか？ 過去チャートをリアルタイムのように再生し、実際のお金を使わずにエントリー・エグジットの練習ができる機能です。ビジュアル再生（自動でバーが進む）と手動トレード（自分のペースで操作）の2モードがあります。 ### バックテストにプログラミングは必要ですか？ 不要です。インジケーター条件をドロップダウンで選ぶだけでエントリー・エグジットルールを設定できます。結果は勝率・プロフィットファクター・最大ドローダウンなどで即座に確認できます。 ### チャートデータはいつまでの期間が見れますか？ 数年分のヒストリカルデータに対応しています。ただしリアルタイム配信ではないため、最新のデータは表示できません。おおむね2週間前からのデータが閲覧可能です。 ### どのデバイスで使えますか？ Mac・Windows・iPad・ChromebookなどモダンブラウザがあればどこでもOKです。さらにiPhoneアプリもあり、ブラウザとアプリはひとつのFormiqアカウントで共通です。同じアカウントでログインすれば、チャート設定・描画・トレード記録・プランがそのまま引き継がれます。 ## 料金に関する質問 ### 無料プランは本当にずっと無料ですか？ はい。無料プランに期限はありません。コアチャート機能、インジケーター2つ、1日3回のバックテスト、トレード記録を無期限でご利用いただけます。 ### 買い切りプランと月額プランの違いは？ 買い切りプランは¥19,800の1回払いで、有効期限も更新もなく、2回目の請求は発生しません。月額プランはご利用のあいだ¥4,000/月です。買い切りは月額のおよそ5ヶ月分にあたり、使える機能はどちらも同じUnlimitedです。月額プランは比較のために残しています。 ### Unlimitedプランにアップグレードするには？ 無料アカウントを作成後、設定画面から「アップグレード」をクリックしてください。お支払いはStripeで安全に処理されます。 ### いつでも解約できますか？ 買い切りプランは解約するものがありません（止めるべき更新が無いためです）。月額プランはアカウント設定からいつでも解約でき、現在の請求期間の終了まではUnlimitedの機能をご利用いただけます。 ### iPhoneアプリでもプランは使えますか？ 使えます。ブラウザとiOSアプリはひとつのFormiqアカウントで共通なので、こちらで購入したUnlimitedは、買い切りプランを含め、アプリで同じアカウントにログインすればそのまま有効です。二重に購入する必要はありません。トレード記録・描画・設定も一緒に引き継がれます。 ### 返金はできますか？ 法律で定められた場合を除き、お支払い済みの料金は返金されません。問題がある場合は support@formiq.jp までご連絡ください。 ## お問い合わせ サポート: support@formiq.jp お問い合わせフォーム: https://formiq.jp/ja/contact ## 他ツールとの比較 Formiq以外のツールに関する記述は各社の公式サイトで確認したものです（確認日は各項目に記載。総覧の5ツールは2026-07-27）。他社の価格は掲載していません（表示通貨が閲覧地域によって変わるため）。比較ページ: https://formiq.jp/ja/compare/forex-backtesting-tools 英語のみの個別比較ページ: https://formiq.jp/compare/fx-replay-alternative, https://formiq.jp/compare/forex-tester-alternative, https://formiq.jp/compare/free-forex-backtesting-no-download ### Formiq 動作環境: ブラウザ、およびiOSアプリ（インストール: 不要） 無料で使える範囲: 期限なしの無料プラン。インジケーター同時2つ、バックテスト1日3回、AIアシスタント1日10回。リプレイ練習とトレードジャーナルは無制限。 リプレイ: あり（自動再生と1本送りの手動モード、無料プランで利用可） このツールが優れている点: 始めるまでの手間が最小。ブラウザのタブとGoogleログインだけで、インストールも支払いも要らない。 向いている人: インストールも支払いもせずに、練習量を積んで単純なルールを検証したい人。 出典: https://formiq.jp/pricing（確認日 2026-07-27） ### Forex Tester Online 動作環境: ブラウザ（インストール: 不要） 無料で使える範囲: 無料デモあり（「クレジットカード不要」と明記）。フル機能は有料サブスクリプション。 リプレイ: あり（製品の中核機能） このツールが優れている点: FX以外への対応幅。株式・指数・商品・ETF・暗号資産まで同じツールで検証できる。 向いている人: 検証専用に作られたツールで、複数の資産クラスにまたがって同じ手法を試したい人。 出典: https://forextester.com/（確認日 2026-07-27） ### Forex Tester 5（デスクトップ版） 動作環境: Windowsデスクトップ（インストール: 必要） 無料で使える範囲: 無料版あり（連続テスト1時間・データ1か月の制限）。フル版は買い切りライセンス。 リプレイ: あり（製品の中核機能） このツールが優れている点: 履歴の深さ（公式表記で25年分のデータ）と、インストール後はオフラインでも動くこと。 向いている人: Windows機で長期のデータを扱いたい、サブスクではなく買い切りが良い、という人。 出典: https://desktop.forextester.com/order（確認日 2026-07-27） ### TradingView 動作環境: ブラウザ、デスクトップアプリ、モバイルアプリ（インストール: ブラウザ版は不要） 無料で使える範囲: 無料のBasicプランは、チャートごとにインジケーター2つ、タブごとにチャート1枚、価格アラート3件、ウォッチリスト1つ（30銘柄まで）、広告あり。公式の料金比較表では、バーリプレイはEssential以上の有料プランの機能として記載されている（Basicは対象外）。 リプレイ: バーリプレイあり（料金表の記載では有料プラン） このツールが優れている点: チャートそのものの完成度。描画ツール、アラート、対応市場の広さ、そして膨大なコミュニティスクリプト資産。 向いている人: 多くの市場を横断して分析したい人。リプレイは数ある機能のひとつ、という使い方に向く。 出典: https://jp.tradingview.com/pricing/（確認日 2026-07-27） ### MT4 / MT5 ストラテジーテスター 動作環境: Windowsデスクトップ（MetaTraderに標準搭載）（インストール: 必要（MetaTrader本体）） 無料で使える範囲: 無料（MetaTrader本体に同梱）。ただし検証を回すにはMQLで書いたEA（エキスパートアドバイザ）が必要。 リプレイ: ビジュアルモードあり（EAで駆動） このツールが優れている点: 戦略の表現力に上限がないこと。そして検証したものを、同じプラットフォームでそのまま実運用につなげられること。 向いている人: MQLを書ける人、あるいは最終的にEAとして動かすことが決まっている人。 出典: https://www.metatrader5.com/en/terminal/help/algotrading/testing（確認日 2026-07-27） ### FX Replay 動作環境: ブラウザ（インストール: 不要） 無料で使える範囲: 無料のBeginnerプランあり。料金ページの表記は「セッション2つ・記録50件・データ保持1週間・セッション期間1か月」で、カスタムインジケーター1つ、ジャーナルのスクリーンショットとチェックリストは各1つ。IntermediateとProは月額または年額。 リプレイ: あり（手動リプレイが製品の中核。チャートはTradingView製） このツールが優れている点: FX以外の市場への対応幅と、スクリーンショットとチェックリストを軸にしたジャーナル。チャートはTradingView製。 向いている人: 複数の市場で執行の練習を積み、セッションごとにスクリーンショット付きの記録を残したい裁量トレーダー。 出典: https://fxreplay.com/pricing（確認日 2026-09-03） ### Forex Simulator（Soft4FX） 動作環境: MetaTrader 4／5の内部（インストール: 必要（MetaTrader本体とシミュレーター）） 無料で使える範囲: 無料デモあり。制限は公式表記で「1シミュレーションにつき5トレードまで」「サンプルデータ2年分のみ」「保存・読み込み不可」。フル版は1ユーザー・MetaTrader口座2つまでの買い切りライセンスで、MT4版とMT5版の両方が含まれます。 リプレイ: あり（製品の中核機能。MetaTraderの中で動く） このツールが優れている点: すでに取引に使っているMetaTraderのチャート・インジケーター・テンプレートそのもので、ティック単位の履歴を使って練習できること。 向いている人: 取引しているMetaTraderから離れずにリプレイ練習をしたい、買い切りが良いという人。 出典: https://soft4fx.com/software/forex-simulator.php（確認日 2026-09-03） ## 通貨ペア別の値動き Formiqが公開ヒストリカルティックデータをM1に集約し、2016〜2025年の完全な暦年のみを対象に算出しました（2026-07-26時点）。変動幅は日中の高値と安値の差をpipsで表したものです。 ### ドル円 — 米ドル／日本円 URL: https://formiq.jp/ja/data/usdjpy 集計期間: 2016-01-04〜2025-12-31（2,595営業日） 1日の平均変動幅: 100.5 pips（中央値 83.3 pips） 最も動く時間帯: 14:00 UTC、平均 27.8 pips 最も動かない時間帯: 21:00 UTC、平均 10.7 pips 最も動くセッション: ロンドン（07:00–16:00 UTC、平均 71.5 pips）。3つのセッションは時間帯が重なるため、合計は1日の変動幅になりません。 過去最大の変動日: 2022-12-20、690.8 pips ### ユーロドル — ユーロ／米ドル URL: https://formiq.jp/ja/data/eurusd 集計期間: 2016-01-04〜2025-12-31（2,595営業日） 1日の平均変動幅: 77.4 pips（中央値 70.4 pips） 最も動く時間帯: 14:00 UTC、平均 24.5 pips 最も動かない時間帯: 22:00 UTC、平均 6.7 pips 最も動くセッション: ロンドン（07:00–16:00 UTC、平均 63.0 pips）。3つのセッションは時間帯が重なるため、合計は1日の変動幅になりません。 過去最大の変動日: 2016-06-24、425.9 pips ### ポンドドル — 英ポンド／米ドル URL: https://formiq.jp/ja/data/gbpusd 集計期間: 2016-01-04〜2025-12-31（2,595営業日） 1日の平均変動幅: 108.1 pips（中央値 94.7 pips） 最も動く時間帯: 14:00 UTC、平均 32.5 pips 最も動かない時間帯: 04:00 UTC、平均 10.4 pips 最も動くセッション: ロンドン（07:00–16:00 UTC、平均 88.5 pips）。3つのセッションは時間帯が重なるため、合計は1日の変動幅になりません。 過去最大の変動日: 2016-06-24、1423.0 pips ### ユーロ円 — ユーロ／日本円 URL: https://formiq.jp/ja/data/eurjpy 集計期間: 2016-01-04〜2025-12-31（2,595営業日） 1日の平均変動幅: 111.1 pips（中央値 95.6 pips） 最も動く時間帯: 14:00 UTC、平均 30.9 pips 最も動かない時間帯: 21:00 UTC、平均 13.3 pips 最も動くセッション: ロンドン（07:00–16:00 UTC、平均 83.6 pips）。3つのセッションは時間帯が重なるため、合計は1日の変動幅になりません。 過去最大の変動日: 2016-06-24、1006.3 pips ### ポンド円（ポン円） — 英ポンド／日本円 URL: https://formiq.jp/ja/data/gbpjpy 集計期間: 2016-01-04〜2025-12-31（2,595営業日） 1日の平均変動幅: 153.9 pips（中央値 131.4 pips） 最も動く時間帯: 14:00 UTC、平均 42.9 pips 最も動かない時間帯: 20:00 UTC、平均 19.3 pips 最も動くセッション: ロンドン（07:00–16:00 UTC、平均 117.9 pips）。3つのセッションは時間帯が重なるため、合計は1日の変動幅になりません。 過去最大の変動日: 2016-06-24、2093.2 pips ### 豪ドル米ドル（オージードル） — 豪ドル／米ドル URL: https://formiq.jp/ja/data/audusd 集計期間: 2016-01-04〜2025-12-31（2,595営業日） 1日の平均変動幅: 66.5 pips（中央値 59.4 pips） 最も動く時間帯: 14:00 UTC、平均 19.3 pips 最も動かない時間帯: 20:00 UTC、平均 7.9 pips 最も動くセッション: ロンドン（07:00–16:00 UTC、平均 47.2 pips）。3つのセッションは時間帯が重なるため、合計は1日の変動幅になりません。 過去最大の変動日: 2020-03-19、456.3 pips ### 豪ドル円（オージー円） — 豪ドル／日本円 URL: https://formiq.jp/ja/data/audjpy 集計期間: 2016-01-04〜2025-12-31（2,595営業日） 1日の平均変動幅: 90.6 pips（中央値 78.0 pips） 最も動く時間帯: 14:00 UTC、平均 24.4 pips 最も動かない時間帯: 20:00 UTC、平均 11.7 pips 最も動くセッション: ロンドン（07:00–16:00 UTC、平均 60.7 pips）。3つのセッションは時間帯が重なるため、合計は1日の変動幅になりません。 過去最大の変動日: 2016-06-24、764.2 pips ### 米ドルカナダドル（ドルカナダ） — 米ドル／カナダドル URL: https://formiq.jp/ja/data/usdcad 集計期間: 2016-01-04〜2025-12-31（2,595営業日） 1日の平均変動幅: 86.3 pips（中央値 76.3 pips） 最も動く時間帯: 14:00 UTC、平均 30.3 pips 最も動かない時間帯: 04:00 UTC、平均 8.2 pips 最も動くセッション: ニューヨーク（12:00–21:00 UTC、平均 69.6 pips）。3つのセッションは時間帯が重なるため、合計は1日の変動幅になりません。 過去最大の変動日: 2020-03-18、484.6 pips ### 米ドルスイスフラン（ドルスイス） — 米ドル／スイスフラン URL: https://formiq.jp/ja/data/usdchf 集計期間: 2016-01-04〜2025-12-31（2,595営業日） 1日の平均変動幅: 65.4 pips（中央値 58.7 pips） 最も動く時間帯: 14:00 UTC、平均 21.5 pips 最も動かない時間帯: 04:00 UTC、平均 6.3 pips 最も動くセッション: ロンドン（07:00–16:00 UTC、平均 54.0 pips）。3つのセッションは時間帯が重なるため、合計は1日の変動幅になりません。 過去最大の変動日: 2025-04-10、370.3 pips ### NZドル米ドル（キウイドル） — NZドル／米ドル URL: https://formiq.jp/ja/data/nzdusd 集計期間: 2016-01-04〜2025-12-31（2,595営業日） 1日の平均変動幅: 63.5 pips（中央値 56.9 pips） 最も動く時間帯: 14:00 UTC、平均 18.0 pips 最も動かない時間帯: 20:00 UTC、平均 8.6 pips 最も動くセッション: ロンドン（07:00–16:00 UTC、平均 44.8 pips）。3つのセッションは時間帯が重なるため、合計は1日の変動幅になりません。 過去最大の変動日: 2020-03-19、447.4 pips --- # Articles ## English articles ### Do Trend Filters Remove False Signals? 1,596 Backtests URL: https://formiq.jp/blog/false-signal-filters Language: en Published: 2026-09-10 Updated: 2026-09-10 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Trend filters, ADX, Moving average slope, False signals, Backtest, USDJPY Evidence: https://formiq.jp/evidence/false-signal-filters Evidence JSON: https://formiq.jp/evidence/false-signal-filters/data.json Also available in: ja — https://formiq.jp/ja/blog/false-signal-filters "The signal fired, but the market was ranging, so it failed. Read ADX, and only trade when it is above 25." Almost every article about removing false signals is built on that sentence. This series has tried the ADX filter on three indicators, GMMA, CCI and the Alligator, and lost pips on all four attempts. Three indicators going wrong does not answer whether the filter works anywhere else. So the same six filters went onto **all 38 settings this site has published**, across three timeframes and two years. With the unfiltered runs, that is 1,596 backtests. **The three ADX levels lowered the count of settings profitable in both 2024 and 2025 on every timeframe. All nine cells.** The 15-minute chart falls from 13 settings to 6, 4 and 7; the hourly from 27 to 18, 10 and 12; the 4-hour from 16 to 14, 11 and 12. The trades were measured the same way. **Whether a filter removed the losing ones was tested against removing the same number at random. Of 1,348 comparisons, 72 fell outside the 95% range, and chance alone puts about 67 outside.** ## ADX and moving-average slope: what the settings mean Both filters come from indicators already in the condition list. **ADX** draws three lines in a panel below price: ADX itself for trend strength, +DI for upward pressure and -DI for downward pressure. The scale runs 0 to 100, and a higher reading is read as a stronger trend. MT4 draws no level lines by default, so the 20 and 25 lines have to be added by hand. **Moving-average slope** draws one ordinary moving average on the price chart and reads only whether the line is higher or lower than it was a few bars ago. The line is visible; the rising-or-falling judgement is not drawn anywhere. | Condition | Parameter | Default | What it does | |---|---|---|---| | ADX | Period | 14 | How many bars the calculation reads | | ADX | Minimum level | 25 | No trade unless ADX is at or above this | | ADX | Reading | Level only | Level only, DI comparison, or DI cross | | MA slope | Period | 20 | Length of the average being drawn | | MA slope | Method | SMA | Simple, exponential, weighted and so on | | MA slope | Lookback | 5 | How many bars back the comparison runs | | MA slope | Direction | Either | Buy only when rising, sell only when falling, or both | ### How each one behaves as a filter Set the ADX reading to level only, and **bars at or above the minimum level allow a buy and a sell, while bars below allow neither**. That is what filtering on trend strength means. The DI comparison allows only buys while +DI is above -DI, and only sells while it is below. The moving-average slope set to either direction works the same way: rising allows buys, falling allows sells. **Neither of those filters on strength. They filter on direction.** Put the filter in the condition list beside the rule already there, set the match to all conditions, and only bars where both point the same way survive. ## How many bars each filter throws away Before judging a filter, count what it can remove. A condition true on most bars removes nothing. Share of 2025 bars where each filter allowed a buy: | Filter | 15-minute | Hourly | 4-hour | |---|---|---|---| | ADX 20 or more | 61.22% | 66.33% | 67.20% | | ADX 25 or more | 40.51% | 47.24% | 37.58% | | ADX 30 or more | 25.15% | 31.40% | 17.58% | | +DI above -DI | 49.85% | 45.87% | 41.24% | | 50-bar average rising | 50.09% | 48.35% | 55.16% | | 200-bar average rising | 48.50% | 54.61% | 60.81% | The ADX levels behave as expected. At 25 they admit 37.58% to 47.24% of bars, and at 30 that falls to between 17.58% and 31.40%. **The three direction filters discard no bars whatsoever.** Either +DI is above -DI or it is below; the average is either rising or falling. The percentages here are the share allowing a buy, and every remaining bar allows a sell. **What they remove is the combination of wanting to buy on a falling bar.** ## What actually follows a high ADX reading The premise is that a strong trend continues. Every 2025 bar was sorted into four ADX bands and measured against the close ten bars later. | Timeframe | ADX | Bars | Move over ten bars | In ATR | Direction continued | |---|---|---|---|---|---| | 15-minute | Under 20 | 9,654 | 19.81 pips | 1.86 | 50.58% | | 15-minute | Over 30 | 6,264 | 20.92 pips | 1.65 | 48.77% | | Hourly | Under 20 | 2,096 | 42.45 pips | 1.91 | 50.52% | | Hourly | Over 30 | 1,947 | 44.52 pips | 1.75 | 47.82% | | 4-hour | Under 20 | 519 | 82.66 pips | 1.81 | 51.64% | | 4-hour | Over 30 | 283 | 95.61 pips | 1.84 | 43.82% | "Direction continued" is the share of bars where the move over the five bars into the bar and the move over the ten bars out of it carried the same sign. **The move does grow a little**, from 19.81 to 20.92 pips on the 15-minute chart and from 82.66 to 95.61 on the 4-hour. Divide the same distance by the bar's own ATR and the 15-minute figure falls from 1.86 to 1.65 and the hourly from 1.91 to 1.75, while the 4-hour rises from 1.81 to 1.84. **A bar with a high ADX reading is itself a larger bar, so the extra pips are mostly the extra bar size.** Direction is the clearer answer. Across two years and three timeframes, all 24 bands landed between 43.82% and 53.45%. Readings over 30 continued more often than readings under 20 in one of the six timeframe-and-year cells: the hourly chart in 2024, at 53.45% against 51.16%. ## How this was measured | Item | Value | |---|---| | Instrument | USD/JPY | | Timeframes | 15-minute, 1-hour, 4-hour | | Period | 1 January 2024 to 31 December 2025 | | Base settings | The 38 settings published in these articles | | Filters | ADX 20, 25 and 30 or more; +DI above -DI; 50-bar and 200-bar averages rising | | Where the filter goes | Entry rules only; exits stay as published | | Control | 1,000 random draws of the filtered count from the unfiltered trades | | Runs | 38 settings x 7 conditions x 3 timeframes x 2 years = 1,596 | | Spread | 0.3 pips, fixed | | Lot size | 0.1 | | Fill | Close of the signal bar | **The base settings were not chosen for this article.** Each published article carries a "run this setting" button, and those are the rules used here. At the time of this run there were 43 of those buttons, and three are left out. The DMI/ADX article already uses ADX, and one rule set holds one ADX configuration. The morning-star and evening-star article joins its two conditions with OR, so a third condition would widen the rule rather than narrow it. This article's own button is a setting with a filter already on it, so it belongs to the results rather than to the inputs. Of the remaining 40, the three Bollinger articles publish identical rules (period 20, 2σ, breakout), which leaves 38. **Exits were not touched.** The advice is about entries, and changing the exit at the same time would measure the exit. ## The same six filters on 38 settings Settings that ended the year with positive pips, out of 38: | Filter | 15-min 2025 | Hourly 2025 | 4-hour 2025 | 15-min 2024 | Hourly 2024 | 4-hour 2024 | |---|---|---|---|---|---|---| | None | 20 | 29 | 20 | 17 | 34 | 31 | | ADX 20 or more | 11 | 26 | 22 | 16 | 27 | 27 | | ADX 25 or more | 11 | 22 | 23 | 14 | 19 | 23 | | ADX 30 or more | 16 | 20 | 21 | 14 | 25 | 24 | | +DI above -DI | 21 | 24 | 13 | 25 | 32 | 33 | | 50-bar average rising | 20 | 28 | 18 | 34 | 30 | 35 | | 200-bar average rising | 22 | 14 | 28 | 21 | 35 | 31 | Read one year at a time and the answer changes cell by cell. ADX 25 took the 4-hour chart from 20 settings to 23 in 2025, and the 50-bar slope took the 15-minute chart from 17 to 34 in 2024. Asking for both years at once takes most of that swing out. **Settings profitable in 2024 and in 2025:** | Filter | 15-minute | Hourly | 4-hour | |---|---|---|---| | None | 13 | 27 | 16 | | ADX 20 or more | 6 | 18 | 14 | | ADX 25 or more | 4 | 10 | 11 | | ADX 30 or more | 7 | 12 | 12 | | +DI above -DI | 15 | 22 | 12 | | 50-bar average rising | 18 | 23 | 17 | | 200-bar average rising | 16 | 13 | 21 | **Every one of the nine ADX cells is below the unfiltered count.** The largest drop is the hourly chart at ADX 25, from 27 settings to 10. The direction filters land on both sides. The 50-bar slope raised the count on the 15-minute chart (13 to 18) and the 4-hour (16 to 17) and lowered it on the hourly (27 to 23). The 200-bar slope also raised the 15-minute and 4-hour counts and lowered the hourly one, from 27 to 13. ## The win rate rises and the pips fall One setting, in detail. The MACD 12/26/9 signal-line crossover from that article, with ADX 25 added, on the hourly chart: | Year | Condition | Trades | Win rate | Pips per trade | Annual pips | |---|---|---|---|---|---| | 2025 | No filter | 451 | 38.58% | +4.93 | +2,223.6 | | 2025 | ADX 25 or more | 204 | 40.69% | +2.21 | +451.0 | | 2024 | No filter | 423 | 39.95% | +7.45 | +3,152.1 | | 2024 | ADX 25 or more | 168 | 46.43% | +8.03 | +1,349.2 | **The win rate rose in both years**, by 2.11 points in 2025 and 6.48 points in 2024. **The annual pips fell in both**, by 1,772.6 and 1,802.9. In 2024 even the pips per trade improved, from 7.45 to 8.03. The year still lost roughly 1,800 pips because the trade count went from 423 to 168. Gaining 0.58 pips on each of 168 trades does not replace the 255 trades that are gone. ## Were the removed trades the losing ones A filter that works removes the losers, so the trades it keeps should beat the full set on pips per trade. Take the same MACD run. On the hourly chart in 2025 it took 451 trades unfiltered; 204 of them were on bars where ADX was at 25 or above, and the other 247 are what the filter removes. | Trades | Count | Win rate | Pips per trade | |---|---|---|---| | All of them, unfiltered | 451 | 38.58% | +4.93 | | Kept by ADX 25 or more | 204 | 40.69% | +2.21 | | Removed by ADX 25 or more | 247 | 36.84% | +7.18 | **The 247 removed trades made +7.18 pips each and the 204 kept ones made +2.21.** The kept side has the higher win rate and the removed side has the money. Removing trades moves that average by itself, though: a smaller sample wanders further from its own mean. **So the comparison is against removing the same number at random.** Drawing 204 of those 451 a thousand times puts pips per trade between -1.38 and +11.55, and the filter's +2.21 sits inside. Win rate lands between 34.31% and 43.63%, and 40.69% sits inside that too. 38 settings times six filters times three timeframes times two years is 1,368 comparisons. In 20 of them the filter kept everything or nothing, leaving nothing to draw against. The other 1,348: | Filter | Compared | Kept beat the full set | Outside the range | Of those, above | |---|---|---|---|---| | ADX 20 or more | 225 | 95 | 11 | 1 | | ADX 25 or more | 227 | 92 | 15 | 8 | | ADX 30 or more | 226 | 103 | 15 | 13 | | +DI above -DI | 215 | 120 | 8 | 4 | | 50-bar average rising | 227 | 147 | 21 | 19 | | 200-bar average rising | 228 | 149 | 2 | 2 | | Total | 1,348 | 706 | 72 | 47 | **Pips per trade beat the full set in 706 comparisons, 52.4%.** A coin lands near the same number. **72 comparisons fell outside the range. The range covers the middle 95%, so sorting nothing at all still puts about 67 outside.** Repeating the measurement on win rate instead of pips gives 67 outside, 44 above and 23 below. These 1,348 comparisons are not independent of one another: the same trades are cut six ways, and the timeframes and years overlap. So 72 against 67 is not a significance test. It says the sorting a filter does is not large enough to separate from chance. ## The settings a filter helped were the ones already losing If the filters sort nothing, something still moved the annual results. The 38 settings were ranked by their unfiltered pips and split at the median. **For each setting, the best of the six filters was chosen after seeing the year.** A filter that cannot help with hindsight cannot help at all. | Timeframe and year | Bottom 19 improved | Median gain | Top 19 improved | Median gain | |---|---|---|---|---| | 15-minute 2025 | 19 | +891.6 pips | 10 | +35.9 pips | | 15-minute 2024 | 19 | +2,228.2 pips | 18 | +976.4 pips | | Hourly 2025 | 17 | +413.1 pips | 1 | -462.1 pips | | Hourly 2024 | 17 | +519.2 pips | 8 | -177.4 pips | | 4-hour 2025 | 19 | +1,012.5 pips | 17 | +248.7 pips | | 4-hour 2024 | 19 | +911.2 pips | 9 | -66.3 pips | **The 19 settings that lost money unfiltered improved in 17 to 19 cases in all six timeframe-and-year cells. The 19 that made money improved in 1 to 18.** The hourly chart in 2025 shows it plainly. The largest gain went to the fair value gap setting, which moved from -3,491.7 pips over 174 trades to -2,230.8 over 83 with ADX 30, an improvement of 1,260.9 pips that is still a losing year. The largest loss went to WaveTrend, which fell from +3,283.4 pips over 867 trades to +1,955.5 over 436 even with the best of the six chosen afterwards. **A filter lifted the settings that were losing and pushed down the settings that were making money.** A filter that sorted trades would work the same way on either half. The count that left the random range said the same thing from a different angle. ## Does last year's filter work this year Every article in this series takes the best choice from one year and applies it to the other. **The filter with the highest 2024 pips was picked for each setting, then run in 2025.** | Timeframe | Settings where 2024 preferred no filter | Beat no filter in 2025 | Ended 2025 profitable | |---|---|---|---| | 15-minute | 1 of 38 | 19 of 38 | 18 of 38 | | Hourly | 13 of 38 | 5 of 38 | 24 of 38 | | 4-hour | 10 of 38 | 14 of 38 | 19 of 38 | The first column is the part worth pausing on: **looking only at 2024, some filter appeared to help in 90 of 114 settings.** Carried into 2025, that choice beat leaving the filter off in 38 of 114. On the hourly chart it was 5 of 38. ## Why the direction filters only worked in 2024 One cell does stand clear. On the 15-minute chart in 2024, the 50-bar slope improved 36 of 38 settings and moved the median year from -205.9 pips to +1,157.3, the largest improvement count among the 36 combinations of six filters, three timeframes and two years. USD/JPY closed 2024 at 157.1975 against 140.8655 at the start, a rise of 1,633.2 pips on the 15-minute closes. In 2025 it went from 157.2375 to 156.67, a fall of 56.8 pips. **"Buy only while the 50-bar average is rising" lets buys through and blocks sells in a year that rose.** The same filter on the same chart in 2025 improved 20 settings and moved the median from +87.1 pips to +32.3. What a direction filter does is not avoid false signals. It leans the rule toward the side price travelled that year, which is useful to anyone who already knows which side that is, and those people have no need of the filter. ## Related reading - [Are DMI/ADX DI Crossovers Profitable? 1,344 Settings Tested](/blog/dmi-adx-settings): the same ADX, measured as a trading rule rather than a filter - [Does Combining RSI and Bollinger Bands Help? 720 Backtests](/blog/rsi-bollinger-combo): where the random-subset control used here was built - [GMMA Tested 216 Ways: Do 12 Moving Averages Beat 2?](/blog/gmma-settings): the article where the ADX filter did the most damage - [Is the Golden Cross Profitable on Forex? 280 Moving Average Tests](/blog/moving-average-cross-settings): one of the 38 settings the filters were applied to - [Do Indicator Settings Transfer Between Pairs? 3,024 Tests](/blog/settings-across-pairs): nearly the same set of settings, crossed with pairs instead of filters ## Notes on scope - USD/JPY only, 2024 and 2025, on the 15-minute, 1-hour and 4-hour charts - Fills at the close, a fixed 0.3 pip spread, 0.1 lots, no slippage - Two of the 38 base settings trade identically: the MACD article's 12/26/9 signal cross and the OsMA article's 12/26/9 matched on trade count and pips in all 42 runs, because OsMA is the distance between the MACD line and its signal, so its zero cross is that crossing - The random ranges come from a fixed seed, so the percentiles reproduce exactly --- ### Is the Inside Bar Breakout Profitable? 192 Backtests URL: https://formiq.jp/blog/inside-outside-bar-settings Language: en Published: 2026-09-10 Updated: 2026-09-10 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Inside bar, Outside bar, Harami, Breakout, Price action, Backtest, USDJPY Evidence: https://formiq.jp/evidence/inside-outside-bar-settings Evidence JSON: https://formiq.jp/evidence/inside-outside-bar-settings/data.json Also available in: ja — https://formiq.jp/ja/blog/inside-outside-bar-settings An inside bar sits entirely within the previous bar's high and low. It is read as a pause, and the usual rule is to **take the direction in which the following bar clears the bar before the pause**. An outside bar is the inverse: one that clears both the previous high and the previous low. The comparison this needs is obvious once stated. **The same level, cleared by the same bar, with the bar before it not contained.** If the containment carries information, the two have to differ. On the hourly chart in 2025, 6,126 bars cleared a two-bar-old high or low. **566 of them, 9.24%, followed a contained bar.** Ten bars later, the break's direction had held for **49.47% of the contained breaks and 50.59% of the rest**. In 2024 it reads 47.68% against 51.19%, the same way round. ## What the inside bar and outside bar actually test Formiq's backtester reads three bars: - **The mother** (two back): the bar whose high and low set the level - **The contained bar** (one back): its high at or below the mother's high, its low at or above the mother's low - **The breaking bar** (now): clearing the mother's high goes long, clearing its low goes short Highs and lows, wicks included. Bodies play no part. The outside bar is counted separately here: a bar that clears **both** the previous bar's high and its low, taken in the direction of its own close. ## Only a tenth of breaks come after an inside bar Counting every bar that cleared a two-bar-old high or low, and splitting out the ones preceded by a contained bar: | Timeframe | Year | All breaks | After an inside bar | Share | Outside bars | |---|---|---|---|---|---| | 15-minute | 2025 | 24498 | 2462 | 10.05% | 2718 | | 15-minute | 2024 | 24132 | 2542 | 10.53% | 2529 | | Hourly | 2025 | 6126 | 566 | 9.24% | 720 | | Hourly | 2024 | 5987 | 625 | 10.44% | 665 | | 4-hour | 2025 | 1555 | 195 | 12.54% | 220 | | 4-hour | 2024 | 1552 | 192 | 12.37% | 208 | **Contained breaks are 9.24 to 12.54% of the total.** The other nine in ten clear the same level with the previous bar sticking out of the mother. ## Settings and how to add a threshold Nothing is drawn: the pattern is a position relationship between bars already on the chart. | Condition | Numeric field | What it fires on | |---|---|---| | Inside bar breakout | none | the bar before sat inside the one before that, and this bar clears it | | Harami | none | this body sits inside the previous body, colour flipped, and smaller | **Neither has a numeric field.** There is no setting for how tightly the bar has to be contained; it either is or it is not. Filters go beside the condition instead. | What to filter on | Condition to add | Setting | |---|---|---| | The breaking bar's body is solid | Marubozu | body ratio 0.7 | | The breaking bar's range is large | Large candle | ATR multiple 1.5, body ratio 0.5 | | Take only the side the average points | MA slope | period 50, compared five bars back | ### The published trading rule Enter at the close of the breaking bar in the direction of the break. The usual exit is a break of the other side, which is what the condition does here too. The usual stop sits at the far end of the mother bar. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Timeframes | 15-minute, hourly, 4-hour | | Period | 1 January 2024 to 31 December 2025 | | Detection | the same code the backtest runs, bar by bar | | Control | the same two-bar-old level cleared, with the previous bar not contained | | Horizon | the close 1, 5, 10 and 20 bars after the breaking bar | | Exits | a break of the other side, and a fixed ten-bar hold | | Conditions | four inside bar, three harami, plus the ten-bar exit: 8 | | Runs | 8 conditions x 3 timeframes x 4 windows x with and without cost = 192 | | Spread | 0.3 pips, fixed | | Lot | 0.1 | | Fill | at the close | ## Against the same break, uncontained | Timeframe | Year | After an inside bar | Hit rate | Uncontained | Hit rate | Difference | |---|---|---|---|---|---|---| | 15-minute | 2025 | 2462 | 50.73% | 22036 | 49.62% | +1.11 | | 15-minute | 2024 | 2542 | 48.23% | 21590 | 49.77% | -1.54 | | Hourly | 2025 | 566 | 49.47% | 5560 | 50.59% | -1.12 | | Hourly | 2024 | 625 | 47.68% | 5362 | 51.19% | -3.51 | | 4-hour | 2025 | 195 | 48.21% | 1360 | 49.63% | -1.43 | | 4-hour | 2024 | 192 | 51.04% | 1360 | 48.75% | +2.29 | **Two cells of six came out ahead, on different charts in different years.** The 15-minute chart is +1.11 then -1.54; the 4-hour chart is -1.43 then +2.29. The hourly chart is negative in both years. In pips the hourly contained break returned +0.58 and +1.29 against +1.89 and +2.14 for the uncontained one. Both are positive; the contained one is smaller. ## What if you fade the break instead? Selling the up-break and buying the down-break, on exactly the same signals: | Timeframe | Year | Take the break | Fade it | Take an ordinary break | |---|---|---|---|---| | 15-minute | 2025 | 50.73% | 49.19% | 49.62% | | 15-minute | 2024 | 48.23% | 51.42% | 49.77% | | Hourly | 2025 | 49.47% | 50.53% | 50.59% | | Hourly | 2024 | 47.68% | 52.16% | 51.19% | | 4-hour | 2025 | 48.21% | 51.79% | 49.63% | | 4-hour | 2024 | 51.04% | 48.96% | 48.75% | Taking the break and fading it are the same trades reversed, so the two columns sum to 100% apart from ties. **The fade is ahead of the break in four cells of six.** Against an ordinary break of the same level the fade is also ahead in four cells, by 0.21 points (4-hour 2024) to 2.16 points (4-hour 2025), and behind in the 15-minute and hourly cells of 2025. The four it wins are not the same four in which it beats the break. So "the inside bar break is a trap, fade it" comes out marginally better than taking the break over these two years, and stays within roughly two points of an ordinary breakout either way. ## Two in a row, three in a row, and both ends at once Splitting by how many bars in a row were contained before the break: | Timeframe | Year | One | Two | Three or more | |---|---|---|---|---| | 15-minute | 2025 | 2248 at 51.11% | 195 at 44.10% | 19 at 73.68% | | 15-minute | 2024 | 2275 at 47.82% | 246 at 51.63% | 21 at 52.38% | | Hourly | 2025 | 498 at 49.40% | 61 at 45.90% | 7 at 85.71% | | Hourly | 2024 | 537 at 48.04% | 80 at 45.00% | 8 at 50.00% | | 4-hour | 2025 | 156 at 48.72% | 34 at 44.12% | 5 at 60.00% | | 4-hour | 2024 | 148 at 51.35% | 43 at 51.16% | 1 at 0.00% | **The three-or-more column holds between 1 and 21 signals.** The hourly 85.71% is six wins out of seven, which decides nothing. Two in a row read lower than one in five cells of six. The longest run of contained bars in the data is five, on the hourly chart in 2024. A breaking bar can also clear the mother's high and its low on the same bar. | Timeframe | Year | After an inside bar | Cleared both ends | Share | Taken as a buy | |---|---|---|---|---|---| | 15-minute | 2025 | 2462 | 262 | 10.64% | 131 at 48.85% | | 15-minute | 2024 | 2542 | 292 | 11.49% | 146 at 52.05% | | Hourly | 2025 | 566 | 66 | 11.66% | 33 at 63.64% | | Hourly | 2024 | 625 | 52 | 8.32% | 26 at 50.00% | | 4-hour | 2025 | 195 | 62 | 31.79% | 31 at 61.29% | | 4-hour | 2024 | 192 | 34 | 17.71% | 17 at 64.71% | **Such a bar satisfies the long condition and the short condition at once.** A single candle does not record which end was reached first, so the order cannot be recovered. Formiq's backtester checks the long side first, which means **every two-ended break is taken as a buy.** On the 4-hour chart in 2025 that is 31.79% of contained breaks. Read as buys, those bars returned between 48.85% and 64.71% on 17 to 146 signals. The direction is a resolution rule rather than a measurement, so treat that column as behaviour to know about rather than as a result. ## What the outside bar does differently Holding the outside bar for ten bars in the direction of its close: | Timeframe | Year | Outside bars | Hit rate | Ordinary break | Hit rate | |---|---|---|---|---|---| | 15-minute | 2025 | 2718 | 48.86% | 22036 | 49.62% | | 15-minute | 2024 | 2529 | 50.85% | 21590 | 49.77% | | Hourly | 2025 | 720 | 51.25% | 5560 | 50.59% | | Hourly | 2024 | 665 | 50.38% | 5362 | 51.19% | | 4-hour | 2025 | 220 | 51.82% | 1360 | 49.63% | | 4-hour | 2024 | 208 | 50.00% | 1360 | 48.75% | **Ahead in four cells of six, by between -0.81 and +2.19 points.** Like the inside bar, it does not separate from an ordinary break by a margin that holds. ## Is a harami the same thing? Both describe a small bar swallowed by a large one, but they complete on different bars. **The harami completes on the contained bar itself; the inside bar breakout completes on the bar after it.** They also test different things: bodies with a colour flip against highs and lows. | Timeframe | Year | Side | Both fired on the same bar | Share of harami signals | |---|---|---|---|---| | 15-minute | 2025 | long | 24 | 1.55% | | 15-minute | 2025 | short | 20 | 1.24% | | Hourly | 2025 | long | 5 | 1.31% | | Hourly | 2025 | short | 5 | 1.31% | | 4-hour | 2025 | long | 3 | 2.80% | | 4-hour | 2025 | short | 0 | 0.00% | **They coincide on 0.91 to 2.80% of harami signals.** Similar names, different moments. ## Trading all 192 runs Entering at the close of the breaking bar and exiting on a break of the other side, hourly: | Condition | 2025 trades | 2025 win rate | 2025 net pips | 2024 trades | 2024 win rate | 2024 net pips | |---|---|---|---|---|---|---| | inside bar breakout | 285 | 44.21% | +384.9 | 301 | 43.19% | +699.9 | | the same, ten-bar exit | 304 | 51.32% | +727.0 | 312 | 45.83% | -360.4 | | the same, body ≥ 0.7 | 115 | 44.35% | +1396.4 | 131 | 45.04% | -391.9 | | the same, range ≥ 1.5 ATR | 33 | 51.52% | +554.6 | 54 | 51.85% | +543.6 | | the same, 50-MA direction | 166 | 39.16% | -587.8 | 180 | 46.11% | +309.2 | | harami | 426 | 50.23% | +62.6 | 432 | 50.23% | -2007.5 | | harami, ten-bar exit | 344 | 47.67% | -662.3 | 362 | 46.13% | -2657.8 | | harami, 50-MA direction | 241 | 48.96% | -546.2 | 243 | 50.62% | -1735.9 | **Two of the eight cleared both years on the hourly chart**: the plain breakout and the ATR-filtered one, the latter on 33 and 54 trades. The 15-minute chart cleared one (the 50-MA direction) and the 4-hour cleared three (the ten-bar exit, the harami, and the harami with the 50-MA direction). **Not one condition cleared both years on even two of the three charts.** ### Splitting 2025 in half | Condition | 2025 | Jan-Jun | Jul-Dec | |---|---|---|---| | inside bar breakout | +384.9 | +762.0 | -484.4 | | the same, ten-bar exit | +727.0 | +440.2 | +281.3 | | the same, body ≥ 0.7 | +1396.4 | +501.3 | +787.8 | | the same, range ≥ 1.5 ATR | +554.6 | +441.7 | +112.9 | | harami | +62.6 | -579.4 | +641.2 | The plain breakout finishes 2025 at +384.9 pips with -484.4 of that in the second half. Three of the five were profitable in both halves. ### Last year's winner, held through this year | Timeframe | 2024 winner | 2024 net pips | 2025 net pips | 2025 rank | 2025 median of 8 | |---|---|---|---|---|---| | 15-minute | harami + 50-MA direction | +1710.7 | -348.4 | 6th | +515.6 | | Hourly | inside bar breakout | +699.9 | +384.9 | 4th | +223.8 | | 4-hour | inside bar breakout | +2999.9 | -779.6 | 6th | -134.7 | The 15-minute and 4-hour picks landed below the following year's median. The 4-hour breakout went from +2,999.9 pips to -779.6. ## Stops, targets and cost | Stop / target | 2025 trades | 2025 win rate | 2025 net pips | 2024 net pips | |---|---|---|---|---| | none (exit on the other side) | 285 | 44.21% | +384.9 | +699.9 | | 30 / 60 | 408 | 34.31% | -300.4 | +564.5 | | 60 / 30 | 410 | 54.88% | -536.6 | -68.4 | | 50 / 50 | 399 | 45.36% | +80.8 | -253.6 | Hourly inside bar breakout. **The 60 / 30 exit produces the highest win rate and loses money in both years.** The entries are identical and the win rate runs from 34.31% to 54.88%. | Timeframe | Condition | Trades | Spread paid | Trades x 0.3 | |---|---|---|---|---| | 15-minute | inside bar breakout | 1184 | 355.2 | 355.2 | | 15-minute | harami | 1720 | 515.9 | 516.0 | | Hourly | inside bar breakout | 285 | 85.5 | 85.5 | | Hourly | harami | 426 | 127.8 | 127.8 | | 4-hour | harami | 103 | 30.9 | 30.9 | Spread paid matched trade count times 0.3 pips. The 15-minute harami still returns -485.8 pips at zero spread, so its loss comes from the entry rather than the trade count. ## Related articles - [Candlestick pattern win rates](/blog/candlestick-patterns-tested): both of these among 17, over the same two years and the same kind of control - [Is the engulfing candle profitable?](/blog/engulfing-settings): the harami's mirror image, measured against colour flips that did not engulf - [Morning star and evening star](/blog/star-patterns-settings): a three-bar reversal taken apart clause by clause - [Do fair value gaps really fill?](/blog/fair-value-gap-settings): another price-action claim measured against ordinary bars ## Notes The data is USD/JPY from 1 January 2024 to 31 December 2025. Fills are at the close, the spread is fixed at 0.3 pips, and the lot is 0.1. Detection uses the same code path as the backtest and reads bars from before the test window. Because the horizons run out to 20 bars, that many bars at the end of each year are excluded. A bar clearing both ends of the mother is counted on both sides in the break-and-fade table and taken as a buy in the backtests. The harami overlap counts bars where both conditions fired in the same direction. --- ### Is the Pin Bar Profitable? 252 Backtests on USD/JPY URL: https://formiq.jp/blog/pin-bar-settings Language: en Published: 2026-09-10 Updated: 2026-09-10 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Pin bar, Wick, Price action, Hammer, Candlestick patterns, Backtest, USDJPY Evidence: https://formiq.jp/evidence/pin-bar-settings Evidence JSON: https://formiq.jp/evidence/pin-bar-settings/data.json Also available in: ja — https://formiq.jp/ja/blog/pin-bar-settings A pin bar has one long wick. A long lower wick is read as buying that came in below, a long upper wick as selling that came in above, and the usual rule is to **enter against the wick**. "Long" is a number here. Formiq's backtester compares the wick with the **body**, at a default of two times, and this article moves that from 1 to 5. One thing came out of the counting first. **At the default of 2, between 27.57% and 30.38% of all bars qualify** — nearly one bar in three. Raising the number barely moves the result. On the hourly chart in 2024 a ratio of 1 reads 51.46% and a ratio of 5 reads 51.69%, while the candidates fall from 2,973 bars to 859. **What moved the result was changing what the wick is measured against.** Comparing it with the bar's whole range instead of its body clears the base rate in five cells of six. ## What the pin bar condition actually tests - **lower wick** ≥ body × ratio, and **lower wick > upper wick** → long - **upper wick** ≥ body × ratio, and **upper wick > lower wick** → short The wick is the part outside the body: from the lower of open and close down to the low, and from the higher of the two up to the high. **The comparison is wick against body.** The range, high to low, never enters it. A bar with a small body clears any ratio easily; a bar with a large body fails it however long the wick is in absolute terms. ## The default calls nearly one bar in three a pin bar Counting bars where one wick is longer than the other and the ratio of 2 is met: | Timeframe | Year | Bars considered | Qualifying at 2 | Share | |---|---|---|---|---| | 15-minute | 2025 | 24760 | 6957 | 28.10% | | 15-minute | 2024 | 24860 | 6853 | 27.57% | | Hourly | 2025 | 6205 | 1757 | 28.32% | | Hourly | 2024 | 6242 | 1862 | 29.83% | | 4-hour | 2025 | 1592 | 469 | 29.46% | | 4-hour | 2024 | 1613 | 490 | 30.38% | **27.57% to 30.38%**: about 1,800 a year on the hourly chart and 6,900 on the 15-minute. A description that fits one bar in three is not filtering much. ## Settings and the one number it takes Nothing is drawn; the pattern is a single candle already on the chart. | Condition | Numeric field | Default | What changes | |---|---|---|---| | Pin bar | wick ratio | 2 | raising it keeps only wicks that are longer relative to the body, and cuts the count | The same value applies to both sides. There is no way to make one side stricter. ### The published trading rule Enter at the close of the pin bar, against the wick: long on a long lower wick, short on a long upper wick. The usual exit is an opposite pin bar, and the usual stop sits at the tip of the wick. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Timeframes | 15-minute, hourly, 4-hour | | Period | 1 January 2024 to 31 December 2025 | | Detection | the same test the backtest runs, rewritten and checked against it | | Control | every bar of the same window, weighted to each group's mix of sides | | Horizon | the close 1, 5, 10 and 20 bars after the signal bar | | Exits | an opposite pin bar, and a fixed ten-bar hold | | Wick ratios | 1, 1.5, 2, 2.5, 3, 4 and 5 | | Runs | 7 ratios x 2 exits and one spread-free pass x 3 timeframes x 4 windows = 252 | | Spread | 0.3 pips, fixed | | Lot | 0.1 | | Fill | at the close | **One note on the control.** The pin bar produces long and short signals, so comparing a short-heavy group with the share of bars that closed higher would penalise it in a year that rose. Each group is therefore compared with a base rate blended to its own count of long and short signals. ## Walking the wick-to-body ratio from 1 to 5 Share of signals that closed against the wick ten bars later, with the gap to the base rate in brackets: | Ratio | Hourly 2025 | Hourly 2024 | 4-hour 2025 | 4-hour 2024 | |---|---|---|---|---| | 1 | 2898 at 50.83% (+0.81) | 2973 at 51.46% (+0.81) | 766 at 52.09% (+2.15) | 789 at 55.64% (+3.70) | | 1.5 | 2188 at 50.78% (+0.76) | 2281 at 51.64% (+1.01) | 584 at 53.42% (+3.49) | 618 at 55.66% (+3.94) | | 2 | 1757 at 50.88% (+0.87) | 1862 at 51.40% (+0.83) | 469 at 52.88% (+2.95) | 490 at 56.12% (+4.18) | | 2.5 | 1476 at 50.68% (+0.66) | 1562 at 51.47% (+0.82) | 385 at 54.29% (+4.36) | 416 at 56.01% (+3.76) | | 3 | 1246 at 50.40% (+0.38) | 1365 at 51.72% (+1.08) | 336 at 55.65% (+5.73) | 350 at 55.43% (+2.86) | | 4 | 977 at 50.05% (+0.03) | 1052 at 51.52% (+0.79) | 270 at 54.81% (+4.88) | 273 at 53.11% (+0.80) | | 5 | 806 at 50.00% (-0.02) | 859 at 51.69% (+1.02) | 227 at 53.74% (+3.83) | 228 at 52.63% (+0.30) | **On the hourly chart in 2024 the gap at a ratio of 5 is +1.02 points, against +0.81 at a ratio of 1.** In 2025 the gap shrinks as the ratio rises, reaching -0.02 at 5. The 4-hour chart peaks at a ratio of 3 in 2025 (+5.73) and at 2 in 2024 (+4.18). **The ratio that worked best does not agree between the two years.** The 15-minute chart is below its base rate at all seven steps: -0.90 to -0.40 in 2025 and -1.06 to -0.06 in 2024. **Raising the ratio does one thing reliably, and that is cutting the count**: 2,898 down to 806 on the hourly chart in 2025. ## Measuring the wick against the range instead The same bars, cut by how much of the bar's high-to-low range the longer wick occupies: | Wick share | Hourly 2025 | Hourly 2024 | 4-hour 2025 | 4-hour 2024 | |---|---|---|---|---| | 30% or more | 4092 at 50.22% (+0.20) | 4208 at 51.43% (+0.73) | 1059 at 51.46% (+1.52) | 1087 at 55.20% (+3.29) | | 40% or more | 3038 at 50.79% (+0.77) | 3159 at 51.47% (+0.75) | 802 at 52.00% (+2.07) | 818 at 55.13% (+3.07) | | 50% or more | 1971 at 50.03% (0.00) | 2125 at 51.44% (+0.63) | 520 at 52.50% (+2.58) | 556 at 56.12% (+3.78) | | 60% or more | 1121 at 50.04% (+0.03) | 1142 at 53.06% (+2.10) | 296 at 55.41% (+5.51) | 300 at 56.00% (+3.25) | | 70% or more | 511 at 53.03% (+3.01) | 510 at 53.53% (+2.23) | 126 at 56.35% (+6.44) | 140 at 55.71% (+1.92) | **The 70% row clears the base rate on the hourly and 4-hour charts in both years.** On the 15-minute chart it clears in 2024 by +0.34 and misses in 2025 by -1.36, so five cells of six. The default wick-to-body ratio of 2 clears in four cells, by 0.83 to 4.18 points. **The range measure wins on both counts: more cells, and larger gaps.** The two rules do not pick the same bars. On the hourly chart in 2025, 1,757 bars meet the ratio of 2, 1,971 have a wick at half the range or more, and 1,506 do both. **251 bars are chosen only by the ratio and 465 only by the range**, so roughly a quarter of each set is unique to it. Formiq's backtester has no "wick at N% of the range" condition, so the range measure here is a count rather than something to trade directly. ## A bar with no body can never qualify The test requires `body > 0`. A bar whose open and close are exactly equal is never a pin bar, however far the wick runs. | Timeframe | Year | Bars considered | Body exactly zero | Share | Median wick share of range | |---|---|---|---|---|---| | 15-minute | 2025 | 24760 | 83 | 0.34% | 66.67% | | 15-minute | 2024 | 24860 | 105 | 0.42% | 66.67% | | Hourly | 2025 | 6205 | 10 | 0.16% | 58.65% | | Hourly | 2024 | 6242 | 13 | 0.21% | 73.18% | | 4-hour | 2025 | 1592 | 2 | 0.13% | 80.26% | | 4-hour | 2024 | 1613 | 4 | 0.25% | 72.83% | They are 0.13% to 0.42% of bars, so not many. What they are, though, is **bars whose wick takes 60 to 80% of the range**, excluded because the divisor is zero. Under a ratio test, a body of zero is not an infinite ratio; it is out of scope. ## How the hammer differs A hammer is the long side of a pin bar with one more requirement: the upper wick must not exceed the body. The long pin bar only asks that the lower wick be the longer of the two. On the hourly chart in 2025 there were 937 long pin bars, of which 306 (32.66%) also met the hammer requirement. Ten bars later, 50.69% of the pin bars had risen against 54.58% of the hammers. **All 306 hammers are inside the 937 pin bars.** They cannot be compared as separate methods. ## Trading all 252 runs Entering at the close and holding until an opposite pin bar, hourly: | Ratio | 2025 trades | 2025 win rate | 2025 net pips | 2024 trades | 2024 win rate | 2024 net pips | |---|---|---|---|---|---|---| | 1 | 1417 | 49.40% | -636.7 | 1468 | 49.86% | +455.4 | | 1.5 | 1083 | 49.86% | -276.3 | 1122 | 50.98% | +1430.8 | | 2 | 884 | 50.68% | +858.9 | 890 | 49.33% | +1560.4 | | 2.5 | 741 | 51.55% | +581.5 | 730 | 48.77% | +753.4 | | 3 | 627 | 51.20% | -105.2 | 630 | 48.73% | +1229.3 | | 4 | 467 | 50.11% | -576.8 | 484 | 48.14% | +612.0 | | 5 | 391 | 50.13% | +339.4 | 394 | 46.70% | +1008.0 | **Three of the seven cleared both years** (2, 2.5 and 5), with 3 and 4 failing in between. The survivors are not a contiguous range. The 4-hour chart cleared six of seven (everything but 2.5). **The 15-minute chart cleared none of the seven, at either exit.** With a fixed ten-bar exit the hourly chart cleared two and the 4-hour four. The 4-hour figures differ sharply between years. At a ratio of 2 the average trade returned +0.60 pips in 2025 and +14.40 in 2024, on 224 and 243 trades. ### Splitting 2025 in half | Ratio | 2025 | Jan-Jun | Jul-Dec | |---|---|---|---| | 1 | -636.7 | -198.3 | -446.1 | | 1.5 | -276.3 | +916.5 | -1269.3 | | 2 | +858.9 | +931.7 | -149.3 | | 2.5 | +581.5 | +486.1 | +18.9 | | 3 | -105.2 | +549.8 | -731.5 | | 4 | -576.8 | +845.1 | -1345.5 | | 5 | +339.4 | +701.3 | -290.2 | **Only the 2.5 ratio was profitable in both halves.** The ratio of 2, which finished 2025 at +858.9 pips, lost 149.3 of them in the second half. ## Last year's best ratio, held through this year | Timeframe | 2024 winner | 2024 net pips | 2025 net pips | 2025 rank | 2025 median of 7 | |---|---|---|---|---|---| | 15-minute | 5 | +551.2 | -3212.9 | 5th | -2830.5 | | Hourly | 2 | +1560.4 | +858.9 | 1st | -105.2 | | 4-hour | 2 | +3498.7 | +133.8 | 6th | +415.7 | Only the hourly pick held its position. The 4-hour ratio of 2 went from +3,498.7 pips to +133.8, sixth of seven. ## Stops, targets and cost | Stop / target | 2025 trades | 2025 win rate | 2025 net pips | 2024 net pips | |---|---|---|---|---| | none (exit on an opposite pin bar) | 884 | 50.68% | +858.9 | +1560.4 | | 30 / 60 | 1116 | 43.46% | +184.6 | +332.3 | | 60 / 30 | 1056 | 54.92% | +79.9 | -362.1 | | 50 / 50 | 1040 | 50.00% | +348.3 | +177.3 | Hourly, ratio 2. **The 60 / 30 exit gives the highest win rate and the only losing year of the four.** The entries are identical; the win rate runs from 43.46% to 54.92%. | Timeframe | Ratio | Trades | Spread paid | Trades x 0.3 | |---|---|---|---|---| | 15-minute | 1 | 5612 | 1683.5 | 1683.6 | | 15-minute | 2 | 3522 | 1056.5 | 1056.6 | | Hourly | 2 | 884 | 265.2 | 265.2 | | 4-hour | 2 | 224 | 67.2 | 67.2 | | 4-hour | 5 | 88 | 26.4 | 26.4 | Spread paid matched trade count times 0.3 pips. **The 15-minute chart at a ratio of 2 still returns -1,505.6 pips at zero spread**, so the loss is in the entry rather than the trade count. ## Related articles - [Candlestick pattern win rates](/blog/candlestick-patterns-tested): the pin bar and the hammer among 17, over the same two years - [Is the engulfing candle profitable?](/blog/engulfing-settings): another body-and-wick threshold, measured against a control - [Is the inside bar breakout profitable?](/blog/inside-outside-bar-settings): an ordinary break of the same level as the control - [Morning star and evening star](/blog/star-patterns-settings): a three-bar reversal taken apart clause by clause - [Do Indicator Settings Transfer Between Pairs? 3,024 Tests](/blog/settings-across-pairs): this setting won the 4-hour chart on USD/JPY in 2024 and came 1st of 42 on EUR/USD as well ## Notes The data is USD/JPY from 1 January 2024 to 31 December 2025. Fills are at the close, the spread is fixed at 0.3 pips, and the lot is 0.1. The detection code is a rewrite of the backtester's own test; at the default ratio of 2 it matches the engine's counts and rates in all twelve cells of three timeframes by two years by two sides. Bars whose upper and lower wicks are exactly equal point in no direction and are excluded. Because the horizons run out to 20 bars, that many bars at the end of each year are excluded from the counts. --- ### Is RCI Below -80 a Buy? 2,484 Backtests on USD/JPY URL: https://formiq.jp/blog/rci-settings Language: en Published: 2026-09-10 Updated: 2026-09-10 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: RCI, Rank correlation, Three lines, Mean reversion, Backtest, USDJPY Evidence: https://formiq.jp/evidence/rci-settings Evidence JSON: https://formiq.jp/evidence/rci-settings/data.json Also available in: ja — https://formiq.jp/ja/blog/rci-settings RCI stands for Rank Correlation Index and runs from -100 to +100. The reading it is always introduced with is "under -80 is oversold, over +80 is overbought". This is 2,484 runs asking whether that reading makes money. First, what the indicator is doing. **RCI reads only the order of the last N closes and throws the size of the move away.** If the last nine closes fell in a straight line, RCI is -100 whether price dropped ten pips or a hundred. So the control is the thing RCI discards: the percentage change across those same nine closes. Three results before the detail. **One. Past ±80 is not a rare state.** RCI sat at or under -80 on 6.99% to 17.69% of bars and at or over +80 on 9.56% to 31.31%. The 5th percentile of the reading runs from -93.33 to -84.96, so **the line spends more time near its ends than near zero**. **Two. The textbook fade came last of the five readings.** Settings that ended both 2024 and 2025 in profit: 4 of 72 on the 15-minute chart, 3 of 72 on the hourly, 9 of 72 on the 4-hour. Read the same threshold as momentum instead, buying over +80, and those become 39, 33 and 6. **Three. The ranking did not beat the magnitude it replaced.** Comparing the bars RCI puts under -80 with the same number of bars taken from the smallest changes, RCI's set won on the next ten bars in 17 of 36 cells. ## What RCI measures RCI builds two rankings over the closes in its window and measures how well they agree. 1. **Rank by time**: number the bars 1, 2, 3 from the oldest. 2. **Rank by price**: number the closes 1, 2, 3 from the lowest. A straight climb puts the oldest bar at the lowest close, the two rankings agree exactly, and RCI reads +100. A straight fall reverses them and reads -100. The formula squares the rank differences and sums them. At period 9 a squared-difference sum of 216 lands the reading exactly on -80. **Two consequences follow from that construction.** **The size of the move is not in the formula.** Only ranks are, so nine bars that fell ten pips and nine bars that fell a hundred read the same if the order was the same. **The reading lands on a lattice.** The squared-difference sum is an integer, so RCI at period 9 can only take values 0.833 apart. Across two years of hourly bars (13,482 of them) it took 313 distinct values. Period 52 has a step of 0.004 and took 11,429. At period 9, the reading was exactly +100 on 51 bars and exactly -100 on 39: nine closes in perfect order, with nowhere further for the line to go. That is 0.67% of bars together. At period 21 and above it never happened. ## RCI settings and how they are read RCI is drawn in a panel under the price chart, scaled -100 to +100. Formiq draws the ±80 lines from the start; MT4 does not, so they have to be added by hand there. The pane takes one to three lines. The three-line version most RCI material is written around uses a short 9, a mid 26 and a long 52. | Where | Setting | Default | What it does | |---|---|---|---| | Chart | Short period | 9 | How many closes are ranked | | Chart | Mid period | 0 | 0 draws nothing; 26 adds the second line | | Chart | Long period | 0 | 0 draws nothing; 52 adds the third line | | Backtest | Reading | Level extreme | One of the five below | | Backtest | Period | 9 | How many closes are ranked | | Backtest | Mid and long | 26 and 52 | Read only by the three-line rule | | Backtest | Direction | Below | Which side the condition opens | | Backtest | Level | -80 | The value the rule compares against | ### The five readings | Reading | Buys when | Sells when | |---|---|---| | Level extreme (fade) | RCI is under the level | RCI is over the level | | Level break (follow) | RCI is over the level | RCI is under the level | | Break out of ±level | RCI crosses up through +level | RCI crosses down through -level | | Level cross | RCI crosses up through the level | RCI crosses down through the level | | Three lines aligned | All three are under -level | All three are over +level | **The fade and the follow are the same number read two ways.** Each opens one side only, so both were swept as a long rule and a short rule separately. The close is a return through a second level: "buy under -80, close at 0" is the published form. The remaining three open both sides, so entry and exit share one rule set. ## How this was measured | Item | Value | |---|---| | Instrument | USD/JPY | | Timeframes | 15-minute, 1-hour, 4-hour | | Period | 1 January 2024 to 31 December 2025, with 2025 also split in halves | | RCI periods | 9, 14, 21, 26, 36, 52 | | Fade and follow levels | 70, 80, 90 to open; 0 and 40 to close | | Break levels | 50, 70, 80, 90 | | Crossing levels | -80, -40, 0, 40, 80 | | Three-line sets | 9/26/52, 9/14/21, 14/26/52 at levels 50, 70, 80 | | Control | The change across the same closes (ROC with a lookback one bar shorter) | | Runs | 207 settings x 3 timeframes x 4 windows = 2,484 | | Spread | 0.3 pips, fixed | | Lot size | 0.1 | | Fill | Close of the signal bar | **How the control was built.** RCI at period 9 reads nine closes, so the move across those same nine closes is the change from the close nine bars back. Formiq's rate-of-change condition at period 8 covers exactly that span. The bars RCI put under -80 were counted, the same number of bars was taken from the small-change end, and both sets were measured to the close ten bars later. ## Past ±80 is not a rare state Before any trading, how often the line reaches the level. Every 2025 bar: | Timeframe | Period | Bars | At or under -80 | At or over +80 | 5th pct | 95th pct | |---|---|---|---|---|---|---| | 15-minute | 9 | 24,893 | 14.96% | 16.47% | -92.08 | 93.33 | | 15-minute | 26 | 24,893 | 12.45% | 13.62% | -89.81 | 91.25 | | 15-minute | 52 | 24,893 | 13.02% | 14.91% | -89.49 | 90.95 | | Hourly | 9 | 6,216 | 15.70% | 16.04% | -93.33 | 93.33 | | Hourly | 26 | 6,216 | 14.03% | 14.43% | -91.59 | 90.29 | | Hourly | 52 | 6,216 | 14.56% | 14.54% | -89.70 | 89.39 | | 4-hour | 9 | 1,600 | 17.69% | 16.88% | -93.33 | 91.67 | | 4-hour | 26 | 1,600 | 9.94% | 16.25% | -87.97 | 90.97 | | 4-hour | 52 | 1,600 | 11.25% | 9.56% | -88.73 | 87.07 | Across all 36 cells of two years, three timeframes and six periods, the reading was at or under -80 on 6.99% to 17.69% of bars and at or over +80 on 9.56% to 31.31%. **At period 9 the two sides together cover 31.28% to 34.57% of bars: one bar in three is in an extreme.** The percentiles are worth a second look. In every cell the 5th percentile sits between -93.33 and -84.96 and the 95th between 87.07 and 96.67. **The line spends more time near ±90 than near zero**, which makes "RCI reached ±80" an ordinary event rather than a signal. ### How often the three lines agree Counting bars where the short 9, mid 26 and long 52 are all outside ±80: | Timeframe | Short line alone | All three | Share of bars | Share of the short line's | |---|---|---|---|---| | 15-minute | 7,824 | 805 | 3.23% | 10.29% | | Hourly | 1,973 | 122 | 1.96% | 6.18% | | 4-hour | 553 | 29 | 1.81% | 5.24% | All three align on 1.81% to 5.14% of bars, and on 5.24% to 14.98% of the bars where the short line alone is outside. **The 4-hour chart offers 29 such bars in a year.** Any figure from a sample that size needs its trade count read beside it. ## What changes when only the order is read Comparing the bars RCI puts under -80 with the same number taken from the smallest changes over the same closes, measured to the close ten bars later. 2025: | Timeframe | Period | Correlation | RCI bars | Overlap | RCI mean | Change mean | All bars | |---|---|---|---|---|---|---|---| | 15-minute | 9 | 0.756 | 3,723 | 62.56% | +0.59 pips | -0.03 pips | -0.03 pips | | 15-minute | 26 | 0.773 | 3,100 | 59.61% | -0.18 pips | +1.73 pips | -0.03 pips | | 15-minute | 52 | 0.785 | 3,242 | 58.95% | -3.17 pips | +0.30 pips | -0.03 pips | | Hourly | 9 | 0.798 | 976 | 64.65% | -4.91 pips | +0.15 pips | -0.07 pips | | Hourly | 26 | 0.810 | 872 | 59.40% | +3.42 pips | +7.57 pips | -0.07 pips | | Hourly | 52 | 0.823 | 905 | 62.98% | +5.29 pips | +5.67 pips | -0.07 pips | | 4-hour | 9 | 0.828 | 283 | 65.37% | -1.86 pips | +7.05 pips | -0.48 pips | | 4-hour | 26 | 0.832 | 159 | 50.31% | +12.96 pips | +13.24 pips | -0.48 pips | | 4-hour | 52 | 0.777 | 180 | 51.11% | +28.25 pips | +9.57 pips | -0.48 pips | **Correlation between RCI and the change runs 0.679 to 0.855.** High, and not 1. **The two sets overlap by 50.31% to 70.83%.** Three to five bars in ten that RCI calls oversold are not in the set the plain change picks. The ranking really is selecting different bars. What it is not doing is selecting better ones. **Across all 36 cells, RCI's set beat the change's set on pips in 17 and on win rate in 18.** Either measure lands where a coin would. ### Did the discarded magnitude carry anything If RCI drops the size of the move, the size might still be informative. Taking only the bars where RCI is at or under -80 and splitting them into quarters by the change over the same nine closes, in 2025: | Timeframe | Quarter by change | Bars | Change range | Win rate | Mean over 10 bars | |---|---|---|---|---|---| | 15-minute | Largest fall | 930 | -1.622% to -0.269% | 52.37% | +0.86 pips | | 15-minute | Second | 930 | -0.269% to -0.174% | 55.70% | +1.75 pips | | 15-minute | Third | 930 | -0.174% to -0.110% | 52.37% | -0.74 pips | | 15-minute | Smallest fall | 933 | -0.110% to -0.007% | 51.98% | +0.51 pips | | Hourly | Largest fall | 244 | -2.047% to -0.535% | 50.82% | -6.63 pips | | Hourly | Second | 244 | -0.532% to -0.377% | 48.36% | -4.19 pips | | Hourly | Third | 244 | -0.377% to -0.250% | 48.36% | -3.01 pips | | Hourly | Smallest fall | 244 | -0.250% to -0.027% | 46.72% | -5.81 pips | **There is no ordered relationship between the size of the fall and what followed.** On the 15-minute chart the second quarter did best at +1.75 pips while the largest falls returned +0.86. All four hourly quarters were negative, the largest falls at -6.63 pips and the smallest at -5.81. So what RCI discards carried nothing here. What it keeps did not do better than what it discards, either. ## Trading all five readings 207 settings, traded. Settings that ended the year with positive pips: ### 2025 | Reading | Settings | 15-minute | Hourly | 4-hour | |---|---|---|---|---| | Level extreme (fade) | 72 | 13 | 34 | 65 | | Level break (follow) | 72 | 54 | 36 | 7 | | Break out of ±level | 24 | 18 | 13 | 3 | | Level cross | 30 | 19 | 21 | 6 | | Three lines aligned | 9 | 6 | 7 | 6 | ### 2024 | Reading | Settings | 15-minute | Hourly | 4-hour | |---|---|---|---|---| | Level extreme (fade) | 72 | 14 | 9 | 12 | | Level break (follow) | 72 | 49 | 63 | 59 | | Break out of ±level | 24 | 18 | 20 | 20 | | Level cross | 30 | 16 | 22 | 22 | | Three lines aligned | 9 | 2 | 2 | 2 | One year at a time, the ordering changes. The fade took 65 of 72 on the 4-hour chart in 2025 and 12 of 72 on the same chart in 2024. The three-line reading had 19 of 27 in profit in 2025 and 6 of 27 in 2024. Requiring both years narrows it. **Settings profitable in 2024 and in 2025:** | Reading | Settings | 15-minute | Hourly | 4-hour | |---|---|---|---|---| | Level extreme (fade) | 72 | 4 | 3 | 9 | | Level break (follow) | 72 | 39 | 33 | 6 | | Break out of ±level | 24 | 14 | 11 | 1 | | Level cross | 30 | 12 | 18 | 5 | | Three lines aligned | 9 | 2 | 2 | 0 | **The textbook fade is the thinnest row on the 15-minute and hourly charts**, at 4 and 3 of 72. Reading the same threshold as momentum instead gives 39 and 33. Splitting 2025 into halves and requiring all four windows narrows it again: | Reading | Settings | 15-minute | Hourly | 4-hour | |---|---|---|---|---| | Level extreme (fade) | 72 | 2 | 0 | 4 | | Level break (follow) | 72 | 8 | 7 | 1 | | Break out of ±level | 24 | 11 | 7 | 0 | | Level cross | 30 | 7 | 8 | 1 | | Three lines aligned | 9 | 2 | 2 | 0 | **All 72 hourly fade settings lost money in at least one of the four windows.** ## The textbook buy, period by period Opening at level 80 and closing on a return to 0, in 2025: | Timeframe | Side | Period | Trades | Win rate | Annual pips | Bars held | |---|---|---|---|---|---|---| | 15-minute | Long | 9 | 750 | 66.13% | +367.4 | 11.2 | | 15-minute | Long | 14 | 456 | 65.13% | -336.6 | 18.5 | | 15-minute | Long | 26 | 226 | 65.93% | -426.2 | 35.2 | | 15-minute | Long | 52 | 116 | 61.21% | -1,074.4 | 78.1 | | 15-minute | Short | 9 | 753 | 63.08% | -1,194.3 | 13.5 | | Hourly | Long | 9 | 176 | 64.20% | -652.9 | 13.7 | | Hourly | Long | 36 | 49 | 67.35% | +322.6 | 50.6 | | Hourly | Short | 21 | 93 | 66.67% | +358.6 | 28.8 | | 4-hour | Long | 21 | 18 | 72.22% | +1,150.3 | 28.3 | | 4-hour | Short | 26 | 23 | 86.96% | +1,396.6 | 28.2 | | 4-hour | Short | 9 | 53 | 52.83% | -441.0 | 12.6 | **Win rates run 52.83% to 86.96%, and all 36 cells of three timeframes, two sides and six periods cleared 50%.** The annual pips column still fills with negative numbers. The 15-minute long side at period 9 won 66.13% for +367.4 pips; the short side of the same setting won 63.08% for -1,194.3. **That is 1,561.7 pips between two sides of one rule.** On 15-minute closes USD/JPY rose 1,633.2 pips through 2024 and fell 56.8 pips through 2025, so the long and the short were reading different markets. The setting closest to the shipped defaults, period 9 at level 80 closing at 0, ranked 179th of 207 by annual pips on the hourly chart's long side in 2025 (176 trades, 64.20%, -652.9 pips), and 185th on the short side in 2024 (200 trades, 55.50%, -1,255.6 pips). **This reading sits below the middle in both years.** ## Three lines worked in 2025 only All 27 settings of three timeframes, three period sets and three levels, in both years: | Timeframe | Periods | Level | 2025 trades | 2025 win | 2025 pips | 2024 trades | 2024 win | 2024 pips | |---|---|---|---|---|---|---|---|---| | 15-minute | 9/26/52 | 70 | 123 | 61.79% | +865.6 | 121 | 66.94% | +1,638.0 | | 15-minute | 9/14/21 | 80 | 244 | 64.34% | +2,788.6 | 216 | 57.87% | -2,607.5 | | Hourly | 9/26/52 | 70 | 35 | 77.14% | +3,340.0 | 22 | 59.09% | +215.5 | | Hourly | 9/14/21 | 80 | 72 | 66.67% | +1,499.1 | 62 | 66.13% | -1,414.5 | | Hourly | 14/26/52 | 70 | 39 | 71.79% | +2,476.5 | 20 | 60.00% | -1,612.6 | | 4-hour | 9/14/21 | 70 | 34 | 79.41% | +3,759.2 | 13 | 30.77% | -3,108.0 | | 4-hour | 9/14/21 | 80 | 19 | 78.95% | +3,282.0 | 9 | 22.22% | -3,224.6 | | 4-hour | 9/26/52 | 50 | 15 | 73.33% | +949.1 | 10 | 40.00% | -3,103.7 | | 4-hour | 14/26/52 | 80 | 2 | 50.00% | -843.2 | 7 | 71.43% | +788.6 | **Four of the 27 ended both years in profit.** All nine 4-hour settings lost money in one of the two. The 4-hour 9/14/21 at level 70 makes it plain. In 2025 it took 34 trades, won 79.41% and made +3,759.2 pips, the best of the 207 settings that year. The same setting in 2024 took 13 trades, won 30.77% and lost 3,108.0 pips, which is 205th of 207. **A rule that trades 13 to 34 times a year moves that far between years on its own.** ## Last year's winner, this year Taking the setting with the highest annual pips in one year and running it in the other, with its rank among the 207: | Timeframe | Picked in | Setting | Pips that year | Pips the other year | Rank | Median that year | |---|---|---|---|---|---|---| | 15-minute | 2024 | Level cross, period 36, level -40 | +3,805.8 | +996.5 | 24th | +89.1 | | 15-minute | 2025 | Three lines, 9/14/21, level 80 | +2,788.6 | -2,607.5 | 205th | -81.2 | | Hourly | 2024 | Level cross, period 21, level -40 | +4,028.1 | +659.6 | 40th | +77.7 | | Hourly | 2025 | Three lines, 9/26/52, level 70 | +3,340.0 | +215.5 | 105th | +226.3 | | 4-hour | 2024 | Break out of ±level, period 21, level 80 | +3,652.5 | -3,086.7 | 206th | -265.1 | | 4-hour | 2025 | Three lines, 9/14/21, level 70 | +3,759.2 | -3,108.0 | 205th | +95.1 | **Three of the six landed 205th or 206th of 207 in the other year.** Two beat that year's median, and both are the 2024 picks on the 15-minute and hourly charts. Those two are the same reading at the same level: a crossing of -40. That is not the ±80 the indicator is introduced with. The only choice that stayed above the median a year later was the one furthest from the textbook. ## Related reading - [Is Buying RSI Below 30 Profitable? 330 Settings Tested on USD/JPY](/blog/rsi-settings): the same fade idea measured on a different oscillator - [Is the CCI ±100 Strategy Profitable? 2,448 Settings Tested](/blog/cci-settings): four of these five readings, on CCI - [Is the Psychological Line 75% Rule Profitable? 5,760 Tests](/blog/psychological-line-settings): a level that turns out to be unreachable as typed - [Do Trend Filters Remove False Signals? 1,596 Backtests](/blog/false-signal-filters): filters applied across every published setting ## Notes on scope - USD/JPY only, 2024 and 2025, on the 15-minute, 1-hour and 4-hour charts - Fills at the close, a fixed 0.3 pip spread, 0.1 lots, no slippage - The three-line reading trades 2 to 45 times a year on the 4-hour chart; win rate and pips from counts that small move a great deal - The control is a single measure of magnitude, not an inventory of everything the ranking discards --- ### Does Combining RSI and Bollinger Bands Help? 720 Backtests URL: https://formiq.jp/blog/rsi-bollinger-combo Language: en Published: 2026-09-10 Updated: 2026-09-10 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: RSI, Bollinger Bands, Indicator combination, Mean reversion, Backtest, USDJPY Evidence: https://formiq.jp/evidence/rsi-bollinger-combo Evidence JSON: https://formiq.jp/evidence/rsi-bollinger-combo/data.json Also available in: ja — https://formiq.jp/ja/blog/rsi-bollinger-combo "Buy when RSI is below 30 and price touches the lower Bollinger band." Stacking two indicators like this is the most common way to build a rule out of two. What the operation does is not in doubt. **Combining creates no signals. It takes the bars where RSI is below 30 and discards the ones that did not touch the band.** On the hourly chart in 2025, RSI closed below 30 on 356 bars. 216 of them, 60.67%, also touched the lower band at two standard deviations. **The other 40% are thrown away, and that is what "filtering" means here.** Throwing bars away moves the win rate on its own, because a smaller sample sits further from its own average. So "the win rate went up when I combined them" only means something if it went up **more than discarding the same number of bars at random would have**. That is the measurement. **Of 108 pairings, 21 fall outside the 95% range of random discards, and 19 of those 21 fall below it.** ## Settings for RSI and the Bollinger band RSI runs between 0 and 100 in a pane under the chart. The Bollinger band draws a moving average with lines a multiple of the standard deviation above and below it, over the price itself. | Condition | Field | Default | What it means | |---|---|---|---| | RSI | period | 14 | how many bars the calculation reads | | RSI | level and comparator | 30, below | below 30 buys; above 70 sells | | Bollinger band | period | 20 | bars in the average and the deviation | | Bollinger band | deviation | 2 | how many standard deviations out the lines sit | | Bollinger band | trigger | touch | a low reaching the lower line buys, a high reaching the upper line sells | **Both are states rather than events.** RSI below 30 holds for as long as it holds; it does not fire once. ### Using both at once Switch both on and set the logic to require all conditions. Only bars where both point the same way remain. RSI below 30 and a lower-band touch both produce long signals only, so the pairing is long-only; the short pairing is built from RSI above 70 and an upper-band touch. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Timeframes | 15-minute, hourly, 4-hour | | Period | 1 January 2024 to 31 December 2025 | | RSI levels | below 25, 30, 35 for longs; above 75, 70, 65 for shorts | | Band deviations | 1.5, 2, 2.5 | | Control | 1,000 random draws of the same size from the same RSI pool | | Horizon | the close 1, 5, 10 and 20 bars after the signal bar | | Exit | the close 10 bars later, for all three | | Conditions | RSI alone 3, band alone 3, combined 9, per side | | Runs | 15 x 2 sides x 3 timeframes x 4 windows x with and without cost = 720 | | Spread | 0.3 pips, fixed | | Lot | 0.1 | | Fill | at the close | **All three use the same ten-bar exit.** The comparison is between three entries, and different exits would turn it into a comparison of exits. The random range is built like this: if 356 bars are below the RSI level and 216 of them also touch the band, then 216 bars are drawn from the 356 at random, a thousand times, and the resulting win rates are sorted to take the 2.5th and 97.5th percentiles. The generator runs from a fixed seed, so the published range is reproducible. ## Against the same number of bars picked at random Hourly, long side. The pool is bars below the RSI level; the pairing is the subset that also touched the band. | Year | RSI | Deviation | Pool | Kept | Kept % | Pool rate | Pairing rate | Random range | |---|---|---|---|---|---|---|---|---| | 2025 | below 25 | 1.5σ | 112 | 106 | 94.64% | 61.61% | 59.43% | 59.43-63.21% | | 2025 | below 25 | 2σ | 112 | 79 | 70.54% | 61.61% | 59.49% | 55.70-67.09% | | 2025 | below 25 | 2.5σ | 112 | 42 | 37.50% | 61.61% | 59.52% | 50.00-71.43% | | 2025 | below 30 | 1.5σ | 356 | 327 | 91.85% | 55.90% | 55.66% | 54.43-57.49% | | 2025 | below 30 | 2σ | 356 | 216 | 60.67% | 55.90% | 56.48% | 51.85-60.19% | | 2025 | below 30 | 2.5σ | 356 | 117 | 32.87% | 55.90% | 54.70% | 48.72-64.10% | | 2025 | below 35 | 1.5σ | 772 | 617 | 79.92% | 54.53% | 53.32% | 52.84-56.24% | | 2025 | below 35 | 2σ | 772 | 395 | 51.17% | 54.53% | 52.41% | 51.14-57.72% | | 2025 | below 35 | 2.5σ | 772 | 216 | 27.98% | 54.53% | 55.09% | 49.07-60.19% | | 2024 | below 25 | 1.5σ | 132 | 123 | 93.18% | 50.76% | 47.97% | 48.78-52.85% | | 2024 | below 25 | 2σ | 132 | 84 | 63.64% | 50.76% | 40.48% | 44.05-57.14% | | 2024 | below 25 | 2.5σ | 132 | 46 | 34.85% | 50.76% | 41.30% | 39.13-63.04% | | 2024 | below 30 | 1.5σ | 309 | 277 | 89.64% | 50.16% | 48.74% | 48.38-51.99% | | 2024 | below 30 | 2σ | 309 | 185 | 59.87% | 50.16% | 45.41% | 45.95-55.14% | | 2024 | below 30 | 2.5σ | 309 | 98 | 31.72% | 50.16% | 52.04% | 42.86-58.16% | | 2024 | below 35 | 1.5σ | 617 | 499 | 80.88% | 48.78% | 49.90% | 46.89-50.70% | | 2024 | below 35 | 2σ | 617 | 321 | 52.03% | 48.78% | 47.35% | 45.17-52.65% | | 2024 | below 35 | 2.5σ | 617 | 175 | 28.36% | 48.78% | 51.43% | 42.86-54.86% | **All nine 2025 pairings land inside the range.** Three of the 2024 pairings land outside it, and all three land below (47.97% against 48.78-52.85%, 40.48% against 44.05-57.14%, 45.41% against 45.95-55.14%). Notice how the range widens. Keeping 94.64% of the pool gives a range 3.78 points wide; keeping 37.50% gives one 21.43 points wide. **The harder the filter, the more the win rate is expected to move on its own.** Across three timeframes and both sides there are 108 pairings. **21 land outside the range: 19 below and 2 above.** Requiring the same timeframe, side and setting to land outside in the same direction in both years leaves 4 of 54, all of them on the hourly short side and all of them below. ## What the band condition selects If the pairing lands low, the band must be selecting a particular kind of bar. Splitting the RSI pool by whether the bar touched the band: | Timeframe | Year | Side | Touched | Run-up | Bar size | Mean RSI | Missed | Run-up | Bar size | Mean RSI | |---|---|---|---|---|---|---|---|---|---|---| | 15-minute | 2025 | long | 780 | 3.07 ATR | 1.56 ATR | 25.04 | 569 | 1.31 ATR | 0.93 ATR | 24.97 | | 15-minute | 2025 | short | 739 | 2.93 ATR | 1.39 ATR | 75.12 | 713 | 1.24 ATR | 0.86 ATR | 74.92 | | Hourly | 2025 | long | 216 | 3.05 ATR | 1.46 ATR | 25.48 | 140 | 1.42 ATR | 0.97 ATR | 26.25 | | Hourly | 2025 | short | 185 | 3.13 ATR | 1.36 ATR | 76.16 | 219 | 1.11 ATR | 0.84 ATR | 75.78 | | Hourly | 2024 | long | 185 | 3.17 ATR | 1.80 ATR | 24.20 | 124 | 1.22 ATR | 0.92 ATR | 25.25 | | Hourly | 2024 | short | 291 | 2.87 ATR | 1.50 ATR | 76.41 | 307 | 1.11 ATR | 0.82 ATR | 75.40 | | 4-hour | 2025 | short | 60 | 3.29 ATR | 1.26 ATR | 74.93 | 58 | 1.26 ATR | 0.88 ATR | 75.61 | | 4-hour | 2024 | short | 106 | 2.70 ATR | 1.40 ATR | 77.88 | 74 | 1.28 ATR | 0.88 ATR | 76.38 | Run-up is how far price had already travelled into the signal over five bars, in ATR: for a long, how far it had already fallen. **The touched half had run 1.80 to 2.82 times as far, on bars 1.43 to 2.03 times as large. The mean RSI of the two halves differs by less than 2.1 points.** So the band is not selecting bars that are more oversold. At the same RSI reading, it selects the ones **falling faster**. In 8 of the 12 cells the touched half had the lower rate ten bars on. On the hourly chart in 2025 the short side reads 32.97% against 54.79%, a gap of 21.82 points. The 4-hour long pool in 2025 is left out of the table: only 11 bars there did not touch. ## Did the win rate rise? Hourly, the RSI level alone against the pairing: | Side | RSI | Deviation | Alone | Paired | Kept % | Alone win rate | Paired win rate | Delta | |---|---|---|---|---|---|---|---|---| | long | below 25 | 1.5σ | 41 | 39 | 95.1% | 70.73% | 69.23% | -1.50 | | long | below 25 | 2σ | 41 | 36 | 87.8% | 70.73% | 69.44% | -1.29 | | long | below 25 | 2.5σ | 41 | 23 | 56.1% | 70.73% | 65.22% | -5.51 | | long | below 30 | 1.5σ | 85 | 82 | 96.5% | 50.59% | 50.00% | -0.59 | | long | below 30 | 2σ | 85 | 74 | 87.1% | 50.59% | 52.70% | +2.11 | | long | below 30 | 2.5σ | 85 | 56 | 65.9% | 50.59% | 51.79% | +1.20 | | long | below 35 | 1.5σ | 145 | 130 | 89.7% | 46.90% | 46.92% | +0.02 | | long | below 35 | 2σ | 145 | 113 | 77.9% | 46.90% | 47.79% | +0.89 | | long | below 35 | 2.5σ | 145 | 87 | 60.0% | 46.90% | 50.57% | +3.67 | | short | above 75 | 1.5σ | 35 | 32 | 91.4% | 37.14% | 31.25% | -5.89 | | short | above 75 | 2σ | 35 | 29 | 82.9% | 37.14% | 34.48% | -2.66 | | short | above 75 | 2.5σ | 35 | 19 | 54.3% | 37.14% | 31.58% | -5.56 | | short | above 70 | 1.5σ | 73 | 63 | 86.3% | 42.47% | 38.10% | -4.37 | | short | above 70 | 2σ | 73 | 53 | 72.6% | 42.47% | 32.08% | -10.39 | | short | above 70 | 2.5σ | 73 | 40 | 54.8% | 42.47% | 32.50% | -9.97 | | short | above 65 | 1.5σ | 146 | 127 | 87.0% | 53.42% | 51.97% | -1.45 | | short | above 65 | 2σ | 146 | 106 | 72.6% | 53.42% | 50.00% | -3.42 | | short | above 65 | 2.5σ | 146 | 78 | 53.4% | 53.42% | 47.44% | -5.98 | **Five of the 18 improved, by between 0.02 and 3.67 points, and all five are long.** All nine short pairings fell, the largest by 10.39 points. How much was discarded does not predict the change. Keeping 95.1% gives -1.50; keeping 53.4% gives -5.98. ## Trading all 720 runs Entering at the close and exiting ten bars later, hourly long side: | Condition | 2025 trades | 2025 win rate | 2025 net pips | 2024 trades | 2024 win rate | 2024 net pips | |---|---|---|---|---|---|---| | RSI below 25 alone | 41 | 70.73% | +695.9 | 34 | 52.94% | -113.3 | | RSI below 30 alone | 85 | 50.59% | +316.4 | 73 | 54.79% | +164.5 | | RSI below 35 alone | 145 | 46.90% | +210.1 | 113 | 48.67% | -529.9 | | 1.5σ touch alone | 284 | 50.00% | -745.5 | 260 | 54.23% | -391.5 | | 2σ touch alone | 214 | 49.53% | +194.1 | 202 | 52.97% | -194.4 | | 2.5σ touch alone | 160 | 50.00% | +446.4 | 151 | 53.64% | +194.6 | | RSI 25 + 1.5σ | 39 | 69.23% | +629.3 | 32 | 50.00% | -381.1 | | RSI 25 + 2σ | 36 | 69.44% | +545.1 | 29 | 51.72% | -163.0 | | RSI 25 + 2.5σ | 23 | 65.22% | +254.7 | 21 | 42.86% | -356.3 | | RSI 30 + 1.5σ | 82 | 50.00% | +439.7 | 71 | 54.93% | +138.8 | | RSI 30 + 2σ | 74 | 52.70% | +385.6 | 62 | 51.61% | +106.5 | | RSI 30 + 2.5σ | 56 | 51.79% | +337.9 | 49 | 51.02% | +63.5 | | RSI 35 + 1.5σ | 130 | 46.92% | -101.6 | 108 | 52.78% | +208.4 | | RSI 35 + 2σ | 113 | 47.79% | +244.8 | 92 | 51.09% | +269.6 | | RSI 35 + 2.5σ | 87 | 50.57% | +335.6 | 80 | 51.25% | +287.4 | **Seven of the 15 cleared both years.** The rest of the grid: | Timeframe | Side | Cleared both years, of 15 | |---|---|---| | 15-minute | long | 0 | | Hourly | long | 7 | | 4-hour | long | 0 | | 15-minute | short | 0 | | Hourly | short | 0 | | 4-hour | short | 2 | **Only the hourly long side and the 4-hour short side produced anything.** In the other four cells nothing survived, alone or paired. ## The two cases where combining rescued a losing level Five of the seven survivors were already profitable in both years using the RSI level alone. **Only two turned a level that lost a year into one that did not.** | Condition | RSI alone, 2024 | Paired, 2024 | RSI alone, 2025 | Paired, 2025 | |---|---|---|---|---| | Hourly, RSI 35 + 2σ | -529.9 | +269.6 | +210.1 | +244.8 | | Hourly, RSI 35 + 2.5σ | -529.9 | +287.4 | +210.1 | +335.6 | Read alone, that looks like the filter earning its place. But both sit inside the random range: the 2024 pool below RSI 35 won 48.78%, the 321 bars kept at 2σ won 47.35%, and the random range is 45.17 to 52.65%. **Discarding the same number of bars at random reaches the same place.** ## The short fade does not get fixed by combining The short side lost almost everywhere, alone and paired. On the hourly chart in 2025 all 15 finished negative, the shallowest being RSI above 65 at -324.6 pips. This series has measured the short fade breaking in a rising market several times, and 2024 rose from 140.87 to 157.20. But 2025 opened at 157.23 and closed at 156.67, and **the short side still lost in all 15 hourly runs**. Direction alone does not account for it. Combining does not soften it either. Hourly 2025, RSI above 70 alone returns -1,040.3 pips; adding the 2σ touch gives -1,158.0, and the win rate falls from 42.47% to 32.08%. ## Cost Spread paid matched trade count times 0.3 pips in all 18 hourly runs. | Side | Condition | Trades | Spread paid | Trades x 0.3 | |---|---|---|---|---| | long | RSI 30 + 1.5σ | 82 | 24.5 | 24.6 | | long | RSI 30 + 2σ | 74 | 22.1 | 22.2 | | long | RSI 35 + 1.5σ | 130 | 38.9 | 39.0 | | short | RSI 70 + 2σ | 53 | 15.8 | 15.9 | | short | RSI 65 + 1.5σ | 127 | 38.0 | 38.1 | Combining does cut the bill. RSI below 35 takes 145 trades and pays 43.5 pips; adding the 2.5σ touch takes 87 and pays 26.1. **The 17.4 pips saved cost 58 trades.** ## Related articles - [Is buying RSI below 30 profitable?](/blog/rsi-settings): the pool used here, tested on its own across 330 settings - [Bollinger Bands: mean reversion or breakout?](/blog/bollinger-band-settings): the band on its own, across 315 settings - [Candlestick pattern win rates](/blog/candlestick-patterns-tested): another case where the answer only appears once a control is placed beside it - [Is the engulfing candle profitable?](/blog/engulfing-settings): filters added to an entry, measured the same way - [Do Trend Filters Remove False Signals? 1,596 Backtests](/blog/false-signal-filters): the random range built here, applied across all 38 published settings ## Notes The data is USD/JPY from 1 January 2024 to 31 December 2025. Fills are at the close, the spread is fixed at 0.3 pips, and the lot is 0.1. RSI uses a period of 14 and the band a period of 20. The random range comes from drawing the pairing's own number of bars out of the pool a thousand times and taking the 2.5th and 97.5th percentiles of the resulting win rates; the generator uses a fixed seed. Where the pairing keeps the whole pool there is nothing to draw, so the pool's own rate is used. Because the horizons run out to 20 bars, that many bars at the end of each year are excluded from the counts. --- ### Do Indicator Settings Transfer Between Pairs? 3,024 Tests URL: https://formiq.jp/blog/settings-across-pairs Language: en Published: 2026-09-10 Updated: 2026-09-10 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Currency pairs, EURUSD, GBPUSD, EURJPY, Backtest, USDJPY Evidence: https://formiq.jp/evidence/settings-across-pairs Evidence JSON: https://formiq.jp/evidence/settings-across-pairs/data.json Also available in: ja — https://formiq.jp/ja/blog/settings-across-pairs Every indicator article here carries a "run this setting" button. **At the time of this run there were 44 of those buttons, and every one held a setting chosen on USD/JPY.** So the question a reader asks next, whether the setting also works on EUR/USD, has never been answered. (Two articles do measure other pairs: the currency-strength one covers 28 of them, and the article on how long RSI stays overbought runs EUR/USD beside USD/JPY. Neither carries a trading rule on its button, so neither is among the 42 below.) Answering it depends on not choosing the settings again. Each article carries a "run this setting" button holding exactly the rules that article published. **Those 42 settings ran again with only the pair and the window changed**: four pairs including EUR/USD, GBP/USD and EUR/JPY, three years from 2023 to 2025, three timeframes. Three results before the detail. **One. The ordering does not carry.** Ranking the 42 settings by annual pips on USD/JPY and comparing with the same 42 ranked on another pair gives correlations from -0.515 to +0.47 across 27 cells, median -0.011. Only 13 of the 27 are above zero. **Two. The count of settings profitable on both pairs is about what two unrelated pairs would give.** If 36 of 42 are profitable on USD/JPY and 5 on EUR/USD, then unrelated pairs would overlap on 36/42 x 5 = 4.3 settings. The measured overlap was 3. Across 27 cells the difference runs -4 to +6, median -0.2. **Three. No setting was profitable on all four pairs in all three years, on any of the three timeframes.** ## Why this test was missing The indicator articles are all built the same way: sweep dozens of settings for one indicator, count how many were profitable across two years, then take the previous year's winner and run it in the next. The across-years check happens every time. **What never happened was an across-pairs check.** The cost of leaving it out shows up immediately. Settings that ended the year with positive pips on the hourly chart, out of 42: | Year | USD/JPY | EUR/USD | GBP/USD | EUR/JPY | |---|---|---|---|---| | 2023 | 22 | 27 | 16 | 31 | | 2024 | 36 | 5 | 6 | 8 | | 2025 | 30 | 24 | 19 | 14 | **In 2024, 36 of 42 settings made money on USD/JPY. The same 42 settings managed 5 on EUR/USD in the same year.** The median year says it too. Pairs move different distances, so each year is divided by that pair's own mean bar range (14-bar true range). The unit is "how many average bars did the year make". | Year | USD/JPY | EUR/USD | GBP/USD | EUR/JPY | |---|---|---|---|---| | 2023 | +3.30 | +6.86 | -10.88 | +29.88 | | 2024 | +47.17 | -51.94 | -59.20 | -25.48 | | 2025 | +24.45 | +8.05 | -6.33 | -28.46 | **USD/JPY is the only pair whose median was positive in all three years.** GBP/USD was negative in all three, EUR/USD in one, EUR/JPY in two. So the 42 settings were chosen inside one pair, in a window where even the median setting made money. That is not a criticism of the settings. **It is a fact about where they were found.** ## How this was measured | Item | Value | |---|---| | Instruments | USD/JPY, EUR/USD, GBP/USD, EUR/JPY | | Timeframes | 15-minute, 1-hour, 4-hour | | Period | 1 January 2023 to 31 December 2025, split by year | | Settings | The 42 published in these articles | | Changed | The pair and the window, nothing else | | Held still | Rules, 0.1 lots, fills at the close, 0.3 pip spread | | Pips | Each pair's own pip (0.01 on a yen pair, 0.0001 on a dollar pair) | | Comparison | Each year also divided by that pair's mean 14-bar range | | Runs | 42 settings x 4 pairs x 3 timeframes x 3 years x 2 spreads = 3,024 | **The settings were not chosen for this article.** Each article's "run this setting" button holds them. There are 44 such buttons, and the three Bollinger articles publish identical rules (period 20, 2σ, breakout), which leaves 42. **The pairs were chosen for a reason.** EUR/USD and GBP/USD share USD/JPY's dollar and drop the yen. EUR/JPY shares the yen and drops the dollar. If USD/JPY settings travel to EUR/JPY but not to EUR/USD, the yen leg is what carries; if it goes the other way round, the dollar does. With no cross in the set, neither could be told apart from "some pairs are just easier". **Pips are not added across pairs untouched.** One hourly bar averages 24.85 pips on USD/JPY, 15.27 on EUR/USD, 17.12 on GBP/USD and 24.72 on EUR/JPY. The same "1,000 pips" means different things, so every annual figure is also shown divided by that range. ## Does the USD/JPY ordering predict another pair's Rank the 42 settings by annual pips on USD/JPY, rank the same 42 by annual pips on another pair, and measure how alike the two orderings are. +1 is the same order, 0 is no relation, -1 is reversed. Tied ranks are averaged. | Timeframe | Year | EUR/USD | GBP/USD | EUR/JPY | |---|---|---|---|---| | 15-minute | 2023 | -0.061 | +0.106 | +0.122 | | 15-minute | 2024 | +0.077 | -0.162 | -0.011 | | 15-minute | 2025 | +0.314 | +0.346 | +0.369 | | Hourly | 2023 | -0.027 | -0.185 | -0.126 | | Hourly | 2024 | -0.270 | -0.072 | -0.218 | | Hourly | 2025 | -0.198 | -0.266 | -0.515 | | 4-hour | 2023 | +0.363 | +0.238 | -0.022 | | 4-hour | 2024 | +0.180 | -0.106 | +0.272 | | 4-hour | 2025 | +0.382 | +0.470 | +0.117 | **The 27 cells run from -0.515 to +0.47 with a median of -0.011.** Thirteen are above zero. The signs cluster by timeframe, which is worth a second look. **All nine hourly cells are negative and seven of the nine 4-hour cells are positive.** Three of the 15-minute cells are negative. A real relationship between what worked on USD/JPY and what works elsewhere would not change sign when the timeframe changes. ### How many would overlap if the pairs were unrelated A correlation on its own is hard to size. Here is the same question in counts. **Suppose 36 of 42 settings were profitable on USD/JPY and 5 on EUR/USD. If the two pairs were entirely unrelated, the number profitable on both would be 36/42 x 5 = 4.3.** The measured count was 3. All 27 cells, measured count first and the unrelated-pairs count second: | Timeframe | Year | EUR/USD | GBP/USD | EUR/JPY | |---|---|---|---|---| | 15-minute | 2023 | 4 / 5.7 | 2 / 2.5 | 10 / 7.9 | | 15-minute | 2024 | 2 / 3.2 | 4 / 5.0 | 4 / 3.6 | | 15-minute | 2025 | 8 / 6.3 | 6 / 4.6 | 9 / 7.4 | | Hourly | 2023 | 12 / 14.1 | 7 / 8.4 | 16 / 16.2 | | Hourly | 2024 | 3 / 4.3 | 3 / 5.1 | 4 / 6.9 | | Hourly | 2025 | 18 / 17.1 | 12 / 13.6 | 6 / 10.0 | | 4-hour | 2023 | 6 / 5.9 | 8 / 5.4 | 6 / 6.3 | | 4-hour | 2024 | 6 / 5.3 | 4 / 5.3 | 20 / 19.8 | | 4-hour | 2025 | 16 / 14.9 | 22 / 16.0 | 8 / 5.7 | **The measured count differs from the unrelated-pairs count by -4 to +6, median -0.2.** Knowing a setting was profitable on USD/JPY barely moves the odds that it is profitable on another pair. ### Sharing the yen does not bring them closer This is what EUR/JPY is in the set for. | Timeframe | Median correlation, EUR/JPY | Median, EUR/USD and GBP/USD | |---|---|---| | 15-minute | +0.122 | +0.0915 | | Hourly | -0.218 | -0.1915 | | 4-hour | +0.117 | +0.3005 | **Sharing the yen did not produce a higher figure.** The 15-minute chart is marginally higher, the hourly marginally lower, and the 4-hour clearly lower at +0.117 against +0.3005. The EUR/JPY column is the middle of three values and the other the middle of six, which is why the second carries a fourth decimal. Neither a shared yen nor a shared dollar carried the ordering. ## Carrying USD/JPY's best setting to the others This series runs the same check across years every time. Here it runs across pairs. **For each timeframe and year, the setting with the highest annual pips on USD/JPY is picked and run on the other three**, with its rank on the new pair. | Timeframe | Year | USD/JPY's best | USD/JPY | EUR/USD | GBP/USD | EUR/JPY | |---|---|---|---|---|---|---| | 15-minute | 2023 | Parabolic SAR | +1,315.1 | -505.1 (24th) | -1,962.0 (34th) | -1,124.4 (36th) | | 15-minute | 2024 | Ichimoku | +2,261.3 | -1,553.8 (34th) | -720.8 (23rd) | -513.7 (14th) | | 15-minute | 2025 | Ichimoku | +2,723.2 | -1,589.4 (34th) | -1,306.5 (22nd) | +739.8 (7th) | | Hourly | 2023 | Supertrend | +1,539.6 | +1,414.8 (2nd) | +177.9 (13th) | +1,346.0 (14th) | | Hourly | 2024 | MACD | +3,152.1 | -389.3 (16th) | -163.2 (8th) | -373.5 (17th) | | Hourly | 2025 | WaveTrend | +3,283.4 | -612.5 (34th) | +66.2 (18th) | -1,421.3 (33rd) | | 4-hour | 2023 | Morning and evening star | +2,449.6 | +534.7 (5th) | +559.9 (6th) | +1,671.7 (4th) | | 4-hour | 2024 | Pin bar | +3,498.7 | +386.9 (1st) | -275.3 (13th) | +804.1 (13th) | | 4-hour | 2025 | Stochastic | +2,959.6 | +211.8 (22nd) | +507.4 (17th) | -1,082.0 (30th) | **Of the 27 carries, 12 made money.** The ranks spread from 1st to 36th. Sixteen beat the new pair's median and seven landed in its top ten. The timeframe splits the result. **The 15-minute chart made money in 1 of 9 carries; the 4-hour in 7 of 9.** The hourly managed 4. Two rows carry the point. The morning-and-evening-star setting that won the 4-hour chart in 2023 made money on EUR/USD, GBP/USD and EUR/JPY alike, ranking 5th, 6th and 4th. WaveTrend, which won the hourly chart in 2025 with +3,283.4 pips on USD/JPY, ranked 34th on EUR/USD and 33rd on EUR/JPY. **The same selection procedure produced one winner that travelled and one that did not.** ## Any setting that works on all four The strictest count the data allows: | Timeframe | Year | Profitable on all four | USD/JPY alone | |---|---|---|---| | 15-minute | 2023 | 0 | 15 | | 15-minute | 2024 | 2 | 19 | | 15-minute | 2025 | 3 | 24 | | Hourly | 2023 | 3 | 22 | | Hourly | 2024 | 1 | 36 | | Hourly | 2025 | 1 | 30 | | 4-hour | 2023 | 2 | 19 | | 4-hour | 2024 | 1 | 32 | | 4-hour | 2025 | 5 | 24 | **Between 0 and 5 of the 42 cleared all four pairs in a given year.** Requiring all four pairs and all three years at once leaves **0 of 42 on every timeframe**. The hourly chart in 2024 makes it plain: 36 of 42 were profitable on USD/JPY alone, and 1 was profitable on all four. ## What the spread removes depends on the timeframe A losing year can be a setting that was wrong or a setting that paid too much to trade. Re-running at a zero spread separates them. Profitable counts at a zero spread first, then at 0.3 pips: | Timeframe | Year | USD/JPY | EUR/USD | GBP/USD | EUR/JPY | |---|---|---|---|---|---| | 15-minute | 2023 | 23 → 15 | 21 → 16 | 13 → 7 | 27 → 22 | | 15-minute | 2024 | 23 → 19 | 14 → 7 | 19 → 11 | 12 → 8 | | 15-minute | 2025 | 31 → 24 | 17 → 11 | 9 → 8 | 14 → 13 | | Hourly | 2023 | 25 → 22 | 29 → 27 | 19 → 16 | 33 → 31 | | Hourly | 2024 | 36 → 36 | 10 → 5 | 7 → 6 | 8 → 8 | | Hourly | 2025 | 31 → 30 | 29 → 24 | 21 → 19 | 14 → 14 | | 4-hour | 2023 | 20 → 19 | 13 → 13 | 13 → 12 | 14 → 14 | | 4-hour | 2024 | 32 → 32 | 7 → 7 | 7 → 7 | 26 → 26 | | 4-hour | 2025 | 24 → 24 | 27 → 26 | 29 → 28 | 11 → 10 | **On the 15-minute chart the 0.3 pip spread lowered the count in all twelve cells, by as much as 8** (USD/JPY 2023, 23 down to 15). **On the 4-hour chart seven of the twelve cells lost nothing at all.** That matches what this series finds every time: cost is trade count times spread, and a shorter timeframe trades more. **It also means the gap between pairs is not a spread effect.** EUR/USD's 5 profitable settings on the hourly chart in 2024 are 10 even with the spread removed. ## Related reading - [Do Trend Filters Remove False Signals? 1,596 Backtests](/blog/false-signal-filters): the same 42 settings, crossed with filters rather than with pairs - [Is the Golden Cross Profitable on Forex? 280 Moving Average Tests](/blog/moving-average-cross-settings): one of the settings carried here - [Is the Ichimoku Three-Signal Buy Profitable? 280 Settings Tested](/blog/ichimoku-settings): the setting that won the 15-minute chart in 2024 and 2025 - [How to Backtest a Forex Strategy Without Coding](/blog/backtest-without-coding): how to re-run one of these settings on a pair of your own - [Do Support and Resistance Lines Work? 5,184 Backtests](/blog/support-resistance-settings): written after this one, and it found a condition that had never taken a trade ## Notes on scope - Four pairs over three years, on the 15-minute, 1-hour and 4-hour charts - Fills at the close, one 0.3 pip spread for every pair, 0.1 lots, no slippage. Real spreads differ by pair, so these figures exist to compare pairs with each other - One published setting per article, not that article's full parameter sweep - EUR/JPY is the only pair here without the dollar in it, so on the question of which leg matters this article can say only that neither carried --- ### Do Support and Resistance Lines Work? 5,184 Backtests URL: https://formiq.jp/blog/support-resistance-settings Language: en Published: 2026-09-10 Updated: 2026-09-10 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Support and resistance, Horizontal lines, Price action, Breakout, Backtest, USDJPY Evidence: https://formiq.jp/evidence/support-resistance-settings Evidence JSON: https://formiq.jp/evidence/support-resistance-settings/data.json Also available in: ja — https://formiq.jp/ja/blog/support-resistance-settings A horizontal line at a price that has already turned is the first thing a discretionary trader draws and the one thing no oscillator describes. The reason it is supposed to work is plain: **price stopped there before.** Put that in a form that can be measured and two questions fall out. One is a dose: if "it held three times" is the reason to trust the line, a line that held five times should hold better than one that held twice. The other is a control: **what happens at a line drawn where nothing ever stopped?** Three results first. **One. The touch count did not matter.** Splitting the bounce signals by how many times the line had already held gives 50.13% for three touches, 49.24% for four and 50.10% for five or more on the 15-minute long side. **Two. Price sits inside a line's tolerance on 41.40% to 65.75% of bars.** Roughly one bar in two, which makes being at a level ordinary rather than special. **Three. Of the four settings that decide what counts as a line, the touch count moved the annual result least** at 357.9 pips, against 1,000.9 for the tolerance band. And one more thing turned up before any of that. **The engine's horizontal-line condition had never taken a single trade.** Two separate faults did it, and either alone was enough. ## The condition had never taken a trade The first sweep for this article returned zero trades on all 2,592 runs. Two reasons. **The first was a missing ATR.** The tolerance is set as a multiple of ATR rather than in pips, because two touches "at the same price" mean different things on gold and on EUR/USD. The engine keeps a list of which conditions need the 14-bar range computed, and the horizontal line was not on it. So the range handed to the level builder was empty, the tolerance band came out as zero, every swing was discarded before it could join a cluster, and the trading test sits behind a check that the band is above zero. **The second was a comparison that can never be true.** Resistance is by construction the nearest line *above* the current close. The break reading asked whether the current close was above resistance. Support is by construction at or below the close, and the short side asked whether the close was below support. A break is a close through the line price was under a bar ago, so that is what it reads now. Both faults present as a property of the market rather than of the code: the condition appears in the editor, the backtest runs, and the report says no trades. This is the fourth fault of that shape this series has found, after the session filter, the fair-value-gap drawing and the GMMA exit. Every figure below is from after the fix. ## What the line is built from Four steps. 1. **Find the swings.** A bar is a swing high if its high is the highest for the given number of bars on each side. Because the right-hand side is needed, **a swing is only confirmed that many bars later**. 2. **Cluster nearby prices.** Two swings closer than the ATR tolerance are the same line, and the line's price moves to the average of what has touched it. 3. **Count the touches.** How many swings joined the cluster. 4. **Offer only the lines that reach the minimum.** The nearest one below the close is support, the nearest above is resistance. **Waiting for confirmation is the part that matters.** A swing is not a swing until the bars to its right exist, so reading a level before then draws support from a low the tester has not seen. It is the most flattering bug a backtest can have. Lines that fall outside the lookback are dropped, and not only for speed. **"This line held three times" means three times inside the lookback.** A price brushed once ten thousand bars ago no longer counts. ## Support and resistance settings, and how they are read **This condition draws nothing on the chart.** It exists inside the backtester, so what follows is about numbers rather than about lines on a screen. | Setting | Default | What it does | |---|---|---| | Reading | Bounce | Bounce (held at the line) or break (closed through it) | | Minimum touches | 3 | How many clustered swings before the line is offered. Two is the floor | | Tolerance | 0.5 | How close two swings must be to count as one line, in ATRs | | Swing strength | 3 | Bars on each side that make a swing | | Lookback | 200 | How far back a line may be drawn from | ### The two readings **Bounce** buys when the bar's low reaches inside support's tolerance and the close finishes above support. It closes at the next line up. The sell side mirrors it at resistance. **Break** buys when the close finishes above the resistance of the previous bar, and closes on the mirror break down through support. Both open either side, so entry and exit share one rule set. ## How this was measured | Item | Value | |---|---| | Instrument | USD/JPY | | Timeframes | 15-minute, 1-hour, 4-hour | | Period | 1 January 2024 to 31 December 2025, with 2025 also split in halves | | Readings | Bounce and break | | Lookback | 100, 200, 400 | | Swing strength | 2, 3, 5, 8 | | Tolerance | 0.25, 0.5, 1.0 ATR | | Minimum touches | 2, 3, 4 | | Controls | Prices 1.5, 2 and 3 ATRs off the line, and a reshuffled close series | | Runs | 2 readings x 3 x 4 x 3 x 3 = 216, x 3 timeframes x 4 windows x 2 spreads = 5,184 | | Spread | 0.3 pips and 0 | | Lot size | 0.1 | | Fill | Close of the signal bar | **Everything runs at a zero spread as well.** The bounce reading trades more than a thousand times a year on the hourly chart, so a losing year has to be separable into "the line did not work" and "the round trips did it". ## Price sits at a line on half the bars Before judging the line, count how many exist and how often price reaches one. At the defaults: lookback 200, swing 3, tolerance 0.5 ATR, three touches. | Timeframe | Year | Bars | Support exists | Price inside the band | Median touches | Max | |---|---|---|---|---|---|---| | 15-minute | 2025 | 24,893 | 81.28% | 59.90% | 4 | 38 | | 15-minute | 2024 | 24,999 | 86.72% | 58.20% | 4 | 24 | | Hourly | 2025 | 6,216 | 85.52% | 65.75% | 5 | 23 | | Hourly | 2024 | 6,250 | 84.24% | 52.64% | 4 | 15 | | 4-hour | 2025 | 1,600 | 85.50% | 61.63% | 5 | 15 | | 4-hour | 2024 | 1,616 | 84.90% | 41.40% | 3 | 8 | **Price is inside a line's tolerance on 41.40% to 65.75% of bars.** "It reached a level" happens on roughly one bar in two. The median touch count runs 3 to 5 and the maximum reaches 38 on the 15-minute chart in 2025. **Lines are not scarce.** ## Does a line that held more times hold better The dose. Bounce signals, split by how many times that line had already held, measured to the close ten bars later. The count is the engine's own, so nothing new is defined here. 2025: | Timeframe | Side | Touches | Signals | Hit rate | Mean over 10 bars | |---|---|---|---|---|---| | 15-minute | Long | 3 | 2,759 | 50.13% | -0.16 pips | | 15-minute | Long | 4 | 1,911 | 49.24% | -0.55 pips | | 15-minute | Long | 5 or more | 4,996 | 50.10% | -0.71 pips | | 15-minute | Short | 3 | 2,640 | 50.23% | +1.19 pips | | 15-minute | Short | 4 | 1,825 | 50.25% | -0.89 pips | | 15-minute | Short | 5 or more | 5,390 | 50.20% | +1.17 pips | | Hourly | Long | 3 | 640 | 50.63% | -0.01 pips | | Hourly | Long | 4 | 402 | 54.48% | +3.18 pips | | Hourly | Long | 5 or more | 1,809 | 51.96% | +0.72 pips | | Hourly | Short | 3 | 711 | 52.04% | +3.12 pips | | Hourly | Short | 4 | 446 | 49.33% | -1.22 pips | | Hourly | Short | 5 or more | 1,607 | 45.24% | -3.77 pips | **All six 15-minute buckets land between 49.24% and 50.25%**, which a coin also does. The hourly long side runs 50.63%, 54.48% and 51.96%, in no order. The short side runs 52.04%, 49.33% and 45.24%: **falling as the count rises.** If "it stopped there before, so it will stop again" is the reason, the number of times it stopped should show up. **It did not.** ## The same lines appear in a reshuffled series Two controls. **The first is a line moved off the real one.** The same test at prices 1.5, 2 and 3 ATRs above support, far enough that the tolerance bands cannot overlap. 2025: | Timeframe | Line | Signals | Hit rate | Close was this many ATRs above the line | |---|---|---|---|---| | 15-minute | Drawn | 9,666 | 49.94% | 0.49 | | 15-minute | +1.5 ATR | 2,748 | 47.49% | 0.46 | | 15-minute | +2 ATR | 1,695 | 47.67% | 0.49 | | Hourly | Drawn | 2,851 | 52.02% | 0.48 | | Hourly | +1.5 ATR | 624 | 50.48% | 0.46 | | Hourly | +2 ATR | 415 | 47.95% | 0.48 | | 4-hour | Drawn | 679 | 53.31% | 0.51 | | 4-hour | +1.5 ATR | 172 | 54.07% | 0.52 | | 4-hour | +2 ATR | 122 | 45.08% | 0.51 | The drawn line beat the offset line on hit rate in 13 of 18 comparisons. **The distance at the moment of the signal does match**: across the nine 2025 rows the close finished 0.48 to 0.51 ATRs above the drawn line and 0.46 to 0.54 above the offset ones. What does not match is the count. **The offset lines fire on 7.8% to 41.6% as many bars as the drawn one.** Support is by construction the nearest line below the close, so price reaches it constantly; a line 1.5 ATRs higher is only reached on bars that travelled that far. **The two sets are not the same size**, so this control is not enough on its own. **The second is a reshuffled series.** Every close-to-close change is kept exactly once and only the order is destroyed, so the drift, the volatility and the fat tails belong to the market. **What does not survive is where price turned.** A reshuffled bar has no wick, so both series are read on the close alone (the engine's bounce reads the low; these figures compare two series rather than report a traded rule). | Timeframe | Year | Series | Signals | Hit rate | Mean over 10 bars | |---|---|---|---|---|---| | 15-minute | 2025 | Real | 5,568 | 49.69% | -0.45 pips | | 15-minute | 2025 | Reshuffled | 2,720 | 50.04% | -1.83 pips | | Hourly | 2025 | Real | 1,637 | 51.13% | -0.51 pips | | Hourly | 2025 | Reshuffled | 542 | 47.23% | -0.15 pips | | 4-hour | 2025 | Real | 376 | 55.59% | +7.53 pips | | 4-hour | 2025 | Reshuffled | 172 | 52.91% | +3.58 pips | | 15-minute | 2024 | Real | 5,553 | 53.97% | +0.80 pips | | 15-minute | 2024 | Reshuffled | 2,328 | 49.44% | -1.00 pips | | Hourly | 2024 | Real | 1,179 | 59.20% | +5.17 pips | | Hourly | 2024 | Reshuffled | 642 | 52.80% | +1.24 pips | | 4-hour | 2024 | Real | 249 | 64.26% | +26.49 pips | | 4-hour | 2024 | Reshuffled | 114 | 36.84% | -24.74 pips | **The real series beat the reshuffled one on hit rate in 5 of 6 cells.** Only the 15-minute chart in 2025 went the other way, at 49.69% against 50.04%. So the order of the closes carries something. The real series' own hit rate still runs from 49.69% to 64.26%, though, and beating a reshuffle is not the same as surviving as a trade. ## Trading both readings 108 settings, traded. Settings that ended the year with positive pips: | Reading | Window | 15-minute | Hourly | 4-hour | |---|---|---|---|---| | Bounce | 2025 | 12 | 21 | 57 | | Bounce | 2024 | 49 | 79 | 91 | | Break | 2025 | 32 | 61 | 30 | | Break | 2024 | 49 | 95 | 92 | **In 2024 the bounce reached 91 of 108 and the break 92 on the 4-hour chart. In 2025 the same cells are 57 and 30.** Requiring both years, and then all four windows including the two halves of 2025: | Reading | Timeframe | Both years | All four windows | |---|---|---|---| | Bounce | 15-minute | 11 | 5 | | Bounce | Hourly | 19 | 7 | | Bounce | 4-hour | 50 | 21 | | Break | 15-minute | 23 | 10 | | Break | Hourly | 58 | 32 | | Break | 4-hour | 23 | 11 | **The strongest cell is the hourly break at 58 of 108, and 32 across all four windows.** The bounce's best is the 4-hour chart at 50 and 21. Win rates and trade sizes at the defaults: | Reading | Timeframe | Year | Trades | Win rate | Mean win | Mean loss | Annual pips | Bars held | |---|---|---|---|---|---|---|---|---| | Bounce | 15-minute | 2025 | 6,372 | 40.68% | +11.11 | -8.56 | -3,544.0 | 5.0 | | Bounce | Hourly | 2025 | 1,799 | 44.41% | +19.66 | -15.86 | -151.3 | 4.2 | | Bounce | 4-hour | 2024 | 261 | 45.59% | +63.49 | -33.52 | +2,794.6 | 7.9 | | Break | 15-minute | 2025 | 3,407 | 28.29% | +22.42 | -9.95 | -2,682.4 | 10.3 | | Break | Hourly | 2025 | 938 | 30.38% | +43.26 | -18.05 | +542.8 | 9.3 | | Break | 4-hour | 2024 | 131 | 38.93% | +121.24 | -46.84 | +2,436.1 | 16.7 | **The break wins 25.45% to 31.08% of its trades** and makes more than twice as much on a winner as it loses on a loser. The bounce wins 38.26% to 42.80% at a ratio of 1.3 to 1.9. ### The bounce leans long **Between 66.06% and 67.73% of the bounce's trades are buys.** For the break it is 50.01% to 50.19%. The engine checks the long side first when a bar satisfies both. A bounce off support and a bounce off resistance can both be true on one bar, and the buy takes it every time. **A rule that is symmetric by description trades long by construction.** The inside-bar article found the same long bias for the same reason. ## The setting that defines a line matters least Each setting moved one at a time, with the median annual pips of every setting group containing that value. Hourly chart, 2025, median in brackets: | Reading | Setting | Profitable count and median per value | |---|---|---| | Bounce | Lookback | 100: 2/36 (-1,291.5) 200: 7/36 (-1,131.0) 400: 12/36 (-384.6) | | Bounce | Swing strength | 2: 7/27 (-658.7) 3: 7/27 (-1,002.9) 5: 5/27 (-977.5) 8: 2/27 (-1,393.9) | | Bounce | Tolerance | 0.25: 15/36 (-315.4) 0.5: 3/36 (-1,316.3) 1.0: 3/36 (-1,049.7) | | Bounce | Minimum touches | 2: 4/36 (-1,020.5) 3: 9/36 (-1,183.6) 4: 8/36 (-825.7) | | Break | Lookback | 100: 13/36 (-521.9) 200: 20/36 (+97.2) 400: 28/36 (+1,002.4) | | Break | Swing strength | 2: 24/27 (+1,148.5) 3: 21/27 (+542.8) 5: 13/27 (-30.6) 8: 3/27 (-822.2) | | Break | Tolerance | 0.25: 20/36 (+261.3) 0.5: 18/36 (-36.0) 1.0: 23/36 (+286.1) | | Break | Minimum touches | 2: 23/36 (+441.4) 3: 22/36 (+423.7) 4: 16/36 (-128.3) | How far each setting moves that median from end to end: | Reading | Lookback | Swing strength | Tolerance | Minimum touches | |---|---|---|---|---| | Bounce | 906.9 | 735.2 | 1,000.9 | 357.9 | | Break | 1,524.3 | 1,970.7 | 322.1 | 569.7 | **For the bounce the touch count moves it least**, at 357.9 pips. For the break it is second smallest at 569.7, behind the tolerance at 322.1. The two that move it most are the break's swing strength (1,970.7) and lookback (1,524.3), and both improve in the direction that **produces more lines**. A swing strength of 2 makes almost every high and low a swing; a lookback of 400 keeps old lines alive. **This measurement cannot separate the lines working from the trade count rising.** What it does show is that the settings which make a line more line-like, how many times it held and how close two touches must be, moved the result less than the settings that make more of them. ## Last year's winner, this year The setting with the highest annual pips in one year, run in the other, with its rank among the 216 (two readings by 108 definitions): | Timeframe | Picked in | Setting | Pips that year | Pips the other year | Rank | Median that year | |---|---|---|---|---|---|---| | 15-minute | 2024 | Break, 100/2/0.5/3 | +3,513.9 | +488.6 | 26th | -1,531.7 | | 15-minute | 2025 | Bounce, 100/5/0.5/4 | +2,194.7 | +2,983.4 | 4th | -113.8 | | Hourly | 2024 | Break, 200/2/1.0/2 | +3,247.4 | +339.8 | 67th | -394.6 | | Hourly | 2025 | Break, 400/2/0.25/4 | +2,323.4 | +377.2 | 148th | +1,033.2 | | 4-hour | 2024 | Break, 200/8/1.0/3 | +3,237.1 | -1,205.0 | 170th | -314.8 | | 4-hour | 2025 | Bounce, 400/8/1.0/4 | +2,569.7 | +1,463.4 | 77th | +962.1 | **Four of the six beat the other year's median.** The 15-minute chart held up in both directions, at 26th and 4th; the 4-hour 2024 pick fell to 170th. The default setting moves just as far. The 4-hour bounce ranked 6th of 216 in 2024 with +2,794.6 pips and 72nd in 2025 with +278.8. The 4-hour break ranked 204th in 2025 at -1,883.7 and 15th in 2024 at +2,436.1. **One setting, one timeframe, 6th and 72nd, 204th and 15th.** ## What the spread takes A losing year can be the line failing or the round trips costing too much. Profitable counts at a zero spread first, then at 0.3 pips: | Reading | Timeframe | Year | Profitable | Median | Median trades | |---|---|---|---|---|---| | Bounce | 15-minute | 2024 | 94 → 49 | +1,339.7 → -99.1 | 4,738.5 | | Bounce | Hourly | 2025 | 39 → 21 | -545.6 → -987.7 | 1,428.0 | | Bounce | 4-hour | 2025 | 61 → 57 | +224.9 → +154.8 | 364.5 | | Bounce | 4-hour | 2024 | 93 → 91 | +912.5 → +845.2 | 196.0 | | Break | 15-minute | 2025 | 57 → 32 | +172.7 → -760.3 | 2,739.5 | | Break | Hourly | 2025 | 66 → 61 | +472.9 → +230.4 | 751.5 | **The 15-minute bounce falls from 94 profitable settings to 49 on the spread alone in 2024**, with the median going from +1,339.7 pips to -99.1. At a median 4,738.5 trades, the spread costs 1,421.5 pips. **On the 4-hour chart it falls from 93 to 91**, because the median is 196 trades. Cost is trade count times spread, which this series finds every time. ## Related reading - [Do Fair Value Gaps Really Fill? 2 Years of USD/JPY Tested](/blog/fair-value-gap-settings): the same "price remembers this level" claim, measured against a control - [Does Fibonacci 61.8% Really Hold? 2 Years of USD/JPY Tested](/blog/fibonacci-retracement-levels): whether a level has an effect of its own, tested by mixing in non-Fibonacci numbers - [Do Double Tops and Bottoms Reverse? 2 Years of USD/JPY Tested](/blog/double-top-bottom-settings): the same swings read as a shape instead of a line - [Do Indicator Settings Transfer Between Pairs? 3,024 Tests](/blog/settings-across-pairs): these settings were chosen on USD/JPY too ## Notes on scope - USD/JPY only, 2024 and 2025, on the 15-minute, 1-hour and 4-hour charts - Fills at the close, spreads of 0.3 pips and 0, 0.1 lots, no slippage - The condition draws nothing on the chart; these are lines computed inside the backtester - The offset control matches the signal-time distance but not the number of signals, and the reshuffled control reads both series on the close alone --- ### Candlestick Pattern Win Rates: 17 Patterns Tested URL: https://formiq.jp/blog/candlestick-patterns-tested Language: en Published: 2026-09-09 Updated: 2026-09-09 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Candlestick patterns, Price action, Engulfing, Hammer, Backtest, USDJPY Evidence: https://formiq.jp/evidence/candlestick-patterns-tested Evidence JSON: https://formiq.jp/evidence/candlestick-patterns-tested/data.json Also available in: ja — https://formiq.jp/ja/blog/candlestick-patterns-tested A candlestick pattern is a name given to the shape of one, two or three bars. The long lower wick is a **hammer**, a bar that swallows the one before it is an **engulfing candle**, three rising bars in a row are **three white soldiers**. Each arrives with a reading attached: price reverses here, or price continues from here. Formiq's backtest carries 17 of them: hammer, shooting star, pin bar, marubozu, large candle, dragonfly doji, gravestone doji, doji, engulfing, harami, tweezers, inside bar break, morning star, evening star, three white soldiers, three black crows, and three same-coloured candles in a row. All 17 were counted on the same charts with the same exit, on USD/JPY through 2024 and 2025. The comparison was fixed before anything was counted. **Every pattern is measured against the bars that formed no pattern.** "Price was higher ten bars after a hammer 54.58% of the time" says nothing on its own, because 50.35% of all hourly bars in 2025 were higher ten bars later. The difference is 4.22 points. That difference did not survive the change of year. **Across 78 cells — 26 pattern directions on three timeframes — the excess over the control was positive in both 2024 and 2025 in 19 of them.** ## What the seventeen patterns actually test Every one of them reduces to the sizes of the body (open to close), the wicks, and the range (high to low). No indicator is calculated. The table gives each condition and how often it occurred on the hourly chart in 2025. | Pattern | Bars | Condition | Long | Short | 2025 rate | |---|---|---|---|---|---| | Hammer | 1 | lower wick ≥ body × 2, upper wick ≤ body | ○ | | 4.92% | | Shooting star | 1 | upper wick ≥ body × 2, lower wick ≤ body | | ○ | 4.17% | | Pin bar | 1 | one wick ≥ body × 2 and longer than the other | ○ | ○ | 28.26% | | Marubozu | 1 | body ÷ range ≥ 0.9 | ○ | ○ | 2.92% | | Large candle | 1 | range ≥ ATR(14) × 1.5 and body ÷ range ≥ 0.7 | ○ | ○ | 4.46% | | Dragonfly doji | 1 | body ÷ range ≤ 0.1, lower wick ≥ range × 0.6, upper ≤ range × 0.05 | ○ | | 0.31% | | Gravestone doji | 1 | body ÷ range ≤ 0.1, upper wick ≥ range × 0.6, lower ≤ range × 0.05 | | ○ | 0.21% | | Doji | 2 | body ÷ range ≤ 0.05; long after a down bar, short after an up bar | ○ | ○ | 5.08% | | Engulfing | 2 | the body covers the previous body and the colour flips | ○ | ○ | 13.50% | | Harami | 2 | the body sits inside the previous body and the colour flips | ○ | ○ | 12.31% | | Tweezers | 2 | two lows (or highs) within 0.3 × the average range, colour flips | ○ | ○ | 35.68% | | Inside bar break | 3 | bar two sits inside bar three, and the current bar breaks out | ○ | ○ | 8.57% | | Morning star | 3 | down bar, small bar, up bar; the outer bodies twice the middle | ○ | | 3.38% | | Evening star | 3 | the morning star inverted | | ○ | 3.35% | | Three white soldiers | 3 | three up bars, each opening at or above and closing above the last | ○ | | 11.63% | | Three black crows | 3 | three down bars, the mirror condition | | ○ | 10.73% | | Three in a row | 3 | three bodies of the same colour | ○ | ○ | 23.42% | The rate is the share of the 6,216 hourly bars of 2025. Tweezers appeared on 35.68% of them and the pin bar on 28.26%, while the gravestone doji appeared 13 times, on 0.21%. **One word covers both a shape that occurs on one bar in three and a shape that occurs 13 times a year.** ## The settings behind each pattern None of them draws anything. These conditions read the candles already on the chart, so no line and no lower pane appears. In the backtest builder each one is a named condition, and the eight with a threshold expose it as a number. | Condition | Field | Default | What changes | |---|---|---|---| | Hammer | lower wick ratio | 2 | raising it keeps only longer lower wicks and cuts the count | | Shooting star | upper wick ratio | 2 | the same filter on the upper wick | | Pin bar | wick ratio | 2 | applies to the long and the short side together | | Marubozu | body ratio | 0.9 | closer to 1 leaves only bars with almost no wick | | Large candle | ATR multiple | 1.5 | how many times the current ATR(14) the range must reach | | Large candle | body ratio | 0.7 | the minimum share of the range the body must fill | | Doji | body ratio | 0.05 | raising it lets bars with visible bodies count as doji | | Tweezers | tolerance | 0.3 | how far apart two lows (or highs) can sit and still match, as a share of the average range | | Three in a row | count | 3 | how many same-coloured bodies are required | The other nine conditions have no numeric field at all. Dragonfly doji, gravestone doji, engulfing, harami, inside bar break, morning star, evening star, three white soldiers and three black crows are defined entirely by the shape. ### The published trading rule The presentation is the same in every case. Reversal shapes (hammer, shooting star, pin bar, dragonfly doji, gravestone doji, doji, engulfing, harami, tweezers, morning star, evening star) are read as a turn and traded against the move that preceded them; continuation shapes (marubozu, large candle, three white soldiers, three black crows, three in a row, inside bar break) are read as a push and traded with it. Every test here enters in the direction the pattern points, at the close of the bar that completed it. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Timeframes | 15-minute, hourly, 4-hour | | Period | 1 January 2024 to 31 December 2025 | | Detection | the same code the backtest runs, bar by bar | | Control | every bar of the same window and timeframe | | Horizon | the close 1, 5, 10 and 20 bars after the signal bar | | Exit | the close 10 bars after entry, with no stop, target or opposite signal | | Spread | 0.3 pips, fixed | | Lot | 0.1 | | Fill | at the close | Fixing the exit at a bar count is what makes the 17 comparable. Eight of them (hammer, dragonfly doji, morning star, three white soldiers, shooting star, gravestone doji, evening star, three black crows) only ever produce one side. A hammer is a buy and there is no hammer that sells, so there is no opposite signal to close against. Giving half the set a signal exit and the other half a clock would turn the comparison into a comparison of exits. ## Seventeen names, but not seventeen different bars Before counting anything forward, the conditions were checked against each other: how often two of them fired on the same bar in the same direction, hourly, 2025. | Inner condition | Outer condition | Shared | Share of the inner | Share of the outer | |---|---|---|---|---| | Hammer (long) | Pin bar (long) | 306 | 100% | 32.66% | | Dragonfly doji | Pin bar (long) | 19 | 100% | 2.03% | | Three white soldiers | Three bullish in a row | 723 | 100% | 95.38% | | Shooting star | Pin bar (short) | 259 | 100% | 31.59% | | Gravestone doji | Pin bar (short) | 13 | 100% | 1.59% | | Three black crows | Three bearish in a row | 667 | 100% | 95.56% | | Gravestone doji | Shooting star | 12 | 92.31% | 4.63% | **All 723 three-soldier signals were also three bullish candles in a row.** There were 758 such runs, so the extra clauses that make a run of three into three white soldiers (each open at or above the previous open, each close above the previous close) removed 35 bars, 4.62% of them. Three black crows behave the same way: 667 of the 698 bearish runs qualified. The hammer is the same story. All 306 hammers were pin bars, 32.66% of the 937 long pin bars. **A hammer is the long side of a pin bar with one clause added: the upper wick must not exceed the body.** The overlap matters when the results arrive. Comparing three white soldiers with a plain run of three is comparing a set of bars with 95% of itself. ## Against the bars that made no pattern, the edge disappears Over the 6,216 hourly bars of 2025, 50.35% closed higher ten bars later and 49.63% closed lower. That is the control. The 26 pattern directions, counted the same way: | Pattern | Side | 2025 count | 2025 rate | vs control | 2024 count | 2024 rate | vs control | |---|---|---|---|---|---|---|---| | Hammer | long | 306 | 54.58% | +4.22 | 347 | 56.48% | +1.57 | | Shooting star | short | 259 | 49.03% | -0.60 | 247 | 44.53% | -0.47 | | Pin bar | long | 937 | 50.69% | +0.34 | 1045 | 56.46% | +1.55 | | Pin bar | short | 820 | 51.10% | +1.47 | 817 | 44.92% | -0.09 | | Marubozu | long | 86 | 63.95% | +13.60 | 114 | 61.40% | +6.49 | | Marubozu | short | 96 | 52.08% | +2.45 | 93 | 40.86% | -4.15 | | Large candle | long | 128 | 61.72% | +11.36 | 136 | 56.62% | +1.71 | | Large candle | short | 149 | 48.32% | -1.31 | 130 | 43.08% | -1.93 | | Dragonfly doji | long | 19 | 47.37% | -2.99 | 24 | 33.33% | -21.58 | | Gravestone doji | short | 13 | 38.46% | -11.17 | 11 | 45.45% | +0.45 | | Doji | long | 158 | 44.94% | -5.42 | 149 | 59.06% | +4.15 | | Doji | short | 158 | 48.73% | -0.90 | 192 | 40.63% | -4.38 | | Engulfing | long | 443 | 52.60% | +2.24 | 415 | 56.14% | +1.23 | | Engulfing | short | 396 | 54.29% | +4.66 | 403 | 49.38% | +4.37 | | Harami | long | 383 | 53.52% | +3.17 | 405 | 53.33% | -1.58 | | Harami | short | 382 | 49.74% | +0.11 | 397 | 45.84% | +0.84 | | Tweezers | long | 1075 | 50.88% | +0.53 | 1035 | 54.11% | -0.81 | | Tweezers | short | 1143 | 49.52% | -0.11 | 1156 | 44.72% | -0.28 | | Inside bar break | long | 276 | 52.54% | +2.18 | 301 | 53.49% | -1.42 | | Inside bar break | short | 257 | 47.86% | -1.77 | 298 | 41.61% | -3.40 | | Morning star | long | 210 | 50.00% | -0.35 | 215 | 53.95% | -0.96 | | Evening star | short | 208 | 50.00% | +0.37 | 217 | 37.33% | -7.68 | | Three white soldiers | long | 723 | 47.30% | -3.05 | 824 | 56.55% | +1.64 | | Three black crows | short | 667 | 49.03% | -0.60 | 654 | 47.09% | +2.09 | | Three in a row | long | 758 | 47.49% | -2.86 | 854 | 56.21% | +1.29 | | Three in a row | short | 698 | 49.28% | -0.35 | 670 | 46.72% | +1.71 | **The morning star finished below its control in both years.** It occurred 210 and 215 times on the hourly chart, and 50.00% and 53.95% of those signals were followed by a higher close ten bars later, against controls of 50.35% and 54.91%. The evening star came in at 37.33% in 2024 against a 45.01% control, 7.68 points short. The two widest gaps, the long marubozu at +13.60 and the long large candle at +11.36, came from 86 and 128 occurrences. Both are continuation shapes: a large bullish body followed by a higher close, not a reversal. Counting the 15-minute and 4-hour charts the same way gives 78 cells. **The excess was positive in both years in 19 of them.** The sign agreed across the two years in 43, against the 39 a coin would give. Restricted to the 57 cells with at least 100 signals in each year, the correlation between the 2024 excess and the 2025 excess is -0.15, and the sign agreed in 31 of the 57. **Beating the control one year did not predict beating it the next.** No pattern direction cleared its control in all six cells (three timeframes × two years). The long hammer, the long large candle and the short harami each managed five. The long hammer's miss was the 15-minute chart in 2025, by 1.13 points. ## A 58% win rate mostly records whether the year went up The control itself moves a long way between years and timeframes. With no pattern involved, entering on every bar and holding a fixed number of bars: | Timeframe | Year | 1 bar | 5 bars | 10 bars | 20 bars | Bars | |---|---|---|---|---|---|---| | 15-minute | 2025 | 50.22% | 50.71% | 50.34% | 50.60% | 24,893 | | 15-minute | 2024 | 51.39% | 52.11% | 52.95% | 54.71% | 24,999 | | Hourly | 2025 | 50.39% | 50.73% | 50.35% | 50.63% | 6,216 | | Hourly | 2024 | 51.70% | 54.66% | 54.91% | 57.38% | 6,250 | | 4-hour | 2025 | 50.78% | 50.97% | 49.56% | 48.68% | 1,600 | | 4-hour | 2024 | 53.53% | 57.12% | 59.16% | 62.31% | 1,616 | USD/JPY opened 2024 at 140.87, reached 161.95 on 3 July and 139.58 on 16 September, and closed the year at 157.20. It opened 2025 at 157.23, bottomed at 139.89 on 22 April, and closed at 156.67. One year ended higher than it began and the other ended where it began, and the table records exactly that. **On the 4-hour chart in 2024, buying any bar at all and holding ten bars was followed by a higher close 59.16% of the time. The same thing in 2025 gives 49.56%.** That 9.6-point gap is close to the widest gap any pattern produced (the long marubozu, 13.60 points). When a bullish pattern is quoted at a 58% win rate, the control has to be quoted beside it before the number means anything. ## Requiring a downtrend before the hammer did not help Most write-ups say the shape alone is not enough: a hammer counts only after a decline, and the identical shape inside an uptrend does not. Each set of signals was split by the close-to-close move over the five bars leading into the pattern — the way the pattern is supposed to reverse, or the other way. The 128 cells where both halves held at least 30 signals: | Pattern | Side | All | Textbook context | Against it | |---|---|---|---|---| | Hammer | long | 54.58% | 54.61% (152) | 54.90% (153) | | Engulfing | long | 52.60% | 49.11% (224) | 56.16% (219) | | Engulfing | short | 54.29% | 53.40% (206) | 55.03% (189) | | Morning star | long | 50.00% | 48.57% (105) | 51.43% (105) | | Evening star | short | 50.00% | 47.52% (101) | 52.34% (107) | | Three white soldiers | long | 47.30% | 49.86% (355) | 44.84% (368) | | Three black crows | short | 49.03% | 44.41% (331) | 53.43% (335) | **The textbook half had the higher rate in 58 of the 128 cells.** On the hourly chart in 2025 the hammer read 54.61% with the decline and 54.90% without it, and the long engulfing was 7.05 points better when the five bars before it had risen. The requirement also halves the number of signals. Halving the count without moving the rate leaves nothing behind. ## Backtested: seven patterns at most cleared both years Counting closes is not trading. Each pattern was entered at the close of the signal bar at 0.3 pips of spread, 0.1 lots, and closed ten bars later. ### Hourly | Pattern | 2025 trades | 2025 win rate | 2025 net pips | 2024 trades | 2024 win rate | 2024 net pips | |---|---|---|---|---|---|---| | Hammer | 202 | 53.96% | +791.5 | 226 | 57.08% | +1721.2 | | Shooting star | 182 | 51.10% | +55.9 | 173 | 46.82% | -68.1 | | Pin bar | 458 | 51.97% | +1229.8 | 462 | 53.03% | +977.2 | | Marubozu | 148 | 58.78% | +624.3 | 147 | 53.74% | +55.1 | | Large candle | 198 | 52.02% | +53.3 | 171 | 52.63% | +327.1 | | Dragonfly doji | 18 | 44.44% | -90.1 | 23 | 34.78% | -232.8 | | Gravestone doji | 13 | 38.46% | -127.1 | 11 | 45.45% | -13.0 | | Doji | 211 | 49.76% | +143.8 | 219 | 49.77% | -502.4 | | Engulfing | 356 | 51.40% | +1000.9 | 368 | 54.08% | +2449.2 | | Harami | 344 | 47.67% | -662.3 | 362 | 46.13% | -2657.8 | | Tweezers | 481 | 51.14% | +1059.0 | 484 | 47.31% | -1114.6 | | Inside bar break | 304 | 51.32% | +727.0 | 312 | 45.83% | -360.4 | | Morning star | 164 | 48.17% | -68.5 | 175 | 54.29% | -281.3 | | Evening star | 161 | 51.55% | +817.6 | 169 | 38.46% | -1619.7 | | Three white soldiers | 265 | 48.68% | +663.8 | 292 | 53.77% | +393.0 | | Three black crows | 259 | 49.81% | +220.2 | 264 | 44.32% | -662.8 | | Three in a row | 407 | 46.68% | -1249.9 | 408 | 50.74% | +1418.5 | Six patterns finished both years in profit on the hourly chart: hammer, pin bar, marubozu, large candle, engulfing and three white soldiers. On the 15-minute chart two did (marubozu and morning star) and on the 4-hour chart seven did (hammer, shooting star, gravestone doji, doji, engulfing, inside bar break and three white soldiers). The 4-hour gravestone doji traded twice in 2025 and once in 2024. **No pattern cleared both years on all three timeframes.** The hammer and the engulfing candle, which both cleared the hourly chart, do not survive the 15-minute one: the hammer went from +1376.9 pips in 2024 to -25.2 in 2025, and the engulfing candle lost money in both years there (-485.2 and -496.7). The 15-minute losses are a trade-count problem. Tweezers traded 1,952 times in a year and paid 585.5 pips of spread doing it. ### Split 2025 in half Every combination that cleared both years, measured again over January to June and July to December of 2025: | Timeframe | Pattern | 2025 | Jan-Jun | Jul-Dec | |---|---|---|---|---| | 15-minute | Marubozu | +365.0 | +677.1 | -312.0 | | 15-minute | Morning star | +416.0 | -342.2 | +758.3 | | Hourly | Hammer | +791.5 | -66.2 | +857.7 | | Hourly | Pin bar | +1229.8 | +782.6 | +439.4 | | Hourly | Marubozu | +624.3 | -81.2 | +705.5 | | Hourly | Large candle | +53.3 | +34.1 | +19.2 | | Hourly | Engulfing | +1000.9 | +395.5 | +604.6 | | Hourly | Three white soldiers | +663.8 | -457.3 | +1121.1 | | 4-hour | Hammer | +122.6 | -516.6 | +655.9 | | 4-hour | Shooting star | +73.8 | +620.3 | -546.5 | | 4-hour | Gravestone doji | +335.3 | +240.8 | +94.5 | | 4-hour | Doji | +1955.3 | +2302.2 | -346.8 | | 4-hour | Engulfing | +395.4 | -66.2 | +619.6 | | 4-hour | Inside bar break | +969.2 | +456.2 | +490.3 | | 4-hour | Three white soldiers | +1136.1 | +454.8 | +674.0 | Six of the 15 combinations were profitable in both halves. The hourly three white soldiers finished 2025 at +663.8 pips and the first half of it at -457.3. Two profitable years is not the same as a consistently profitable rule. ## Buying last year's winner Take whichever pattern earned the most in 2024 and hold it through 2025. | Timeframe | 2024 winner | 2024 net pips | 2025 net pips | 2025 rank | 2025 median of 17 | |---|---|---|---|---|---| | 15-minute | Hammer | +1376.9 | -25.2 | 7th | -372.7 | | Hourly | Engulfing | +2449.2 | +1000.9 | 3rd | +220.2 | | 4-hour | Hammer | +1641.7 | +122.6 | 7th | +73.8 | All three beat the following year's median, which is the opposite of what most indicators in this series have done. The amounts shrank in every case: +1376.9 pips became -25.2 on the 15-minute chart. Seventh place still beats the median there because 11 of the 17 patterns lost money on the 15-minute chart in 2025. ## Does three in a row need to be three? Three white soldiers, three black crows, the morning star and the evening star are all defined on three bars. Running the count from two to six changes how many same-coloured bodies are required before entry. | Timeframe | Count | 2025 trades | 2025 net pips | 2024 trades | 2024 net pips | |---|---|---|---|---|---| | Hourly | 2 | 517 | -778.8 | 524 | -699.5 | | Hourly | 3 | 407 | -1249.9 | 408 | +1418.5 | | Hourly | 4 | 261 | -484.0 | 291 | +527.5 | | Hourly | 5 | 152 | +155.1 | 166 | +2253.7 | | Hourly | 6 | 84 | +32.2 | 90 | +710.2 | Five and six cleared both years on the hourly chart. **Three was the worst of the five on that chart: +1418.5 pips in 2024 and -1249.9 in 2025, the widest swing in the ladder.** Neither the 15-minute nor the 4-hour chart produced a count that cleared both years, so 2 of the 15 combinations survived and both of them sat at five or six bars. Longer runs are rarer. The hourly six-bar run traded 84 times in 2025, and +32.2 pips over 84 trades is 0.38 pips each. At a 0.7-pip spread instead of 0.3, that year finishes negative. ## Stops, targets and cost ### The win rate is set by the exit The same entry, with the exit changed. Hourly hammer: | Stop / target | 2025 trades | 2025 win rate | 2025 net pips | 2024 net pips | |---|---|---|---|---| | none (10-bar exit) | 202 | 53.96% | +791.5 | +1721.2 | | 30 / 60 | 231 | 43.29% | +213.1 | +464.5 | | 60 / 30 | 230 | 62.61% | +318.5 | +369.2 | | 50 / 50 | 224 | 52.23% | +49.6 | +492.4 | Identical entries, and the win rate runs from 43.29% to 62.61%. **A stop at twice the target raises the win rate and lowers the net pips in both years.** A pattern advertised at "60% winners" is describing an exit as much as an entry. The trade count rises from 202 to about 230 because a stop or target ends a position early, which frees the engine to take signals it would otherwise have sat through. ### Cost Spread paid matched trade count × 0.3 pips in every one of the 51 timeframe-pattern runs. | Timeframe | Pattern | Trades | Spread paid | Trades × 0.3 | |---|---|---|---|---| | 15-minute | Tweezers | 1952 | 585.5 | 585.6 | | 15-minute | Pin bar | 1838 | 551.3 | 551.4 | | 15-minute | Engulfing | 1455 | 436.5 | 436.5 | | Hourly | Tweezers | 481 | 144.3 | 144.3 | | 4-hour | Tweezers | 124 | 37.2 | 37.2 | Removing the spread entirely flipped the sign of 2 runs out of 51: the 15-minute hammer and the 15-minute large candle. **The rest of the losses come from the entry, not from paying for it too often.** ## Related articles - [Is the engulfing candle profitable?](/blog/engulfing-settings): one of these 17 taken apart, against the colour flips that did not engulf - [Do morning and evening stars reverse price?](/blog/star-patterns-settings): the three-bar pair taken apart, starting with the gap they are named for - [Is the inside bar breakout profitable?](/blog/inside-outside-bar-settings): taking the break and fading it, against an ordinary break of the same level - [Is the pin bar profitable?](/blog/pin-bar-settings): the wick measured against the body and against the range, on the same bars - [Do fair value gaps really get filled?](/blog/fair-value-gap-settings): the same control group applied to another price-action claim - [Double tops and double bottoms tested](/blog/double-top-bottom-settings): reversal rates for multi-bar shapes over the same two years - [What Heikin Ashi actually changes](/blog/heikin-ashi-settings): a chart that rewrites the candles themselves - [SMC: structure breaks and liquidity sweeps](/blog/smc-settings): another vocabulary laid over the same price data ## Notes The data is USD/JPY from 1 January 2024 to 31 December 2025. Fills are at the close, the spread is fixed at 0.3 pips, and the lot is 0.1. Detection uses the same code path as the backtest and reads bars from before the test window. Because the horizons run 1, 5, 10 and 20 bars forward, that many bars at the end of each year are excluded from the counts. The dragonfly doji and the gravestone doji occur around 20 times a year on the hourly chart and between 1 and 6 times on the 4-hour chart, so their figures move a long way on a single trade. --- ### Is the Engulfing Candle Profitable? 396 Backtests on USD/JPY URL: https://formiq.jp/blog/engulfing-settings Language: en Published: 2026-09-09 Updated: 2026-09-09 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Engulfing, Price action, Candlestick patterns, False break, Backtest, USDJPY Evidence: https://formiq.jp/evidence/engulfing-settings Evidence JSON: https://formiq.jp/evidence/engulfing-settings/data.json Also available in: ja — https://formiq.jp/ja/blog/engulfing-settings An engulfing candle is two bars. The second one is the opposite colour to the first, and its body covers the first one's body. A large up bar after a down bar is read as a buy; a large down bar after an up bar is read as a sell. Two separate things are packed into that. **The colour flips, and the body covers.** Colour flips are not rare: the hourly chart of 2025 held 3,166 of them, and only 839 covered the previous body. So the comparison here is not against all bars. It is against the 2,327 pairs that flipped colour and failed to cover. If the covering carries the pattern, those two groups have to behave differently. **On the hourly chart they did.** Ten bars later the engulfing signals had moved their own way 53.40% of the time against 49.12% for the plain flip, and 2024 gave the same ordering at 52.81% against 48.98%. **On the 15-minute chart the gap is +0.50 points in one year and -0.37 in the other.** ## What the engulfing condition actually tests The test Formiq's backtester runs looks at the bodies only. - **Long**: the previous bar closed down, this bar closed up, this bar opened at or below the previous close and closed at or above the previous open - **Short**: the mirror of that Wicks play no part. The shape that also covers the previous high and low is a different one, the outside bar, and both are counted separately here. Write-ups usually add three requirements: cover the wicks too, make the new body much larger than the old one, and take it only after a move in the opposite direction. All three are measured below. ## Settings and how to add a threshold Nothing is drawn. The pattern is two candles that are already on the chart, so no line and no lower pane appears. ### The condition has no numbers | Condition | Numeric field | Default | |---|---|---| | Engulfing | none | fires whenever the body covers the previous body | **There is no field for how much bigger the new body has to be.** The condition either fires or it does not, about 800 times a year on the hourly chart, 3,300 on the 15-minute and 200 on the 4-hour. ### Adding a threshold with a second condition To filter by size, switch on a second condition beside it. Every filter in this article was built that way. | What to filter on | Condition to add | Setting | |---|---|---| | The engulfing bar's body is solid | Marubozu | body ratio 0.5 to 0.9 | | The engulfing bar's range is large | Large candle | ATR multiple 1.0 to 2.0, body ratio 0.5 | | Take only the side the average points | MA slope | period 20 or 50, compared five bars back | The marubozu condition is **body divided by range, at or above a threshold**, so beside the engulfing condition it keeps the signals whose body is solid rather than mostly wick. The MA slope condition passes longs while the average is rising and shorts while it is falling, which turns the pattern into a with-the-trend entry. ### The published trading rule Enter at the close of the bar that completes the pattern, in the direction it points. The usual exit is the opposite engulfing candle, and the usual stop is the far end of the engulfing bar: the low for a long, the high for a short. Both are measured here. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Timeframes | 15-minute, hourly, 4-hour | | Period | 1 January 2024 to 31 December 2025 | | Detection | the same code the backtest runs, bar by bar | | Control | pairs that flipped colour without covering the previous body | | Horizon | the close 1, 5, 10 and 20 bars after the signal bar | | Exits | the opposite engulfing candle, and a fixed ten-bar hold | | Conditions | the plain pattern plus 10 filters, 11 in total | | Runs | 11 conditions x 2 exits and one spread-free pass x 3 timeframes x 4 windows = 396 | | Spread | 0.3 pips, fixed | | Lot | 0.1 | | Fill | at the close | **The filters gate the entry only. The exit is always the plain pattern.** If a filter gated the exit as well, a stricter filter would leave positions open longer and the comparison would become a comparison of holding times. ## Against the colour flip that did not engulf The share of signals whose close ten bars later had moved the pattern's way: | Timeframe | Year | Engulfing | Hit rate | Colour flip only | Hit rate | Difference | |---|---|---|---|---|---|---| | 15-minute | 2025 | 3301 | 50.08% | 9182 | 49.58% | +0.50 | | 15-minute | 2024 | 3274 | 49.97% | 9532 | 50.34% | -0.37 | | Hourly | 2025 | 839 | 53.40% | 2327 | 49.12% | +4.28 | | Hourly | 2024 | 818 | 52.81% | 2297 | 48.98% | +3.83 | | 4-hour | 2025 | 198 | 54.04% | 601 | 50.25% | +3.79 | | 4-hour | 2024 | 215 | 47.91% | 561 | 50.62% | -2.72 | **The hourly chart is the only one where the difference is positive in both years.** The 15-minute chart stays inside half a point either way, so whether the body was covered did not separate anything. The 4-hour chart is +3.79 in 2025 and -2.72 in 2024, changing sign between the years. In pips the hourly gap is wider than the hit rate suggests. Engulfing signals moved an average of +4.28 pips (2025) and +4.59 pips (2024) over ten bars, while plain colour flips moved -1.58 and -2.39. The differences are 5.87 and 6.98 pips. ## Does covering the wicks help? Splitting the signals into bars that covered the previous range entirely (the outside bar) and bars that covered only the body: | Timeframe | Year | Outside bar | Hit rate | Body only | Hit rate | |---|---|---|---|---|---| | 15-minute | 2025 | 919 | 48.97% | 2382 | 50.50% | | 15-minute | 2024 | 868 | 50.69% | 2406 | 49.71% | | Hourly | 2025 | 229 | 51.97% | 610 | 53.93% | | Hourly | 2024 | 199 | 51.26% | 619 | 53.31% | | 4-hour | 2025 | 61 | 52.46% | 137 | 54.74% | | 4-hour | 2024 | 67 | 46.27% | 148 | 48.65% | **Body-only was the higher of the two in five of the six cells.** The exception is the 15-minute chart in 2024, by 0.98 points. On the hourly chart body-only led by 1.96 and 2.05 points. Covering the wicks requires the bar's high and low to clear the previous bar's, which admits bars that travelled a long way in both directions. The test does not identify why that reads worse. What it shows is that adding the requirement cut the signal count to about a quarter without raising the hit rate. ## Keeping only the solid bodies Filtering on the engulfing bar itself: body divided by range at 0.7 or more, so wicks take up no more than 30% of the bar. | Timeframe | Year | All engulfing | Hit rate | Body fills 70%+ | Hit rate | |---|---|---|---|---|---| | 15-minute | 2025 | 3301 | 50.08% | 1264 | 51.82% | | 15-minute | 2024 | 3274 | 49.97% | 1271 | 50.59% | | Hourly | 2025 | 839 | 53.40% | 282 | 56.38% | | Hourly | 2024 | 818 | 52.81% | 264 | 55.30% | | 4-hour | 2025 | 198 | 54.04% | 65 | 56.92% | | 4-hour | 2024 | 215 | 47.91% | 70 | 50.00% | **All six cells improved**, by between 0.62 points (15-minute, 2024) and 2.98 points (hourly, 2025). Of every filter measured here, this is the only one with no exception across three timeframes and two years. The signal count falls to roughly a third: 839 becomes 282 on the hourly chart, 198 becomes 65 on the 4-hour. ## How often is it a false signal? The usual stop sits at the far end of the engulfing bar, so that is the level to test. **For a bullish engulfing, does price break the bar's low first, or travel the same distance upward first?** No exit and no holding period enters the count. | Timeframe | Year | Group | Signals | Broke the far end first | Broke on the next bar | |---|---|---|---|---|---| | Hourly | 2025 | Engulfing | 840 | 51.73% | 21.79% | | Hourly | 2025 | Colour flip only | 2331 | 55.95% | 32.43% | | Hourly | 2025 | Body fills 70%+ | 283 | 50.35% | 10.60% | | Hourly | 2024 | Engulfing | 818 | 48.40% | 20.54% | | Hourly | 2024 | Colour flip only | 2297 | 53.77% | 32.35% | | Hourly | 2024 | Body fills 70%+ | 264 | 43.73% | 15.53% | Across all six cells the plain pattern broke first 48.40 to 53.27% of the time, the colour flip 53.67 to 57.28%, and the solid-bodied pattern 43.73 to 50.71%. **The plain pattern was above half in five of the six.** With a stop and a target the same distance apart, it is close to a coin toss. Next-bar failures run 17.21 to 21.79% for the plain pattern, 32.35 to 34.40% for the colour flip and 10.60 to 15.53% once the body has to fill 70% of the range. **Roughly one engulfing signal in five loses its far end on the very next bar**, and the median failure arrives two bars in. A false break is therefore the ordinary case rather than an accident. Filtering on the body brings it down to about one in ten, at the cost of two thirds of the signals. ## Reversal or continuation? Splitting each set by the close-to-close move over the five bars leading into the pair: against the pattern's direction is the textbook reversal, with it is a continuation. | Timeframe | Year | Reversal | Hit rate | Continuation | Hit rate | |---|---|---|---|---|---| | 15-minute | 2025 | 1701 | 50.44% | 1593 | 49.72% | | 15-minute | 2024 | 1595 | 50.60% | 1676 | 49.40% | | Hourly | 2025 | 430 | 51.16% | 408 | 55.64% | | Hourly | 2024 | 406 | 49.75% | 411 | 55.72% | | 4-hour | 2025 | 88 | 56.82% | 109 | 51.38% | | 4-hour | 2024 | 109 | 53.21% | 106 | 42.45% | **The answer flips with the timeframe.** The hourly chart favoured continuation in both years by 4.48 and 5.97 points. The 4-hour chart favoured reversal in both years by 5.44 and 10.76. The 15-minute chart favoured reversal by 0.72 and 1.20. Taking only engulfing candles that follow a move the other way is therefore a decision that cannot be made before the timeframe is. Added on the hourly chart over these two years, it keeps the weaker half. ## Trading all 396 runs Entering at the close of the completing bar and holding until the opposite engulfing candle, at 0.3 pips of spread and 0.1 lots, on the hourly chart: | Filter | 2025 trades | 2025 win rate | 2025 net pips | 2024 trades | 2024 win rate | 2024 net pips | |---|---|---|---|---|---|---| | none (plain pattern) | 439 | 43.28% | +813.8 | 426 | 42.96% | +2180.8 | | body / range ≥ 0.5 | 334 | 45.21% | +319.0 | 341 | 44.28% | +2339.3 | | body / range ≥ 0.6 | 280 | 47.14% | +761.3 | 279 | 43.73% | +2237.3 | | body / range ≥ 0.7 | 210 | 47.62% | +1182.6 | 204 | 44.12% | +2249.1 | | body / range ≥ 0.8 | 135 | 53.33% | +1478.2 | 112 | 42.86% | +1091.8 | | body / range ≥ 0.9 | 29 | 51.72% | +145.7 | 46 | 43.48% | +949.9 | | range ≥ 1.0 x ATR | 222 | 50.00% | +615.1 | 204 | 47.06% | +1924.6 | | range ≥ 1.5 x ATR | 110 | 48.18% | +44.6 | 100 | 46.00% | +1186.3 | | range ≥ 2.0 x ATR | 40 | 40.00% | -121.8 | 47 | 46.81% | +1085.1 | | 20-period MA direction | 254 | 44.49% | +643.8 | 243 | 41.56% | +1894.2 | | 50-period MA direction | 232 | 46.12% | +1226.9 | 233 | 42.92% | +1608.7 | **Ten of the eleven conditions finished both years in profit on the hourly chart.** The exception is the 2.0 x ATR filter, which took 40 trades in 2025. The other two charts are not like that. Two of eleven cleared both years on the 15-minute chart (the 2.0 x ATR filter and the 20-period MA direction), and four on the 4-hour (the plain pattern, the 0.5 and 0.6 body ratios, and the 50-period MA direction). The plain pattern lost money in both years on the 15-minute chart, at -679.7 and -725.2 pips. ### With a fixed ten-bar exit instead Closing after ten bars rather than waiting for the opposite pattern also left ten of eleven profitable in both years on the hourly chart, and the win rates line up more cleanly. | Filter | 2025 trades | 2025 win rate | 2025 net pips | 2024 trades | 2024 win rate | 2024 net pips | |---|---|---|---|---|---|---| | none (plain pattern) | 356 | 51.40% | +1000.9 | 368 | 54.08% | +2449.2 | | body / range ≥ 0.5 | 284 | 51.76% | +884.0 | 306 | 51.31% | +1248.3 | | body / range ≥ 0.6 | 251 | 54.18% | +727.7 | 256 | 50.78% | +1628.3 | | body / range ≥ 0.7 | 194 | 57.22% | +1656.6 | 179 | 55.31% | +1212.8 | | body / range ≥ 0.8 | 132 | 58.33% | +1104.6 | 104 | 57.69% | +920.6 | | body / range ≥ 0.9 | 32 | 56.25% | +155.6 | 45 | 53.33% | +556.3 | Raising the body ratio from 0.5 to 0.8 lifts the win rate from 51.76% to 58.33% in 2025 and from 51.31% to 57.69% in 2024, while the trade count falls from 284 to 132. At 0.9 the 2025 sample is 32 trades and the year returns +155.6 pips. ### Splitting 2025 in half | Filter | 2025 | Jan-Jun | Jul-Dec | |---|---|---|---| | none (plain pattern) | +813.8 | +659.3 | +116.0 | | body / range ≥ 0.5 | +319.0 | +76.4 | +204.2 | | body / range ≥ 0.6 | +761.3 | +253.5 | +469.4 | | body / range ≥ 0.7 | +1182.6 | +305.3 | +838.8 | | body / range ≥ 0.8 | +1478.2 | +618.7 | +821.1 | | body / range ≥ 0.9 | +145.7 | +243.2 | -97.5 | Five of the six were profitable in both halves, which is more than most indicators in this series manage. The 0.9 ratio is the one that broke, at -97.5 pips in the second half. ## Last year's winner, held through this year Taking whichever condition earned the most in 2024 and running it unchanged through 2025: | Timeframe | 2024 winner | 2024 net pips | 2025 net pips | 2025 rank | 2025 median of 11 | |---|---|---|---|---|---| | 15-minute | 50-period MA direction | +1680.6 | -444.3 | 9th | +63.7 | | Hourly | body / range ≥ 0.5 | +2339.3 | +319.0 | 8th | +643.8 | | 4-hour | body / range ≥ 0.8 | +1405.9 | -261.1 | 8th | -114.1 | **All three finished below the following year's median.** Ten of the eleven conditions made money in both years on the hourly chart, and the one that earned most in 2024 still landed 8th of 11 in 2025. Ranking the filters by last year's profit did not help over these two years. ## Stops, targets and cost ### The exit sets the win rate | Entry | Stop / target | 2025 trades | 2025 win rate | 2025 net pips | 2024 net pips | |---|---|---|---|---|---| | plain pattern | none | 439 | 43.28% | +813.8 | +2180.8 | | plain pattern | 30 / 60 | 614 | 35.18% | +448.2 | +1600.7 | | plain pattern | 60 / 30 | 639 | 53.68% | +515.4 | +274.1 | | plain pattern | 50 / 50 | 593 | 41.48% | +47.4 | +875.1 | | body ≥ 0.7 | none | 210 | 47.62% | +1182.6 | +2249.1 | | body ≥ 0.7 | 30 / 60 | 246 | 39.02% | +878.0 | +1194.5 | | body ≥ 0.7 | 60 / 30 | 251 | 56.18% | +513.6 | +989.8 | | body ≥ 0.7 | 50 / 50 | 244 | 45.08% | +590.2 | +1397.2 | Same entries, and the plain pattern's win rate moves from 35.18% to 53.68%. **The 60 / 30 exit that produces the highest win rate also produces the smallest 2024 result of the four, at +274.1 pips.** A "60% win rate engulfing strategy" can be manufactured from the exit alone. ### Cost | Timeframe | Filter | Trades | Spread paid | Break-even spread | |---|---|---|---|---| | 15-minute | none | 1795 | 538.5 | -0.08 | | 15-minute | body ≥ 0.7 | 944 | 283.2 | 1.58 | | Hourly | none | 439 | 131.7 | 2.15 | | Hourly | body ≥ 0.7 | 210 | 63.0 | 5.93 | | Hourly | body ≥ 0.8 | 135 | 40.5 | 11.25 | | 4-hour | none | 99 | 29.6 | 13.07 | Spread paid matched trade count times 0.3 pips in every run. The break-even spread is the width at which the 2025 result reaches zero. **The plain pattern on the 15-minute chart breaks even at -0.08 pips, so it loses at zero spread too.** The trade count is not what sinks it; the entry is. The hourly chart with a 0.8 body ratio has room out to 11.25 pips, on 135 trades a year. ## Related articles - [Candlestick pattern win rates](/blog/candlestick-patterns-tested): 17 patterns including this one, over the same two years and the same kind of control - [Do fair value gaps really fill?](/blog/fair-value-gap-settings): another price-action claim measured against ordinary bars - [Do morning and evening stars reverse price?](/blog/star-patterns-settings): a three-bar reversal, measured the same way - [Double tops and double bottoms tested](/blog/double-top-bottom-settings): reversal rates for multi-bar shapes over the same two years - [What Heikin Ashi actually changes](/blog/heikin-ashi-settings): a chart that rewrites the candles themselves ## Notes The data is USD/JPY from 1 January 2024 to 31 December 2025. Fills are at the close, the spread is fixed at 0.3 pips, and the lot is 0.1. Detection uses the same code path as the backtest and reads bars from before the test window. Because the horizons run out to 20 bars, that many bars at the end of each year are excluded from the counts. In the false-break test, a bar that touches both the far end and the target cannot be ordered from a candle, so it is counted as the failure. The body-ratio filters are built from the marubozu condition, so the engulfing test itself is unchanged. --- ### MACD's Three Readings Tested: 2,052 Settings on USD/JPY URL: https://formiq.jp/blog/macd-settings Language: en Published: 2026-09-09 Updated: 2026-09-09 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: MACD, histogram, signal line, moving average, oscillator, backtest, USD/JPY Evidence: https://formiq.jp/evidence/macd-settings Evidence JSON: https://formiq.jp/evidence/macd-settings/data.json Also available in: ja — https://formiq.jp/ja/blog/macd-settings MACD is the gap between two exponential moving averages, drawn as a line under the price chart. Fast 12, slow 26, signal 9 are so widely used that most platforms ship them as the default, and the pane shows three things: the MACD line, its signal line, and a histogram. Three trading rules go with it. The MACD line crossing its signal line, the MACD line crossing zero, and the histogram turning. Formiq's backtester lists all three as readings, so the plan was to measure each of them. **They are three different rules, and their characters are opposites.** The histogram turn won more than half its trades in all 57 settings on the 1-hour chart, and 41 of those settings still finished the year negative. The signal-line cross won between 33.54% and 40.46% and finished 2024 and 2025 in profit in all 57. The zero-line cross produced identical trades to a plain EMA crossover in all 228 runs. ## What MACD measures Three steps: 1. **MACD line** = fast EMA − slow EMA (12 and 26 by default) 2. **Signal line** = an EMA of the MACD line (9 by default) 3. **Histogram** = MACD line − signal line The first line is the whole indicator. It is a difference between two averages, so it is positive when the fast average sits above the slow one. The units are price units: on USD/JPY it says how many sen apart the two averages are. Half the results below are already decided by those three lines. ### The histogram turn enters before the crossing The histogram is the MACD line minus the signal line, so the bar where it crosses zero is the bar where the MACD line crosses its signal line. **Read as a crossing, the histogram is the signal-line cross with a different picture in front of you.** The turn is somewhere else. It is the bar where a falling histogram stops falling while it is still below zero: three bars, a trough for a buy and a peak for a sell. It completes before zero is reached, so it anticipates the crossing. The three readings on the 1-hour chart in 2025, all at 12/26/9: | Reading | Trades | Win rate | Annual pips | Average win | Average loss | Bars held | |---|---|---|---|---|---|---| | Signal-line cross | 451 | 38.58% | +2,223.6 | 68.00 pips | −34.69 pips | 19.4 | | Histogram turn | 373 | 56.84% | +761.1 | 46.71 pips | −56.78 pips | 23.4 | | Zero-line cross | 200 | 35.00% | +689.5 | 95.08 pips | −45.89 pips | 43.6 | **The reading with the highest win rate has the smallest average win.** The turn collects 46.71 pips and gives back 56.78, the only one of the three whose losses are the larger side. Entering early means the opposite turn arrives and closes the position before the move has extended. ### The zero-line cross is an EMA crossover The MACD line is the fast EMA minus the slow EMA. So **the MACD line crossing above zero is the fast EMA crossing above the slow EMA** — an ordinary golden cross, with the subtraction done for you. The same 19 period pairs were run twice: once as a MACD zero-line cross, once through the moving-average rule with both averages set to EMA. | Rules compared | Runs matching on trades and annual pips | |---|---| | Zero-line cross vs EMA crossover | 228 of 228 | No difference there either. A strategy built on the MACD zero line is a moving-average crossover strategy that carries an extra indicator. This is the fifth time in this series that an indicator's rule turned out to be another indicator: the CCI zero cross, the psychological line, Heikin-Ashi candles and moving-average disparity came first. ### Three readings, one of them another indicator Two conditions here belong to the MACD: the **signal-line cross** and the **histogram turn**. The zero-line numbers are reported alongside them, but they are also the numbers for an EMA crossover. ## MACD settings and how to read them ### Where it appears Not on the candles. A separate pane opens below the price chart with three things drawn in it: - **MACD line** (blue by default) - **Signal line** (orange by default) - **Histogram**, coloured differently above and below zero (green up, red down by default) Zero is the reference level. MT4's built-in MACD is built differently. It computes the signal line with a simple moving average and draws the MACD line itself as the histogram, so it does not look like this one. The three steps above are what this article and TradingView's MACD use. ### The settings dialog | Field | Default | What it sets | |---|---|---| | Fast | 12 | Period of the faster EMA in the MACD line | | Slow | 26 | Period of the slower EMA in the MACD line | | Signal | 9 | Period that smooths the MACD line into the signal line | The backtester adds a **reading**: signal-line cross, zero-line cross, or histogram. ### The signal period is inert under one of the readings A rule that never looks at the signal line cannot be changed by the signal period. Here are the 57 settings on the 1-hour chart in 2025, grouped by signal period, as medians. | Signal period | Signal-line cross | Histogram turn | Zero-line cross | |---|---|---|---| | 5 | +2,055.4 pips (614 trades avg) | −1,494.1 pips (438 trades avg) | +625.8 pips (211 trades avg) | | 9 | +2,054.3 pips (480 trades avg) | −152.2 pips (376 trades avg) | +625.8 pips (211 trades avg) | | 14 | +1,654.3 pips (408 trades avg) | −111.7 pips (332 trades avg) | +625.8 pips (211 trades avg) | The right-hand column is the same value three times. On the 15-minute chart all three rows give +1,531.0 pips over 861 trades, and on the 4-hour chart all three give −1,290.0 pips over 66 trades. **Choosing the zero-line reading retires one of the three numbers the dialog lets you set.** The signal period does reach the two readings that use the signal line, and what it changes is frequency: the crossing takes 614 trades at a signal of 5 against 408 at 14. ### The published rules and the conditions tested here | Common name | Condition measured here | |---|---| | MACD crossover | MACD line crosses above the signal line to buy, below to sell | | Zero-line break | MACD line crosses above zero to buy, below to sell | | Histogram turn | Histogram turns up while still below zero to buy, turns down while still above zero to sell (judged over three bars) | Positions are held until the opposite signal. Stops and targets are measured separately further down. ## How this was measured | Item | Value | |---|---| | Instrument | USD/JPY | | Period | 1 Jan to 31 Dec 2024, and 1 Jan to 31 Dec 2025 | | Timeframes | 15-minute, 1-hour, 4-hour | | Fast | 5, 8, 12, 16, 24 | | Slow | 17, 26, 35, 52 | | Signal | 5, 9, 14 | | Combinations | 19 pairs with fast < slow, times 3 signal periods = 57 settings | | Readings | Signal-line cross, zero-line cross, histogram | | Fills | At the close | | Spread | 0.3 pips, fixed | | Lot | 0.1 | | Exit | On the opposite signal | 2,052 runs in total: 57 settings x 3 readings x 3 timeframes x 4 period windows (2024, 2025, and the two halves of 2025). ## The timeframe reverses the answer For each reading and timeframe, here is how many of the 57 settings finished both 2024 and 2025 in profit. | Timeframe | Reading | Profitable in both years | 2025 median | 2024 median | Trades per year | Mean win rate | |---|---|---|---|---|---|---| | 15-minute | Signal-line cross | 10 / 57 | +365.3 pips | −1,359.9 pips | 2,125 | 33.04% | | 15-minute | Histogram turn | 0 / 57 | −2,535.6 pips | −795.0 pips | 1,549 | 57.62% | | 15-minute | Zero-line cross | 54 / 57 | +1,531.0 pips | +1,659.5 pips | 861 | 29.57% | | 1-hour | Signal-line cross | 57 / 57 | +1,972.5 pips | +2,127.4 pips | 501 | 36.95% | | 1-hour | Histogram turn | 1 / 57 | −433.7 pips | −1,143.0 pips | 382 | 57.12% | | 1-hour | Zero-line cross | 48 / 57 | +625.8 pips | +1,103.2 pips | 211 | 33.96% | | 4-hour | Signal-line cross | 27 / 57 | +762.2 pips | +137.3 pips | 128 | 44.12% | | 4-hour | Histogram turn | 19 / 57 | −401.6 pips | +1,301.0 pips | 100 | 50.90% | | 4-hour | Zero-line cross | 0 / 57 | −1,290.0 pips | +2,398.0 pips | 66 | 26.95% | **The signal-line cross put all 57 settings in profit across both years on the 1-hour chart, and only 10 of 57 on the 15-minute.** The 15-minute median for 2024 is negative, so more than half of those settings spent the year losing money. **The histogram turn has the highest mean win rate of the three readings on every chart, and reached both years in profit in 0 of 57 settings on the 15-minute and 1 of 57 on the 1-hour.** Only the 4-hour recovers to 19 of 57, and its 2025 median is still −401.6 pips. The zero-line cross runs the other way: 54 of 57 on the 15-minute chart, none at all on the 4-hour. Its 4-hour numbers are the largest of any cell in 2024 at +2,398.0 pips and negative in 2025 at −1,290.0 pips. Trade counts explain part of that. The 4-hour zero-line cross takes 66 trades a year. USD/JPY rose 1,632.5 pips over 2024, so a small number of trades could catch a lot of it. 2025 opened at 157.23 and closed at 156.67, which is −55.7 pips for anyone holding. **A rule that trades 66 times a year mostly inherits the direction of the year.** The 1-hour signal-line cross takes 501 trades on average and still returned a median +1,972.5 pips in the year that finished where it started. ## 12/26/9 is not a special setting Here is the default measured against the other 56, with its rank in each cell. | Timeframe | Reading | 2025 | 2024 | |---|---|---|---| | 15-minute | Signal-line cross | 1,938 trades, 33.69%, +1,093.6 pips (7th) | −1,359.9 pips (29th) | | 15-minute | Histogram turn | 1,493 trades, 57.74%, −2,695.2 pips (32nd) | −538.9 pips (24th) | | 15-minute | Zero-line cross | 849 trades, 29.33%, +1,212.0 pips (41st) | +2,225.3 pips (11th) | | 1-hour | Signal-line cross | 451 trades, 38.58%, +2,223.6 pips (16th) | +3,152.1 pips (2nd) | | 1-hour | Histogram turn | 373 trades, 56.84%, +761.1 pips (2nd) | −1,966.7 pips (49th) | | 1-hour | Zero-line cross | 200 trades, 35.00%, +689.5 pips (23rd) | +623.6 pips (50th) | | 4-hour | Signal-line cross | 107 trades, 46.73%, +1,299.1 pips (11th) | +88.1 pips (30th) | | 4-hour | Histogram turn | 88 trades, 48.86%, −109.8 pips (24th) | +1,630.9 pips (18th) | | 4-hour | Zero-line cross | 71 trades, 19.72%, −2,089.3 pips (53rd) | +2,690.3 pips (17th) | Every rank is out of 57. **The same 12/26/9 placed 2nd in 2024 on the 1-hour signal-line cross and 50th in 2024 on the 1-hour zero-line cross.** On the 1-hour histogram turn it went from 2nd in 2025 at +761.1 pips to 49th in 2024 at −1,966.7. On the 4-hour zero-line cross it placed 53rd in 2025 and lost 2,089.3 pips. On the 15-minute signal-line cross its 2024 rank of 29th is the median of the 57, which is another way of saying the default was neither good nor bad there. **Fix the timeframe and the reading before arguing about the periods.** Copying someone else's 12/26/9 does not reproduce their trades. ## A faster fast period returned more The 57 settings on the 1-hour signal-line cross, grouped by fast period, as 2025 medians. Every slow and signal period is present in each group. | Fast | 2025 median | Trades per year | Profitable in both years | |---|---|---|---| | 5 | +2,243.0 pips | 674 | 12 / 12 | | 8 | +2,044.4 pips | 549 | 12 / 12 | | 12 | +2,033.4 pips | 469 | 12 / 12 | | 16 | +1,886.5 pips | 424 | 12 / 12 | | 24 | +1,492.3 pips | 351 | 9 / 9 | The order is clean, and no group fails: every fast period put all of its settings in profit across both years. The slow period does not order itself the same way. | Slow | 2025 median | Trades per year | |---|---|---| | 17 | +2,224.6 pips | 582 | | 26 | +2,011.4 pips | 503 | | 35 | +1,755.8 pips | 482 | | 52 | +1,889.3 pips | 452 | It falls to 35 and recovers at 52 while trade counts keep falling, so frequency alone does not account for it. This aggregation cannot identify the reason. ## Taking each year's best setting into the other year For each timeframe, the setting with the highest annual pips in one year, run unchanged in the other. Signal-line cross. | Timeframe | Year and setting | Its own year | The other year | |---|---|---|---| | 15-minute | 2024, 12/52/14 | +1,273.1 pips (1st) | +561.8 pips (18th) | | 15-minute | 2025, 16/52/5 | +1,404.2 pips (1st) | −1,025.3 pips (28th) | | 1-hour | 2024, 8/17/9 | +3,261.2 pips (1st) | +2,526.5 pips (9th) | | 1-hour | 2025, 8/17/5 | +3,376.8 pips (1st) | +1,450.5 pips (51st) | | 4-hour | 2024, 5/17/5 | +1,515.2 pips (1st) | +308.5 pips (45th) | | 4-hour | 2025, 16/17/9 | +1,862.9 pips (1st) | +259.7 pips (25th) | Ranks are out of 57. Five of the six fell from 1st to somewhere between 9th and 51st. **The 1-hour winner of 2025, 8/17/5, placed 51st in 2024**, and still made 1,450.5 pips there. Because all 57 settings on the 1-hour signal-line cross were profitable in both years, **drawing the worst one still leaves you in profit**. Losing the ranking and losing money are two different failures, and only the first happened in that cell. The 15-minute chart is not like that. Its 2025 winner, 16/52/5, lost 1,025.3 pips in 2024, so a bad pick there changes the sign. ## The signal-line cross holds every win rate near 37% The 57 settings on the 1-hour chart in 2025 have win rates between 33.54% and 40.46%. **All 57 finished in profit while winning fewer than half their trades.** The best, middle and worst by annual pips: | Setting | Trades | Win rate | Annual pips | Average win | Average loss | Per trade | Max drawdown | |---|---|---|---|---|---|---|---| | 8/17/5 | 723 | 37.07% | +3,376.8 | 56.16 pips | −25.66 pips | 4.67 pips | 1,284.5 pips | | 8/52/14 | 407 | 37.10% | +1,972.5 | 70.38 pips | −33.81 pips | 4.84 pips | 829.2 pips | | 24/26/14 | 316 | 37.34% | +627.8 | 77.17 pips | −42.82 pips | 1.98 pips | 1,440.5 pips | "Per trade" is the win rate applied to the average win and loss. Multiply it by the trade count and you land on the year: 723 trades at 4.67 pips is about 3,376 pips against a measured +3,376.8. **All three win rates round to 37%, and their order runs backwards against the pips.** The worst performer, 24/26/14, has both the highest win rate and the largest average win at 77.17 pips. Two separate things divide them. **The top two earn almost the same per trade, 4.67 and 4.84 pips, so their gap in annual pips is a gap in frequency: 723 trades against 407.** The bottom one earns 1.98 pips per trade, less than half of either. Its ratio of average win to average loss is 1.80 against 2.08 and 2.19, falling in the same order as the annual pips. Drawdown does not follow frequency: the 316-trade setting has the deepest at 1,440.5 pips and the 407-trade setting the shallowest at 829.2. ## Filters, stops and cost All measured on the 1-hour chart with 12/26/9 on the signal-line cross. Unfiltered, that setting took 451 trades for +2,223.6 pips in 2025 and 423 trades for +3,152.1 pips in 2024. ### The ADX filter fails for the fifth time Only trading while ADX sits above a level, at four levels: | Condition | 2025 | 2024 | |---|---|---| | No filter | 451 trades, +2,223.6 pips | 423 trades, +3,152.1 pips | | ADX above 15 | 28 trades, +2,435.1 pips | 30 trades, −2,061.5 pips | | ADX above 20 | 50 trades, +288.4 pips | 58 trades, +950.2 pips | | ADX above 25 | 57 trades, +1,053.9 pips | 53 trades, +984.1 pips | | ADX above 30 | 59 trades, +790.9 pips | 50 trades, +982.0 pips | Every level cuts 451 trades down to between 28 and 59. ADX above 15 beats the unfiltered rule in 2025 at +2,435.1 pips and loses 2,061.5 pips in 2024. **No level improved both years.** This series has now tried an ADX filter five times with the same outcome each time. Trade counts rise with the level rather than falling: 28, 50, 57, 59 in 2025. The same condition gates both entries and exits, so a stricter filter changes where positions are closed as well as opened, but this aggregation cannot identify the reason. ### Stops and targets | Setting | 2025 | 2024 | |---|---|---| | None (hold to the opposite signal) | 451 trades, 38.58%, +2,223.6 | +3,152.1 | | Stop 20, target 40 | 404 trades, 33.91%, +455.1 | +1,075.7 | | Stop 30, target 60 | 397 trades, 35.52%, +1,232.9 | +366.8 | | Stop 50, target 100 | 396 trades, 37.63%, +1,906.2 | +1,975.9 | | Stop 30, no target | 394 trades, 31.73%, +2,101.9 | +1,426.4 | | No stop, target 60 | 451 trades, 43.46%, +1,570.3 | +2,326.1 | Figures are pips. **A 60-pip target alone lifts the win rate from 38.58% to 43.46% and drops the year from +2,223.6 to +1,570.3 pips.** On a rule whose average win is 56 to 77 pips, capping wins at 60 removes the part that pays for the losses. Only stop 50 with target 100 stayed near the unfiltered result in both years, and even that gave up over 300 pips in 2025. ### Cost The same trades at different spreads, 2025, 451 trades: | Spread | Annual pips | |---|---| | 0.0 pips | +2,358.9 | | 0.3 pips | +2,223.6 | | 0.6 pips | +2,088.3 | | 1.0 pips | +1,907.9 | | 1.5 pips | +1,682.4 | | 2.0 pips | +1,456.9 | **Each extra pip of spread costs 451 pips**, which is the trade count. As with every other indicator in this series, the spread bill is trades times spread. Dividing the zero-spread result of +2,358.9 pips by 451 trades gives 5.2 pips per trade, so the setting turns negative once the spread passes 5.2 pips. The more a setting trades, the less room it has. The same 12/26/9 on the 15-minute signal-line cross took 1,938 trades in 2025, and the same calculation puts its break-even spread at 0.86 pips. **Dropping one timeframe cuts the spread it can absorb by a factor of six.** ## Related reading - [Moving Average Crossover Settings Compared](/blog/moving-average-cross-settings): the same trades as this article's zero-line cross, measured from the moving-average side - [SMA vs EMA: Which Moving Average Is More Profitable? 672 Tests](/blog/sma-vs-ema): the exponential average MACD is built from, against the other ways of averaging - [OsMA Settings Compared](/blog/osma-settings): the gap between the MACD line and its signal line, promoted to an indicator of its own - [CCI Settings Compared](/blog/cci-settings): the first time a zero-line cross turned out to be a moving-average cross - [Moving Average Disparity Settings Compared](/blog/ma-disparity-settings): the fourth - [Do Indicator Settings Transfer Between Pairs? 3,024 Tests](/blog/settings-across-pairs): this setting won the hourly chart on USD/JPY in 2024 with +3,152.1 pips and lost money on all three other pairs ## Notes USD/JPY only, over 2024 and 2025. Fills are at the close with a fixed 0.3 pip spread and no slippage, so live fills and commissions will differ. USD/JPY opened 2024 at 140.87 and closed at 157.20, which is +1,632.5 pips for anyone simply holding, so trend-following rules had a favourable stretch in the sample. 2025 opened at 157.23 and closed at 156.67 for −55.7 pips, with a high-to-low range of 1,899.5 pips. The two identity checks compared trade counts, win rates and annual pips between rules run over the same period on the same fill terms. They did not match trades one by one. --- ### Morning Star and Evening Star: Do They Reverse? 240 Backtests URL: https://formiq.jp/blog/star-patterns-settings Language: en Published: 2026-09-09 Updated: 2026-09-09 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Morning star, Evening star, Candlestick patterns, Price action, Backtest, USDJPY Evidence: https://formiq.jp/evidence/star-patterns-settings Evidence JSON: https://formiq.jp/evidence/star-patterns-settings/data.json Also available in: ja — https://formiq.jp/ja/blog/star-patterns-settings A morning star is three candles. The first falls, the second has a small body, the third rises and closes past the midpoint of the first body. The evening star inverts every part of that. In the Sakata tradition both belong to the *santen* or three-river family. The name comes from the middle candle floating free of the two around it, the way a single star sits in the dawn sky. That picture requires a **gap** on each side of it. So that was counted first. **Across three timeframes and two years of USD/JPY, 4,558 signals fired and 0 of them had a gap on both sides.** Measured against the colour sequence it is built on, a down bar then anything then an up bar, the pattern came out ahead in **one cell of six**. ## What the morning star and evening star actually test For a morning star, Formiq's backtester requires three things at once: 1. **The colour frame**: the first bar falls and the third rises. The middle bar's colour is not checked. 2. **A small middle**: the middle body is under half the first body and under half the third body. 3. **The recovery**: the third close passes the midpoint of the first body. The evening star inverts all of it. **No gap is tested.** The condition reads open and close positions only; whether the middle bar is separated from the first is never asked. This article counts the gaps, then removes and moves the other clauses one at a time. ## Is there really a gap? Two definitions were counted. A **body gap** puts the middle body entirely below the first bar's close (for a morning star) and the third body entirely above the middle body. A **full gap** requires the same of the wicks, which is what leaves visible white space on a chart. | Timeframe | Year | Signals | Body gap | Share | Full gap | |---|---|---|---|---|---| | 15-minute | 2025 | 1714 | 36 | 2.10% | 0 | | 15-minute | 2024 | 1789 | 26 | 1.45% | 0 | | Hourly | 2025 | 418 | 5 | 1.20% | 0 | | Hourly | 2024 | 432 | 6 | 1.39% | 0 | | 4-hour | 2025 | 106 | 1 | 0.94% | 0 | | 4-hour | 2024 | 99 | 1 | 1.01% | 0 | **The full gap is zero in all six cells**, 0 out of 4,558. The reason is the trading day. Currency does not close overnight and reopen in the morning; it runs continuously from Monday to Friday, so each bar opens at almost exactly the previous close. | Timeframe | Year | Bars | Open differs from previous close | Differs by 1 pip or more | Largest | |---|---|---|---|---|---| | 15-minute | 2025 | 24903 | 93.04% | 0.97% | 204.05 pips | | 15-minute | 2024 | 24999 | 94.21% | 0.85% | 94.80 pips | | Hourly | 2025 | 6226 | 93.74% | 2.28% | 204.05 pips | | Hourly | 2024 | 6250 | 93.81% | 2.05% | 94.80 pips | | 4-hour | 2025 | 1610 | 92.98% | 3.42% | 204.05 pips | | 4-hour | 2024 | 1616 | 94.25% | 3.28% | 94.80 pips | Some difference exists on 93 to 94% of bars, but **a full pip of separation appears on 0.85 to 3.42%**. The rest is under a pip. The largest break on all three charts is the same bar in each year: 204.05 pips on 5 October 2025 and 94.80 pips on 27 October 2024, both at the Sunday reopen. Waiting for a star that is genuinely separated therefore leaves a handful of candidates a year, all at the weekly reopen. **The rest of this article measures the gapless shape, which is the one that actually gets traded.** ## Settings and how to add a threshold Nothing is drawn. Three candles that are already on the chart make the pattern, so no line and no lower pane appears. ### Neither condition has numbers | Condition | Numeric field | What it fires on | |---|---|---| | Morning star | none | down, small, up, with the third close past the midpoint of the first body | | Evening star | none | up, small, down, with the third close below that midpoint | **Both thresholds are fixed in the code**: the middle body at half, the recovery at the midpoint. This article moves them from outside instead. | What to filter on | Condition to add | Setting | |---|---|---| | The third body is solid | Marubozu | body ratio 0.7 | | The third bar's range is large | Large candle | ATR multiple 1.5, body ratio 0.5 | | Take only the side the average points | MA slope | period 50, compared five bars back | ### The published trading rule Enter at the close of the third bar, long on a morning star and short on an evening star. **Each one only produces a single side**, so a rule using just the morning star never sells. Switching both on lets the morning star close shorts and the evening star close longs. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Timeframes | 15-minute, hourly, 4-hour | | Period | 1 January 2024 to 31 December 2025 | | Detection | the same code the backtest runs, bar by bar | | Control | every instance of the colour sequence, whatever else it does | | Horizon | the close 1, 5, 10 and 20 bars after the signal bar | | Exits | the opposite star, and a fixed ten-bar hold | | Conditions | both together, each alone, and each with three filters: 10 | | Runs | 10 conditions x 3 timeframes x 4 windows x with and without cost = 240 | | Spread | 0.3 pips, fixed | | Lot | 0.1 | | Fill | at the close | ## Against the colour sequence alone The pattern adds two clauses on top of "a down bar, then anything, then an up bar". Here is what that base does by itself. | Timeframe | Year | Colour sequence | Hit rate | Star pattern | Hit rate | Difference | |---|---|---|---|---|---|---| | 15-minute | 2025 | 12337 | 49.74% | 1714 | 50.35% | +0.61 | | 15-minute | 2024 | 12606 | 49.82% | 1789 | 49.02% | -0.80 | | Hourly | 2025 | 3142 | 50.22% | 418 | 50.00% | -0.22 | | Hourly | 2024 | 3166 | 48.70% | 432 | 45.60% | -3.10 | | 4-hour | 2025 | 802 | 50.87% | 106 | 45.28% | -5.59 | | 4-hour | 2024 | 782 | 49.49% | 99 | 38.38% | -11.10 | **Only the 15-minute chart in 2025 came out ahead, by 0.61 points.** The other five are negative, and the 4-hour chart in 2024 is 11.10 points down. A colour sequence landing near 50% is unremarkable: a down bar followed two bars later by an up bar happens more than 3,000 times a year on the hourly chart. What the table shows is that adding two clauses, which cuts the count to about a seventh, produced a lower number than the count it started from. ## How far does the third bar have to recover? The shipped rule puts the third close past the midpoint of the first body. Cutting that threshold at six levels, with the small-middle requirement left in place: | Recovery required | 2025 signals | 2025 hit rate | 2024 signals | 2024 hit rate | |---|---|---|---|---| | more than 0% | 570 | 50.53% | 578 | 45.50% | | more than 25% | 522 | 51.53% | 533 | 45.97% | | more than 50% (shipped) | 418 | 50.00% | 432 | 45.60% | | more than 61.8% | 365 | 50.14% | 387 | 45.99% | | more than 75% | 325 | 50.46% | 348 | 45.40% | | more than 100% | 267 | 50.19% | 280 | 46.07% | Hourly chart. **Within a year the whole ladder spans 1.53 points in 2025 and 0.67 in 2024.** The same threshold moves 4.12 to 5.56 points when the year changes. The shipped 50% level reads 50.00% and 45.60%. Signals fall from 570 to 267 across the ladder. Cutting more of them did not raise the rate. ## Does the middle bar's size or colour matter? Removing the other clause, "the middle body is under half of each outer body": | Timeframe | Year | Colour sequence | Small middle only | Recovery only | Both (shipped) | |---|---|---|---|---|---| | 15-minute | 2025 | 49.74% (12337) | 50.78% (2233) | 49.87% (7670) | 50.35% (1714) | | 15-minute | 2024 | 49.82% (12606) | 49.27% (2330) | 49.55% (7721) | 49.02% (1789) | | Hourly | 2025 | 50.22% (3142) | 50.70% (574) | 50.47% (1922) | 50.00% (418) | | Hourly | 2024 | 48.70% (3166) | 45.52% (580) | 48.96% (1924) | 45.60% (432) | | 4-hour | 2025 | 50.87% (802) | 48.65% (148) | 48.99% (494) | 45.28% (106) | | 4-hour | 2024 | 49.49% (782) | 41.73% (139) | 47.63% (464) | 38.38% (99) | **The recovery-only column tracks the colour sequence almost exactly.** The six differences run from -1.88 to +0.26 points, and on the hourly chart they are 0.25 and 0.26. That clause removes about 40% of the signals and leaves the rate where it was. Keeping only the small-middle clause is slightly above the sequence on the 15-minute and hourly charts in 2025 and below it everywhere else. Combining the two, which is the shipped condition, is not better than either alone. The middle bar's colour was counted too. It ran opposite to the third bar in 41.41 to 46.06% of signals. | Timeframe | Year | Middle opposite | Hit rate | Middle same | Hit rate | |---|---|---|---|---|---| | 15-minute | 2025 | 753 | 52.06% | 961 | 49.01% | | 15-minute | 2024 | 782 | 49.36% | 1007 | 48.76% | | Hourly | 2025 | 183 | 50.82% | 235 | 49.36% | | Hourly | 2024 | 199 | 49.25% | 233 | 42.49% | | 4-hour | 2025 | 46 | 36.96% | 60 | 51.67% | | 4-hour | 2024 | 41 | 34.15% | 58 | 41.38% | The opposite-coloured middle leads on the 15-minute and hourly charts in all four cells and trails on the 4-hour in both. **Four of six, but not consistent across timeframes.** ## Trading all 240 runs Closing ten bars after entry. A single star produces one side only, so there is no opposite signal to close against. | Condition | 2025 trades | 2025 win rate | 2025 net pips | 2024 trades | 2024 win rate | 2024 net pips | |---|---|---|---|---|---|---| | both (exit on the opposite star) | 233 | 44.64% | -386.7 | 242 | 43.39% | -2240.9 | | both (ten-bar exit) | 249 | 50.60% | +559.7 | 267 | 46.44% | -1529.2 | | morning star alone | 164 | 48.17% | -68.5 | 175 | 54.29% | -281.3 | | morning star + body ≥ 0.7 | 71 | 54.93% | -93.8 | 80 | 58.75% | +510.2 | | morning star + range ≥ 1.5 ATR | 33 | 60.61% | +157.7 | 31 | 67.74% | +213.7 | | morning star + 50-MA direction | 82 | 42.68% | +107.2 | 108 | 57.41% | +100.0 | | evening star alone | 161 | 51.55% | +817.6 | 169 | 38.46% | -1619.7 | | evening star + body ≥ 0.7 | 72 | 45.83% | +184.3 | 68 | 41.18% | -1081.7 | | evening star + range ≥ 1.5 ATR | 33 | 36.36% | -227.9 | 26 | 34.62% | -443.8 | | evening star + 50-MA direction | 87 | 50.57% | +149.8 | 69 | 42.03% | -1097.0 | Hourly chart. **Two of the ten conditions cleared both years**, both of them the morning star with a filter, and the ATR version took 33 and 31 trades. The 15-minute chart cleared three (the morning star alone, with the body filter, and with the 50-MA direction) and the 4-hour cleared two (the morning star and the evening star, each with the body filter). **No condition cleared both years on all three charts.** ## Subtracting the market's own drift The morning star only buys and the evening star only sells. Buying repeatedly in a year that rose pays regardless of the pattern, so what an unconditional ten-bar hold would have returned is subtracted from the per-trade result. | Timeframe | Ten-bar drift, 2025 | Ten-bar drift, 2024 | |---|---|---| | 15-minute | -0.03 pips | +0.66 pips | | Hourly | -0.07 pips | +2.53 pips | | 4-hour | -0.48 pips | +9.49 pips | USD/JPY rose through 2024, so on the 4-hour chart holding any ten bars returned 9.49 pips on average. The evening star's 2024 result carries that headwind in full. | Condition | 2025 per trade | after the drift | 2024 per trade | after the drift | |---|---|---|---|---| | morning star alone | -0.42 | -0.35 | -1.61 | -4.14 | | morning star + range ≥ 1.5 ATR | +4.78 | +4.85 | +6.89 | +4.36 | | morning star + 50-MA direction | +1.31 | +1.38 | +0.93 | -1.61 | | evening star alone | +5.08 | +5.01 | -9.58 | -7.05 | | evening star + body ≥ 0.7 | +2.56 | +2.49 | -15.91 | -13.37 | The hourly evening star lost 9.58 pips a trade in 2024. Removing the 2.53 pips of drift still leaves **-7.05 pips**. **After the drift, four of the 30 timeframe-condition pairs kept a positive per-trade result in both years**: the morning star with the body filter on the 15-minute and 4-hour charts, the morning star with the ATR filter on the hourly, and the evening star with the body filter on the 4-hour. All four take between 16 and 80 trades a year. ## Last year's winner, held through this year | Timeframe | 2024 winner | 2024 net pips | 2025 net pips | 2025 rank | 2025 median of 10 | |---|---|---|---|---|---| | 15-minute | morning star + body ≥ 0.7 | +418.3 | +23.5 | 7th | +135.6 | | Hourly | morning star + body ≥ 0.7 | +510.2 | -93.8 | 8th | +128.5 | | 4-hour | evening star + body ≥ 0.7 | +772.7 | +612.6 | 1st | -65.6 | The 15-minute and hourly picks fell below the following year's median. The 4-hour pick stayed first, on 17 trades in 2025 and 15 in 2024. ## Stops, targets and cost | Entry | Stop / target | 2025 trades | 2025 win rate | 2025 net pips | 2024 net pips | |---|---|---|---|---|---| | morning star | none | 164 | 48.17% | -68.5 | -281.3 | | morning star | 30 / 60 | 187 | 34.22% | -510.1 | +153.6 | | morning star | 60 / 30 | 176 | 56.82% | -427.1 | +372.0 | | morning star | 50 / 50 | 178 | 49.44% | -66.7 | -51.6 | | evening star | none | 161 | 51.55% | +817.6 | -1619.7 | | evening star | 30 / 60 | 186 | 39.78% | +426.0 | -913.7 | | evening star | 60 / 30 | 180 | 61.67% | +160.8 | -1038.4 | | evening star | 50 / 50 | 177 | 48.59% | +121.4 | -1262.2 | **A 60 / 30 exit lifts the evening star to a 61.67% win rate and still loses 1,038.4 pips in 2024.** The entries are identical; the ratio of stop to target sets the win rate. | Timeframe | Condition | Trades | Spread paid | Trades x 0.3 | |---|---|---|---|---| | 15-minute | both (ten-bar exit) | 1043 | 312.8 | 312.9 | | 15-minute | morning star alone | 664 | 199.2 | 199.2 | | Hourly | both (ten-bar exit) | 249 | 74.7 | 74.7 | | Hourly | morning star alone | 164 | 49.2 | 49.2 | | 4-hour | morning star alone | 38 | 11.4 | 11.4 | Spread paid matched trade count times 0.3 pips everywhere. The hourly morning star still returns -19.3 pips at zero spread, so the loss comes from the entry rather than from paying for it too often. ## Related articles - [Candlestick pattern win rates](/blog/candlestick-patterns-tested): these two among 17, over the same two years and the same kind of control - [Is the engulfing candle profitable?](/blog/engulfing-settings): a two-bar reversal measured against the colour flips that did not engulf - [Double tops and double bottoms tested](/blog/double-top-bottom-settings): a larger reversal shape over the same two years - [Do fair value gaps really fill?](/blog/fair-value-gap-settings): the gap counted here, promoted into a trading condition - [Do Indicator Settings Transfer Between Pairs? 3,024 Tests](/blog/settings-across-pairs): this setting won the 4-hour chart on USD/JPY in 2023 and is one of the few that also made money on EUR/USD, GBP/USD and EUR/JPY ## Notes The data is USD/JPY from 1 January 2024 to 31 December 2025. Fills are at the close, the spread is fixed at 0.3 pips, and the lot is 0.1. Detection uses the same code path as the backtest and reads bars from before the test window. Because the horizons run out to 20 bars, that many bars at the end of each year are excluded. The body gap is measured on opens and closes; the full gap on highs and lows. The drift subtraction averages the ten-bar close-to-close move over every bar of the window, then removes it from long-only results and adds it to short-only ones. --- ### Best Stochastic Settings: 2,700 Tests of K/D and Smoothing URL: https://formiq.jp/blog/stochastic-settings Language: en Published: 2026-09-09 Updated: 2026-09-09 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Stochastic, Settings, Win rate, Oscillators, Backtesting, USDJPY Also available in: ja — https://formiq.jp/ja/blog/stochastic-settings The stochastic oscillator measures where the close sits within a recent high/low range. Its K and D lines appear below the price chart. Traders can read the 20/80 levels or trade crosses between the lines, but those rules take different positions and need separate tests. Searching for **the best stochastic settings for a 1-hour chart** also requires separating the K lookback from K smoothing. A longer lookback changes the range being measured; more smoothing averages the individual K readings. The D period controls another average of K. Across 2,700 USD/JPY runs, **43 of 75 hourly crossover settings were profitable in both years**. All 15 K/D combinations with smoothing set to 20 were profitable in both years. Short fades produced no settings profitable in both years on any tested timeframe. ## How the stochastic oscillator calculates K and D Raw K subtracts the lookback's lowest low from the current close, divides by the lookback's highest high minus lowest low, and multiplies by 100. A close near the top of that range produces a reading near 100; a close near the bottom produces a reading near 0. The definition tested here takes a simple average of the individual raw K readings over the smoothing period. That average is the displayed K line. D is a further simple average of the displayed K values. If the lookback has identical highs and lows, the raw reading is set to 50. A reading above 80 describes the close's position. It does not establish that price has started falling. An advance can keep closing near its recent highs, leaving K elevated while a short position loses money. ## Stochastic settings: K period, D period and smoothing Adding Stochastic from Formiq's indicator list draws the lines in a separate pane below the candles. The calculation and level fields are: | Field | Default | Meaning | | --- | --- | --- | | K period | 14 | Bars used to find the highest high and lowest low | | D period | 3 | Number of smoothed K values averaged for D | | Slowing | 3 | Number of raw K values averaged; called K smoothing in this article | | Upper level | 80 | Upper guide drawn in the pane | | Lower level | 20 | Lower guide drawn in the pane | K is yellow by default and D is blue. Line colors and width can be changed without changing the calculation. The pane includes the 20 and 80 guides. Chart guides and backtest trading levels are configured separately. **14/3/3 means K lookback / D period / K smoothing**, in that order. Smoothing of 1 leaves raw K unchanged. A longer smoothing period averages over more readings. Increasing D smooths the D line without changing K. The tested entry and exit rules were: | Trading rule | Entry | Exit | | --- | --- | --- | | Long fade | Buy while K is below 20 | Close when K exceeds 80 | | Short fade | Sell while K exceeds 80 | Close when K falls below 20 | | K/D cross | Buy when K crosses above D; sell when K crosses below D | Close on the opposite cross | The fades enter while K is outside a level; they do not wait for K to return inside the band. Crosses require an actual intersection and have no 20/80 restriction. Long and short fades were run independently, not combined into one portfolio. ## How this was measured All trading results were measured with Formiq's backtesting engine. | Item | Setting | | --- | --- | | Pair | USD/JPY | | Main period | 2025-01-01 to 2025-12-31 | | Separate-year check | 2024-01-01 to 2024-12-31 | | Within-year checks | 2025-01-01 to 06-30 and 2025-07-01 to 12-31, each run independently | | Timeframes | 15-minute, 1-hour, 4-hour | | K lookbacks | 5, 9, 14, 21, 50 | | K smoothing periods | 1, 3, 5, 10, 20 | | Crossover D periods | 2, 3, 5 | | Fade level pairs | 10/90, 20/80, 30/70; D held at 3 | | Run count | 75 settings per trading rule; 3 rules × 3 timeframes × 4 periods = 2,700 runs | | Exit | Opposite cross for crossovers; opposite level for fades; remaining positions closed at period end | | Execution and costs | Candle close, fixed 0.3-pip spread, zero slippage | | Position size | 0.1 lot | | Stops and targets | Off in the main sweep; tested separately | Varying D would duplicate level-only settings, so those duplicates were excluded. Net pips were calculated from entry and exit prices. Extra filter, stop, target and spread runs are outside the 2,700-run count. ## Does slower smoothing stop K reaching 20 and 80? With K lookback 14 and D period 3, the 1-hour series contained 6,226 valid bars in 2025. The table counts bars outside each threshold. | K smoothing | Bars below 20 | Bars above 80 | Share below 20 | Share above 80 | | --- | --- | --- | --- | --- | | 1 | 1,238 | 1,534 | 19.88% | 24.64% | | 3 | 1,091 | 1,480 | 17.52% | 23.77% | | 5 | 969 | 1,323 | 15.56% | 21.25% | | 10 | 689 | 1,022 | 11.07% | 16.42% | | 20 | 280 | 512 | 4.50% | 8.22% | Increasing smoothing from 1 to 20 reduced the number of bars below 20 from 1,238 to 280. Smoothing of 20 still reached both thresholds, so a longer smoothing period did not make 20/80 unreachable in this sample. These are bar counts, not entry counts. Several consecutive bars can remain outside a threshold, and bars reached while a position is open do not create another trade. ## Which timeframe stayed profitable across years? Each trading rule has 75 distinct settings. A setting with no trades and zero net pips is not counted as profitable. | Timeframe | Trading rule | Profitable in 2024 | Profitable in 2025 | Profitable in both | 2025 median net (pips) | | --- | --- | --- | --- | --- | --- | | 15-minute | K/D cross | 2/75 | 4/75 | 1/75 | −2,190.0 | | 15-minute | Long fade | 28/75 | 17/75 | 11/75 | −712.6 | | 15-minute | Short fade | 0/75 | 16/75 | 0/75 | −662.4 | | 1-hour | K/D cross | 49/75 | 52/75 | 43/75 | +834.5 | | 1-hour | Long fade | 42/75 | 36/75 | 26/75 | 0.0 | | 1-hour | Short fade | 2/75 | 35/75 | 0/75 | −145.5 | | 4-hour | K/D cross | 62/75 | 67/75 | 56/75 | +1,143.7 | | 4-hour | Long fade | 29/75 | 59/75 | 18/75 | +643.2 | | 4-hour | Short fade | 1/75 | 56/75 | 0/75 | +560.7 | The 15-minute crossover's 2025 median was −2,190.0 pips, compared with +1,143.7 pips on the 4-hour chart. The direction of a fade also matters: no short-fade setting was profitable in both years on any timeframe. Splitting 2025 into halves left 54 of 75 hourly crossover settings profitable in the first half and 44 in the second. Annual profitability does not describe every part of the year. Because each run closes remaining positions at its own end date, adding the independent half-year results does not exactly reproduce the annual result. ## Should you increase smoothing on the 1-hour chart? Each row holds the same five K lookbacks and three D periods, changing only K smoothing. The annual means below are averages across 15 settings, not the return of a combined trading account. | K smoothing | Profitable in both | 2024 mean annual net (pips) | 2025 mean annual net (pips) | | --- | --- | --- | --- | | 1 | 2/15 | −1,121.3 | −1,347.3 | | 3 | 6/15 | −391.9 | +492.3 | | 5 | 8/15 | +592.7 | −40.5 | | 10 | 12/15 | +2,061.8 | +1,295.9 | | 20 | 15/15 | +3,443.8 | +1,744.1 | Settings profitable in both years increased from 2 with smoothing 1 to 15 with smoothing 20. The 2025 mean did not improve at every step: smoothing 3 produced +492.3 pips, while smoothing 5 produced −40.5 pips. ### Long fades had the most two-year profitable settings at smoothing 5 For fades, each row combines five K lookbacks with three level pairs. D stays fixed because the entry and exit rules do not use it. | K smoothing | Profitable in both | 2024 mean annual net (pips) | 2025 mean annual net (pips) | | --- | --- | --- | --- | | 1 | 2/15 | −225.3 | −745.9 | | 3 | 3/15 | −57.9 | −387.7 | | 5 | 8/15 | +217.0 | −186.3 | | 10 | 7/15 | +258.5 | +347.9 | | 20 | 6/15 | +231.1 | +369.6 | Long fades had 8 settings profitable in both years at smoothing 5, versus 6 at smoothing 20. The hourly crossover's result at smoothing 20 therefore does not transfer directly to a level fade. Short fades had no settings profitable in both years at any smoothing period: | K smoothing | Profitable in both | 2024 mean annual net (pips) | 2025 mean annual net (pips) | | --- | --- | --- | --- | | 1 | 0/15 | −1,833.6 | −657.6 | | 3 | 0/15 | −1,630.1 | −318.9 | | 5 | 0/15 | −1,338.3 | −113.9 | | 10 | 0/15 | −1,184.6 | +301.6 | | 20 | 0/15 | −941.4 | +201.6 | ## Default 14/3/3 turned negative on the hourly chart The default K/D crossover, separated by timeframe and year: | Timeframe | Year | Trades | Win rate | Annual net (pips) | | --- | --- | --- | --- | --- | | 15-minute | 2024 | 6,136 | 37.24% | −1,547.6 | | 15-minute | 2025 | 6,051 | 38.13% | −1,661.5 | | 1-hour | 2024 | 1,525 | 38.36% | +686.6 | | 1-hour | 2025 | 1,510 | 37.35% | −49.7 | | 4-hour | 2024 | 359 | 43.18% | +3,093.0 | | 4-hour | 2025 | 383 | 42.82% | +2,959.6 | The hourly 14/3/3 cross made money in 2024 but lost 49.7 pips in 2025. The same periods were profitable in both years on the 4-hour chart. A period recommendation needs to specify the timeframe it was tested on. ## Did the previous year's best setting beat the next year's median? For each timeframe, the setting with the highest annual net pips was applied unchanged to the other year. Selecting in 2024 and testing in 2025 is the forward check. Selecting in 2025 and looking back at 2024 checks historical consistency; it is not a forward prediction. | Timeframe | Selection year | K/D/smoothing | Selection-year net (pips) | Test year | Trades | Test-year net (pips) | Test-year median (pips) | | --- | --- | --- | --- | --- | --- | --- | --- | | 1-hour | 2024 | 21/3/20 | +5,095.2 | 2025 | 657 | +1,920.5 | +834.5 | | 1-hour | 2025 | 14/3/20 | +3,211.3 | 2024 | 745 | +3,323.9 | +799.4 | | 4-hour | 2024 | 50/3/5 | +4,481.9 | 2025 | 263 | +598.0 | +1,143.7 | | 4-hour | 2025 | 21/5/10 | +3,354.8 | 2024 | 175 | −1,686.5 | +1,857.5 | On the hourly chart, the 2024 winner, 21/3/20, earned +1,920.5 pips in 2025 and exceeded that year's +834.5-pip median. On the 4-hour chart, the 2024 winner, 50/3/5, returned +598.0 pips in 2025, below the +1,143.7-pip median. The 4-hour setting with the highest 2025 net, 21/5/10, lost 1,686.5 pips when applied to 2024. Searching many settings makes the largest observed profit a selected result, which is why the other-year figures are shown alongside it. ## A win rate above 60% still left both fades negative The comparison below uses the 1-hour chart in 2025, periods 14/3/3 and fade levels 20/80. | Trading rule | Trades | Win rate | Average win (pips) | Average loss (pips) | Annual net (pips) | | --- | --- | --- | --- | --- | --- | | K/D cross | 1,510 | 37.35% | +32.72 | −19.56 | −49.7 | | Long fade | 106 | 60.38% | +46.83 | −85.05 | −574.9 | | Short fade | 107 | 61.68% | +53.62 | −98.85 | −513.9 | The long fade won 60.38% of 106 trades, but its average winner was +46.83 pips against an average loser of −85.05 pips. Those larger losses left annual net at −574.9 pips. The short fade also lost money despite winning 61.68% of 107 trades. Long-only and short-only trades are exposed differently to the market's direction. Profitable long-fade settings alone do not establish that the stochastic threshold predicted a reversal beyond that exposure. ## Could filters or stops reduce the losses? The following variants use the hourly 14/3/3 crossover. ADX uses period 14 with no directional requirement and filters entries only. Exits still use the opposite K/D cross. The time restriction also applies only to new entries. | Added condition | Year | Trades | Annual net (pips) | | --- | --- | --- | --- | | No added condition | 2024 | 1,525 | +686.6 | | No added condition | 2025 | 1,510 | −49.7 | | ADX ≥ 20 | 2024 | 1,005 | +355.7 | | ADX ≥ 20 | 2025 | 1,007 | +136.3 | | ADX ≥ 25 | 2024 | 698 | +194.0 | | ADX ≥ 25 | 2025 | 705 | +639.6 | | 08:00–16:00 UTC | 2024 | 524 | −360.9 | | 08:00–16:00 UTC | 2025 | 520 | +1,132.3 | | 30-pip stop / 60-pip target | 2024 | 1,351 | −620.9 | | 30-pip stop / 60-pip target | 2025 | 1,327 | −1,622.4 | | 100-pip stop / 200-pip target | 2024 | 1,505 | +356.4 | | 100-pip stop / 200-pip target | 2025 | 1,503 | −41.5 | An ADX threshold of 25 produced +639.6 pips over 705 trades in 2025, but reduced 2024 net compared with the unfiltered rule. The UTC session filter turned 2025 positive while turning 2024 negative. Showing only the improved year would give a different account of the filter. A 30-pip stop and 60-pip target left both years negative. An opposite cross can close a position before its stop or target is reached. If one candle touches both the stop and target, the stop takes precedence. ### A 0.3-pip spread erased the default crossover's gross profit These are 2025 annual results for hourly 14/3/3. Spread and net profit are both expressed in pips. | Trading rule | Trades | Annual net, spread 0 | Annual net, spread 0.3 | Annual net, spread 1.0 | | --- | --- | --- | --- | --- | | K/D cross | 1,510 | +403.4 | −49.7 | −1,106.7 | | Long fade | 106 | −543.1 | −574.9 | −649.1 | | Short fade | 107 | −481.8 | −513.9 | −588.8 | The crossover earned +403.4 pips at zero spread. Charging a 0.3-pip spread across 1,510 trades left −49.7 pips. Its break-even spread was 0.27 pips when displayed to two decimal places; that displayed value is not a guarantee that a spread of exactly 0.27 remains profitable. The long and short fades were already negative at zero spread. Transaction costs cannot account for their losses on their own. ## Related - [Stochastic RSI settings](/blog/stoch-rsi-settings): applying a range calculation to RSI instead of price - [RSI periods and threshold tests](/blog/rsi-settings): separating level fades from crosses - [Moving average disparity settings](/blog/ma-disparity-settings): measuring threshold reach before trading it ## Notes The sample covers USD/JPY in 2024 and 2025, with candle-close fills and fixed spreads. Variable spreads, execution differences and swap charges are excluded. Pre-start candles are provided for indicator calculation, and positions still open at each period's end are closed. Calculation needs enough history for K, smoothing and D, plus earlier values to identify a cross. --- ### Do Double Tops and Bottoms Reverse? 2 Years of USD/JPY Tested URL: https://formiq.jp/blog/double-top-bottom-settings Language: en Published: 2026-09-05 Updated: 2026-09-05 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: double top, double bottom, chart patterns, neckline, reversal, backtest, USDJPY Also available in: ja — https://formiq.jp/ja/blog/double-top-bottom-settings A double top has two highs near the same price with a low between them. A double bottom turns the geometry upside down: two lows with an intermediate high. The usual reading expects price to continue in the reversal direction after it crosses the neckline. Hindsight makes the two highs or lows look unambiguous. At the time of the first high or low, the bars to its right do not exist. This test waits for those bars to close, then starts measuring when the close moves below the intermediate low for a double top or above the intermediate high for a double bottom. With five confirming bars on each side of a pivot, only H1 moved in the reversal direction more than half the time in both 2024 and 2025. The 10-bar rates were 49.71% and 50.15% on M15, and 46.88% and 31.43% on H4. ## What double tops and double bottoms measure These are three-point price patterns, not oscillators with a formula. The mechanical double-top definition used here has four steps: 1. Find a swing high with the same number of lower highs on its left and right 2. Require two swing highs near the same price and a swing low between them 3. Use the intermediate low as the neckline 4. Mark the pattern formed when a close falls below that neckline The double-bottom definition reverses the highs and lows. Its intermediate high becomes the neckline, and the pattern is formed on a close above it. The two highs or lows may differ by 1.2 ATR(14), clamped between 0.1% and 1.0% of price. Pattern height must be at least 0.3 ATR, and the two pivots may be no more than 300 bars apart. Those rules match the current chart overlay. ### A confirmed shape and a neckline break happen at different times A five-bar swing is not knowable until five later bars have closed. With Swing Period at 30, the delay is 30 bars. That means 7.5 hours on M15, 30 hours on H1 and 120 hours on H4. The research separates entry when the second pivot becomes knowable from entry after the neckline close. No trade is moved back to the high or low where a pivot is later found. ## Double top and bottom settings and signals Formiq draws the patterns over the candles in the main price pane. | Platform | Where to find it | |---|---| | Formiq | Indicator list → Chart Patterns | | TradingView | Indicators → Technicals → Patterns, then Double Top / Double Bottom or All Chart Patterns | | MT4 / MT5 | Draw the two highs or lows and the neckline manually with horizontal or trend lines | [TradingView's own double-top definition](https://www.tradingview.com/support/solutions/43000653211-chart-pattern-double-top/) also calls the pattern formed when a close falls below the intermediate low. Merely seeing a possible second high is an earlier state. ### The one input is Swing Period | Input | Default | Range | Meaning | |---|---:|---:|---| | Swing Period | 30 | 5 to 100 | Number of bars required on each side before a high or low becomes a confirmed pivot | A lower value finds smaller price turns and draws more patterns. A higher value keeps larger formations, but it also increases the confirmation delay. Double tops are red and double bottoms are green. The overlay draws the two highs or lows, a dashed neckline and a green dotted target. The target is one pattern height beyond the neckline in the expected reversal direction. ### The common trading rule | Pattern | Entry | Target | Failure level | |---|---|---|---| | Double top | Sell after a close below the intermediate low | One pattern height below the neckline | 20% of pattern height above the average of the two highs | | Double bottom | Buy after a close above the intermediate high | One pattern height above the neckline | 20% of pattern height below the average of the two lows | The strategy sells double tops and buys double bottoms. An opposite pattern closes the trade, and the strategy does not open a new position on the same bar that closes the old one. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Period | 1 January to 31 December 2025, repeated on 2024 | | Timeframes | M15 / H1 / H4 | | Swing Period | 5 / 10 / 15 / 20 / 30 / 40 / 50 / 75 | | Entry timing | Second pivot confirmed / close beyond neckline | | Grid | 8 pivot strengths × 2 entry timings = 16 settings; 192 rows over three timeframes and four windows | | Exit | Opposite pattern; no stop or target in the base run | | Spread and fill | 0.3 pips, filled at the close | | Position size | 0.1 lot | | Look-ahead control | A pivot is unavailable until all right-side confirmation bars have closed | For the 10-bar measurement, a fall after a double top and a rise after a double bottom count as positive. The test covers full-year 2024 and 2025 windows and also splits 2025 into halves. ## Do these patterns really reverse? Only H1 cleared a 50% reversal rate in both years. The table uses five-bar pivots, waits for a neckline close, and then holds each pattern for 10 bars. It counts each formation before removing overlapping trades. | Timeframe | Year | Patterns | Moved in reversal direction | Mean move after 10 bars | |---|---:|---:|---:|---:| | M15 | 2024 | 690 | 49.71% | +0.47 pips | | M15 | 2025 | 680 | 50.15% | +1.27 pips | | H1 | 2024 | 124 | 56.45% | +1.82 pips | | H1 | 2025 | 162 | 51.85% | +4.66 pips | | H4 | 2024 | 32 | 46.88% | −6.06 pips | | H4 | 2025 | 35 | 31.43% | −19.14 pips | H1 stayed over 50%, but its double tops and double bottoms did not contribute equally. In 2024, 55 hourly double tops had a 43.64% reversal rate and a mean of −11.86 pips. The 69 double bottoms reached 66.67% and +12.73 pips. Combining them into 56.45% hides that split. H4 did not support a reversal claim. In 2025, 20 double tops reversed 35.00% of the time and 15 double bottoms reversed 26.67%. Both means pointed away from the expected direction. ## How often does the target get hit? M15 and H1 held at 64% to 66% in both years; H4 did not carry across. The standard target is one pattern height beyond the neckline. I followed that target and the failure level for up to 300 bars, using five-bar pivots. | Timeframe | Year | Formed patterns | Target first | Failure first | Same bar / unresolved | |---|---:|---:|---:|---:|---:| | M15 | 2024 | 690 | 443 (64.20%) | 238 | 7 / 2 | | M15 | 2025 | 681 | 439 (64.46%) | 238 | 0 / 4 | | H1 | 2024 | 124 | 81 (65.32%) | 42 | 1 / 0 | | H1 | 2025 | 162 | 107 (66.05%) | 54 | 0 / 1 | | H4 | 2024 | 32 | 23 (71.88%) | 9 | 0 / 0 | | H4 | 2025 | 36 | 18 (50.00%) | 15 | 0 / 3 | M15 and H1 stayed between 64% and 66% in both years. The reward and risk are not the same size, however. The target is one pattern height from the neckline; the failure level is roughly 1.2 pattern heights from entry. Reading the target percentage as an ordinary win rate would ignore that difference. ## What Swing Period should you use? A longer one removes the H4 observations first. The default of 30 is compared with the minimum of 5. All rows enter after the neckline and exit on an opposite pattern. | Timeframe | Swing Period | 2024 trades | 2024 net pips | 2025 trades | 2025 net pips | |---|---:|---:|---:|---:|---:| | M15 | 5 | 225 | +459.3 | 225 | +275.1 | | M15 | 30 | 19 | −1,405.4 | 15 | −284.5 | | H1 | 5 | 37 | +613.3 | 53 | −562.7 | | H1 | 30 | 3 | +677.6 | 5 | −235.6 | | H4 | 5 | 12 | −187.2 | 13 | −763.2 | | H4 | 30 | 0 | 0.0 | 3 | +225.3 | Swing Period 30 found no completed H4 trade in 2024 and only three in 2025. The +225.3 pips and 33.33% win rate from three trades cannot carry the same weight as the 225 M15 trades. Across all eight Swing Period values, the number profitable in both years was 2/8 on M15, 0/8 on H1 and 2/8 on H4 after a neckline close. The two H4 survivors still had very few trades, so their count is not evidence that H4 was the better chart. ## Is waiting for the neckline safer? It changed the sign on M15 and nowhere else. Swing Period stays at 5; only entry timing changes. | Timeframe | Year | Second pivot confirmed: trades / net pips | After neckline: trades / net pips | |---|---:|---:|---:| | M15 | 2024 | 499 / −589.9 | 225 / +459.3 | | M15 | 2025 | 469 / −195.4 | 225 / +275.1 | | H1 | 2024 | 86 / +1,135.1 | 37 / +613.3 | | H1 | 2025 | 105 / −404.7 | 53 / −562.7 | | H4 | 2024 | 21 / −1,667.2 | 12 / −187.2 | | H4 | 2025 | 27 / −643.3 | 13 / −763.2 | Waiting for the neckline changed M15 from a loss to a profit in both years. H1 and H4 did not show the same improvement. The neckline cut trade count roughly in half, but lower frequency did not make every timeframe profitable. ## Does last year's best still hold? Three of six stayed profitable. The best 2024 Swing Period was moved unchanged to 2025, and the best 2025 value was moved to 2024. Entry remains after the neckline. | Timeframe | Selection year | Swing Period | Net pips in selection year | Net pips in other year | 2025 median | |---|---:|---:|---:|---:|---:| | M15 | 2024 | 5 | +459.3 | 2025: +275.1 | −360.6 | | M15 | 2025 | 10 | +685.5 | 2024: +288.3 | −360.6 | | H1 | 2024 | 30 | +677.6 | 2025: −235.6 | −262.6 | | H1 | 2025 | 50 | +1,875.0 | 2024: −443.2 | −262.6 | | H4 | 2024 | 10 | +777.5 | 2025: +676.7 | +568.8 | | H4 | 2025 | 15 | +886.9 | 2024: −809.1 | +568.8 | Three of the six transfers remained profitable. Splitting 2025 in half left only five of 24 timeframe-and-setting cells profitable in both halves: 2/8 on M15, 1/8 on H1 and 2/8 on H4. ## Why a 60% win rate lost money The average win never reached the average loss. These are the default Swing Period 30 results after waiting for the neckline. | Timeframe | Year | Trades | Win rate | Average win | Average loss | Net pips | |---|---:|---:|---:|---:|---:|---:| | M15 | 2025 | 15 | 33.33% | +227.16 | −142.03 | −284.5 | | M15 | 2024 | 19 | 31.58% | +195.01 | −198.11 | −1,405.4 | | H1 | 2025 | 5 | 60.00% | +274.30 | −529.25 | −235.6 | | H1 | 2024 | 3 | 66.67% | +544.30 | −411.00 | +677.6 | | H4 | 2025 | 3 | 33.33% | +687.90 | −231.27 | +225.3 | | H4 | 2024 | 0 | n/a | n/a | n/a | 0.0 | The five H1 trades in 2025 won 60% of the time and lost 235.6 pips. Three wins averaged +274.30 pips; two losses averaged −529.25. Five trades are also too few to treat 60% as a setting-selection statistic. ## Do ADX and session filters help? Neither improved both years the same way. Filters were applied to the five-bar neckline entry, which had enough trades to compare. Adding ADX(14) at 20 or higher to M15 raised 2024 from +459.3 to +1,088.1 pips, then changed 2025 from +275.1 to −1,005.1. The Tokyo window from 00:00 to 09:00 UTC returned +684.2 pips from 25 H1 trades in 2024 and +679.1 from 32 in 2025. That session was selected after inspecting five filter candidates, so it still needs a separate period before it can be treated as a rule. A fixed 30-pip stop and 60-pip target also changed with timeframe. | Timeframe | 2024 | 2025 | |---|---:|---:| | M15 | +1,345.8 pips (393 trades) | +103.9 pips (425 trades) | | H1 | +67.1 pips (101 trades) | +297.2 pips (141 trades) | | H4 | −39.3 pips (31 trades) | −258.3 pips (36 trades) | The same 30/60-pip exit is wide relative to M15 and narrow relative to H4. This table cannot separate the pattern from the mismatched fixed-pip scale as the source of the H4 loss. ## How much spread can it take? M15 broke even at 1.52 pips. The five-bar M15 neckline strategy took 225 trades in 2025. It returned +342.6 pips at zero spread, +275.1 at 0.3 pips, +117.6 at 1.0 pip and −107.4 at 2.0 pips. One spread is charged per completed trade, so the break-even spread is 342.6 ÷ 225 = 1.52 pips. The same setting still returned +76.8 pips at a 2.0-pip spread in 2024. Cost tolerance did not carry over unchanged either. ## Related reading - [SMC structure breaks and liquidity sweeps](/blog/smc-settings): another test built from confirmed swing highs and lows - [Do fair value gaps really fill?](/blog/fair-value-gap-settings): a popular price-action claim compared with ordinary bars - [Fibonacci retracement levels](/blog/fibonacci-retracement-levels): pivot confirmation delay and reactions at projected levels - [Candlestick pattern win rates](/blog/candlestick-patterns-tested): one- to three-bar shapes over the same two years and the same control - [Do Support and Resistance Lines Work? 5,184 Backtests](/blog/support-resistance-settings): the same swings read as a horizontal line, bounce and break ## Notes The sample is USD/JPY from 1 January 2024 through 31 December 2025. Trades fill at the close with a 0.3-pip spread and 0.1-lot size. Bars before each annual window supply warm-up history. No pivot is used before its right-side confirmation bars have closed. When target and failure levels are both touched inside one candle, the outcome remains in a separate same-bar category because OHLC data cannot order the two touches. --- ### Is SMC Profitable? 144 Backtests on USD/JPY URL: https://formiq.jp/blog/smc-settings Language: en Published: 2026-09-04 Updated: 2026-09-04 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: SMC, Smart Money Concepts, BOS, liquidity sweep, market structure, indicator, backtest, USDJPY Also available in: ja — https://formiq.jp/ja/blog/smc-settings Smart Money Concepts (SMC) divides price action into ideas such as **market structure, liquidity sweeps, order blocks and fair value gaps (FVGs)**. Unlike RSI, SMC has no single accepted formula. Two indicators can use the same BOS or order-block label while disagreeing about swing confirmation, candle bodies and wicks. This test therefore covers two rules that can be reproduced from OHLC data: **a close through a confirmed swing** and **a wick through a confirmed swing followed by a close back inside**. The rules do not identify institutional orders. They translate two SMC labels into observable price patterns. Across USD/JPY in 2024 and 2025, **11 of 12 hourly structure-break settings were profitable in both years**. No M15 or H1 liquidity-sweep setting survived both years. ## What this SMC test measures There is no standard SMC calculation, so the swing definition comes first: 1. A swing high must be strictly higher than the same number of bars on each side. 2. A swing low must be strictly lower than the same number of bars on each side. 3. The swing becomes actionable only after the right-hand bars have closed. 4. Equal highs or lows do not qualify as swings. With three bars on each side, a pivot is not known until three bars after the plotted swing. Marking the pivot on the earlier candle is useful for drawing a chart, but entering there would use future information. Every trade in this test waits for confirmation. ### A structure break closes beyond the swing An upside structure break occurs when the close crosses above the latest confirmed swing high. The rule opens a long, then exits and reverses when a close crosses below the confirmed swing low. The downside rule is the mirror image. Some SMC implementations separate BOS from CHoCH according to the prior trend state. That classification depends on which earlier swings define the state. This article avoids assigning the label after the fact and tests one condition directly: whether the close crossed a confirmed swing. ### A liquidity sweep wicks through and closes back inside A bullish liquidity sweep trades below the latest confirmed swing low but closes above it. A bearish sweep trades above the confirmed swing high and closes below it. OHLC records that candle shape, but it does not show whether stop orders were present or who traded. “Liquidity sweep” refers to the price rule defined above, not an observation of trader identity or the order book. ## SMC settings and how to read them MT4 and MT5 do not include one standard SMC indicator. Third-party implementations use different definitions and inputs, so matching names do not guarantee matching signals. In Formiq, select “SMC market structure” from the backtest condition list. The condition does not add a new line to the chart. It evaluates the relationship between each candle and the latest confirmed swings. | Setting | Default | What it changes | |---|---:|---| | Reading | Structure break | Close through a swing, or wick through and close back inside | | Bars on each side of a swing | 3 | How many left- and right-hand bars confirm the pivot | Increasing the swing setting from 2 to 8 removes smaller pivots and reduces the number of trades. On the 2025 hourly structure-break rule, swing 2 produced 290 trades, swing 3 produced 201, swing 5 produced 127, and swing 8 produced 89–95. The research also varied an internal expiry window that is not shown in the settings screen. Windows of 50, 100 and 200 bars gave the same result at swing 2 and swing 3. Some swing-8 results changed between 50 and 100 bars, while 100 and 200 were identical in every tested cell. Because the window did not change the default swing-3 result, it is not exposed as a user setting. ## How this was measured | Item | Test condition | |---|---| | Market | USD/JPY | | Timeframes | M15, H1 and H4 | | Period | 2024 and 2025; 2025 also split into halves | | Readings | Structure break and liquidity sweep | | Bars on each side | 2, 3, 5 and 8 | | Swing validity | 50, 100 and 200 bars | | Main test count | 2 rules x 4 settings x 3 settings x 3 timeframes x 2 years = 144 conditions | | Execution | Enter at the signal-bar close; exit on the opposite signal | | Costs | Fixed 0.3-pip spread, no slippage | | Position size | 0.1 lot | Each rule and timeframe has 12 parameter settings. The out-of-sample check also selects and transfers settings only within those same 12 choices. ## Which timeframe did SMC work on? Eleven of 12 hourly structure-break settings survived both years. A setting counts as profitable when annual net pips are above zero. The final column is the median annual net across the 12 settings in 2025. | Trading rule | Timeframe | Profitable in 2025 | Profitable in 2024 | Profitable in both | 2025 median net | |---|---|---:|---:|---:|---:| | Structure break | M15 | 11/12 | 9/12 | 8/12 | +865.3 pips | | Structure break | H1 | 12/12 | 11/12 | 11/12 | +1,096.3 pips | | Structure break | H4 | 3/12 | 12/12 | 3/12 | −1,951.4 pips | | Liquidity sweep | M15 | 3/12 | 3/12 | 0/12 | −1,160.9 pips | | Liquidity sweep | H1 | 0/12 | 6/12 | 0/12 | −574.2 pips | | Liquidity sweep | H4 | 9/12 | 6/12 | 3/12 | +1,390.6 pips | The hourly structure break kept 11 of 12 settings profitable in both years. The default swing-3, 200-bar setting made 2,177.2 pips in 2024 and 2,214.2 pips in 2025. The H4 structure break moved the other way: all 12 settings were profitable in 2024, but only three were profitable in 2025, when the median fell to −1,951.4 pips. A structure-break label alone does not carry the same result across timeframes. ## Is a high win rate enough? It is not: the liquidity sweep won over 57% and still lost money. The following table uses the default swing-3, 200-bar settings in 2025. | Trading rule | Timeframe | Trades | Win rate | Average win | Average loss | Annual net | |---|---|---:|---:|---:|---:|---:| | Structure break | M15 | 832 | 35.82% | +49.44 pips | −25.35 pips | +1,194.9 pips | | Structure break | H1 | 201 | 40.80% | +100.25 pips | −50.48 pips | +2,214.2 pips | | Structure break | H4 | 59 | 38.98% | +98.15 pips | −107.99 pips | −1,630.4 pips | | Liquidity sweep | M15 | 1,040 | 57.88% | +21.31 pips | −32.51 pips | −1,410.5 pips | | Liquidity sweep | H1 | 244 | 56.97% | +45.01 pips | −79.45 pips | −2,086.4 pips | | Liquidity sweep | H4 | 56 | 58.93% | +100.44 pips | −90.15 pips | +1,241.1 pips | The M15 and H1 structure breaks won only 36–41% of their trades, but their average wins were about twice their average losses. The M15 and H1 liquidity sweeps won more than 57% and still lost money because their average losses were larger. The H4 liquidity sweep made 1,241.1 pips from 56 trades. That sample is too small to select a setting from the win rate alone before testing another year or market. ## Did the prior year's winner transfer? Only on the H1 and H4 structure breaks. The best 2024 setting was applied to 2025, and the best 2025 setting was applied to 2024. Each cell below reports net pips in the year receiving the setting. | Trading rule | Timeframe | 2024 winner in 2025 | 2025 winner in 2024 | |---|---|---:|---:| | Structure break | M15 | −199.0 pips | −282.3 pips | | Structure break | H1 | +2,214.2 pips | +2,177.2 pips | | Structure break | H4 | +847.5 pips | +1,905.4 pips | | Liquidity sweep | M15 | −2,234.2 pips | −1,578.4 pips | | Liquidity sweep | H1 | −1,052.6 pips | +377.0 pips | | Liquidity sweep | H4 | −1,020.8 pips | −1,786.5 pips | Swing 3 with a 50-bar validity window was the best H1 structure-break setting in both years. Swing 2 with a 50-bar window was also the best H4 structure-break setting in both years. The M15 structure-break winner and the M15 and H4 liquidity-sweep winners all turned negative when transferred. The halves of 2025 were uneven. The default H1 structure break made 260.9 pips in the first half and 1,929.2 in the second. The default M15 liquidity sweep moved from +54.8 pips in the first half to −1,467.7 in the second. ## Do ADX and FVG filters help? Neither raised hourly net pips. Adding ADX(14) to the default H1 structure break reduced 2025 net from 2,214.2 pips to 919.2 at ADX 20 and 293.2 at ADX 25. The trade count fell from 201 to 151 and 114. Requiring the structure break and a newly formed FVG on the same candle produced the following hourly results. | Minimum FVG size | Trades | Win rate | Annual net | |---|---:|---:|---:| | Any size | 93 | 43.01% | +811.9 pips | | 0.25 ATR | 42 | 47.62% | +377.7 pips | | 0.50 ATR | 22 | 36.36% | −82.6 pips | The FVG filter raised the win rate in one row but never exceeded the unfiltered 2,214.2-pip net. This test requires a structure break and FVG creation on the same candle. It does not model a sequence of sweep, structure shift and later return to the gap. ## Do fixed stops and targets help? Not across all three timeframes. Adding a 30-pip stop and 60-pip target changed the M15 liquidity sweep from −1,410.5 to +599.7 pips in 2025. The same exit lost 442.7 pips on H1 and 615.2 on H4. The 20/40 and 50/100 exits also failed to improve all three timeframes. For the H1 structure break, opposite-signal exits made 2,214.2 pips. Fixed 20/40, 30/60 and 50/100 exits made 98.1, 991.5 and 1,342.5 pips. Every fixed exit reduced net in that cell. ## Which rules pay the most spread? The M15 rules, which take the most trades. Moving the spread from zero to 0.3 pips reduced 2025 net by trade count multiplied by 0.3 pips, within 0.1 pip after rounding. | Trading rule | Timeframe | Trades | Zero spread | 0.3-pip spread | |---|---|---:|---:|---:| | Structure break | M15 | 832 | +1,444.5 pips | +1,194.9 pips | | Structure break | H1 | 201 | +2,274.4 pips | +2,214.2 pips | | Liquidity sweep | M15 | 1,040 | −1,098.5 pips | −1,410.5 pips | | Liquidity sweep | H1 | 244 | −2,013.2 pips | −2,086.4 pips | The M15 liquidity sweep remained 1,098.5 pips underwater before spread, so costs do not explain the whole loss. The H1 structure break remained positive at a 1.0-pip spread, with 2,073.5 net pips. Slippage and live fill differences were not included. ## Related articles - [Double top and bottom reversal test](/blog/double-top-bottom-settings): the same confirmed swing points measured as two-turn reversal patterns - [Do fair value gaps really fill?](/blog/fair-value-gap-settings): an SMC-related pattern compared with ordinary candles - [Fibonacci retracement levels tested](/blog/fibonacci-retracement-levels): another popular chart structure tested against nearby controls - [How to build a backtest without code](/blog/backtest-without-coding): repeat the structure-break and liquidity-sweep rules on another market ## Notes The sample covers USD/JPY from January 1, 2024 through December 31, 2025. Trades use the signal-bar close, a fixed 0.3-pip spread and 0.1 lot. A swing is never used before its right-hand confirmation bars have closed. OHLC cannot recover the intrabar path or trader identity, so the test does not claim to observe institutional orders. --- ### ATR Periods and Comparison Intervals: 720 Tests URL: https://formiq.jp/blog/atr-settings Language: en Published: 2026-09-02 Updated: 2026-09-02 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: ATR, Average True Range, volatility, indicator settings, backtest, USDJPY Also available in: ja — https://formiq.jp/ja/blog/atr-settings Average True Range (ATR) measures how far price moves per bar. It ignores whether the move was up or down and turns intrabar range and price gaps into one volatility line. Period 14 is the familiar default. But across 60 combinations of periods 5 to 50, comparison windows from 1 to 20 bars, and rising or falling ATR, **the comparison window changed the result more than the ATR period did.** On the hourly chart, the one-bar comparison averaged −562.6 pips in 2024 and −1,149.2 in 2025. Falling ATR over ten bars with period 20 made +3,022.6 pips in 2024 and +2,184.5 in 2025. A default-like period 14, five-bar comparison with rising ATR lost 639.6 and 393.1 pips. Period 20 is not a universal answer: **changing the period alone did not repair a weak reading of the line.** ## What ATR measures ATR starts with True Range, the largest of three distances: 1. Current high minus current low 2. The distance between the current high and the previous close 3. The distance between the current low and the previous close If the current candle spans ten pips but opened away from the previous close, True Range keeps the larger gap-inclusive distance. [MetaTrader 5's ATR documentation](https://www.metatrader5.com/en/terminal/help/indicators/oscillators/atr) gives the same three-part calculation. Formiq smooths True Range with Wilder's recurrence. For period N, the newest True Range receives a weight of 1/N and the previous ATR receives (N−1)/N. | Period | Weight of newest bar | Bars until old influence halves | |---:|---:|---:| | 5 | 20.0% | 3.11 | | 14 | 7.1% | 9.35 | | 50 | 2.0% | 34.31 | ATR(14) is not a simple average that forgets everything after 14 bars. It absorbs 7.1% of each new True Range while the older influence decays gradually. ### ATR has no trade direction True Range uses absolute distances, so ATR cannot fall below zero. A crash and a rally can both lift it. [TradingView's official ATR guide](https://www.tradingview.com/support/solutions/43000501823-average-true-range-atr/) explicitly describes ATR as a volatility measure rather than a price-direction indicator. The tested rule therefore did not invent a side from ATR. When ATR was above or below its value a specified number of bars earlier, **price rising over those same bars opened a long and price falling opened a short.** The opposite price direction closed the trade while the ATR regime still held. ## ATR settings and signals ATR appears as one line in a separate pane below price. It has no standard 30/70-style guide levels. | Platform | Where to add it | |---|---| | MT4 / MT5 | Insert → Indicators → Oscillators → Average True Range | | TradingView | Search Indicators for Average True Range | | Formiq | Pick ATR from the indicator list | ### Chart settings | Input | Default | What it changes | |---|---:|---| | Period | 14 | How quickly True Range is smoothed | | Line colour and width | Platform-dependent | Appearance only | The standard MetaTrader ATR calculation exposes the period. TradingView starts at 14 and lets the smoothing change among RMA, SMA, EMA and WMA. Formiq exposes the period and line appearance and uses Wilder smoothing for the calculation. ### Backtest settings | Input | Starting value | What it does | |---|---:|---| | Reading | ATR rising | Chooses whether ATR must be above or below its earlier value | | Period | 14 | Sets the response speed of the ATR line | | Comparison bars | 5 | Looks this many bars back for both ATR and price direction | The common interpretation is to read rising ATR as expanding movement and falling ATR as contraction. Neither supplies a side; price direction, a breakout, or another directional rule is still needed. Across the 60 settings, changing the period left mean hourly trade count between 815 and 828 in 2025. Changing the comparison window moved it from 1,569.8 trades at one bar to 337.8 at 20 bars. **The comparison window, not the period, was the main trade-frequency control.** ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Window | 2025, repeated in 2024 and split into first and second half of 2025 | | Timeframes | 15-minute / 1-hour / 4-hour | | Readings | ATR rising / ATR falling; price direction from the same comparison window | | ATR periods | 5, 10, 14, 20, 30, 50 | | Comparison windows | 1, 3, 5, 10, 20 bars | | Runs | 60 settings per cell; 3 timeframes × 4 windows = 720 | | Exit | Opposite signal; entry and exit share one rule set | | Fill and cost | Close, fixed 0.3-pip spread, 0.1 lot | | Stops and targets | None in the main sweep; tested separately | Pips were recalculated from each trade's fill prices. ## Which timeframe worked for ATR? Twenty-nine of 60 hourly settings were profitable in both 2024 and 2025, against 22 on the four-hour chart and 6 on the 15-minute. | Timeframe | Profitable 2024 | Profitable 2025 | Both years | 2024 median | 2025 median | |---|---:|---:|---:|---:|---:| | 15-minute | 24/60 | 9/60 | 6/60 | −312.8 | −1,205.8 | | 1-hour | 36/60 | 39/60 | **29/60** | +532.5 | +954.9 | | 4-hour | 50/60 | 22/60 | 22/60 | +1,674.2 | −711.6 | Only the hourly chart had a positive median in both years. The 15-minute chart traded often and its median setting lost in both years. Four-hour settings reversed from 50 of 60 profitable in 2024 to 22 of 60 in 2025. ## Period or comparison window: which matters? The comparison window separated results more than the period did. Each row averages 12 hourly settings across both readings and all six periods. | Comparison bars | Mean 2024 | Mean 2025 | Mean trades 2025 | |---:|---:|---:|---:| | 1 | −562.6 | −1,149.2 | 1,569.8 | | 3 | +794.5 | +1,146.4 | 940.8 | | 5 | +163.4 | +676.6 | 727.0 | | 10 | +869.6 | −40.6 | 517.8 | | 20 | +773.9 | +856.9 | 337.8 | The one-bar comparison lost in both years and traded 4.6 times as often as the 20-bar version. Reclassifying ATR as up or down on every small one-bar change generated too much turnover. Mean trade count by ATR period stayed between 815 and 828 in 2025. Period 5 led mean net profit in both years at +743.5 and +561.7 pips, but other periods did not preserve their ordering: period 30 moved from −106.2 to +310.3, while period 14 fell from +475.2 to +200.0. **Nothing in the grid made 14 uniquely strong.** ### Falling ATR had a positive mean in both years | Reading | Mean 2024 | Profitable 2024 | Mean 2025 | Profitable 2025 | |---|---:|---:|---:|---:| | ATR rising | −97.9 | 13/30 | +245.9 | 19/30 | | ATR falling | +913.5 | 23/30 | +350.1 | 20/30 | Following price while volatility expanded had a negative mean in 2024. Falling ATR averaged a profit in both years, but the 2025 profitable counts were nearly tied at 20 of 30 versus 19 of 30. The ATR regime alone did not select a robust setting. ## Does last year's winner win again? Two of six transfers lost money. The best setting in one year was run unchanged in the other: | Timeframe | Selection year | Setting | Selection year | Other year | |---|---:|---|---:|---:| | 15-minute | 2025 | Rising / 10 / 10 | +1,226.1 | 2024 +970.0 | | 15-minute | 2024 | Rising / 14 / 20 | +2,503.8 | 2025 **−738.7** | | 1-hour | 2025 | Falling / 10 / 5 | +2,541.7 | 2024 +341.8 | | 1-hour | 2024 | Falling / 20 / 10 | +3,022.6 | 2025 +2,184.5 | | 4-hour | 2025 | Falling / 10 / 3 | +2,809.6 | 2024 +2,359.1 | | 4-hour | 2024 | Falling / 5 / 1 | +4,225.7 | 2025 **−596.7** | The 2025 hourly winner stayed positive in 2024 at +341.8 pips, but finished below the 2024 median of +532.5. The hourly falling-ATR, period-20, ten-bar setting made +1,166.9 pips in the first half of 2025 and +993.5 in the second. Its survival was not confined to one half, but period 20 cannot be separated from the falling reading, ten-bar comparison and hourly chart that produced it. ## Is the default period 14 enough? Pairing the default period 14 with rising ATR left one profitable cell out of six. The chart default period with the strategy builder's five-bar starting window: | Timeframe | Year | Trades | Win rate | Net | |---|---:|---:|---:|---:| | 15-minute | 2024 | 2,111 | 35.20% | +68.4 pips | | 15-minute | 2025 | 2,292 | 33.77% | −807.1 pips | | 1-hour | 2024 | 577 | 34.32% | −639.6 pips | | 1-hour | 2025 | 574 | 32.40% | −393.1 pips | | 4-hour | 2024 | 127 | 35.43% | −558.6 pips | | 4-hour | 2025 | 139 | 37.41% | −860.1 pips | A common chart period is not automatically a trading-rule default. ## Why a low win rate still won The win rate stayed below 40% while the mean win ran more than twice the mean loss. The hourly falling-ATR, period-20, ten-bar setting: | Year | Trades | Win rate | Mean win | Mean loss | Net | |---|---:|---:|---:|---:|---:| | 2024 | 628 | 35.51% | +54.66 pips | −22.63 pips | +3,022.6 pips | | 2025 | 601 | 36.94% | +51.75 pips | −24.55 pips | +2,184.5 pips | Ranking ATR settings by win rate would discard that payoff difference. ## Do ADX and stops help? None of the five additions beat the baseline in both years. The same hourly setting with one addition at a time: | Addition | 2024 | 2025 | |---|---:|---:| | None | +3,022.6 | +2,184.5 | | ADX(14) at 20 or above | +646.9 | −572.2 | | Exit after 24 bars | +2,840.5 | +2,482.2 | | Stop 50 / target 100 pips | +2,942.2 | +1,472.7 | | Stop 1 ATR / target 2 ATR | +1,940.0 | +554.9 | | Stop 2 ATR / target 3 ATR | +1,456.6 | +1,151.7 | The 24-bar exit added 297.7 pips in 2025 but removed 182.1 in 2024. ADX and all four stop/target cases failed to beat the baseline in both years. An ATR-sized stop adapts the unit to current volatility; it does not create an exit edge by itself. In this sample, the fixed 50/100-pip exit finished ahead of both ATR exits in both years, and all three trailed the signal-only baseline. ## How much does the spread cost? A one-pip spread removed 600.9 pips from 601 trades. The same 2025 candidate: | Spread | Trades | Net | |---:|---:|---:| | 0 | 601 | +2,364.7 pips | | 0.3 | 601 | +2,184.5 pips | | 1.0 | 601 | +1,763.8 pips | | 2.0 | 601 | +1,162.8 pips | The break-even spread was 3.93 pips. Cost declines by trades multiplied by spread, so the high-turnover one-bar settings reach their limit sooner. ## Related articles - [Supertrend settings: the multiplier moved the result](/blog/supertrend-settings) - [Bollinger BandWidth settings and the squeeze](/blog/bollinger-bandwidth-settings) - [Moving average disparity: percent versus ATR units](/blog/ma-disparity-settings) ## Notes - The sample runs from 1 January 2024 through 31 December 2025; 2025 was also split into halves - Fills use the close and the main sweep fixes spread at 0.3 pips - ATR uses Wilder smoothing with warmup before each measured window - MetaTrader, TradingView and Formiq expose different smoothing choices, so match the method before moving a numerical setting between platforms --- ### Are DMI/ADX DI Crossovers Profitable? 1,344 Settings Tested URL: https://formiq.jp/blog/dmi-adx-settings Language: en Published: 2026-09-02 Updated: 2026-09-02 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: DMI/ADX, ADX, DMI, +DI, −DI, trend, indicator, backtest, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/dmi-adx-settings Directional Movement Index (DMI) puts three lines in one pane: +DI for positive directional movement, −DI for negative directional movement, and ADX for the smoothed separation between them. The DI lines supply direction. ADX uses an absolute difference, so it measures strength without distinguishing an uptrend from a downtrend. Period 14 with ADX at least 25 is the familiar setup. On USD/JPY in 2025, adding that threshold cut the period-14 DI crosses from **2,184 to 339 on M15, 534 to 101 on H1, and 141 to 19 on H4**. Only 13.5% to 18.9% of the crosses remained. The smaller set was not consistently better. The 14/25 DI cross made money in both years only on M15. H1 lost 68.1 pips in 2025 after gaining 137.5 in 2024. Used as a filter on a 10/20 SMA cross, ADX 25 reduced 2025 net from +377.4 to −922.1 pips. ## What DMI/ADX measures The calculation starts with changes in consecutive highs and lows. 1. Measure how far the current high rose above the previous high. 2. Measure how far the previous low stood above the current low. 3. Keep only the larger positive move as +DM or −DM. 4. Divide the smoothed directional moves by true range to produce +DI and −DI. 5. Calculate DX as `|+DI − −DI| ÷ (+DI + −DI) × 100`, then smooth DX into ADX. Taking the absolute value removes direction. **ADX at 30 does not mean price is rising. It means the two DI lines have separated.** The [MetaTrader 5 ADX documentation](https://www.metatrader5.com/en/terminal/help/indicators/trend_indicators/admi) describes the basic period-14 rule as buying when +DI is above −DI and selling when −DI moves above +DI. MetaTrader's [Wilder calculation reference](https://www.metatrader5.com/en/terminal/help/indicators/trend_indicators/admiw) shows the full sequence from +DM, −DM and true range through DI, DX and ADX. ## DMI/ADX settings and signals DMI is drawn in **a separate pane below the candles**, with one ADX line and two DI lines. All three run from 0 to 100. | Platform | Where to find it | |---|---| | MT4 / MT5 | Insert → Indicators → Trend → Average Directional Movement Index | | TradingView | Search the indicator list for Directional Movement Index (DMI) | | Browser (Formiq) | Pick ADX from the indicator list | TradingView separates ADX from DMI. Its [official DMI reference](https://www.tradingview.com/support/solutions/43000502250-directional-movement-dmi/) directs readers to DMI when all three lines are needed. MT4 and MT5 show the three lines together; MT5 also offers a separate ADX Wilder calculation. ### Chart settings | Field | Default | What it changes | |---|---:|---| | Period | 14 | Smoothing length for +DM, −DM, true range and DX | | ADX line | Platform colour | Trend strength without direction | | +DI line | Platform colour | Relative strength of positive directional movement | | −DI line | Platform colour | Relative strength of negative directional movement | | Levels | Optional | Reference lines such as 20 or 25 | A shorter period reacts sooner and produces more DI crosses. A longer period smooths the lines but reacts later to short trends. Changing the threshold does not change the lines; it changes which signals are accepted. ### Backtest settings | Field | Default | What it changes | |---|---:|---| | Period | 14 | Number of bars in the DMI/ADX calculation | | Minimum | 25 | Allows a new entry only while ADX is at or above this level | | Direction | Filter only | Keeps the direction chosen by another rule and adds only the ADX threshold | The direction menu has three choices: - **Filter only:** require the ADX threshold but take direction from another rule, such as a moving-average cross - **+DI/−DI cross:** buy when +DI crosses above −DI and sell on the reverse cross, provided ADX meets the threshold - **+DI/−DI comparison:** while ADX meets the threshold, buy when +DI is higher and sell when −DI is higher; entry does not have to occur on the crossing bar The sweep tests the DI cross and ADX acceleration as separate trading rules. ADX used only as a filter is tested by adding it to a 10/20 SMA cross. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Window | 2025-01-01 to 2025-12-31, with 2024 run identically for comparison | | Timeframes | 15-minute / 1-hour / 4-hour | | Readings | +DI/−DI cross / +DI/−DI comparison | | Periods | 5 / 7 / 10 / **14** / 20 / 28 / 40 / 50 | | Minimum ADX | 0 / 15 / 20 / **25** / 30 / 35 / 40 | | Combinations | 112 per cell; 3 timeframes × 4 windows = 1,344 runs | | Exit | The opposite DI signal; entry and exit share the same rule set | | Stops and targets | None in the main sweep, tested separately below | | Spread and fill | 0.3 pips fixed, filled at the close | | Size | 0.1 lot | Level 0 is the no-ADX-filter control. The 2025 sample is also split into two halves so a full-year total cannot hide a reversal within the year. ## What does the ADX 25 level mean? It keeps about half the market while removing more than four in five DI crosses. The table counts period-14 bars above each threshold and the DI crosses that survived level 25. | Timeframe | Year | ADX ≥ 20 bars | ADX ≥ 25 bars | ADX ≥ 30 bars | All DI crosses | Crosses at ADX ≥ 25 | |---|---:|---:|---:|---:|---:|---:| | M15 | 2024 | 63.65% | 43.47% | 27.93% | 2,136 | 349 | | M15 | 2025 | 61.22% | 40.51% | 25.15% | 2,184 | 339 | | H1 | 2024 | 64.82% | 44.91% | 29.49% | 501 | 93 | | H1 | 2025 | 66.33% | 47.24% | 31.40% | 534 | 101 | | H4 | 2024 | 73.82% | 50.43% | 30.45% | 142 | 35 | | H4 | 2025 | 67.20% | 37.58% | 17.58% | 141 | 19 | ADX was at least 25 on 37.58% to 50.43% of all bars, yet only 13.5% to 24.6% of DI crosses met the same threshold. A DI cross marks a change in direction; ADX is a smoothed measure of separation and often remains low on the crossing bar. The timing mismatch also appears in the slope. At the 2025 crossing bars, ADX was rising on only 218 of 2,184 M15 crosses, 53 of 534 H1 crosses and 11 of 141 H4 crosses. ## How many settings held for two years? Between 3 and 16 of 56. Each reading has eight periods by seven thresholds. ### DI cross | Timeframe | Profitable 2025 | 2025 median | Profitable 2024 | 2024 median | Profitable both years | |---|---:|---:|---:|---:|---:| | M15 | 17/56 | −40.1 pips | 17/56 | −189.8 pips | **10/56** | | H1 | 27/56 | 0.0 pips | 19/56 | −37.9 pips | **9/56** | | H4 | 6/56 | −167.1 pips | 29/56 | +3.2 pips | **4/56** | ### DI comparison | Timeframe | Profitable 2025 | 2025 median | Profitable 2024 | 2024 median | Profitable both years | |---|---:|---:|---:|---:|---:| | M15 | 18/56 | −335.5 pips | 26/56 | −53.2 pips | **13/56** | | H1 | 23/56 | −79.4 pips | 32/56 | +118.4 pips | **16/56** | | H4 | 3/56 | −817.8 pips | 38/56 | +359.2 pips | **3/56** | The H4 DI-comparison grid moved from 38 profitable settings in 2024 to three in 2025. Changing the reading did not remove the timeframe reversal. ## How did the defaults 14 and 25 do? The DI cross held its ground while DI comparison changed sign on every timeframe. The default combination produced the following runs. | Reading | Timeframe | 2025 trades | 2025 win rate | 2025 net | 2024 trades | 2024 net | |---|---|---:|---:|---:|---:|---:| | DI cross | M15 | 339 | 32.45% | **+141.1 pips** | 349 | **+403.3 pips** | | DI cross | H1 | 101 | 33.66% | −68.1 pips | 93 | +137.5 pips | | DI cross | H4 | 19 | 42.11% | +433.3 pips | 35 | +152.0 pips | | DI comparison | M15 | 686 | 32.65% | −391.2 pips | 679 | +570.3 pips | | DI comparison | H1 | 181 | 34.25% | −388.5 pips | 179 | +72.3 pips | | DI comparison | H4 | 45 | 24.44% | −2,277.2 pips | 59 | +1,042.7 pips | The H4 cross was profitable in both years, but only 19 trades occurred in 2025 and 35 in 2024. M15 supplied 339 and 349 trades and stayed profitable. DI comparison changed sign between the two years on every timeframe. ## Where should the ADX level sit? Nothing marked 25 as a boundary. These are hourly marginal means across all eight periods. Settings with no trades remain in the average at zero pips. ### DI-cross threshold | Minimum ADX | Mean net 2025 | Mean net 2024 | Mean trades 2025 | |---:|---:|---:|---:| | 0 | −37.4 pips | **+842.1 pips** | 541.1 | | 15 | **+447.4 pips** | −320.5 pips | 381.1 | | 20 | +231.5 pips | −416.4 pips | 265.6 | | 25 | +94.6 pips | −239.8 pips | 170.5 | | 30 | +17.4 pips | −120.1 pips | 101.4 | | 35 | +7.6 pips | −122.5 pips | 57.8 | | 40 | +39.5 pips | +13.9 pips | 31.8 | Level 0 averaged +842.1 pips in 2024 and −37.4 in 2025. Level 15 reversed the pattern, gaining 447.4 pips in 2025 and losing 320.5 in 2024. Level 40 was positive in both averages, but traded only 31.8 times per setting in 2025. Nothing special happened at 25. ### DI-comparison threshold | Minimum ADX | Mean net 2025 | Mean net 2024 | Mean trades 2025 | |---:|---:|---:|---:| | 0 | −84.0 pips | +788.4 pips | 542.1 | | 15 | −36.1 pips | −14.9 pips | 417.5 | | 20 | −317.6 pips | +52.2 pips | 326.8 | | 25 | −215.6 pips | +80.9 pips | 248.1 | | 30 | −178.2 pips | −47.7 pips | 183.5 | | 35 | −295.6 pips | −107.8 pips | 136.4 | | 40 | −138.7 pips | +22.3 pips | 98.8 | Every DI-comparison threshold had a negative mean in 2025. Raising the minimum did not recover the loss. Periods 5 and 7 were the only smoothing lengths with positive hourly marginal means in both years for both readings. The DI-cross means at period 5 were +347.9 pips in 2025 and +225.0 in 2024; period 7 returned +307.1 and +71.2. DI comparison returned +833.4 and +663.0 at period 5, then +437.7 and +476.5 at period 7. The shorter setting also averaged 596.1 cross trades and 729.3 comparison trades in 2025, so its result includes a much larger spread bill. ## Did last year's winners hold? Four of six lost money the next year. Each 2024 winner was carried unchanged into 2025. | Reading | Timeframe | Period | Level | 2024 net | 2025 net | 2025 median | |---|---|---:|---:|---:|---:|---:| | DI cross | M15 | 28 | 0 | +1,305.0 pips | +366.0 pips | −40.1 pips | | DI cross | H1 | 50 | 0 | +1,513.1 pips | **−1,309.9 pips** | 0.0 pips | | DI cross | H4 | 20 | 0 | +1,738.5 pips | **−752.8 pips** | −167.1 pips | | DI comparison | M15 | 20 | 20 | +1,630.4 pips | +86.0 pips | −335.5 pips | | DI comparison | H1 | 50 | 0 | +1,391.5 pips | **−1,352.7 pips** | −79.4 pips | | DI comparison | H4 | 20 | 15 | +1,677.3 pips | **−1,329.7 pips** | −817.8 pips | The two M15 winners remained positive, but retained only 22.4% and 5.3% of their 2024 net. Both H1 and H4 winners fell below their 2025 median under both readings. ## Is win rate enough to judge? It is not. The 14/25 hourly DI cross made 101 trades in 2025, won 33.66%, and lost 68.1 pips. Its average win was +54.05 pips and average loss −28.44 pips. Winning only one trade in three nearly broke even because the average win was about 1.9 times the average loss. The 2025 H4 result looks better at 42.11% and +433.3 pips, but it contains only 19 trades. DI comparison at the same setting made 45 trades, won 24.44%, and lost 2,277.2 pips. Trade count, average win and loss, and annual net are needed alongside the win rate. ## Does ADX work as a filter? It reduced profit on a 10/20 SMA cross in both years. Filter only was added to the entry of a 10/20 SMA cross. The exit remained the unfiltered opposite cross. | Year | ADX entry filter | Trades | Win rate | Annual net | |---:|---|---:|---:|---:| | 2025 | None | 358 | 39.11% | **+377.4 pips** | | 2025 | At least 20 | 197 | 36.55% | −80.9 pips | | 2025 | At least 25 | 117 | 31.62% | −922.1 pips | | 2025 | At least 30 | 77 | 32.47% | −358.8 pips | | 2024 | None | 315 | 41.59% | **+2,888.4 pips** | | 2024 | At least 20 | 169 | 42.01% | +1,349.6 pips | | 2024 | At least 25 | 106 | 48.11% | +1,073.2 pips | | 2024 | At least 30 | 59 | 44.07% | +462.8 pips | Level 25 raised the 2024 win rate from 41.59% to 48.11% while cutting net from +2,888.4 to +1,073.2 pips. All three thresholds turned the profitable 2025 baseline into a loss. Waiting for an already strong reading removed profitable early entries as well as losing crosses. ## Do sessions and stops help? Neither improved both years the same way. The variants use the 14/25 hourly DI cross. | Variant | 2025 trades | 2025 net | 2024 trades | 2024 net | |---|---:|---:|---:|---:| | Baseline | 101 | −68.1 pips | 93 | +137.5 pips | | Tokyo, UTC 0–8 | 33 | −457.5 pips | 37 | +351.1 pips | | London and New York, UTC 7–21 | 66 | +511.1 pips | 56 | −7.6 pips | | Stop 30 / target 60 | 93 | +603.0 pips | 87 | +253.2 pips | | Stop 50 / target 100 | 97 | +612.3 pips | 87 | −246.1 pips | | Stop 100 / target 200 | 100 | +170.6 pips | 90 | −274.5 pips | | Stop 50 / target 50 | 97 | +700.5 pips | 88 | +31.4 pips | The London and New York window gained 511.1 pips in 2025 and lost 7.6 in 2024. Stop 30 / target 60 beat the baseline in both years. It was selected from four tested exit pairs and still needs another period for confirmation. ## Is the spread to blame? It is not. The 14/25 hourly DI cross lost 37.8 pips at zero spread and 68.1 pips at 0.3. The 30.3-pip difference equals 101 trades times 0.3 pips, so transaction cost alone does not explain the loss. In 2024, the same rule returned +165.3 pips at zero spread, +137.5 at 0.3, +25.9 at 1.5, and −20.6 at 2.0. Its break-even spread was about 1.78 pips. The losing 2025 run has no positive break-even spread. ## Related articles - [Moving-average cross settings](/blog/moving-average-cross-settings): the 10/20 SMA baseline used for the filter-only test - [Parabolic SAR settings](/blog/parabolic-sar-settings): another trend rule whose net fell after adding ADX - [CCI settings](/blog/cci-settings): ADX thresholds tested on three readings of one trend indicator - [Alligator settings](/blog/alligator-settings): the base rate of a three-line trend state measured before using it as a filter - [Do Trend Filters Remove False Signals? 1,596 Backtests](/blog/false-signal-filters): the same ADX used as a filter on all 38 published settings - [Build and test the conditions without code](/blog/backtest-without-coding): combine DMI/ADX with another rule on a different pair or window ## Notes The sample uses USD/JPY in 2024 and 2025, close fills, a fixed 0.3-pip spread and 0.1 lot. ADX(14) needs 28 bars before its first value in this implementation. The main sweep exits on the opposite DI signal; fixed stops and targets are measured separately. --- ### Is the Ichimoku Three-Signal Buy Profitable? 280 Settings Tested URL: https://formiq.jp/blog/ichimoku-settings Language: en Published: 2026-09-02 Updated: 2026-09-04 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Ichimoku, three-signal alignment, time theory, Tenkan-sen, Kijun-sen, Senkou span, Chikou span, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/ichimoku-settings Ichimoku puts Tenkan-sen, Kijun-sen, two Senkou spans and Chikou span over the candles. The names make it look more complicated than the calculation: most of the lines are midpoints of a recent high-low range, and three of the five are shifted horizontally. Across USD/JPY from 2022 through 2025, the **standard 9/26/52 three-signal alignment made +558.1, −1,012.7, +536.0 and +1,085.2 pips on H1**. Three aligned conditions did not survive all four years. The standard price/Kijun cross, by contrast, was positive in each year. I also fixed a mechanical swing rule and measured the proposed time-theory change dates. Combining bars 9, 17, 26, 33, 42, 52, 65 and 76 produced a local-turn rate of 34.19% to 35.49% across the three timeframes. The matched dates three bars earlier or later produced 34.23% to 35.04%, so the full set added no distinct edge. Count 26 stood out on its own: 39.52% of its H1 candidates were local turns, versus 32.59% for the matched controls. ## What the five Ichimoku lines measure | Line | Standard period | Calculation | Plotted at | |---|---:|---|---| | Tenkan-sen | 9 | (9-bar highest high + lowest low) / 2 | Current bar | | Kijun-sen | 26 | (26-bar highest high + lowest low) / 2 | Current bar | | Senkou Span A | 9 and 26 | (Tenkan + Kijun) / 2 | 26 bars ahead | | Senkou Span B | 52 | (52-bar highest high + lowest low) / 2 | 26 bars ahead | | Chikou span | 26 | Current close | 26 bars back | The area between Senkou A and B is the Kumo, or cloud. The usual reading is bullish above the cloud, bearish below it, and undecided inside it. ### Senkou spans are not forecasts Plotting a line 26 bars ahead does not put future prices into its calculation. The Senkou B value visible at the current x-position was calculated 26 bars ago from the 52 bars available then. The shift changes where the value is drawn, not what the indicator knows. ### Bringing Chikou back to the present Chikou plots the current close 26 bars back. Comparing Chikou with the price at its plotted position is therefore identical to comparing the current close with price 26 bars ago. The stored Chikou values matched the close 26 bars later without a mismatch in any of the 12 timeframe/year cells. That identity is not predictive evidence; it confirms that a current close has been moved into the past. A test without look-ahead bias uses `current close > close 26 bars ago` at the current bar. ## Do the time-theory change dates work? Across all eight counts they matched ordinary dates. Ichimoku is broader than its five plotted lines. Its framework includes time theory, wave theory and price-observation theory. Time theory counts bars from a significant high or low. The [IFTA 2025 syllabus](https://www.ifta.org/assets/docs/IFTA_CFTe_Syllabus_Reading_Material.pdf) lists 9, 17, 26, 33, 42, 51, 65 and 76, among others. Some general guides use 52 instead, so instructional sources do not agree on that count. A candidate change date is not a directional signal or a guaranteed reversal. Some interpretations also count acceleration in the same direction or an extension in timing as change. I kept the outcome narrower here: **did a mechanical local high or low occur near the scheduled bar?** ### The anchor and outcome were fixed before reading the results | Item | Mechanical definition | |---|---| | Anchor | The bar is the unique highest high or lowest low among the five bars on either side; confirmed five bars later | | Change-date counts | 9, 17, 26, 33, 42, 52, 65 and 76, counting the anchor as bar one | | Hit | The scheduled bar or either adjacent bar is a local high or low under the same definition | | Control | The same anchor, but three bars before and three bars after each named count | | Duplicates | Multiple projections onto the same scheduled bar count once; 2022 through 2025 are also reported separately | | Uncertainty | A 95% interval for the difference from 5,000 month-block resamples | The anchor is known before the shortest nine-bar target. In code, inclusive counting makes the target `anchor + count - 1`. The primary analysis used 52, while substituting 51 was kept as a sensitivity check. ### About 34% to 35% were local turns, but so were the controls | Timeframe | Candidate bars | Candidate local turns | Candidate turn rate | Control turn rate | 95% interval for difference | |---|---:|---:|---:|---:|---:| | M15 | 63,192 | 21,650 | 34.26% | 34.23% | −0.21 to +0.29 pp | | H1 | 15,923 | 5,444 | 34.19% | 34.84% | −1.17 to −0.13 pp | | H4 | 4,142 | 1,470 | 35.49% | 35.04% | −0.66 to +1.58 pp | With all eight counts active, the one-bar windows around candidates covered 96.26% to 96.73% of all candles. Almost the entire chart became a candidate, so a raw hit rate cannot show that time theory selected the changes. Replacing 52 with 51 left differences of only +0.17 pp on M15, −0.17 pp on H1 and +0.48 pp on H4. ### Count 26 beat its H1 and H4 controls Restricting the set to basic counts 9, 17 and 26 put H1 at 35.93% against 33.58%, a +2.35 pp difference with a +1.14 to +3.54 pp 95% interval. M15 differed by +0.57 pp and H4 by −0.08 pp. Most of the H1 difference came from count 26. Across four years, 1,195 of 3,024 H1 candidates were local turns, or 39.52%. The matched 23- and 29-bar controls produced 32.59%, a +6.93 pp difference with a +5.10 to +8.75 pp interval. H4 produced 39.34% against 32.00%; M15 produced 34.61% against 34.62%. | Timeframe | Year | Count-26 turn rate | Control turn rate | Difference | |---|---:|---:|---:|---:| | H1 | 2022 | 39.24% | 33.87% | +5.37 pp | | H1 | 2023 | 39.97% | 32.46% | +7.52 pp | | H1 | 2024 | 40.22% | 32.18% | +8.04 pp | | H1 | 2025 | 38.64% | 31.84% | +6.79 pp | | H4 | 2022 | 36.95% | 33.79% | +3.16 pp | | H4 | 2023 | 35.64% | 29.53% | +6.11 pp | | H4 | 2024 | 45.41% | 31.37% | +14.04 pp | | H4 | 2025 | 38.95% | 33.24% | +5.71 pp | Changing the anchor window to three, five or ten bars on either side left count 26 ahead by +3.99 to +7.13 pp on H1 and +3.25 to +8.64 pp on H4. This is not a test of direction, profit, acceleration or extension. It is also one result among several counts and timeframes, so a separate period and pair need to reproduce it before “count 26 predicts a reversal” is justified. Equal-interval projections, which carry the distance between past swing points forward from another swing, are not included because they require a separate rule for choosing those swing pairs. ## Ichimoku settings and trading rules In MT4 and MT5, use Insert → Indicators → Trend → Ichimoku Kinko Hyo. It draws five lines over price and fills the area between Senkou A and B. Colours vary by platform and template, so use the line names rather than colour alone. | Setting | Default | Meaning | |---|---:|---| | Tenkan-sen | 9 | Bars in the conversion-line range | | Kijun-sen | 26 | Bars in the base-line range; also the displacement | | Senkou Span B | 52 | Bars in the slower leading-span range | Chikou has no separate period setting. The Kijun value of 26 also sets its displacement. ### Four rules, including the three-signal alignment | Reading | Long | Short | Periods it actually uses | |---|---|---|---| | Tenkan/Kijun | Tenkan crosses above Kijun | Tenkan crosses below Kijun | Tenkan and Kijun | | Price/Kijun | Close crosses above Kijun | Close crosses below Kijun | Kijun only | | Price/cloud | Close is above the cloud | Close is below the cloud | All three inputs | | Three signals | Tenkan above Kijun, close above cloud, and current close above the close one Kijun period ago | All three conditions reversed | All three inputs | The long state is the bullish three-signal alignment, commonly called *sanyaku koten*; the inverse is *sanyaku gyakuten*. It means all three conditions are true at once, not that three crossovers must happen on the same bar. Each position exits on the opposite signal. Price/cloud opens nothing inside the cloud, and the three-signal rule opens nothing until all three conditions agree. Choose the reading before tuning the inputs. Changing Tenkan or Senkou B cannot affect a price/Kijun cross, and Senkou B cannot affect a Tenkan/Kijun cross. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Window | 2022-01-01 to 2025-12-31, measured as four separate calendar years | | Timeframes | M15, H1 and H4 | | Tenkan | 5, 7, 9, 12, 18 | | Kijun | 20, 22, 26, 30, 36 | | Senkou Span B | 40, 44, 52, 60, 72 | | Settings | 25 Tenkan/Kijun, 5 price/Kijun, 125 price/cloud and 125 three-signal: 280 per timeframe | | Exit | Opposite signal, using the same rule set as entry | | Stops and targets | None in the main sweep; tested separately below | | Spread and fill | 0.3 pips fixed, filled at the close | | Size | 0.1 lot | Every setting was also run on the first and second halves of 2025. All live settings produced both long and short trades. ## Which reading held for four years? All five H1 price/Kijun inputs stayed positive in every year. The denominator in each cell is the number of inputs that can change that reading. | Timeframe | Reading | 2022 | 2023 | 2024 | 2025 | All four | 2025 median | |---|---|---:|---:|---:|---:|---:|---:| | M15 | Tenkan/Kijun | 20/25 | 8/25 | 25/25 | 24/25 | 4/25 | +1,315.9 pips | | M15 | Price/Kijun | 4/5 | 0/5 | 5/5 | 5/5 | 0/5 | +541.8 pips | | M15 | Price/cloud | 125/125 | 103/125 | 125/125 | 125/125 | 103/125 | +1,148.8 pips | | M15 | Three signals | 102/125 | 84/125 | 125/125 | 101/125 | 68/125 | +859.7 pips | | H1 | Tenkan/Kijun | 25/25 | 19/25 | 17/25 | 11/25 | 7/25 | −35.5 pips | | H1 | Price/Kijun | 5/5 | 5/5 | 5/5 | 5/5 | 5/5 | +1,236.6 pips | | H1 | Price/cloud | 120/125 | 57/125 | 125/125 | 120/125 | 54/125 | +437.7 pips | | H1 | Three signals | 125/125 | 7/125 | 110/125 | 87/125 | 5/125 | +259.4 pips | | H4 | Tenkan/Kijun | 15/25 | 2/25 | 24/25 | 3/25 | 0/25 | −984.0 pips | | H4 | Price/Kijun | 4/5 | 0/5 | 4/5 | 4/5 | 0/5 | +255.8 pips | | H4 | Price/cloud | 125/125 | 48/125 | 125/125 | 0/125 | 0/125 | −2,310.6 pips | | H4 | Three signals | 125/125 | 14/125 | 125/125 | 0/125 | 0/125 | −2,136.8 pips | Across all four years, 103 M15 price/cloud settings, 68 M15 three-signal settings and all five H1 price/Kijun inputs stayed positive. H1 three signals fell from 125 positive variants in 2022 to seven in 2023. No H4 setting remained positive in every year. ## Why does the Kijun stay flat? It held still on roughly six bars in ten. Ichimoku lines do not average closes. Their midpoint remains unchanged until the rolling high or low changes enough to move it. | Timeframe | Year | Tenkan unchanged | Kijun unchanged | Price inside cloud | |---|---:|---:|---:|---:| | M15 | 2022 | 35.37% | 60.14% | 14.72% | | M15 | 2023 | 34.57% | 59.74% | 14.82% | | M15 | 2024 | 34.59% | 59.30% | 14.12% | | M15 | 2025 | 33.81% | 58.39% | 14.89% | | H1 | 2022 | 35.98% | 59.74% | 13.14% | | H1 | 2023 | 36.47% | 61.25% | 14.43% | | H1 | 2024 | 35.92% | 60.16% | 13.01% | | H1 | 2025 | 34.92% | 57.65% | 14.25% | | H4 | 2022 | 35.36% | 58.00% | 9.93% | | H4 | 2023 | 36.36% | 60.35% | 10.50% | | H4 | 2024 | 36.88% | 58.54% | 10.89% | | H4 | 2025 | 34.66% | 59.94% | 15.16% | Kijun was unchanged on 57.7% to 61.3% of bars across the 12 timeframe/year cells. The H1 price/Kijun rule won 26.26% of its 2025 trades, with a +67.47-pip average win against a −21.62-pip average loss. ## Is last year's winner next year's? It finished below the next year's median. Each 2024 winner was carried into 2025 on H1. | Reading | Best 2024 input | 2024 | Same input in 2025 | 2025 median | |---|---|---:|---:|---:| | Tenkan/Kijun | 5/22 | +2,954.4 pips | −152.0 pips | −35.5 pips | | Price/Kijun | Kijun 20 | +3,224.8 pips | +1,072.0 pips | +1,236.6 pips | | Price/cloud | 5/30/72 | +2,069.9 pips | −10.3 pips | +437.7 pips | | Three signals | 9/22/72 | +1,736.0 pips | +824.1 pips | +259.4 pips | The winners from the three single-condition readings finished below their 2025 medians. The three-signal winner, 9/22/72, stayed above its median at +824.1 pips but ranked only 24th of 125 in 2025. The prior winner did not remain the winner. ## How did the standard 9/26/52 do? The three-signal rule led on both win rate and net. Standard 9/26/52 on H1 in 2025: | Reading | Trades | Win rate | Average win | Average loss | Net | |---|---:|---:|---:|---:|---:| | Tenkan/Kijun | 270 | 40.37% | +65.67 pips | −44.05 pips | +67.0 pips | | Price/Kijun | 556 | 26.26% | +67.47 pips | −21.62 pips | +985.3 pips | | Price/cloud | 141 | 35.46% | +103.26 pips | −49.23 pips | +683.3 pips | | Three signals | 99 | 43.43% | +103.63 pips | −60.20 pips | +1,085.2 pips | Price/Kijun won only about one trade in four and still made +985.3 pips. Three signals reduced activity to 99 trades and led all four readings in both win rate and net pips. Its 125-setting median was only +259.4 pips, however, so the standard input was stronger than the rule's typical tested input. ## What do ADX and stops add? The effect changed with the reading and the year. The additions below use standard 9/26/52 on H1. | Reading | Addition | 2024 trades and net | 2025 trades and net | |---|---|---:|---:| | Tenkan/Kijun | None | 269, +2,041.0 pips | 270, +67.0 pips | | Tenkan/Kijun | ADX at least 25 | 40, +1,091.3 pips | 39, +753.0 pips | | Tenkan/Kijun | 50-pip stop, 100-pip target | 267, +2,310.8 pips | 260, +2,013.8 pips | | Price/Kijun | None | 483, +3,121.8 pips | 556, +985.3 pips | | Price/Kijun | ADX at least 25 | 52, +251.2 pips | 52, +820.3 pips | | Price/Kijun | 50-pip stop, 100-pip target | 451, +2,303.1 pips | 521, +2,264.6 pips | | Price/cloud | None | 135, +902.4 pips | 141, +683.3 pips | | Price/cloud | ADX at least 25 | 64, −13.9 pips | 65, −996.8 pips | | Price/cloud | 50-pip stop, 100-pip target | 395, +935.4 pips | 355, +1,558.2 pips | | Three signals | None | 105, +536.0 pips | 99, +1,085.2 pips | | Three signals | ADX at least 25 | 53, +447.0 pips | 56, −590.1 pips | | Three signals | 50-pip stop, 100-pip target | 309, +901.8 pips | 292, +1,046.0 pips | ADX improved the 2025 Tenkan/Kijun result but damaged price/cloud in both years and made three signals negative in 2025. The stop and target also reduced the 2025 three-signal result by 39.2 pips. Neither addition had one effect across every rule and year. ## Which reading is most cost-sensitive? The Tenkan/Kijun cross. Only the spread changes here, using H1 in 2025 with standard settings. | Reading | Trades | 0 pips | 0.3 pips | 0.6 pips | Break-even spread | |---|---:|---:|---:|---:|---:| | Tenkan/Kijun | 270 | +148.0 | +67.0 | −14.0 | 0.55 pips | | Price/Kijun | 556 | +1,152.0 | +985.3 | +818.5 | 2.07 pips | | Price/cloud | 141 | +725.5 | +683.3 | +641.0 | 5.15 pips | | Three signals | 99 | +1,114.8 | +1,085.2 | +1,055.5 | 11.26 pips | The Tenkan/Kijun cross turned negative at a 0.6-pip spread. Standard periods alone do not provide enough margin to ignore the actual cost of the pair and trading session. ## Related tests - [Moving-average crossover settings](/blog/moving-average-cross-settings): the same two-line crossover condition built from close averages - [Alligator settings and forward shifts](/blog/alligator-settings): another multi-line indicator that moves values along the time axis - [Parabolic SAR settings](/blog/parabolic-sar-settings): another trend rule where win and loss size mattered more than win rate - [Do Indicator Settings Transfer Between Pairs? 3,024 Tests](/blog/settings-across-pairs): this setting won the 15-minute chart on USD/JPY in both 2024 and 2025, and ranked 34th of 42 on EUR/USD in both ## Notes The figures use Formiq's backtester on USD/JPY from 2022 through 2025, with close fills and a fixed 0.3-pip spread. History starts in December 2021 so the 52-bar calculation and 26-bar displacement are ready before the first measurement window. Chikou comparisons were expressed as current close against the close 26 bars ago to avoid look-ahead bias. --- ### Perfect Order Moving Averages: 2,400 Settings Tested URL: https://formiq.jp/blog/perfect-order-settings Language: en Published: 2026-09-02 Updated: 2026-09-02 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: perfect order, moving averages, settings, trend, backtest, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/perfect-order-settings Perfect order uses three moving averages with different lookback lengths. A bullish order has the short average above the middle average and the middle above the long average. A bearish order reverses the stack. Because every line is a moving average, the state completes after price has already moved. Across 40 period sets and five moving-average types, **157 of 200 settings were profitable in both 2024 and 2025 on M15, against 87 on H1 and 10 on H4**. The four-hour chart went from 184 profitable settings in 2024 to 15 in 2025. Longer bars did not make the rule more stable. ## What perfect order measures There is no separate perfect-order formula. The bullish state is: `short moving average > middle moving average > long moving average` Reverse both inequalities for the bearish state. If either pair falls out of order, the state has ended. The default 10/25/75 SMA stack was present on 62.84% to 63.55% of M15 bars in 2024 and 2025. The hourly share was 63.12% to 63.46%. A condition present for most of the chart is not a selective entry filter, so the test enters only on the first close that completes the stack and exits on the first close that breaks it. ## Perfect order settings and signals The three lines are drawn over the candles in the main chart. Perfect order has no oscillator pane or horizontal levels. In a bullish stack, the short, middle and long lines run from top to bottom. A bearish stack reverses them. Formiq exposes four settings: | Setting | Default | What it changes | |---|---:|---| | Short | 10 | The fastest moving average | | Middle | 25 | The moving average between the fast and slow lines | | Long | 75 | The slowest moving average and broadest trend measure | | Type | SMA | The calculation: SMA, EMA, WMA, SMMA or HMA | The backtest rule matches the chart reading directly. | State | Action | |---|---| | Short > middle > long first completes | Buy | | Short < middle < long first completes | Sell | | Either relationship breaks against the open position | Exit | Shorter periods complete and break more often, raising trade count. Changing the moving-average type changes the stack even when the three periods stay fixed. The periods also need to remain in ascending order; otherwise the names short, middle and long no longer match their calculation. ## How this was measured | Item | Test condition | |---|---| | Pair | USD/JPY | | Timeframes | M15, H1 and H4 | | Periods | Short 5, 10, 20, 25; middle 20, 25, 50, 75; long 50, 75, 100, 200; filtered to 40 ascending triples | | Moving averages | SMA, EMA, WMA, SMMA and HMA | | Windows | 2024, 2025, first half of 2025, second half of 2025 | | Entry | Close of the first bar with all three averages ordered | | Exit | Close of the first bar that breaks the position's order | | Cost | 0.3-pip spread, no slippage | | Size | 0.1 lot | Each window contains 200 settings. Three timeframes and four windows produce 2,400 backtests. Net pips were recalculated from the actual fills. ## Does it work on four-hour bars? M15 kept 157 settings in both years; H4 kept 10. | Timeframe | Profitable in 2024 | Profitable in 2025 | Profitable in both | |---|---:|---:|---:| | M15 | 182/200 | 165/200 | 157/200 | | H1 | 193/200 | 91/200 | 87/200 | | H4 | 184/200 | 15/200 | 10/200 | M15 retained the most settings, with a median 2025 result of +646.8 pips. The hourly median fell from +1,407.2 pips in 2024 to -36.4 in 2025. H4 reversed from +1,588.0 to -812.0 pips. Lower trade count did not protect the slower chart from a change in year. The median H4 setting made 29 to 48 trades per year, too few observations to smooth out that reversal. ## SMA or EMA: which is better? Neither: EMA led on M15 while HMA led on H1. The table counts period sets that were profitable in both years. Each moving-average type has 40 sets. | Moving average | M15 | H1 | H4 | |---|---:|---:|---:| | SMA | 37/40 | 12/40 | 0/40 | | EMA | 40/40 | 13/40 | 0/40 | | WMA | 34/40 | 26/40 | 2/40 | | SMMA | 32/40 | 0/40 | 5/40 | | HMA | 14/40 | 36/40 | 3/40 | EMA kept all 40 M15 settings, while HMA kept 36 of 40 on H1. Neither result transfers cleanly to H4, where every type kept five settings or fewer. The period results have the same limitation. A long period of 100 retained 53 of 65 M15 settings and 28 of 65 H1 settings, but only 5 of 65 on H4. The number 100 did not carry the same result from one timeframe to another. ## What periods should you use? The default 10/25/75 SMA survived on M15 and H1. | Timeframe | Year | Trades | Win rate | Net pips | |---|---:|---:|---:|---:| | M15 | 2024 | 728 | 38.74% | +2,622.5 | | M15 | 2025 | 749 | 35.51% | +646.8 | | H1 | 2024 | 182 | 34.07% | +780.2 | | H1 | 2025 | 186 | 40.86% | +770.0 | | H4 | 2024 | 43 | 46.51% | +1,714.5 | | H4 | 2025 | 49 | 32.65% | -904.3 | The default was profitable in both years on M15 and H1. H4 reversed from +1,714.5 pips in 2024 to -904.3 in 2025. Splitting 2025 also exposes movement hidden by the annual totals: H1 made -146.7 pips in the first half and +892.6 in the second, while H4 moved from +389.1 to -1,286.4. ## Does a one-year winner hold? Each table selects the best setting in one year and carries it unchanged into the other year. ### Select in 2024, test in 2025 | Timeframe | Setting | 2024 | 2025 | 2025 median | |---|---|---:|---:|---:| | M15 | 10/50/100 WMA | +2,866.4 | +880.9 | +646.8 | | H1 | 10/20/200 WMA | +3,215.1 | +319.1 | -36.4 | | H4 | 10/20/75 EMA | +2,861.7 | -1,043.4 | -812.0 | ### Select in 2025, test in 2024 | Timeframe | Setting | 2025 | 2024 | 2024 median | |---|---|---:|---:|---:| | M15 | 20/25/50 WMA | +2,267.6 | +1,229.3 | +1,272.2 | | H1 | 5/25/200 HMA | +2,164.5 | +130.9 | +1,407.2 | | H4 | 10/20/50 HMA | +1,053.1 | -242.0 | +1,588.0 | The 2024 M15 winner stayed above the 2025 median. The reverse selection missed the 2024 median by 42.9 pips. The 2025 H1 winner made +130.9 pips in 2024, far below that year's +1,407.2 median. Both H4 winners lost money in the other year. ## Why a 34% win rate still profited The default H1 setting won 34.07% of its 182 trades in 2024 and made +780.2 pips. Its average winner was +91.88 pips and its average loss was -40.97. In 2025, a 40.86% win rate, +69.78 average winner and -41.21 average loss produced +770.0 pips. Perfect order takes small losses when a stack breaks and depends on the less frequent runs that stay ordered. Optimizing win rate alone can cut those long trades. Trade count, average win, average loss and annual net need to be read together. ## Do ADX and fixed exits help? Adding ADX 25 to the default produced +1,826.7 and -117.6 pips on M15, +935.0 and -295.1 on H1, and +2,079.1 and -2,433.7 on H4 for 2024 and 2025 respectively. Reducing the number of entries did not improve both years. A 50-pip stop with a 100-pip target raised M15 to +2,807.5 pips in 2024 and +781.2 in 2025. On H4, it reduced 2024 to +1,186.8 while lifting 2025 to +120.2. The same fixed exit did not have one effect across timeframes. The difference between zero spread and 0.3 pip matched trade count multiplied by 0.3. In 2025, the default moved from +871.5 to +646.8 pips on M15, +825.7 to +770.0 on H1, and -889.7 to -904.3 on H4. The H4 loss remained without spread, so transaction cost was not its cause. ## Related articles [Moving-average crosses](/blog/moving-average-cross-settings) test the relationship between two lines; perfect order requires all three lines to be stacked. [SMA versus EMA](/blog/sma-vs-ema) compares a change in calculation with a change in period. [GMMA](/blog/gmma-settings) tests whether a 12-line ribbon adds information when lines are removed. [The Alligator](/blog/alligator-settings) also uses three ordered averages, but displaces them forward on the chart. ## Notes - The sample is USD/JPY in 2024 and 2025. Only 2025 was also split into halves. - Signals and fills use the close, with a fixed 0.3-pip spread. - Up to 200 earlier bars are loaded before each window so the long average exists when the test starts. - Pullback entries during an existing order, moving-average slope and spacing between the lines were not tested. --- ### MA Disparity Mean Reversion: 4,320 Settings Tested URL: https://formiq.jp/blog/ma-disparity-settings Language: en Published: 2026-08-29 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: MA Disparity, oscillator, mean reversion, indicator, backtest, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/ma-disparity-settings The disparity index measures how far the close sits from its own moving average, as a percentage of that average. There is one line of arithmetic behind it: **disparity = (close − moving average) / moving average × 100** Some platforms plot `close / average × 100` instead, which is the same series with 100 added to it. A reading of +5 here is a reading of 105 there. The level quoted with it is almost always the same one: **a 25-period average, and 5% as the point where price has gone too far and should come back.** That number comes from daily equity charts. So the first thing worth measuring is whether it exists here. **On USD/JPY across 2024 and 2025, on 15-minute, hourly and 4-hour bars, not one bar printed a 25-period disparity of 5%.** ## What the MA disparity actually measures The numerator is the distance from the average. The denominator is the price level at the time. Zero means the close is sitting on the line. That is what separates it from RSI or Stochastics. Those are squeezed into a 0-to-100 frame, so 70 means the same thing on every instrument. This one is not squeezed into anything: **it is a fraction of the current price, so the range it reaches depends on the market and on the timeframe.** Measured on a 25-period simple moving average: | Timeframe | Median | Nine bars in ten | 2024 max | 2025 max | |---|---|---|---|---| | 15-minute | 0.09% | 0.27% | 1.94% | 1.94% | | Hourly | 0.19% | 0.56% | 2.78% | 1.67% | | 4-hour | 0.41% | 1.05% | 4.54% | 2.35% | Median and the nine-in-ten figure are 2025. **The same "5%" is an extreme a 4-hour chart did not reach in two years, and 2.6 times the largest reading a 15-minute chart produced at all.** ## Does ±5% ever arrive on USD/JPY? Lengthen the period and 5% does appear. Over both years, three settings out of the thirty measured got there: | Timeframe | Period | Year | Bars past 5% | Largest reading that year | |---|---:|---:|---:|---:| | 4-hour | 200 | 2024 | 4.02% | 9.60% | | 4-hour | 75 | 2024 | 0.39% | 7.08% | | Hourly | 200 | 2024 | 0.22% | 6.25% | All three are 2024. USD/JPY rose 1,632 pips that year with a 2,237-pip range; 2025 fell 56 pips with a 1,900-pip range. **In these two years, only periods 75 and above in 2024 crossed 5%.** At the period the level is actually published with, 25, even 3% was reached in only one of six cells: 4-hour bars in 2024, on 0.38% of them. ## Is percent of price a stable unit? It is not. Two measurements say what that unit costs in practice. ### One percent is a different number of pips every month USD/JPY traded between 139.723 and 161.940 in 2024. **The same 1% gap is 139.7 pips at the low end of that year and 161.9 pips at the high end.** In 2025 the range was 139.923 to 158.663, so 1% ran from 139.9 to 158.7 pips. ### The same level fires two to three and a half times as often in one quarter as another This matters more. A percent threshold and an ATR threshold were each calibrated on the whole year **to fire on exactly 20% of its bars**, then read quarter by quarter. | Timeframe | Year | Percent: quietest quarter → busiest | ATR: quietest quarter → busiest | |---|---:|---|---| | 15-minute | 2024 | 11.4% → 32.4% (2.83x) | 18.0% → 21.9% (1.22x) | | 15-minute | 2025 | 12.5% → 28.6% (2.30x) | 18.0% → 21.8% (1.22x) | | Hourly | 2024 | 12.3% → 32.8% (2.66x) | 17.2% → 21.6% (1.26x) | | Hourly | 2025 | 12.8% → 31.3% (2.44x) | 16.2% → 22.6% (1.40x) | | 4-hour | 2024 | 9.7% → 34.5% (3.56x) | 18.5% → 20.9% (1.13x) | | 4-hour | 2025 | 14.9% → 30.5% (2.05x) | 16.3% → 22.7% (1.39x) | **The percent unit swung more in all six cells.** Its quietest quarter came in under 10% and its busiest went over 34%. The ATR unit stayed between 16 and 23 everywhere. ### The steadier unit did not make more money The trading results break that expectation: if percent is the wrong unit, ATR should trade better. Across 90 cells (the same reading, period, timeframe and year, with each unit allowed **its own best level**), ATR finished ahead in 26. Comparing medians rather than bests, ATR won 33 of 90. The reason is trade count. Holding the fire rate steady means firing in quiet markets too. On 15-minute bars the long fade took a median of 29 trades a year in percent and 161 in ATR. As the cost section below shows, cost is trade count times spread, so **steadying the rate is also multiplying the bill.** ## Does it return to zero because price returns? The average catches up, not the price. "It always comes back to zero" is true. What comes back is the question. Every bar that pushed the reading past a level was followed forward to the bar the sign flipped, and that round trip was split into **the part price moved and the part the average moved**. The two shares sum to 100%. | Timeframe | Year | Price's share | Average's share | Bars to zero | |---|---:|---:|---:|---:| | 15-minute | 2024 | −1.6% | 101.6% | 20.2 | | 15-minute | 2025 | −0.7% | 100.7% | 19.5 | | Hourly | 2024 | −26.8% | 126.8% | 20.1 | | Hourly | 2025 | −4.0% | 104.0% | 16.7 | | 4-hour | 2024 | −24.1% | 124.1% | 20.6 | | 4-hour | 2025 | +9.2% | 90.8% | 17.3 | Period 25, level 0.25% (0.5% on 4-hour bars). Across all 18 cells, which add periods 10 and 75, **the price share was negative in 13.** The average's share ran from 61.6% to 139.6%. Read plainly: **price does not have to move at all for the reading to return to zero, because the average is moving towards price the whole time.** The "reversion" the indicator is named for is mostly arithmetic. Counting episodes instead of averaging them changes the picture. Price did end up back in the trade's favour in 52.0% to 78.1% of round trips. **More than half come back; the mean is still negative because the ones that do not come back travel much further.** The price series therefore shows a high return rate but a negative mean outcome before any trading rule is applied. ### Shuffling the bars changes almost nothing Each close-to-close change was kept exactly once and put in a random order, so the volatility, the drift and the fat tails all belong to the market and only the sequence does not. Period 25, 2025: | Timeframe | Nine-in-ten (real / shuffled) | Share past 0.25% | Bars to zero | |---|---|---|---| | 15-minute | 0.269% / 0.258% | 11.9% / 10.9% | 19.5 / 20.7 | | Hourly | 0.558% / 0.521% | 39.6% / 39.5% | 16.7 / 17.1 | | 4-hour | 1.051% / 0.998% | 69.5% / 67.3% | 15.2 / 12.7 | **Nearly identical.** The round trip is a property of subtracting a trailing mean of closes from a close, not a property of this market. One column does separate them. The 99th percentile was 0.563%, 1.129% and 1.997% on real bars against 0.462%, 0.899% and 1.573% on shuffled ones. **The extremes belong to the market; the round trip does not.** ## What is the zero-line cross? It is a moving average crossover. The denominator is a price and prices are positive, so **the sign of the reading is the sign of (close − average)**. Zero disagreements over three timeframes and five periods, and zero again between the percent and ATR versions of the same reading. The backtest says the same thing. Running the zero cross against a crossover of the close (a 1-period simple average) and the same N-period average: | Period | Trades under both trading rules | Annual net under both trading rules | |---:|---:|---:| | 5 | 1,688 | +1,323.4 pips | | 10 | 1,053 | +2,007.6 pips | | 25 | 600 | +2,227.8 pips | | 75 | 335 | +787.0 pips | | 200 | 268 | −1,966.8 pips | Hourly bars, 2025. Swapping the simple average for an exponential, weighted, smoothed or Hull average produced the same identity in all five cases. **Reading the zero line is reading a moving average crossover.** Adding the disparity has not added a condition. One cell out of sixty did diverge between the two units: 4-hour bars, 2024, period 200. The ATR basis divides by an average true range over the same window and Wilder's smoothing needs its own window before that settles, so it is given twice the warmup and starts later. The sign never changed; the line simply began on a different bar. ## Adding it to a chart Stated as facts: - **MT4 / MT5** ship no indicator by this name. What they do ship is **Envelopes** (Insert > Indicators > Trend > Envelopes), which draws lines at the average times (1 ± d/100): the same calculation, drawn on the price pane instead of underneath it. For the oscillator form, install a distributed .mq4 or .mq5 file. - **TradingView** has community scripts published as "Disparity Index". It is not in the built-in list. - **Formiq (browser)** has it in the indicator list. Period, guide lines and the unit (percent or multiples of ATR) are settings, and the backtest side accepts four readings: level, level fade, break out of ±level, and the zero cross. ### The envelope reads wicks; the disparity reads closes The lower envelope band sits at `average × (1 − d/100)`, so **touching the lower band and printing a reading below −d% are the same condition**, differing only in which price is judged. The envelope looks at the bar's low and high; the disparity looks at its close. Entering at the lower line and exiting at the upper one, period 25: | Signal price | Year | Timeframe | Level | Side | Trades | Annual net | |---|---:|---|---:|---|---:|---:| | Envelope (wick) | 2025 | Hourly | 0.5% | Long | 40 | +26.8 pips | | Disparity (close) | 2025 | Hourly | 0.5% | Long | 31 | +769.9 pips | | Envelope (wick) | 2024 | Hourly | 0.5% | Long | 46 | +873.0 pips | | Disparity (close) | 2024 | Hourly | 0.5% | Long | 29 | +169.5 pips | Over 36 cells (three timeframes, two years, three levels, both directions) **the close never traded more often than the wick: 36 of 36.** It finished ahead in 20 of 36. **Judging on closes reliably cuts the trade count and does not reliably improve the result.** ## How this was tested | Item | Value | |---|---| | Pair | USD/JPY | | Timeframes | 15-minute, hourly, 4-hour | | Periods | 5, 10, 25, 75, 200 | | Levels (percent) | 0.15, 0.3, 0.5, 0.8, 1.2 | | Levels (ATR) | 1.0, 1.5, 2.0, 2.5, 3.0 | | Exit levels (percent) | 0 (back to the average), 0.15, 0.3 | | Exit levels (ATR) | 0, 0.75, 1.5 | | Averages | simple, exponential, weighted, smoothed, Hull | | Windows | 2024, 2025, first half 2025, second half 2025 | | Spread | 0.3 pips | | Lot | 0.1 | | Fills | at the close | | Stops and targets | none in the sweep, measured separately | | Combinations | 4,320, plus 360 for the averaging grid | Three readings were run: 1. **Fade**: past the level, take the other side, and give it back when the reading returns to the average. This opens buys or sells but never both, so the long fade and short fade were swept as separate trading rules. 2. **Break**: up through +level opens a long, down through −level a short. Symmetric, so entry and exit share one rule set. 3. **Zero cross**: through zero one way opens, through it the other way reverses. ## How do the three readings differ? The fade won around 65% of its trades, the zero cross 24%, and only the hourly zero cross held. 2025, measured in percent: | Reading | Timeframe | Profitable | Median | Mean win rate | Median trades | |---|---|---|---|---|---| | Fade, long | 15-minute | 35/75 | +0.0 pips | 65.2% | 29 | | Fade, long | Hourly | 42/75 | +110.5 pips | " | 30 | | Fade, long | 4-hour | 60/75 | +466.2 pips | " | 20 | | Fade, short | 15-minute | 31/75 | −76.5 pips | 66.4% | 23 | | Fade, short | Hourly | 53/75 | +201.4 pips | " | 27 | | Fade, short | 4-hour | 63/75 | +718.9 pips | " | 21 | | Break | 15-minute | 15/25 | +300.0 pips | 36.7% | 37 | | Break | Hourly | 9/25 | −749.4 pips | " | 42 | | Break | 4-hour | 8/25 | −851.6 pips | " | 30 | | Zero cross | 15-minute | 2/5 | −641.6 pips | 23.8% | 2,704 | | Zero cross | Hourly | 4/5 | +1,323.4 pips | " | 600 | | Zero cross | 4-hour | 1/5 | −918.6 pips | " | 149 | Mean win rate pools the three timeframes. **The fade wins about two trades in three and finishes ahead in most cells; the zero cross wins one in four and only survives on hourly bars.** Counting settings that were profitable in *both* years cuts that down hard: 19, 13 and 7 of 75 for the long fade, and 11, 17 and 11 for the short. ## Does last year's best still work? Only one of the four transferred. For each row, the setting with the largest 2025 annual net was applied unchanged to 2024. | Reading | Timeframe | Period | Entry level | Exit level | 2025 trades | 2025 annual net | 2024 annual net | |---|---|---:|---:|---:|---:|---:|---:| | Fade, long | 15-minute | 200 | 0.8% | 0.15% | 34 | +1,368.6 pips | +118.9 pips | | Fade, long | Hourly | 200 | 0.5% | 0% | 32 | +1,327.8 pips | −511.2 pips | | Fade, long | 4-hour | 75 | 0.5% | 0.3% | 18 | +1,373.6 pips | −394.8 pips | | Fade, short | 4-hour | 25 | 0.8% | 0.3% | 21 | +1,419.2 pips | −1,840.1 pips | Only the 15-minute long setting remained profitable in 2024. In the other direction, the 15-minute long fade at period 5 with 0.15% → 0.3% made +2,155.5 pips in 2024 and +255.6 pips in 2025. The median across the same 75 settings in 2025 was 0.0 pips. ## Is the 65% win rate real? On 4-hour bars in 2024, not one of 72 settings kept an edge once the drift came out. The fade only ever opens one side. **Hold a long for b bars in a year that rises and it collects the rise, whatever opened it.** So the drift an unconditional position of the same length would have earned (the window's total move divided by its bar count, times the bars held) was subtracted from every setting. | Timeframe | Year | Side | Pips per trade | Drift | Indicator edge | Settings with a positive edge | |---|---:|---|---:|---:|---:|---:| | 4-hour | 2025 | Long | +29.20 | −1.24 | +30.44 | 60/69 | | 4-hour | 2025 | Short | +40.61 | +1.40 | +39.21 | 60/70 | | 4-hour | 2024 | Long | −22.54 | +46.07 | −68.61 | **0/72** | | Hourly | 2024 | Long | −10.91 | +21.22 | −32.13 | 5/72 | | Hourly | 2025 | Long | +10.70 | −0.74 | +11.44 | 42/63 | **On 4-hour bars in 2024, not one of 72 settings had a positive edge left after the drift came out.** The ATR version was 0 of 57. 2025 drifted 56 pips and the fade's edge remained positive in 60 of 69 four-hour long settings. In 2024, after a 1,632-pip rise, it was positive in 0 of 72. **The 2025 fade result did not carry into 2024.** The short side is the mirror. The 4-hour short in 2024 lost 74.08 pips a trade, of which 76.04 was the drift, leaving an edge of +1.96. **Losing money selling into a rising year is not the indicator's doing.** ## Which average to measure from The reading is a distance from a moving average, so which average is a real setting. Five were run with everything else held. As lines first: the share of bars on which each disagrees with the simple average about which side of zero price is on, period 25, 2025. | Average | 15-minute | Hourly | 4-hour | |---|---|---|---| | Exponential | 7.0% | 6.5% | 5.3% | | Weighted | 8.2% | 8.0% | 8.3% | | Smoothed | 15.8% | 15.6% | 12.0% | | Hull | 36.8% | 37.1% | 37.9% | **The sign of the Hull-based disparity disagreed with the simple-average disparity on 36.8% to 37.9% of bars.** This is not a cosmetic choice. Across 72 cells (four readings, three timeframes, two years and three periods), the average with the largest annual net was Hull in 24 cells, smoothed in 14, weighted in 13, simple in 11 and exponential in 10. **Every method produced the largest result in at least ten cells, so the simple average was not consistently superior.** ## Filters, stops and the spread ### Cost is trade count times spread No article in this series has found an exception, and there is none here: the two columns agree to the tenth of a pip both are rounded to. | Trading rule | Period | Level | Timeframe | Year | Trades | Annual net at spread 0 | Annual net at spread 0.3 | Difference | |---|---:|---:|---|---:|---:|---:|---:|---:| | Zero cross | 25 | 0 | 15-minute | 2025 | 2,704 | +169.6 pips | −641.6 pips | 811.2 pips | | Zero cross | 25 | 0 | Hourly | 2025 | 600 | +2,407.7 pips | +2,227.8 pips | 179.9 pips | | Break | 25 | ±0.5% | Hourly | 2025 | 62 | −1,662.1 pips | −1,680.6 pips | 18.5 pips | **The 15-minute zero cross makes +169.6 pips at zero spread.** Its break-even spread is 0.063 pips, so transaction cost turns the rule negative. The same setting on hourly bars breaks even at 4.013 pips. ### What the filters did Hourly, period 25, zero cross: | Condition | Year | Trades | Annual net | |---|---:|---:|---:| | Baseline | 2025 | 600 | +2,227.8 pips | | Baseline | 2024 | 601 | +2,546.5 pips | | ADX ≥ 25 | 2025 | 176 | +554.8 pips | | ADX ≥ 25 | 2024 | 161 | +24.9 pips | | London + NY | 2025 | 328 | +1,793.1 pips | | London + NY | 2024 | 353 | +892.6 pips | | Tokyo | 2025 | 247 | +564.3 pips | | Tokyo | 2024 | 210 | +2,427.2 pips | | Stop 40 pips | 2025 | 558 | +2,461.7 pips | | Stop 40 pips | 2024 | 559 | +3,612.6 pips | | Target 40 pips | 2025 | 599 | +2,403.1 pips | | Target 40 pips | 2024 | 598 | +1,223.9 pips | **No filter improved both years.** ADX reduced annual net in both. London and New York produced the larger session result in 2025 (+1,793.1 versus +564.3 for Tokyo), while Tokyo produced the larger result in 2024 (+2,427.2 versus +892.6). **A 40-pip stop improved both.** On 15-minute bars a 20-pip stop lifts −641.6 pips to −108.6 (still a loss). The zero cross wins a quarter of its trades with an average win more than three times its average loss, so a stop cuts very few winners. ## Notes - **One pair.** The unit is a fraction of price, so the 0.27% / 0.56% / 1.05% figures are USD/JPY's, for these two years, and do not transfer to a pair trading at a different level. - **The shuffled control is a close-only series.** Each bar's high and low collapse onto its close, so it cannot be used for anything that reads wicks, including the envelope comparison above. - **Fills are at the close with no slippage.** A setting that trades 2,700 times a year on 15-minute bars will execute worse than this in practice. ## Related tests - [Bollinger band settings](/blog/bollinger-band-settings): the same fade-or-follow pair of readings, with the band set by standard deviation - [Moving average cross settings](/blog/moving-average-cross-settings): what the zero line here turned out to be - [CCI settings](/blog/cci-settings): distance from an average, divided by the average distance - [Psychological line settings](/blog/psychological-line-settings), where subtracting the drift was first used in this series --- ### Is Trading Renko Reversals Profitable? 1,080 Settings Tested URL: https://formiq.jp/blog/renko-chart-settings Language: en Published: 2026-08-29 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Renko, Brick Charts, Trend Following, Indicator, Backtest, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/renko-chart-settings A Renko chart drops time from the axis. A brick appears when price has moved a fixed distance, and a quiet hour leaves no mark at all. That is what makes the trends look so clean. It has one setting: the brick height. Except that choosing the height does not finish the chart. **The levels snap to multiples of the brick, and where that ladder of multiples starts is a second choice: one every implementation makes silently.** A 20-pip brick has twenty possible grids. Same 20-pip bricks, same rule, same year of USD/JPY 15-minute bars. Only the grid moved. Net pips ran from +221.9 to +3,082.5. Trade counts ran from 1,004 to 1,057 and were essentially unchanged. The same amount of trading, and a 13.9-fold difference in the result. ## How a Renko chart is built Only the close matters. Say the level sits at 148.00 with a 10-pip brick. A close at or above 148.10 stacks a brick upward and moves the level to 148.10. A close at or below 147.90 stacks one down. A bar that closes at 148.05 draws nothing at all. Three consequences follow. ### The level only lands on the grid Levels snap to multiples of the brick. With a 10-pip brick they can only be …147.90 / 148.00 / 148.10… **So where does that ladder start?** Every 10 pips from 148.00, or every 10 pips from 148.03? Both are a "10-pip brick chart", both are correct, and **no charting package I have looked at exposes the choice.** Ten pips means ten possible grids; twenty means twenty. The implementation picks one and never mentions it. ### One bar can complete several bricks If the close is 35 pips from the level, that bar stacks three. The number of bricks does not match the number of bars. On USD/JPY in 2025 with 10-pip bricks, bars completing two or more bricks numbered 1,189 on 15-minute, 1,260 on hourly and 721 on 4-hour data. **By the moment you can see "three bricks in a row", all three may have arrived together.** ### It does forget its start date, and the cost is measured in price Heikin Ashi never forgets where it was started: the [Heikin Ashi article](/blog/heikin-ashi-settings) measures the halving. Renko does forget. The level is an integer address on a grid, so two series that reach the same address agree exactly from that bar onward. How long that takes depends on the brick, because what is really needed is a distance in price. | Timeframe | Brick | Median | 95th percentile | |---|---:|---:|---:| | 15-min | 5 pips | 2 bars | 8 bars | | 15-min | 20 pips | 11 bars | 54 bars | | 15-min | 50 pips | 36 bars | 124 bars | | Hourly | 5 pips | 1 bar | 3 bars | | Hourly | 20 pips | 4 bars | 15 bars | | Hourly | 50 pips | 13.5 bars | 36 bars | | 4-hour | 5 pips | 1 bar | 2 bars | | 4-hour | 20 pips | 2 bars | 5 bars | | 4-hour | 50 pips | 4 bars | 14 bars | Formiq's backtester warms up eight bars per pip of brick (131 bars at 10 pips, 451 at 50) which covers the 95th percentile above. ## Renko settings and how to read them Renko is not a line beside the candles. It **replaces** them, so nothing is added in a pane below. | Environment | Where to switch it on | |---|---| | TradingView | the chart-type menu at the top of the chart | | MT4 / MT5 | not a built-in chart type. It comes from an offline-chart script or a custom indicator that draws the bricks on a normal chart | | In a browser | [pick Renko from Formiq's chart types](/chart) | ### What the settings dialog offers | Item | Default | What it does | |---|---|---| | Brick height | 10 pips in Formiq (1 to 200) | how far price must move to stack one brick | | How the height is set | fixed | TradingView also offers an ATR-derived height, which makes the box size change from period to period | | Price it is built from | the close | closes only, or highs and lows as well. Implementations differ | | Bricks to reverse | 1 in Formiq's backtester (1 to 10) | how much counter-move a turn has to give back before it is drawn. MT4-style charts usually use 2 | | Bricks to wait for | 3 in Formiq's backtester (1 to 50) | how many bricks in one direction before the rule trades | **One item is missing from that table: which multiples of the brick the levels land on.** No environment offers it as a setting; the implementation picks one silently. That is what this article is about, and Formiq's backtester is the only place here that lets you type it, as a grid offset, defaulting to 0 pips. ### What appears on screen Bricks stack vertically and **change colour with direction.** Quiet stretches add no bricks, so a span of clock time can leave a gap with nothing drawn in it. No level lines are added. ### How it is usually traded The most widely published rule is **follow the direction when a brick changes direction.** This article measures that as the brick flip. The other common one is **wait for N bricks in the same direction before trading**, so a single counter-brick does not turn the position. This article measures waits of two to five bricks. The test separates trading the first change in brick direction from waiting for two to five bricks in the same direction. ### Which parameter changes the trade count **A bigger brick trades less and a smaller one trades more.** On 15-minute bars a 5-pip brick took 4,841 trades a year and a 50-pip brick took 252. Requiring two bricks to reverse cuts the count to between 43% and 84% on top of that. One warning about TradingView. **Leaving the chart type on Renko while a strategy runs prices the fills at brick values too.** The rest of this article is a measurement of exactly that difference. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Timeframes | 15-minute, hourly, 4-hour | | Windows | 2025 (main), 2024 (out of sample), first and second half of 2025 | | Brick height | 5, 10, 15, 20, 30, 50 pips | | Bricks to reverse | 1 (draw each brick as it comes), 2 (the traditional rule) | | Grid offset | ten steps from 0 to one brick | | Readings | brick flip; N bricks in one direction (N = 2 to 5) | | Exit | hold to the opposite signal — the exit rule is the entry rule | | Spread | 0.3 pips | | Lot | 0.1 | | Fills | the bar's close, never the brick level | | Combinations | 720 main + 360 grid = 1,080 | Measured with Formiq's backtester. Entering the same settings reproduces the same numbers. **Every fill above comes from the candle, not the brick.** The second half of the article measures why. ## Why the same settings give different results An invisible grid origin moves them: ten grids at one brick height ran +222 to +3,083 pips. Brick height, reading, year and timeframe all held. Only the grid moves: eighteen cells. | Timeframe | Year | Brick | Net across ten grids | Spread | The grid a chart draws | |---|---|---|---|---|---| | 15-min | 2025 | 10 pips | −719.9 to +724.5 | 1,444.4 | −676.5 | | 15-min | 2025 | 20 pips | +221.9 to +3,082.5 | 2,860.6 | +1,381.7 | | Hourly | 2025 | 10 pips | +398.2 to +3,162.3 | 2,764.1 | +2,368.6 | | Hourly | 2025 | 20 pips | +1,310.9 to +3,805.9 | 2,495.0 | +1,310.9 | | 4-hour | 2025 | 30 pips | −906.9 to +853.9 | 1,760.8 | −906.9 | | 4-hour | 2024 | 10 pips | +706.4 to +1,640.6 | 934.2 | +925.2 | Across all eighteen cells the grid spread ran from **934.2 to 2,860.6 pips**. In five of the eighteen, the sign of the year depends on which grid you got. The chart's offset-zero grid produced the smallest annual net in three cells and the largest in one. The 15-minute 10-pip cell in 2025 is the sharpest: one of the ten grids finished profitable, and it was not the one the chart draws. Compare the hidden grid offset with the visible brick-height parameter. | Timeframe | Year | Spread across ten grids | Spread across six brick heights | Ratio | |---|---|---|---|---| | 15-min | 2025 | 1,444.4 to 2,860.6 | 4,471.5 | 0.32 to 0.64 | | 15-min | 2024 | 2,402.8 to 2,782.7 | 1,873.8 | 1.28 to 1.49 | | Hourly | 2025 | 2,112.8 to 2,764.1 | 3,740.0 | 0.56 to 0.74 | | 4-hour | 2024 | 934.2 to 1,875.2 | 3,119.6 | 0.30 to 0.60 | Ratios run 0.30 to 1.49, and **in 3 of 18 cells the grid mattered more than the brick height did.** The setting you cannot reach moves the answer by the same order of magnitude as the setting you can. ## Can you just tune the grid? Not from one year's result. Taking the best offset of 2024 into 2025: | Timeframe | Brick | Offset selected in 2024 | 2024 annual net | 2025 annual net | 2025 median | |---|---:|---:|---:|---:|---:| | 15-min | 10 pips | 0 pips | +1,738.7 pips | −676.5 pips | −497.0 pips | | 15-min | 20 pips | 16 pips | +2,471.1 pips | +1,569.2 pips | +1,818.5 pips | | 15-min | 30 pips | 18 pips | +3,483.2 pips | +1,207.6 pips | +1,723.7 pips | | Hourly | 10 pips | 8 pips | +1,480.0 pips | +398.2 pips | +2,272.9 pips | | Hourly | 20 pips | 14 pips | +2,661.7 pips | +2,038.7 pips | +2,345.9 pips | | Hourly | 30 pips | 18 pips | +1,975.8 pips | +1,789.2 pips | +1,561.4 pips | | 4-hour | 10 pips | 3 pips | +1,640.6 pips | +161.0 pips | +908.0 pips | | 4-hour | 20 pips | 2 pips | +2,387.8 pips | +1,102.3 pips | +1,132.1 pips | | 4-hour | 30 pips | 18 pips | +1,134.3 pips | +91.4 pips | −16.7 pips | Seven of nine offsets selected from 2024 annual net fell below the 2025 median for the same timeframe and brick height. **The grid moves the result substantially, but the prior year's annual net did not select the next year's grid.** There is one practical instruction in this. **A single Renko backtest is a point inside a band roughly a thousand to three thousand pips wide.** Run several grids and keep what survives all of them; a number from one grid is a number from one grid. ## Does Renko really remove time? It does not. The bricks themselves contain none. But **the bricks are built from bar closes**, so which bars they were built from stays in the picture. USD/JPY 2025, the same 10-pip brick throughout. | Timeframe | Bars | Bricks | Bars completing one | Mean move per bar | Brick ÷ move | Bricks reversing | |---|---|---|---|---|---|---| | 15-min | 24,903 | 8,041 | 25.2% | 6.07 pips | 1.65 | 30.5% | | Hourly | 6,226 | 5,270 | 47.6% | 12.20 pips | 0.82 | 22.3% | | 4-hour | 1,610 | 3,298 | 68.9% | 24.67 pips | 0.41 | 14.5% | Identical price action, 2.4 times as many bricks. On 4-hour bars two bars in three complete a brick, so almost no compression is happening; on 15-minute bars three in four draw nothing. And the share of bricks that reverse is set not by the brick height but by **the brick height divided by the typical bar move.** Where that column reads 1.65 / 0.82 / 0.41, the reversal rate reads 30.5 / 22.3 / 14.5%. That is the whole reason a small brick looks like a beautiful trend. Choosing a brick height is choosing a coarseness **relative to a timeframe**, not escaping one. ## Do three bricks mean a trend? Shuffled price changes made runs just as long. The comparison series is the year's bar-to-bar changes **put in a random order**. The size of each move, the spread of the big days and the year's total rise are all the market's; only the order they arrive in is destroyed. The same bricks are then built on top. Across three timeframes and five brick heights, the share of bricks pointing the other way from the one before ran 8.5% to 41.6% on the real year and 8.5% to 43.4% on the shuffles. **The largest disagreement in any cell is 2.4 points.** Mean run length differs by −0.19 to +0.16 bricks, and total brick counts by a factor of 0.96 to 1.014. | Timeframe | Brick | Price changes | Bricks reversing | Mean bricks in one direction | |---|---:|---|---:|---:| | 15-min | 10 pips | real | 30.5% | 3.28 | | 15-min | 10 pips | shuffled | 30.2% | 3.31 | | Hourly | 10 pips | real | 22.3% | 4.48 | | Hourly | 10 pips | shuffled | 22.6% | 4.43 | | 4-hour | 5 pips | real | 8.5% | 11.78 | | 4-hour | 5 pips | shuffled | 8.5% | 11.80 | | 4-hour | 30 pips | real | 27.7% | 3.60 | | 4-hour | 30 pips | shuffled | 27.2% | 3.68 | A series with its order completely destroyed draws the same picture. **The run lengths a Renko chart displays are set by the ratio of brick height to bar move, not by any persistence in price.** The mechanism is in the construction. With the level at 148.00 and price at 148.09, stacking upward needs a close of 148.10 and stacking downward needs 147.90. **Continuation is near, reversal is far.** That asymmetry manufactures trends out of sequence-free noise. ## So is Renko meaningless? The picture survived shuffling, but the results did not. The same rule was backtested on the shuffled series, ten shuffles averaged per cell. | Timeframe | Year | Brick | Trades | Net on real changes | Net on shuffled changes | Difference | |---|---|---:|---:|---:|---:|---:| | 15-min | 2025 | 20 pips | 1,007 | +1,381.7 pips | −959.2 pips | +2,340.9 pips | | 15-min | 2024 | 10 pips | 2,287 | +1,738.7 pips | −1,507.9 pips | +3,246.6 pips | | Hourly | 2025 | 10 pips | 1,176 | +2,368.6 pips | −1,110.9 pips | +3,479.5 pips | | Hourly | 2024 | 30 pips | 373 | +679.9 pips | +170.4 pips | +509.5 pips | | 4-hour | 2025 | 30 pips | 228 | −906.9 pips | +155.9 pips | −1,062.8 pips | | 4-hour | 2024 | 20 pips | 267 | +2,155.8 pips | +85.1 pips | +2,070.7 pips | Across three timeframes, two years and three brick heights (eighteen cells) **the real year finished ahead in 17.** The shuffled series itself finished profitable in 9 of 18, roughly a coin toss. Two sentences carry the section. **The arrangement of the bricks is not information**: shuffling reproduces it. **The trading result is**: what shuffling destroys is how far price travels between one brick and the next. Which is another way of saying that what to read is not how many bricks came in a row, but how many pips arrived between them. ## Why Renko backtests look profitable Repricing at the brick level improved all thirty cells. A brick level is a price something genuinely traded at. That is the difference from Heikin Ashi, whose close is an average of four prices and none of them. It does not help. A brick is only complete when the bar closes, and by then the price has moved past the level. | Timeframe | Brick | Mean overshoot | Median | As a share of the brick | |---|---|---|---|---| | 15-min | 10 pips | 3.76 pips | 3.20 pips | 37.6% | | Hourly | 20 pips | 7.60 pips | 6.55 pips | 38.0% | | 4-hour | 30 pips | 11.80 pips | 10.60 pips | 39.3% | | 4-hour | 50 pips | 18.43 pips | 15.75 pips | 36.9% | Across three timeframes and five brick heights (fifteen cells) the overshoot lands **on the profitable side, without exception.** Bricks stack in the direction price moved and stop short of it, so buying there buys below the market and selling there sells above it. One hundred percent of the time. The same trades, priced both ways. Not one character of the strategy changes. | Timeframe | Year | Brick | Trades | Net at the bar's close | Net at the brick level | |---|---|---:|---:|---:|---:| | 15-min | 2025 | 5 pips | 4,841 | −786.1 pips | +19,815 pips | | 15-min | 2025 | 10 pips | 2,452 | +59.1 pips | +18,040 pips | | Hourly | 2025 | 20 pips | 602 | +1,491.5 pips | +10,500 pips | | 4-hour | 2025 | 30 pips | 228 | −838.6 pips | +4,440 pips | | 4-hour | 2024 | 50 pips | 112 | −799.4 pips | +3,600 pips | Across three timeframes, two years and five brick heights (thirty cells) **all 30 improved, and 4 turned a losing year into a winning one.** The lift per trade ran from 4.26 to 39.28 pips. It has a closed form. The overshoot is collected once at each end of a trade, so **lift per trade = mean overshoot × 2**; measured over eighteen cells the ratio came out between 0.908 and 1.024. That is *cost = trades × spread* with the sign reversed, and it grows with the brick. Heikin Ashi's close is a price that never existed. Renko's is a price that existed and is gone. **Neither is a price you can fill at.** ## What brick size should you use? No single height held: the best one changes with the year and the timeframe. Grid unmoved, brick flip, one brick to reverse. Annual net first. | Timeframe | Year | 5 pips | 10 pips | 15 pips | 20 pips | 30 pips | 50 pips | |---|---|---:|---:|---:|---:|---:|---:| | 15-min | 2025 | −2,238.4 | −676.5 | +1,018.4 | +1,381.7 | +2,233.1 | +333.4 | | 15-min | 2024 | −135.1 | +1,738.7 | +1,062.3 | +668.3 | +1,080.4 | +800.1 | | Hourly | 2025 | −655.7 | +2,368.6 | +3,084.3 | +1,310.9 | +1,698.6 | +57.7 | | Hourly | 2024 | +456.3 | +498.3 | +2,177.7 | +1,026.0 | +679.9 | +130.7 | | 4-hour | 2025 | +759.0 | +1,105.3 | +2,288.0 | +1,356.8 | −906.9 | +284.6 | | 4-hour | 2024 | +1,861.6 | +925.2 | +2,286.6 | +2,155.8 | +391.8 | −833.0 | Trade counts for the same settings. | Timeframe | Year | 5 pips | 10 pips | 15 pips | 20 pips | 30 pips | 50 pips | |---|---|---:|---:|---:|---:|---:|---:| | 15-min | 2025 | 4,841 | 2,452 | 1,526 | 1,007 | 535 | 252 | | 15-min | 2024 | 4,508 | 2,287 | 1,423 | 960 | 572 | 251 | | Hourly | 2025 | 1,918 | 1,176 | 808 | 602 | 370 | 196 | | Hourly | 2024 | 1,741 | 1,117 | 765 | 563 | 373 | 187 | | 4-hour | 2025 | 612 | 481 | 364 | 312 | 228 | 122 | | 4-hour | 2024 | 562 | 442 | 343 | 267 | 188 | 112 | The height with the best annual net was 15 pips on hourly bars in both years, but 30 pips on 15-minute bars in 2025 and 10 pips in 2024. Smaller bricks trade more: 5 pips on 15-minute bars takes 4,841 trades, pays 1,452 pips of spread at 0.3 pips, and finished 2025 at −2,238.4. Widening to 50 pips cuts the year to 112–252 trades, and 4-hour 2024 took 112 of them for −833.0 pips. Formiq defaults to 10 pips. Reading the brick flip with one brick to reverse, 2025 came to −676.5 pips on 15-minute bars, +2,368.6 on hourly and +1,105.3 on 4-hour. The same brick height changed sign and amount with the timeframe. The reversal rule was measured too. One brick draws each brick as it arrives; two bricks is the traditional rule MT4 and the textbooks use, where a turn has to give back two bricks before it is drawn. | Timeframe | Year | Brick | Bricks to reverse | Trades | Win rate | Annual net | |---|---|---:|---:|---:|---:|---:| | 15-min | 2025 | 10 pips | 1 | 2,452 | 35.7% | −676.5 pips | | 15-min | 2025 | 10 pips | 2 | 1,256 | 37.2% | +1,520.4 pips | | 15-min | 2025 | 50 pips | 1 | 252 | 33.7% | +333.4 pips | | 15-min | 2025 | 50 pips | 2 | 113 | 31.9% | −1,821.8 pips | | Hourly | 2025 | 5 pips | 1 | 1,918 | 35.7% | −655.7 pips | | Hourly | 2025 | 5 pips | 2 | 1,338 | 37.6% | +1,524.4 pips | | 4-hour | 2025 | 15 pips | 1 | 364 | 39.0% | +2,288.0 pips | | 4-hour | 2025 | 15 pips | 2 | 266 | 36.8% | −297.4 pips | Across three timeframes, two years and six brick heights (thirty-six cells) **the traditional rule finished ahead in 16.** A dead heat overall, with a clear split inside it: **two bricks to reverse helps small bricks and hurts large ones.** It cuts the trade count to between 43% and 84% across those thirty-six cells, which is relief where there was too much trading and starvation where there was already too little. ## How many bricks should you wait? Two was best in three of four rows, and five lost in all three. The other reading waits for N bricks in one direction before trading. Waiting for one is the plain brick flip, so that column matches what was already measured. Annual net on 20-pip bricks with one brick to reverse. | Timeframe | Year | No wait | 2 bricks | 3 bricks | 4 bricks | 5 bricks | |---|---|---:|---:|---:|---:|---:| | 15-min | 2025 | +1,381.7 | +2,070.3 | +1,287.4 | −1,117.5 | −2,351.8 | | 15-min | 2024 | +668.3 | +1,668.5 | +1,018.7 | −414.6 | −1,837.9 | | Hourly | 2025 | +1,310.9 | +1,728.2 | −12.3 | +252.3 | −1,918.0 | | Hourly | 2024 | +1,026.0 | +187.9 | +443.9 | +557.5 | +718.8 | Trade counts for the same settings. Waiting longer trades less. | Timeframe | Year | No wait | 2 bricks | 3 bricks | 4 bricks | 5 bricks | |---|---|---:|---:|---:|---:|---:| | 15-min | 2025 | 1,007 | 437 | 244 | 162 | 104 | | 15-min | 2024 | 960 | 453 | 255 | 165 | 114 | | Hourly | 2025 | 602 | 310 | 201 | 123 | 89 | | Hourly | 2024 | 563 | 311 | 195 | 129 | 88 | Annual net was highest at a two-brick wait for M15 2025, M15 2024 and H1 2025, then fell to −2,351.8, −1,837.9 and −1,918.0 pips at five bricks. H1 2024 instead increased through the five-brick setting, so no wait count ranked best in all four rows. Overall, waiting cost money. The 2025 median annual net for the brick flip was +1,166.0 pips on 15-minute bars, +1,611.5 on hourly and +786.1 on 4-hour; for the waiting reading it was −106.6, +290.6 and −289.3. Because a single bar can complete several bricks, the wait is counted as "reached N bricks", not "is exactly N bricks". A bar that stacks three at once satisfies a two-brick wait on the way past. ## Does last year's best still work? 2024's three winners took 17 to 30 trades and all lost in 2025. Of the 60 settings per timeframe, the best of one year measured in the other. | Timeframe | Selected in | Bricks waited | Brick | Bricks to reverse | Net that year | Trades that year | Tested on | Net that year | |---|---:|---:|---:|---:|---:|---:|---:|---:| | 15-min | 2025 | 3 | 10 pips | 2 | +2,621.3 pips | 753 | 2024 | +1,427.6 pips | | 15-min | 2024 | 5 | 50 pips | 1 | +3,255.7 pips | 17 | 2025 | −805.8 pips | | Hourly | 2025 | 3 | 5 pips | 1 | +3,306.8 pips | 1,000 | 2024 | +839.0 pips | | Hourly | 2024 | 4 | 50 pips | 2 | +2,866.8 pips | 27 | 2025 | −499.0 pips | | 4-hour | 2025 | none | 15 pips | 1 | +2,288.0 pips | 364 | 2024 | +2,286.6 pips | | 4-hour | 2024 | 5 | 30 pips | 1 | +2,484.9 pips | 30 | 2025 | −790.1 pips | The three settings selected from 2024 took 17, 27 and 30 trades. **All three lost in 2025: −805.8, −499.0 and −790.1 pips.** The large 2024 annual nets came from small trade samples and did not continue into the next year. The four-hour setting selected from 2025 took 364 trades and also made +2,286.6 pips in 2024. Annual net comparisons need the trade count beside them. ## Why a sub-40% win rate profited Wins ran 1.77 times the size of the losses. Averaged over the 2025 settings that took five or more trades. | Reading | Timeframe | Mean win rate | Mean win | Mean loss | |---|---|---:|---:|---:| | Brick flip | 15-min | 36.5% | +57.6 pips | −32.0 pips | | Brick flip | Hourly | 37.2% | +69.0 pips | −38.9 pips | | Brick flip | 4-hour | 38.1% | +85.5 pips | −53.3 pips | | Waiting for N bricks | 15-min | 36.8% | +91.1 pips | −60.7 pips | | Waiting for N bricks | 4-hour | 37.4% | +107.7 pips | −82.7 pips | Every row wins under 40% of its trades. The hourly brick flip is profitable anyway because its +69.0-pip average win is 1.77 times its −38.9-pip average loss. On win rate alone this looks like a losing indicator. ## Filters, stops and cost Built on hourly bars, 20-pip bricks, the brick flip. | Variant | Year | Trades | Annual net | |---|---:|---:|---:| | Baseline | 2025 | 602 | +1,310.9 pips | | Baseline | 2024 | 563 | +1,026.0 pips | | ADX ≥ 20 | 2025 | 395 | +834.1 pips | | ADX ≥ 20 | 2024 | 393 | −29.8 pips | | ADX ≥ 30 | 2025 | 191 | +806.6 pips | | ADX ≥ 30 | 2024 | 184 | +176.0 pips | | London + New York (UTC 7–21) | 2025 | 346 | +636.3 pips | | London + New York (UTC 7–21) | 2024 | 348 | +1,013.0 pips | | Tokyo (UTC 0–8) | 2025 | 227 | +371.1 pips | | Tokyo (UTC 0–8) | 2024 | 197 | −144.5 pips | | SL 50 / TP 100 | 2025 | 536 | +1,006.7 pips | | SL 50 / TP 100 | 2024 | 481 | +1,569.5 pips | | SL 100 / TP 200 | 2025 | 598 | +1,198.0 pips | | SL 100 / TP 200 | 2024 | 544 | +1,667.1 pips | | Time exit, 24 bars | 2025 | 596 | +1,947.3 pips | | Time exit, 24 bars | 2024 | 561 | +1,117.5 pips | ADX lowered the annual net in both years: 2025 fell from +1,310.9 to +834.1 pips at ADX 20 or more, and 2024 from +1,026.0 to −29.8. Restricting the hours did not help either: 2025 came to +636.3 and +371.1 pips, both under the baseline. Stops and targets raised 2024 (+1,026.0 to +1,569.5 pips) and lowered 2025 (+1,310.9 to +1,006.7). The only variant that improved 2025 was the 24-bar time exit, at +1,947.3 pips, or +636.4 over the baseline: though in 2024 it added only +91.5. Cost tracked the trade count exactly. | Spread | Net | Cost vs zero | Trades × spread | |---|---|---|---| | 0 pips | +1,491.5 | — | — | | 0.3 pips | +1,310.9 | 180.6 | 180.6 | | 1.0 pips | +889.5 | 602.0 | 602.0 | | 3.0 pips | −314.5 | 1,806.0 | 1,806.0 | It matches 602 trades times the spread, to the pip. The spread at which the annual net reaches zero was 2.48 pips for the hourly 20-pip flip, 5.87 with two bricks to reverse, 1.67 on 15-minute bars, and 0.24 for the hourly setting that waits three bricks. **That last one loses its profit to a quarter-pip of extra spread.** ## Related reading - [Heikin Ashi settings](/blog/heikin-ashi-settings): the other indicator in this series that rewrites the candles. Its calculated close may never have traded; a Renko close may have traded before the source bar ended. In both cases, filling at the transformed close overstates what could have been executed - [Moving average crossover settings](/blog/moving-average-cross-settings): another trend-following rule, measured with moving-average lines instead of bricks - [Psychological line settings](/blog/psychological-line-settings): another indicator on a ladder, where the number you type is not the rule you get - [GMMA settings](/blog/gmma-settings), where stacking a trend filter on a trend rule made it worse, exactly as it does here - [Fair value gap settings](/blog/fair-value-gap-settings): the other article here where adding something to compare against erased the indicator's own claim ## Notes - The sample covers USD/JPY in 2024 and 2025; it does not include other pairs or periods - **Only ten of the possible grids.** A 20-pip brick has twenty; ten were measured. The spreads quoted are a floor - **The shuffled series throws away the shape of the bar.** It reconnects closes only, so highs and lows carry no information in it. It is a valid comparison only for rules that read closes - **Two reversal rules.** Some implementations require three bricks to turn - **Renko itself differs between implementations.** Building from closes and building from highs and lows are both in use; this test builds from closes - **No execution detail.** Spread is fixed at 0.3 pips, with no widening around releases and no rejected fills --- ### How Long Does RSI Stay Overbought? 3 Years of USD/JPY Tested URL: https://formiq.jp/blog/rsi-overbought-oversold-sticking Language: en Published: 2026-08-29 Updated: 2026-09-02 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: RSI, overbought, oversold, RSI 70/30, mean reversion, backtesting, EUR/USD Also available in: ja — https://formiq.jp/ja/blog/rsi-overbought-oversold-sticking RSI above 70 is called overbought, and RSI below 30 is called oversold. Those labels sound like a reversal is due. In a strong move, however, the line can remain beyond its threshold for several bars while price continues in the same direction. With hourly RSI 14 in 2023–2025, an upper-zone run continued into the next bar **60.3% of the time on USD/JPY and 67.3% on EUR/USD**. The lower-zone rates were 60.3% and 55.8%. Remaining extreme for one more bar was more common than leaving immediately on both pairs. Long runs were less transferable. At least ten upper-zone bars occurred in 19.2% of USD/JPY runs, against 6.3% below 30. The EUR/USD rates were 10.1% and 8.1%. The large upper-versus-lower gap in USD/JPY did not appear on the second pair. Lengths 7, 9, 14, 21, 30, and 50 were also compared. Persistence did not move steadily as the length changed. The repeatable change on both pairs was how rarely a longer RSI reached 70 or 30. ## What RSI overbought and oversold actually mean RSI converts average gains and average losses over the last N bars into a value from 0 to 100. RSI = 100 − 100 ÷ (1 + average gain ÷ average loss) [TradingView's RSI documentation](https://www.tradingview.com/support/solutions/43000502338-relative-strength-index-rsi/) describes readings above 70 as overbought and readings below 30 as oversold. The calculation itself only reports the balance of recent gains and losses. Crossing 70 or 30 does not put a reversal condition into the formula. "Staying overbought" in this test does not mean RSI printed exactly 100. It means the line closed at or above 70 for consecutive bars. Likewise, an oversold run consists of consecutive closes at or below 30. With RSI 14, neither pair printed exactly 100 or 0 on any of the three timeframes during the three years. ## RSI settings and how a run was counted The baseline uses the conventional RSI 14 with 70 and 30 as the two boundaries. | Input | Baseline | Meaning in this test | |---|---:|---| | Length | 14 | Number of bars used to smooth gains and losses | | Overbought | 70 | A close at or above 70 belongs to the upper zone | | Oversold | 30 | A close at or below 30 belongs to the lower zone | | Observation | Bar close | The zone is known only after the bar closes | | Run length | One bar or more | The first bar inside the zone counts as bar one | If RSI reads “68 → 72 → 76 → 71 → 66,” the overbought run lasts three bars. The first 72 is bar one, 76 is bar two, 71 is bar three, and the run ends at 66. A one-bar run leaves the zone immediately. A two-bar run remains there on the next bar. RSI is normally shown in a separate pane below price. Set the length to 14 and add horizontal levels at 70 and 30. The main [RSI settings test](/blog/rsi-settings) covers platform setup, conventional 30/70 fades, and rules that trade a cross of the selected level. This article isolates the duration of each visit beyond 30 or 70. ## How this was measured | Item | Value | |---|---| | Markets | USD/JPY / EUR/USD | | Window | January 1, 2023 to December 31, 2025 | | Timeframes | Fifteen-minute / hourly / four-hour | | USD/JPY bars with RSI 14 defined | 74,800 / 18,701 / 4,835 | | EUR/USD bars with RSI 14 defined | 74,797 / 18,701 / 4,835 | | RSI lengths | 7 / 9 / **14** / 21 / 30 / 50 | | Level pairs | 80/20 / 75/25 / **70/30** / 65/35 / 60/40 | | Run starts | Previous bar outside, current bar first inside the zone | | Run ends | First bar back outside the zone | | Persistence | Share lasting at least 2 / 3 / 5 / 10 bars | | Fade direction | Short the upper zone, buy the lower zone | | Execution | Next bar's open using recorded bid and ask prices | | Pip size | USD/JPY 0.01 yen / EUR/USD 0.0001 dollars | | Holding periods | 1 / 3 / 5 / 10 bars | One continuous visit to a zone counts as one run. Ten consecutive bars above 70 are one observation, not ten. A run that had not left the zone by year-end was excluded because its final length was unknown. The trading comparison separates a fade entered as the run begins from one entered after RSI leaves the zone. Both use the next bar's open, after the signal bar is complete. Every run is treated as an independent event, so summed pips are not a compounded portfolio return and trades can overlap in time. ## Does RSI stay overbought next bar? It did in every sample. The following tables pool 2023 through 2025 for RSI 14 at 70/30. ### USD/JPY | Timeframe | Zone | Runs | 2+ bars | 3+ bars | 5+ bars | 10+ bars | Longest | |---|---|---:|---:|---:|---:|---:|---:| | Fifteen-minute | 70 or higher | 1,154 | 62.3% | 43.2% | 25.4% | 10.2% | 37 bars | | Fifteen-minute | 30 or lower | 1,049 | 58.4% | 38.5% | 20.7% | 7.8% | 43 bars | | Hourly | 70 or higher | 312 | 60.3% | 45.8% | 30.8% | 19.2% | 39 bars | | Hourly | 30 or lower | 300 | 60.3% | 41.0% | 21.7% | 6.3% | 27 bars | | Four-hour | 70 or higher | 103 | 70.9% | 51.5% | 28.2% | 8.7% | 32 bars | | Four-hour | 30 or lower | 53 | 67.9% | 54.7% | 26.4% | 5.7% | 23 bars | ### EUR/USD | Timeframe | Zone | Runs | 2+ bars | 3+ bars | 5+ bars | 10+ bars | Longest | |---|---|---:|---:|---:|---:|---:|---:| | Fifteen-minute | 70 or higher | 1,071 | 56.2% | 38.2% | 20.9% | 6.8% | 36 bars | | Fifteen-minute | 30 or lower | 984 | 57.2% | 38.3% | 20.2% | 6.1% | 31 bars | | Hourly | 70 or higher | 306 | 67.3% | 47.7% | 25.5% | 10.1% | 24 bars | | Hourly | 30 or lower | 283 | 55.8% | 37.1% | 22.6% | 8.1% | 25 bars | | Four-hour | 70 or higher | 82 | 65.9% | 48.8% | 31.7% | 9.8% | 25 bars | | Four-hour | 30 or lower | 76 | 59.2% | 46.1% | 17.1% | 2.6% | 24 bars | Across all twelve pair, timeframe, and zone samples, next-bar persistence ranged from 55.8% to 70.9%. Remaining beyond 30 or 70 for one more bar was always more common than leaving immediately. Four-hour samples are small. The smallest contains 53 USD/JPY lower-zone runs. A few percentage points between pairs therefore should not be read without the event counts. ## How long can RSI stay overbought? Long runs stood out on USD/JPY and much less on EUR/USD. The hourly results below separate each year. | Market | Year | Zone | Runs | 2+ bars | 5+ bars | 10+ bars | Longest | |---|---:|---|---:|---:|---:|---:|---:| | USD/JPY | 2023 | 70 or higher | 105 | 62.9% | 30.5% | 20.0% | 19 bars | | USD/JPY | 2023 | 30 or lower | 88 | 62.5% | 28.4% | 10.2% | 18 bars | | USD/JPY | 2024 | 70 or higher | 122 | 62.3% | 32.8% | 18.0% | 39 bars | | USD/JPY | 2024 | 30 or lower | 92 | 58.7% | 18.5% | 5.4% | 27 bars | | USD/JPY | 2025 | 70 or higher | 85 | 54.1% | 28.2% | 20.0% | 30 bars | | USD/JPY | 2025 | 30 or lower | 120 | 60.0% | 19.2% | 4.2% | 14 bars | | EUR/USD | 2023 | 70 or higher | 98 | 65.3% | 28.6% | 16.3% | 23 bars | | EUR/USD | 2023 | 30 or lower | 95 | 60.0% | 25.3% | 10.5% | 16 bars | | EUR/USD | 2024 | 70 or higher | 87 | 60.9% | 26.4% | 10.3% | 21 bars | | EUR/USD | 2024 | 30 or lower | 116 | 50.9% | 19.0% | 6.0% | 23 bars | | EUR/USD | 2025 | 70 or higher | 121 | 73.6% | 22.3% | 5.0% | 24 bars | | EUR/USD | 2025 | 30 or lower | 72 | 58.3% | 25.0% | 8.3% | 25 bars | On USD/JPY, 18.0% to 20.0% of upper-zone runs lasted at least ten bars in each year. The lower-zone range was 4.2% to 10.2%. Pooled EUR/USD rates were 10.1% and 8.1%, leaving much less difference between the two sides. EUR/USD also shows why next-bar persistence and long persistence are separate quantities. Its upper next-bar rate rose from 60.9% in 2024 to 73.6% in 2025 while its ten-bar rate fell from 10.3% to 5.0%. ## Does 80/20 avoid the problem? It does not: every next-bar rate stayed above 53%. These are hourly RSI 14 runs pooled across all three years. | Market | Levels | Upper runs | Upper 2+ bars | Lower runs | Lower 2+ bars | |---|---:|---:|---:|---:|---:| | USD/JPY | 80 / 20 | 76 | 57.9% | 47 | 55.3% | | USD/JPY | 75 / 25 | 163 | 66.3% | 135 | 53.3% | | USD/JPY | 70 / 30 | 312 | 60.3% | 300 | 60.3% | | USD/JPY | 65 / 35 | 564 | 64.7% | 483 | 63.1% | | USD/JPY | 60 / 40 | 800 | 68.0% | 681 | 64.8% | | EUR/USD | 80 / 20 | 52 | 53.8% | 39 | 53.8% | | EUR/USD | 75 / 25 | 156 | 62.2% | 120 | 59.2% | | EUR/USD | 70 / 30 | 306 | 67.3% | 283 | 55.8% | | EUR/USD | 65 / 35 | 521 | 60.3% | 535 | 59.8% | | EUR/USD | 60 / 40 | 746 | 63.1% | 743 | 65.1% | At 80/20, every next-bar rate remained above 53%. The samples fell to 76 and 47 runs on USD/JPY and 52 and 39 on EUR/USD. A handful of events can move percentages sharply at those sizes. Wider zones naturally produce more runs because values nearer 50 now count as extreme. Comparing the rates without the level width and run count would give the wider settings more meaning than they contain. ## Does RSI length change persistence? It changed the number of events, not their length. The 70/30 lines stay fixed while hourly RSI length changes. Upper and lower runs are combined. | Market | Length | Bars beyond 70/30 | Runs | 2+ bars | 5+ bars | 10+ bars | Longest | |---|---:|---:|---:|---:|---:|---:|---:| | USD/JPY | 7 | 27.91% | 1,557 | 57.7% | 23.2% | 6.6% | 25 bars | | USD/JPY | 9 | 22.23% | 1,149 | 60.1% | 25.2% | 8.4% | 28 bars | | USD/JPY | 14 | 13.31% | 612 | 60.3% | 26.3% | 12.9% | 39 bars | | USD/JPY | 21 | 6.98% | 303 | 64.7% | 28.7% | 13.2% | 30 bars | | USD/JPY | 30 | 3.20% | 132 | 56.8% | 34.1% | 15.9% | 29 bars | | USD/JPY | 50 | 0.66% | 23 | 56.5% | 34.8% | 26.1% | 22 bars | | EUR/USD | 7 | 26.38% | 1,552 | 57.0% | 21.2% | 6.3% | 22 bars | | EUR/USD | 9 | 20.68% | 1,157 | 57.0% | 22.5% | 7.5% | 23 bars | | EUR/USD | 14 | 11.48% | 589 | 61.8% | 24.1% | 9.2% | 25 bars | | EUR/USD | 21 | 5.54% | 267 | 61.8% | 25.5% | 12.4% | 30 bars | | EUR/USD | 30 | 2.10% | 111 | 52.3% | 22.5% | 7.2% | 29 bars | | EUR/USD | 50 | 0.39% | 14 | 64.3% | 21.4% | 21.4% | 22 bars | As length rises from 7 to 50, complete runs fall from 1,557 to 23 on USD/JPY and from 1,552 to 14 on EUR/USD. Smoothing pulls a longer RSI toward 50, so it reaches either outer zone less often. Next-bar persistence ranges from 56.5% to 64.7% on USD/JPY and from 52.3% to 64.3% on EUR/USD without a steady direction. The length-50 samples are too small to interpret the 21.4% to 26.1% ten-bar rates as a stable feature. **Signal frequency, not persistence, is the effect of RSI length that repeated across the two pairs.** ## Does anything survive a change of pair? The five EUR/USD survivors did not. Each market was tested across six lengths, two entry timings, and four holding periods on each timeframe. That produces 48 conditions per timeframe and 144 per market. | Scope | Positive in all three years | Location | |---|---:|---| | USD/JPY | 1/144 | One hourly condition | | EUR/USD | 5/144 | Five four-hour conditions | | Both pairs across all six market-year cells | **0/144** | No shared condition | These are all conditions with positive mean pips per event in each of the three years. | Market | Timeframe | Length | Entry | Hold | 2023 | 2024 | 2025 | |---|---|---:|---|---:|---:|---:|---:| | USD/JPY | Hourly | 50 | After zone ends | 10 bars | 9 / +12.54 | 12 / +16.82 | 2 / +10.75 | | EUR/USD | Four-hour | 7 | After zone ends | 5 bars | 141 / +2.51 | 147 / +0.34 | 143 / +1.44 | | EUR/USD | Four-hour | 9 | As zone begins | 10 bars | 108 / +11.21 | 109 / +6.93 | 103 / +0.36 | | EUR/USD | Four-hour | 9 | After zone ends | 3 bars | 108 / +3.81 | 109 / +1.30 | 103 / +1.44 | | EUR/USD | Four-hour | 9 | After zone ends | 5 bars | 108 / +5.89 | 109 / +0.46 | 103 / +2.69 | | EUR/USD | Four-hour | 14 | As zone begins | 10 bars | 54 / +6.97 | 59 / +0.95 | 44 / +11.01 | Each year cell shows events followed by mean pips per event. Four of the five EUR/USD conditions contain a yearly mean between +0.34 and +0.95 pips. A modest cost or sample change can erase margins of that size. The signs changed when the market changed. The EUR/USD length-7 exit-and-hold-five condition averaged +3.55, −5.69, and +1.02 pips on USD/JPY. The lone USD/JPY length-50 survivor averaged −15.90, +2.60, and −27.66 pips on EUR/USD. A three-year result from one pair did not identify a setting that survived the other. ## Fade at once, or wait for the exit? Neither held for three years. The table combines upper-zone shorts and lower-zone buys on hourly RSI 14. Values are average pips per event using recorded bid and ask prices. | Market | Entry | Hold | 2023 | 2024 | 2025 | |---|---|---:|---:|---:|---:| | USD/JPY | As zone begins | 1 bar | −1.75 | +1.68 | −1.69 | | USD/JPY | As zone begins | 3 bars | −2.17 | +1.01 | −1.67 | | USD/JPY | As zone begins | 5 bars | −1.97 | +0.26 | −2.29 | | USD/JPY | As zone begins | 10 bars | −1.48 | −2.81 | −3.87 | | USD/JPY | After zone ends | 1 bar | −3.02 | −0.51 | −0.96 | | USD/JPY | After zone ends | 3 bars | −2.27 | −0.59 | −2.61 | | USD/JPY | After zone ends | 5 bars | −0.33 | −3.70 | −3.13 | | USD/JPY | After zone ends | 10 bars | −1.89 | −6.14 | +2.00 | | EUR/USD | As zone begins | 1 bar | −1.39 | +0.47 | −1.53 | | EUR/USD | As zone begins | 3 bars | −0.71 | +0.17 | −0.88 | | EUR/USD | As zone begins | 5 bars | −0.96 | +0.73 | −0.49 | | EUR/USD | As zone begins | 10 bars | +2.48 | +3.88 | −2.66 | | EUR/USD | After zone ends | 1 bar | −0.75 | −0.93 | −1.77 | | EUR/USD | After zone ends | 3 bars | +0.86 | −0.81 | −3.63 | | EUR/USD | After zone ends | 5 bars | +2.18 | +1.13 | −2.44 | | EUR/USD | After zone ends | 10 bars | +4.44 | +3.26 | −2.90 | The five-bar event counts were 193, 214, and 205 on USD/JPY, and 193, 203, and 193 on EUR/USD. Waiting for EUR/USD to leave the zone looked favourable in 2023 and 2024, then turned to −2.44 pips in 2025. Across all three timeframes, neither five-bar timing was positive in all three years on either pair. Leaving the zone can be a confirmation event, but it was not a complete trading edge here. The 70 and 30 lines mark a state, not a reversal time. A practical test must separate entering the zone from leaving it and then require the result to survive changes of year and market. None of the 144 settings finished positive in all six USD/JPY and EUR/USD market-year cells. ## Related tests The main [RSI settings test](/blog/rsi-settings) compares 330 combinations of 30/70 fading and 50-line momentum. This follow-up isolates how long each overbought or oversold visit lasts. [Stochastic RSI settings](/blog/stoch-rsi-settings) apply another range calculation to RSI itself. The [Bollinger Bands fade-versus-breakout test](/blog/bollinger-band-settings) asks the same broader question on price: whether an extreme should be faded immediately or followed as a breakout. ## Notes - The sample contains USD/JPY and EUR/USD in 2023–2025. Other markets and regimes can have different run lengths. - RSI uses closing prices. Intrabar touches of 70 or 30 that did not survive to the close are excluded. - Runs already open at the start of a year and runs still open at year-end are excluded because their full duration is unknown within that year's sample. - The 80/20 and RSI 50 samples are small, so their percentages are uncertain. - Each run is treated as an independent event. Trades can overlap, and summed pips are not a portfolio return. - Trend, volatility and session filters, stops, and profit targets were not tested. --- ### Bollinger Bands Period: Is 20 Really Best? 3,024 Settings Tested URL: https://formiq.jp/blog/bollinger-band-period-settings Language: en Published: 2026-08-28 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: bollinger bands, period, lookback, breakout, mean reversion, backtest, USDJPY Also available in: ja — https://formiq.jp/ja/blog/bollinger-band-period-settings The “Period” or “Length” box in Bollinger Bands determines how many bars build both the centre line and its standard deviation. Twenty is the conventional default, but 20 bars on a 15-minute chart cover five hours while 20 four-hour bars cover 80 hours. We tested 12 periods from 5 through 200, seven deviations from 0.5σ through 2σ, and three readings of the bands. The grid ran on 15-minute, hourly and four-hour USD/JPY bars across 2024, 2025 and both halves of 2025: **3,024 backtests** in all. There was no single best bar count. Averaging the breakout across deviations and years, period 50 led M15, period 20 led H1 and period 5 led H4. The counts look unrelated, but they cover **12.5, 20 and 20 hours**. The most useful result is therefore not “always use 20”; it is to compare periods as elapsed market time. ## What the Bollinger Bands period changes For period N and deviation k, the bands are built in three steps: 1. Average the last N closes to produce the centre line 2. Measure the population standard deviation of those same N closes 3. Add and subtract k times that standard deviation A longer period makes the centre line slower and keeps earlier volatility in the band width for longer. The measured effect on hourly USD/JPY in 2025 at 1σ was substantial. | Period | Mean half-width | Closes inside | Annual trades | |---:|---:|---:|---:| | 5 | 12.79 pips | 47.1% | 934 | | 20 | 28.51 pips | 45.0% | 277 | | 50 | 46.54 pips | 45.8% | 103 | | 100 | 65.62 pips | 44.9% | 68 | | 200 | 88.68 pips | 50.3% | 46 | The band became almost seven times wider and trades fell by about 95%. Yet containment stayed between 44.9% and 50.3%. **Increasing period did not steadily increase the percentage of closes inside 1σ**, because both width and the horizon used to estimate the mean changed. ### Short periods can make a deviation unreachable There is a mathematical boundary when the current close is one of the N values used in a population standard deviation. The largest standardized distance any one value can have from that sample's mean is **√(N−1)**. At period 5, the ceiling is √4 = 2σ. Because this test requires a close to be strictly outside the band, period 5 and 2σ cannot trigger. The result was zero trades in all 12 combinations of three timeframes and four windows. Period and deviation cannot be chosen independently. A short enough period paired with a distant enough band can be an impossible setting, not merely an unprofitable one. ## Bollinger Bands period settings and signals MT4 and MT5 place Bollinger Bands under Insert → Indicators → Trend. TradingView and Formiq expose the same core inputs. | Field | Conventional value | What it controls | |---|---:|---| | Period / Length | 20 | Number of prices in the mean and standard deviation | | Shift / Offset | 0 | Horizontal displacement of all three lines | | Deviations / StdDev | 2.000 | Multiple of standard deviation above and below the mean | | Apply to / Source | Close | Price used in the calculation; this test uses closes | | Style / Color | Any | Display only; it does not affect the calculation | A shorter period follows price faster and creates more signals. A longer one smooths the line and tends to lengthen trades. Deviation moves the two outer bands nearer to or farther from the mean while holding the period fixed. This study measured three trading interpretations. | Trading rule | Entry | Exit | |---|---|---| | Fade long | Buy the close when the low touches the lower band | Close when the opposite 1σ band is touched | | Fade short | Sell the close when the high touches the upper band | Close when the opposite 1σ band is touched | | Breakout | Buy a close above the upper band and sell a close below the lower band | Reverse on an opposite close | The earlier deviation study found that 0.5σ through 1.25σ survived changes of year and period, with many 2σ settings surviving as well. This period test gives the seven values from 0.5σ through 2σ equal weight so that a handful of distant-band trades cannot choose the winning lookback. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Windows | Calendar 2025, calendar 2024, and the two halves of 2025 | | Timeframes | 15-minute / one-hour / four-hour | | 2025 bars | 24,903 M15, 6,226 H1, 1,610 H4 | | Periods | 5, 10, 14, 20, 25, 30, 40, 50, 75, 100, 150, 200 | | Deviations | 0.5σ, 0.75σ, 1σ, 1.25σ, 1.5σ, 1.75σ, 2σ | | Grid | 12 periods × 7 deviations × 3 trading rules = 252 settings, on 3 timeframes and 4 windows: 3,024 runs | | Costs | 0.3-pip spread, zero slippage, 0.1 lots | | Fills | Signal-bar close; opposite-side reversals may occur on the exit bar, while same-side re-entry waits for a later bar | | Stops and targets | None in the main grid; added separately below | | Measurement | Formiq's production backtester, with pips re-derived from every fill | Each period is judged by the mean of its seven deviations and by how many of those deviations were profitable in both calendar years. ## What is the best Bollinger period? No single bar count won: it changed with the timeframe. These are the periods with the highest two-year deviation mean for the breakout. | Timeframe | Leading period | Elapsed time | 2025 mean | 2024 mean | Positive both | Mean 2025 trades | |---|---:|---:|---:|---:|---:|---:| | 15-minute | 50 | 12.5 hours | +1,368.1 | +1,656.0 | 6 / 7 | 463.0 | | One-hour | 20 | 20 hours | +1,067.1 | +2,058.3 | 7 / 7 | 245.6 | | Four-hour | 5 | 20 hours | +1,260.0 | +1,470.4 | 6 / 6 | 173.1 | Period 5 has only six active deviations because its 2σ breakout is impossible. Across all 84 settings, 59 M15, 53 H1 and 26 H4 settings were profitable in both years. The three winners do not support a universal period of 20. They support a more limited observation: **the best lookbacks in this grid clustered from half a day to roughly one day**. ## Is 20 the right choice on H1? Period 20 led the hourly two-year mean, with all seven deviations profitable in both years. The hourly breakout provides the largest common sample for comparing all 12 periods. Each row below averages seven deviations. | Period | 2025 mean | 2024 mean | Two-year mean | Mean 2025 trades | Positive both | |---:|---:|---:|---:|---:|---:| | 5 | +1,174.4 | +1,211.6 | +1,193.0 | 692.7 | 6 / 6 | | 10 | +1,560.6 | +665.4 | +1,113.0 | 437.3 | 6 / 7 | | 14 | +1,405.8 | +1,165.9 | +1,285.8 | 329.9 | 6 / 7 | | **20** | **+1,067.1** | **+2,058.3** | **+1,562.7** | **245.6** | **7 / 7** | | 25 | +799.2 | +885.8 | +842.5 | 203.4 | 6 / 7 | | 30 | +1,427.1 | +562.1 | +994.6 | 169.0 | 6 / 7 | | 40 | +1,161.1 | +889.6 | +1,025.3 | 127.6 | 7 / 7 | | 50 | +1,040.7 | +1,103.8 | +1,072.3 | 104.1 | 6 / 7 | | 75 | −494.3 | +1,905.9 | +705.8 | 78.6 | 2 / 7 | | 100 | −1,092.3 | +1,957.2 | +432.5 | 63.3 | 1 / 7 | | 150 | −1,910.7 | +2,109.6 | +99.4 | 47.7 | 0 / 7 | | 200 | −2,022.6 | +1,663.4 | −179.6 | 38.3 | 0 / 7 | Period 20 had both the highest two-year mean and seven of seven deviations positive in both years. Period 40 also retained seven, but its two-year mean was +1,025.3 pips. **If one hourly value has to be the starting point, 20 has the best support in this sample.** Periods 75 and above profited strongly in 2024 but lost in 2025, while averaging only 38 to 79 annual trades. The long period did not simply remove noise; a few extended moves came to dominate its result. ### Where the default 20 and 2σ landed | Year | Trades | Win rate | PF | Net pips | |---|---:|---:|---:|---:| | 2025 | 132 | 43.18% | 1.235 | +1,177.5 | | 2024 | 137 | 41.61% | 1.270 | +1,434.6 | The conventional setting was profitable in both years and above the median, but it did not lead either one. A defensible period does not make its conventional deviation automatically optimal. With period 20 and 1σ, the breakout rule took 277 trades and made +1,052.5 pips in 2025, then 280 and +2,071.7 in 2024. ## Can you pick the best single year? The largest single-year result did not repeat. - Period 10 averaged +1,560.6 pips in 2025 and +665.4 pips in 2024 - Period 150 averaged +2,109.6 pips in 2024 and −1,910.7 pips in 2025 Optimizing period and deviation together was still less portable. | Selection year | Setting | Trades in selection year | Net in selection year | Check year | Net in check year | |---:|---|---:|---:|---:|---:| | 2025 | Period 10, 1.75σ | 271 | +2,753.0 pips | 2024 | +686.0 pips | | 2024 | Period 150, 0.5σ | 53 | +3,164.9 pips | 2025 | −1,772.1 pips | Choosing the single largest historical cell is therefore hard to defend. Period 20 is a better baseline because nearby deviations survived both years, not because its best cell topped the table. ## Does period 20 hold over six months? It does not: only two of its seven deviations were profitable in both halves. Splitting 2025 changes which definition of “best” wins. | Period | First-half mean | Second-half mean | Positive in both halves | |---:|---:|---:|---:| | 10 | +745.8 | +788.9 | 7 / 7 | | 14 | +779.2 | +603.3 | 5 / 7 | | **20** | **−211.0** | **+1,258.5** | **2 / 7** | | 40 | +227.8 | +916.3 | 6 / 7 | | 50 | +596.3 | +448.2 | 6 / 7 | All seven period-20 deviations were positive in both calendar years, yet only two were positive in both halves of 2025. Period 10 retained seven of seven; periods 40 and 50 retained six. **The highest two-year profit and the most stable half-year result are different objectives.** Use 20 as the hourly annual baseline, but put 10, 40 and 50 beside it when shorter-window stability matters. ## Win rate explained one year, not the next The 83 active hourly breakout settings in 2025 averaged 214 trades, from 19 to 1,258. Their mean win rate was 36.3%, mean winner +112.8 pips and mean loser −67.9 pips. Win rate moved in the same direction as net result within 2025. Yet period 10 fell from a +1,560.6-pip mean in 2025 to +665.4 in 2024, while period 150 changed from +2,109.6 to −1,910.7. **One year's win rate does not select a lookback for another year.** The fade led elsewhere. Only 11 of 84 hourly fade-long settings and none of the 84 fade-short settings were profitable in both years. A period chosen for the breakout cannot be carried over to mean reversion without retesting the direction. ## Do filters and costs change the answer? They did not. We used the hourly period 20 and 1σ breakout as a baseline, then changed one practical condition at a time. Each cell is trades and annual net pips. | Condition | Year | Trades | Net | |---|---:|---:|---:| | Plain | 2025 | 277 | +1,052.5 pips | | Plain | 2024 | 280 | +2,071.7 pips | | ADX ≥ 20 | 2025 | 188 | −1,033.2 pips | | ADX ≥ 20 | 2024 | 185 | −3.9 pips | | Tokyo hours | 2025 | 164 | +519.4 pips | | Tokyo hours | 2024 | 162 | +2,139.6 pips | | London and New York hours | 2025 | 207 | +1,073.5 pips | | London and New York hours | 2024 | 218 | +2,154.3 pips | | SL 30 / TP 60 | 2025 | 398 | +1,687.6 pips | | SL 30 / TP 60 | 2024 | 419 | +1,467.1 pips | | SL 50 / TP 100 | 2025 | 323 | +1,157.3 pips | | SL 50 / TP 100 | 2024 | 345 | +1,896.2 pips | | SL 100 / TP 200 | 2025 | 297 | +1,340.5 pips | | SL 100 / TP 200 | 2024 | 292 | +1,987.6 pips | | 24-bar time exit | 2025 | 334 | +989.1 pips | | 24-bar time exit | 2024 | 337 | +1,559.0 pips | Fixed stops and targets are executable levels measured from the filled entry and are booked at that configured distance when reached. ADX thresholds of 20, 25 and 30 all trailed the unfiltered breakout. London and New York hours improved it slightly in both years, but not by enough to replace the period conclusion. No fixed stop-and-target combination beat the opposite-band exit in both years. The 277 signals stayed fixed while spread changed. Net profit was +1,135.6 pips at zero spread, +1,052.5 at 0.3, +858.6 at 1, +581.6 at 2 and +304.6 at 3 pips. The simple break-even spread was **4.1 pips**. Shortening period increases frequency, so costs need to be checked again rather than borrowed from period 20. ## Related tests [Which Bollinger Bands standard deviation works best?](/blog/bollinger-band-sigma-settings) holds the period range tighter and compares 0.5σ through 4σ. [315 Bollinger Band backtests](/blog/bollinger-band-settings) broadens the entry and exit combinations. [Moving-average crossover settings](/blog/moving-average-cross-settings) show how period affects two lines, while [SMA versus EMA](/blog/sma-vs-ema) isolates the averaging method at the same period. [Backtesting without code](/blog/backtest-without-coding) shows how to rebuild these conditions on another pair. ## Notes - One pair and two calendar years - The best period changed by timeframe and cannot be assumed to transfer to another pair or to daily bars - Selecting from 12 periods and seven deviations creates multiple-comparison uplift. The opposite year is a check, not untouched future data - Period 5 and 2σ is impossible under this population-standard-deviation implementation. A sample deviation or a calculation that excludes the current close changes the boundary - Touches use highs and lows; breakouts use closes. If one bar touches both sides, a fade may reverse to the opposite side at that close, but it cannot exit and reopen the same side until a later bar - Fills use bar closes and a fixed spread. Changing spreads, limit-order slippage and intrabar path are not reproduced - An open position is closed at the final bar of each window, so the two half-year results do not add exactly to the calendar-year result --- ### Does the Bollinger BandWidth Squeeze Work? 4,032 Settings Tested URL: https://formiq.jp/blog/bollinger-bandwidth-settings Language: en Published: 2026-08-28 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Bollinger BandWidth, BBW, squeeze, volatility, indicator settings, backtest, USDJPY Also available in: ja — https://formiq.jp/ja/blog/bollinger-bandwidth-settings Bollinger BandWidth turns the distance between the upper and lower Bollinger Bands into one volatility line. A low reading says the bands are tight; a high one says they are wide. It says **nothing about whether price is moving up or down**. The usual search for the “best BBW settings” starts at 20 periods, two standard deviations and a squeeze defined by the lowest BandWidth in 125 bars. Across 4,032 USD/JPY runs, the convention did not transfer cleanly. **A fixed BBW number changed meaning by almost five times across the three timeframes, and the 125-bar low produced fewer two-year hourly winners than a 20- or 50-bar low.** Period 20 did stand out on the hourly chart: 42 of 64 combinations were profitable in both years. But the default-like combination of 20, 2, a 125-bar low and a 75-bar breakout window took 58 trades and lost 70.6 pips in 2025. One good marginal is not a universal preset. ## What Bollinger BandWidth measures The calculation starts with ordinary Bollinger Bands: 1. Middle band: a simple moving average of the close 2. Upper and lower bands: middle ± deviation × standard deviation 3. **BandWidth = (upper − lower) ÷ middle × 100** If the deviation multiplier is `k`, the numerator is `2 × k × standard deviation`. With the familiar two-deviation setting, BBW is `4 × standard deviation ÷ 20-bar average × 100`. Dividing by the middle band makes the width a percentage of price rather than a raw number of yen or dollars. It does not make every timeframe comparable. Twenty 15-minute bars cover five hours; twenty four-hour bars cover 80 hours, so the standard deviations being measured are different objects. ### Volatility does not supply a side We measured every fresh 125-bar BBW low and the close-to-close move over the next 20 bars in 2025. | Timeframe | Squeeze lows | Mean signed move | Mean absolute move | All-bar absolute move | |---|---:|---:|---:|---:| | M15 | 615 | −5.03 pips | 27.40 | 29.39 | | H1 | 129 | +0.45 pips | 76.90 | 61.73 | | H4 | 38 | −15.35 pips | 110.89 | 119.69 | Only the hourly chart moved more than its all-bar baseline after a squeeze, by 24.6%. M15 and H4 moved less. More importantly, the signed means did not point consistently anywhere: −5.03, +0.45 and −15.35 pips. That is why the trading rule below waits for price. A close through the upper band supplies a long; a close through the lower band supplies a short. BBW identifies the volatility setup, not the direction. ## Bollinger BandWidth settings and signals BandWidth is drawn as **one line in a separate pane below price**. It is bounded by zero but has no universal 30/70-style reference lines. | Platform | Where it is found | |---|---| | MT4 / MT5 | Install a custom BBW indicator; only Bollinger Bands are built in | | TradingView | Search Indicators for Bollinger BandWidth | | Browser (Formiq) | Pick “BB Width” from the indicator list | The standard MetaTrader lists contain Bollinger Bands and Standard Deviation, not BandWidth. Custom MT4/MT5 versions therefore vary in their fields and guide lines; there is no standard BBW level drawn by default. ### The chart inputs [TradingView's official Bollinger BandWidth guide](https://www.tradingview.com/support/solutions/43000501972-bollinger-bandwidth-bbw/) lists the following defaults. Formiq exposes period and deviation, fixes the source to close and the average to SMA, and lets the ordinary line colour and thickness controls handle appearance. | Input | Default | What it changes | |---|---:|---| | Length | 20 | Bars used for both the average and standard deviation | | Source | Close | Price series entering the calculation | | StdDev | 2 | Standard deviations between the middle and each outer band | | Highest Expansion Length | 125 | Window used to show the recent BBW maximum | | Lowest Contraction Length | 125 | Window used to show the recent BBW minimum | The last two are reference windows; they do not change the BBW line. [John Bollinger’s own BandWidth note](https://www.bollingerbands.com/_files/ugd/58be43_d09c50b6e8ea4afd9af0523ef94de876.pdf) describes the popular squeeze as a 125-period low. ### The backtest inputs BBW is now available directly in the strategy builder. | Input | Starting value | What it does | |---|---:|---| | Trigger | Band break after squeeze | Waits for price to confirm the side | | Period | 20 | Builds BBW and the matching Bollinger Bands | | Deviation | 2 | Sets both the line scale and the outer bands | | Contraction lookback | 125 | Bars over which BBW must be the minimum | | Breakout wait | 75 bars | How long that squeeze remains eligible | | Width level | 0.5 | Used only by the absolute width-cross trigger | The two selectable readings are deliberately explicit: - **Band break after squeeze:** once BBW makes a lookback low, buy the next upper-band close or sell the next lower-band close inside the wait window - **Width crosses level:** when BBW crosses upward through a fixed number, use close above or below the middle band to pick long or short The second reading is the control group for a fixed threshold. For ordinary use, the relative squeeze is the more defensible starting point. Period, lookback and wait decide what “unusually narrow” means; deviation also moves the price band that confirms the break. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Window | 2025-01-01 to 2025-12-31, with 2024 run identically for comparison | | Timeframes | 15-minute / 1-hour / 4-hour | | Squeeze grid | 4 periods × 4 deviations × 4 contraction windows × 4 waits = 256 | | Fixed-level control | 4 periods × 4 deviations × 5 levels = 80 | | Periods | 10 / **20** / 30 / 50 | | Deviations | 1 / 1.5 / **2** / 2.5 | | Contraction lookbacks | 20 / 50 / **125** / 250 bars | | Breakout waits | 10 / 20 / 50 / **75 bars** | | Fixed levels | 0.1 / 0.2 / 0.4 / 0.8 / 1.6 | | Total | 336 settings × 3 timeframes × 4 windows = 4,032 runs | | Exit | The opposite signal; entry and exit use the same rule set | | Stops and targets | Off in the main grid; tested separately below | | Spread and fill | 0.3 pips fixed, filled at the close | | Size | 0.1 lot | A rolling low alone never opens a trade. Price must subsequently close through an outer band. This keeps the line’s volatility information separate from the direction supplied by price. ## What BBW level means a squeeze? No shared level does: the median moved 4.7-fold across timeframes. Period 20, deviation 2, every bar in 2025: | Timeframe | 10th percentile | Median | 90th percentile | Full range | |---|---:|---:|---:|---:| | M15 | 0.1392 | **0.2943** | 0.6611 | 0.0384–3.5979 | | H1 | 0.3303 | **0.6439** | 1.3723 | 0.1082–3.7455 | | H4 | 0.7642 | **1.3693** | 2.5363 | 0.3544–4.6219 | The median rises from 0.2943 to 1.3693, a factor of 4.7. A threshold of 0.4 is above the M15 median, just above the hourly tenth percentile, and below even the H4 tenth percentile. One number has become “wide,” “narrow” and “extremely narrow” at once. The fixed-level control reflects that mismatch. Settings profitable in both years fell from 36 of 80 on M15 to 22 of 80 on H1 and 14 of 80 on H4. The 2025 median was +201.6 pips on M15 and zero on H1 and H4, where many level combinations never traded. **A fixed BBW threshold is not portable.** Narrowness needs a percentile, rolling low or other comparison drawn from the same market and timeframe. ## How many bars should the squeeze use? The published 125-bar low was the weakest, profitable in both years in 16 of 64. The 256 relative-squeeze settings: | Timeframe | Year | Profitable settings | Median net | Positive in both years | |---|---:|---:|---:|---:| | M15 | 2025 | 159/256 | +324.1 pips | **115/256** | | M15 | 2024 | 177/256 | +661.1 pips | **115/256** | | H1 | 2025 | 166/256 | +490.5 pips | **106/256** | | H1 | 2024 | 162/256 | +458.4 pips | **106/256** | | H4 | 2025 | 111/256 | −283.9 pips | **65/256** | | H4 | 2024 | 168/256 | +566.7 pips | **65/256** | M15 and H1 each put more than half the settings above zero in an individual year, but only 45% and 41% survived both years. H4 inverted from a +566.7-pip median in 2024 to −283.9 in 2025. That is not evidence that every squeeze is profitable. It is evidence that changing the year and timeframe removes more than half the candidates. H4 averaged only 22 trades per setting, so its brightest rows were also its least dependable. ### The published 125-bar low was not special Hourly marginals, 64 settings per row: | Contraction lookback | 2025 mean | 2024 mean | Profitable in both | 2025 mean trades | |---:|---:|---:|---:|---:| | 20 | **+916.0** | +665.7 | **37/64** | 143.2 | | 50 | +144.6 | **+753.7** | 31/64 | 100.0 | | 125 | −13.4 | +322.4 | **16/64** | 51.5 | | 250 | +399.3 | −125.4 | 22/64 | 28.4 | The conventional 125-bar row had the fewest two-year winners. A longer lookback makes a squeeze rarer, but rarity is not an edge by itself. At 250 bars, the mean count was down to 28.4 trades and the 2024 marginal was already negative. ## Which period should you use? Period 20 was the only hourly row above +800 pips in both years, at 42 of 64. The same hourly grid grouped by period, 64 settings per row: | Period | 2025 mean | 2024 mean | Profitable in both | 2025 mean trades | |---:|---:|---:|---:|---:| | 10 | −7.3 | +392.2 | 22/64 | 130.8 | | **20** | **+1,101.1** | **+815.3** | **42/64** | 80.6 | | 30 | +612.8 | −345.8 | 19/64 | 65.3 | | 50 | −260.1 | +754.8 | 23/64 | 46.4 | Period 20 was the only row above +800 pips in both annual means, and 42 of 64 settings stayed profitable. Period 30 averaged +612.8 pips in 2025 and −345.8 in 2024; period 50 averaged −260.1 in 2025 and +754.8 in 2024. The breakout wait mattered too. The two-year profitable counts for waits of 10, 20, 50 and 75 bars were 19, 25, 33 and 29 of 64. Their 2024 means were −155.2, +27.5, +991.8 and +752.4 pips. A confirmation can be too early, but an old squeeze can also be kept alive too long. ## What does the deviation setting do? It leaves the line's shape alone and moves the band price has to break. Every deviation-2 BBW value was twice its deviation-1 counterpart, with maximum error below 4.1e-14 on all three timeframes. The bars at which local maxima, local minima and rolling lows occur do not change. That creates an exact identity in the fixed-level control: - deviation 1 / level 0.1 equals deviation 2 / level 0.2 - deviation 1 / level 0.2 equals deviation 2 / level 0.4 - deviation 1 / level 0.4 equals deviation 2 / level 0.8 - deviation 1 / level 0.8 equals deviation 2 / level 1.6 Across four periods, three timeframes and four windows, all **192 matched pairs** had identical trades, win rates and net pips. When the rule only reads BBW crossing a number, scaling both fields changes nothing. The squeeze-break rule is different because deviation also moves the outer price bands. Its hourly deviation-1 marginal averaged +735.3 pips in 2025 and +763.1 in 2024, with 38 of 64 profitable in both. Deviation 2 averaged +225.3 and +116.3, with 22 of 64. The squeeze dates are unchanged; the price required to confirm them is not. ## What happens if you keep the defaults? No timeframe was profitable in both years. Period 20, deviation 2, a 125-bar low and a 75-bar breakout window combines the familiar chart defaults with the strategy builder’s starting wait. | Timeframe | Year | Trades | Win rate | Net | |---|---:|---:|---:|---:| | M15 | 2025 | 314 | 35.35% | +88.9 pips | | M15 | 2024 | 298 | 30.54% | −293.5 pips | | H1 | 2025 | 58 | 41.38% | −70.6 pips | | H1 | 2024 | 71 | 40.85% | +909.0 pips | | H4 | 2025 | 26 | 15.38% | −1,307.7 pips | | H4 | 2024 | 20 | 40.00% | +570.4 pips | No timeframe was profitable in both years. On H1, keeping period 20, deviation 2 and the 125-bar low but shortening the wait to 50 bars produced 34 trades and +3,043.1 pips in 2025, then 55 and +1,459.6 in 2024. Extending eligibility to 75 bars turned 2025 into −70.6. **How long a squeeze remains valid matters as much as how it is identified.** ## Does a one-year winner hold? The two hourly settings stayed profitable but shrank; both four-hour settings turned negative. | Timeframe | Selection year | Period | Deviation | Low lookback | Wait | Trades in selection year | Net in selection year | Net in check year | |---|---:|---:|---:|---:|---:|---:|---:|---:| | H1 | 2025 | 20 | 1 | 20 bars | 10 bars | 167 | +3,868.5 pips | 2024 +1,589.6 pips | | H1 | 2024 | 50 | 1 | 125 bars | 75 bars | 63 | +3,025.0 pips | 2025 +450.2 pips | | H4 | 2025 | — | — | — | — | 3 | +2,392.6 pips | 2024 −889.8 pips | | H4 | 2024 | — | — | — | — | 10 | +4,025.5 pips | 2025 −710.7 pips | Both hourly settings remained profitable in the check year, but at smaller totals. Both four-hour settings turned negative, and their selection-year samples contained only three and ten trades. ## Can win rate decide this? It pointed the same way as profit here, and still was not enough on its own. The default-like hourly row took 58 trades at a 41.38% win rate. Its mean winner was +134.46 pips and mean loser −96.99. Winners were 1.39 times larger, but 24 wins against 34 losses still summed to −70.6 pips. **Win rate, payoff size and sample size have to be read together.** ## Filters, exits and cost These follow-up runs keep the default-like hourly 20 / 2 / 125 / 75 setup fixed. ### ADX and sessions | Entry filter | Year | Trades | Net | |---|---:|---:|---:| | None | 2025 | 58 | −70.6 pips | | None | 2024 | 71 | +909.0 pips | | ADX ≥ 20 | 2025 | 43 | +289.7 pips | | ADX ≥ 20 | 2024 | 56 | +583.1 pips | | ADX ≥ 25 | 2025 | 32 | +284.4 pips | | ADX ≥ 25 | 2024 | 44 | +17.6 pips | | ADX ≥ 30 | 2025 | 21 | +219.5 pips | | ADX ≥ 30 | 2024 | 28 | +793.8 pips | | 00:00–08:00 UTC only | 2025 | 32 | **+821.5 pips** | | 00:00–08:00 UTC only | 2024 | 39 | +1,020.8 pips | | 07:00–21:00 UTC only | 2025 | 50 | +6.0 pips | | 07:00–21:00 UTC only | 2024 | 57 | **+1,518.3 pips** | ADX turned 2025 positive, but every threshold reduced 2024 from its +909.0-pip baseline. Both session windows improved both annual totals, although the smaller window leaves only 32 and 39 trades and was selected after seeing these years. ### Stops, targets and time | Exit | Year | Trades | Net | |---|---:|---:|---:| | Opposite BBW signal | 2025 | 58 | −70.6 pips | | Opposite BBW signal | 2024 | 71 | +909.0 pips | | Stop 30 / target 60 | 2025 | 131 | +570.0 pips | | Stop 30 / target 60 | 2024 | 134 | +694.0 pips | | Stop 50 / target 100 | 2025 | 104 | +489.7 pips | | Stop 50 / target 100 | 2024 | 115 | +1,055.2 pips | | Stop 100 / target 200 | 2025 | 74 | **+1,556.3 pips** | | Stop 100 / target 200 | 2024 | 85 | **+1,181.4 pips** | | Exit after 24 bars | 2025 | 103 | +826.0 pips | | Exit after 24 bars | 2024 | 110 | +976.2 pips | The 100/200 exit and the 24-bar exit beat the opposite-signal exit in both years. In 2025 the opposite-signal position stayed open for 149 bars on average, so shortening the exit changes the holding period substantially. These are exploratory comparisons chosen from four candidates, not an untouched validation set. ### Spread At zero spread the row already lost **53.2 pips over 58 trades**. At 0.3 it lost 70.6; the 17.4-pip difference is exactly `58 × 0.3`. It fell to −111.2 at 1.0 pips and −227.2 at 3.0. The arithmetic break-even spread is −0.92 pips. A trader cannot receive a negative spread, so there is **no non-negative break-even spread** for this row. Cost made it worse, but cost did not create the loss. ## Related tests [Bollinger Band fades versus breakouts](/blog/bollinger-band-settings) compares the two price readings of the underlying bands. [The period sweep from 5 to 200](/blog/bollinger-band-period-settings) and [the deviation sweep from 0.5σ to 4σ](/blog/bollinger-band-sigma-settings) isolate the two fields that build BBW’s numerator. [Squeeze Momentum settings](/blog/squeeze-momentum-settings) define contraction as Bollinger Bands moving inside a Keltner Channel, which is not the same event as a 125-bar BandWidth low. [Building a backtest without code](/blog/backtest-without-coding) is the route to repeat the new BBW condition on another pair or timeframe. ## Notes - It covers USD/JPY in 2024 and 2025. Equities, crypto and other FX pairs may have different BBW distributions and breakout behaviour. - “Squeeze” means a rolling N-bar low followed by an outer-band close inside a bounded wait. Percentile lows and Bollinger-inside-Keltner are different squeeze definitions and were not tested here. - The grid is limited to the listed periods, deviations, lookbacks, waits and fixed levels; it is not a continuous optimisation. - The line uses close, SMA and population standard deviation. Other applied prices and average types were not tested. - Several H4 settings took only three to ten trades in a year, leaving their results highly uncertain. - Fills are at the close with a fixed 0.3-pip spread. Slippage, variable intraday cost and financing are excluded. --- ### Is Currency Strength Trading Profitable? 72 Settings Tested URL: https://formiq.jp/blog/currency-strength-settings Language: en Published: 2026-08-28 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Currency Strength, strongest currency, weakest currency, currency momentum, forex strategy, backtest Also available in: ja — https://formiq.jp/ja/blog/currency-strength-settings Buying the strongest currency and selling the weakest sounds like a way to put both sides of the FX market behind one trade. Across all 28 pairs formed by eight major currencies, the result was much less convenient: **the rule did not win as stated**. We tested H1, H4 and D1 charts, lookbacks from 5 to 80 bars and holding periods from 1 to 24 bars. None of the 72 strongest-versus-weakest momentum settings was profitable after spread in both 2024 and 2025. A familiar hourly starting point, 14 bars of strength followed by a four-bar hold, took 1,553 trades in 2025, won 39.6% and lost 6,236.2 basis points. We then stopped forcing the trade to use only first and eighth place. First-versus-seventh, second-versus-eighth, second-versus-seventh and an equal-weight four-pair top-two/bottom-two basket were added. Momentum remained weak, but **reversing daily 20-bar strength and holding for 12 bars** was positive in 2024 and 2025 under all five selection methods. Spreading capital across all 12 possible start phases left second-versus-eighth as the stronger single-pair candidate and the four-pair basket as the diversified candidate. “The strongest pair” needs one clarification. The meter ranks **currencies**, not pairs. The traded pair is the combination of the top-ranked and bottom-ranked currencies. Confusing that cross-sectional rank with the rate of change on one open chart turns two different measurements into one. ## What a currency strength meter measures Every currency price is a ratio against another currency. A rising USD/JPY chart cannot tell us by itself whether the dollar is broadly strong or the yen alone is weak. A strength meter combines several bilateral rates into one index for each currency. The [BIS effective exchange-rate framework](https://data.bis.org/topics/EER) also summarizes a currency through geometric averages of bilateral rates. Its weights come from trade and its purpose is macroeconomic. Short-term trading meters have no single standard: some equal-weight pairs, some normalize by volatility, and others measure distance from a moving average. This test fixed one transparent definition: 1. Use every one of the 28 pairs among USD, EUR, JPY, GBP, AUD, NZD, CAD and CHF. 2. Calculate each pair's log return over the last N bars. 3. Give that return a positive sign for the base currency and a negative sign for the quote currency. 4. Average the seven readings attached to each currency. 5. Rank the eight averages from strongest to weakest. In compact form: `strength(c) = 1/7 × Σ[oriented N-bar return]` A rise in EUR/USD adds to EUR and subtracts from USD. A rise in USD/JPY adds to USD and subtracts from JPY. Averaging seven opponents keeps one unusual bilateral move from owning the whole score. ### Strength has no universal scale Changing the universe or weights changes the line. An equal-weight basket of eight liquid currencies and a trade-weighted basket of 40 partners are both valid measures of “strength,” but they answer different questions. Zero here means relative to the eight-currency average; it does not mean absolute economic strength. Before comparing colours or ranks, check the **currency universe, lookback, return definition and weighting method**. A reading of +2.0 from two different meters need not represent the same move. ## Currency strength settings and practical use Four fields reach the trading result directly. | Input | Values in this test | What it changes | |---|---|---| | Currency universe | Eight majors | Who competes for the ranks | | Lookback | 5 / 10 / **14** / 20 / 40 / 80 bars | How far back strength is measured | | Rebalance interval | Same as holding period | How often the strongest and weakest are selected again | | Weighting | Equal across seven opponents | How much each bilateral rate contributes | A short lookback promotes the latest shock. A long lookback preserves moves lasting days or months. “Fourteen bars” is 14 hours on H1 and roughly three trading weeks on D1, so matching the number does not match the strategy. The holding period is also the exit and rebalance interval. A one-bar hold switches pairs every bar; a 24-bar hold ignores rank changes until the next scheduled rebalance. In the 2025 hourly grid, mean annual spread cost fell from 15,723.7 basis points at a one-bar hold to 2,981.4 at 24 bars. Rebalance frequency is not a cosmetic setting. ### MT4, MT5, TradingView and Formiq | Platform | How Currency Strength is handled | |---|---| | MT4 / MT5 | Install a custom indicator and inspect its formula | | TradingView | Public scripts differ in universe, formula and normalization | | Formiq | The separate pane shows how the base currency of the open pair moved against its quote currency | [MetaTrader 5's official indicator help](https://www.metatrader5.com/en/terminal/help/charts_analysis/indicators) separates 38 built-in indicators from custom indicators. Currency Strength is not a standard entry, so the shared name does not guarantee a shared calculation. Formiq's current chart line is a simple pair-relative reading: `(current ÷ N bars ago − 1) × 100`, with 14 as the starting period. **This article tests a different calculation:** it synchronizes 28 pairs to build an eight-currency cross-sectional rank. One pair's line cannot identify first and eighth place across the full universe. ## How this was measured | Item | Value | |---|---| | Currencies | USD / EUR / JPY / GBP / AUD / NZD / CAD / CHF | | Pair universe | All 28 combinations of those currencies | | Window | 2024-01-01 to 2025-12-31 | | Timeframes | H1 / H4 / D1 | | Synchronized bars | 12,439 / 3,222 / 627 | | Strength lookbacks | 5 / 10 / **14** / 20 / 40 / 80 bars | | Holding periods | 1 / **4** / 12 / 24 bars | | Momentum side | Buy strongest, sell weakest | | Reversal control | Buy weakest, sell strongest | | Added rank selections | 1 vs 8 / 1 vs 7 / 2 vs 8 / 2 vs 7 / top two × bottom two | | Entry | Next open after the signal bar closes | | Exit | Open after the fixed hold; rank again immediately | | Cost | Recorded bid/ask at each fill; no commission, slippage or financing | | Exposure | Equal notional for one pair; total notional split equally in the four-pair basket | | Initial runs | 3 timeframes × 6 lookbacks × 4 holds × 4 windows × 2 directions = **576** | | Rank-extension runs | The same grid × five selection methods = **2,880** | Strength uses only closes available when the signal bar finishes. The trade begins at the next open, so it never receives the close that created the rank as a fill. In the added top-two/bottom-two method, the total notional is split into quarters across all four combinations of ranks 1–2 and 7–8. Returns are averaged, not summed at four times the exposure. Returns are in basis points so JPY and non-JPY pairs can be combined. One basis point is 0.01%. Annual figures sum trade percentage returns and are not compounded account returns. Besides the two calendar years, the first and second halves of 2025 were checked. Parameter selection uses 2024 before looking at 2025, and the strength-gap filter below takes its threshold from 2024 only. The extension also shifts a 12-bar holding schedule through all offsets from zero to 11, separating a result tied to one convenient start date from a phase-neutral result. ## Does buying the strongest work? Each timeframe contains 24 momentum settings: six lookbacks by four holds. Medians are annual net basis points after recorded spread. | Timeframe | Year | Profitable settings | Median net | Positive in both years | |---|---:|---:|---:|---:| | H1 | 2024 | 1/24 | −2,771.6 bps | **0/24** | | H1 | 2025 | 0/24 | −4,987.7 bps | **0/24** | | H4 | 2024 | 12/24 | −296.7 bps | **0/24** | | H4 | 2025 | 0/24 | −1,989.0 bps | **0/24** | | D1 | 2024 | 3/24 | −734.2 bps | **0/24** | | D1 | 2025 | 4/24 | −1,177.8 bps | **0/24** | | Total | 2024 | 16/72 | — | **0/72** | | Total | 2025 | 4/72 | — | **0/72** | Half the H4 grid was profitable in 2024; all 24 settings lost in 2025. The four daily winners in 2025 were each negative under the same setting in 2024. Removing spread left only one of 72 settings gross-profitable in both years. H4 with a 10-bar lookback and a 12-bar hold made 359.6 and 29.1 basis points gross. Recorded bid/ask reduced those numbers to +107.2 and −245.1, removing the two-year survivor. ### The best 2025 setting failed in 2024 The 2025 momentum leader was D1 with an 80-bar lookback and 24-bar hold, at +883.3 basis points. It lost 720.1 in 2024. The runner-up, D1 20/24, made 857.0 in 2025 after losing 1,670.8 in 2024. Those rows prove that a profitable setting can be found after opening the test year. They do not provide a setting selected before that year began. ## How did the 14/4 default do? The hourly 14-bar lookback with a four-bar hold produced the following split: | Year | Trades | Win rate | PF | Before spread | Spread cost | After spread | Mean win | Mean loss | |---:|---:|---:|---:|---:|---:|---:|---:|---:| | 2024 | 1,552 | 43.8% | 0.73 | −889.7 bps | −2,908.0 bps | **−3,797.7 bps** | +15.26 bps | −16.24 bps | | 2025 | 1,553 | 39.6% | 0.63 | −1,249.1 bps | −4,987.1 bps | **−6,236.2 bps** | +17.38 bps | −18.08 bps | The 2025 win rate fell below 40%, and the mean loss was slightly larger than the mean win. Gross performance was already −1,249.1 basis points, so spread cannot carry all the blame. The two independently reset halves were negative too: −4,323.0 in the first half and −964.7 in the second. The second half was barely positive before spread at +65.4, but could not absorb 1,030.1 basis points of cost. ### Reversal flips gross direction, not cost Reversing the same pairs at the same times produced +1,249.1 gross basis points in 2025. The same 4,987.0-bp spread bill left it at −3,737.9 net. Momentum and reversal gross results are nearly opposite signs. Their transaction costs are not. **Trying both sides does not guarantee one net winner; spread creates a region in which both lose.** ## Did last year's best settings hold? The five best 2024 momentum settings were carried unchanged into 2025. | Timeframe | Lookback | Hold | 2024 net | 2025 net | |---|---:|---:|---:|---:| | H4 | 10 bars | 24 bars | +1,199.0 bps | −1,289.6 bps | | H4 | 14 bars | 24 bars | +984.4 bps | −1,676.2 bps | | H4 | 40 bars | 12 bars | +793.5 bps | −1,835.1 bps | | H4 | 40 bars | 24 bars | +711.6 bps | −2,243.8 bps | | H4 | 5 bars | 24 bars | +665.6 bps | −1,645.5 bps | Their mean 2025 result was −1,738.0 basis points, with zero profitable settings. Selection should therefore check whether the same condition remains positive in the check year, not whether it merely loses less than neighbouring settings. ## Do gap filters, longer holds or reversal rescue it? ### A larger strength gap did not predict continuation We admitted trades only when the strongest-minus-weakest gap exceeded the 75th percentile learned from 2024. The threshold was never fit on 2025. For H1 14/4, filtering reduced 1,553 trades and −6,236.2 basis points to 430 trades and −2,744.1. Total loss fell because exposure fell, but average return worsened from −4.016 to −6.382 basis points per trade and PF slipped from 0.63 to 0.59. For H4 10/12, 133 trades and −245.1 became 39 trades and −153.2. Average return fell from −1.843 to −3.927 and PF from 0.94 to 0.91. A more dramatic rank gap was not a continuation filter in these rows. ### Longer holds cut turnover but did not repair gross return These are 2025 hourly means across the six lookbacks. | Hold | Mean trades | Before spread | Spread cost | After spread | |---:|---:|---:|---:|---:| | 1 bar | 6,214 | −575.6 | −15,723.7 | −16,299.4 | | 4 bars | 1,553 | −1,167.3 | −5,037.3 | −6,204.6 | | 12 bars | 517 | −704.9 | −3,388.1 | −4,093.0 | | 24 bars | 258 | −649.5 | −2,981.4 | −3,630.8 | Longer holds traded less and lost less. Gross return remained negative at every holding-period marginal, so reducing turnover did not create an edge by itself. ### Six reversal settings survived both years Buying the weakest and selling the strongest left 6 of 72 settings profitable in both years. The standout was H4 with a 20-bar lookback and 24-bar hold. | Year | Trades | Win rate | PF | Net return | |---|---:|---:|---:|---:| | 2024 | 66 | 39.4% | 1.15 | +375.3 bps | | 2025 | 66 | 59.1% | 2.45 | +2,454.9 bps | That row was selected after inspecting 576 runs. More importantly, the five best reversal settings selected on 2024 all lost in 2025, averaging −1,745.4 basis points. The test found pockets of short-horizon mean reversion, not a validated instruction to reverse every meter. ## Do the second-highest and second-lowest ranks help? We moved each edge of the rank inward by one place and reran the same 72 settings under five selection methods. “Top two × bottom two” is an equal-weight portfolio of four pairs. The table counts settings that were profitable after spread in both 2024 and 2025. | Rank selection | Momentum | Reversal | |---|---:|---:| | First vs eighth | 0/72 | 6/72 | | First vs seventh | 1/72 | 4/72 | | Second vs eighth | 1/72 | **12/72** | | Second vs seventh | 2/72 | 6/72 | | Equal-weight top two × bottom two | 0/72 | 1/72 | Moving to the second rank did not rescue continuation: momentum survived in at most two of 72 settings. The weakness was not only a first-versus-eighth outlier problem. The improvement clustered on reversal. ### Daily 20/12 reversal survived all five selections Selling the high-ranked side, buying the low-ranked side and holding for 12 daily bars was profitable in both years under every selection. These first results rebalance every 12 bars from each calendar year's first eligible entry. | Reversal selection | 2024 | 2025 | |---|---:|---:| | Sell first, buy eighth | +526.4 bps | +783.1 bps | | Sell first, buy seventh | +567.6 bps | +1,009.2 bps | | Sell second, buy eighth | +620.9 bps | +993.2 bps | | Sell second, buy seventh | **+642.5 bps** | **+1,207.0 bps** | | Sell top two, buy bottom two | +589.4 bps | +998.1 bps | We also checked the nine-setting D1 neighbourhood formed by lookbacks 14, 20 and 40 and holds 4, 12 and 24. Daily 20/12 was the only cell in which all five selection methods were profitable in both years. Daily 20/4 had zero and 20/24 had one, so the neighbouring hold periods did not reproduce the 20/12 result. ### Start phase changes a single schedule materially A 12-bar hold selects different trades depending on which January session starts the schedule. Across all offsets from zero to 11, the top-two/bottom-two basket ranged from −321.6 to +970.9 basis points in 2024 and from −748.0 to +998.1 in 2025. Choosing the convenient phase would overstate the result. We therefore divided capital into 12 sleeves, started one sleeve at each daily phase and held each for 12 bars. A sleeve receives one-twelfth of total capital; the four-pair method divides that sleeve equally again. The table shows mean annual net return across the 12 phases, including recorded spread. | Reversal selection | 2024 | 2025 | Phases positive both years | |---|---:|---:|---:| | Sell first, buy eighth | +499.9 bps | +450.6 bps | 7/12 | | Sell first, buy seventh | +454.8 bps | +125.9 bps | 5/12 | | Sell second, buy eighth | +315.6 bps | **+603.4 bps** | **9/12** | | Sell second, buy seventh | +266.1 bps | +291.4 bps | 5/12 | | Sell top two, buy bottom two | **+384.1 bps** | **+367.8 bps** | **8/12** | Second-versus-eighth had the most single-pair phases profitable in both years. The basket reduced the worst 2025 phase to −748.0 basis points, versus −1,410.1 for second-versus-eighth. The rolling basket made 336.3 basis points in the first half of 2025 and only 31.5 in the second. Two positive calendar years therefore do not establish regime independence. ### The candidate rule that remained The least-fragile procedure inside this sample is: 1. Synchronize daily data for all 28 pairs among the eight currencies. 2. Calculate equal-weight strength over the last 20 bars. 3. Reverse the rank: sell the high side and buy the low side. 4. Hold for 12 bars and refresh one-twelfth of capital each day. 5. Use second-versus-eighth for one pair, or split a sleeve equally across all four top-two/bottom-two pairs. This condition was positive in both years, but adjacent settings broke and it is not a profit-guaranteeing preset. Financing matters over a roughly 12-session hold and is not in the result. Requiring a strength gap above the 2024 median left the four-pair basket with just seven trades and +1,017.3 basis points in 2025. That sample is too sparse to promote the gap filter into the rule. ## Is this academic currency momentum? The BIS working paper [Currency Momentum Strategies](https://www.bis.org/publ/work366.pdf) reports a significant winner-minus-loser spread across a broad currency sample. It also finds that transaction costs explain part of the return and discusses limits that make the premium hard to exploit. That result need not conflict with this test: - The study forms broad diversified portfolios; the initial rule held one first-versus-eighth pair and the extension reaches only a four-pair top-two/bottom-two basket. - Academic formation and holding periods are mainly monthly; this grid runs from hourly bars to daily bars over 1–24 bars. - The study uses a long, broad sample; this test covers eight majors in 2024–2025. Evidence for “currency momentum” is not evidence that buying the strongest currency over 14 hours and holding it for four hours works. Ranking method, diversification, horizon and cost define the strategy. ## Related tests [Moving-average cross settings](/blog/moving-average-cross-settings) measure time-series momentum inside one pair. Currency strength is different because it compares several currencies at the same timestamp. [RSI periods and 30/70 levels](/blog/rsi-settings) test both continuation and reversal after a recent move. [Building a backtest without code](/blog/backtest-without-coding) shows how to retest the selected pair's entry and exit conditions. The 28-pair ranking itself is a synchronized multi-series calculation and should not be presented as a single-chart condition. ## Notes - The universe contains eight major currencies and their 28 pairs. It excludes emerging-market currencies, metals, equity indices and crypto. - The window is only 2024–2025 and does not represent long-run currency momentum or every monetary-policy regime. - Strength is equal-weighted across seven opponents. Trade weights, volatility scaling, carry adjustment and moving-average distance would produce different meters. - The extension covers ranks 1–2 and 7–8 plus an equal-weight four-pair basket. Broader long-short portfolios and volatility weighting were not tested. - Exits are fixed holds. Stops, targets, rank-cross exits and trailing exits are outside this grid. - Recorded bid/ask is included; commission, slippage, financing and size-dependent execution are not. - Daily 20/12 was also discovered from the 2024–2025 comparison. Start phases and adjacent settings were checked, but 2026 and later remain untouched rather than validated. --- ### Do Fair Value Gaps Really Fill? 2 Years of USD/JPY Tested URL: https://formiq.jp/blog/fair-value-gap-settings Language: en Published: 2026-08-28 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Fair Value Gap, FVG, price action, indicator, backtest, USDJPY Also available in: ja — https://formiq.jp/ja/blog/fair-value-gap-settings A fair value gap (FVG) is a three-candle pattern. If the first candle's high and the third candle's low do not overlap, the price range between them is read as territory the middle candle crossed without trading back against it: an imbalance price is said to return to. It does return. On USD/JPY, 96.2-99.4% of gaps were revisited. The problem is that this number **says nothing about fair value gaps.** A bullish gap's ceiling is the third candle's low, so "the gap was filled" is word-for-word the same event as "price came back to that candle's low." And every candle has a low. ## What a fair value gap actually tests One comparison between two candles, done once: - **Bullish**: third candle's low > first candle's high. The band sits between those two prices - **Bearish**: third candle's high < first candle's low. Same, inverted The middle candle does not appear. It is not in the expression, so a wide-range bar and a doji produce the same verdict. How often the condition fires on USD/JPY: | Timeframe | Year | Bars | Gaps | Share of bars | Band width (median) | |---|---|---|---|---|---| | 15-minute | 2025 | 24,903 | 5,610 | 22.53% | 2.95 pips | | 15-minute | 2024 | 24,999 | 5,455 | 21.82% | 2.70 pips | | Hourly | 2025 | 6,226 | 1,312 | 21.07% | 6.85 pips | | Hourly | 2024 | 6,250 | 1,271 | 20.34% | 5.80 pips | | 4-hour | 2025 | 1,610 | 368 | 22.86% | 15.75 pips | | 4-hour | 2024 | 1,616 | 369 | 22.83% | 13.50 pips | **One bar in five opens a gap.** This is not a rare configuration. On the 15-minute chart, 82-89 gaps a year were under a tenth of a pip wide, which draws as a dashed line rather than a band. ## Do fair value gaps really fill? 99% do, and so does every other bar. The fill rate is compared with the rate at which price came back to the low or high of bars that opened *no* gap: bars where the first and third candles overlapped. | Timeframe | Year | Bars that opened a gap | Every other bar | |---|---|---|---| | 15-minute | 2025 | 99.36% | 99.37% | | 15-minute | 2024 | 99.19% | 99.39% | | Hourly | 2025 | 98.55% | 98.71% | | Hourly | 2024 | 98.27% | 98.82% | | 4-hour | 2025 | 97.01% | 97.18% | | 4-hour | 2024 | 96.21% | 97.79% | That table is the figure at the top. **In all six cells the gap bar is the lower of the two.** The median wait before the level was reached was two bars for both groups. "Fair value gaps get filled" is a correct observation, but not because they are gaps. In a market that ranges, the high and low of any bar is eventually traded through. The indicator has put a name on that general property. ## How long does a gap take to fill? The median wait is two bars. Timing the fill makes it harder to use, not easier. | Timeframe | Year | Median | On the next bar | Within 3 bars | 90th percentile wait | |---|---|---|---|---|---| | 15-minute | 2025 | 2 bars | 44.03% | 63.31% | 44 bars | | 15-minute | 2024 | 2 bars | 46.07% | 64.02% | 41 bars | | Hourly | 2025 | 2 bars | 45.32% | 62.57% | 43 bars | | Hourly | 2024 | 2 bars | 43.55% | 62.37% | 40 bars | | 4-hour | 2025 | 2 bars | 45.10% | 63.59% | 29 bars | | 4-hour | 2024 | 2 bars | 41.97% | 63.10% | 54 bars | **More than four in ten are filled on the bar immediately after the gap forms.** The band appears on the chart the moment the third candle closes and is gone on the next one: that is the single most common outcome. "Wait for price to return to an unfilled gap, then enter" does not survive this distribution. Close to half of the gaps are finished between the band appearing and an order reaching the market. Some do stay open. Across all of 2025, the gaps still unfilled at the end of the year numbered 36 on the 15-minute chart (of 5,610), 19 hourly (of 1,312) and 11 on the 4-hour (of 368). ## Does price stop inside the band? It does not: the band has no inside. A band has width, so "touched the edge" and "filled to the far side" should be different events. Measuring how far the first bar that arrives reaches into the band: 0 is the near edge, 1 the far one: | Timeframe | Touches | Depth reached (median) | Reached halfway | Closed it fully | Overshot by half a band | |---|---|---|---|---|---| | 15-minute | 5,574 | 1.12 | 70.58% | 53.07% | 42.07% | | Hourly | 1,293 | 1.06 | 66.82% | 51.35% | 40.84% | | 4-hour | 357 | 0.97 | 66.11% | 49.30% | 39.78% | **The median is about 1.0.** The typical behaviour is that the arriving bar goes straight through, and about half close the band completely in that single bar. Four in ten carry half a band's width past it. Price entering the band and stalling there did not show up in the data. ## How to show fair value gaps on a chart None of the platforms ship one, so every route takes a step: - **MT4 / MT5**: no fair value gap in the bundled indicator list. You put a custom indicator (.mq4 / .mq5) into the indicators folder and add it from the Navigator - **TradingView**: not in the built-in indicator list either; you search the public scripts and add one - **Formiq**: in the chart's indicator settings under the pattern group, with a toggle for showing only unfilled gaps and a minimum width in pips Those two settings are the ones that matter. Gaps open on a fifth of all bars and 99% of them fill, so drawing every gap fills the chart with history rather than levels. ## Does a bigger middle candle help? The test never reads it. The published reading calls the middle candle *displacement* and wants it large. The test does not look at it. Whether that costs anything is measurable. Splitting the middle candle's body into ATR quartiles, then measuring how often price moved the gap's way over the ten bars after it returned to the level: | Timeframe | Q1 (smallest) | Q2 | Q3 | Q4 (largest) | |---|---|---|---|---| | 15-minute | 49.68% | 47.02% | 51.76% | 49.32% | | Hourly | 53.11% | 53.73% | 47.52% | 41.93% | | 4-hour | 44.32% | 47.19% | 49.44% | 52.81% | Bigger is not better. The best quartile is Q3 on the 15-minute chart, Q2 on the hourly (where it *falls* as the body grows) and Q4 on the 4-hour. **The direction changes with the timeframe.** The boundary itself is worth one more step. Measuring the overlap between the first and third candles as a signed multiple of ATR, and keeping the negative side: the near misses, where the two candles do overlap and there is no gap: Nothing happens on either side of zero. The per-bin figures scatter between 41.46% and 55.87% across the 24 cells (three timeframes by eight bins), but the step across zero is −1.81 points on the 15-minute chart, +2.81 hourly and −4.35 on the 4-hour. **The signs do not agree.** The line the indicator draws sits partway along a continuum. ## How this was measured The strategy test below uses the price measurements above. | Item | Value | |---|---| | Pair | USD/JPY | | Timeframes | 15-minute / hourly / 4-hour | | Windows | 2025 (main), 2024 (out of sample), first and second half of 2025 | | Spread | 0.3 pips | | Lot | 0.1 | | Exit | Opposite signal (stops and targets measured separately) | | Readings | Return to the gap / gap opens | | Minimum gap size | 0, 0.25, 0.5, 1, 1.5, 2 x ATR | | Valid for | 5, 10, 20, 50, 100 bars (return-to-the-gap only) | Two readings of one pattern: - **Return to the gap**: when price comes back to a band that is still unfilled, enter the way the band points (bullish gap = buy). This is the textbook reading - **Gap opens**: enter on the bar that creates the gap, in its direction. The same object read as displacement The size floor is in ATR rather than pips because the median band runs from 2.95 pips on the 15-minute chart to 15.75 on the 4-hour. No single pip figure straddles all three. ## Which timeframe worked? Not one of the 30 hourly retracement settings was profitable in both years. Settings that finished 2025 in profit: | Reading | Timeframe | Year | Profitable settings | Positive in both years | 2025 median | |---|---|---:|---:|---:|---:| | Return to the gap | 15-minute | 2025 | 13/30 | 6/30 | −361.7 pips | | Return to the gap | 15-minute | 2024 | 20/30 | 6/30 | — | | Return to the gap | Hourly | 2025 | 6/30 | 0/30 | −1,817.6 pips | | Return to the gap | Hourly | 2024 | 20/30 | 0/30 | — | | Return to the gap | 4-hour | 2025 | 5/30 | 5/30 | −1,842.8 pips | | Return to the gap | 4-hour | 2024 | 22/30 | 5/30 | — | | Gap opens | 15-minute | 2025 | 2/6 | 1/6 | −513.5 pips | | Gap opens | 15-minute | 2024 | 3/6 | 1/6 | — | | Gap opens | Hourly | 2025 | 3/6 | 3/6 | −79.7 pips | | Gap opens | Hourly | 2024 | 6/6 | 3/6 | — | | Gap opens | 4-hour | 2025 | 1/6 | 1/6 | −1,584.7 pips | | Gap opens | 4-hour | 2024 | 6/6 | 1/6 | — | **On the hourly chart, not one of the 30 return-to-the-gap settings was profitable in both years.** The two years disagree sharply. In 2024 at least half the settings were profitable in every group; in 2025 the largest count was 13/30. Two years are not enough to determine which state is more likely to persist. The condition editor opens with a 0.25x ATR floor that remains valid for 50 bars: | Reading | Timeframe | Year | Trades | Win rate | PF | Net | |---|---|---:|---:|---:|---:|---:| | Return to the gap | 15-minute | 2025 | 1,334 | 47.08% | 1.078 | +1,251.5 pips | | Return to the gap | 15-minute | 2024 | 1,278 | 48.51% | 1.075 | +1,235.2 pips | | Return to the gap | Hourly | 2025 | 321 | 41.43% | 0.689 | −3,023.1 pips | | Return to the gap | 4-hour | 2025 | 96 | 43.75% | 0.636 | −2,107.5 pips | | Gap opens | 15-minute | 2025 | 1,211 | 37.90% | 1.052 | +840.7 pips | | Gap opens | 4-hour | 2024 | 73 | 41.10% | 1.751 | +2,852.4 pips | Only the 15-minute chart is positive in both years, and it does so on more than 1,200 trades rather than on a handful. The size floor is clearest there: | Minimum size | Mean net | Mean trades | Profitable | |---|---|---|---| | 0 (off) | +381.5 pips | 2,314 | 3/5 | | 0.25x ATR | +1,382.9 pips | 1,229 | 5/5 | | 0.5x ATR | +2,228.5 pips | 605 | 5/5 | | 1x ATR | −606.9 pips | 133 | 0/5 | | 1.5x ATR | −1,302.1 pips | 49 | 0/5 | | 2x ATR | −949.1 pips | 29 | 0/5 | 15-minute chart, 2025, return-to-the-gap. Discarding the thin gaps works; discarding too many leaves nothing. ## Did last year's best still work? All six lost money the next year. The test this series runs every time: take the best setting from one year and apply it, unchanged, to the other. | Reading | Timeframe | Minimum size | Validity | 2024 net | 2025 net | |---|---|---:|---:|---:|---:| | Return to the gap | 15-minute | 1x ATR | 100 bars | +3,713.3 pips | −1,172.2 pips | | Return to the gap | Hourly | 1x ATR | 100 bars | +3,968.4 pips | −2,744.2 pips | | Return to the gap | 4-hour | 1.5x ATR | 20 bars | +2,946.3 pips | −1,765.1 pips | | Gap opens | 15-minute | 1x ATR | — | +2,445.9 pips | −136.7 pips | | Gap opens | Hourly | 1x ATR | — | +4,186.1 pips | −1,017.3 pips | | Gap opens | 4-hour | 0.25x ATR | — | +2,852.4 pips | −1,694.4 pips | **All six lost money.** The hourly return-to-the-gap setting changed from +3,968.4 pips in 2024 to −2,744.2 in 2025; the hourly gap-open setting changed from +4,186.1 to −1,017.3. ## Did any setting win half its trades? Restricted to settings that took at least 20 trades, **zero** settings in any of the six groups had a win rate above 50%. | Reading | Timeframe | Cells | Mean win rate | Mean win | Mean loss | |---|---|---|---|---|---| | Return to the gap | 15-minute | 30 | 43.56% | 76.93 pips | −72.33 pips | | Return to the gap | Hourly | 20 | 40.00% | 79.40 pips | −83.68 pips | | Return to the gap | 4-hour | 20 | 37.43% | 101.46 pips | −132.37 pips | | Gap opens | 15-minute | 6 | 38.78% | 77.89 pips | −72.53 pips | | Gap opens | Hourly | 4 | 38.52% | 92.30 pips | −67.71 pips | | Gap opens | 4-hour | 4 | 34.38% | 137.85 pips | −120.82 pips | Despite the mean-reversion wording of "return to the gap," its mean win rate was only 37.43% to 43.56%. On hourly and four-hour bars, the average loss was also larger than the average win. The hit rate does not establish a mean-reversion edge here. ## Filters, stops and the spread Added to the editor's opening setting, 2025 throughout. An ADX filter only trimmed the one group that was profitable: | Condition | Trades | Win rate | Net | |---|---|---|---| | Bare | 1,334 | 47.08% | +1,251.5 pips | | ADX(14) ≥ 20 | 945 | 46.24% | +537.2 pips | | ADX(14) ≥ 25 | 662 | 46.53% | +288.0 pips | | 07-16 UTC only | 640 | 46.09% | −1,013.2 pips | | 12-21 UTC only | 574 | 49.13% | +888.9 pips | One session window survives and the other does not. 12-21 UTC keeps +888.9 pips on 57% of the trades, while 07-16 UTC flips the sign to −1,013.2 pips. A filter that works one way and breaks the other is a partition of the sample, not a reason to use it. Stops and targets add a little on the 15-minute chart (+1,354.1 pips at 30/60, +1,312.0 at 50/100, against +1,251.5 bare) and shrink the losses on the 4-hour (−2,107.5 bare, −132.0 at 30/60). They make the losing groups lose less; they do not make them win. The spread behaves the way it has in every article in this series: | Reading | Timeframe | Trades | Spread 0 | 0.3 pips | 1.0 pips | Break-even | |---|---|---|---|---|---|---| | Return to the gap | 15-minute | 1,334 | +1,651.6 pips | +1,251.5 pips | +317.7 pips | 1.238 pips | | Gap opens | 15-minute | 1,211 | +1,204.0 pips | +840.7 pips | −7.0 pips | 0.994 pips | | Return to the gap | Hourly | 321 | −2,926.8 pips | −3,023.1 pips | −3,247.8 pips | — | **What the spread takes is the trade count times the spread.** Predicted and observed agreed within 0.1 pips in all six cases. The two profitable 15-minute settings disappear above 1.238 and 0.994 pips of spread: real USD/JPY leaves room, but not much for something trading more than 1,200 times a year. For related shapes, [SMC structure breaks and liquidity sweeps](/blog/smc-settings) separate the broader framework into fixed price rules. Promoting a drawing tool to a testable condition is the same move as [Fibonacci retracement](/blog/fibonacci-retracement-levels); the same control group is applied to [seventeen candlestick patterns](/blog/candlestick-patterns-tested); counting a base rate before evaluating a filter comes from [GMMA](/blog/gmma-settings); and an indicator turning out to be something already known is [CCI](/blog/cci-settings). ## Notes - **USD/JPY only.** Another pair may differ, and band width in particular scales with how much a pair moves - **Two years, 2024 and 2025**, are included - **Exits use the opposite signal.** People who trade this pattern often place a stop beyond the far edge of the band, or hold to a prior swing. Those alternative exit rules are not measured - **"Returned" means touched the near edge.** Some definitions require a close inside the band, which would change the figures - **The session filter uses one of the two available windows.** Other hours than the two tested here exist - **Higher-timeframe gaps traded on a lower chart are not covered.** Each timeframe is measured within itself --- ### Is the Heikin Ashi Color Flip Profitable? 960 Settings Tested URL: https://formiq.jp/blog/heikin-ashi-settings Language: en Published: 2026-08-28 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Heikin Ashi, Candlesticks, Trend Following, Indicator, Backtest, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/heikin-ashi-settings Heikin Ashi does not draw a line next to the candles. It **replaces the candles**. Two lines of arithmetic do the whole job: the close becomes the average of the bar's four prices, the open becomes the midpoint of the previous Heikin Ashi bar. The chart comes out visibly smoother. Twenty-one indicators into this series, it is the first thing measured that rewrites the bars themselves, and doing that raises a problem none of the other twenty had. **The Heikin Ashi close is a price nobody traded at.** The same 6,107 trades, priced at the real close and then at the Heikin Ashi close. Nothing else changed. ## What Heikin Ashi calculates Two lines per bar: - **Heikin Ashi close** = (open + high + low + close) ÷ 4 - **Heikin Ashi open** = (previous Heikin Ashi open + previous Heikin Ashi close) ÷ 2 The high and low take the outer of those two against the bar's real high and low. Three properties follow, and no other indicator in this series has them. ### The Heikin Ashi close is not a price An average of four prices is not any of the four. **There is no moment inside the bar when the market was at that number.** How far away it sits: | Timeframe | Distance from the real close (median) | Mean | Top 10% | Max | |---|---|---|---|---| | 15-minute | 2.22 pips | 3.09 pips | 6.70 pips | 78.5 pips | | 1-hour | 4.55 pips | 6.22 pips | 13.46 pips | 113.2 pips | | 4-hour | 9.26 pips | 12.58 pips | 27.34 pips | 147.6 pips | USD/JPY, 2025. And **the error has a direction.** Of the green Heikin Ashi bars, the ones whose Heikin Ashi close sits *below* the real close are 63.7% to 67.3% across three timeframes and two years. When the signal says buy, the recorded price is on the cheap side. ### Heikin Ashi never forgets its own start date The Heikin Ashi open comes from the bar before it, which came from the bar before that. **There is no period setting because it uses every bar there is.** The past never drops out, but it halves every bar. Recomputing the same bars from a start date 500 bars later: | Timeframe | Gap at the start | Bars to agree within 1 pip | 0.1 pip | 0.01 pip | 0.001 pip | |---|---|---|---|---|---| | 15-minute | 4.76 pips | 3 | 6 | 9 | 13 | | 1-hour | 7.06 pips | 3 | 7 | 10 | 13 | | 4-hour | 7.30 pips | 3 | 7 | 10 | 13 | **The ratio between one bar's gap and the next was 0.5 exactly, on all 16 to 17 bars where it was large enough to measure.** Subtract the two recursions and only the halved difference survives, so this is arithmetic rather than a finding. It has two practical consequences. **Changing the start date changes the values**: though thirteen bars close the gap to three decimal places, so it rarely shows. And **there is no answer to "how much history does it need"**: fifty bars is plenty in practice, but no number makes the seed exactly zero. ### The Heikin Ashi colour is a moving average crossover The Heikin Ashi open is the midpoint of the previous open and close, which makes it **an exponential moving average of the Heikin Ashi close with a smoothing factor of 0.5**. A factor of 0.5 is a 3-period EMA. So **"the bar is green" means "(O+H+L+C)/4 is above its own 3-period EMA as of the previous bar"**: nothing more. Measured across three timeframes and two years: **zero colour disagreements and zero disagreements about which bar the colour changed on**. (The open itself differs by at most 0.08 to 0.48 pips, which is the seed still decaying.) **A Heikin Ashi candle is a price/average cross drawn in the shape of a candle.** The last section of this article measures what that costs. ## How to add Heikin Ashi to a chart | Platform | Steps | |---|---| | MT4 / MT5 | Heiken Ashi ships with the platform. Navigator → Indicators → Heiken Ashi, dragged onto the chart | | TradingView | Pick Heikin Ashi from the chart-type menu at the top of the chart | | Browser (Formiq) | Available as a chart type. The backtest exposes four readings (colour flip, N-bar run, wickless bar, doji) and four calculation variants | **TradingView is where the care is needed.** Leaving the chart type set to Heikin Ashi while a strategy runs prices the **fills** at Heikin Ashi values too. That gap is exactly what this article measures. MT4 works the other way: Heiken Ashi only draws, so `Close[]` stays the real close. But MT4 has no built-in function returning the series, so exporting a rule to an EA means **writing a loop that rebuilds the bars**. ## How Heikin Ashi was tested | Item | Value | |---|---| | Pair | USD/JPY | | Window | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Readings | Colour flip / N bars of one colour (1–5) / wickless bar / doji (body ratio 0.05, 0.1, 0.2) | | Variants | Standard / better (2-bar average) / smoothed (3, 5, 10) / double smoothed (3, 5, 10) | | Combinations | 80 per cell. 3 timeframes × 4 windows = 960 in total | | Exit | Opposite signal. All four readings are symmetric, so entry and exit share a rule set | | **Fills** | **The real candle's close. Heikin Ashi decides only when to trade** | | Stops and targets | Off (measured separately below) | | Spread | 0.3 pips fixed | | Size | 0.1 lots | **Filling at the real candles is the premise of this test.** The transform decides *when*; the market decides *at what*. The next section measures what happens when that premise is dropped. ## Why the backtest looks profitable Because the fill is written at the Heikin Ashi close. The strategy does not change at all. **Only which close the fill is written down at.** The spread is zero on both sides so nothing but the price differs. The figure at the top of this article is the result: grey is the real close, red is the Heikin Ashi close, and both lines are the same 6,107 trades. | Reading | Timeframe | Year | Real close | Heikin Ashi close | Difference | |---|---|---|---|---|---| | Colour flip | 15-minute | 2025 | −523.1 | **+34,120.0** | +34,643.1 | | Colour flip | 15-minute | 2024 | −376.0 | **+34,013.3** | +34,389.3 | | Colour flip | 1-hour | 2025 | +1,617.7 | **+18,228.6** | +16,611.0 | | Colour flip | 4-hour | 2025 | +910.7 | **+10,203.5** | +9,292.8 | | Run of 3 | 1-hour | 2025 | +1,276.5 | +3,875.8 | +2,599.3 | | Wickless bar | 1-hour | 2025 | +1,123.1 | +6,708.6 | +5,585.6 | | Doji (0.1) | 15-minute | 2025 | −2,450.0 | **+2,996.1** | +5,446.1 | **Across 36 cells (six readings × three timeframes × two years) the Heikin Ashi price was kinder in 35. In 15 of them it turned a loss into a profit.** ### Where the difference comes from Every trade has an entry and an exit, and both fills include the gap between the Heikin Ashi close and the real close. Across three timeframes and two years, the Heikin Ashi close was below the real close on 63.7% to 67.3% of bullish candles. That lowers simulated buy prices and raises simulated sell prices. | Timeframe | Year | Mean gap | × 2 | Measured lift per trade | Ratio | |---|---|---|---|---|---| | 15-minute | 2025 | 3.09 | 6.18 | 5.67 | 0.917 | | 1-hour | 2025 | 6.22 | 12.44 | 10.73 | 0.863 | | 4-hour | 2025 | 12.58 | 25.17 | 24.33 | 0.967 | | 4-hour | 2024 | 12.05 | 24.10 | 23.58 | 0.978 | **The lift per trade runs at about twice the average gap**: 0.86 to 0.98 times it, measured. Twenty articles in this series have confirmed that *cost = trades × spread*; this is the same law with the sign reversed. **Trades × roughly twice the gap is the profit that was invented.** Which is why the 15-minute colour flip is the most spectacular of them: 6,107 trades a year, 5.67 pips of fiction each. ## Does Heikin Ashi really cut noise? It does. Heikin Ashi is sold on cutting noise and holding a trend. Measured, that part is true. | Timeframe | Year | Heikin Ashi mean run | Ordinary candles | Colour changes (HA / ordinary) | |---|---|---|---|---| | 15-minute | 2025 | 4.08 | 1.98 | 6,107 / 12,576 | | 1-hour | 2025 | 4.02 | 1.95 | 1,548 / 3,185 | | 4-hour | 2025 | 4.20 | 2.00 | 383 / 806 | **The mean run roughly doubles and the colour changes roughly halve**, on all six cells. Runs that end after a single bar fall from 49.2–50.7% to 20.8–24.3%. **The smoothing being real and the smoothing being profitable are separate questions.** The rest of the article is the second one. ### The wickless bar happens on half the bars A green Heikin Ashi bar with no lower wick is supposed to mark a strong trend. Count it first: | Timeframe | Share of bars with no wick | |---|---| | 15-minute | 46.9–47.3% | | 1-hour | 46.2–46.8% | | 4-hour | 47.7–49.4% | **Roughly every second bar.** A condition that holds half the year separates nothing. It is the second time this series has found one: GMMA's "long ribbon in order" held on 85% of bars. The doji is rarer: a body under 10% of the range occurred on 10.5–12.5% of bars. ## The four readings, backtested The strategy results use **fills from the real candles** and the standard Heikin Ashi variant throughout. | Timeframe | Year | Colour flip | Run of 3 | Wickless | Doji (0.1) | |---|---|---|---|---|---| | 15-minute | 2025 | −2,355.2 | −1,602.5 | −535.2 | −2,905.4 | | 15-minute | 2024 | −2,200.9 | −1,308.4 | −1,260.5 | −3,417.0 | | 1-hour | 2025 | **+1,153.3** | **+1,094.1** | **+912.8** | **+788.6** | | 1-hour | 2024 | **+101.3** | **+2,194.2** | **+3,895.2** | −1,548.6 | | 4-hour | 2025 | **+796.1** | −834.3 | −566.5 | −1,244.3 | | 4-hour | 2024 | **+3,051.6** | −302.1 | **+2,972.9** | **+682.9** | Trade counts on hourly bars in 2025: 1,548 for the flip, 608 for the run of 3, 701 for the wickless bar, 434 for the doji. **All twelve 15-minute cells lost.** No reading rescues it. The colour flip trades 6,107 times a year there, so 0.3 pips of spread alone is 1,832 pips while the rule already loses 523.1 pips at zero spread. Only the doji shows a high win rate (52.76% on hourly bars in 2025, against 35–37% for the rest), but its annual net was lower than the trend readings shown in the table. **The higher win rate did not produce the higher annual net.** ## Which timeframe did it work on? Only the hourly chart held together. | Reading | 15-minute | 1-hour | 4-hour | |---|---|---|---| | Colour flip | 1/8 | **7/8** | 5/8 | | N-bar run | 11/40 | **32/40** | 7/40 | | Wickless | 3/8 | 3/8 | 0/8 | | Doji | 1/24 | 8/24 | 11/24 | | **All 80** | **16/80** | **50/80** | **23/80** | 2025 medians: −509.3 pips on 15-minute, +850.7 on hourly, +5.3 on 4-hour. **Hourly bars are the only place Heikin Ashi worked honestly**: 63 of 80 settings ahead in 2025 and 50 of 80 ahead in both years. Few indicators in this series have produced a timeframe that held together this well. ## How many bars should you wait? Waiting longer helped on the hourly chart and hurt on the four-hour. Waiting for N same-coloured bars, N from 1 to 5. **A run of 1 is the colour flip**, so the ladder lands on something already measured (the sweep asserts the two are identical). | Timeframe | Year | 1 | 2 | 3 | 4 | 5 | |---|---|---|---|---|---|---| | 15-minute | 2025 | −2,355.2 | −1,090.9 | −1,602.5 | −401.9 | +85.7 | | 15-minute | 2024 | −2,200.9 | −1,729.8 | −1,308.4 | −774.6 | −1,851.0 | | 1-hour | 2025 | +1,153.3 | **+2,415.4** | +1,094.1 | +1,375.8 | +442.4 | | 1-hour | 2024 | +101.3 | +2,545.3 | +2,194.2 | +2,227.4 | **+3,392.5** | | 4-hour | 2025 | +796.1 | **+1,478.1** | −834.3 | −993.7 | −496.5 | | 4-hour | 2024 | **+3,051.6** | +1,379.9 | −302.1 | −1,328.7 | −1,227.3 | **It is not monotonic.** On hourly bars, waiting raises the win rate (38.5% to 42.5%) and cuts the trades (824 to 336) without moving net pips much. Only the both-years count comes out cleanly there: 7/8 at a run of 1, **8/8** at a run of 5. On 4-hour bars, none of the eight settings that waited three bars or more stayed profitable in both years. On hourly bars, all eight settings that waited five bars did. The effect of the waiting rule reversed between the two timeframes. ## Which calculation variant should you use? The more smoothing, the less it resembles the standard. Standard, better (average two bars, then transform), smoothed and double smoothed. What changes first: | Variant | Colour changes (hourly, 2025) | Mean run | Bars whose colour differs from standard | |---|---|---|---| | Standard | 1,548 | 4.02 | 0% | | Better | 1,232 | 5.05 | 12.3% | | Smoothed (5) | 718 | 8.67 | 27.9% | | Double smoothed (5) | 557 | 11.18 | 39.4% | **More smoothing means longer runs and a different indicator.** The double-smoothed series is a different colour on four bars in ten. Results (mean of the flip and the run readings, hourly): | Variant | 2025 | 2024 | |---|---|---| | Standard | +1,272 | +1,760 | | Better | +1,127 | +2,021 | | Smoothed (3) | **+1,543** | **+2,506** | | Smoothed (10) | +267 | +3,200 | | Double smoothed (10) | +282 | +1,406 | **Smoothed (3) made +1,543 pips in 2025 and +2,506 in 2024.** Smoothed (10) made +267 in 2025 and +3,200 in 2024. On 15-minute bars only double smoothed (10) cleared both years (+1,706 and +1,697), while cutting the trades to 1,198. ## Does last year's best transfer? Only one of the three cleared the median of the year it moved into. | Timeframe | Reading | Variant | 2024 net | 2025 net | 2025 median | |---|---|---|---:|---:|---:| | 15-minute | Wickless | Smoothed (5) | +2,921.9 pips | −2,445.0 pips | −509.3 pips | | 1-hour | Run of 2 | Smoothed (10) | +4,030.9 pips | −209.5 pips | +850.7 pips | | 4-hour | Doji (0.2) | Smoothed (5) | +4,237.9 pips | +437.8 pips | +5.3 pips | Only the four-hour setting beat its 2025 median. Hourly bars had 50 of 80 settings positive in both years, yet the setting that made +4,030.9 pips in 2024 lost 209.5 in 2025. ## Does it have to be Heikin Ashi? The colour turned out to be (O+H+L+C)/4 against its own 3-period EMA. So the question is what the transform adds over **running that same cross on ordinary candles**. The control is an EMA(1)/EMA(3) crossover on the close: the same two-line crossover condition without the Heikin Ashi transform. It was compared across 48 cells covering three timeframes, two years and eight variants. | | Heikin Ashi colour flip | EMA 1/3 cross on the close | |---|---|---| | Trades (hourly, 2025, standard) | 1,548 | 2,082 | | Win rate | 35.66% | 32.37% | | Net pips | +1,153.3 | +476.0 | Over all 48: - **Fewer trades in 48 of 48** - **Higher win rate in 46 of 48** - **More net pips in 31 of 48** **The transform is reliable on trade count and win rate, and better than a coin flip on net pips.** This is the third time this series has run that comparison (CCI's typical price came out at 25 of 36, the psychological line's bar counting at 16 of 36) and **31 of 48 is the best of the three.** The split by timeframe matters, though. 15-minute bars favour the transform in 16 of 16 cells and hourly bars in 13 of 16; **4-hour bars in only 2 of 16.** At 4 hours there was no reason to use it. The 15-minute clean sweep means "lost less", not "won": the EMA 1/3 control made −4,981.6 pips there in 2025 against the colour flip's −2,355.2. Both lose. ## Filters, stops and cost ### The ADX filter broke it again | Reading | Timeframe | Year | ADX condition | Trades | Net | |---|---|---:|---|---:|---:| | Colour flip | Hourly | 2025 | None | 1,548 | +1,153.3 pips | | Colour flip | Hourly | 2025 | ≥ 20 | 1,001 | +838.2 pips | | Colour flip | Hourly | 2025 | ≥ 25 | 705 | +422.0 pips | | Colour flip | Hourly | 2025 | ≥ 30 | 468 | +89.4 pips | | Colour flip | Hourly | 2024 | None | 1,525 | +101.3 pips | | Colour flip | Hourly | 2024 | ≥ 20 | 976 | −1,437.0 pips | | Colour flip | Hourly | 2024 | ≥ 25 | 671 | −1,043.0 pips | | Colour flip | Hourly | 2024 | ≥ 30 | 449 | −1,213.9 pips | **Monotonically worse in 2025, straight into the red in 2024.** Adding a trend-strength filter to a trend rule is now the fourth failure of its kind here, after GMMA, CCI and the psychological line. ### Session Tokyo hours (UTC 0–8) reduced the hourly colour flip from +1,153.3 to +646.5 in 2025, but increased it from +101.3 to **+1,558.7** in 2024. London and New York hours reduced it in both years (+1,153.3 → +257.4, +101.3 → −700.2). ### Stops and targets A 100-pip stop with a 200-pip target left the hourly colour flip at +1,050.2 in 2025 (baseline +1,153.3) and +20.4 in 2024 (baseline +101.3). **Nothing here counts as an improvement.** A 30-pip stop with a 60-pip target dropped 2025 to −191.1. ### Cost Net pips were exactly linear in the spread: **pips lost = trades × spread** held without error on all five settings. | Setting | Trades | Spread 0 | 0.3 | 1.0 | Break-even | |---|---|---|---|---|---| | Colour flip, 15-minute | 6,107 | −523.1 | −2,355.2 | −6,630.1 | never | | Colour flip, 1-hour | 1,548 | +1,617.7 | +1,153.3 | +69.7 | 1.05 pips | | Run of 3, 1-hour | 608 | +1,276.5 | +1,094.1 | +668.5 | **2.10 pips** | | Doji (0.1), 1-hour | 434 | +918.8 | +788.6 | +484.8 | **2.12 pips** | **The readings that trade less break even later.** The 15-minute colour flip loses at zero spread, so cost is not its problem. ## Related reading - [Psychological line settings](/blog/psychological-line-settings): the same "does it have to be this indicator" test, on one that throws away the size of every move - [CCI settings](/blog/cci-settings), where the zero line turned out to be a moving average crossover - [Moving average cross settings](/blog/moving-average-cross-settings), where this article's control ends up - [GMMA line count](/blog/gmma-settings): the first "a condition that holds 85% of the time separates nothing" - [Renko chart settings](/blog/renko-chart-settings): the second indicator here that rewrites the candles. Its close is a real price the bar has already left ## Notes - **One pair, two years.** The gap between the Heikin Ashi close and the real one scales with how far price moves, so a different instrument changes the size of the invented profit too - **The fill experiment ran at zero spread.** A real account pays spread on either calculation, but the gap between them is unchanged - **The spread is fixed at 0.3 pips.** The hourly colour flip breaks even at 1.05, so that assumption is doing real work - **Exits are on the opposite signal.** No stop-and-target combination improved anything by a margin worth reporting - **The MT4 export covers the standard variant only.** The three smoothed variants need a rolling average of each component, so the export falls back to standard and says so in the emitted code - **The doji ratio and the run length were swept over three and five values.** 960 combinations comes mostly from the eight variants, so the resolution within any one reading is coarse --- ### Is the Psychological Line 75% Rule Profitable? 5,760 Tests URL: https://formiq.jp/blog/psychological-line-settings Language: en Published: 2026-08-28 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Psychological Line, Oscillator, Mean Reversion, Trend Following, Indicator, Backtest, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/psychological-line-settings The psychological line does one thing: it **counts how many of the last N bars closed up** and turns that into a percentage. The usual setup is 12 bars, with 75% marked as overbought and 25% as oversold. What separates it from every other oscillator is that it is a **count, not an average**. A bar that rose 50 pips and a bar that rose a tenth of a pip are the same tally mark. Nineteen indicators into this series, it is the first one that throws the size of a move away entirely. Two consequences follow. The reading can only land on **N+1 values**, and the level you type does not sit on any of them. Ask a 12-bar line for "above 75" and what you get is 10 bars out of 12: **83.3%**. Across the six periods tested here, a typed 75 never meant 75%. ## What the psychological line measures The formula is one line: **Psychological line = (up bars in the last N ÷ N) × 100** No average, no standard deviation, no smoothing. Which is exactly why it behaves unlike anything else on the chart. ### The level you type is not the level you get A 12-bar psychological line can only produce thirteen values: 0 / 8.3 / 16.7 / 25.0 / 33.3 / 41.7 / **50.0** / 58.3 / 66.7 / **75.0** / 83.3 / 91.7 / 100 75.0 is on that list. But "above 75" excludes 75.0 itself, so the first value that qualifies is the next rung up: 10 of 12, or 83.3%. Change the period and the landing point moves. | Period | Values it can reach | What "75" demands | Actual level | |---|---|---|---| | 6 | 7 | 5 of 6 | 83.3% | | 10 | 11 | 8 of 10 | 80.0% | | 12 | 13 | 10 of 12 | 83.3% | | 14 | 15 | 11 of 14 | 78.6% | | 20 | 21 | 16 of 20 | 80.0% | | 26 | 27 | 20 of 26 | 76.9% | **None of the six is 75%.** That is the figure above: grey rungs are the values the period can reach, and the blue one is where a typed 75 lands. Some entered levels produce exactly the same condition. On a 12-bar line, 75, 76, 80, and 83 all require 10 rising bars out of 12. Lowering the level to 74 changes the requirement to 9 of 12, or 75.0%. **Nine percentage-point steps can leave the rule unchanged, while the next step changes the required bar count.** | Period | Levels that are one rule (around 75) | What they mean | |---|---|---| | 6 | 67–83 | 5 of 6 = 83.3% | | 12 | 75–83 | 10 of 12 = 83.3% | | 14 | 72–78 | 11 of 14 = 78.6% | | 26 | 74–76 | 20 of 26 = 76.9% | At period 26, levels 74 and 75 both require 20 rising bars out of 26, while level 70 requires 19. At period 12, levels 75 and 80 both require 10 rising bars. ### Identical rows show up inside the sweep The discrete steps change the trading results. The 15-minute mean-reversion buy at period 12 in 2025 was swept over five entry levels and three exit levels. | Entry → exit | Trades | Net pips | |---|---|---| | 70 → 40 | 363 | +142.0 | | 74 → 40 | 363 | +142.0 | | 70 → 50 | 314 | +697.7 | | 74 → 50 | 314 | +697.7 | | 75 → 50 | 119 | +250.6 | | 80 → 50 | 119 | +250.6 | | 90 → 50 | 22 | +207.7 | **Fifteen settings, nine distinct results.** The same thing happens across the whole test: of 5,760 combinations, only 3,548 produced a different outcome. **38.4% of the grid duplicates another row.** ### The period sets the resolution The ladder has period + 1 rungs, so a short period gives a coarse line. A 6-bar psychological line has seven readings, and typing any of the 101 levels from 0 to 100 selects one of only seven rules. At period 26 there are 27. Fewer still turn up in practice. Counted on USD/JPY in 2025 at period 12: | Timeframe | Rungs seen | Readings that never occurred | |---|---|---| | 15-minute | 12 of 13 | 0 bars up (all twelve down) | | 1-hour | 12 of 13 | 12 bars up (all twelve up) | | 4-hour | 10 of 13 | 0, 1 and 12 bars up | And **about one bar in five sits exactly on 50** on every timeframe (15-minute 22.65–23.71%, hourly 21.93–22.14%, 4-hour 21.59–21.76%). That matters for the midline. A cross at 50 and a cross at 55 are **the same rule going up and different rules coming down**: both need the count to rise to 7 of 12, but the downward leg needs it to fall to 6 at level 50 and to 7 at level 55. On hourly bars in 2025 the up leg fired 365 times at both levels; the down leg fired 345 times at 50 and 365 at 55. Trade volume moves with it. Average hourly trade counts were 378 at level 50, 703 at 55 and 622 at 45. **Moving the level five points nearly doubles the trading.** ### There are two definitions of an up bar Count bars that closed above the previous close, or bars that closed above their own open. Any bar that opens away from the last close splits the two. | Timeframe | Bars where the two disagree | Correlation of the lines | Bars that swap sides of 75 | Bars that swap sides of 50 | |---|---|---|---|---| | 15-minute | 1.24% | 0.976 | 0.32% | 2.79% | | 1-hour | 1.17% | 0.977 | 0.35% | 2.75% | | 4-hour | 0.81% | 0.984 | 0.38% | 2.19% | **They are genuinely different series.** They are not different results. Across 2,134 live settings run both ways, counting bull bodies finished ahead in 1,006: 47.1%, which is a coin flip. ## Adding the psychological line to a chart It came out of the Japanese equity market, so it is frequently absent from the built-in lists on non-Japanese platforms. | Platform | Steps | |---|---| | MT4 / MT5 | Not in the standard oscillator list. Drop a custom indicator (.mq4 / .mq5) into the Indicators folder and load it | | TradingView | Not in the built-in indicator list. Search the community scripts | | Browser (Formiq) | Built in. Period, both guide levels and the up-bar definition are settings, and the backtest exposes four readings | Exporting to MT4 means there is no `iCCI`-style built-in to call, so **the count has to be written as a loop over the last N bars**. The export used for this test does exactly that. ## How this was measured Before the numbers, the setup. Entering the same settings reproduces the same results. | Item | Value | |---|---| | Pair | USD/JPY | | Window | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Readings | Fade (opposite direction past the level) / band break (in the direction of the break) / midline cross | | Periods | 6, 10, 12, 14, 20, 26 | | Levels | Fade: entry 70, 74, 75, 80, 90 × exit 40, 50, 60. Break: 60, 70, 75, 80, 90. Midline: 40, 45, 50, 55, 60 | | Up-bar definition | Above the previous close / above its own open | | Combinations | 240 per cell per definition. 3 timeframes × 4 windows = 5,760 in total | | Exit | Opposite signal. The fade alone carries its own exit level | | Stops and targets | Off (measured separately below) | | Spread | 0.3 pips fixed, filled at the close | | Size | 0.1 lots | The fade points one way ("below 25 is a buy" only ever produces buys), so **the long fade and short fade were swept as separate trading rules**. The break and the midline are symmetric, so entry and exit share a rule set. Levels are handled as a reflected pair: typing 75 also draws 25. ## Is a reading above 75% unusual? It is not: the distribution is almost exactly a coin's. The premise behind the psychological line is that a run of up bars means the crowd has become too optimistic. If that were true, the direction of one bar would say something about the next, and **the distribution of up-bar counts would not look like chance**. So it was counted directly. The share of bars on each of the thirteen rungs, against what twelve tosses of a coin with the same up-bar rate would give. **The two ladders differ by 0.97 percentage points in total** on 15-minute bars in 2025: 2.19 on hourly bars, 2.79 on 15-minute bars in 2024. The reach rates are close to the theoretical values too. | Timeframe / year | Up-bar rate | Above 75 (measured / coin) | 75 or above (measured / coin) | Inside 75–25 | |---|---|---|---|---| | 15-minute 2025 | 50.22% | 1.76% / 2.00% | 7.02% / 7.51% | 85.97% | | 1-hour 2025 | 50.39% | 1.74% / 2.06% | 6.98% / 7.69% | 86.24% | | 4-hour 2025 | 50.78% | 2.32% / 2.19% | 8.39% / 8.08% | 86.67% | | 4-hour 2024 | 53.50% | 3.30% / 3.37% | 13.09% / 11.31% | 80.11% | Look at **"above 75" against "75 or above"**: 1.76% and 7.02% for the same number. **The comparison operator changes the frequency by a factor of four.** Textbooks write "75% or more"; most rule editors write "above 75". At period 12 that difference is exactly one bar. ### Does an up bar make the next one more likely? If the distribution matches a coin's, bar direction has no memory. Measured directly: | Timeframe / year | Up after an up bar | Up after a down bar | Lag-1 correlation | |---|---|---|---| | 15-minute 2025 | 49.59% | 50.85% | −0.0126 | | 15-minute 2024 | 49.76% | 53.11% | −0.0335 | | 1-hour 2025 | 49.17% | 51.62% | −0.0245 | | 1-hour 2024 | 51.42% | 51.95% | −0.0053 | | 4-hour 2025 | 50.67% | 50.95% | −0.0027 | | 4-hour 2024 | 54.75% | 52.00% | **+0.0274** | **Five of six are negative.** An up bar is very slightly less likely to be followed by another. That points the same way the indicator's premise does, but at **one to three points**, nowhere near enough to pull the distribution off the coin. Which sets the expectation: if there is anything here, it is on the **fade** side, and it is thin. ## Should you fade it or follow it? For the first time in this series, the fade beat the trend readings. The same period 12 and 75/25 pair, read three ways. The fade column shows the buy side and the sell side separately. | Timeframe | Year | Fade (buy / sell) | Band break | Midline | |---|---|---|---|---| | 15-minute | 2025 | **+250.6** / −265.6 | −2,428.2 | −1,099.2 | | 15-minute | 2024 | **+322.6** / **+644.9** | −97.2 | **+238.3** | | 1-hour | 2025 | **+82.5** / **+228.3** | −903.2 | **+272.6** | | 1-hour | 2024 | **+273.6** / −376.1 | **+725.8** | **+1,169.5** | | 4-hour | 2025 | **+566.9** / **+316.6** | −3,263.3 | **+65.6** | | 4-hour | 2024 | **+33.2** / −1,091.6 | **+2,663.8** | **+886.3** | Trade counts matter here. On hourly bars in 2025 the fade took 26 buys and 28 sells, the break 32, the midline 402. **The fade trades a few dozen times a year.** **This is the first indicator in the series where mean reversion beat trend following.** The fade buy is positive in 6 of 6 cells, the fade sell in 3 of 6, and the band break in 2 of 6. In the previous eighteen articles, trend-following rules had the stronger result, so this deserves a closer look. ## Was that the indicator or the trend? The reversal contribution was larger only in 2025. The fade buy enters after a run of down bars. In a rising year that earns money regardless of what opened the trade, so **the drift has to be separated from the indicator**. Each trade's result was compared with what an unconditional position of the same length would have made: the year's total move divided by its bar count, times the bars held. Whatever is left is the indicator's share. | Timeframe | Year | Pips per trade | Of which drift | Of which the indicator | Settings with a positive share | |---|---|---|---|---|---| | 15-minute | 2025 | +4.77 | −0.06 | **+4.83** | 77 of 81 | | 15-minute | 2024 | +2.00 | +1.64 | +0.36 | 45 of 72 | | 1-hour | 2025 | +8.07 | −0.23 | **+8.31** | 33 of 60 | | 1-hour | 2024 | +9.62 | +6.29 | +3.34 | 25 of 60 | | 4-hour | 2025 | +47.40 | −0.66 | **+48.07** | 49 of 51 | | 4-hour | 2024 | −21.06 | +25.34 | **−46.40** | 8 of 57 | Fade buys, averaged over settings with at least five trades. **2025 had almost no drift, so what is left is the indicator**, and 77 of 81 settings kept something. **It disappears in 2024.** The share falls to +0.36 on 15-minute bars and to −46.40 on 4-hour bars, where 49 of 57 settings are negative. USD/JPY rose 1,632 pips in 2024 and finished 2025 down 56. **"Overdone means it snaps back" only held in the market that snapped back.** The cell where the lag-1 correlation turned most positive (+0.0274, 4-hour 2024) is the same cell where the fade broke worst. ## Which settings held for two years? The 15-minute fade buy led at 60 of 90. To avoid choosing from one year's result, the table counts how many settings were profitable in both 2024 **and** 2025. | Reading | 15-minute | 1-hour | 4-hour | |---|---|---|---| | Fade buy | **60/90** | 32/90 | 23/90 | | Fade sell | 24/90 | 12/90 | 5/90 | | Band break | 5/30 | 8/30 | 5/30 | | Midline | 1/30 | **18/30** | 14/30 | **The 15-minute fade buy survived best at 60 of 90.** The fade sell managed 41 of 270. A buy/sell gap that wide, measured over two years one of which trended hard, **cannot be separated from the direction of the market**. **The midline only survives on hourly bars**: 1 of 30 on 15-minute, where it trades 729 to 3,968 times a year. ## How to choose period and level Shorter periods survived more often, and nothing argued for moving the midline off 50. Period results are split into the fade, band-break, and midline-cross rules because changing period affects their trade counts and annual net differently. **Fade buy, 15-minute, by period** | Period | 2025 mean | Mean trades | Profitable 2025 | Both years | |---|---|---|---|---| | 6 | +355 | 581 | 15/15 | **15/15** | | 10 | +366 | 185 | 15/15 | 11/15 | | 12 | +248 | 178 | 13/15 | 10/15 | | 14 | +371 | 130 | 14/15 | 8/15 | | 20 | +174 | 39 | 11/15 | 11/15 | | 26 | +191 | 18 | 12/15 | 5/15 | **Every one of the fifteen settings at period 6 finished ahead in both years.** Lengthening the period thins the trades, and by 18 a year nothing survives 2024. **Midline, 1-hour, by level** | Level | 2025 mean | 2024 mean | Mean trades | Both years | |---|---|---|---|---| | 40 | +272 | +1,743 | 464 | **5/6** | | 45 | +650 | +609 | 622 | 3/6 | | 50 | +411 | +1,419 | 378 | **5/6** | | 55 | +384 | +920 | 703 | 4/6 | | 60 | +54 | +14 | 483 | 1/6 | The trade counts are not ordered by level. **50 is the quietest at 378 and 55 the busiest at 703**, because 50 sits exactly on a rung and one bar in five parks there, which suppresses crossings. **There was no reason to move the midline off 50, but moving it changes the trading volume by nearly half.** ## Did the prior year's best hold? Two of five cleared the median of the year they moved into. The test at the centre of this series. | Reading | Timeframe | Period | Level | 2024 net | 2025 net | 2025 median | |---|---|---:|---|---:|---:|---:| | Midline | 1-hour | 12 | 40 | +3,191.5 pips | +102.0 pips | +234.4 pips | | Midline | 15-minute | 26 | 45 | +2,486.6 pips | −1,449.8 pips | −1,081.4 pips | | Midline | 4-hour | 6 | 40 | +3,341.4 pips | +158.5 pips | −211.8 pips | | Band break | 1-hour | 10 | 60 / 40 | +2,557.1 pips | −1,292.4 pips | −465.0 pips | | Fade buy | 15-minute | 6 | 70→50 | +1,114.2 pips | +252.2 pips | +250.6 pips | **Two of five exceeded their 2025 median.** The fade buy cleared its median by only 1.6 pips: +252.2 against +250.6. The hourly midline made +102.0 against a +234.4 median, and the 15-minute midline made −1,449.8 against −1,081.4. ## Is counting bars worth more than measuring the move? The psychological line asks how many of the last N bars rose. A matched test asked **how much price rose over the same window**: the close against the close N bars back, which is momentum. The two answer differently. They disagreed on the direction on **23.2% to 25.9% of bars** (periods 12 and 26, both years, 15-minute and hourly). Of the bars where the count said up, 14.8% to 19.7% sat below where price was N bars earlier. The correlation between the two ran 0.58 to 0.69. So does counting help? The midline cross was run against the sign of momentum over the same period, across 36 cells of timeframe × year × period. | | Counting bars | Measuring the move | |---|---|---| | Trades (1-hour, 2025, period 12) | 402 | 795 | | Win rate | 38.06% | 34.97% | | Net pips | +272.6 | +1,535.3 | Over all 36 cells: - **Fewer trades in 36 of 36** - **Higher win rate in 34 of 36** - **More net pips in only 16 of 36** **Counting halves the trading and lifts the win rate, and then loses the coin flip on net pips.** The equivalent comparison in the CCI article came out at 25 of 36, so this does not even reach that. Throwing the size away does do something: a small move only ever counts as one vote, so the line reverses on noise less often. **Whether the trades it removes were the bad ones is not something this test could answer.** ## Filters, stops and cost ### ADX | Reading | Timeframe | Year | ADX condition | Trades | Net | |---|---|---:|---|---:|---:| | Midline | 1-hour | 2025 | None | 402 | +272.6 pips | | Midline | 1-hour | 2025 | ≥ 20 | 277 | −99.5 pips | | Midline | 1-hour | 2025 | ≥ 25 | 196 | −906.0 pips | | Midline | 1-hour | 2025 | ≥ 30 | 144 | −726.8 pips | | Midline | 15-minute | 2024 | None | 1,541 | +238.3 pips | | Midline | 15-minute | 2024 | ≥ 20 | 1,003 | +559.7 pips | | Midline | 15-minute | 2024 | ≥ 25 | 703 | +274.6 pips | | Midline | 15-minute | 2024 | ≥ 30 | 442 | **+747.6 pips** | **The direction does not hold.** Hourly 2025 gets worse, 15-minute 2024 gets better. No filter moved both years the same way. ### Session Tokyo hours (UTC 0–8) improved the midline in three of four cells. | Setting | Baseline | Tokyo only | London + NY (UTC 7–21) | |---|---|---|---| | Midline, 15-minute, 2025 | −1,099.2 | **+544.5** | −808.8 | | Midline, 15-minute, 2024 | +238.3 | **+649.8** | +218.8 | | Midline, 1-hour, 2025 | +272.6 | **+505.4** | −100.2 | | Midline, 1-hour, 2024 | +1,169.5 | +213.3 | +846.6 | **The one cell it hurt was the most profitable one.** London and New York hours improved one of four. Part of what Tokyo did was simply trade less: 15-minute 2025 fell from 1,551 trades to 605. ### Stops and targets On the hourly midline, a 30-pip stop with a 90-pip target took 2025 from +272.6 to +946.4 and 2024 from +1,169.5 to +1,971.2: **one of the few changes that helped in both years**. On 15-minute bars it lifted 2025 from −1,099.2 to −339.8, still a loss. ### Cost Net pips were exactly linear in the spread. **Pips lost = trade count × spread** held without error on all five settings. | Setting | Trades | Spread 0 | 0.3 | 1.0 | Break-even | |---|---|---|---|---|---| | Midline, 1-hour | 402 | +393.2 | +272.6 | −8.8 | **0.98 pips** | | Midline, 15-minute | 1,551 | −633.9 | −1,099.2 | −2,184.9 | never | | Fade sell, 1-hour | 28 | +236.7 | +228.3 | +208.7 | 8.45 pips | | Band break, 15-minute | 147 | −2,384.1 | −2,428.2 | −2,531.1 | never | **The hourly midline breaks even at 0.98 pips**, so it does not survive a one-pip spread. The fade sell's 8.45 looks comfortable until you notice it comes from 28 trades a year. Note that the **15-minute midline and band break lose money at zero spread**. Trade count is not the explanation there. ### Settings that never fire Push the level far enough and the line stops reaching it. Settings with fewer than five trades in 2025: | Timeframe | Band break | Fade | |---|---|---| | 15-minute | 2 of 30 | 18 of 180 | | 1-hour | 9 of 30 | 48 of 180 | | 4-hour | 11 of 30 | 74 of 180 | **On 4-hour bars, four out of ten fade settings produced no trades.** Period 20 with level 90 never fired on any timeframe, which follows from the ladder, since at period 20 a level of 90 asks for 19 bars out of 20. ## Related reading - [CCI settings](/blog/cci-settings): the same "is the indicator doing anything" test, where CCI's zero line turned out to be a moving average crossover - [RSI settings](/blog/rsi-settings): a comparison of fading a level with trading a cross through the selected level - [Bollinger band settings](/blog/bollinger-band-settings), where splitting a band into "fade the touch" and "follow the break" started - [Moving average cross settings](/blog/moving-average-cross-settings): the measuring-the-move side of the comparison above - [Heikin Ashi settings](/blog/heikin-ashi-settings): the same comparison again, where the transform did clear a coin flip on net pips - [Renko chart settings](/blog/renko-chart-settings): another indicator confined to a ladder, where the ladder's own position is not a setting either - [Is RCI Below -80 a Buy? 2,484 Backtests on USD/JPY](/blog/rci-settings): another reading that drops the size of the move, measured against the size it drops ## Notes - **One pair, two years.** The psychological line came out of equity markets, and a different instrument's bar-size distribution changes which rungs appear at all - **2024 rose 1,632 pips and 2025 fell 56.** The reversal share was positive only in 2025, and the result has been checked on these two years only - **Many fade settings barely trade** (about 25 a year on hourly bars, about 10 on 4-hour). Some show a 100% win rate off five trades, and 48 to 74 of 180 settings fired fewer than five times - **The spread is fixed at 0.3 pips.** The hourly midline breaks even at 0.98, so that assumption is doing real work - **Exits are on the opposite signal.** A 30-pip stop with a 90-pip target helped in both years; nothing else did - **The two up-bar definitions could not be separated on results**: 47.1% against 52.9% across 2,134 settings. They are different lines (0.68% to 1.24% of bars disagree), so the choice still needs to be made and kept --- ### SMA vs EMA: Which Moving Average Is More Profitable? 672 Tests URL: https://formiq.jp/blog/sma-vs-ema Language: en Published: 2026-08-28 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: moving average, SMA, EMA, indicator, parameters, backtest, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/sma-vs-ema Every guide to moving averages eventually reaches the same fork. Simple or exponential? The stock answers are "the EMA reacts faster, so use it for short-term trading" and "the SMA is smoother, so use it for the long term." Neither is usually accompanied by a number. So we ran the comparison directly: **the same period pair, the same timeframe, the same year, with only the averaging method swapped.** Fifty-six fast/slow combinations, three timeframes and four windows gives 672 head-to-heads on USD/JPY. **The EMA finished ahead in 318 of them: 47.32%.** There were no ties. Split into twelve cells, the EMA's win count ranged from 15 to 39 out of 56, and no cell separates the two methods. That is not the same as saying the choice does not matter. Swapping the method at fixed periods moved net pips by a median of 491.1. **The change is real; its direction is what cannot be predicted.** That combination is the finding, and the rest of this article is the mechanism behind it. Two properties turned out to belong to one method and not the other. One is the SMA's, one is the EMA's, and both are arithmetic rather than market behaviour. ## What the two averages actually weight Both are a weighted sum of past closes. The difference is how the weight is handed out. - **Simple (SMA)**: add the last n closes and divide by n. Every one of those n bars carries the same weight, and nothing older is read at all. - **Exponential (EMA)**: weight the newest bar most and every older bar by a constant fraction less. The weight shrinks but never reaches zero. The weights can be recovered rather than assumed. Build a flat series, move a single bar up by one unit, and whatever comes out of the average is that bar's weight at each age. Running that through the same `calcSMA` and `calcEMA` the backtests use gives this: | Period | Newest bar, SMA | Newest bar, EMA | Ratio | Centre of gravity | EMA weight inside its own period | How far back the EMA reads | |---|---|---|---|---|---|---| | 10 | 10% | 18.182% | 1.818× | 4.5 bars, both | 86.56% | 69 bars | | 20 | 5% | 9.524% | 1.905× | 9.5 bars, both | 86.49% | 139 bars | | 25 | 4% | 7.692% | 1.923× | 12 bars, both | 86.48% | 173 bars | | 50 | 2% | 3.922% | 1.961× | 24.5 bars, both | 86.47% | 346 bars | | 200 | 0.5% | 0.995% | 1.99× | 99.5 bars, both | 86.47% | 1,382 bars | ("How far back the EMA reads" is the oldest bar still carrying more than a millionth of the newest bar's weight.) The quiet row is the middle one. **The centre of gravity is identical at every period.** For an n-period average of either kind it lands on (n−1)/2 bars back: 9.5 for period 20, 99.5 for period 200, matching to the decimal in all seven periods tested. So the EMA is not reading newer information on average. The *average age* of what it weights is the same. What differs is the spread: it doubles the weight on today and pays for that by keeping weight outside its own window. A 20-period EMA holds 86.49% of its weight inside 20 bars and scatters the other 13.51% across older ones, and that 86% figure barely moves with the period (86.47% to 86.83%). Both descriptions refer to the same weight profile: a 20-period EMA assigns 1.905 times as much weight to the newest bar as the SMA does, while leaving 13.51% of its weight outside the stated 20-bar period. ## Does the EMA really react faster? It depends on the measurement, and on short pairs it did not. If the centre of gravity is shared, where does the speed show up? Three measurements, three different answers. **Distance from the close.** How far the line typically sits from price, hourly USD/JPY, 2025: | Period | SMA | EMA | Ratio | |---|---|---|---| | 5 | 13.81 pips | 11.34 pips | 0.8214 | | 10 | 22.11 pips | 18.63 pips | 0.8428 | | 20 | 33.55 pips | 28.46 pips | 0.8483 | | 25 | 38.08 pips | 32.34 pips | 0.8493 | | 50 | 55.10 pips | 46.68 pips | 0.8472 | | 200 | 97.66 pips | 85.62 pips | 0.8767 | The EMA is closer to price in every one of the 42 cells measured (three timeframes × two years × seven periods), by a ratio between 0.8126 and 0.8937. Around 85%, near enough, whatever the period and timeframe. That is one real sense of "faster." **How often the two lines cross.** Same periods, one method against the other: | Timeframe | Pair | SMA crossings | EMA crossings | |---|---|---|---| | M15 | 10 / 20 | 1,402 | 1,053 | | M15 | 20 / 50 | 524 | 470 | | H1 | 10 / 20 | 358 | 264 | | H1 | 25 / 75 | 99 | 75 | | H4 | 10 / 20 | 87 | 75 | | H4 | 50 / 200 | 14 | 8 | (USD/JPY, 2025) **The supposedly faster method crosses less often.** Across 36 cells the simple pair crossed more in 31, fewer in 3, and the same number in 2. The next section explains why. **Which crossover comes first.** Matching same-direction crossovers closest-pair-first, so that neither method's extra crossings drag the answer: | Timeframe | Pair | EMA first | Median gap | |---|---|---|---| | M15 | 10 / 20 | 51.78% | 1 bar earlier | | M15 | 50 / 200 | 66.4% | 3 bars earlier | | H1 | 5 / 20 | 49.02% | level | | H1 | 50 / 200 | 70% | 6.5 bars earlier | | H4 | 12 / 25 | 50.79% | 1 bar earlier | | H4 | 20 / 50 | 76.47% | 4 bars earlier | (USD/JPY, 2025) Over all 36 cells the EMA crossover came first between **41.36% and 76.47%** of the time, above half in 26 of them. It genuinely leads on long pairs like 50/200. On a 10/20 pair, which method fires first is close to a coin toss. **"The EMA gives you the signal earlier" did not hold on short pairs.** On 50/200, five of the six timeframe-and-year cells put the median lead at 3 to 7 bars; the sixth (four-hour, 2025, where the EMA pair crossed 8 times) came in half a bar late. ## Why the two cross at different rates Only the simple average turns against price. The crossing-count gap has an exact cause. One bar of SMA movement is this, and it is an identity rather than an approximation: ``` SMA today − SMA yesterday = (close today − close n bars ago) / n ``` Checked against the shipped function on three timeframes at periods 10, 20, 50 and 200, the largest disagreement over those twelve cells was under 0.0000000000037 pips. Rounding, and nothing else. Read what it says: **whether the SMA turns up or down is decided by comparing today's close to the close n bars ago.** That older bar is nowhere on the chart. The bar dropping out of the window has as much say in today's direction as the bar arriving. The EMA's step is a different identity, verified on the same twelve cells to the same precision: ``` EMA today − EMA yesterday = k × (close today − EMA yesterday) k = 2 / (period + 1) ``` The bracket is "is today's close above or below the line". So **the EMA turns up whenever the close is above it and down whenever it is below, always.** That difference is countable. Bars where the line turned one way while the close sat on the other side: | Timeframe | Period | Bars | SMA turned against price | Share | EMA | |---|---|---|---|---|---| | M15 | 20 | 24,903 | 4,007 | 16.09% | 0 | | M15 | 200 | 24,903 | 4,488 | 18.02% | 0 | | H1 | 20 | 6,226 | 992 | 15.93% | 0 | | H1 | 200 | 6,226 | 1,296 | 20.82% | 0 | | H4 | 20 | 1,610 | 262 | 16.27% | 0 | | H4 | 50 | 1,610 | 345 | 21.43% | 0 | (USD/JPY, 2025) **Zero in all 42 cells for the EMA**, and it will be zero on any instrument and any year, because the definition forbids it. The SMA ran between 10.95% and 21.43%, roughly one bar in six. This drop-off is what inflates the SMA's crossing count. Two simple averages of different lengths shed different old bars at different moments, so they change direction for reasons that have nothing to do with each other, and they meet more often. Whether that helps or hurts is a separate question, and the backtests below are the answer to it. One thing that does *not* follow: the EMA changes direction more often, not less. In 41 of 42 cells it turned more frequently: at period 5 on 15-minute bars, 26.82% of bars against the SMA's 21.2%. **The EMA turns more; the EMA pair crosses less.** Two different measurements, both true. ## Is the EMA affected by your chart history? It is. The EMA has an instability the SMA does not. Today's value is built from yesterday's, so **wherever the calculation was started leaks into every value that follows.** Two EMAs seeded at different points, measured bar by bar: | Period | Gap ratio per bar | (n−1)/(n+1) | Half-life | To 1/10 | To 1/100 | SMA exact after | |---|---|---|---|---|---|---| | 10 | 0.818182 | 0.818182 | 3.5 bars | 12 bars | 23 bars | 9 bars | | 20 | 0.904762 | 0.904762 | 6.9 bars | 24 bars | 47 bars | 19 bars | | 50 | 0.960784 | 0.960784 | 17.3 bars | 58 bars | 116 bars | 49 bars | | 200 | 0.99005 | 0.99005 | 69.3 bars | 231 bars | 461 bars | 199 bars | The ratio matches the closed form exactly, **and it never reaches zero.** The SMA is identical from the moment its window clears the join, which is its period, and stays identical after that. In practical terms, here is what a chart draws on 1 January 2025 depending on how much history it loaded: | Timeframe | Bars of history | EMA(200) difference | Bars until under 0.1 pip | SMA(200) difference | |---|---|---|---|---| | M15 | 200 | 7.705 pips | 435 | 0 | | M15 | 1,000 | 0.001 pips | immediately | 0 | | H1 | 200 | 49.083 pips | 620 | 0 | | H1 | 600 | 0.380 pips | 134 | 0 | | H1 | 1,000 | 0.008 pips | immediately | 0 | | H4 | 200 | 47.173 pips | 616 | 0 | | H4 | 1,000 | 0.011 pips | immediately | 0 | **An hourly EMA(200) drawn from exactly 200 bars sits 49.083 pips away from the same EMA(200) with adequate history**, and takes 620 more bars (over a month of hourly data) to converge. The SMA(200) agrees from its very first value. Two things follow. Load three to five times the period in history before trusting a long EMA: at 1,000 hourly bars the error is down to 0.008 pips. And when two platforms disagree about where an EMA sits, neither is necessarily wrong: they may simply have started in different places. This is not unique to the EMA. Anything defined from its own previous value behaves this way; [Heikin Ashi](/blog/heikin-ashi-settings) candles halve their seed's influence every single bar. ## Adding a moving average to your chart Moving averages ship with every platform. What varies is the list of methods and the default. | Platform | How to add it | Methods offered | |---|---|---| | MT4 / MT5 | Drag Moving Average from the Navigator onto the chart | Simple, exponential, smoothed, linear weighted | | TradingView | Add Moving Average from the indicator search | Simple, exponential and others, depending on the script | | Browser (Formiq) | Add MA in the indicator settings; several lines can be stacked | Simple, exponential, weighted, smoothed, Hull | The line is drawn over the candles, not in a pane below. The settings dialog holds four fields: | Field | Default | What it changes | |---|---|---| | Period | 20 | How many bars are averaged. Longer sits further from price and turns less often | | Type | SMA | How the weight is handed out. This is the field the article is about | | Colour | assigned per line | Only matters once two or more lines are on the chart | | Line width | 1 | Display only; the calculation is unaffected | To turn two lines into a rule, the backtest conditions take a trigger and both legs. **The fast and the slow leg take their method separately**, which is measured later in this article. | Field | Meaning | |---|---| | Trigger | Crossover, or price against a single average | | Fast period and type | The quicker line | | Slow period and type | The slower line | The common rule is to buy when the fast line crosses above the slow one and sell when it crosses below: the golden cross and the dead cross. Every number here uses that rule. Only two fields are worth tuning: the period and the type. **The period is the decision that comes first**, and it is covered in [280 moving-average crossover settings tested](/blog/moving-average-cross-settings). Here the periods are held and only the type moves. ## How this was tested | Item | Value | |---|---| | Instrument | USD/JPY | | Period | 2025-01-01 to 2025-12-31, with 2024 and each half of 2025 as separate windows | | Timeframes | M15 / H1 / H4 | | Bars | 24,903 on M15, 6,226 on H1, 1,610 on H4 in 2025 | | Long | Close of the bar on which the fast average crosses above the slow one | | Short | Close of the bar on which it crosses below | | Exit | The opposite crossover, reversing on the same close. No stop, target or time exit | | Spread | 0.3 pips, no slippage, 0.1 lots, charged on both the exit and the new entry | | Grid | Eight fast periods (5, 8, 10, 12, 15, 20, 25, 50) × eight slow (20, 25, 30, 40, 50, 75, 100, 200), keeping the 56 where fast is shorter. Run in both methods and paired one to one, over three timeframes and four windows: 672 head-to-heads | | Control | Consecutive-integer ladders — fast 5 to 30 with slow fixed at 50, and slow 20 to 60 with fast fixed at 10 — in both methods. 780 rungs | | Measurement | Run through Formiq's backtester, with pips re-derived from each trade's fill prices | Fast and slow use the same method except in the mixing section. Holding to the opposite crossover with no stop keeps everything except the method out of the comparison; stops are added separately below. ## SMA or EMA: which should you use? The 672 head-to-head runs came out close to a draw. Same periods, same timeframe, same window, method swapped. This is the table the chart at the top draws. | Timeframe | Window | EMA ahead | EMA mean | SMA mean | Median gap | Worst | Best | |---|---|---|---|---|---|---|---| | M15 | 2025 | 29 / 56 | +1,134.4 | +1,064.1 | +16.3 | −1,396.5 | +1,767.2 | | M15 | 2024 | 26 / 56 | +1,529.4 | +1,632.0 | −127.3 | −2,201.6 | +1,457.7 | | M15 | H1 2025 | 31 / 56 | +213.5 | +76.8 | +92.3 | −1,035.4 | +1,549.7 | | M15 | H2 2025 | 20 / 56 | +903.9 | +978.1 | −280.4 | −1,160.1 | +1,934.7 | | H1 | 2025 | 26 / 56 | +76.7 | +62.9 | −142.4 | −1,337.9 | +2,121.2 | | H1 | 2024 | 32 / 56 | +1,503.2 | +1,198.7 | +241.1 | −1,761.6 | +3,542.2 | | H1 | H1 2025 | 39 / 56 | −46.9 | −265.1 | +203.1 | −651.9 | +1,705.4 | | H1 | H2 2025 | 15 / 56 | +179.9 | +363.8 | −218.9 | −1,055.2 | +1,465.6 | | H4 | 2025 | 22 / 56 | −1,410.4 | −1,162.9 | −195.8 | −3,096.9 | +1,807.8 | | H4 | 2024 | 35 / 56 | +2,429.9 | +1,965.0 | +284.6 | −1,114.5 | +4,517.7 | | H4 | H1 2025 | 27 / 56 | −214.7 | −189.3 | −60.7 | −1,573.9 | +1,809.5 | | H4 | H2 2025 | 16 / 56 | −1,075.0 | −857.3 | −187.4 | −1,624.5 | +598.1 | (Net pips) **318 of 672 across all twelve cells: 47.32%.** No pair produced identical net pips in the two methods. The two hourly half-year rows are worth reading together. The EMA produced the larger annualised net in 39 of 56 period pairs during the first half of 2025, but in only 15 of 56 during the second half on the same instrument and timeframe. A different question (how many settings finished both years in profit) gives a different-looking result: | Timeframe | Method | Up in 2025 | Up in 2024 | Both years | Median 2025 | Median 2024 | |---|---|---|---|---|---|---| | M15 | SMA | 51 / 56 | 53 / 56 | 48 / 56 | +1,004.1 | +1,650.5 | | M15 | EMA | 52 / 56 | 56 / 56 | 52 / 56 | +1,133.7 | +1,516.4 | | H1 | SMA | 35 / 56 | 44 / 56 | 28 / 56 | +371.6 | +1,201.0 | | H1 | EMA | 34 / 56 | 56 / 56 | 34 / 56 | +395.4 | +1,348.0 | | H4 | SMA | 3 / 56 | 53 / 56 | 2 / 56 | −1,224.6 | +2,129.6 | | H4 | EMA | 2 / 56 | 56 / 56 | 2 / 56 | −1,542.7 | +2,456.2 | By this count the EMA leaves slightly more standing (52 against 48, 34 against 28, and 2 against 2). **"More EMA settings survive both years" and "more EMA settings out-earn their SMA twin" are not the same statement**, and only the first one is true here. ## Does last year's winner win this year? For each period pair, take the method that won in 2024 and check whether it won again in 2025. | Timeframe | EMA won in 2024 | EMA won in 2025 | Same method both years | Agreement | Both halves of 2025 | |---|---|---|---|---|---| | M15 | 26 / 56 | 29 / 56 | 19 / 56 | 33.93% | 25 / 56 | | H1 | 32 / 56 | 26 / 56 | 26 / 56 | 46.43% | 30 / 56 | | H4 | 35 / 56 | 22 / 56 | 25 / 56 | 44.64% | 25 / 56 | **33.93% on 15-minute bars is below a coin flip.** Neither of the other two reaches 50% either, and cutting the year in half gives the same 25-to-30 range. ## How the common pairs performed The most-quoted period pairs, method by method. | Pair | Timeframe | Year | SMA annual net | EMA annual net | Larger annual net | |---|---|---:|---:|---:|---| | 5/20 | H1 | 2025 | −484.9 pips | +1,636.3 pips | EMA | | 5/20 | H1 | 2024 | +3,744.7 pips | +2,293.3 pips | SMA | | 10/20 | M15 | 2025 | +1,015.0 pips | +2,052.3 pips | EMA | | 10/20 | M15 | 2024 | +1,600.9 pips | +1,183.0 pips | SMA | | 10/20 | H4 | 2025 | −1,336.2 pips | −664.6 pips | EMA | | 10/20 | H4 | 2024 | −1,245.7 pips | +3,272.0 pips | EMA | | 12/25 | M15 | 2025 | +2,572.9 pips | +1,436.6 pips | SMA | | 12/25 | M15 | 2024 | +1,894.7 pips | +2,102.1 pips | EMA | | 20/50 | H1 | 2025 | −220.5 pips | +720.1 pips | EMA | | 20/50 | H1 | 2024 | −34.6 pips | +944.5 pips | EMA | | 25/75 | M15 | 2025 | −1,023.1 pips | +691.3 pips | EMA | | 25/75 | M15 | 2024 | +3,613.7 pips | +2,753.1 pips | SMA | | 50/200 | H1 | 2025 | −1,340.4 pips | −943.0 pips | EMA | | 50/200 | H1 | 2024 | +2,435.4 pips | +1,541.5 pips | SMA | (Seven of the eighteen combinations.) Over all eighteen (six pairs × three timeframes) the EMA was ahead in 11 in 2025, 9 in 2024, and **in both years in 3.** The 15-minute 10/20 is the clearest case. In 2025 the EMA produced more than 1,000 pips above the SMA; in 2024 the SMA produced more than 400 pips above the EMA on the same currency pair. ## Does the averaging method matter more than the period? "The EMA wins 47.32% of the time" is also what you would see if the method changed nothing at all. Those two situations need separating, and separating them takes a control. So: two ladders of consecutive integers. Fast 5 through 30 with the slow leg pinned at 50, and slow 20 through 60 with the fast leg pinned at 10. At every rung, compare how far swapping the method moved net pips against how far moving the period by a single bar moved it. | Timeframe | Ladder | Median swap | Median one-bar step | Swap moved more | |---|---|---|---|---| | M15 | fast 5–30 | 317.0 pips | 275.5 pips | 15 / 25 | | M15 | slow 20–60 | 322.6 pips | 173.7 pips | 26 / 40 | | H1 | fast 5–30 | 270.6 pips | 224.1 pips | 16 / 25 | | H1 | slow 20–60 | 601.0 pips | 146.9 pips | 35 / 40 | | H4 | fast 5–30 | 364.7 pips | 229.3 pips | 14 / 25 | | H4 | slow 20–60 | 374.9 pips | 122.9 pips | 34 / 40 | (The six 2025 rows; the totals below cover all 24 cells) **Across 780 rungs the swap moved the result further than the one-bar step in 593: 76.03%.** The medians are 350.7 pips for the swap against 154.2 for the step. Put in the period's own units: how many bars would the period have to move to shift the result as much as the method did? **The median answer is two.** One bar sufficed at 268 of 804 rungs, five or more were needed at 208, and at 70 rungs twenty bars still were not enough. So: **swapping SMA for EMA is worth about two bars of period. That is not nothing. Which way it moves you is unknowable in advance.** The [Fibonacci retracement test](/blog/fibonacci-retracement-levels) needed the same control-group design for a different reason. When there are only two candidates, counting how often one wins proves nothing on its own. A control beside it is what makes the count mean something. ## Mixing the two methods across the two lines The fast and slow legs can take different methods, so six pairs × four combinations × two years = 36 cells. | Timeframe | Window | Pair | Both SMA | Both EMA | Fast EMA, slow SMA | Fast SMA, slow EMA | |---|---|---|---|---|---|---| | M15 | 2025 | 10 / 20 | +1,015.0 | +2,052.3 | +922.3 | +1,549.2 | | M15 | 2025 | 20 / 50 | +1,774.9 | +1,108.3 | +2,139.3 | +149.2 | | H1 | 2025 | 12 / 25 | +958.0 | +970.9 | +1,431.9 | +917.7 | | H1 | 2025 | 25 / 75 | −56.8 | +708.9 | +995.9 | −270.2 | | H1 | 2024 | 12 / 25 | +1,882.8 | +440.6 | +2,169.5 | −75.8 | | H4 | 2024 | 10 / 20 | −1,245.7 | +3,272.0 | +231.8 | +2,692.6 | (Net pips. Six of the 36 cells) All four combinations landed on the same side of zero in 21 of 36 cells, and the spread between the best and worst of the four ran to a median of 1,399.7 pips, widest 4,517.7. **Mixing moves the result substantially.** What it does *not* do is swap the roles of the two lines. [The moving-average crossover article](/blog/moving-average-cross-settings) found that a 20-period Hull average sits closer to price than a 10-period simple one, so giving the slow leg a Hull average makes the "slow" line the fast one. An EMA only closes about 15% of the distance, which is not enough: | Timeframe | Pair | Period ratio | SMA on the fast leg | EMA on the slow leg | Slow line closer? | |---|---|---|---|---|---| | H1 | 10 / 20 | 2.0× | 22.11 pips | 28.46 pips | no | | H1 | 20 / 50 | 2.5× | 33.55 pips | 46.68 pips | no | | H1 | 50 / 200 | 4.0× | 55.10 pips | 85.62 pips | no | (USD/JPY, 2025) In all eighteen combinations the slow EMA stayed further from price than the fast SMA. **The inversion needs the periods within 1.3× to 1.45× of each other** (holding the fast leg at SMA 20 and lengthening the slow one bar at a time, the boundary fell at 1.3× on M15 and H1, and 1.45× on H4). It is worth checking only for pairs as close together as 20 and 25. ## Do trades and win rate change? Both change, and the lower win rate kept the better net. | Pair | Timeframe | SMA trades | EMA trades | SMA win rate | EMA win rate | |---|---|---|---|---|---| | 5 / 20 | M15 | 1,616 | 1,513 | 33.11% | 28.42% | | 10 / 20 | M15 | 1,402 | 1,053 | 37.52% | 29.44% | | 12 / 25 | M15 | 1,106 | 849 | 37.25% | 30.04% | | 10 / 20 | H1 | 358 | 264 | 39.11% | 32.58% | | 12 / 25 | H1 | 274 | 204 | 41.97% | 36.27% | | 25 / 75 | H1 | 99 | 75 | 44.44% | 42.67% | | 50 / 200 | H4 | 14 | 8 | 50.00% | 37.50% | (USD/JPY, 2025) Over all 672 head-to-heads the EMA took **fewer trades in 405** and had a **lower win rate in 561**. Among the eighteen named combinations its win rate was lower in all eighteen. Net pips still finish level because the EMA's average trade is bigger. On the 15-minute 10/20 the SMA made +1,015.0 pips over 1,402 trades (+0.72 each) and the EMA +2,052.3 over 1,053 (+1.95 each). **The SMA had the higher win rate in 561 of 672 comparisons, while annual net was split 354 to 318 in the SMA's favour.** The same mismatch runs through this series: [Williams %R](/blog/williams-r-settings) wins 55–79% of its trades and loses money. The full comparison contains the return, trade count and win-rate differences: | | Simple (SMA) | Exponential (EMA) | |---|---|---| | Centre of gravity | (n−1)/2 bars back | (n−1)/2 bars back — identical | | Weight on the newest bar | 1/n | about 2/n | | Reads past its own period | no | yes, 13.5% of its weight | | Average distance from the close | baseline | 0.81 to 0.90 of it | | Can turn against the close | yes, on 10.95–21.43% of bars | never | | Depends on where it was started | no, exact after n bars | yes, halving every ~0.35 × period bars | | Trade count | higher, in 405 of 672 | lower | | Win rate | higher, in 561 of 672 | lower | | Net pips | 354 of 672 | 318 of 672 | ## Filters, stops and targets | Timeframe | Fast/slow | Method | Year | Filter | Trades | Win rate | Net pips | Per trade | |---|---|---|---:|---|---:|---:|---:|---:| | M15 | 10/20 | SMA | 2025 | none | 1,402 | 37.52% | +1,015.0 | +0.72 | | M15 | 10/20 | SMA | 2025 | ADX ≥ 25 | 354 | 34.75% | −271.0 | −0.77 | | M15 | 10/20 | SMA | 2025 | Tokyo hours | 460 | 38.91% | +309.7 | +0.67 | | M15 | 10/20 | EMA | 2025 | none | 1,053 | 29.44% | +2,052.3 | +1.95 | | M15 | 10/20 | EMA | 2025 | ADX ≥ 25 | 154 | 25.97% | −185.5 | −1.20 | | M15 | 10/20 | EMA | 2025 | Tokyo hours | 356 | 31.18% | +1,469.8 | +4.13 | | H1 | 12/25 | SMA | 2025 | none | 274 | 41.97% | +958.0 | +3.50 | | H1 | 12/25 | SMA | 2025 | ADX ≥ 25 | 80 | 37.50% | −400.4 | −5.00 | | H1 | 12/25 | EMA | 2025 | none | 204 | 36.27% | +970.9 | +4.76 | | H1 | 12/25 | EMA | 2025 | ADX ≥ 25 | 41 | 29.27% | −1,050.0 | −25.61 | (Ten of the 24 filter runs) **11 of 24 improved pips per trade.** The ADX filter improved 2 of its 8 and turned all four 2025 runs from profit into loss. The Tokyo-session filter improved 4 of 8. **Filters did not behave differently by method.** Where ADX hurt, it hurt both, and in the same direction. | Timeframe | Fast/slow | Method | Year | Exit | Trades | Net pips | |---|---|---|---:|---|---:|---:| | M15 | 10/20 | SMA | 2025 | crossover only | 1,402 | +1,015.0 | | M15 | 10/20 | SMA | 2025 | SL 50 / TP 100 | 1,372 | +535.1 | | M15 | 10/20 | EMA | 2025 | crossover only | 1,053 | +2,052.3 | | M15 | 10/20 | EMA | 2025 | SL 50 / TP 100 | 1,025 | +1,459.8 | | H1 | 12/25 | SMA | 2025 | crossover only | 274 | +958.0 | | H1 | 12/25 | SMA | 2025 | 24-bar time exit | 268 | +1,808.1 | | H1 | 12/25 | EMA | 2025 | crossover only | 204 | +970.9 | | H1 | 12/25 | EMA | 2025 | SL 50 / TP 100 | 185 | +1,789.8 | (Eight of the 32 exit runs) **6 of 32 beat holding to the opposite crossover**: three on the SMA side and three on the EMA side. No difference by method here either. ## The spread, and where each method breaks even | Timeframe | Fast/slow | Method | Trades | Spread | Annual net | |---|---|---|---:|---:|---:| | M15 | 10/20 | SMA | 1,402 | 0.0 pips | +1,435.6 pips | | M15 | 10/20 | SMA | 1,402 | 0.3 pips | +1,015.0 pips | | M15 | 10/20 | SMA | 1,402 | 0.6 pips | +594.4 pips | | M15 | 10/20 | SMA | 1,402 | 1.0 pips | +33.6 pips | | M15 | 10/20 | SMA | 1,402 | 1.5 pips | −667.4 pips | | M15 | 10/20 | SMA | 1,402 | 2.0 pips | −1,368.4 pips | | M15 | 10/20 | EMA | 1,053 | 0.0 pips | +2,368.2 pips | | M15 | 10/20 | EMA | 1,053 | 0.3 pips | +2,052.3 pips | | M15 | 10/20 | EMA | 1,053 | 0.6 pips | +1,736.4 pips | | M15 | 10/20 | EMA | 1,053 | 1.0 pips | +1,315.2 pips | | M15 | 10/20 | EMA | 1,053 | 1.5 pips | +788.7 pips | | M15 | 10/20 | EMA | 1,053 | 2.0 pips | +262.2 pips | | H1 | 12/25 | SMA | 274 | 0.0 pips | +1,040.1 pips | | H1 | 12/25 | SMA | 274 | 0.3 pips | +958.0 pips | | H1 | 12/25 | SMA | 274 | 0.6 pips | +875.8 pips | | H1 | 12/25 | SMA | 274 | 1.0 pips | +766.2 pips | | H1 | 12/25 | SMA | 274 | 1.5 pips | +629.2 pips | | H1 | 12/25 | SMA | 274 | 2.0 pips | +492.2 pips | | H1 | 12/25 | EMA | 204 | 0.0 pips | +1,032.1 pips | | H1 | 12/25 | EMA | 204 | 0.3 pips | +970.9 pips | | H1 | 12/25 | EMA | 204 | 0.6 pips | +909.7 pips | | H1 | 12/25 | EMA | 204 | 1.0 pips | +828.1 pips | | H1 | 12/25 | EMA | 204 | 1.5 pips | +726.1 pips | | H1 | 12/25 | EMA | 204 | 2.0 pips | +624.1 pips | Moving the spread from 0 to 2.0 pips costs the 15-minute SMA run 2,804.0 pips, which is exactly 1,402 × 2.0. The EMA loses 2,106.0 (1,053 × 2.0), the hourly SMA 547.9 against 274 × 2.0, and the hourly EMA 408.0 (204 × 2.0). **Cost equals trade count times spread holds through the method swap**, as it has in every article in this series. | Timeframe | Fast/slow | Method | Trades | Net pips at zero spread | Break-even spread | |---|---|---|---:|---:|---:| | M15 | 10/20 | SMA | 1,402 | +1,435.6 | 1.02 pips | | M15 | 10/20 | EMA | 1,053 | +2,368.2 | 2.25 pips | | H1 | 12/25 | SMA | 274 | +1,040.1 | 3.80 pips | | H1 | 12/25 | EMA | 204 | +970.9 | 5.06 pips | **The 15-minute 10/20 breaks even at 1.02 pips as an SMA and 2.25 as an EMA.** A third fewer trades buys more than twice the headroom. This is not an edge in expected value: that part is a tie. It is tolerance: when the spread widens, or trades cluster into expensive hours, the simple version goes underwater first. The EMA took fewer trades in 405 of 672 head-to-heads, so the direction generalises even if the size does not. What survives as a practical difference is trade count, and cost tolerance with it. The EMA trades less in 405 of 672 head-to-heads and breaks even at a wider spread. Since the result cannot decide it, **the spread you actually pay and how often you want to trade are the only grounds this test supports.** The period is the decision that comes first, so [the 280-setting crossover test](/blog/moving-average-cross-settings) is worth reading before this one. If you want to run the same comparison on your own pair and dates, [it can be done without writing code](/blog/backtest-without-coding). ## Notes - One instrument (USD/JPY) and two years. Nothing here promises the same on another pair or another year - Only simple and exponential were compared. Weighted, smoothed and Hull are in [the crossover article](/blog/moving-average-cross-settings) - The weights, the centre of gravity and the two step formulas follow from arithmetic and hold on any data. **The distances, crossing counts and results are USD/JPY over two years** - Entries and exits are taken at the bar close. Real fills differ - The spread is held at 0.3 pips throughout. Real spreads move with the session and the calendar - The seeding measurement is read at one moment, 1 January 2025. Another date gives a different size of error, though the decay rate is date-independent --- ### Alligator Periods and Shifts: 1,584 Settings Tested URL: https://formiq.jp/blog/alligator-settings Language: en Published: 2026-08-27 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Alligator, Bill Williams, moving average, trend, indicator, backtest, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/alligator-settings The Alligator is three smoothed moving averages, named the **jaw (13), teeth (8) and lips (5)**. When they are tangled the alligator is said to be sleeping; when they stack up in order and fan apart it is eating, and that is the trend you are supposed to ride. One thing separates it from other three-average rules: **each line is displaced forward: eight, five and three bars into the future.** Nineteen indicators into this series, nothing else has moved a line along the time axis. **The displacement is not a forecast. It is a delay.** A line drawn eight bars ahead carries the value it had eight bars ago, so reading it today means reading a staler average, never a newer one. And the second thing the measurement said before any trading happened: **the alligator is eating for four bars in five.** ## What the Alligator is made of All three lines are built the same way: 1. **Median price** = (high + low) ÷ 2 2. A smoothed moving average (SMMA) of it over 13, 8 or 5 bars 3. Each line drawn 8, 5 or 3 bars to the right Three arbitrary choices sit in there: the median price, the smoothed average, and the displacement. Each one is checkable. ### The shift is a lag The implementation is literal: compute the 13-bar SMMA at bar `i`, then draw that value at bar `i + 8`. Read the other way round, the jaw drawn at bar `i` is **the average computed at bar `i − 8`**. No future data enters. Only age does. Comparing the displaced lines against the undisplaced ones, shifted back by their own displacement: | Timeframe | Year | Values compared | Mismatches | |---|---|---|---| | M15 | 2024 | 74,958 | 0 | | H1 | 2024 | 18,711 | 0 | | H4 | 2024 | 4,809 | 0 | | M15 | 2025 | 74,670 | 0 | | H1 | 2025 | 18,639 | 0 | | H4 | 2025 | 4,791 | 0 | **A displaced line is an undisplaced line fetched from that far back.** Everything below is about whether the delay is worth anything. ### Median price, and the smoothing Swapping the two other choices out, with no displacement, in 2025: | What changed | Correlation of the lips | Bars where the stacking order disagrees | |---|---|---| | Median price → close | 1.0000 to 0.9999 | 3.7% to 4.6% | | SMMA → EMA | 0.9998 to 0.9999 | 26.7% to 28.5% | | SMMA → SMA | 0.9989 to 0.9999 | 41.4% to 45.2% | **A correlation of 0.999 and an order that disagrees on nearly half the bars.** The three lines sit close together, so lines that overlap almost perfectly can still swap places constantly. Reading the correlation and concluding "close enough, the setting doesn't matter" gets this exactly backwards. The price input barely matters (3.7–4.6%); the averaging method matters a lot (26.7–45.2%). **The smoothed average is the choice doing work here.** ## Setting it up and reading it It ships with everything. Adding it draws **three lines over the price candles.** | Platform | Where to add it | |---|---| | MT4 / MT5 | Insert → Indicators → Bill Williams → Alligator | | TradingView | Search the indicator list for Williams Alligator | | Browser (Formiq) | Pick "Alligator" from the indicator list | ### The eight fields in the settings dialog MT4 opens it pre-filled with these. **Every platform exposes the same eight.** | Field | Default | What it does | |---|---|---| | Jaw period | 13 | The slowest line: how many bars get averaged | | Jaw shift | 8 | How many bars to the right that line is drawn | | Teeth period | 8 | The middle line's period | | Teeth shift | 5 | The middle line's displacement | | Lips period | 5 | The fastest line's period | | Lips shift | 3 | The fastest line's displacement | | MA method | Smoothed | A smoothed average, slower to react than a simple one | | Apply to | Median price (HL/2) | The midpoint of the bar's range, not the close | MT4 colours them **jaw blue, teeth red, lips green** by default. The names are only a metaphor; the maths is identical for all three. **Slowest to fastest: jaw, teeth, lips.** ### Three lines, three market states Everything the Alligator says comes from where the three lines sit relative to each other. | State | What you see | Williams' name for it | |---|---|---| | Sleeping | The three lines tangled and flat | The alligator sleeps. Stay out | | Awakening | The lines starting to separate | The mouth opens. Get ready | | Eating | Lips above teeth above jaw, or the reverse, and spreading | The alligator eats. Ride it | **Eating upward** means lips (green) on top and jaw (blue) at the bottom; downward is the mirror. Where this article says "stacked in order", that is the state it means. ### How people trade it The most widely published rule is: **buy when the lips (5) cross above the teeth (8), sell when they cross below.** This article measures that as "lips crossing teeth". Two other readings are common: - **Ride the stack**: instead of the crossing moment, hold while the three stay in order - **Ride the break**: buy when a candle closes above all three lines, sell when it closes below The lips-teeth cross, three-line alignment, and price outside all three lines are tested as separate trading rules. ### Which fields are worth touching **Periods change how fast the lines move; shifts slide them sideways.** The method and the applied price are editable too, and the article compares those on the lines themselves further down: swapping the smoothing changes the stacking order on 26.7–45.2% of bars, swapping the price input on 3.7–4.6%. **Where the shifts can be set to zero, that is the "three plain moving averages" version.** This article compares the two. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Window | 2025-01-01 to 2025-12-31, with 2024 run identically for comparison | | Timeframes | 15-minute / 1-hour / 4-hour | | Readings | Lips crossing teeth / all three stacked in order / price closing outside all three | | Period sets | Fibonacci ladder 5/3/2, 8/5/3, **13/8/5**, 21/13/8, 34/21/13; controls 12/7/4, 12/7/6, 14/9/4, 14/9/6; jaw-only 11/8/5, 16/8/5 — eleven in all | | Displacement | none 0/0/0, half 4/3/2, **published 8/5/3**, double 16/10/6 | | Combinations | 132 per cell; 3 timeframes × 4 windows = 1,584 runs | | Exit | The opposite signal; entry and exit share one rule set | | Stops and targets | None in the main sweep, measured separately below | | Spread | 0.3 pips fixed, filled at the close | | Size | 0.1 lot | All three readings are symmetric (each one opens longs and shorts off mirror conditions) so entry and exit share a rule set. ## How long does the alligator sleep? For 16% to 19% of the year: the mouth was open on more than 80% of bars in all six cells. The share of bars with the three lines stacked in one direction, lips through jaw: | Timeframe | Year | Stacked up | Stacked down | In order | Tangled | |---|---|---|---|---|---| | M15 | 2024 | 46.47% | 34.08% | 80.54% | 19.46% | | H1 | 2024 | 50.27% | 30.58% | 80.85% | 19.15% | | H4 | 2024 | 56.52% | 27.44% | 83.96% | 16.04% | | M15 | 2025 | 41.79% | 38.83% | 80.62% | 19.38% | | H1 | 2025 | 41.65% | 39.20% | 80.86% | 19.14% | | H4 | 2025 | 38.30% | 42.58% | 80.88% | 19.12% | **Over 80% in all six cells**, and 85.9–88.3% with the displacement removed. "Stay out while the alligator sleeps" means staying out for 16–19% of the year. That is not a selective filter. **The GMMA long ribbon was likewise in order on 85% of bars.** Count how often a state occurs before measuring the trades it separates. The up-and-down split tracks the market honestly. USD/JPY rose 1,632 pips in 2024 and the hourly split was 50.27% to 30.58%; 2025 closed 56 pips lower and the split was 41.65% to 39.20%. ## Should you remove the displacement? In 16 of 18 cells, removing it made more money. Same reading, same periods, displacement varied: | Reading | M15 | H1 | H4 | |---|---|---|---| | Lips crossing teeth | 13/22 | 19/22 | 20/22 | | All three in order | 14/22 | 5/22 | 12/22 | | Price outside all three | 1/22 | 8/22 | 11/22 | Each cell is 11 period sets × 2 years, counting how often **removing the displacement finished with more pips**. **Across all 198 it is 103: a coin toss.** The displacement has no general effect. But it splits by reading. **On the published lips-and-teeth entry it is 52 of 66 in favour of removing it.** On price-outside-all-three it is 20 of 66, so there the displacement earns its place. The concrete version, hourly chart, periods 13/8/5, lips crossing teeth: | Displacement | Year | Trades | Win rate | Net | |---|---:|---:|---:|---:| | None | 2025 | 298 | 34.56% | **+742.5 pips** | | None | 2024 | — | — | **+2,944.9 pips** | | Half (4/3/2) | 2025 | 374 | 44.39% | +397.9 pips | | Half (4/3/2) | 2024 | — | — | +2,462.2 pips | | Published (8/5/3) | 2025 | 404 | 43.07% | +115.5 pips | | Published (8/5/3) | 2024 | — | — | +973.7 pips | | Double (16/10/6) | 2025 | 370 | 48.11% | +332.1 pips | | Double (16/10/6) | 2024 | — | — | +1,298.8 pips | Williams' displacement produced less net profit than no displacement in both years. In 2025, trades rose from 298 to 404 and win rate from 34.56% to 43.07%, while net fell from +742.5 to +115.5 pips. The break-even spread makes it plainer. **With the shift, 0.59 pips. Without it, 2.79.** The first does not survive a 0.6-pip spread. ### Price alone says the same thing With no strategy in the way: every bar where the lips crossed the teeth, the move over the next 10, 20 and 50 bars, minus the unconditional drift over the same horizon. **The undisplaced lines gave the better forward move in 16 of 18 cells.** They also cross less often: 294 times against 410 on the hourly chart in 2024. Delaying both lines produces more crossings carrying less. ## Does the jaw period matter? The standard entry never reads it. "Buy when the lips cross the teeth" reads the lips and the teeth. The jaw is not in it. Three period sets sharing the teeth and lips, with the jaw moved: | Reading | 11/8/5 | 13/8/5 | 16/8/5 | |---|---|---|---| | Lips crossing teeth (H1, 2025) | +397 | +397 | +397 | | All three in order | +357 | +306 | +483 | | Price outside all three | +1,495 | +1,337 | +935 | **Identical to the pip on the first row.** The 13-bar line the indicator is named around plays no part in its most popular entry. The editor now hides the jaw period and jaw shift when the lips-teeth cross is selected, because neither parameter can change that rule's trades. ## Do the Fibonacci numbers matter? Jaw 13, teeth 8, lips 5; shifts 8, 5, 3. Every one is a Fibonacci number. The [Fibonacci retracement test](/blog/fibonacci-retracement-levels) used the same control method: build a group with **no Fibonacci number in it** and compare like against like. The controls here are 12/7/4, 12/7/6, 14/9/4 and 14/9/6: the four corners around Williams' teeth and lips. **Matched one-to-one across every trigger, timeframe, displacement and year, 13/8/5 finished ahead in 136 of 288.** Below half. | Timeframe | Year | Fibonacci ladder (5 sets) | Controls (4 sets) | |---|---|---|---| | M15 | 2025 | +435 | +865 | | M15 | 2024 | +937 | +1,178 | | H1 | 2025 | +707 | +734 | | H1 | 2024 | +1,399 | +1,293 | | H4 | 2025 | −982 | −952 | | H4 | 2024 | +1,709 | +2,055 | The controls are ahead in five of six; the lone exception is the hourly 2024 row, +1,399 against +1,293. **No effect specific to the Fibonacci numbers showed up**, matching the retracement article's result for the ratios. ## Which timeframe can you use it on? The hourly price-through-all-three reading was profitable in 44 of 44. Settings profitable in 2024 *and* 2025: | Reading | M15 | H1 | H4 | |---|---|---|---| | Lips crossing teeth | 35/44 | 39/44 | 4/44 | | All three in order | 41/44 | 27/44 | 1/44 | | Price outside all three | 22/44 | **44/44** | 9/44 | **Every hourly setting of the price-breakout reading made money in both years**: the most uniform result in this series. The four-hour chart inverted again: 40/44, 42/44 and 44/44 in 2024 became 4/44, 1/44 and 9/44 in 2025. GMMA and CCI both did this too. ### Except in the first half of 2025 The same counts with 2025 split down the middle, hourly chart: | Reading | Window | Profitable settings | |---|---|---:| | Lips crossing teeth | 2024 | 42/44 | | Lips crossing teeth | 2025 | 41/44 | | Lips crossing teeth | First half 2025 | **11/44** | | Lips crossing teeth | Second half 2025 | 44/44 | | All three in order | 2024 | 34/44 | | All three in order | 2025 | 31/44 | | All three in order | First half 2025 | **10/44** | | All three in order | Second half 2025 | 34/44 | | Price outside all three | 2024 | 44/44 | | Price outside all three | 2025 | 44/44 | | Price outside all three | First half 2025 | **18/44** | | Price outside all three | Second half 2025 | 43/44 | **44 of 44 is an 18-of-44 half added to a 43-of-44 half.** "Profitable in both years" and "profitable throughout" are different claims. Concretely: 13/8/5 with the published shift, price outside all three, hourly: 361 trades and +969.3 pips over 2025, made of **−673.5 over 212 trades in the first half and +1,618.7 over 150 in the second.** All three readings decline together, so this is not specific to one reading. The annual figure hides the difference between the two halves. ## Why the lowest win rate won Its mean win ran 2.6 times its mean loss. Hourly chart, 2025, settings with at least one trade: | Reading | Mean win rate | Mean win | Mean loss | Mean hold | Profitable | |---|---|---|---|---|---| | Lips crossing teeth | 42.8% | +56.5 | −38.6 | 26.7 bars | 41/44 | | All three in order | 41.1% | +76.8 | −50.7 | 51.9 bars | 31/44 | | Price outside all three | 30.4% | +76.7 | −29.2 | 24.6 bars | 44/44 | **Price outside all three has the lowest win rate and 44 of 44 profitable settings**, because its average win is 2.6 times its average loss. ## Does a wider mouth precede a bigger move? Entry bars were split into quartiles by the distance between the outermost lines, then measured by the absolute move over the next 20 bars: | Timeframe | Narrowest | Q2 | Q3 | Widest | All bars | |---|---|---|---|---|---| | M15 | 24.69 | 28.39 | 28.86 | **34.75** | 29.17 | | H1 | 57.53 | 61.91 | 59.42 | **73.06** | 62.98 | | H4 | 116.48 | 125.60 | 127.92 | **133.64** | 125.92 | **The widest quartile leads on all three timeframes**, and the narrowest is below the all-bars average everywhere. Read it carefully, though. The lines are apart *because price just moved*, so this is volatility persisting, not direction being predicted, and the edge is +19% on the 15-minute chart. ## Filters, stops and cost ### An ADX filter breaks it, for the fourth time | Reading | Displacement | Timeframe | Year | ADX condition | Trades | Net | |---|---|---|---:|---|---:|---:| | Lips × teeth | On | H1 | 2025 | None | 404 | +115.5 pips | | Lips × teeth | On | H1 | 2025 | ≥ 20 | 182 | +121.8 pips | | Lips × teeth | On | H1 | 2025 | ≥ 25 | 100 | **−1,175.1 pips** | | Lips × teeth | On | H1 | 2025 | ≥ 30 | 50 | −695.5 pips | | Lips × teeth | Off | H1 | 2025 | None | 298 | +742.5 pips | | Lips × teeth | Off | H1 | 2025 | ≥ 20 | 128 | −116.2 pips | | Lips × teeth | Off | H1 | 2025 | ≥ 25 | 73 | −137.6 pips | | Lips × teeth | Off | H1 | 2025 | ≥ 30 | 35 | −101.7 pips | | Price outside all three | On | H1 | 2025 | None | 361 | +969.3 pips | | Price outside all three | On | H1 | 2025 | ≥ 20 | 224 | +715.2 pips | | Price outside all three | On | H1 | 2025 | ≥ 25 | 161 | −568.6 pips | | Price outside all three | On | H1 | 2025 | ≥ 30 | 111 | −1,056.5 pips | Filtering a trend rule for trend duplicates the condition. GMMA and CCI found it first; this is the fourth time. ### Stops and targets **A 100-pip stop with a 200-pip target improved both years on the crossing entry**: no displacement went +742.5 → +2,242.8 in 2025 and +2,944.9 → +3,172.2 in 2024, and with the displacement +115.5 → +624.6 and +973.7 → +1,232.6. It does not carry to the other readings. Price-outside-all-three went +969.3 → +1,255.3 in 2025 but +2,161.6 → +2,002.2 in 2024. **Of the four settings measured, only the two crossing variants improved in both years.** ### Cost Net pips fall linearly with the spread. **Pips lost = trades × spread**, with no error. | Setting | Trades | Spread 0 | 0.3 | 1.0 | Break-even | |---|---|---|---|---|---| | Lips × teeth, displaced, H1 | 404 | +236.7 | +115.5 | −167.3 | **0.59 pips** | | Lips × teeth, no shift, H1 | 298 | +831.9 | +742.5 | +533.9 | **2.79 pips** | | Price outside all three, H1 | 361 | +1,077.6 | +969.3 | +716.6 | **2.99 pips** | **Removing the displacement moved the break-even from 0.59 pips to 2.79.** 106 fewer trades to pay for, and more pips in each one. ## Related - [GMMA line count](/blog/gmma-settings): its long ribbon is in order 85% of the time; the same base-rate problem - [Fibonacci retracements](/blog/fibonacci-retracement-levels), where the control-group method in this article comes from - [Moving average crossover periods](/blog/moving-average-cross-settings): how much the averaging method is worth on its own - [CCI and its ±100 band](/blog/cci-settings): the same approach of swapping out one arbitrary choice inside a formula at a time - [Ichimoku settings and its five lines](/blog/ichimoku-settings): another shifted indicator, translated back into rules without look-ahead bias ## Notes - **One pair, two years.** The Alligator came from commodity futures; on another market even the share of time the mouth is open would change - **Only four displacements were tried** (0/0/0, 4/3/2, 8/5/3, 16/10/6), not a continuous sweep - **The averaging method and price input were measured on the lines only, never traded.** Switching SMMA to EMA changes the stacking order on 27–29% of bars, so it must change the results too: this article does not have those numbers - **The spread is fixed at 0.3 pips.** A setting with 0.59 pips of room does not survive that assumption breaking - **The mouth-width result has no direction in it.** It is an absolute move, so it says "bigger", not "up" - **The four-hour samples are small** (25 to 290 trades a year depending on the reading). It reversed completely between 2024 and 2025, and two years cannot say which was the exception --- ### Bollinger Bands: 1, 2 or 3 Standard Deviations? Backtest Results URL: https://formiq.jp/blog/bollinger-band-sigma-settings Language: en Published: 2026-08-27 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: bollinger bands, standard deviation, sigma, breakout, mean reversion, backtest, USDJPY Also available in: ja — https://formiq.jp/ja/blog/bollinger-band-sigma-settings The Bollinger Band dialog has a box called “Deviation” or “Standard Deviations.” Two is the familiar default, but whether 1σ, 2σ or 3σ is best depends first on what you do at the band. We tested USD/JPY from 0.5σ through 4σ in 0.25 steps, crossed with seven periods from 10 to 50. A lower-band fade, an upper-band fade and a close-based breakout were run on 15-minute, hourly and four-hour bars across 2024, 2025, and the two halves of 2025: **3,780 backtests** in all. The short answer is that **1σ is a defensible starting point for an hourly breakout**. From 0.5σ through 1.25σ, all seven periods were profitable in both 2024 and 2025. At 1σ, the seven-period mean was +1,568.8 pips in 2024 and +1,081.9 in 2025. At 1.75σ it was +1,850.2 in 2025 but only +590.2 in 2024. That does not make 1σ universally optimal. It means that, among these candidates, **1σ made the best compromise between profit, survival across periods and transfer to another year**. The fade did not produce a repeatable “best sigma.” ## What Bollinger Bands sigma changes Let SMA(n) be the simple moving average of the last n closes, SD(n) their standard deviation and k the deviation setting. The bands are: **Upper = SMA(n) + k × SD(n)** **Lower = SMA(n) − k × SD(n)** The 2 in the settings dialog does not mean 2%. It multiplies the rolling standard deviation by two. Raising sigma pushes both bands away from the mean, producing fewer touches and fewer close-based breaks. Raising period also tends to reduce frequency because the centre line reacts more slowly. | Setting | Where it sits | Role in this hourly test | |---|---|---| | 1σ | Close to the mean | Breakout candidate with many signals | | 2σ | The conventional width | Many breakout settings survived both years | | 3σ | Far from the mean | Few trades, highly dependent on rare moves | | 4σ | Reached very rarely | Only a handful of trades per year | The often-quoted 68%, 95% and 99.7% figures for 1σ, 2σ and 3σ describe a normal distribution of independent observations. A market has serial dependence, trends and jumps, so those percentages are not market signal frequencies. On 6,226 hourly USD/JPY bars in 2025, using period 20, **45.0% of closes finished inside 1σ, 87.1% inside 2σ and 99.1% inside 3σ**. That leaves 12.9% outside 2σ. Merely reaching 2σ does not establish that price is abnormal or due to reverse. ## The settings dialog For the conventional 20-period, 2σ display in MT4 or MT5, the inputs are below. TradingView and Formiq expose the corresponding period and deviation fields. | Field | Example | What it controls | |---|---:|---| | Period | 20 | Lookback for the moving average and standard deviation | | Shift | 0 | Horizontal displacement; normally zero | | Deviations | 2.000 | Multiplier applied to standard deviation | | Apply to | Close | Price used in the calculation; this test uses closes | | Style / Color | Any | Display only; it does not change the calculation | Formiq's backtester adds a trigger choice. | Trigger | Rule | |---|---| | Band touch | Buy when the low reaches the lower band, or sell when the high reaches the upper band | | Band breakout | Buy on a close above the upper band, or sell on a close below the lower band | The same 20 and 2σ can give opposite results when a touch is faded and a close outside is followed. **The trading direction has to be fixed before sigma can be compared.** ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Windows | 1 Jan to 31 Dec 2025; 2024 and the two halves of 2025 as comparisons | | Timeframes | 15-minute / one-hour / four-hour | | 2025 bars | 24,903 M15, 6,226 H1, 1,610 H4 | | Periods | 10, 14, 20, 25, 30, 40, 50 | | Deviations | 0.5σ to 4σ in 0.25 steps, 15 values | | Fade long | Buy the close of a bar whose low touched the lower band; exit at the close of a bar touching the opposite 1σ band | | Fade short | Sell the close of a bar whose high touched the upper band; exit at the close of a bar touching the opposite 1σ band | | Breakout | Buy a close above the upper band, sell a close below the lower band, reverse on the opposite close | | Grid | 7 periods × 15 deviations × 3 trading rules = 315 settings, on 3 timeframes and 4 windows: 3,780 runs | | Costs | 0.3-pip spread, zero slippage, 0.1 lots | | Fills | Signal-bar close; opposite-side reversals may occur on the exit bar, while same-side re-entry waits for a later bar | | Stops and targets | None in the main grid; added separately below | | Measurement | Formiq's production backtester, with pips re-derived from every fill | The fade's opposite band stays fixed at 1σ so the experiment changes the entry sigma only. Moving the entry and exit together would not reveal which one caused the result. ## Which sigma works best? On the hourly chart the 0.5σ to 1.25σ band was the steadiest. Each row below averages its seven periods. “Positive both” counts the periods that made money in both 2024 and 2025. | Sigma | Mean pips 2025 | Mean pips 2024 | Mean 2025 trades | Positive both | |---:|---:|---:|---:|---:| | 0.5σ | +1,069.1 | +1,430.3 | 363.4 | 7 / 7 | | 0.75σ | +1,113.0 | +1,217.8 | 301.1 | 7 / 7 | | 1σ | +1,081.9 | +1,568.8 | 260.4 | 7 / 7 | | 1.25σ | +999.7 | +1,088.6 | 222.4 | 7 / 7 | | 1.5σ | +1,368.1 | +693.0 | 188.7 | 5 / 7 | | 1.75σ | +1,850.2 | +590.2 | 151.7 | 5 / 7 | | 2σ | +979.6 | +742.2 | 129.0 | 6 / 7 | | 2.25σ | −18.9 | +1,530.9 | 97.7 | 3 / 7 | | 2.5σ | −812.6 | +895.6 | 65.0 | 2 / 7 | | 2.75σ | −1,133.3 | +990.7 | 40.0 | 1 / 7 | | 3σ | −124.9 | +462.2 | 21.7 | 3 / 7 | | 3.25σ | +772.2 | +114.0 | 12.7 | 2 / 7 | | 3.5σ | +509.1 | +210.8 | 6.3 | 4 / 7 | | 3.75σ | +329.7 | +697.9 | 4.1 | 4 / 7 | | 4σ | +394.5 | +989.9 | 2.3 | 4 / 7 | All seven periods survived both years from 0.5σ through 1.25σ. **At 1σ, the means were +1,568.8 pips in 2024 and +1,081.9 in 2025.** Two sigma was also viable, with six of seven periods positive in both years, but its result depended more on period. The apparent return to four survivors from 3.5σ upward needs the trade column beside it. Those settings averaged 6.3, 4.1 and 2.3 trades in 2025. **A few successful outliers are not the same as a stable setting.** The whole 105-setting breakout grid changed with timeframe. | Timeframe | Year | Profitable settings | Median net | Positive in both years | |---|---:|---:|---:|---:| | M15 | 2025 | 55 / 105 | +138.5 pips | 44 / 105 | | M15 | 2024 | 65 / 105 | +404.3 pips | 44 / 105 | | H1 | 2025 | 75 / 105 | +672.9 pips | 67 / 105 | | H1 | 2024 | 86 / 105 | +902.5 pips | 67 / 105 | | H4 | 2025 | 32 / 105 | −270.8 pips | 27 / 105 | | H4 | 2024 | 84 / 105 | +926.6 pips | 27 / 105 | Hourly bars retained 67 settings in both years. Four-hour bars went from 84 winners in 2024 to 32 in 2025, which is why the answer here is explicitly an hourly one. ## Are 2 standard deviations any good? At period 20 alone, 1.75σ beat both 1σ and 2σ in both years. Fixing the period makes the frequency change easier to see. | Sigma | Year | Trades | Win rate | PF | Net | |---:|---:|---:|---:|---:|---:| | 1σ | 2025 | 277 | 38.27% | 1.137 | +1,052.5 pips | | 1σ | 2024 | 280 | 34.29% | 1.281 | +2,071.7 pips | | 1.75σ | 2025 | 158 | 44.94% | 1.348 | +1,846.0 pips | | 1.75σ | 2024 | 169 | 40.24% | 1.407 | +2,331.5 pips | | 2σ | 2025 | 132 | 43.18% | 1.235 | +1,177.5 pips | | 2σ | 2024 | 137 | 41.61% | 1.270 | +1,434.6 pips | | 2.75σ | 2025 | 48 | 41.67% | 0.703 | −955.6 pips | | 2.75σ | 2024 | 48 | 50.00% | 1.219 | +791.9 pips | | 3σ | 2025 | 24 | 37.50% | 1.017 | +33.7 pips | | 3σ | 2024 | 34 | 44.12% | 1.255 | +780.1 pips | Across all seven periods in 2024, however, 1.75σ averaged +590.2 pips versus +1,568.8 at 1σ. **A result after fixing one period is a different comparison from a result averaged across periods.** ## Which sigma for mean reversion? No sigma was stable for the fade. The 105 hourly settings point in a different direction when the band is faded. | Trading rule | Positive 2025 | Median | Positive 2024 | Median | Positive both | |---|---:|---:|---:|---:|---:| | Fade long | 43 / 105 | −114.4 pips | 48 / 105 | −20.5 pips | 23 / 105 | | Fade short | 11 / 105 | −483.9 pips | 25 / 105 | −995.2 pips | 1 / 105 | | Breakout | 75 / 105 | +672.9 pips | 86 / 105 | +902.5 pips | 67 / 105 | Widening the entry often made the fade-long loss smaller. Averaged over seven periods, 1σ made −811.6 pips in 2025 and −111.6 in 2024. At 2.75σ the figures were +230.2 and −45.3; at 3σ, +35.6 and −12.7. The fade-long result at 2.75σ was +230.2 pips in 2025 and −45.3 in 2024; at 3σ it was +35.6 and −12.7. The short fade retained just one of 105 settings in both years. **A wider band may simply trade less and therefore lose less; it did not establish a repeatable mean-reversion setting.** ## Does one year's best sigma hold? The single-year maximum shrank the next year, while 1σ held in both. Choosing sigma by the largest 2025 mean selects 1.75σ at +1,850.2 pips. In 2024 it made +590.2. The 1σ setting made +1,568.8 in 2024 and +1,081.9 in 2025. | Selection year | Sigma | Net in selection year | Check year | Net in check year | |---:|---:|---:|---:|---:| | 2025 | 1.75σ | +1,850.2 pips | 2024 | +590.2 pips | | 2024 | 1σ | +1,568.8 pips | 2025 | +1,081.9 pips | Optimizing period and sigma together is less stable still. Period 10 and 1.75σ made +2,753.0 pips in 2025 but +686.0 in 2024. Period 25 and 2.25σ made +2,793.5 in 2024 but +231.3 in 2025. That is why the conclusion favors several nearby settings that stayed positive in both years over one maximum cell. ## Can win rate choose sigma? No. Averaged over all 105 hourly settings in 2025, the fade long won 61.9%, the fade short 61.2% and the breakout 41.0%. | Trading rule | Mean win rate | Mean win | Mean loss | |---|---:|---:|---:| | Fade long | 61.9% | +39.2 pips | −65.9 pips | | Fade short | 61.2% | +40.2 pips | −84.8 pips | | Breakout | 41.0% | +188.8 pips | −93.0 pips | The fade is right more often but loses more when wrong. The breakout's average winner is about twice its average loser. Looking only for the sigma with the highest hit rate discards that payoff shape. ## Filters, exits and costs We added common filters to the period 20 and 1σ breakout used as the robust baseline. | Condition | Year | Trades | Annual net | |---|---:|---:|---:| | Plain | 2025 | 277 | +1,052.5 pips | | Plain | 2024 | 280 | +2,071.7 pips | | ADX ≥ 20 | 2025 | 188 | −1,033.2 pips | | ADX ≥ 20 | 2024 | 185 | −3.9 pips | | ADX ≥ 25 | 2025 | 134 | −1,572.2 pips | | ADX ≥ 25 | 2024 | 126 | +218.7 pips | | ADX ≥ 30 | 2025 | 94 | −972.7 pips | | ADX ≥ 30 | 2024 | 88 | −231.8 pips | | Tokyo hours | 2025 | 164 | +519.4 pips | | Tokyo hours | 2024 | 162 | +2,139.6 pips | | London and New York hours | 2025 | 207 | +1,073.5 pips | | London and New York hours | 2024 | 218 | +2,154.3 pips | None of the ADX thresholds beat the baseline in both years. London and New York hours improved it slightly in both, from +1,052.5 to +1,073.5 in 2025 and +2,071.7 to +2,154.3 in 2024, but that session result needs more pairs and years before it becomes a setting recommendation. | Exit | Year | Annual net | |---|---:|---:| | Opposite 1σ only | 2025 | +1,052.5 pips | | Opposite 1σ only | 2024 | +2,071.7 pips | | SL 30 / TP 60 | 2025 | +1,687.6 pips | | SL 30 / TP 60 | 2024 | +1,467.1 pips | | SL 50 / TP 100 | 2025 | +1,157.3 pips | | SL 50 / TP 100 | 2024 | +1,896.2 pips | | SL 100 / TP 200 | 2025 | +1,340.5 pips | | SL 100 / TP 200 | 2024 | +1,987.6 pips | | 24-bar time exit | 2025 | +989.1 pips | | 24-bar time exit | 2024 | +1,559.0 pips | Fixed stops and targets are executable levels measured from the filled entry. When reached, they fill at the configured distance without another exit-side adjustment being deducted from that level. Fixed stops and targets improved 2025 at the cost of 2024. None beat the plain rule in both years. With spread varied on the 2025 period 20 and 1σ breakout, net profit moved from +1,135.6 pips at zero to +1,052.5 at 0.3, +858.6 at 1, +581.6 at 2 and +304.6 at 3 pips. Its simple break-even spread was **about 4.10 pips**. One sigma trades often, so low cost is part of the setting. [Which Bollinger Bands period works best?](/blog/bollinger-band-period-settings) compares lookbacks from 5 through 200 and tests whether the best bar count changes with timeframe. The broader comparison between fading and following the bands is in [315 Bollinger Band backtests](/blog/bollinger-band-settings). [RSI periods and 30/70 levels](/blog/rsi-settings) also show why a higher fade win rate need not produce more profit. [MA disparity settings](/blog/ma-disparity-settings) put the band at a percentage of price rather than a standard deviation, while [backtesting without code](/blog/backtest-without-coding) shows how to rebuild the condition on your own pair. ## Notes - One pair, USD/JPY, and two independent calendar years. Nothing here guarantees the same order on other pairs or regimes - Picking a winner from a period-and-sigma grid creates a multiple-comparison uplift. The other year checks it, but is not an untouched future sample - Three sigma and wider settings have small samples. A 4σ mean of 2.3 trades cannot be compared with 260.4 trades at 1σ at equal confidence - Touch rules use highs and lows; breakouts use closes. Counting wick breaks as signals would produce a different test - Fills are at bar closes and spread is fixed. Live fills and changing spreads are not reproduced - The fade exit stays at the opposite 1σ band to isolate entry sigma. A study that optimizes the exit as well answers a different question --- ### Is the CCI ±100 Strategy Profitable? 2,448 Settings Tested URL: https://formiq.jp/blog/cci-settings Language: en Published: 2026-08-27 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: CCI, oscillator, mean reversion, trend following, indicator, backtest, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/cci-settings CCI (Commodity Channel Index) measures how far price has travelled from its own average, divided by how far it usually travels. Two lines are drawn at **+100 and −100**, and crossing them is described as the market having gone too far. What you do next is where the advice splits. One school **fades** the ±100 line: past it, take the other side. The other school **follows** it: a break through +100 is a buy. Same line, opposite trades, and the indicator's definition does not settle which is right. There is a quieter assumption underneath both. The ±100 lines are supposed to be **rare**. The 0.015 constant in the formula is usually explained as the number that keeps roughly 70–80% of readings inside the band. **The line sat outside ±100 for about 40% of the year.** ## What CCI actually measures Three steps: 1. **Typical price** = (high + low + close) ÷ 3 2. **Mean deviation** = the average distance of the last N typical prices from their own average 3. **CCI** = (typical price − average) ÷ (0.015 × mean deviation) The numerator is how far price has gone; the denominator is how far it normally goes. CCI is **a multiple of a normal move**. Three arbitrary choices sit inside that: the typical price, the mean deviation rather than a standard deviation, and the constant. Each one is checkable. ### Changing the 0.015 constant is equivalent to changing the level 0.015 is a divisor, so changing it rescales the whole line. **A different constant is the same indicator read at a different level**, not a different indicator. | Constant | Equivalent level at 0.015 | Largest error | |---|---|---| | 0.010 | 66.7 | 1.1e-13 | | 0.015 | 100 | 0 | | 0.020 | 133.3 | 5.7e-14 | | 0.030 | 200 | 0 | The errors are floating-point rounding. There are not two settings here; there is one. ### The period sets a hard ceiling on the level Deviations from a mean sum to zero, so the positive ones and the negative ones total the same amount. A single bar can hold at most all of the positive half, which caps |CCI| at **period ÷ 0.03**. | Period | Ceiling | Largest reading in 2025 (H1) | |---|---|---| | 7 | 233.3 | 233.3 | | 10 | 333.3 | 333.3 | | 14 | 466.7 | 466.7 | | 20 | 666.7 | 507.2 | | 30 | 1000.0 | 596.4 | | 50 | 1666.7 | 592.1 | **This is not a theoretical bound the market stays away from.** Nine of the eighteen timeframe-and-period cells hit their ceiling exactly during the year. The practical consequence: shorten the period and the higher levels stop existing. In this sweep, seven settings per cell (period 7 waiting at ±250) never fired once, in all twelve timeframe-and-window cells. **Period and level are not independent.** ### Mean deviation or standard deviation Swapping the denominator for a standard deviation (H1, 2025): | Period | Correlation | Bars that disagree about ±100 | Bars that disagree about zero | |---|---|---|---| | 7 | 0.9961 | 9.73% | 0% | | 14 | 0.9950 | 9.11% | 0% | | 50 | 0.9957 | 9.57% | 0% | **Not one bar disagrees about the zero line.** The denominator is never negative, so no measure of spread can change the sign. It only moves the scale, and there it flips the ±100 reading on about one bar in ten. ## Setting it up and reading it It ships with everything. CCI draws in **its own panel below the price chart**, not over the candles. | Platform | Where to add it | |---|---| | MT4 / MT5 | Insert → Indicators → Oscillators → Commodity Channel Index | | TradingView | Search the indicator list for Commodity Channel Index | | Browser (Formiq) | Pick "CCI" from the indicator list | ### The fields in the settings dialog | Field | Default | What it does | |---|---|---| | Period | 14 | How many bars go into the calculation. Some traders use 20 | | Apply to | Typical price (HLC/3) | The average of high, low and close — not the close | | Levels | −100 and +100 | Two horizontal lines drawn inside the panel; the decision points | **MT4 sometimes opens with no level lines at all.** Add −100 and 100 on the "Levels" tab, otherwise nothing in this article is visible on your screen. ### What appears One line moving up and down inside the panel. **Unlike RSI or the stochastic, it has no ceiling and no floor**: it can travel far above or below zero. | Where the line is | The usual reading | |---|---| | Above +100 | Overbought, or a strong push upward | | Between −100 and +100 | Neutral, nothing to do | | Below −100 | Oversold, or a strong push downward | | Crossing zero | The direction turned | The "or" is there because one rule fades readings outside ±100 while another trades a break through ±100. This article measures the fade and the ±100 breakout separately. ### How people trade it | Name | Entry | The idea | |---|---|---| | Fade | Buy below −100, sell above +100 | It went too far, so it comes back | | Breakout (±100) | Buy when it breaks above +100, sell when it breaks below −100 | Breaking out means momentum | | Zero cross | Buy crossing above 0, sell crossing below | The direction changed | **Looking at the same ±100 lines, the fade and the breakout place opposite trades.** The indicator's definition does not settle which is right, so both were run under identical conditions here. ### Period and level cannot be chosen independently Period and level look like two settings, but as the article shows below, **the period also caps the level**: a 7-bar CCI can never travel past ±233.3. Changing the applied price to the close gives a different line too: the ±100 state disagrees on 14.68% of bars at period 7 and 5.00% at period 50. Not a difference to wave away. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Window | 2025-01-01 to 2025-12-31, with 2024 run identically for comparison | | Timeframes | 15-minute / 1-hour / 4-hour | | Readings | Fade (take the other side past ±level) / breakout (take the side it broke) / zero cross | | Periods | 7, 10, 14, 20, 30, 50 | | Levels | Fade: entry 100/150/200/250 × exit 0/50/100. Breakout: 50/100/150/200/250. Zero cross: −100/−50/0/50/100 | | Combinations | 204 per cell; 3 timeframes × 4 windows = 2,448 runs | | Exit | The opposite signal. Only the fade carries a separate exit level | | Stops and targets | None in the main sweep, measured separately below | | Spread | 0.3 pips fixed, filled at the close | | Size | 0.1 lot | The fade points one way ("below −100 is a buy" only ever buys), so the **long fade and short fade were swept as separate trading rules**. The breakout and the zero cross are symmetric, so entry and exit share one rule set. ## What does price do after ±100? Before backtesting anything, look at price alone. No entry rule, no exit rule, so nothing is left but **whether ±100 says anything**. Every bar where a 14-period CCI broke out of ±100, and every bar where it came back inside, is collected, and the move over the next 10, 20 and 50 bars is measured. Raw moves would be meaningless: USD/JPY rose 1,632 pips in 2024, so any long-side number is mostly drift. The unconditional average move over the same horizon is subtracted, leaving **how much more the move was than an average bar's**. | Timeframe | After breaking out (+10 / +20 / +50 bars) | After coming back inside | Events | |---|---|---|---| | M15 | +0.46 / +0.93 / +2.04 | −0.86 / −1.01 / −1.92 | 5,776 | | H1 | +2.51 / +3.93 / +2.89 | −2.61 / −3.06 / −3.17 | 1,396 | | H4 | −3.89 / −3.14 / −9.79 | +2.91 / +5.02 / +12.50 | 346 | Both years and both directions pooled. **On the 15-minute and hourly charts, breaking out is followed by more move than usual and coming back inside by less.** Split by year, timeframe and horizon, the break-out cells beat the drift in 22 of 24 and the come-back cells fell short of it in 24 of 24. **On four-hour bars, moves after a ±100 breakout were below the unconditional average, while moves after a return inside ±100 were above it.** Those results held in only 6 of 12 and 5 of 12 year-direction-horizon cells, respectively, across 346 events. So ±100 leans towards *keep going*, not *too far*. At least on the two faster charts. ## Mean reversion or trend following? The fade lost in all six cells, while the breakout and zero cross mostly profited. Period 14 and level 100, read three ways. The fade column is the long and the short added together. | Timeframe | Year | Fade | ±100 breakout | Zero cross | |---|---|---|---|---| | M15 | 2025 | −1,175.2 | **+643.4** | −1,136.4 | | M15 | 2024 | −805.2 | −203.2 | −910.7 | | H1 | 2025 | −1,193.7 | **+424.8** | **+2,294.4** | | H1 | 2024 | −2,249.8 | **+1,839.9** | **+2,579.8** | | H4 | 2025 | −156.2 | **+577.5** | **+569.5** | | H4 | 2024 | −995.8 | **+863.8** | **+1,363.3** | **The fade lost in all six cells.** The breakout made money in five, the zero cross in four. The trading result agrees with the price-only measurement above. Trade counts, hourly 2025: the fade takes 210 longs and 212 shorts, the breakout 321, the zero cross 693. ### Why the fade loses Across the fade settings, **the average loss was larger than the average win despite the higher win rate.** | Reading | Timeframe | Mean win rate | Mean win | Mean loss | Profitable in 2025 | |---|---|---|---|---|---| | Fade, long | H1 | 59.9% | +44.4 | −68.7 | 27/69 | | Fade, short | H1 | 56.7% | +44.0 | −83.1 | 2/69 | | ±100 breakout | H1 | 40.1% | +101.6 | −71.8 | 20/29 | | Zero cross | H1 | 31.2% | +59.5 | −23.3 | 27/30 | Settings with at least one trade are included. The fade averaged a win rate above 60%, but its average loss was larger than its average win. The trend readings won less often and retained a larger payoff per winner. The short fade is the worst of the set: on the 15-minute chart, **0 of 69 live settings finished 2025 in profit**. That is not just the trend, either; 2025 closed 56 pips below where it opened. ## Which combination held for two years? The hourly zero cross led at 22 of 30. One year on its own is a selection problem, so here is the count of settings profitable in 2024 *and* 2025. | Reading | M15 | H1 | H4 | |---|---|---|---| | Fade, long | 2/72 | 8/72 | 22/72 | | Fade, short | 0/72 | 0/72 | 1/72 | | ±100 breakout | 17/30 | 20/30 | 11/30 | | Zero cross | 6/30 | 22/30 | 13/30 | **The hourly zero cross is the most durable at 22 of 30.** The short fade survives 1 of 216. The four-hour long fade is the one respectable fade number at 22/72, but 56 of 69 live settings were profitable in 2025 alone. **A reading that works for six settings in ten one year is not a reading that works.** ## What period should you use? The answer changes with the trading rule, so period alone does not settle it. Period results are split into the zero-cross, ±100 breakout, and fade rules because changing period affects their annual net and profitable-setting counts differently. **Zero cross, 15-minute chart, by period** | Period | Mean 2025 | Mean trades | Profitable 2025 | Profitable 2024 | |---|---|---|---|---| | 7 | −1,121 | 4,551 | 0/5 | 0/5 | | 14 | −730 | 3,030 | 0/5 | 0/5 | | 30 | +328 | 1,956 | 3/5 | 3/5 | | 50 | **+1,255** | 1,419 | 5/5 | 4/5 | **On the 15-minute chart the zero cross improves the longer you make it,** and the reason is the trade count. Period 7 at level 0 trades 4,533 times a year, which is 1,359.9 pips of spread at 0.3 pips. It lost 1,699.3, so **four fifths of the loss is the cost of trading at all.** The same rule at period 50 takes 1,469 trades and finishes +614.1 in 2025 and +189.5 in 2024. **±100 breakout, hourly, by level** | Level | Mean 2025 | Mean trades | Profitable both years | |---|---|---|---| | ±50 | +1,551 | 404 | 6/6 | | ±100 | +1,395 | 292 | 6/6 | | ±150 | +1,519 | 182 | 6/6 | | ±200 | −936 | 85 | 2/6 | | ±250 | −675 | 29 | 0/6 | Flat from ±50 to ±150, then it falls apart. ±250 fires 29 times a year. **Zero cross, hourly, by level** | Level | Mean 2025 | Mean 2024 | Profitable both years | |---|---|---|---| | −100 | +166 | +1,892 | 3/6 | | −50 | +1,325 | +2,309 | 6/6 | | 0 | +1,548 | +1,817 | 6/6 | | +50 | +1,763 | +653 | 4/6 | | +100 | +1,710 | −425 | 3/6 | **No reason to move the line off zero.** Only −50 and 0 are 6/6. The positive offsets look better in 2025 and collapse in 2024. ## Does one year's best carry over? Only two of four beat the median of the year they were moved into. The check at the centre of this series. | Reading | Timeframe | 2024 setting | 2024 net | 2025 net | 2025 median | |---|---|---|---:|---:|---:| | ±100 breakout | H1 | 20 @ ±100 | +3,393.7 pips | +1,048.5 pips | +1,113.0 pips | | Zero cross | H1 | 30 @ −50 | +3,046.7 pips | +1,729.0 pips | +1,294.1 pips | | ±100 breakout | M15 | 30 @ ±50 | +3,250.8 pips | +1,717.2 pips | +549.8 pips | | Fade, long | H4 | 10 / 200→100 | +946.1 pips | +176.7 pips | +387.5 pips | Two of the four exceeded the 2025 median. The hourly breakout finished 64.5 pips below its median, at +1,048.5 versus +1,113.0. The four-hour fade made +176.7 against a +387.5 median. For scale, the best four-hour breakout of 2025 was period 50 at ±250, +1,777.3 pips on **2 trades**. Numbers like that cannot choose a setting. ## What does the CCI zero line mean? It is a moving average crossover. CCI's denominator is 0.015 × mean deviation, and a mean deviation is never negative. So **the sign of CCI is exactly the sign of (typical price − its own simple moving average)**. Measured across periods 7 to 50: zero sign disagreements, zero crossing-bar disagreements. The best-performing reading in this article is therefore reproducible without CCI. What CCI contributes is the **typical price**. Comparing it against the identical rule reading the close (an SMA(1)/SMA(N) crossover, since SMA(1) is the close) over all 36 matched runs: | | CCI zero cross | Close / SMA cross | |---|---|---| | Trades (H1, 2025, period 14) | 693 | 885 | | Win rate | 30.45% | 26.44% | | Net pips | +2,294.4 | +1,801.6 | Across all 36: - **Fewer trades in 36 of 36** - **Higher win rate in 36 of 36** (by 1.5 to 8.0 points) - **More net pips in only 25 of 36** **(H+L+C)/3 filters whipsaw.** A close-only rule counts every bar that pokes through the average and comes back; averaging the bar's own range means it never poked through. That effect is completely consistent: no exceptions in 36 runs. But **some of the trades it removed were good ones**, which is what the other 11 runs are. The filtering is certain; the improvement is not. ## Filters, stops and cost ### An ADX filter breaks the trend readings The ADX filter reduced both trend-rule results. For the M15 breakout, every tested ADX threshold changed +643.4 pips into a loss. For the H1 zero cross, ADX ≥ 20 reduced +2,294.4 to +786.0 pips, with higher thresholds reducing it further. | Reading | Timeframe | Year | ADX condition | Trades | Net | |---|---|---:|---|---:|---:| | ±100 breakout | M15 | 2025 | None | 1,298 | +643.4 pips | | ±100 breakout | M15 | 2025 | ≥ 20 | 854 | −183.8 pips | | ±100 breakout | M15 | 2025 | ≥ 25 | 577 | −618.9 pips | | ±100 breakout | M15 | 2025 | ≥ 30 | 358 | −229.5 pips | | Zero cross | H1 | 2025 | None | 693 | +2,294.4 pips | | Zero cross | H1 | 2025 | ≥ 20 | 399 | +786.0 pips | | Zero cross | H1 | 2025 | ≥ 25 | 285 | +617.2 pips | | Zero cross | H1 | 2025 | ≥ 30 | 179 | −19.7 pips | **Adding a trend filter to a trend rule is a duplicated condition.** The GMMA test found the same thing; this is the third time in the series. The four-hour fade is the exception (2025 improves from +796.2 to +1,091.2 at ADX ≥ 30, on 16 trades) but **the same filter takes 2024 from +26.6 to −527.9.** It does not cross the year boundary. ### Sessions The Tokyo window (UTC 0–8) helped the breakout: M15 2025 +643.4 → +877.0, M15 2024 −203.2 → **+1,718.7**, H1 2025 +424.8 → +1,146.8. H1 2024 slipped from +1,839.9 to +1,746.0, so three of four. It did not help the zero cross (H1 2025 +2,294.4 → +571.3). ### Stops and targets On the hourly zero cross, a 100-pip stop with a 200-pip target lifts 2025 to +2,811.4 but drops 2024 to +2,095.4 against a +2,579.8 baseline. A 24-bar time exit gives +3,588.2 in 2025 and +2,303.0 in 2024. **Nothing improved both years.** ### Cost Net pips fall linearly with the spread. **Pips lost = trades × spread** held with no error on all four settings. | Setting | Trades | Spread 0 | 0.3 | 1.0 | Break-even | |---|---|---|---|---|---| | ±100 breakout, M15 | 1,298 | +1,032.8 | +643.4 | −265.2 | 0.80 pips | | ±100 breakout, H1 | 321 | +521.1 | +424.8 | +200.1 | 1.62 pips | | Zero cross, H1 | 693 | +2,502.2 | +2,294.4 | +1,809.3 | **3.61 pips** | **The hourly zero cross breaks even at 3.61 pips, the widest margin in this series.** The 15-minute breakout has 0.80 pips of room and does not survive a one-pip spread. ## Related - [RSI period and levels](/blog/rsi-settings): a comparison of fading a selected level with trading a cross through it - [Bollinger Band deviation](/blog/bollinger-band-settings): the first article to split a band into a fade and a breakout - [GMMA line count](/blog/gmma-settings): another trend rule that an ADX filter destroys - [The Alligator's forward shift](/blog/alligator-settings): the same one-choice-at-a-time approach, applied to a displacement along the time axis - [Moving average crossover periods](/blog/moving-average-cross-settings), where this article's zero cross ends up - [Psychological line settings](/blog/psychological-line-settings): the same "is the indicator doing anything" test, run on an indicator that discards the size of every move - [Is RCI Below -80 a Buy? 2,484 Backtests on USD/JPY](/blog/rci-settings): four of these five readings again, on a rank correlation - [Heikin Ashi settings](/blog/heikin-ashi-settings): the Heikin Ashi colour turns out to be a price/average cross too - [MA disparity settings](/blog/ma-disparity-settings): the same distance from an average, read as a raw percentage with no divisor in the way ## Notes - **One pair, two years.** CCI was built for commodity futures; on a different market even the share of time outside ±100 would change - **The spread is fixed at 0.3 pips.** In practice it widens around releases, and a setting with 0.80 pips of room does not survive that - **Exits are the opposite signal.** Adding stops and targets produces the separate numbers above, and none of them improved both years - **The four-hour samples are small**: about 40 breakout trades a year, about 20 for the fade. The four-hour chart also inverted in the price-only measurement, but on 6 of 12 cells, which decides nothing - **"The typical price helps" is consistent for trade count and win rate only.** For net pips it is 25 of 36, and whether that counts as helping depends on which column you read --- ### Parabolic SAR Settings by Timeframe: 36 Tests URL: https://formiq.jp/blog/parabolic-sar-settings Language: en Published: 2026-08-27 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Parabolic SAR, PSAR, indicator settings, trend following, backtest, USDJPY Also available in: ja — https://formiq.jp/ja/blog/parabolic-sar-settings Most Parabolic SAR settings guides begin and end with **0.02 and 0.20**. One number is labelled Step, the other Maximum, and both look precise enough to be settled facts. They are not equivalent controls. Step changes the starting speed and every later acceleration. Maximum only caps that acceleration. To separate them, I tested six values of each on USD/JPY: 36 combinations, three timeframes, two full years, and both halves of 2025. The useful headline is not one winning pair. **Hourly bars kept 23 of 36 settings positive in both full years.** On four-hour bars, the 0.02/0.20 default made +637.7 pips in 2025 and lost 108.2 pips in 2024. On hourly 2025, it lost 1,032.5 pips in the first half and made 1,847.7 in the second. ## What Parabolic SAR is measuring SAR moves a dot from its current position toward the extreme price reached in the active trend: **next SAR = current SAR + acceleration factor × (extreme price − current SAR)** In an up move, the extreme is the highest high since that move began. In a down move, it is the lowest low. Each new extreme raises the acceleration factor by Step until Maximum stops it. When price reaches the dot and the dot switches sides, the trend state reverses. At 0.02/0.20, the factor starts at 0.02. New extremes take it through 0.04, 0.06 and so on, reaching 0.20 after nine updates. The two settings therefore have distinct jobs: - **Step** sets the initial factor and the amount added after each new extreme - **Maximum** limits how fast the dot can catch price late in a sustained move This article tests flips at the bar close. It does not assume that an order filled at the plotted SAR price. That distinction matters whenever a bar passes through the dot before closing. ## How to read Parabolic SAR settings The indicator draws **one sequence of dots over the price chart**. It has no lower pane and no reference levels. | Platform | Where to add it | |---|---| | MT4 / MT5 | Insert → Indicators → Trend → Parabolic SAR | | TradingView | Search indicators for Parabolic SAR | | Formiq | Choose Parabolic SAR from the indicator list | ### Every field in the settings dialog | Field | Default | What it changes | |---|---:|---| | Step | 0.02 | Initial acceleration factor and the increment after each new extreme | | Maximum | 0.20 | Ceiling on the acceleration factor | The 0.02 is not 2% of price. It advances the dot by 2% of the remaining distance between the current dot and the extreme. At the 0.20 ceiling, that advance is capped at 20% of the remaining distance per bar. ### What appears on screen Dots below price are the up state; dots above price are the down state. Platforms may use different colors, but color is not part of the formula. There are no default 30/70-style levels to add. ### The trading rule tested here | Flip | Action | |---|---| | Dot moves from above price to below | Buy; hold until the dot flips above | | Dot moves from below price to above | Sell; hold until the dot flips below | That is the Stop and Reverse part of the name. The opposite signal closes the current position and starts one in the other direction. Long and short use the same rule. ### Which control is worth changing Raising Step makes the dots react sooner, so it also increases the number of reversals. Raising Maximum only matters after enough new extremes have pushed the factor toward that ceiling. The hourly 2025 sweep makes the difference visible. Mean trade count rose from 215.3 at Step 0.005 to 777.7 at 0.05. Maximum 0.30 averaged 608.8 trades and Maximum 0.50 averaged 609.8. **Step was the main frequency control.** ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Test windows | 2025-01-01 to 2025-12-31; 2024 under the same rules; 2025 split into first and second halves | | Timeframes | 15-minute / hourly / four-hour | | Step values | 0.005 / 0.01 / **0.02** / 0.03 / 0.04 / 0.05 | | Maximum values | 0.05 / 0.10 / 0.15 / **0.20** / 0.30 / 0.50 | | Grid | 6 steps × 6 ceilings = 36 settings; 432 runs across three timeframes and four windows | | Long entry | SAR flips from above price to below; fill at that bar's close | | Short entry | SAR flips from below price to above; fill at that bar's close | | Exit | Opposite flip, using the same rule set for entry and exit | | Stops and targets | Off in the sweep; tested separately below | | Cost | Flat 0.3-pip spread, zero slippage | | Size | 0.1 lot | | Measurement | Run through Formiq's backtest engine, with pips recalculated from every fill price | ## Which timeframe can you use? | Timeframe | Positive in 2025 | 2025 median | Positive in 2024 | 2024 median | Positive in both | |---|---:|---:|---:|---:|---:| | 15-minute | 11/36 | −433.3 pips | 21/36 | +248.8 pips | 11/36 | | **Hourly** | **29/36** | **+928.8 pips** | **30/36** | **+1,385.1 pips** | **23/36** | | Four-hour | 8/36 | −372.4 pips | 32/36 | +679.4 pips | 4/36 | Hourly bars have the broadest result, not merely the best outlier. Most settings were positive in each year, and 23 stayed positive in both. The 0.03/0.10 hourly setting made +3,910.5 pips in 2024 and +981.4 pips in 2025. A large profitable-setting count does not mean each setting will produce a similar annual net in both years. Four-hour bars changed from 32 positive settings in 2024 to eight in 2025. ## What do step and maximum change? The next table averages across all six Maximum values on hourly bars. | Step | Mean 2025 trades | Mean 2025 net | Mean 2024 net | |---:|---:|---:|---:| | 0.005 | 215.3 | +1,240.3 pips | −1,138.4 pips | | 0.01 | 342.5 | −144.9 pips | +1,535.3 pips | | **0.02** | 504.2 | +573.4 pips | +2,333.2 pips | | 0.03 | 611.5 | +1,246.6 pips | +2,335.3 pips | | 0.04 | 701.7 | +597.2 pips | +1,749.9 pips | | 0.05 | 777.7 | +1,459.8 pips | +923.4 pips | Taking Step from 0.005 to 0.05 raised frequency 3.6 times. Net pips did not rise with it. The slowest step averaged +1,240.3 pips in 2025 and −1,138.4 in 2024. Maximum has a saturation point in trade count. Its 2025 mean rose from 316.3 trades at 0.05 to 479.8 at 0.10, then flattened: 608.8 at 0.30 and 609.8 at 0.50. A high ceiling does little whenever the live acceleration factor rarely reaches it. ## Are the 0.02/0.20 defaults good? | Timeframe | Year | Trades | Win rate | PF | Annual net | |---|---:|---:|---:|---:|---:| | 15-minute | 2025 | 2,290 | 37.51% | 0.984 | −375.3 pips | | 15-minute | 2024 | 2,107 | 39.16% | 1.040 | +870.8 pips | | **Hourly** | **2025** | **547** | **40.04%** | **1.074** | **+886.8 pips** | | **Hourly** | **2024** | **531** | **38.79%** | **1.240** | **+2,683.6 pips** | | Four-hour | 2025 | 138 | 46.38% | 1.110 | +637.7 pips | | Four-hour | 2024 | 130 | 35.38% | 0.982 | −108.2 pips | Hourly is the only timeframe where the default made money in both years. Annual net ranged from +886.8 pips in 2025 to +2,683.6 in 2024. The four-hour default made +637.7 pips in 2025 and −108.2 pips in 2024. One year alone cannot settle whether to use it. ## Does one year's best carry over? Take each year's winner and move it untouched into the other year: | Timeframe | Selection year | Setting | Selection-year annual net | Other-year annual net | |---|---:|---|---:|---:| | 15-minute | 2025 | 0.02/0.05 | +1,147.3 pips | 2024: +350.9 pips | | 15-minute | 2024 | 0.005/0.05 | +2,852.8 pips | 2025: +678.8 pips | | Hourly | 2025 | 0.05/0.10 | +2,062.7 pips | 2024: +2,969.8 pips | | Hourly | 2024 | 0.03/0.10 | +3,910.5 pips | 2025: +981.4 pips | | Four-hour | 2025 | 0.02/0.20 | +637.7 pips | 2024: −108.2 pips | | Four-hour | 2024 | 0.005/0.05 | +2,369.4 pips | 2025: −452.2 pips | On 15-minute and hourly bars, the selected setting stayed profitable in the other year. On four-hour bars, both selected settings lost in the other year. One setting remaining profitable does not show that the other settings will transfer in the same way. The shorter split is also different. Hourly 0.02/0.20 made 281 trades and lost 1,032.5 pips in the first half of 2025, then made 1,847.7 pips over 266 trades in the second half. The full-year +886.8 pips combines two halves with opposite signs. ## Why a 40% win rate was enough The hourly default in 2025 made 547 trades at a 40.04% win rate and PF 1.074. Its average winner was +58.71 pips; its average loser was −36.49. The payoff gap, not a majority of winning trades, produced the +886.8 pips. The default won 38.79% in 2024 and made +2,683.6 pips, versus a higher 40.04% win rate and only +886.8 pips in 2025. A 1.25-point increase in win rate coincided with 1,796.8 fewer annual net pips. ## Does an ADX filter help? These filters apply only when entering the hourly default Parabolic SAR reversal rule. | Entry condition | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | SAR alone | 2025 | 547 | 40.04% | +886.8 pips | | SAR alone | 2024 | 531 | 38.79% | +2,683.6 pips | | ADX ≥ 20 | 2025 | 357 | 38.66% | −814.5 pips | | ADX ≥ 20 | 2024 | 351 | 36.47% | +979.4 pips | | ADX ≥ 25 | 2025 | 263 | 41.83% | −601.6 pips | | ADX ≥ 25 | 2024 | 243 | 38.27% | +140.2 pips | | ADX ≥ 30 | 2025 | 178 | 38.20% | −906.7 pips | | ADX ≥ 30 | 2024 | 158 | 38.61% | +431.5 pips | | Tokyo hours | 2025 | 207 | 42.51% | +158.2 pips | | Tokyo hours | 2024 | 202 | 38.12% | +585.2 pips | | London/NY hours | 2025 | 307 | 38.76% | +982.7 pips | | London/NY hours | 2024 | 311 | 40.84% | +2,465.3 pips | All three ADX thresholds turned 2025 negative and cut the 2024 result. Adding a second trend condition to a trend reversal signal removed useful entries. London/NY hours nudged 2025 up to +982.7 pips but pulled 2024 down to +2,465.3. No filter beat the baseline in both years. ## Do stops and timed exits help? | Exit | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | Opposite flip | 2025 | 547 | 40.04% | +886.8 pips | | Opposite flip | 2024 | 531 | 38.79% | +2,683.6 pips | | Stop 30 / target 60 | 2025 | 459 | 38.13% | +1,317.5 pips | | Stop 30 / target 60 | 2024 | 445 | 36.18% | +823.0 pips | | Stop 50 / target 100 | 2025 | 471 | 39.49% | +1,047.9 pips | | Stop 50 / target 100 | 2024 | 464 | 37.07% | +970.2 pips | | Stop 100 / target 200 | 2025 | 541 | 40.30% | +1,600.4 pips | | Stop 100 / target 200 | 2024 | 515 | 38.64% | +2,302.3 pips | | Stop 50 / target 50 | 2025 | 470 | 44.89% | +1,092.1 pips | | Stop 50 / target 50 | 2024 | 468 | 41.88% | +138.0 pips | | Time exit after 24 bars | 2025 | 539 | 40.07% | +1,187.7 pips | | Time exit after 24 bars | 2024 | 522 | 38.89% | +2,153.1 pips | Every exit variant stayed positive in both years. None beat the opposite-flip baseline in both. Stop 100 / target 200 raised 2025 to +1,600.4 pips but lowered 2024 to +2,302.3. ## How much spread can it take? Only the spread changes here: hourly 0.02/0.20, 2025. | Spread | Trades | Net pips | |---:|---:|---:| | 0.0 pips | 547 | +1,050.9 | | 0.3 pips | 547 | +886.8 | | 0.6 pips | 547 | +722.7 | | 1.0 pips | 547 | +503.9 | | 1.5 pips | 547 | +230.4 | | 2.0 pips | 547 | −43.1 | | 3.0 pips | 547 | −590.1 | The loss from zero spread is exactly trades × spread. Dividing the gross +1,050.9 pips by 547 trades gives a **1.92-pip break-even spread**. On 15-minute bars, the default traded 2,290 times and lost 375.3 pips. Once flips become that frequent, cost matters before fine differences between acceleration settings do. ## Related tests - [Supertrend settings](/blog/supertrend-settings): another flip-based trend rule, with period and multiplier separated over 36 settings - [Alligator settings](/blog/alligator-settings): three displaced lines used to read trend instead of a trailing dot - [Moving average crossover settings](/blog/moving-average-cross-settings): the same hold-to-opposite-signal structure applied to MA periods - [How to backtest without code](/blog/backtest-without-coding): build the conditions used in this article through the interface - [Do Indicator Settings Transfer Between Pairs? 3,024 Tests](/blog/settings-across-pairs): this setting won the 15-minute chart in 2023 on USD/JPY and ranked 24th, 34th and 36th of 42 on the other three pairs ## Notes - One pair and two full years: USD/JPY in 2024 and 2025 - Parabolic SAR implementations can differ in their initial direction, order of extreme updates and prior-bar clamps. These numbers use Formiq's implementation - Orders fill at the close of the flip bar, not at the plotted SAR value - Spread is flat at 0.3 pips and slippage is zero in the main sweep - The 36-setting grid brackets the common default; it does not test every possible decimal - The four-hour default made 130 to 138 trades a year, so its win rate and PF come from fewer trades than the hourly and 15-minute versions --- ### Does Fibonacci 61.8% Really Hold? 2 Years of USD/JPY Tested URL: https://formiq.jp/blog/fibonacci-retracement-levels Language: en Published: 2026-08-26 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Fibonacci, retracement, support and resistance, indicator, backtest, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/fibonacci-retracement-levels A Fibonacci retracement is a drawing tool. You anchor it to a swing low and a swing high, and it puts horizontal lines at 23.6%, 38.2%, 50%, 61.8% and 78.6% of that range. The premise is that **pullbacks tend to stop at those lines**. So which of the five stops price most often? This article measures it. Testing only those five would not answer the question because one of five levels must rank first even when none is special. Every depth from 10.0% to 90.0% was therefore measured in 0.2% steps, which is **401 levels**, to test whether the canonical ratios had higher reaction rates than neighbouring depths. The data is USD/JPY over 2024 and 2025 on the 15-minute, hourly and four-hour charts: 6,266 automatically detected swings. **Past the 30% mark, the share of pullbacks that ended at a level is flat between 3.5% and 4.4%.** The five Fibonacci ratios beat the depths within three points of them by an average of minus 0.037 points: no different from the six non-Fibonacci controls at plus 0.046. **Depth mattered. The ratio did not.** ## What a Fibonacci retracement measures The arithmetic is one subtraction and one multiplication. With a swing low L and a swing high H, the level at depth d sits at `H − (H − L) × d/100`. The 61.8% line is the price 61.8% of the range below the high. So the tool measures **how far price has given back a recent one-way move**, as a fraction: never as a distance. The swings this test detected had a median range of 38.4 pips on the 15-minute chart, 85.8 pips on the hourly and 178.7 pips on the four-hour, all for 2025. The same "61.8%" is a 24-pip pullback in one case and a 110-pip one in another. The two choices are **which depth to read** and **which two points anchor the tool**. This test repeated every depth with five swing definitions: a bar had to beat 3, 5, 8, 13, or 21 bars on each side to count as a swing point. Reach rate changed by more than ten percentage points across those definitions, so the swing-detection rule must be chosen before the retracement depth. ### More than half of all retracements go all the way back Before asking where pullbacks stop, it is worth counting how they end at all. Across the 6,266 swings: | Ending | Swings | Share | |---|---|---| | Turned, and the swing extended in its original direction | 2,955 | 47.2% | | Retraced fully to the swing's origin, erasing it | 3,295 | 52.6% | | Still undecided after 200 bars | 16 | 0.3% | **A slim majority of swings are completely unwound.** Any statement about where a pullback stops sits on top of that: more often than not, it does not stop. ## How to draw Fibonacci retracement levels Every platform ships the tool. | Platform | Steps | |---|---| | MT4 / MT5 | Insert → Fibonacci → Retracement, then drag from the swing low to the swing high | | TradingView | Pick Fib Retracement from the drawing tool list | | Browser (Formiq) | In the drawing tools; on the backtest side the swing is detected automatically so a level can be used as a rule | The only difference between drawing by hand and detecting automatically is **who picks the two anchors**. This article uses automatic detection: a bar that beats the N bars on each side of it becomes a swing point, highs and lows are forced to alternate, and the most recent leg between them is the swing. **That method has a confirmation lag.** A bar is only known to be a swing high once N later bars have failed to exceed it, so the line cannot be drawn until N bars after the top. How much that costs is measured further down. ## How this was measured The settings, at the resolution a reader needs to reproduce them. | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31, with 2024 run identically for comparison | | Timeframes | M15 / H1 / H4 | | Swing detection | A bar beating the 5 bars on each side is a swing point; highs and lows alternate; the latest leg is used. Repeated with 3, 8, 13 and 21 as a sensitivity check | | Swings | 6,266 (M15 4,752 / H1 1,176 / H4 338, both years) | | Depth grid | 10.0% to 90.0% in 0.2% steps = 401 levels. 23.6, 38.2, 50, 61.8 and 78.6 land on the grid exactly | | Turn rate | Of the swings that reached a depth, the share whose pullback ended within 5 points of it. Also computed at 2.5 and 10 points | | Trading test | 9 levels (5 Fibonacci, 4 controls) x 5 swing definitions x 2 triggers = 90 settings, per timeframe, per window | | Stops and targets | None, except in the section that measures them | | Spread | 0.3 pips, zero slippage, 0.1 lot, charged on both fills | | Measurement | Run through Formiq's backtest engine; pips recomputed from each individual fill | The four control depths (30%, 45%, 70% and 80%) are the point of the design. **If a Fibonacci ratio has an edge that belongs to the number rather than to the depth, it has to show up as a gap against the ordinary depth next door.** ## How deep the pullbacks actually go Start with reach: the share of pullbacks that got that deep at all. | Depth | Reached | Swings | |---|---|---| | 23.6% | 95.9% | 6,012 | | 30% | 92.5% | 5,798 | | 38.2% | 87.2% | 5,462 | | 45% | 82.3% | 5,157 | | 50% | 78.8% | 4,936 | | 55% | 75.4% | 4,725 | | 61.8% | 71.4% | 4,474 | | 70% | 66.8% | 4,184 | | 78.6% | 62.3% | 3,902 | | 80% | 61.5% | 3,855 | **Perfectly monotonic.** Deeper is rarer, and nothing bends at a Fibonacci ratio. The 23.6% line is touched on 19 swings out of 20; the 78.6% line is missed on two out of three. That table alone explains something. **Shallow levels feel like they work because price visits them constantly.** Draw a line at 23.6% and it will be touched almost every time, and a touch is what gets remembered. ## Which level reacts most? The highest was 45%, which is not a Fibonacci ratio. Dividing by the reach count asks a different question: for each depth, **of the swings that got there, what share ended their pullback within five points of it?** This number has no threshold to tune: it depends only on whether the level was reached and where the move stopped. | Depth | Turn rate | 95% interval | Turned / reached | |---|---|---|---| | 20% | 2.05% | ±0.36 | 125 / 6,096 | | 23.6% | 2.73% | ±0.41 | 164 / 6,012 | | 30% | 3.55% | ±0.48 | 206 / 5,798 | | 38.2% | 3.94% | ±0.52 | 215 / 5,462 | | 45% | 4.29% | ±0.55 | 221 / 5,157 | | 50% | 4.27% | ±0.56 | 211 / 4,936 | | 55% | 3.92% | ±0.55 | 185 / 4,725 | | 61.8% | 4.22% | ±0.59 | 189 / 4,474 | | 70% | 3.78% | ±0.58 | 158 / 4,184 | | 78.6% | 3.95% | ±0.61 | 154 / 3,902 | | 80% | 3.76% | ±0.60 | 145 / 3,855 | **The highest is 45%, which is not a Fibonacci ratio.** Then 50%, then 61.8%. The ordering should not be trusted, though. **Every depth from 38.2% down sits inside every other one's interval.** The gap between 45% at 4.29% and 78.6% at 3.95% is 0.34 points, smaller than either interval. What this measurement supports is "past 38.2%, they are all about the same", and nothing finer. The only real structure is at the shallow end. **20% and 23.6% are clearly lower**, at 2.05% and 2.73%. A pullback that has given back a fifth of the move has rarely finished. The figure at the top is the whole curve. Plotted across all 401 depths it climbs from 10% to about 30% and then runs flat in a band between 3.5% and 4.4% all the way to 90%. **None of the five Fibonacci depths had a higher reaction rate than the neighbouring tested depths.** Widening the window that counts as "ended here" does not change it: | Window | 23.6% | 38.2% | 50% | 61.8% | 78.6% | |---|---|---|---|---|---| | ±2.5 points | 1.1% | 2.3% | 2.0% | 2.1% | 1.9% | | ±5 points | 2.7% | 3.9% | 4.3% | 4.2% | 3.9% | | ±10 points | 6.0% | 8.4% | 8.0% | 8.0% | 7.8% | ## Is 61.8% a special level? Against the depth next door, no ratio-specific effect showed up. If deeper levels turn more often, the depth effect has to be separated from the number. So each level's turn rate was compared against **the mean of the depths within three points of it, excluding the 0.8 points either side**: near-identical depth, different number. Averaged over the twelve cells (three timeframes x two years x up and down legs): | Level | Gap vs neighbours | Ahead in | |---|---|---| | 20% | −0.078 points | 6 / 12 | | 23.6% (ratio) | −0.196 points | 5 / 12 | | 30% | +0.160 points | 6 / 12 | | 38.2% (ratio) | −0.077 points | 6 / 12 | | 45% | +0.033 points | 5 / 12 | | 50% (ratio) | −0.105 points | 7 / 12 | | 55% | −0.071 points | 4 / 12 | | 61.8% (ratio) | −0.040 points | 6 / 12 | | 70% | +0.091 points | 8 / 12 | | 78.6% (ratio) | +0.234 points | 7 / 12 | | 80% | +0.144 points | 7 / 12 | Pooled: **the five Fibonacci ratios average minus 0.037 points and beat their neighbours in 31 of 60 cells. The six controls average plus 0.046 points and 36 of 72.** Both are coin flips. Put at its plainest: **61.8% turns 4.22% of the pullbacks that reach it, and 62.0% turns 4.30%.** So does 61.0%, at 4.22%. The 38.2% level turns 3.94%, against 3.92% at 37.0%, 3.97% at 39.0% and 4.07% at 40.0%. **Moving the line one notch sideways changes nothing.** Within this test, **there is no evidence that a Fibonacci ratio stops price better than the depth beside it.** ## Is price past the level before you can draw? For the shallow levels, yes. One more number, and it may be the one that matters most in practice. A swing high is confirmed five bars after the top, using this test's default. The line cannot be drawn before that. **In those five bars, price has usually already gone through the shallow levels.** Share of swings whose level had already been traded through by the time the swing was confirmed, averaged over the twelve cells: | Depth | Already passed | |---|---| | 23.6% | 94.1% | | 30% | 88.8% | | 38.2% | 81.3% | | 45% | 73.4% | | 50% | 67.9% | | 61.8% | 52.5% | | 70% | 43.4% | | 78.6% | 35.0% | | 80% | 34.0% | **On 94 swings in 100, the 23.6% level had already been passed by the time the swing could be drawn.** At 38.2% it is 81.3%. Only at 61.8% does it fall to about half. Drawing by hand does not escape this. A top is only recognisable as a top after price has left it. **Planning to buy a shallow retracement assumes you can place the order before you know it is a retracement**, and that is true whether a person or an algorithm picks the swing. ## Does trading a level make money Everything so far describes price. This section trades it. Two readings are commonly taught: - **Touch the level and take the original direction**: buy the dip inside a rally - **Close through the level and take the break**: sell when a rally's 61.8% gives way The touch rule trades in the original swing direction, while the break rule trades through the level in the retracement direction. Nine depths x five swing definitions x two triggers were run across three timeframes and four windows. Settings profitable in **both** 2024 and 2025, out of 45 per cell: | Timeframe | Buy the level | Trade the break | |---|---|---| | M15 | 2 / 45 | 20 / 45 | | H1 | 6 / 45 | 14 / 45 | | H4 | 4 / 45 | 3 / 45 | **Buying the level barely survives.** On the 15-minute chart only 2 of 45 did: 78.6% and 80%, both with the 21-bar swing definition. The medians were minus 1,095.7 pips in 2025 and minus 2,153.6 pips in 2024. **Trading the break kept 20 of 45 on the 15-minute chart**, with medians of plus 631.8 pips in 2025 and plus 512.2 in 2024. That is the most stable cell in the study. The ratios have no edge here either. For buying the level, the five Fibonacci depths returned a median of minus 742.4 pips over 150 settings against minus 668.5 for the four controls over 120. For the break, Fibonacci returned plus 118.4 against plus 222.4. **The controls are slightly ahead in both.** Matched one against one (same timeframe, year, trigger and swing definition) **the Fibonacci level beat the ordinary depth next to it in 143 comparisons out of 300.** ## Buy the pullback or trade the break? The touch rule won 18 points more often and still finished 835 pips behind. For settings with at least 20 trades, mean win rate and median annual net were: | Trading rule | Mean win rate | Median net pips per year | Settings | |---|---|---|---| | Buy the level | 61.17% | −629.1 | 256 | | Trade the break | 43.23% | +206.1 | 217 | **Buying the level won 17.94 percentage points more often but its median annual net was 835.2 pips lower.** The hourly default at 61.8% on the 5-bar swing shows the difference in trade size. In 2025 buying the level took 292 trades at a 65.75% win rate for plus 921.1 pips. Its average loss of 74.68 pips was about 1.7 times its average win of 43.69. The break trigger took 112 trades at 35.71%, but its average win of 110.23 pips was about 1.8 times its average loss of 60.07. [Bollinger Bands](/blog/bollinger-band-settings) and [RSI](/blog/rsi-settings) produced the same split. **Choosing levels by win rate alone misidentifies which ones made more pips.** ## Does last year's best carry forward? Three of six stayed profitable. For each timeframe and trigger, the setting with the largest 2024 net result was applied unchanged to 2025: | Timeframe | Trigger | Setting selected in 2024 | 2024 trades | 2024 annual net | 2025 annual net | |---|---|---|---:|---:|---:| | 15-minute | Touch | 50%, 3-bar swing | 2,491 | +445.8 pips | −846.4 pips | | 15-minute | Break | 80%, 3-bar swing | 491 | +3,922.6 pips | −1,410.7 pips | | 1-hour | Touch | 30%, 21-bar swing | 150 | +1,909.5 pips | +320.2 pips | | 1-hour | Break | 80%, 13-bar swing | 35 | +3,335.4 pips | −251.4 pips | | 4-hour | Touch | 23.6%, 8-bar swing | 94 | +2,891.4 pips | +692.2 pips | | 4-hour | Break | 23.6%, 21-bar swing | 7 | +3,775.7 pips | +148.7 pips | Three of the six settings remained profitable in 2025. The 15-minute break setting moved from plus 3,922.6 pips to minus 1,410.7 pips. The four-hour break setting had only seven trades in 2024, so its plus 3,775.7-pip result is not evidence of stable performance. ## Filters and stops Applied to the hourly 61.8% level on the 5-bar swing. | Condition | Trigger | Year | Trades | Annual net | |---|---|---:|---:|---:| | None | Touch | 2025 | 292 | +921.1 pips | | None | Touch | 2024 | 255 | −1,846.5 pips | | None | Break | 2025 | 112 | +84.6 pips | | None | Break | 2024 | 101 | −622.8 pips | | ADX ≥ 20 | Touch | 2025 | 76 | +149.9 pips | | ADX ≥ 20 | Touch | 2024 | 64 | +642.8 pips | | ADX ≥ 20 | Break | 2025 | 29 | −887.4 pips | | ADX ≥ 20 | Break | 2024 | 25 | +218.8 pips | | ADX ≥ 25 | Touch | 2025 | 74 | +3.1 pips | | ADX ≥ 25 | Touch | 2024 | 69 | −713.9 pips | | ADX ≥ 25 | Break | 2025 | 28 | −536.8 pips | | ADX ≥ 25 | Break | 2024 | 25 | +282.6 pips | | ADX ≥ 30 | Touch | 2025 | 66 | −308.9 pips | | ADX ≥ 30 | Touch | 2024 | 53 | −785.0 pips | | ADX ≥ 30 | Break | 2025 | 25 | −1,018.2 pips | | ADX ≥ 30 | Break | 2024 | 16 | +1,068.7 pips | | London and New York | Touch | 2025 | 250 | −58.4 pips | | London and New York | Touch | 2024 | 223 | −2,132.8 pips | | London and New York | Break | 2025 | 82 | −515.6 pips | | London and New York | Break | 2024 | 79 | −1,194.6 pips | | Tokyo | Touch | 2025 | 185 | −611.2 pips | | Tokyo | Touch | 2024 | 178 | −811.2 pips | | Tokyo | Break | 2025 | 58 | −108.4 pips | | Tokyo | Break | 2024 | 49 | +176.8 pips | **The ADX filter reduced the touch rule from +921.1 to +3.1 pips in 2025 at ADX 25, but changed it from −1,846.5 to +642.8 in 2024 at ADX 20.** One year shows deterioration and the other shows improvement, so the filter did not produce a consistent result. The session filter was worse than no filter in three of the four columns, and London/New York was worse in all four. Stops and targets: | Exit rule | Trigger | Year | Trades | Annual net | |---|---|---:|---:|---:| | Opposite signal | Touch | 2025 | 292 | +921.1 pips | | Opposite signal | Touch | 2024 | 255 | −1,846.5 pips | | Opposite signal | Break | 2025 | 112 | +84.6 pips | | Opposite signal | Break | 2024 | 101 | −622.8 pips | | Stop 30 / target 60 | Touch | 2025 | 743 | −1,145.5 pips | | Stop 30 / target 60 | Touch | 2024 | 650 | −568.6 pips | | Stop 30 / target 60 | Break | 2025 | 182 | +170.7 pips | | Stop 30 / target 60 | Break | 2024 | 163 | +164.7 pips | | Stop 50 / target 100 | Touch | 2025 | 525 | +436.7 pips | | Stop 50 / target 100 | Touch | 2024 | 492 | −1,968.8 pips | | Stop 50 / target 100 | Break | 2025 | 154 | +303.6 pips | | Stop 50 / target 100 | Break | 2024 | 141 | −64.5 pips | | Stop 100 / target 200 | Touch | 2025 | 381 | +1,269.4 pips | | Stop 100 / target 200 | Touch | 2024 | 351 | −2,208.7 pips | | Stop 100 / target 200 | Break | 2025 | 130 | +29.8 pips | | Stop 100 / target 200 | Break | 2024 | 119 | +441.9 pips | | 24-bar time exit | Touch | 2025 | 382 | +923.2 pips | | 24-bar time exit | Touch | 2024 | 338 | −1,865.9 pips | | 24-bar time exit | Break | 2025 | 162 | +303.3 pips | | 24-bar time exit | Break | 2024 | 145 | −28.6 pips | **A 30-pip stop with a 60-pip target was the only exit that was profitable in both years** (+170.7 and +164.7), and only on the break trigger. Nothing rescued the touch trigger's 2024. ## What it pays in spread Re-running 2025 on the hourly chart with only the spread changed: | Spread | Touch 61.8% (292 trades) | Break 61.8% (112 trades) | |---|---|---| | 0.0 pips | +1,008.7 | +118.2 | | 0.3 pips | +921.1 | +84.6 | | 0.6 pips | +833.5 | +51.0 | | 1.0 pips | +716.7 | +6.2 | | 1.5 pips | +570.7 | −49.8 | | 2.0 pips | +424.7 | −105.8 | Going from 0 to 2.0 pips costs the touch rule 584.0 pips (292 × 2.0) and the break rule 224.0 (112 × 2.0). **Both match trade count times spread exactly**: a relationship that has held in all sixteen articles in this series. The break rule goes underwater somewhere past 1.0 pips. **112 trades a year is not many, but 118.2 pips of gross profit is not much either.** "61.8% works because it is special" is not supported here. But **deeper levels do turn more often than shallow ones**, so this is not a result against using the tool as a depth scale. [Bollinger Bands](/blog/bollinger-band-settings) and [RSI](/blog/rsi-settings) also produced high win rates alongside negative annual net because average losses exceeded average wins. [Moving average crosses](/blog/moving-average-cross-settings) ask the same question of an event rather than a level, and [GMMA](/blog/gmma-settings) also uses non-standard controls to test whether adding more lines improves results. [Fair value gaps](/blog/fair-value-gap-settings) similarly converts a drawing pattern into a testable condition and compares it with ordinary-bar controls, which were filled just as often as the gaps. If you would rather check this on your own pair and period, [building the conditions without writing code](/blog/backtest-without-coding) is the route. The [Alligator](/blog/alligator-settings) uses 13, 8 and 5 for the same reason. Matched by trigger, timeframe, displacement and year against four sets containing no Fibonacci number, 13/8/5 produced more annual net pips in 136 of 288 comparisons. ## Notes - One pair, USD/JPY, and two years. Nothing here guarantees the same behaviour on another pair or another period - **Swings are detected, not drawn.** A person choosing anchors by hand may pick different waves, and this test cannot measure that choice. Five detection widths were run and the ordering of the turn rates held across all of them - **The 5-point window that counts as "ended here" is arbitrary.** It was also computed at 2.5 and 10 points, and the relationship between levels did not change - The turn rate asks whether the pullback finished at the level. A pause that resumed lower is not counted as a turn - The 401 depths overlap, so neighbouring figures are not independent. The maximum cannot be read as the result of 401 independent trials - Entries and exits use bar closes. Real fills differ - The spread is held at 0.3 pips throughout. Real spreads move with the session and with releases --- ### GMMA Tested 216 Ways: Do 12 Moving Averages Beat 2? URL: https://formiq.jp/blog/gmma-settings Language: en Published: 2026-08-26 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: GMMA, moving averages, ribbon, indicator, backtest, USD/JPY Also available in: ja — https://formiq.jp/ja/blog/gmma-settings GMMA (the Guppy Multiple Moving Average) draws twelve exponential moving averages at once. The short ribbon (3, 5, 8, 10, 12, 15) is usually described as the short-term traders and the long ribbon (30, 35, 40, 45, 50, 60) as the longer-term investors. The claim is not that a fast average crossing a slow one is worth trading. Two lines already say that. What GMMA claims is that **twelve lines show you something two cannot**: whether both groups agree, or whether one of them is hesitating. **Each ribbon's mean period is pinned at 9 and 45 while the number of lines drops from six to one.** At one line apiece the ribbons disappear and what is left is a plain EMA 9 / EMA 45 crossover. Same speed, no fan. Any difference belongs to the fan. The answer first. **The test found no benefit from twelve lines.** One line beat six in 21 of 36 matched runs; median net was +1,313.5 pips with one line and +1,371.5 with six. ## What GMMA measures There is no GMMA formula. It is exponential moving averages, and the only decision is **which periods, and how many of them.** The ladder used here keeps the mean period fixed as the count falls: | Lines | Short ribbon | Long ribbon | |---|---|---| | 6 (Guppy's own) | 3, 5, 8, 10, 12, 15 | 30, 35, 40, 45, 50, 60 | | 5 | 3, 6, 9, 12, 15 | 30, 38, 45, 53, 60 | | 4 | 3, 7, 11, 15 | 30, 40, 50, 60 | | 3 | 3, 9, 15 | 30, 45, 60 | | 2 | 3, 15 | 30, 60 | | 1 | 9 | 45 | Every row's short ribbon averages 9.0 periods (8.83 on the six-line row, which uses Guppy's exact numbers) and every long ribbon averages 45.0 (43.33 on the six-line row). **The one-line row and the six-line row are looking at averages of the same speed.** That matters. Thinning from the front (keeping only 3, 5 and 8) would drop the mean period to 5.3, and then a worse result could be blamed either on the missing fan or on the extra speed. Holding the endpoints fixed removes the second explanation. ### A one-line GMMA is an EMA crossover With one line per ribbon, the "cross" trigger compares the short average against the long average, which is **an EMA 9 / EMA 45 golden and dead cross**. The test confirms it exactly. On the 2025 hourly chart, the one-line GMMA took 184 trades for +393.5 pips. An EMA 9 / 45 crossover built separately took 184 trades for +393.5 pips. **Thinning the ribbon terminates in the crossover everybody already knew.** ## How to add GMMA to a chart It is twelve moving averages, so it can always be assembled by hand. | Platform | Steps | |---|---| | MT4 / MT5 | Not included. Either stack twelve moving averages and set the periods and colours, or find a GMMA file to install | | TradingView | Add Guppy Multiple Moving Average from the indicator search | | Browser (Formiq) | Included; the backtest side exposes all three triggers and the number of lines per ribbon | **If you are placing twelve averages by hand, the periods are yours to choose.** On the evidence in the table below, thinning them while holding the mean period barely moves the result, so the work of placing twelve is not, in this test, buying anything. ## How this was measured The settings, at the resolution a reader needs to reproduce them. | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31, with 2024 run identically for comparison | | Timeframes | M15 / H1 / H4 | | Triggers | Cross (the two averages meet), Expansion (the gap widens), All-above (every short line clear of every long line) | | Lines per ribbon | 1, 2, 3, 4, 5, 6, with the mean period held at 9 and 45 | | Combinations | 3 triggers x 6 counts = 18 per timeframe per window; 216 runs in total | | Exit | The opposite signal — entry and exit share one rule set | | Stops and targets | None, except in the section that measures them | | Spread | 0.3 pips, zero slippage, 0.1 lot, charged on both fills | | Ribbon state | For each trade, the gap between ribbons, both ribbon widths, and whether the long ribbon was in strict order, read on the bar before the fill (5,217 records) | | Measurement | Run through Formiq's backtest engine; pips recomputed from each individual fill | ## Do you need all twelve lines? Thinning the ribbon to one line cost nothing. Mean net pips across all three triggers and both years, by line count: | Lines | M15 | H1 | H4 | |---|---|---|---| | 1 | +1,719 | +680 | +487 | | 2 | +1,830 | +441 | −1 | | 3 | +1,797 | +617 | +70 | | 4 | +1,742 | +538 | +86 | | 5 | +1,749 | +560 | +65 | | 6 | +1,880 | +645 | +138 | **No column climbs.** M15 wanders between 1,719 and 1,880; H1 between 441 and 680. Matched one against one (same timeframe, year and trigger) **one line beat six in 21 of 36 comparisons**, with a mean difference of +124.9 pips in the one-line direction. By median: one line +1,313.5 pips, six lines +1,371.5. **Whatever the other ten lines add, it is smaller than this test can measure.** The figure at the top is that table. The four-hour line jumps at one line, but that is the four-hour chart inverting between the two years, not the fan doing something. ## Which timeframe does GMMA work on? Only the 15-minute chart held together. Settings profitable in **both** 2024 and 2025, out of 18 per timeframe: | Timeframe | Year | Profitable settings | Median net | Positive in both years | |---|---:|---:|---:|---:| | M15 | 2025 | 18 / 18 | +1,703.6 pips | **18 / 18** | | M15 | 2024 | 18 / 18 | +1,917.9 pips | **18 / 18** | | H1 | 2025 | 14 / 18 | +221.9 pips | 14 / 18 | | H1 | 2024 | 18 / 18 | +1,067.7 pips | 14 / 18 | | H4 | 2025 | 0 / 18 | −1,719.0 pips | 0 / 18 | | H4 | 2024 | 18 / 18 | +2,100.2 pips | 0 / 18 | **Every 15-minute setting made money in both years**: the most uniform cell in the seventeen articles of this series. The published trend-following setting (cross, six lines) took 831 trades in 2025 at a 27.32% win rate for +1,703.6 pips, with an average win of 55.02 pips against an average loss of 17.86. **One average win covered about 3.1 average losses.** Trades were held 42 bars (ten and a half hours) so this is not scalping. **The four-hour chart inverted completely.** All 18 settings made money in 2024 (median +2,100.2) and all 18 lost in 2025 (median −1,719.0). The two-line cross changed from +2,334.6 pips in 2024 to −2,419.9 in 2025. ## The three triggers compared Three readings of the same ribbons, both years pooled: | Trigger | Mean win rate | Median net pips | Mean trades | Mean bars held | |---|---|---|---|---| | Cross | 26.52% | +1,468.4 | 364 | 42.7 | | Expansion | 29.14% | +1,313.5 | 321 | 48.3 | | All-above | 32.76% | +1,345.4 | 209 | 76.1 | **All-above trades least often (57% of the cross's count), wins most often (6.2 points above the cross), and holds longest**, exactly what tightening a condition produces. The cross has the highest median, and the three are within 150 pips of each other. ## Does a wide ribbon mean a strong trend? It does raise the win rate. All 5,217 trades were split into quartiles by **the gap between the ribbons on the bar before the fill**: | Timeframe | Q1 (narrowest) | Q2 | Q3 | Q4 (widest) | |---|---|---|---|---| | M15 (995 each) | 24.32% | 28.04% | 29.05% | **32.96%** | | H1 (239 each) | 22.18% | 27.20% | 35.98% | **35.98%** | | H4 (69 each) | 21.74% | 15.94% | 30.43% | **34.78%** | **The widest quartile won most often on every timeframe**, by 8.64 points on M15, 13.80 on H1 and 13.04 on H4. The relationship held on M15, H1 and H4. Win rate and money part company, though: | Timeframe | Q1 | Q2 | Q3 | Q4 | |---|---|---|---|---| | M15 | +3.96 pips | +2.58 | +2.85 | **+2.06** | | H1 | −3.68 pips | +1.83 | +7.08 | **+12.15** | | H4 | −5.62 pips | −34.82 | +24.73 | **+34.48** | **M15 points the other way.** The win rate climbs while the average win falls from 3.96 to 2.06 pips: entering once the ribbons have already separated is more likely to work and has less left to capture. On the hourly and four-hour charts both measures move together. ## Is the long ribbon's order useful? It told us nothing here. A common reading is that **a long ribbon in strict order means a real trend**: investors agreeing, 30 above 35 above 40 and so on. The long ribbon was already in order on most bars: | Timeframe and year | Bars | Long ribbon in order | |---|---|---| | M15 2025 | 24,903 | 85.18% | | M15 2024 | 24,999 | 85.39% | | H1 2025 | 6,226 | 86.35% | | H1 2024 | 6,250 | 85.74% | | H4 2025 | 1,610 | 85.09% | | H4 2024 | 1,616 | 90.47% | **It holds on about 85 bars in 100.** Six EMAs of similar length rarely fall out of sequence. **A condition true 85% of the year cannot filter anything.** Splitting trades by it does not point one way either: | Timeframe | In order | Not in order | |---|---|---| | M15 | 3,020 trades, 28.48% win, +1.70 pips | 961 trades, 28.93% win, **+6.39 pips** | | H1 | 742 trades, 27.49% win, −0.81 pips | 216 trades, 39.81% win, **+20.71 pips** | | H4 | 225 trades, 24.44% win, **+5.78 pips** | 53 trades, 30.19% win, −8.90 pips | **Two of the three timeframes did better when the ribbon was *not* in order.** On the hourly chart that is 12.32 points of win rate and 21.52 pips a trade. The four-hour chart disagrees on 53 trades. ## Does compression precede a big move? It did not. The other piece of folklore: **the ribbon squeezes, then a big move follows.** That is a statement about price rather than a trading rule, so no backtest is needed. Adding both ribbon widths together, the **narrowest tenth of bars** counts as compressed. Mean absolute move over the following 24 bars: | Timeframe and year | After a squeeze | After every other bar | |---|---|---| | M15 2025 | 30.61 pips (2,491 bars) | 32.71 pips (22,388 bars) | | M15 2024 | 23.56 pips (2,500) | 32.94 pips (22,499) | | H1 2025 | 68.27 pips (618) | 68.41 pips (5,584) | | H1 2024 | 47.60 pips (626) | 73.69 pips (5,624) | | H4 2025 | 85.83 pips (147) | 134.22 pips (1,439) | | H4 2024 | 153.56 pips (162) | 145.58 pips (1,454) | **Five of the six moved less after a squeeze.** Widening the horizon to 6, 12, 24 and 48 bars gives the same picture: **in 21 of 24 combinations the post-squeeze move is the smaller one.** Quiet following quiet is an ordinary property of price and has nothing to do with GMMA. **"Compressed means coiled" came out backwards here.** One thing did point somewhere, and it is direction rather than size. Measuring the next 24 bars along the side the ribbons were already leaning, the 15-minute post-squeeze move is +6.31 pips against +0.24 for every other bar. Small, but persistent in the existing lean. The hourly 2024 cell reads −8.50 and disagrees, so this does not survive both years. ## Does the 2024 winner hold in 2025? The 15-minute one did; the four-hour one changed sign. | Timeframe | Trigger | Lines | 2024 trades | 2024 net | 2025 net | |---|---|---:|---:|---:|---:| | M15 | All-above | 2 | 436 | +2,304.6 pips | +1,404.1 pips | | H1 | Cross | 1 | 173 | +1,209.7 pips | +393.5 pips | | H4 | Cross | 2 | 41 | +2,334.6 pips | −2,419.9 pips | All 18 M15 settings were positive in both years, and this setting also retained +1,404.1 pips in 2025. On H4, the same setting changed from +2,334.6 to −2,419.9 pips. ## Filters and stops Applied to the 15-minute cross on six lines. Trade count, then net pips. | Condition | Year | Trades | Net | |---|---:|---:|---:| | None | 2025 | 831 | +1,703.6 pips | | None | 2024 | 821 | +2,067.4 pips | | ADX ≥ 20 | 2025 | 113 | −769.3 pips | | ADX ≥ 20 | 2024 | 120 | +735.1 pips | | ADX ≥ 25 | 2025 | 68 | −474.8 pips | | ADX ≥ 25 | 2024 | 75 | −352.6 pips | | ADX ≥ 30 | 2025 | 20 | −199.6 pips | | ADX ≥ 30 | 2024 | 30 | −335.3 pips | | London and New York | 2025 | 441 | +1,524.9 pips | | London and New York | 2024 | 435 | +382.4 pips | | Tokyo | 2025 | 321 | +799.8 pips | | Tokyo | 2024 | 310 | +1,186.2 pips | **The ADX filter did more damage than anything else in this test.** It cuts 831 trades to 113 and turns +1,703.6 pips into −769.3. At ADX 25 both years lose. **A trend filter on a trend indicator subtracts**, which reads as the two conditions asking the same question twice. The session filters both came in under the baseline without turning negative. Stops and targets: | Exit rule | Year | Trades | Net | |---|---:|---:|---:| | Opposite signal | 2025 | 831 | +1,703.6 pips | | Opposite signal | 2024 | 821 | +2,067.4 pips | | Stop 20 / target 40 | 2025 | 730 | +1,115.7 pips | | Stop 20 / target 40 | 2024 | 731 | +335.8 pips | | Stop 30 / target 60 | 2025 | 768 | +2,039.9 pips | | Stop 30 / target 60 | 2024 | 757 | +1,020.0 pips | | Stop 50 / target 100 | 2025 | 813 | +1,512.3 pips | | Stop 50 / target 100 | 2024 | 796 | +757.3 pips | | 48-bar time exit | 2025 | 827 | +2,036.1 pips | | 48-bar time exit | 2024 | 815 | +1,409.4 pips | **Nothing beat the baseline in both years.** The 30/60 stop and the 48-bar exit each improve 2025 and take about a thousand pips out of 2024. **The baseline depends on an average win of about 55 pips against an average loss of about 18 pips, and a fixed exit removes part of that difference.** ## What it pays in spread Re-running 2025 on the 15-minute chart with only the spread changed: | Spread | Net pips | |---|---| | 0.0 | +1,952.9 | | 0.3 | +1,703.6 | | 0.6 | +1,454.3 | | 1.0 | +1,121.9 | | 1.5 | +706.4 | | 2.0 | +290.9 | | 3.0 | −540.1 | The trade count stays at 831. Going from 0 to 3.0 pips costs 2,493.0 pips, which is **exactly 831 × 3.0**: the identity that has held in all seventeen articles. **Break-even sits at 2.35 pips**, between +290.9 at 2.0 and −540.1 at 3.0. Eight hundred trades a year on a 15-minute chart still leaves room at a realistic USD/JPY spread. This is the first 15-minute trading rule in the series with enough gross profit to remain positive after those costs. Its average holding time is 42 bars, or ten and a half hours, so the average move per trade is many times the spread. The part about twelve lines revealing something is not supported. **The gap between the ribbons, read as a win-rate gauge, is.** [Moving average crosses](/blog/moving-average-cross-settings) measure the crossing itself rather than the fan around it. [Perfect order](/blog/perfect-order-settings) requires all three averages to be stacked and found a different surviving calculation on each timeframe. [Bollinger Bands](/blog/bollinger-band-settings) and [RSI](/blog/rsi-settings) also produced profitable rules with low win rates because their average wins exceeded their average losses. [Fibonacci retracements](/blog/fibonacci-retracement-levels) test each level against neighbouring non-Fibonacci levels. [CCI](/blog/cci-settings) also lost annual net when an ADX filter was added. The base-rate problem repeats in [the Alligator](/blog/alligator-settings), whose three lines are stacked in order for 80% of the year, and reaches its limit in [fair value gaps](/blog/fair-value-gap-settings), where the famous 99% fill rate turned out to be 99% on ordinary bars too. If you would rather check this on your own pair, [the conditions can be built without writing code](/blog/backtest-without-coding). ## Notes - One pair, USD/JPY, and two years. Nothing here guarantees the same behaviour elsewhere - **The two years are very different.** 2024 rose 1,632 pips within a 2,237-pip range; 2025 finished 56 pips lower within a 1,900-pip range. A trend follower profitable in both years on M15 is still resting on those two years - **There is one way of thinning the ribbon here.** Endpoints (3–15 and 30–60) are held and the interior is spaced evenly, so the mean period does not move. Thinning from the front or keeping only the middle would give different numbers - The six-line row uses Guppy's exact periods, so its means are 8.83 and 43.33 rather than the 9.0 and 45.0 of every other row - Ribbon state is read one bar before the fill, so the trade's own move cannot leak into the feature - "Compressed" is the narrowest tenth within each period: a relative definition. An absolute width would give different counts - Entries and exits use bar closes. Real fills differ - The spread is held at 0.3 pips throughout. Real spreads move with the session and with releases - [Volatility varies by hour](/data), which the session filter results inherit --- ### Bollinger Bands: Mean Reversion or Breakout? 315 Settings Tested URL: https://formiq.jp/blog/bollinger-band-settings Language: en Published: 2026-08-25 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: bollinger bands, mean reversion, breakout, indicator, settings, backtest, USDJPY Also available in: ja — https://formiq.jp/ja/blog/bollinger-band-settings Bollinger Bands are a moving average with a band above and below it, set at some multiple of how much price has been varying. Most price action stays inside that envelope. Two common trading rules are **fade the band**, which treats a touch as a reversal, and **follow the break**, which treats a close outside as continuation. Which one is right depends on who you ask. So we measured both, on the same pair, over the same period, paying the same costs. For 2025 (1 January to 31 December) on USD/JPY, seven periods crossed with five deviations, plus four choices of where the fade takes profit, gives **140 fade settings, run separately for the long and the short side**, and **35 breakout settings**: 315 in all. Each was run on three timeframes, and the whole grid was run again on 2024. The answer first. **On the hourly chart, the fade averaged a 60.4% win rate and came through both years in only 25 of 140 settings; the breakout averaged 40.4% and came through in 23 of 35.** Selling the upper band did worse still: **zero settings out of 140 were profitable in both years** on either the 15-minute or the hourly chart. Ranked by win rate, the order reverses. ## What Bollinger Bands measure The middle line is a simple moving average (**period**). The bands sit above and below it at a multiple of the standard deviation of closes over that same period (**deviation**). The usual claim is that 1σ holds about 68% of the action and 2σ about 95%. That is what a normal distribution would do. Counting the actual bars gives something else: here are the 6,226 hourly USD/JPY closes of 2025 against a 20-period band: | Deviation | Closes finishing inside | |---|---| | 1σ | 45.0% | | 1.5σ | 70.4% | | 2σ | 87.1% | | 2.5σ | 95.1% | | 3σ | 99.1% | **Every band holds less than the textbook says.** At 2σ, roughly one bar in eight closes outside. A fade built on "price rarely goes there" is going to meet price out there more often than the assumption allows. The bands themselves are a ruler that stretches with the market: wide when it is moving, narrow when it is not. The same 2σ is a different number of pips on different days. The shared parameters are period and deviation. The fade also has an exit deviation. | Parameter | What raising it does | |---|---| | Period | The middle line dulls. Fewer touches | | Deviation | The bands move further out. Fewer touches and fewer breaks | | Take-profit deviation (fade only) | Puts the far band further away. Fewer trades, each held longer | The counts bear it out. Averaged over the fade-long settings on the hourly chart in 2025, a period of 10 traded 319 times a year and a period of 50 traded 53. By deviation it runs 213 trades at 1σ down to 77 at 3σ. Raising period or deviation reduces trade count and therefore reduces the total spread paid. ## How to put them on a chart Bollinger Bands ship with every platform. | Platform | Steps | |---|---| | MT4 / MT5 | Pick Bollinger Bands from the indicator list and set the period and deviation | | TradingView | Add Bollinger Bands from the indicator search and set the period and deviation | | Browser (Formiq) | Built in; the backtest side switches between fading a touch and following a break with a trigger | Drawing them is the same everywhere. **The decision that matters is whether the rule trades a touch or a close outside the band.** This article tests a band-touch fade and a close breakout separately. ## How this was tested Entering this table reproduces these results. | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / one-hour / four-hour | | Bars tested | 24,903 on M15, 6,226 on H1, 1,610 on H4 for 2025 | | Fade, long | Buy the close of a bar whose low reached the lower band; close it when a high reaches the upper band | | Fade, short | Sell the close of a bar whose high reached the upper band; close it when a low reaches the lower band | | Breakout | A close above the upper band goes long, a close below the lower band goes short, and the opposite break reverses at that same close | | Stops and targets | None in the main grid; measured separately below | | Spread | 0.3 pips, zero slippage, 0.1 lots. Charged on the exit and the new entry both | | Fade grid | 7 periods (10, 14, 20, 25, 30, 40, 50) × 5 deviations (1, 1.5, 2, 2.5, 3σ) × 4 take-profit deviations (0.5, 1, 1.5, 2σ) = 140, run separately long and short | | Breakout grid | 7 periods × 5 deviations = 35 | | Measurement | Run through Formiq's backtester, with pips re-derived from each trade's fill prices | The long fade and short fade are reported separately because a band touch is directional: buying the lower band and selling the upper one produce entirely different results. ## Fade the band or trade the break? The fade won more often per trade; the breakout kept more settings profitable in both years. | Trading rule | Timeframe | Profitable 2025 | 2025 median | Profitable 2024 | 2024 median | Both years | Mean win rate | |---|---|---|---|---|---|---|---| | Fade long | M15 | 22 / 140 | −503 pips | 44 / 140 | −223 pips | 7 / 140 | 64.9% | | Fade long | H1 | 39 / 140 | −252 pips | 82 / 140 | +106 pips | 25 / 140 | 60.4% | | Fade long | H4 | 104 / 140 | +511 pips | 36 / 140 | −308 pips | 15 / 140 | 69.5% | | Fade short | M15 | 4 / 140 | −661 pips | 2 / 140 | −1,376 pips | 0 / 140 | 63.9% | | Fade short | H1 | 21 / 140 | −489 pips | 15 / 140 | −1,023 pips | 0 / 140 | 60.8% | | Fade short | H4 | 99 / 140 | +526 pips | 6 / 140 | −1,247 pips | 1 / 140 | 63.4% | | Breakout | M15 | 25 / 35 | +432 pips | 18 / 35 | +33 pips | 16 / 35 | 38.3% | | Breakout | H1 | 25 / 35 | +789 pips | 30 / 35 | +838 pips | 23 / 35 | 40.4% | | Breakout | H4 | 12 / 35 | −1,025 pips | 28 / 35 | +861 pips | 9 / 35 | 36.3% | (Mean win rate is over the 2025 settings; the settings that never took a trade are left out of that column.) The figure at the top plots the hourly row of that table, one dot per setting: win rate across, net pips up. **The red and purple dots (fading) gather to the bottom right; the blue ones (following) to the top left.** The settings that are right more often are the ones below zero. **Selling the upper band produced no setting at all that was profitable in both years** on the 15-minute or hourly chart. USD/JPY rose 1,632 pips through 2024, so repeatedly selling upper-band touches kept the short fade positioned against that rise. ## Why the higher win rate loses The two sides of the trade are different sizes. Averaged over every 2025 setting: | Trading rule | Timeframe | Mean win | Mean loss | |---|---|---|---| | Fade long | H1 | +41.3 pips | −68.6 pips | | Fade short | H1 | +45.1 pips | −83.8 pips | | Breakout | H1 | +123.8 pips | −80.9 pips | A fade wins a little when the touch snaps back and loses a lot when price simply keeps going: the exit waits for the far band, so a trade that was wrong stays wrong for a while. The loss runs 1.7 to 1.9 times the win. **A 60% hit rate does not survive winners worth 60% of the losers.** The breakout is the mirror image: wrong more often than not, but the trades that work are the ones that run. The relationship mostly holds on the other timeframes. The fade-long wins 20.8 and loses 40.6 pips on M15, wins 84.9 and loses 140.9 on H4. The breakout wins 66.5 against 42.5 on M15, and 286.1 against 136.4 on H4. There is one exception. **On the four-hour chart the short fade wins 99.3 pips against a 94.5-pip loss**: the only row where the winner is the larger side. It still came through both years in 1 setting out of 140: a favourable size ratio did not cover what 2024's rise took out of it. ## The textbook 20 and 2σ, used both ways | Use | Timeframe | 2025 | 2024 | |---|---|---|---| | Fade long, exit at 1σ | M15 | 667 trades, 65.07%, −411.1 pips | 642 trades, 67.76%, −613.3 pips | | Fade long, exit at 1σ | H1 | 167 trades, 63.47%, +71.0 pips | 157 trades, 68.79%, −597.6 pips | | Fade long, exit at 1σ | H4 | 40 trades, 77.50%, +1,030.3 pips | 34 trades, 67.65%, −389.0 pips | | Breakout | M15 | 570 trades, 37.72%, +292.5 pips | 566 trades, 35.87%, +344.4 pips | | Breakout | H1 | 132 trades, 43.18%, +1,177.5 pips | 137 trades, 41.61%, +1,434.6 pips | | Breakout | H4 | 41 trades, 19.51%, −2,126.7 pips | 34 trades, 35.29%, +775.5 pips | **The only rows profitable in both years are the 15-minute and hourly breakouts.** Faded, the same two numbers lost on M15 in both years; on H1 they made a small profit in 2025 and lost in 2024. The four-hour chart changes sign between years on both sides. Its breakout took 41 trades in 2025 at a 19.51% win rate for −2,126.7 pips; that is not a sample worth drawing a conclusion from. ## How to choose period and deviation The only fade settings that survived were the slow, rarely touched ones. Period and deviation are varied one at a time on the hourly chart and averaged over the remaining settings. The last column counts settings profitable in both 2024 and 2025. **Breakout, by period** | Period | 2025 mean | 2024 mean | Both years | |---|---|---|---| | 10 | +520 pips | +732 pips | 2 / 5 | | 14 | +243 pips | +1,221 pips | 3 / 5 | | 20 | +885 pips | +1,489 pips | 5 / 5 | | 25 | +596 pips | +870 pips | 5 / 5 | | 30 | +804 pips | +287 pips | 2 / 5 | | 40 | +452 pips | +803 pips | 4 / 5 | | 50 | −12 pips | +704 pips | 2 / 5 | **Periods 20 and 25 went 5 for 5.** The textbook period really is near the top on the breakout side. **Breakout, by deviation** | Deviation | 2025 mean | 2024 mean | Both years | |---|---|---|---| | 1σ | +1,082 pips | +1,569 pips | 7 / 7 | | 1.5σ | +1,368 pips | +693 pips | 5 / 7 | | 2σ | +980 pips | +742 pips | 6 / 7 | | 2.5σ | −813 pips | +896 pips | 2 / 7 | | 3σ | −125 pips | +462 pips | 3 / 7 | **A 1σ deviation went 7 for 7.** Narrower bands break more often, which gives the setting a sample to stand on. Past 2.5σ a break becomes rare and the year starts deciding the result. **Fade, by deviation** | Deviation | 2025 mean | 2024 mean | Both years | |---|---|---|---| | 1σ | −741 pips | −27 pips | 1 / 28 | | 1.5σ | −532 pips | +5 pips | 4 / 28 | | 2σ | −252 pips | +83 pips | 6 / 28 | | 2.5σ | −44 pips | +27 pips | 6 / 28 | | 3σ | −14 pips | +107 pips | 8 / 28 | For the fade, wider deviations reduced the average loss. Every deviation still averaged negative in 2025, and a 1σ fade was profitable in both years in only one of its 28 settings. Period runs the same way. On the hourly chart the fade at period 50 came through both years 7 times out of 20; at period 20, 0 times out of 20. **What survives on the fade side is the setting that almost never fires.** ## Does last year's pick hold up? The best fade of 2024 finished below the 2025 median; the best breakout stayed above it. | Trading rule | Setting | Selection year | Selection-year net | Trades | Test year | Test-year net | Test-year median | |---|---|---:|---:|---:|---:|---:|---:| | Fade long, H1 | period 40, 2σ → 1.5σ | 2024 | +1,029.9 pips | 67 | 2025 | −268.0 pips | −251.7 pips | | Breakout, H1 | period 20, 1.5σ | 2024 | +2,637.8 pips | 195 | 2025 | +1,488.1 pips | +788.6 pips | The fade setting lost 268.0 pips in 2025, below that year's −251.7-pip median. The breakout setting made +1,488.1 pips, above the +788.6-pip median. The direction does not run both ways. Period 10 at 1.5σ made +1,838.4 pips over 363 trades in 2025 and lost 156.5 pips in 2024. **Even for the breakout, one year's annual net did not select a setting that worked in the other year.** ## Filters, stops and targets | Trading rule | Condition | Year | Trades | Annual net | |---|---|---:|---:|---:| | Fade long 20/2σ→1σ | Unfiltered | 2025 | 167 | +71.0 pips | | Fade long 20/2σ→1σ | Unfiltered | 2024 | 157 | −597.6 pips | | Fade long 20/2σ→1σ | ADX ≥ 20 | 2025 | 120 | +385.6 pips | | Fade long 20/2σ→1σ | ADX ≥ 20 | 2024 | 111 | −728.0 pips | | Fade long 20/2σ→1σ | ADX ≥ 25 | 2025 | 97 | +735.5 pips | | Fade long 20/2σ→1σ | ADX ≥ 25 | 2024 | 75 | −1,055.9 pips | | Fade long 20/2σ→1σ | ADX ≥ 30 | 2025 | 73 | +836.7 pips | | Fade long 20/2σ→1σ | ADX ≥ 30 | 2024 | 53 | −879.3 pips | | Fade long 20/2σ→1σ | London and New York hours | 2025 | 140 | +309.8 pips | | Fade long 20/2σ→1σ | London and New York hours | 2024 | 131 | −691.5 pips | | Fade long 20/2σ→1σ | Tokyo hours | 2025 | 91 | +2.2 pips | | Fade long 20/2σ→1σ | Tokyo hours | 2024 | 81 | −727.9 pips | | Breakout 20/2σ | Unfiltered | 2025 | 132 | +1,177.5 pips | | Breakout 20/2σ | Unfiltered | 2024 | 137 | +1,434.6 pips | | Breakout 20/2σ | ADX ≥ 20 | 2025 | 94 | +703.6 pips | | Breakout 20/2σ | ADX ≥ 20 | 2024 | 106 | −322.6 pips | | Breakout 20/2σ | ADX ≥ 25 | 2025 | 69 | −199.2 pips | | Breakout 20/2σ | ADX ≥ 25 | 2024 | 74 | −120.3 pips | | Breakout 20/2σ | ADX ≥ 30 | 2025 | 52 | −445.6 pips | | Breakout 20/2σ | ADX ≥ 30 | 2024 | 49 | +263.3 pips | | Breakout 20/2σ | London and New York hours | 2025 | 114 | −54.1 pips | | Breakout 20/2σ | London and New York hours | 2024 | 113 | +1,599.8 pips | | Breakout 20/2σ | Tokyo hours | 2025 | 84 | +78.2 pips | | Breakout 20/2σ | Tokyo hours | 2024 | 79 | +449.2 pips | The ADX filter turns the fade positive in 2025 and deepens its loss in 2024. Added to the breakout it **came in under the unfiltered version in both years.** Screening a trend-following rule for trend strength did not help it. | Trading rule | Exit rule | Year | Annual net | |---|---|---:|---:| | Fade long 20/2σ→1σ | Opposite band only | 2025 | +71.0 pips | | Fade long 20/2σ→1σ | Opposite band only | 2024 | −597.6 pips | | Fade long 20/2σ→1σ | SL 30 / TP 60 | 2025 | −230.9 pips | | Fade long 20/2σ→1σ | SL 30 / TP 60 | 2024 | −530.5 pips | | Fade long 20/2σ→1σ | SL 50 / TP 100 | 2025 | −197.8 pips | | Fade long 20/2σ→1σ | SL 50 / TP 100 | 2024 | −531.7 pips | | Fade long 20/2σ→1σ | SL 100 / TP 200 | 2025 | −179.3 pips | | Fade long 20/2σ→1σ | SL 100 / TP 200 | 2024 | +41.2 pips | | Fade long 20/2σ→1σ | Time exit, 24 bars | 2025 | +303.4 pips | | Fade long 20/2σ→1σ | Time exit, 24 bars | 2024 | −635.2 pips | | Breakout 20/2σ | Opposite band only | 2025 | +1,177.5 pips | | Breakout 20/2σ | Opposite band only | 2024 | +1,434.6 pips | | Breakout 20/2σ | SL 30 / TP 60 | 2025 | +193.6 pips | | Breakout 20/2σ | SL 30 / TP 60 | 2024 | +231.7 pips | | Breakout 20/2σ | SL 50 / TP 100 | 2025 | +54.5 pips | | Breakout 20/2σ | SL 50 / TP 100 | 2024 | +1,121.3 pips | | Breakout 20/2σ | SL 100 / TP 200 | 2025 | +1,328.6 pips | | Breakout 20/2σ | SL 100 / TP 200 | 2024 | +1,128.8 pips | | Breakout 20/2σ | Time exit, 24 bars | 2025 | +1,413.8 pips | | Breakout 20/2σ | Time exit, 24 bars | 2024 | +1,536.2 pips | Fixed stops and targets are executable levels measured from the filled entry. A 60-pip target therefore closes at exactly +60 pips when reached; the engine does not subtract a second exit-side adjustment from that configured level. **Only the 24-bar time exit beat the opposite-band exit in both years, and only on the breakout** (+1,177.5 → +1,413.8 and +1,434.6 → +1,536.2). Tightening the stop makes the breakout worse in a straight line because the rule depends on the few trades that continue far enough to offset its losses, and a close stop removes those trades. Nothing helped the fade in both years. ## How much does spread take? About 40% of the fade's profit. | Spread | Fade long 20/2σ→1σ (167 trades) | Breakout 20/2σ (132 trades) | |---|---|---| | 0.0 pips | +121.1 | +1,217.1 | | 0.3 pips | +71.0 | +1,177.5 | | 0.6 pips | +20.9 | +1,137.9 | | 1.0 pips | −45.9 | +1,085.1 | | 1.5 pips | −129.4 | +1,019.1 | | 2.0 pips | −212.9 | +953.1 | **At a zero spread the fade makes 121.1 pips; at 0.3 pips it keeps 71.0.** Costs remove about 40% of the pre-cost profit: the 50.1-pip difference is exactly 167 trades × 0.3 pips. The breakout still keeps 953.1 pips at a 2.0-pip spread. It loses 264.0 pips going from 0 to 2.0, which is 132 × 2.0. The same relationship held without error in [QQE](/blog/qqe-settings) and [moving average crossovers](/blog/moving-average-cross-settings). Choosing the fade because it wins more often is not something this test supports. [The fine sweep from 0.5σ through 4σ](/blog/bollinger-band-sigma-settings) takes the separate question of which deviation works best. [The 125-bar BBW squeeze test](/blog/bollinger-bandwidth-settings) uses the distance between those bands as a volatility condition. [Williams %R also produced high win rates without correspondingly high annual net results](/blog/williams-r-settings). [The RSI fade and level-cross test](/blog/rsi-settings) found mean win rates of 65.3% for the fade and 23.7% for the cross; [using a band touch to filter that fade](/blog/rsi-bollinger-combo) is measured separately. [Fibonacci retracements](/blog/fibonacci-retracement-levels) compare buying a level with trading its break across 401 depths. [MA disparity settings](/blog/ma-disparity-settings) sets the same two lines by a percentage of price instead of a standard deviation, and judges them on closes rather than wicks. [Building the rules without writing code](/blog/backtest-without-coding) is how to repeat any of this on your own pair. ## Notes - One pair, USD/JPY, and two years. Nothing here guarantees the same pattern elsewhere - The fade holds until the far band with no stop, so a trade that goes wrong shows its full loss. Stops are measured in their own section - The breakout reads closes. A bar whose wick pierced a band but whose close came back inside is not a signal - Entries and exits fill at bar closes. Real fills differ - The spread is a flat 0.3 pips throughout. Real spreads move with the session and around releases - Four-hour settings trade between 4 and 153 times a year on the fade side, and one breakout setting took no trades at all. Win rates and profit factors at those counts are wide numbers before anything else is said --- ### Is the Golden Cross Profitable on Forex? 280 Moving Average Tests URL: https://formiq.jp/blog/moving-average-cross-settings Language: en Published: 2026-08-25 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: moving average, golden cross, indicator, settings, backtest, USDJPY Also available in: ja — https://formiq.jp/ja/blog/moving-average-cross-settings Use two moving averages: buy when the fast line crosses above the slow line and sell when it crosses below. Those are the golden-cross and dead-cross rules tested here. The variables are the two periods, the averaging method and the timeframe. This article writes a fast period of 10 and a slow period of 20 as 10/20. We tested 56 period pairs with five averaging methods, giving 280 settings. Each was backtested on 15-minute, one-hour and four-hour USD/JPY data for 2024 and 2025. The largest 2025 annual net result was 12/50 weighted (WMA) on the 15-minute chart: 751 trades, a 33.95% win rate and +2,978.0 pips. On the four-hour chart, however, profitable settings fell from 243 of 280 in 2024 to 60 in 2025. **The timeframe and whether price sustained one direction mattered more than selecting one period pair.** ## What a crossover actually measures A moving average is the last n bars smoothed into one line. Lengthen the period and the line drifts further from price and takes longer to change direction. Watching two of them cross means watching **which smoothing is on top**, and taking the moment they change places. They change places when recent price has moved far enough away from the older average, which makes a crossover a proxy for "the trend may have turned". The rule has three parameters. | Parameter | What raising or changing it does | |---|---| | Fast period | The fast line reacts later and crosses less often | | Slow period | The slow line reacts later and holds trades longer | | Averaging method | Changes the line's distance from price and the crossing count | On the one-hour chart in 2025, a slow period of 20 averaged 398 trades and a slow period of 200 averaged 108. Hull moving averages (HMA) averaged 455 trades and smoothed moving averages (SMMA) 86. Both periods and methods change trade frequency. ## How to put two moving averages on a chart Moving averages overlay the price chart. Add one fast line and one slow line, then set the period and method of each. | Platform | Steps | |---|---| | MT4 / MT5 | Add Moving Average twice. Choose simple, exponential, smoothed or linear weighted | | TradingView | Add Moving Average twice and set the period and method of each | | Browser (Formiq) | Add two moving averages. Choose SMA, EMA, WMA, SMMA or HMA | The five methods tested are simple (SMA), exponential (EMA), weighted (WMA), smoothed (SMMA) and Hull (HMA). HMA is not built into MT4 or MT5. ## How this was tested The test used these conditions. | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / one-hour / four-hour | | Bars tested | 24,903 on M15, 6,226 on H1, 1,610 on H4 for 2025 | | Long | Close of the bar where the fast average crosses above the slow one | | Short | Close of the bar where the fast average crosses below | | Exit | The opposite crossover, reversing into the new direction at the same close. No stop, target or time limit | | Spread | 0.3 pips, zero slippage, 0.1 lots. Charged on the exit and the new entry both | | Grid | 8 fast periods (5, 8, 10, 12, 15, 20, 25, 50) × 8 slow periods (20, 25, 30, 40, 50, 75, 100, 200), keeping the 56 pairs where fast is shorter × 5 averaging methods = 280, per timeframe per year | | Measurement | Run through Formiq's backtester, with pips re-derived from each trade's fill prices | The main grid uses the same method on both lines. Mixed methods are tested separately. Holding to the opposite crossover with no stop keeps this a test of the crossover rather than of the crossover plus an exit. Stops and targets get their own section. ## Which timeframe should you use? The four-hour chart varied most between years, leaving 32 of 280 profitable in both. The same 280 settings were run on each timeframe. | Timeframe | Profitable in 2025 | 2025 median | Profitable in 2024 | 2024 median | Profitable in both | |---|---|---|---|---|---| | 15-minute | 231/280 | +810 pips | 234/280 | +1,420 pips | 205/280 | | One-hour | 188/280 | +445 pips | 259/280 | +1,671 pips | 172/280 | | Four-hour | 60/280 | −1,153 pips | 243/280 | +2,151 pips | 32/280 | On the four-hour chart, profitable settings fell from 243/280 in 2024 to 60/280 in 2025. The median annual net result reversed from +2,151 pips to −1,153, and only 32/280 were profitable in both years. The number profitable in both years was 205/280 on the 15-minute chart and 172/280 on the one-hour chart. More settings held up across both years than on the four-hour chart. The path of price differed between the two years. USD/JPY rose 1,632 pips from the start to the end of 2024. In 2025, the year-end price was 56 pips below the year-opening price. The high-to-low ranges were still similar at 2,237 and 1,900 pips. **Price sustained an upward direction in 2024, while in 2025 it moved widely but finished near where it began.** A moving-average crossover benefits from a sustained direction, and that difference affected the four-hour results most. ## How the common period pairs performed These are the annual net results for common period pairs. | Combination | Timeframe | 2025 | 2024 | |---|---|---:|---:| | 5/20 SMA | 15-minute | +264.5 pips | +646.1 pips | | 5/20 SMA | One-hour | −484.9 pips | +3,744.7 pips | | 10/20 SMA | 15-minute | +1,015.0 pips | +1,600.9 pips | | 10/20 SMA | One-hour | +377.4 pips | +2,888.4 pips | | 12/25 EMA | 15-minute | +1,436.6 pips | +2,102.1 pips | | 12/25 EMA | One-hour | +970.9 pips | +440.6 pips | | 20/50 SMA | 15-minute | +1,774.9 pips | +1,464.8 pips | | 20/50 SMA | One-hour | −220.5 pips | −34.6 pips | | 25/75 SMA | 15-minute | −1,023.1 pips | +3,613.7 pips | | 25/75 SMA | One-hour | −56.8 pips | −253.1 pips | | 50/200 SMA | 15-minute | +629.3 pips | +301.2 pips | | 50/200 SMA | One-hour | −1,340.4 pips | +2,435.4 pips | On the 15-minute chart, 25/75 SMA moved from +3,613.7 pips in 2024 to −1,023.1 in 2025. On the one-hour chart, 5/20 SMA moved from +3,744.7 to −484.9 pips. **The familiar period pairs did not repeat the same result in both years.** The 10/20 SMA default was profitable in both years on the 15-minute and one-hour charts. On the four-hour chart it lost in both years: −1,336.2 pips in 2025 and −1,245.7 pips in 2024. ## Did the 50/200 golden cross work? Its result rests on too few trades to judge. The common 50/200 SMA pair produced these results. | Timeframe | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | 15-minute | 2025 | 147 | 38.10% | +629.3 pips | | 15-minute | 2024 | 147 | 29.93% | +301.2 pips | | One-hour | 2025 | 47 | 31.91% | −1,340.4 pips | | One-hour | 2024 | 27 | 55.56% | +2,435.4 pips | | Four-hour | 2025 | 14 | 50.00% | +173.6 pips | | Four-hour | 2024 | 7 | 71.43% | +1,357.5 pips | The 71.43% win rate on the four-hour chart in 2024 is five wins from seven trades. That sample is too small to establish an edge. On the one-hour chart, annual net pips reversed from +2,435.4 in 2024 to −1,340.4 in 2025. ## Is a longer slow line better? Lengthening it widened the gap between the two years. The next table groups the one-hour results by slow period. | Slow period | 2025 mean | 2024 mean | Profitable in both | |---|---|---|---| | 20 | +736 pips | +1,899 pips | 20 / 25 | | 25 | +992 pips | +1,798 pips | 30 / 30 | | 30 | +822 pips | +1,474 pips | 31 / 35 | | 40 | +709 pips | +1,388 pips | 26 / 35 | | 50 | +567 pips | +1,468 pips | 27 / 35 | | 75 | −129 pips | +1,293 pips | 17 / 40 | | 100 | −158 pips | +1,679 pips | 17 / 40 | | 200 | −1,179 pips | +1,964 pips | 4 / 40 | All 30 settings with a slow period of 25 were profitable in both years. At a slow period of 200, only 4 of 40 were. The slow-200 mean changed from +1,964 pips in 2024 to −1,179 in 2025. **Longer slow periods produced a wider difference between the two years.** On the 15-minute chart, 24 of 40 settings with a slow period of 200 were profitable in both years. The effect was smaller than on the one-hour chart, so the timeframe also changed the result. ## Does last year's best repeat? It did not. The best 2024 setting on each timeframe was run unchanged in 2025. | Timeframe | Best 2024 setting | 2024 annual net | 2025 annual net | 2025 median | |---|---|---|---|---| | 15-minute | 20/75 SMA | +3,995.2 pips (387 trades) | −234.6 pips | +809.8 pips | | One-hour | 5/25 SMA | +3,828.5 pips (309 trades) | +583.3 pips | +445.0 pips | | Four-hour | 8/50 WMA | +3,378.8 pips (44 trades) | −1,367.2 pips | −1,152.9 pips | The 2024 winner lost money in 2025 on the 15-minute and four-hour charts and finished below the 2025 median. The one-hour winner remained profitable but only slightly exceeded the median. **Choosing the previous year's winner did not reproduce the previous year's result.** ## Which moving average type is best? It changes with the timeframe: EMA on M15, HMA on H1, SMMA on H4. The table counts settings that were profitable in both 2024 and 2025. Each method has 56 settings. | Method | 15-minute | One-hour | Four-hour | |---|---|---|---| | SMA | 48/56 | 28/56 | 2/56 | | EMA | 52/56 | 34/56 | 2/56 | | WMA | 49/56 | 39/56 | 8/56 | | SMMA | 43/56 | 22/56 | 13/56 | | HMA | 13/56 | 49/56 | 7/56 | The largest counts were EMA at 52/56 on the 15-minute chart, HMA at 49/56 on the one-hour chart and SMMA at 13/56 on the four-hour chart. HMA fell to 13/56 on the 15-minute chart and averaged 1,852 trades there, compared with 324 to 875 for the other four methods. Method selection therefore needs the profitable-setting count and trade count for the specific timeframe. ### SMA versus EMA at identical periods An SMA gives the last n closes equal weight and drops anything older. An EMA gives more weight to recent closes and lets older observations fade rather than disappear at a fixed boundary. That is the basis for calling the EMA "faster", but a faster response does not automatically produce a better trading result. To separate method from every other choice, we paired SMA and EMA runs with the same fast period, slow period, timeframe and test window. The comparison contains 672 one-to-one matches. The EMA produced more annual net pips in 318 of 672 comparisons and the SMA in 354. The EMA took fewer trades in 405. It tended to reduce frequency, but it did not finish ahead on annual net pips. | Timeframe | EMA profitable both years | SMA profitable both years | 2024 winner also won in 2025 | |---|---:|---:|---:| | M15 | 52 / 56 | 48 / 56 | 19 / 56 | | H1 | 34 / 56 | 28 / 56 | 26 / 56 | | H4 | 2 / 56 | 2 / 56 | 25 / 56 | The EMA left slightly more settings profitable in both years on M15 and H1, while H4 was tied. The method with higher annual net pips in 2024 repeated in fewer than half of the 56 period pairs on every timeframe. The 10/20 pair on M15 shows the reversal plainly: in 2025 the SMA made +1,015.0 pips and the EMA +2,052.3; in 2024 the SMA made +1,600.9 and the EMA +1,183.0. **The winning method changed with the year, so last year's result is not a sound reason to lock in EMA or SMA.** The full [SMA versus EMA test](/blog/sma-vs-ema) covers the weight profiles, response timing and sensitivity to starting history behind these results. ## What if the two lines use different types? The longer period can move faster, reversing the roles of the two lines. We tested the 25 fast-method and slow-method combinations on the one-hour chart. The figure shows each moving average's mean distance from the close. A 20-period HMA sat 18.7 pips from the close, while a 10-period SMA sat 22.1 pips away. The longer HMA was closer to price, so using HMA only on the slow line of a 10/20 pair made that line react faster. On the one-hour chart in 2024, 10/20 with SMA fast and HMA slow lost 3,393.4 pips over 385 trades. Reversing the methods made +3,739.2 pips over 397 trades. **When the two lines use different methods, the periods alone do not identify which line reacts faster.** ## Does a higher win rate mean more profit? Not consistently. On the 15-minute chart, the highest win rate came from 15/20 HMA: 44.22% over 2,680 trades, yet it lost 755.1 pips. The annual-net leader, 12/50 WMA, won only 33.95% of its trades. Its average winner was +56.74 pips and its average loser −23.17, so the size of the winners offset the lower win rate. The four-hour results even include a 100% win rate, but it came from only two trades. **Win rate needs to be read together with trade count and annual net pips.** ## Filters, stops and targets Two 15-minute settings, with the filters that usually get suggested. | Condition | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | 10/20 SMA | 2025 | 1,402 | 37.52% | +1,015.0 pips | | 10/20 SMA | 2024 | 1,397 | 38.08% | +1,600.9 pips | | + ADX ≥ 20 | 2025 | 674 | 35.31% | −495.4 pips | | + ADX ≥ 20 | 2024 | 705 | 37.02% | +923.1 pips | | + ADX ≥ 25 | 2025 | 354 | 34.75% | −271.0 pips | | + ADX ≥ 25 | 2024 | 400 | 40.25% | +1,847.9 pips | | + ADX ≥ 30 | 2025 | 170 | 34.71% | +195.7 pips | | + ADX ≥ 30 | 2024 | 217 | 37.33% | +703.8 pips | | + London and New York hours (UTC 7–21) | 2025 | 802 | 37.16% | +846.9 pips | | + London and New York hours (UTC 7–21) | 2024 | 797 | 37.77% | +41.1 pips | | + Tokyo hours (UTC 0–8) | 2025 | 460 | 38.91% | +309.7 pips | | + Tokyo hours (UTC 0–8) | 2024 | 443 | 39.73% | +2,194.9 pips | ADX ≥ 25 increased annual net pips from +1,600.9 to +1,847.9 in 2024, but changed +1,015.0 to −271.0 in 2025. Restricting trades to Tokyo hours also helped in 2024 and hurt in 2025. **Neither filter improved both years.** Stops, targets and a time exit were tested on two 15-minute settings. | Fast/slow | Method | Year | Exit rule | Annual net | |---|---|---:|---|---:| | 10/20 | SMA | 2025 | Opposite crossover only | +1,015.0 pips | | 10/20 | SMA | 2024 | Opposite crossover only | +1,600.9 pips | | 10/20 | SMA | 2025 | SL 30 / TP 60 | +260.0 pips | | 10/20 | SMA | 2024 | SL 30 / TP 60 | +187.9 pips | | 10/20 | SMA | 2025 | SL 50 / TP 100 | +535.1 pips | | 10/20 | SMA | 2024 | SL 50 / TP 100 | +961.9 pips | | 10/20 | SMA | 2025 | SL 30 / TP 90 | +615.6 pips | | 10/20 | SMA | 2024 | SL 30 / TP 90 | +1,167.5 pips | | 10/20 | SMA | 2025 | Time exit, 24 bars | +621.7 pips | | 10/20 | SMA | 2024 | Time exit, 24 bars | +273.0 pips | | 12/50 | WMA | 2025 | Opposite crossover only | +2,978.0 pips | | 12/50 | WMA | 2024 | Opposite crossover only | +2,103.7 pips | | 12/50 | WMA | 2025 | SL 30 / TP 60 | +1,865.5 pips | | 12/50 | WMA | 2024 | SL 30 / TP 60 | +599.0 pips | | 12/50 | WMA | 2025 | SL 50 / TP 100 | +2,498.5 pips | | 12/50 | WMA | 2024 | SL 50 / TP 100 | +774.7 pips | | 12/50 | WMA | 2025 | SL 30 / TP 90 | +2,314.2 pips | | 12/50 | WMA | 2024 | SL 30 / TP 90 | +1,405.1 pips | | 12/50 | WMA | 2025 | Time exit, 24 bars | +641.2 pips | | 12/50 | WMA | 2024 | Time exit, 24 bars | −108.5 pips | For both settings and in both years, the opposite-crossover exit produced the largest annual net result. On the one-hour chart, however, a 30-pip stop and 90-pip target on 10/20 SMA made +1,260.9 pips in 2025 versus +377.4 unmodified. The same exit underperformed in 2024. The result changed with the timeframe and year. ## What does the spread take? Re-running 2025 with nothing but the spread changed: | Timeframe | Fast/slow | Method | Trades | Spread | Annual net | |---|---|---|---:|---:|---:| | M15 | 10/20 | SMA | 1,402 | 0.0 pips | +1,435.6 pips | | M15 | 10/20 | SMA | 1,402 | 0.3 pips | +1,015.0 pips | | M15 | 10/20 | SMA | 1,402 | 0.6 pips | +594.4 pips | | M15 | 10/20 | SMA | 1,402 | 1.0 pips | +33.6 pips | | M15 | 10/20 | SMA | 1,402 | 1.5 pips | −667.4 pips | | M15 | 10/20 | SMA | 1,402 | 2.0 pips | −1,368.4 pips | | M15 | 12/50 | WMA | 751 | 0.0 pips | +3,203.3 pips | | M15 | 12/50 | WMA | 751 | 0.3 pips | +2,978.0 pips | | M15 | 12/50 | WMA | 751 | 0.6 pips | +2,752.7 pips | | M15 | 12/50 | WMA | 751 | 1.0 pips | +2,452.3 pips | | M15 | 12/50 | WMA | 751 | 1.5 pips | +2,076.8 pips | | M15 | 12/50 | WMA | 751 | 2.0 pips | +1,701.3 pips | | H1 | 10/20 | SMA | 358 | 0.0 pips | +484.8 pips | | H1 | 10/20 | SMA | 358 | 0.3 pips | +377.4 pips | | H1 | 10/20 | SMA | 358 | 0.6 pips | +270.0 pips | | H1 | 10/20 | SMA | 358 | 1.0 pips | +126.8 pips | | H1 | 10/20 | SMA | 358 | 1.5 pips | −52.2 pips | | H1 | 10/20 | SMA | 358 | 2.0 pips | −231.2 pips | | H4 | 50/200 | SMA | 14 | 0.0 pips | +177.7 pips | | H4 | 50/200 | SMA | 14 | 0.3 pips | +173.6 pips | | H4 | 50/200 | SMA | 14 | 0.6 pips | +169.4 pips | | H4 | 50/200 | SMA | 14 | 1.0 pips | +163.8 pips | | H4 | 50/200 | SMA | 14 | 1.5 pips | +156.8 pips | | H4 | 50/200 | SMA | 14 | 2.0 pips | +149.8 pips | Increasing the spread from 0 to 2.0 pips reduced the 15-minute 10/20 SMA result by 2,804.0 pips, exactly 1,402 trades × 2.0 pips. **Spread cost scales with trade count.** The 15-minute 10/20 SMA was near break-even at a 1.0-pip spread and lost money at 1.5. The lower-frequency 12/50 WMA still returned +1,701.3 pips at a 2.0-pip spread. Short-timeframe results need to be checked against the actual spread paid. ## Related reading - [SMA versus EMA](/blog/sma-vs-ema) - [GMMA settings tested](/blog/gmma-settings) - [Moving-average disparity settings tested](/blog/ma-disparity-settings) - [DMI/ADX periods and thresholds tested](/blog/dmi-adx-settings) ## Notes - The test covers USD/JPY in 2024 and 2025. Other pairs and periods were not tested - Entries and exits use bar closes. Actual fill prices may differ - The spread is fixed at 0.3 pips. Real spreads change by session and around economic releases - A slow period of 200 uses 200 bars from before the test window. A chart with less history may not produce the same line --- ### Is Buying RSI Below 30 Profitable? 330 Settings Tested on USD/JPY URL: https://formiq.jp/blog/rsi-settings Language: en Published: 2026-08-25 Updated: 2026-09-02 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: RSI, indicators, backtesting, USD/JPY, mean reversion, trend following Evidence: https://formiq.jp/evidence/rsi-settings Evidence JSON: https://formiq.jp/evidence/rsi-settings/data.json Also available in: ja — https://formiq.jp/ja/blog/rsi-settings RSI compares how much price rose against how much it fell over the last N bars and turns that into a number between 0 and 100. Almost every description of it says the same thing: **below 30 is oversold, above 70 is overbought**. A second, quieter reading treats **50 as the dividing line**: above it the buyers are ahead, below it the sellers are. The first is a mean-reversion tool. The second is a trend-following one. The same line produces opposite rules, and the usual advice does not say which reading the settings belong to. So both were measured side by side, on the same pair, the same years and the same costs. On USD/JPY in 2025, **150 fade settings run long and short separately** (6 periods × 5 entry levels × 5 exit levels) plus **30 midline-cross settings** (6 periods × 5 levels): 330 in all, on the fifteen-minute, hourly and four-hour charts, then repeated over 2024. On the hourly chart the fade won **65.3%** of its trades on average and survived both years in **31 of 150** settings. The midline cross won **23.7%** and survived in **17 of 30**. Not one cross setting won more than three trades in ten. ## What RSI actually measures Take the last N bars. Average the size of the up moves, average the size of the down moves, and express the up average as a percentage of the total. The only calculation parameter is **period**. - Nothing but up bars gives 100 - Nothing but down bars gives 0 - Up and down in balance gives 50 So 50 is the point where gains and losses cancel, and distance from it measures how lopsided the recent bars have been. "Overbought" and "oversold" are an interpretation laid on top of that; the arithmetic only reports the imbalance. ### A longer period does not reach 30 Longer periods create a practical problem. Averaging over more bars mixes the ups and downs together and pulls the number toward 50. The count below covers every hourly bar of 2025 on which RSI was defined: | Period | Bars below 30 | Bars above 70 | Lowest value all year | Highest | |---|---|---|---|---| | 7 | 13.97% | 14.13% | 1.0 | 97.0 | | 9 | 10.81% | 11.20% | 2.4 | 94.5 | | 14 | 5.73% | 6.54% | 8.0 | 89.5 | | 21 | 1.80% | 4.17% | 16.4 | 84.7 | | 30 | 0.39% | 2.03% | 24.9 | 80.8 | | 50 | 0.00% | 0.32% | 31.2 | 74.7 | **A period-50 RSI never once closed below 30 in all of 2025.** Its lowest reading was 31.2. "Use a longer period for cleaner oversold signals" describes something that does not happen: the signals do not get cleaner, they stop existing. On the four-hour chart both period 30 and period 50 stayed above 30 for the entire year. The sweep shows the consequence directly: on the four-hour chart, **50 of the 150 long-fade settings never took a single trade**, and 20 did nothing on the hourly chart either. ## Putting RSI on a chart | Platform | How | |---|---| | MT4 / MT5 | Relative Strength Index in the Navigator's indicator list; set the period | | TradingView | Search the indicator list for RSI and set the period | | Browser (Formiq) | On the chart by default; the backtester switches between the level reading and the level cross with a trigger | Drawing it is the same everywhere. **The decision that matters is whether the rule trades a level being reached or a cross through that level.** This article tests the fade and the level-cross rule separately. ## How this was measured Enter the settings below and the same numbers come out. | Item | Value | |---|---| | Pair | USD/JPY | | Window | 2025-01-01 to 2025-12-31, with 2024 run identically for comparison | | Timeframes | M15 / H1 / H4 | | Bars tested | 24,903 on M15, 6,226 on H1, 1,610 on H4 in 2025 | | Fade, long | Buy the close of the bar where RSI drops below the entry level; sell the close of the bar where it rises above the exit level | | Fade, short | The long reflected around 50: sell above 100 − entry, cover below 100 − exit | | Midline cross | Buy when RSI closes through the level from below, sell when it closes through from above, reversing on the same bar | | Stops and targets | None (measured in their own section) | | Spread | 0.3 pips, no slippage, 0.1 lots, charged on entry and exit | | Fade grid | 6 periods (7, 9, 14, 21, 30, 50) × 5 entry levels (20, 25, 30, 35, 40) × 5 exit levels (50, 55, 60, 65, 70) = 150, run long and short separately | | Cross grid | 6 periods × 5 levels (40, 45, 50, 55, 60) = 30 | | Measurement | Run through Formiq's backtester; pips recomputed from each trade's own fill prices | The long fade and short fade are reported separately because a level condition is directional. "Buy below 30" and "sell above 70" produce entirely different results. ## Mean reversion or trend following? The higher-win-rate fade came through both years less often than the level cross. | Trading rule | Timeframe | Profitable 2025 | 2025 median | Profitable 2024 | 2024 median | Both years | Mean win rate | |---|---|---|---|---|---|---|---| | Fade long | M15 | 77 / 150 | +5 pips | 67 / 150 | −45 pips | 29 / 150 | 64.6% | | Fade long | H1 | 78 / 150 | +11 pips | 58 / 150 | −65 pips | 31 / 150 | 65.3% | | Fade long | H4 | 81 / 150 | +95 pips | 36 / 150 | −211 pips | 13 / 150 | 67.6% | | Fade short | M15 | 42 / 150 | −264 pips | 33 / 150 | −367 pips | 5 / 150 | 57.7% | | Fade short | H1 | 48 / 150 | −137 pips | 23 / 150 | −727 pips | 9 / 150 | 61.2% | | Fade short | H4 | 84 / 150 | +272 pips | 17 / 150 | −1,075 pips | 1 / 150 | 61.0% | | Midline cross | M15 | 8 / 30 | −347 pips | 12 / 30 | −408 pips | 4 / 30 | 22.3% | | Midline cross | H1 | 25 / 30 | +830 pips | 22 / 30 | +846 pips | 17 / 30 | 23.7% | | Midline cross | H4 | 7 / 30 | −885 pips | 28 / 30 | +1,388 pips | 6 / 30 | 23.0% | (Mean win rate covers the 2025 settings that took at least one trade. The median is over every setting on that timeframe.) The figure at the top plots the hourly rows one point each, leaving out settings that traded fewer than ten times: win rate across, net pips up. **The red and purple fade settings spread across a 40–90% band; the blue cross settings sit in a narrow 17–30% band.** The medians still come out at +830 pips for the cross, +11 for the long fade and −137 for the short. Selling the overbought line is the weakest of the three, surviving both years in 9 of 150 hourly settings. USD/JPY rose 1,632 pips through 2024, and a rule that sells strength gets run over in a year like that. ## Does a longer RSI period help? Not uniformly: a longer period moves the fade and the level cross in opposite directions. The clearest result in the sweep is what the period does. Settings profitable in **both** 2024 and 2025, hourly chart: | Period | Fade long | Midline cross | |---|---|---| | 7 | 2 / 25 | 4 / 5 | | 9 | 3 / 25 | 5 / 5 | | 14 | 7 / 25 | 3 / 5 | | 21 | 7 / 25 | 2 / 5 | | 30 | 8 / 25 | 2 / 5 | | 50 | 4 / 25 | 1 / 5 | **The long fade had the most settings profitable in both years at period 30 (8 of 25), while the level cross had the most at period 9 (5 of 5).** At period 50 those counts fell to 4 of 25 for the fade and 1 of 5 for the cross. For the long fade, the profitable-setting count rose from 2 of 25 at period 7 to 8 of 25 at period 30, then fell to 4 of 25 at period 50. Longer periods made readings below 30 less frequent, and period 50 never fell below 30 on the 2025 hourly chart. This test does not identify why the count rose through period 30; it does show that the count halved when period increased from 30 to 50. For the level cross, all five tested levels were profitable in both years at period 9. The count fell to 3, 2, 2, and 1 as period increased through 14, 21, 30, and 50. Period 9 traded 775 times a year on the hourly chart and period 50 traded 231 times, so fewer trades gave individual wins and losses more influence over annual net. Levels behave more simply: | Fade entry level | Both years | |---|---| | 20 | 6 / 30 | | 25 | 13 / 30 | | 30 | 4 / 30 | | 35 | 2 / 30 | | 40 | 6 / 30 | | Cross level | Both years | |---|---| | 40 | 3 / 6 | | 45 | 5 / 6 | | 50 | 4 / 6 | | 55 | 4 / 6 | | 60 | 1 / 6 | The fade's exit level survived best at 70 (9 of 30, against 6 at level 50 and 5 at 65). **The back half of "buy at 30, sell at 70" holds up in this test even where the front half does not.** ## What the standard 14 / 30 / 70 did The most-quoted settings, run both ways. The cross column uses the same period 14 with the level at 50. | Reading | Timeframe | 2025 | 2024 | |---|---|---|---| | Fade long (30 → 70) | M15 | 105 trades, 63.81% win, +332.1 pips | 93 trades, 75.27% win, +9.0 pips | | Fade long (30 → 70) | H1 | 25 trades, 60.00% win, −250.7 pips | 22 trades, 63.64% win, +26.1 pips | | Fade long (30 → 70) | H4 | 9 trades, 66.67% win, +1,148.4 pips | 7 trades, 85.71% win, +209.2 pips | | Fade short (70 → 30) | M15 | 106 trades, 62.26% win, +435.5 pips | 94 trades, 50.00% win, −1,545.0 pips | | Fade short (70 → 30) | H1 | 25 trades, 64.00% win, −256.1 pips | 22 trades, 40.91% win, −1,429.7 pips | | Fade short (70 → 30) | H4 | 9 trades, 66.67% win, +1,059.2 pips | 8 trades, 50.00% win, −1,150.7 pips | | Cross the 50 line | M15 | 2,840 trades, 22.64% win, −265.7 pips | 2,820 trades, 21.99% win, +94.9 pips | | Cross the 50 line | H1 | 638 trades, 24.14% win, +1,272.0 pips | 681 trades, 22.17% win, +666.0 pips | | Cross the 50 line | H4 | 164 trades, 27.44% win, −790.1 pips | 141 trades, 23.40% win, +1,798.5 pips | **Three of those nine rows came through both years: the long fade on M15 and H4, and the cross on H1.** The 2024 fifteen-minute row is the one worth staring at. **93 trades, 75.27% of them winners, +9.0 pips for the year.** Three winners in every four, and what is left over after a year is nine pips. On the hourly chart the standard fade fires 25 times a year. **There is not much to say about a sample that size.** A 60% win rate there means fifteen winners and ten losers. The hourly cross took 638 trades for +1,272.0 pips, then 681 for +666.0 pips. The count is two orders of magnitude larger, and the year-to-year swing correspondingly smaller. ## Why the higher win rate loses The two sides of the trade are different sizes. Averaged over every 2025 setting: | Trading rule | Timeframe | Mean win | Mean loss | Mean bars held | |---|---|---|---|---| | Fade long | H1 | +88.6 pips | −120.8 pips | 188.9 | | Fade short | H1 | +80.7 pips | −129.6 pips | 273.4 | | Midline cross | H1 | +71.5 pips | −20.5 pips | 21.6 | **The cross wins 3.5 times what it loses.** At a 23.7% win rate, 0.237 × 71.5 − 0.763 × 20.5 still comes out positive. It rides a move as long as the line stays on one side of the level and reverses on the same bar when it does not, so a wrong call is cut early. For the fade, the average win is only 0.73 times the average loss. It collects small profits when price returns to the exit level and takes a large loss when price does not come back. The holding times differ just as sharply: 188.9 hourly bars for the fade, about eight days, against 21.6 for the cross. The same relationship holds on the other timeframes. The long fade ran +49.6 / −73.1 on M15 and +126.2 / −196.2 on H4; the cross ran +36.8 / −10.4 and +120.1 / −42.8. ## Can you pick settings from last year? The best fade of 2024 lost in 2025; the best cross fell but stayed profitable. | Trading rule | Setting | Selection year | Selection-year net | Trades | Test year | Test-year net | Test-year median | |---|---|---:|---:|---:|---:|---:|---:| | Fade long, H1 | period 14, 40 → 70 | 2024 | +1,139.8 pips | 44 | 2025 | −798.8 pips | +11.2 pips | | Midline cross, H1 | period 9, level 45 | 2024 | +3,601.3 pips | 725 | 2025 | +1,254.9 pips | +830.3 pips | The period-14, 40 → 70 long fade made +1,139.8 pips in 2024 from 44 trades at a 77.27% win rate, then lost 798.8 pips in 2025. It also fell below the 2025 median of +11.2 pips. The period-9, level-45 cross fell from +3,601.3 pips in 2024 to +1,254.9 pips in 2025, but remained above the 2025 median of +830.3 pips. Read the other way round, period 9 at level 40 made +1,815.4 pips over 672 trades in 2025 and +1,003.6 pips in 2024. These two cross settings were profitable in both years even though their annual net changed. ## Filters, stops and targets Two representative hourly settings with the usual additions bolted on. | Trading rule | Condition | Year | Trades | Annual net | |---|---|---:|---:|---:| | Fade long 14, 30→70 | Baseline | 2025 | 25 | −250.7 pips | | Fade long 14, 30→70 | Baseline | 2024 | 22 | +26.1 pips | | Fade long 14, 30→70 | ADX ≥ 20 | 2025 | 25 | −225.0 pips | | Fade long 14, 30→70 | ADX ≥ 20 | 2024 | 21 | −99.2 pips | | Fade long 14, 30→70 | ADX ≥ 25 | 2025 | 25 | −58.2 pips | | Fade long 14, 30→70 | ADX ≥ 25 | 2024 | 20 | −164.8 pips | | Fade long 14, 30→70 | ADX ≥ 30 | 2025 | 22 | −200.8 pips | | Fade long 14, 30→70 | ADX ≥ 30 | 2024 | 16 | −403.2 pips | | Fade long 14, 30→70 | London + NY | 2025 | 22 | −561.0 pips | | Fade long 14, 30→70 | London + NY | 2024 | 20 | +100.2 pips | | Fade long 14, 30→70 | Tokyo | 2025 | 21 | −113.6 pips | | Fade long 14, 30→70 | Tokyo | 2024 | 18 | +53.1 pips | | Cross 14 at 50 | Baseline | 2025 | 638 | +1,272.0 pips | | Cross 14 at 50 | Baseline | 2024 | 681 | +666.0 pips | | Cross 14 at 50 | ADX ≥ 20 | 2025 | 290 | +389.4 pips | | Cross 14 at 50 | ADX ≥ 20 | 2024 | 334 | −827.0 pips | | Cross 14 at 50 | ADX ≥ 25 | 2025 | 163 | +710.9 pips | | Cross 14 at 50 | ADX ≥ 25 | 2024 | 165 | +94.6 pips | | Cross 14 at 50 | ADX ≥ 30 | 2025 | 79 | +420.8 pips | | Cross 14 at 50 | ADX ≥ 30 | 2024 | 89 | −152.8 pips | | Cross 14 at 50 | London + NY | 2025 | 370 | +1,669.9 pips | | Cross 14 at 50 | London + NY | 2024 | 400 | −117.9 pips | | Cross 14 at 50 | Tokyo | 2025 | 247 | +284.3 pips | | Cross 14 at 50 | Tokyo | 2024 | 236 | +1,254.5 pips | **The ADX filter left the cross worse off in both years.** It does raise efficiency per trade (profit factor goes from 1.146 to 1.322 at ADX ≥ 25) but the total falls from +1,272.0 to +710.9 in 2025 and from +666.0 to +94.6 in 2024. Adding a trend-strength filter to the level-cross rule reduced annual net in both years. The London and New York filter produced +1,669.9 pips in 2025 and −117.9 in 2024. Tokyo produced +284.3 pips in 2025 and +1,254.5 in 2024. **Either session choice can appear superior when selected from only one year.** | Trading rule | Exit rule | Year | Annual net | |---|---|---:|---:| | Fade long 14, 30→70 | None | 2025 | −250.7 pips | | Fade long 14, 30→70 | None | 2024 | +26.1 pips | | Fade long 14, 30→70 | SL 30 / TP 60 | 2025 | −360.0 pips | | Fade long 14, 30→70 | SL 30 / TP 60 | 2024 | +90.0 pips | | Fade long 14, 30→70 | SL 50 / TP 100 | 2025 | −243.6 pips | | Fade long 14, 30→70 | SL 50 / TP 100 | 2024 | +189.2 pips | | Fade long 14, 30→70 | SL 100 / TP 200 | 2025 | −513.0 pips | | Fade long 14, 30→70 | SL 100 / TP 200 | 2024 | −731.2 pips | | Fade long 14, 30→70 | Time exit, 24 bars | 2025 | −536.7 pips | | Fade long 14, 30→70 | Time exit, 24 bars | 2024 | −878.1 pips | | Cross 14 at 50 | None | 2025 | +1,272.0 pips | | Cross 14 at 50 | None | 2024 | +666.0 pips | | Cross 14 at 50 | SL 30 / TP 60 | 2025 | +1,427.1 pips | | Cross 14 at 50 | SL 30 / TP 60 | 2024 | −312.7 pips | | Cross 14 at 50 | SL 50 / TP 100 | 2025 | +1,964.5 pips | | Cross 14 at 50 | SL 50 / TP 100 | 2024 | −300.1 pips | | Cross 14 at 50 | SL 100 / TP 200 | 2025 | +2,122.6 pips | | Cross 14 at 50 | SL 100 / TP 200 | 2024 | +761.1 pips | | Cross 14 at 50 | Time exit, 24 bars | 2025 | +2,960.5 pips | | Cross 14 at 50 | Time exit, 24 bars | 2024 | −84.7 pips | **Only the widest pair beat the baseline in both years for the cross** (+2,122.6 and +761.1). Tight stops helped in 2025 and turned 2024 negative. The 24-bar time exit produced the best single number in the table and the worst kind of evidence for it: 2024 came out below baseline. Nothing improved the fade in both years. ## Is the spread what loses the money? For the level cross yes, for the fade no. | Spread | Fade long 14 / 30→70 (25 trades) | Cross 14 @ 50 (638 trades) | |---|---|---| | 0.0 pips | −243.2 | +1,463.4 | | 0.3 pips | −250.7 | +1,272.0 | | 0.6 pips | −258.2 | +1,080.6 | | 1.0 pips | −268.2 | +825.4 | | 1.5 pips | −280.7 | +506.4 | | 2.0 pips | −293.2 | +187.4 | **At zero spread the fade still loses 243.2 pips.** Going from 0 to 2.0 pips costs it 50.0 pips: 25 trades times 2.0. Removing transaction cost therefore does not make the fade profitable; the trades themselves remain negative before spread. The cross is the case the identity does describe. It pays 191.4 pips at 0.3 (638 × 0.3), 13% of its +1,463.4 gross, and 1,276.0 pips at 2.0, which leaves +187.4. The same arithmetic held exactly in the [Bollinger Band](/blog/bollinger-band-settings) and [moving average cross](/blog/moving-average-cross-settings) runs. Choosing RSI mean reversion because it wins often is not supported by this test. A separate [RSI overbought and oversold duration test](/blog/rsi-overbought-oversold-sticking) measures how often the line remains beyond 70 or 30 on the next bar. [Williams %R also produced high win rates without correspondingly high annual net results](/blog/williams-r-settings). The [Bollinger Band article](/blog/bollinger-band-settings) compares a band-touch fade with a close breakout, and [combining the two](/blog/rsi-bollinger-combo) is measured against discarding the same number of bars at random. [QQE](/blog/qqe-settings) smooths RSI and [Stochastic RSI](/blog/stoch-rsi-settings) runs a stochastic over it; both were measured the same way. [Fibonacci retracements](/blog/fibonacci-retracement-levels) compare buying a price level with trading its break. [CCI](/blog/cci-settings) has no ceiling on its scale, and its fade lost in all six cells. If you want to check this on your own pair and your own window, [the conditions can be assembled without writing code](/blog/backtest-without-coding). ## Notes - One pair, two years. Nothing here promises the same behaviour on another pair or another year - The fade holds until the exit level arrives, with no stop. A wrong call shows up at full size. Stops were measured separately - **Many fade settings trade too rarely to judge.** The hourly median is 29 trades a year, and 50 of the 150 four-hour settings never traded at all. Win rates and profit factors on counts like those are wide numbers - Entries and exits are taken at bar closes. Real fills differ - Spread is a flat 0.3 pips throughout. Real spreads move with the session and around releases --- ### Accelerator Oscillator by Timeframe: 27 Settings Tested URL: https://formiq.jp/blog/ac-oscillator-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Accelerator Oscillator, Bill Williams, indicators, parameters, USDJPY Also available in: ja — https://formiq.jp/ja/blog/ac-oscillator-settings The Accelerator Oscillator measures the acceleration of the Awesome Oscillator: not how strong the move is, but whether that strength is building or fading. I tested three parameters across 27 combinations on USD/JPY through the whole of 2025, on three timeframes, then ran the same 27 on 2024. **On hourly bars, 24 of 27 were profitable in both years. On 15-minute bars, none were.** ## Which timeframe does it work on? | Timeframe | Year | Profitable settings | Median annual net | Median annual trades | |---|---:|---:|---:|---:| | 15-minute | 2025 | 7 / 27 | −437 pips | 2,934 | | 15-minute | 2024 | 0 / 27 | −1,707 pips | — | | 1-hour | 2025 | 26 / 27 | +1,292 pips | 711 | | 1-hour | 2024 | 25 / 27 | +1,814 pips | — | | 4-hour | 2025 | 24 / 27 | +658 pips | 184 | | 4-hour | 2024 | 16 / 27 | +274 pips | — | At about 2,930 trades a year, a 0.3 pip spread costs roughly 880 pips, and nothing here earned past that. **All 27 of them lost in 2024, so not one was profitable in both years.** Across both years, 24 of 27 hourly settings and 16 of 27 four-hour settings were profitable. Four-hour had 24 profitable settings in 2025 and 16 in 2024. ## Did the best setting hold in 2024? | Setting | Year | Annual net | |---|---:|---:| | Default 5/34/5 | 2025 | +2,139.4 pips | | Default 5/34/5 | 2024 | +1,436.8 pips | | 2025 maximum, 3/34/3 | 2025 | +3,359.1 pips | | 2025 maximum, 3/34/3 | 2024 | +1,108.6 pips | | 2024 maximum, 3/55/8 | 2025 | +1,031.6 pips | | 2024 maximum, 3/55/8 | 2024 | +3,400.9 pips | | 2025 minimum, 8/21/3 | 2025 | −175.0 pips | | 2025 minimum, 8/21/3 | 2024 | +2,716.1 pips | The 3/34/3 setting fell from +3,359.1 pips in 2025 to +1,108.6 pips in 2024. The 8/21/3 setting moved the other way, from −175.0 pips to +2,716.1 pips. **One year's net result did not identify the next year's setting.** The default 5/34/5 remained profitable in both years at +2,139.4 and +1,436.8 pips. Hourly trade counts ran 434 to 1,208. Shrinking the fast term raises the frequency. ## Are the defaults good enough? | Setting | Year | Annual net | |---|---:|---:| | Default 5/34/5 | 2025 | +2,192.3 pips | | Default 5/34/5 | 2024 | +274.2 pips | | 2024 maximum, 5/21/3 | 2025 | +1,988.2 pips | | 2024 maximum, 5/21/3 | 2024 | +2,053.6 pips | | 2025 minimum, 8/55/3 | 2025 | −901.1 pips | | 2025 minimum, 8/55/3 | 2024 | −1,227.5 pips | The default produced the largest annual net among the 27 four-hour settings in 2025. It took 190 trades at a 46.84% win rate. The hourly default took 711 trades at 41.49%, so the smaller four-hour sample is not enough to treat the win-rate difference as a stable property. ## Can win rate choose the setting? Across the 27 hourly settings in 2025, win rates ran from 38.34% to 43.79%, a 5.45-point range, while annual net ranged from −175.0 to +3,359.1 pips. **Win rate alone did not separate the profitable settings.** Holding in the direction of a zero-line break puts it in the same family as [OsMA](/blog/osma-settings) and [Supertrend](/blog/supertrend-settings): frequent small losses, occasional large wins, a win rate under 50% by design. ## Adding the Accelerator Oscillator to a chart | Platform | How to add it | |---|---| | MT4 / MT5 | Built in — Navigator, Bill Williams, Accelerator Oscillator | | TradingView | In the indicator list as "Accelerator Oscillator" | | Browser (Formiq) | In the indicator list, with fast, slow and smoothing adjustable | ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Buy | AC crosses above zero, filled at that bar's close | | Sell | It crosses below zero | | Exit | The opposite zero crossing only — no stop, no target, no time exit | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Combinations | 3 fast (3, 5, 8) × 3 slow (21, 34, 55) × 3 smoothing (3, 5, 8) = 27 | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | The same method applied to [OsMA](/blog/osma-settings), [QQE](/blog/qqe-settings), [Aroon](/blog/aroon-settings), [RVI](/blog/rvi-settings), [Supertrend](/blog/supertrend-settings), [Williams %R](/blog/williams-r-settings) and [MFI](/blog/mfi-settings). ## Notes - One pair, two years - 27 combinations is a coarser grid than the 54 to 240 used for other indicators in this series - AC inherits whatever the Awesome Oscillator implementation does. These figures come from Formiq's - Entries and exits both fill at bar closes - The spread is modelled as a flat 0.3 pips; the 15-minute result depends heavily on that - Four-hour settings trade 61 to 161 times a year: wide numbers at that count --- ### Aroon Periods and Timeframes: 84 Settings Tested URL: https://formiq.jp/blog/aroon-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Aroon, indicators, parameters, backtesting, USDJPY Also available in: ja — https://formiq.jp/ja/blog/aroon-settings Aroon looks at one thing: how many bars ago the recent high and the recent low were set. Not how far price moved, not how fast. Only when. Does that plainness work? I swept 84 configurations (period, trigger, and threshold) across the whole of 2025 on USD/JPY, on three timeframes, then ran all of it again on 2024. **Aroon on its own was weak over these two years.** On hourly bars in 2025, 43 of 84 configurations finished positive with a median of +6 pips. That is a coin flip. One result changed sharply. **On four-hour bars, period 40 with a threshold of 70 made +4,091.3 pips in 2024 and lost 1,664.5 pips in 2025.** Across the same timeframe, profitable configurations fell from 61 in 2024 to 17 in 2025. ## What Aroon measures Two lines: - **Aroon Up**: how long ago the highest high of the last N bars occurred. A new high today reads 100; a high set N bars ago reads 0 - **Aroon Down**: the same for the lowest low In an uptrend, highs keep being made, so Up pins near 100 and Down falls away. **Both lines measure elapsed time and nothing else**: no range, no volatility. There are two ways to read them, and they are not the same trade: | Trigger | Buy when | Nature | |---|---|---| | **Crossover** | Up crosses above Down | An event, on one bar | | **Level** | Up at or above a threshold while Down is below it | A state, true for as long as it lasts | So the trigger is swept here alongside the numbers. | Parameter | What raising it does | |---|---| | Period | Widens the window the high and low are found in. Fewer reversals, fewer signals | | Trigger | Crossover or level — a choice, not a number | | Threshold | The level mode's bar. Higher is stricter, so fewer signals | ## Adding Aroon to a chart | Platform | How to add it | |---|---| | MT4 | Not included. Download an .ex4 or .mq4 into `MQL4/Indicators`, restart | | MT5 | Included — Navigator, under Oscillators | | TradingView | In the indicator list as "Aroon" | | Browser (Formiq) | In the indicator list; the period is adjustable in its settings | ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Buy (crossover) | Up crosses above Down, filled at that bar's close | | Buy (level) | Up at or above the threshold while Down is below it | | Sell | The mirror of each | | Exit | The opposite reading only — no stop, no target, no time exit; the reversing bar opens the other side | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Configurations | 12 periods (7–75) × [crossover, or one of 6 thresholds 50–100] = 84, per timeframe and per year | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | ## Which timeframe can you use? | Timeframe | Year | Profitable settings | Median net | Positive in both years | |---|---:|---:|---:|---:| | 15-minute | 2025 | 63 / 84 | +452 pips | 48 / 84 | | 15-minute | 2024 | 63 / 84 | +1,955 pips | 48 / 84 | | 1-hour | 2025 | 43 / 84 | +6 pips | 33 / 84 | | 1-hour | 2024 | 63 / 84 | +1,429 pips | 33 / 84 | | 4-hour | 2025 | 17 / 84 | −1,207 pips | 3 / 84 | | 4-hour | 2024 | 61 / 84 | +1,649 pips | 3 / 84 | Read the four-hour row twice. In 2024, 61 of 84 configurations made money at a median of +1,649 pips. **In 2025, 17 did, at a median of −1,207 pips**, and only 3 of the 84 came through both years. Best timeframe to worst, in one year. **Aroon had more settings profitable in both years on shorter timeframes**: 48 configurations on 15-minute bars against 3 on four-hour. The 15-minute result also carries the highest trade count and therefore the largest spread cost, measured below. ## Is the four-hour result trustworthy? The best four-hour configuration of 2024 was period 40 with a threshold of 70. | Year | Trades | Win rate | Profit factor | Net | |---:|---:|---:|---:|---:| | 2024 | 27 | 62.96% | 4.335 | **+4,091.3 pips** | | 2025 | 40 | 25.00% | 0.533 | **−1,664.5 pips** | A 63.0% win rate and a 4.3 profit factor became a 25.0% win rate, a 0.533 profit factor and a 1,664.5-pip loss the following year. That was not one unlucky configuration: The figure at the top plots one point per configuration, with 2024 net on the horizontal axis and 2025 net on the vertical axis. On four-hour bars, 61 configurations were profitable in 2024, 17 in 2025 and only 3 in both years. One number in that table deserves more weight than the profit factor: **27 trades.** At 27, a single win or loss moves the win rate by 3.7 points. Four-hour configurations traded between 15 and 188 times a year, most of them in that region. The count is not enough for 63.0% to be a property of the setting. ## What period should you use? In crossover mode the period is the only variable, so those twelve runs can be laid side by side: | Period | 2025 | 2024 | |---|---|---| | **7** | **+638 pips** | **+1,136 pips** | | **9** | **+360 pips** | **+1,589 pips** | | **12** | **+47 pips** | **+1,620 pips** | | 14 | −325 pips | +2,014 pips | | 18 | −1,682 pips | +1,781 pips | | 20 | −621 pips | +2,066 pips | | **25** | **+381 pips** | **+578 pips** | | 30 | +220 pips | −944 pips | | 40 | +1,042 pips | −1,802 pips | | **50** | **+37 pips** | **+332 pips** | | 60 | −922 pips | −215 pips | | 75 | −1,857 pips | +1,696 pips | Six of the twelve changed sign between the years. One lost in both (60). **Five finished positive in both: 7, 9, 12, 25 and 50**: the short end of the range, plus 25 and 50. Period 7 made +637.8 pips in 2025 and +1,135.9 in 2024, the largest two-year pair among those five settings. Period 25 (Aroon's conventional default) is positive in both years too, but **its standing among the five swaps around**: second behind period 7 in 2025, second from last in 2024. | Year | Trades | Win rate | Profit factor | Net | Max drawdown | |---:|---:|---:|---:|---:|---:| | 2025 | 228 | 41.23% | 1.056 | +380.6 pips | 1,530.0 pips | | 2024 | 237 | 36.29% | 1.075 | +578.4 pips | 1,440.8 pips | A profit factor of about 1.06 both years, with a drawdown 4.0 times the profit in 2025 and 2.5 times in 2024. **Positive in both years, yes, but thin enough that it is hard to call it an edge.** Looking across all 84, there was not much better on offer. ## Can win rate choose the setting? The highest hourly win rate in 2025 came from period 30 with threshold 90, at 49.62%. The same setting recorded a 36.43% win rate and +335.4 pips in 2024. **One year's win rate does not determine the following year's result.** Win rates spanned 32.50% to 49.62%. ## Do filters help? Added to period 25, crossover: | Condition | Year | Trades | Win rate | Net | |---|---:|---:|---:|---:| | Aroon alone | 2025 | 228 | 41.23% | +380.6 pips | | Aroon alone | 2024 | 237 | 36.29% | +578.4 pips | | ADX ≥ 20 | 2025 | 108 | 40.74% | +46.0 pips | | ADX ≥ 20 | 2024 | 123 | 33.33% | +38.8 pips | | ADX ≥ 25 | 2025 | 57 | 36.84% | −698.7 pips | | ADX ≥ 25 | 2024 | 66 | 37.88% | −465.7 pips | | ADX ≥ 30 | 2025 | 30 | 40.00% | −246.3 pips | | ADX ≥ 30 | 2024 | 33 | 39.39% | +59.5 pips | | Tokyo hours only (UTC 0–8) | 2025 | 71 | 47.89% | +696.3 pips | | Tokyo hours only (UTC 0–8) | 2024 | 75 | 37.33% | −80.1 pips | | London/NY hours only (UTC 7–21) | 2025 | 149 | 36.91% | −1,076.4 pips | | London/NY hours only (UTC 7–21) | 2024 | 162 | 37.04% | +849.7 pips | **Every ADX threshold reduced net profit in both years**, and ADX ≥ 25 turned both of them negative. Screening for trend strength simply did not work on this combination. Tokyo hours increased annual net in 2025, while London and New York hours increased it in 2024. QQE showed the same year split. ## Do stops and targets help? | Exit rule | Year | Trades | Win rate | Net | |---|---:|---:|---:|---:| | Opposite reading only | 2025 | 228 | 41.23% | +380.6 pips | | Opposite reading only | 2024 | 237 | 36.29% | +578.4 pips | | Stop 30 / target 60 | 2025 | 212 | 35.38% | +217.9 pips | | Stop 30 / target 60 | 2024 | 228 | 39.47% | +1,084.2 pips | | Stop 50 / target 100 | 2025 | 213 | 40.85% | +1,703.7 pips | | Stop 50 / target 100 | 2024 | 223 | 37.22% | +1,058.6 pips | | Stop 100 / target 200 | 2025 | 213 | 41.78% | +1,716.4 pips | | Stop 100 / target 200 | 2024 | 223 | 38.12% | +1,888.2 pips | | Stop 50 / target 50 | 2025 | 212 | 50.94% | +845.4 pips | | Stop 50 / target 50 | 2024 | 230 | 48.26% | +795.2 pips | | Time exit after 24 bars | 2025 | 220 | 48.18% | +2,100.2 pips | | Time exit after 24 bars | 2024 | 240 | 42.50% | +1,324.0 pips | **Four of the five added exits beat the plain opposite-reading exit in both years**: stop 100 / target 200 at +1,716.4 and +1,888.2, and the 24-bar time exit at +2,100.2 and +1,324.0. Read the other way round, that says **Aroon's own exit is the weak part**. Waiting for the opposite reading gives back what the move produced. Stop 50 / target 50 also lifts the win rate to about 50% in both years (50.94% and 48.26%). ## How much does the spread take? The rule from the QQE test holds here: the more often a setting trades, the more the spread decides it. | Timeframe | Period | Reading | Trades | Spread | Annual net | |---|---:|---|---:|---:|---:| | 1-hour | 30 | Level 100 | 121 | 0.0 pips | +2,380.0 pips | | 1-hour | 30 | Level 100 | 121 | 0.3 pips | +2,343.7 pips | | 1-hour | 30 | Level 100 | 121 | 0.6 pips | +2,307.4 pips | | 1-hour | 30 | Level 100 | 121 | 1.0 pips | +2,259.0 pips | | 1-hour | 30 | Level 100 | 121 | 1.5 pips | +2,198.5 pips | | 1-hour | 30 | Level 100 | 121 | 2.0 pips | +2,138.0 pips | | 15-minute | 25 | Crossover | 1,031 | 0.0 pips | +1,101.1 pips | | 15-minute | 25 | Crossover | 1,031 | 0.3 pips | +791.9 pips | | 15-minute | 25 | Crossover | 1,031 | 0.6 pips | +482.6 pips | | 15-minute | 25 | Crossover | 1,031 | 1.0 pips | +70.2 pips | | 15-minute | 25 | Crossover | 1,031 | 1.5 pips | −445.3 pips | | 15-minute | 25 | Crossover | 1,031 | 2.0 pips | −960.8 pips | Across the full range the hourly setting gives up 242 pips; the 15-minute one gives up 2,062 and **turns negative at 1.5 pips.** Which matters because the 15-minute chart was Aroon's one consistent timeframe. That consistency sits on a high trade count, and it does not survive a wider spread. The same relationship holds for the other indicators tested this way ([RVI](/blog/rvi-settings), [Supertrend](/blog/supertrend-settings)). On this evidence Aroon reads better as one input among several than as something to trade by itself. That four of five stop-and-target variants beat its own exit in both years points the same way. [Building conditions without code](/blog/backtest-without-coding) is the quickest way to see what adding it to something else does on your own pair. ## Notes - One pair, two years - Four-hour configurations trade 15 to 188 times a year. Win rates and profit factors at that count are wide numbers before anything else is said about them - Aroon's calculation varies little between tools, but the trigger logic (how a cross is defined, whether a threshold comparison is inclusive) does. These figures come from Formiq's implementation - Entries and exits both fill at bar closes - The spread is modelled as a flat 0.3 pips throughout; real spreads move with the session and around releases - [Volatility varies by hour of the day](/data), which the session-filter rows are exposed to --- ### MFI Oversold Entry Frequency: 240 Settings Tested URL: https://formiq.jp/blog/mfi-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: MFI, Money Flow Index, indicators, volume, USDJPY Also available in: ja — https://formiq.jp/ja/blog/mfi-settings MFI, the Money Flow Index, is usually introduced as "RSI weighted by volume." Below 20 is oversold and above 80 overbought, the same reading as RSI with volume behind it. I tested it on USD/JPY across the whole of 2025: six periods × five entry levels × four exit levels × two directions, 240 settings, on three timeframes, then the same 240 again on 2024. **The first finding came before any question of performance. On forex's higher timeframes, this indicator hardly trades at all.** On four-hour bars, **48 of the 240 settings took no trades in a year** and 137 took fewer than ten. On hourly bars, 20 took none and 81 took fewer than ten. Before arguing about the best period, there is the question of whether **there is a sample to argue about.** ## Does MFI work on forex? For MFI to drop below 20, the volume-weighted total of down bars has to dominate the up bars across the whole lookback. That is a narrower condition than RSI's, so the extremes are reached less often. Raise the timeframe and there are fewer bars for the condition to occur in. A year of four-hour bars is about 1,610 candles. Choose period 28 with a level of 10 there and going a whole year without the condition lining up is an ordinary outcome, not a freak one. | Timeframe | Median trades a year (of 240 settings) | Settings with 0 trades | Settings under 10 trades | |---|---|---|---| | 15-minute | 80 | 8 / 240 | 54 / 240 | | 1-hour | 23 | 20 / 240 | 81 / 240 | | 4-hour | 5 | 48 / 240 | 137 / 240 | (2025, long and short sides combined.) ## Does the volume premise hold? One more thing worth stating before the results. MFI's calculation takes volume as an input. In equities that volume is consolidated at an exchange and comes close to meaning "how much money moved." **Forex has no such consolidated figure.** What appears on a chart is what a single source observed. So when MFI is used on forex, the weighting that gives the indicator its name rests on weaker ground than the textbook description implies. Every number below sits on that footing. ## Adding MFI to a chart | Platform | How to add it | |---|---| | MT4 / MT5 | Built in — Navigator, Oscillators, Money Flow Index | | TradingView | In the indicator list as "Money Flow Index" | | Browser (Formiq) | In the indicator list, with an adjustable period | ## How this was measured The rule is **directional**: "below 20" can only be a buy and "above 80" can only be a sell, so the long fade and short fade are tested as separate trading rules. | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Long side | Buy when MFI falls below the entry level (10–30), close when it rises above the exit level (50–90) | | Short side | The mirror (sell at 70–90, close at 10–50) | | Stops and targets | None | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Combinations | 6 periods (7–40) × 5 entry levels × 4 exit levels × 2 directions = 240, per timeframe and per year | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | ## How did the textbook 14, 20/80 do? Period 14, buy below 20, close above 80: the version everyone quotes. | Timeframe | Side | Year | Trades | Win rate | Annual net | |---|---|---:|---:|---:|---:| | 15-minute | Long | 2025 | 145 | 61.38% | +1,379.1 pips | | 15-minute | Long | 2024 | 113 | 62.83% | −654.2 pips | | 15-minute | Short | 2025 | 146 | 60.27% | +1,426.3 pips | | 15-minute | Short | 2024 | 113 | 50.44% | −2,236.8 pips | | 1-hour | Long | 2025 | 37 | 56.76% | −108.0 pips | | 1-hour | Long | 2024 | 27 | 77.78% | +245.4 pips | | 4-hour | Long | 2025 | 9 | 77.78% | +570.4 pips | | 4-hour | Long | 2024 | 6 | 50.00% | −277.0 pips | The 15-minute long setup had similar win rates in the two years, 61.38% and 62.83%, while annual net reversed from +1,379.1 to −654.2 pips. Each dot is one of the 120 long-side settings: 2024 across, 2025 up. The red dot is the textbook setting, sitting top-left: heavily negative one year, heavily positive the next. Across the whole cloud, larger 2024 results do not visibly line up with larger 2025 results. The textbook setting is one of the points that changes sign. ## Do the long and short sides differ? | Timeframe | Side | Year | Profitable settings | Median annual net | Profitable in both years | |---|---|---:|---:|---:|---:| | 15-minute | Long | 2025 | 74 / 120 | +35 pips | 50 / 120 | | 15-minute | Long | 2024 | 77 / 120 | +142 pips | 50 / 120 | | 15-minute | Short | 2025 | 80 / 120 | +97 pips | 7 / 120 | | 15-minute | Short | 2024 | 11 / 120 | −1,195 pips | 7 / 120 | | 1-hour | Long | 2025 | 46 / 120 | −31 pips | 31 / 120 | | 1-hour | Long | 2024 | 56 / 120 | 0 pips | 31 / 120 | | 1-hour | Short | 2025 | 37 / 120 | −186 pips | 3 / 120 | | 1-hour | Short | 2024 | 5 / 120 | −1,214 pips | 3 / 120 | | 4-hour | Long | 2025 | 77 / 120 | +156 pips | 53 / 120 | | 4-hour | Long | 2024 | 65 / 120 | +84 pips | 53 / 120 | | 4-hour | Short | 2025 | 64 / 120 | +33 pips | 7 / 120 | | 4-hour | Short | 2024 | 20 / 120 | −603 pips | 7 / 120 | **The short side breaks in 2024 on all three timeframes.** USD/JPY rose 1,632 pips that year, and selling "overbought" through that does not survive. [The same thing happened to Williams %R](/blog/williams-r-settings). The long side hovers around zero: +35 and +142 pips on 15-minute, −31 and 0 on hourly. **Neither winning nor losing much** is what this indicator did on its own. ## What about settings with enough trades? Restricting to settings that took at least 20 trades (hourly long side, 2025, 65 settings): - Win rate range: 47.5% to 72.73% - Median net: **−304 pips** - Profitable: 13 of 65 The high win rates did not produce positive median net. As in the Williams %R test, the counter-trend exit closes after a return toward the centre, limiting the average win while a continued move can expand the average loss. Note also that including the near-empty settings puts rows with 0% and 100% win rates into the table. A win rate computed from one or two trades a year is not a win rate in any useful sense. **Read every win rate here next to its trade count.** Using MFI as a standalone signal on forex is not supported by these numbers. On a timeframe where it fires often enough, and alongside a separate judgement about direction, is as far as this goes. The same method applied to [Williams %R](/blog/williams-r-settings), [QQE](/blog/qqe-settings), [Aroon](/blog/aroon-settings), [RVI](/blog/rvi-settings) and [Supertrend](/blog/supertrend-settings). ## Notes - One pair, two years - **Forex volume is not a market-wide figure.** MFI's calculation rests on it - Entries and exits both fill at bar closes - Settings with very few trades are included in the aggregates; the figures excluding them are given separately above --- ### OsMA Settings by Timeframe: 54 Settings Tested URL: https://formiq.jp/blog/osma-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: OsMA, MACD, indicators, parameters, USDJPY Also available in: ja — https://formiq.jp/ja/blog/osma-settings OsMA (Moving Average of Oscillator) plots the gap between the MACD line and its signal line as a histogram. Above zero means the MACD line is above its signal; below means the reverse. I tested three parameters across 54 combinations on USD/JPY through the whole of 2025, on three timeframes, then ran the same 54 on 2024. **On hourly bars, all 54 were profitable in both years.** Across the six indicators tested this way, no other produced a clean sweep in even one year. But that is not the end of it: **the same 54 settings on 15-minute bars managed one profitable in both years.** ## Which timeframe suits OsMA? | Timeframe | Year | Profitable settings | Median annual net | Median annual trades | |---|---:|---:|---:|---:| | 15-minute | 2025 | 39 / 54 | +433 pips | 1,943 | | 15-minute | 2024 | 13 / 54 | −968 pips | — | | 1-hour | 2025 | 54 / 54 | +1,874 pips | 460 | | 1-hour | 2024 | 54 / 54 | +2,135 pips | — | | 4-hour | 2025 | 43 / 54 | +590 pips | 109 | | 4-hour | 2024 | 29 / 54 | +97 pips | — | Thirteen of 54 in 2024 on 15-minute bars. At around 1,940 trades a year, the spread alone is a large bill: [the same relationship held for every indicator in this series](/blog/rvi-settings). Four-hour bars cut the count to about 110 trades, leaving a smaller sample. Only 22 of 54 settings were profitable in both years, below the hourly result. ## Which setting is best? | Setting | Year | Annual net | |---|---:|---:| | Default 12/26/9 | 2025 | +2,223.6 pips | | Default 12/26/9 | 2024 | +3,152.1 pips | | 2025 maximum, 6/18/9 | 2025 | +3,133.1 pips | | 2025 maximum, 6/18/9 | 2024 | +2,453.6 pips | | 2025 minimum, 24/26/15 | 2025 | +575.2 pips | | 2025 minimum, 24/26/15 | 2024 | +2,912.7 pips | The third row is the one that matters. **The worst hourly setting still made +575.2 pips in 2025 and +2,912.7 in 2024.** There was never a question of picking a profitable setting versus an unprofitable one. The settings changed order between years, but all 54 remained profitable in both, so the profitable-versus-unprofitable decision did not change. Trade counts across the 54 ran from 260 to 831 a year. The parameters move the frequency; profit survives at every frequency here. ## Why a 36.9% win rate profited Across the 54 hourly settings in 2025, win rates ran from 34.02% to 39.89%, while annual net ranged from +575.2 to +3,133.1 pips. Holding in the direction of a zero-line break means losing often and small while the winners run. It is the exact inverse of [what Williams %R showed](/blog/williams-r-settings). **Win rate alone does not compare annual net results.** ## Adding OsMA to a chart | Platform | How to add it | |---|---| | MT4 / MT5 | Built in — Navigator, Oscillators, Moving Average of Oscillator | | TradingView | Add MACD and read its histogram (the same calculation) | | Browser (Formiq) | In the indicator list, with fast, slow and signal all adjustable | ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Buy | OsMA crosses above zero, filled at that bar's close | | Sell | It crosses below zero | | Exit | The opposite zero crossing only — no stop, no target, no time exit | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Combinations | 5 fast (6–24) × 4 slow (18–60) × 3 signal (5–15), keeping fast < slow: 54 | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | ## Can four-hour settings be chosen? | Setting | Year | Annual net | |---|---:|---:| | 2025 maximum, 9/18/9 | 2025 | +2,184.6 pips | | 2025 maximum, 9/18/9 | 2024 | −1,052.9 pips | | 2024 maximum, 24/60/15 | 2025 | −1,525.0 pips | | 2024 maximum, 24/60/15 | 2024 | +1,871.7 pips | The 24/60/15 setting moved from +1,871.7 pips in 2024 to −1,525.0 pips in 2025. The 9/18/9 setting moved from −1,052.9 pips to +2,184.6 pips. Median annual trade count was 109, and one year did not identify the next year's setting. The same indicator changes character with the timeframe, including whether choosing settings means anything at all. The same method applied to [QQE](/blog/qqe-settings), [Aroon](/blog/aroon-settings), [RVI](/blog/rvi-settings), [Supertrend](/blog/supertrend-settings), [Williams %R](/blog/williams-r-settings) and [MFI](/blog/mfi-settings). ## Notes - One pair, two years - OsMA inherits whatever the MACD implementation does, including how its EMAs are seeded. These figures come from Formiq's - Entries and exits both fill at bar closes - The spread is modelled as a flat 0.3 pips; the 15-minute result depends heavily on that - Four-hour settings trade 69 to 220 times a year: wide numbers at that count --- ### QQE Settings Compared: 200 Settings Tested URL: https://formiq.jp/blog/qqe-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: QQE, indicators, parameters, backtesting, USDJPY Also available in: ja — https://formiq.jp/ja/blog/qqe-settings Search for QQE settings and you land on the same three numbers almost every time: RSI period 14, smoothing 5, factor 4.236. Most tools ship with them. So are they the right ones? I backtested 200 parameter combinations on USD/JPY across the whole of 2025, ran the same 200 on three timeframes, and then ran all of it again on 2024 to see whether anything survived the change of year. The headline answer: **the largest 2025 annual net came from RSI period 17, smoothing 3, factor 4.236: 465 trades for +3,341.0 pips.** A different combination made +4,027.1 pips in 2024, then returned +1,186.1 pips in 2025, below that year's +1,822.3-pip median. Choosing from one year's annual net can leave the next year below the middle of the same 200-setting grid. ## What QQE is measuring QQE (Quantitative Qualitative Estimation) does not read RSI directly. It smooths RSI, then compares that smoothed line against a band built from the line's own volatility. 1. Calculate RSI (**RSI period**) 2. Smooth it with an EMA (**smoothing, or SF**) 3. Take the bar-to-bar change of that smoothed line and smooth it too 4. Multiply by a coefficient to set the band width (**factor**, 4.236 by default) 5. Read trend direction from which side of the band the smoothed RSI sits on QQE has three parameters: RSI period, smoothing, and factor. | Parameter | What raising it does | |---|---| | RSI period | Dulls the underlying RSI. Fewer false turns, slower reaction | | Smoothing (SF) | Flattens the line. Fewer signals | | Factor | Widens the band. Fewer signals, each trade held longer | The measured version: on hourly bars in 2025, smoothing of 2 averaged 689 trades a year and smoothing of 12 averaged 369. Raising factor from 2 to 5.236 took the average from 656 trades to 391. Raising RSI period, smoothing, or factor reduced trade count and therefore reduced the total spread paid. ## Adding QQE to a chart QQE is not built into every platform, so the route differs. | Platform | How to add it | |---|---| | MT4 / MT5 | Not included. Download an .ex4 or .mq4 file, drop it into `MQL4/Indicators` (`MQL5/Indicators` on MT5), restart the terminal | | TradingView | Search the indicator list for "QQE" and add a community script | | Browser (Formiq) | Included; pick it from the indicator list | One caveat worth stating up front: **QQE implementations differ.** Whether the smoothing uses an EMA or Wilder's method, and how the band's trailing rule is written, changes where the line sits. The numbers here come from Formiq's implementation, and another script fed the same three parameters may not produce the same trades. ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Buy | QQE line crosses above the signal line, filled at that bar's close | | Sell | QQE line crosses below the signal line, filled at that bar's close | | Exit | The opposite crossover only — no stop, no target, no time exit | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Combinations | 8 RSI periods (6–25) × 5 smoothing values (2–12) × 5 factors (2–5.236) = 200, run per timeframe and per year | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | Holding to the opposite crossover with no stop keeps the test about QQE and nothing else. Stops and targets are measured separately further down. ## Are the defaults good enough? The 14 / 5 / 4.236 defaults cleared the 200-setting median in both years. Across the 200 hourly settings, RSI period 17, smoothing 3 and factor 4.236 produced the largest 2025 annual net: 465 trades, +3,341.0 pips, a 42.37% win rate and a 1.348 PF. Its 2024 annual net was +2,553.6 pips. The defaults, 14 / 5 / 4.236, returned +2,182.3 pips in 2025 and +2,309.7 pips in 2024. The medians across the 200 settings were +1,822.3 and +2,088 pips. Showing the maximum beside the defaults and the median keeps all three comparisons on the same metric. The losing end is worth the same attention. The worst 2025 combination was RSI 6 / smoothing 2 / factor 4.236 at −1,516.0 pips, and it lost 580.9 pips in 2024 as well: **the only one of the 200 hourly combinations to lose in both years**. It is the most sensitive setting on the grid: the shortest RSI period paired with the least smoothing. ## What are the best QQE settings? The largest result of one year finished 636.2 pips below the next year's median. **RSI 10, smoothing 8 and factor 5.236 made +4,027.1 pips in 2024, then +1,186.1 pips in 2025.** The 2025 median across all 200 settings was +1,822.3 pips, so selecting that setting from its 2024 annual net left it 636.2 pips below the next year's median. The figure at the top plots one dot per setting, with 2024 annual net pips on the horizontal axis and 2025 annual net pips on the vertical axis. If the same settings produced similar annual net pips in both years, the dots would cluster along a rising line. Instead, they are widely scattered. None of which means QQE failed. On hourly bars, 194 of the 200 combinations were profitable in 2025 and 188 in 2024, with **183 profitable in both years**. The indicator worked across this two-year window on this pair. What did not work was the search for the single best number. ## Which timeframe suits QQE? 183 of 200 hourly settings were profitable in both years, against 3 on the 15-minute chart. The same 200 combinations, changing only the timeframe: | Timeframe | Year | Profitable settings | Median annual net | Profitable both years | |---|---:|---:|---:|---:| | 15-minute | 2025 | 78 / 200 | −305 pips | 3 / 200 | | 15-minute | 2024 | 4 / 200 | −2,031 pips | 3 / 200 | | 1-hour | 2025 | 194 / 200 | +1,822 pips | 183 / 200 | | 1-hour | 2024 | 188 / 200 | +2,088 pips | 183 / 200 | | 4-hour | 2025 | 197 / 200 | +1,385 pips | 136 / 200 | | 4-hour | 2024 | 138 / 200 | +642 pips | 136 / 200 | On 15-minute bars in 2024, **four combinations of 200 finished positive**. All four sit at the blunt end of the grid (RSI period 20 or 25, smoothing 8 or 12) and three of them were profitable in 2025 as well. Four-hour bars beat hourly on the 2025 count, then fell to 138 in 2024. Counting only the settings that came through both years, hourly leads with 183 of 200 against 136. Holding the parameters at the defaults and moving nothing but the timeframe: | Timeframe | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | 15-minute | 2025 | 1,710 | 37.37% | −460.6 pips | | 15-minute | 2024 | 1,739 | 38.59% | −1,008.1 pips | | 1-hour | 2025 | 415 | 42.89% | +2,182.3 pips | | 1-hour | 2024 | 407 | 43.73% | +2,309.7 pips | | 4-hour | 2025 | 105 | 49.52% | +1,855.6 pips | | 4-hour | 2024 | 107 | 32.71% | +130.8 pips | ## Was the 15-minute chart losing before costs? Not because the indicator degrades. Because of how many times the cost is paid. Half of it was, and the other half was not: the two years answer differently. Set the spread to zero and the result is the gross edge before costs. Medians are across all 200 combinations. | Timeframe | Year | Gross | Net | |---|---:|---:|---:| | 15-minute | 2025 | +181 pips | −305 pips | | 15-minute | 2024 | −1,251 pips | −2,031 pips | | 1-hour | 2025 | +1,976 pips | +1,822 pips | | 1-hour | 2024 | +2,257 pips | +2,088 pips | | 4-hour | 2025 | +1,428 pips | +1,385 pips | | 4-hour | 2024 | +681 pips | +642 pips | **In 2025 the 15-minute chart was positive before costs.** The gross median was +181 pips with 116 of 200 combinations in profit. Charging 0.3 pips drops the median to −305 and the count to 78. **The median crosses zero at about 0.12 pips of spread**: below any spread that actually exists. **In 2024 it was already at −1,251 pips gross.** That year it lost on the trades themselves, and the spread only deepened it. "The 15-minute chart loses to costs" is true of 2025 here and false of 2024. The cost itself behaves predictably. Taking the spread from 0 to 2.0 pips moves the medians from +181 to −3,381 on 15-minute bars, +1,976 to +969 on hourly, and +1,428 to +1,140 on four-hour. **At 1.5 pips every one of the 200 fifteen-minute combinations is negative, while 196 of the four-hour ones are still positive.** What you give up is the trade count times the spread, exactly. This is not specific to QQE. [RVI](/blog/rvi-settings) trades 1,094 times a year on the same terms and gives up 2,188 pips going from 0 to 2.0; [Supertrend](/blog/supertrend-settings) trades 152 times and gives up 304. Both match trade count × spread without error, which is why checking it against your own account's spread is worth more than another pass over the parameters. ## Which parameters held across both years? Smoothing and RSI period, and nothing else. Each RSI-period, smoothing, and factor value was averaged over every combination of the other two parameters. Smoothing and RSI period were the two parameters whose annual-net ordering was consistent in 2024 and 2025. | Smoothing | 2025 average | 2024 average | |---|---|---| | 2 | +923 pips | +563 pips | | 3 | +1,807 pips | +1,428 pips | | 5 | +2,167 pips | +2,024 pips | | 8 | +1,895 pips | +2,691 pips | | 12 | +1,431 pips | +2,793 pips | **Smoothing of 2 came last in both years**, at under half the return of the better values. | RSI period | 2025 average | 2024 average | |---|---|---| | 6 | +1,173 pips | +1,266 pips | | 10 | +1,645 pips | +1,876 pips | | 14 | +1,736 pips | +1,937 pips | | 20 | +1,790 pips | +2,177 pips | | 25 | +1,807 pips | +2,206 pips | **A short RSI period came last in both years too**: monotonically improving out to 25 in 2024, and flattening out from 17 onward in 2025. Both findings say the same thing: do not set these short. Factor swapped optima: 2.618 averaged best in 2025 (+1,840 pips), 4.236 in 2024 (+2,255 pips). It does have one consistent effect across both years: **raising it raises the win rate**, from 39.8% to 43.5% in 2025 and 38.8% to 43.2% in 2024, while trade count falls from 656 to 391. Factor trades frequency for win rate. It does not maximise profit. ## Does a higher win rate make more money? It did not. Across the 200 hourly combinations in 2025, win rates ranged from 37.09% to 47.32%. The filter comparison below shows what happened to trade count and annual net when the win rate changed. Adding the filters people usually reach for, on top of the defaults: | Condition | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | QQE alone | 2025 | 415 | 42.89% | +2,182.3 pips | | QQE alone | 2024 | 407 | 43.73% | +2,309.7 pips | | Only when ADX ≥ 20 | 2025 | 261 | 41.76% | +830.9 pips | | Only when ADX ≥ 20 | 2024 | 248 | 44.76% | +1,652.9 pips | | Only when ADX ≥ 25 | 2025 | 198 | 43.43% | +774.9 pips | | Only when ADX ≥ 25 | 2024 | 175 | 45.14% | −257.4 pips | | Only when ADX ≥ 30 | 2025 | 128 | 42.19% | +582.1 pips | | Only when ADX ≥ 30 | 2024 | 122 | 42.62% | −70.9 pips | | ADX ≥ 20 plus DI agreement | 2025 | 91 | 35.16% | −408.4 pips | | ADX ≥ 20 plus DI agreement | 2024 | 83 | 43.37% | +686.5 pips | | Tokyo hours only (UTC 0–8) | 2025 | 167 | 47.90% | +1,281.0 pips | | Tokyo hours only (UTC 0–8) | 2024 | 151 | 47.68% | +702.6 pips | | London/NY hours only (UTC 7–21) | 2025 | 223 | 38.57% | +240.4 pips | | London/NY hours only (UTC 7–21) | 2024 | 222 | 44.14% | +1,582.7 pips | **Not one of the six beat the unfiltered version in both years.** Tightening ADX cuts the trade count and cuts the total with it; at ADX ≥ 25 the 2024 result crosses into negative territory at −257.4 pips. The filters increased win rate but reduced annual net. Restricting to Tokyo hours lifts the win rate from 42.89% to 47.90%, and ADX ≥ 25 lifts it to 43.43%. Both filters remove trades, and neither increases annual net in both years. ## Stops and targets Same pattern again. Adding exits to RSI 17 / smoothing 3 / factor 4.236, with the crossover exit still active: | Exit rule | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | Opposite crossover only | 2025 | 465 | 42.37% | +3,341.0 pips | | Opposite crossover only | 2024 | 479 | 40.71% | +2,553.6 pips | | Stop 50 / target 100 | 2025 | 420 | 41.67% | +2,674.1 pips | | Stop 50 / target 100 | 2024 | 429 | 38.93% | +1,487.7 pips | | Stop 50 / target 50 | 2025 | 425 | 48.24% | +2,300.5 pips | | Stop 50 / target 50 | 2024 | 432 | 43.06% | +85.7 pips | | Stop 100 / target 200 | 2025 | 459 | 42.48% | +3,182.5 pips | | Stop 100 / target 200 | 2024 | 469 | 39.66% | +1,108.7 pips | | Time exit after 24 bars | 2025 | 457 | 43.11% | +3,519.5 pips | | Time exit after 24 bars | 2024 | 469 | 41.15% | +2,609.5 pips | **All three stop-and-target pairs came in below the plain crossover exit in both years.** Stop 50 / target 50 pushes the 2025 win rate to 48.24% and still gives back a thousand pips; in 2024 it finishes at +85.7, which is almost nothing. The 24-bar time exit is the exception: **it beat the baseline in both years**, +3,341.0 to +3,519.5 and +2,553.6 to +2,609.5. It is the only addition in this article that improved both. Nothing in this test rewarded narrowing QQE down to one triple of numbers. In the plain [RSI test](/blog/rsi-settings), the fade had the most settings profitable in both years at period 30, while the level-cross rule had the most at period 9. For QQE, timeframe, avoiding extreme parameter values, and cost per trade mattered more than selecting one exact triple. If you want to check the same thing on your own pair and dates, [building the rules without writing code](/blog/backtest-without-coding) takes about as long as reading this article. ## Notes - The sample covers USD/JPY in 2024 and 2025; it does not include other pairs or periods - QQE implementations vary between scripts; these figures come from Formiq's and may not match another script fed the same parameters - Four-hour settings range from 67 to 257 trades a year, so win rates and profit factors from the smaller samples vary more - Entries and exits both fill at bar closes. Real fills differ - The spread is modelled as a flat 0.3 pips for the whole period. Real spreads move with the session and around data releases --- ### RVI Periods and Spread Costs: 2 Years Tested URL: https://formiq.jp/blog/rvi-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: RVI, indicators, parameters, spread, USDJPY Also available in: ja — https://formiq.jp/ja/blog/rvi-settings RVI (the Relative Vigor Index) divides how far the close finished from the open by the bar's high-to-low range. The idea behind it: in a rising market, bars should close near their highs. So what does changing its period do? I backtested ten periods, from 4 to 55, across the whole of 2025 on USD/JPY, on three timeframes, then ran all of it again on 2024. The finding that shaped this article: **widening the period from 4 to 55 barely changed how often RVI traded.** On hourly bars, period 4 produced 1,336 trades a year and period 55 produced 1,078, a reduction of less than 30% despite a fourteen-fold increase in period. And at more than five hundred trades a year, frequency is not a detail. It decides the result. ## What RVI measures For each bar, two quantities: - **Numerator**: close minus open, weighted 1:2:2:1 across the last four bars - **Denominator**: high minus low, weighted the same way Divide one by the other and you have RVI, moving roughly between −1 and +1. Bars that keep closing near their highs push it positive. That value is then **summed over N bars** to make the plotted line, and **N is the only place the period appears.** There is a second line: the signal. It is RVI itself, weighted 1:2:2:1 over the last four bars. **The period does not enter this calculation at all.** Trades come from the two lines crossing. So **one line can be smoothed by the period while the other always chases it within four bars.** The two can never drift apart for long, which means how often they cross is set by the signal line's construction, not by the period. The measurements matched the theory. ## Adding RVI to a chart | Platform | How to add it | |---|---| | MT4 / MT5 | Built in — Navigator, Oscillators, Relative Vigor Index | | TradingView | In the indicator list as "Relative Vigor Index" | | Browser (Formiq) | In the indicator list, with an adjustable period | ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Buy | RVI crosses above its signal line, filled at that bar's close | | Sell | RVI crosses below it | | Exit | The opposite crossover only — no stop, no target, no time exit | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Periods tested | 4 / 6 / 8 / 10 / 14 / 18 / 24 / 30 / 40 / 55, per timeframe and per year | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | ## Does a longer period trade less? The figure at the top of this article draws the table below. | Period | Year | Trades | Annual net | |---:|---:|---:|---:| | 4 | 2025 | 1,336 | +511.0 pips | | 4 | 2024 | 1,325 | +944.8 pips | | 6 | 2025 | 1,110 | +38.6 pips | | 6 | 2024 | 1,155 | −1,250.0 pips | | 8 | 2025 | 1,048 | +272.4 pips | | 8 | 2024 | 1,153 | −1,545.5 pips | | 10 | 2025 | 1,094 | +2,233.9 pips | | 10 | 2024 | 1,144 | +184.6 pips | | 14 | 2025 | 1,075 | +1,853.1 pips | | 14 | 2024 | 1,126 | +1,910.1 pips | | 18 | 2025 | 1,043 | −183.8 pips | | 18 | 2024 | 1,114 | −831.9 pips | | 24 | 2025 | 1,081 | +2,810.0 pips | | 24 | 2024 | 1,099 | −991.4 pips | | 30 | 2025 | 1,081 | +1,489.3 pips | | 30 | 2024 | 1,086 | −430.2 pips | | 40 | 2025 | 1,094 | +2,764.9 pips | | 40 | 2024 | 1,089 | +289.0 pips | | 55 | 2025 | 1,078 | +3,071.8 pips | | 55 | 2024 | 1,109 | +606.5 pips | Read the trade columns downward: everything sits between 1,043 and 1,336. Fourteen times the period, under thirty per cent of the frequency. The P&L columns span more than 4,600 pips, from −1,546 to +3,072, and the two years disagree about which end is which. Period 18 lost in both. Period 8 made +272 pips in 2025 and lost 1,546 in 2024. The period therefore **left trading frequency nearly unchanged while annual net moved widely.** One year's result is a weak basis for selecting it. Five of the ten were profitable in both years: 4, 10, 14, 40 and 55. Period 14 produced +1,853.1 and +1,910.1 pips. The default period 10 produced +2,233.9 and +184.6 pips. ## Can you use RVI on 15-minute bars? Not on this evidence. Results by timeframe: | Timeframe | Year | Profitable periods | Median annual net | Median annual trades | |---|---:|---:|---:|---:| | 15-minute | 2025 | 2 / 10 | −683 pips | ~4,400 | | 15-minute | 2024 | 0 / 10 | −1,510 pips | — | | 1-hour | 2025 | 9 / 10 | +1,671 pips | ~1,080 | | 1-hour | 2024 | 5 / 10 | −123 pips | — | | 4-hour | 2025 | 5 / 10 | +15 pips | ~275 | | 4-hour | 2024 | 10 / 10 | +927 pips | — | Hourly and four-hour swap places between the years. Fifteen-minute is bad in both, and the trade count is why. ## Does the spread decide it? It does. Same year, same settings, changing only the spread: | Spread | Timeframe | Period | Trades | Annual net | |---:|---|---:|---:|---:| | 0.0 pips | 1-hour | 10 | 1,094 | +2,562.1 pips | | 0.3 pips | 1-hour | 10 | 1,094 | +2,233.9 pips | | 0.6 pips | 1-hour | 10 | 1,094 | +1,905.7 pips | | 1.0 pips | 1-hour | 10 | 1,094 | +1,468.1 pips | | 1.5 pips | 1-hour | 10 | 1,094 | +921.1 pips | | 2.0 pips | 1-hour | 10 | 1,094 | +374.1 pips | | 0.0 pips | 15-minute | 10 | 4,375 | +396.5 pips | | 0.3 pips | 15-minute | 10 | 4,375 | −916.0 pips | | 0.6 pips | 15-minute | 10 | 4,375 | −2,228.5 pips | | 1.0 pips | 15-minute | 10 | 4,375 | −3,978.5 pips | | 1.5 pips | 15-minute | 10 | 4,375 | −6,166.0 pips | | 2.0 pips | 15-minute | 10 | 4,375 | −8,353.5 pips | The 15-minute version makes **+396.5 pips with the spread set to zero.** Charging even 0.3 turns it negative; at 2.0 it loses 8,353.5. The hourly version is not safe either: across the same range it gives up **2,188 pips**, taking +2,562.1 down to +374.1. Those numbers follow a rule. Same pair, same year, same timeframe, only the indicator differs. | Indicator | Trades a year | Pips lost, 0 → 2.0 spread | |---|---|---| | Supertrend 10/3 | 152 | 304 pips | | Aroon 25 crossover | 228 | 456 pips | | QQE 17/3/4.236 | 465 | 930 pips | | **RVI 10** | **1,094** | **2,188 pips** | Every one is exactly **trades × 2.0 pips**. Trivial arithmetic, which is precisely why it never fails to apply. **Choosing an indicator is choosing how much spread you will pay in a year.** RVI trades more often than any of the others here, so it is the most exposed of them. ## Can win rate choose the period? It cannot. Across the ten hourly periods in 2025, win rates ran from 41.61% to 44.94%, while annual net ranged from −183.8 to +3,071.8 pips. Filters, added to period 10: | Condition | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | RVI alone | 2025 | 1,094 | 42.23% | +2,233.9 pips | | RVI alone | 2024 | 1,144 | 42.05% | +184.6 pips | | ADX ≥ 20 | 2025 | 701 | 43.08% | +1,431.2 pips | | ADX ≥ 20 | 2024 | 713 | 43.34% | −385.6 pips | | ADX ≥ 25 | 2025 | 488 | 44.06% | +1,691.3 pips | | ADX ≥ 25 | 2024 | 500 | 41.80% | −449.7 pips | | ADX ≥ 30 | 2025 | 313 | 44.41% | +915.3 pips | | ADX ≥ 30 | 2024 | 324 | 41.98% | +125.5 pips | | Tokyo hours only | 2025 | 369 | 41.46% | +315.0 pips | | Tokyo hours only | 2024 | 389 | 40.87% | +640.1 pips | | London/NY hours only | 2025 | 636 | 42.45% | +1,876.6 pips | | London/NY hours only | 2024 | 643 | 43.86% | +741.4 pips | **None of them beat the unfiltered 2025 result of +2,233.9 pips.** ADX at 20 and 25 lifts the win rate a point or two and takes 2024 from +184.6 into the red. Signs are a different matter. **Three conditions stayed positive in both years** (ADX ≥ 30 (+915.3 and +125.5), Tokyo hours (+315.0 and +640.1) and London/NY hours (+1,876.6 and +741.4)) each by cutting the trade count to between 30% and 60% of the original. Given that frequency is this indicator's weak point, a filter that removes trades and keeps the sign is worth something. ## Stops and targets | Exit rule | Year | Trades | Annual net | |---|---:|---:|---:| | Opposite crossover only | 2025 | 1,094 | +2,233.9 pips | | Opposite crossover only | 2024 | 1,144 | +184.6 pips | | Stop 30 / target 60 | 2025 | 980 | +1,371.0 pips | | Stop 30 / target 60 | 2024 | 1,044 | −1,188.5 pips | | Stop 50 / target 100 | 2025 | 1,031 | +1,144.0 pips | | Stop 50 / target 100 | 2024 | 1,080 | −972.7 pips | | Stop 100 / target 200 | 2025 | 1,082 | +2,618.3 pips | | Stop 100 / target 200 | 2024 | 1,131 | −332.6 pips | | Stop 50 / target 50 | 2025 | 1,027 | +901.5 pips | | Stop 50 / target 50 | 2024 | 1,076 | −1,658.6 pips | | Time exit after 24 bars | 2025 | 1,094 | +2,222.0 pips | | Time exit after 24 bars | 2024 | 1,144 | +184.6 pips | **All four stop-and-target variants lost money in 2024.** The best of them in 2025 (stop 100, target 200) returned −332.6 the year before. RVI trades are short, averaging 8.0 bars to the exit. Layering a fixed pip distance on top means the price band decides the trade before the indicator does. **The 24-bar time exit changes almost nothing** (identical to the baseline in 2024 and 11.9 pips away from it in 2025) for the same reason: a trade that ends in eight bars rarely reaches a twenty-four-bar limit. Holding to the opposite crossover was steadier across both years. For the same test applied to other indicators, see [QQE across 200 combinations](/blog/qqe-settings), [Aroon across 84](/blog/aroon-settings) and [Supertrend across 36](/blog/supertrend-settings). The conclusions differ by indicator, which is the point of running the same method on each. ## Notes - One pair, two years - RVI varies little between implementations, but signal-line conventions differ. These figures come from Formiq's - Entries and exits both fill at bar closes - The spread is modelled as flat for the whole period; real spreads move with the session, and high-frequency settings feel that most - Four-hour configurations trade about 275 times a year, in a narrow 271–333 band; the period made no difference to frequency there either --- ### Squeeze Momentum by Timeframe: 45 Settings Tested URL: https://formiq.jp/blog/squeeze-momentum-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Squeeze Momentum, indicators, parameters, backtesting, USDJPY Also available in: ja — https://formiq.jp/ja/blog/squeeze-momentum-settings Squeeze Momentum detects a "squeeze" (Bollinger Bands contracting inside the Keltner Channel) and takes the direction the market expands in when that squeeze releases. Contraction, then expansion. I tested three parameters across 45 combinations on USD/JPY through the whole of 2025, on three timeframes, then ran the same 45 on 2024. **On four-hour bars, the profitable count fell from 26 of 45 settings in 2024 to 5 in 2025.** Median annual net moved from +251 to −461 pips, and only 1 of 45 settings was profitable in both years. ## Which timeframe can you use it on? 29 of 45 settings were profitable in both years on the 15-minute chart, against 24 on hourly and 1 on four-hour. | Timeframe | Profitable 2025 | 2025 median | Profitable 2024 | 2024 median | Both years | Trades a year | |---|---|---|---|---|---|---| | **15-minute** | 37 / 45 | +273 pips | 36 / 45 | +468 pips | **29 / 45** | ~366 | | 1-hour | 30 / 45 | +168 pips | 36 / 45 | +814 pips | 24 / 45 | ~81 | | 4-hour | **5 / 45** | −461 pips | 26 / 45 | +251 pips | **1 / 45** | ~21 | The number profitable in both years was 29 of 45 on 15-minute bars, compared with 24 on hourly and 1 on four-hour. The trade count is why. Waiting for a squeeze to release produces few signals to begin with: 81 a year on hourly, about 21 on four-hour. **The problem here is too few trades, not too many.** Even on 15-minute bars it takes about 366 trades a year, under a tenth of what the other indicators produce there. Light on cost, with enough sample to read. ## Does the winner hold the next year? On hourly bars it did. | Setting | 2025 annual net | 2024 annual net | |---|---:|---:| | Default 20/2/1.5 | +398.3 pips | +1,530.5 pips | | 10/2/1 | +1,234.2 pips | +1,536.6 pips | | 20/2.5/2 | +624.7 pips | +2,732.0 pips | | 14/1.5/1.5 | −1,152.8 pips | +2,029.9 pips | The 10/2/1 setting took 225 trades and made +1,234.2 pips in 2025, then +1,536.6 pips in 2024. The 20/2.5/2 setting fell from +2,732.0 pips in 2024 to +624.7 pips in 2025. The default is positive in both, but its +398.3 pips in 2025 sits more than 800 pips behind the top of the table. ## How did the four-hour default do? It won 4 of its 23 trades. | Setting | 2025 annual net | 2024 annual net | |---|---:|---:| | Default 20/2/1.5 | −967.3 pips | +5.6 pips | | 20/2.5/1.5 | +295.5 pips | −188.9 pips | | 10/1.5/1.5 | −1,147.7 pips | +1,683.0 pips | | 20/1.5/2 | −2,399.9 pips | +1,351.4 pips | The default took **23 trades at a 17.39% win rate** in 2025: four winners out of 23, which is all that number says. The 20/2.5/1.5 setting made +295.5 pips in 2025 from **11 trades**, with a 54.55% win rate and 2.507 profit factor. Those figures came from six wins, and its 2024 annual net was −188.9 pips. Four-hour settings trade 0 to 80 times a year, and **8 of the 45 never traded at all.** One of 45 was profitable in both years. ## What do the three settings change? The period sets the value; the two multipliers only decide when you enter. The three parameters do not do the same kind of work. - **Period**: feeds the momentum calculation. It sets the value - **Bollinger multiplier and Keltner multiplier**: used only to detect the squeeze. They do not change the momentum value So the multipliers decide *when* a trade happens and the period decides *which way*. Tightening them (wider Bollinger, narrower Keltner) makes the squeeze harder to satisfy and cuts the signal count. On hourly bars in 2025 the trade count ranged from 0 to 319: tighten too far and 3 of the 45 settings never trade in a year. ## What win rate should you expect? The median win rate across the 45 hourly settings in 2025 was 41.2%. On four-hour bars, even the 29 settings with at least 10 trades ranged from 16.67% to 54.55%. **At about 21 trades a year, win rate cannot be interpreted without trade count and annual net pips.** ## Squeeze Momentum settings and display | Platform | How to add it | |---|---| | MT4 / MT5 | Not included. Download an .ex4 or .mq4 into `MQL4/Indicators` (`MQL5/Indicators` on MT5), restart | | TradingView | In the indicator list as "Squeeze Momentum" | | Browser (Formiq) | In the indicator list, with the period and both band multipliers adjustable | ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Buy | On the bar the squeeze releases, if momentum is positive | | Sell | On the bar the squeeze releases, if momentum is negative | | Exit | When momentum crosses zero against the position — no stop, no target, no time exit | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Combinations | 5 periods (10–40) × 3 Bollinger multipliers (1.5, 2, 2.5) × 3 Keltner multipliers (1, 1.5, 2) = 45 | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | The same method applied to [OsMA](/blog/osma-settings), [Accelerator Oscillator](/blog/ac-oscillator-settings), [Stoch RSI](/blog/stoch-rsi-settings), [STC](/blog/stc-settings), [WaveTrend](/blog/wavetrend-settings), [QQE](/blog/qqe-settings), [Aroon](/blog/aroon-settings), [RVI](/blog/rvi-settings), [Supertrend](/blog/supertrend-settings), [Williams %R](/blog/williams-r-settings) and [MFI](/blog/mfi-settings). ## Notes - One pair, two years - Squeeze Momentum implementations differ (how the momentum regression is taken, which midline it uses). These figures come from Formiq's - Entries are restricted to the bar a squeeze releases on; momentum turns during a squeeze are not taken - Entries and exits both fill at bar closes - The spread is modelled as a flat 0.3 pips - Four-hour settings trade 0 to 80 times a year and 8 of 45 never trade. Win rates and profit factors at that count are wide numbers --- ### STC Periods and Entry Levels: 72 Settings Tested URL: https://formiq.jp/blog/stc-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: STC, Schaff Trend Cycle, indicators, parameters, USDJPY Also available in: ja — https://formiq.jp/ja/blog/stc-settings The Schaff Trend Cycle runs a double stochastic over MACD and bounds the result between 0 and 100: built to turn faster than MACD while cutting the whipsaws a range produces. I tested four parameters across 72 combinations on USD/JPY through the whole of 2025, on three timeframes, then ran the same 72 on 2024. The profitable count across both years was 10 of 72 settings on 15-minute bars, 49 on hourly bars and 36 on four-hour bars. Four-hour 10/30/5 at level 25 made +3,399.7 pips in 2025 and +3,014.6 in 2024, while the four-hour default lost 593.4 pips in 2024. ## How did the default 23/50/10 do? Its annual net moved just 3.4 pips between the two years. | Setting | 2025 annual net | 2024 annual net | |---|---:|---:| | Default 23/50/10, level 25 | +1,718.3 pips | +1,721.7 pips | | 23/80/10, level 40 | +3,024.7 pips | +2,730.4 pips | | 23/80/20, level 10 | −1,670.8 pips | +713.3 pips | **The default's two years differ by 3.4 pips**: +1,718.3 and +1,721.7. The 23/80/10, level-40 setting made +3,024.7 pips in 2025, which was 1,306.4 pips more than the default. A default that was profitable in both years was not the setting with the largest annual net. ## Which setting performed best? On four-hour bars, **10/30/5 at level 25 made +3,399.7 pips in 2025 and +3,014.6 pips in 2024**, the largest annual net among the 72 settings in each year. The sample covers two years of one currency pair. This one setting having the largest annual net twice does not establish that it will do so on another pair or period. For what it is worth, here is the setting itself: | Metric | 2025 | 2024 | |---|---|---| | Trades | 164 | 165 | | Win rate | 50.00% | 40.61% | | Profit factor | 1.657 | 1.575 | | Net | +3,399.7 pips | +3,014.6 pips | About 165 trades a year is high for a four-hour setting. Unlike the 15- and 27-trade profit factors seen on four-hour [Aroon](/blog/aroon-settings) and [Supertrend](/blog/supertrend-settings), this is not a number that thin sampling explains away. ## Which timeframe does STC work on? The hourly chart kept the most settings profitable in both years, 49 of 72. | Timeframe | Profitable 2025 | 2025 median | Profitable 2024 | 2024 median | Both years | Trades a year | |---|---|---|---|---|---|---| | 15-minute | 33 / 72 | −54 pips | 19 / 72 | −1,244 pips | 10 / 72 | ~1,400 | | **1-hour** | 62 / 72 | +1,198 pips | 59 / 72 | +1,234 pips | **49 / 72** | ~340 | | 4-hour | 54 / 72 | +925 pips | 52 / 72 | +642 pips | 36 / 72 | ~90 | At 1,400 trades a year the cost bites again, leaving 10 of 72 on 15-minute bars. Four-hour falls to about 90 trades and 36 of 72, below hourly's 49. ## Does the default work on four-hour? It made money in 2025 and lost in 2024. | Setting | 2025 annual net | 2024 annual net | |---|---:|---:| | Default 23/50/10, level 25 | +1,722.8 pips | −593.4 pips | | 10/30/5, level 25 | +3,399.7 pips | +3,014.6 pips | | 40/80/20, level 40 | −875.6 pips | +1,367.1 pips | On hourly bars the default's two annual results differ by only 3.4 pips. On four-hour bars it made +1,722.8 pips in 2025 and lost 593.4 pips in 2024. Four-hour win rates span 37.50% to 64.29%, which is what 90 trades a year looks like. ## How wide do win rates spread? Across the 72 settings in 2025, hourly win rates ran from 32.98% to 52.40%, with a 42.9% median. Four-hour win rates spanned 37.50% to 64.29%. Four-hour settings averaged about 90 trades a year, so any high win rate needs to be read with its trade count and annual net pips. ## Adding STC to a chart | Platform | How to add it | |---|---| | MT4 / MT5 | Not included. Download an .ex4 or .mq4 into `MQL4/Indicators` (`MQL5/Indicators` on MT5), restart | | TradingView | In the indicator list as "Schaff Trend Cycle" | | Browser (Formiq) | In the indicator list, with fast, slow and stoch period adjustable | ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Buy | STC crosses above the level (10, 25 or 40), filled at that bar's close | | Sell | STC crosses below "100 − level" | | Exit | The opposite level crossing only — no stop, no target, no time exit | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Combinations | 3 fast (10, 23, 40) × 3 slow (30, 50, 80) × 3 stoch periods (5, 10, 20) × 3 levels, keeping fast < slow: 72 | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | The same method applied to [OsMA](/blog/osma-settings), [Accelerator Oscillator](/blog/ac-oscillator-settings), [Stoch RSI](/blog/stoch-rsi-settings), [QQE](/blog/qqe-settings), [Aroon](/blog/aroon-settings), [RVI](/blog/rvi-settings), [Supertrend](/blog/supertrend-settings), [Williams %R](/blog/williams-r-settings) and [MFI](/blog/mfi-settings). ## Notes - One pair, two years - STC implementations differ (the smoothing constants inside the double stochastic). These figures come from Formiq's - Entries and exits both fill at bar closes - The spread is modelled as a flat 0.3 pips - Four-hour settings trade 57 to 172 times a year: wide numbers at that count --- ### Stoch RSI Crossovers by Timeframe: 36 Settings Tested URL: https://formiq.jp/blog/stoch-rsi-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Stoch RSI, Stochastic RSI, indicators, parameters, USDJPY Also available in: ja — https://formiq.jp/ja/blog/stoch-rsi-settings Stoch RSI runs a stochastic over RSI. Rather than reading RSI's level, it reads where RSI sits inside its own range over the last N bars: a second layer of processing that makes it quicker to turn. I tested four parameters across 36 combinations on USD/JPY through the whole of 2025, on three timeframes, then ran the same 36 on 2024. **On hourly bars, only 3 of 36 were positive in both years.** Twenty were positive in 2025 and seven in 2024, but almost never the same ones. On four-hour bars, **30 of 36 settings were positive in both years**, compared with only 3 of 36 on hourly bars despite using the same Stochastic RSI parameters. ## How does the timeframe change it? The result reversed between timeframes. | Timeframe | Year | Profitable settings | Median annual net | Median annual trades | |---|---:|---:|---:|---:| | 15-minute | 2025 | 0 / 36 | −1,139 pips | ~4,850 | | 15-minute | 2024 | 3 / 36 | −1,205 pips | — | | 1-hour | 2025 | 20 / 36 | +162 pips | ~1,230 | | 1-hour | 2024 | 7 / 36 | −1,008 pips | — | | 4-hour | 2025 | 30 / 36 | +1,261 pips | ~300 | | 4-hour | 2024 | 36 / 36 | +2,534 pips | — | The trade count explains it. Two layers of smoothing produce **a great many signals**: about 4,850 a year on 15-minute bars. For comparison, on the same terms [QQE](/blog/qqe-settings) took 465 trades a year and [Supertrend](/blog/supertrend-settings) took 152. Spread cost scales exactly with the count. At 4,850 trades a 0.3 pip spread is about 1,455 pips a year; at 1,230 it is 369. **Only at the four-hour chart's 300 trades does the frequency start to fit the cost.** ## Is it worth tuning on hourly bars? Changing the settings did not pay. | Setting | Year | Annual net | |---|---:|---:| | Default 14/14/3/3 | 2025 | +764.5 pips | | Default 14/14/3/3 | 2024 | −1,811.3 pips | | 2025 maximum, 14/21/3/5 | 2025 | +1,560.4 pips | | 2025 maximum, 14/21/3/5 | 2024 | −2,011.2 pips | | 2024 maximum, 14/7/5/3 | 2025 | −2,200.9 pips | | 2024 maximum, 14/7/5/3 | 2024 | +858.4 pips | | 2025 minimum, 7/7/5/5 | 2025 | −3,221.7 pips | | 2025 minimum, 7/7/5/5 | 2024 | −1,009.9 pips | The 14/7/5/3 setting moved from +858.4 pips in 2024 to −2,200.9 pips in 2025. The 14/21/3/5 setting moved from +1,560.4 pips in 2025 to −2,011.2 pips in 2024. **The setting selected in either year lost money in the other.** ## Can four-hour settings be selected? Four-hour bars profited, but one year still did not select the next. | Setting | Year | Annual net | |---|---:|---:| | Default 14/14/3/3 | 2025 | +1,437.3 pips | | Default 14/14/3/3 | 2024 | +2,430.4 pips | | 2025 maximum, 21/21/3/3 | 2025 | +2,333.0 pips | | 2025 maximum, 21/21/3/3 | 2024 | +2,205.8 pips | | 2024 maximum, 7/21/5/5 | 2025 | −863.4 pips | | 2024 maximum, 7/21/5/5 | 2024 | +3,799.1 pips | | 2025 minimum, 7/14/5/5 | 2025 | −1,389.5 pips | | 2025 minimum, 7/14/5/5 | 2024 | +1,918.2 pips | The 7/21/5/5 setting moved from +3,799.1 pips in 2024 to −863.4 pips in 2025. The 21/21/3/3 setting remained above +2,200 pips in both years. All 36 settings were positive in 2024. The default 14/14/3/3 remained **above +1,400 pips in both years.** ## Does win rate track the result? It did not. Across the 36 hourly settings in 2025, win rates ran from 38.17% to 41.84%, while annual net ranged from −3,221.7 to +1,560.4 pips. On four-hour bars the win-rate range widened to 36.72%–45.85%. Win rate alone still does not identify the annual result. ## Adding Stoch RSI to a chart | Platform | How to add it | |---|---| | MT4 / MT5 | Not included. Download an .ex4 or .mq4 into `MQL4/Indicators` (`MQL5/Indicators` on MT5), restart | | TradingView | In the indicator list as "Stochastic RSI" | | Browser (Formiq) | In the indicator list, with RSI period, stoch period, %K and %D adjustable | ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Buy | %K crosses above %D, filled at that bar's close | | Sell | %K crosses below %D | | Exit | The opposite crossover only — no stop, no target, no time exit | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Combinations | 3 RSI periods (7, 14, 21) × 3 stoch periods (7, 14, 21) × 2 %K (3, 5) × 2 %D (3, 5) = 36 | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | Using this indicator means raising the timeframe until the trade count is bearable. [The same constraint governs RVI](/blog/rvi-settings). The [RSI](/blog/rsi-settings) underneath it signals orders of magnitude less often without the two extra layers. Also tested this way: [OsMA](/blog/osma-settings), [Accelerator Oscillator](/blog/ac-oscillator-settings), [QQE](/blog/qqe-settings), [Aroon](/blog/aroon-settings), [Supertrend](/blog/supertrend-settings), [Williams %R](/blog/williams-r-settings) and [MFI](/blog/mfi-settings). ## Related - [Stochastic periods, smoothing and win rates](/blog/stochastic-settings): level fades and K/D crosses calculated from price ## Notes - One pair, two years - Stoch RSI implementations differ (how RSI is smoothed, how %K is taken). These figures come from Formiq's - Only the %K/%D crossover is tested; level readings such as 20/80 are not, and would reduce the trade count - Entries and exits both fill at bar closes - The spread is modelled as a flat 0.3 pips, which matters more here than for any low-frequency indicator - Four-hour settings trade 230 to 397 times a year --- ### Supertrend Periods and Multipliers: 36 Settings Tested URL: https://formiq.jp/blog/supertrend-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Supertrend, indicators, parameters, backtesting, USDJPY Also available in: ja — https://formiq.jp/ja/blog/supertrend-settings Supertrend arrives with two numbers attached: an ATR period of 10 and a multiplier of 3. Both are editable, but they do not affect the result equally. I moved both. Six periods by six multipliers is 36 combinations, backtested across the whole of 2025 on USD/JPY, on three timeframes, then run again on 2024. **The period did almost nothing.** Taking it from 5 to 30 (six times wider) moved the hourly trade count between 205 and 212 a year, and the average result by 406 pips. The multiplier did the work. From 1.5 to 5, the trade count went from 408 a year to 80. ## What Supertrend calculates 1. **Average True Range over N bars** ← the period 2. Bands placed above and below the midpoint at ATR × multiplier ← the multiplier 3. Price breaking a band sets the trend direction; the line trails on the other side ATR period and multiplier affect the band differently: - Changing the **period** changes how many bars go into the ATR. On USD/JPY the ATR at period 5 and at period 30 are close to each other, so **the band width hardly moves** - Changing the **multiplier** scales that ATR directly. **The band width moves with it** Band width sets how often price breaks through. So the multiplier alone sets the trade count. The measurements say the same thing, and the figure at the top of this article is that comparison. | Position | Trades a year, changing the period | Trades a year, changing the multiplier | |---|---|---| | 1st | period 5 → 212 | multiplier 1.5 → 408 | | 2nd | period 7 → 207 | multiplier 2 → 291 | | 3rd | period 10 → 205 | multiplier 2.5 → 201 | | 4th | period 14 → 205 | multiplier 3 → 159 | | 5th | period 20 → 206 | multiplier 4 → 104 | | 6th | period 30 → 207 | multiplier 5 → 80 | (Hourly bars, 2025. Period values are averaged over all six multipliers, and multiplier values over all six periods.) ## Adding Supertrend to a chart | Platform | How to add it | |---|---| | MT4 / MT5 | Not included. Download an .ex4 or .mq4 into `MQL4/Indicators` (`MQL5/Indicators` on MT5), restart | | TradingView | In the indicator list as "Supertrend" | | Browser (Formiq) | In the indicator list, with period and multiplier both adjustable | ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Buy | Supertrend flips upward, filled at that bar's close | | Sell | It flips downward | | Exit | The opposite flip only — no stop, no target, no time exit | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Combinations | 6 periods (5–30) × 6 multipliers (1.5–5) = 36, per timeframe and per year | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | ## Period or multiplier: which matters? The multiplier moved the result. Hourly bars, 2025, net pips: | Period ＼ Multiplier | 1.5 | 2 | 2.5 | 3 | 4 | 5 | |---|---|---|---|---|---|---| | 5 | −162 | +370 | **+1,927** | +1,009 | +1,132 | +562 | | 7 | −99 | +342 | +1,091 | +1,068 | +797 | +267 | | 10 | −56 | −188 | +726 | **+1,122** | +1,011 | +292 | | 14 | −122 | +497 | +911 | +674 | +1,567 | +478 | | 20 | +396 | −164 | +889 | −142 | +1,311 | +109 | | 30 | +1,024 | −349 | +1,817 | −297 | +1,292 | +52 | Reading down a column produces no pattern. Reading across, **four of the six entries in the 1.5 column are negative**, and 2.5 and 4 are the good ones. The same grid for 2024: | Period ＼ Multiplier | 1.5 | 2 | 2.5 | 3 | 4 | 5 | |---|---|---|---|---|---|---| | 5 | +347 | +1,722 | +1,091 | +840 | +1,224 | +1,408 | | 7 | +1,422 | +1,706 | +319 | +570 | +1,425 | +1,428 | | 10 | +1,575 | +1,401 | −269 | **+1,145** | +1,956 | +2,075 | | 14 | +1,413 | −497 | −257 | +1,531 | +1,236 | +2,232 | | 20 | +1,775 | −856 | −165 | +172 | +966 | +1,749 | | 30 | +2,187 | −421 | −258 | −242 | −4 | +2,664 | **The 2.5 column averages worst here (+77 pips) and the 5 column best (+1,926).** In 2025 that same 2.5 column leads at +1,227, with 5 third from bottom at +293. **One multiplier is the best of the six in one year and the worst in the other.** | Multiplier | Year | Mean annual net | Annual trades | |---:|---:|---:|---:| | 1.5 | 2025 | +164 pips | 408 | | 1.5 | 2024 | +1,453 pips | — | | 2 | 2025 | +85 pips | 290 | | 2 | 2024 | +509 pips | — | | 2.5 | 2025 | +1,227 pips | 201 | | 2.5 | 2024 | +77 pips | — | | 3 | 2025 | +572 pips | 159 | | 3 | 2024 | +669 pips | — | | 4 | 2025 | +1,185 pips | 104 | | 4 | 2024 | +1,134 pips | — | | 5 | 2025 | +293 pips | 80 | | 5 | 2024 | +1,926 pips | — | ## Is the 10 and 3 default good enough? | Timeframe | Year | Trades | Annual net | |---|---:|---:|---:| | 15-minute | 2025 | 659 | +1,064.3 pips | | 15-minute | 2024 | 613 | +1,849.0 pips | | 1-hour | 2025 | 152 | +1,121.8 pips | | 1-hour | 2024 | 155 | +1,145.0 pips | | 4-hour | 2025 | 42 | −99.1 pips | | 4-hour | 2024 | 34 | +2,960.5 pips | On 15-minute and hourly bars the default was **profitable in both years**. Of the indicators tested this way, Supertrend is the only one whose default held up that cleanly. The period-30, multiplier-5 setting fell from +2,663.7 pips in 2024 to +52.2 pips in 2025. The setting with the largest annual net in one year was not necessarily the largest in the next. ## Which timeframe suits Supertrend? | Timeframe | Year | Profitable settings | Profitable in both years | |---|---:|---:|---:| | 15-minute | 2025 | 26 / 36 | 21 / 36 | | 15-minute | 2024 | 26 / 36 | 21 / 36 | | 1-hour | 2025 | 27 / 36 | 21 / 36 | | 1-hour | 2024 | 27 / 36 | 21 / 36 | | 4-hour | 2025 | 10 / 36 | 10 / 36 | | 4-hour | 2024 | 36 / 36 | 10 / 36 | Only four-hour collapses, and there is a reason. The best four-hour combination of 2024 was period 30 with a multiplier of 5: | Year | Trades | Win rate | Profit factor | Annual net | |---:|---:|---:|---:|---:| | 2024 | 15 | 60.00% | 5.038 | +3,556.4 pips | | 2025 | 28 | 21.43% | 0.337 | −2,235.9 pips | Fifteen trades. A profit factor of 5.04 means nine wins out of fifteen and nothing more. The next year it lost 2,235.9 pips. Four-hour combinations trade 26 to 110 times a year, and raising the multiplier lowers that further (the multiplier-5 row runs 26 to 29) so **multiplier 5 on four-hour bars is structurally a setting whose numbers swing.** The better it looks, the less it means. ## How much spread can it absorb? Trading rarely is this indicator's practical advantage. | Indicator (hourly, 2025) | Trades a year | Pips lost, 0 → 2.0 spread | |---|---|---| | **Supertrend 10/3** | **152** | **304 pips** | | Aroon 25 crossover | 228 | 456 pips | | QQE 17/3/4.236 | 465 | 930 pips | | RVI 10 | 1,094 | 2,188 pips | Every one is exactly trades × spread. Supertrend trades least of the four, so it pays least. | Spread | Timeframe | Setting | Trades | Annual net | |---:|---|---|---:|---:| | 0.0 pips | 1-hour | 10/3 | 152 | +1,167.4 pips | | 0.3 pips | 1-hour | 10/3 | 152 | +1,121.8 pips | | 1.0 pips | 1-hour | 10/3 | 152 | +1,015.4 pips | | 2.0 pips | 1-hour | 10/3 | 152 | +863.4 pips | | 0.0 pips | 15-minute | 10/3 | 659 | +1,262.0 pips | | 0.3 pips | 15-minute | 10/3 | 659 | +1,064.3 pips | | 1.0 pips | 15-minute | 10/3 | 659 | +603.0 pips | | 2.0 pips | 15-minute | 10/3 | 659 | −56.0 pips | The hourly version keeps +863.4 pips even at a 2.0 pip spread. The 15-minute version crosses just below zero. ## Can win rate choose the setting? Not on its own. On hourly bars in 2025, larger multipliers reduced the trade count and included many of the larger win-rate and annual-net results. But the multiplier-5 mean was +293 pips in 2025 and +1,926 pips in 2024, so the size of that result still depended on the year. Filters, added to the default: | Condition | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | Supertrend alone | 2025 | 152 | 44.08% | +1,121.8 pips | | Supertrend alone | 2024 | 155 | 39.35% | +1,145.0 pips | | ADX ≥ 20 | 2025 | 81 | 45.68% | +673.0 pips | | ADX ≥ 20 | 2024 | 80 | 36.25% | +825.9 pips | | ADX ≥ 25 | 2025 | 54 | 44.44% | +362.8 pips | | ADX ≥ 25 | 2024 | 47 | 38.30% | −318.9 pips | | ADX ≥ 30 | 2025 | 32 | 50.00% | +243.2 pips | | ADX ≥ 30 | 2024 | 31 | 35.48% | −328.6 pips | | Tokyo hours only | 2025 | 48 | 45.83% | +582.1 pips | | Tokyo hours only | 2024 | 55 | 38.18% | −79.3 pips | | London/NY hours only | 2025 | 103 | 43.69% | +571.7 pips | | London/NY hours only | 2024 | 100 | 42.00% | +1,378.7 pips | ADX lowered the result in both years. Screening for trend strength on something that is already a trend-following signal removed profit, not risk. **Two conditions stayed positive in both years**: ADX ≥ 20 (+673.0 and +825.9) and London/NY hours (+571.7 and +1,378.7). Both land below the unfiltered +1,121.8 in 2025, so they hold the sign without adding anything. ## Stops and targets | Exit rule | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | Opposite flip only | 2025 | 152 | 44.08% | +1,121.8 pips | | Opposite flip only | 2024 | 155 | 39.35% | +1,145.0 pips | | Stop 30 / target 60 | 2025 | 139 | 37.41% | +486.1 pips | | Stop 30 / target 60 | 2024 | 146 | 34.25% | +107.3 pips | | Stop 50 / target 100 | 2025 | 136 | 39.71% | +1,103.2 pips | | Stop 50 / target 100 | 2024 | 141 | 36.17% | +238.8 pips | | Stop 100 / target 200 | 2025 | 144 | 45.83% | +2,952.4 pips | | Stop 100 / target 200 | 2024 | 141 | 39.72% | +1,291.8 pips | | Stop 50 / target 50 | 2025 | 138 | 55.07% | +817.2 pips | | Stop 50 / target 50 | 2024 | 145 | 51.72% | +524.0 pips | | Time exit after 24 bars | 2025 | 149 | 52.35% | +2,035.9 pips | | Time exit after 24 bars | 2024 | 152 | 46.05% | +942.9 pips | **All six made money in both years**: a first in this series. And **stop 100 / target 200 beat holding to the opposite flip in both years** (+2,952.4 and +1,291.8). A wide stop with a wide target did better than waiting for the indicator to turn. Stop 50 / target 50 puts the win rate above 50% in both years (55.07% and 51.72%) without adding profit. [The ATR period and comparison-window test](/blog/atr-settings) isolates the same volatility input and asks whether period 14 works as a trading-rule default. The same method applied to [Parabolic SAR](/blog/parabolic-sar-settings), [QQE](/blog/qqe-settings), [Aroon](/blog/aroon-settings), [RVI](/blog/rvi-settings) and [moving average crossovers](/blog/moving-average-cross-settings). [Building the rules without code](/blog/backtest-without-coding) is how to repeat any of it on your own pair. ## Notes - One pair, two years - Supertrend implementations differ in how the ATR is averaged (simple or Wilder) and in the band's trailing rule. These figures come from Formiq's - Entries and exits both fill at bar closes - The spread is modelled as a flat 0.3 pips for the whole period - Four-hour combinations trade 26 to 110 times a year. Win rates and profit factors at that count are wide numbers before anything else is said about them --- ### WaveTrend Settings and Timeframes: 30 Settings Tested URL: https://formiq.jp/blog/wavetrend-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: WaveTrend, indicators, parameters, backtesting, USDJPY Also available in: ja — https://formiq.jp/ja/blog/wavetrend-settings WaveTrend measures how far the bar midpoint has moved from its own average, divided by the typical size of that gap. Standardising the deviation gives two lines, and their crossings are the signal. I tested two parameters across 30 combinations on USD/JPY through the whole of 2025, on three timeframes, then ran the same 30 on 2024. **All 30 were profitable on hourly bars in 2025, and all 30 were profitable on four-hour bars in 2024.** On 15-minute bars, **not one was profitable in both years.** ## Which timeframe does WaveTrend work on? Not the 15-minute chart, where no setting finished profitable. | Timeframe | Year | Profitable settings | Median annual net | Median annual trades | |---|---:|---:|---:|---:| | 15-minute | 2025 | 2 / 30 | −2,142 pips | 3,598 | | 15-minute | 2024 | 1 / 30 | −953 pips | — | | 1-hour | 2025 | 30 / 30 | +2,466 pips | 838 | | 1-hour | 2024 | 24 / 30 | +1,185 pips | — | | 4-hour | 2025 | 29 / 30 | +1,318 pips | 224 | | 4-hour | 2024 | 30 / 30 | +1,566 pips | — | At 3,600 trades a year, **only one of 30 made money in 2024** on 15-minute bars, and none survived both years. A 0.3 pip spread costs about 1,080 pips at that count, and nothing here earned past it. Hourly and four-hour results were much stronger. On four-hour bars, 29 of 30 settings were profitable in both years, among the highest survival counts of the nine indicators tested this way. ## Is the 10/21 default good enough? The two years disagreed. | Setting | Year | Annual net | |---|---:|---:| | Default 10/21 | 2025 | +3,283.4 pips | | Default 10/21 | 2024 | +1,076.6 pips | | 2025 maximum, 20/15 | 2025 | +3,496.9 pips | | 2025 maximum, 20/15 | 2024 | +606.9 pips | | 2024 maximum, 30/30 | 2025 | +2,248.7 pips | | 2024 maximum, 30/30 | 2024 | +2,938.9 pips | | 2025 minimum, 5/10 | 2025 | +1,339.6 pips | | 2025 minimum, 5/10 | 2024 | −1,266.5 pips | The 2025 difference is 213.5 pips. In 2024, the default finished 469.7 pips ahead of 20/15. **One year did not justify changing away from the default.** The smallest hourly result in 2025 was still +1,339.6 pips. The same 5/10 setting lost −1,266.5 pips in 2024. ## Does four-hour tell the same story? There the default was profitable in both years. | Setting | Year | Annual net | |---|---:|---:| | Default 10/21 | 2025 | +1,262.2 pips | | Default 10/21 | 2024 | +1,373.5 pips | | 2025 maximum, 20/21 | 2025 | +2,476.8 pips | | 2025 maximum, 20/21 | 2024 | +506.1 pips | | 2024 maximum, 5/10 | 2025 | +1,232.4 pips | | 2024 maximum, 5/10 | 2024 | +3,321.3 pips | | 2025 minimum, 30/45 | 2025 | −98.2 pips | | 2025 minimum, 30/45 | 2024 | +1,009.8 pips | The default produced +1,262.2 pips in 2025 and +1,373.5 pips in 2024. The 30/45 setting changed sign, from −98.2 pips to +1,009.8 pips. ## Can win rate pick the setting? It cannot. Across the 30 hourly settings in 2025, win rates ran from 33.48% to 39.45%, while annual net ranged from +1,339.6 to +3,496.9 pips. On 15-minute bars, 28 of 30 settings lost money in 2025. The 3,598-trade median and its cost matter more than a higher win rate. ## Adding WaveTrend to a chart | Platform | How to add it | |---|---| | MT4 / MT5 | Not included. Download an .ex4 or .mq4 into `MQL4/Indicators` (`MQL5/Indicators` on MT5), restart | | TradingView | In the indicator list as "WaveTrend" | | Browser (Formiq) | In the indicator list, with channel and average periods adjustable | ## How this was measured | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Buy | wt1 crosses above wt2, filled at that bar's close | | Sell | wt1 crosses below wt2 | | Exit | The opposite crossover only — no stop, no target, no time exit | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Combinations | 6 channel periods (5–30) × 5 average periods (10–45) = 30 | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | The same method applied to [OsMA](/blog/osma-settings), [Accelerator Oscillator](/blog/ac-oscillator-settings), [Stoch RSI](/blog/stoch-rsi-settings), [STC](/blog/stc-settings), [QQE](/blog/qqe-settings), [Aroon](/blog/aroon-settings), [RVI](/blog/rvi-settings), [Supertrend](/blog/supertrend-settings), [Williams %R](/blog/williams-r-settings) and [MFI](/blog/mfi-settings). ## Notes - One pair, two years - WaveTrend implementations differ (the constants used to standardise the deviation, the smoothing type). These figures come from Formiq's - Only the two-line crossover is tested; overbought and oversold bands such as ±60 are not - Entries and exits both fill at bar closes - The spread is modelled as a flat 0.3 pips; the 15-minute result depends heavily on that - Four-hour settings trade 147 to 325 times a year --- ### Williams %R Mean Reversion: 240 Settings Tested URL: https://formiq.jp/blog/williams-r-settings Language: en Published: 2026-08-24 Updated: 2026-08-30 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: Williams %R, indicators, win rate, backtesting, USDJPY Also available in: ja — https://formiq.jp/ja/blog/williams-r-settings Williams %R places the current close inside the high-low range of the last N bars, on a scale from 0 to −100. Below −80 is called oversold, above −20 overbought. I tested exactly that reading. On USD/JPY across the whole of 2025: six periods × five entry levels × four exit levels × two directions, 240 settings in all, on three timeframes, then the same 240 again on 2024. **The win rates came out high. The results came out negative.** Across the 120 long-side settings on hourly bars in 2025 the median win rate was 59.0%, and **only 10 of the 120 finished positive.** The textbook setup, period 14 buying below −80 and closing above −20, **won 68.42% of its trades in 2024 and lost 569.3 pips.** Each dot is one setting: win rate across, net pips up. The win rates spread from 49% to 67% and almost the entire cloud sits under the zero line. ## Why a 68% win rate still loses The arithmetic is visible. | Timeframe | Year | Setting | Win rate | Average win | Average loss | Annual net | |---|---:|---|---:|---:|---:|---:| | 1-hour | 2025 | Period 14, −80/−20 | 58.11% | +39.8 pips | −72.7 pips | −1,082.8 pips | | 1-hour | 2024 | Period 14, −80/−20 | 68.42% | +44.9 pips | −110.8 pips | −569.3 pips | | 1-hour | 2025 | Setting with 67.05% win rate | 67.05% | +28.3 pips | −74.1 pips | −953.7 pips | | 4-hour | 2024 | Period 14, −80/−20 | 79.41% | +84.9 pips | −267.0 pips | +423.5 pips | Look at the last row. **A 79.41% win rate produced 423.5 pips**, because the losing fifth of trades averaged 3.1 times the size of the winners. The exit rule creates this payoff distribution. A long trade enters below −80 and closes only after %R reaches −20, so a bounce closes the gain while a continued decline leaves the loss open. The result is frequent small wins and occasional large losses. The high win rate follows from those frequent small bounces; it does not establish positive annual net. **None of the 120 hourly settings in 2025 reached a 70% win rate.** The maximum was 67.05%, and that setting still lost 953.7 pips. ## Adding Williams %R to a chart | Platform | How to add it | |---|---| | MT4 / MT5 | Built in — Navigator, Oscillators, Williams' Percent Range | | TradingView | In the indicator list as "Williams %R" | | Browser (Formiq) | In the indicator list, with an adjustable period | ## How this was measured The rule here is **directional**: "below −80" can only be a buy condition and "above −20" can only be a sell. The long fade and short fade were therefore tested as separate trading rules. | Item | Value | |---|---| | Pair | USD/JPY | | Period | 2025-01-01 to 2025-12-31 (2024 run identically for comparison) | | Timeframes | 15-minute / 1-hour / 4-hour | | Bars tested | 6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour | | Long side | Buy when %R falls below the entry level (−95 to −70), close when it rises above the exit level (−50 to −10) | | Short side | The mirror of that on the 0 side (sell at −5 to −30, close at −50 to −90) | | Stops and targets | None, except in the section that measures them | | Costs | 0.3 pip spread, zero slippage, 0.1 lot | | Combinations | 6 periods (7–40) × 5 entry levels × 4 exit levels × 2 directions = 240, per timeframe and per year | | Method | Executed in Formiq's backtester; pips recomputed from each trade's fill prices | ## Do the long and short sides differ? | Timeframe | Side | Year | Profitable settings | Median annual net | Median win rate | |---|---|---:|---:|---:|---:| | 15-minute | Long | 2025 | 3 / 120 | −661 pips | 64.3% | | 15-minute | Long | 2024 | 54 / 120 | −226 pips | — | | 15-minute | Short | 2025 | 7 / 120 | −777 pips | 63.3% | | 15-minute | Short | 2024 | 2 / 120 | −1,169 pips | — | | 1-hour | Long | 2025 | 10 / 120 | −720 pips | 59.0% | | 1-hour | Long | 2024 | 47 / 120 | −201 pips | — | | 1-hour | Short | 2025 | 10 / 120 | −633 pips | 59.3% | | 1-hour | Short | 2024 | 0 / 120 | −1,442 pips | — | | 4-hour | Long | 2025 | 89 / 120 | +572 pips | 64.1% | | 4-hour | Long | 2024 | 71 / 120 | +123 pips | — | | 4-hour | Short | 2025 | 78 / 120 | +333 pips | 57.6% | | 4-hour | Short | 2024 | 0 / 120 | −1,396 pips | — | Two things stand out. **The short side was wiped out in 2024**: zero of 120 on both hourly and four-hour bars. USD/JPY rose 1,632 pips that year. **Selling "overbought" through a rising market produces exactly this.** A counter-trend indicator cannot be judged apart from the direction the market took. **Only the four-hour long side worked**: 89 of 120 in 2025, 71 in 2024, and 40 profitable in both. It is the one usable cell in the whole test. On the four-hour long side, 40 of 120 settings were profitable in both years. Many were profitable in only one year, so one year's result did not identify the next year's period and levels. ## Which timeframe actually worked? | Period | Year | Mean annual net | Mean win rate | |---:|---:|---:|---:| | 7 | 2025 | −83 pips | 57.1% | | 7 | 2024 | +549 pips | 73.3% | | 9 | 2025 | −52 pips | 56.7% | | 9 | 2024 | +520 pips | 77.3% | | 14 | 2025 | +203 pips | 58.9% | | 14 | 2024 | +197 pips | 74.1% | | 21 | 2025 | +711 pips | 66.8% | | 21 | 2024 | −138 pips | 66.4% | | 28 | 2025 | +1,143 pips | 70.7% | | 28 | 2024 | −215 pips | 65.1% | | 40 | 2025 | +935 pips | 76.5% | | 40 | 2024 | −148 pips | 63.4% | Long periods produced the larger results in 2025, while short periods produced the larger results in 2024. **Only period 14 was positive in both years** (+203 and +197). Deeper entry levels helped in 2025 (−70 gave +135 pips, −95 gave +757) and made almost no difference in 2024 (+143 and +71). Exit levels showed no consistent pattern either year. ## Does a stop fix it? If the losses are the problem, cut them. I tried. | Exit rule | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | No stop | 2025 | 148 | 58.11% | −1,082.8 pips | | No stop | 2024 | 133 | 68.42% | −569.3 pips | | Stop 30 / target 60 | 2025 | 376 | 33.51% | −1,501.2 pips | | Stop 30 / target 60 | 2024 | 297 | 37.04% | −667.1 pips | | Stop 50 / target 100 | 2025 | 281 | 43.77% | −1,788.9 pips | | Stop 50 / target 100 | 2024 | 242 | 45.45% | −900.4 pips | | Stop 20 / target 40 | 2025 | 476 | 29.62% | −1,331.7 pips | | Stop 20 / target 40 | 2024 | 370 | 32.43% | −709.5 pips | | Stop 50 / target 50 | 2025 | 304 | 49.67% | −1,234.4 pips | | Stop 50 / target 50 | 2024 | 260 | 50.77% | −889.9 pips | | Time exit after 24 bars | 2025 | 184 | 58.15% | −989.6 pips | | Time exit after 24 bars | 2024 | 152 | 65.13% | −729.6 pips | **Every one lost.** The four stop-and-target pairs also lost more than doing nothing, in both years. The 24-bar time exit is the single exception, and only in one direction: −989.6 pips in 2025 against −1,082.8 without it, then −729.6 against −569.3 in 2024. The average loss does shrink, from −110.8 pips to −30.0. But the win rate collapses with it, 68.42% down to 37.04%, and the trade-off goes the wrong way. A stop converted one large loss into a run of small ones. The size of the losses comes from waiting for an exit condition that may not arrive. Changing the stop distance does not change that design. ## Is the spread to blame? | Spread | H1 long, 14, −80/−20 (148 trades) | |---|---| | 0.0 pips | −1,038.4 | | 0.3 pips | −1,082.8 | | 1.0 pips | −1,186.4 | | 2.0 pips | −1,334.4 | At a zero spread it still loses 1,038.4 pips. **This is not a cost problem.** [For higher-frequency indicators the spread does decide the outcome](/blog/rvi-settings), but 148 trades a year changes the total by only 296 pips between 0 and 2.0 pips of spread. The 2024 default averaged +44.9 pips per win and −110.8 per loss, so removing trading cost does not repair the payoff imbalance. ## Can the four-hour long side be narrowed? Narrowing the one cell that worked: | Condition | Year | Trades | Win rate | Annual net | |---|---:|---:|---:|---:| | Period 14, −80/−20 | 2025 | 36 | 61.11% | +168.1 pips | | Period 14, −80/−20 | 2024 | 34 | 79.41% | +423.5 pips | | Period 28, −80/−20 | 2025 | 24 | 70.83% | +912.1 pips | | Period 28, −80/−20 | 2024 | 15 | 60.00% | −785.0 pips | | London/NY hours only | 2025 | 36 | 66.67% | +609.5 pips | | London/NY hours only | 2024 | 28 | 75.00% | +16.8 pips | | Tokyo hours only | 2025 | 30 | 63.33% | +363.0 pips | | Tokyo hours only | 2024 | 23 | 69.57% | −1.0 pips | The session filters improve 2025 and take 2024 to roughly zero. **The only variant positive in both years is the unfiltered period 14**, based on 15 to 36 trades a year, which is thin evidence in itself. These numbers support using Williams %R as one input after direction has been decided elsewhere, rather than letting it choose direction by itself. The same method applied to [QQE](/blog/qqe-settings), [Aroon](/blog/aroon-settings), [RVI](/blog/rvi-settings) and [Supertrend](/blog/supertrend-settings), where lower win rates accompanied better annual net results. [Bollinger Bands](/blog/bollinger-band-settings) compare a band-touch fade with a band breakout on the same indicator. ## Notes - One pair, two years - Williams %R varies little between implementations, though conventions differ on whether level comparisons are inclusive. These figures come from Formiq's - Entries and exits both fill at bar closes - Spread is fixed at 0.3 pips throughout and slippage is set to zero - Four-hour settings trade 9 to 92 times a year: thin for reading win rates --- ### Forex Practice on iPhone: Replay and Trade Notes URL: https://formiq.jp/blog/forex-practice-iphone-free Language: en Published: 2026-03-21 Updated: 2026-09-06 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: iPhone, free, mobile, practice Also available in: ja — https://formiq.jp/ja/blog/forex-practice-iphone-free You can practice on historical charts with Formiq's iOS app. Choose a currency pair and timeframe, then decide on entry and exit conditions before starting replay. ## Open the app and choose a historical period Install Formiq from the [App Store](https://apps.apple.com/jp/app/formiq-fx-backtest-tester/id6759964075) and sign in. Select a pair, timeframe and historical period, then start replay. Begin by advancing candles manually and checking the order and exit controls. Use familiar periods to learn the controls and unseen periods to assess trading decisions. ## Keep prices and order levels readable Overlay indicators and subpanes reduce the space available for the price chart. If candles or order levels are hard to read, reduce the display or adjust the zoom. Removing a condition to fit the screen changes the strategy being practiced. If a decision requires several indicators, use a larger screen such as a computer or iPad. ## Plan the trade and record the decision Before entering, identify the condition that triggered the trade and the stop and target locations. After closing, note whether you followed the plan. Profit and loss alone cannot establish that you followed the rule. Leave time to review even a short session. The [practice review guide](/blog/1000-replay-trades) explains how to record entries and skipped setups. ## Review history on a computer Using the same account lets you share chart settings and trade records with the browser version. After switching devices, confirm that the latest records have appeared before reviewing them. A larger screen makes it easier to examine several indicators or longer periods alongside trade history. Check the app and browser interfaces for the particular features you need. ## Check the free plan The [pricing page](/pricing) and the app's plan screen show feature availability and usage limits. Try the displays and controls required for your practice before purchasing. Replay uses historical data and simulated orders. It does not place live orders or trade real funds. --- ### Free Forex Testing Software: Features and Limits URL: https://formiq.jp/blog/forex-tester-free Language: en Published: 2026-03-17 Updated: 2026-09-06 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: forex tester, free, replay practice, backtesting Also available in: ja — https://formiq.jp/ja/blog/forex-tester-free Choose a free forex testing tool according to the task: practicing discretionary entries in replay or calculating trades from predefined rules. Each task needs different controls, and a free plan may restrict them differently. ## Choose replay or automated backtesting Replay hides future candles while you decide when to enter and exit. An automated backtest calculates trades according to conditions specified in advance. For discretionary practice, check candle controls and trade records. For rule comparisons, check whether you can express the entry conditions, trading costs and exit methods you need. ## Read the free plan's limits | Item | What to check | |---|---| | Expiry | A time-limited trial or an ongoing free plan | | Historical data | Required pairs, timeframes and date ranges | | Usage | Replay and backtest run limits | | Conditions and display | Supported entry rules and indicators | | Records | Saved trades and space for decision notes | A free label alone does not establish that a tool supports your practice routine. The [forex testing tool comparison](/compare/forex-backtesting-tools) links to the vendors and their terms. ## Replay a historical chart in Formiq Formiq runs in a browser. Open the [chart](/chart), select a pair and timeframe, and start replay. You can use automatic playback or advance candles manually. Decide on entry and exit conditions before placing an order. After the session, add the reason for each decision to your trade records. The [replay practice guide](/blog/1000-replay-trades) explains the review process. ## Enter conditions for a backtest Formiq's backtester accepts indicator conditions and stop-loss and take-profit settings. Check trade count and entry and exit locations as well as net profit. Confirm that the available conditions can express your rule before testing. The [no-code backtesting guide](/blog/backtest-without-coding) includes an input example. ## Try the required controls before upgrading Formiq's free plan limits simultaneous indicators and backtest runs, among other features. Check the [pricing page](/pricing) for current allowances and try the controls needed for your practice. The chart uses historical prices. Replay orders are simulated and do not move real funds. --- ### Forex Replay Practice: Rules, Trade Notes and Review URL: https://formiq.jp/blog/1000-replay-trades Language: en Published: 2026-03-14 Updated: 2026-09-06 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: replay, forex simulator, practice, trading psychology Also available in: ja — https://formiq.jp/ja/blog/1000-replay-trades Forex replay hides future candles so you can make a trading decision before seeing the outcome. Write down entry and exit rules before starting, then record whether you followed them. Keeping execution mistakes separate from profit and loss makes the review more useful. ## Choose a rule and a historical period Select a currency pair, timeframe and date range. Define your entry, stop-loss and take-profit conditions before starting. If the rule depends on a reversal, specify the candle pattern or price level that counts as a reversal. If you change a rule during practice, group trades before and after the change separately. Combining trades taken under different conditions makes it harder to assess the change. ## Decide before advancing the chart In manual mode, decide whether to enter or skip using only the candles currently visible. Record the reason and exit conditions before advancing. For skipped entries, note what was missing: price had not reached support, for example, or an indicator condition had not triggered. A later price rise does not make a skipped entry a mistake if the setup did not meet your rule. ## Record execution separately from profit and loss Keep the date, pair, timeframe, direction, entry and exit prices, profit or loss, and reason for each trade. Note any departure from the plan, such as moving a stop after entering. You can add memos to Formiq's trade records. Review profitable trades that broke your rules as well as losing trades. A profit alone does not show whether the decision is repeatable. ## Read win rate alongside average profit and loss Win rate alone cannot explain total profit. Check average winning and losing trade sizes and the trade count. Also check whether a few trades account for most of the profit. Grouping trades by pair or session can reveal where losses occurred. Small groups need more observations before you can draw conclusions about a session. The [hourly range data](/data) compares price movement; it does not establish that your entry rule is profitable. ## Check a period you have not seen Repeated practice on the same chart can make the next price move familiar. That may help with learning the controls, but assessing trading decisions requires a period you have not already reviewed. After changing a rule, try it on a separate period. Compare profit and loss and adherence to the rule separately. A higher practice count does not establish that a strategy is profitable. ## Start a replay in Formiq Open the [chart](/chart), choose a pair and timeframe, and start replay. After the session, review your trade records and add notes about the decisions. Check the [pricing page](/pricing) for the free plan's limits. Replay trades are simulated. Practice does not establish that you will make the same decisions with real money at risk or receive the same execution prices in a live account. --- ### How to Backtest a Forex Strategy Without Coding URL: https://formiq.jp/blog/backtest-without-coding Language: en Published: 2026-03-10 Updated: 2026-09-06 Author: Yuichi Saiki — https://formiq.jp/about/yuichi-saiki Tags: backtesting, no-code, strategy, beginner Also available in: ja — https://formiq.jp/ja/blog/backtest-without-coding Formiq's backtester lets you build trading rules by selecting indicator conditions. Choose a currency pair, timeframe, date range and trading costs, then inspect both aggregate results and individual trades. ## Set entry and exit conditions The following is an input example, not a strategy with verified profitability. | Item | Example setting | |---|---| | Long entry | EMA(9) crosses above EMA(21), and RSI(14) is greater than 40 | | Stop loss | 30 pips | | Take profit | 60 pips | Check the pair, timeframe, date range and spread before running. When comparing a change, keep the other settings fixed. ## Read the result metrics together Check net profit, win rate, average winning and losing trade sizes, profit factor, maximum drawdown and trade count. A high win rate can still produce a net loss if losing trades are much larger than winning trades. Profit factor divides gross profit by the absolute value of gross loss. A high value does not establish stability when there are few trades or when profit comes mainly from one large price move. ## Inspect entries and exits on the chart Open trades from the results on the chart and check whether entry and exit locations match the intended rule. Aggregate statistics can conceal input mistakes or trades concentrated in a short historical period. Also inspect candles that reach both a stop and a target. The order of price movements within the candle can affect the result, depending on the backtest's execution assumptions. ## Separate selection from evaluation Repeatedly changing parameters on one date range and choosing the most profitable result increases the chance of selecting conditions that fit that particular history. After selecting settings, run them unchanged on a separate period. The [QQE study](/blog/qqe-settings) compares a setting selected in one year with its results in another year. Save the second result even when performance deteriorates. ## Check whether the builder can express your rule The builder supports its available conditions and exit methods. If your rule requires external data or custom order management, check whether those requirements are supported. Substituting a similar condition tests the substitute, not the original rule. ## Check usage and data availability The [pricing page](/pricing) lists the free plan's run limits. Check the chart for the historical date range available for your test. Backtests calculate results from past prices and execution assumptions. They do not guarantee future profit or live execution prices. [Open the chart to set conditions](/chart). --- ## 日本語記事 ### ダマシ回避フィルターは効くのか？1,596通り検証した結果 URL: https://formiq.jp/ja/blog/false-signal-filters Language: ja Published: 2026-09-10 Updated: 2026-09-10 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ダマシ回避, ADX, 移動平均線の傾き, トレンドフィルター, フィルター, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/false-signal-filters Evidence JSON: https://formiq.jp/ja/evidence/false-signal-filters/data.json Also available in: en — https://formiq.jp/blog/false-signal-filters 「シグナルが出ても、レンジ相場ならダマシになる。だからADXでトレンドの強さを測って、25以上のときだけ売買する」。ダマシを減らす方法の説明は、たいていこの形をしています。 このシリーズでは、ADXのフィルターをGMMA・CCI・アリゲーターの3つの記事で試して、4回とも成績が落ちました。ただ、3つの指標で悪くなったことは「どの指標に掛けても悪くなる」の証明にはなりません。 そこで、**公開済みの記事に載せた設定38通りへ、同じ6つのフィルターをまとめて掛けました。** 3つの時間足と2年分、フィルターなしを含めて1,596通りです。 **ADXの3つの水準（20・25・30以上）は、3つの時間足すべてで「2024年と2025年の両方で黒字だった設定の数」を減らしました。9区分すべてです。** 15分足は13通りから6・4・7通りへ、1時間足は27通りから18・10・12通りへ、4時間足は16通りから14・11・12通りへ落ちています。 そして、**フィルターが落とした売買が本当に負けていた売買なのかを、同じ回数をランダムに落とした場合と比べました。1,348通りの比較のうち範囲の外へ出たのは72通り。何も仕分けていなくても、偶然で約67通りは外へ出ます。** ## ADXと移動平均線の傾きの設定と読み方 この記事で使うフィルターは2つの指標から作ります。どちらもFormiqのバックテスト条件にそのまま入っています。 **ADX**は価格チャートの下の別枠に3本の線で表示されます。ADX本体（トレンドの強さ）、+DI（上昇の勢い）、−DI（下降の勢い）です。ADXは0から100の値を取り、大きいほどトレンドが強いと読みます。MT4では水準線が初期状態では引かれないので、20や25の線は自分で追加します。 **移動平均線の傾き**は価格チャートに重なる普通の移動平均線を1本引き、その線が数本前と比べて上がっているか下がっているかだけを見ます。線そのものは表示されますが、傾きの判定は画面に出ません。 | 条件 | 数値の項目 | 初期設定 | 成立の中身 | |---|---|---|---| | ADX | 期間 | 14 | 何本分の値動きから計算するか | | ADX | 最低水準 | 25 | ADXがこの値以上でなければ売買しない | | ADX | 読み方 | 水準のみ | 水準のみ／DIの比較／DIのクロスの3つから選ぶ | | 移動平均線の傾き | 期間 | 20 | 何本の平均を引くか | | 移動平均線の傾き | 計算方法 | SMA | 単純・指数・加重などから選ぶ | | 移動平均線の傾き | 判定の間隔 | 5 | 何本前と比べて上下を決めるか | | 移動平均線の傾き | 向き | どちらでも | 上向きのみ買い／下向きのみ売り／どちらでも | ### フィルターとして使うときの読み方 ADXの読み方を「水準のみ」にすると、**ADXが最低水準以上の足では買いと売りの両方を認め、下回っている足ではどちらも認めません**。これが「トレンドの強さで絞る」という意味です。 「DIの比較」にすると、+DIが−DIより上の足では買いだけ、下の足では売りだけを認めます。移動平均線の傾きを「どちらでも」にした場合も同じ作りで、上向きなら買いだけ、下向きなら売りだけです。**この2つはトレンドの強さではなく向きで絞ります。** 条件一覧で元の条件と一緒に有効にし、判定を「すべて満たす」にすると、両方が同じ向きを指した足だけが残ります。 ## ダマシ回避フィルターは何本の足を捨てているのか フィルターを評価する前に、そもそも何本を落とせるのかを数えます。ほとんどの足で成立する条件は、何も絞れません。 2025年の全足に対して、それぞれのフィルターが売買を認めた割合です。 | フィルター | 15分足 買いを認めた | 1時間足 買いを認めた | 4時間足 買いを認めた | |---|---|---|---| | ADX 20以上 | 61.22% | 66.33% | 67.20% | | ADX 25以上 | 40.51% | 47.24% | 37.58% | | ADX 30以上 | 25.15% | 31.40% | 17.58% | | +DIが−DIより上 | 49.85% | 45.87% | 41.24% | | 50本移動平均が上向き | 50.09% | 48.35% | 55.16% | | 200本移動平均が上向き | 48.50% | 54.61% | 60.81% | ADXは水準を上げるほど通す足が減ります。25以上にすると37.58%から47.24%の足しか通さず、30以上なら17.58%から31.40%まで落ちます。 **向きで絞る3つは、足を1本も捨てていません。** +DIと−DIはどちらかが必ず上ですし、移動平均線もどちらかへ傾いています。ここに並べた割合は買いを認めた足のぶんで、残りの足では売りを認めています。**捨てているのは「買いたいのに下向き」という組み合わせだけです。** ## ADXが高い足のあとに何が起きているか フィルターの前提は「トレンドが強い足のあとは、その向きに動きが続く」です。2025年の全足をADXの値で4つに分け、10本後の終値までに何pips動いたかを測りました。 | 時間足 | ADX | 本数 | 10本後までの値幅 | ATR比 | 直前の向きに続いた割合 | |---|---|---|---|---|---| | 15分足 | 20未満 | 9,654本 | 19.81pips | 1.86 | 50.58% | | 15分足 | 30超 | 6,264本 | 20.92pips | 1.65 | 48.77% | | 1時間足 | 20未満 | 2,096本 | 42.45pips | 1.91 | 50.52% | | 1時間足 | 30超 | 1,947本 | 44.52pips | 1.75 | 47.82% | | 4時間足 | 20未満 | 519本 | 82.66pips | 1.81 | 51.64% | | 4時間足 | 30超 | 283本 | 95.61pips | 1.84 | 43.82% | 「直前の向きに続いた割合」は、その足までの5本の値動きと、その足から10本後までの値動きが同じ符号だった割合です。 **値幅は確かに少し広がります。**15分足で19.81pipsから20.92pips、4時間足で82.66pipsから95.61pipsです。ところが同じ値幅をその足のATRで割ると、15分足は1.86から1.65へ、1時間足は1.91から1.75へ縮みます（4時間足だけは1.81から1.84へ増えます）。**ADXが高い足はその足自体が大きいので、pipsで見た値幅の広がりは、足の大きさの広がりでほとんど説明がつきます。** **向きのほうはもっとはっきりしています。**2つの年と3つの時間足を合わせた24区分で、続いた割合は43.82%から53.45%の間に収まりました。ADXが30を超えた足のほうが20未満の足より続きやすかったのは、6区分のうち1つ（1時間足の2024年、53.45%対51.16%）だけです。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 2024年1月1日〜2025年12月31日 | | 元の設定 | 公開済みの記事に載せた38通り | | フィルター | ADX20・25・30以上、+DIが−DIより上、50本と200本の移動平均が上向き | | 足す場所 | エントリー条件のみ。決済は記事が公開した通り | | 比べる相手 | フィルターなしの売買から同じ回数をランダムに選んだ1,000回 | | 回した数 | 38通り×7条件×3時間足×2年＝1,596通り | | スプレッド | 0.3pips固定 | | ロット | 0.1 | | 約定 | 終値 | **元の設定はこちらで選んでいません。**公開済みの記事それぞれに「この設定で走らせる」ボタンが付いていて、その設定をそのまま使いました。検証した時点で公開済みのボタンは43件で、そのうち3件を対象から外しました。DMI/ADXの記事はすでにADXを使っていて、1つの条件セットにADXの設定は1組しか入りません。明けの明星・宵の明星の記事は2つの条件をORで結んでいるので、3つ目を足すと絞り込みではなく追加になります。この記事のボタンはフィルターを掛けたあとの設定なので、掛ける前の一覧には入りません。残る40件のうち、ボリンジャーバンドの3記事が同じ設定（期間20・2σのブレイク）なので1つにまとめ、38通りとしました。 **決済は動かしていません。**「エントリーのダマシを減らす」という話なので、エントリー条件にだけフィルターを足します。決済も同時に変えると、エントリーではなく決済を測ることになります。 ## 38通りの設定に同じフィルターを掛ける 年間損益がプラスで終わった設定の数です。38通り中の数で書きます。 | フィルター | 15分足 2025 | 1時間足 2025 | 4時間足 2025 | 15分足 2024 | 1時間足 2024 | 4時間足 2024 | |---|---|---|---|---|---|---| | なし | 20 | 29 | 20 | 17 | 34 | 31 | | ADX 20以上 | 11 | 26 | 22 | 16 | 27 | 27 | | ADX 25以上 | 11 | 22 | 23 | 14 | 19 | 23 | | ADX 30以上 | 16 | 20 | 21 | 14 | 25 | 24 | | +DIが−DIより上 | 21 | 24 | 13 | 25 | 32 | 33 | | 50本移動平均が上向き | 20 | 28 | 18 | 34 | 30 | 35 | | 200本移動平均が上向き | 22 | 14 | 28 | 21 | 35 | 31 | 1年だけを見ると、良くなった区分も悪くなった区分もあります。4時間足の2025年はADX25以上で20通りから23通りへ増え、15分足の2024年は50本の傾きで17通りから34通りへ増えました。 そこで2つの年をそろえます。**2024年と2025年の両方で黒字だった設定の数です。** | フィルター | 15分足 | 1時間足 | 4時間足 | |---|---|---|---| | なし | 13 | 27 | 16 | | ADX 20以上 | 6 | 18 | 14 | | ADX 25以上 | 4 | 10 | 11 | | ADX 30以上 | 7 | 12 | 12 | | +DIが−DIより上 | 15 | 22 | 12 | | 50本移動平均が上向き | 18 | 23 | 17 | | 200本移動平均が上向き | 16 | 13 | 21 | **ADXの行は9つのマス全部でフィルターなしを下回りました。**いちばん減ったのは1時間足のADX25以上で、27通りから10通りです。 向きで絞る3つはばらけています。50本の傾きは15分足と4時間足で増やし（13→18、16→17）、1時間足で減らしました（27→23）。200本の傾きも15分足と4時間足で増やし、1時間足では27通りから13通りへ減らしました。 ## 勝率は上がるのに年間損益が減る MACDの記事に載せた12/26/9のシグナル線とのクロスに、ADX25以上のフィルターを足しました。1時間足です。 | 年 | 条件 | 取引回数 | 勝率 | 1回あたり | 年間損益 | |---|---|---|---|---|---| | 2025 | フィルターなし | 451回 | 38.58% | +4.93pips | +2,223.6pips | | 2025 | ADX25以上 | 204回 | 40.69% | +2.21pips | +451.0pips | | 2024 | フィルターなし | 423回 | 39.95% | +7.45pips | +3,152.1pips | | 2024 | ADX25以上 | 168回 | 46.43% | +8.03pips | +1,349.2pips | **2年とも勝率は上がりました。**2025年は2.11ポイント、2024年は6.48ポイントです。**2年とも年間損益は減りました。**2025年は1,772.6pips、2024年は1,802.9pipsの減少です。 2024年は1回あたりの損益も+7.45pipsから+8.03pipsへ上がっています。それでも年間で1,800pips近く減るのは、取引回数が423回から168回へ落ちるからです。1回あたりが0.58pips良くなっても、255回ぶんの利益は戻りません。 ## 落とした売買は、本当に負けていた売買か **フィルターが落とした売買が負けていた売買なら、残った売買の1回あたりの損益は全体より良くなるはずです。** 同じMACDで確かめます。1時間足の2025年、フィルターなしの451回のうちADX25以上を満たしていたのは204回で、残る247回が落ちた売買です。 | 売買 | 回数 | 勝率 | 1回あたり | |---|---|---|---| | フィルターなしの全部 | 451回 | 38.58% | +4.93pips | | ADX25以上で残った | 204回 | 40.69% | +2.21pips | | ADX25未満で落ちた | 247回 | 36.84% | +7.18pips | **落ちた247回は1回あたり+7.18pips、残った204回は+2.21pipsでした。**勝率は残った側のほうが高いのに、稼いでいたのは落とした側です。 ただし、売買を減らせば1回あたりの平均は動きます。標本が小さくなるほど、その平均から離れやすくなるからです。**だから比べる相手は「同じ回数をランダムに落とした場合」です。**この451回から204回を無作為に取り出す操作を1,000回繰り返すと、1回あたりの損益は−1.38pipsから+11.55pipsの間に収まりました。ADX25以上の+2.21pipsはその中です。勝率のほうも34.31%から43.63%の範囲で、40.69%は中に入ります。 38通り×6フィルター×3時間足×2年で1,368通り。うち20通りは全部通すか全部落とすかで、比べる相手がありません。残り1,348通りの結果です。 | フィルター | 比べた数 | 1回あたりが全体を上回った | 範囲の外へ出た | うち上側 | |---|---|---|---|---| | ADX 20以上 | 225 | 95 | 11 | 1 | | ADX 25以上 | 227 | 92 | 15 | 8 | | ADX 30以上 | 226 | 103 | 15 | 13 | | +DIが−DIより上 | 215 | 120 | 8 | 4 | | 50本移動平均が上向き | 227 | 147 | 21 | 19 | | 200本移動平均が上向き | 228 | 149 | 2 | 2 | | 合計 | 1,348 | 706 | 72 | 47 | **1回あたりの損益が全体を上回ったのは706通り、52.4%です。**コインを投げても同じくらいになります。 **範囲の外へ出たのは72通り。範囲は中央の95%なので、何も仕分けていなくても偶然で約67通りが外へ出ます。**勝率で同じことをすると、外へ出たのは67通り（上側44、下側23）でした。 なお、この1,348通りの比較は互いに独立ではありません。同じ売買を6つのフィルターで切り分けていますし、時間足と年もまたいでいます。ですから「72対67」は検定の結果ではなく、**フィルターの仕分けが偶然と見分けのつく大きさになっていない**という読み方です。 ## 助かったのは、フィルターなしで負けていた設定だった 仕分けていないのなら、フィルターが年の結果を動かしたぶんは何なのか。38通りをフィルターなしの年間損益で並べ、下位19通りと上位19通りに割って数えました。**それぞれの設定について、6つのフィルターのうち結果を見てから一番良かったものを選んでいます。**後知恵で選んでも改善しないなら、そのフィルターは本当に使えません。 | 時間足・年 | 下位19通りで改善した数 | 中央値の変化 | 上位19通りで改善した数 | 中央値の変化 | |---|---|---|---|---| | 15分足 2025 | 19 | +891.6pips | 10 | +35.9pips | | 15分足 2024 | 19 | +2,228.2pips | 18 | +976.4pips | | 1時間足 2025 | 17 | +413.1pips | 1 | −462.1pips | | 1時間足 2024 | 17 | +519.2pips | 8 | −177.4pips | | 4時間足 2025 | 19 | +1,012.5pips | 17 | +248.7pips | | 4時間足 2024 | 19 | +911.2pips | 9 | −66.3pips | **フィルターなしで損していた19通りは、6区分すべてで17通りから19通りが改善しました。稼いでいた19通りは1通りから18通りです。** 1時間足の2025年が分かりやすい例です。いちばん助かったのはフェアバリューギャップの設定で、−3,491.7pips（174回）がADX30以上で−2,230.8pips（83回）になりました。1,260.9pipsの改善ですが、まだ赤字です。逆にいちばん落ちたのはウェーブトレンドで、+3,283.4pips（867回）が、6つのうち一番良いものを選んでも+1,955.5pips（436回）でした。 **フィルターは、負けていた設定を持ち上げ、稼いでいた設定を押し下げました。**売買を仕分けているなら、フィルターなしの成績がどちら側の設定でも同じように効くはずです。範囲の外へ出た数が偶然と変わらなかったことと、同じ結果を別の角度から見ています。 ## 前年に効いたフィルターを翌年も使えるか このシリーズでは毎回、片方の年で一番だった設定を相手の年に当てています。フィルターでも同じことをしました。**2024年の成績が一番良かったフィルターを設定ごとに選び、それを2025年に使います。** | 時間足 | 2024年に「なし」が一番だった設定 | 2025年にフィルターなしを上回った設定 | 2025年に黒字で終わった設定 | |---|---|---|---| | 15分足 | 1／38 | 19／38 | 18／38 | | 1時間足 | 13／38 | 5／38 | 24／38 | | 4時間足 | 10／38 | 14／38 | 19／38 | 左の列が示しているのは、**2024年のデータだけを見れば、114通りのうち90通りで「何かフィルターを足したほうが良かった」ように見えるということです。** その選択を翌年に持ち込むと、フィルターなしを上回ったのは114通り中38通りでした。1時間足に至っては38通り中5通りです。 ## 方向のフィルターが2024年だけ強かった理由 はっきり効いた区分が1つあります。15分足の2024年で、50本移動平均の傾きは38通り中36通りを改善し、年間損益の中央値を−205.9pipsから+1,157.3pipsへ動かしました。この36通りは、6つのフィルター×3時間足×2年の36区分で最も多い改善数です。 この年の米ドル／円は、15分足の終値で140.8655から157.1975まで、1,633.2pips上昇しています。2025年は157.2375から156.67で、−56.8pipsでした。 **「50本平均が上向きのときだけ買う」は、上げ相場では買いを通して売りを止めるルールです。**同じフィルターを2025年の15分足に掛けると改善は20通りにとどまり、中央値も+87.1pipsから+32.3pipsへ下がりました。 向きのフィルターがしているのは、ダマシを避けることではなく、**その年に価格が進んだ側へ寄せることです。**どちらへ進むかを先に知っていれば効きますが、それが分かっているならフィルターは要りません。 ## 関連記事 - [DMI/ADXのDIクロスは勝てる？1,344通り検証した結果](/ja/blog/dmi-adx-settings)：この記事でフィルターに使ったADXそのものを、売買ルールとして測っています - [RSIとボリンジャーバンドの組み合わせを720通り検証](/ja/blog/rsi-bollinger-combo)：ランダムに間引いた場合と比べる型は、この記事で作ったものです - [GMMAは移動平均線2本と何が違う？216通り検証](/ja/blog/gmma-settings)：ADXフィルターがこの検証で一番壊した条件だった記事です - [移動平均線のゴールデンクロスは勝てる？280通り検証した結果](/ja/blog/moving-average-cross-settings)：フィルターを掛けた38通りに入っている設定のひとつです - [ドル円の設定は他の通貨ペアで通用する？3,024通り検証](/ja/blog/settings-across-pairs)：ほぼ同じ設定の集合に、フィルターではなく通貨ペアを掛けた検証です ## 補足 - 対象は米ドル／円のみです。2024年と2025年の2年分で、時間足は15分足・1時間足・4時間足です - 約定は終値、スプレッドは0.3pips固定、ロットは0.1、スリッページはありません - 元の設定38通りのうち、MACDの記事（シグナルクロス12/26/9）とOsMAの記事（12/26/9）は42回の実行すべてで取引回数も損益も一致しました。OsMAはMACDとシグナル線の差なので、そのゼロクロスはMACDのシグナルクロスと同じ足になります - ランダムの範囲は固定した種の乱数から作っているので、同じ数字が再現できます --- ### インサイドバーのブレイク勝率を192通り検証した結果 URL: https://formiq.jp/ja/blog/inside-outside-bar-settings Language: ja Published: 2026-09-10 Updated: 2026-09-10 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: インサイドバー, アウトサイドバー, はらみ足, ブレイク, プライスアクション, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/inside-outside-bar-settings Evidence JSON: https://formiq.jp/ja/evidence/inside-outside-bar-settings/data.json Also available in: en — https://formiq.jp/blog/inside-outside-bar-settings インサイドバーは、直前の足の高値と安値の中に完全に収まった足です。値動きが止まった合図と読まれ、**その次の足が母足の外へ抜けた方向に乗る**のが一般的な使い方になります。アウトサイドバーはその反対で、直前の足の高値と安値の両方を超えた足です。 この形には測るべき相手がはっきりしています。**同じ高値を抜いたけれど、直前の足が収まっていなかったブレイクです。** インサイドバーが意味を持つなら、この2つは違う結果になるはずです。 1時間足の2025年で、2本前の高値か安値を抜いた足は6,126回ありました。そのうち**直前の足が収まっていたのは566回、9.24%です。** そして10本後にブレイク方向へ動いていた割合は、**インサイドバー発が49.47%、普通のブレイクが50.59%**でした。2024年も47.68%対51.19%で、同じ向きです。 ## インサイドバーとアウトサイドバーは何を見ているのか Formiqのバックテストが使っている判定は、3本の足を見ます。 - **母足**（2本前）：高値と安値の基準になる足 - **内側の足**（1本前）：高値が母足の高値以下、安値が母足の安値以上 - **抜けた足**（今の足）：母足の高値を上抜けたら買い、母足の安値を下抜けたら売り ヒゲを含む高値と安値で判定します。実体は見ません。 アウトサイドバーはこの記事では別に数えました。**直前の足の高値を上回り、かつ安値を下回った足**で、終値の向きをその足の向きとしています。 ## ブレイクの1割しかインサイドバー発ではない 2本前の高値か安値を抜いた足を全部数え、そのうち直前の足が収まっていたものを分けます。 | 時間足 | 年 | 抜けた足の合計 | インサイドバー発 | その割合 | アウトサイドバー | |---|---|---|---|---|---| | 15分足 | 2025年 | 24498回 | 2462回 | 10.05% | 2718回 | | 15分足 | 2024年 | 24132回 | 2542回 | 10.53% | 2529回 | | 1時間足 | 2025年 | 6126回 | 566回 | 9.24% | 720回 | | 1時間足 | 2024年 | 5987回 | 625回 | 10.44% | 665回 | | 4時間足 | 2025年 | 1555回 | 195回 | 12.54% | 220回 | | 4時間足 | 2024年 | 1552回 | 192回 | 12.37% | 208回 | **インサイドバー発のブレイクは、全体の9.24〜12.54%です。** 残りの約9割は、直前の足が母足からはみ出していたブレイクになります。 ## インサイドバーの設定と読み方 チャートに線や別枠は増えません。ローソク足そのものの位置関係なので、見るのは価格チャートだけです。 | 条件 | 数値の項目 | 成立の中身 | |---|---|---| | インサイドバーのブレイク | なし | 1本前が2本前の高値安値に収まり、今の足がその外へ抜けた | | はらみ足 | なし | 今の足の実体が直前の実体に収まり、色が反対で実体が小さい | **どちらも数値の項目はありません。** 「どれだけ小さく収まればインサイドバーか」を決める設定はなく、収まったかどうかの2択です。 絞り込みたい場合は条件を横に足します。この記事の検証もその形で作りました。 | 足したい条件 | 組み合わせる条件 | 設定 | |---|---|---| | 抜けた足の実体がヒゲに対して太い | 丸坊主 | 実体比率0.7 | | 抜けた足の値幅が大きい | 大陽線・大陰線 | ATR倍率1.5、実体比率0.5 | | 移動平均の向きに合わせる | 移動平均の傾き | 期間50、判定5本前と比較 | ### 一般に使われている売買ルール 抜けた足の終値で、抜けた方向に入ります。決済は反対側へ抜けたときで、Formiqの条件も同じ形です。損切りは母足の反対側の端に置く形が紹介されます。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 2024年1月1日〜2025年12月31日 | | 成立の判定 | Formiqのバックテスト機能と同じ判定で、1本ずつ数える | | 比べる相手 | 同じ2本前の高値・安値を抜いたが、直前の足が収まっていなかったブレイク | | 先の見方 | 抜けた足の終値から1本後・5本後・10本後・20本後の終値 | | 決済 | 反対側へ抜けるまで持つ形と、10本後の終値で決済する形 | | 条件の数 | インサイドバー4通り＋はらみ足3通り＋10本決済＝8通り | | 回した数 | 8通り×3時間足×4期間×手数料あり／なし＝192通り | | スプレッド | 0.3pips固定 | | ロット | 0.1 | | 約定 | 終値 | ## 同じ高値を抜いた普通のブレイクと比べると 10本後にブレイク方向へ動いていた割合です。 | 時間足 | 年 | インサイドバー発 | その割合 | 普通のブレイク | その割合 | 差 | |---|---|---|---|---|---|---| | 15分足 | 2025年 | 2462回 | 50.73% | 22036回 | 49.62% | +1.11 | | 15分足 | 2024年 | 2542回 | 48.23% | 21590回 | 49.77% | -1.54 | | 1時間足 | 2025年 | 566回 | 49.47% | 5560回 | 50.59% | -1.12 | | 1時間足 | 2024年 | 625回 | 47.68% | 5362回 | 51.19% | -3.51 | | 4時間足 | 2025年 | 195回 | 48.21% | 1360回 | 49.63% | -1.43 | | 4時間足 | 2024年 | 192回 | 51.04% | 1360回 | 48.75% | +2.29 | **上回ったのは6通り中2通りで、しかも別の時間足・別の年です。** 15分足は2025年に+1.11ポイント、翌年に−1.54ポイント。4時間足は2025年に−1.43ポイント、翌年に+2.29ポイントでした。1時間足は2年ともマイナスです。 pipsで見ると1時間足は、インサイドバー発が+0.58pipsと+1.29pips、普通のブレイクが+1.89pipsと+2.14pipsでした。**どちらもプラスですが、直前の足が収まっていたほうが小さくなっています。** ## 逆に張ったらどうなるか ブレイクに乗るのではなく、抜けた反対側に入る形です。同じ566回を反対向きに測ります。 | 時間足 | 年 | 順張り | 逆張り | 普通のブレイクに乗る | |---|---|---|---|---| | 15分足 | 2025年 | 50.73% | 49.19% | 49.62% | | 15分足 | 2024年 | 48.23% | 51.42% | 49.77% | | 1時間足 | 2025年 | 49.47% | 50.53% | 50.59% | | 1時間足 | 2024年 | 47.68% | 52.16% | 51.19% | | 4時間足 | 2025年 | 48.21% | 51.79% | 49.63% | | 4時間足 | 2024年 | 51.04% | 48.96% | 48.75% | 順張りと逆張りは同じ取引を裏返したものなので、2つを足すと100%になります（引き分けの分だけ差が出ます）。**逆張りが順張りを上回ったのは6通り中4通りです。** 比べる相手を普通のブレイクに戻すと、逆張りが上回ったのも6通り中4通りで、差は+0.21ポイント（4時間足2024年）から+2.16ポイント（4時間足2025年）でした。下回った2通りは15分足の2025年と1時間足の2025年です。ただし上回った4通りと、順張りを上回った4通りは同じ組み合わせではありません。 つまり「インサイドバーのブレイクはダマシだから逆に張る」という形は、この2年では順張りより少しだけ高く出ますが、同じ高値を抜いた普通のブレイクに対しては2ポイント以内の差にとどまり、6通りのうち2通りでは下回っています。 ## 2本・3本と続いたとき、両端を抜けたとき インサイドバーが何本続いたあとのブレイクかで分けます。 | 時間足 | 年 | 1本 | 2本 | 3本以上 | |---|---|---|---|---| | 15分足 | 2025年 | 2248回・51.11% | 195回・44.10% | 19回・73.68% | | 15分足 | 2024年 | 2275回・47.82% | 246回・51.63% | 21回・52.38% | | 1時間足 | 2025年 | 498回・49.40% | 61回・45.90% | 7回・85.71% | | 1時間足 | 2024年 | 537回・48.04% | 80回・45.00% | 8回・50.00% | | 4時間足 | 2025年 | 156回・48.72% | 34回・44.12% | 5回・60.00% | | 4時間足 | 2024年 | 148回・51.35% | 43回・51.16% | 1回・0.00% | **3本以上の列は1回から21回しかありません。** 1時間足の85.71%は7回のうち6回という意味で、この本数から何かを決めることはできません。2本続いた場合は6通り中5通りで1本より低くなっています。 インサイドバーが最も長く続いたのは、1時間足の2024年の5本でした。 もう1つ、抜けた足が母足の高値と安値を**同じ足で両方抜ける**ことがあります。 | 時間足 | 年 | インサイドバー発 | 両端を抜けた | その割合 | 両端の買いの割合 | |---|---|---|---|---|---| | 15分足 | 2025年 | 2462回 | 262回 | 10.64% | 131回・48.85% | | 15分足 | 2024年 | 2542回 | 292回 | 11.49% | 146回・52.05% | | 1時間足 | 2025年 | 566回 | 66回 | 11.66% | 33回・63.64% | | 1時間足 | 2024年 | 625回 | 52回 | 8.32% | 26回・50.00% | | 4時間足 | 2025年 | 195回 | 62回 | 31.79% | 31回・61.29% | | 4時間足 | 2024年 | 192回 | 34回 | 17.71% | 17回・64.71% | **この足は買いと売りの両方に当てはまります。** ローソク足1本からは高値と安値のどちらが先だったか分からないので、順序を決められません。Formiqのバックテストは買いを先に判定するため、**両端を抜けた足はすべて買いとして処理されます。** 4時間足の2025年ではインサイドバー発の31.79%がこれに当たります。 買いとして扱ったときの10本後は、17回から146回のサンプルで48.85%から64.71%まで散らばりました。件数が少なく、方向の決め方も測定の結果ではないので、この列は成績ではなく仕様として読んでください。 ## アウトサイドバーは何が違うのか 直前の足の高値と安値を両方超えた足を、終値の向きに10本持ちます。 | 時間足 | 年 | アウトサイドバー | その割合 | 普通のブレイク | その割合 | |---|---|---|---|---|---| | 15分足 | 2025年 | 2718回 | 48.86% | 22036回 | 49.62% | | 15分足 | 2024年 | 2529回 | 50.85% | 21590回 | 49.77% | | 1時間足 | 2025年 | 720回 | 51.25% | 5560回 | 50.59% | | 1時間足 | 2024年 | 665回 | 50.38% | 5362回 | 51.19% | | 4時間足 | 2025年 | 220回 | 51.82% | 1360回 | 49.63% | | 4時間足 | 2024年 | 208回 | 50.00% | 1360回 | 48.75% | **上回ったのは6通り中4通りで、差は−0.81ポイントから+2.19ポイントの間です。** インサイドバーと同じく、普通のブレイクからはっきり離れる値にはなりませんでした。 ## はらみ足とインサイドバーは同じものか どちらも「小さい足が大きい足に収まる」形ですが、成立する足が違います。**はらみ足は内側の足が出たその足で成立し、インサイドバーのブレイクは次の足で成立します。** 判定するものも違い、はらみ足は実体どうしの包含と色の反転、インサイドバーは高値安値の包含です。 | 時間足 | 年 | 向き | 同じ足で両方成立 | はらみ足のうちの割合 | |---|---|---|---|---| | 15分足 | 2025年 | 買い | 24回 | 1.55% | | 15分足 | 2025年 | 売り | 20回 | 1.24% | | 1時間足 | 2025年 | 買い | 5回 | 1.31% | | 1時間足 | 2025年 | 売り | 5回 | 1.31% | | 4時間足 | 2025年 | 買い | 3回 | 2.80% | | 4時間足 | 2025年 | 売り | 0回 | 0.00% | **同じ足で両方が成立したのは、はらみ足の0.91〜2.80%です。** 名前が似ていても、売買条件としては別の瞬間を指しています。 ## 192通りを売買した結果 抜けた足の終値で入り、反対側へ抜けるまで持ちます。 | 条件 | 2025年 回数 | 2025年 勝率 | 2025年 年間損益 | 2024年 回数 | 2024年 勝率 | 2024年 年間損益 | |---|---|---|---|---|---|---| | インサイドバーのブレイク | 285回 | 44.21% | +384.9pips | 301回 | 43.19% | +699.9pips | | 同・10本で決済 | 304回 | 51.32% | +727.0pips | 312回 | 45.83% | -360.4pips | | 同・実体0.7以上 | 115回 | 44.35% | +1396.4pips | 131回 | 45.04% | -391.9pips | | 同・ATR1.5倍以上 | 33回 | 51.52% | +554.6pips | 54回 | 51.85% | +543.6pips | | 同・50本平均の向き | 166回 | 39.16% | -587.8pips | 180回 | 46.11% | +309.2pips | | はらみ足 | 426回 | 50.23% | +62.6pips | 432回 | 50.23% | -2007.5pips | | はらみ足・10本で決済 | 344回 | 47.67% | -662.3pips | 362回 | 46.13% | -2657.8pips | | はらみ足・50本平均の向き | 241回 | 48.96% | -546.2pips | 243回 | 50.62% | -1735.9pips | 1時間足です。**2024年と2025年の両方で黒字になったのは8通り中2通り**で、そのままのインサイドバーのブレイクと、ATR1.5倍以上を足したものでした。ATR版は2025年に33回、2024年に54回の売買です。 15分足では8通り中1通り（50本平均の向き）、4時間足では3通り（10本決済、はらみ足、はらみ足＋50本平均の向き）でした。**2つ以上の時間足で2年とも黒字になった条件は1つもありません。** ### 2025年を前半と後半に割ると | 条件 | 2025年通年 | 1〜6月 | 7〜12月 | |---|---|---|---| | インサイドバーのブレイク | +384.9pips | +762.0pips | -484.4pips | | 同・10本で決済 | +727.0pips | +440.2pips | +281.3pips | | 同・実体0.7以上 | +1396.4pips | +501.3pips | +787.8pips | | 同・ATR1.5倍以上 | +554.6pips | +441.7pips | +112.9pips | | はらみ足 | +62.6pips | -579.4pips | +641.2pips | そのままのインサイドバーのブレイクは、通年+384.9pipsですが後半だけなら−484.4pipsです。前半と後半の両方が黒字だったのは、10本決済・実体0.7以上・ATR1.5倍以上の3つでした。 ### 前年に一番稼いだ条件を翌年へ移すと | 時間足 | 2024年の1位 | 2024年の年間損益 | 2025年の年間損益 | 2025年の順位 | 2025年の8通りの中央値 | |---|---|---|---|---|---| | 15分足 | はらみ足＋50本平均の向き | +1710.7pips | -348.4pips | 6位 | +515.6pips | | 1時間足 | インサイドバーのブレイク | +699.9pips | +384.9pips | 4位 | +223.8pips | | 4時間足 | インサイドバーのブレイク | +2999.9pips | -779.6pips | 6位 | -134.7pips | 15分足と4時間足は翌年に中央値を下回りました。4時間足は+2,999.9pipsが−779.6pipsになっています。 ## 損切り・利確とコスト | 損切り・利確 | 2025年 回数 | 2025年 勝率 | 2025年 損益 | 2024年 損益 | |---|---|---|---|---| | なし（反対側へ抜けるまで） | 285回 | 44.21% | +384.9pips | +699.9pips | | 30・60 | 408回 | 34.31% | -300.4pips | +564.5pips | | 60・30 | 410回 | 54.88% | -536.6pips | -68.4pips | | 50・50 | 399回 | 45.36% | +80.8pips | -253.6pips | 1時間足のインサイドバーのブレイクです。**勝率が最も高い60・30は、2年とも赤字でした。** 入る足は同じで、勝率だけが34.31%から54.88%まで動きます。 | 時間足 | 条件 | 取引回数 | 支払ったスプレッド | 回数×0.3pips | |---|---|---|---|---| | 15分足 | インサイドバーのブレイク | 1184回 | 355.2pips | 355.2pips | | 15分足 | はらみ足 | 1720回 | 515.9pips | 516.0pips | | 1時間足 | インサイドバーのブレイク | 285回 | 85.5pips | 85.5pips | | 1時間足 | はらみ足 | 426回 | 127.8pips | 127.8pips | | 4時間足 | はらみ足 | 103回 | 30.9pips | 30.9pips | 支払ったスプレッドは取引回数×0.3pipsと一致しました。15分足のはらみ足はスプレッドを0にしても−485.8pipsで、回数の多さではなく入り方から赤字になっています。 ## 関連記事 - [ローソク足パターン17種の勝率](/ja/blog/candlestick-patterns-tested)：この2つを含む17種を、同じ2年・同じ対照群で横断的に並べています - [包み足の勝率とだまし](/ja/blog/engulfing-settings)：はらみ足の反対にあたる形を、色が変わっただけの足と比べています - [明けの明星・宵の明星の勝率](/ja/blog/star-patterns-settings)：3本でできる反転の形を、同じ対照群の置き方で測っています - [フェアバリューギャップは本当に埋まるのか](/ja/blog/fair-value-gap-settings)：同じく普通の足を比べる相手に置いて、プライスアクションを測っています ## 補足 対象は米ドル／円の2024年1月1日から2025年12月31日です。終値約定、スプレッド0.3pips固定、0.1ロットで計算しました。成立の判定にはFormiqのバックテスト機能と同じ処理を使い、対象期間の前の足も読み込んでいます。1本後から20本後までを見るため、年の終わりからその本数ぶんは対象から外しています。母足の両端を同じ足で抜けたケースは、順張りと逆張りの表では買いと売りの両方に数え、売買では買いとして処理しています。はらみ足との重なりは、同じ足で両方の条件が同じ向きに成立した回数です。 --- ### ピンバーの勝率とヒゲ比率を252通り検証した結果 URL: https://formiq.jp/ja/blog/pin-bar-settings Language: ja Published: 2026-09-10 Updated: 2026-09-10 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ピンバー, ヒゲ, プライスアクション, ローソク足, ハンマー, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/pin-bar-settings Evidence JSON: https://formiq.jp/ja/evidence/pin-bar-settings/data.json Also available in: en — https://formiq.jp/blog/pin-bar-settings ピンバーは、片側に長いヒゲを持つ足です。長い下ヒゲは下で買われた跡、長い上ヒゲは上で売られた跡と読まれ、**ヒゲと反対の方向に入る**のが一般的な使い方になります。 「長い」の基準は数字で決まっています。Formiqのバックテストは**ヒゲが実体の何倍あるか**で判定し、初期設定は2倍です。この記事はその数字を1倍から5倍まで動かしました。 まず分かったことが1つあります。**初期設定の2倍では、全部の足の27.6〜30.4%が該当します。** 3本に1本近くです。 そして倍率を上げても、成績はほとんど動きませんでした。1時間足の2024年なら、1倍で51.46%、5倍で51.69%。取引の候補は2,973回から859回へ3分の1以下に減るのに、割合は0.3ポイントも動きません。 **動いたのは、ヒゲの測り方を変えたときでした。** 実体との比ではなく、その足の値幅に対する割合で測り直すと、6通り中5通りで全部の足を上回ります。 ## ピンバーは何を見ているのか Formiqのバックテストが使っている判定です。 - **下ヒゲ** ≧ 実体 × ヒゲ倍率、かつ **下ヒゲ > 上ヒゲ** → 買い - **上ヒゲ** ≧ 実体 × ヒゲ倍率、かつ **上ヒゲ > 下ヒゲ** → 売り ヒゲは実体の外側の部分で、下ヒゲは始値と終値の低いほうから安値まで、上ヒゲは高いほうから高値までです。 **比べているのはヒゲと実体です。** 値幅（高値から安値まで）は判定に入っていません。実体が小さければ、ヒゲが短くても倍率を軽く超えます。逆に実体が大きい足は、ヒゲが絶対的に長くても倍率に届きません。 ## 初期設定では3本に1本近くがピンバーになる ヒゲ倍率2で成立した足を数えます。片側のヒゲがもう片側より長い足だけが対象です。 | 時間足 | 年 | 対象の足 | 倍率2で成立 | その割合 | |---|---|---|---|---| | 15分足 | 2025年 | 24760本 | 6957本 | 28.10% | | 15分足 | 2024年 | 24860本 | 6853本 | 27.57% | | 1時間足 | 2025年 | 6205本 | 1757本 | 28.32% | | 1時間足 | 2024年 | 6242本 | 1862本 | 29.83% | | 4時間足 | 2025年 | 1592本 | 469本 | 29.46% | | 4時間足 | 2024年 | 1613本 | 490本 | 30.38% | **27.57%から30.38%です。** 年に1時間足で1,800回前後、15分足で6,900回前後成立します。3本に1本近くが当てはまる形は、絞り込みとしては緩いほうです。 ## ピンバーの設定と読み方 チャートに線や別枠は増えません。ローソク足そのものの形なので、見るのは価格チャートだけです。 | 条件 | 数値の項目 | 初期設定 | 変えると何が変わるか | |---|---|---|---| | ピンバー | ヒゲ倍率 | 2 | 上げるほど実体に対して長いヒゲだけが残り、成立回数が減る | 買い側と売り側に同じ倍率がかかります。片側だけを厳しくする設定はありません。 ### 一般に使われている売買ルール ピンバーが確定した足の終値で、ヒゲと反対の方向に入ります。下ヒゲが長ければ買い、上ヒゲが長ければ売りです。決済は反対向きのピンバーが出るまで持つ形が多く、損切りはヒゲの先端に置きます。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 2024年1月1日〜2025年12月31日 | | 成立の判定 | Formiqのバックテスト機能と同じ判定で、1本ずつ数える | | 比べる相手 | 同じ期間の全部の足。買いと売りの構成に合わせて加重する | | 先の見方 | 成立した足の終値から1本後・5本後・10本後・20本後の終値 | | 決済 | 反対向きのピンバーまで持つ形と、10本後の終値で決済する形 | | ヒゲ倍率 | 1・1.5・2・2.5・3・4・5の7通り | | 回した数 | 7通り×決済2種と手数料なし1種×3時間足×4期間＝252通り | | スプレッド | 0.3pips固定 | | ロット | 0.1 | | 約定 | 終値 | **比べる相手の作り方だけ補足します。** ピンバーは買いと売りの両方を作るので、上げた年の全部の足と比べるときに買いの基準だけを使うと、売りが多い集団は不当に低く見えます。そこで各集団の買いと売りの数に合わせて、上昇した割合と下落した割合を混ぜた値を基準にしました。 ## ヒゲ比率を1から5まで動かすと 10本後にヒゲと反対の方向へ動いていた割合です。カッコ内は全部の足との差です。 | 倍率 | 1時間足2025年 | 1時間足2024年 | 4時間足2025年 | 4時間足2024年 | |---|---|---|---|---| | 1倍 | 2898回・50.83%（+0.81） | 2973回・51.46%（+0.81） | 766回・52.09%（+2.15） | 789回・55.64%（+3.70） | | 1.5倍 | 2188回・50.78%（+0.76） | 2281回・51.64%（+1.01） | 584回・53.42%（+3.49） | 618回・55.66%（+3.94） | | 2倍 | 1757回・50.88%（+0.87） | 1862回・51.40%（+0.83） | 469回・52.88%（+2.95） | 490回・56.12%（+4.18） | | 2.5倍 | 1476回・50.68%（+0.66） | 1562回・51.47%（+0.82） | 385回・54.29%（+4.36） | 416回・56.01%（+3.76） | | 3倍 | 1246回・50.40%（+0.38） | 1365回・51.72%（+1.08） | 336回・55.65%（+5.73） | 350回・55.43%（+2.86） | | 4倍 | 977回・50.05%（+0.03） | 1052回・51.52%（+0.79） | 270回・54.81%（+4.88） | 273回・53.11%（+0.80） | | 5倍 | 806回・50.00%（−0.02） | 859回・51.69%（+1.02） | 227回・53.74%（+3.83） | 228回・52.63%（+0.30） | **1時間足の2024年は、倍率を5倍にしても差が+1.02ポイントで、1倍の+0.81ポイントとほとんど同じです。** 2025年は上げるほど下がり、5倍で−0.02ポイントになりました。 4時間足では2025年が3倍で最大（+5.73）、2024年が2倍で最大（+4.18）になります。**同じ時間足でも、良かった倍率が年をまたいで一致しません。** 15分足は7段階すべてで全部の足を下回りました。2025年は−0.78から−0.40、2024年は−1.06から−0.06です。 **倍率を上げる効果は、回数が減ることのほうがはっきりしています。** 1時間足の2025年で2,898回から806回へ、4分の1近くまで落ちます。 ## ヒゲを「値幅の何割か」で測ると 同じ足を別の物差しで切り直します。長いほうのヒゲが、その足の値幅（高値から安値まで）の何割を占めるかです。 | 条件 | 1時間足2025年 | 1時間足2024年 | 4時間足2025年 | 4時間足2024年 | |---|---|---|---|---| | 30%以上 | 4092回・50.22%（+0.20） | 4208回・51.43%（+0.73） | 1059回・51.46%（+1.52） | 1087回・55.20%（+3.29） | | 40%以上 | 3038回・50.79%（+0.77） | 3159回・51.47%（+0.75） | 802回・52.00%（+2.07） | 818回・55.13%（+3.07） | | 50%以上 | 1971回・50.03%（0.00） | 2125回・51.44%（+0.63） | 520回・52.50%（+2.58） | 556回・56.12%（+3.78） | | 60%以上 | 1121回・50.04%（+0.03） | 1142回・53.06%（+2.10） | 296回・55.41%（+5.51） | 300回・56.00%（+3.25） | | 70%以上 | 511回・53.03%（+3.01） | 510回・53.53%（+2.23） | 126回・56.35%（+6.44） | 140回・55.71%（+1.92） | **70%以上の行は、1時間足でも4時間足でも2年とも全部の足を上回りました。** 15分足でも2024年は+0.34ポイントで、下回ったのは2025年の−1.36ポイントだけです。6通り中5通りです。 初期設定のヒゲ倍率2が上回ったのは6通り中4通りで、幅は+0.83から+4.18ポイントでした。**値幅で70%という切り方のほうが、上回った回数も差の大きさも上です。** 2つの物差しは同じ足を選びません。1時間足の2025年で、倍率2に当てはまるのは1,757本、ヒゲが値幅の50%以上に当てはまるのは1,971本、両方に当てはまるのは1,506本でした。**倍率だけが選ぶ足が251本、値幅だけが選ぶ足が465本あります。** どちらも「ヒゲが長い足」と呼びますが、指している足は4分の1ほど違います。 なおFormiqのバックテストに「ヒゲが値幅の何割以上」という条件はありません。この記事の値幅比は数えた結果で、そのまま売買条件にはできません。 ## 実体がゼロの足はピンバーにならない 判定には `実体 > 0` という条件が入っています。始値と終値がぴったり同じ足は、ヒゲがどれだけ長くてもピンバーになりません。 | 時間足 | 年 | 対象の足 | 実体ゼロの足 | その割合 | ヒゲが値幅に占める割合の中央値 | |---|---|---|---|---|---| | 15分足 | 2025年 | 24760本 | 83本 | 0.34% | 66.67% | | 15分足 | 2024年 | 24860本 | 105本 | 0.42% | 66.67% | | 1時間足 | 2025年 | 6205本 | 10本 | 0.16% | 58.65% | | 1時間足 | 2024年 | 6242本 | 13本 | 0.21% | 73.18% | | 4時間足 | 2025年 | 1592本 | 2本 | 0.13% | 80.26% | | 4時間足 | 2024年 | 1613本 | 4本 | 0.25% | 72.83% | 対象の足に占める割合は0.13%から0.42%で、数としては多くありません。ただし**ヒゲが値幅の6割から8割を占める足**が、実体がゼロという理由だけで外れています。倍率で測る限り、実体がゼロの足は「倍率が無限大」ではなく「対象外」です。 ## ハンマーとは何が違うのか ハンマーは、ピンバーの買い側に条件を1つ足したものです。ピンバーの買いは「下ヒゲが上ヒゲより長い」ことを求め、ハンマーは「上ヒゲが実体以下」まで求めます。 1時間足・2025年で、ピンバーの買いは937回、そのうちハンマーの条件も満たしたのは306回で32.66%でした。10本後に上昇していた割合は、ピンバーの買いが50.69%、ハンマーが54.58%です。 **ハンマーの306回はすべてピンバーの買いに含まれます。** 別々の手法として比べることはできません。 ## 252通りを売買した結果 成立した足の終値で入り、反対向きのピンバーが出るまで持ちます。 | 倍率 | 2025年 回数 | 2025年 勝率 | 2025年 年間損益 | 2024年 回数 | 2024年 勝率 | 2024年 年間損益 | |---|---|---|---|---|---|---| | 1倍 | 1417回 | 49.40% | -636.7pips | 1468回 | 49.86% | +455.4pips | | 1.5倍 | 1083回 | 49.86% | -276.3pips | 1122回 | 50.98% | +1430.8pips | | 2倍 | 884回 | 50.68% | +858.9pips | 890回 | 49.33% | +1560.4pips | | 2.5倍 | 741回 | 51.55% | +581.5pips | 730回 | 48.77% | +753.4pips | | 3倍 | 627回 | 51.20% | -105.2pips | 630回 | 48.73% | +1229.3pips | | 4倍 | 467回 | 50.11% | -576.8pips | 484回 | 48.14% | +612.0pips | | 5倍 | 391回 | 50.13% | +339.4pips | 394回 | 46.70% | +1008.0pips | 1時間足です。**2024年と2025年の両方で黒字になったのは7通り中3通り**（2倍・2.5倍・5倍）で、間の3倍と4倍は飛ばされています。連続した範囲になりません。 4時間足では7通り中6通りが両年黒字でした（2.5倍以外）。**15分足は7通りすべてが2年とも赤字です。** 10本後に決済する形でも15分足は7通りすべてが赤字で、1時間足は2通り、4時間足は4通りが両年黒字でした。 4時間足の数字は年で大きく違います。倍率2の1取引あたりは、2025年が+0.60pips、2024年が+14.40pipsでした。取引回数は224回と243回でほぼ同じです。 ### 2025年を前半と後半に割ると | 倍率 | 2025年通年 | 1〜6月 | 7〜12月 | |---|---|---|---| | 1倍 | -636.7pips | -198.3pips | -446.1pips | | 1.5倍 | -276.3pips | +916.5pips | -1269.3pips | | 2倍 | +858.9pips | +931.7pips | -149.3pips | | 2.5倍 | +581.5pips | +486.1pips | +18.9pips | | 3倍 | -105.2pips | +549.8pips | -731.5pips | | 4倍 | -576.8pips | +845.1pips | -1345.5pips | | 5倍 | +339.4pips | +701.3pips | -290.2pips | **前半と後半の両方が黒字だったのは2.5倍だけです。** 通年で黒字だった2倍も、後半は−149.3pipsでした。 ## 前年に一番稼いだ比率を翌年へ移すと | 時間足 | 2024年の1位 | 2024年の年間損益 | 2025年の年間損益 | 2025年の順位 | 2025年の7通りの中央値 | |---|---|---|---|---|---| | 15分足 | 5倍 | +551.2pips | -3212.9pips | 5位 | -2830.5pips | | 1時間足 | 2倍 | +1560.4pips | +858.9pips | 1位 | -105.2pips | | 4時間足 | 2倍 | +3498.7pips | +133.8pips | 6位 | +415.7pips | 1時間足だけは1位のままでした。4時間足は+3,498.7pipsが+133.8pipsになり、7通り中6位です。 ## 損切り・利確とコスト | 損切り・利確 | 2025年 回数 | 2025年 勝率 | 2025年 損益 | 2024年 損益 | |---|---|---|---|---| | なし（反対のピンバーまで） | 884回 | 50.68% | +858.9pips | +1560.4pips | | 30・60 | 1116回 | 43.46% | +184.6pips | +332.3pips | | 60・30 | 1056回 | 54.92% | +79.9pips | -362.1pips | | 50・50 | 1040回 | 50.00% | +348.3pips | +177.3pips | 1時間足・倍率2です。**勝率が最も高い60・30は、2024年が−362.1pipsで4通りのうち唯一の赤字でした。** 入る足は同じで、勝率は43.46%から54.92%まで動きます。 | 時間足 | 倍率 | 取引回数 | 支払ったスプレッド | 回数×0.3pips | |---|---|---|---|---| | 15分足 | 1倍 | 5612回 | 1683.5pips | 1683.6pips | | 15分足 | 2倍 | 3522回 | 1056.5pips | 1056.6pips | | 1時間足 | 2倍 | 884回 | 265.2pips | 265.2pips | | 4時間足 | 2倍 | 224回 | 67.2pips | 67.2pips | | 4時間足 | 5倍 | 88回 | 26.4pips | 26.4pips | 支払ったスプレッドは取引回数×0.3pipsと一致しました。**15分足は倍率2でスプレッドを0にしても−1,505.6pipsで、回数の多さではなく入り方から赤字になっています。** ## 関連記事 - [ローソク足パターン17種の勝率](/ja/blog/candlestick-patterns-tested)：ピンバーとハンマーを含む17種を、同じ2年で横断的に並べています - [包み足の勝率とだまし](/ja/blog/engulfing-settings)：同じく実体とヒゲの比率で絞り込みを作り、対照群と比べています - [インサイドバーのブレイク勝率](/ja/blog/inside-outside-bar-settings)：同じ水準を抜いた普通のブレイクを比べる相手にしています - [明けの明星・宵の明星の勝率](/ja/blog/star-patterns-settings)：3本でできる反転の形を、条件1つずつに分解しています - [ドル円の設定は他の通貨ペアで通用する？3,024通り検証](/ja/blog/settings-across-pairs)：この設定は2024年の4時間足でドル円1位で、ユーロドルでも42通り中1位でした ## 補足 対象は米ドル／円の2024年1月1日から2025年12月31日です。終値約定、スプレッド0.3pips固定、0.1ロットで計算しました。成立の判定はFormiqのバックテスト機能と同じ条件を書き直したもので、初期設定の倍率2では3つの時間足×2年×買い売りの12通りすべてで、エンジンが数えた回数と割合に一致しています。上下のヒゲが同じ長さの足は、どちらの方向も指さないため対象から外しました。1本後から20本後までを見るため、年の終わりからその本数ぶんは対象から外しています。 --- ### RCI「−80で買い」は本当に勝てる？2,484通り検証 URL: https://formiq.jp/ja/blog/rci-settings Language: ja Published: 2026-09-10 Updated: 2026-09-10 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: RCI, 順位相関, 3本表示, 逆張り, 順張り, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/rci-settings Evidence JSON: https://formiq.jp/ja/evidence/rci-settings/data.json Also available in: en — https://formiq.jp/blog/rci-settings RCIは「順位相関指数」の略で、−100から+100の間を動きます。定番の読み方は「−80以下で売られすぎ、+80以上で買われすぎ」。ここでは、その定番が本当に勝てるのかを2,484通り回して確かめました。 先に、RCIが何をしている指標なのかをはっきりさせておきます。**RCIは終値の並び順だけを見て、値幅を捨てます。** 過去9本が一直線に下げていれば、その9本で10pips下げても100pips下げてもRCIは同じ−100です。 そこで対照群を、RCIが捨てているもの、つまり同じ9本の変化率に置きました。 結果を3つ先に書きます。 **1つ目。±80は珍しい状態ではありません。** RCIが−80以下だった足は全体の6.99%から17.69%、+80以上は9.56%から31.31%です。分布の5%地点は−93.33から−84.96で、**RCIは真ん中よりも両端にいる時間のほうが長い**指標でした。 **2つ目。教科書どおりの「−80以下で買い」は、5つの読み方のうち最も弱い結果です。** 2024年と2025年の両方で黒字だった設定は、15分足で72通り中4通り、1時間足で3通り、4時間足で9通り。同じ水準を逆に読んで「+80以上で買い」にすると、39通り・33通り・6通りになります。 **3つ目。順位付けは変化率に勝っていません。** RCIが−80以下になった足と、同じ本数だけ変化率の小さい側から取った足を比べると、10本後の損益でRCIが上回ったのは36区分中17区分でした。 ## RCIは何を測っているのか RCIは、窓の中の終値について2つの順位を作り、その一致度を測ります。 1. **時間の順位**：古い足から1、2、3…と番号を振る 2. **価格の順位**：終値の安い順に1、2、3…と番号を振る 一直線に上げていれば、いちばん古い足がいちばん安いので2つの順位は完全に一致し、RCIは+100になります。一直線に下げていれば完全に逆順で−100です。 計算式は順位差を2乗して足したものを使います。期間9で順位差の2乗和が216になると、ちょうど−80になります。 **この作りから2つのことが出てきます。** **値幅は式に入りません。** 順位しか使わないので、9本で10pips下げた並びと100pips下げた並びは、並び順が同じなら同じ値になります。 **取れる値が飛び飛びになります。** 順位差の2乗和は整数なので、期間9のRCIは0.833刻みの値しか取りません。1時間足の2年分（13,482本）で実際に現れた値は313種類だけでした。期間52なら刻みは0.004で、11,429種類まで細かくなります。 期間9では、**+100ちょうどが51本、−100ちょうどが39本**ありました。9本すべてが前の足より高い（または低い）並びで、指標がそれ以上動けない状態です。合わせて全体の0.67%です。期間21以上では1本もありません。 ## RCIの設定と読み方 RCIは価格チャートの下の別枠に、−100から+100の目盛りで表示されます。Formiqでは±80の水準線が初めから引かれます。MT4で水準線が出ない場合は自分で追加します。 チャートに出す線は1本から3本まで選べます。日本語のRCI解説でよく見る「3本表示」は、短期9・中期26・長期52を重ねたものです。 | 場所 | 項目 | 初期設定 | 中身 | |---|---|---|---| | チャート | 短期 | 9 | 何本の終値を並べるか | | チャート | 中期 | 0 | 0なら描かない。26にすると2本目が出る | | チャート | 長期 | 0 | 0なら描かない。52にすると3本目が出る | | バックテスト | 読み方 | 水準の行き過ぎ | 下の5つから選ぶ | | バックテスト | 期間 | 9 | 何本の終値を並べるか | | バックテスト | 中期・長期 | 26・52 | 3本そろいのときだけ読まれる | | バックテスト | 向き | 以下 | 買いと売りのどちら側で成立させるか | | バックテスト | 水準 | −80 | 判定に使う値 | ### 5つの読み方 | 読み方 | 買いが成立する条件 | 売りが成立する条件 | |---|---|---| | 水準の行き過ぎ（逆張り） | RCIが水準より下 | RCIが水準より上 | | 水準を超えた側（順張り） | RCIが水準より上 | RCIが水準より下 | | ±水準の抜け | +水準を上抜けた足 | −水準を下抜けた足 | | 水準跨ぎ | 水準を上抜けた足 | 水準を下抜けた足 | | 3本そろい | 3本すべてが−水準より下 | 3本すべてが+水準より上 | **「水準の行き過ぎ」と「水準を超えた側」は、同じ数字を逆向きに読んだ2つです。** どちらも買いか売りの一方しか作らないので、この検証では買い側と売り側を別々に回しました。決済は水準を戻したところに置きます。「−80以下で買って、0に戻ったら決済」が定番の形です。 残る3つは両方向の売買を作るので、エントリーと決済に同じ条件を使います。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 時間足 | 15分足・1時間足・4時間足 | | 期間（対象年） | 2024年1月1日〜2025年12月31日。2025年は前半後半にも分割 | | RCIの期間 | 9・14・21・26・36・52 | | 逆張り・順張りの水準 | 成立が70・80・90、決済が0・40 | | ±水準の抜け | 50・70・80・90 | | 水準跨ぎ | −80・−40・0・40・80 | | 3本そろい | 9/26/52・9/14/21・14/26/52を水準50・70・80で | | 比べる相手 | 同じ終値の範囲の変化率（ROCの期間をひとつ短くしたもの） | | 回した数 | 207通り×3時間足×4期間＝2,484通り | | スプレッド | 0.3pips固定 | | ロット | 0.1 | | 約定 | 終値 | **対照群の作り方を書いておきます。** 期間9のRCIは終値9本を読むので、同じ9本の値動きは「9本前の終値からの変化率」です。Formiqの条件一覧にある変化率（ROC）で期間8を指定すると、同じ範囲になります。RCIが−80以下になった本数を数え、変化率の小さい側から同じ本数を取って、10本後までの値動きを比べました。 ## ±80は珍しい状態ではない まず、RCIがどれくらいの頻度で水準に届くのかを数えます。2025年の全足です。 | 時間足 | 期間 | 本数 | −80以下 | +80以上 | 5%地点 | 95%地点 | |---|---|---|---|---|---|---| | 15分足 | 9 | 24,893本 | 14.96% | 16.47% | −92.08 | 93.33 | | 15分足 | 26 | 24,893本 | 12.45% | 13.62% | −89.81 | 91.25 | | 15分足 | 52 | 24,893本 | 13.02% | 14.91% | −89.49 | 90.95 | | 1時間足 | 9 | 6,216本 | 15.70% | 16.04% | −93.33 | 93.33 | | 1時間足 | 26 | 6,216本 | 14.03% | 14.43% | −91.59 | 90.29 | | 1時間足 | 52 | 6,216本 | 14.56% | 14.54% | −89.70 | 89.39 | | 4時間足 | 9 | 1,600本 | 17.69% | 16.88% | −93.33 | 91.67 | | 4時間足 | 26 | 1,600本 | 9.94% | 16.25% | −87.97 | 90.97 | | 4時間足 | 52 | 1,600本 | 11.25% | 9.56% | −88.73 | 87.07 | 2年分・3時間足・6期間の36区分では、−80以下が6.99%から17.69%、+80以上が9.56%から31.31%でした。**期間9では両側を合わせて31.28%から34.57%、つまり3本に1本で「行き過ぎ」が成立しています。** 5%地点と95%地点にも注目してください。どの区分でも5%地点は−93.33から−84.96、95%地点は87.07から96.67です。**RCIは0の近くよりも±90の近くにいる時間のほうが長い**ので、「±80に届いた」は特別な出来事ではありません。 ### 3本そろいはどれくらい起きるか 短期9・中期26・長期52が全部±80の外にある足を数えました。 | 時間足 | 短期だけが外 | 3本とも外 | 全体に対する割合 | 短期だけの何% | |---|---|---|---|---| | 15分足 | 7,824本 | 805本 | 3.23% | 10.29% | | 1時間足 | 1,973本 | 122本 | 1.96% | 6.18% | | 4時間足 | 553本 | 29本 | 1.81% | 5.24% | 3本そろいは足の1.81%から5.14%です。短期だけが外に出た足のうち、3本そろうのは5.24%から14.98%でした。**4時間足の2025年は1年で29本しかありません。** 数字が派手に見えても、その裏の取引回数を必ず確認する必要がある水準です。 ## 順位だけを見ると何が変わるのか RCIが捨てている値幅と比べます。RCIが−80以下になった足と、同じ本数だけ変化率の小さい側から取った足の、10本後までの値動きです。2025年です。 | 時間足 | 期間 | 相関 | RCIの本数 | 重なり | RCIの平均 | 変化率の平均 | 全足の平均 | |---|---|---|---|---|---|---|---| | 15分足 | 9 | 0.756 | 3,723本 | 62.56% | +0.59pips | −0.03pips | −0.03pips | | 15分足 | 26 | 0.773 | 3,100本 | 59.61% | −0.18pips | +1.73pips | −0.03pips | | 15分足 | 52 | 0.785 | 3,242本 | 58.95% | −3.17pips | +0.30pips | −0.03pips | | 1時間足 | 9 | 0.798 | 976本 | 64.65% | −4.91pips | +0.15pips | −0.07pips | | 1時間足 | 26 | 0.810 | 872本 | 59.40% | +3.42pips | +7.57pips | −0.07pips | | 1時間足 | 52 | 0.823 | 905本 | 62.98% | +5.29pips | +5.67pips | −0.07pips | | 4時間足 | 9 | 0.828 | 283本 | 65.37% | −1.86pips | +7.05pips | −0.48pips | | 4時間足 | 26 | 0.832 | 159本 | 50.31% | +12.96pips | +13.24pips | −0.48pips | | 4時間足 | 52 | 0.777 | 180本 | 51.11% | +28.25pips | +9.57pips | −0.48pips | **RCIと変化率の相関は0.679から0.855です。** 高いですが1ではありません。 **重なりは50.31%から70.83%でした。** つまりRCIが「売られすぎ」と呼ぶ足の3割から5割は、変化率で選ぶと入ってこない別の足です。順位付けは確かに違う足を選んでいます。 ただし、選び分けた結果が良かったわけではありません。**2年分・3時間足・6期間の36区分で、RCIの側が変化率の側を上回ったのは17区分、勝率では18区分です。** どちらを使っても同じくらい、という結果でした。 ### 捨てた値幅は何か教えていたのか RCIが値幅を捨てているなら、捨てた値幅に情報が残っているかもしれません。RCIが−80以下の足だけを取り出し、同じ9本の変化率で4つに分けました。2025年です。 | 時間足 | 変化率の区分 | 本数 | 変化率の幅 | 10本後の勝率 | 10本後の平均 | |---|---|---|---|---|---| | 15分足 | 最も下げた4分の1 | 930本 | −1.622%〜−0.269% | 52.37% | +0.86pips | | 15分足 | 2番目 | 930本 | −0.269%〜−0.174% | 55.70% | +1.75pips | | 15分足 | 3番目 | 930本 | −0.174%〜−0.110% | 52.37% | −0.74pips | | 15分足 | 最も下げていない4分の1 | 933本 | −0.110%〜−0.007% | 51.98% | +0.51pips | | 1時間足 | 最も下げた4分の1 | 244本 | −2.047%〜−0.535% | 50.82% | −6.63pips | | 1時間足 | 2番目 | 244本 | −0.532%〜−0.377% | 48.36% | −4.19pips | | 1時間足 | 3番目 | 244本 | −0.377%〜−0.250% | 48.36% | −3.01pips | | 1時間足 | 最も下げていない4分の1 | 244本 | −0.250%〜−0.027% | 46.72% | −5.81pips | **変化率の大小と10本後の値動きの間に、順番のある関係は出ていません。** 15分足では2番目の区分が最も良く（+1.75pips）、最も下げた区分は+0.86pipsです。1時間足は4区分すべてマイナスで、最も下げた区分が−6.63pips、最も下げていない区分が−5.81pipsでした。 **RCIが捨てているものは、この検証では何も教えていませんでした。** ただし残している順位のほうも、36区分中17区分でしか変化率を上回っていません。捨てた側と残した側のどちらも、10本後の値動きを説明していないという結果です。 ## 5つの読み方を売買した結果 207通りを売買しました。年間損益がプラスで終わった設定の数です。 ### 2025年 | 読み方 | 設定数 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---|---| | 水準の行き過ぎ（逆張り） | 72 | 13 | 34 | 65 | | 水準を超えた側（順張り） | 72 | 54 | 36 | 7 | | ±水準の抜け | 24 | 18 | 13 | 3 | | 水準跨ぎ | 30 | 19 | 21 | 6 | | 3本そろい | 9 | 6 | 7 | 6 | ### 2024年 | 読み方 | 設定数 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---|---| | 水準の行き過ぎ（逆張り） | 72 | 14 | 9 | 12 | | 水準を超えた側（順張り） | 72 | 49 | 63 | 59 | | ±水準の抜け | 24 | 18 | 20 | 20 | | 水準跨ぎ | 30 | 16 | 22 | 22 | | 3本そろい | 9 | 2 | 2 | 2 | 1年ずつ見ると入れ替わります。逆張りは2025年の4時間足で72通り中65通りが黒字ですが、2024年の同じ4時間足では12通りです。3本そろいは2025年に27通り中19通りが黒字で、2024年は6通りでした。 そこで2つの年をそろえます。**2024年と2025年の両方で黒字だった設定の数です。** | 読み方 | 設定数 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---|---| | 水準の行き過ぎ（逆張り） | 72 | 4 | 3 | 9 | | 水準を超えた側（順張り） | 72 | 39 | 33 | 6 | | ±水準の抜け | 24 | 14 | 11 | 1 | | 水準跨ぎ | 30 | 12 | 18 | 5 | | 3本そろい | 9 | 2 | 2 | 0 | **定番の逆張りは、15分足と1時間足で最も少ない読み方です。** 72通りのうち4通りと3通りしか2年連続で黒字になりませんでした。同じ水準を逆に読んだだけの順張りは39通りと33通りです。 さらに2025年を前半と後半に割り、4つの期間すべてで黒字だった設定も数えました。 | 読み方 | 設定数 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---|---| | 水準の行き過ぎ（逆張り） | 72 | 2 | 0 | 4 | | 水準を超えた側（順張り） | 72 | 8 | 7 | 1 | | ±水準の抜け | 24 | 11 | 7 | 0 | | 水準跨ぎ | 30 | 7 | 8 | 1 | | 3本そろい | 9 | 2 | 2 | 0 | **1時間足の逆張りは、72通りすべてが4つの期間のどこかで赤字でした。** ## 教科書どおりの「−80で買い」を期間別に見る 水準80で成立し、0に戻ったら決済する形です。2025年です。 | 時間足 | 向き | 期間 | 取引回数 | 勝率 | 年間損益 | 平均保有 | |---|---|---|---|---|---|---| | 15分足 | 買い | 9 | 750回 | 66.13% | +367.4pips | 11.2本 | | 15分足 | 買い | 14 | 456回 | 65.13% | −336.6pips | 18.5本 | | 15分足 | 買い | 26 | 226回 | 65.93% | −426.2pips | 35.2本 | | 15分足 | 買い | 52 | 116回 | 61.21% | −1,074.4pips | 78.1本 | | 15分足 | 売り | 9 | 753回 | 63.08% | −1,194.3pips | 13.5本 | | 1時間足 | 買い | 9 | 176回 | 64.20% | −652.9pips | 13.7本 | | 1時間足 | 買い | 36 | 49回 | 67.35% | +322.6pips | 50.6本 | | 1時間足 | 売り | 21 | 93回 | 66.67% | +358.6pips | 28.8本 | | 4時間足 | 買い | 21 | 18回 | 72.22% | +1,150.3pips | 28.3本 | | 4時間足 | 売り | 26 | 23回 | 86.96% | +1,396.6pips | 28.2本 | | 4時間足 | 売り | 9 | 53回 | 52.83% | −441.0pips | 12.6本 | **勝率は52.83%から86.96%です。36通り（3時間足×2方向×6期間）すべてで50%を超えました。** それでも年間損益はマイナスの行が並びます。 15分足の買い・期間9は勝率66.13%で+367.4pipsですが、同じ設定の売り側は勝率63.08%で−1,194.3pipsです。**買いと売りで1,561.7pipsの差が出ています。** 15分足の終値で見ると米ドル／円は2024年に1,633.2pips上げ、2025年は56.8pips下げました。買い側と売り側は同じ設定でも別の相場を見ています。 初期設定に近い「期間9・水準80・決済0」の順位も出しておきます。1時間足の2025年の買いは、207通りを年間損益順に並べて179位（176回・勝率64.20%・−652.9pips）でした。2024年の売りは185位（200回・勝率55.50%・−1,255.6pips）です。**この読み方はどちらの年でも中位より下にいます。** ## 3本そろいは2025年だけ強かった 27通り（3時間足×3組×3水準）を両年で並べます。 | 時間足 | 3本の期間 | 水準 | 2025年 回数 | 2025年 勝率 | 2025年 損益 | 2024年 回数 | 2024年 勝率 | 2024年 損益 | |---|---|---|---|---|---|---|---|---| | 15分足 | 9/26/52 | 70 | 123回 | 61.79% | +865.6pips | 121回 | 66.94% | +1,638.0pips | | 15分足 | 9/14/21 | 80 | 244回 | 64.34% | +2,788.6pips | 216回 | 57.87% | −2,607.5pips | | 1時間足 | 9/26/52 | 70 | 35回 | 77.14% | +3,340.0pips | 22回 | 59.09% | +215.5pips | | 1時間足 | 9/14/21 | 80 | 72回 | 66.67% | +1,499.1pips | 62回 | 66.13% | −1,414.5pips | | 1時間足 | 14/26/52 | 70 | 39回 | 71.79% | +2,476.5pips | 20回 | 60.00% | −1,612.6pips | | 4時間足 | 9/14/21 | 70 | 34回 | 79.41% | +3,759.2pips | 13回 | 30.77% | −3,108.0pips | | 4時間足 | 9/14/21 | 80 | 19回 | 78.95% | +3,282.0pips | 9回 | 22.22% | −3,224.6pips | | 4時間足 | 9/26/52 | 50 | 15回 | 73.33% | +949.1pips | 10回 | 40.00% | −3,103.7pips | | 4時間足 | 14/26/52 | 80 | 2回 | 50.00% | −843.2pips | 7回 | 71.43% | +788.6pips | **27通りのうち、両年とも黒字だったのは4通りです。** 4時間足は9通り全部が片方の年で赤字でした。 4時間足の9/14/21・水準70が分かりやすい例です。2025年は34回・勝率79.41%・+3,759.2pipsで、この年の1位です。同じ設定の2024年は13回・勝率30.77%・−3,108.0pipsで、207通り中205位でした。**取引回数が年間13回から34回しかない設定なので、勝率も損益も年ごとに大きく振れます。** ## 前年の1位を翌年に当てる 片方の年で年間損益が1位だった設定を、相手の年に当てます。207通りの中での順位で示します。 | 時間足 | 選んだ年 | 設定 | 選んだ年の損益 | 相手の年の損益 | 相手の年の順位 | 相手の年の中央値 | |---|---|---|---|---|---|---| | 15分足 | 2024年 | 水準跨ぎ・期間36・水準−40 | +3,805.8pips | +996.5pips | 24位 | +89.1pips | | 15分足 | 2025年 | 3本そろい・9/14/21・水準80 | +2,788.6pips | −2,607.5pips | 205位 | −81.2pips | | 1時間足 | 2024年 | 水準跨ぎ・期間21・水準−40 | +4,028.1pips | +659.6pips | 40位 | +77.7pips | | 1時間足 | 2025年 | 3本そろい・9/26/52・水準70 | +3,340.0pips | +215.5pips | 105位 | +226.3pips | | 4時間足 | 2024年 | ±水準の抜け・期間21・水準80 | +3,652.5pips | −3,086.7pips | 206位 | −265.1pips | | 4時間足 | 2025年 | 3本そろい・9/14/21・水準70 | +3,759.2pips | −3,108.0pips | 205位 | +95.1pips | **6件のうち3件が、相手の年で205位から206位に落ちています。** 中央値を上回ったのは15分足と1時間足の2024年側の2件だけでした。 上位に残った2件はどちらも「水準跨ぎ・水準−40」です。−40は教科書の±80ではありません。2024年に1位だった設定が、翌年も中央値より上にいた唯一の読み方が、定番から最も遠い水準だったという結果でした。 ## 関連記事 - [RSI「30以下で買い」は本当に勝てる？330通り検証した結果](/ja/blog/rsi-settings)：同じ「行き過ぎで逆張り」を別の指標で測っています - [CCI「±100で逆張り」は勝てる？2,448通り検証した結果](/ja/blog/cci-settings)：この記事と同じ5つの読み方のうち4つを、CCIで比べています - [サイコロジカルラインの75%逆張りは勝てる？5,760通りで検証](/ja/blog/psychological-line-settings)：打ち込んだ水準が実際には成立しない例です - [ダマシ回避フィルターは効くのか？1,596通り検証した結果](/ja/blog/false-signal-filters)：公開済みの設定へフィルターを横断適用した検証です ## 補足 - 対象は米ドル／円のみです。2024年と2025年の2年分で、時間足は15分足・1時間足・4時間足です - 約定は終値、スプレッドは0.3pips固定、ロットは0.1、スリッページはありません - 3本そろいは4時間足で年間2回から45回しか売買しません。回数の少ない行の勝率と損益は大きく振れます - 対照群に使った変化率は1つだけです。RCIが捨てている情報を網羅したものではありません --- ### RSIとボリンジャーバンドの組み合わせを720通り検証 URL: https://formiq.jp/ja/blog/rsi-bollinger-combo Language: ja Published: 2026-09-10 Updated: 2026-09-10 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: RSI, ボリンジャーバンド, 組み合わせ, 逆張り, フィルター, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/rsi-bollinger-combo Evidence JSON: https://formiq.jp/ja/evidence/rsi-bollinger-combo/data.json Also available in: en — https://formiq.jp/blog/rsi-bollinger-combo 「RSIが30以下で、価格がボリンジャーバンドの下限に触れたら買い」。指標を2つ重ねるやり方の中でいちばんよく見る形です。 ただ、この操作が何をしているかははっきりしています。**組み合わせは新しい売買を作りません。RSIが30以下の足のうち、バンドに触れていないものを捨てているだけです。** 1時間足の2025年で、RSIが30を下回った足は356本ありました。バンドの下限（2σ）にも触れていたのは216本、60.67%です。**残りの40%を捨てる操作を「絞り込み」と呼んでいます。** そして、足を捨てれば勝率は動きます。母集団が小さくなるほど、その平均から離れやすくなるからです。だから「組み合わせたら勝率が上がった」が意味を持つのは、**同じ本数をランダムに捨てたときより上がったとき**だけです。 そこを測りました。**108通りのうち、ランダムに間引いた場合の95%の範囲から外れたのは21通り。うち19通りは下側でした。** ## RSIとボリンジャーバンドの設定と読み方 RSIは0から100の間を動くオシレーターで、下の別枠に表示されます。ボリンジャーバンドは移動平均の上下に標準偏差の倍数だけ離れた線を引くもので、価格チャートに重なります。 | 条件 | 数値の項目 | 初期設定 | 成立の中身 | |---|---|---|---| | RSI | 期間 | 14 | 何本分の値動きから計算するか | | RSI | 水準・比較 | 30・以下 | 30を下回っていれば買い。70以上なら売り | | ボリンジャーバンド | 期間 | 20 | 移動平均と標準偏差を何本で計算するか | | ボリンジャーバンド | 偏差 | 2 | 中心から標準偏差の何倍離すか | | ボリンジャーバンド | 読み方 | タッチ | 安値が下限に触れたら買い、高値が上限に触れたら売り | **どちらも「以下なら買い」という状態の条件で、その足で新しく起きた出来事ではありません。** RSIが30を下回っている間はずっと成立し続けます。 ### 2つを同時に使うとどうなるか 条件一覧で両方を有効にして、判定を「すべて満たす」にします。すると**両方が同じ向きを指した足**だけが残ります。RSIの買い（30以下）とバンドの買い（下限タッチ）はどちらも買いしか作らないので、組み合わせも買い専用になります。売りは水準70以上と上限タッチの組で別に作ります。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 2024年1月1日〜2025年12月31日 | | RSIの水準 | 買いは25・30・35以下、売りは75・70・65以上 | | バンドの偏差 | 1.5・2・2.5 | | 比べる相手 | 同じRSIの母集団から同じ本数をランダムに選んだ1,000回 | | 先の見方 | 成立した足の終値から1本後・5本後・10本後・20本後の終値 | | 決済 | 10本後の終値で固定 | | 条件の数 | RSI単体3＋バンド単体3＋組み合わせ9＝15通り（買い売り別） | | 回した数 | 15通り×買い売り×3時間足×4期間×手数料あり／なし＝720通り | | スプレッド | 0.3pips固定 | | ロット | 0.1 | | 約定 | 終値 | **決済は3つとも10本後で固定しました。** RSI単体・バンド単体・組み合わせの3つを比べるのが目的なので、決済を変えると入り方ではなく決済の比較になってしまいます。 **ランダムの作り方も書いておきます。** RSIが30以下の足が356本、そのうちバンドにも触れたのが216本なら、356本から216本を無作為に取り出す操作を1,000回繰り返し、その1,000個の勝率を並べて下から2.5%と97.5%の値を範囲としました。乱数は固定の種から作っているので、同じ数字が再現できます。 ## ランダムに間引いた場合と比べる 1時間足の買い側です。「母集団」はRSIが水準を下回った足、「組み合わせ」はそのうちバンドにも触れた足です。 | 年 | RSI水準 | 偏差 | 母集団 | 残した数 | 残した割合 | 母集団の割合 | 組み合わせの割合 | ランダムの範囲 | |---|---|---|---|---|---|---|---|---| | 2025 | 25以下 | 1.5σ | 112本 | 106本 | 94.64% | 61.61% | 59.43% | 59.43〜63.21% | | 2025 | 25以下 | 2σ | 112本 | 79本 | 70.54% | 61.61% | 59.49% | 55.70〜67.09% | | 2025 | 25以下 | 2.5σ | 112本 | 42本 | 37.50% | 61.61% | 59.52% | 50.00〜71.43% | | 2025 | 30以下 | 1.5σ | 356本 | 327本 | 91.85% | 55.90% | 55.66% | 54.43〜57.49% | | 2025 | 30以下 | 2σ | 356本 | 216本 | 60.67% | 55.90% | 56.48% | 51.85〜60.19% | | 2025 | 30以下 | 2.5σ | 356本 | 117本 | 32.87% | 55.90% | 54.70% | 48.72〜64.10% | | 2025 | 35以下 | 1.5σ | 772本 | 617本 | 79.92% | 54.53% | 53.32% | 52.84〜56.24% | | 2025 | 35以下 | 2σ | 772本 | 395本 | 51.17% | 54.53% | 52.41% | 51.14〜57.72% | | 2025 | 35以下 | 2.5σ | 772本 | 216本 | 27.98% | 54.53% | 55.09% | 49.07〜60.19% | | 2024 | 25以下 | 1.5σ | 132本 | 123本 | 93.18% | 50.76% | 47.97% | 48.78〜52.85% | | 2024 | 25以下 | 2σ | 132本 | 84本 | 63.64% | 50.76% | 40.48% | 44.05〜57.14% | | 2024 | 25以下 | 2.5σ | 132本 | 46本 | 34.85% | 50.76% | 41.30% | 39.13〜63.04% | | 2024 | 30以下 | 1.5σ | 309本 | 277本 | 89.64% | 50.16% | 48.74% | 48.38〜51.99% | | 2024 | 30以下 | 2σ | 309本 | 185本 | 59.87% | 50.16% | 45.41% | 45.95〜55.14% | | 2024 | 30以下 | 2.5σ | 309本 | 98本 | 31.72% | 50.16% | 52.04% | 42.86〜58.16% | | 2024 | 35以下 | 1.5σ | 617本 | 499本 | 80.88% | 48.78% | 49.90% | 46.89〜50.70% | | 2024 | 35以下 | 2σ | 617本 | 321本 | 52.03% | 48.78% | 47.35% | 45.17〜52.65% | | 2024 | 35以下 | 2.5σ | 617本 | 175本 | 28.36% | 48.78% | 51.43% | 42.86〜54.86% | **2025年の9通りはすべて範囲の中に収まりました。** 2024年は3通りが範囲の外に出ていますが、3つとも下側です（47.97%対48.78〜52.85%、40.48%対44.05〜57.14%、45.41%対45.95〜55.14%）。 範囲の広さに注目してください。残した割合が94.64%のときの範囲は59.43〜63.21%の3.8ポイント幅ですが、37.50%まで絞ると50.00〜71.43%の21.4ポイント幅になります。**絞るほど、勝率は「動いて当たり前」になります。** 3つの時間足と買い売りを全部合わせると108通りです。**範囲の外に出たのは21通りで、19通りが下、上に出たのは2通りだけでした。** さらに、同じ時間足・同じ向き・同じ設定で、2024年と2025年の両方が同じ側に外れたのは54通り中4通りです。4つとも1時間足の売り側で、4つとも下側でした。 ## バンドはどんな足を選んでいるのか 下振れするなら、バンドの条件は何か特定の性質を持つ足を選んでいるはずです。RSIが30以下（売りは70以上）の足を、バンドに触れたかどうかで2つに分けました。 | 時間足 | 年 | 向き | 触れた本数 | 直前5本の値動き | 足の大きさ | RSIの平均 | 触れなかった本数 | 直前5本の値動き | 足の大きさ | RSIの平均 | |---|---|---|---|---|---|---|---|---|---|---| | 15分足 | 2025 | 買い | 780本 | 3.07 ATR | 1.56 ATR | 25.04 | 569本 | 1.31 ATR | 0.93 ATR | 24.97 | | 15分足 | 2025 | 売り | 739本 | 2.93 ATR | 1.39 ATR | 75.12 | 713本 | 1.24 ATR | 0.86 ATR | 74.92 | | 1時間足 | 2025 | 買い | 216本 | 3.05 ATR | 1.46 ATR | 25.48 | 140本 | 1.42 ATR | 0.97 ATR | 26.25 | | 1時間足 | 2025 | 売り | 185本 | 3.13 ATR | 1.36 ATR | 76.16 | 219本 | 1.11 ATR | 0.84 ATR | 75.78 | | 1時間足 | 2024 | 買い | 185本 | 3.17 ATR | 1.80 ATR | 24.20 | 124本 | 1.22 ATR | 0.92 ATR | 25.25 | | 1時間足 | 2024 | 売り | 291本 | 2.87 ATR | 1.50 ATR | 76.41 | 307本 | 1.11 ATR | 0.82 ATR | 75.40 | | 4時間足 | 2025 | 売り | 60本 | 3.29 ATR | 1.26 ATR | 74.93 | 58本 | 1.26 ATR | 0.88 ATR | 75.61 | | 4時間足 | 2024 | 売り | 106本 | 2.70 ATR | 1.40 ATR | 77.88 | 74本 | 1.28 ATR | 0.88 ATR | 76.38 | 「直前5本の値動き」は、シグナルの向きと逆方向へ何ATR分進んでいたかです。買いなら、直前5本で何ATR下げてきたかを表します。 **触れた側は、触れなかった側より1.80〜2.82倍速く動いています。** 足そのものの大きさも1.43〜2.03倍です。一方で**RSIの平均値の差は2.1ポイント未満**でした。 つまりバンドの条件は「もっと売られている足」を選んでいるのではありません。RSIの水準は同じまま、**もっと速く落ちている足**を選んでいます。12通りのうち8通りで、触れた側のほうが10本後の割合が低くなりました。1時間足・2025年の売り側では32.97%対54.79%と、21.82ポイントの差がついています。 なお4時間足・2025年の買いは、触れなかった側が11本しかありません。この行だけ表から外しました。 ## 勝率は上がったのか 1時間足で、RSI単体と組み合わせの勝率を並べます。 | 向き | RSI水準 | 偏差 | 単体の取引 | 組み合わせの取引 | 残した割合 | 単体の勝率 | 組み合わせの勝率 | 差 | |---|---|---|---|---|---|---|---|---| | 買い | 25以下 | 1.5σ | 41回 | 39回 | 95.1% | 70.73% | 69.23% | -1.50 | | 買い | 25以下 | 2σ | 41回 | 36回 | 87.8% | 70.73% | 69.44% | -1.29 | | 買い | 25以下 | 2.5σ | 41回 | 23回 | 56.1% | 70.73% | 65.22% | -5.51 | | 買い | 30以下 | 1.5σ | 85回 | 82回 | 96.5% | 50.59% | 50.00% | -0.59 | | 買い | 30以下 | 2σ | 85回 | 74回 | 87.1% | 50.59% | 52.70% | +2.11 | | 買い | 30以下 | 2.5σ | 85回 | 56回 | 65.9% | 50.59% | 51.79% | +1.20 | | 買い | 35以下 | 1.5σ | 145回 | 130回 | 89.7% | 46.90% | 46.92% | +0.02 | | 買い | 35以下 | 2σ | 145回 | 113回 | 77.9% | 46.90% | 47.79% | +0.89 | | 買い | 35以下 | 2.5σ | 145回 | 87回 | 60.0% | 46.90% | 50.57% | +3.67 | | 売り | 75以上 | 1.5σ | 35回 | 32回 | 91.4% | 37.14% | 31.25% | -5.89 | | 売り | 75以上 | 2σ | 35回 | 29回 | 82.9% | 37.14% | 34.48% | -2.66 | | 売り | 75以上 | 2.5σ | 35回 | 19回 | 54.3% | 37.14% | 31.58% | -5.56 | | 売り | 70以上 | 1.5σ | 73回 | 63回 | 86.3% | 42.47% | 38.10% | -4.37 | | 売り | 70以上 | 2σ | 73回 | 53回 | 72.6% | 42.47% | 32.08% | -10.39 | | 売り | 70以上 | 2.5σ | 73回 | 40回 | 54.8% | 42.47% | 32.50% | -9.97 | | 売り | 65以上 | 1.5σ | 146回 | 127回 | 87.0% | 53.42% | 51.97% | -1.45 | | 売り | 65以上 | 2σ | 146回 | 106回 | 72.6% | 53.42% | 50.00% | -3.42 | | 売り | 65以上 | 2.5σ | 146回 | 78回 | 53.4% | 53.42% | 47.44% | -5.98 | **18通りのうち、勝率が上がったのは5通りです。** 5つとも買い側で、上げ幅は+0.02から+3.67ポイントでした。**売り側は9通りすべてで下がり、最大の下げは10.39ポイントです。** 残した割合と勝率の差の間に、はっきりした関係は出ていません。95.1%残しても−1.50ポイント、53.4%まで絞っても−5.98ポイントです。 ## 720通りを売買した結果 成立した足の終値で入り、10本後の終値で決済します。1時間足の買い側です。 | 条件 | 2025年 回数 | 2025年 勝率 | 2025年 年間損益 | 2024年 回数 | 2024年 勝率 | 2024年 年間損益 | |---|---|---|---|---|---|---| | RSI25以下だけ | 41回 | 70.73% | +695.9pips | 34回 | 52.94% | -113.3pips | | RSI30以下だけ | 85回 | 50.59% | +316.4pips | 73回 | 54.79% | +164.5pips | | RSI35以下だけ | 145回 | 46.90% | +210.1pips | 113回 | 48.67% | -529.9pips | | 1.5σタッチだけ | 284回 | 50.00% | -745.5pips | 260回 | 54.23% | -391.5pips | | 2σタッチだけ | 214回 | 49.53% | +194.1pips | 202回 | 52.97% | -194.4pips | | 2.5σタッチだけ | 160回 | 50.00% | +446.4pips | 151回 | 53.64% | +194.6pips | | RSI25＋1.5σ | 39回 | 69.23% | +629.3pips | 32回 | 50.00% | -381.1pips | | RSI25＋2σ | 36回 | 69.44% | +545.1pips | 29回 | 51.72% | -163.0pips | | RSI25＋2.5σ | 23回 | 65.22% | +254.7pips | 21回 | 42.86% | -356.3pips | | RSI30＋1.5σ | 82回 | 50.00% | +439.7pips | 71回 | 54.93% | +138.8pips | | RSI30＋2σ | 74回 | 52.70% | +385.6pips | 62回 | 51.61% | +106.5pips | | RSI30＋2.5σ | 56回 | 51.79% | +337.9pips | 49回 | 51.02% | +63.5pips | | RSI35＋1.5σ | 130回 | 46.92% | -101.6pips | 108回 | 52.78% | +208.4pips | | RSI35＋2σ | 113回 | 47.79% | +244.8pips | 92回 | 51.09% | +269.6pips | | RSI35＋2.5σ | 87回 | 50.57% | +335.6pips | 80回 | 51.25% | +287.4pips | **2024年と2025年の両方で黒字になったのは15通り中7通りです。** 内訳はRSI30以下だけ、2.5σだけ、RSI30との組み合わせ3通り、RSI35との組み合わせ2通りでした。 他の時間足と向きは次のとおりです。 | 時間足 | 向き | 15通り中の両年黒字 | |---|---|---| | 15分足 | 買い | 0 | | 1時間足 | 買い | 7 | | 4時間足 | 買い | 0 | | 15分足 | 売り | 0 | | 1時間足 | 売り | 0 | | 4時間足 | 売り | 2 | **両年黒字が出たのは、1時間足の買いと4時間足の売りだけです。** 残り4つの組み合わせでは、単体も組み合わせも1つも残りませんでした。 ## 組み合わせが単体を救った2例 7通りのうち5通りは、組み合わせる前のRSI単体でもすでに両年黒字です。**単体では片方の年が赤字だったのに、組み合わせて両年黒字になったのは2通りだけでした。** | 条件 | RSI単体の2024年 | 組み合わせの2024年 | RSI単体の2025年 | 組み合わせの2025年 | |---|---|---|---|---| | 1時間足・RSI35＋2σ | -529.9pips | +269.6pips | +210.1pips | +244.8pips | | 1時間足・RSI35＋2.5σ | -529.9pips | +287.4pips | +210.1pips | +335.6pips | 2024年の−529.9pipsが+269.6pipsと+287.4pipsになっています。ここだけ見れば組み合わせが効いたように見えます。 ただし、この2通りはどちらもランダムの範囲の中に収まっています。2024年のRSI35以下の母集団は48.78%で、2σに絞った321本は47.35%、ランダムの範囲は45.17〜52.65%でした。**同じ本数をランダムに捨てても届く場所です。** ## 逆張りの売りは組み合わせても直らない 売り側は、単体でも組み合わせでもほぼ全滅でした。1時間足の2025年は15通りすべてが赤字で、最も浅い赤字がRSI65以上の−324.6pipsです。 このシリーズでは、逆張りの売りが上げ相場で壊れることを繰り返し測ってきました。2024年のドル円は140.87円から157.20円へ上げた年です。ただし2025年は157.23円で始まって156.67円で終わった年で、**それでも売り側は1時間足で15通りすべてが赤字**でした。相場の向きだけでは説明がつきません。 そして組み合わせは、この赤字を浅くしていません。1時間足・2025年のRSI70以上は単体で−1,040.3pips、2σを足すと−1,158.0pipsです。勝率も42.47%から32.08%へ10.39ポイント下がりました。 ## コスト 支払ったスプレッドは、取引回数×0.3pipsと一致しました。1時間足の18通りすべてで計算どおりです。 | 向き | 条件 | 取引回数 | 支払ったスプレッド | 回数×0.3pips | |---|---|---|---|---| | 買い | RSI30＋1.5σ | 82回 | 24.5pips | 24.6pips | | 買い | RSI30＋2σ | 74回 | 22.1pips | 22.2pips | | 買い | RSI35＋1.5σ | 130回 | 38.9pips | 39.0pips | | 売り | RSI70＋2σ | 53回 | 15.8pips | 15.9pips | | 売り | RSI65＋1.5σ | 127回 | 38.0pips | 38.1pips | 組み合わせると取引回数が減るので、コストも減ります。1時間足の買いでRSI35以下は145回で43.5pips、2.5σを足すと87回で26.1pipsです。**17.4pipsのコストは浮きますが、その代わり58回の売買を捨てています。** ## 関連記事 - [RSI「30以下で買い」は本当に勝てる？](/ja/blog/rsi-settings)：この記事で母集団に使ったRSI単体を、330通りで検証しています - [ボリンジャーバンド「2σで逆張り」は勝てる？](/ja/blog/bollinger-band-settings)：バンド単体の逆張りと順張りを315通りで比較しています - [ローソク足パターン17種の勝率](/ja/blog/candlestick-patterns-tested)：同じく「対照群を置かないと差が見えない」型の検証です - [包み足の勝率とだまし](/ja/blog/engulfing-settings)：絞り込みを足したときに何が起きるかを、別の形で測っています - [ダマシ回避フィルターは効くのか？1,596通り検証した結果](/ja/blog/false-signal-filters)：ここで作ったランダムの範囲を、公開済み38通りの設定へ広げています ## 補足 対象は米ドル／円の2024年1月1日から2025年12月31日です。終値約定、スプレッド0.3pips固定、0.1ロットで計算しました。RSIは期間14、ボリンジャーバンドは期間20です。ランダムの範囲は、母集団から組み合わせと同じ本数を無作為に取り出す操作を1,000回繰り返し、勝率を並べて下から2.5%と97.5%の位置を取ったものです。乱数は固定の種から作っているので同じ値が再現できます。母集団と組み合わせの本数が同じになる場合は取り出す余地がないため、母集団そのものの値を範囲としています。1本後から20本後までを見るため、年の終わりからその本数ぶんは対象から外しています。 --- ### ドル円の設定は他の通貨ペアで通用する？3,024通り検証 URL: https://formiq.jp/ja/blog/settings-across-pairs Language: ja Published: 2026-09-10 Updated: 2026-09-10 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: 通貨ペア, ユーロドル, ポンドドル, ユーロ円, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/settings-across-pairs Evidence JSON: https://formiq.jp/ja/evidence/settings-across-pairs/data.json Also available in: en — https://formiq.jp/blog/settings-across-pairs このサイトの指標検証記事には、記事ごとに「この設定で走らせる」ボタンが付いています。**検証した時点でそのボタンは44件あり、どれも米ドル／円で選んだ設定です。** だから読者が次に聞くこと、「この設定はユーロドルでも使えるのか」に、まだ誰も答えていませんでした。 （別の通貨ペアを測った記事も2本あります。通貨強弱の記事は28ペアを、RSIの買われすぎ・売られすぎの継続時間の記事はユーロドルと並べて測っています。ただしこの2本はボタンに売買ルールを持たないので、下の42通りには入っていません。） 答えを出すには、設定を選び直さないことが大事です。各記事には「この設定で走らせる」ボタンが付いていて、記事が公開した設定がそのまま入っています。**その42通りを、通貨ペアと期間だけ差し替えて走らせました。** ユーロ／米ドル、英ポンド／米ドル、ユーロ／円を加えた4ペア、2023年から2025年の3年、3つの時間足です。 結果を先に3つ書きます。 **1つ目。順位はまったく引き継がれません。** 42設定をドル円の年間損益順に並べた順位と、他ペアの順位の相関は27区分で−0.515から+0.47、中央値は−0.011でした。プラスだったのは27区分中13です。 **2つ目。「両方のペアで黒字」の数は、2つのペアが無関係な場合の予測とほとんど同じです。** たとえばドル円で36通り、ユーロドルで5通りが黒字なら、無関係なら重なりは36÷42×5＝4.3通り。実測は3通りでした。27区分で差は−4から+6、中央値は−0.2です。 **3つ目。4ペア全部と3年全部で黒字だった設定は、3つの時間足すべてで0本でした。** ## なぜこの検証が必要だったのか 指標検証記事はどれも同じ形をしています。ある指標の設定を何十通りも回し、2年分で黒字だったものを数え、前年の1位を翌年に当てて確かめる。年をまたぐ検定は毎回やっています。 **やっていなかったのは、通貨ペアをまたぐ検定です。** そして下の表を見ると、その省略には代償がありました。 1時間足で年間損益がプラスで終わった設定の数を、ペアと年で並べます。42通り中の数です。 | 年 | ドル円 | ユーロドル | ポンドドル | ユーロ円 | |---|---|---|---|---| | 2023 | 22 | 27 | 16 | 31 | | 2024 | 36 | 5 | 6 | 8 | | 2025 | 30 | 24 | 19 | 14 | **2024年のドル円は42通り中36通りが黒字です。同じ42通りが、同じ年のユーロドルでは5通りしか黒字になりません。** 年間損益の中央値でも同じことが出ます。ペアごとに値幅が違うので、各ペアの1本あたり平均値幅（14本ATR）で割った数字で並べます。単位は「平均的な1本の値幅の何本ぶん稼いだか」です。 | 年 | ドル円 | ユーロドル | ポンドドル | ユーロ円 | |---|---|---|---|---| | 2023 | +3.30 | +6.86 | −10.88 | +29.88 | | 2024 | +47.17 | −51.94 | −59.20 | −25.48 | | 2025 | +24.45 | +8.05 | −6.33 | −28.46 | **3年すべてで中央値がプラスだったのは、ドル円だけです。** ポンドドルは3年すべてマイナス、ユーロドルは2年、ユーロ円は1年でした。 42通りの設定は、この1つのペアの、この3年の窓の中で選ばれていたことになります。設定が悪いという話ではありません。**探した場所が、たまたま「中央値でも稼げた」場所だったということです。** ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円・ユーロ／米ドル・英ポンド／米ドル・ユーロ／円 | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 2023年1月1日〜2025年12月31日（年ごとに分割） | | 設定 | 公開済みの記事に載せた42通り | | 差し替えたもの | 通貨ペアと期間だけ | | そのままのもの | 売買ルール・ロット0.1・終値約定・スプレッド0.3pips固定 | | pipsの数え方 | 各ペア自身のpip（円ペアは0.01、ドルペアは0.0001） | | 比べ方 | 各ペアの14本ATRの平均で割った数字を併記 | | 回した数 | 42通り×4ペア×3時間足×3年×スプレッド2種＝3,024通り | **設定はこちらで選んでいません。** 各記事の「この設定で走らせる」ボタンが持っている設定をそのまま使いました。ボタンは44件ありますが、ボリンジャーバンドの3記事が同じ設定（期間20・2σのブレイク）なので1つにまとめ、42通りとしています。 **ペアの選び方には理由があります。** ユーロドルとポンドドルはドル円とドルを共有し、円を持ちません。ユーロ円は円を共有し、ドルを持ちません。もしドル円の設定がユーロ円へは通用してユーロドルへは通用しないなら、効いているのは円のほうだと言えます。逆なら、ドルのほうです。ドル以外を含むペアが1つも無いと、この2つを「ペアによって難易度が違う」から切り離せません。 **pipsをそのまま足し合わせて比べていない**ことも書いておきます。1時間足1本の平均値幅は、ドル円24.85pips、ユーロドル15.27pips、ポンドドル17.12pips、ユーロ円24.72pipsです。同じ「1,000pips」がペアによって別の意味になるので、年間損益はこの値幅で割った数字も併記しました。 ## ドル円の順位は他のペアの順位を予測するか 42設定をドル円の年間損益順に並べ、同じ42設定を他ペアの年間損益順に並べて、2つの並びがどれくらい似ているかを測ります。+1なら同じ順番、0なら無関係、−1なら逆順です。同順位は平均して扱いました。 | 時間足 | 年 | ユーロドル | ポンドドル | ユーロ円 | |---|---|---|---|---| | 15分足 | 2023 | −0.061 | +0.106 | +0.122 | | 15分足 | 2024 | +0.077 | −0.162 | −0.011 | | 15分足 | 2025 | +0.314 | +0.346 | +0.369 | | 1時間足 | 2023 | −0.027 | −0.185 | −0.126 | | 1時間足 | 2024 | −0.270 | −0.072 | −0.218 | | 1時間足 | 2025 | −0.198 | −0.266 | −0.515 | | 4時間足 | 2023 | +0.363 | +0.238 | −0.022 | | 4時間足 | 2024 | +0.180 | −0.106 | +0.272 | | 4時間足 | 2025 | +0.382 | +0.470 | +0.117 | **27区分の範囲は−0.515から+0.47、中央値は−0.011です。** プラスだったのは13区分でした。 時間足で符号が偏っているのも見ておいてください。**1時間足は9区分すべてがマイナスで、4時間足は9区分中7がプラスです。** 15分足は3つがマイナスです。もし「ドル円で良かった設定は他ペアでも良い」という関係が本当にあるなら、時間足を変えただけで符号が反転することはありません。 ### 無関係だとしたらいくつ重なるか 相関の数字だけでは、それが大きいのか小さいのか判断しにくいです。もっと読みやすい形にします。 **ドル円で黒字だった設定が42通り中36通り、ユーロドルで5通りだったとします。2つのペアが完全に無関係なら、両方で黒字になる設定は36÷42×5＝4.3通りのはずです。** 実測は3通りでした。 これを27区分で並べます。左が実測、右が無関係な場合の予測です。 | 時間足 | 年 | ユーロドル | ポンドドル | ユーロ円 | |---|---|---|---|---| | 15分足 | 2023 | 4／5.7 | 2／2.5 | 10／7.9 | | 15分足 | 2024 | 2／3.2 | 4／5.0 | 4／3.6 | | 15分足 | 2025 | 8／6.3 | 6／4.6 | 9／7.4 | | 1時間足 | 2023 | 12／14.1 | 7／8.4 | 16／16.2 | | 1時間足 | 2024 | 3／4.3 | 3／5.1 | 4／6.9 | | 1時間足 | 2025 | 18／17.1 | 12／13.6 | 6／10.0 | | 4時間足 | 2023 | 6／5.9 | 8／5.4 | 6／6.3 | | 4時間足 | 2024 | 6／5.3 | 4／5.3 | 20／19.8 | | 4時間足 | 2025 | 16／14.9 | 22／16.0 | 8／5.7 | **実測と予測の差は−4から+6で、中央値は−0.2でした。** つまり「ドル円で黒字だった」という情報は、そのペアで黒字になる確率をほとんど動かしません。 ### 円を共有しても近づかない ユーロ円を入れた理由に戻ります。円を共有していれば近づくのかを見ます。 | 時間足 | ユーロ円との相関の中央値 | ユーロドルとポンドドルの中央値 | |---|---|---| | 15分足 | +0.122 | +0.0915 | | 1時間足 | −0.218 | −0.1915 | | 4時間足 | +0.117 | +0.3005 | **円を共有しているほうが高い、という結果にはなりませんでした。** 15分足はわずかに高く、1時間足はわずかに低く、4時間足では0.117対0.3005で明確に低いです。ユーロ円側は3つの値の中央、ドル側は6つの値の中央なので、桁を4つまで出しています。ドルを共有しても円を共有しても、順位は引き継がれていません。 ## ドル円の1位を他のペアへ持ち込む このシリーズが年をまたいで毎回やっている検定を、ペアをまたいでやります。**その時間足・その年にドル円で年間損益1位だった設定を選び、他の3ペアに当てます。** 相手のペアでの順位も出します。 | 時間足 | 年 | ドル円で1位だった設定 | ドル円 | ユーロドル | ポンドドル | ユーロ円 | |---|---|---|---|---|---|---| | 15分足 | 2023 | パラボリックSAR | +1,315.1 | −505.1（24位） | −1,962.0（34位） | −1,124.4（36位） | | 15分足 | 2024 | 一目均衡表 | +2,261.3 | −1,553.8（34位） | −720.8（23位） | −513.7（14位） | | 15分足 | 2025 | 一目均衡表 | +2,723.2 | −1,589.4（34位） | −1,306.5（22位） | +739.8（7位） | | 1時間足 | 2023 | スーパートレンド | +1,539.6 | +1,414.8（2位） | +177.9（13位） | +1,346.0（14位） | | 1時間足 | 2024 | MACD | +3,152.1 | −389.3（16位） | −163.2（8位） | −373.5（17位） | | 1時間足 | 2025 | ウェーブトレンド | +3,283.4 | −612.5（34位） | +66.2（18位） | −1,421.3（33位） | | 4時間足 | 2023 | 明けの明星・宵の明星 | +2,449.6 | +534.7（5位） | +559.9（6位） | +1,671.7（4位） | | 4時間足 | 2024 | ピンバー | +3,498.7 | +386.9（1位） | −275.3（13位） | +804.1（13位） | | 4時間足 | 2025 | ストキャスティクス | +2,959.6 | +211.8（22位） | +507.4（17位） | −1,082.0（30位） | **27回の持ち込みのうち、黒字で終わったのは12回です。** 順位は1位から36位まで散らばりました。相手のペアの中央値を上回ったのは16回、上位10位に入ったのは7回です。 時間足で結果が分かれます。**15分足は9回中1回しか黒字になりませんでした。4時間足は9回中7回です。** 1時間足は4回でした。 いちばん良く伝わる2行を並べます。2023年の4時間足で1位だった明けの明星・宵の明星は、ユーロドル・ポンドドル・ユーロ円のすべてで黒字になり、順位はユーロドル5位・ポンドドル6位・ユーロ円4位でした。2025年の1時間足で1位だったウェーブトレンドは、ドル円で+3,283.4pipsを出しながら、ユーロドルで34位、ユーロ円で33位です。**同じ手続きで選んだ1位が、片方は移り、片方は移りません。** ## 4ペア全部で黒字だった設定はあるか いちばん厳しい条件で数えます。 | 時間足 | 年 | 4ペア全部で黒字 | ドル円だけ | |---|---|---|---| | 15分足 | 2023 | 0 | 15 | | 15分足 | 2024 | 2 | 19 | | 15分足 | 2025 | 3 | 24 | | 1時間足 | 2023 | 3 | 22 | | 1時間足 | 2024 | 1 | 36 | | 1時間足 | 2025 | 1 | 30 | | 4時間足 | 2023 | 2 | 19 | | 4時間足 | 2024 | 1 | 32 | | 4時間足 | 2025 | 5 | 24 | **42通り中、4ペア全部で黒字だったのは年によって0通りから5通りです。** そして4ペア全部と3年全部を同時に満たした設定は、**3つの時間足すべてで0通り**でした。 1時間足の2024年が分かりやすいです。ドル円だけなら36通りが黒字ですが、4ペア全部で黒字だったのは1通りだけです。 ## スプレッドで消える分は時間足で決まる 年間損益がマイナスなのは、設定が外れているからかコストを払いすぎているからか。スプレッドを0にして走らせ直し、切り分けます。左が0のとき、右が0.3pipsのときの黒字設定数です。 | 時間足 | 年 | ドル円 | ユーロドル | ポンドドル | ユーロ円 | |---|---|---|---|---|---| | 15分足 | 2023 | 23→15 | 21→16 | 13→7 | 27→22 | | 15分足 | 2024 | 23→19 | 14→7 | 19→11 | 12→8 | | 15分足 | 2025 | 31→24 | 17→11 | 9→8 | 14→13 | | 1時間足 | 2023 | 25→22 | 29→27 | 19→16 | 33→31 | | 1時間足 | 2024 | 36→36 | 10→5 | 7→6 | 8→8 | | 1時間足 | 2025 | 31→30 | 29→24 | 21→19 | 14→14 | | 4時間足 | 2023 | 20→19 | 13→13 | 13→12 | 14→14 | | 4時間足 | 2024 | 32→32 | 7→7 | 7→7 | 26→26 | | 4時間足 | 2025 | 24→24 | 27→26 | 29→28 | 11→10 | **15分足では0.3pipsのスプレッドが12区分すべてで黒字設定を減らし、最大で8通り減らしました**（2023年のドル円、23通りから15通り）。**4時間足では12区分のうち7区分で1通りも減りません。** これはこのシリーズが毎回確認している「コスト＝取引回数×スプレッド」と同じ形です。時間足が短いほど回数が多く、スプレッドで消える分が大きくなります。**通貨ペアを変えたときの差は、スプレッドでは説明できません。** 2024年の1時間足でユーロドルが5通りしか黒字にならないのは、スプレッドを0にしても10通りだからです。 ## 関連記事 - [ダマシ回避フィルターは効くのか？1,596通り検証した結果](/ja/blog/false-signal-filters)：同じ42設定に、通貨ペアではなくフィルターを横断適用しています - [移動平均線のゴールデンクロスは勝てる？280通り検証した結果](/ja/blog/moving-average-cross-settings)：この検証に入っている設定のひとつです - [一目均衡表「三役好転で買い」は勝てる？280通り検証した結果](/ja/blog/ichimoku-settings)：15分足の2024年と2025年でドル円1位だった設定の記事です - [FXバックテストのやり方：プログラミング不要で手法を検証](/ja/blog/backtest-without-coding)：同じ設定を別の通貨ペアで走らせ直す手順です - [水平線トレードは本当に勝てる？5,184通り検証](/ja/blog/support-resistance-settings)：この検証のあとに書いた記事で、エンジンの水平線の条件が一度も売買していなかったことが分かりました ## 補足 - 対象は4通貨ペア・3年間です。時間足は15分足・1時間足・4時間足です - 約定は終値、スプレッドは全ペア共通で0.3pips固定、ロットは0.1、スリッページはありません。実際のスプレッドはペアごとに違うので、この数字はペア間の比較のためのものです - 設定は各記事の代表1通りだけで、その記事が回したパラメータ全体ではありません - ドル以外の通貨同士の組み合わせはユーロ／円1本だけです。円とドルのどちらが効くかについて、この記事が言えるのは「どちらも引き継がれていない」までです --- ### 水平線トレードは本当に勝てる？5,184通り検証 URL: https://formiq.jp/ja/blog/support-resistance-settings Language: ja Published: 2026-09-10 Updated: 2026-09-10 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: 水平線, サポートライン, レジスタンスライン, プライスアクション, ブレイク, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/support-resistance-settings Evidence JSON: https://formiq.jp/ja/evidence/support-resistance-settings/data.json Also available in: en — https://formiq.jp/blog/support-resistance-settings 「何度も反発している価格に線を引く」。オシレーターが1つも説明しない、裁量トレーダーが最初に描くものです。効くとされる理由もはっきりしています。**その価格で以前も止まったから。** 検証できる形にすると、確かめるべきことが2つ出てきます。1つは用量です。「3回止まった線だから信頼できる」なら、5回止まった線は2回の線より止まるはずです。もう1つは対照群で、**誰も止まらなかった価格に同じ線を引いたらどうなるか**です。 先に結果を3つ書きます。 **1つ目。タッチ回数は効きませんでした。** 線で反発した足を、その線が何回持ちこたえていたかで分けると、15分足の買いは3回で50.13%、4回で49.24%、5回以上で50.10%です。 **2つ目。価格が線に来ている足は、全体の41.40%から65.75%あります。** おおむね2本に1本で、特別な状態ではありません。 **3つ目。線の定義を決める4つの設定のうち、年間損益をいちばん動かさなかったのがタッチ回数でした。** いちばん動かしたのは許容幅で、中央値の振れ幅は1,000.9pips対357.9pipsです。 そしてこの検証を始めたときに分かったことがもう1つあります。**エンジンの水平線の条件は、一度も売買していませんでした。** 2つの理由があり、どちらか一方だけでも売買は0件になります。 ## エンジンで水平線が一度も売買していなかった この記事のために最初のスイープを回したところ、2,592通りすべてが取引0件でした。原因は2つです。 **1つ目は、ATRが渡っていなかったことです。** 水平線の許容幅は「ATRの何倍」で指定します。ゴールドとユーロドルで「同じ価格」の意味が違うので、pipsではなくATR倍率にしてあります。ところがエンジンには「どの条件がATRを必要とするか」の一覧があり、水平線がそこに入っていませんでした。結果、許容幅の計算に渡るATRが空になり、幅が0になります。幅が0だと、線を作る段階でスイングがすべて捨てられ、売買判定も`幅が0より大きい`という条件の後ろにあるので通りません。 **2つ目は、抜けの判定が成立しない形で書かれていたことです。** レジスタンスは定義上「いまの終値より上でいちばん近い線」です。その線に対して「いまの終値がレジスタンスより上か」を聞いていました。**この比較は絶対に真になりません。** サポート側も同じで、「いまの終値がサポートより下か」を聞いていました。抜けは、1本前の足が下にいた線を越えたかどうかで判定するのが正しいので、そう直しました。 どちらも「相場の性質」に見える形で壊れます。条件は画面に出ていて、バックテストは走り、レポートは「取引0件」と表示します。このシリーズが同じ形の不具合を見つけたのは、時間帯フィルター・フェアバリューギャップの描画・GMMAの決済に続いて4件目です。 以下の数字はすべて修正後のものです。 ## 水平線とは何を指しているのか エンジンが線を引く手順は4段です。 1. **スイングを見つける。** 左右それぞれ指定本数のあいだ、その足の高値がいちばん高い（安値がいちばん安い）ならスイングです。左右が必要なので、スイングは**その足から指定本数あとに確定します** 2. **近い価格をまとめる。** 2つのスイングがATR倍率の許容幅より近ければ、同じ線として扱います。線の価格は、まとまったスイングの平均へ寄せます 3. **タッチ回数を数える。** まとまった回数がその線のタッチ回数です 4. **指定回数に達した線だけを出す。** いまの終値より下でいちばん近いものをサポート、上でいちばん近いものをレジスタンスとして返します **確定を待つのが大事です。** スイングは右側の足がそろって初めてスイングなので、確定前の足で線を読むと、テスターがまだ見ていない安値からサポートを引くことになります。バックテストでいちばん都合のいい不具合がこれです。 遡る本数の外に出た線は捨てられます。これは速さのためだけではありません。**「この線は3回持ちこたえた」の3回は、遡る本数の中で数えた3回です。** 1万本前に一度かすっただけの価格は、もう数に入りません。 ## 水平線の設定と読み方 **この条件はチャートに線を描きません。** バックテストの中だけで計算されます。MT4のように画面に引いた線を見ながら判断する形にはなっていないので、以下は数字の上での話です。 | 項目 | 初期設定 | 中身 | |---|---|---| | 読み方 | 反発 | 反発（線で止まったら線の側へ）／抜け（線を越えたら越えた側へ） | | 最低タッチ回数 | 3 | 何回まとまった線から使うか。2が下限 | | 許容幅 | 0.5 | 2つのスイングを同じ線とみなす距離。ATRの倍率 | | スイング判定 | 3 | 左右それぞれ何本を見てスイングと決めるか | | 遡る本数 | 200 | 線を引ける過去の範囲 | ### 2つの読み方 **反発**は、足の安値がサポートの許容幅に入り、終値がサポートより上で終わったら買います。決済は1つ上のレジスタンスに届いたときです。売りはレジスタンスで鏡になります。 **抜け**は、終値が1本前のレジスタンスを越えたら買います。決済は反対側、つまりサポートを終値で下抜けたときです。 どちらも買いと売りの両方を作るので、エントリーと決済に同じ条件を使いました。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 2024年1月1日〜2025年12月31日。2025年は前半後半にも分割 | | 読み方 | 反発・抜け | | 遡る本数 | 100・200・400 | | スイング判定 | 2・3・5・8 | | 許容幅 | 0.25・0.5・1.0ATR | | 最低タッチ回数 | 2・3・4 | | 比べる相手 | 線から1.5・2・3ATR離した価格、および並べ替えた終値の系列 | | 回した数 | 2読み方×3×4×3×3＝216通り×3時間足×4期間×スプレッド2種＝5,184通り | | スプレッド | 0.3pips固定と0 | | ロット | 0.1 | | 約定 | 終値 | **スプレッド0でも回しています。** 反発の読み方は1時間足で年1,000回を超えるので、負けたときに「線が効かない」のか「往復のコストで負けた」のかを分けないと読めません。 ## 価格が線に来ている足は半分ある 線を評価する前に、線がどれくらいあって、価格がどれくらい線に来ているかを数えます。初期設定（遡る200本・スイング3・許容幅0.5ATR・3タッチ）です。 | 時間足 | 年 | 足の数 | サポートが引ける足 | 価格が線の幅に入った足 | タッチ回数の中央値 | 最大 | |---|---|---|---|---|---|---| | 15分足 | 2025 | 24,893本 | 81.28% | 59.90% | 4回 | 38回 | | 15分足 | 2024 | 24,999本 | 86.72% | 58.20% | 4回 | 24回 | | 1時間足 | 2025 | 6,216本 | 85.52% | 65.75% | 5回 | 23回 | | 1時間足 | 2024 | 6,250本 | 84.24% | 52.64% | 4回 | 15回 | | 4時間足 | 2025 | 1,600本 | 85.50% | 61.63% | 5回 | 15回 | | 4時間足 | 2024 | 1,616本 | 84.90% | 41.40% | 3回 | 8回 | **価格が線の許容幅に入っている足は41.40%から65.75%です。** 「水平線に来た」は、2本に1本で起きています。 タッチ回数の中央値は3回から5回で、最大は15分足の2025年で38回です。**線そのものは足りないどころか、たくさんあります。** ## 何度も反発した線は強いのか 用量を見ます。反発の判定が成立した足を、その線が**それまでに何回持ちこたえていたか**で分け、10本後の終値までの値動きを測ります。回数はエンジンが数えているものをそのまま使っているので、新しい定義は入っていません。2025年です。 | 時間足 | 向き | タッチ回数 | 成立した足 | 10本後の勝率 | 10本後の平均 | |---|---|---|---|---|---| | 15分足 | 買い | 3回 | 2,759本 | 50.13% | −0.16pips | | 15分足 | 買い | 4回 | 1,911本 | 49.24% | −0.55pips | | 15分足 | 買い | 5回以上 | 4,996本 | 50.10% | −0.71pips | | 15分足 | 売り | 3回 | 2,640本 | 50.23% | +1.19pips | | 15分足 | 売り | 4回 | 1,825本 | 50.25% | −0.89pips | | 15分足 | 売り | 5回以上 | 5,390本 | 50.20% | +1.17pips | | 1時間足 | 買い | 3回 | 640本 | 50.63% | −0.01pips | | 1時間足 | 買い | 4回 | 402本 | 54.48% | +3.18pips | | 1時間足 | 買い | 5回以上 | 1,809本 | 51.96% | +0.72pips | | 1時間足 | 売り | 3回 | 711本 | 52.04% | +3.12pips | | 1時間足 | 売り | 4回 | 446本 | 49.33% | −1.22pips | | 1時間足 | 売り | 5回以上 | 1,607本 | 45.24% | −3.77pips | **15分足は6つの区分すべてが49.24%から50.25%の中に収まりました。** コイン投げと区別がつきません。1時間足の買いは3回50.63%・4回54.48%・5回以上51.96%で、順番になっていません。売りは52.04%・49.33%・45.24%で、**回数が増えるほど下がっています。** 「その価格で以前も止まったから止まる」が理由なら、止まった回数は効くはずです。**この検証では効きませんでした。** ## 並べ替えた価格でも同じ線が引ける 対照群を2つ置きます。 **1つ目は、線から離した価格です。** 同じ判定を、サポートから1.5・2・3ATR上の価格で走らせます。線とは重ならない距離なので、許容幅も重なりません。2025年です。 | 時間足 | 線 | 成立した足 | 10本後の勝率 | 終値が線の何ATR上だったか | |---|---|---|---|---| | 15分足 | 引いた線 | 9,666本 | 49.94% | 0.49 | | 15分足 | +1.5ATR | 2,748本 | 47.49% | 0.46 | | 15分足 | +2ATR | 1,695本 | 47.67% | 0.49 | | 1時間足 | 引いた線 | 2,851本 | 52.02% | 0.48 | | 1時間足 | +1.5ATR | 624本 | 50.48% | 0.46 | | 1時間足 | +2ATR | 415本 | 47.95% | 0.48 | | 4時間足 | 引いた線 | 679本 | 53.31% | 0.51 | | 4時間足 | +1.5ATR | 172本 | 54.07% | 0.52 | | 4時間足 | +2ATR | 122本 | 45.08% | 0.51 | 引いた線が離した価格を勝率で上回ったのは18通り中13通りでした。**成立した瞬間の距離はそろっています。**終値が線の何ATR上で終わったかは、2025年の9行で引いた線が0.48〜0.51、離した価格が0.46〜0.54です。 そろっていないのは回数です。**離した価格で成立した足の数は、引いた線の7.8%から41.6%しかありません。** サポートは定義上「いまの終値より下でいちばん近い線」なので、価格はそこへ何度も来ます。1.5ATR上に置いた線は、価格がそこまで上がった足でしか成立しません。**同じ足数で比べたわけではない**ので、これだけでは足りません。 **2つ目は、終値の変化を並べ替えた系列です。** 終値から終値への変化を1回ずつ残して順番だけ崩すので、上昇の勢いも変動の大きさも太い尾も相場のものです。**残らないのは、価格が反転した位置だけです。** 並べ替えた足は実体も影も持たないので、判定は両方の系列で終値だけを読む形にそろえました（エンジンの反発は安値を読みます。ここでの数字は取引した結果ではなく、系列同士の比較です）。 | 時間足 | 年 | 系列 | 成立した足 | 10本後の勝率 | 10本後の平均 | |---|---|---|---|---|---| | 15分足 | 2025 | 実際 | 5,568本 | 49.69% | −0.45pips | | 15分足 | 2025 | 並べ替え | 2,720本 | 50.04% | −1.83pips | | 1時間足 | 2025 | 実際 | 1,637本 | 51.13% | −0.51pips | | 1時間足 | 2025 | 並べ替え | 542本 | 47.23% | −0.15pips | | 4時間足 | 2025 | 実際 | 376本 | 55.59% | +7.53pips | | 4時間足 | 2025 | 並べ替え | 172本 | 52.91% | +3.58pips | | 15分足 | 2024 | 実際 | 5,553本 | 53.97% | +0.80pips | | 15分足 | 2024 | 並べ替え | 2,328本 | 49.44% | −1.00pips | | 1時間足 | 2024 | 実際 | 1,179本 | 59.20% | +5.17pips | | 1時間足 | 2024 | 並べ替え | 642本 | 52.80% | +1.24pips | | 4時間足 | 2024 | 実際 | 249本 | 64.26% | +26.49pips | | 4時間足 | 2024 | 並べ替え | 114本 | 36.84% | −24.74pips | **実際の価格が並べ替えた価格を勝率で上回ったのは6区分中5区分です。** 15分足の2025年だけが逆でした（49.69%対50.04%）。 つまり、**順番には何かあります。** ただし実際の系列の勝率自体は49.69%から64.26%で、49.69%の側にいる区分もあります。並べ替えを上回ることと、売買して残ることは別の話です。 ## 2つの読み方を売買した結果 108通りを売買しました。年間損益がプラスで終わった設定の数です。 | 読み方 | 期間 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---|---| | 反発 | 2025年 | 12 | 21 | 57 | | 反発 | 2024年 | 49 | 79 | 91 | | 抜け | 2025年 | 32 | 61 | 30 | | 抜け | 2024年 | 49 | 95 | 92 | **2024年は反発が91/108、抜けが92/108（4時間足）まで届きます。2025年は57と30です。** 年で大きく振れます。 2つの年をそろえ、さらに2025年を前半後半に割った4つの期間すべてで黒字だった数も出します。 | 読み方 | 時間足 | 2年連続で黒字 | 4期間すべてで黒字 | |---|---|---|---| | 反発 | 15分足 | 11 | 5 | | 反発 | 1時間足 | 19 | 7 | | 反発 | 4時間足 | 50 | 21 | | 抜け | 15分足 | 23 | 10 | | 抜け | 1時間足 | 58 | 32 | | 抜け | 4時間足 | 23 | 11 | **いちばん多いのは1時間足の抜けで108通り中58通り、4期間すべてでも32通りです。** 反発は4時間足で50通り・21通りでした。 勝率と1回あたりの幅も出しておきます。初期設定（遡る200本・スイング3・許容幅0.5ATR・3タッチ）です。 | 読み方 | 時間足 | 年 | 取引回数 | 勝率 | 平均利益 | 平均損失 | 年間損益 | 平均保有 | |---|---|---|---|---|---|---|---|---| | 反発 | 15分足 | 2025 | 6,372回 | 40.68% | +11.11pips | −8.56pips | −3,544.0pips | 5.0本 | | 反発 | 1時間足 | 2025 | 1,799回 | 44.41% | +19.66pips | −15.86pips | −151.3pips | 4.2本 | | 反発 | 4時間足 | 2024 | 261回 | 45.59% | +63.49pips | −33.52pips | +2,794.6pips | 7.9本 | | 抜け | 15分足 | 2025 | 3,407回 | 28.29% | +22.42pips | −9.95pips | −2,682.4pips | 10.3本 | | 抜け | 1時間足 | 2025 | 938回 | 30.38% | +43.26pips | −18.05pips | +542.8pips | 9.3本 | | 抜け | 4時間足 | 2024 | 131回 | 38.93% | +121.24pips | −46.84pips | +2,436.1pips | 16.7本 | **抜けの勝率は25.45%から31.08%です。** 3回に1回も当たりませんが、平均利益が平均損失の2倍を超えます。反発は勝率38.26%から42.80%で、利益と損失の比が1.3倍から1.9倍です。 ### 反発は買いに偏る **反発の読み方は、取引の66.06%から67.73%が買いでした。** 抜けは50.01%から50.19%です。 エンジンは1つの足で買いと売りの両方が成立したとき、買いを先に見ます。反発ではサポートの反発とレジスタンスの反発が同じ足で成立することがあり、そのとき必ず買いになります。**上下対称のはずのルールが、上に偏った売買になります。** インサイドバーの記事でも同じ形が出ました。 ## 線らしさを決める設定がいちばん効かない 4つの設定を1つずつ動かし、その値を含む設定群の年間損益の中央値を並べます。1時間足の2025年です。括弧の中が中央値です。 | 読み方 | 設定 | 値ごとの黒字数と中央値 | |---|---|---| | 反発 | 遡る本数 | 100：2/36（−1,291.5）　200：7/36（−1,131.0）　400：12/36（−384.6） | | 反発 | スイング判定 | 2：7/27（−658.7）　3：7/27（−1,002.9）　5：5/27（−977.5）　8：2/27（−1,393.9） | | 反発 | 許容幅 | 0.25：15/36（−315.4）　0.5：3/36（−1,316.3）　1.0：3/36（−1,049.7） | | 反発 | タッチ回数 | 2：4/36（−1,020.5）　3：9/36（−1,183.6）　4：8/36（−825.7） | | 抜け | 遡る本数 | 100：13/36（−521.9）　200：20/36（+97.2）　400：28/36（+1,002.4） | | 抜け | スイング判定 | 2：24/27（+1,148.5）　3：21/27（+542.8）　5：13/27（−30.6）　8：3/27（−822.2） | | 抜け | 許容幅 | 0.25：20/36（+261.3）　0.5：18/36（−36.0）　1.0：23/36（+286.1） | | 抜け | タッチ回数 | 2：23/36（+441.4）　3：22/36（+423.7）　4：16/36（−128.3） | 中央値の振れ幅で並べ替えます。 | 読み方 | 遡る本数 | スイング判定 | 許容幅 | タッチ回数 | |---|---|---|---|---| | 反発 | 906.9 | 735.2 | 1,000.9 | 357.9 | | 抜け | 1,524.3 | 1,970.7 | 322.1 | 569.7 | **反発ではタッチ回数がいちばん動きません**（357.9pips）。抜けでは許容幅（322.1pips）に次いで小さく、569.7pipsです。 いちばん動くのは、抜けではスイング判定（1,970.7pips）と遡る本数（1,524.3pips）です。しかもどちらも**線の本数が増える方向が良い**という向きです。スイング判定を左右2本にすると、ほとんどの高値安値がスイングになります。遡る本数を400にすると、古い線も生き残ります。 **線が効いたのか、線が増えて取引が増えたのかは、この集計では分けられません。** 分かるのは、線を「らしく」する設定（何回止まったか、どれだけ近ければ同じ線か）よりも、線を多くする設定のほうが結果を動かしたということです。 ## 前年の1位を翌年に当てる 片方の年で年間損益が1位だった設定を相手の年に当てます。216通り（2読み方×108）の中での順位です。 | 時間足 | 選んだ年 | 設定 | 選んだ年の損益 | 相手の年の損益 | 相手の年の順位 | 相手の年の中央値 | |---|---|---|---|---|---|---| | 15分足 | 2024年 | 抜け・100/2本/0.5/3回 | +3,513.9pips | +488.6pips | 26位 | −1,531.7pips | | 15分足 | 2025年 | 反発・100/5本/0.5/4回 | +2,194.7pips | +2,983.4pips | 4位 | −113.8pips | | 1時間足 | 2024年 | 抜け・200/2本/1.0/2回 | +3,247.4pips | +339.8pips | 67位 | −394.6pips | | 1時間足 | 2025年 | 抜け・400/2本/0.25/4回 | +2,323.4pips | +377.2pips | 148位 | +1,033.2pips | | 4時間足 | 2024年 | 抜け・200/8本/1.0/3回 | +3,237.1pips | −1,205.0pips | 170位 | −314.8pips | | 4時間足 | 2025年 | 反発・400/8本/1.0/4回 | +2,569.7pips | +1,463.4pips | 77位 | +962.1pips | **6件のうち相手の年の中央値を上回ったのは4件です。** 15分足はどちらの向きでも上位に残り（26位・4位）、4時間足の2024年は170位まで落ちました。 初期設定に近い「遡る200本・スイング3・許容幅0.5・3タッチ」の順位も出しておきます。反発の4時間足は2024年が216通り中6位（+2,794.6pips）で、2025年は72位（+278.8pips）です。抜けの4時間足は2025年が204位（−1,883.7pips）で、2024年は15位（+2,436.1pips）でした。**同じ設定が同じ時間足で、年をまたぐと6位と72位、204位と15位に動きます。** ## スプレッドで消える分 年間損益がマイナスなのは線が効かないからかコストのせいか。スプレッドを0にして走らせ直します。左が0、右が0.3pipsのときの黒字設定数です。 | 読み方 | 時間足 | 年 | 黒字設定数 | 中央値 | 取引回数の中央値 | |---|---|---|---|---|---| | 反発 | 15分足 | 2024 | 94→49 | +1,339.7→−99.1 | 4,738.5回 | | 反発 | 1時間足 | 2025 | 39→21 | −545.6→−987.7 | 1,428.0回 | | 反発 | 4時間足 | 2025 | 61→57 | +224.9→+154.8 | 364.5回 | | 反発 | 4時間足 | 2024 | 93→91 | +912.5→+845.2 | 196.0回 | | 抜け | 15分足 | 2025 | 57→32 | +172.7→−760.3 | 2,739.5回 | | 抜け | 1時間足 | 2025 | 66→61 | +472.9→+230.4 | 751.5回 | **15分足の反発は、2024年に0.3pipsのスプレッドだけで94通りから49通りへ落ちます。** 中央値も+1,339.7pipsから−99.1pipsです。取引回数の中央値が4,738.5回なので、スプレッドだけで1,421.5pipsを払っています。 **4時間足では93通りから91通りへしか減りません。** 取引回数の中央値が196回だからです。このシリーズが毎回確認している「コスト＝取引回数×スプレッド」の形どおりでした。 ## 関連記事 - [フェアバリューギャップ「埋まる」は本当？2年分の足で検証した結果](/ja/blog/fair-value-gap-settings)：同じ「価格が覚えている水準」型の主張を、対照群を置いて測っています - [フィボナッチ「61.8%で反発」は本当？FXで2年分検証した結果](/ja/blog/fibonacci-retracement-levels)：水準に固有の効果があるかを、非フィボナッチ数を混ぜて測った記事です - [ダブルトップ・ダブルボトムは反転する？2年分検証した結果](/ja/blog/double-top-bottom-settings)：同じスイングから作る形を、別の読み方で測っています - [ドル円の設定は他の通貨ペアで通用する？3,024通り検証](/ja/blog/settings-across-pairs)：この記事の設定もドル円で選んだものです ## 補足 - 対象は米ドル／円のみです。2024年と2025年の2年分で、時間足は15分足・1時間足・4時間足です - 約定は終値、スプレッドは0.3pips固定と0の2種類、ロットは0.1、スリッページはありません - この条件はチャートに線を描きません。数字はバックテストの中で計算した線のものです - 離した価格の対照群は、成立した瞬間の距離はそろっていますが、成立する足の数が引いた線の7.8〜41.6%です。並べ替えた系列の対照群は、両方を終値だけで判定しています --- ### ローソク足パターン17種の勝率を普通の足と比べた URL: https://formiq.jp/ja/blog/candlestick-patterns-tested Language: ja Published: 2026-09-09 Updated: 2026-09-09 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ローソク足, プライスアクション, チャートパターン, 酒田五法, 包み足, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/candlestick-patterns-tested Evidence JSON: https://formiq.jp/ja/evidence/candlestick-patterns-tested/data.json Also available in: en — https://formiq.jp/blog/candlestick-patterns-tested ローソク足パターンは、1本から3本のローソク足の形に名前を付けたものです。長い下ヒゲの**ハンマー**、前の足を包む**包み足**、陽線が3本続く**赤三兵**。どれも「ここで反転する」「ここから続く」という読み方とセットで紹介されます。 Formiqのバックテストには、この種類の条件が17個入っています。ハンマー、流れ星、ピンバー、丸坊主、大陽線・大陰線、トンボ、トウバ、十字線、包み足、はらみ足、毛抜き、インサイドバーのブレイク、明けの明星、宵の明星、赤三兵、三羽烏、そして3本連続の陽線・陰線です。17個すべてを同じ期間・同じ決済で並べて、ドル円の2024年と2025年で数えました。 比べる相手を先に決めました。**そのパターンを作らなかった足です。** 「ハンマーの10本後に上がっていた割合は54.58%」だけでは何も言えません。同じ期間の全部の足を10本後まで追うと50.35%が上がっていたからです。差は4.22ポイントです。 そして、この差は年が変わると残りませんでした。**時間足3種×26通りの成立方向、計78通りのうち、2024年と2025年の両方でこの差がプラスだったのは19通りです。** ## ローソク足パターンは何を見ているのか どのパターンも、実体（始値と終値の差）、ヒゲ（高値・安値と実体の差）、値幅（高値と安値の差）の大小関係だけで決まります。指標の計算は入りません。17種の判定条件と、1時間足・2025年に成立した回数です。 | パターン | 何本使うか | 判定条件 | 買い | 売り | 2025年の成立率 | |---|---|---|---|---|---| | ハンマー | 1本 | 下ヒゲ≧実体×2、かつ上ヒゲ≦実体 | ○ | | 4.92% | | 流れ星 | 1本 | 上ヒゲ≧実体×2、かつ下ヒゲ≦実体 | | ○ | 4.17% | | ピンバー | 1本 | 片側のヒゲ≧実体×2、かつそのヒゲが反対側より長い | ○ | ○ | 28.26% | | 丸坊主 | 1本 | 実体÷値幅≧0.9 | ○ | ○ | 2.92% | | 大陽線・大陰線 | 1本 | 値幅≧ATR(14)×1.5、かつ実体÷値幅≧0.7 | ○ | ○ | 4.46% | | トンボ | 1本 | 実体÷値幅≦0.1、下ヒゲ≧値幅×0.6、上ヒゲ≦値幅×0.05 | ○ | | 0.31% | | トウバ | 1本 | 実体÷値幅≦0.1、上ヒゲ≧値幅×0.6、下ヒゲ≦値幅×0.05 | | ○ | 0.21% | | 十字線 | 2本 | 実体÷値幅≦0.05。直前が陰線なら買い、陽線なら売り | ○ | ○ | 5.08% | | 包み足 | 2本 | 今の足の実体が直前の足の実体を包み、色が反対 | ○ | ○ | 13.50% | | はらみ足 | 2本 | 今の足の実体が直前の足の実体に収まり、色が反対 | ○ | ○ | 12.31% | | 毛抜き | 2本 | 2本の安値（高値）の差≦平均値幅×0.3、かつ色が反対 | ○ | ○ | 35.68% | | インサイドバーのブレイク | 3本 | 2本前の高値安値に1本前が収まり、今の足がその外へ抜ける | ○ | ○ | 8.57% | | 明けの明星 | 3本 | 陰線→小さい足→陽線。両端の実体が中央の2倍超、3本目が1本目の実体の中値超 | ○ | | 3.38% | | 宵の明星 | 3本 | 明けの明星の上下反転 | | ○ | 3.35% | | 赤三兵 | 3本 | 陽線3本。始値が前の足の始値以上、終値が前の足の終値超 | ○ | | 11.63% | | 三羽烏 | 3本 | 陰線3本。始値が前の足の始値以下、終値が前の足の終値未満 | | ○ | 10.73% | | 3本連続の陽線・陰線 | 3本 | 同じ色の実体が3本続く | ○ | ○ | 23.42% | 成立率は1時間足・2025年の6,216本に対する割合です。毛抜きが35.68%、ピンバーが28.26%ある一方で、トウバは0.21%（13回）しかありません。**同じ「パターン」という言葉でも、3本に1本の足と、年に13回の足が混じっています。** ## 17種類の設定と読み方 チャート上では、どれもローソク足そのものの形なので、線や別枠が増えることはありません。バックテストの条件一覧に名前が並び、しきい値を持つものだけ数値の項目が出ます。 ### 数値を変えられるのは8種類だけ | 条件 | 項目 | 初期設定 | 変えると何が変わるか | |---|---|---|---| | ハンマー | 下ヒゲ倍率 | 2 | 上げると下ヒゲの長い足だけが残り、成立回数が減る | | 流れ星 | 上ヒゲ倍率 | 2 | 同じく上ヒゲの長さで絞る | | ピンバー | ヒゲ倍率 | 2 | 買い側と売り側の両方に同じ倍率がかかる | | 丸坊主 | 実体比率 | 0.9 | 1に近づけるほどヒゲの無い足だけになる | | 大陽線・大陰線 | ATR倍率 | 1.5 | 値幅がその時点のATR(14)の何倍必要かを決める | | 大陽線・大陰線 | 実体比率 | 0.7 | 値幅のうち実体が占める割合の下限 | | 十字線 | 実体比率 | 0.05 | 上げるほど実体の小さくない足も十字線に含まれる | | 毛抜き | 一致幅 | 0.3 | 2本の安値（高値）がどれだけ離れていても「同じ」とみなすかを、平均値幅に対する割合で決める | | 3本連続の陽線・陰線 | 本数 | 3 | 何本続けば成立とするか | 残りの9種（トンボ、トウバ、包み足、はらみ足、インサイドバーのブレイク、明けの明星、宵の明星、赤三兵、三羽烏）に数値の項目はありません。形の定義がそのまま条件になっています。 ### 一般に使われている売買ルール 紹介のされ方はどれも同じ形です。反転型（ハンマー、流れ星、ピンバー、トンボ、トウバ、十字線、包み足、はらみ足、毛抜き、明けの明星、宵の明星）は「その方向へ反転するので逆に入る」、継続型（丸坊主、大陽線・大陰線、赤三兵、三羽烏、3本連続、インサイドバーのブレイク）は「その方向へ続くので順に入る」と読みます。 この記事で測る売買は、反転型も継続型も**そのパターンが指す向きに、成立した足の終値で入る**という1つの形にそろえました。反転型なら反転すると読まれる方向、継続型なら続くと読まれる方向です。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 2024年1月1日〜2025年12月31日 | | 成立の判定 | Formiqのバックテスト機能と同じ判定で、1本ずつ数える | | 比べる相手 | 同じ期間・同じ時間足の全部の足 | | 先の見方 | 成立した足の終値から1本後・5本後・10本後・20本後の終値 | | 売買の決済 | 入った足から10本後の終値で決済。損切り・利確・反対シグナルは使わない | | スプレッド | 0.3pips固定 | | ロット | 0.1 | | 約定 | 終値 | 17種のうち8種（ハンマー、トンボ、明けの明星、赤三兵、流れ星、トウバ、宵の明星、三羽烏）は、買いか売りの一方しか作りません。ハンマーは買いの合図であって、売りのハンマーというものが無いので、反対シグナルで決済することができません。**決済を本数で固定したのはそのためです。** 8種だけを本数で決済し、残りを反対シグナルで決済すると、17種の比較が決済の比較になってしまいます。 ## 名前は17個あるが、指していた足は重なっていた 数える前に、17個が別々の足を指しているかを確かめました。1時間足・2025年で、2つの条件が同じ足の同じ向きに成立した回数を数えます。 | 内側の条件 | 外側の条件 | 重なった回数 | 内側のうち重なった割合 | 外側のうち重なった割合 | |---|---|---|---|---| | ハンマー（買い） | ピンバー（買い） | 306回 | 100% | 32.66% | | トンボ（買い） | ピンバー（買い） | 19回 | 100% | 2.03% | | 赤三兵 | 3本連続の陽線 | 723回 | 100% | 95.38% | | 流れ星（売り） | ピンバー（売り） | 259回 | 100% | 31.59% | | トウバ（売り） | ピンバー（売り） | 13回 | 100% | 1.59% | | 三羽烏 | 3本連続の陰線 | 667回 | 100% | 95.56% | | トウバ（売り） | 流れ星（売り） | 12回 | 92.31% | 4.63% | **赤三兵の723回は、1回残らず「陽線が3本続いた足」でもありました。** そして陽線が3本続いた足は758回あったので、赤三兵の追加条件（2本目・3本目の始値が前の足の始値以上、終値が前の足の終値超）が落としたのは35回、4.62%です。三羽烏も同じで、陰線3本連続698回のうち667回が三羽烏でした。 ハンマーも同じ形です。ハンマーの306回はすべてピンバーの買い側で、ピンバーの買い937回の32.66%にあたります。**ハンマーは、ピンバーの買い側に「上ヒゲが実体より短い」を足したものです。** 赤三兵と3本連続の陽線を別々の手法として比べても、比べているのは同じ足の95%です。以降の表でこの2つの成績が違って見えても、違いを作れるのは3本連続にしか入っていない35本だけです。 ## パターンが出なかった足と比べたらどうなるか 1時間足・2025年の6,216本を全部たどると、10本後の終値が上がっていた足は50.35%、下がっていた足は49.63%でした。これが対照群です。17種26通りの成立方向を、同じ数え方で並べます。 | パターン | 向き | 2025年の回数 | 2025年の割合 | 対照群との差 | 2024年の回数 | 2024年の割合 | 対照群との差 | |---|---|---|---|---|---|---|---| | ハンマー | 買い | 306回 | 54.58% | +4.22 | 347回 | 56.48% | +1.57 | | 流れ星 | 売り | 259回 | 49.03% | -0.60 | 247回 | 44.53% | -0.47 | | ピンバー | 買い | 937回 | 50.69% | +0.34 | 1045回 | 56.46% | +1.55 | | ピンバー | 売り | 820回 | 51.10% | +1.47 | 817回 | 44.92% | -0.09 | | 丸坊主 | 買い | 86回 | 63.95% | +13.60 | 114回 | 61.40% | +6.49 | | 丸坊主 | 売り | 96回 | 52.08% | +2.45 | 93回 | 40.86% | -4.15 | | 大陽線・大陰線 | 買い | 128回 | 61.72% | +11.36 | 136回 | 56.62% | +1.71 | | 大陽線・大陰線 | 売り | 149回 | 48.32% | -1.31 | 130回 | 43.08% | -1.93 | | トンボ | 買い | 19回 | 47.37% | -2.99 | 24回 | 33.33% | -21.58 | | トウバ | 売り | 13回 | 38.46% | -11.17 | 11回 | 45.45% | +0.45 | | 十字線 | 買い | 158回 | 44.94% | -5.42 | 149回 | 59.06% | +4.15 | | 十字線 | 売り | 158回 | 48.73% | -0.90 | 192回 | 40.63% | -4.38 | | 包み足 | 買い | 443回 | 52.60% | +2.24 | 415回 | 56.14% | +1.23 | | 包み足 | 売り | 396回 | 54.29% | +4.66 | 403回 | 49.38% | +4.37 | | はらみ足 | 買い | 383回 | 53.52% | +3.17 | 405回 | 53.33% | -1.58 | | はらみ足 | 売り | 382回 | 49.74% | +0.11 | 397回 | 45.84% | +0.84 | | 毛抜き | 買い | 1075回 | 50.88% | +0.53 | 1035回 | 54.11% | -0.81 | | 毛抜き | 売り | 1143回 | 49.52% | -0.11 | 1156回 | 44.72% | -0.28 | | インサイドバーのブレイク | 買い | 276回 | 52.54% | +2.18 | 301回 | 53.49% | -1.42 | | インサイドバーのブレイク | 売り | 257回 | 47.86% | -1.77 | 298回 | 41.61% | -3.40 | | 明けの明星 | 買い | 210回 | 50.00% | -0.35 | 215回 | 53.95% | -0.96 | | 宵の明星 | 売り | 208回 | 50.00% | +0.37 | 217回 | 37.33% | -7.68 | | 赤三兵 | 買い | 723回 | 47.30% | -3.05 | 824回 | 56.55% | +1.64 | | 三羽烏 | 売り | 667回 | 49.03% | -0.60 | 654回 | 47.09% | +2.09 | | 3本連続の陽線・陰線 | 買い | 758回 | 47.49% | -2.86 | 854回 | 56.21% | +1.29 | | 3本連続の陽線・陰線 | 売り | 698回 | 49.28% | -0.35 | 670回 | 46.72% | +1.71 | **明けの明星は2年とも対照群を下回りました。** 1時間足で210回と215回出て、10本後に上昇していた割合は50.00%と53.95%。同じ期間の全部の足はそれぞれ50.35%と54.91%です。宵の明星も、2024年は37.33%に対して対照群が45.01%で、7.68ポイント下回りました。 差が大きく出た2つは丸坊主の買い（+13.60）と大陽線・大陰線の買い（+11.36）で、成立回数はそれぞれ86回と128回です。丸坊主も大陽線も反転型ではなく、**大きな陽線が出た次にさらに上がっていたという継続型**です。 15分足と4時間足も同じ数え方で並べると、時間足3種×26通り＝78通りになります。**2024年と2025年の両方で対照群を上回ったのは19通りでした。** 符号が2年で一致したのは43通りで、コイン投げの39通りとほとんど変わりません。 1年あたり100回以上出た57通りに絞って、2024年の差と2025年の差を突き合わせると、相関は−0.15でした。符号が一致したのは57通り中31通りです。**前の年に対照群を上回ったかどうかは、次の年に上回るかどうかを言い当てませんでした。** 6通り（3つの時間足×2年）すべてで対照群を上回った成立方向はありません。最も多かったハンマーの買いと、大陽線・大陰線の買いと、はらみ足の売りが5通りです。ハンマーの買いは15分足の2025年だけ1.13ポイント下回りました。 ## 「勝率58%」はその年に上がったかどうかで決まっていた 対照群そのものが、年と時間足で大きく動きます。パターンを何も使わず、全部の足から入って一定本数持ったときに上昇していた割合です。 | 時間足 | 年 | 1本後 | 5本後 | 10本後 | 20本後 | 対象の足 | |---|---|---|---|---|---|---| | 15分足 | 2025 | 50.22% | 50.71% | 50.34% | 50.60% | 24,893本 | | 15分足 | 2024 | 51.39% | 52.11% | 52.95% | 54.71% | 24,999本 | | 1時間足 | 2025 | 50.39% | 50.73% | 50.35% | 50.63% | 6,216本 | | 1時間足 | 2024 | 51.70% | 54.66% | 54.91% | 57.38% | 6,250本 | | 4時間足 | 2025 | 50.78% | 50.97% | 49.56% | 48.68% | 1,600本 | | 4時間足 | 2024 | 53.53% | 57.12% | 59.16% | 62.31% | 1,616本 | 2024年のドル円は140.87円で始まり、7月3日に161.95円、9月16日に139.58円を付けて157.20円で終わりました。2025年は157.23円で始まり、4月22日の139.89円を安値に156.67円で終わっています。**2024年は上げて終わった年、2025年は戻って終わった年**で、この違いがそのまま表に出ています。**4時間足なら、何も条件を付けずに買って10本持つだけで、2024年は59.16%が上昇していました。同じことを2025年にやると49.56%です。** この9.6ポイントは、17種26通りで最も大きかった差（丸坊主の買いの13.60ポイント）に迫る大きさです。買いのパターンの勝率が58%と紹介されているとき、その数字がパターンから来たのか、測った年が上げ相場だったから来たのかは、対照群を並べないと分かりません。 ## 下降のあとのハンマーに限っても改善しなかった 紹介記事の多くは、形だけでは足りないと書いています。ハンマーは下降のあとに出てはじめてハンマーで、上昇の途中に出た同じ形は数えない、という条件です。 パターンが始まる直前の5本の終値の変化で、成立回数を2つに分けました。反転すべき向きに動いていた側（買いのパターンなら直前5本が下げていた）と、逆に動いていた側です。両方が30回以上あった128通りで比べます。 | パターン | 向き | 全体 | 直前が教科書どおり | 逆向き | |---|---|---|---|---| | ハンマー | 買い | 54.58% | 54.61%（152回） | 54.90%（153回） | | 包み足 | 買い | 52.60% | 49.11%（224回） | 56.16%（219回） | | 包み足 | 売り | 54.29% | 53.40%（206回） | 55.03%（189回） | | 明けの明星 | 買い | 50.00% | 48.57%（105回） | 51.43%（105回） | | 宵の明星 | 売り | 50.00% | 47.52%（101回） | 52.34%（107回） | | 赤三兵 | 買い | 47.30% | 49.86%（355回） | 44.84%（368回） | | 三羽烏 | 売り | 49.03% | 44.41%（331回） | 53.43%（335回） | **教科書どおりの側が上回ったのは、128通り中58通りです。** 半分を下回りました。1時間足・2025年に限れば、ハンマーは54.61%と54.90%でほぼ同じ、包み足の買いは直前が上げていた側のほうが7.05ポイント高くなっています。 直前の値動きを条件に足すと成立回数は半分になります。回数が半分になって割合が変わらないなら、条件を足した意味は残りません。 ## 17種を同じ決済で売買するとどうなるか ここまでは終値の位置を数えただけです。スプレッド0.3pips・0.1ロットで、成立した足の終値から入り、10本後の終値で決済しました。 ### 1時間足 | パターン | 2025年 取引回数 | 2025年 勝率 | 2025年 年間損益 | 2024年 取引回数 | 2024年 勝率 | 2024年 年間損益 | |---|---|---|---|---|---|---| | ハンマー | 202回 | 53.96% | +791.5pips | 226回 | 57.08% | +1721.2pips | | 流れ星 | 182回 | 51.10% | +55.9pips | 173回 | 46.82% | -68.1pips | | ピンバー | 458回 | 51.97% | +1229.8pips | 462回 | 53.03% | +977.2pips | | 丸坊主 | 148回 | 58.78% | +624.3pips | 147回 | 53.74% | +55.1pips | | 大陽線・大陰線 | 198回 | 52.02% | +53.3pips | 171回 | 52.63% | +327.1pips | | トンボ | 18回 | 44.44% | -90.1pips | 23回 | 34.78% | -232.8pips | | トウバ | 13回 | 38.46% | -127.1pips | 11回 | 45.45% | -13.0pips | | 十字線 | 211回 | 49.76% | +143.8pips | 219回 | 49.77% | -502.4pips | | 包み足 | 356回 | 51.40% | +1000.9pips | 368回 | 54.08% | +2449.2pips | | はらみ足 | 344回 | 47.67% | -662.3pips | 362回 | 46.13% | -2657.8pips | | 毛抜き | 481回 | 51.14% | +1059.0pips | 484回 | 47.31% | -1114.6pips | | インサイドバーのブレイク | 304回 | 51.32% | +727.0pips | 312回 | 45.83% | -360.4pips | | 明けの明星 | 164回 | 48.17% | -68.5pips | 175回 | 54.29% | -281.3pips | | 宵の明星 | 161回 | 51.55% | +817.6pips | 169回 | 38.46% | -1619.7pips | | 赤三兵 | 265回 | 48.68% | +663.8pips | 292回 | 53.77% | +393.0pips | | 三羽烏 | 259回 | 49.81% | +220.2pips | 264回 | 44.32% | -662.8pips | | 3本連続の陽線・陰線 | 407回 | 46.68% | -1249.9pips | 408回 | 50.74% | +1418.5pips | 2024年と2025年の両方で黒字になったのは、1時間足では6種（ハンマー、ピンバー、丸坊主、大陽線・大陰線、包み足、赤三兵）です。15分足では2種（丸坊主、明けの明星）、4時間足では7種（ハンマー、流れ星、トウバ、十字線、包み足、インサイドバーのブレイク、赤三兵）でした。ただし4時間足のトウバは2025年に2回、2024年に1回しか売買がありません。 **3つの時間足すべてで両年とも黒字だったパターンは1つもありません。** 1時間足で両年黒字だったハンマーと包み足も、15分足では残りません。ハンマーは15分足で2024年+1376.9pipsに対して2025年−25.2pips、包み足は15分足で2年とも赤字（2024年−485.2pips、2025年−496.7pips）でした。 15分足で赤字が並ぶ理由は取引回数です。毛抜きは1年で1,952回売買していて、スプレッドだけで585.5pipsを払っています。 ### 2025年を前半と後半に割ると 両年とも黒字だった組み合わせを、2025年の1〜6月と7〜12月で割り直しました。 | 時間足 | パターン | 2025年通年 | 1〜6月 | 7〜12月 | |---|---|---|---|---| | 15分足 | 丸坊主 | +365.0pips | +677.1pips | -312.0pips | | 15分足 | 明けの明星 | +416.0pips | -342.2pips | +758.3pips | | 1時間足 | ハンマー | +791.5pips | -66.2pips | +857.7pips | | 1時間足 | ピンバー | +1229.8pips | +782.6pips | +439.4pips | | 1時間足 | 丸坊主 | +624.3pips | -81.2pips | +705.5pips | | 1時間足 | 大陽線・大陰線 | +53.3pips | +34.1pips | +19.2pips | | 1時間足 | 包み足 | +1000.9pips | +395.5pips | +604.6pips | | 1時間足 | 赤三兵 | +663.8pips | -457.3pips | +1121.1pips | | 4時間足 | ハンマー | +122.6pips | -516.6pips | +655.9pips | | 4時間足 | 流れ星 | +73.8pips | +620.3pips | -546.5pips | | 4時間足 | トウバ | +335.3pips | +240.8pips | +94.5pips | | 4時間足 | 十字線 | +1955.3pips | +2302.2pips | -346.8pips | | 4時間足 | 包み足 | +395.4pips | -66.2pips | +619.6pips | | 4時間足 | インサイドバーのブレイク | +969.2pips | +456.2pips | +490.3pips | | 4時間足 | 赤三兵 | +1136.1pips | +454.8pips | +674.0pips | 15組のうち、前半と後半の両方が黒字だったのは6組です。1時間足の赤三兵は通年+663.8pipsですが、前半だけなら−457.3pipsでした。**2年とも黒字だった15組のうち、半年に割ると9組は片側が赤字です。** ## 前年の1位を翌年も持ち続けたらどうなるか 2024年に最も稼いだパターンを選び、そのまま2025年に持ち越します。 | 時間足 | 2024年の1位 | 2024年の年間損益 | 2025年の年間損益 | 2025年の順位 | 2025年の17種の中央値 | |---|---|---|---|---|---| | 15分足 | ハンマー | +1376.9pips | -25.2pips | 7位 | -372.7pips | | 1時間足 | 包み足 | +2449.2pips | +1000.9pips | 3位 | +220.2pips | | 4時間足 | ハンマー | +1641.7pips | +122.6pips | 7位 | +73.8pips | 3つとも、翌年は17種の中央値を上回りました。このシリーズの他の指標では、前年に一番稼いだ設定が翌年に中央値を下回ることのほうが多かったので、ここは違う結果です。ただし損益は3つとも大きく縮んでいて、15分足では+1376.9pipsが−25.2pipsになりました。15分足で7位でも中央値を上回るのは、2025年に17種のうち11種が赤字だったからです。 ## 3本連続は3本である必要があったのか 赤三兵、三羽烏、明けの明星、宵の明星は、どれも3本の足で定義されています。3という数字を2本から6本まで動かしました。同じ色の実体が何本続いたら入るか、という条件です。 | 時間足 | 本数 | 2025年 取引回数 | 2025年 年間損益 | 2024年 取引回数 | 2024年 年間損益 | |---|---|---|---|---|---| | 1時間足 | 2本 | 517回 | -778.8pips | 524回 | -699.5pips | | 1時間足 | 3本 | 407回 | -1249.9pips | 408回 | +1418.5pips | | 1時間足 | 4本 | 261回 | -484.0pips | 291回 | +527.5pips | | 1時間足 | 5本 | 152回 | +155.1pips | 166回 | +2253.7pips | | 1時間足 | 6本 | 84回 | +32.2pips | 90回 | +710.2pips | 1時間足で両年とも黒字になったのは5本と6本です。**3本は、2024年に+1418.5pips、2025年に−1249.9pipsで、5通りのうち最も振れた本数でした。** 15分足と4時間足では、2本から6本のどれも両年黒字になりません。3つの時間足を合わせた15通りのうち、両年黒字は1時間足の5本と6本の2通りだけです。 本数を増やすと取引回数が落ちます。1時間足の6本は2025年に84回で、+32.2pipsは1回あたり0.38pipsです。スプレッドが0.3pipsではなく0.7pipsだったら、この年は赤字で終わります。 ## 損切り・利確とコスト ### 勝率は決済で決まる 同じ入り方に、損切りと利確を付け替えました。1時間足のハンマーです。 | 損切り・利確 | 2025年 取引回数 | 2025年 勝率 | 2025年 年間損益 | 2024年 年間損益 | |---|---|---|---|---| | なし（10本後に決済） | 202回 | 53.96% | +791.5pips | +1721.2pips | | 損切り30・利確60 | 231回 | 43.29% | +213.1pips | +464.5pips | | 損切り60・利確30 | 230回 | 62.61% | +318.5pips | +369.2pips | | 損切り50・利確50 | 224回 | 52.23% | +49.6pips | +492.4pips | 入る足は同じで、勝率が43.29%から62.61%まで動きます。**損切りを利確の2倍にすれば勝率は上がり、年間損益はどちらも減りました。** 「このパターンは勝率6割」という紹介を読むときは、決済の条件を一緒に見る必要があります。 取引回数が202回から230回前後に増えるのは、損切りや利確で早く決済が終わると、10本待つ場合には見送っていた次の成立に入れるようになるためです。 ### コスト 支払ったスプレッドは、取引回数×0.3pipsと一致しました。1時間足・2025年の17種で、計算どおりでなかったものはありません。 | 時間足 | パターン | 取引回数 | 支払ったスプレッド | 回数×0.3pips | |---|---|---|---|---| | 15分足 | 毛抜き | 1952回 | 585.5pips | 585.6pips | | 15分足 | ピンバー | 1838回 | 551.3pips | 551.4pips | | 15分足 | 包み足 | 1455回 | 436.5pips | 436.5pips | | 1時間足 | 毛抜き | 481回 | 144.3pips | 144.3pips | | 4時間足 | 毛抜き | 124回 | 37.2pips | 37.2pips | スプレッドを0にすると符号が変わったのは、3つの時間足×17種＝51通りのうち2通りだけでした（15分足のハンマーと、15分足の大陽線・大陰線）。**残りの赤字は、回数の多さではなく入り方そのものから来ています。** ## 関連記事 - [包み足の勝率とだまし](/ja/blog/engulfing-settings)：この17種のうち1つを、色が変わっただけの足を比べる相手にして深掘りしています - [明けの明星・宵の明星の勝率](/ja/blog/star-patterns-settings)：3本でできる2種を、窓の有無から条件1つずつまで分解しています - [インサイドバーのブレイク勝率](/ja/blog/inside-outside-bar-settings)：ブレイクに乗る形と逆に張る形を、普通のブレイクと比べています - [ピンバーの勝率とヒゲ比率](/ja/blog/pin-bar-settings)：ヒゲの長さを実体比と値幅比の2通りで測り直しています - [フェアバリューギャップは本当に埋まるのか](/ja/blog/fair-value-gap-settings)：同じく普通の足を比べる相手に置いて、有名なプライスアクションを測っています - [ダブルトップ・ダブルボトムは反転する？](/ja/blog/double-top-bottom-settings)：複数の足でできる形の反転率を、同じ2年で検証しています - [平均足は何を変えているのか](/ja/blog/heikin-ashi-settings)：ローソク足そのものを書き換える表示方法を測っています - [SMCの構造ブレイクと流動性スイープ](/ja/blog/smc-settings)：同じ値動きを別の名前で切り分ける手法を検証しています ## 補足 対象は米ドル／円の2024年1月1日から2025年12月31日です。終値約定、スプレッド0.3pips固定、0.1ロットで計算しました。成立の判定にはFormiqのバックテスト機能と同じ処理を使い、対象期間の前の足も読み込んでいます。1本後・5本後・10本後・20本後を見るため、年の終わりからその本数ぶんは対象から外しています。トンボとトウバは1時間足で年20回前後、4時間足で年1〜6回しか成立しないため、数値の振れが大きくなります。 --- ### 包み足の勝率とだましを396通り検証した結果 URL: https://formiq.jp/ja/blog/engulfing-settings Language: ja Published: 2026-09-09 Updated: 2026-09-09 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: 包み足, プライスアクション, ローソク足, 反転, だまし, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/engulfing-settings Evidence JSON: https://formiq.jp/ja/evidence/engulfing-settings/data.json Also available in: en — https://formiq.jp/blog/engulfing-settings 包み足は2本のローソク足でできています。前の足と色が反対で、実体（始値と終値の間）が前の足の実体を覆っている形です。陰線の次に来た大きな陽線は買い、陽線の次に来た大きな陰線は売り、と紹介されます。 この形には条件が2つ入っています。**色が変わることと、前の足を包むことです。** 色が変わるだけなら珍しくありません。1時間足の2025年で、色が変わった2本組は3,166回ありました。そのうち包んでいたのは839回です。 なので、比べる相手を「色が変わったのに包まなかった2,327回」に置きました。包むことに意味があるなら、この2つの成績は違うはずです。 **1時間足では違いました。** 10本後にパターンの向きへ動いていた割合は、包み足が53.40%、色が変わっただけの足が49.12%です。2024年も52.81%対48.98%で同じ向きでした。**15分足では+0.50ポイントと−0.37ポイントで、差がありません。** ## 包み足は何を見ているのか Formiqのバックテストが使っている判定は、実体だけを見ます。 - **買い**：前の足が陰線、今の足が陽線で、今の足の始値が前の足の終値以下、終値が前の足の始値以上 - **売り**：前の足が陽線、今の足が陰線で、今の足の始値が前の足の終値以上、終値が前の足の始値以下 ヒゲは判定に入りません。高値と安値まで前の足を覆っている形は別にあって、アウトサイドバーと呼ばれます。この記事では両方を数えて比べます。 紹介記事が足す条件は、だいたい3つです。ヒゲまで包んでいること、包んだ足が前の足よりずっと大きいこと、そして下降のあとに出た陽の包み足であること。3つとも測りました。 ## 包み足の設定と読み方 チャートに線や別枠は増えません。ローソク足そのものの形なので、見るのは価格チャートだけです。 ### 設定ダイアログに並ぶ項目 | 条件 | 数値の項目 | 初期設定 | |---|---|---| | 包み足 | なし | 実体が前の足の実体を覆っていれば成立 | **数値の項目はありません。** 「どれくらい大きく包めば包み足か」を決める設定が無いので、成立するかしないかの2択です。1時間足で年800回強、15分足で年3,300回前後、4時間足で年200回前後が成立します。 ### しきい値を足したいとき 包む大きさで絞りたい場合は、条件を2つ同時に有効にします。この記事の検証もその形で作りました。 | 足したい条件 | 組み合わせる条件 | 設定 | |---|---|---| | 包んだ足の実体がヒゲに対して太い | 丸坊主 | 実体比率0.5〜0.9 | | 包んだ足の値幅が大きい | 大陽線・大陰線 | ATR倍率1.0〜2.0、実体比率0.5 | | 移動平均の向きに合わせる | 移動平均の傾き | 期間20または50、判定5本前と比較 | 「丸坊主」は**実体÷値幅がしきい値以上**という条件です。包み足と同時に有効にすると、包んだうえで実体が太い足だけが残ります。「移動平均の傾き」は上向きのとき買いだけ、下向きのとき売りだけを通すので、包み足を移動平均の向きにそろえた売買になります。 ### 一般に使われている売買ルール 包み足が出た足の終値で、その向きに入ります。決済は反対向きの包み足が出るまで持つ形が多く、損切りは包み足の反対側の端（買いなら安値、売りなら高値）に置きます。この記事はどちらも測りました。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 2024年1月1日〜2025年12月31日 | | 成立の判定 | Formiqのバックテスト機能と同じ判定で、1本ずつ数える | | 比べる相手 | 色が変わったのに前の足を包まなかった2本組 | | 先の見方 | 成立した足の終値から1本後・5本後・10本後・20本後の終値 | | 決済 | 反対向きの包み足まで持つ形と、10本後の終値で決済する形の2つ | | 条件の数 | 包み足だけ＋10通りの絞り込み＝11通り | | 回した数 | 11通り×決済2種と手数料なし1種×3時間足×4期間＝396通り | | スプレッド | 0.3pips固定 | | ロット | 0.1 | | 約定 | 終値 | **絞り込みは入口だけに使い、決済はいつも素の包み足にしました。** 絞り込んだ条件を決済にも使うと、厳しい条件ほど反対シグナルが出にくくなって保有が長くなります。そうすると入口の比較ではなく保有時間の比較になってしまいます。 ## 色が変わっただけの足と比べたらどうなるか 10本後にパターンの向きへ動いていた割合です。 | 時間足 | 年 | 包み足の回数 | 包み足の割合 | 色が変わっただけの回数 | その割合 | 差 | |---|---|---|---|---|---|---| | 15分足 | 2025年 | 3301回 | 50.08% | 9182回 | 49.58% | +0.5 | | 15分足 | 2024年 | 3274回 | 49.97% | 9532回 | 50.34% | -0.37 | | 1時間足 | 2025年 | 839回 | 53.40% | 2327回 | 49.12% | +4.28 | | 1時間足 | 2024年 | 818回 | 52.81% | 2297回 | 48.98% | +3.83 | | 4時間足 | 2025年 | 198回 | 54.04% | 601回 | 50.25% | +3.79 | | 4時間足 | 2024年 | 215回 | 47.91% | 561回 | 50.62% | -2.72 | **2024年と2025年の両方で差がプラスだったのは1時間足だけです。** 15分足は2年とも±0.5ポイント以内で、包んでいるかどうかが結果を分けていません。4時間足は2025年に+3.79ポイント、2024年に−2.72ポイントで、年ごとに符号が変わりました。 pipsで見ると差はもっとはっきりします。1時間足の包み足は10本後までに平均+4.28pips（2025年）と+4.59pips（2024年）動き、色が変わっただけの足は−1.58pipsと−2.39pipsでした。差は5.87pipsと6.98pipsです。 ## ヒゲまで包むと成績は上がるのか ヒゲまで包んだ足（アウトサイドバー）と、実体だけ包んだ足に分けます。 | 時間足 | 年 | ヒゲまで包んだ回数 | その割合 | 実体だけ包んだ回数 | その割合 | |---|---|---|---|---|---| | 15分足 | 2025年 | 919回 | 48.97% | 2382回 | 50.50% | | 15分足 | 2024年 | 868回 | 50.69% | 2406回 | 49.71% | | 1時間足 | 2025年 | 229回 | 51.97% | 610回 | 53.93% | | 1時間足 | 2024年 | 199回 | 51.26% | 619回 | 53.31% | | 4時間足 | 2025年 | 61回 | 52.46% | 137回 | 54.74% | | 4時間足 | 2024年 | 67回 | 46.27% | 148回 | 48.65% | **実体だけ包んだ足のほうが高かったのは6通り中5通りです。** ヒゲまで包んだ足のほうが高かったのは15分足の2024年だけで、そこも50.69%対49.71%の0.98ポイント差でした。1時間足では2年とも2ポイント前後、実体だけのほうが上です。 ヒゲまで覆うには、その足の高値と安値が前の足を超えている必要があります。上下どちらにも大きく動いた足なので、方向が決まりきっていない足が混ざります。ただしこの検証では、ヒゲまで包む条件で割合が下がる理由までは特定できません。分かるのは、この条件を足して回数を4分の1に減らしても割合が上がらなかったことです。 ## 実体が値幅の7割以上ある包み足だけを残す 包んだ足そのものの形で絞ります。実体÷値幅が0.7以上、つまりヒゲが値幅の3割以下の足です。 | 時間足 | 年 | 包み足すべて | その割合 | 実体が7割以上 | その割合 | |---|---|---|---|---|---| | 15分足 | 2025年 | 3301回 | 50.08% | 1264回 | 51.82% | | 15分足 | 2024年 | 3274回 | 49.97% | 1271回 | 50.59% | | 1時間足 | 2025年 | 839回 | 53.40% | 282回 | 56.38% | | 1時間足 | 2024年 | 818回 | 52.81% | 264回 | 55.30% | | 4時間足 | 2025年 | 198回 | 54.04% | 65回 | 56.92% | | 4時間足 | 2024年 | 215回 | 47.91% | 70回 | 50.00% | **6通りすべてで上がりました。** 測った絞り込みのうち、これだけが3つの時間足でも2つの年でも例外なく効いています。上げ幅は0.62ポイント（15分足2024年）から2.98ポイント（1時間足2025年）です。 回数は3分の1前後に減ります。1時間足では839回が282回、4時間足では198回が65回になりました。 ## だましはどれくらいの割合で起きるか 損切りを包み足の反対側の端に置く形が一般的なので、そこを基準にします。**買いの包み足なら、その足の安値を割るのが先か、同じ幅だけ上がるのが先か**を数えました。決済も保有期間も入らない数え方です。 | 時間足 | 年 | 種類 | 回数 | 逆行が先だった割合 | 翌足で逆行した割合 | |---|---|---|---|---|---| | 1時間足 | 2025年 | 包み足 | 840回 | 51.73% | 21.79% | | 1時間足 | 2025年 | 色が変わっただけ | 2331回 | 55.95% | 32.43% | | 1時間足 | 2025年 | 実体が7割以上の包み足 | 283回 | 50.35% | 10.60% | | 1時間足 | 2024年 | 包み足 | 818回 | 48.40% | 20.54% | | 1時間足 | 2024年 | 色が変わっただけ | 2297回 | 53.77% | 32.35% | | 1時間足 | 2024年 | 実体が7割以上の包み足 | 264回 | 43.73% | 15.53% | 3つの時間足×2年の6通りで見ると、包み足の逆行先は48.40〜53.27%、色が変わっただけの足は53.67〜57.28%、実体が7割以上の包み足は43.73〜50.71%でした。**包み足が50%を超えたのは6通り中5通りです。** 損切りと利確を同じ幅に置くと、ほぼコイン投げになります。 翌足で逆行した割合は、包み足が17.21〜21.79%、色が変わっただけが32.35〜34.40%、実体が7割以上が10.60〜15.53%でした。**素の包み足はおよそ5回に1回、次の1本で反対側の端を割られます。** 逆行までの本数の中央値は2本です。 つまり「だまし」は珍しい事故ではありません。反対側の端に損切りを置いた場合、5回に1回は次の足で終わり、半分弱は同じ幅に届く前に終わります。実体が7割以上の包み足に絞ると10.60〜15.53%まで下がりますが、そのぶん回数が3分の1になります。 ## 下降のあとの陽の包み足は、上昇のあとより良いか 2本組が始まる直前5本の終値の変化で分けました。買いの包み足なら、直前が下げていた側（教科書どおりの逆張り）と、上げていた側（順張り）です。 | 時間足 | 年 | 直前が逆方向（逆張り）回数 | その割合 | 直前が同方向（順張り）回数 | その割合 | |---|---|---|---|---|---| | 15分足 | 2025年 | 1701回 | 50.44% | 1593回 | 49.72% | | 15分足 | 2024年 | 1595回 | 50.60% | 1676回 | 49.40% | | 1時間足 | 2025年 | 430回 | 51.16% | 408回 | 55.64% | | 1時間足 | 2024年 | 406回 | 49.75% | 411回 | 55.72% | | 4時間足 | 2025年 | 88回 | 56.82% | 109回 | 51.38% | | 4時間足 | 2024年 | 109回 | 53.21% | 106回 | 42.45% | **時間足で答えが逆になりました。** 1時間足では2年とも順張り側が高く、差は4.48ポイントと5.97ポイントです。4時間足では2年とも逆張り側が高く、差は5.44ポイントと10.76ポイントでした。15分足は2年とも逆張り側がわずかに高く、差は0.72ポイントと1.20ポイントです。 つまり「下降のあとの包み足だけを買う」という条件は、この2年のドル円では時間足を決めてからでないと足せません。1時間足でそれを足すと、逆に勝率の低い側だけを残すことになります。 ## 396通りを売買した結果 成立した足の終値で入り、反対向きの包み足が出るまで持ちます。スプレッド0.3pips、0.1ロットです。 | 足す条件 | 2025年 回数 | 2025年 勝率 | 2025年 年間損益 | 2024年 回数 | 2024年 勝率 | 2024年 年間損益 | |---|---|---|---|---|---|---| | なし（包み足だけ） | 439回 | 43.28% | +813.8pips | 426回 | 42.96% | +2180.8pips | | 実体÷値幅 0.5以上 | 334回 | 45.21% | +319.0pips | 341回 | 44.28% | +2339.3pips | | 実体÷値幅 0.6以上 | 280回 | 47.14% | +761.3pips | 279回 | 43.73% | +2237.3pips | | 実体÷値幅 0.7以上 | 210回 | 47.62% | +1182.6pips | 204回 | 44.12% | +2249.1pips | | 実体÷値幅 0.8以上 | 135回 | 53.33% | +1478.2pips | 112回 | 42.86% | +1091.8pips | | 実体÷値幅 0.9以上 | 29回 | 51.72% | +145.7pips | 46回 | 43.48% | +949.9pips | | 値幅がATRの1.0倍以上 | 222回 | 50.00% | +615.1pips | 204回 | 47.06% | +1924.6pips | | 値幅がATRの1.5倍以上 | 110回 | 48.18% | +44.6pips | 100回 | 46.00% | +1186.3pips | | 値幅がATRの2.0倍以上 | 40回 | 40.00% | -121.8pips | 47回 | 46.81% | +1085.1pips | | 20本移動平均の向き | 254回 | 44.49% | +643.8pips | 243回 | 41.56% | +1894.2pips | | 50本移動平均の向き | 232回 | 46.12% | +1226.9pips | 233回 | 42.92% | +1608.7pips | **1時間足では11通り中10通りが2年とも黒字でした。** 赤字が出たのは値幅がATRの2.0倍以上という条件だけで、2025年に40回しか売買していません。 15分足と4時間足は別です。15分足で2年とも黒字だったのは11通り中2通り（ATR2.0倍以上と20本移動平均の向き）、4時間足は4通り（絞り込みなし、実体0.5以上、実体0.6以上、50本移動平均の向き）でした。素の包み足は15分足で2年とも赤字（−679.7pipsと−725.2pips）です。 ### 決済を10本後に固定すると 反対シグナルを待たずに10本で切る形でも、1時間足の11通り中10通りが2年とも黒字でした。勝率の並びはこちらのほうがきれいです。 | 足す条件 | 2025年 回数 | 2025年 勝率 | 2025年 年間損益 | 2024年 回数 | 2024年 勝率 | 2024年 年間損益 | |---|---|---|---|---|---|---| | なし（包み足だけ） | 356回 | 51.40% | +1000.9pips | 368回 | 54.08% | +2449.2pips | | 実体÷値幅 0.5以上 | 284回 | 51.76% | +884.0pips | 306回 | 51.31% | +1248.3pips | | 実体÷値幅 0.6以上 | 251回 | 54.18% | +727.7pips | 256回 | 50.78% | +1628.3pips | | 実体÷値幅 0.7以上 | 194回 | 57.22% | +1656.6pips | 179回 | 55.31% | +1212.8pips | | 実体÷値幅 0.8以上 | 132回 | 58.33% | +1104.6pips | 104回 | 57.69% | +920.6pips | | 実体÷値幅 0.9以上 | 32回 | 56.25% | +155.6pips | 45回 | 53.33% | +556.3pips | 実体比率を0.5から0.8へ上げると、勝率は2025年に51.76%から58.33%、2024年に51.31%から57.69%まで上がります。回数は284回から132回へ半分以下になりました。0.9まで上げると2025年は32回しか売買がなく、年間損益も+155.6pipsまで落ちます。 ### 2025年を前半と後半に割ると | 足す条件 | 2025年通年 | 1〜6月 | 7〜12月 | |---|---|---|---| | なし（包み足だけ） | +813.8pips | +659.3pips | +116.0pips | | 実体÷値幅 0.5以上 | +319.0pips | +76.4pips | +204.2pips | | 実体÷値幅 0.6以上 | +761.3pips | +253.5pips | +469.4pips | | 実体÷値幅 0.7以上 | +1182.6pips | +305.3pips | +838.8pips | | 実体÷値幅 0.8以上 | +1478.2pips | +618.7pips | +821.1pips | | 実体÷値幅 0.9以上 | +145.7pips | +243.2pips | -97.5pips | 6通り中5通りが前半も後半も黒字でした。このシリーズでは半分に割ると片側が赤字になる指標が多く、6通り中5通りが残ったのは多いほうです。落ちたのは実体0.9以上で、後半は−97.5pipsでした。 ## 前年に一番稼いだ条件を翌年へ移すと 2024年に最も稼いだ条件を選び、そのまま2025年に持ち越します。 | 時間足 | 2024年の1位 | 2024年の年間損益 | 2025年の年間損益 | 2025年の順位 | 2025年の11通りの中央値 | |---|---|---|---|---|---| | 15分足 | 50本移動平均の向き | +1680.6pips | -444.3pips | 9位 | +63.7pips | | 1時間足 | 実体÷値幅 0.5以上 | +2339.3pips | +319.0pips | 8位 | +643.8pips | | 4時間足 | 実体÷値幅 0.8以上 | +1405.9pips | -261.1pips | 8位 | -114.1pips | **3つとも翌年は中央値を下回りました。** 1時間足は11通り中10通りが2年とも黒字なのに、その中で2024年の1位を選ぶと2025年は8位になります。前年の順位で絞り込みを選ぶ意味は、この2年では出ていません。 ## 損切り・利確とコスト ### 損切りと利確を置くと | エントリー | 損切り・利確 | 2025年 回数 | 2025年 勝率 | 2025年 損益 | 2024年 損益 | |---|---|---|---|---|---| | 包み足だけ | なし | 439回 | 43.28% | +813.8pips | +2180.8pips | | 包み足だけ | 30・60 | 614回 | 35.18% | +448.2pips | +1600.7pips | | 包み足だけ | 60・30 | 639回 | 53.68% | +515.4pips | +274.1pips | | 包み足だけ | 50・50 | 593回 | 41.48% | +47.4pips | +875.1pips | | 実体7割以上 | なし | 210回 | 47.62% | +1182.6pips | +2249.1pips | | 実体7割以上 | 30・60 | 246回 | 39.02% | +878.0pips | +1194.5pips | | 実体7割以上 | 60・30 | 251回 | 56.18% | +513.6pips | +989.8pips | | 実体7割以上 | 50・50 | 244回 | 45.08% | +590.2pips | +1397.2pips | 入る足は変えずに決済だけを変えると、素の包み足の勝率は35.18%から53.68%まで動きます。**そして勝率が一番高い60・30は、2024年の損益が+274.1pipsで4通りのうち最も少ない値です。** 「勝率5割超の包み足手法」を作ること自体は決済だけでできます。 ### コスト | 時間足 | 足す条件 | 取引回数 | 支払ったスプレッド | 損益分岐スプレッド | |---|---|---|---|---| | 15分足 | なし | 1795回 | 538.5pips | -0.08pips | | 15分足 | 実体0.7以上 | 944回 | 283.2pips | 1.58pips | | 1時間足 | なし | 439回 | 131.7pips | 2.15pips | | 1時間足 | 実体0.7以上 | 210回 | 63.0pips | 5.93pips | | 1時間足 | 実体0.8以上 | 135回 | 40.5pips | 11.25pips | | 4時間足 | なし | 99回 | 29.6pips | 13.07pips | 支払ったスプレッドは、取引回数×0.3pipsと一致しました。損益分岐スプレッドは、そこまで広がると2025年の年間損益が0になる値です。 **15分足の素の包み足は−0.08pipsで、スプレッドを0にしても赤字でした。** 回数が多いから負けているのではなく、入り方そのものが負けています。逆に1時間足で実体0.8以上に絞ると11.25pipsまで余裕がありますが、そのぶん年135回しか売買がありません。 ## 関連記事 - [ローソク足パターン17種の勝率](/ja/blog/candlestick-patterns-tested)：包み足を含む17種を、同じ2年・同じ対照群で横断的に並べています - [フェアバリューギャップは本当に埋まるのか](/ja/blog/fair-value-gap-settings)：同じく普通の足を比べる相手に置いて、プライスアクションを測っています - [明けの明星・宵の明星の勝率](/ja/blog/star-patterns-settings)：3本でできる反転の形を、同じ2年・同じ対照群の置き方で測っています - [ダブルトップ・ダブルボトムは反転する？](/ja/blog/double-top-bottom-settings)：複数の足でできる形の反転率を、同じ2年で検証しています - [平均足は何を変えているのか](/ja/blog/heikin-ashi-settings)：ローソク足そのものを書き換える表示方法を測っています ## 補足 対象は米ドル／円の2024年1月1日から2025年12月31日です。終値約定、スプレッド0.3pips固定、0.1ロットで計算しました。成立の判定にはFormiqのバックテスト機能と同じ処理を使い、対象期間の前の足も読み込んでいます。1本後から20本後までを見るため、年の終わりからその本数ぶんは対象から外しています。逆行の判定では、同じ足で反対側の端と目標の両方に触れた場合は逆行として数えました。実体比率の条件は「丸坊主」を同時に有効にして作っているため、包み足の判定そのものは変わりません。 --- ### MACDの3つの読み方と12/26/9を2,052通り検証 URL: https://formiq.jp/ja/blog/macd-settings Language: ja Published: 2026-09-09 Updated: 2026-09-09 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: MACD, ヒストグラム, シグナル線, 移動平均線, オシレーター, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/macd-settings Evidence JSON: https://formiq.jp/ja/evidence/macd-settings/data.json Also available in: en — https://formiq.jp/blog/macd-settings MACD（マックディー）は、期間の違う2本の指数移動平均の差を線にした指標です。短期12・長期26・シグナル9という数字がほぼ初期設定として使われていて、チャートの下の別枠に、MACD線・シグナル線・ヒストグラムの3つが表示されます。 売買ルールも3つ知られています。MACD線がシグナル線を抜けたところ、MACD線が0を抜けたところ、ヒストグラムが反転したところです。Formiqのバックテストにもこの3つが読み方として並んでいるので、まずそれぞれの成績を測るつもりで回しました。 **3つは別々の売買になりますが、性格が正反対でした。** ヒストグラムの反転は1時間足の57通りすべてで勝率50%を超え、そのうち41通りが年間ではマイナスです。シグナル線とのクロスは勝率が33.54〜40.46%しかないのに、57通りすべてが2024年と2025年の両方でプラスでした。そしてゼロラインクロスは、指数移動平均どうしのクロスと228通りすべてで一致しています。 ## MACDは何を測っているのか 計算は3段です。 1. **MACD線** ＝ 短期の指数移動平均 − 長期の指数移動平均（初期設定では12と26） 2. **シグナル線** ＝ MACD線の指数移動平均（初期設定では9） 3. **ヒストグラム** ＝ MACD線 − シグナル線 1行目がこの指標の本体です。2本の平均の差なので、短期が長期より上にいれば正、下にいれば負になります。値そのものは価格の単位で、ドル円なら「短期の平均が長期より何銭離れているか」を表します。 ここまでで、後の検証結果の半分は決まってしまいます。順に見ます。 ### ヒストグラムの反転は、クロスより早く入る ヒストグラムはMACD線からシグナル線を引いた値です。0を上に抜けた足は、MACD線がシグナル線を上に抜けた足そのものなので、**ヒストグラムをクロスの合図として読むと、シグナル線とのクロスと同じ売買になります。** 反転はそこではありません。ヒストグラムがまだマイナスのうちに、縮み始めた足を指します。3本を見て、下がり続けていたヒストグラムが上向いた谷が買い、上がり続けていた山が売りです。0を抜けるより前に成立するので、クロスを先回りする形になります。 1時間足・2025年・12/26/9で、3つを並べます。 | 読み方 | 取引回数 | 勝率 | 年間損益 | 平均の勝ち幅 | 平均の負け幅 | 平均保有 | |---|---|---|---|---|---|---| | シグナル線とのクロス | 451回 | 38.58% | +2,223.6pips | 68.00pips | −34.69pips | 19.4本 | | ヒストグラムの反転 | 373回 | 56.84% | +761.1pips | 46.71pips | −56.78pips | 23.4本 | | ゼロラインクロス | 200回 | 35.00% | +689.5pips | 95.08pips | −45.89pips | 43.6本 | **勝率が最も高い読み方の、平均の勝ち幅が最も小さくなっています。** 反転は46.71pipsを取って56.78pipsを失う形で、**負けのほうが大きい唯一の読み方**です。早く入るぶん、伸びる前に反対の反転が出て決済されます。 ### ゼロラインクロスは、指数移動平均どうしのクロスだった MACD線は短期の指数移動平均から長期の指数移動平均を引いた値です。したがって**MACD線が0を上に抜けることは、短期の指数移動平均が長期を上に抜けることと同じ条件**になります。ゴールデンクロスそのものです。 同じ19組の期間を、MACDのゼロラインクロスと、移動平均線クロス（平均の種類を指数移動平均にしたもの）の両方で回して突き合わせました。 | 突き合わせた組み合わせ | 取引回数と年間損益が一致 | |---|---| | ゼロラインクロス と 指数移動平均どうしのクロス | 228通り中228通り | こちらも差はありません。MACDのゼロラインで売買する設計は、指標を1つ減らしても同じ売買になります。 このシリーズで「その指標である必要があるのか」を確かめたのは5件目です。CCIのゼロラインクロス、サイコロジカルライン、平均足、移動平均線乖離率に続きます。今回は指標の3つの読み方のうち1つがこれに当たりました。 ### 読み方は3つ、うち1つは別の指標 整理すると、MACD固有の売買条件は**シグナル線とのクロス**と**ヒストグラムの反転**の2つです。ゼロラインクロスも数字は載せますが、それは指数移動平均どうしのクロスの数字でもあります。 ## MACDの設定と読み方 ### どこに表示されるか ローソク足には重なりません。価格チャートの下に別枠が付き、そこに3つが描かれます。 - **MACD線**（初期設定は青） - **シグナル線**（初期設定はオレンジ） - **ヒストグラム**（0より上と下で色が変わる。初期設定は上昇が緑、下降が赤） 0の水平線が基準になります。MT4の標準MACDだけは作りが違い、シグナル線を単純移動平均で計算し、ヒストグラムをMACD線そのものとして描くため、見た目が変わります。この記事とTradingViewのMACDは上の3段の計算です。 ### 設定ダイアログに並ぶ項目 | 項目 | 初期設定 | 何を決めるか | |---|---|---| | 短期 | 12 | MACD線の速いほうの指数移動平均の期間 | | 長期 | 26 | MACD線の遅いほうの指数移動平均の期間 | | シグナル | 9 | MACD線を平滑してシグナル線を作る期間 | バックテスト側には、これに**読み方**（シグナル線とのクロス／ゼロラインクロス／ヒストグラム）が加わります。 ### シグナル期間は、読み方によっては何も変えない シグナル線を使わない読み方では、シグナル期間を動かしても計算に届きません。1時間足・2025年の57通りを、シグナル期間ごとに中央値で並べます。 | シグナル期間 | シグナル線とのクロス | ヒストグラムの反転 | ゼロラインクロス | |---|---|---|---| | 5 | +2,055.4pips（平均614回） | −1,494.1pips（平均438回） | +625.8pips（平均211回） | | 9 | +2,054.3pips（平均480回） | −152.2pips（平均376回） | +625.8pips（平均211回） | | 14 | +1,654.3pips（平均408回） | −111.7pips（平均332回） | +625.8pips（平均211回） | ゼロラインクロスの列は3行とも同じ値です。15分足では3行とも+1,531.0pips・平均861回、4時間足では3行とも−1,290.0pips・平均66回で、時間足を変えても同じでした。**ゼロラインクロスを選んだ時点で、設定できる3つの数字のうち1つは効かなくなります。** シグナル線を読む2つには効きます。効き方は取引回数で、シグナル線とのクロスは5から14へ伸ばすと平均614回から408回へ減ります。 ### 一般に使われている売買ルールと、この記事で測った条件 | 一般的な呼び方 | この記事で測った条件 | |---|---| | ゴールデンクロス・デッドクロス | MACD線がシグナル線を上に抜けたら買い、下に抜けたら売り | | 0ラインブレイク | MACD線が0を上に抜けたら買い、下に抜けたら売り | | ヒストグラムの反転 | ヒストグラムがマイナスのまま上向いたら買い、プラスのまま下向いたら売り（3本で判定） | 決済は反対のシグナルが出るまで持ち続けます。損切りと利確を付けた場合は後ろの節で別に測りました。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2024年1月1日〜12月31日、2025年1月1日〜12月31日 | | 時間足 | 15分足、1時間足、4時間足 | | 短期 | 5・8・12・16・24 | | 長期 | 17・26・35・52 | | シグナル | 5・9・14 | | 組み合わせ | 短期<長期の19組×シグナル3種＝57通り | | 読み方 | シグナル線とのクロス、ゼロラインクロス、ヒストグラム | | 約定 | 終値 | | スプレッド | 0.3pips固定 | | ロット | 0.1 | | 決済 | 反対のシグナルが出るまで | 合計2,052通りを回しています（57通り×読み方3種×時間足3種×期間4区分）。 期間の内訳は2024年・2025年・2025年前半・2025年後半です。 ## 時間足を変えると結果が入れ替わる 読み方と時間足の組み合わせごとに、2024年と2025年の両方でプラスになった設定の数を数えました。 | 時間足 | 読み方 | 2年ともプラス | 2025年の中央値 | 2024年の中央値 | 平均取引回数 | 平均勝率 | |---|---|---|---|---|---|---| | 15分足 | シグナル線とのクロス | 10 / 57 | +365.3pips | −1,359.9pips | 2,125回 | 33.04% | | 15分足 | ヒストグラムの反転 | 0 / 57 | −2,535.6pips | −795.0pips | 1,549回 | 57.62% | | 15分足 | ゼロラインクロス | 54 / 57 | +1,531.0pips | +1,659.5pips | 861回 | 29.57% | | 1時間足 | シグナル線とのクロス | 57 / 57 | +1,972.5pips | +2,127.4pips | 501回 | 36.95% | | 1時間足 | ヒストグラムの反転 | 1 / 57 | −433.7pips | −1,143.0pips | 382回 | 57.12% | | 1時間足 | ゼロラインクロス | 48 / 57 | +625.8pips | +1,103.2pips | 211回 | 33.96% | | 4時間足 | シグナル線とのクロス | 27 / 57 | +762.2pips | +137.3pips | 128回 | 44.12% | | 4時間足 | ヒストグラムの反転 | 19 / 57 | −401.6pips | +1,301.0pips | 100回 | 50.90% | | 4時間足 | ゼロラインクロス | 0 / 57 | −1,290.0pips | +2,398.0pips | 66回 | 26.95% | **シグナル線とのクロスは、1時間足では57通り全部が2年ともプラスで、15分足では57通り中10通りまで落ちます。** 15分足の2024年は中央値がマイナスで、57通りのうち過半数が1年を通してマイナスで終わりました。 **ヒストグラムの反転は、平均勝率が3つの読み方で最も高いのに、2年ともプラスになったのは15分足で0通り、1時間足で1通りです。** 4時間足だけ19通りまで戻りますが、そこも2025年の中央値は−401.6pipsでした。 ゼロラインクロスは逆向きです。15分足では54通りが2年ともプラスなのに、4時間足では0通りです。4時間足のゼロラインクロスは2024年に中央値+2,398.0pipsと最も大きい一方、2025年は−1,290.0pipsでした。年で符号が変わっています。 取引回数を見ると理由の一部が分かります。4時間足のゼロラインクロスは年66回しかありません。2024年のドル円は買い持ちで+1,632.5pipsの上昇があったので、少ない回数でも大きく取れました。2025年は始値157.23に対して終値156.67で、買い持ちなら−55.7pipsです。**回数の少ない設定は、その年の方向にそのまま左右されます。** 一方、1時間足のシグナル線とのクロスは平均501回あり、2025年の中央値も+1,972.5pipsでした。年間の上下がほぼ同値で終わった年に、この数字が出ています。 ## 12/26/9は特別な数字ではなかった 初期設定として広く使われている12/26/9を、同じ57通りの中で順位まで含めて測りました。 | 時間足 | 読み方 | 2025年 | 2024年 | |---|---|---|---| | 15分足 | シグナル線とのクロス | 1,938回・勝率33.69%・+1,093.6pips（7位） | −1,359.9pips（29位） | | 15分足 | ヒストグラムの反転 | 1,493回・勝率57.74%・−2,695.2pips（32位） | −538.9pips（24位） | | 15分足 | ゼロラインクロス | 849回・勝率29.33%・+1,212.0pips（41位） | +2,225.3pips（11位） | | 1時間足 | シグナル線とのクロス | 451回・勝率38.58%・+2,223.6pips（16位） | +3,152.1pips（2位） | | 1時間足 | ヒストグラムの反転 | 373回・勝率56.84%・+761.1pips（2位） | −1,966.7pips（49位） | | 1時間足 | ゼロラインクロス | 200回・勝率35.00%・+689.5pips（23位） | +623.6pips（50位） | | 4時間足 | シグナル線とのクロス | 107回・勝率46.73%・+1,299.1pips（11位） | +88.1pips（30位） | | 4時間足 | ヒストグラムの反転 | 88回・勝率48.86%・−109.8pips（24位） | +1,630.9pips（18位） | | 4時間足 | ゼロラインクロス | 71回・勝率19.72%・−2,089.3pips（53位） | +2,690.3pips（17位） | 順位はいずれも57通り中のものです。**同じ12/26/9が、1時間足のシグナル線とのクロスでは2024年に2位、1時間足のゼロラインクロスでは2024年に50位でした。** ヒストグラムの反転では2025年に2位（+761.1pips）だったものが2024年には49位（−1,966.7pips）です。 4時間足のゼロラインクロスに至っては2025年に53位・−2,089.3pipsです。 15分足のシグナル線とのクロスでは、2024年の12/26/9がちょうど29位で、これは57通りの中央値そのものです。この設定が特に良くも悪くもなかった年でした。 **12/26/9という数字を選ぶ前に、時間足と読み方を決める必要があります。** 期間だけを他人と揃えても、同じ売買にはなりません。 ## 短期を伸ばすほど成績が下がった 1時間足・シグナル線とのクロスの57通りを、短期の値ごとに中央値で並べます。長期とシグナルは全部の組み合わせが入っています。 | 短期 | 2025年の中央値 | 平均取引回数 | 2年ともプラス | |---|---|---|---| | 5 | +2,243.0pips | 674回 | 12 / 12 | | 8 | +2,044.4pips | 549回 | 12 / 12 | | 12 | +2,033.4pips | 469回 | 12 / 12 | | 16 | +1,886.5pips | 424回 | 12 / 12 | | 24 | +1,492.3pips | 351回 | 9 / 9 | 短い順に大きく、長くするほど下がります。ただし**5列目はどの短期でも全部が2年ともプラス**で、優劣は付いていても失敗はしていません。 長期のほうは順番になりません。 | 長期 | 2025年の中央値 | 平均取引回数 | |---|---|---| | 17 | +2,224.6pips | 582回 | | 26 | +2,011.4pips | 503回 | | 35 | +1,755.8pips | 482回 | | 52 | +1,889.3pips | 452回 | 35で一度下がって52で戻ります。取引回数は単調に減っているので、回数だけでは説明が付きません。この集計だけでは理由を特定できません。 ## 片方の年で1位だった設定を、もう片方の年に当てる 各時間足で、片方の年の年間損益が最も大きかった設定を選び、そのまま相手の年に当てました。シグナル線とのクロスです。 | 時間足 | 選んだ年と設定 | 選んだ年 | 相手の年 | |---|---|---|---| | 15分足 | 2024年の12/52/14 | +1,273.1pips（1位） | +561.8pips（18位） | | 15分足 | 2025年の16/52/5 | +1,404.2pips（1位） | −1,025.3pips（28位） | | 1時間足 | 2024年の8/17/9 | +3,261.2pips（1位） | +2,526.5pips（9位） | | 1時間足 | 2025年の8/17/5 | +3,376.8pips（1位） | +1,450.5pips（51位） | | 4時間足 | 2024年の5/17/5 | +1,515.2pips（1位） | +308.5pips（45位） | | 4時間足 | 2025年の16/17/9 | +1,862.9pips（1位） | +259.7pips（25位） | 順位はいずれも57通り中です。6件のうち5件で、1位だった設定が相手の年では9位から51位まで落ちました。**1時間足の2025年1位だった8/17/5は、2024年には51位です。** それでも2024年の年間損益は+1,450.5pipsでプラスでした。 1時間足のシグナル線とのクロスは57通り全部が2年ともプラスなので、**最下位を引いてもプラスで終わります**。順位が翌年に残らないことと、選んだ設定が翌年にマイナスになることは別です。この時間足と読み方では前者だけが起きました。 15分足は違います。2025年1位の16/52/5は2024年に−1,025.3pipsで、選び方を間違えると符号が変わります。 ## シグナル線とのクロスは勝率が全部37%前後だった 1時間足・2025年・シグナル線とのクロスの57通りは、勝率が33.54%から40.46%の間に収まりました。そして**57通り全部が勝率50%未満のままプラスで終わっています**。 年間損益で並べた上・中央・下の3つを比べます。 | 設定 | 取引回数 | 勝率 | 年間損益 | 平均の勝ち幅 | 平均の負け幅 | 1回あたり | 最大ドローダウン | |---|---|---|---|---|---|---|---| | 8/17/5 | 723回 | 37.07% | +3,376.8pips | 56.16pips | −25.66pips | 4.67pips | 1,284.5pips | | 8/52/14 | 407回 | 37.10% | +1,972.5pips | 70.38pips | −33.81pips | 4.84pips | 829.2pips | | 24/26/14 | 316回 | 37.34% | +627.8pips | 77.17pips | −42.82pips | 1.98pips | 1,440.5pips | 「1回あたり」は勝率と平均の勝ち幅・負け幅から計算した1取引あたりのpipsです。取引回数を掛けると年間損益とほぼ一致します（723回×4.67pipsで約3,376pips、実測は+3,376.8pips）。 **勝率は3つとも37%台で、順番は年間損益と逆です。** 最下位の24/26/14が最も高い勝率で、最も大きい平均の勝ち幅（77.17pips）を持っています。 差が付いた場所は2つあります。**上位2つは1回あたりが4.67pipsと4.84pipsでほぼ同じで、年間損益の差は取引回数の差（723回と407回）です。** 一方、最下位の24/26/14だけは1回あたりが1.98pipsで、上の2つの半分以下でした。勝ち幅と負け幅の比は2.19倍・2.08倍・1.80倍で、年間損益の順番に下がっています。 最大ドローダウンは回数順ではありません。316回の24/26/14が1,440.5pipsで最も深く、407回の8/52/14が829.2pipsで最も浅くなっています。 ## フィルター・損切り利確・コスト 以下は1時間足・12/26/9・シグナル線とのクロスで測りました。フィルターなしの成績は2025年が451回・+2,223.6pips、2024年が423回・+3,152.1pipsです。 ### ADXフィルターは5度目の失敗 トレンドが出ているときだけ売買する条件として、ADXが一定の水準以上のときに限る形を4通り試しました。 | 条件 | 2025年 | 2024年 | |---|---|---| | フィルターなし | 451回・+2,223.6pips | 423回・+3,152.1pips | | ADX 15以上 | 28回・+2,435.1pips | 30回・−2,061.5pips | | ADX 20以上 | 50回・+288.4pips | 58回・+950.2pips | | ADX 25以上 | 57回・+1,053.9pips | 53回・+984.1pips | | ADX 30以上 | 59回・+790.9pips | 50回・+982.0pips | どの水準も取引回数を451回から28〜59回まで落とします。2025年だけを見るとADX 15以上が+2,435.1pipsでフィルターなしを上回りますが、同じ条件の2024年は−2,061.5pipsです。**2年とも改善した水準はありません。** このシリーズでADXフィルターを試したのは5度目で、5度とも同じ形になりました。 取引回数は水準を上げるほど増えています。2025年で28回・50回・57回・59回です。フィルターを厳しくすると回数が減るという向きにはなっていません。エントリーとポジションを閉じる判断の両方に同じ条件が掛かるため、条件が厳しいほど売買の区切り方が変わりますが、この集計だけでは理由を特定できません。 ### 損切り・利確 | 設定 | 2025年 | 2024年 | |---|---|---| | なし（反対シグナルまで） | 451回・勝率38.58%・+2,223.6pips | +3,152.1pips | | 損切り20・利確40 | 404回・勝率33.91%・+455.1pips | +1,075.7pips | | 損切り30・利確60 | 397回・勝率35.52%・+1,232.9pips | +366.8pips | | 損切り50・利確100 | 396回・勝率37.63%・+1,906.2pips | +1,975.9pips | | 損切り30・利確なし | 394回・勝率31.73%・+2,101.9pips | +1,426.4pips | | 損切りなし・利確60 | 451回・勝率43.46%・+1,570.3pips | +2,326.1pips | 数字はpipsです。**利確60だけを付けると勝率は38.58%から43.46%へ上がりますが、年間損益は+2,223.6pipsから+1,570.3pipsへ下がります。** 勝ち幅が2倍ある売買ルールで勝ちを60pipsで切ると、上の節で見た「平均の勝ち幅56〜77pips」を削ることになります。 2年とも元の成績に近いのは損切り50・利確100だけで、それでも2025年は+1,906.2pipsと下がりました。 ### コスト スプレッドを変えて同じ売買を回します。2025年・451回です。 | スプレッド | 年間損益 | |---|---| | 0.0pips | +2,358.9pips | | 0.3pips | +2,223.6pips | | 0.6pips | +2,088.3pips | | 1.0pips | +1,907.9pips | | 1.5pips | +1,682.4pips | | 2.0pips | +1,456.9pips | **スプレッドが1pips増えるごとに451pips減ります。** これは取引回数と同じ数字です。このシリーズで測った他の指標と同じで、スプレッドで失うpipsは取引回数×スプレッドになります。 スプレッド0での年間損益+2,358.9pipsを451回で割ると1回あたり5.2pipsです。スプレッドがこれを超えると、この設定はマイナスになります。 取引回数の多い設定ほど、この余裕は小さくなります。同じ12/26/9でも15分足のシグナル線とのクロスは2025年に1,938回あり、同じ計算での損益分岐スプレッドは0.86pipsです。**時間足を1つ下げると、耐えられるスプレッドが6分の1になります。** ## 関連記事 - [移動平均線クロスの設定を比較](/ja/blog/moving-average-cross-settings)：この記事のゼロラインクロスと同じ売買を、移動平均線の側から測っています - [移動平均線はSMAとEMAどっちが勝てる？672通り検証した結果](/ja/blog/sma-vs-ema)：MACDが使っている指数移動平均を、他の平均の作り方と比べています - [OsMAの設定を比較](/ja/blog/osma-settings)：MACD線とシグナル線の差を指標として独立させたものです - [CCIの設定を比較](/ja/blog/cci-settings)：ゼロラインクロスが移動平均のクロスと同じになった1件目です - [移動平均線乖離率の設定を比較](/ja/blog/ma-disparity-settings)：同じ突き合わせの4件目です - [ドル円の設定は他の通貨ペアで通用する？3,024通り検証](/ja/blog/settings-across-pairs)：この設定は2024年の1時間足でドル円1位（+3,152.1pips）でしたが、他の3ペアはすべて赤字でした ## 補足 対象は米ドル／円のみ、期間は2024年と2025年の2年です。約定は終値、スプレッドは0.3pips固定、スリッページなしで計算しています。実際の取引では約定価格と手数料が変わります。 2024年のドル円は始値140.87から終値157.20で、買い持ちなら+1,632.5pipsでした。順張りの売買ルールに有利な期間が含まれています。2025年は始値157.23から終値156.67で−55.7pipsですが、年間の高安の幅は1,899.5pipsありました。 読み方の一致を確かめた突き合わせは、同じ期間・同じ約定条件で両方を回して、取引回数・勝率・年間損益を比べたものです。取引の1件ずつを照合したものではありません。 --- ### 明けの明星・宵の明星の勝率を240通り検証した結果 URL: https://formiq.jp/ja/blog/star-patterns-settings Language: ja Published: 2026-09-09 Updated: 2026-09-09 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: 明けの明星, 宵の明星, 酒田五法, 三川, ローソク足, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/star-patterns-settings Evidence JSON: https://formiq.jp/ja/evidence/star-patterns-settings/data.json Also available in: en — https://formiq.jp/blog/star-patterns-settings 明けの明星は3本のローソク足でできた形です。1本目が陰線、2本目が実体の小さい足、3本目が陽線で、3本目が1本目の実体の半分より上まで戻ったところで成立します。宵の明星は上下をそのまま反転させた形です。酒田五法では**三川**の一種として紹介されます。 名前の由来は、真ん中の足が上下から離れて浮いていることです。夜明け前の空に1つだけ残る星が、その形に見えるという説明が付きます。つまり本来は、2本目の前後に**窓**が空いていることが条件に入っています。 そこを最初に数えました。**ドル円の2年間、3つの時間足で成立した4,558回のうち、上下に窓が空いていたものは0回です。** そして、この形が「陰線→なにか→陽線」という色の並びそのものより優れているかを測ると、**6通り中5通りで下回りました。** ## 明けの明星と宵の明星は何を見ているのか Formiqのバックテストが使っている判定は、明けの明星なら次の3つを同時に満たす形です。 1. **色の並び**：1本目が陰線、3本目が陽線（2本目の色は問わない） 2. **真ん中が小さい**：2本目の実体が、1本目の実体の半分未満かつ3本目の実体の半分未満 3. **戻り**：3本目の終値が、1本目の実体の中値より上 宵の明星はすべて反転させます。1本目が陽線、3本目が陰線で、3本目の終値が1本目の実体の中値より下です。 **窓は判定に入っていません。** 判定に入っているのは終値と始値の位置関係だけで、2本目が1本目から離れているかどうかは見ていません。この記事では、窓を数えたうえで、条件1つずつを外したり動かしたりして比べます。 ## 窓は本当に開いているのか 2つの数え方で窓を数えました。**実体の窓**は、2本目の実体が1本目の終値より完全に下（明けの明星の場合）にあり、3本目の実体が2本目の実体より完全に上にある形です。**値幅の窓**は、ヒゲも含めて離れている形で、チャートに白い隙間が見えるのはこちらです。 | 時間足 | 年 | 成立した回数 | 実体の窓 | その割合 | 値幅の窓 | |---|---|---|---|---|---| | 15分足 | 2025年 | 1714回 | 36回 | 2.10% | 0回 | | 15分足 | 2024年 | 1789回 | 26回 | 1.45% | 0回 | | 1時間足 | 2025年 | 418回 | 5回 | 1.20% | 0回 | | 1時間足 | 2024年 | 432回 | 6回 | 1.39% | 0回 | | 4時間足 | 2025年 | 106回 | 1回 | 0.94% | 0回 | | 4時間足 | 2024年 | 99回 | 1回 | 1.01% | 0回 | **値幅の窓は6通りすべてで0回でした。** 合計4,558回のうち0回です。 理由は為替の取引時間にあります。株式のように毎日取引が止まって翌朝また始まるのではなく、平日は24時間つながっているので、足の始値は前の足の終値とほぼ同じ価格になります。 | 時間足 | 年 | 足の数 | 始値が直前の終値と違った割合 | 1pips以上離れた割合 | 最大の開き | |---|---|---|---|---|---| | 15分足 | 2025年 | 24903本 | 93.04% | 0.97% | 204.05pips | | 15分足 | 2024年 | 24999本 | 94.21% | 0.85% | 94.80pips | | 1時間足 | 2025年 | 6226本 | 93.74% | 2.28% | 204.05pips | | 1時間足 | 2024年 | 6250本 | 93.81% | 2.05% | 94.80pips | | 4時間足 | 2025年 | 1610本 | 92.98% | 3.42% | 204.05pips | | 4時間足 | 2024年 | 1616本 | 94.25% | 3.28% | 94.80pips | 始値が直前の終値と少しでも違う足は93〜94%あります。ただし**1pips以上離れた足は0.85〜3.42%**しかありません。残りは1pips未満のずれです。最大の開きは3つの時間足とも同じ足で、2025年10月5日の204.05pipsと、2024年10月27日の94.80pipsでした。どちらも日曜の取引再開時です。 つまり為替で「窓を空けて離れた星」を待つと、週明けの数本しか候補が残りません。**この記事の残りは、窓を外した形、つまり実際に売買条件として使われている形を測っています。** ## 明けの明星の設定と読み方 チャートに線や別枠は増えません。ローソク足そのものの形なので、見るのは価格チャートだけです。 ### 設定ダイアログに並ぶ項目 | 条件 | 数値の項目 | 成立の中身 | |---|---|---| | 明けの明星 | なし | 陰線→小さい足→陽線で、3本目が1本目の実体の中値より上 | | 宵の明星 | なし | 陽線→小さい足→陰線で、3本目が1本目の実体の中値より下 | **どちらも数値の項目はありません。** 「真ん中の足がどれくらい小さければいいか」も「3本目がどこまで戻ればいいか」も、2倍と中値で固定されています。この記事はその2つを外側から動かして測りました。 ### しきい値を足したいとき | 足したい条件 | 組み合わせる条件 | 設定 | |---|---|---| | 3本目の実体がヒゲに対して太い | 丸坊主 | 実体比率0.7 | | 3本目の値幅が大きい | 大陽線・大陰線 | ATR倍率1.5、実体比率0.5 | | 移動平均の向きに合わせる | 移動平均の傾き | 期間50、判定5本前と比較 | ### 一般に使われている売買ルール 3本目が確定した足の終値で、その向きに入ります。明けの明星は買い、宵の明星は売りです。**片方しか作らない形なので、明けの明星だけを条件にすると買いしか出ません。** 両方を同時に有効にすると、明けの明星で買って宵の明星で決済する、両方向の売買になります。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 2024年1月1日〜2025年12月31日 | | 成立の判定 | Formiqのバックテスト機能と同じ判定で、1本ずつ数える | | 比べる相手 | 同じ色の並び（陰線→なにか→陽線、およびその反転）すべて | | 先の見方 | 成立した足の終値から1本後・5本後・10本後・20本後の終値 | | 決済 | 反対の星が出るまで持つ形と、10本後の終値で決済する形 | | 条件の数 | 両方同時＋片方ずつ＋絞り込み＝10通り | | 回した数 | 10通り×3時間足×4期間×手数料あり／なし＝240通り | | スプレッド | 0.3pips固定 | | ロット | 0.1 | | 約定 | 終値 | ## 同じ色の並びと比べたらどうなるか 明けの明星は「陰線→なにか→陽線」の上に条件を2つ足した形です。土台の色の並びだけでどうなるかを先に置きます。 | 時間足 | 年 | 色の並びの回数 | その割合 | 明けの明星・宵の明星 | その割合 | 差 | |---|---|---|---|---|---|---| | 15分足 | 2025年 | 12337回 | 49.74% | 1714回 | 50.35% | +0.61 | | 15分足 | 2024年 | 12606回 | 49.82% | 1789回 | 49.02% | -0.80 | | 1時間足 | 2025年 | 3142回 | 50.22% | 418回 | 50.00% | -0.22 | | 1時間足 | 2024年 | 3166回 | 48.70% | 432回 | 45.60% | -3.10 | | 4時間足 | 2025年 | 802回 | 50.87% | 106回 | 45.28% | -5.59 | | 4時間足 | 2024年 | 782回 | 49.49% | 99回 | 38.38% | -11.10 | **上回ったのは15分足の2025年だけで、そこも+0.61ポイントです。** 残り5通りはマイナスで、4時間足の2024年は11.10ポイント下回りました。 色の並びだけで50%前後になるのは当然です。1本目が陰線で3本目が陽線という並びは、1時間足で年3,000回以上あります。**2つの条件を足して回数を7分の1前後に絞った結果が、絞る前より低いというのが、この表の中身です。** ## 3本目はどこまで戻せば成立とするのか 判定に入っている「1本目の実体の中値より上」を動かします。1本目の実体をどれだけ戻したかを0%超から100%超まで6段階に切り、真ん中の足が小さいという条件はそのまま残しました。 | 戻りの条件 | 2025年 回数 | 2025年 割合 | 2024年 回数 | 2024年 割合 | |---|---|---|---|---| | 0%超（戻れば何でも） | 570回 | 50.53% | 578回 | 45.50% | | 25%超 | 522回 | 51.53% | 533回 | 45.97% | | 50%超（初期の判定） | 418回 | 50.00% | 432回 | 45.60% | | 61.8%超 | 365回 | 50.14% | 387回 | 45.99% | | 75%超 | 325回 | 50.46% | 348回 | 45.40% | | 100%超（1本目を全部戻す） | 267回 | 50.19% | 280回 | 46.07% | 1時間足の数字です。**同じ年の中では1.53ポイントと0.67ポイントしか動きません。** 一方、同じ戻りの条件でも年が変われば4.1〜5.6ポイント動きます。50%超の初期設定は、2025年に50.00%、2024年に45.60%でした。 回数は570回から267回まで半分以下に減ります。**減らした分だけ良くなる、という関係は出ていません。** ## 真ん中の足の大きさと色は効いたのか もう1つの条件、「2本目の実体が両隣の半分未満」を外します。 | 時間足 | 年 | 色の並びだけ | 真ん中の条件だけ | 戻りの条件だけ | 両方（初期の判定） | |---|---|---|---|---|---| | 15分足 | 2025年 | 49.74%（12337回） | 50.78%（2233回） | 49.87%（7670回） | 50.35%（1714回） | | 15分足 | 2024年 | 49.82%（12606回） | 49.27%（2330回） | 49.55%（7721回） | 49.02%（1789回） | | 1時間足 | 2025年 | 50.22%（3142回） | 50.70%（574回） | 50.47%（1922回） | 50.00%（418回） | | 1時間足 | 2024年 | 48.70%（3166回） | 45.52%（580回） | 48.96%（1924回） | 45.60%（432回） | | 4時間足 | 2025年 | 50.87%（802回） | 48.65%（148回） | 48.99%（494回） | 45.28%（106回） | | 4時間足 | 2024年 | 49.49%（782回） | 41.73%（139回） | 47.63%（464回） | 38.38%（99回） | **戻りの条件だけを残した列は、色の並びだけの列とほとんど同じです。** 6通りの差は−1.88から+0.26ポイントで、1時間足では0.25と0.26ポイントしかありません。戻りの条件は、回数を6割に減らして割合を変えていません。 真ん中の条件だけを残した列は、15分足と1時間足の2025年ではわずかに上、それ以外では下です。**この2つを組み合わせた初期の判定が、片方ずつより良くなる関係は出ていません。** 真ん中の足の色も測りました。3本目と反対の色（明けの明星なら2本目が陰線）だったのは、成立回数の41.41〜46.06%です。 | 時間足 | 年 | 3本目と反対の色 | その割合 | 3本目と同じ色 | その割合 | |---|---|---|---|---|---| | 15分足 | 2025年 | 753回 | 52.06% | 961回 | 49.01% | | 15分足 | 2024年 | 782回 | 49.36% | 1007回 | 48.76% | | 1時間足 | 2025年 | 183回 | 50.82% | 235回 | 49.36% | | 1時間足 | 2024年 | 199回 | 49.25% | 233回 | 42.49% | | 4時間足 | 2025年 | 46回 | 36.96% | 60回 | 51.67% | | 4時間足 | 2024年 | 41回 | 34.15% | 58回 | 41.38% | 15分足と1時間足では反対の色が4通りとも上、4時間足では2通りとも下でした。**6通りのうち4通りで上ですが、時間足でそろっていません。** ## 240通りを売買した結果 10本後の終値で決済する形です。片方の星だけを条件にすると買いか売りの一方しか出ないので、反対シグナルでの決済ができません。 | 条件 | 2025年 回数 | 2025年 勝率 | 2025年 年間損益 | 2024年 回数 | 2024年 勝率 | 2024年 年間損益 | |---|---|---|---|---|---|---| | 両方（反対の星まで持つ） | 233回 | 44.64% | -386.7pips | 242回 | 43.39% | -2240.9pips | | 両方（10本で決済） | 249回 | 50.60% | +559.7pips | 267回 | 46.44% | -1529.2pips | | 明けの明星だけ | 164回 | 48.17% | -68.5pips | 175回 | 54.29% | -281.3pips | | 明けの明星＋実体0.7以上 | 71回 | 54.93% | -93.8pips | 80回 | 58.75% | +510.2pips | | 明けの明星＋ATR1.5倍以上 | 33回 | 60.61% | +157.7pips | 31回 | 67.74% | +213.7pips | | 明けの明星＋50本平均の向き | 82回 | 42.68% | +107.2pips | 108回 | 57.41% | +100.0pips | | 宵の明星だけ | 161回 | 51.55% | +817.6pips | 169回 | 38.46% | -1619.7pips | | 宵の明星＋実体0.7以上 | 72回 | 45.83% | +184.3pips | 68回 | 41.18% | -1081.7pips | | 宵の明星＋ATR1.5倍以上 | 33回 | 36.36% | -227.9pips | 26回 | 34.62% | -443.8pips | | 宵の明星＋50本平均の向き | 87回 | 50.57% | +149.8pips | 69回 | 42.03% | -1097.0pips | 1時間足です。**2024年と2025年の両方で黒字になったのは10通り中2通り**で、どちらも明けの明星に絞り込みを足したものです。ATR1.5倍以上は2025年に33回、2024年に31回しか売買がありません。 15分足では10通り中3通り（明けの明星だけ、実体0.7以上、50本平均の向き）、4時間足では2通り（明けの明星＋実体0.7以上、宵の明星＋実体0.7以上）でした。**3つの時間足すべてで2年とも黒字だったものはありません。** ## 相場の流れを引くとどう変わるか 明けの明星は買いしか、宵の明星は売りしか作りません。上げた年に買い続ければ、パターンと関係なくプラスになります。そこで**1取引あたりの損益から、同じ10本を無条件に持っていたら得られた分を引きました。** | 時間足 | 2025年の10本あたり | 2024年の10本あたり | |---|---|---| | 15分足 | -0.03pips | +0.66pips | | 1時間足 | -0.07pips | +2.53pips | | 4時間足 | -0.48pips | +9.49pips | 2024年は上げた年なので、4時間足では何もしなくても10本で平均9.49pips得られました。宵の明星の2024年の成績には、この逆風がそのまま入っています。 1時間足の宵の明星は、2024年に1取引あたり−9.58pipsでした。ここから−2.53pipsを引くと**−7.05pips**です。流れを引いても赤字は消えません。 | 条件 | 2025年 1取引あたり | 流れを引くと | 2024年 1取引あたり | 流れを引くと | |---|---|---|---|---| | 明けの明星だけ | -0.42pips | -0.35pips | -1.61pips | -4.14pips | | 明けの明星＋ATR1.5倍以上 | +4.78pips | +4.85pips | +6.89pips | +4.36pips | | 明けの明星＋50本平均の向き | +1.31pips | +1.38pips | +0.93pips | -1.61pips | | 宵の明星だけ | +5.08pips | +5.01pips | -9.58pips | -7.05pips | | 宵の明星＋実体0.7以上 | +2.56pips | +2.49pips | -15.91pips | -13.37pips | **流れを引いたあと、1取引あたりが2年ともプラスだったのは、時間足3種×10通り＝30通りのうち4通りです。** 15分足の明けの明星＋実体0.7以上、1時間足の明けの明星＋ATR1.5倍以上、4時間足の明けの明星＋実体0.7以上、4時間足の宵の明星＋実体0.7以上でした。いずれも年16〜80回の売買です。 ## 前年に一番稼いだ条件を翌年へ移すと | 時間足 | 2024年の1位 | 2024年の年間損益 | 2025年の年間損益 | 2025年の順位 | 2025年の10通りの中央値 | |---|---|---|---|---|---| | 15分足 | 明けの明星＋実体0.7以上 | +418.3pips | +23.5pips | 7位 | +135.6pips | | 1時間足 | 明けの明星＋実体0.7以上 | +510.2pips | -93.8pips | 8位 | +128.5pips | | 4時間足 | 宵の明星＋実体0.7以上 | +772.7pips | +612.6pips | 1位 | -65.6pips | 15分足と1時間足は翌年に中央値を下回りました。4時間足だけは1位のままですが、2025年は17回、2024年は15回の売買です。 ## 損切り・利確とコスト ### 損切りと利確を置くと | エントリー | 損切り・利確 | 2025年 回数 | 2025年 勝率 | 2025年 損益 | 2024年 損益 | |---|---|---|---|---|---| | 明けの明星 | なし | 164回 | 48.17% | -68.5pips | -281.3pips | | 明けの明星 | 30・60 | 187回 | 34.22% | -510.1pips | +153.6pips | | 明けの明星 | 60・30 | 176回 | 56.82% | -427.1pips | +372.0pips | | 明けの明星 | 50・50 | 178回 | 49.44% | -66.7pips | -51.6pips | | 宵の明星 | なし | 161回 | 51.55% | +817.6pips | -1619.7pips | | 宵の明星 | 30・60 | 186回 | 39.78% | +426.0pips | -913.7pips | | 宵の明星 | 60・30 | 180回 | 61.67% | +160.8pips | -1038.4pips | | 宵の明星 | 50・50 | 177回 | 48.59% | +121.4pips | -1262.2pips | **宵の明星に損切り60・利確30を付けると勝率は61.67%になりますが、2024年は−1,038.4pipsです。** 入る足は同じで、勝率は決済の比で決まります。 ### コスト | 時間足 | 条件 | 取引回数 | 支払ったスプレッド | 回数×0.3pips | |---|---|---|---|---| | 15分足 | 両方（10本で決済） | 1043回 | 312.8pips | 312.9pips | | 15分足 | 明けの明星だけ | 664回 | 199.2pips | 199.2pips | | 1時間足 | 両方（10本で決済） | 249回 | 74.7pips | 74.7pips | | 1時間足 | 明けの明星だけ | 164回 | 49.2pips | 49.2pips | | 4時間足 | 明けの明星だけ | 38回 | 11.4pips | 11.4pips | 支払ったスプレッドは取引回数×0.3pipsと一致しました。1時間足の明けの明星はスプレッドを0にしても−19.3pipsで、回数の多さではなく入り方から赤字になっています。 ## 関連記事 - [ローソク足パターン17種の勝率](/ja/blog/candlestick-patterns-tested)：この2つを含む17種を、同じ2年・同じ対照群で横断的に並べています - [包み足の勝率とだまし](/ja/blog/engulfing-settings)：2本でできる反転の形を、色が変わっただけの足と比べています - [ダブルトップ・ダブルボトムは反転する？](/ja/blog/double-top-bottom-settings)：もっと大きな反転の形を同じ2年で検証しています - [フェアバリューギャップは本当に埋まるのか](/ja/blog/fair-value-gap-settings)：この記事で数えた「窓」を、条件として売買した場合を測っています - [ドル円の設定は他の通貨ペアで通用する？3,024通り検証](/ja/blog/settings-across-pairs)：この設定は2023年の4時間足でドル円1位で、ユーロドル・ポンドドル・ユーロ円のすべてでも黒字になった数少ない例です ## 補足 対象は米ドル／円の2024年1月1日から2025年12月31日です。終値約定、スプレッド0.3pips固定、0.1ロットで計算しました。成立の判定にはFormiqのバックテスト機能と同じ処理を使い、対象期間の前の足も読み込んでいます。1本後から20本後までを見るため、年の終わりからその本数ぶんは対象から外しています。窓の判定は、実体の窓が始値と終値の位置、値幅の窓が高値と安値の位置で行いました。相場の流れを引く計算では、対象期間の全部の足について10本後の終値との差を平均し、買いからは引き、売りには足しています。 --- ### ストキャスティクス設定：2,700通りでスロー化を検証 URL: https://formiq.jp/ja/blog/stochastic-settings Language: ja Published: 2026-09-09 Updated: 2026-09-09 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ストキャスティクス, 設定, 勝率, オシレーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/stochastic-settings ストキャスティクスは、過去の高値・安値の幅に対して、終値がどの位置にあるかを示す指標です。下の別枠に％Kと％Dを表示し、20や80の水準、線の交差を売買の判断に使います。 設定で混同しやすいのは、**％Kの期間と、％Kを滑らかにする期間が別の項目であること**です。％Dも加わるため、期間の数字だけを見ても、どの計算を遅くしたのかは分かりません。 ドル円2,700通りを測ると、**1時間足のクロス読みは75通り中43通りが両年黒字**でした。スロー化の期間20に限れば15通りすべてが両年黒字です。一方、売り逆張りはどの時間足でも両年黒字がありませんでした。 ## ストキャスティクスは何を測っているのか 平滑化前の％Kは「終値から期間内の最安値を引き、期間内の最高値と最安値の差で割って100を掛けた値」です。終値が期間内の高値に近ければ100に、安値に近ければ0に近づきます。 この記事では、各足で計算した％Kをスロー化の期間で単純平均し、その線を％Kとして読みます。％Dは、平滑化した％Kを％D期間でさらに単純平均した線です。期間内の高値と安値が同じ場合、平滑化前の値は50とします。 **高い水準は、下落開始を意味しません。** 価格が上昇しながら高値付近で終わる足が続けば、％Kも高いままです。80を超えたという位置の情報と、売ったあとに下がるかという問いは分けて測る必要があります。 ## ストキャスティクスの設定と読み方 Formiqのインジケーター一覧からストキャスティクスを追加すると、ローソク足の下の別枠に表示されます。設定画面の計算・水準項目は次のとおりです。 | 項目 | 初期設定 | 意味 | | --- | --- | --- | | %K期間 | 14 | 高値・安値を探す本数 | | %D期間 | 3 | 平滑化した％Kを平均する本数 | | スローイング | 3 | 平滑化前の％Kを平均する本数。本文ではスロー化の期間と呼ぶ | | 上の水準 | 80 | 別枠に描く上側の目安 | | 下の水準 | 20 | 別枠に描く下側の目安 | ％Kの初期色は黄色、％Dは青色です。線の色と太さも変更できますが、計算値は変わりません。20と80の水準線は表示されます。チャートの水準線とバックテストの売買条件は別に設定します。 本文の**14/3/3は、％K期間／％D期間／スロー化の期間**の順です。スロー化の期間1なら追加の平均を取らず、期間を長くするほど短い変化が平均に吸収されます。％D期間を長くすると、％Kを変えずに％Dだけが滑らかになります。 今回は水準による逆張りとクロスを、別々の売買ルールとして測りました。 | 売買ルール | エントリー | 決済 | | --- | --- | --- | | 買い逆張り | ％Kが20未満なら買い | ％Kが80を超えたら決済 | | 売り逆張り | ％Kが80を超えたら売り | ％Kが20未満になったら決済 | | クロス | ％Kが％Dを上抜けたら買い、下抜けたら売り | 反対向きの交差で決済 | 水準読みは、水準内へ戻るのを待つルールではありません。水準の外にある状態でエントリーします。クロスは交差した足でのみ判定し、20/80の制限を付けません。買いと売りの逆張りは別々に実行し、損益を合算した戦略としては扱いません。 ## 検証の条件 Formiqのバックテスト機能で実測しました。 | 項目 | 条件 | | --- | --- | | 通貨ペア | 米ドル／円 | | 検証期間 | 2025-01-01〜2025-12-31 | | 年を変えた再検証 | 2024-01-01〜2024-12-31 | | 年内の分割 | 2025-01-01〜06-30、2025-07-01〜12-31。各期間を独立して実行 | | 時間足 | 15分足・1時間足・4時間足 | | %K期間 | 5・9・14・21・50 | | スロー化の期間 | 1・3・5・10・20 | | クロスの%D期間 | 2・3・5 | | 逆張りの水準ペア | 10/90・20/80・30/70。%D期間は3に固定 | | 設定数 | 各売買ルール75通り。3ルール×3時間足×4期間で2,700通り | | 決済 | クロスは反対の交差。逆張りは反対側の水準。期間末の保有分は強制決済 | | 約定と費用 | 終値約定、スプレッド0.3pips固定、スリッページ0 | | 数量 | 0.1ロット | | 損切り・利確 | 基本検証では使わず、追加検証で比較 | 水準読みでは売買に使わない％D期間を動かして件数を増やしていません。損益は約定価格からpipsで計算しています。フィルター・損切り・利確・スプレッドを変えた追加検証は、2,700通りに含めません。 ## 遅くすると20や80に届かなくなるのか 1時間足・2025年・％K期間14について、有効な6,226本のうち何本が水準の外にあったかを数えました。％D期間は3です。 | スロー化の期間 | 20未満の足 | 80超の足 | 20未満の割合 | 80超の割合 | | --- | --- | --- | --- | --- | | 1 | 1,238 | 1,534 | 19.88% | 24.64% | | 3 | 1,091 | 1,480 | 17.52% | 23.77% | | 5 | 969 | 1,323 | 15.56% | 21.25% | | 10 | 689 | 1,022 | 11.07% | 16.42% | | 20 | 280 | 512 | 4.50% | 8.22% | スロー化の期間を1から20へ伸ばすと、20未満の足は1,238本から280本に減りました。ただし、**期間20でも20未満と80超の足は残っています**。「遅くすれば水準に届かない」と一律には言えません。 この表はエントリーの回数ではありません。保有中に水準の外へ出ても新しい取引は作られず、同じ水準の外に留まる複数の足も数えています。 ## どの時間足で利益が続いたか 各売買ルールを75通りずつ比べた結果です。取引がなく損益0の設定は、黒字に含めません。 | 時間足 | 売買ルール | 2024年の黒字設定 | 2025年の黒字設定 | 両年黒字 | 2025年の年間損益中央値（pips） | | --- | --- | --- | --- | --- | --- | | 15分足 | クロス | 2/75 | 4/75 | 1/75 | −2,190.0 | | 15分足 | 買い逆張り | 28/75 | 17/75 | 11/75 | −712.6 | | 15分足 | 売り逆張り | 0/75 | 16/75 | 0/75 | −662.4 | | 1時間足 | クロス | 49/75 | 52/75 | 43/75 | +834.5 | | 1時間足 | 買い逆張り | 42/75 | 36/75 | 26/75 | 0.0 | | 1時間足 | 売り逆張り | 2/75 | 35/75 | 0/75 | −145.5 | | 4時間足 | クロス | 62/75 | 67/75 | 56/75 | +1,143.7 | | 4時間足 | 買い逆張り | 29/75 | 59/75 | 18/75 | +643.2 | | 4時間足 | 売り逆張り | 1/75 | 56/75 | 0/75 | +560.7 | 15分足のクロスは2025年の年間損益中央値が−2,190.0pipsでした。4時間足は+1,143.7pipsで、両年黒字の設定も多くなっています。水準の逆張りでは、買いと売りを分けることが欠かせません。**売り逆張りは全時間足で両年黒字が0通り**でした。 1時間足のクロスを2025年の前半と後半に分けると、黒字はそれぞれ75通り中54通りと44通りです。年間の黒字設定数だけでは、途中の損失期間は分かりません。なお、保有分を各期間末に決済するため、前後半を足しても年間損益と厳密には一致しません。 ## 1時間足ではスロー化を長くするべきか ％K期間と％D期間の組み合わせを揃え、スロー化の期間だけを比較します。各行は％K期間5種×％D期間3種の15通りです。 | スロー化の期間 | 両年黒字 | 2024年の年間損益平均（pips） | 2025年の年間損益平均（pips） | | --- | --- | --- | --- | | 1 | 2/15 | −1,121.3 | −1,347.3 | | 3 | 6/15 | −391.9 | +492.3 | | 5 | 8/15 | +592.7 | −40.5 | | 10 | 12/15 | +2,061.8 | +1,295.9 | | 20 | 15/15 | +3,443.8 | +1,744.1 | スロー化の期間1では両年黒字が2通り、期間20では15通りでした。ただし、2025年の年間損益平均は期間3の+492.3pipsから期間5の−40.5pipsへ下がっています。期間を少し長くするたびに改善したわけではありません。 ### 買い逆張りは期間5で両年黒字が最多 逆張りは％Dの代わりに水準ペアを変えています。各行は％K期間5種×水準ペア3種の15通りです。 | スロー化の期間 | 両年黒字 | 2024年の年間損益平均（pips） | 2025年の年間損益平均（pips） | | --- | --- | --- | --- | | 1 | 2/15 | −225.3 | −745.9 | | 3 | 3/15 | −57.9 | −387.7 | | 5 | 8/15 | +217.0 | −186.3 | | 10 | 7/15 | +258.5 | +347.9 | | 20 | 6/15 | +231.1 | +369.6 | 買い逆張りで両年黒字が最も多かったのは期間5の8通りで、期間20では6通りでした。クロスで得られた「期間20の15通りが両年黒字」という結果は、逆張りにそのまま当てはまりません。 売り逆張りでは、スロー化の期間を変えても両年黒字はありませんでした。 | スロー化の期間 | 両年黒字 | 2024年の年間損益平均（pips） | 2025年の年間損益平均（pips） | | --- | --- | --- | --- | | 1 | 0/15 | −1,833.6 | −657.6 | | 3 | 0/15 | −1,630.1 | −318.9 | | 5 | 0/15 | −1,338.3 | −113.9 | | 10 | 0/15 | −1,184.6 | +301.6 | | 20 | 0/15 | −941.4 | +201.6 | ## 初期設定14/3/3は1時間足で翌年赤字 クロス読みの初期設定を、時間足と年ごとに並べます。 | 時間足 | 年 | 取引回数 | 勝率 | 年間損益（pips） | | --- | --- | --- | --- | --- | | 15分足 | 2024 | 6,136 | 37.24% | −1,547.6 | | 15分足 | 2025 | 6,051 | 38.13% | −1,661.5 | | 1時間足 | 2024 | 1,525 | 38.36% | +686.6 | | 1時間足 | 2025 | 1,510 | 37.35% | −49.7 | | 4時間足 | 2024 | 359 | 43.18% | +3,093.0 | | 4時間足 | 2025 | 383 | 42.82% | +2,959.6 | 1時間足の14/3/3は、2024年には黒字でしたが2025年は−49.7pipsになりました。4時間足は両年とも黒字です。同じ期間の数字でも、使う時間足まで決めてから検証する必要があります。 ## 前年にいちばん利益が出た設定は翌年も使えたか クロスの年間損益が最大だった設定を選び、設定を変えずに別の年へ適用しました。2024年で選んで2025年を測る行が、時系列に沿った確認です。逆方向の行は、2025年にいちばん利益が出た設定が過去にも利益を出していたかを調べています。 | 時間足 | 設定を選んだ年 | %K/%D/スロー化 | 選択年の年間損益（pips） | 再検証年 | 取引回数 | 再検証年の年間損益（pips） | 再検証年の中央値（pips） | | --- | --- | --- | --- | --- | --- | --- | --- | | 1時間足 | 2024 | 21/3/20 | +5,095.2 | 2025 | 657 | +1,920.5 | +834.5 | | 1時間足 | 2025 | 14/3/20 | +3,211.3 | 2024 | 745 | +3,323.9 | +799.4 | | 4時間足 | 2024 | 50/3/5 | +4,481.9 | 2025 | 263 | +598.0 | +1,143.7 | | 4時間足 | 2025 | 21/5/10 | +3,354.8 | 2024 | 175 | −1,686.5 | +1,857.5 | 1時間足では、2024年にいちばん利益が出た設定21/3/20が2025年も+1,920.5pipsとなり、中央値+834.5pipsを上回りました。ところが4時間足の2024年にいちばん利益が出た設定50/3/5は、2025年に+598.0pipsで中央値+1,143.7pipsを下回っています。 4時間足で2025年に最も利益が大きかった21/5/10も、2024年へ戻すと−1,686.5pipsでした。設定を多数比較したあとの最大の年間損益には選択の偏りがあるため、別の年の成績を併記しています。 ## 勝率60%台でも逆張りは赤字だった 1時間足・2025年・14/3/3を比較します。逆張りの水準は20/80です。 | 売買ルール | 取引回数 | 勝率 | 平均利益（pips） | 平均損失（pips） | 年間損益（pips） | | --- | --- | --- | --- | --- | --- | | クロス | 1,510 | 37.35% | +32.72 | −19.56 | −49.7 | | 買い逆張り | 106 | 60.38% | +46.83 | −85.05 | −574.9 | | 売り逆張り | 107 | 61.68% | +53.62 | −98.85 | −513.9 | 買い逆張りは106回で勝率60.38%ですが、平均利益+46.83pipsに対して平均損失は−85.05pipsでした。勝った回数が多くても、負けたときの幅が大きく、年間損益は−574.9pipsです。売り逆張りも107回で勝率61.68%ながら赤字でした。 買い逆張りと売り逆張りは、相場の上昇・下落の影響をそれぞれ受けます。買い側の黒字を、そのまま「水準に反転を予測する効果があった」とは解釈していません。 ## フィルターと損切りで赤字は減ったか 1時間足のクロス14/3/3に条件を追加しました。ADXは期間14で方向を指定せず、エントリー時だけ判定します。決済は反対のクロスを保ちます。時間帯の制限も新規エントリーだけです。 | 追加条件 | 年 | 取引回数 | 年間損益（pips） | | --- | --- | --- | --- | | 追加条件なし | 2024 | 1,525 | +686.6 | | 追加条件なし | 2025 | 1,510 | −49.7 | | ADX ≥ 20 | 2024 | 1,005 | +355.7 | | ADX ≥ 20 | 2025 | 1,007 | +136.3 | | ADX ≥ 25 | 2024 | 698 | +194.0 | | ADX ≥ 25 | 2025 | 705 | +639.6 | | UTC 8〜16時 | 2024 | 524 | −360.9 | | UTC 8〜16時 | 2025 | 520 | +1,132.3 | | 損切り30・利確60 | 2024 | 1,351 | −620.9 | | 損切り30・利確60 | 2025 | 1,327 | −1,622.4 | | 損切り100・利確200 | 2024 | 1,505 | +356.4 | | 損切り100・利確200 | 2025 | 1,503 | −41.5 | ADXを25以上に絞ると、2025年は705回で+639.6pipsになりましたが、2024年の年間損益は追加条件なしより小さくなりました。UTCの時間帯制限は2025年を黒字にしても、2024年を赤字にしています。改善した年だけを抜き出すと、フィルターの評価が変わってしまいます。 損切り30・利確60では両年とも赤字です。損切り・利確に届く前に反対のクロスが出れば、シグナルによって決済します。両方の価格に同じ足で届く場合は、損切りを先に扱います。 ### 初期設定のクロスはスプレッド0.3で黒字が消えた 1時間足・2025年・14/3/3の年間損益です。表の費用と損益の単位はいずれもpipsです。 | 売買ルール | 取引回数 | スプレッド0の年間損益 | スプレッド0.3の年間損益 | スプレッド1.0の年間損益 | | --- | --- | --- | --- | --- | | クロス | 1,510 | +403.4 | −49.7 | −1,106.7 | | 買い逆張り | 106 | −543.1 | −574.9 | −649.1 | | 売り逆張り | 107 | −481.8 | −513.9 | −588.8 | クロスはスプレッド0なら+403.4pipsですが、1,510回の取引に費用が掛かると、0.3pipsでは−49.7pipsになりました。損益分岐スプレッドは小数第2位で0.27pipsです。この値は表示精度に整えた値であり、0.27pipsまで必ず黒字という意味ではありません。 逆張りは買いも売りもスプレッド0で赤字でした。逆張りの損失は、スプレッドだけでは説明できません。 ## 関連記事 - [ストキャスRSIの設定検証](/ja/blog/stoch-rsi-settings)：価格の高安幅ではなく、RSIの変動幅を使う指標 - [RSIの水準と期間の検証](/ja/blog/rsi-settings)：水準による逆張りとクロスを分けて比較 - [移動平均線乖離率の検証](/ja/blog/ma-disparity-settings)：売買の前に水準への到達頻度を確認 ## 補足 対象はドル円の2024年と2025年で、終値約定・固定スプレッドの検証です。実際のスプレッド拡大、約定のずれ、スワップは含めていません。計算には開始日前のローソク足も渡し、期間末に残ったポジションは決済しています。％K・スロー化・％Dを計算できる本数に加えて、交差を判定する過去の足が必要です。 --- ### ダブルトップ・ダブルボトムは反転する？2年分検証した結果 URL: https://formiq.jp/ja/blog/double-top-bottom-settings Language: ja Published: 2026-09-05 Updated: 2026-09-05 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ダブルトップ, ダブルボトム, チャートパターン, ネックライン, 反転, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/double-top-bottom-settings ダブルトップは、近い価格に2つの高値があり、その間に安値がある形です。ダブルボトムは上下を反転させた形で、2つの安値と中間高値からできています。どちらも、ネックラインを抜けたあとに値動きが反転方向へ続くと読まれます。 問題は、チャートを後から見ると高値と安値がはっきり見えることです。実際の時点では、右側の足が閉じるまでスイング高値・安値は確定しません。この記事では右側の確認を待ち、ダブルトップなら終値が中間安値を下回った足、ダブルボトムなら終値が中間高値を上回った足から測りました。 スイング確認を左右5本にすると、ネックライン抜けから10本後に反転方向へ進んだ割合が2024年と2025年の両方で50%を超えたのは1時間足だけでした。15分足は49.71%と50.15%、4時間足は46.88%と31.43%です。 ## ダブルトップ・ダブルボトムとは 計算式を持つオシレーターではなく、3つの価格を結ぶチャートパターンです。 ダブルトップは次の順で判定しました。 1. 左右に同じ本数だけ、自分より高い高値がないスイング高値を探す 2. 2つのスイング高値が近い価格にあり、その間にスイング安値があることを確認する 3. 中間安値をネックラインにする 4. 終値がネックラインを下回ったら形が完成したと扱う ダブルボトムは高値と安値を入れ替えます。中間高値がネックラインになり、終値が上回った足で完成です。 2つの高値・安値の許容差はATR(14)の1.2倍を使い、価格の0.1〜1.0%に収めました。パターンの高さは0.3 ATR以上、2つのスイングの間隔は最大300本です。これらは現在のチャート表示と同じ判定です。 ### 形の確定とネックライン抜けは別の時点 スイング確認5本なら、2つ目の高値・安値は5本後まで確定しません。Swing Periodが30なら30本後です。15分足の30本は7時間30分、1時間足は30時間、4時間足は120時間に相当します。 形を見つけた時点で売買する読み方と、ネックラインを終値で抜けるまで待つ読み方を分けました。後から確定したスイングを高値・安値の足へさかのぼって売買する処理は入れていません。 ## ダブルトップ・ダブルボトムの設定と読み方 Formiqでは、ローソク足と同じ価格チャート上へ表示します。 | 環境 | 表示方法 | |---|---| | Formiq | インジケーター一覧 → チャートパターン | | TradingView | インジケーター → Technicals → Patterns。Double Top / Double BottomまたはAll Chart Patterns | | MT4 / MT5 | 高値・安値とネックラインへ水平線やトレンドラインを手動で引く | [TradingViewの公式説明](https://www.tradingview.com/support/solutions/43000653211-chart-pattern-double-top/)でも、終値が中間安値を下回った時点をダブルトップの完成としています。2つ目の高値が見えただけの段階とは分けられています。 ### 設定はSwing Periodの1項目 | 項目 | 初期設定 | 選択範囲 | 意味 | |---|---:|---:|---| | Swing Period | 30 | 5〜100 | 高値・安値の左右で、何本ずつ比較してスイングを確定するか | 数値を小さくすると小さな上下動まで拾うため、表示回数が増えます。数値を大きくすると大きな形だけが残りますが、確定までの待ち時間も長くなります。 チャートではダブルトップを赤、ダブルボトムを緑で表示します。2つの高値・安値を結ぶ線、破線のネックライン、緑の点線で目標価格が描かれます。目標価格は、ネックラインからパターンの高さと同じ値幅だけ反転方向へ進んだ水準です。 ### 一般的な売買ルール | パターン | 売買開始 | 目標 | 形の失敗を判断する位置 | |---|---|---|---| | ダブルトップ | 終値が中間安値を下回ったら売る | ネックラインからパターンの高さぶん下 | 2つの高値の平均より、パターン高の20%ぶん上 | | ダブルボトム | 終値が中間高値を上回ったら買う | ネックラインからパターンの高さぶん上 | 2つの安値の平均より、パターン高の20%ぶん下 | 売買検証では、ダブルトップで売り、ダブルボトムで買いました。反対側のパターンが成立したら決済し、同じ足では新しい売買を始めていません。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01〜2025-12-31。比較用に2024年も同条件 | | 時間足 | 15分足 / 1時間足 / 4時間足 | | Swing Period | 5 / 10 / 15 / 20 / 30 / 40 / 50 / 75 | | 売買開始 | 形の確定時 / ネックラインの終値抜け後 | | 組み合わせ | 8確認本数 × 2開始時点 = 16設定。3時間足 × 4期間で192行 | | 決済 | 反対側のパターンまで。損切り・利確なし | | スプレッド | 0.3pips固定。約定は終値 | | ロット | 0.1ロット | | 先読み防止 | スイングは右側の確認本数が閉じた後だけ利用 | 10本後の値動きは、ダブルトップなら下落、ダブルボトムなら上昇をプラスとして集計しました。2024年と2025年の年間集計に加え、2025年を前半と後半にも分けています。 ## 本当に反転するのか 10本後の反転率が両年とも50%を超えたのは、1時間足だけでした。スイング確認5本でネックラインを抜けたあと、10本保有した値動きです。売買の重複を除く前に、パターン1件ずつを数えています。 | 時間足 | 年 | パターン数 | 反転方向へ進んだ割合 | 10本後の平均値幅 | |---|---:|---:|---:|---:| | 15分足 | 2024 | 690 | 49.71% | +0.47pips | | 15分足 | 2025 | 680 | 50.15% | +1.27pips | | 1時間足 | 2024 | 124 | 56.45% | +1.82pips | | 1時間足 | 2025 | 162 | 51.85% | +4.66pips | | 4時間足 | 2024 | 32 | 46.88% | −6.06pips | | 4時間足 | 2025 | 35 | 31.43% | −19.14pips | 1時間足は両年とも50%を超えました。ただしダブルトップとダブルボトムの内訳は同じではありません。1時間足・2024年の10本後は、ダブルトップ55件が43.64%・平均−11.86pips、ダブルボトム69件が66.67%・平均+12.73pipsでした。ダブルトップとダブルボトムをまとめた56.45%だけでは、この差が隠れます。 4時間足は反転を支持しません。2025年はダブルトップ20件が35.00%、ダブルボトム15件が26.67%で、どちらも平均値幅が反転方向と逆でした。 ## 目標まで届くのはどれくらいか 15分足と1時間足は両年とも64〜66%で、4時間足だけ年をまたげませんでした。ネックラインからパターンの高さと同じ値幅へ置く目標と、反対側の失敗水準のどちらへ先に触れたかを300本まで追いました。スイング確認5本です。 | 時間足 | 年 | 完成した形 | 目標へ先に到達 | 失敗水準が先 | 同じ足 / 未到達 | |---|---:|---:|---:|---:|---:| | 15分足 | 2024 | 690 | 443（64.20%） | 238 | 7 / 2 | | 15分足 | 2025 | 681 | 439（64.46%） | 238 | 0 / 4 | | 1時間足 | 2024 | 124 | 81（65.32%） | 42 | 1 / 0 | | 1時間足 | 2025 | 162 | 107（66.05%） | 54 | 0 / 1 | | 4時間足 | 2024 | 32 | 23（71.88%） | 9 | 0 / 0 | | 4時間足 | 2025 | 36 | 18（50.00%） | 15 | 0 / 3 | 15分足と1時間足は両年とも64〜66%でした。ただし、目標までの距離はネックラインからパターン高の1倍、失敗水準までの距離は約1.2倍です。目標到達率だけを勝率として読んでも、同じ利益幅と損失幅にはなりません。 ## Swing Periodは何本がいいか 伸ばすと4時間足の件数が先になくなります。初期設定の30と、最小値5を比べます。表はネックライン抜け後に売買し、反対側のパターンで決済した年間損益です。 | 時間足 | Swing Period | 2024年の取引回数 | 2024年の年間損益 | 2025年の取引回数 | 2025年の年間損益 | |---|---:|---:|---:|---:|---:| | 15分足 | 5 | 225 | +459.3pips | 225 | +275.1pips | | 15分足 | 30 | 19 | −1,405.4pips | 15 | −284.5pips | | 1時間足 | 5 | 37 | +613.3pips | 53 | −562.7pips | | 1時間足 | 30 | 3 | +677.6pips | 5 | −235.6pips | | 4時間足 | 5 | 12 | −187.2pips | 13 | −763.2pips | | 4時間足 | 30 | 0 | 0.0pips | 3 | +225.3pips | Swing Period 30は、4時間足・2024年で完成形が0件でした。4時間足・2025年も3取引しかありません。+225.3pipsや勝率33.33%を、15分足225取引の数字と同じ重さでは扱えません。 8通りのSwing Periodで両年とも黒字だったのは、ネックライン抜け後の15分足が2/8、1時間足が0/8、4時間足が2/8です。4時間足の2設定も取引回数が少ないため、黒字設定数だけで4時間足を選ぶ根拠にはなりません。 ## ネックラインを待つほうが安全か 15分足では損益の向きが入れ替わりましたが、1時間足と4時間足では同じ改善が起きていません。Swing Period 5で、売買を始める時点だけを変えました。 | 時間足 | 年 | 形の確定時：取引回数 / 年間損益 | ネックライン後：取引回数 / 年間損益 | |---|---:|---:|---:| | 15分足 | 2024 | 499 / −589.9pips | 225 / +459.3pips | | 15分足 | 2025 | 469 / −195.4pips | 225 / +275.1pips | | 1時間足 | 2024 | 86 / +1,135.1pips | 37 / +613.3pips | | 1時間足 | 2025 | 105 / −404.7pips | 53 / −562.7pips | | 4時間足 | 2024 | 21 / −1,667.2pips | 12 / −187.2pips | | 4時間足 | 2025 | 27 / −643.3pips | 13 / −763.2pips | ネックラインを待つと15分足は両年とも赤字から黒字へ変わりました。1時間足と4時間足では同じ改善が起きていません。ネックライン確認は売買回数をほぼ半分にしますが、待てばどの時間足でも安全になるわけではありません。 ## 去年よかった本数は今年も勝てるか 移した先でも黒字だったのは6通り中3通りです。2024年にいちばん良かった確認本数を2025年へ、2025年にいちばん良かった本数を2024年へ移しました。売買開始はネックライン抜け後です。 | 時間足 | 選んだ年 | Swing Period | 選んだ年の年間損益 | 移した年の年間損益 | 2025年中央値 | |---|---:|---:|---:|---:|---:| | 15分足 | 2024 | 5 | +459.3pips | 2025年 +275.1pips | −360.6pips | | 15分足 | 2025 | 10 | +685.5pips | 2024年 +288.3pips | −360.6pips | | 1時間足 | 2024 | 30 | +677.6pips | 2025年 −235.6pips | −262.6pips | | 1時間足 | 2025 | 50 | +1,875.0pips | 2024年 −443.2pips | −262.6pips | | 4時間足 | 2024 | 10 | +777.5pips | 2025年 +676.7pips | +568.8pips | | 4時間足 | 2025 | 15 | +886.9pips | 2024年 −809.1pips | +568.8pips | 移した先でも黒字だったのは6通り中3通りです。2025年を前半と後半に分けると、両方で黒字だったのは24区分中5区分でした。内訳は15分足2/8、1時間足1/8、4時間足2/8です。 ## 勝率60%でも負けるのはなぜか 1回あたりの勝ち幅が負け幅に届かないからです。初期設定のSwing Period 30でネックラインを待った実測です。 | 時間足 | 年 | 取引回数 | 勝率 | 平均利益 | 平均損失 | 年間損益 | |---|---:|---:|---:|---:|---:|---:| | 15分足 | 2025 | 15 | 33.33% | +227.16pips | −142.03pips | −284.5pips | | 15分足 | 2024 | 19 | 31.58% | +195.01pips | −198.11pips | −1,405.4pips | | 1時間足 | 2025 | 5 | 60.00% | +274.30pips | −529.25pips | −235.6pips | | 1時間足 | 2024 | 3 | 66.67% | +544.30pips | −411.00pips | +677.6pips | | 4時間足 | 2025 | 3 | 33.33% | +687.90pips | −231.27pips | +225.3pips | | 4時間足 | 2024 | 0 | 対象外 | 対象外 | 対象外 | 0.0pips | 1時間足・2025年は勝率60%でも−235.6pipsです。利益3回の平均が+274.30pips、損失2回の平均が−529.25pipsだったためです。5取引しかないので、勝率を設定選びに使える標本でもありません。 ## ADXや時間帯で改善するか 年をまたいで同じ改善にはなりませんでした。取引回数を確保するため、Swing Period 5・ネックライン抜け後へフィルターを足しました。 15分足にADX(14)が20以上を加えると、2024年は+459.3pipsから+1,088.1pipsへ増えましたが、2025年は+275.1pipsから−1,005.1pipsへ変わりました。1時間足の東京時間（00〜09 UTC）は2024年25取引・+684.2pips、2025年32取引・+679.1pipsでした。ただし、5つの候補を見たあとで残った時間帯なので、別期間での検定前に採用できる数字ではありません。 固定の損切り30pips・利確60pipsも時間足で結果が分かれました。 | 時間足 | 2024年 | 2025年 | |---|---:|---:| | 15分足 | +1,345.8pips（393取引） | +103.9pips（425取引） | | 1時間足 | +67.1pips（101取引） | +297.2pips（141取引） | | 4時間足 | −39.3pips（31取引） | −258.3pips（36取引） | 同じ30/60pipsは、15分足では広め、4時間足では狭めです。4時間足の赤字を、ダブルトップやダブルボトムそのものと固定幅のどちらが作ったかは、この表だけでは分けられません。 ## スプレッドはどこまで耐えられるか 15分足の損益分岐は1.52pipsでした。Swing Period 5・ネックライン抜け後の15分足は、2025年に225取引ありました。スプレッド0で+342.6pips、0.3pipsで+275.1pips、1.0pipsで+117.6pips、2.0pipsで−107.4pipsです。 取引1回につきスプレッドが1回分かかるため、損益分岐は342.6÷225＝1.52pipsです。2024年はスプレッド2.0pipsでも+76.8pipsでした。コストへの耐性も年によって同じではありません。 ## 関連記事 - [SMCの構造ブレイクと流動性スイープ](/ja/blog/smc-settings)：同じスイング高値・安値を使い、抜けとだましを別々に検証しています - [フェアバリューギャップは本当に埋まるのか](/ja/blog/fair-value-gap-settings)：有名なプライスアクションを通常の足と比較しています - [フィボナッチ・リトレースメントの水準検証](/ja/blog/fibonacci-retracement-levels)：スイング確定の遅れと、目標水準への反応を測っています - [ローソク足パターン17種の勝率](/ja/blog/candlestick-patterns-tested)：1〜3本でできる形を、同じ2年・同じ対照群で並べています - [水平線トレードは本当に勝てる？5,184通り検証](/ja/blog/support-resistance-settings)：同じスイングから作る水平線を、反発と抜けの2つの読み方で測っています ## 補足 対象はUSDJPYの2024年1月1日から2025年12月31日です。終値約定、スプレッド0.3pips、0.1ロットで計算しました。スイング判定には対象期間前の足を使っています。右側の確認本数が閉じる前の高値・安値は利用していません。同じ足で目標と失敗水準の両方へ触れた場合は順序を決められないため、別区分にしています。 --- ### 【FX検証】SMCは使える？1,032条件で検証 URL: https://formiq.jp/ja/blog/smc-settings Language: ja Published: 2026-09-04 Updated: 2026-09-04 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: SMC, Smart Money Concepts, BOS, オーダーブロック, 流動性スイープ, 市場構造, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/smc-settings SMC（Smart Money Concepts）は、値動きを市場構造・流動性スイープ・オーダーブロック・フェアバリューギャップ（FVG）などに分けて読む考え方です。ただし、RSIのように計算式が1つに決まったインジケーターではありません。同じBOSやオーダーブロックでも、山谷の決め方やヒゲを含めるかで結果が変わります。 そこで今回は、OHLCだけで同じ答えを再計算できる3つに絞りました。**確定済みの山谷を終値で抜ける構造ブレイク**、**ヒゲだけが山谷を抜けて終値が戻る流動性スイープ**、**値動きが確定した後に直前の反対色ローソク帯へ初めて戻るオーダーブロック**です。機関投資家の注文を観測したのではなく、SMCで使われる名前を価格パターンへ置き換えた検証です。 ドル円の2024年と2025年を調べると、**1時間足の構造ブレイクは12設定中11設定が両方の年で黒字**でした。オーダーブロックの検出定義は、15分足で108設定中40設定が両年黒字でしたが、1時間足は4設定です。流動性スイープは15分足と1時間足で両年黒字が0設定でした。 ## SMCは何を測っているのか SMCには統一された計算式がありません。この記事では、まず山谷を次のように固定しました。 1. ある足の高値が、左右に置いた同じ本数の足よりすべて高ければ山とする 2. ある足の安値が、左右に置いた同じ本数の足よりすべて低ければ谷とする 3. 右側の足が閉じた時点で初めて、その山谷を売買に使う 4. 同値があれば、どちらも山谷にしない 左右3本で山を判定する場合、山が見えるのは3本後です。チャートでは山だった足へ印を置けますが、その足で売買すると未来の3本を先に見たことになります。この検証は3本後まで待ち、そこから先の値動きだけを使いました。 ### 構造ブレイクは終値で山谷を抜ける 上方向の構造ブレイクは、終値が直近の確定した山を上抜けた足です。買いで入り、終値が確定した谷を下抜けたら決済して売りへ切り替えます。下方向は逆です。 SMCではBOSとCHoCHを分ける説明もありますが、どの過去構造を基準にするかは実装ごとに異なります。この記事は方向転換を後から分類せず、**確定した山谷を終値で抜けたか**だけを測ります。 ### 流動性スイープはヒゲで抜けて終値が戻る 買いの流動性スイープは、安値が直近の確定した谷を下回り、同じ足の終値が谷より上へ戻った形です。売りは、高値が山を上回り、終値が山より下へ戻った形です。 この形から分かるのはローソク足の位置だけです。ストップ注文が実際に置かれていたか、誰が売買したかはOHLCから確認できません。この記事の「流動性スイープ」は、注文主体ではなく、この価格条件の名前です。 ### オーダーブロックは確定後の初回回帰だけを使う オーダーブロックには統一された数式がありません。ICTの初期講義では、候補となる反対色ローソクの高値または安値を後続足が抜いてから、候補足へ戻る形を説明しています。現在のSMCインジケーターには、確定済みの山谷を終値で抜けたBOSの直前から反対色ローソクを探す実装もあります。この2つは同じ条件ではありません。 検証では、候補足のヒゲを抜く原典型、候補足を終値で抜く保守型、確定スイングを終値で抜くBOS型を分けました。価格帯は候補足の実体または高値から安値です。確定した足と同じ足では戻りを判定せず、次の足から初回タッチ、価格帯の50%、同方向のローソクで引ける反応足を別々に数えます。 オーダーブロックと呼んでいるのは、OHLC上の反対色ローソク価格帯です。スポットFXは取引所へ注文が集まる市場ではないため、この価格帯から機関投資家の注文や未約定数量を確認することはできません。 ## SMCの設定と読み方 SMCはMT4・MT5の標準インジケーターには入っていません。外部インジケーターごとに定義と設定項目が違うため、名前が同じでも売買条件が同じとは限りません。 Formiqでは、バックテストの条件一覧から「SMC（市場構造）」を選びます。チャートへ新しい線を重ねる条件ではなく、確定した山谷と現在のローソク足の位置関係を売買判定に使います。 | 項目 | 初期設定 | 意味 | |---|---:|---| | 読み方 | 構造ブレイク | 終値で抜けるか、ヒゲで抜けて戻るか | | 山谷を確定する片側の本数 | 3 | 山谷の左右に何本ずつ置くか | | オーダーブロックの確定条件 | 確定した山谷を終値で抜ける | 候補足のヒゲ抜け・終値抜け・BOSのどれで価格帯を確定するか | | 価格帯の範囲 | 候補足の実体 | 始値と終値の間か、高値から安値までか | | 戻りの判定 | 価格帯へ触れ、同方向の足で引ける | 初回タッチ、50%到達、反応足のどれで売買するか | | 反対色ローソクを探す本数 | 5 | 確定足より前を何本探すか | | 最低変位 | 候補足の実体2倍 | 候補足から確定足の終値までに必要な距離 | | 無効化 | 終値が価格帯の遠端を抜ける | ヒゲまたは終値のどちらで価格帯を捨てるか | | FVGを必須にする | オフ | 候補足から確定までにFVGが必要か | | 価格帯の有効本数 | 50 | 確定後、初回回帰を何本待つか | 片側の本数を2から8へ増やすと、小さな山谷が減り、売買回数も減ります。2025年の1時間足・構造ブレイクでは、片側2本が290回、3本が201回、5本が127回、8本が89〜95回でした。 検証では、設定画面に出していない「古い山谷を50・100・200本のどこで捨てるか」も比較しました。片側2〜3本では同じ結果でした。片側8本では50本と100本で一部の結果が変わりましたが、100本と200本は検証した全区分で同じでした。初期設定の片側3本で結果が変わらないため、この値は操作項目にしていません。 ## 検証の条件 | 項目 | 条件 | |---|---| | 通貨ペア | USDJPY | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 2024年・2025年。2025年は前半と後半も別集計 | | 読み方 | 構造ブレイク・流動性スイープ・オーダーブロック | | 山谷を確定する片側の本数 | 2・3・5・8 | | 山谷の有効本数 | 50・100・200 | | 構造ブレイク・流動性スイープ | 2ルール×4設定×3設定×3時間足×2年＝144条件 | | オーダーブロックの検出定義 | 3確定条件×2価格帯×3探索本数×3変位×FVG有無×3時間足×2年＝648条件 | | オーダーブロックの戻り方 | 3判定×2無効化×3期限×2価格帯×3時間足×2年＝216条件 | | オーダーブロックのスイング強度 | 4設定×3時間足×2年＝24条件 | | 年間検証セルの合計 | 1,032条件 | | 売買 | シグナル足の終値で入り、反対シグナルで決済 | | コスト | スプレッド0.3pips、スリッページ0 | | ロット | 0.1ロット | 片側の本数と有効本数の組み合わせは、各ルール・各時間足につき12設定です。2024年で選んだ設定を2025年へ移す検定も、同じ12設定の中で行いました。 ## どの時間足で勝てたか 構造ブレイクの1時間足で、12設定中11設定が両年黒字でした。各年の年間損益が0pipsを超えた設定を数えます。中央値は2025年の12設定を年間損益順に並べた中央です。 | 売買ルール | 時間足 | 2025年に黒字 | 2024年に黒字 | 両年黒字 | 2025年の年間損益中央値 | |---|---|---:|---:|---:|---:| | 構造ブレイク | 15分足 | 11/12 | 9/12 | 8/12 | +865.3pips | | 構造ブレイク | 1時間足 | 12/12 | 11/12 | 11/12 | +1,096.3pips | | 構造ブレイク | 4時間足 | 3/12 | 12/12 | 3/12 | −1,951.4pips | | 流動性スイープ | 15分足 | 3/12 | 3/12 | 0/12 | −1,160.9pips | | 流動性スイープ | 1時間足 | 0/12 | 6/12 | 0/12 | −574.2pips | | 流動性スイープ | 4時間足 | 9/12 | 6/12 | 3/12 | +1,390.6pips | 構造ブレイクの1時間足は、12設定中11設定が両年黒字でした。片側3本・有効200本の初期設定も、2024年+2,177.2pips、2025年+2,214.2pipsです。 同じ構造ブレイクでも4時間足は、2024年に12/12設定が黒字だったものが2025年は3/12設定へ減りました。2025年の中央値も−1,951.4pipsです。「構造ブレイク」という名前だけでは時間足をまたいで同じ結果になりません。 ## 勝率が高ければ使えるのか 使えません。流動性スイープは勝率57%を超えても、平均損失のほうが大きく年間損益は赤字でした。片側3本・有効200本の2025年を、取引回数、勝率、平均利益、平均損失と一緒に見ます。 | 売買ルール | 時間足 | 取引回数 | 勝率 | 平均利益 | 平均損失 | 年間損益 | |---|---|---:|---:|---:|---:|---:| | 構造ブレイク | 15分足 | 832 | 35.82% | +49.44pips | −25.35pips | +1,194.9pips | | 構造ブレイク | 1時間足 | 201 | 40.80% | +100.25pips | −50.48pips | +2,214.2pips | | 構造ブレイク | 4時間足 | 59 | 38.98% | +98.15pips | −107.99pips | −1,630.4pips | | 流動性スイープ | 15分足 | 1,040 | 57.88% | +21.31pips | −32.51pips | −1,410.5pips | | 流動性スイープ | 1時間足 | 244 | 56.97% | +45.01pips | −79.45pips | −2,086.4pips | | 流動性スイープ | 4時間足 | 56 | 58.93% | +100.44pips | −90.15pips | +1,241.1pips | 構造ブレイクは15分足と1時間足で勝率36〜41%でも、平均利益が平均損失のおよそ2倍あり、年間損益が黒字でした。流動性スイープは15分足と1時間足で勝率57%を超えても、平均損失のほうが大きく、年間損益は赤字です。 4時間足の流動性スイープは+1,241.1pipsですが、56回しかありません。次の年や別の通貨ペアへ広げる前に、この勝率だけを根拠に設定を選ぶことはできません。 ## 去年よかった設定は今年も残るか 残ったのは構造ブレイクの1時間足と4時間足だけでした。2024年にいちばん良かった設定を2025年へ、2025年にいちばん良かった設定を2024年へ移しました。表の数値は移した先の年間損益です。 | 売買ルール | 時間足 | 2024年の1位を2025年へ | 2025年の1位を2024年へ | |---|---|---:|---:| | 構造ブレイク | 15分足 | −199.0pips | −282.3pips | | 構造ブレイク | 1時間足 | +2,214.2pips | +2,177.2pips | | 構造ブレイク | 4時間足 | +847.5pips | +1,905.4pips | | 流動性スイープ | 15分足 | −2,234.2pips | −1,578.4pips | | 流動性スイープ | 1時間足 | −1,052.6pips | +377.0pips | | 流動性スイープ | 4時間足 | −1,020.8pips | −1,786.5pips | 構造ブレイクの1時間足では、両年とも片側3本・有効50本が1位でした。4時間足も片側2本・有効50本が両年の1位です。一方、15分足の構造ブレイクと、流動性スイープの15分足・4時間足は、前年にいちばん良かった設定が翌年に赤字へ変わりました。 年内でも差があります。1時間足の構造ブレイク初期設定は、2025年前半+260.9pips、後半+1,929.2pipsでした。15分足の流動性スイープ初期設定は、前半+54.8pipsから後半−1,467.7pipsへ変わっています。 ## ADXやFVGを足すと増えるか 増えませんでした。2025年の初期設定へADX(14)を足すと、1時間足の構造ブレイクは+2,214.2pipsから、ADX 20以上で+919.2pips、25以上で+293.2pipsへ減りました。条件を厳しくするほど取引回数も201回から151回、114回へ減っています。 構造ブレイクと同じ足でFVGができることを条件にすると、1時間足は次のように変わりました。 | FVGの最小幅 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:| | FVGの大きさを問わない | 93 | 43.01% | +811.9pips | | 0.25 ATR以上 | 42 | 47.62% | +377.7pips | | 0.50 ATR以上 | 22 | 36.36% | −82.6pips | FVGを足すと勝率が一時的に上がっても、年間損益は単独の+2,214.2pipsを超えませんでした。この記事で測ったのは「構造ブレイクとFVG発生が同じ足」という条件です。流動性スイープ、構造転換、FVGへの戻りを順番に待つ別のSMC手順とは異なります。 ## 損切り利確で改善するか 3つの時間足で同じ改善にはなりませんでした。15分足の流動性スイープへ損切り30pips・利確60pipsを付けると、2025年は−1,410.5pipsから+599.7pipsへ変わりました。しかし、1時間足は−442.7pips、4時間足は−615.2pipsです。損切り20・利確40、損切り50・利確100も含め、3時間足で同じ改善にはなりませんでした。 構造ブレイクの1時間足は、反対シグナル決済の+2,214.2pipsに対し、損切り20・利確40が+98.1pips、損切り30・利確60が+991.5pips、損切り50・利確100が+1,342.5pipsでした。固定幅の決済は、この区分では利益を減らしています。 ## スプレッドの影響が大きいのはどれか 取引回数が最も多い15分足です。スプレッドを0から0.3pipsへ上げると、2025年の年間損益の減少幅は取引回数×0.3pipsと0.1pips以内で一致しました。 | 売買ルール | 時間足 | 取引回数 | スプレッド0 | スプレッド0.3 | |---|---|---:|---:|---:| | 構造ブレイク | 15分足 | 832 | +1,444.5pips | +1,194.9pips | | 構造ブレイク | 1時間足 | 201 | +2,274.4pips | +2,214.2pips | | 流動性スイープ | 15分足 | 1,040 | −1,098.5pips | −1,410.5pips | | 流動性スイープ | 1時間足 | 244 | −2,013.2pips | −2,086.4pips | 15分足の流動性スイープはスプレッド0でも−1,098.5pipsです。赤字の理由をコストだけでは説明できません。反対に、1時間足の構造ブレイクはスプレッド1.0pipsでも+2,073.5pipsでした。ただし、スリッページと実運用の約定差はこの検証に含めていません。 ## 関連記事 - [ダブルトップ・ダブルボトムの反転検証](/ja/blog/double-top-bottom-settings)：同じスイング高値・安値を2点の反転パターンとして測っています - [フェアバリューギャップは本当に埋まるのか](/ja/blog/fair-value-gap-settings)：SMCでも使われるFVGを、普通の足と比べています - [フィボナッチ・リトレースメントの水準検証](/ja/blog/fibonacci-retracement-levels)：有名な水準を隣の数値と比較した記事です - [プログラミングなしでバックテスト条件を組む方法](/ja/blog/backtest-without-coding)：この記事の構造ブレイクと流動性スイープを自分の通貨ペアで試せます ## 補足 対象はUSDJPYの2024年1月1日から2025年12月31日です。売買はシグナル足の終値、スプレッド0.3pips、0.1ロットで計算しました。山谷は右側の確認本数が閉じるまで利用していません。OHLCだけでは足の中の値動きの順序や注文主体を判定できないため、この記事は機関投資家の注文を観測した検証ではありません。 --- ### ATRの期間と比較する足の間隔を720回検証 URL: https://formiq.jp/ja/blog/atr-settings Language: ja Published: 2026-09-02 Updated: 2026-09-02 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ATR, Average True Range, ボラティリティ, インジケーター, 設定, バックテスト, USDJPY Also available in: en — https://formiq.jp/blog/atr-settings ATR（Average True Range、平均的な真の値幅）は、価格が1本あたりどれくらい動いているかを測る指標です。上昇と下降を区別せず、窓開けも含めた値幅を1本の線にします。 期間14が初期設定として広く使われています。しかし、USD/JPYで期間5〜50、比較する本数1〜20、ATR上昇中と低下中を組み合わせた60設定を測ると、**期間より「何本前のATRと比べるか」で結果が大きく変わりました。** 1時間足の1本比較は2024年平均−562.6pips、2025年平均−1,149.2pipsです。 期間20・10本比較・ATR低下中は、1時間足で2024年+3,022.6pips、2025年+2,184.5pipsでした。一方、初期設定に近い期間14・5本比較・ATR上昇中は−639.6pipsと−393.1pipsです。期間20を万能設定とする結果ではなく、**ATRの期間だけを変えても読み方の違いは埋まりませんでした。** ## ATRは何を測っているのか ATRは、まず各ローソク足のTrue Range（真の値幅）を求めます。次の3つで最も大きい値です。 1. 当日の高値 − 安値 2. 当日の高値と前の終値の差 3. 当日の安値と前の終値の差 高値と安値だけなら10pipsでも、前の終値から離れて始まれば、その空白を含むほうがTrue Rangeになります。[MetaTrader 5公式ヘルプ](https://www.metatrader5.com/en/terminal/help/indicators/oscillators/atr)も同じ3候補を示しています。 FormiqのATRは、Wilder式で平滑化します。期間をNとすると、新しいTrue Rangeの重みは1/N、前のATRの重みは(N−1)/Nです。 | 期間 | 新しい足の重み | 古い影響が半分になるまで | |---:|---:|---:| | 5 | 20.0% | 3.11本 | | 14 | 7.1% | 9.35本 | | 50 | 2.0% | 34.31本 | 期間14は「直近14本の単純平均」ではありません。新しい足を7.1%ずつ取り込み、過去の影響を少しずつ薄めます。 ### ATRには売買方向がない True Rangeは3候補を絶対値で比べるため、ATRは0未満になりません。暴落でも急騰でも値幅が広がれば上がります。[TradingView公式ガイド](https://www.tradingview.com/support/solutions/43000501823-average-true-range-atr/)も、ATRは価格方向を示さず、ボラティリティだけを測ると説明しています。 そこで本検証は、ATRだけで買いと売りを決めていません。ATRが指定本数前より上昇または低下しているとき、**同じ本数で価格が上がっていれば買い、下がっていれば売り**としました。決済は逆方向の条件です。 ## ATRの設定と読み方 ATRは価格チャートの下の別枠に、1本の線で表示されます。RSIの30・70のような標準水準線はありません。 | 画面 | 追加方法 | |---|---| | MT4 / MT5 | 挿入 → インディケータ → オシレーター → Average True Range | | TradingView | インジケーター検索でAverage True Rangeを追加 | | Formiq | インジケーター一覧からATRを追加 | ### 表示側の設定 | 項目 | 初期設定 | 意味 | |---|---:|---| | 期間 | 14 | True Rangeを平滑化する速さ | | 線の色・太さ | 画面による | ATR線の見た目だけを変える | MetaTrader標準ATRの計算項目は期間だけです。TradingViewは期間14に加えて、平滑化をRMA・SMA・EMA・WMAから選べます。Formiqは期間と線の見た目を設定し、計算はWilder式に固定しています。 ### バックテスト側の設定 | 項目 | 初期値 | 売買での意味 | |---|---:|---| | 読み方 | ATR上昇中 | 指定本数前よりATRが上か下か | | 期間 | 14 | ATR線の反応速度 | | 比較する本数 | 5 | ATRと価格を何本前から比べるか | 一般的な読み方は、ATR上昇を値動きの拡大、ATR低下を値動きの縮小として使う方法です。ただしATR上昇だけでは買いと売りが決まらないので、価格方向、ブレイク、移動平均線などを別に必要とします。 今回の60設定では、期間を変えても平均取引回数は1時間足2025年で815〜828回に収まりました。比較本数は1本の1,569.8回から20本の337.8回まで減りました。**売買頻度を大きく変えたのはATR期間ではなく比較本数でした。** ## 検証の条件 | 項目 | 条件 | |---|---| | 通貨ペア | USD/JPY | | 期間 | 2025年。2024年を同条件で比較し、2025年前半・後半も分割 | | 時間足 | 15分足・1時間足・4時間足 | | 読み方 | ATR上昇中／ATR低下中。価格方向は同じ比較本数の終値差 | | ATR期間 | 5・10・14・20・30・50 | | 比較本数 | 1・3・5・10・20 | | 設定数 | 1区分60設定、3時間足×4期間区分で720回 | | 決済 | 反対シグナル。エントリーと決済は同じルールセット | | 約定 | 終値、スプレッド0.3pips固定、0.1ロット | | 損切り・利確 | 主比較ではなし。別に測定 | 損益は約定価格からpipsへ再計算しました。 ## ATRはどの時間足で使えたか 2024年と2025年の両方で黒字だったのは、1時間足が60設定中29、4時間足が22、15分足が6でした。 | 時間足 | 2024年黒字 | 2025年黒字 | 両年黒字 | 2024年中央値 | 2025年中央値 | |---|---:|---:|---:|---:|---:| | 15分足 | 24/60 | 9/60 | 6/60 | −312.8 | −1,205.8 | | 1時間足 | 36/60 | 39/60 | **29/60** | +532.5 | +954.9 | | 4時間足 | 50/60 | 22/60 | 22/60 | +1,674.2 | −711.6 | 1時間足だけは両年とも中央値が黒字でした。15分足は取引回数が多く、60設定の中央値が両年赤字です。4時間足は2024年の50/60から2025年の22/60へ反転しました。 ## 期間と比較本数、どちらが効くか 損益を動かしたのは期間ではなく比較本数でした。1時間足の60設定を、比較本数ごとに12設定ずつ平均しています。 | 比較本数 | 2024年平均損益 | 2025年平均損益 | 2025年平均取引回数 | |---:|---:|---:|---:| | 1 | −562.6 | −1,149.2 | 1,569.8 | | 3 | +794.5 | +1,146.4 | 940.8 | | 5 | +163.4 | +676.6 | 727.0 | | 10 | +869.6 | −40.6 | 517.8 | | 20 | +773.9 | +856.9 | 337.8 | 1本比較は両年とも赤字で、取引回数は20本比較の4.6倍でした。ATRが1本前より少し上か下かを毎足読み直すと、売買が増えすぎます。 期間別の平均取引回数は2025年に815〜828回でほぼ同じでした。平均損益の首位は2024年が期間5の+743.5pips、2025年も期間5の+561.7pipsですが、期間30は−106.2pipsから+310.3pipsへ入れ替わり、期間14も+475.2pipsから+200.0pipsへ縮みました。**14が特別に強い結果は出ていません。** ### ATR低下中は両年の平均が黒字 | 読み方 | 2024年平均損益 | 2024年黒字 | 2025年平均損益 | 2025年黒字 | |---|---:|---:|---:|---:| | ATR上昇中 | −97.9 | 13/30 | +245.9 | 19/30 | | ATR低下中 | +913.5 | 23/30 | +350.1 | 20/30 | 「値動きが拡大しているときに方向へ乗る」という読み方は、2024年の平均が赤字でした。ATR低下中のほうが両年で平均黒字ですが、2025年の黒字数は20/30対19/30でほぼ同じです。増減だけで設定を決められる差ではありません。 ## 去年の1位は今年も勝てるか 6件中2件が翌年赤字でした。各時間足で一方の年にいちばん良かった設定を、もう一方の年へそのまま当てています。 | 時間足 | 選んだ年 | 設定 | 選んだ年 | 相手の年 | |---|---:|---|---:|---:| | 15分足 | 2025 | 上昇・10・10 | +1,226.1 | 2024年 +970.0 | | 15分足 | 2024 | 上昇・14・20 | +2,503.8 | 2025年 **−738.7** | | 1時間足 | 2025 | 低下・10・5 | +2,541.7 | 2024年 +341.8 | | 1時間足 | 2024 | 低下・20・10 | +3,022.6 | 2025年 +2,184.5 | | 4時間足 | 2025 | 低下・10・3 | +2,809.6 | 2024年 +2,359.1 | | 4時間足 | 2024 | 低下・5・1 | +4,225.7 | 2025年 **−596.7** | 1時間足の2025年首位は2024年にも+341.8pipsでしたが、2024年の60設定中央値+532.5pipsを下回りました。 1時間足の「低下・期間20・10本比較」は2025年前半+1,166.9pips、後半+993.5pipsで、年内も両方黒字でした。1年の成績で選ぶのがいつも失敗するわけではありませんが、期間20だけを抜き出すのではなく、ATR低下中・10本比較・1時間足まで一組で読む必要があります。 ## 初期設定の期間14はどうだったか 初期設定の期間14に初期比較5本とATR上昇中を組み合わせると、6区分のうち黒字は15分足2024年の1つだけでした。 | 時間足 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | 15分足 | 2024 | 2,111 | 35.20% | +68.4pips | | 15分足 | 2025 | 2,292 | 33.77% | −807.1pips | | 1時間足 | 2024 | 577 | 34.32% | −639.6pips | | 1時間足 | 2025 | 574 | 32.40% | −393.1pips | | 4時間足 | 2024 | 127 | 35.43% | −558.6pips | | 4時間足 | 2025 | 139 | 37.41% | −860.1pips | 期間14が一般的でも、「ATR上昇中に価格方向へ乗る」売買ルールの初期設定にはなりません。 ## 勝率が低いのに勝てるのはなぜか 勝率は4割未満でも、平均利益が平均損失の2倍を超えていたからです。両年黒字だった1時間足の「低下・期間20・10本比較」を見ます。 | 年 | 取引回数 | 勝率 | 平均利益 | 平均損失 | 年間損益 | |---|---:|---:|---:|---:|---:| | 2024 | 628 | 35.51% | +54.66pips | −22.63pips | +3,022.6pips | | 2025 | 601 | 36.94% | +51.75pips | −24.55pips | +2,184.5pips | ATR設定を勝率だけで比べると、この損益幅の差が消えます。 ## ADXや損切り利確を足すと良くなるか 足した5つは、どれも両年の基準値を超えませんでした。同じ「低下・期間20・10本比較・1時間足」に条件を足しています。 | 追加条件 | 2024年 | 2025年 | |---|---:|---:| | なし | +3,022.6 | +2,184.5 | | ADX(14)が20以上 | +646.9 | −572.2 | | 24本で時間決済 | +2,840.5 | +2,482.2 | | 損切り50・利確100 | +2,942.2 | +1,472.7 | | 損切りATR1倍・利確ATR2倍 | +1,940.0 | +554.9 | | 損切りATR2倍・利確ATR3倍 | +1,456.6 | +1,151.7 | 24本決済は2025年を+297.7pips増やしましたが、2024年を−182.1pips減らしました。ADXと4つの損切り利確は、どれも両年の基準値を上回っていません。 ATRで損切り幅を変動させても、自動的に固定pipsより良くなるわけではありません。ATRは距離の単位を相場の値幅へ合わせますが、出口の優位性までは作りません。 ## スプレッドはいくら削るか 601回の売買では、スプレッド1pipsで年間600.9pips減りました。2025年の同じ設定です。 | スプレッド | 取引回数 | 年間損益 | |---:|---:|---:| | 0 | 601 | +2,364.7pips | | 0.3 | 601 | +2,184.5pips | | 1.0 | 601 | +1,763.8pips | | 2.0 | 601 | +1,162.8pips | 損益分岐は3.93pipsです。コストは取引回数×スプレッドで減り、1本比較のように売買回数が増える設定ほど先に不利になります。 ## 関連記事 - [スーパートレンド設定：期間より倍率が結果を変えた](/ja/blog/supertrend-settings) - [BBW設定：固定水準とスクイーズを検証](/ja/blog/bollinger-bandwidth-settings) - [移動平均線乖離率：％とATR倍を比較](/ja/blog/ma-disparity-settings) ## 補足 - 対象期間は2024年1月1日〜2025年12月31日。2025年は前半・後半にも分割しました - 約定は終値、スプレッドは主比較で0.3pips固定です - ATRはWilder式で、各期間の前にウォームアップを置いています - MetaTrader・TradingView・Formiqでは平滑化方式や利用できる設定が異なるため、別実装へ数値を移すときは方式を確認してください --- ### DMI/ADXのDIクロスは勝てる？1,344通り検証した結果 URL: https://formiq.jp/ja/blog/dmi-adx-settings Language: ja Published: 2026-09-02 Updated: 2026-09-02 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: DMI/ADX, ADX, DMI, +DI, −DI, トレンド, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/dmi-adx-settings DMI（Directional Movement Index、方向性指数）は、上方向の値動きを表す+DI、下方向を表す−DI、両者の開きを平滑化したADXの3本でトレンドを読みます。+DIと−DIが方向を決め、ADXは上昇か下落かを区別せず、トレンドの強さだけを表します。 期間14・ADX 25以上はよく使われる設定です。ところが2025年のドル円では、水準25を課すと期間14のDIクロスが**15分足2,184回中339回、1時間足534回中101回、4時間足141回中19回**まで減りました。残った割合は13.5〜18.9%です。 売買結果も「25以上ならだましが減る」とは揃いません。期間14・水準25のDIクロスは、15分足だけが2024年と2025年の両方で黒字でした。1時間足は2025年−68.1pips、2024年+137.5pipsです。水準25を移動平均線クロスのフィルターにした場合も、2025年は+377.4pipsから−922.1pipsへ下がりました。 ## DMI/ADXは何を測っているのか 計算は、隣り合う足の高値と安値から始まります。 1. 当日の高値が前の足の高値よりどれだけ上がったかを測る 2. 前の足の安値が当日の安値よりどれだけ高かったかを測る 3. 大きかった側だけを+DMまたは−DMとして残す 4. +DMと−DMを値幅（TR）で割り、平滑化して+DIと−DIにする 5. `|+DI − −DI| ÷（+DI + −DI）× 100`をDXとし、さらに平滑化してADXにする 絶対値を取るため、+DIが大きくても−DIが大きくてもADXは上がります。**ADX 30は上昇を意味しません。2本のDIが大きく離れている状態を意味します。** [MetaTrader 5の公式解説](https://www.metatrader5.com/ja/terminal/help/indicators/trend_indicators/admi)は、期間14の+DIと−DIを比較し、+DIが上なら買い、−DIが上なら売りという読み方を示しています。[Wilder版の計算資料](https://www.metatrader5.com/en/terminal/help/indicators/trend_indicators/admiw)では、+DM・−DM・TRを平滑化してDIを求め、DIの差からDX、ADXへ進む式を確認できます。 ## DMI/ADXの設定と読み方 DMI/ADXはローソク足へ重ねず、**チャート下の別枠にADX、+DI、−DIの3本**を表示します。0〜100の範囲で動きます。 | 環境 | 追加する場所 | |---|---| | MT4 / MT5 | 挿入 → インディケータ → トレンド → Average Directional Movement Index | | TradingView | インジケーター検索から Directional Movement Index（DMI） | | ブラウザ（Formiq） | インジケーター一覧から「ADX」 | TradingViewではADX単体とDMIが別項目です。[TradingViewの公式DMI解説](https://www.tradingview.com/support/solutions/43000502250-directional-movement-dmi/)にあるように、3本を同時に見る場合はDMIを選びます。MT4・MT5では3本が同じ別枠へ表示され、MT5には通常のADXとADX Wilderがあります。 ### チャート側の設定 | 項目 | 初期値 | 意味 | |---|---:|---| | 期間 | 14 | +DM、−DM、TR、DXを平滑化する本数 | | ADX線 | プラットフォームの初期色 | トレンドの強さ。上昇・下落の方向は示さない | | +DI線 | プラットフォームの初期色 | 上方向の値動きが相対的に優勢かを示す | | −DI線 | プラットフォームの初期色 | 下方向の値動きが相対的に優勢かを示す | | 水準線 | 任意 | 20や25など、強さを区切る目安を追加する | 期間を短くすると3本とも反応が速くなり、DIクロスが増えます。長くすると線は滑らかになりますが、短いトレンドへの反応は遅くなります。水準は線の計算を変えず、どの売買候補を残すかだけを変えます。 ### バックテスト条件の設定 | 項目 | 初期値 | 意味 | |---|---:|---| | 期間 | 14 | DMI/ADXを計算する本数 | | 最低水準 | 25 | ADXがこの値以上のときだけ新規売買を許す | | 方向判定 | フィルターのみ | 別の売買条件が決めた方向を変えず、ADXの強さだけを足す | 方向判定は3通りあります。 - **フィルターのみ**：ADXが最低水準以上かだけを見る。単独では売買方向を決められないため、移動平均線クロスなどと組み合わせる - **+DI/−DIクロス**：+DIが−DIを上抜けた足で買い、下抜けた足で売る。ADXが最低水準以上のクロスだけを採る - **+DI/−DI比較**：ADXが最低水準以上で+DIが上なら買い、−DIが上なら売る。クロスの瞬間を逃しても条件を満たせる この記事では、DIクロスとADX加速を独立した売買ルールとして測り、ADXをフィルターとして使う場合は10/20・SMAクロスへ追加して測ります。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01〜2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 読み方 | +DI/−DIクロス / +DI/−DI比較 | | 計算期間 | 5 / 7 / 10 / **14** / 20 / 28 / 40 / 50 | | ADX最低水準 | 0 / 15 / 20 / **25** / 30 / 35 / 40 | | 組み合わせ | 112通り × 時間足3種 × 期間4区分 = 1,344回 | | 決済 | 反対方向のDIシグナルまで。エントリーと決済に同じルールセットを使う | | 損切り・利確 | 使わない（別の節で測定） | | スプレッド | 0.3pips固定。約定は終値 | | ロット | 0.1ロット | 水準0はADXで絞らない対照です。2025年は前半と後半にも分け、1年の合計だけで設定を選ばないようにしました。 ## ADXの25は何を意味するか 市場の半分前後を残しながら、DIクロスの4分の3以上を落とす線です。期間14のADXが各水準以上だった足と、DIクロスの回数を数えました。 | 時間足 | 年 | ADX 20以上の足 | ADX 25以上の足 | ADX 30以上の足 | 全DIクロス | 水準25を通ったクロス | |---|---:|---:|---:|---:|---:|---:| | 15分足 | 2024 | 63.65% | 43.47% | 27.93% | 2,136 | 349 | | 15分足 | 2025 | 61.22% | 40.51% | 25.15% | 2,184 | 339 | | 1時間足 | 2024 | 64.82% | 44.91% | 29.49% | 501 | 93 | | 1時間足 | 2025 | 66.33% | 47.24% | 31.40% | 534 | 101 | | 4時間足 | 2024 | 73.82% | 50.43% | 30.45% | 142 | 35 | | 4時間足 | 2025 | 67.20% | 37.58% | 17.58% | 141 | 19 | ADX 25以上の足自体は37.58〜50.43%あり、珍しくありません。それでもDIクロスと同時に満たす回数は全クロスの13.5〜24.6%です。**水準25は市場の半分前後を残しながら、クロスの4分の3以上を落とします。** DIが交差する時点ではADXがまだ低い場合が多いためです。 実際、2025年にDIが交差した足でADXが前の足より上がっていたのは、15分足218/2,184、1時間足53/534、4時間足11/141でした。クロスは方向の入れ替わり、ADXは平滑化された開きなので、同時に立ち上がるとは限りません。 ## 2年とも勝てた設定はどれだけあるか 56設定のうち3〜16設定です。期間8通り×水準7通り、各56設定の分布です。 ### DIクロス | 時間足 | 2025年黒字 | 2025年中央値 | 2024年黒字 | 2024年中央値 | 両年黒字 | |---|---:|---:|---:|---:|---:| | 15分足 | 17/56 | −40.1pips | 17/56 | −189.8pips | **10/56** | | 1時間足 | 27/56 | 0.0pips | 19/56 | −37.9pips | **9/56** | | 4時間足 | 6/56 | −167.1pips | 29/56 | +3.2pips | **4/56** | ### DI比較 | 時間足 | 2025年黒字 | 2025年中央値 | 2024年黒字 | 2024年中央値 | 両年黒字 | |---|---:|---:|---:|---:|---:| | 15分足 | 18/56 | −335.5pips | 26/56 | −53.2pips | **13/56** | | 1時間足 | 23/56 | −79.4pips | 32/56 | +118.4pips | **16/56** | | 4時間足 | 3/56 | −817.8pips | 38/56 | +359.2pips | **3/56** | 4時間足は読み方を変えても2024年から2025年へ反転しました。DI比較は2024年に38/56が黒字でも、2025年は3/56です。期間や水準を選ぶ前に、年をまたいだ時点で候補の多くが消えています。 ## 初期設定の期間14・水準25はどうか DIクロスは両年黒字が残り、DI比較はすべての時間足で符号が入れ替わりました。初期値の実測です。 | 読み方 | 時間足 | 2025年取引数 | 2025年勝率 | 2025年損益 | 2024年取引数 | 2024年損益 | |---|---|---:|---:|---:|---:|---:| | DIクロス | 15分足 | 339 | 32.45% | **+141.1pips** | 349 | **+403.3pips** | | DIクロス | 1時間足 | 101 | 33.66% | −68.1pips | 93 | +137.5pips | | DIクロス | 4時間足 | 19 | 42.11% | +433.3pips | 35 | +152.0pips | | DI比較 | 15分足 | 686 | 32.65% | −391.2pips | 679 | +570.3pips | | DI比較 | 1時間足 | 181 | 34.25% | −388.5pips | 179 | +72.3pips | | DI比較 | 4時間足 | 45 | 24.44% | −2,277.2pips | 59 | +1,042.7pips | DIクロスの4時間足は両年黒字ですが、2025年19回、2024年35回です。15分足は339回と349回あり、両年とも黒字でした。反対にDI比較はすべての時間足で2024年と2025年の符号が入れ替わりました。 ## 水準はどこに置くのがいいか 水準25だけが境目になる形はありませんでした。1時間足の周辺平均です。各行は8期間の平均で、設定が出さなかった売買も0pipsとして含めています。 ### DIクロスの最低水準 | 最低水準 | 2025年平均損益 | 2024年平均損益 | 2025年平均取引数 | |---:|---:|---:|---:| | 0 | −37.4pips | **+842.1pips** | 541.1 | | 15 | **+447.4pips** | −320.5pips | 381.1 | | 20 | +231.5pips | −416.4pips | 265.6 | | 25 | +94.6pips | −239.8pips | 170.5 | | 30 | +17.4pips | −120.1pips | 101.4 | | 35 | +7.6pips | −122.5pips | 57.8 | | 40 | +39.5pips | +13.9pips | 31.8 | 水準0は2024年+842.1pipsでも2025年−37.4pips、水準15は2025年+447.4pipsでも2024年−320.5pipsでした。両年平均がプラスだった水準40は、平均取引数が31.8回まで減っています。水準25だけが境目になる形はありません。 ### DI比較の最低水準 | 最低水準 | 2025年平均損益 | 2024年平均損益 | 2025年平均取引数 | |---:|---:|---:|---:| | 0 | −84.0pips | +788.4pips | 542.1 | | 15 | −36.1pips | −14.9pips | 417.5 | | 20 | −317.6pips | +52.2pips | 326.8 | | 25 | −215.6pips | +80.9pips | 248.1 | | 30 | −178.2pips | −47.7pips | 183.5 | | 35 | −295.6pips | −107.8pips | 136.4 | | 40 | −138.7pips | +22.3pips | 98.8 | DI比較は7水準すべてで2025年の平均が赤字です。最低水準を上げても、2025年の年間損益はプラスへ戻りませんでした。 計算期間では、短い5と7だけが両方の年でプラスの周辺平均になりました。DIクロスの期間5は2025年+347.9pips・2024年+225.0pips、期間7は+307.1・+71.2pipsです。DI比較の期間5は+833.4・+663.0pips、期間7は+437.7・+476.5pipsでした。ただし期間5の2025年平均取引数はDIクロス596.1回、DI比較729.3回です。短い期間の差には、取引回数とコストも含まれます。 ## 去年いちばん稼いだ設定はどうなるか 6通り中4通りで翌年赤字でした。2024年の年間損益が最大だった設定を、そのまま2025年へ移します。 | 読み方 | 時間足 | 期間 | 水準 | 2024年損益 | 2025年損益 | 2025年中央値 | |---|---|---:|---:|---:|---:|---:| | DIクロス | 15分足 | 28 | 0 | +1,305.0pips | +366.0pips | −40.1pips | | DIクロス | 1時間足 | 50 | 0 | +1,513.1pips | **−1,309.9pips** | 0.0pips | | DIクロス | 4時間足 | 20 | 0 | +1,738.5pips | **−752.8pips** | −167.1pips | | DI比較 | 15分足 | 20 | 20 | +1,630.4pips | +86.0pips | −335.5pips | | DI比較 | 1時間足 | 50 | 0 | +1,391.5pips | **−1,352.7pips** | −79.4pips | | DI比較 | 4時間足 | 20 | 15 | +1,677.3pips | **−1,329.7pips** | −817.8pips | 15分足の2設定は2025年も黒字でしたが、利益は2024年の22.4%と5.3%まで縮みました。1時間足と4時間足は、両方の読み方で2025年の中央値も下回っています。 ## 勝率だけで判断できるか できません。期間14・水準25のDIクロス、1時間足2025年は101回、勝率33.66%、−68.1pipsでした。勝った取引の平均は+54.05pips、負けた取引の平均は−28.44pipsです。勝率が3分の1でも、1回の勝ちが負けの約1.9倍あるため損益はほぼ均衡しました。 4時間足2025年は勝率42.11%、19回、+433.3pipsです。勝率だけなら良く見えますが、19回では設定差と偶然を分けにくくなります。DI比較の同じ設定は45回、勝率24.44%、−2,277.2pipsでした。読むべき組み合わせは勝率だけでなく、取引回数、平均利益・平均損失、年間損益です。 ## ADXをフィルターに使うとどうなるか 10/20・SMAクロスへ足すと、両年とも利益が減りました。「フィルターのみ」を入口へ追加し、決済は元の反対クロスのままにしました。 | 年 | ADX条件 | 取引数 | 勝率 | 年間損益 | |---:|---|---:|---:|---:| | 2025 | なし | 358 | 39.11% | **+377.4pips** | | 2025 | 20以上 | 197 | 36.55% | −80.9pips | | 2025 | 25以上 | 117 | 31.62% | −922.1pips | | 2025 | 30以上 | 77 | 32.47% | −358.8pips | | 2024 | なし | 315 | 41.59% | **+2,888.4pips** | | 2024 | 20以上 | 169 | 42.01% | +1,349.6pips | | 2024 | 25以上 | 106 | 48.11% | +1,073.2pips | | 2024 | 30以上 | 59 | 44.07% | +462.8pips | 2024年は水準25で勝率が41.59%から48.11%へ上がりましたが、年間損益は+2,888.4pipsから+1,073.2pipsへ減りました。2025年は3水準すべて赤字です。強い場面だけを残す操作が、利益の大きい初動まで落としています。 ## 時間帯や損切りで改善するか 年をまたいで同じ改善にはなりませんでした。期間14・水準25の1時間足DIクロスへ追加します。 | 条件 | 2025年取引数 | 2025年損益 | 2024年取引数 | 2024年損益 | |---|---:|---:|---:|---:| | 追加なし | 101 | −68.1pips | 93 | +137.5pips | | 東京時間（UTC 0〜8時） | 33 | −457.5pips | 37 | +351.1pips | | ロンドン・NY時間（UTC 7〜21時） | 66 | +511.1pips | 56 | −7.6pips | | 損切り30・利確60 | 93 | +603.0pips | 87 | +253.2pips | | 損切り50・利確100 | 97 | +612.3pips | 87 | −246.1pips | | 損切り100・利確200 | 100 | +170.6pips | 90 | −274.5pips | | 損切り50・利確50 | 97 | +700.5pips | 88 | +31.4pips | 時間帯は2025年にロンドン・NY時間が+511.1pipsでも、2024年は−7.6pipsでした。損切り30・利確60だけは両年で基準を上回りました。ただし比較した4組から選んだ結果であり、別期間での確認は残っています。 ## スプレッドのせいで負けたのか 違います。期間14・水準25の1時間足DIクロスは、2025年にスプレッド0で−37.8pips、0.3pipsで−68.1pipsでした。差は101回×0.3pipsの30.3pipsです。コストをゼロにしても赤字なので、2025年の負けをスプレッドだけでは説明できません。 2024年はスプレッド0で+165.3pips、0.3pipsで+137.5pips、1.5pipsで+25.9pips、2.0pipsで−20.6pipsでした。損益分岐は約1.78pipsです。2025年側には正の損益分岐スプレッドがありません。 ## 関連記事 - [移動平均線クロスの設定検証](/ja/blog/moving-average-cross-settings)：ADXを足した基準側の10/20・SMAクロスを期間と計算方法ごとに比較 - [パラボリックSARの設定検証](/ja/blog/parabolic-sar-settings)：同じトレンド系指標で、ADXフィルターが両年の損益を下げた例 - [CCIの設定検証](/ja/blog/cci-settings)：トレンド系の読み方へADXを足したときの取引数と損益を比較 - [アリゲーターの設定検証](/ja/blog/alligator-settings)：3本の線が順番に並ぶ状態の基準率を測定 - [ダマシ回避フィルターは効くのか？1,596通り検証した結果](/ja/blog/false-signal-filters)：同じADXを、公開済み38通りの設定へフィルターとして掛けた横断検証 - [プログラミングなしで条件を組んで検証する](/ja/blog/backtest-without-coding)：DMI/ADXを別条件と組み合わせる手順 ## 補足 2024年と2025年の米ドル／円を終値で約定し、0.3pips固定、0.1ロットで測定しました。ADX(14)は最初の値に28本分の計算が必要です。主検証は反対のDIシグナルで決済し、固定の損切り・利確は別に測っています。 --- ### 一目均衡表「三役好転で買い」は勝てる？280通り検証した結果 URL: https://formiq.jp/ja/blog/ichimoku-settings Language: ja Published: 2026-09-02 Updated: 2026-09-04 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: 一目均衡表, 三役好転, 時間論, 転換線, 基準線, 先行スパン, 遅行スパン, USD/JPY Also available in: en — https://formiq.jp/blog/ichimoku-settings 一目均衡表は、転換線、基準線、先行スパンA・B、遅行スパンをローソク足に重ねる指標です。名前は多いものの、使っている計算は「一定期間の最高値と最安値の中間」と「線を26本ずらして描く」の2つに整理できます。 USD/JPYの2022〜2025年で4つの読み方を測ると、**1時間足・標準9/26/52の三役は、2022年+558.1pips、2023年−1,012.7pips、2024年+536.0pips、2025年+1,085.2pips**でした。3条件が揃っても4年連続では残りません。一方、価格と基準線の標準設定は4年すべて黒字でした。 時間論の変化日も高安の選び方を固定して測りました。9・17・26・33・42・52・65・76本目を合わせた候補足の局所高安率は、3つの時間足で34.19〜35.49%でした。ただし3本前・3本後の対照足も34.23〜35.04%で、8数値全体に固有の上乗せは確認できませんでした。26本目だけは1時間足で39.52%、対照足で32.59%となり、別に見る価値がある結果でした。 ## 一目均衡表の5本は何を測っているのか | 線 | 標準期間 | 計算 | 描く位置 | |---|---:|---|---| | 転換線 | 9 | 過去9本の（最高値＋最安値）÷2 | 現在 | | 基準線 | 26 | 過去26本の（最高値＋最安値）÷2 | 現在 | | 先行スパンA | 9と26 | （転換線＋基準線）÷2 | 26本先 | | 先行スパンB | 52 | 過去52本の（最高値＋最安値）÷2 | 26本先 | | 遅行スパン | 26 | 現在の終値 | 26本前 | 先行スパンAとBの間が「雲」です。価格が雲の上なら上向き、下なら下向き、雲の中なら方向が定まっていないと読むのが一般的です。 ### 先行スパンは予測値ではない 先行スパンは26本先に描かれますが、未来の価格は計算に入りません。たとえば現在のチャート上に見える先行スパンBは、**26本前の時点で過去52本を調べた中間値**です。右へずらすのは表示方法であり、予測計算ではありません。 ### 遅行スパンを現在の判断へ戻す 遅行スパンは現在の終値を26本前に描きます。そのため、26本前の位置にある遅行スパンと当時の価格を比べることは、現在の終値と26本前の価格を比べることと同じです。 実装上も、遅行スパンの各表示値は26本後の終値と一致し、12の時間足・年区分で比較した値に不一致はありませんでした。これは未来を当てた結果ではなく、**現在の終値を過去へ移しているという恒等関係**です。先読みを避けるなら、現在足では「現在の終値＞26本前の終値」の形で使います。 ## 時間論の変化日は当たるのか 8数値全体では、局所高安ができる割合が通常の日と変わりませんでした。一目均衡表は5本の線だけではありません。体系には時間論・波動論・値幅観測論があり、時間論では重要な高値や安値を起点にローソク足の本数を数えます。[IFTAの2025年教材](https://www.ifta.org/assets/docs/IFTA_CFTe_Syllabus_Reading_Material.pdf)は9・17・26・33・42・51・65・76などを挙げています。一方、一般向けの解説には51ではなく52を載せた資料もあり、数値が一致していません。 ここでいう「変化日」は、上昇か下落かを決める売買シグナルではありません。反転のほか、同じ方向への加速や変化時期の延長と読む場合もあります。以下では意味を広げず、**予定した足の近くに局所的な高値または安値ができたか**だけを測りました。 ### 起点と的中条件を結果を見る前に固定した | 項目 | 機械的な定義 | |---|---| | 起点 | その足が前後5本の中で唯一の最高値または最安値。5本後に確定 | | 変化日の数値 | 9・17・26・33・42・52・65・76。起点を1本目として数える | | 的中 | 予定足の前後1本以内に、同じ定義の高値または安値がある | | 対照 | 同じ起点から各数値の3本前と3本後を予定足にする | | 集計 | 同じ予定足への重複は1本にまとめ、2022〜2025年を年別にも集計 | | 不確実性 | 月単位で5,000回再標本化した差の95%区間 | この定義なら、起点は最短の9本目より前に確定します。起点と終点は両方数えるため、コード上の予定足は「起点＋数値−1」です。51と52の食い違いは、52を主集計、51を感度確認に分けました。 ### 候補足の34〜35%に局所高安ができたが、対照足も同率 | 時間足 | 候補足 | 局所高安になった候補足 | 候補足の局所高安率 | 対照足の局所高安率 | 差の95%区間 | |---|---:|---:|---:|---:|---:| | 15分足 | 63,192 | 21,650 | 34.26% | 34.23% | −0.21〜+0.29pt | | 1時間足 | 15,923 | 5,444 | 34.19% | 34.84% | −1.17〜−0.13pt | | 4時間足 | 4,142 | 1,470 | 35.49% | 35.04% | −0.66〜+1.58pt | 8数値をすべて使うと、前後1本を含む候補範囲は全ローソク足の96.26〜96.73%を覆いました。ほぼ全期間が候補になるため、的中率だけを見ても時間論が変化を選び分けたとは言えません。51へ置き換えても、候補足と対照足の差は15分足+0.17pt、1時間足−0.17pt、4時間足+0.48ptにとどまりました。 ### 26本目は1時間足と4時間足で対照を上回った 基本数値9・17・26だけに絞ると、1時間足は候補足35.93%、対照足33.58%で+2.35ptでした。差の95%区間は+1.14〜+3.54ptです。15分足は+0.57pt、4時間足は−0.08ptでした。 差の大半は26本目にありました。1時間足の26本目は4年合計で1,195/3,024本が局所高安となり、39.52%でした。23本目・29本目を使った対照足は32.59%で、差は+6.93pt、95%区間は+5.10〜+8.75ptです。4時間足も39.34%対32.00%でしたが、15分足は34.61%対34.62%でした。 | 時間足 | 年 | 26本目の局所高安率 | 対照足の局所高安率 | 差 | |---|---:|---:|---:|---:| | 1時間足 | 2022 | 39.24% | 33.87% | +5.37pt | | 1時間足 | 2023 | 39.97% | 32.46% | +7.52pt | | 1時間足 | 2024 | 40.22% | 32.18% | +8.04pt | | 1時間足 | 2025 | 38.64% | 31.84% | +6.79pt | | 4時間足 | 2022 | 36.95% | 33.79% | +3.16pt | | 4時間足 | 2023 | 35.64% | 29.53% | +6.11pt | | 4時間足 | 2024 | 45.41% | 31.37% | +14.04pt | | 4時間足 | 2025 | 38.95% | 33.24% | +5.71pt | 26本目の差は起点を前後3本・5本・10本で定義し直しても、1時間足で+3.99〜+7.13pt、4時間足で+3.25〜+8.64ptでした。ただし、これは売買方向、損益、加速や延長を測った結果ではありません。複数の数値と時間足を比較した中の結果なので、別期間と別通貨で同じ差が残るまでは「26本目なら反転する」とは判断できません。また、過去の高安間隔を次の高安から投影する対等数値は、どの高安を結ぶかという別の定義が必要なため、この集計には含めていません。 ## 一目均衡表の設定と読み方 MT4・MT5では「挿入 → インディケータ → トレンド → Ichimoku Kinko Hyo」から追加します。チャート上に5本の線が重なり、先行スパンAとBの間が塗られます。色はプラットフォームやテンプレートで変わるため、色ではなく線名で確認してください。 | 設定項目 | 初期設定 | 意味 | |---|---:|---| | Tenkan-sen | 9 | 転換線が調べる本数 | | Kijun-sen | 26 | 基準線が調べる本数。先行・遅行のずらし幅にも使う | | Senkou Span B | 52 | 先行スパンBが調べる本数 | 遅行スパンに独立した期間設定はありません。基準線の26がずらし幅を兼ねます。 ### 三役好転・三役逆転を含む4つの売買ルール | 読み方 | 買い | 売り | 実際に使う期間 | |---|---|---|---| | 転換線×基準線 | 転換線が基準線を上抜け | 転換線が基準線を下抜け | 転換線・基準線 | | 価格×基準線 | 終値が基準線を上抜け | 終値が基準線を下抜け | 基準線だけ | | 価格×雲 | 終値が雲より上 | 終値が雲より下 | 転換線・基準線・先行スパンB | | 三役 | 転換線＞基準線、終値＞雲、現在終値＞基準線期間前の終値がすべて成立 | 3条件がすべて反対 | 3期間すべて | 三役の買い側が三役好転、売り側が三役逆転です。3つが同じ足で同時にクロスすることではなく、**3条件が同時に成立している状態**を指します。どのルールも反対シグナルで決済します。価格と雲のルールは雲の中で、三役は3条件が揃わない間、新規売買を始めません。 設定を変えるなら、先に読み方を決めます。価格と基準線のクロスでは転換線9と先行スパンB52を変えても結果は変わりません。転換線と基準線のクロスでも先行スパンB52は使われません。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2022-01-01〜2025-12-31。各年を同条件で集計 | | 時間足 | 15分足・1時間足・4時間足 | | 転換線 | 5・7・9・12・18 | | 基準線 | 20・22・26・30・36 | | 先行スパンB | 40・44・52・60・72 | | 設定数 | 転換線×基準線25、価格×基準線5、価格×雲125、三役125。1時間足あたり280設定 | | 決済 | 反対シグナルまで。エントリーと決済に同じルールを使用 | | 損切り・利確 | なし。別節で追加検証 | | スプレッド・約定 | 0.3pips固定・終値約定 | | ロット | 0.1ロット | 各設定は2025年前半と後半にも分けて測りました。買いと売りは全設定で両方成立しています。 ## どの読み方が4年とも勝てたか 1時間足の基準線クロスは5設定すべてが4年とも黒字でした。各セルの分母は、その読み方で意味のある設定数です。 | 時間足 | 読み方 | 2022年 | 2023年 | 2024年 | 2025年 | 4年すべて | 2025年中央値 | |---|---|---:|---:|---:|---:|---:|---:| | 15分足 | 転換線×基準線 | 20/25 | 8/25 | 25/25 | 24/25 | 4/25 | +1,315.9pips | | 15分足 | 価格×基準線 | 4/5 | 0/5 | 5/5 | 5/5 | 0/5 | +541.8pips | | 15分足 | 価格×雲 | 125/125 | 103/125 | 125/125 | 125/125 | 103/125 | +1,148.8pips | | 15分足 | 三役 | 102/125 | 84/125 | 125/125 | 101/125 | 68/125 | +859.7pips | | 1時間足 | 転換線×基準線 | 25/25 | 19/25 | 17/25 | 11/25 | 7/25 | −35.5pips | | 1時間足 | 価格×基準線 | 5/5 | 5/5 | 5/5 | 5/5 | 5/5 | +1,236.6pips | | 1時間足 | 価格×雲 | 120/125 | 57/125 | 125/125 | 120/125 | 54/125 | +437.7pips | | 1時間足 | 三役 | 125/125 | 7/125 | 110/125 | 87/125 | 5/125 | +259.4pips | | 4時間足 | 転換線×基準線 | 15/25 | 2/25 | 24/25 | 3/25 | 0/25 | −984.0pips | | 4時間足 | 価格×基準線 | 4/5 | 0/5 | 4/5 | 4/5 | 0/5 | +255.8pips | | 4時間足 | 価格×雲 | 125/125 | 48/125 | 125/125 | 0/125 | 0/125 | −2,310.6pips | | 4時間足 | 三役 | 125/125 | 14/125 | 125/125 | 0/125 | 0/125 | −2,136.8pips | 4年すべて黒字だった設定は、15分足の価格と雲103設定、三役68設定、1時間足の価格と基準線5設定などです。1時間足の三役は2022年に125設定すべてが黒字でも、2023年は7設定まで減りました。4時間足には4年連続黒字の設定がありません。 ## 基準線はなぜ横ばいが多いのか 終値平均ではなく、期間の高安が更新されないと動かないからです。約6割の足が横ばいでした。期間中に最高値か最安値が更新されないかぎり、中間値が横ばいになりやすい計算です。 | 時間足 | 年 | 転換線が前の足と同値 | 基準線が前の足と同値 | 雲の中にいた足 | |---|---:|---:|---:|---:| | 15分足 | 2022 | 35.37% | 60.14% | 14.72% | | 15分足 | 2023 | 34.57% | 59.74% | 14.82% | | 15分足 | 2024 | 34.59% | 59.30% | 14.12% | | 15分足 | 2025 | 33.81% | 58.39% | 14.89% | | 1時間足 | 2022 | 35.98% | 59.74% | 13.14% | | 1時間足 | 2023 | 36.47% | 61.25% | 14.43% | | 1時間足 | 2024 | 35.92% | 60.16% | 13.01% | | 1時間足 | 2025 | 34.92% | 57.65% | 14.25% | | 4時間足 | 2022 | 35.36% | 58.00% | 9.93% | | 4時間足 | 2023 | 36.36% | 60.35% | 10.50% | | 4時間足 | 2024 | 36.88% | 58.54% | 10.89% | | 4時間足 | 2025 | 34.66% | 59.94% | 15.16% | 基準線は12区分すべてで57.7〜61.3%の足が横ばいでした。価格と基準線のクロスは2025年の勝率26.26%でしたが、平均の勝ち幅+67.47pipsに対して平均の負け幅は−21.62pipsでした。 ## 去年よかった読み方は今年もよいか なりませんでした。1時間足で、2024年にいちばん良かった設定をそのまま2025年へ移します。 | 読み方 | 2024年にいちばん良かった設定 | 2024年 | 同じ設定の2025年 | 2025年の中央値 | |---|---|---:|---:|---:| | 転換線×基準線 | 5/22 | +2,954.4pips | −152.0pips | −35.5pips | | 価格×基準線 | 基準線20 | +3,224.8pips | +1,072.0pips | +1,236.6pips | | 価格×雲 | 5/30/72 | +2,069.9pips | −10.3pips | +437.7pips | | 三役 | 9/22/72 | +1,736.0pips | +824.1pips | +259.4pips | 単独3ルールで前年にいちばん良かった設定は、翌年の中央値を下回りました。三役で前年にいちばん良かった9/22/72は翌年も+824.1pipsで、中央値を上回ったものの125設定中24位です。1年の首位がそのまま翌年の首位になったわけではありません。 ## 標準設定9/26/52はどうか 三役が勝率と年間損益の両方で4ルール中最大でした。標準9/26/52を1時間足・2025年で比べます。 | 読み方 | 取引回数 | 勝率 | 勝ちの平均 | 負けの平均 | 年間損益 | |---|---:|---:|---:|---:|---:| | 転換線×基準線 | 270 | 40.37% | +65.67pips | −44.05pips | +67.0pips | | 価格×基準線 | 556 | 26.26% | +67.47pips | −21.62pips | +985.3pips | | 価格×雲 | 141 | 35.46% | +103.26pips | −49.23pips | +683.3pips | | 三役 | 99 | 43.43% | +103.63pips | −60.20pips | +1,085.2pips | 価格と基準線のクロスは4回に1回ほどしか勝っていませんが、+985.3pipsでした。三役は取引回数を99回まで絞り、勝率と年間損益の両方で4ルール中最大です。ただし125設定の中央値は+259.4pipsなので、標準9/26/52だけを見て三役全体を評価すると上振れを含みます。 ## ADXや損切りを足すとどうなるか 効果は読み方と年で変わり、ADX25は価格と雲を両年赤字にしました。標準9/26/52の1時間足へ条件を追加した結果です。 | 読み方 | 追加条件 | 2024年（取引回数・損益） | 2025年（取引回数・損益） | |---|---|---:|---:| | 転換線×基準線 | なし | 269回・+2,041.0pips | 270回・+67.0pips | | 転換線×基準線 | ADX25以上 | 40回・+1,091.3pips | 39回・+753.0pips | | 転換線×基準線 | 損切り50・利確100 | 267回・+2,310.8pips | 260回・+2,013.8pips | | 価格×基準線 | なし | 483回・+3,121.8pips | 556回・+985.3pips | | 価格×基準線 | ADX25以上 | 52回・+251.2pips | 52回・+820.3pips | | 価格×基準線 | 損切り50・利確100 | 451回・+2,303.1pips | 521回・+2,264.6pips | | 価格×雲 | なし | 135回・+902.4pips | 141回・+683.3pips | | 価格×雲 | ADX25以上 | 64回・−13.9pips | 65回・−996.8pips | | 価格×雲 | 損切り50・利確100 | 395回・+935.4pips | 355回・+1,558.2pips | | 三役 | なし | 105回・+536.0pips | 99回・+1,085.2pips | | 三役 | ADX25以上 | 53回・+447.0pips | 56回・−590.1pips | | 三役 | 損切り50・利確100 | 309回・+901.8pips | 292回・+1,046.0pips | ADX25以上は転換線と基準線の2025年を改善しましたが、価格と雲は両年、三役は2025年に悪化しました。損切り50・利確100も三役の2025年を+39.2pips減らしています。追加条件の効果は読み方と年で変わりました。 ## スプレッドに弱いのはどれか 転換線と基準線のクロスです。1時間足・2025年・標準設定でスプレッドだけを変えました。 | 読み方 | 取引回数 | 0pips | 0.3pips | 0.6pips | 損益分岐スプレッド | |---|---:|---:|---:|---:|---:| | 転換線×基準線 | 270 | +148.0 | +67.0 | −14.0 | 0.55pips | | 価格×基準線 | 556 | +1,152.0 | +985.3 | +818.5 | 2.07pips | | 価格×雲 | 141 | +725.5 | +683.3 | +641.0 | 5.15pips | | 三役 | 99 | +1,114.8 | +1,085.2 | +1,055.5 | 11.26pips | 転換線と基準線のクロスは、スプレッドが0.6pipsになると赤字です。標準設定を使う場合でも、通貨ペアや時間帯の実際のコストを入れずに判定できる余裕はありません。 ## 関連記事 - [移動平均線クロスの設定検証](/ja/blog/moving-average-cross-settings)：2本の線が交差する同じ形を、終値平均で比較 - [アリゲーターの使い方と設定](/ja/blog/alligator-settings)：複数の線と前方シフトを使う指標との違い - [パラボリックSARの設定検証](/ja/blog/parabolic-sar-settings)：勝率より勝ち幅と負け幅が効いた別のトレンド指標 - [ドル円の設定は他の通貨ペアで通用する？3,024通り検証](/ja/blog/settings-across-pairs)：この設定は2024年と2025年の15分足でドル円1位でしたが、ユーロドルではどちらの年も34位でした ## 補足 検証はFormiqのバックテスト機能で、2022〜2025年のUSD/JPYを終値約定・0.3pips固定で測定しました。先行スパンを現在位置で使うには52本の計算期間と26本の描画ずらしが必要なため、2021年12月から過去データを読み込んでいます。遅行スパンの直接比較は先読みを避け、現在の終値と26本前の終値に置き換えています。 --- ### パーフェクトオーダーの設定比較：2,400通り検証 URL: https://formiq.jp/ja/blog/perfect-order-settings Language: ja Published: 2026-09-02 Updated: 2026-09-02 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: パーフェクトオーダー, 移動平均線, 設定, トレンド, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/perfect-order-settings パーフェクトオーダーは、期間の違う3本の移動平均線が短期・中期・長期の順に並んだ状態です。短期線が最上部なら上昇、最下部なら下降と読みます。移動平均線を使うため、ローソク足の動きに遅れて完成します。 ドル円で40通りの期間組み合わせと5種類の移動平均を比べると、**2024年と2025年の両方で黒字だった設定は15分足157/200、1時間足87/200、4時間足10/200**でした。15分足は多くの設定が残りましたが、4時間足は2024年の184設定から2025年の15設定へ減っています。時間足を長くすれば安定する、という結果ではありません。 ## パーフェクトオーダーは何を測っているのか 計算するのは3本とも普通の移動平均です。上昇のパーフェクトオーダーは次の並びです。 `短期の移動平均 > 中期の移動平均 > 長期の移動平均` 下降は不等号が反対になります。3本のうち1組でも順番が崩れれば、パーフェクトオーダーではありません。 初期設定の10/25/75・SMAは、2024年と2025年の15分足で62.84〜63.55%の足に成立していました。1時間足も63.12〜63.46%です。成立している時間が長いため、この記事では並びが完成した最初の終値で入り、崩れた終値で決済します。成立中の足を毎回新しいシグナルとは数えません。 ## パーフェクトオーダーの設定と読み方 パーフェクトオーダーはローソク足と同じメインチャートへ3本の線を重ねます。別枠のオシレーターや水準線はありません。上昇時は短期線、中期線、長期線の順に上から並び、下降時は上下が反対になります。 Formiqの条件設定にある項目は次の4つです。 | 設定項目 | 初期設定 | 意味 | |---|---:|---| | 短期 | 10 | 値動きへ最も速く反応する移動平均の期間 | | 中期 | 25 | 短期線と長期線の間に置く期間 | | 長期 | 75 | 大きな方向を表す、最も遅い移動平均の期間 | | 種類 | SMA | SMA・EMA・WMA・SMMA・HMAから計算方法を選ぶ | この記事の売買ルールは、設定画面の判定と1対1で対応しています。 | 状態 | 売買 | |---|---| | 短期線 > 中期線 > 長期線が完成 | 買い | | 短期線 < 中期線 < 長期線が完成 | 売り | | 保有方向の並びが1組でも崩れる | 決済 | 期間を短くすると完成と崩れが増え、取引回数も増えます。移動平均の種類を変えると、同じ期間でも線の反応速度と並びが変わります。短期・中期・長期は必ず小さい順に設定しないと、名前と計算の役割が一致しません。 ## 検証の条件 | 項目 | 条件 | |---|---| | 通貨ペア | USDJPY | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 短期5・10・20・25、中期20・25・50・75、長期50・75・100・200から、短期 < 中期 < 長期になる40組 | | 移動平均 | SMA・EMA・WMA・SMMA・HMAの5種類 | | 検証区分 | 2024年、2025年、2025年前半、2025年後半 | | エントリー | 3本の並びが完成した足の終値 | | 決済 | 保有方向の並びが崩れた足の終値 | | コスト | スプレッド0.3pips、スリッページ0 | | ロット | 0.1 | 1区分につき200設定です。3時間足と4区分を合わせて2,400回バックテストしました。損益は実際の約定価格からpipsへ再計算しています。 ## 4時間足でも使えるか 両年黒字は15分足の157設定に対し、4時間足は10設定でした。 | 時間足 | 2024年に黒字 | 2025年に黒字 | 両方の年で黒字 | |---|---:|---:|---:| | 15分足 | 182/200 | 165/200 | 157/200 | | 1時間足 | 193/200 | 91/200 | 87/200 | | 4時間足 | 184/200 | 15/200 | 10/200 | 15分足は2年を通して残った設定が最も多く、2025年の年間損益中央値も+646.8pipsでした。1時間足は2024年の中央値+1,407.2pipsから2025年−36.4pipsへ低下しています。4時間足は+1,588.0pipsから−812.0pipsへ反転しました。 長い時間足ほど取引回数は少なくなりますが、少ない取引回数が設定の安定につながったわけではありません。4時間足の中央値は年29〜48回で、年の違いを吸収するには少なすぎました。 ## SMAとEMAはどちらがいいか 時間足で変わり、15分足はEMA、1時間足はHMAが最多でした。移動平均の種類ごとに、2024年と2025年の両方で黒字だった設定を数えました。各種類には40通りの期間組み合わせがあります。 | 移動平均 | 15分足 | 1時間足 | 4時間足 | |---|---:|---:|---:| | SMA | 37/40 | 12/40 | 0/40 | | EMA | 40/40 | 13/40 | 0/40 | | WMA | 34/40 | 26/40 | 2/40 | | SMMA | 32/40 | 0/40 | 5/40 | | HMA | 14/40 | 36/40 | 3/40 | 15分足ではEMAが40/40、1時間足ではHMAが36/40で最多でした。EMAを選べばよい、HMAを選べばよい、という一つの答えにはなりません。4時間足ではどの種類も5/40以下です。 期間だけを見ても同じです。両年黒字が最も多かった長期線は15分足で100の53/65、1時間足で100の28/65でした。ただし4時間足の長期100は5/65です。期間100という数字だけを別の時間足へ移しても、同じ結果にはなりませんでした。 ## 期間は何がよいか 初期設定の10/25/75・SMAは、15分足と1時間足で両年黒字でした。 | 時間足 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | 15分足 | 2024 | 728 | 38.74% | +2,622.5pips | | 15分足 | 2025 | 749 | 35.51% | +646.8pips | | 1時間足 | 2024 | 182 | 34.07% | +780.2pips | | 1時間足 | 2025 | 186 | 40.86% | +770.0pips | | 4時間足 | 2024 | 43 | 46.51% | +1,714.5pips | | 4時間足 | 2025 | 49 | 32.65% | −904.3pips | 初期設定は15分足と1時間足で両年黒字でした。4時間足は2024年の+1,714.5pipsから2025年−904.3pipsへ反転しています。2025年を前半と後半に分けても、1時間足は前半−146.7pips、後半+892.6pips。4時間足は前半+389.1pips、後半−1,286.4pipsでした。年間が黒字でも、半年ごとの動きは揃っていません。 ## 去年よかった設定を今年に当てると 翌年の中央値を安定して上回りませんでした。各年でいちばん稼いだ設定を、変えずに反対の年へ移しました。 ### 2024年にいちばん稼いだ設定を2025年へ移す | 時間足 | 設定 | 2024年 | 2025年 | 2025年中央値 | |---|---|---:|---:|---:| | 15分足 | 10/50/100・WMA | +2,866.4 | +880.9 | +646.8 | | 1時間足 | 10/20/200・WMA | +3,215.1 | +319.1 | −36.4 | | 4時間足 | 10/20/75・EMA | +2,861.7 | −1,043.4 | −812.0 | ### 2025年にいちばん稼いだ設定を2024年へ移す | 時間足 | 設定 | 2025年 | 2024年 | 2024年中央値 | |---|---|---:|---:|---:| | 15分足 | 20/25/50・WMA | +2,267.6 | +1,229.3 | +1,272.2 | | 1時間足 | 5/25/200・HMA | +2,164.5 | +130.9 | +1,407.2 | | 4時間足 | 10/20/50・HMA | +1,053.1 | −242.0 | +1,588.0 | 15分足では2024年にいちばん稼いだ設定が翌年の中央値を上回りましたが、2025年のほうを2024年へ戻すと中央値を42.9pips下回りました。1時間足で2025年にいちばん稼いだ設定は反対の年で+130.9pipsとなり、2024年中央値+1,407.2pipsには届きません。4時間足はどちらの年のものも反対の年で赤字でした。 ## 勝率34%でも勝てるのはなぜか 初期設定の1時間足は、2024年の勝率が34.07%でも+780.2pipsでした。平均利益+91.88pipsに対して平均損失は−40.97pipsです。2025年は勝率40.86%、平均利益+69.78pips、平均損失−41.21pipsで+770.0pipsでした。 パーフェクトオーダーは小さな崩れで損切りし、並びが長く続いた取引で取り返す形です。勝率だけを上げると、長く伸びる取引を途中で切る可能性があります。取引回数、平均利益、平均損失、年間損益を一緒に確認する必要があります。 ## ADXや損切り利確で改善するか 初期設定へADX 25以上を追加すると、15分足は2024年+1,826.7pips、2025年−117.6pips。1時間足は+935.0pips、−295.1pips。4時間足は+2,079.1pips、−2,433.7pipsでした。取引を絞っても2年を通した改善にはなっていません。 損切り50・利確100は、15分足を2024年+2,807.5pips、2025年+781.2pipsへ改善しました。一方、4時間足は2024年+1,186.8pipsへ低下し、2025年は+120.2pipsへ改善しています。同じ損切り利確を全時間足へ付ける根拠にはなりません。 スプレッド0から0.3pipsへ上げた差は、取引回数×0.3pipsと一致します。2025年の初期設定は15分足で+871.5pipsから+646.8pips、1時間足で+825.7pipsから+770.0pips、4時間足で−889.7pipsから−904.3pipsです。4時間足の赤字はスプレッドをゼロにしても残るため、コストだけが原因ではありません。 ## 関連記事 [移動平均線クロス](/blog/moving-average-cross-settings)は2本の交差を測り、パーフェクトオーダーは3本すべての並びを測ります。[SMAとEMAの比較](/blog/sma-vs-ema)では、計算方法を交換したときの差を期間変更と比べています。[GMMA](/blog/gmma-settings)は12本の移動平均線を使い、本数を減らしたときに情報が残るかを検証しています。[アリゲーター](/blog/alligator-settings)も3本の並びを使いますが、線を前方へずらす点が異なります。 ## 補足 - 対象はUSDJPYの2024年と2025年。2025年だけ前半と後半も分けました - シグナル判定と約定は終値、スプレッドは0.3pips固定です - 各区分の前に最大200本を読み込み、長期移動平均を計算してから検証期間へ入っています - 並びが完成した後の押し目買い・戻り売り、移動平均の傾き、線どうしの間隔は検証していません --- ### 移動平均線乖離率の逆張り：4,320通りの設定比較 URL: https://formiq.jp/ja/blog/ma-disparity-settings Language: ja Published: 2026-08-29 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: 移動平均線乖離率, オシレーター, 逆張り, 順張り, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/ma-disparity-settings 移動平均線乖離率は、終値が移動平均線からどれだけ離れているかを、移動平均に対する割合で表したものです。式は1行しかありません。 **乖離率 =（終値 − 移動平均）÷ 移動平均 × 100** 紹介のされ方はほぼ決まっています。**期間25の移動平均に対して±5%まで離れたら行き過ぎで、そこから平均に戻ってくる**。日本語で書かれた解説のほとんどがこの数字を挙げます。 そこから測りました。**ドル円の2024年と2025年、15分足・1時間足・4時間足のどれでも、期間25の乖離率が±5%に届いた足は1本もありません。** ## 移動平均線乖離率は何を測っているのか 分子は「価格が平均からどれだけ離れたか」、分母は「そのときの価格の水準」です。単位は％で、ゼロは価格が移動平均線の上に乗っている状態を指します。 ここが他のオシレーターと違う点です。RSIやストキャスティクスは0から100の枠に押し込まれていて、どの通貨ペアでも70や30という数字が同じ意味を持ちます。乖離率は押し込まれていません。**そのときの価格に対する割合なので、届く範囲は市場と時間足によって変わります。** 期間25の単純移動平均で実測すると、こうなります。 | 時間足 | 中央値 | 9割の足が入る範囲 | 2024年の最大 | 2025年の最大 | |---|---|---|---|---| | 15分足 | 0.09% | 0.27% | 1.94% | 1.94% | | 1時間足 | 0.19% | 0.56% | 2.78% | 1.67% | | 4時間足 | 0.41% | 1.05% | 4.54% | 2.35% | 中央値と9割の範囲は2025年の値です。**同じ「±5%」という水準が、4時間足では2年に1度も届かない極端値で、15分足では最大値の2.6倍あります。** ### ±5%はどこから来たのか 株式市場です。25日移動平均に対する乖離率が±5%を超えたら過熱、という使われ方が日本の株式で定着していて、それがそのまま為替の解説に移っています。**日足の株価と15分足の通貨ペアでは、平均から離れる幅が2桁違います。** 期間を伸ばせば届きます。2年間で±5%を超えた足があったのは3通りだけでした。 | 時間足 | 期間 | 年 | ±5%を超えた足 | その年の最大 | |---|---:|---:|---:|---:| | 4時間足 | 200 | 2024年 | 4.02% | 9.60% | | 4時間足 | 75 | 2024年 | 0.39% | 7.08% | | 1時間足 | 200 | 2024年 | 0.22% | 6.25% | どれも2024年です。2024年のドル円は年間で1,632pips上昇し、値幅は2,237pipsありました。2025年は年間で56pips下落し、値幅は1,900pipsです。**この2年間で±5%を超えたのは、2024年の期間75以上だけでした。** ## ％という単位は安定しているか 安定しません。同じ1%が、価格水準によって別の距離になります。「価格に対する割合」が実務上何を表すのか、2つの角度から測りました。 ### 1%が何pipsかは、年の中で16%変わる 2024年のドル円は139.723円から161.940円まで動いています。**1%という同じ乖離幅が、年の安いところでは139.7pips、高いところでは161.9pipsです。** 同じ水準を打ち込んでいるつもりで、実際には別の距離を待っていることになります。 2025年は139.923円から158.663円で、1%は139.9pipsから158.7pipsでした。 ### 同じ水準が、四半期によって2倍から3.6倍の頻度で成立する pipsの話より効くのはこちらです。％で測った水準と、同じ乖離をATR倍（その期間の平均的な値幅の何倍か）で測った水準を並べて、**どちらも「その年の2割の足で成立する」ように合わせてから**、四半期ごとの成立率を比べました。 | 時間足 | 年 | ％：最も静かな四半期→最も忙しい四半期 | ATR倍：最も静かな四半期→最も忙しい四半期 | |---|---:|---|---| | 15分足 | 2024年 | 11.4% → 32.4%（2.83倍） | 18.0% → 21.9%（1.22倍） | | 15分足 | 2025年 | 12.5% → 28.6%（2.30倍） | 18.0% → 21.8%（1.22倍） | | 1時間足 | 2024年 | 12.3% → 32.8%（2.66倍） | 17.2% → 21.6%（1.26倍） | | 1時間足 | 2025年 | 12.8% → 31.3%（2.44倍） | 16.2% → 22.6%（1.40倍） | | 4時間足 | 2024年 | 9.7% → 34.5%（3.56倍） | 18.5% → 20.9%（1.13倍） | | 4時間足 | 2025年 | 14.9% → 30.5%（2.05倍） | 16.3% → 22.7%（1.39倍） | **6区分すべてで、％のほうが振れました。** ％で打ち込んだ水準は、静かな四半期にはほとんど成立せず、荒れた四半期には3倍の頻度で成立します。ATR倍で打ち込んだ水準は16〜23%の間に収まりました。 ### ただし、成績は％のほうが良かった ここで話が折り返します。単位として素直なのはATR倍のはずなので、成績もそちらが上だと予想して測りました。**結果は逆です。** 同じ読み方・同じ期間・同じ時間足・同じ年で、それぞれの単位に**自分にいちばん都合のいい水準を選ばせて**比べた90区分のうち、ATR倍が勝ったのは26区分でした。中央値どうしの比較でも33区分です。 理由は取引回数です。ATR倍は静かな相場でも同じ頻度で成立するので、そのぶん売買が増えます。15分足の買いの逆張りでは、取引回数の中央値が％で年29回、ATR倍で161回でした。**この記事の最後にあるとおりコストは取引回数×スプレッドなので、成立を安定させることは、成立を増やすことでもあります。** 安定した単位が儲かる単位とは限りませんでした。 ## ゼロに戻るのは価格が戻るからか 戻しているのは価格ではなく、追いついてくる移動平均のほうでした。「乖離率は必ずゼロに戻る」という説明は、実測でも正しいです。問題はその戻り方です。 水準を超えた足から、乖離率が符号を変えるまでを1回の往復として、**その往復を「価格が動いた分」と「移動平均が追いついた分」に分けました。** 2つの寄与を足すと100%になります。 | 時間足 | 年 | 価格の寄与 | 平均の寄与 | 往復の本数 | |---|---:|---:|---:|---:| | 15分足 | 2024年 | −1.6% | 101.6% | 20.2本 | | 15分足 | 2025年 | −0.7% | 100.7% | 19.5本 | | 1時間足 | 2024年 | −26.8% | 126.8% | 20.1本 | | 1時間足 | 2025年 | −4.0% | 104.0% | 16.7本 | | 4時間足 | 2024年 | −24.1% | 124.1% | 20.6本 | | 4時間足 | 2025年 | +9.2% | 90.8% | 17.3本 | 期間25、水準0.25%（4時間足は0.5%）での値です。期間10と75も含めた18区分では、**価格の寄与がマイナスだったのは13区分**でした。平均の寄与は61.6%から139.6%の範囲です。 読み方はこうです。**価格が動かなくても、移動平均のほうが価格に近づいていくので、乖離率はゼロに戻ります。** 「行き過ぎたら戻る」の「戻る」は、多くの場合は価格の話ではありません。 回数で見ると印象が変わります。価格が入った方向に戻したのは、18区分で52.0%から78.1%の往復でした。**半分以上の往復では価格も戻っています。それでも平均するとマイナスなのは、戻らなかった往復のほうが大きく動いているからです。** 勝率が高くて損益がマイナスになる形が、値動きの側にそのまま出ています。 ### 順番を壊しても同じことが起きる 1本ごとの終値の変化はそのままに、並び順だけをランダムに入れ替えた系列（値幅も年間の上昇幅もテールも本物のまま）で、同じ測定をしました。2025年、期間25です。 | 時間足 | 9割の範囲（実データ／並べ替え） | 0.25%超えの割合 | ゼロに戻るまでの本数 | |---|---|---|---| | 15分足 | 0.269% ／ 0.258% | 11.9% ／ 10.9% | 19.5本 ／ 20.7本 | | 1時間足 | 0.558% ／ 0.521% | 39.6% ／ 39.5% | 16.7本 ／ 17.1本 | | 4時間足 | 1.051% ／ 0.998% | 69.5% ／ 67.3% | 15.2本 ／ 12.7本 | **ほとんど同じです。** 乖離率が広がってゼロに戻る往復は、相場の性質ではなく、終値から終値の移動平均を引くという計算の性質でした。 違いが出たのは端のほうだけです。99%点は実データが0.563%・1.129%・1.997%、並べ替えが0.462%・0.899%・1.573%で、**大きく離れる場面だけは本物の相場のほうが多くあります。** 往復は計算のもので、極端値は相場のものです。 ## ゼロラインクロスは何なのか 移動平均線クロスそのものです。乖離率の分母は価格で、価格は負になりません。ということは、**乖離率の符号は（終値 − 移動平均）の符号と一致します。** 実測でも、時間足3種×期間5種の15通りで不一致は0件でした。％で測ってもATR倍で測っても符号は同じで、こちらも15通りで不一致0件です。 バックテストでも確かめました。乖離率のゼロクロスと、終値（＝期間1の単純移動平均）と期間Nの移動平均のクロスを、それぞれ1つの手法として回します。 | 期間 | 両手法の取引回数 | 両手法の年間損益 | |---:|---:|---:| | 5 | 1,688回 | +1,323.4pips | | 10 | 1,053回 | +2,007.6pips | | 25 | 600回 | +2,227.8pips | | 75 | 335回 | +787.0pips | | 200 | 268回 | −1,966.8pips | 1時間足・2025年。平均の種類を単純移動平均から指数移動平均・加重移動平均・平滑移動平均・ハル移動平均に変えても、5種すべてで取引数もpipsも一致しました。 **乖離率のゼロラインを見ることは、移動平均線クロスを見ることです。** 別の指標を足したことにはなりません。 なお単位の違いがゼロクロスの結果を変えた区分は、60区分のうち1つだけありました。4時間足・2024年・期間200です。ATR倍のほうはATRが落ち着くまでのウォームアップを2倍取るので、線の始まりが遅れて最初の数回の売買がずれます。符号が変わったわけではありません。 ## チャートに表示する方法 事実として並記します。 - **MT4／MT5**：乖離率という名前の標準インジケーターはありません。標準で入っているのは**エンベロープ**（挿入→インディケータ→トレンド→Envelopes）で、これは移動平均の±d%に線を引いたものです。つまり同じ計算を、別枠のオシレーターではなく価格チャート上の2本の線として描いています。乖離率そのものを別枠に出したい場合は、配布されている .mq4 / .mq5 ファイルを探してインストールします。 - **TradingView**：インジケーター検索で「Disparity Index」と入れると公開スクリプトが見つかります。標準搭載のリストにはありません。 - **Formiq（ブラウザ）**：インジケーター一覧の「移動平均線乖離率」から追加します。期間・水準線・測る単位（％／ATR倍）を設定できます。バックテスト側でも条件として使えて、読み方は4種（水準・水準の行き過ぎ・±水準の抜け・ゼロラインクロス）です。 ### エンベロープとの違いは、ヒゲか終値か エンベロープの下限は `移動平均 ×（1 − d/100）` なので、**下限にタッチすることと、乖離率が −d% を下回ることは同じ条件です。** 違うのは判定に使う価格だけで、エンベロープは足の安値と高値を見て、乖離率は終値を見ます。 下限で入って上限で決済する形に揃えて、期間25で並べました。 | 判定 | 年 | 時間足 | 水準 | 売買方向 | 取引回数 | 年間損益 | |---|---:|---|---:|---|---:|---:| | エンベロープ（ヒゲ） | 2025年 | 1時間足 | 0.5% | 買い | 40回 | +26.8pips | | 乖離率（終値） | 2025年 | 1時間足 | 0.5% | 買い | 31回 | +769.9pips | | エンベロープ（ヒゲ） | 2024年 | 1時間足 | 0.5% | 買い | 46回 | +873.0pips | | 乖離率（終値） | 2024年 | 1時間足 | 0.5% | 買い | 29回 | +169.5pips | 時間足3種×2年×水準3種×売買2方向の36区分で、**終値判定の取引回数がヒゲ判定を上回った区分は0でした（36区分すべてで同じか少ない）。** 一方で損益が上回ったのは36区分中20です。**終値で判定すると売買は確実に減りますが、減った先が良いとは限りませんでした。** ## 検証の条件 読者が同じ数字を出せる粒度で書きます。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 5・10・25・75・200 | | 水準（％） | 0.15・0.3・0.5・0.8・1.2 | | 水準（ATR倍） | 1.0・1.5・2.0・2.5・3.0 | | 決済水準（％） | 0（平均に戻る）・0.15・0.3 | | 決済水準（ATR倍） | 0・0.75・1.5 | | 平均の種類 | 単純・指数・加重・平滑・ハル | | 検証期間 | 2024年／2025年／2025年前半／2025年後半 | | スプレッド | 0.3pips | | ロット | 0.1 | | 約定 | 終値 | | 損切り・利確 | 無し（別途セクションで測定） | | 組み合わせ数 | 4,320（平均の種類は別枠で360） | 読み方は3つに分けました。 1. **逆張り**：水準を超えたら反対側に入り、平均に戻ったら決済する。買いと売りのどちらか一方しか作らないので、買い側と売り側を別の手法として回しています 2. **順張り**：+水準を上抜けたら買い、−水準を下抜けたら売り。対称なのでエントリーと決済に同じ条件を入れます 3. **ゼロラインクロス**：ゼロを上抜けたら買い、下抜けたら売り ## 3つの読み方はどう違うか 逆張りは勝率65%前後で黒字が多く、ゼロクロスは勝率24%で1時間足だけが残ります。2025年、％で測った場合です。 | 読み方 | 時間足 | 黒字 | 中央値 | 平均勝率 | 中央の取引回数 | |---|---|---|---|---|---| | 逆張り・買い | 15分足 | 35/75 | +0.0pips | 65.2% | 29回 | | 逆張り・買い | 1時間足 | 42/75 | +110.5pips | 〃 | 30回 | | 逆張り・買い | 4時間足 | 60/75 | +466.2pips | 〃 | 20回 | | 逆張り・売り | 15分足 | 31/75 | −76.5pips | 66.4% | 23回 | | 逆張り・売り | 1時間足 | 53/75 | +201.4pips | 〃 | 27回 | | 逆張り・売り | 4時間足 | 63/75 | +718.9pips | 〃 | 21回 | | 順張り | 15分足 | 15/25 | +300.0pips | 36.7% | 37回 | | 順張り | 1時間足 | 9/25 | −749.4pips | 〃 | 42回 | | 順張り | 4時間足 | 8/25 | −851.6pips | 〃 | 30回 | | ゼロクロス | 15分足 | 2/5 | −641.6pips | 23.8% | 2,704回 | | ゼロクロス | 1時間足 | 4/5 | +1,323.4pips | 〃 | 600回 | | ゼロクロス | 4時間足 | 1/5 | −918.6pips | 〃 | 149回 | 平均勝率は時間足をまたいだ全体の値です。**逆張りは勝率65%前後で黒字が多く、ゼロクロスは勝率24%で1時間足だけが残ります。** シリーズ12本目までで見えていた「逆張りは勝率が高くて負ける」という形が、2025年の逆張りには当てはまりません。 当てはまらない理由は、次の2つの節にあります。 ## 去年勝てた設定は今年も勝てるか ほとんど残りませんでした。2025年の年間損益が最も大きかった設定を、同じ条件のまま2024年に当て直します。 | 読み方 | 時間足 | 期間 | エントリー水準 | 決済水準 | 2025年の取引回数 | 2025年の年間損益 | 2024年の年間損益 | |---|---|---:|---:|---:|---:|---:|---:| | 逆張り・買い | 15分足 | 200 | 0.8% | 0.15% | 34回 | +1,368.6pips | +118.9pips | | 逆張り・買い | 1時間足 | 200 | 0.5% | 0% | 32回 | +1,327.8pips | −511.2pips | | 逆張り・買い | 4時間足 | 75 | 0.5% | 0.3% | 18回 | +1,373.6pips | −394.8pips | | 逆張り・売り | 4時間足 | 25 | 0.8% | 0.3% | 21回 | +1,419.2pips | −1,840.1pips | 4設定のうち、2024年も黒字だったのは15分足の買いだけです。逆方向でも、2024年に+2,155.5pipsだった15分足・逆張り買い・期間5・0.15%→0.3%は、2025年には+255.6pipsまで減りました。2025年の同じ75設定の中央値は0.0pipsです。 ## 勝率65%の逆張りは本物か 2024年の4時間足では、相場の流れを引くと72通りのうち1つも上乗せが残りませんでした。逆張りは買いと売りのどちらか一方しか作りません。**上昇した年に買いを持てば、何をきっかけに買っても上昇分は取れます。** そこで1取引あたりの損益から、「同じ本数だけ無条件に持っていたら得られた分」（年間の値幅 ÷ 足数 × 平均保有本数）を引きました。 | 時間足 | 年 | 売買方向 | 1取引の損益 | 相場の流れ | 指標の上乗せ | 上乗せがプラスの設定 | |---|---:|---|---:|---:|---:|---:| | 4時間足 | 2025年 | 買い | +29.20pips | −1.24pips | +30.44pips | 60/69 | | 4時間足 | 2025年 | 売り | +40.61pips | +1.40pips | +39.21pips | 60/70 | | 4時間足 | 2024年 | 買い | −22.54pips | +46.07pips | −68.61pips | **0/72** | | 1時間足 | 2024年 | 買い | −10.91pips | +21.22pips | −32.13pips | 5/72 | | 1時間足 | 2025年 | 買い | +10.70pips | −0.74pips | +11.44pips | 42/63 | **2024年の4時間足では、72通りのうち1つも上乗せがプラスになりませんでした。** ATR倍で測った場合も57通り中0通りです。 2025年はドル円が年間56pips下落しただけの横ばいの年で、そこでは逆張りの上乗せがはっきり残ります。2024年は1,632pips上昇した年で、そこでは全滅します。**「乖離率の逆張りは効く」と書けるかどうかは、どちらの年を見たかで決まっていました。** 売り側は逆の非対称を持ちます。2024年の4時間足の売りは1取引−74.08pipsですが、そのうち−76.04pipsが相場の流れで、上乗せは+1.96pipsです。**上昇した年に売り続けた損は、指標のせいではありません。** ## どの平均から測るか 乖離率は「移動平均からの乖離」なので、移動平均線の種類も設定結果を変えるパラメータです。期間とその他の条件を固定し、5種類の移動平均線を比べました。 2025年・期間25について、単純移動平均とゼロの上下が食い違う足の割合を比べました。 | 平均の種類 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---| | 指数移動平均 | 7.0% | 6.5% | 5.3% | | 加重移動平均 | 8.2% | 8.0% | 8.3% | | 平滑移動平均 | 15.8% | 15.6% | 12.0% | | ハル移動平均 | 36.8% | 37.1% | 37.9% | **ハル移動平均は、4割近い足で単純移動平均と反対側を指します。** 平均の種類は表示の好みではなく、別の条件です。 成績では、読み方4種×時間足3種×2年×期間3種の72区分ごとに、年間損益が最も大きい平均を確認しました。該当した区分数はハル24・平滑14・加重13・単純11・指数10です。**どの平均も10区分以上で最大となり、単純移動平均だけが一貫して優位という結果ではありませんでした。** ## フィルターと損切りとコスト ### コストは取引回数×スプレッド シリーズを通して例外が出ていない関係です。ここでも、表示している桁（0.1pips）まで一致しました。 | 手法 | 期間 | 水準 | 時間足 | 年 | 取引回数 | スプレッド0の年間損益 | スプレッド0.3の年間損益 | 差 | |---|---:|---:|---|---:|---:|---:|---:|---:| | ゼロクロス | 25 | 0 | 15分足 | 2025年 | 2,704回 | +169.6pips | −641.6pips | 811.2pips | | ゼロクロス | 25 | 0 | 1時間足 | 2025年 | 600回 | +2,407.7pips | +2,227.8pips | 179.9pips | | 順張り | 25 | ±0.5% | 1時間足 | 2025年 | 62回 | −1,662.1pips | −1,680.6pips | 18.5pips | **15分足のゼロクロスは、スプレッド0なら+169.6pipsです。** 損益分岐スプレッドは0.063pipsで、実質的に「コストで負けている」手法でした。1時間足の同じ設定は損益分岐が4.013pipsあります。 ### フィルターと損切り 1時間足・期間25のゼロクロスで測りました。 | 条件 | 年 | 取引回数 | 年間損益 | |---|---:|---:|---:| | そのまま | 2025年 | 600回 | +2,227.8pips | | そのまま | 2024年 | 601回 | +2,546.5pips | | ADX 25以上 | 2025年 | 176回 | +554.8pips | | ADX 25以上 | 2024年 | 161回 | +24.9pips | | ロンドン＋NY時間 | 2025年 | 328回 | +1,793.1pips | | ロンドン＋NY時間 | 2024年 | 353回 | +892.6pips | | 東京時間 | 2025年 | 247回 | +564.3pips | | 東京時間 | 2024年 | 210回 | +2,427.2pips | | 損切り40pips | 2025年 | 558回 | +2,461.7pips | | 損切り40pips | 2024年 | 559回 | +3,612.6pips | | 利確40pips | 2025年 | 599回 | +2,403.1pips | | 利確40pips | 2024年 | 598回 | +1,223.9pips | **ADXも時間帯フィルターも、2年ともに改善した組み合わせはありません。** ADXはどちらの年も成績を下げました。時間帯は2025年ならロンドン＋NY、2024年なら東京で、逆を向いています。 **損切り40pipsだけが2年とも改善しました。** 15分足のゼロクロスでも、損切り20pipsで−641.6pipsが−108.6pipsまで戻ります（黒字にはなりません）。ゼロクロスは勝率24%・平均利益が平均損失の3倍以上という形なので、損切りを置いても勝ちトレードがほとんど切れません。 ## 補足 - **通貨ペアは米ドル／円1つだけです。** 乖離率の単位は価格に対する割合なので、価格水準の違う通貨ペアでは水準がそのまま移りません。上に書いた0.27%・0.56%・1.05%という数字は、ドル円のこの2年間の数字です - **順番を壊した対照群は終値だけの系列です。** 足の中の高値と安値は捨てているので、ヒゲを読むルールには使えません。この記事でエンベロープと比べた部分には適用していません - **約定は終値で、スリッページを0としています。** 15分足で年2,700回売買する設定では、実際の執行はこの数字より悪くなります ## 関連する検証 - [ボリンジャーバンドの設定](/ja/blog/bollinger-band-settings)：同じ「バンドに触れたら逆張り／抜けたら順張り」の形を、標準偏差で測った場合 - [移動平均線クロスの設定](/ja/blog/moving-average-cross-settings)：この記事のゼロラインクロスと同じもの - [CCIの設定](/ja/blog/cci-settings)：平均からの乖離を、乖離の平均で割って目盛りにした場合 - [サイコロジカルラインの設定](/ja/blog/psychological-line-settings)：相場の流れを引き算する測り方を最初に使った記事 --- ### 練行足「反転で売買」は本当に勝てる？1,080通り検証した結果 URL: https://formiq.jp/ja/blog/renko-chart-settings Language: ja Published: 2026-08-29 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: 練行足, レンコ足, 順張り, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/renko-chart-settings 練行足（れんこうあし、Renko）は、時間を捨てたチャートです。値動きが決めた幅ぶん動いたときだけブロックが1つ増え、動かない時間は何も描かれません。だから縦にきれいなブロックが並び、トレンドが見えます。 設定はふつう1つだけです。ブロックの大きさ。 ですが、ブロックの大きさを決めても、チャートはまだ決まりません。**水準が乗る目盛りをどこに置くかという選択が残っていて、それはどのチャートも黙って決めています。** 20pipsのブロックなら、目盛りの置きかたは20通りあります。 同じ20pipsのブロック・同じルール・同じドル円2025年の15分足で、目盛りだけを10通り動かしました。 損益は+221.9pipsから+3,082.5pipsまで開きました。取引回数は1,004回から1,057回で、ほぼ変わっていません。同じだけ売買して、結果だけが13.9倍違います。 ## 練行足は何を計算しているのか 必要なのは終値だけです。 いま水準が148.00にあって、ブロックが10pipsだとします。終値が148.10以上になったらブロックを1つ上に積み、水準は148.10へ動きます。147.90以下なら下に1つ。148.05で終わった足には、何も描かれません。 終値だけでブロックを作るため、水準・時間・ブロックの並びに3つの性質が生まれます。 ### 水準は目盛りの上にしか乗らない 水準はブロックの倍数の上にしか乗りません。10pipsのブロックなら、水準は…147.90／148.00／148.10…のどれかです。 **ではその目盛りの列は、どこを起点に並んでいるのか。** 148.00から10pipsおきなのか、148.03から10pipsおきなのか。どちらも同じ「10pipsのブロック」で、どちらも正しく、**設定として出しているチャートを見たことがありません。** 10pipsのブロックなら目盛りは10通り、20pipsなら20通り。実装が黙って1つ選んでいます。 ### 1本の足で、ブロックが何個も出る 終値が水準から35pips離れていれば、その足で3個積まれます。ブロックの数は足の数と一致しません。 ドル円2025年・10pipsのブロックで、ブロックが2個以上出た足は15分足で1,189本、1時間足で1,260本、4時間足で721本ありました。**「3ブロック続いた」を見た瞬間には、その3つが同時に確定していることがあります。** ### 開始日は忘れる。ただし何本かかるかはブロック次第 平均足は自分の開始日を忘れません（[平均足の記事](/ja/blog/heikin-ashi-settings)で測りました）。練行足は忘れます。水準は目盛りの上の整数番地なので、2つの系列が同じ番地に乗った瞬間から完全に一致します。 何本かかるかは、ブロックの大きさで変わります。ドル円2025年、開始位置を変えて突き合わせた実測です。 | 時間足 | ブロック | 中央値 | 95%点 | |---|---:|---:|---:| | 15分足 | 5pips | 2本 | 8本 | | 15分足 | 20pips | 11本 | 54本 | | 15分足 | 50pips | 36本 | 124本 | | 1時間足 | 5pips | 1本 | 3本 | | 1時間足 | 20pips | 4本 | 15本 | | 1時間足 | 50pips | 13.5本 | 36本 | | 4時間足 | 5pips | 1本 | 2本 | | 4時間足 | 20pips | 2本 | 5本 | | 4時間足 | 50pips | 4本 | 14本 | 一致に必要なのは価格の移動距離であって本数ではないので、ブロックを大きくするほど本数は伸びます。Formiqのバックテストは、この実測に合わせてブロック1pipsあたり8本の助走を取っています（10pipsなら131本、50pipsなら451本）。 ## 練行足の設定と読み方 練行足は線ではなく**ローソク足そのものの代わり**に描かれます。下に別枠は増えません。 | 環境 | 追加する場所 | |---|---| | TradingView | チャート上部の足種メニューから Renko | | MT4 / MT5 | 標準のチャート種類に無い。オフラインチャートを作るスクリプトか、ブロックを描くカスタムインジケーターを入れる | | ブラウザ（Formiq） | [チャート種類から練行足を選ぶ](/chart) | ### 設定ダイアログに並ぶ項目 | 項目 | 初期設定 | 意味 | |---|---|---| | ブロックの大きさ | Formiqは10pips（1〜200pips） | 何pips動いたらブロックを1つ積むか | | ブロックの決め方 | 固定値 | TradingViewは固定値のほかにATRから自動で決める方式も選べる。ATRを選ぶと、大きさが期間ごとに変わる | | 元にする価格 | 終値 | 終値だけで組むか、高値と安値も見るか。実装で分かれる | | 反転に必要なブロック数 | Formiqのバックテストは1（1〜10） | 逆方向にブロックを積むのに、何ブロックぶん戻す必要があるか。MT4系の作図は2が多い | | 待つブロック数 | Formiqのバックテストは3（1〜50） | 同じ向きが何ブロック続いたら売買するか | **この表に無い項目が1つあります。水準がブロックの何倍の位置に乗るか（目盛りの起点）です。** どの環境も設定として出しておらず、実装が黙って1つ選んでいます。この記事の中心はそこで、Formiqのバックテストだけは「目盛りのずらし」（初期設定の0pips）として入力できるようにしてあります。 ### 画面に何が出るか ブロックが縦に積まれ、**上向きと下向きで色が変わります。** 動かない時間はブロックが増えないので、同じ時刻の幅でも何も描かれない区間ができます。水準線は引かれません。 ### 売買のルールとして使われている形 もっとも広く紹介されているのは、**ブロックの向きが変わったら、その向きについていく**という形です。この記事はこれを「ブロックの向きの変化」として測っています。 もう1つよく使われるのが、**同じ向きがNブロック続いてから乗る**形です。1ブロックだけの反転に振り回されないための待ちで、この記事では2〜5ブロックまで測りました。 ブロックの向きが変わったら売買するルールと、同じ向きが2〜5ブロック続いてから売買するルールを、別々に検証しています。 ### どこをいじるべきか **ブロックを大きくすると取引回数が減り、小さくすると増えます。** 15分足では5pipsで年4,841回、50pipsで252回まで変わりました。反転に必要なブロック数を2にすると、取引回数はさらに43〜84%へ減ります。 TradingViewで1つ注意が要ります。**チャート種類を練行足にしたままストラテジーを動かすと、約定価格までブロックの値になります。** 後半で測っている差がそのまま乗ります。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | USD/JPY | | 時間足 | 15分足・1時間足・4時間足 | | 期間 | 2025年（本命）／2024年（アウトオブサンプル）／2025年前半・後半 | | ブロックの大きさ | 5・10・15・20・30・50pips | | 反転に必要なブロック | 1（出たまま描く）・2（伝統的な作図） | | 目盛りのずらし | 0からブロック幅まで10段階 | | 読み方 | ブロックの向きの変化／同じ向きがNブロック（N＝2〜5） | | 決済 | 反対のシグナルまで保有（エントリーと同じルール） | | スプレッド | 0.3pips | | ロット | 0.1 | | 約定 | 足の終値（ブロックの終値ではありません） | | 通り数 | 720（本体）＋360（目盛り）＝1,080 | Formiqのバックテスト機能で実測したものです。同じ設定を入力すれば同じ数字になります。 **約定はすべて足の終値です。** ブロックの終値ではありません。理由は後半で測ります。 ## 同じ設定でも損益が変わるのはなぜか 画面に出ない目盛りの起点が動くからです。ブロックの大きさ・読み方・年・時間足を固定して目盛りだけを動かすと、年間損益は+221.9から+3,082.5pipsまで変わりました。18通りの実測です。 | 時間足 | 年 | ブロック | 10通りの損益 | 幅 | チャートが描く目盛り | |---|---|---|---|---|---| | 15分足 | 2025 | 10pips | −719.9〜+724.5pips | 1,444.4 | −676.5pips | | 15分足 | 2025 | 20pips | +221.9〜+3,082.5pips | 2,860.6 | +1,381.7pips | | 1時間足 | 2025 | 10pips | +398.2〜+3,162.3pips | 2,764.1 | +2,368.6pips | | 1時間足 | 2025 | 20pips | +1,310.9〜+3,805.9pips | 2,495.0 | +1,310.9pips | | 4時間足 | 2025 | 30pips | −906.9〜+853.9pips | 1,760.8 | −906.9pips | | 4時間足 | 2024 | 10pips | +706.4〜+1,640.6pips | 934.2 | +925.2pips | 18通り全部で、目盛りの幅は**934.2pipsから2,860.6pips**でした。18通り中5通りでは、目盛り次第で黒字にも赤字にもなります。チャートが描く目盛り（ずらし0）は、18通り中3通りで年間損益が最小、1通りで最大でした。 15分足・2025年・10pipsは極端です。10通りの目盛りのうち黒字は1通りだけで、チャートが描く目盛りはその1通りではありませんでした。 ブロックの大きさを6通り変えたときの幅と比べます。 | 時間足 | 年 | 目盛り10通りの幅 | ブロック6通りの幅 | 比 | |---|---|---|---|---| | 15分足 | 2025 | 1,444.4〜2,860.6pips | 4,471.5pips | 0.32〜0.64 | | 15分足 | 2024 | 2,402.8〜2,782.7pips | 1,873.8pips | 1.28〜1.49 | | 1時間足 | 2025 | 2,112.8〜2,764.1pips | 3,740.0pips | 0.56〜0.74 | | 4時間足 | 2024 | 934.2〜1,875.2pips | 3,119.6pips | 0.30〜0.60 | 比は0.30から1.49、**18通り中3通りでは目盛りのずれのほうがブロック幅より大きい**という結果でした。設定できるブロック幅だけでなく、画面に表示されない目盛りの起点も同じpips単位でブロックの位置を変えています。 ## 良い目盛りを選んでおけばいいのか 前年の成績からは選べませんでした。「では良い目盛りを探せばいい」に見えます。2024年でいちばん良かった目盛りを、2025年に当てました。 | 時間足 | ブロック | 2024年に選んだずらし | 2024年の年間損益 | 2025年の年間損益 | 2025年の中央値 | |---|---:|---:|---:|---:|---:| | 15分足 | 10pips | 0pips | +1,738.7pips | −676.5pips | −497.0pips | | 15分足 | 20pips | 16pips | +2,471.1pips | +1,569.2pips | +1,818.5pips | | 15分足 | 30pips | 18pips | +3,483.2pips | +1,207.6pips | +1,723.7pips | | 1時間足 | 10pips | 8pips | +1,480.0pips | +398.2pips | +2,272.9pips | | 1時間足 | 20pips | 14pips | +2,661.7pips | +2,038.7pips | +2,345.9pips | | 1時間足 | 30pips | 18pips | +1,975.8pips | +1,789.2pips | +1,561.4pips | | 4時間足 | 10pips | 3pips | +1,640.6pips | +161.0pips | +908.0pips | | 4時間足 | 20pips | 2pips | +2,387.8pips | +1,102.3pips | +1,132.1pips | | 4時間足 | 30pips | 18pips | +1,134.3pips | +91.4pips | −16.7pips | 9通り中7通りで、2024年に選んだ目盛りの2025年年間損益は、同年の10目盛りの中央値を下回りました。**目盛りは結果を大きく動かしますが、前年の年間損益から翌年の目盛りを選ぶ根拠は得られませんでした。** 実務としての意味は1つです。**練行足のバックテスト結果は、そのまま信じるには幅が広すぎます。** 目盛りを何通りか動かして、全部の目盛りで残るかどうかを見るしかありません。1通りだけ回して出た数字は、この幅の中のどこか1点です。 ## 練行足に時間足は関係ないのか 関係します。ブロックの中に時間が入っていないのは事実です。ですが**ブロックを作るのに使うのは足の終値**なので、どの足で組むかは残ります。 ドル円2025年、同じ10pipsのブロックです。 | 時間足 | 足の本数 | 出たブロック | ブロックが出た足 | 1本の平均値幅 | ブロック÷値幅 | 逆を向く割合 | |---|---|---|---|---|---|---| | 15分足 | 24,903 | 8,041 | 25.2% | 6.07pips | 1.65 | 30.5% | | 1時間足 | 6,226 | 5,270 | 47.6% | 12.20pips | 0.82 | 22.3% | | 4時間足 | 1,610 | 3,298 | 68.9% | 24.67pips | 0.41 | 14.5% | 同じ年の同じ値動きで、ブロックの数が2.4倍違います。4時間足では3本に2本がブロックを出すので、圧縮はほとんどかかっていません。15分足では4本に3本が空振りです。 そして直前と反対方向へ進むブロックの割合は、**ブロック幅そのものより、ブロック幅をローソク足1本の平均値幅で割った比と連動しました。** この比が1.65、0.82、0.41と下がるにつれて、反対方向へ進む割合も30.5%、22.3%、14.5%へ下がりました。小さいブロックほど同じ方向へ連続しやすくなるため、チャート上ではトレンドが続いているように見えます。 つまりブロックの大きさを選ぶことは、**その時間足に対する相対的な粗さを選ぶこと**であって、時間足から自由になることではありません。 ## 3本続いたらトレンドで合っているか 値動きを1本ずつ並べ替えた系列からも、連続は同じだけ出ました。比べる相手は、同じ年の値動きを1本ずつ**ばらばらに並べ替えた系列**です。1本あたりの値幅も、大きく動いた日の分布も、1年の上昇幅も本物のままで、並んでいる順番だけが壊れています。そこから同じブロックを組みます。 3つの時間足×5つのブロック幅の15通りで測りました。1つ前と逆を向くブロックの割合は、実際の値動きで8.5〜41.6%、並べ替えた値動きで8.5〜43.4%です。**15通りのうち差が最も大きい組でも2.4ポイント**でした。同じ向きが続く平均ブロック数の差は−0.19から+0.16、ブロックの総数の比は0.96から1.014です。 | 時間足 | ブロック | 値動き | 逆を向く割合 | 同じ向きが続く平均ブロック数 | |---|---:|---|---:|---:| | 15分足 | 10pips | 実際 | 30.5% | 3.28 | | 15分足 | 10pips | 並べ替え | 30.2% | 3.31 | | 1時間足 | 10pips | 実際 | 22.3% | 4.48 | | 1時間足 | 10pips | 並べ替え | 22.6% | 4.43 | | 4時間足 | 5pips | 実際 | 8.5% | 11.78 | | 4時間足 | 5pips | 並べ替え | 8.5% | 11.80 | | 4時間足 | 30pips | 実際 | 27.7% | 3.60 | | 4時間足 | 30pips | 並べ替え | 27.2% | 3.68 | 順番を完全に壊した系列が、同じ絵を出します。**練行足が見せている連続の長さは、値動きの連続性ではなく、ブロックの大きさと1本の値幅の比で決まっている**ということです。 理由は作りかたにあります。水準が148.00で価格が148.09にいるとき、上にブロックを積むには終値が148.10になれば足りますが、下に積むには147.90まで落ちる必要があります。**同じ向きに進むほうが、向きを変えるより近い。** 上と下で必要な値幅が違うので、順番の無い値動きからでもブロックは続きます。 ## では練行足に意味は無いのか 見た目は同じでも、成績には差が出ました。並べ替えた系列を同じ条件でバックテストして、実際の値動きと比べます。1つの組み合わせにつき並べ替えを10回行い、その平均です。 | 時間足 | 年 | ブロック | 取引回数 | 実際の年間損益 | 並べ替えの年間損益 | 差 | |---|---:|---:|---:|---:|---:|---:| | 15分足 | 2025年 | 20pips | 1,007回 | +1,381.7pips | −959.2pips | +2,340.9pips | | 15分足 | 2024年 | 10pips | 2,287回 | +1,738.7pips | −1,507.9pips | +3,246.6pips | | 1時間足 | 2025年 | 10pips | 1,176回 | +2,368.6pips | −1,110.9pips | +3,479.5pips | | 1時間足 | 2024年 | 30pips | 373回 | +679.9pips | +170.4pips | +509.5pips | | 4時間足 | 2025年 | 30pips | 228回 | −906.9pips | +155.9pips | −1,062.8pips | | 4時間足 | 2024年 | 20pips | 267回 | +2,155.8pips | +85.1pips | +2,070.7pips | 3つの時間足×2年×3つのブロック幅の18通りのうち、**17通りで実際の値動きが勝ちました。** 並べ替えた値動き自体が黒字になったのは18通り中9通りで、平均するとほぼ横ばいです。 **ブロックの並びかたは情報を持っていません。** 順番を壊しても同じだけ出ます。**一方で、成績には差が出ます。** ブロックとブロックのあいだで価格がどれだけ動くかは、順番を壊すと変わるからです。 読むべきはブロックが何個続いたかではなく、そのあいだに何pips動いたか、ということでもあります。 ## バックテストの成績が良く見える理由 ブロックの終値で約定させると、30通り全部で成績が上がります。ブロックの終値は目盛りの上の価格です。平均足の終値と違って、**その値段では確かに取引されています。** ただし、それは足の途中の話です。 ブロックが確定するのは足が引けたときで、そのとき価格はもう水準を通り過ぎています。行き過ぎ幅を測りました。 | 時間足 | ブロック | 平均の行き過ぎ | 中央値 | ブロック比 | |---|---|---|---|---| | 15分足 | 10pips | 3.76pips | 3.20pips | 37.6% | | 1時間足 | 20pips | 7.60pips | 6.55pips | 38.0% | | 4時間足 | 30pips | 11.80pips | 10.60pips | 39.3% | | 4時間足 | 50pips | 18.43pips | 15.75pips | 36.9% | 3つの時間足×5つのブロック幅の15通りすべてで、**行き過ぎは必ず有利な側**でした。ブロックは動いた方向へ、手前で止まるように積まれるので、そこで買えば実勢より安く、売れば高くなります。100%です。例外はありません。 同じ売買を両方の価格で計算しました。戦略は1文字も変えていません。 | 時間足 | 年 | ブロック | 取引回数 | 足の終値での年間損益 | ブロックの終値での年間損益 | |---|---|---|---|---|---| | 15分足 | 2025年 | 5pips | 4,841回 | −786.1pips | +19,815pips | | 15分足 | 2025年 | 10pips | 2,452回 | +59.1pips | +18,040pips | | 1時間足 | 2025年 | 20pips | 602回 | +1,491.5pips | +10,500pips | | 4時間足 | 2025年 | 30pips | 228回 | −838.6pips | +4,440pips | | 4時間足 | 2024年 | 50pips | 112回 | −799.4pips | +3,600pips | 3つの時間足×2年×5つのブロック幅の30通りを測って、**30通りすべてでブロックの終値のほうが良く、うち4通りは赤字が黒字に反転**しました。1取引あたりの上乗せは4.26pipsから39.28pipsです。 上乗せの大きさには式があります。行き過ぎは入口と出口で1回ずつ拾うので、**1取引あたりの上乗せ＝平均の行き過ぎ×2**。3つの時間足×2年×3つのブロック幅の18通りで測った比は0.908から1.024でした。「コスト＝取引回数×スプレッド」の符号違いで、ブロックを大きくするほど1取引あたりが増えます。 平均足では終値が4本値の平均という架空の価格でした。練行足の終値は実在した価格です。**それでも約定には使えません。実在した価格と、いま約定できる価格は別物です。** ## ブロック幅は何pipsがいいか 年と時間足で入れ替わるので、1つには決まりませんでした。目盛りをずらさず、反転に1ブロックの作図で、ブロックの向きの変化を6通りのブロック幅で測っています。 年間損益です。 | 時間足 | 年 | 5pips | 10pips | 15pips | 20pips | 30pips | 50pips | |---|---|---:|---:|---:|---:|---:|---:| | 15分足 | 2025年 | −2,238.4pips | −676.5pips | +1,018.4pips | +1,381.7pips | +2,233.1pips | +333.4pips | | 15分足 | 2024年 | −135.1pips | +1,738.7pips | +1,062.3pips | +668.3pips | +1,080.4pips | +800.1pips | | 1時間足 | 2025年 | −655.7pips | +2,368.6pips | +3,084.3pips | +1,310.9pips | +1,698.6pips | +57.7pips | | 1時間足 | 2024年 | +456.3pips | +498.3pips | +2,177.7pips | +1,026.0pips | +679.9pips | +130.7pips | | 4時間足 | 2025年 | +759.0pips | +1,105.3pips | +2,288.0pips | +1,356.8pips | −906.9pips | +284.6pips | | 4時間足 | 2024年 | +1,861.6pips | +925.2pips | +2,286.6pips | +2,155.8pips | +391.8pips | −833.0pips | 同じ設定の取引回数です。 | 時間足 | 年 | 5pips | 10pips | 15pips | 20pips | 30pips | 50pips | |---|---|---:|---:|---:|---:|---:|---:| | 15分足 | 2025年 | 4,841回 | 2,452回 | 1,526回 | 1,007回 | 535回 | 252回 | | 15分足 | 2024年 | 4,508回 | 2,287回 | 1,423回 | 960回 | 572回 | 251回 | | 1時間足 | 2025年 | 1,918回 | 1,176回 | 808回 | 602回 | 370回 | 196回 | | 1時間足 | 2024年 | 1,741回 | 1,117回 | 765回 | 563回 | 373回 | 187回 | | 4時間足 | 2025年 | 612回 | 481回 | 364回 | 312回 | 228回 | 122回 | | 4時間足 | 2024年 | 562回 | 442回 | 343回 | 267回 | 188回 | 112回 | 年間損益が最大だったブロック幅は、1時間足では2025年も2024年も15pipsですが、15分足では2025年が30pips、2024年が10pipsです。ブロックを小さくすると取引回数が増えます。15分足の5pipsは年4,841回で、スプレッド0.3pipsだけで1,452pipsを払い、2025年は−2,238.4pipsでした。逆に50pipsまで広げると年112〜252回まで減り、4時間足2024年は112回で−833.0pipsです。 Formiqの初期設定は10pipsです。ブロックの向きの変化を、反転に1ブロックの作図で回すと、2025年の年間損益は15分足−676.5pips、1時間足+2,368.6pips、4時間足+1,105.3pipsでした。同じ10pipsでも、時間足によって符号と金額が変わります。 反転に必要なブロック数も測りました。1ブロックは出たまま描く作図、2ブロックはMT4や教科書が使う伝統的な作図で、逆方向はブロック2つぶん戻さないと描かれません。 | 時間足 | 年 | ブロック | 反転に必要なブロック数 | 取引回数 | 勝率 | 年間損益 | |---|---|---:|---:|---:|---:|---:| | 15分足 | 2025年 | 10pips | 1 | 2,452回 | 35.7% | −676.5pips | | 15分足 | 2025年 | 10pips | 2 | 1,256回 | 37.2% | +1,520.4pips | | 15分足 | 2025年 | 50pips | 1 | 252回 | 33.7% | +333.4pips | | 15分足 | 2025年 | 50pips | 2 | 113回 | 31.9% | −1,821.8pips | | 1時間足 | 2025年 | 5pips | 1 | 1,918回 | 35.7% | −655.7pips | | 1時間足 | 2025年 | 5pips | 2 | 1,338回 | 37.6% | +1,524.4pips | | 4時間足 | 2025年 | 15pips | 1 | 364回 | 39.0% | +2,288.0pips | | 4時間足 | 2025年 | 15pips | 2 | 266回 | 36.8% | −297.4pips | 3つの時間足×2年×6つのブロック幅の36通りのうち、**伝統的な作図が勝ったのは16通り**でした。ほぼ引き分けです。ただし内訳は分かれていて、**小さいブロックでは2ブロック必要な作図が、大きいブロックでは1ブロックの作図が上回ります。** 2ブロック必要にすると取引回数は36通りで43〜84%に減るので、回数が多すぎる側では助けになり、もともと少ない側では削りすぎになります。 ## 何ブロック待ってから売買するか 4通りのうち3通りで2ブロック待つのが一番よく、5ブロック待つとその3通りとも赤字でした。同じ向きがNブロック続くのを待ってから売買する読み方です。Nが1のときはブロックの向きの変化そのものなので、待たない場合の数字は既に測ったものと一致します。20pipsのブロックを、反転に1ブロックの作図で回した年間損益です。 | 時間足 | 年 | 待たない | 2ブロック | 3ブロック | 4ブロック | 5ブロック | |---|---|---:|---:|---:|---:|---:| | 15分足 | 2025年 | +1,381.7pips | +2,070.3pips | +1,287.4pips | −1,117.5pips | −2,351.8pips | | 15分足 | 2024年 | +668.3pips | +1,668.5pips | +1,018.7pips | −414.6pips | −1,837.9pips | | 1時間足 | 2025年 | +1,310.9pips | +1,728.2pips | −12.3pips | +252.3pips | −1,918.0pips | | 1時間足 | 2024年 | +1,026.0pips | +187.9pips | +443.9pips | +557.5pips | +718.8pips | 同じ設定の取引回数です。待つほど減ります。 | 時間足 | 年 | 待たない | 2ブロック | 3ブロック | 4ブロック | 5ブロック | |---|---|---:|---:|---:|---:|---:| | 15分足 | 2025年 | 1,007回 | 437回 | 244回 | 162回 | 104回 | | 15分足 | 2024年 | 960回 | 453回 | 255回 | 165回 | 114回 | | 1時間足 | 2025年 | 602回 | 310回 | 201回 | 123回 | 89回 | | 1時間足 | 2024年 | 563回 | 311回 | 195回 | 129回 | 88回 | 4通りのうち3通りで、2ブロック待つのが一番でした。その3通りは5ブロック待つと−2,351.8pips・−1,837.9pips・−1,918.0pipsの赤字です。ただし2024年の1時間足だけは逆で、5ブロックまで待つほど年間損益が伸びました。 全体でも待つほうが悪くなっています。2025年の年間損益の中央値は、ブロックの向きの変化が15分足+1,166.0pips・1時間足+1,611.5pips・4時間足+786.1pipsなのに対し、Nブロック待つ読み方は15分足−106.6pips・1時間足+290.6pips・4時間足−289.3pipsでした。 なお1本の足で複数のブロックが出るので、連続数は「ちょうどNブロック」ではなく「Nブロックに到達した足」で数えています。3個まとめて出た足は、2ブロック待つ条件も飛び越えて成立させます。 ## 去年うまくいった設定は今年も使えるか 2024年に成績のよかった3設定は取引17〜30回で、翌年は3つとも赤字でした。各時間足60通りのうち、片方の年でいちばん良かった設定を相手の年で見ます。 | 時間足 | 選んだ年 | 待つブロック数 | ブロック | 反転に必要なブロック数 | 選んだ年の年間損益 | 選んだ年の取引回数 | 当てた年 | 当てた年の年間損益 | |---|---:|---:|---:|---:|---:|---:|---:|---:| | 15分足 | 2025年 | 3 | 10pips | 2 | +2,621.3pips | 753回 | 2024年 | +1,427.6pips | | 15分足 | 2024年 | 5 | 50pips | 1 | +3,255.7pips | 17回 | 2025年 | −805.8pips | | 1時間足 | 2025年 | 3 | 5pips | 1 | +3,306.8pips | 1,000回 | 2024年 | +839.0pips | | 1時間足 | 2024年 | 4 | 50pips | 2 | +2,866.8pips | 27回 | 2025年 | −499.0pips | | 4時間足 | 2025年 | 待たない | 15pips | 1 | +2,288.0pips | 364回 | 2024年 | +2,286.6pips | | 4時間足 | 2024年 | 5 | 30pips | 1 | +2,484.9pips | 30回 | 2025年 | −790.1pips | 2024年の年間損益が最大だった3設定は、取引回数が17回・27回・30回でした。**3設定とも2025年は−805.8pips、−499.0pips、−790.1pipsの赤字です。** 少ない取引回数から出た2024年の年間損益は翌年に続きませんでした。 一方、4時間足で2025年に選んだ設定は364回取引し、2024年も+2,286.6pipsでした。年間損益を比べるときは、同時に取引回数を確認する必要があります。 ## 勝率4割弱でも黒字になる理由 勝ち幅が負け幅の約1.7倍あったからです。2025年の、取引が5回以上あった設定の平均を見ます。 | 読み方 | 時間足 | 平均勝率 | 平均の勝ち幅 | 平均の負け幅 | |---|---|---:|---:|---:| | ブロックの向きの変化 | 15分足 | 36.5% | +57.6pips | −32.0pips | | ブロックの向きの変化 | 1時間足 | 37.2% | +69.0pips | −38.9pips | | ブロックの向きの変化 | 4時間足 | 38.1% | +85.5pips | −53.3pips | | Nブロック待つ | 15分足 | 36.8% | +91.1pips | −60.7pips | | Nブロック待つ | 4時間足 | 37.4% | +107.7pips | −82.7pips | 勝率はどれも4割未満です。1時間足のブロックの向きの変化は、勝ち幅+69.0pipsが負け幅−38.9pipsの1.77倍あるので、勝率37.2%でも年間損益が黒字になります。勝率だけを見ると負けている指標に見えます。 ## フィルター・損切り利確・コスト 1時間足・20pipsのブロック・ブロックの向きの変化を土台に測りました。 | 条件 | 年 | 取引回数 | 年間損益 | |---|---:|---:|---:| | 素のまま | 2025年 | 602回 | +1,310.9pips | | 素のまま | 2024年 | 563回 | +1,026.0pips | | ADX 20以上 | 2025年 | 395回 | +834.1pips | | ADX 20以上 | 2024年 | 393回 | −29.8pips | | ADX 30以上 | 2025年 | 191回 | +806.6pips | | ADX 30以上 | 2024年 | 184回 | +176.0pips | | ロンドン＋NY（UTC 7-21） | 2025年 | 346回 | +636.3pips | | ロンドン＋NY（UTC 7-21） | 2024年 | 348回 | +1,013.0pips | | 東京（UTC 0-8） | 2025年 | 227回 | +371.1pips | | 東京（UTC 0-8） | 2024年 | 197回 | −144.5pips | | 損切り50・利確100 | 2025年 | 536回 | +1,006.7pips | | 損切り50・利確100 | 2024年 | 481回 | +1,569.5pips | | 損切り100・利確200 | 2025年 | 598回 | +1,198.0pips | | 損切り100・利確200 | 2024年 | 544回 | +1,667.1pips | | 24本で時間決済 | 2025年 | 596回 | +1,947.3pips | | 24本で時間決済 | 2024年 | 561回 | +1,117.5pips | ADXは両方の年で年間損益を下げました。2025年は+1,310.9pipsが+834.1pips（ADX 20以上）へ、2024年は+1,026.0pipsが−29.8pipsへ落ちています。時間帯で絞っても、2025年は+636.3pips・+371.1pipsとどちらも素のままを下回りました。 損切りと利確は2024年で年間損益を上げ（+1,026.0pips→+1,569.5pips）、2025年では下げました（+1,310.9pips→+1,006.7pips）。改善したのは24本での時間決済だけで、2025年は+1,947.3pipsと素のままより+636.4pipsです。ただし2024年は+1,117.5pipsで+91.5pipsにとどまります。 コストは回数どおりでした。 | スプレッド | 年間損益 | 0pipsとの差 | 取引回数×スプレッド | |---:|---:|---:|---:| | 0pips | +1,491.5pips | 0.0pips | 0.0pips | | 0.3pips | +1,310.9pips | 180.6pips | 180.6pips | | 1.0pips | +889.5pips | 602.0pips | 602.0pips | | 3.0pips | −314.5pips | 1,806.0pips | 1,806.0pips | 取引回数602回にスプレッドを掛けた値と、誤差なく一致します。年間損益がゼロになるスプレッドは、1時間足20pipsで2.48pips、同じ設定を反転に2ブロック必要な作図にすると5.87pips、15分足20pipsで1.67pips、1時間足で3ブロック続くのを待つ設定は0.24pipsでした。**3ブロック待つ設定は、スプレッドが0.25pips広がるだけで黒字が消えます。** ## 関連記事 - [平均足の終値](/ja/blog/heikin-ashi-settings)：同じく足そのものを書き換える指標です。あちらの終値は4本値の平均という架空の価格で、こちらは実在するが過ぎ去った価格。約定を分けるべき理由が2通りあります - [移動平均線クロスの期間](/ja/blog/moving-average-cross-settings)：練行足の1ブロック順張りと同じ「動いた方向についていく」形を、線で測ったものです - [サイコロジカルラインの75%](/ja/blog/psychological-line-settings)：こちらも階段状の指標で、打ち込んだ数字がそのままルールにならない例です - [GMMAの本数](/ja/blog/gmma-settings)：順張りの指標に順張りのフィルターを重ねると悪化する、という同じ結果が出ています - [フェアバリューギャップの設定](/ja/blog/fair-value-gap-settings)：比べる相手を用意したら指標の主張そのものが消えた、という同じ形の記事です ## 補足 - 対象はドル円の2024年と2025年です。他の通貨ペアや期間は含みません - **目盛りは10段階しか測っていません。** 20pipsのブロックには20通りの目盛りがありますが、そのうち10通りです。幅の実測は下限として読んでください - **並べ替えた系列は、足の中の形を捨てています。** 終値だけをつなぎ直しているので、高値と安値の情報は入っていません。終値だけを読むルールと比べる用途にしか使えません - **反転の作図は2種類だけです。** 実装によっては3ブロック戻さないと反転しないものもあります - **練行足そのものの描き方も、実装で違います。** 終値で組むか高値安値で組むかは分かれていて、この検証は終値で組んでいます - **注文の細かい滑りは入れていません。** スプレッドは固定0.3pipsで、指標発表時の広がりや約定拒否は反映していません --- ### RSI「70以上は売り」は本当？3年分の張り付きを検証した結果 URL: https://formiq.jp/ja/blog/rsi-overbought-oversold-sticking Language: ja Published: 2026-08-29 Updated: 2026-09-02 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: RSI, 買われすぎ, 売られすぎ, 張り付き, 70/30, 逆張り, バックテスト, EUR/USD Also available in: en — https://formiq.jp/blog/rsi-overbought-oversold-sticking RSIが70を超えたら買われすぎ、30を下回ったら売られすぎと呼ばれます。この言葉からは、価格が行き過ぎてすぐ反転するように聞こえます。しかし、強い相場ではRSIが水準内に何本もとどまる「張り付き」が起こります。 期間14の1時間足を2023〜2025年で調べると、70以上へ入った次の足も70以上だった割合は、**ドル円60.3%、ユーロドル67.3%**でした。30以下でも、次の足まで続いた割合はドル円60.3%、ユーロドル55.8%です。2通貨ペアとも、入った直後に水準外へ戻るケースより、もう1本残るケースのほうが多くなりました。 長い張り付きには通貨ペア差があります。10本以上続いた割合は、ドル円では70以上19.2%、30以下6.3%でしたが、ユーロドルでは10.1%と8.1%です。ドル円だけを見て「買われすぎ側は長く続く」と一般化することはできません。 期間は7、9、14、21、30、50も比較しました。期間を長くしても張り付き確率は一定方向に変わらず、明確に減ったのは70や30へ到達する回数でした。 ## RSIの買われすぎ・売られすぎ水準とは RSIは、直近N本の平均上昇幅と平均下落幅を使い、値動きの偏りを0〜100で表します。 RSI = 100 − 100 ÷（1 + 平均上昇幅 ÷ 平均下落幅） [TradingViewのRSI解説](https://www.tradingview.com/support/solutions/43000502338-relative-strength-index-rsi/)では、70を超える値を買われすぎ、30を下回る値を売られすぎとする一般的な読み方が説明されています。ただし、RSIの計算式が示すのは上昇幅と下落幅の比率です。70や30に到達したこと自体が、次の足で価格が反転する条件ではありません。 この記事でいう「張り付き」は、RSIが正確に100や0になることではありません。**70以上または30以下に、終値ベースで何本連続してとどまったか**を数えます。期間14では、今回調べた2通貨ペア、3つの時間足、3年間を通じて、正確な100または0は一度もありませんでした。 ## RSIの設定と張り付きの読み方 中心にしたのは、一般的な期間14と70/30です。 | 項目 | 基準値 | 今回の意味 | |---|---:|---| | 期間 | 14 | 直近の値動きを平滑化する本数 | | 買われすぎ水準 | 70 | RSIが70以上の足を買われすぎ側に数える | | 売られすぎ水準 | 30 | RSIが30以下の足を売られすぎ側に数える | | 判定時刻 | 足の終値 | 足が確定してから水準内かを判断する | | 張り付きの長さ | 1本以上 | 最初に水準へ入った足を1本目とする | たとえばRSIが「68 → 72 → 76 → 71 → 66」と動けば、70以上の張り付きは3本です。最初の72を1本目とし、76、71まで数え、66で終了します。最初の足だけで70を下回れば1本、次の足も70以上なら2本以上です。 チャート上で確認する場合は、RSIの期間を14、上側の水準を70、下側を30に設定します。Formiq、MT4・MT5、TradingViewでは、RSIはローソク足の下にある別枠へ表示されます。既存の[RSI設定検証](/ja/blog/rsi-settings)では、表示方法と30/70の逆張り、50ラインを使う順張りを詳しく扱っています。 ## 検証の条件 | 項目 | 値 | |---|---| | 対象 | USD/JPY / EUR/USD | | 期間 | 2023-01-01〜2025-12-31 | | 時間足 | 15分足 / 1時間足 / 4時間足 | | ドル円でRSI期間14を計算できた足 | 74,800 / 18,701 / 4,835本 | | ユーロドルでRSI期間14を計算できた足 | 74,797 / 18,701 / 4,835本 | | RSIの期間 | 7 / 9 / **14** / 21 / 30 / 50 | | 水準 | 80/20 / 75/25 / **70/30** / 65/35 / 60/40 | | 張り付き開始 | 前の足では水準外、現在の足で初めて水準内に入る | | 張り付き終了 | RSIが水準外へ戻った最初の足 | | 継続確率 | 2本以上 / 3本以上 / 5本以上 / 10本以上続いた回数の割合 | | 逆張り | 70以上では売り、30以下では買い | | 約定 | シグナル足の次の足の始値。実際の売値と買値を使用 | | 1pip | ドル円0.01円 / ユーロドル0.0001ドル | | 保有期間 | 1 / 3 / 5 / 10本 | 張り付きは、1回の連続した区間を1件として数えました。70以上に10本とどまっても、10件ではなく1件です。年末まで水準外へ戻らなかった区間は、最終的な長さが分からないため集計から除外しています。 逆張りの比較では、水準へ入った直後に取引する方法と、水準外へ戻るまで待つ方法を分けました。どちらもシグナルが確定した足では取引せず、次の足の始値で売買します。各区間を独立した事例として集計しているため、合計pipsは複利運用した口座の利回りではありません。 ## 次の足も70以上のままか 12通りすべてで半数を超えました。期間14、70/30の設定について、2023〜2025年の張り付きを合計しています。 ### ドル円 | 時間足 | 側 | 区間数 | 2本以上 | 3本以上 | 5本以上 | 10本以上 | 最長 | |---|---|---:|---:|---:|---:|---:|---:| | 15分足 | 70以上 | 1,154 | 62.3% | 43.2% | 25.4% | 10.2% | 37本 | | 15分足 | 30以下 | 1,049 | 58.4% | 38.5% | 20.7% | 7.8% | 43本 | | 1時間足 | 70以上 | 312 | 60.3% | 45.8% | 30.8% | 19.2% | 39本 | | 1時間足 | 30以下 | 300 | 60.3% | 41.0% | 21.7% | 6.3% | 27本 | | 4時間足 | 70以上 | 103 | 70.9% | 51.5% | 28.2% | 8.7% | 32本 | | 4時間足 | 30以下 | 53 | 67.9% | 54.7% | 26.4% | 5.7% | 23本 | ### ユーロドル | 時間足 | 側 | 区間数 | 2本以上 | 3本以上 | 5本以上 | 10本以上 | 最長 | |---|---|---:|---:|---:|---:|---:|---:| | 15分足 | 70以上 | 1,071 | 56.2% | 38.2% | 20.9% | 6.8% | 36本 | | 15分足 | 30以下 | 984 | 57.2% | 38.3% | 20.2% | 6.1% | 31本 | | 1時間足 | 70以上 | 306 | 67.3% | 47.7% | 25.5% | 10.1% | 24本 | | 1時間足 | 30以下 | 283 | 55.8% | 37.1% | 22.6% | 8.1% | 25本 | | 4時間足 | 70以上 | 82 | 65.9% | 48.8% | 31.7% | 9.8% | 25本 | | 4時間足 | 30以下 | 76 | 59.2% | 46.1% | 17.1% | 2.6% | 24本 | 12通りの時間足と側の組み合わせすべてで、次の足まで続く割合は55.8〜70.9%でした。70や30へ入った時点で「行き過ぎたから次は戻る」と決めるより、**もう1本は同じ水準内にいるケースのほうが多い**という結果です。 4時間足は区間数が少なく、最も少ないのはドル円の30以下53件です。通貨ペアを比較するときも、数ポイントの差だけで優劣を決めず、区間数を一緒に見る必要があります。 ## 買われすぎは何本続くのか 長い張り付きはドル円で目立ち、ユーロドルでは差が縮まりました。1時間足の結果を年ごとに分けます。 | 通貨ペア | 年 | 側 | 区間数 | 2本以上 | 5本以上 | 10本以上 | 最長 | |---|---:|---|---:|---:|---:|---:|---:| | ドル円 | 2023 | 70以上 | 105 | 62.9% | 30.5% | 20.0% | 19本 | | ドル円 | 2023 | 30以下 | 88 | 62.5% | 28.4% | 10.2% | 18本 | | ドル円 | 2024 | 70以上 | 122 | 62.3% | 32.8% | 18.0% | 39本 | | ドル円 | 2024 | 30以下 | 92 | 58.7% | 18.5% | 5.4% | 27本 | | ドル円 | 2025 | 70以上 | 85 | 54.1% | 28.2% | 20.0% | 30本 | | ドル円 | 2025 | 30以下 | 120 | 60.0% | 19.2% | 4.2% | 14本 | | ユーロドル | 2023 | 70以上 | 98 | 65.3% | 28.6% | 16.3% | 23本 | | ユーロドル | 2023 | 30以下 | 95 | 60.0% | 25.3% | 10.5% | 16本 | | ユーロドル | 2024 | 70以上 | 87 | 60.9% | 26.4% | 10.3% | 21本 | | ユーロドル | 2024 | 30以下 | 116 | 50.9% | 19.0% | 6.0% | 23本 | | ユーロドル | 2025 | 70以上 | 121 | 73.6% | 22.3% | 5.0% | 24本 | | ユーロドル | 2025 | 30以下 | 72 | 58.3% | 25.0% | 8.3% | 25本 | ドル円では、70以上が10本以上続いた割合が毎年18.0〜20.0%でした。30以下では4.2〜10.2%です。一方、ユーロドルの3年合計は70以上10.1%、30以下8.1%で、上下の差が小さくなりました。 ユーロドルでも年ごとの振れがあります。70以上が次の足まで続いた割合は2024年60.9%から2025年73.6%へ上がりましたが、10本以上の割合は10.3%から5.0%へ下がりました。次の足まで続く割合と、長く続く割合は同じ動きをしていません。 ## 80/20にすれば避けられるか 避けられません。80/20まで厳しくしても、次の足まで水準内に残った割合は53%以上でした。期間14の1時間足で、水準を変えた結果です。2023〜2025年を合計しています。 | 通貨ペア | 水準 | 上側の区間数 | 上側が2本以上 | 下側の区間数 | 下側が2本以上 | |---|---:|---:|---:|---:|---:| | ドル円 | 80 / 20 | 76 | 57.9% | 47 | 55.3% | | ドル円 | 75 / 25 | 163 | 66.3% | 135 | 53.3% | | ドル円 | 70 / 30 | 312 | 60.3% | 300 | 60.3% | | ドル円 | 65 / 35 | 564 | 64.7% | 483 | 63.1% | | ドル円 | 60 / 40 | 800 | 68.0% | 681 | 64.8% | | ユーロドル | 80 / 20 | 52 | 53.8% | 39 | 53.8% | | ユーロドル | 75 / 25 | 156 | 62.2% | 120 | 59.2% | | ユーロドル | 70 / 30 | 306 | 67.3% | 283 | 55.8% | | ユーロドル | 65 / 35 | 521 | 60.3% | 535 | 59.8% | | ユーロドル | 60 / 40 | 746 | 63.1% | 743 | 65.1% | 80/20まで厳しくしても、次の足まで水準内に残った割合は両通貨ペアの上下すべてで53%以上です。ただし、対象はドル円76件・47件、ユーロドル52件・39件まで減ります。厳しい水準ほど、数件の違いで確率が大きく動きます。 水準を60/40へ広げると区間数が増えます。強いシグナルが増えたのではなく、中立に近い値まで同じ区間に含めたためです。水準の違いを比べるときは、確率だけでなく、水準内に含める範囲と区間数も確認します。 ## 期間を変えると張り付きも変わるか 変わったのは張り付きの長さではなく、発生回数でした。70/30を固定し、1時間足の期間を7、9、14、21、30、50へ変えました。買われすぎ側と売られすぎ側を合計した3年間の結果です。 | 通貨ペア | 期間 | 水準内だった足 | 区間数 | 2本以上 | 5本以上 | 10本以上 | 最長 | |---|---:|---:|---:|---:|---:|---:|---:| | ドル円 | 7 | 27.91% | 1,557 | 57.7% | 23.2% | 6.6% | 25本 | | ドル円 | 9 | 22.23% | 1,149 | 60.1% | 25.2% | 8.4% | 28本 | | ドル円 | 14 | 13.31% | 612 | 60.3% | 26.3% | 12.9% | 39本 | | ドル円 | 21 | 6.98% | 303 | 64.7% | 28.7% | 13.2% | 30本 | | ドル円 | 30 | 3.20% | 132 | 56.8% | 34.1% | 15.9% | 29本 | | ドル円 | 50 | 0.66% | 23 | 56.5% | 34.8% | 26.1% | 22本 | | ユーロドル | 7 | 26.38% | 1,552 | 57.0% | 21.2% | 6.3% | 22本 | | ユーロドル | 9 | 20.68% | 1,157 | 57.0% | 22.5% | 7.5% | 23本 | | ユーロドル | 14 | 11.48% | 589 | 61.8% | 24.1% | 9.2% | 25本 | | ユーロドル | 21 | 5.54% | 267 | 61.8% | 25.5% | 12.4% | 30本 | | ユーロドル | 30 | 2.10% | 111 | 52.3% | 22.5% | 7.2% | 29本 | | ユーロドル | 50 | 0.39% | 14 | 64.3% | 21.4% | 21.4% | 22本 | 期間を7から50へ延ばすと、連続した区間はドル円で1,557件から23件、ユーロドルで1,552件から14件へ減りました。期間が長いほどRSIは50付近に寄りやすくなり、70や30へ届きにくくなります。 一方、次の足まで続く確率はドル円56.5〜64.7%、ユーロドル52.3〜64.3%で、期間に沿って増減していません。期間50ではドル円23区間、ユーロドル14区間しかないため、21.4〜26.1%となった10本以上の割合を長期設定の特徴とは判断できません。 **期間を変えたときに2通貨ペアで共通して変わったのは、張り付きの長さではなく発生回数です。** ## 通貨ペアを替えても残るのか ユーロドルで3年連続プラスだった5条件は、ドル円で再現しませんでした。各通貨ペアで、期間6種類、売買を始めるタイミング2種類、保有期間4種類を組み合わせました。各時間足48条件、3つの時間足で144条件です。 | 対象 | 3年すべてプラス | 内訳 | |---|---:|---| | ドル円 | 1/144 | 1時間足1条件 | | ユーロドル | 5/144 | 4時間足5条件 | | ドル円・ユーロドル × 3年 | **0/144** | 共通する条件なし | 3年とも1回あたりの平均損益がプラスだった条件をすべて示します。 | 通貨ペア | 時間足 | 期間 | 売買開始 | 保有 | 2023年 | 2024年 | 2025年 | |---|---|---:|---|---:|---:|---:|---:| | ドル円 | 1時間足 | 50 | 水準外へ戻った後 | 10本 | 9回 / +12.54 | 12回 / +16.82 | 2回 / +10.75 | | ユーロドル | 4時間足 | 7 | 水準外へ戻った後 | 5本 | 141回 / +2.51 | 147回 / +0.34 | 143回 / +1.44 | | ユーロドル | 4時間足 | 9 | 水準へ入った直後 | 10本 | 108回 / +11.21 | 109回 / +6.93 | 103回 / +0.36 | | ユーロドル | 4時間足 | 9 | 水準外へ戻った後 | 3本 | 108回 / +3.81 | 109回 / +1.30 | 103回 / +1.44 | | ユーロドル | 4時間足 | 9 | 水準外へ戻った後 | 5本 | 108回 / +5.89 | 109回 / +0.46 | 103回 / +2.69 | | ユーロドル | 4時間足 | 14 | 水準へ入った直後 | 10本 | 54回 / +6.97 | 59回 / +0.95 | 44回 / +11.01 | 表は「取引回数 / 1回あたりの平均pips」です。ユーロドルで残った5条件はすべて4時間足でしたが、2024年または2025年の平均が+0.34〜+0.95pipsにとどまる条件が4つあります。わずかなコストや対象期間の違いで、プラスとマイナスが入れ替わり得る小ささです。 通貨ペアを替えると、プラスとマイナスが実際に入れ替わりました。ユーロドルの期間7・水準外へ戻った後・5本保有は、ドル円では2023年+3.55、2024年−5.69、2025年+1.02pipsです。ドル円で残った期間50の条件も、ユーロドルでは2023年−15.90、2024年+2.60、2025年−27.66pipsでした。**片方の通貨ペアで3年続いた設定を、そのままもう片方へ移せる結果ではありません。** ## すぐ売るか、戻るまで待つか どちらも3年は続きませんでした。1時間足でRSIを期間14、70/30に設定し、買われすぎでは売り、売られすぎでは買いました。表は実際の売値と買値を使った1回あたりの平均pipsです。 | 通貨ペア | 売買開始 | 保有 | 2023年 | 2024年 | 2025年 | |---|---|---:|---:|---:|---:| | ドル円 | 水準へ入った直後 | 1本 | −1.75 | +1.68 | −1.69 | | ドル円 | 水準へ入った直後 | 3本 | −2.17 | +1.01 | −1.67 | | ドル円 | 水準へ入った直後 | 5本 | −1.97 | +0.26 | −2.29 | | ドル円 | 水準へ入った直後 | 10本 | −1.48 | −2.81 | −3.87 | | ドル円 | 水準外へ戻った後 | 1本 | −3.02 | −0.51 | −0.96 | | ドル円 | 水準外へ戻った後 | 3本 | −2.27 | −0.59 | −2.61 | | ドル円 | 水準外へ戻った後 | 5本 | −0.33 | −3.70 | −3.13 | | ドル円 | 水準外へ戻った後 | 10本 | −1.89 | −6.14 | +2.00 | | ユーロドル | 水準へ入った直後 | 1本 | −1.39 | +0.47 | −1.53 | | ユーロドル | 水準へ入った直後 | 3本 | −0.71 | +0.17 | −0.88 | | ユーロドル | 水準へ入った直後 | 5本 | −0.96 | +0.73 | −0.49 | | ユーロドル | 水準へ入った直後 | 10本 | +2.48 | +3.88 | −2.66 | | ユーロドル | 水準外へ戻った後 | 1本 | −0.75 | −0.93 | −1.77 | | ユーロドル | 水準外へ戻った後 | 3本 | +0.86 | −0.81 | −3.63 | | ユーロドル | 水準外へ戻った後 | 5本 | +2.18 | +1.13 | −2.44 | | ユーロドル | 水準外へ戻った後 | 10本 | +4.44 | +3.26 | −2.90 | 5本保有の取引回数は、ドル円が2023年193回、2024年214回、2025年205回、ユーロドルが193回、203回、193回です。ユーロドルで水準外へ戻るまで待つ方法は2023年と2024年にプラスでしたが、2025年は−2.44pipsへ反転しました。 時間足別に5本保有の結果を比べても、3年すべてプラスだった時間足は、どちらの通貨ペアにもありません。RSIが水準外へ戻ったことは確認材料にはなりますが、それだけで利益の出る逆張りにはなりませんでした。 70や30は、反転の時刻を知らせる線ではありません。逆張りに使う場合は、水準へ入った足と水準外へ戻った足を分け、期間や通貨ペアを替えたときにも平均損益が残るかを確認する必要があります。今回の144条件には、ドル円とユーロドルの両方で2023〜2025年のすべてがプラスになった設定はありませんでした。 ## 関連記事 [RSIの期間・30/70・50ラインを330通り検証した記事](/ja/blog/rsi-settings)では、逆張りと順張りの年間損益を比較しています。今回の記事は、そのうち30/70へ入った1回ごとの継続時間に焦点を当てました。 [ストキャスRSIの設定検証](/ja/blog/stoch-rsi-settings)では、RSIをさらに過去の高値・安値に対して正規化した指標を扱っています。[ボリンジャーバンドの逆張りと順張り](/ja/blog/bollinger-band-settings)でも、行き過ぎをすぐ売買する方法と、抜けた方向へ乗る方法を分けて検証しています。 ## 補足 - 対象はドル円とユーロドルの2023〜2025年です。株価指数、暗号資産、他の通貨ペアでは張り付きの頻度が変わります。 - RSIは終値から計算しています。高値や安値の途中で一時的に70や30へ触れたケースは含みません。 - 前年から続いていた区間と、年末まで水準内に残った区間は、対象年だけでは長さを確定できないため除外しました。 - 80/20や期間50は区間数が少なく、確率の不確実性が大きくなります。 - 逆張りの比較は各区間を独立した事例として扱っています。複数の取引が時間的に重なる場合があるため、合計pipsは口座の運用成績ではありません。 - トレンド、値動きの大きさ、時間帯による絞り込み、損切り、利確は検証していません。 --- ### ボリンジャーバンドの期間20は本当に最適？3,024回検証の結果 URL: https://formiq.jp/ja/blog/bollinger-band-period-settings Language: ja Published: 2026-08-28 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ボリンジャーバンド, 期間, 設定, 順張り, 逆張り, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/bollinger-band-period-settings ボリンジャーバンドの「期間」は、中央の移動平均と標準偏差を何本の終値から計算するかを決めます。初期設定は20ですが、15分足の20本と4時間足の20本では、振り返る時間が5時間と80時間でまったく違います。 そこで、期間を5〜200の12段階、偏差を0.5〜2σの7段階にして、逆張り買い・逆張り売り・終値ブレイクの3つの売買ルールを測りました。結論から言うと、**全時間足に共通する一番いい期間はありません。** 2年平均が最大だったのは、15分足50本、1時間足20本、4時間足5本です。バー数は50・20・5と離れていますが、実時間に直すと12.5・20・20時間でした。**期間のバー数だけでなく、時間足を掛けた計算範囲も確認する必要があります。** ## ボリンジャーバンドの期間は何を変えるのか 計算は3段階です。 1. 直近N本の終値を平均して中央線を作る 2. 同じN本が中央線からどれだけ散らばったかを標準偏差で測る 3. 中央線の上下に「標準偏差×偏差」を足し引きする 期間を長くすると中央線の反応が遅くなり、過去の大きな値動きがバンド幅に長く残ります。1時間足・2025年・1σで実測するとこうなりました。 | 期間 | 平均片幅 | バンド内で終えた割合 | 年間取引数 | |---:|---:|---:|---:| | 5 | 12.79pips | 47.1% | 934 | | 20 | 28.51pips | 45.0% | 277 | | 50 | 46.54pips | 45.8% | 103 | | 100 | 65.62pips | 44.9% | 68 | | 200 | 88.68pips | 50.3% | 46 | 幅は約7倍になり、取引数は20分の1まで減りました。一方、1σの内側で終えた割合は44.9〜50.3%です。**期間を長くするとバンド内に収まりやすくなる、とは限りません。** 幅と同時に、平均との差を測る時間の長さも変わるからです。 ### 短い期間には届かない偏差がある 終値を含むN本で平均との差と母標準偏差を計算すると、1本の終値が取れる標準化偏差の上限は **√(N−1)** です。 期間5の上限は√4＝2σです。この記事のブレイク条件は「終値がバンドを厳密に超える」なので、期間5・2σは数学的に成立しません。実測でも、3時間足×4期間区分の12通りすべてが0回でした。 短い期間で遠い偏差を組み合わせると、悪い設定になる前に**売買できない設定**になります。期間を短くするとバンドが遠い偏差へ届かないため、期間と偏差は独立して選べません。 ## ボリンジャーバンドの期間設定と読み方 ボリンジャーバンドはローソク足と同じ場所に、中央線・上バンド・下バンドの3本が重なって表示されます。 | 環境 | 追加する場所 | |---|---| | MT4 / MT5 | 挿入 → インディケータ → トレンド → Bollinger Bands | | TradingView | インジケーター検索から Bollinger Bands | | ブラウザ（Formiq） | インジケーター一覧から「ボリンジャーバンド」 | ### 設定画面に並ぶ項目 | 項目 | 初期設定 | 意味 | |---|---:|---| | Period | 20 | 平均と標準偏差を計算する本数 | | Shift | 0 | 3本の線を左右へ何本ずらすか | | Deviations | 2.000 | 標準偏差を何倍して上下へ置くか | | Apply to | Close | 計算に使う価格。今回の検証は終値 | | Style / Color | 任意 | 線の色・太さ・種類。計算結果は変わらない | 期間を短くすると線は価格へ速く追いつき、売買が増えます。長くすると線は滑らかになり、1回の取引が長くなります。偏差は同じ期間の中でバンドを近づけたり遠ざけたりする設定です。 ### この記事で測る3つの売買ルール | 売買ルール | エントリー | 決済 | |---|---|---| | 逆張り買い | 安値が下バンドへ触れた足の終値で買う | 反対側の1σバンドへ触れた足の終値 | | 逆張り売り | 高値が上バンドへ触れた足の終値で売る | 反対側の1σバンドへ触れた足の終値 | | 順張り | 終値が上バンドを抜けたら買い、下バンドを抜けたら売り | 終値が反対側のバンドを抜けたらドテン | 偏差記事で0.5〜1.25σが年をまたいで残り、2σも多くの期間で残りました。今回は遠いバンドの少数取引に期間の答えを決めさせないため、0.5〜2σだけを同じ重みで使っています。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間区分 | 2025年、比較用の2024年、2025年前半・後半 | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 2025年の本数 | 15分足24,903本、1時間足6,226本、4時間足1,610本 | | 期間 | 5・10・14・20・25・30・40・50・75・100・150・200 | | 偏差 | 0.5・0.75・1・1.25・1.5・1.75・2σ | | 売買ルール | 逆張り買い / 逆張り売り / 終値ブレイク | | 組み合わせ | 12期間×7偏差×3売買ルール＝252設定。3時間足×4期間区分で3,024回 | | コスト | スプレッド0.3pips、スリッページ0、0.1ロット | | 約定 | シグナル足の終値。決済後の反対方向へのドテンは同じ足、同方向への再エントリーは次の足以降 | | 損切り・利確 | 主検証では使わない。別セクションで追加測定 | | 測定 | Formiqのバックテスト機能で実行し、全約定価格からpipsを再集計 | 期間だけの偶然を減らすため、各期間は偏差7種の平均と、7種のうち何種が両方の年で黒字だったかで比べます。 ## 期間は何が一番いいか 時間足ごとに入れ替わり、1つには決まりませんでした。順張りについて、偏差7種の2年平均が最大だった期間です。 | 時間足 | 首位の期間 | 実時間 | 2025年平均 | 2024年平均 | 両年黒字 | 2025年平均取引数 | |---|---:|---:|---:|---:|---:|---:| | 15分足 | 50 | 12.5時間 | +1,368.1 | +1,656.0 | 6 / 7 | 463.0 | | 1時間足 | 20 | 20時間 | +1,067.1 | +2,058.3 | 7 / 7 | 245.6 | | 4時間足 | 5 | 20時間 | +1,260.0 | +1,470.4 | 6 / 6 | 173.1 | 4時間足の期間5は2σが成立しないので、分母は売買できた6偏差です。3つの首位をバー数だけで見ると50・20・5で共通点がありません。実時間では12.5〜20時間に集まります。 全84設定のうち両方の年で黒字だった数は、15分足59、1時間足53、4時間足26でした。**期間20が強かったのは1時間足であり、期間20そのものが普遍的だったわけではありません。** ## 1時間足なら期間20でいいか 1時間足では期間20が2年平均で首位、7偏差すべてが両年黒字でした。順張りを、期間ごとに偏差7種で平均しています。 | 期間 | 2025年平均 | 2024年平均 | 2年平均 | 2025年平均取引数 | 両年黒字 | |---:|---:|---:|---:|---:|---:| | 5 | +1,174.4 | +1,211.6 | +1,193.0 | 692.7 | 6 / 6 | | 10 | +1,560.6 | +665.4 | +1,113.0 | 437.3 | 6 / 7 | | 14 | +1,405.8 | +1,165.9 | +1,285.8 | 329.9 | 6 / 7 | | **20** | **+1,067.1** | **+2,058.3** | **+1,562.7** | **245.6** | **7 / 7** | | 25 | +799.2 | +885.8 | +842.5 | 203.4 | 6 / 7 | | 30 | +1,427.1 | +562.1 | +994.6 | 169.0 | 6 / 7 | | 40 | +1,161.1 | +889.6 | +1,025.3 | 127.6 | 7 / 7 | | 50 | +1,040.7 | +1,103.8 | +1,072.3 | 104.1 | 6 / 7 | | 75 | −494.3 | +1,905.9 | +705.8 | 78.6 | 2 / 7 | | 100 | −1,092.3 | +1,957.2 | +432.5 | 63.3 | 1 / 7 | | 150 | −1,910.7 | +2,109.6 | +99.4 | 47.7 | 0 / 7 | | 200 | −2,022.6 | +1,663.4 | −179.6 | 38.3 | 0 / 7 | 20期間は2年平均が最高で、7偏差すべてが両方の年で黒字です。40期間も7 / 7ですが、2年平均は+1,025.3pipsでした。**1時間足で一つだけ出発点を選ぶなら20**という答えになります。 ただし75以上は、2024年に大きく勝ち、2025年に負けています。取引数も年38〜79回まで減るので、長期化でノイズが消えたのではなく、特定の長い値動きを少数回拾った形です。 ### 初期設定の20・2σの位置 | 年 | 取引数 | 勝率 | PF | 損益 | |---|---:|---:|---:|---:| | 2025 | 132 | 43.18% | 1.235 | +1,177.5pips | | 2024 | 137 | 41.61% | 1.270 | +1,434.6pips | 初期設定は首位ではありませんが、両年とも黒字で中央値より上です。期間20が妥当でも、偏差2σまで自動的に一番良いとは限りません。偏差を1σにすると、2025年277回・+1,052.5pips、2024年280回・+2,071.7pipsでした。 ## 去年よかった期間を選べばいいか 1年だけで見た最大値は、別の年で再現しませんでした。 - 期間10：2025年平均+1,560.6pips、2024年平均+665.4pips - 期間150：2024年平均+2,109.6pips、2025年平均−1,910.7pips 期間と偏差を同時に選ぶと差はさらに広がります。 | 選択年 | 設定 | 選択年の取引数 | 選択年の損益 | 確認年 | 確認年の損益 | |---:|---|---:|---:|---:|---:| | 2025 | 10期間・1.75σ | 271回 | +2,753.0pips | 2024 | +686.0pips | | 2024 | 150期間・0.5σ | 53回 | +3,164.9pips | 2025 | −1,772.1pips | 1年だけの最高値をそのまま採用するより、20期間のように近い偏差がまとめて残った場所を出発点にするほうが、選び方として無理がありません。 ## 期間20は半年でも安定するか 年間では7偏差すべてが黒字でも、前半と後半の両方で黒字だったのは2偏差でした。2025年を前半と後半に分け、偏差7種の平均を比べます。 | 期間 | 2025年前半 | 2025年後半 | 両方の半年で黒字 | |---:|---:|---:|---:| | 10 | +745.8 | +788.9 | 7 / 7 | | 14 | +779.2 | +603.3 | 5 / 7 | | **20** | **−211.0** | **+1,258.5** | **2 / 7** | | 40 | +227.8 | +916.3 | 6 / 7 | | 50 | +596.3 | +448.2 | 6 / 7 | 年間では20期間が7 / 7でも、前半と後半の両方で黒字だったのは2偏差です。期間10は7 / 7、40と50は6 / 7でした。 **「2年平均が一番」と「半年ごとに安定」は別の基準です。** 年間損益を重視するなら20、短い区間で崩れにくいことを重視するなら10・40・50も並べて確認する必要があります。 ## 勝率だけでは期間を選べない 1時間足・2025年の売買があった83設定は、平均214回（19〜1,258回）取引しました。平均勝率は36.3%、平均利益+112.8pips、平均損失−67.9pipsです。 勝率は設定ごとの損益と同じ方向へ動きましたが、期間10は2025年の平均+1,560.6pipsから2024年の+665.4pipsへ低下し、期間150は+2,109.6pipsから−1,910.7pipsへ反転しました。**同じ年の勝率だけでは、別の年に使う期間は選べません。** 逆張りは反対でした。1時間足で両年黒字だった設定は、買い11 / 84、売り0 / 84です。長い期間の逆張りは2025年だけ良く見えるものが増えましたが、2024年へ移りませんでした。期間の答えは順張りと逆張りで共有できません。 ## フィルターやコストで変わるか 期間の答えは変わりませんでした。1時間足・20期間・1σの順張りを基準に、入口と出口を変えました。各セルは「取引数・年間損益」です。 | 条件 | 年 | 取引数 | 損益 | |---|---:|---:|---:| | 基準 | 2025 | 277回 | +1,052.5pips | | 基準 | 2024 | 280回 | +2,071.7pips | | ADX 20以上 | 2025 | 188回 | −1,033.2pips | | ADX 20以上 | 2024 | 185回 | −3.9pips | | 東京時間だけ | 2025 | 164回 | +519.4pips | | 東京時間だけ | 2024 | 162回 | +2,139.6pips | | ロンドン・NY時間 | 2025 | 207回 | +1,073.5pips | | ロンドン・NY時間 | 2024 | 218回 | +2,154.3pips | | 損切り30・利確60 | 2025 | 398回 | +1,687.6pips | | 損切り30・利確60 | 2024 | 419回 | +1,467.1pips | | 損切り50・利確100 | 2025 | 323回 | +1,157.3pips | | 損切り50・利確100 | 2024 | 345回 | +1,896.2pips | | 損切り100・利確200 | 2025 | 297回 | +1,340.5pips | | 損切り100・利確200 | 2024 | 292回 | +1,987.6pips | | 24本で時間決済 | 2025 | 334回 | +989.1pips | | 24本で時間決済 | 2024 | 337回 | +1,559.0pips | 固定の損切り・利確はエントリー約定価格から指定pips離れた約定水準として扱い、到達時はその水準どおりに決済します。 ADXは20・25・30の3水準すべてで基準を下回りました。ロンドン・NY時間だけは両年でわずかに上回りましたが、期間20を別の期間へ置き換えるほどの差ではありません。固定の損切り・利確も、両方の年で基準を上回る組み合わせはありませんでした。 スプレッド感応度は次の通りです。277回の売買シグナルは変わりません。 | スプレッド | 損益 | |---:|---:| | 0pips | +1,135.6pips | | 0.3pips | +1,052.5pips | | 1pips | +858.6pips | | 2pips | +581.6pips | | 3pips | +304.6pips | 損益分岐は4.1pipsです。期間を短くすると売買が増えるため、同じ偏差でもコスト耐性は下がりやすくなります。 ## 関連記事 [ボリンジャーバンドは何シグマがいいか](/ja/blog/bollinger-band-sigma-settings)では、期間を10〜50に置いて偏差0.5〜4σを比較しています。[逆張りと順張りの315通り検証](/ja/blog/bollinger-band-settings)は、入口と出口の組み合わせを広げた検証です。 期間が2本の移動平均線へどう効くかは[移動平均線クロスの設定](/ja/blog/moving-average-cross-settings)、同じ期間でSMAとEMAの差を測った結果は[SMAとEMAはどちらがいいか](/ja/blog/sma-vs-ema)にまとめています。条件を自分の通貨ペアで再現する手順は[プログラミングなしのバックテスト](/ja/blog/backtest-without-coding)で確認できます。 ## 補足 - 米ドル／円の2024年・2025年だけです - いちばん良い期間は時間足で変わりました。ほかの通貨ペアや日足へそのまま移せません - 12期間×7偏差から選ぶため、1年だけの最高値には複数比較による上振れが含まれます。確認年も完全に未使用の将来データではありません - 期間5・2σのように、計算上成立しない組み合わせがあります。別の標準偏差定義や、終値を計算へ含めない実装では境界が変わります - バンドタッチは高値・安値、ブレイクは終値で判定しています。同じ足が上下へ触れた場合、決済後の同方向への再エントリーは行わず、反対方向へのドテンだけを同じ終値で許可します - 約定は足の終値、スプレッドは固定です。変動スプレッド、指値の滑り、足の中の値動き順序は再現していません - 期間区分の末尾では保有中の取引を終値で決済します。年間と半年の合計は完全には一致しません --- ### BBWのスクイーズは本当に使える？4,032通り検証した結果 URL: https://formiq.jp/ja/blog/bollinger-bandwidth-settings Language: ja Published: 2026-08-28 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: BBW, ボリンジャーバンド幅, Bollinger BandWidth, スクイーズ, ボラティリティ, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/bollinger-bandwidth-settings BBW（Bollinger BandWidth、ボリンジャーバンド幅）は、ボリンジャーバンドの上限と下限がどれだけ離れているかを1本の線にした指標です。線が低ければ値動きが縮み、高ければ広がっています。価格の上昇・下落ではなく、**値動きの大きさだけ**を測ります。 よく使われる読み方は「過去125本で最も狭くなった後のブレイク」です。ところがドル円で4,032通りを測ると、**BBWの絶対値は時間足が変わるだけで約4.7倍ずれ、125本安値も他の比較本数を上回りませんでした。** 1時間足では期間20が最も残りましたが、初期設定に近い期間20・偏差2・125本安値・75本待ちは2025年に58回、−70.6pipsです。 もう一つ、設定画面で別々に見える偏差と固定水準は、BBWの線だけを読む場合、同じ比率で変えると同じ判定になります。**偏差を2倍にすると線も2倍です。偏差と水準を一緒に2倍にした192組は、取引まで完全に一致しました。** ## BBW（ボリンジャーバンド幅）とは 計算はボリンジャーバンドから3段階です。 1. 中心線：終値の単純移動平均 2. 上限・下限：中心線 ± 偏差 × 標準偏差 3. **BBW =（上限 − 下限）÷ 中心線 × 100** 偏差を `k` とすると、上限と下限の差は `2 × k × 標準偏差` です。したがってBBWは、価格に対して最近の標準偏差が何%あるかを表しています。期間20・偏差2なら、線は `4 × 標準偏差 ÷ 20本平均 × 100` です。 この割り算には意味があります。ドル円が100円のときと150円のときで、同じ1円幅をそのまま比べず、中心価格に対する割合へ直します。ただし時間足が変わると20本が覆う実時間も変わるため、値は揃いません。 ### 線は値動きの向きを持たない BBWが上がっても、価格が上がったとは限りません。急落でも標準偏差は大きくなり、線は上がります。実測した125本安値から20本後の平均値幅は次のとおりでした。 | 時間足 | 125本安値の回数 | 20本後の平均変化 | 絶対値 | 全足の絶対値 | |---|---:|---:|---:|---:| | 15分足 | 615 | −5.03pips | 27.40 | 29.39 | | 1時間足 | 129 | +0.45pips | 76.90 | 61.73 | | 4時間足 | 38 | −15.35pips | 110.89 | 119.69 | 1時間足だけはスクイーズ後の値幅が全足平均を24.6%上回りました。一方、15分足と4時間足は下回っています。平均変化も−5.03、+0.45、−15.35pipsと揃いません。**スクイーズは動く大きさの候補であって、買いか売りかを決める線ではありません。** この記事の売買ルールは、上側バンドを終値で抜けたら買い、下側を抜けたら売りとして、足りない向きを価格から補います。 ## BBWの設定と読み方 BBWはローソク足に重ならず、**チャート下の別枠に1本の線**として表示されます。0を下限にして動きますが、RSIの30・70のような全銘柄共通の水準線はありません。 | 環境 | 追加する場所 | |---|---| | MT4 / MT5 | BBWは標準指標にないため、カスタム指標を追加する | | TradingView | インジケーター検索から Bollinger BandWidth | | ブラウザ（Formiq） | インジケーター一覧から「BBバンド幅」 | MT4・MT5に標準で入っているのは3本を価格上へ描くBollinger Bandsです。BBWを追加するカスタム指標は配布元によって項目や水準線が違います。**MT4の標準状態にBBWの水準線はありません。** ### チャート側の設定 [TradingViewのBollinger BandWidth公式解説](https://www.tradingview.com/support/solutions/43000501972-bollinger-bandwidth-bbw/)では、次の値が初期設定です。Formiqでは期間と偏差を変更でき、適用価格は終値、移動平均はSMAに固定しています。 | 項目 | 初期設定 | 意味 | |---|---:|---| | Length（期間） | 20 | 中心線と標準偏差を何本で計算するか | | Source（適用価格） | Close | 高値・安値ではなく終値を使う | | StdDev（偏差） | 2 | 標準偏差を何倍して上下バンドを置くか | | Highest Expansion Length | 125 | 直近の最大BBWを比べる本数 | | Lowest Contraction Length | 125 | 直近の最小BBWを比べる本数 | 最後の2つは線の計算を変えず、過去125本の最大・最小を案内します。[John Bollinger本人のBandWidth資料](https://www.bollingerbands.com/_files/ugd/58be43_d09c50b6e8ea4afd9af0523ef94de876.pdf)でも、代表的なスクイーズを125期間の安値としています。 ### バックテスト条件の設定 今回の追加で、BBWを売買条件として直接選べるようにしました。 | 項目 | 初期値 | 意味 | |---|---:|---| | トリガー | スクイーズ後のバンド抜け | 125本安値の後に外側バンドを終値で抜ける形 | | 期間 | 20 | BBWと外側バンドの計算本数 | | 偏差 | 2 | バンドを置く標準偏差の倍数 | | 安値を比べる本数 | 125 | 何本の中で最小ならスクイーズとするか | | ブレイクを待つ本数 | 75 | スクイーズから何本後までのバンド抜けを採るか | | 幅の水準 | 0.5 | 「幅の水準上抜け」を選んだときだけ使う固定値 | 画面に出るトリガーは2つです。 - **スクイーズ後のバンド抜け**：BBWが比較範囲の最小を更新した後、待機本数以内に上限を終値で抜けたら買い、下限を抜けたら売り - **幅の水準上抜け**：BBWが固定水準を上抜けた足で、終値が中心線より上なら買い、下なら売り 後者は固定水準が時間足をまたげるかを見る対照群です。通常はスクイーズ側を使い、期間・比較本数・待機本数を先に調整します。偏差は線の高さだけでなく、売買方向を確定する外側バンドの位置も変えるため、スクイーズ側では結果に届きます。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | スクイーズ | 期間4種 × 偏差4種 × 安値比較4種 × 待機4種 = 256通り | | 固定水準の対照群 | 期間4種 × 偏差4種 × 水準5種 = 80通り | | 期間 | 10 / **20** / 30 / 50 | | 偏差 | 1 / 1.5 / **2** / 2.5 | | 安値比較 | 20 / 50 / **125** / 250本 | | 待機 | 10 / 20 / 50 / **75本** | | 固定水準 | 0.1 / 0.2 / 0.4 / 0.8 / 1.6 | | 組み合わせ | 336通り × 時間足3種 × 期間4区分 = 4,032回 | | 決済 | 反対方向の同じシグナルまで。エントリーと決済に同じルールセットを使う | | 損切り・利確 | 使わない（別セクションで測定） | | スプレッド | 0.3pips固定。約定は終値 | | ロット | 0.1ロット | 「125本で最小になっただけ」では売買しません。その後に価格が上側か下側のバンドを終値で抜けて初めて入ります。スクイーズそのものに売買方向がないためです。 ## BBWの水準はいくつが目安か 共通の目安はありません。中央値は時間足のあいだで4.7倍違いました。期間20・偏差2のBBWを、2025年の全足で数えています。 | 時間足 | 下位10% | 中央値 | 上位10% | 最小〜最大 | |---|---:|---:|---:|---:| | 15分足 | 0.1392 | **0.2943** | 0.6611 | 0.0384〜3.5979 | | 1時間足 | 0.3303 | **0.6439** | 1.3723 | 0.1082〜3.7455 | | 4時間足 | 0.7642 | **1.3693** | 2.5363 | 0.3544〜4.6219 | 中央値は0.2943から1.3693へ**4.7倍**です。固定水準0.4は、15分足では中央値より上、1時間足では下位10%の少し上、4時間足では下位10%より下です。同じ水準が「広い」「普通」「狭い」を全部指してしまいます。 固定水準の対照群にも出ました。両年黒字だった数は、15分足36/80、1時間足22/80、4時間足14/80。2025年の中央値は15分足+201.6pipsですが、1時間足と4時間足は0pipsでした。水準が高すぎて1回も売買しない設定が混ざるためです。 **BBWは0.4以下ならスクイーズ、のような共通値では読めません。** 同じ銘柄・時間足の過去値と比べて、相対的に狭いかを判断する必要があります。 ## スクイーズは何本で判定するか 公表されている125本は、両年黒字16/64で最少でした。スクイーズ256通りの成績です。 | 時間足 | 年 | 黒字設定 | 損益中央値 | 両年とも黒字 | |---|---:|---:|---:|---:| | 15分足 | 2025 | 159/256 | +324.1pips | **115/256** | | 15分足 | 2024 | 177/256 | +661.1pips | **115/256** | | 1時間足 | 2025 | 166/256 | +490.5pips | **106/256** | | 1時間足 | 2024 | 162/256 | +458.4pips | **106/256** | | 4時間足 | 2025 | 111/256 | −283.9pips | **65/256** | | 4時間足 | 2024 | 168/256 | +566.7pips | **65/256** | 15分足と1時間足は過半数が各年で黒字ですが、同じ設定が両年黒字だった数は45%と41%です。4時間足は2024年の中央値+566.7pipsから2025年−283.9pipsへ反転しました。 この表は「スクイーズなら勝てる」ではなく、**時間足と年をまたぐと候補が半分以下になる**ことを示します。4時間足は平均22回しか取引せず、上位の数字も少数取引で膨らんでいました。 ### 125本だけは特別ではなかった 1時間足で、安値を比べる本数ごとの周辺平均です。各行は期間4種 × 偏差4種 × 待機4種の64通りです。 | 安値比較 | 2025年 平均 | 2024年 平均 | 両年黒字 | 2025年 平均取引数 | |---:|---:|---:|---:|---:| | 20 | **+916.0** | +665.7 | **37/64** | 143.2 | | 50 | +144.6 | **+753.7** | 31/64 | 100.0 | | 125 | −13.4 | +322.4 | **16/64** | 51.5 | | 250 | +399.3 | −125.4 | 22/64 | 28.4 | 公表されている125本は、両年黒字16/64で最少です。長く比べるほど珍しいスクイーズになりますが、珍しさが利益には直結しません。250本は平均取引数28.4回まで減り、2024年平均が赤字でした。 ## おすすめの期間はどれか 1時間足では期間20だけが両年平均で+800pipsを超え、42/64が両年黒字でした。同じ1時間足で、期間だけをまとめます。各行64通りです。 | 期間 | 2025年 平均 | 2024年 平均 | 両年黒字 | 2025年 平均取引数 | |---:|---:|---:|---:|---:| | 10 | −7.3 | +392.2 | 22/64 | 130.8 | | **20** | **+1,101.1** | **+815.3** | **42/64** | 80.6 | | 30 | +612.8 | −345.8 | 19/64 | 65.3 | | 50 | −260.1 | +754.8 | 23/64 | 46.4 | 期間20だけが両年平均で+800pipsを超え、42/64が両年黒字です。期間30は2025年に+612.8でも2024年に−345.8、期間50は反対でした。 待機本数では50本が33/64で最多でした。10本は2024年平均−155.2pips、20本は+27.5、50本は+991.8、75本は+752.4です。**早すぎるブレイクだけでも、いつまでも古いスクイーズを有効にしても残りません。** ## 偏差を変えると何が変わるか 線の形は変わらず、価格が抜ける外側バンドの位置だけが動きます。偏差1と2のBBWを全点で突き合わせると、偏差2の線は偏差1の線のほぼ正確な2倍でした。最大誤差は3時間足とも4.1e-14以下です。極大値・極小値になる足と、125本安値になる足は変わりません。 固定水準の読み方では、これは完全な重複になります。 - 偏差1・水準0.1 と 偏差2・水準0.2 - 偏差1・水準0.2 と 偏差2・水準0.4 - 偏差1・水準0.4 と 偏差2・水準0.8 - 偏差1・水準0.8 と 偏差2・水準1.6 期間4種 × 時間足3種 × 期間4区分で192組を比べ、取引数・勝率・損益の不一致は**0組**でした。BBWの固定水準だけを見るなら、偏差を2倍にして水準も2倍にする操作は何も変えません。 ただしスクイーズ後のバンド抜けでは、偏差が外側バンドの位置を変えます。1時間足の周辺平均は偏差1が2025年+735.3・2024年+763.1pipsで両年黒字38/64、偏差2は+225.3・+116.3で22/64です。BBWが125本安値になる足は同じでも、価格が抜ける外側バンドの位置が変わるため、売買結果は一致しません。 ## 初期設定のままだとどうなるか 両年黒字になった時間足はありませんでした。期間20・偏差2・125本安値・75本待ちです。チャートの初期設定と、今回追加した条件の初期値を合わせた形です。 | 時間足 | 年 | 取引数 | 勝率 | 損益 | |---|---:|---:|---:|---:| | 15分足 | 2025 | 314回 | 35.35% | +88.9pips | | 15分足 | 2024 | 298回 | 30.54% | −293.5pips | | 1時間足 | 2025 | 58回 | 41.38% | −70.6pips | | 1時間足 | 2024 | 71回 | 40.85% | +909.0pips | | 4時間足 | 2025 | 26回 | 15.38% | −1,307.7pips | | 4時間足 | 2024 | 20回 | 40.00% | +570.4pips | 両年黒字の時間足はありません。1時間足は125本安値から50本以内に限定すると、2025年34回・+3,043.1pips、2024年55回・+1,459.6pipsでしたが、75本まで待つと2025年に−70.6pipsへ落ちました。**125本安値という入口だけでなく、その情報をいつまで有効とするかが結果を変えます。** ## 去年の数字で設定を決めていいか 1時間足の2設定は黒字のまま縮み、4時間足の2設定は赤字へ反転しました。 | 時間足 | 選択年 | 期間 | 偏差 | 安値比較 | 待機 | 選択年の取引数 | 選択年の損益 | 確認年の損益 | |---|---:|---:|---:|---:|---:|---:|---:|---:| | 1時間足 | 2025 | 20 | 1 | 20本 | 10本 | 167回 | +3,868.5pips | 2024年 +1,589.6pips | | 1時間足 | 2024 | 50 | 1 | 125本 | 75本 | 63回 | +3,025.0pips | 2025年 +450.2pips | | 4時間足 | 2025 | 対象外 | 対象外 | 対象外 | 対象外 | 3回 | +2,392.6pips | 2024年 −889.8pips | | 4時間足 | 2024 | 対象外 | 対象外 | 対象外 | 対象外 | 10回 | +4,025.5pips | 2025年 −710.7pips | 1時間足の2設定は確認年も黒字でしたが、損益は小さくなりました。4時間足の2設定は確認年に赤字へ反転しています。また、4時間足は取引が3回と10回しかなく、設定の再現性を判断できる件数ではありません。 ## 勝率は判断材料になるか 今回は損益と同じ向きでしたが、それだけでは足りませんでした。初期設定に近い1時間足は58回、勝率41.38%、平均利益+134.46pips、平均損失−96.99pipsでした。勝ちは負けの1.39倍ありますが、34敗に対して24勝しかなく、合計は−70.6pipsです。**平均利益が大きいだけでも、勝率が高いだけでも足りません。回数と両方の幅を一緒に見る必要があります。** ## フィルター・損切り利確・コスト 損切り・利確・スプレッドの比較では、初期設定に近い1時間足、期間20・偏差2・125本安値・75本待ちに固定します。 ### ADXと時間帯 | 条件 | 年 | 取引数 | 損益 | |---|---:|---:|---:| | 追加なし | 2025 | 58回 | −70.6pips | | 追加なし | 2024 | 71回 | +909.0pips | | ADX 20以上 | 2025 | 43回 | +289.7pips | | ADX 20以上 | 2024 | 56回 | +583.1pips | | ADX 25以上 | 2025 | 32回 | +284.4pips | | ADX 25以上 | 2024 | 44回 | +17.6pips | | ADX 30以上 | 2025 | 21回 | +219.5pips | | ADX 30以上 | 2024 | 28回 | +793.8pips | | 0〜8時UTCだけ | 2025 | 32回 | **+821.5pips** | | 0〜8時UTCだけ | 2024 | 39回 | +1,020.8pips | | 7〜21時UTCだけ | 2025 | 50回 | +6.0pips | | 7〜21時UTCだけ | 2024 | 57回 | **+1,518.3pips** | ADXは2025年を黒字にしましたが、3水準とも2024年の+909.0pipsを下回ります。時間帯は両方とも各年を改善しましたが、0〜8時UTCは取引が32回と39回です。後から選んだ区分なので、別期間での確認が必要です。 ### 損切り・利確 | 決済 | 年 | 取引数 | 損益 | |---|---:|---:|---:| | 反対シグナル | 2025 | 58回 | −70.6pips | | 反対シグナル | 2024 | 71回 | +909.0pips | | 損切り30・利確60 | 2025 | 131回 | +570.0pips | | 損切り30・利確60 | 2024 | 134回 | +694.0pips | | 損切り50・利確100 | 2025 | 104回 | +489.7pips | | 損切り50・利確100 | 2024 | 115回 | +1,055.2pips | | 損切り100・利確200 | 2025 | 74回 | **+1,556.3pips** | | 損切り100・利確200 | 2024 | 85回 | **+1,181.4pips** | | 24本で時間決済 | 2025 | 103回 | +826.0pips | | 24本で時間決済 | 2024 | 110回 | +976.2pips | 損切り100・利確200と24本決済は両方の年で反対シグナル決済を上回りました。スクイーズの反対方向が出るまで平均149本持っていた2025年に、出口を短くした効果です。ただしこの4候補から良いものを選んだ結果で、未検証の年へそのまま持ち出せる保証はありません。 ### スプレッド この設定はスプレッド0でも**−53.2pips（58回）**でした。0.3pipsでは−70.6pipsで、差の17.4pipsは `58回 × 0.3pips` と一致します。1.0pipsでは−111.2、3.0pipsでは−227.2pipsです。 損益分岐スプレッドを計算すると−0.92pipsになります。負のスプレッドは受け取れないので、実務上の損益分岐は**存在しません**。この行はコストで沈んだのではなく、売買そのものがスプレッド0で負けています。 ## 関連記事 [ボリンジャーバンドの逆張りと順張り](/ja/blog/bollinger-band-settings)は、価格がバンドに触れたら反対方向へ売買するルールと、バンドを抜けた方向へ売買するルールを比較しています。[期間を5〜200本で比べた検証](/ja/blog/bollinger-band-period-settings)と[偏差を0.5〜4σで比べた検証](/ja/blog/bollinger-band-sigma-settings)では、BBWの分子を決める期間と偏差を個別に測りました。 [スクイーズモメンタム](/ja/blog/squeeze-momentum-settings)は、ボリンジャーバンドがケルトナーチャネルの内側に入る別のスクイーズ定義です。BBWの125本安値とは同じ足を選びません。[プログラミングなしで条件を組んで検証する方法](/ja/blog/backtest-without-coding)では、今回追加したBBW条件を自分の通貨ペア・時間足で組み直せます。 ## 補足 - 米ドル／円の2024年と2025年だけです。他の通貨ペア、株式、暗号資産ではBBWの分布もブレイク後の動きも変わります。 - スクイーズを「過去N本の最小値」と定義しました。下位10%や一定日数の低位、ケルトナーチャネルとの比較は別の条件です。 - 期間・偏差・比較本数・待機本数は記事の範囲に限定しています。隣の数値を含む連続最適化ではありません。 - 終値、SMA、母標準偏差で計算しています。適用価格や平均の種類を変えたBBWは測っていません。 - 4時間足には年3〜10回しか取引しない設定があり、結果の不確実性が大きくなります。 - 約定は終値、スプレッドは0.3pips固定です。スリッページ、時間帯別の変動コスト、保有中の金利は含みません。 --- ### 通貨強弱「最強を買い最弱を売る」は勝てる？72通り検証した結果 URL: https://formiq.jp/ja/blog/currency-strength-settings Language: ja Published: 2026-08-28 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: 通貨強弱, Currency Strength, 最強通貨, 最弱通貨, モメンタム, FX手法, バックテスト Also available in: en — https://formiq.jp/blog/currency-strength-settings 通貨強弱で最も強い通貨を買い、最も弱い通貨を売れば、勢いのある方向へ素直に乗れそうです。しかし、28通貨ペアの値動きから主要8通貨を順位付けし、2024年と2025年のデータで検証したところ、**この方法だけでは利益を残せませんでした**。 1時間足・4時間足・日足、計算期間5〜80本、保有期間1〜24本の72条件で、**最強通貨を買い、最弱通貨を売って両年とも黒字になった条件は0**です。スプレッドを引く前でも、両年プラスは1条件だけでした。1時間足で過去14本の強弱を計算し、4本保有する条件では、2025年に1,553回取引し、勝率39.6%、損益−6,236.2bpsでした。 そこで、最強通貨と最弱通貨だけでなく、2番目に強い通貨や2番目に弱い通貨も組み合わせて検証しました。順張りの成績はほとんど改善しませんでしたが、**過去20営業日で強かった通貨を売り、弱かった通貨を買って約12営業日保有する逆張り**は、検証した5通りの組み合わせすべてで両年黒字でした。売買開始日を分散した場合、単一ペアでは「2番目に強い通貨を売り、最弱通貨を買う」方法、複数ペアでは「上位2通貨を売り、下位2通貨を買う」方法が比較的安定しています。 ここで順位を付ける対象は通貨ペアではなく、EURやUSDといった**個々の通貨**です。たとえば最強通貨がEUR、最弱通貨がUSDなら、取引するのはEUR/USDになります。1つのチャートの上昇率と、8通貨を比べた強弱ランキングは別物です。 ## 通貨強弱とは何を測る指標か 通貨は必ず別の通貨との比率で表示されます。ドル円が上がっただけでは、ドルが全面的に強いのか、円だけが弱いのか分かりません。そこで複数の相手に対する変化をまとめ、1通貨ごとの指数にするのが通貨強弱です。 [BISの実効為替レート](https://data.bis.org/topics/EER)も、複数の為替レートを組み合わせて1通貨の対外的な価値を算出します。ただし、BISの指標は貿易比率を重みに使う経済指標です。短期売買用の通貨強弱メーターには共通規格がなく、各通貨ペアを同じ比率で扱うもの、値動きの大きさを調整するもの、移動平均からの乖離を見るものなどがあります。 今回は8通貨を対等に比べるため、相手となる7通貨への変化率を同じ比率で平均しました。 1. USD、EUR、JPY、GBP、AUD、NZD、CAD、CHFの全28通貨ペアを使う 2. 各ペアの直近N本の対数変化率を計算する 3. 通貨ペアの左側にある通貨にはプラス、右側にある通貨にはマイナスの符号を付ける 4. 各通貨について、相手となる7通貨への変化率を平均する 5. 8通貨の最高点を最強、最低点を最弱とする 式を短く書くと、通貨 `c` の強さは次です。 `強さ(c) = 1/7 × Σ［向きを揃えたN本変化率］` たとえばEUR/USDが上昇した場合は、EURのスコアに加え、USDのスコアから引きます。USD/JPYが上昇した場合は、USDのスコアに加え、JPYのスコアから引きます。7通貨に対する変化を平均するため、特定の通貨ペアだけが大きく動いた影響を抑えられます。 ### 通貨強弱に共通の尺度はない 通貨強弱の値は、対象にする通貨や計算方法を変えるだけで変わります。主要8通貨を同じ比率で平均した指標と、40通貨を貿易比率で加重した指標は、どちらも通貨の強さを表しますが、同じ尺度ではありません。また、今回の0は8通貨の平均より強いか弱いかを示す基準であり、0を上回れば絶対的に強いという意味ではありません。 したがって、メーターを選ぶときは色や順位より先に、**対象通貨、計算期間、変化率の種類、加重方法**を確認します。異なるメーターに表示された+2.0同士を、そのまま比較することはできません。 ## 通貨強弱の設定と使い方 売買結果を大きく左右する設定は、主に次の4項目です。 | 項目 | 今回の値 | 変わるもの | |---|---|---| | 比較する通貨 | 主要8通貨 | 順位を付ける対象 | | 計算期間 | 5 / 10 / **14** / 20 / 40 / 80本 | 何本前からの強弱を見るか | | 更新間隔 | 保有期間ごと | 何本おきに最強・最弱を選び直すか | | 平均方法 | 相手7通貨を同じ比率で扱う | 各通貨ペアをどの程度重視するか | 計算期間が短いほど直近の急変に敏感になり、長いほど数日から数か月の流れを反映しやすくなります。ただし、同じ「14本」でも、1時間足なら14時間、日足なら約3週間です。本数が同じでも時間足が違えば、別の戦略になります。 保有期間は、決済のタイミングとランキングを更新する間隔を兼ねます。1本保有なら毎足ペアを選び直し、24本保有なら途中で順位が変わっても24本後まで保有します。今回の1時間足では、年間の平均コストが1本保有の15,723.7bpsから、24本保有では2,981.4bpsまで減りました。どのくらいの頻度でペアを入れ替えるかは、結果を大きく左右します。 ### MT4・MT5・TradingView・Formiqの違い | 環境 | 通貨強弱の扱い | |---|---| | MT4 / MT5 | 標準指標ではない。カスタム指標を追加し、計算式を確認する | | TradingView | 公開スクリプトごとに対象通貨・式・正規化が違う | | Formiq | 開いているペアの基軸通貨が決済通貨に対して何%動いたかを別枠へ表示する | [MetaTrader 5公式ヘルプ](https://www.metatrader5.com/en/terminal/help/charts_analysis/indicators)では、38種類の標準指標とカスタム指標を分けて説明しています。Currency Strengthは標準指標に含まれないため、同じ名前の指標でも計算方法が同じとは限りません。 Formiqのチャートに表示される通貨強弱は、開いている1通貨ペアについて `（現在値 ÷ N本前 − 1）×100` を描く簡易指標です。計算期間は初期値14で変更できます。一方、**この記事では28通貨ペアを同じ時刻でそろえ、8通貨の相対順位を計算しています**。1通貨ペアのチャートだけでは、8通貨の中でどれが最も強く、どれが最も弱いかは分かりません。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨 | USD / EUR / JPY / GBP / AUD / NZD / CAD / CHF | | 通貨ペア | 上記8通貨の全28組 | | 期間 | 2024-01-01〜2025-12-31 | | 時間足 | 1時間足 / 4時間足 / 日足 | | 全通貨ペアがそろった足数 | 12,439 / 3,222 / 627本 | | 計算期間 | 5 / 10 / **14** / 20 / 40 / 80本 | | 保有期間 | 1 / **4** / 12 / 24本 | | 順張り | 最強通貨を買い、最弱通貨を売る | | 逆張り | 最弱通貨を買い、最強通貨を売る | | 検証した組み合わせ | 1位と8位 / 1位と7位 / 2位と8位 / 2位と7位 / 上位2と下位2 | | エントリー | シグナル足確定後、次の足の始値 | | 決済 | 指定した本数を保有した後の始値。決済時に通貨を選び直す | | コスト | 各足の売値と買値を使う。手数料・スリッページ・スワップポイントは含めない | | 投資額 | 単一ペアは毎回同額。4組を売買する場合は資金を均等配分 | | 最初の検証 | 時間足3種類 × 計算期間6種類 × 保有期間4種類 × 検証期間4区分 × 売買方向2種類 = **576通り** | | 組み合わせの追加検証 | 上記条件 × 5種類の組み合わせ = **2,880通り** | 通貨強弱は、シグナルが確定するまでの終値だけで計算し、次の足の始値で取引します。シグナルの計算に使った終値で、そのまま約定したことにはしていません。上位2通貨と下位2通貨を使う検証では、作れる4組へ資金を4分の1ずつ配分しました。4組の損益を単純に足して、投資額を4倍に見せることはしていません。 異なる通貨ペアの成績を比較できるよう、損益はbpsで集計しました。1bpは0.01%です。表に示す年間bpsは各取引の変化率を合計したもので、複利運用した口座の利回りではありません。 年間成績だけでなく、2025年前半と後半の成績も確認しました。条件を選ぶときは2024年の結果だけを使い、その条件を2025年で確認しています。強弱差フィルターの基準値も2024年のデータだけから決めました。また、12本保有では売買を始める日を1本ずつずらした12通りを試し、特定の開始日にだけ有利な結果になっていないかを確認しました。 ## 最強を買い最弱を売れば勝てるか 勝てませんでした。順張り72条件の年間結果です。各時間足では、計算期間6種と保有期間4種を組み合わせた24通りを試しました。表の中央値は、スプレッドを含む年間合計bpsです。 | 時間足 | 年 | 黒字条件 | 損益中央値 | 両年とも黒字 | |---|---:|---:|---:|---:| | 1時間足 | 2024 | 1/24 | −2,771.6bps | **0/24** | | 1時間足 | 2025 | 0/24 | −4,987.7bps | **0/24** | | 4時間足 | 2024 | 12/24 | −296.7bps | **0/24** | | 4時間足 | 2025 | 0/24 | −1,989.0bps | **0/24** | | 日足 | 2024 | 3/24 | −734.2bps | **0/24** | | 日足 | 2025 | 4/24 | −1,177.8bps | **0/24** | | 合計 | 2024 | 16/72 | 対象外 | **0/72** | | 合計 | 2025 | 4/72 | 対象外 | **0/72** | 4時間足は2024年に半数の条件が黒字でしたが、2025年は24通りすべて赤字です。日足で2025年に黒字だった4条件も、同じ条件では2024年に損失が出ていました。 スプレッドを除いても、両年黒字は72通り中1通りだけです。4時間足の10本計算・12本保有は、スプレッドを引く前なら2024年+359.6bps、2025年+29.1bpsでした。しかし、実際の売値と買値で計算すると+107.2bpsと−245.1bpsになり、両年黒字ではなくなります。 ### 2025年にいちばん良かった条件は2024年に通用しなかった 2025年に最も成績が良かった順張りは、日足で過去80本の強弱を計算し、24本保有する条件の+883.3bpsでした。しかし、同じ条件は2024年に−720.1bpsです。2番目に良かった日足20本・24本保有も、2025年は+857.0bpsでしたが、2024年は−1,670.8bpsでした。 2025年だけを見れば、利益が出た条件はあります。しかし、2025年の結果を見てから選んだ条件なので、今後も同じように機能する根拠にはなりません。 ## 初期設定の14本・4本はどうか 初期設定に近い、1時間足で過去14本の強弱を計算し、4本保有する条件を詳しく見ます。 | 年 | 取引数 | 勝率 | PF | スプレッド前 | スプレッドコスト | スプレッド後 | 平均利益 | 平均損失 | |---:|---:|---:|---:|---:|---:|---:|---:|---:| | 2024 | 1,552 | 43.8% | 0.73 | −889.7bps | −2,908.0bps | **−3,797.7bps** | +15.26bps | −16.24bps | | 2025 | 1,553 | 39.6% | 0.63 | −1,249.1bps | −4,987.1bps | **−6,236.2bps** | +17.38bps | −18.08bps | 2025年は平均利益より平均損失が少し大きく、勝率も4割を切りました。コストを入れる前から−1,249.1bpsなので、負けをスプレッドだけのせいにはできません。 2025年を前半と後半に分けても、前半−4,323.0bps、後半−964.7bpsで、どちらも赤字です。後半はスプレッドを引く前なら+65.4bpsでしたが、1,030.1bpsの取引コストを補えませんでした。 ### 売買を逆にしても取引コストはなくならない 同じペアと時刻で売買を逆にすると、2025年のスプレッド前は+1,249.1bpsです。しかしコストは同じ4,987.0bpsかかるため、スプレッド後は−3,737.9bpsでした。 スプレッドを引く前の損益は、順張りと逆張りでほぼ正負が逆になります。しかし、どちらの売買でもスプレッドは支払います。**売買を逆にすれば必ず利益になるわけではなく、スプレッドが粗利益を上回れば、順張りも逆張りも最終損益はマイナスになります。** ## 去年よかった条件は今年も勝てるか 5条件とも翌年に負けました。2024年に成績が良かった順張り5条件を、同じ設定のまま2025年でも検証しました。 | 時間足 | 計算期間 | 保有期間 | 2024年の損益 | 2025年の損益 | |---|---:|---:|---:|---:| | 4時間足 | 10本 | 24本 | +1,199.0bps | −1,289.6bps | | 4時間足 | 14本 | 24本 | +984.4bps | −1,676.2bps | | 4時間足 | 40本 | 12本 | +793.5bps | −1,835.1bps | | 4時間足 | 40本 | 24本 | +711.6bps | −2,243.8bps | | 4時間足 | 5本 | 24本 | +665.6bps | −1,645.5bps | 5条件の2025年平均は−1,738.0bpsで、黒字は1つもありません。前年に良かったという理由だけでは採用せず、同じ条件が確認年にも黒字かを直接見る必要があります。 ## 条件を変えれば成績は改善するか ### 強弱差が大きい場面だけでも改善しなかった 「最強通貨と最弱通貨の差が大きく開いたときだけ取引する」方法も試しました。2025年の結果を見て基準を決めないよう、2024年の強弱差で上位25%に入る値を基準にし、そのまま2025年へ適用しています。 1時間足の14本計算・4本保有では、すべて取引すると1,553回で−6,236.2bpsでした。条件を満たす430回だけに絞ると−2,744.1bpsです。取引回数が減った分だけ合計損失は小さくなりましたが、1回あたりの平均損益は−4.016bpsから−6.382bpsへ悪化し、PFも0.63から0.59へ下がりました。 4時間足の10本計算・12本保有でも、133回・−245.1bpsが39回・−153.2bpsになっただけです。1回あたりの平均損益は−1.843bpsから−3.927bpsへ悪化し、PFも0.94から0.91へ下がりました。**強弱差が大きいほど、その後も同じ方向へ動きやすいという結果にはなっていません。** ### 長く保有するとコストは減るが、収益性までは改善しない 1時間足の2025年、計算期間6種の平均です。 | 保有期間 | 平均取引数 | スプレッド前 | コスト | スプレッド後 | |---:|---:|---:|---:|---:| | 1本 | 6,214 | −575.6 | −15,723.7 | −16,299.4 | | 4本 | 1,553 | −1,167.3 | −5,037.3 | −6,204.6 | | 12本 | 517 | −704.9 | −3,388.1 | −4,093.0 | | 24本 | 258 | −649.5 | −2,981.4 | −3,630.8 | 保有期間を長くすると通貨ペアを入れ替える回数が減り、合計損失も小さくなります。それでも、6つの計算期間を平均した損益は、スプレッドを引く前からすべてマイナスです。頻繁な売買がコストを増やしていたのは確かですが、取引回数を減らすだけでは利益を生み出せませんでした。 ### 逆張りでは6条件が両年黒字 最弱通貨を買い、最強通貨を売る逆張りでは、72条件中6条件が両年黒字でした。そのうち、4時間足の20本計算・24本保有は次の成績です。 | 年 | 取引数 | 勝率 | PF | 損益 | |---|---:|---:|---:|---:| | 2024 | 66 | 39.4% | 1.15 | +375.3bps | | 2025 | 66 | 59.1% | 2.45 | +2,454.9bps | ただし、この条件は576通りの結果をすべて見た後で見つけたものです。2024年の時点で逆張り上位5条件を選び、同じ条件を2025年で検証すると、5つすべてが赤字で、平均は−1,745.4bpsでした。一部の条件で短期的な反転傾向は見られましたが、単純に売買を逆にすれば勝てるとは言えません。 ## 2番目に強い・弱い通貨も使うと改善するか 最強通貨と最弱通貨だけでなく、2番目に強い通貨と2番目に弱い通貨も組み合わせ、同じ72条件を5通りの方法で検証しました。「上位2×下位2」では、上位2通貨と下位2通貨から作れる4組へ資金を均等に配分します。表は、スプレッドを差し引いた後に2024年と2025年の両方で黒字になった条件数です。 | 順位の組み方 | 順張り | 逆張り | |---|---:|---:| | 最強 × 最弱 | 0/72 | 6/72 | | 最強 × 2番目に弱い | 1/72 | 4/72 | | 2番目に強い × 最弱 | 1/72 | **12/72** | | 2番目に強い × 2番目に弱い | 2/72 | 6/72 | | 上位2 × 下位2（4組へ均等配分） | 0/72 | 1/72 | 2番目に強い通貨や2番目に弱い通貨を使っても、順張りで両年黒字になったのは72条件中、最大でも2条件です。最強と最弱だけを選ぶ方法が極端すぎたのではなく、直近の強弱をそのまま追いかける順張り自体が、この2年間では安定しませんでした。成績が改善したのは逆張りです。 ### 日足20本計算・12本保有の逆張りは、すべての組み合わせで黒字 過去20営業日の日足から強弱を計算し、順位の高い通貨を売って低い通貨を買い、12本後（約12営業日後）に決済する方法は、5通りの組み合わせすべてで両年黒字でした。各年の最初の取引可能日から12本ごとに通貨を選び直した結果は次のとおりです。 | 逆張りの組み方 | 2024年 | 2025年 | |---|---:|---:| | 最強を売る／最弱を買う | +526.4bps | +783.1bps | | 最強を売る／2番目に弱い通貨を買う | +567.6bps | +1,009.2bps | | 2番目に強い通貨を売る／最弱を買う | +620.9bps | +993.2bps | | 2番目に強い通貨を売る／2番目に弱い通貨を買う | **+642.5bps** | **+1,207.0bps** | | 上位2通貨を売る／下位2通貨を買う | +589.4bps | +998.1bps | 近い設定でも同じ結果になるかを確かめるため、計算期間14・20・40本と、保有期間4・12・24本を組み合わせた日足9条件も比べました。5通りの通貨の組み合わせすべてが両年黒字になったのは、20本計算・12本保有だけです。20本計算・4本保有では該当なし、20本計算・24本保有では1通りだけでした。**少し設定を変えるだけで結果が崩れており、パラメーターへの依存は強めです。** ### 売買を始める日によって成績が変わる 日足で12本ごとに通貨を選び直す場合、1月の何日から売買を始めるかによって、その後の取引日が変わります。売買開始日を1営業日ずつずらした12通りを試すと、上位2通貨を売り、下位2通貨を買う方法の年間損益は、2024年が−321.6〜+970.9bps、2025年が−748.0〜+998.1bpsでした。特定の開始日だけを選ぶと、実際よりも安定した方法に見える可能性があります。 そこで資金を12等分し、毎日そのうち1つを新しい取引へ振り向け、約12営業日後に決済する方法を試しました。上位2通貨と下位2通貨を使う場合は、その日の投資分をさらに4組へ均等に配分します。表は12通りの開始タイミングの平均で、スプレッドを差し引いた後の年間損益です。 | 逆張りの組み方 | 2024年 | 2025年 | 両年黒字の開始パターン | |---|---:|---:|---:| | 最強を売る／最弱を買う | +499.9bps | +450.6bps | 7/12 | | 最強を売る／2番目に弱い通貨を買う | +454.8bps | +125.9bps | 5/12 | | 2番目に強い通貨を売る／最弱を買う | +315.6bps | **+603.4bps** | **9/12** | | 2番目に強い通貨を売る／2番目に弱い通貨を買う | +266.1bps | +291.4bps | 5/12 | | 上位2通貨を売る／下位2通貨を買う | **+384.1bps** | **+367.8bps** | **8/12** | 1組だけを取引する方法では、「2番目に強い通貨を売り、最弱通貨を買う」組み合わせが最も安定し、12通り中9通りの開始パターンで両年黒字でした。上位2通貨を売り、下位2通貨を買う方法では、2025年に最も成績が悪かった開始パターンが−748.0bpsでした。「2番目に強い通貨を売り、最弱通貨を買う」方法の−1,410.1bpsよりも、開始日による振れは小さくなっています。 開始日を分散した上位2・下位2の方法は、2025年前半が+336.3bps、後半が+31.5bpsでした。後半はほぼ横ばいなので、どのような相場でも利益が出るとは言えません。 ### 今回の検証で比較的安定していた条件 今回の結果を具体的な手順にすると、次のようになります。 1. 主要8通貨から作れる28通貨ペアの日足を、同じ時刻でそろえる 2. 各通貨について、相手となる7通貨への過去20営業日の変化率を同じ比率で平均する 3. 順位の高い通貨を売り、順位の低い通貨を買う 4. 資金を12等分し、毎日1/12ずつ新規取引へ振り向けて約12営業日後に決済する 5. 1組だけ取引するなら2番目に強い通貨を売って最弱通貨を買い、分散するなら上位2通貨を売って下位2通貨を買う4組へ均等に投資する これは**今回の候補で両年黒字だった条件**ですが、近い設定へ変えると崩れており、今後の利益を保証するものではありません。約12営業日にわたって保有するため、今回含めていないスワップポイントの影響も受けます。また、強弱差が2024年の中央値以上のときだけ取引すると、上位2・下位2の方法は2025年に7回・+1,017.3bpsでした。しかし、取引回数が少なすぎるため、有効な追加条件とは判断しませんでした。 ## 研究の通貨モメンタムと同じか [BISのワーキングペーパー「Currency Momentum Strategies」](https://www.bis.org/publ/work366.pdf)では、多数の通貨を過去の収益率で順位付けし、上位通貨を買って下位通貨を売る戦略に統計的に有意な収益があったと報告しています。一方で、収益の一部は取引コストで失われ、実際の運用には制約があることも指摘しています。 今回の結果と矛盾するとは限りません。違いは少なくとも3つあります。 - 先行研究は多数の通貨へ分散するのに対し、今回は最大でも上位2通貨と下位2通貨から作る4組だけを取引している - 先行研究は月次データで強弱を計算して保有するのに対し、今回は1時間足から日足まで、1〜24本の保有期間を試している - 先行研究は多くの通貨を長期間検証しているのに対し、今回は主要8通貨の2024〜2025年だけを対象にしている 通貨モメンタムを支持する研究があっても、「過去14時間で最も強かった通貨を4時間買えば利益になる」とは限りません。順位の計算方法、同時に保有する通貨数、時間軸、取引コストが違えば、売買戦略としては別物です。 ## 関連記事 [移動平均線クロスの設定検証](/ja/blog/moving-average-cross-settings)は、1通貨ペアの過去の値動きからトレンドを判断する方法です。今回の通貨強弱は、同じ時刻に複数の通貨を比べて順位を付ける点が異なります。 [RSIの期間と30/70を比べた検証](/ja/blog/rsi-settings)では、直近の値動きを順張りと逆張りの両方で検証しています。[プログラミングなしでバックテストする方法](/ja/blog/backtest-without-coding)では、1通貨ペアのエントリー条件や決済方法を組み立てる手順を紹介しています。今回のように28通貨ペアを同時に順位付けする方法は、1つのチャートだけを使う通常のバックテストとは分けて考える必要があります。 ## 補足 - 対象は主要8通貨と、その全28通貨ペアです。新興国通貨、金、株価指数、暗号資産は含みません。 - 検証期間は2024年と2025年の2年間だけです。異なる金融政策や長期的な相場環境でも同じ結果になるとは限りません。 - 各通貨の強弱は、相手となる7通貨への変化率を同じ比率で平均しています。貿易比率、値動きの大きさ、金利差、移動平均からの乖離などを使う指標では結果が変わります。 - 最強・2番目に強い通貨と、最弱・2番目に弱い通貨の組み合わせまでを検証しました。3番目以降の通貨を含む、より広く分散した方法や、値動きの大きさに応じて投資額を変える方法は試していません。 - 決済は指定した本数を保有した後に行います。損切り、利確、順位が逆転したときの決済、トレーリングストップは含みません。 - 実際の売値と買値の差は反映しましたが、手数料、スリッページ、保有中のスワップポイント、注文量による約定価格の違いは含みません。 - 日足20本計算・12本保有の逆張りも、2024〜2025年の結果を比較して見つけた条件です。開始日を変えた検証と近い設定の比較は行いましたが、2026年以降のデータではまだ確認していません。 --- ### フェアバリューギャップ「埋まる」は本当？2年分の足で検証した結果 URL: https://formiq.jp/ja/blog/fair-value-gap-settings Language: ja Published: 2026-08-28 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: フェアバリューギャップ, FVG, プライスアクション, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/fair-value-gap-settings フェアバリューギャップ（Fair Value Gap、FVG）は、ローソク足3本で判定するパターンです。1本目の高値と3本目の安値が重ならなければ、その間の値幅は「真ん中の足が一気に通り抜けて、売り買いが釣り合わないまま残った領域」だと読みます。価格はそこに戻ってくる、というのが公開されている説明です。 戻ってきます。ドル円で数えると96.2〜99.4%が戻られていました。 問題は、その数字が**フェアバリューギャップについて何も言っていない**ことです。強気ギャップの天井は3本目の足の安値そのものなので、「ギャップが埋まった」は「その足の安値に価格が戻った」と一字一句同じ出来事です。そして安値は、どのローソク足にもあります。 ## フェアバリューギャップとは 判定は3本の足の比較1回だけです。 - **強気**：3本目の安値 > 1本目の高値。帯はその2つの価格のあいだ - **弱気**：3本目の高値 < 1本目の安値。帯はその2つのあいだ 真ん中の足は出てきません。式に入っていないので、そこが大陽線でも十字線でも判定は変わりません。 ドル円で、この条件が成立する頻度を数えました。 | 時間足 | 年 | 足の本数 | ギャップ | 足に対する割合 | 帯の幅（中央値） | |---|---|---|---|---|---| | 15分足 | 2025 | 24,903 | 5,610 | 22.53% | 2.95pips | | 15分足 | 2024 | 24,999 | 5,455 | 21.82% | 2.70pips | | 1時間足 | 2025 | 6,226 | 1,312 | 21.07% | 6.85pips | | 1時間足 | 2024 | 6,250 | 1,271 | 20.34% | 5.80pips | | 4時間足 | 2025 | 1,610 | 368 | 22.86% | 15.75pips | | 4時間足 | 2024 | 1,616 | 369 | 22.83% | 13.50pips | **5本に1本の足がギャップを開けています。** 珍しい形ではありません。15分足では幅が0.1pipsに満たないものが年82〜89件あって、これはチャートに描いても帯にならず点線1本に見えます。 ## 本当に埋まるのか 99%が埋まります。ただしギャップを作らなかった足も、同じだけ埋まりました。ギャップが埋まった割合と、ギャップを作らなかった足の安値・高値に価格が戻った割合を並べます。後者は「1本目と3本目が重なっていた足」、つまりフェアバリューギャップではない足です。 | 時間足 | 年 | ギャップの足 | それ以外の足 | |---|---|---|---| | 15分足 | 2025 | 99.36% | 99.37% | | 15分足 | 2024 | 99.19% | 99.39% | | 1時間足 | 2025 | 98.55% | 98.71% | | 1時間足 | 2024 | 98.27% | 98.82% | | 4時間足 | 2025 | 97.01% | 97.18% | | 4時間足 | 2024 | 96.21% | 97.79% | 冒頭の図がこの表です。**6区分すべてで、ギャップの足のほうがわずかに低い。** 戻られるまでの本数の中央値も、両方とも2本で一致しました。 「フェアバリューギャップは埋まる」は正しい観測ですが、埋まるのはギャップだからではありません。レンジのある相場では、どの足の高安もいずれ通過されます。この指標は、その一般的な性質に名前を付けたものです。 ## 埋まるまで何本かかるのか 中央値は2本です。待ち時間を数えると、さらに使いにくくなります。 | 時間足 | 年 | 中央値 | 次の1本で | 3本以内 | 上位10%の待ち | |---|---|---|---|---|---| | 15分足 | 2025 | 2本 | 44.03% | 63.31% | 44本 | | 15分足 | 2024 | 2本 | 46.07% | 64.02% | 41本 | | 1時間足 | 2025 | 2本 | 45.32% | 62.57% | 43本 | | 1時間足 | 2024 | 2本 | 43.55% | 62.37% | 40本 | | 4時間足 | 2025 | 2本 | 45.10% | 63.59% | 29本 | | 4時間足 | 2024 | 2本 | 41.97% | 63.10% | 54本 | **4割強が、ギャップの成立した次の1本で埋まっています。** 3本目の足が終わった瞬間にチャートへ帯が現れて、次の足で消える、という動きが最頻です。 「未約定のギャップに価格が戻ったら入る」という手順は、この分布の上では成立しません。帯を認識してから注文を置くまでのあいだに、半分近くが終わっています。 埋まらずに残るものもあります。2025年の1年間を通して最後まで開いていたのは、15分足で36件（5,610件中）、1時間足で19件（1,312件中）、4時間足で11件（368件中）でした。 ## 帯の内側で止まるのか 止まりません。最初の1本で半分が閉じ切ります。帯には幅があります。「天井に触れる」と「底まで埋める」は別の出来事のはずです。最初に触れた1本が、帯のどこまで入ったかを測りました。帯の幅を1として、0が天井、1が反対側です。 | 時間足 | 触れた回数 | 到達した深さ（中央値） | 半分まで | 完全に埋めた | 帯の1.5倍まで抜けた | |---|---|---|---|---|---| | 15分足 | 5,574 | 1.12 | 70.58% | 53.07% | 42.07% | | 1時間足 | 1,293 | 1.06 | 66.82% | 51.35% | 40.84% | | 4時間足 | 357 | 0.97 | 66.11% | 49.30% | 39.78% | **中央値がほぼ1.0です。** 触れた足がそのまま帯を通り抜けるのが真ん中の挙動で、約半分は1本で完全に埋め切ります。4割は帯を半分ぶん行き過ぎます。 帯の内側に価格が入って止まる、という描像は出てきませんでした。 ## チャートに表示する方法 標準では入っていないので、どのツールでも一手間かかります。 - **MT4 / MT5**：同梱のインジケーター一覧にフェアバリューギャップはありません。カスタムインジケーター（.mq4 / .mq5）を `MQL4/Indicators` などに置いてから、ナビゲーターで追加します - **TradingView**：標準のインジケーター一覧には無く、公開スクリプトから探して追加します - **Formiq**：チャートのインジケーター設定にパターンの分類で入っています。埋まっていないギャップだけを表示する切り替えと、最小の幅（pips）の指定ができます 表示結果を変えるのは、「埋まっていないギャップだけを表示する」切り替えと、表示するギャップの最小幅です。ギャップは足の2割で生まれて99%が埋まるので、すべて表示すると過去の帯でチャートが埋まります。 ## 真ん中の足が大きいほうが効くか 判定式は真ん中の足を読んでいません。公開されている説明では、真ん中の足は「displacement（勢いのある足）」と呼ばれ、大きいほど良いとされます。判定式はそこを読みません。読まないことが損なのかを測りました。 真ん中の足の実体をATR比で4等分して、価格が水準に戻ったあと10本でギャップの向きに動いていた割合を出します。 | 時間足 | Q1（最小） | Q2 | Q3 | Q4（最大） | |---|---|---|---|---| | 15分足 | 49.68% | 47.02% | 51.76% | 49.32% | | 1時間足 | 53.11% | 53.73% | 47.52% | 41.93% | | 4時間足 | 44.32% | 47.19% | 49.44% | 52.81% | 実体が大きいほど良い、にはなりませんでした。15分足はQ3が最大、1時間足はQ2が最大で大きいほど下がり、4時間足はQ4が最大です。**時間足ごとに向きが違います。** もう一段踏み込んで、判定の境目そのものを見ます。1本目と3本目の重なりをATR比の符号付きで測り、マイナス（重なっている＝ギャップではない）も含めて区切りました。 ゼロの左右で何も起きていません。区分ごとの数字は41.46〜55.87%（3時間足×8区分の24通り）に散らばりますが、ゼロをまたぐ段差は15分足が−1.81ポイント、1時間足が+2.81ポイント、4時間足が−4.35ポイントで、**符号が揃いません**。 インジケーターが引いている線は、連続した量の途中に置かれた線でした。 ## 検証の条件 価格上の性質を確認した同じギャップを、売買条件として測ります。 | 項目 | 値 | |---|---| | 通貨ペア | ドル円 | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 期間 | 2025年（本命）、2024年（検定用）、2025年の前半と後半 | | スプレッド | 0.3pips | | ロット | 0.1 | | 決済 | 反対シグナル（損切り・利確は別途） | | 読み方 | 戻りを取る / 発生に乗る | | ギャップの最小幅 | ATRの0倍・0.25倍・0.5倍・1倍・1.5倍・2倍 | | 有効期間 | 5・10・20・50・100本（戻りを取る側のみ） | 読み方は2つに分けました。 - **戻りを取る**：まだ埋まっていない帯に価格が戻ったら、帯の向きに入る（強気ギャップ＝買い）。教科書どおりの読み方です - **発生に乗る**：ギャップが生まれた足で、その向きに入る。同じ形を勢いとして読みます 最小幅をpipsではなくATR倍にしたのは、中央値が15分足の2.95pipsから4時間足の15.75pipsまで動くからです。1つのpips値では3つの時間足をまたげません。 ## どの時間足で使えたか 1時間足の「戻りを取る」は、両方の年で黒字だった設定が1つもありませんでした。2025年に黒字だった設定の数です。 | 読み方 | 時間足 | 年 | 黒字設定 | 両年とも黒字 | 2025年の中央値 | |---|---|---:|---:|---:|---:| | 戻りを取る | 15分足 | 2025 | 13/30 | 6/30 | −361.7pips | | 戻りを取る | 15分足 | 2024 | 20/30 | 6/30 | 対象外 | | 戻りを取る | 1時間足 | 2025 | 6/30 | 0/30 | −1,817.6pips | | 戻りを取る | 1時間足 | 2024 | 20/30 | 0/30 | 対象外 | | 戻りを取る | 4時間足 | 2025 | 5/30 | 5/30 | −1,842.8pips | | 戻りを取る | 4時間足 | 2024 | 22/30 | 5/30 | 対象外 | | 発生に乗る | 15分足 | 2025 | 2/6 | 1/6 | −513.5pips | | 発生に乗る | 15分足 | 2024 | 3/6 | 1/6 | 対象外 | | 発生に乗る | 1時間足 | 2025 | 3/6 | 3/6 | −79.7pips | | 発生に乗る | 1時間足 | 2024 | 6/6 | 3/6 | 対象外 | | 発生に乗る | 4時間足 | 2025 | 1/6 | 1/6 | −1,584.7pips | | 発生に乗る | 4時間足 | 2024 | 6/6 | 1/6 | 対象外 | **1時間足の「戻りを取る」は、30通りのうち両方の年で黒字だったものが1つもありません。** 2024年は6区分すべてで半分かそれ以上が黒字（20/30・20/30・22/30・3/6・6/6・6/6）でしたが、2025年は最大でも13/30です。この2年だけでは、どちらの状態が続きやすいかは判断できません。 条件エディタが開いたときの設定（最小幅ATR0.25倍・有効期間50本）の成績を出します。 | 読み方 | 時間足 | 年 | 取引 | 勝率 | PF | 損益 | |---|---|---:|---:|---:|---:|---:| | 戻りを取る | 15分足 | 2025 | 1,334 | 47.08% | 1.078 | +1,251.5pips | | 戻りを取る | 15分足 | 2024 | 1,278 | 48.51% | 1.075 | +1,235.2pips | | 戻りを取る | 1時間足 | 2025 | 321 | 41.43% | 0.689 | −3,023.1pips | | 戻りを取る | 4時間足 | 2025 | 96 | 43.75% | 0.636 | −2,107.5pips | | 発生に乗る | 15分足 | 2025 | 1,211 | 37.90% | 1.052 | +840.7pips | | 発生に乗る | 4時間足 | 2024 | 73 | 41.10% | 1.751 | +2,852.4pips | 15分足だけが2年ともプラスです。取引回数も1,200を超えていて、少ない回数で作られた数字ではありません。 最小幅の効き方は15分足で最もはっきりしています。 | 最小幅 | 平均損益 | 平均取引数 | 黒字だった設定 | |---|---|---|---| | 0倍（無効） | +381.5pips | 2,314 | 3/5 | | ATR 0.25倍 | +1,382.9pips | 1,229 | 5/5 | | ATR 0.5倍 | +2,228.5pips | 605 | 5/5 | | ATR 1倍 | −606.9pips | 133 | 0/5 | | ATR 1.5倍 | −1,302.1pips | 49 | 0/5 | | ATR 2倍 | −949.1pips | 29 | 0/5 | 15分足・2025年・「戻りを取る」の数字です。細いギャップを捨てるところまでは効いて、捨てすぎると残りません。 ## 去年よかった設定は今年どうなるか 6通りすべて翌年に負けました。このシリーズで毎回やる検定です。片方の年でいちばん良かった設定を、そのまま相手の年に当てます。 | 読み方 | 時間足 | 最小幅 | 有効期間 | 2024年の損益 | 2025年の損益 | |---|---|---:|---:|---:|---:| | 戻りを取る | 15分足 | ATR 1倍 | 100本 | +3,713.3pips | −1,172.2pips | | 戻りを取る | 1時間足 | ATR 1倍 | 100本 | +3,968.4pips | −2,744.2pips | | 戻りを取る | 4時間足 | ATR 1.5倍 | 20本 | +2,946.3pips | −1,765.1pips | | 発生に乗る | 15分足 | ATR 1倍 | 対象外 | +2,445.9pips | −136.7pips | | 発生に乗る | 1時間足 | ATR 1倍 | 対象外 | +4,186.1pips | −1,017.3pips | | 発生に乗る | 4時間足 | ATR 0.25倍 | 対象外 | +2,852.4pips | −1,694.4pips | **6通りすべてが赤字になりました。** 1時間足の「戻りを取る」は2024年の+3,968.4pipsから2025年の−2,744.2pipsへ、「発生に乗る」は+4,186.1pipsから−1,017.3pipsへ変わりました。 ## 勝率5割を超えた設定はあるか 年20回以上取引した設定に限ると、6つの区分すべてで勝率が50%を超えた設定は**ゼロ件**でした。 | 読み方 | 時間足 | 対象 | 平均勝率 | 平均利益 | 平均損失 | |---|---|---|---|---|---| | 戻りを取る | 15分足 | 30通り | 43.56% | 76.93pips | −72.33pips | | 戻りを取る | 1時間足 | 20通り | 40.00% | 79.40pips | −83.68pips | | 戻りを取る | 4時間足 | 20通り | 37.43% | 101.46pips | −132.37pips | | 発生に乗る | 15分足 | 6通り | 38.78% | 77.89pips | −72.53pips | | 発生に乗る | 1時間足 | 4通り | 38.52% | 92.30pips | −67.71pips | | 発生に乗る | 4時間足 | 4通り | 34.38% | 137.85pips | −120.82pips | 「戻りを取る」という名前でも、勝率は37.43〜43.56%でした。さらに1時間足と4時間足では平均損失が平均利益より大きく、勝率だけで逆張りの優位性を確認できる結果ではありません。 ## フィルター・損切り・スプレッド 条件エディタの初期設定に足して測りました。すべて2025年です。 ADXフィルターは、唯一黒字だった15分足の「戻りを取る」を削るだけでした。 | 条件 | 取引 | 勝率 | 損益 | |---|---|---|---| | そのまま | 1,334 | 47.08% | +1,251.5pips | | ADX(14) 20以上 | 945 | 46.24% | +537.2pips | | ADX(14) 25以上 | 662 | 46.53% | +288.0pips | | 07-16 UTC のみ | 640 | 46.09% | −1,013.2pips | | 12-21 UTC のみ | 574 | 49.13% | +888.9pips | 時間帯は片方だけ残ります。12-21 UTC は取引を6割弱に減らして損益を+888.9pipsに保ちますが、07-16 UTC は符号が反転して−1,013.2pipsです。片方が効いてもう片方が壊れる形は、フィルターというより期間の切り分けに近いので、そのまま使う根拠にはなりません。 損切り・利確は15分足でわずかに上積みし（SL30/TP60で+1,354.1pips、SL50/TP100で+1,312.0pips、そのままが+1,251.5pips）、赤字だった4時間足では損失を縮めます（そのまま−2,107.5pips → SL30/TP60で−132.0pips）。赤字を小さくしているだけで、黒字にはなりません。 スプレッドは、このシリーズで毎回そうなる形になりました。 | 読み方 | 時間足 | 取引 | スプレッド0 | 0.3pips | 1.0pips | 損益分岐 | |---|---|---|---|---|---|---| | 戻りを取る | 15分足 | 1,334 | +1,651.6pips | +1,251.5pips | +317.7pips | 1.238pips | | 発生に乗る | 15分足 | 1,211 | +1,204.0pips | +840.7pips | −7.0pips | 0.994pips | | 戻りを取る | 1時間足 | 321 | −2,926.8pips | −3,023.1pips | −3,247.8pips | なし | **スプレッドで失うpipsは、取引回数×スプレッドと一致します。** 6通りすべてで、予測と実測の差は0.1pips以内でした。15分足で黒字だった2つは、スプレッドが1.238pipsと0.994pipsを超えると消えます。実際のドル円ならまだ余裕がありますが、年1,200回以上取引する設定の余裕としては薄いほうです。 関連する記事として、FVGを含む枠組みを価格パターンへ分解した[SMCの構造ブレイクと流動性スイープ](/ja/blog/smc-settings)があります。同じ「描画ツールを条件に昇格させて測る」形は[フィボナッチ・リトレースメント](/ja/blog/fibonacci-retracement-levels)で作りました。同じ対照群の置き方でローソク足の形を17種まとめて測ったものが[ローソク足パターン17種の勝率](/ja/blog/candlestick-patterns-tested)です。基準率を先に数えてから条件を評価する形は[GMMA](/ja/blog/gmma-settings)、指標が別の既知のものと同じだった例は[CCI](/ja/blog/cci-settings)にあります。勝率と損益の向きについては[ボリンジャーバンド](/ja/blog/bollinger-band-settings)が対照になります。 ## 補足 - **ドル円1本のみ。** 通貨ペアを変えると変わる可能性があります。とくに帯の幅は値動きの大きさに直結します - **2024年と2025年の2年だけ**です - **決済は反対シグナル。** 実際にこのパターンを使う人は、帯の反対側に損切りを置いたり、直近の高安まで持ったりします。その形は測っていません - **「戻る」の定義は帯の端に触れた時点。** 終値で判定する流儀もあり、その場合の数字は変わります - **時間帯フィルターは2つの窓のうち片方だけを設定して測っています。** 実運用でどの時間に絞るかは、ここで測った2つ以外にもあります - **上位足のギャップを下位足で使う形は測っていません。** この検証は各時間足の中で完結しています --- ### 平均足「色が変わったら売買」は勝てる？960通り検証した結果 URL: https://formiq.jp/ja/blog/heikin-ashi-settings Language: ja Published: 2026-08-28 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: 平均足, ローソク足, 順張り, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/heikin-ashi-settings 平均足（へいきんあし、Heikin Ashi）は、ローソク足を書き換えたものです。インジケーターのように足の横に線を引くのではなく、**足そのものを別の数字に置き換えます**。 計算は2行だけです。終値は4本値の平均、始値は1本前の平均足の真ん中。それだけで、チャートはずっと滑らかになります。 このシリーズで足そのものを書き換える対象を測るのは、21本目にして初めてです。そして測ってみると、他の20本には無かった問題が出てきました。**平均足の終値は、誰も売買していない価格**です。 同じ6,107回の売買を、実際の終値で計算するか、平均足の終値で計算するかだけ変えました。 ## 平均足は何を計算しているのか 1本の足につき2行です。 - **平均足の終値** =（始値＋高値＋安値＋終値）÷ 4 - **平均足の始値** =（1本前の平均足の始値＋1本前の平均足の終値）÷ 2 高値と安値は、この2つと実際の高値・安値のうち外側を取ります。 この2行から、他のインジケーターには無い性質が3つ出てきます。 ### 平均足の終値は、価格ではない 4本値の平均は、4つのどれでもありません。**その値段で取引された瞬間は、その足の中に存在しません。** どれくらい離れているかを測りました。 | 時間足 | 実際の終値との差（中央値） | 平均 | 上位10% | 最大 | |---|---|---|---|---| | 15分足 | 2.22pips | 3.09pips | 6.70pips | 78.5pips | | 1時間足 | 4.55pips | 6.22pips | 13.46pips | 113.2pips | | 4時間足 | 9.26pips | 12.58pips | 27.34pips | 147.6pips | ドル円2025年の数字です。しかも**ずれる向きが偏っています。** 陽線の平均足のうち、平均足の終値が実際の終値より**下**にあったものは63.7〜67.3%（時間足3種×2年の6通り）。買いのシグナルが出たとき、平均足の終値は実勢より安いほうに寄っているということです。 ### 平均足は、自分の開始日を忘れない 平均足の始値は、1本前の平均足から作られます。その1本前も、さらに1本前から作られます。**期間を指定する項目がないのは、計算開始以降の平均足が連続して引き継がれるから**です。 過去の影響は消えません。ただし毎回半分になります。同じ足を、500本ずらした別の日から計算し直して比べました。 | 時間足 | 開始時点の差 | 1pips以内になるまで | 0.1pips | 0.01pips | 0.001pips | |---|---|---|---|---|---| | 15分足 | 4.76pips | 3本 | 6本 | 9本 | 13本 | | 1時間足 | 7.06pips | 3本 | 7本 | 10本 | 13本 | | 4時間足 | 7.30pips | 3本 | 7本 | 10本 | 13本 | **1本進むごとの差の比は、測れた16〜17本すべてで0.5ちょうど**でした。2つの漸化式を引き算すると差の項だけが残って半分になるので、当然ではあります。 実務上の意味は2つです。**開始日を変えると平均足の値は変わります**（3桁の精度でも13本で追いつくので、ふつうは気になりません）。そして**平均足には「何本ぶんのデータが要るか」の答えが無い**。50本あれば実用上は十分ですが、厳密にゼロになる本数はありません。 ### 平均足の色は、移動平均のクロスそのもの ここが一番はっきりした結果です。 平均足の始値は「1本前の始値と終値の平均」なので、**平均足の終値を係数0.5で指数平滑したもの**と同じです。係数0.5の指数移動平均は、期間3のEMAです。 つまり **平均足が陽線であること＝（4本値の平均）がその3期間EMAを1本前の時点で上回っていること**、そのものです。 実測でも、時間足3種×2年の6通りで**色の食い違いは0件、色が変わった足の食い違いも0件**でした（始値の誤差は最大0.08〜0.48pips、これは開始時点の名残です）。 **平均足のローソク足は、価格と移動平均のクロスをローソク足の形で描いたもの**だ、と言えます。この記事の後半で、そのことが成績にどう出るかを測ります。 ## 平均足をチャートに表示する方法 | 環境 | 手順 | |---|---| | MT4 / MT5 | Heiken Ashi が最初から同梱されている。ナビゲーター → インディケータ → Heiken Ashi をチャートにドラッグする | | TradingView | チャート上部の足種メニューから平均足（Heikin Ashi）を選ぶ | | ブラウザ（Formiq） | チャート種類から選べる。バックテスト側では読み方4種（色の変化・N本連続・ヒゲの無い足・同時線）と計算方式4種を条件にできる | **TradingViewで注意が要るのはここです。** チャート種類を平均足にしたままストラテジーを動かすと、**約定価格まで平均足の値になります。** この記事で測っているのは、まさにその差です。 MT4は逆で、Heiken Ashi は描画するだけなので `Close[]` は実際の終値のままです。ただしMT4には平均足を返す組み込み関数が無いので、EAに書き出すときは**直近N本から自分で組み立てるループ**を書くことになります。 ## 検証の条件 数字を見る前に、何をどう測ったかを書いておきます。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 読み方 | 色の変化 ／ 同じ色がN本連続（1〜5） ／ ヒゲの無い足 ／ 同時線（実体比率0.05・0.1・0.2） | | 計算方式 | 標準 ／ ベター（2本平均） ／ 平滑化（3・5・10） ／ 二重平滑化（3・5・10） | | 組み合わせ | 1区分あたり80通り。時間足3種 × 期間4区分で合計960通り | | 決済 | 反対シグナルまで。4つの読み方はすべて対称なので、エントリーと決済に同じルールを使う | | **約定** | **実際の足の終値。平均足はいつ売買するかだけを決める** | | 損切り・利確 | 使わない（別セクションで測定） | | スプレッド | 0.3pips固定 | | ロット | 0.1ロット | **約定を実際の足で取ることが、この検証の前提です。** 平均足は「いつ」を決め、「いくらで」は市場が決めます。次のセクションは、その前提を外すと何が起きるかの測定です。 ## 成績が良く見えるのはなぜか 約定価格に平均足の終値を使うからです。戦略は一切変えません。**約定価格をどちらの終値で書くかだけ**を変えます。スプレッドは両方とも0にして、価格の差だけが残るようにしました。 冒頭の図がその結果です。灰色が実際の終値、赤が平均足の終値。同じ6,107回の売買です。 | 読み方 | 時間足 | 年 | 実際の終値 | 平均足の終値 | 差 | |---|---|---|---|---|---| | 色の変化 | 15分足 | 2025 | −523.1 | **+34,120.0** | +34,643.1 | | 色の変化 | 15分足 | 2024 | −376.0 | **+34,013.3** | +34,389.3 | | 色の変化 | 1時間足 | 2025 | +1,617.7 | **+18,228.6** | +16,611.0 | | 色の変化 | 4時間足 | 2025 | +910.7 | **+10,203.5** | +9,292.8 | | N本連続(3) | 1時間足 | 2025 | +1,276.5 | +3,875.8 | +2,599.3 | | ヒゲの無い足 | 1時間足 | 2025 | +1,123.1 | +6,708.6 | +5,585.6 | | 同時線(0.1) | 15分足 | 2025 | −2,450.0 | **+2,996.1** | +5,446.1 | **6つの読み方×3時間足×2年の36通りのうち、平均足の終値で計算したほうが良くなったのが35通り。** そのうち**15通りは、赤字が黒字に変わりました。** ### 差がどこから来るのか 差が生まれる理由は、約定が入口と出口の2回あり、そのたびに平均足の終値と実勢の差を1回ずつ加算するためです。陽線では、平均足の終値が実際の終値より安かった足が、時間足3種×2年の6区分で63.7〜67.3%を占めました。そのため、平均足の終値で計算すると、買いは実勢より安く、売りは実勢より高く約定したことになります。 | 時間足 | 年 | 実勢との差の平均 | ×2 | 実測の1取引あたりの差 | 比 | |---|---|---|---|---|---| | 15分足 | 2025 | 3.09 | 6.18 | 5.67 | 0.917 | | 1時間足 | 2025 | 6.22 | 12.44 | 10.73 | 0.863 | | 4時間足 | 2025 | 12.58 | 25.17 | 24.33 | 0.967 | | 4時間足 | 2024 | 12.05 | 24.10 | 23.58 | 0.978 | **1取引あたりの上乗せは、実勢との差のおよそ2倍**（実測で0.86〜0.98倍）になります。このシリーズでは「コスト＝取引回数×スプレッド」を20本で確かめてきましたが、これはその逆向きの版です。**取引回数×（実勢との差の約2倍）が、そのまま架空の利益になります。** だから15分足の色の変化がいちばん派手になります。年6,107回売買して、1回あたり5.67pipsを拾うからです。 ## 本当にノイズは減るのか 減ります。平均足が売られている理由は「ノイズが減ってトレンドに乗り続けられる」です。ここは測ると本当でした。 | 時間足 | 年 | 平均足の連続本数 | ふつうの足 | 色が変わる回数（平均足 / ふつう） | |---|---|---|---|---| | 15分足 | 2025 | 4.08 | 1.98 | 6,107 / 12,576 | | 1時間足 | 2025 | 4.02 | 1.95 | 1,548 / 3,185 | | 4時間足 | 2025 | 4.20 | 2.00 | 383 / 806 | **平均の連続本数はおよそ2倍、色が変わる回数はおよそ半分。** 6通り（時間足3種×2年）すべてで同じ方向でした。1本で色が変わってしまう割合は、ふつうの足の49.2〜50.7%に対して平均足は20.8〜24.3%です。 **平滑化が本物であることと、平均足で利益が出ることは別**です。実際の売買結果を測ります。 ### ヒゲの無い足は、半分の足で起きる 「陽線で下ヒゲが無ければ強いトレンド」という読み方があります。下ヒゲのない陽線が出る頻度を数えました。 | 時間足 | ヒゲの無い足の割合 | |---|---| | 15分足 | 46.9〜47.3% | | 1時間足 | 46.2〜46.8% | | 4時間足 | 47.7〜49.4% | **ほぼ2本に1本です。** 年の半分で成立する条件は、何も絞り込みません。GMMAの「長期リボンの整列が85%の足で成立していた」と同じ形で、これで2件目になります。 同時線（実体が小さい足）は、実体比率0.1で10.5〜12.5%でした。こちらは十分に珍しい条件です。 ## 色が変わったら順張りで合っているか 1時間足は4つの読み方とも2025年が黒字で、15分足は12通りすべて赤字でした。標準の平均足でシグナルを作り、**実際の足で約定**させた結果です。 | 時間足 | 年 | 色の変化 | N本連続(3) | ヒゲの無い足 | 同時線(0.1) | |---|---|---|---|---|---| | 15分足 | 2025 | −2,355.2 | −1,602.5 | −535.2 | −2,905.4 | | 15分足 | 2024 | −2,200.9 | −1,308.4 | −1,260.5 | −3,417.0 | | 1時間足 | 2025 | **+1,153.3** | **+1,094.1** | **+912.8** | **+788.6** | | 1時間足 | 2024 | **+101.3** | **+2,194.2** | **+3,895.2** | −1,548.6 | | 4時間足 | 2025 | **+796.1** | −834.3 | −566.5 | −1,244.3 | | 4時間足 | 2024 | **+3,051.6** | −302.1 | **+2,972.9** | **+682.9** | 取引数は1時間足2025年で、色の変化1,548回・連続(3)608回・ヒゲ701回・同時線434回です。 **15分足は12通り中12通りが赤字**でした。読み方を変えても救えません。理由は取引回数で、色の変化は年6,107回売買します。スプレッド0.3pipsだけで1,832pipsを払うので、**スプレッド0でも−523.1pips**の系にそれが乗ります。 同時線だけ勝率が高く出ます（1時間足2025年で52.76%、他は35〜37%）。**逆張り的な読み方は勝率が高く、損益では勝てない。** このシリーズで21本続けて同じ形です。 ## どの時間足で機能したか 1時間足だけです。 | 読み方 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---| | 色の変化 | 1/8 | **7/8** | 5/8 | | N本連続 | 11/40 | **32/40** | 7/40 | | ヒゲの無い足 | 3/8 | 3/8 | 0/8 | | 同時線 | 1/24 | 8/24 | 11/24 | | **80通り全体** | **16/80** | **50/80** | **23/80** | 2025年の中央値は15分足−509.3pips、1時間足+850.7pips、4時間足+5.3pips。 **平均足がまともに機能したのは1時間足だけ**でした。80通り中63通りが2025年に黒字、50通りが両方の年で黒字です。このシリーズで、ひとつの時間足がこれだけ揃って残った例は多くありません。 ## 何本続いたら入るのがいいか 1時間足では待つほど残り、4時間足では3本以上待つと残りませんでした。「同じ色がN本続いたら入る」のNを1から5まで動かします。**N＝1は色の変化そのもの**なので、はしごの端が既知のものに着地します（スイープで完全一致を確認しています）。 | 時間足 | 年 | 1本 | 2本 | 3本 | 4本 | 5本 | |---|---|---|---|---|---|---| | 15分足 | 2025 | −2,355.2 | −1,090.9 | −1,602.5 | −401.9 | +85.7 | | 15分足 | 2024 | −2,200.9 | −1,729.8 | −1,308.4 | −774.6 | −1,851.0 | | 1時間足 | 2025 | +1,153.3 | **+2,415.4** | +1,094.1 | +1,375.8 | +442.4 | | 1時間足 | 2024 | +101.3 | +2,545.3 | +2,194.2 | +2,227.4 | **+3,392.5** | | 4時間足 | 2025 | +796.1 | **+1,478.1** | −834.3 | −993.7 | −496.5 | | 4時間足 | 2024 | **+3,051.6** | +1,379.9 | −302.1 | −1,328.7 | −1,227.3 | **単調ではありません。** 1時間足では待つほど勝率が上がり（38.5%→42.5%）取引数が減りますが（824→336）、損益はそう動きません。両方の年で黒字だった数だけは1時間足できれいに出ます。1本7/8、5本**8/8**です。 4時間足では、3本以上待った8設定はすべて両年黒字ではありませんでした。1時間足では5本待つ8設定がすべて両年黒字だったため、待機本数の効果は時間足によって逆になりました。 ## 計算方式はどれを選ぶか 平滑化するほど標準とは別物になります。標準・ベター（2本平均してから平均足）・平滑化・二重平滑化の4種類で、足の色と実際の終値との差がどう変わるかを比べます。 | 方式 | 色が変わる回数（1時間足2025年） | 平均連続本数 | 標準と色が違う足 | |---|---|---|---| | 標準 | 1,548 | 4.02 | 0% | | ベター | 1,232 | 5.05 | 12.3% | | 平滑化(5) | 718 | 8.67 | 27.9% | | 二重平滑化(5) | 557 | 11.18 | 39.4% | **平滑化するほど連続本数が伸び、標準とは別物になります。** 二重平滑化は4割の足で色が違います。 成績（色の変化とN本連続の平均、1時間足）： | 方式 | 2025年 | 2024年 | |---|---|---| | 標準 | +1,272 | +1,760 | | ベター | +1,127 | +2,021 | | 平滑化(3) | **+1,543** | **+2,506** | | 平滑化(10) | +267 | +3,200 | | 二重平滑化(10) | +282 | +1,406 | **平滑化(3)は2025年+1,543pips、2024年+2,506pips**でした。一方、平滑化(10)は2025年+267pips、2024年+3,200pipsです。15分足では二重平滑化(10)だけが両年黒字（+1,706pips、+1,697pips）で、取引数は1,198回まで減りました。 ## 去年いちばんの設定は今年どうなるか 3通りのうち、当てた年の中央値を上回ったのは4時間足だけでした。 | 時間足 | 読み方 | 計算方式 | 2024年の損益 | 2025年の損益 | 2025年の中央値 | |---|---|---|---:|---:|---:| | 15分足 | ヒゲの無い足 | 平滑化(5) | +2,921.9pips | −2,445.0pips | −509.3pips | | 1時間足 | N本連続(2) | 平滑化(10) | +4,030.9pips | −209.5pips | +850.7pips | | 4時間足 | 同時線(0.2) | 平滑化(5) | +4,237.9pips | +437.8pips | +5.3pips | 中央値を上回ったのは4時間足だけです。1時間足は80通り中50通りが両年黒字でしたが、2024年に+4,030.9pipsだった設定は2025年に−209.5pipsでした。 ## 平均足でなければならないのか 平均足の色は、（4本値の平均）とその3期間EMAの比較そのものでした。だとすれば、**ふつうのローソク足で同じクロスを回したら何が違うのか**を測る必要があります。 対照はEMA(1)とEMA(3)のクロス（終値）です。形は同じ、変換だけ無し。時間足3種×2年×計算方式8種の48通りで突き合わせました。 | | 平均足の色の変化 | 終値のEMA1/3クロス | |---|---|---| | 取引数（1時間足・2025年・標準） | 1,548 | 2,082 | | 勝率 | 35.66% | 32.37% | | 損益 | +1,153.3 | +476.0 | 48通り全体では、 - **取引数が少ないのは48通り中48通り** - **勝率が高いのは48通り中46通り** - **損益で上回ったのは48通り中31通り** **変換は取引数と勝率では確実に効き、損益でも半分より上に出ました。** CCIの典型価格（25/36）とサイコロジカルラインの本数（16/36）に続いて3件目の比較ですが、**31/48＝64.6%はこの3つの中で最も良い数字**です。 ただし内訳を見ると時間足で割れます。15分足は16通り中16通り、1時間足は16通り中13通りで変換のほうが上ですが、**4時間足は16通り中2通り**しかありません。4時間足では平均足にする意味が見つかりませんでした。 なお15分足の16/16は「負けが小さい」という意味です。対照のEMA1/3クロスが2025年に−4,981.6pipsで、平均足の色の変化が−2,355.2pips。どちらも赤字です。 ## フィルター・損切り利確・コスト ### ADXフィルターは、また壊した | 読み方 | 時間足 | 年 | ADX条件 | 取引数 | 損益 | |---|---|---:|---|---:|---:| | 色の変化 | 1時間足 | 2025 | なし | 1,548回 | +1,153.3pips | | 色の変化 | 1時間足 | 2025 | 20以上 | 1,001回 | +838.2pips | | 色の変化 | 1時間足 | 2025 | 25以上 | 705回 | +422.0pips | | 色の変化 | 1時間足 | 2025 | 30以上 | 468回 | +89.4pips | | 色の変化 | 1時間足 | 2024 | なし | 1,525回 | +101.3pips | | 色の変化 | 1時間足 | 2024 | 20以上 | 976回 | −1,437.0pips | | 色の変化 | 1時間足 | 2024 | 25以上 | 671回 | −1,043.0pips | | 色の変化 | 1時間足 | 2024 | 30以上 | 449回 | −1,213.9pips | **2025年は単調に削れ、2024年は赤字に落ちます。** トレンド系の読み方にトレンドの強さのフィルターを足すと壊れる、はこれで4件目です（GMMA・CCI・サイコロジカルラインに続く）。 ### 時間帯 東京時間（UTC 0〜8）は年で逆を向きました。1時間足の色の変化は2025年が+1,153.3 → +646.5と悪化、2024年が+101.3 → **+1,558.7**と改善です。ロンドン＋ニューヨーク時間は両方の年で悪化しました（+1,153.3 → +257.4、+101.3 → −700.2）。 ### 損切り・利確 1時間足の色の変化に損切り100・利確200を足すと2025年は+1,050.2（ベース+1,153.3）、2024年は+20.4（ベース+101.3）。**どれも改善と呼べる幅ではありません。** 損切り30・利確60は2025年を−191.1まで落とします。 ### コスト 損益はスプレッドに対してきれいに直線で、**失ったpips＝取引回数×スプレッド**が5つの設定すべてで誤差なく成立しました。 | 設定 | 取引数 | スプレッド0 | 0.3 | 1.0 | 損益分岐 | |---|---|---|---|---|---| | 色の変化・15分足 | 6,107 | −523.1 | −2,355.2 | −6,630.1 | 成立せず | | 色の変化・1時間足 | 1,548 | +1,617.7 | +1,153.3 | +69.7 | 1.05pips | | N本連続(3)・1時間足 | 608 | +1,276.5 | +1,094.1 | +668.5 | **2.10pips** | | 同時線(0.1)・1時間足 | 434 | +918.8 | +788.6 | +484.8 | **2.12pips** | **取引数を減らす読み方ほど損益分岐が高くなります。** 15分足の色の変化はスプレッド0でも赤字なので、コストの問題ではありません。 ## 関連記事 - [サイコロジカルラインの75%](/ja/blog/psychological-line-settings)：同じ「その指標である必要があるのか」を、値幅を捨てた指標で測っています - [CCIの±100](/ja/blog/cci-settings)：ゼロラインが移動平均線クロスそのものだった回です - [移動平均線クロスの期間](/ja/blog/moving-average-cross-settings)：この記事の対照群が行き着く先です - [GMMAの本数](/ja/blog/gmma-settings)：「年の85%で成立する条件は何も絞らない」の初出です - [練行足の目盛り](/ja/blog/renko-chart-settings)：足そのものを書き換える指標の2本目です。あちらの終値は実在した価格ですが、確定した時点では通り過ぎています ## 補足 - **通貨ペアはドル円だけ、期間は2年だけ**です。平均足の終値と実勢の差は値動きの大きさに比例するので、対象が変われば架空の利益の大きさも変わります - **約定価格の実験はスプレッド0で行いました。** 実際の環境ではどちらの計算にもスプレッドが乗りますが、差そのものは変わりません - **スプレッドは0.3pips固定**です。1時間足の色の変化は1.05pipsで損益分岐するので、この前提が崩れると成立しません - **決済は反対シグナルまで**です。損切りと利確を足した組み合わせは、どれも改善と呼べる幅になりませんでした - **MT4への書き出しは標準の平均足だけ**です。平滑化系の3方式は各値に移動平均が要るので、書き出しでは標準に落として、そのことをコードに明記しています - **同時線の実体比率と連続本数は、それぞれ3通り・5通りしか振っていません。** 計算方式8種との組み合わせで960通りですが、1つの読み方あたりの解像度は高くありません --- ### サイコロジカルラインの75%逆張りは勝てる？5,760通りで検証 URL: https://formiq.jp/ja/blog/psychological-line-settings Language: ja Published: 2026-08-28 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: サイコロジカルライン, オシレーター, 逆張り, 順張り, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/psychological-line-settings サイコロジカルラインは、**直近N本のうち何本が上がったかを数えて、割合にしただけ**のオシレーターです。定番は12本で、75%を超えたら買われすぎ、25%を下回ったら売られすぎ、と紹介されます。 他のオシレーターとの違いは、平均ではなく**数**であることです。50pips上げた足も0.1pips上げた足も、同じ1本として数えます。値幅を完全に捨てる指標は、このシリーズで19本目にして初めてでした。 そこから2つの性質が出てきます。ひとつは、**読み取り値がN+1個の段しか取らない**こと。もうひとつは、**その段の位置が水準の打ち込み方とずれる**ことです。 期間12で「75より上」と指定すると、判定されるのは12本中10本、つまり83.3%です。**打ち込んだ75は、どの期間でも75%になりませんでした。** ## サイコロジカルラインは何を数えているのか 計算は1行です。 **サイコロジカルライン ＝（直近N本のうち上がった本数 ÷ N）× 100** 平均も、標準偏差も、平滑化も入りません。だからこそ、他のオシレーターには無い癖が2つ出ます。 ### 打ち込んだ水準は、そのままの水準にならない 期間12のサイコロジカルラインが取れる値は13個だけです。 0 / 8.3 / 16.7 / 25.0 / 33.3 / 41.7 / **50.0** / 58.3 / 66.7 / **75.0** / 83.3 / 91.7 / 100 75.0はこの中にあります。ところが「75より上」という条件は、75.0そのものを含みません。ひとつ上の段、12本中10本＝83.3%が最初に該当する値になります。 期間を変えると、着地する段も変わります。 | 期間 | 取りうる値の数 | 「75」が要求する本数 | 実際の水準 | |---|---|---|---| | 6 | 7 | 6本中5本 | 83.3% | | 10 | 11 | 10本中8本 | 80.0% | | 12 | 13 | 12本中10本 | 83.3% | | 14 | 15 | 14本中11本 | 78.6% | | 20 | 21 | 20本中16本 | 80.0% | | 26 | 27 | 26本中20本 | 76.9% | **6通りのどれでも75%になりません。** 冒頭の図がこれを描いたものです。灰色の目盛りがその期間で取りうる値、青がそこに「75」が落ちる段です。 もっと厄介なのは、**動かしても何も起きない範囲がある**ことです。期間12では、水準を75にしても76にしても83にしても、判定は12本中10本のまま変わりません。ところが74に下げると、12本中9本＝75.0%に切り替わります。**9動かしても同じで、1動かすと変わる**ということです。 | 期間 | 同じルールになる水準の幅（75付近） | その中身 | |---|---|---| | 6 | 67〜83 | 6本中5本 = 83.3% | | 12 | 75〜83 | 12本中10本 = 83.3% | | 14 | 72〜78 | 14本中11本 = 78.6% | | 26 | 74〜76 | 26本中20本 = 76.9% | 期間26では74と75が同じで、70は違います。**期間12とは食い違う組み合わせが同じになります。** ### スイープの中に、まったく同じ行が並ぶ これは理屈だけの話ではありません。15分足・2025年・期間12の逆張り買いを、5つの入口水準×3つの出口水準で回した結果です。 | 入口 → 出口 | 取引数 | 損益 | |---|---|---| | 70 → 40 | 363 | +142.0 | | 74 → 40 | 363 | +142.0 | | 70 → 50 | 314 | +697.7 | | 74 → 50 | 314 | +697.7 | | 75 → 50 | 119 | +250.6 | | 80 → 50 | 119 | +250.6 | | 90 → 50 | 22 | +207.7 | **15通り打ち込んで、違う結果は9通りしか出ません。** 検証全体でも同じことが起きていて、5,760通りのうち違う結果になったのは3,548通りでした。**打ち込んだ組み合わせの38.4%が、別の行の複製です。** ### 期間が目盛りの細かさを決める 段の数は期間+1なので、期間を短くすると線は粗くなります。期間6のサイコロジカルラインは7段しかなく、0から100まで101通りの水準を打ち込んでも、区別できるルールは7通りです。期間26なら27通りになります。 実際に現れた段はもっと少なくなります。2025年のドル円・期間12で数えました。 | 時間足 | 現れた段 | 一度も出なかった読み取り値 | |---|---|---| | 15分足 | 13段中12段 | 0本（12本すべて陰線） | | 1時間足 | 13段中12段 | 12本（12本すべて陽線） | | 4時間足 | 13段中10段 | 0本・1本・12本 | そして**ちょうど50に座っている足が、どの時間足でも2割**あります（15分足22.65〜23.71%、1時間足21.93〜22.14%、4時間足21.59〜21.76%）。線が真ん中に張り付いている時間が、それだけあるということです。 これは中央線の使い方に直接効きます。水準50の中央線クロスと水準55の中央線クロスは、**上抜けについてはまったく同じルールですが、下抜けが違います**。どちらも「12本中7本以上に増えたら上抜け」ですが、下抜けは50なら「6本以下に減ったとき」、55なら「7本以下に減ったとき」だからです。1時間足2025年で上抜けはどちらも365回、下抜けは50が345回、55が365回でした。 売買量にも出ます。1時間足の平均取引数は水準50で378回、55で703回、45で622回です。**水準を5動かすと売買が倍近くになります。** ### 「上がった日」の定義が2つある 前の足の終値より高く引けた足を数えるか、陽線（自分の始値より高く引けた足）を数えるか。窓が開けば別の足になります。 | 時間足 | 上下が食い違う足 | 2本の線の相関 | 75の内外が入れ替わる足 | 50の上下が入れ替わる足 | |---|---|---|---|---| | 15分足 | 1.24% | 0.976 | 0.32% | 2.79% | | 1時間足 | 1.17% | 0.977 | 0.35% | 2.75% | | 4時間足 | 0.81% | 0.984 | 0.38% | 2.19% | **別の系列ではあります。** ただし成績では選べませんでした。両方の数え方で回した2,134通りのうち、陽線で数えたほうが上回ったのは1,006通り、47.1%です。コインを投げたのと変わりません。 ## チャートに表示する方法 もともと日本の株式市場で使われてきた指標なので、海外製プラットフォームの初期搭載リストには入っていないことが多いです。 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準オシレーターには入っていない。カスタムインジケーター（.mq4 / .mq5）を Indicators フォルダに置いて読み込む | | TradingView | 標準のインジケーター一覧には無い。コミュニティスクリプトを検索して追加する | | ブラウザ（Formiq） | 最初から入っている。期間・上下2本の水準線・上昇の数え方を設定でき、バックテスト側では読み方4種を条件にできる | MT4向けに書き出す場合、`iCCI` のような組み込み関数が無いので、**直近N本を数えるループを自分で書くことになります**。この検証で使ったエクスポートもそうしています。 ## 検証の条件 数字を見る前に、何をどう測ったかを書いておきます。この表と同じ設定を入れれば、同じ結果が出ます。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 読み方 | 逆張り（水準の外側で反対方向）／ 水準の抜け（抜けた方向）／ 中央線（水準を上下に抜ける） | | 期間 | 6・10・12・14・20・26 | | 水準 | 逆張り：入口70・74・75・80・90 × 出口40・50・60／ 抜け：60・70・75・80・90／ 中央線：40・45・50・55・60 | | 上昇の数え方 | 前の足の終値より上 / 陽線 の2通り | | 組み合わせ | 1区分あたり240通り × 数え方2種。時間足3種 × 期間4区分で合計5,760通り | | 決済 | 反対シグナルまで。逆張りだけは出口の水準を別に指定する | | 損切り・利確 | 使わない（別セクションで測定） | | スプレッド | 0.3pips固定。約定は終値 | | ロット | 0.1ロット | 逆張りは買いと売りのどちらか一方しか作りません（「25より下は買い」は買いしか作らない）ので、**買い逆張りと売り逆張りを別々の売買ルールとして回しました。** 抜けと中央線は対称なので、エントリーと決済に同じルールセットを使います。水準は上下1組で扱い、75と指定すれば25も同時に引かれます。 ## 75%を超えるのは珍しいか 珍しくありません。上昇本数の分布はコイン投げとほとんど同じでした。サイコロジカルラインの前提は「上げが続くと群衆が強気に傾きすぎる」です。だとすれば、上げた足の後は上げやすい（または下げやすい）はずで、**上昇本数の分布はコイン投げからずれるはず**です。 実際の上昇本数と、同じ上昇率のコインを12回投げた場合の本数を比べました。 13段のそれぞれに何%の足がいたか、そして同じ上昇率のコインを12回投げたらどうなるかです。**2つのはしごの差は、合計で0.97ポイント**しかありませんでした（15分足2025年）。1時間足2025年で2.19ポイント、15分足2024年で2.79ポイントです。 水準に届く頻度も、ほぼ理論値どおりでした。 | 時間足・年 | 足の上昇率 | 75より上（実測 / コイン） | 75以上（実測 / コイン） | 75〜25の内側 | |---|---|---|---|---| | 15分足 2025 | 50.22% | 1.76% / 2.00% | 7.02% / 7.51% | 85.97% | | 1時間足 2025 | 50.39% | 1.74% / 2.06% | 6.98% / 7.69% | 86.24% | | 4時間足 2025 | 50.78% | 2.32% / 2.19% | 8.39% / 8.08% | 86.67% | | 4時間足 2024 | 53.50% | 3.30% / 3.37% | 13.09% / 11.31% | 80.11% | ここで**「より上」と「以上」の差**を見てください。同じ75で、1.76%と7.02%です。**不等号の書き方だけで頻度が4倍になります。** 教科書は「75%以上」と書き、多くのツールの条件式は「75より上」です。期間12ではその差がちょうど1本ぶんになります。 ### 上げた足の後は、上げやすいのか 分布が一致するなら、足の向きに記憶が無いということです。直接測りました。 | 時間足・年 | 上げた足の次が上げる確率 | 下げた足の次が上げる確率 | 1本前との相関 | |---|---|---|---| | 15分足 2025 | 49.59% | 50.85% | −0.0126 | | 15分足 2024 | 49.76% | 53.11% | −0.0335 | | 1時間足 2025 | 49.17% | 51.62% | −0.0245 | | 1時間足 2024 | 51.42% | 51.95% | −0.0053 | | 4時間足 2025 | 50.67% | 50.95% | −0.0027 | | 4時間足 2024 | 54.75% | 52.00% | **+0.0274** | **6通り中5通りでマイナス**です。上げた足の次は、わずかに上げにくい。サイコロジカルラインの前提が言っていることと向きは同じですが、**効果は1〜3ポイント**で、分布をコインから引き離すには足りません。 つまり冒頭の期待値はこうなります。**この指標が拾えるものがあるとすれば、それは「行き過ぎたら戻る」側であって、しかも極めて薄い。** ## 逆張りと順張りのどちらで使うか このシリーズで初めて、逆張りが順張りに勝ちました。同じ期間12・水準75/25を、3通りに読ませます。逆張りは買い側と売り側の合計です。 | 時間足 | 年 | 逆張り（買い / 売り） | 水準の抜け | 中央線 | |---|---|---|---|---| | 15分足 | 2025 | **+250.6** / −265.6 | −2,428.2 | −1,099.2 | | 15分足 | 2024 | **+322.6** / **+644.9** | −97.2 | **+238.3** | | 1時間足 | 2025 | **+82.5** / **+228.3** | −903.2 | **+272.6** | | 1時間足 | 2024 | **+273.6** / −376.1 | **+725.8** | **+1,169.5** | | 4時間足 | 2025 | **+566.9** / **+316.6** | −3,263.3 | **+65.6** | | 4時間足 | 2024 | **+33.2** / −1,091.6 | **+2,663.8** | **+886.3** | 取引数を添えます。1時間足2025年で、逆張りは買い26回・売り28回、抜けは32回、中央線は402回です。**逆張りの取引数は年に数十回しかありません。** **このシリーズで初めて、逆張りが順張りに勝ちました。** 12通り中、逆張りは買いが6/6、売りが3/6で黒字。水準の抜けは6通り中2通りです。18本の記事で「逆張りは勝率が高く損益はマイナス」を繰り返してきたので、ここは詳しく見る必要があります。 ## 逆張りが勝ったのは指標のおかげか 「行き過ぎたら戻る」は、戻る相場でしか成立しませんでした。逆張りの買いは、下げが続いた後に買います。上昇する年ならそれだけで儲かります。**買いが儲かったのが指標のおかげなのか、相場の向きのおかげなのかを分ける**必要があります。 そこで、各取引の損益から「同じ本数だけ何もせず持っていたら得られた分」を引きました。年の値幅を足数で割ったものが1本あたりの流れなので、それに保有本数を掛ければ出ます。残りが指標の取り分です。 | 時間足 | 年 | 1取引あたり | うち相場の流れ | うち指標の取り分 | 取り分がプラスの設定 | |---|---|---|---|---|---| | 15分足 | 2025 | +4.77 | −0.06 | **+4.83** | 81通り中77 | | 15分足 | 2024 | +2.00 | +1.64 | +0.36 | 72通り中45 | | 1時間足 | 2025 | +8.07 | −0.23 | **+8.31** | 60通り中33 | | 1時間足 | 2024 | +9.62 | +6.29 | +3.34 | 60通り中25 | | 4時間足 | 2025 | +47.40 | −0.66 | **+48.07** | 51通り中49 | | 4時間足 | 2024 | −21.06 | +25.34 | **−46.40** | 57通り中8 | 逆張り買い、取引が5回以上あった設定の平均です。**2025年は流れがほぼゼロなので、出た数字はすべて指標の取り分です。** 15分足では81通り中77通りでプラスでした。 **2024年になると消えます。** 15分足で+0.36、4時間足では−46.40まで落ち、57通り中49通りがマイナスになります。2024年のドル円は年間+1,632pips上げた年で、2025年は−56pipsの横ばいです。 **「行き過ぎたら戻る」は、戻る相場でしか成立しませんでした。** 1本前の値動きとの相関が+0.0274と6区分中で最も高かった4時間足・2024年では、買い逆張りから相場上昇分を除いた平均損益が1取引あたり−46.40pipsとなり、プラスだった設定は57通り中8通りだけでした。 ## 2年とも勝てたのはどれか 15分足の逆張り買いが60/90で最多でした。片方の年だけの結果は選び方の問題になるので、2024年と2025年の両方で黒字だった設定を数えます。 | 読み方 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---| | 逆張り・買い | **60/90** | 32/90 | 23/90 | | 逆張り・売り | 24/90 | 12/90 | 5/90 | | 水準の抜け | 5/30 | 8/30 | 5/30 | | 中央線 | 1/30 | **18/30** | 14/30 | **15分足の逆張り買いが60/90で最も残りました。** 逆張りの売りは270通り中41通りしかありません。買いと売りでこれだけ差が付く形は、上昇年を含む2年ぶんしか見ていないので、**相場の向きの影響を切り離せていません**。 **中央線は1時間足でだけ残ります。** 15分足では30通り中1通りです。理由は取引数で、15分足の中央線は年に729〜3,968回売買します。 ## 期間と水準はどう選ぶか 短い期間ほど残り、中央線を50から動かす理由は見つかりませんでした。読み方ごとに分けて出します。 **逆張り・買い・15分足（期間別）** | 期間 | 2025年の平均 | 平均取引数 | 黒字 2025 | 両方の年で黒字 | |---|---|---|---|---| | 6 | +355 | 581 | 15/15 | **15/15** | | 10 | +366 | 185 | 15/15 | 11/15 | | 12 | +248 | 178 | 13/15 | 10/15 | | 14 | +371 | 130 | 14/15 | 8/15 | | 20 | +174 | 39 | 11/15 | 11/15 | | 26 | +191 | 18 | 12/15 | 5/15 | **期間6は15通りすべてが両方の年で黒字**でした。期間を伸ばすほど取引数が落ち、年18回まで減ると2024年に残らなくなります。 **中央線・1時間足（水準別）** | 水準 | 2025年の平均 | 2024年の平均 | 平均取引数 | 両方の年で黒字 | |---|---|---|---|---| | 40 | +272 | +1,743 | 464 | **5/6** | | 45 | +650 | +609 | 622 | 3/6 | | 50 | +411 | +1,419 | 378 | **5/6** | | 55 | +384 | +920 | 703 | 4/6 | | 60 | +54 | +14 | 483 | 1/6 | 取引数が水準の順に並んでいないことに注目してください。**50が378回で最少、55が703回で最多**です。50はちょうど目盛りの上にあり、そこに座っている足が2割あるので、抜けたと判定される回数が減ります。**中央線を50から動かす理由は見つかりませんでしたが、動かすと売買量が倍近く変わります。** ## 去年よかった条件は今年も通用するか 5通り中2通りが、当てた年の中央値を上回りました。このシリーズの中心にある検定です。 | 読み方 | 時間足 | 期間 | 水準 | 2024年の損益 | 2025年の損益 | 2025年の中央値 | |---|---|---:|---|---:|---:|---:| | 中央線 | 1時間足 | 12 | 40 | +3,191.5pips | +102.0pips | +234.4pips | | 中央線 | 15分足 | 26 | 45 | +2,486.6pips | −1,449.8pips | −1,081.4pips | | 中央線 | 4時間足 | 6 | 40 | +3,341.4pips | +158.5pips | −211.8pips | | 水準の抜け | 1時間足 | 10 | 60・40 | +2,557.1pips | −1,292.4pips | −465.0pips | | 逆張り・買い | 15分足 | 6 | 70→50 | +1,114.2pips | +252.2pips | +250.6pips | **5通り中2通りが2025年の中央値を上回りました。** 15分足の逆張り買いは+252.2pipsで、中央値+250.6pipsとの差は1.6pipsです。1時間足の中央線は+102.0pipsに対して中央値+234.4pips、15分足の中央線は−1,449.8pipsに対して−1,081.4pipsでした。 ## 本数を数える意味はあるのか サイコロジカルラインは「直近N本のうち何本上げたか」を見ます。**同じ範囲について「結局いくら上げたか」を見る指標がモメンタム**で、N本前の終値との差を測ります。 2つは同じ質問に別の答えを出します。実測では、**23.2〜25.9%の足で向きが食い違いました**（期間12・26、両年、15分足と1時間足）。上げた本数が過半数なのに価格はN本前より安い、という足が、その中の14.8〜19.7%です。相関は0.58〜0.69でした。 では、数えたほうが良いのか。中央線クロス（本数）と、同じ期間のモメンタムの符号（値幅）を、時間足3種×2年×期間6種の36通りで突き合わせました。 | | 本数を数える | 値幅を測る | |---|---|---| | 取引数（1時間足・2025年・期間12） | 402 | 795 | | 勝率 | 38.06% | 34.97% | | 損益 | +272.6 | +1,535.3 | 36通り全部で見るとこうなりました。 - **取引数が少ないのは36通り中36通り** - **勝率が高いのは36通り中34通り** - **損益で上回ったのは36通り中16通り** **数えると売買が半分になり、勝率は上がり、損益は五分五分を下回ります。** CCIの検証では同じ形の比較で損益25/36だったので、そこにも届きません。 値幅を捨てることには効果があります。小さな上げ下げを1票ずつしか数えないので、ノイズで反転する回数が減る。**減らした結果が良いかどうかは、この検証では言えませんでした。** ## フィルター・損切り利確・コスト ### ADXフィルター | 読み方 | 時間足 | 年 | ADX条件 | 取引数 | 損益 | |---|---|---:|---|---:|---:| | 中央線 | 1時間足 | 2025 | なし | 402回 | +272.6pips | | 中央線 | 1時間足 | 2025 | 20以上 | 277回 | −99.5pips | | 中央線 | 1時間足 | 2025 | 25以上 | 196回 | −906.0pips | | 中央線 | 1時間足 | 2025 | 30以上 | 144回 | −726.8pips | | 中央線 | 15分足 | 2024 | なし | 1,541回 | +238.3pips | | 中央線 | 15分足 | 2024 | 20以上 | 1,003回 | +559.7pips | | 中央線 | 15分足 | 2024 | 25以上 | 703回 | +274.6pips | | 中央線 | 15分足 | 2024 | 30以上 | 442回 | **+747.6pips** | **方向がそろいませんでした。** 1時間足2025年は下げ、15分足2024年は上げます。どちらの年も同じ向きに動いたフィルターはありませんでした。 ### 時間帯 東京時間（UTC 0〜8）だけで売買させると、中央線は4通り中3通りで改善しました。 | 設定 | ベース | 東京時間のみ | ロンドン＋NY（UTC 7〜21） | |---|---|---|---| | 中央線・15分足・2025 | −1,099.2 | **+544.5** | −808.8 | | 中央線・15分足・2024 | +238.3 | **+649.8** | +218.8 | | 中央線・1時間足・2025 | +272.6 | **+505.4** | −100.2 | | 中央線・1時間足・2024 | +1,169.5 | +213.3 | +846.6 | **唯一悪化したのが、いちばん儲かっていた区分**です。ロンドン＋ニューヨーク時間は4通り中1通りしか改善しませんでした。東京時間が効いたのは「取引数が減ったから」でもあり、15分足2025年は1,551回から605回に落ちています。 ### 損切り・利確 中央線・1時間足で、損切り30・利確90を足すと2025年は+272.6が+946.4に、2024年は+1,169.5が+1,971.2になります。**両方の年で改善した数少ない変更**です。15分足では2025年が−1,099.2から−339.8まで戻るものの、赤字のままでした。 ### コスト 損益はスプレッドに対してきれいに直線でした。**失ったpips＝取引回数×スプレッド**が、5つの設定すべてで誤差なく成立します。 | 設定 | 取引数 | スプレッド0 | 0.3 | 1.0 | 損益分岐 | |---|---|---|---|---|---| | 中央線・1時間足 | 402 | +393.2 | +272.6 | −8.8 | **0.98pips** | | 中央線・15分足 | 1,551 | −633.9 | −1,099.2 | −2,184.9 | 成立せず | | 逆張り売り・1時間足 | 28 | +236.7 | +228.3 | +208.7 | 8.45pips | | 水準の抜け・15分足 | 147 | −2,384.1 | −2,428.2 | −2,531.1 | 成立せず | **1時間足の中央線は0.98pipsで損益分岐**します。スプレッドが1pips開く環境では成立しません。逆張り売りの8.45pipsは大きく見えますが、年28回しか売買しない設定の数字です。 なお**15分足の中央線と水準の抜けは、スプレッドを0にしても赤字**です。取引回数のせいではありません。 ### 一度も発動しない設定 水準を上げすぎると、線が届かなくなります。2025年に5回未満しか売買しなかった設定です。 | 時間足 | 水準の抜け | 逆張り | |---|---|---| | 15分足 | 30通り中2 | 180通り中18 | | 1時間足 | 30通り中9 | 180通り中48 | | 4時間足 | 30通り中11 | 180通り中74 | **4時間足では逆張りの4割が売買にならない**ということです。期間20・水準90は、どの時間足でも一度も発動しませんでした。上で見たとおり水準90は期間20で「20本中19本」を要求するので、当然ではあります。 ## 関連記事 - [CCIの±100](/ja/blog/cci-settings)：「その指標である必要があるのか」を同じ形で測っています。CCIのゼロラインは移動平均線クロスそのものでした - [RSIの期間と水準](/ja/blog/rsi-settings)：RSIが設定水準に到達したら売買する逆張りと、設定水準を抜けた方向へ売買する順張りを比較しています - [ボリンジャーバンドの偏差](/ja/blog/bollinger-band-settings)：バンドを「触れたら逆張り」と「抜けたら順張り」に分けた最初の記事です - [移動平均線クロスの期間](/ja/blog/moving-average-cross-settings)：値幅で測る側の代表です - [平均足の設定](/ja/blog/heikin-ashi-settings)：同じ形の比較で、こちらは変換が損益でも半分より上に出ました - [練行足の目盛り](/ja/blog/renko-chart-settings)：こちらも階段の上しか動かない指標で、階段の置き場所そのものが設定に出ていません - [RCI「−80で買い」は本当に勝てる？2,484通り検証](/ja/blog/rci-settings)：同じく値幅を捨てる指標で、捨てた値幅を対照群に置いています ## 補足 - **通貨ペアはドル円だけ、期間は2年だけ**です。サイコロジカルラインは株式市場で使われてきた指標で、値幅の分布が違う対象では段の出方そのものが変わります - **2024年は年間+1,632pips、2025年は−56pips**でした。逆張りの取り分がプラスだったのは2025年だけで、この結果は値動きの異なる2年間でしか確認していません - **逆張りの取引数が少ない設定が多い**（1時間足で年25回前後、4時間足で年10回前後）。勝率100%の設定が出ていますが5回の話です。180通り中48〜74通りは5回未満で、判定できていません - **スプレッドは0.3pips固定**です。1時間足の中央線は0.98pipsで損益分岐するので、この前提が崩れると成立しません - **決済は反対シグナルまで**です。損切り30・利確90だけは両方の年で改善しましたが、他の組み合わせは片方の年でしか良くなりませんでした - **上昇の数え方の違いは、成績では判定できていません**。2,134通りで47.1%対52.9%です。線としては別物（食い違う足が0.68〜1.24%）なので、どちらを使うかは決めておく必要はあります --- ### 移動平均線はSMAとEMAどっちが勝てる？672通り検証した結果 URL: https://formiq.jp/ja/blog/sma-vs-ema Language: ja Published: 2026-08-28 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: 移動平均線, SMA, EMA, インジケーター, パラメーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/sma-vs-ema 移動平均線を引くとき、単純移動平均（SMA）と指数移動平均（EMA）のどちらを選ぶかで迷う人は多くいます。「EMAのほうが反応が速いから短期向き」「SMAのほうが素直だから長期向き」。どちらもよく見かける説明ですが、実測値が添えられていることはほとんどありません。 そこで、**同じ期間・同じ時間足のまま、種類だけを入れ替えた672通り**をひとつずつバックテストしました。短期8種と長期8種の組み合わせ56通りを、15分足・1時間足・4時間足の3つと、2024年・2025年・2025年前半・2025年後半の4区分で回し、SMAとEMAを1対1で突き合わせています。 **EMAの損益が上回ったのは672通り中318通り、47.32%でした。** 引き分けはありません。12の区分に分けても、EMAが勝った数は56通り中15〜39のあいだに収まり、どの区分でも片方がもう片方を引き離しませんでした。 ただし「どちらでも同じ」という意味ではありません。同じ期間のまま種類を入れ替えると損益は中央値で491.1pips動きます。**大きく動くのに、どちらへ動くかは事前に決まらない**という形が、この検証では最もはっきり出ました。 以下、なぜそうなるのかを線そのものの性質から順に見ていきます。SMAにあってEMAにない性質と、EMAにあってSMAにない性質が1つずつ見つかりました。 ## SMAとEMAは、何が違って何が同じか どちらも「直近の価格をならした線」ですが、ならし方が違います。 - **単純移動平均（SMA）**：直近n本の終値を足してnで割る。n本すべてに同じ重みが掛かり、n本より前は一切見ない - **指数移動平均（EMA）**：新しい足ほど重く、古い足ほど軽く見る。重みは1本古くなるごとに一定の比率で小さくなり、ゼロにはならない 言葉で書くとこれだけですが、実際にどの足がどれだけ効いているかは測れます。全部の終値を同じ値にした系列を作り、1本だけを1円だけ上げて線を計算すると、出てきた差がそのままその足の重みになります。期間20で測った結果が次の図です。 | 期間 | 直近の足の重み（SMA） | 直近の足の重み（EMA） | 比 | 重心（SMA） | 重心（EMA） | EMAが期間内に置く重み | EMAが読む長さ | |---|---|---|---|---|---|---|---| | 10 | 10% | 18.182% | 1.818倍 | 4.5本前 | 4.5本前 | 86.56% | 69本 | | 20 | 5% | 9.524% | 1.905倍 | 9.5本前 | 9.5本前 | 86.49% | 139本 | | 25 | 4% | 7.692% | 1.923倍 | 12本前 | 12本前 | 86.48% | 173本 | | 50 | 2% | 3.922% | 1.961倍 | 24.5本前 | 24.5本前 | 86.47% | 346本 | | 200 | 0.5% | 0.995% | 1.99倍 | 99.5本前 | 99.5本前 | 86.47% | 1,382本 | （「EMAが読む長さ」は、重みが直近の足の100万分の1より大きい足の本数） ここに、この記事のいちばん地味で、いちばん効く事実があります。**重心はどの期間でも完全に一致します。** 期間nのSMAもEMAも、重みの重心は (n−1)÷2 本前です。期間20なら両方とも9.5本前、期間200なら両方とも99.5本前。7つの期間すべてで小数点以下まで一致しました。 つまり、**EMAはSMAより「新しい情報を見ている」わけではありません。** 重みの平均的な古さは同じで、配り方だけが違います。EMAは直近の足を約2倍重く見て、そのかわり期間の外側にも重みを残します。期間20のEMAが20本以内に置いている重みは86.49%で、残りの13.51%は21本前より古い足に散っています。この86%台という数字は期間を変えてもほとんど動きません（86.47〜86.83%）。 「EMAは期間20でも実質もっと長い期間を見ている」という言い方と、「EMAのほうが速い」という言い方は、両方とも同じ重み配分の別々の側面を指しています。片方だけを取り出すと結論が反対になります。 ## EMAは本当に反応が速いのか 測り方によって答えが変わり、短期の組み合わせでは速くなりませんでした。重心が同じなら、速さの違いはどこに出るのでしょうか。3つの測り方で確かめました。 **1つ目は、線が終値からどれだけ離れたところを通るか**です。ドル円の1時間足・2025年で測った平均距離です。 | 期間 | SMAの平均距離 | EMAの平均距離 | 比 | |---|---|---|---| | 5 | 13.81pips | 11.34pips | 0.8214 | | 10 | 22.11pips | 18.63pips | 0.8428 | | 20 | 33.55pips | 28.46pips | 0.8483 | | 25 | 38.08pips | 32.34pips | 0.8493 | | 50 | 55.10pips | 46.68pips | 0.8472 | | 200 | 97.66pips | 85.62pips | 0.8767 | **EMAのほうが価格に近いところを通ります。** 時間足3つ・年2つ・期間7つの42区分すべてで近く、比は0.8126〜0.8937に収まりました。期間や時間足を変えても、だいたい85%前後です。EMAの「速さ」は、価格との距離にも表れています。 **2つ目は、2本の線がクロスする回数**です。同じ期間の組み合わせで、SMA同士のクロスとEMA同士のクロスを数えました。 | 時間足 | 組み合わせ | SMAのクロス回数 | EMAのクロス回数 | |---|---|---|---| | 15分足 | 10 / 20 | 1,402回 | 1,053回 | | 15分足 | 20 / 50 | 524回 | 470回 | | 1時間足 | 10 / 20 | 358回 | 264回 | | 1時間足 | 25 / 75 | 99回 | 75回 | | 4時間足 | 10 / 20 | 87回 | 75回 | | 4時間足 | 50 / 200 | 14回 | 8回 | （ドル円2025年） **速いはずのEMAのほうが、クロスは少なくなります。** 36区分のうちSMAのほうが多かったのが31区分、EMAのほうが多かったのが3区分、同数が2区分でした。理由は次の節で書きます。 **3つ目は、どちらのクロスが先に成立するか**です。同じ向きのクロス同士を、距離の近いペアから順に突き合わせて数えました。 | 時間足 | 組み合わせ | EMAが先だった割合 | 差の中央値 | |---|---|---|---| | 15分足 | 10 / 20 | 51.78% | 1本早い | | 15分足 | 50 / 200 | 66.4% | 3本早い | | 1時間足 | 5 / 20 | 49.02% | 差なし | | 1時間足 | 50 / 200 | 70% | 6.5本早い | | 4時間足 | 12 / 25 | 50.79% | 1本早い | | 4時間足 | 20 / 50 | 76.47% | 4本早い | （ドル円2025年） 36区分の全体では、EMAが先だった割合は**41.36%から76.47%**でした。半分を超えたのは26区分です。長い期間の組み合わせほどEMAが先行しますが、10と20のような短い組み合わせでは、EMAが先になるかSMAが先になるかは、ほぼコイン投げです。 **「EMAのほうが早くサインが出る」は、短期の組み合わせでは成立しませんでした。** 成立するのは50と200のような長い2本のときで、6区分中5区分が中央値3〜7本ぶん早く、残る1区分（4時間足の2025年、EMA側のクロスが8回）は0.5本遅いという結果です。 ## クロス回数が違うのはなぜか 単純移動平均だけが、価格と反対に曲がるからです。SMAの1本あたりの動きは、次の式そのものです。 ``` 今日のSMA − 前日のSMA ＝（今日の終値 − n本前の終値）÷ n ``` これは近似ではなく等式です。15分足・1時間足・4時間足の期間10・20・50・200の12区分で、実際の線と突き合わせたところ、最大のずれは0.0000000000037pips未満でした。丸め誤差だけです。 この式が意味するのは、**SMAが上を向くか下を向くかは、今日の終値と「n本前の終値」の比較で決まる**ということです。n本前の終値は、チャートの上でどこにも表示されていません。窓から抜け落ちる足が、今日の線の向きを半分決めています。 EMAの1本あたりの動きは違います。 ``` 今日のEMA − 前日のEMA ＝ k ×（今日の終値 − 前日のEMA）　　k ＝ 2 ÷（期間 + 1） ``` こちらも等式で、同じ12区分で最大のずれは0.0000000000037pips未満でした。右辺の括弧の中は「今日の終値が線の上にあるか下にあるか」そのものです。つまり、**EMAは終値が線より上にあれば必ず上を向き、下にあれば必ず下を向きます。** この違いは数えられます。「線が下を向いたのに、終値は線より上にあった」という足がどれだけあったかを測りました。 | 時間足 | 期間 | 判定した足 | SMAが価格と反対に曲がった足 | 割合 | EMAで同じことが起きた足 | |---|---|---|---|---|---| | 15分足 | 20 | 24,903本 | 4,007本 | 16.09% | 0本 | | 15分足 | 200 | 24,903本 | 4,488本 | 18.02% | 0本 | | 1時間足 | 20 | 6,226本 | 992本 | 15.93% | 0本 | | 1時間足 | 200 | 6,226本 | 1,296本 | 20.82% | 0本 | | 4時間足 | 20 | 1,610本 | 262本 | 16.27% | 0本 | | 4時間足 | 50 | 1,610本 | 345本 | 21.43% | 0本 | （ドル円2025年） **42区分すべてで、EMAは0本でした。** 定義からそうなるので、他の年でも他の通貨ペアでも0本になります。SMAのほうは10.95%から21.43%で、**おおむね6本に1本**の頻度です。 この「n本前の足が抜けたせいで線が曲がる」動きが、SMA同士のクロスを増やしています。速い線と遅い線が、それぞれ別のタイミングで別の理由で曲がるので、交差する機会が増えるという形です。 ただし、クロスが早いことと、売買で有利になることは別の話です。実際の売買結果を比べます。 なお、線が向きを変える回数そのものはEMAのほうが多くなります。42区分中41区分でEMAのほうが多く、15分足の期間5ではSMAが21.2%、EMAが26.82%の足で向きが変わりました。**線の向きが変わる回数はEMAのほうが多いのに、2本のクロス回数はSMAのほうが多い。** 別々のものを測っているので、両方とも成立します。 ## EMAは表示本数で値が変わるのか 変わります。EMAには、SMAにない不安定さがあります。前日のEMAから今日のEMAを作る計算なので、**いちばん最初の1本をどこから始めたかの影響が、いつまでも残ります。** 始点をずらした2本のEMAの差が、1本進むごとにどれだけ縮むかを測りました。 | 期間 | 1本あたりの縮小率 | （期間−1）÷（期間+1） | 半分になるまで | 10分の1まで | 100分の1まで | SMAが完全一致するまで | |---|---|---|---|---|---|---| | 10 | 0.818182 | 0.818182 | 3.5本 | 12本 | 23本 | 9本 | | 20 | 0.904762 | 0.904762 | 6.9本 | 24本 | 47本 | 19本 | | 50 | 0.960784 | 0.960784 | 17.3本 | 58本 | 116本 | 49本 | | 200 | 0.99005 | 0.99005 | 69.3本 | 231本 | 461本 | 199本 | 縮小率は理論値とぴったり一致します。**そして、ゼロにはなりません。** SMAのほうは期間ぶんの本数が過ぎた時点で完全に一致し、それ以降は差が消えます。 実務的にどれくらいの大きさなのかを、実際のドル円で測りました。2025年1月1日の時点で、その足のEMA(200)が、読み込んだ過去データの長さによってどれだけ違うかです。 | 時間足 | 読み込んだ本数 | EMA(200)の差 | 0.1pips未満になるまで | SMA(200)の差 | |---|---|---|---|---| | 15分足 | 200本 | 7.705pips | 435本 | 0 | | 15分足 | 1,000本 | 0.001pips | すぐ | 0 | | 1時間足 | 200本 | 49.083pips | 620本 | 0 | | 1時間足 | 600本 | 0.380pips | 134本 | 0 | | 1時間足 | 1,000本 | 0.008pips | すぐ | 0 | | 4時間足 | 200本 | 47.173pips | 616本 | 0 | | 4時間足 | 1,000本 | 0.011pips | すぐ | 0 | **1時間足でEMA(200)を、ちょうど200本ぶんのデータから描くと、十分な履歴から描いた線と49.083pips違います。** 一致に近づくまで620本、つまり1か月ぶん以上かかります。SMA(200)は最初の1本目から完全に一致しています。 実用上の意味は2つあります。ひとつは、**長い期間のEMAを使うなら、期間の3倍から5倍の履歴を読み込んでおく**こと。1時間足の200なら1,000本で差は0.008pipsまで落ちます。もうひとつは、**別のツールで同じEMAを引いて数字が合わなくても、必ずしもどちらかが間違っているわけではない**ということです。始点が違えば線も違います。 この性質はEMAだけのものではなく、前の値から次の値を作る計算をしている指標に共通します。[平均足](/ja/blog/heikin-ashi-settings)の始値も同じ形で、こちらは1本ごとに差がちょうど半分になります。 ## チャートに表示する方法 移動平均はどのプラットフォームにも標準で入っています。違うのは、選べる計算方法の種類と、初期設定です。 | 環境 | 手順 | 選べる種類 | |---|---|---| | MT4 / MT5 | ナビゲータの指標一覧から Moving Average をチャートにドラッグする | 単純・指数・平滑・線形加重の4つ | | TradingView | インジケーター検索から Moving Average を追加する | 単純・指数など（スクリプトによる） | | ブラウザ（Formiq） | インジケーター設定の MA を追加する。複数本を並べられる | 単純・指数・加重・平滑・ハルの5つ | 線はローソク足に重なって描かれます。下の別枠には出ません。設定画面に並ぶ項目は次の4つです。 | 項目 | 初期設定 | 何が変わるか | |---|---|---| | 期間 | 20 | ならす本数。増やすほど価格から離れ、向きが変わりにくくなる | | 種類 | SMA（単純） | 重みの配り方。この記事が測っているのはここ | | 色 | 線ごとに自動 | 2本以上引くとき、速い側と遅い側を見分けるため | | 線の太さ | 1 | 表示のみ。計算には影響しない | 2本引いてクロスを売買ルールにする場合、バックテストの条件では次の4つを指定します。**短期と長期で別々の種類を選べます**（この記事の後半で、混ぜたときに何が起きるかを測っています）。 | 項目 | 意味 | |---|---| | トリガー | 「クロス」か「価格と移動平均の位置関係」か | | 短期の期間・種類 | 速いほうの線 | | 長期の期間・種類 | 遅いほうの線 | いちばん一般的な使い方は、短い線が長い線を上に抜けたら買い、下に抜けたら売る、というものです。ゴールデンクロスとデッドクロスの名前で呼ばれます。この記事の数字は、すべてこのルールで測っています。 設定できる項目は期間と移動平均線の種類です。**先に期間を決める必要があります。** 期間の選び方そのものは[移動平均線クロスの設定を280通り検証した記事](/ja/blog/moving-average-cross-settings)で扱っているので、ここでは期間を固定したまま種類だけを変えます。 ## 検証の条件 数字を見る前に、何をどう測ったかを書いておきます。この表と同じ設定を入れれば、同じ結果が出ます。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年、および2025年の前半・後半） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 2025年で15分足24,903本、1時間足6,226本、4時間足1,610本 | | 買い | 短期の移動平均が長期を上抜けた足の終値 | | 売り | 短期が長期を下抜けた足の終値 | | 決済 | 反対側のクロス。同じ足の終値でそのまま反対方向へ建て直す（ドテン）。損切り・利確・時間決済は使わない | | スプレッド | 0.3pips（スリッページ0、0.1ロット）。決済と新規の両方に掛かる | | 試した組み合わせ | 短期8種（5・8・10・12・15・20・25・50）× 長期8種（20・25・30・40・50・75・100・200）のうち短期が長期より短い56通り。これを単純・指数の2種類で回して1対1で比較。時間足3つ × 区分4つで672通り | | 対照群 | 期間を1本ずつ動かした連続の梯子（短期5〜30・長期50固定と、短期10固定・長期20〜60）を両方の種類で。780組 | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | 短期と長期には同じ計算方法を使っています。片方だけ変えた場合は後の「短期と長期で種類を変えると何が起きるか」で別に測りました。 損切りを置かず反対クロスまで持つのは、種類の違い以外を混ぜないためです。損切りを付けた場合は「フィルターと損切り利確」で別に測っています。 ## SMAとEMA、どちらを使うべきか 672通りの総当たりは、ほぼ引き分けでした。同じ期間・同じ時間足のまま種類だけ入れ替えた結果です。冒頭の図がこの表です。 | 時間足 | 区分 | EMAが勝った数 | EMAの平均損益 | SMAの平均損益 | 差の中央値 | 差の最小 | 差の最大 | |---|---|---|---|---|---|---|---| | 15分足 | 2025年 | 29 / 56 | +1,134.4 | +1,064.1 | +16.3 | −1,396.5 | +1,767.2 | | 15分足 | 2024年 | 26 / 56 | +1,529.4 | +1,632.0 | −127.3 | −2,201.6 | +1,457.7 | | 15分足 | 2025年前半 | 31 / 56 | +213.5 | +76.8 | +92.3 | −1,035.4 | +1,549.7 | | 15分足 | 2025年後半 | 20 / 56 | +903.9 | +978.1 | −280.4 | −1,160.1 | +1,934.7 | | 1時間足 | 2025年 | 26 / 56 | +76.7 | +62.9 | −142.4 | −1,337.9 | +2,121.2 | | 1時間足 | 2024年 | 32 / 56 | +1,503.2 | +1,198.7 | +241.1 | −1,761.6 | +3,542.2 | | 1時間足 | 2025年前半 | 39 / 56 | −46.9 | −265.1 | +203.1 | −651.9 | +1,705.4 | | 1時間足 | 2025年後半 | 15 / 56 | +179.9 | +363.8 | −218.9 | −1,055.2 | +1,465.6 | | 4時間足 | 2025年 | 22 / 56 | −1,410.4 | −1,162.9 | −195.8 | −3,096.9 | +1,807.8 | | 4時間足 | 2024年 | 35 / 56 | +2,429.9 | +1,965.0 | +284.6 | −1,114.5 | +4,517.7 | | 4時間足 | 2025年前半 | 27 / 56 | −214.7 | −189.3 | −60.7 | −1,573.9 | +1,809.5 | | 4時間足 | 2025年後半 | 16 / 56 | −1,075.0 | −857.3 | −187.4 | −1,624.5 | +598.1 | （数字はpips） **12区分すべてを合わせて、EMAが勝ったのは672通り中318通り、47.32%です。** 完全に同じ損益になった組み合わせは1つもありませんでした。 区分ごとに見ると、1時間足の2025年前半は39/56でEMA寄り、同じ1時間足の2025年後半は15/56でSMA寄りです。**同じ時間足・同じ56通りで、半年をまたぐと勝つ側が入れ替わっています。** 「2年とも黒字だった組み合わせ」という別の見方でも並べておきます。 | 時間足 | 種類 | 2025年に黒字 | 2024年に黒字 | 両方の年で黒字 | 2025年の中央値 | 2024年の中央値 | |---|---|---|---|---|---|---| | 15分足 | SMA | 51 / 56 | 53 / 56 | 48 / 56 | +1,004.1 | +1,650.5 | | 15分足 | EMA | 52 / 56 | 56 / 56 | 52 / 56 | +1,133.7 | +1,516.4 | | 1時間足 | SMA | 35 / 56 | 44 / 56 | 28 / 56 | +371.6 | +1,201.0 | | 1時間足 | EMA | 34 / 56 | 56 / 56 | 34 / 56 | +395.4 | +1,348.0 | | 4時間足 | SMA | 3 / 56 | 53 / 56 | 2 / 56 | −1,224.6 | +2,129.6 | | 4時間足 | EMA | 2 / 56 | 56 / 56 | 2 / 56 | −1,542.7 | +2,456.2 | この見方だとEMAがわずかに多く残ります（15分足52対48、1時間足34対28、4時間足は2対2）。ただし1組ごとの勝敗では引き分けなので、**「EMAのほうが黒字の設定が多い」と「EMAのほうが損益が良い設定が多い」は同時には成立しません。** 前者は成立し、後者は成立しないという結果です。 ## 去年勝ったほうは今年も勝つか 3つの時間足すべてで、的中はコイン投げの50%に届きませんでした。それぞれの期間の組み合わせについて、2024年に勝った種類を選び、それが2025年にも勝ったかを数えました。 | 時間足 | 2024年にEMAが勝った数 | 2025年にEMAが勝った数 | 2年とも同じ側が勝った数 | 一致率 | 2025年の前半と後半で一致 | |---|---|---|---|---|---| | 15分足 | 26 / 56 | 29 / 56 | 19 / 56 | 33.93% | 25 / 56 | | 1時間足 | 32 / 56 | 26 / 56 | 26 / 56 | 46.43% | 30 / 56 | | 4時間足 | 35 / 56 | 22 / 56 | 25 / 56 | 44.64% | 25 / 56 | **15分足では33.93%です。コイン投げの50%を下回っています。** 1時間足と4時間足も50%に届きません。半年で切っても25/56から30/56で、同じ範囲です。 よく見かける組み合わせで具体的に見ると、こうなります。 | 組み合わせ | 時間足 | 年 | SMAの年間損益 | EMAの年間損益 | 年間損益が大きい側 | |---|---|---:|---:|---:|---| | 5/20 | 1時間足 | 2025年 | −484.9pips | +1,636.3pips | EMA | | 5/20 | 1時間足 | 2024年 | +3,744.7pips | +2,293.3pips | SMA | | 10/20 | 15分足 | 2025年 | +1,015.0pips | +2,052.3pips | EMA | | 10/20 | 15分足 | 2024年 | +1,600.9pips | +1,183.0pips | SMA | | 10/20 | 4時間足 | 2025年 | −1,336.2pips | −664.6pips | EMA | | 10/20 | 4時間足 | 2024年 | −1,245.7pips | +3,272.0pips | EMA | | 12/25 | 15分足 | 2025年 | +2,572.9pips | +1,436.6pips | SMA | | 12/25 | 15分足 | 2024年 | +1,894.7pips | +2,102.1pips | EMA | | 20/50 | 1時間足 | 2025年 | −220.5pips | +720.1pips | EMA | | 20/50 | 1時間足 | 2024年 | −34.6pips | +944.5pips | EMA | | 25/75 | 15分足 | 2025年 | −1,023.1pips | +691.3pips | EMA | | 25/75 | 15分足 | 2024年 | +3,613.7pips | +2,753.1pips | SMA | | 50/200 | 1時間足 | 2025年 | −1,340.4pips | −943.0pips | EMA | | 50/200 | 1時間足 | 2024年 | +2,435.4pips | +1,541.5pips | SMA | （全18通りのうち代表7組み合わせ） 18通り（6組み合わせ × 3時間足）のうち、EMAが2025年に上回ったのが11通り、2024年に上回ったのが9通り、**両方の年で上回ったのは3通り**でした。 15分足の10と20は、2025年にEMAが1,000pips以上勝ち、2024年にはSMAが400pips以上勝っています。同じ設定・同じ通貨ペアで、年が変わると符号が反対になりました。 ## 種類と期間、どちらの影響が大きいか 「EMAが勝ったのは47.32%」という結果だけを見ると、種類の選択には意味がないように読めます。ですが、勝率が半分になるのは、**そもそも種類を変えても結果が動かない**場合も同じです。変更幅を区別して測る必要があります。 そこで、期間を1本ずつ動かした連続の梯子を組みました。短期を5から30まで1ずつ（長期は50に固定）、長期を20から60まで1ずつ（短期は10に固定）。各段で「種類を入れ替えたときに動いた損益」と「期間を1本ずらしたときに動いた損益」を並べます。 | 時間足 | 区分 | 梯子 | 種類の入れ替え（中央値） | 期間1本（中央値） | 種類のほうが大きかった段 | |---|---|---|---|---|---| | 15分足 | 2025年 | 短期5〜30 | 317.0pips | 275.5pips | 15 / 25 | | 15分足 | 2025年 | 長期20〜60 | 322.6pips | 173.7pips | 26 / 40 | | 1時間足 | 2025年 | 短期5〜30 | 270.6pips | 224.1pips | 16 / 25 | | 1時間足 | 2025年 | 長期20〜60 | 601.0pips | 146.9pips | 35 / 40 | | 4時間足 | 2025年 | 短期5〜30 | 364.7pips | 229.3pips | 14 / 25 | | 4時間足 | 2025年 | 長期20〜60 | 374.9pips | 122.9pips | 34 / 40 | （2025年ぶんの6行。全24区分の集計は下） **24区分780段の全体では、種類の入れ替えのほうが大きく動いた段が593段、76.03%でした。** 中央値は種類が350.7pips、期間1本が154.2pipsです。 もう少し直感的な形にするため、「その段で種類を入れ替えたのと同じだけ損益を動かすには、期間を何本ずらす必要があるか」も数えました。**中央値は2本**です。1本で足りた段が804段中268段、5本以上必要だった段が208段、20本ずらしても届かなかった段が70段ありました。 **SMAをEMAに替えることは、期間を2本ずらすくらいの大きさの変更です。小さくはありません。ただし、どちらへ動くかは事前に決まりません。** これは[フィボナッチ・リトレースメント](/ja/blog/fibonacci-retracement-levels)の検証で使ったのと同じ考え方です。候補が2つしかないとき、片方が勝つ回数を数えるだけでは何も分かりません。**隣に対照群を置いて、初めて「勝ち負けに意味があるか」が判定できます。** ## よく使う組み合わせで比べる 種類ごとの取引回数と勝率を、代表的な設定で並べます。 | 組み合わせ | 時間足 | SMAの回数 | EMAの回数 | SMAの勝率 | EMAの勝率 | |---|---|---|---|---|---| | 5 / 20 | 15分足 | 1,616回 | 1,513回 | 33.11% | 28.42% | | 10 / 20 | 15分足 | 1,402回 | 1,053回 | 37.52% | 29.44% | | 12 / 25 | 15分足 | 1,106回 | 849回 | 37.25% | 30.04% | | 10 / 20 | 1時間足 | 358回 | 264回 | 39.11% | 32.58% | | 12 / 25 | 1時間足 | 274回 | 204回 | 41.97% | 36.27% | | 25 / 75 | 1時間足 | 99回 | 75回 | 44.44% | 42.67% | | 50 / 200 | 4時間足 | 14回 | 8回 | 50.00% | 37.50% | （ドル円2025年） 672通り全体では、**EMAのほうが取引回数が少なかったのが405通り、勝率が低かったのが561通り**です。代表18通りに絞ると、勝率は18通り全部でEMAのほうが低くなりました。 それでも損益が引き分けになるのは、EMAの1回あたりの利益がSMAより大きいからです。15分足の10と20では、SMAが1,402回で+1,015.0pips（1回あたり+0.72pips）、EMAが1,053回で+2,052.3pips（1回あたり+1.95pips）でした。 **勝率で種類を選ぶとSMAになりますが、年間損益ではSMAとEMAに明確な優劣がありません。** 高い勝率が大きな年間損益を意味しない結果は、[ウィリアムズ%R](/ja/blog/williams-r-settings)でも確認され、勝率55〜79%の設定でも年間損益はマイナスでした。 ## 短期と長期で種類を変えると何が起きるか 短期をEMA、長期をSMAにするような使い方も見かけるので、6つの組み合わせ × 4通りの掛け合わせを2年ぶん、計36区分で回しました。 | 時間足 | 区分 | 組み合わせ | 両方SMA | 両方EMA | 短期EMA・長期SMA | 短期SMA・長期EMA | |---|---|---|---|---|---|---| | 15分足 | 2025年 | 10 / 20 | +1,015.0 | +2,052.3 | +922.3 | +1,549.2 | | 15分足 | 2025年 | 20 / 50 | +1,774.9 | +1,108.3 | +2,139.3 | +149.2 | | 1時間足 | 2025年 | 12 / 25 | +958.0 | +970.9 | +1,431.9 | +917.7 | | 1時間足 | 2025年 | 25 / 75 | −56.8 | +708.9 | +995.9 | −270.2 | | 1時間足 | 2024年 | 12 / 25 | +1,882.8 | +440.6 | +2,169.5 | −75.8 | | 4時間足 | 2024年 | 10 / 20 | −1,245.7 | +3,272.0 | +231.8 | +2,692.6 | （数字はpips。36区分のうち代表6つ） 4通りの掛け合わせが全部同じ符号だった区分は36区分中21区分、4通りのあいだの幅は中央値で1,399.7pips、最大で4,517.7pipsでした。**混ぜると結果は動きます。** ただし、[移動平均線クロスの記事](/ja/blog/moving-average-cross-settings)で見つかった「長期側に速い種類を置くと2本の役割が入れ替わる」という現象は、**SMAとEMAのあいだでは起きません。** EMAは同じ期間のSMAの0.85倍の距離しか縮めないので、期間が2倍離れていれば追いつけないからです。 | 時間足 | 組み合わせ | 期間の比 | SMA（短期側）の距離 | EMA（長期側）の距離 | 長期側のほうが価格に近いか | |---|---|---|---|---|---| | 1時間足 | 10 / 20 | 2.0倍 | 22.11pips | 28.46pips | いいえ | | 1時間足 | 20 / 50 | 2.5倍 | 33.55pips | 46.68pips | いいえ | | 1時間足 | 50 / 200 | 4.0倍 | 55.10pips | 85.62pips | いいえ | （ドル円2025年） 代表18通り（6組み合わせ × 3時間足）すべてで、長期側のEMAは短期側のSMAより遠いままでした。**役割が入れ替わるのは、期間の比が1.3倍から1.45倍まで近づいたときです**（短期をSMA 20に固定して長期を1本ずつ伸ばすと、15分足と1時間足は1.3倍、4時間足は1.45倍が境目でした）。20と25のような近い組み合わせを混ぜて使う場合だけ、確認する意味があります。 ## 取引回数や勝率は変わるか 変わりますが、勝率が低いほうが年間損益は良くなりました。ここまでで、種類ごとの性質は次のように整理できます。 | | 単純移動平均（SMA） | 指数移動平均（EMA） | |---|---|---| | 重心 | (n−1)÷2 本前 | (n−1)÷2 本前（同じ） | | 直近の足の重み | 1÷n | 約2÷n | | 期間より前を読むか | 読まない | 読む（重みの13.5%） | | 終値からの平均距離 | 基準 | 0.81〜0.90倍 | | 価格と反対に曲がるか | 10.95〜21.43%の足で曲がる | 起こらない | | 起動時の履歴に依存するか | しない（n本で一致） | する（差が半分になるのは期間の約0.35倍の本数） | | 取引回数 | 多い（672通り中405通りで） | 少ない | | 勝率 | 高い（672通り中561通りで） | 低い | | 損益 | 引き分け（318通り対354通り） | 引き分け | **勝率が高いほうと損益が良いほうが一致していません。** SMAは672通り中561通りで勝率が高いのに、損益では負け越していません（318対354）。回数が多く、1回あたりが小さいという形です。 取引回数の差は、コストに影響します。 ## フィルターと損切り利確 代表2組に、よく紹介される条件を足して測りました。 | 時間足 | 短期／長期 | 種類 | 年 | 条件 | 回数 | 勝率 | 損益 | 1回あたり | |---|---|---|---:|---|---:|---:|---:|---:| | 15分足 | 10／20 | SMA | 2025年 | なし | 1,402回 | 37.52% | +1,015.0 | +0.72 | | 15分足 | 10／20 | SMA | 2025年 | ADX 25以上 | 354回 | 34.75% | −271.0 | −0.77 | | 15分足 | 10／20 | SMA | 2025年 | 東京時間 | 460回 | 38.91% | +309.7 | +0.67 | | 15分足 | 10／20 | EMA | 2025年 | なし | 1,053回 | 29.44% | +2,052.3 | +1.95 | | 15分足 | 10／20 | EMA | 2025年 | ADX 25以上 | 154回 | 25.97% | −185.5 | −1.20 | | 15分足 | 10／20 | EMA | 2025年 | 東京時間 | 356回 | 31.18% | +1,469.8 | +4.13 | | 1時間足 | 12／25 | SMA | 2025年 | なし | 274回 | 41.97% | +958.0 | +3.50 | | 1時間足 | 12／25 | SMA | 2025年 | ADX 25以上 | 80回 | 37.50% | −400.4 | −5.00 | | 1時間足 | 12／25 | EMA | 2025年 | なし | 204回 | 36.27% | +970.9 | +4.76 | | 1時間足 | 12／25 | EMA | 2025年 | ADX 25以上 | 41回 | 29.27% | −1,050.0 | −25.61 | （損益はpips。2024年ぶんを含む24通りの一部） **24通りのうち1回あたりが改善したのは11通りです。** ADXフィルターは8通り中2通りしか改善せず、2025年は4通り全部で損益を黒字から赤字にしています。東京時間フィルターは8通り中4通りで改善しました。 **フィルターの効き方が種類によって変わる、という結果は出ませんでした。** ADXが効かないのはSMAでもEMAでも同じで、効かない方向も同じです。 損切りと利確も足しました。 | 時間足 | 短期／長期 | 種類 | 年 | 決済ルール | 回数 | 損益 | |---|---|---|---:|---|---:|---:| | 15分足 | 10／20 | SMA | 2025年 | 反対クロスのみ | 1,402回 | +1,015.0 | | 15分足 | 10／20 | SMA | 2025年 | 損切り50・利確100 | 1,372回 | +535.1 | | 15分足 | 10／20 | EMA | 2025年 | 反対クロスのみ | 1,053回 | +2,052.3 | | 15分足 | 10／20 | EMA | 2025年 | 損切り50・利確100 | 1,025回 | +1,459.8 | | 1時間足 | 12／25 | SMA | 2025年 | 反対クロスのみ | 274回 | +958.0 | | 1時間足 | 12／25 | SMA | 2025年 | 24本で時間決済 | 268回 | +1,808.1 | | 1時間足 | 12／25 | EMA | 2025年 | 反対クロスのみ | 204回 | +970.9 | | 1時間足 | 12／25 | EMA | 2025年 | 損切り50・利確100 | 185回 | +1,789.8 | （損益はpips。全32通りの一部） **32通りのうち、反対クロスまで持つ形を上回ったのは6通りです。** SMAで3通り、EMAで3通りでした。ここでも種類による差は出ていません。 ## スプレッドはどこで損益分岐か 種類の選択が実務ではっきり効くのは、ここだけでした。スプレッドを変えて2025年を回し直します。 | 時間足 | 短期／長期 | 種類 | 取引回数 | スプレッド | 年間損益 | |---|---|---|---:|---:|---:| | 15分足 | 10／20 | SMA | 1,402回 | 0.0pips | +1,435.6pips | | 15分足 | 10／20 | SMA | 1,402回 | 0.3pips | +1,015.0pips | | 15分足 | 10／20 | SMA | 1,402回 | 0.6pips | +594.4pips | | 15分足 | 10／20 | SMA | 1,402回 | 1.0pips | +33.6pips | | 15分足 | 10／20 | SMA | 1,402回 | 1.5pips | −667.4pips | | 15分足 | 10／20 | SMA | 1,402回 | 2.0pips | −1,368.4pips | | 15分足 | 10／20 | EMA | 1,053回 | 0.0pips | +2,368.2pips | | 15分足 | 10／20 | EMA | 1,053回 | 0.3pips | +2,052.3pips | | 15分足 | 10／20 | EMA | 1,053回 | 0.6pips | +1,736.4pips | | 15分足 | 10／20 | EMA | 1,053回 | 1.0pips | +1,315.2pips | | 15分足 | 10／20 | EMA | 1,053回 | 1.5pips | +788.7pips | | 15分足 | 10／20 | EMA | 1,053回 | 2.0pips | +262.2pips | | 1時間足 | 12／25 | SMA | 274回 | 0.0pips | +1,040.1pips | | 1時間足 | 12／25 | SMA | 274回 | 0.3pips | +958.0pips | | 1時間足 | 12／25 | SMA | 274回 | 0.6pips | +875.8pips | | 1時間足 | 12／25 | SMA | 274回 | 1.0pips | +766.2pips | | 1時間足 | 12／25 | SMA | 274回 | 1.5pips | +629.2pips | | 1時間足 | 12／25 | SMA | 274回 | 2.0pips | +492.2pips | | 1時間足 | 12／25 | EMA | 204回 | 0.0pips | +1,032.1pips | | 1時間足 | 12／25 | EMA | 204回 | 0.3pips | +970.9pips | | 1時間足 | 12／25 | EMA | 204回 | 0.6pips | +909.7pips | | 1時間足 | 12／25 | EMA | 204回 | 1.0pips | +828.1pips | | 1時間足 | 12／25 | EMA | 204回 | 1.5pips | +726.1pips | | 1時間足 | 12／25 | EMA | 204回 | 2.0pips | +624.1pips | 0から2.0pipsに動かしたときに失う額は、15分足のSMAで2,804.0pips。これは1,402回 × 2.0pipsとぴったり一致します。EMAは2,106.0pips（1,053回 × 2.0）、1時間足のSMAは547.9pips（274回 × 2.0の548.0に対して丸めぶんの差）、EMAは408.0pips（204回 × 2.0）でした。**コスト＝取引回数×スプレッドは、種類を変えても成立します。** 損益がゼロになるスプレッドを計算するとこうなります。 | 時間足 | 短期／長期 | 種類 | 回数 | スプレッド0での損益 | 損益分岐点 | |---|---|---|---:|---:|---:| | 15分足 | 10／20 | SMA | 1,402回 | +1,435.6pips | 1.02pips | | 15分足 | 10／20 | EMA | 1,053回 | +2,368.2pips | 2.25pips | | 1時間足 | 12／25 | SMA | 274回 | +1,040.1pips | 3.80pips | | 1時間足 | 12／25 | EMA | 204回 | +970.9pips | 5.06pips | **15分足の10と20は、SMAなら1.02pipsで損益がゼロ、EMAなら2.25pipsまで持ちます。** 回数が3割少ないぶん、スプレッドに対する余裕が倍以上あります。 この差は損益の期待値ではなく、**期待値のばらつきに対する耐性**です。損益そのものは引き分けでも、スプレッドが広がる時間帯に取引が集中したり、口座のスプレッドが想定より広かったりしたときに、先に赤字になるのはSMA側です。672通り中405通りでEMAのほうが回数が少ないので、この向きはある程度一般的です。 実務的な差として残ったのは、取引回数とスプレッド耐性だけでした。EMAのほうが回数が少なく（672通り中405通り）、損益分岐点のスプレッドが広くなります。損益で選べない以上、**自分の口座のスプレッドと、どれだけ売買したいかで選ぶ**のが、この検証から言える唯一の選び方です。 期間の選び方のほうが先に決まる問題なので、[移動平均線クロスの設定を280通り検証した記事](/ja/blog/moving-average-cross-settings)から読むと、この記事の位置づけが分かりやすくなります。[プログラミングなしで条件を組んで検証する](/ja/blog/backtest-without-coding)なら、この記事の表と同じことを自分の通貨ペア・自分の期間で回して確かめられます。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです。他の通貨ペアや他の年で同じ傾向になる保証はありません - 比べたのは単純と指数の2種類だけです。加重（WMA）・平滑（SMMA）・ハル（HMA）を含めた5種類の比較は[移動平均線クロスの記事](/ja/blog/moving-average-cross-settings)にあります - 重心・重み・向きの計算は数式から出る性質なので通貨ペアや年に依存しませんが、**終値からの距離やクロス回数はドル円の2年ぶんの実測**です - エントリーも決済も足の終値で行っています。実際の約定はここからずれます - スプレッドは全期間0.3pips固定として計算しています。実際のスプレッドは時間帯と指標発表で変動します - 起動時の履歴の影響は、2025年1月1日の1点で測っています。別の日を選べば差の大きさは変わります（縮小率のほうは日付に依存しません） --- ### アリゲーターの期間とずらし幅を1,584通り検証 URL: https://formiq.jp/ja/blog/alligator-settings Language: ja Published: 2026-08-27 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: アリゲーター, ビル・ウィリアムズ, 移動平均線, トレンド, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/alligator-settings アリゲーター（Alligator）は、ビル・ウィリアムズが作った3本の移動平均線です。**顎（13）・歯（8）・唇（5）**と名前が付いていて、3本が絡み合っていれば「ワニが寝ている」、順番に並んで開いていけば「ワニが食べている」と読みます。 ただの3本の移動平均線と違うところが1つあります。**3本がそれぞれ8本・5本・3本ぶん、右に（未来の方向に）ずらして描かれている**ことです。このシリーズで測った19本のうち、時間軸そのものを動かす指標はアリゲーターが初めてです。 先にそこから測りました。**ずらしは予測ではなく、遅れです。** 8本先に置かれた線は、8本前に計算した値をそこに置いたものなので、今日それを読むことは8本前の平均を読むことです。未来を見ているわけではありません。 そして、もう一つ。**ワニが食べている時間は、年の8割ありました。** ## アリゲーターは何を測っているのか 3本とも同じ作り方です。 1. **中央値** =（高値＋安値）÷2 2. その平滑移動平均（SMMA）を、13本・8本・5本で計算する 3. できた3本を、それぞれ8本・5本・3本ぶん右にずらして描く 平滑移動平均は前の値に今の値を1本ぶん足していく形なので、単純移動平均より反応が鈍くなります。 恣意的な選択が3つ入っています。中央値を使うこと、SMMAを使うこと、そしてずらすことです。順に測りました。 ### ずらしは遅れである 「右にずらす」の中身はこうです。13本のSMMAをi番目の足で計算し、その値を **i＋8番目の足の位置に置く**。読むときは逆で、i番目の足に描かれている顎は、**i−8番目の足で計算された値**です。 つまり**未来のデータは一切入っていません。** 入っているのは古さだけです。 実測でも一致しました。ずらした3本と、ずらさない3本を突き合わせます。 | 時間足 | 年 | 突き合わせた値の数 | 不一致 | |---|---|---|---| | 15分足 | 2024 | 74,958 | 0 | | 1時間足 | 2024 | 18,711 | 0 | | 4時間足 | 2024 | 4,809 | 0 | | 15分足 | 2025 | 74,670 | 0 | | 1時間足 | 2025 | 18,639 | 0 | | 4時間足 | 2025 | 4,791 | 0 | **ずらした線は、ずらさない線をそのぶん過去から持ってきたものです。** この記事の残りは「その遅れに価値があるか」を測っています。 ### 中央値と終値、SMMAと他の平均 分子の作り方を差し替えて比べました（ずらしなし・2025年）。 | 変えたもの | 唇の相関 | 3本の並び順が食い違う足 | |---|---|---| | 中央値 → 終値 | 1.0000〜0.9999 | 3.7〜4.6% | | SMMA → EMA | 0.9998〜0.9999 | 26.7〜28.5% | | SMMA → SMA | 0.9989〜0.9999 | 41.4〜45.2% | **相関が0.999でも、並び順は半分近く食い違います。** 3本が近い場所にいるので、線としてはほぼ重なっていても、どちらが上かは頻繁に入れ替わるからです。相関を見て「ほとんど同じだから設定は関係ない」と結論すると間違えます。 中央値と終値の差は小さく（3.7〜4.6%）、平均の種類の差は大きい（26.7〜45.2%）。**アリゲーターで効いている選択はSMMAのほうです。** ## アリゲーターの設定と読み方 主要な環境には最初から入っています。追加すると、**チャートの上に3本の線が重なって表示されます。** | 環境 | 追加する場所 | |---|---| | MT4 / MT5 | 挿入 → インディケータ → ビル・ウィリアムズ → Alligator | | TradingView | インジケーター検索から Williams Alligator | | ブラウザ（Formiq） | インジケーター一覧から「アリゲーター」 | ### 設定ダイアログに並ぶ8つの項目 MT4で追加すると、次の値が入った状態で開きます。**どの環境でもこの8つは同じです。** | 項目 | 初期設定 | 意味 | |---|---|---| | Jaw period（顎の期間） | 13 | いちばん遅い線。何本ぶんの価格を平均するか | | Jaw shift（顎のずらし） | 8 | その線を右に何本ずらして描くか | | Teeth period（歯の期間） | 8 | 真ん中の線の期間 | | Teeth shift（歯のずらし） | 5 | 真ん中の線のずらし | | Lips period（唇の期間） | 5 | いちばん速い線の期間 | | Lips shift（唇のずらし） | 3 | いちばん速い線のずらし | | MA method（平均の種別） | Smoothed | 平滑移動平均。単純平均より反応が鈍い | | Apply to（適用価格） | Median price (HL/2) | 高値と安値の真ん中を使う。終値ではない | MT4の初期色は**顎が青、歯が赤、唇が緑**です。線の名前は太さや役割の比喩で、計算はどれも同じです。**遅い順に顎・歯・唇**とだけ覚えれば読めます。 ### 3本の並びで相場を3つに分ける アリゲーターの読み方は、3本の位置関係だけで決まります。 | 状態 | 3本の見え方 | ビル・ウィリアムズの言い方 | |---|---|---| | 寝ている | 3本が絡み合って横ばい | ワニが眠っている。手を出さない | | 目覚める | 3本が離れはじめる | ワニが口を開く。準備する | | 食べている | 唇＞歯＞顎（またはその逆）に並んで開く | ワニが食べている。その方向に乗る | **上向きに食べている**＝唇（緑）がいちばん上、顎（青）がいちばん下。**下向き**はその逆です。この記事で「順番に並んでいる」と書いているのは、この状態を指しています。 ### 売買のルールとして使われている形 もっとも広く紹介されているのは、**唇（5）が歯（8）を上に抜けたら買い、下に抜けたら売り**という形です。この記事はこれを「唇×歯のクロス」として測っています。 ほかに2つ、よく使われる読み方があります。 - **3本が順番に並んだら乗る**：クロスの瞬間ではなく、並んだ状態が続くあいだ持つ - **価格が3本すべての外に出たら乗る**：ローソク足の終値が3本より上なら買い、下なら売り 唇と歯のクロス、3本の並び順、価格が3本すべての外に出る条件を、それぞれ別の売買ルールとして検証しました。 ### どこをいじるべきか **期間を変えれば線の速さが変わり、ずらしを変えれば線が横に動きます。** 平均の種別と適用価格も変更できます。平均の種別を変えると並び順が変わる足は26.7〜45.2%、適用価格を変えた場合は3.7〜4.6%ありました。 **ずらし幅を0にできる環境なら、3本の線は前方へ移動しない通常の移動平均線になります。** この記事では、一般的な前方ずらしを使う設定と、ずらし幅0の設定を比べています。 ## 検証の条件 数字を見る前に、何をどう測ったかを書いておきます。この表と同じ設定を入れれば、同じ結果が出ます。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 読み方 | 唇×歯のクロス ／ 3本が順番に並ぶ ／ 価格が3本を抜ける | | 期間の組 | フィボナッチ列 5/3/2・8/5/3・**13/8/5**・21/13/8・34/21/13 ／ 対照群 12/7/4・12/7/6・14/9/4・14/9/6 ／ 顎だけ動かす 11/8/5・16/8/5 の計11組 | | ずらし | なし 0/0/0 ／ 半分 4/3/2 ／ **本来 8/5/3** ／ 倍 16/10/6 | | 組み合わせ | 1区分あたり132通り。時間足3種 × 期間4区分で合計1,584通り | | 決済 | 反対シグナルまで。エントリーと決済に同じルールセットを使う | | 損切り・利確 | 使わない（別セクションで測定） | | スプレッド | 0.3pips固定。約定は終値 | | ロット | 0.1ロット | 3つの読み方はどれも対称（買いと売りが同じ形で出る）なので、エントリーと決済に同じルールセットを使いました。 ## 寝ているのはどれくらいの期間か 年の16〜19%だけです。6通りすべてで、口が開いている足が80%を超えました。3本が「順番に並んでいる」（唇＞歯＞顎、またはその逆）足の割合です。 | 時間足 | 年 | 上向きに並ぶ | 下向きに並ぶ | 並んでいる合計 | 絡んでいる | |---|---|---|---|---|---| | 15分足 | 2024 | 46.47% | 34.08% | 80.54% | 19.46% | | 1時間足 | 2024 | 50.27% | 30.58% | 80.85% | 19.15% | | 4時間足 | 2024 | 56.52% | 27.44% | 83.96% | 16.04% | | 15分足 | 2025 | 41.79% | 38.83% | 80.62% | 19.38% | | 1時間足 | 2025 | 41.65% | 39.20% | 80.86% | 19.14% | | 4時間足 | 2025 | 38.30% | 42.58% | 80.88% | 19.12% | **6通りすべてで80%を超えました。** ずらしを外すと85.9〜88.3%まで上がります。「ワニが寝ているあいだは手を出さない」というルールは、**年の16〜19%だけ手を出さない**ということです。 これは絞り込みの条件になりません。**GMMAの検証でも、長期リボンの整列が85%の足で成立していて同じ結論になりました。** 状態を条件にする前に、その状態がどれくらい珍しいのかを数える必要があります。 上向きと下向きの内訳は素直に相場を映しています。2024年のドル円は年間1,632pips上げていて、1時間足では50.27%対30.58%。2025年は年間−56pipsの横ばいで、41.65%対39.20%とほぼ半々でした。 ## ずらしは外したほうがいいか 18通り中16通りで、外したほうが損益は上でした。同じ読み方・同じ期間で、ずらし幅だけを変えて突き合わせます。 | 読み方 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---| | 唇×歯のクロス | 13/22 | 19/22 | 20/22 | | 3本が順番に並ぶ | 14/22 | 5/22 | 12/22 | | 価格が3本を抜ける | 1/22 | 8/22 | 11/22 | 各セルは「11組の期間 × 2年」で、**ずらしを外したほうが損益で上回った数**です。 **全体では198通り中103通り。** 半分です。ずらしそのものに一般的な効果は検出できませんでした。 ただし読み方で向きが割れます。**公表されている「唇が歯を抜ける」エントリーに限れば66通り中52通りで、外したほうが良い**。逆に「価格が3本を抜ける」は66通り中20通りで、ずらしたままのほうが良い。 いちばん具体的な形で書きます。1時間足・期間13/8/5・唇×歯のクロスです。 | ずらし | 年 | 取引数 | 勝率 | 損益 | |---|---:|---:|---:|---:| | なし | 2025 | 298 | 34.56% | **+742.5pips** | | なし | 2024 | 対象外 | 対象外 | **+2,944.9pips** | | 半分（4/3/2） | 2025 | 374 | 44.39% | +397.9pips | | 半分（4/3/2） | 2024 | 対象外 | 対象外 | +2,462.2pips | | 本来（8/5/3） | 2025 | 404 | 43.07% | +115.5pips | | 本来（8/5/3） | 2024 | 対象外 | 対象外 | +973.7pips | | 倍（16/10/6） | 2025 | 370 | 48.11% | +332.1pips | | 倍（16/10/6） | 2024 | 対象外 | 対象外 | +1,298.8pips | 本来のずらしは、ずらしなしより両年とも損益が小さくなりました。2025年は取引数が298回から404回へ増え、勝率は34.56%から43.07%へ上がりましたが、損益は+742.5pipsから+115.5pipsへ減っています。 損益分岐スプレッドで見るともっとはっきりします。**ずらしありは0.59pips、ずらしなしは2.79pips。** 前者はスプレッドが0.6pips開いた時点で成立しません。 ### 値動きだけで見ても同じ 売買を挟まずに測ります。唇が歯を抜けた足を全部拾い、そこから10・20・50本先の値幅から同区間の無条件平均を引きます（2024年は1,632pips上げているので、生の値幅では買い方向が必ず大きく出るため）。 ずらしありとずらしなしを別々に集計すると、**ずらさないほうが良かったのは18通り中16通り**でした。ずらすと交差の回数も増えます（1時間足2024年で410回対294回）。**遅らせた線どうしを比べると、交差が増えて中身が薄くなる**、という形です。 ## 顎の期間は結果に効くのか 標準のエントリーでは読まれていません。「唇が歯を抜けたら」というルールは、唇と歯しか読みません。顎は入っていません。 期間11/8/5・13/8/5・16/8/5の3組は、歯と唇が同じで顎だけ違います。回すとこうなりました。 | 読み方 | 11/8/5 | 13/8/5 | 16/8/5 | |---|---|---|---| | 唇×歯のクロス（1時間足2025） | +397 | +397 | +397 | | 3本が順番に並ぶ | +357 | +306 | +483 | | 価格が3本を抜ける | +1,495 | +1,337 | +935 | **クロスの行は3つとも同じ数字です。** 1回も違いません。ワニの名前の由来である13本の線は、いちばん普及しているエントリー方法では一度も読まれていません。 この検証ではそれを踏まえて、エディタ側でも**クロスを選んだときは顎の期間とずらし幅を隠す**ようにしました。売買結果を変えない設定項目は表示しないほうが、条件を誤解せずに済みます。 ## 13・8・5はフィボナッチ数だから効くのか 顎13・歯8・唇5、ずらし8・5・3。全部フィボナッチ数です。 [フィボナッチ・リトレースメントの検証](/ja/blog/fibonacci-retracement-levels)で使った型をそのまま当てました。**フィボナッチ数を1つも含まない対照群**を作って、隣同士で比べます。対照群は12/7/4・12/7/6・14/9/4・14/9/6の4組で、歯と唇の4隅を囲む形です。 **13/8/5が対照群を上回ったのは288通り中136通り。** 半分を下回りました。 | 時間足 | 年 | フィボナッチ列5組の平均 | 対照群4組の平均 | |---|---|---|---| | 15分足 | 2025 | +435 | +865 | | 15分足 | 2024 | +937 | +1,178 | | 1時間足 | 2025 | +707 | +734 | | 1時間足 | 2024 | +1,399 | +1,293 | | 4時間足 | 2025 | −982 | −952 | | 4時間足 | 2024 | +1,709 | +2,055 | 6通り中5通りで対照群のほうが上です（唯一の例外は1時間足2024年の+1,399対+1,293）。**フィボナッチ数であることに固有の効果は検出できませんでした。** 深さで測ったリトレースメントの記事と同じ結論です。 ## どの時間足で使えるか 1時間足の「価格が3本を抜ける」が44通り中44通りで、最も揃いました。両方の年で黒字だった設定の数です。 | 読み方 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---| | 唇×歯のクロス | 35/44 | 39/44 | 4/44 | | 3本が順番に並ぶ | 41/44 | 27/44 | 1/44 | | 価格が3本を抜ける | 22/44 | **44/44** | 9/44 | **1時間足の「価格が3本を抜ける」が44通り中44通り**、2024年も2025年も全部黒字です。このシリーズで最も揃った数字でした。 4時間足はまた反転しました。2024年は3つの読み方が40/44・42/44・44/44だったのに、2025年は4/44・1/44・9/44です。GMMAとCCIでも同じことが起きています。 ### ただし2025年の前半だけを見ると 年間の数字を半分に割ります。1時間足、44通り中の黒字数です。 | 読み方 | 集計期間 | 黒字設定 | |---|---|---:| | 唇×歯のクロス | 2024年 | 42/44 | | 唇×歯のクロス | 2025年 | 41/44 | | 唇×歯のクロス | 2025年前半 | **11/44** | | 唇×歯のクロス | 2025年後半 | 44/44 | | 3本が順番に並ぶ | 2024年 | 34/44 | | 3本が順番に並ぶ | 2025年 | 31/44 | | 3本が順番に並ぶ | 2025年前半 | **10/44** | | 3本が順番に並ぶ | 2025年後半 | 34/44 | | 価格が3本を抜ける | 2024年 | 44/44 | | 価格が3本を抜ける | 2025年 | 44/44 | | 価格が3本を抜ける | 2025年前半 | **18/44** | | 価格が3本を抜ける | 2025年後半 | 43/44 | **44/44は、18/44の半年と43/44の半年を足したものでした。** 「2年とも黒字」と「ずっと黒字」は別のことです。 具体的な設定で書くと、期間13/8/5・ずらし8/5/3・価格が3本を抜ける・1時間足は、2025年通年で361回・+969.3pipsですが、**前半は212回で−673.5pips、後半は150回で+1,618.7pips**です。 3つの読み方が同時に減っているため、特定の読み方だけの問題ではありません。年単位の集計では、この前半と後半の差が見えなくなります。 ## 勝率が低いほうが勝てるのはなぜか 平均利益が平均損失の2.6倍あるからです。1時間足2025年、売買の生きている設定の平均を見ます。 | 読み方 | 平均勝率 | 平均利益 | 平均損失 | 平均保有 | 黒字 | |---|---|---|---|---|---| | 唇×歯のクロス | 42.8% | +56.5 | −38.6 | 26.7本 | 41/44 | | 3本が順番に並ぶ | 41.1% | +76.8 | −50.7 | 51.9本 | 31/44 | | 価格が3本を抜ける | 30.4% | +76.7 | −29.2 | 24.6本 | 44/44 | **勝率が最も低い「価格が3本を抜ける」が、黒字設定44/44でした。** 平均利益が平均損失の2.6倍あるためです。 ## 口が広いと大きく動くのか 「口が開いていれば強いトレンド」という主張は、唯一そのまま成立しました。 3本の最大と最小の差でエントリー時点を四分位に分け、その後20本の値幅（絶対値）を測ります。 | 時間足 | 最も狭い | 第2 | 第3 | 最も広い | 全足の平均 | |---|---|---|---|---|---| | 15分足 | 24.69 | 28.39 | 28.86 | **34.75** | 29.17 | | 1時間足 | 57.53 | 61.91 | 59.42 | **73.06** | 62.98 | | 4時間足 | 116.48 | 125.60 | 127.92 | **133.64** | 125.92 | **3つの時間足すべてで、最も広い区分の後がいちばん大きく動きます。** 最も狭い区分はどれも全足平均を下回ります。 ただし読み方には注意が必要です。3本が離れているのは**直前に価格が動いたから**なので、これは「動いた後は動く」という値動きの性質（ボラティリティの持続）であって、方向を当てているわけではありません。差の大きさも15分足で+19%です。 ## フィルター・損切り利確・コスト ### ADXフィルターは4度目の失敗 | 読み方 | ずらし | 時間足 | 年 | ADX条件 | 取引数 | 損益 | |---|---|---|---:|---|---:|---:| | 唇×歯 | あり | 1時間足 | 2025 | なし | 404回 | +115.5pips | | 唇×歯 | あり | 1時間足 | 2025 | 20以上 | 182回 | +121.8pips | | 唇×歯 | あり | 1時間足 | 2025 | 25以上 | 100回 | **−1,175.1pips** | | 唇×歯 | あり | 1時間足 | 2025 | 30以上 | 50回 | −695.5pips | | 唇×歯 | なし | 1時間足 | 2025 | なし | 298回 | +742.5pips | | 唇×歯 | なし | 1時間足 | 2025 | 20以上 | 128回 | −116.2pips | | 唇×歯 | なし | 1時間足 | 2025 | 25以上 | 73回 | −137.6pips | | 唇×歯 | なし | 1時間足 | 2025 | 30以上 | 35回 | −101.7pips | トレンド系にトレンドフィルターを足すと条件が重複します。GMMA・CCIに続いて3指標目、4回目です。 ### 損切り・利確 **損切り100・利確200は、唇×歯のクロスでは両方の年を改善しました。** ずらしなしが2025年 +742.5 → **+2,242.8**、2024年 +2,944.9 → +3,172.2。ずらしありも2025年 +115.5 → +624.6、2024年 +973.7 → +1,232.6です。 ただし読み方を変えると成立しません。「価格が3本を抜ける」1時間足は2025年こそ+969.3 → +1,255.3ですが、2024年は+2,161.6 → +2,002.2と落ちます。**測った4つの設定のうち、両方の年で改善したのはクロスの2つだけ**でした。 ### コスト 損益はスプレッドに対してきれいに直線でした。**失ったpips＝取引回数×スプレッド**が誤差なく成立します。 | 設定 | 取引数 | スプレッド0 | 0.3 | 1.0 | 損益分岐 | |---|---|---|---|---|---| | 唇×歯・ずらしあり・1時間足 | 404 | +236.7 | +115.5 | −167.3 | **0.59pips** | | 唇×歯・ずらしなし・1時間足 | 298 | +831.9 | +742.5 | +533.9 | **2.79pips** | | 価格が3本を抜ける・1時間足 | 361 | +1,077.6 | +969.3 | +716.6 | **2.99pips** | **ずらしを外すだけで、損益分岐が0.59pipsから2.79pipsに上がりました。** 取引回数が106回減るぶんだけコストが軽くなり、1回あたりの利益が伸びるからです。 ## 関連記事 - [GMMAの本数](/ja/blog/gmma-settings)：「長期リボンの整列」も85%の足で成立していました。同じ基準率の話です - [フィボナッチ・リトレースメント](/ja/blog/fibonacci-retracement-levels)：この記事の対照群の作り方は、そちらで確立したものです - [移動平均線クロスの期間](/ja/blog/moving-average-cross-settings)：平均の種類がどれだけ効くかを、単体で測っています - [CCIの±100](/ja/blog/cci-settings)：式の中の恣意的な選択を1つずつ差し替える型は、そちらと同じです - [一目均衡表の転換線・基準線・先行スパン・遅行スパン](/ja/blog/ichimoku-settings)：線を前後へずらす別の指標を、先読みなしの条件へ戻して検証しています ## 補足 - **通貨ペアはドル円だけ、期間は2年だけ**です。アリゲーターは商品先物のトレーダーが作った指標なので、対象が変われば口が開いている時間の割合も変わります - **ずらしの検証は0/0/0・4/3/2・8/5/3・16/10/6の4通り**だけです。連続的に振ってはいません - **平均の種別と適用価格は線の比較でだけ測っていて、売買では測っていません。** SMMAをEMAに替えると並び順が27〜29%の足で変わるので、売買結果も変わるはずですが、この記事はその数字を持っていません - **スプレッドは0.3pips固定**です。損益分岐0.59pipsの設定は、この前提が崩れると成立しません - **口の広がりと次の値幅の関係は方向を含みません。** 絶対値で測っているので、「大きく動く」であって「上に動く」ではありません - **4時間足の標本は小さい**（読み方によって年25〜290回）。2024年と2025年で完全に反転しましたが、どちらが例外なのかはこの2年では決められません --- ### ボリンジャーバンドは1σ・2σ・3σどれが勝てる？検証した結果 URL: https://formiq.jp/ja/blog/bollinger-band-sigma-settings Language: ja Published: 2026-08-27 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ボリンジャーバンド, シグマ, 標準偏差, 順張り, 逆張り, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/bollinger-band-sigma-settings ボリンジャーバンドの設定画面には「偏差」や「Deviations」という欄があります。定番は2ですが、1σ、2σ、3σのどれを使うべきかは、売買の向きで答えが変わります。 ドル円で0.5〜4σを0.25刻みにし、期間10〜50の7種と組み合わせました。逆張り買い、逆張り売り、終値のバンド抜けについて、15分足・1時間足・4時間足、2024年・2025年と2025年前半・後半を合わせて**3,780回**バックテストしています。 結論から書くと、**1時間足の順張りなら1σが比較の基準**になりました。0.5〜1.25σは期間7種すべてが2024年と2025年の両方で黒字です。1σの7期間平均は2024年+1,568.8pips、2025年+1,081.9pipsでした。一方、1.75σは2025年の+1,850.2pipsに対し、2024年は+590.2pipsです。 これは「どの相場でも1σが最強」という意味ではありません。**今回の候補の中で、利益、期間を変えたときの生存率、別の年への移りやすさをまとめて見ると、1σが最も無理のない出発点だった**という答えです。逆張りには、両年で再現した「一番いいシグマ」は見つかりませんでした。 ## ボリンジャーバンドのシグマは何を変えるのか 期間をn、終値の単純移動平均をSMA(n)、標準偏差をSD(n)、偏差の倍率をkとすると、バンドは次の位置に引かれます。 **上バンド = SMA(n) + k × SD(n)** **下バンド = SMA(n) − k × SD(n)** 設定欄の2は「2%」ではなく、標準偏差を2倍するという意味です。シグマを大きくするとバンドは遠くなり、タッチも終値抜けも減ります。期間を大きくしても平均線の反応が遅くなるため、取引回数は減りやすくなります。 よく使う設定の読み方は次のとおりです。 | 設定 | バンドの位置 | 今回の1時間足での用途 | |---|---|---| | 1σ | 平均線に近い | 順張りの基準候補。シグナルは多い | | 2σ | 定番の幅 | 順張りは両年黒字の設定が多い | | 3σ | かなり遠い | 取引が少なく、少数の値動きに左右される | | 4σ | ほとんど届かない | 年数回で判断することになる | 「1σに約68%、2σに約95%、3σに約99.7%が入る」という説明は、独立した値が正規分布に従う場合の目安です。実際の相場には連続性、トレンド、急変があるため、その数字を出現率として置くことはできません。 期間20のドル円1時間足を2025年の6,226本で数えると、終値が内側だった割合は**1σで45.0%、2σで87.1%、3σで99.1%**でした。2σでも12.9%は外側です。2σに触れたこと自体を「異常だから戻る」とみなすには、前提が足りません。 ## 設定画面の見方 MT4・MT5で期間20、2σを表示する場合は、次の設定になります。TradingViewやFormiqでも対応する項目は同じです。 | 設定欄 | 入力例 | 役割 | |---|---:|---| | Period | 20 | 移動平均と標準偏差を計算する本数 | | Shift | 0 | バンドを左右へずらす本数。通常は0 | | Deviations | 2.000 | 標準偏差の倍率 | | Apply to | Close | 計算に使う価格。今回の検証は終値 | | Style / Color | 任意 | 表示だけを変え、計算結果は変えない | Formiqのバックテストでは、期間と偏差に加えてトリガーを選びます。 | トリガー | ルール | |---|---| | バンドタッチ | 安値が下バンドに触れたら買い、または高値が上バンドに触れたら売り | | バンドブレイク | 終値が上バンドを上抜けたら買い、下バンドを下抜けたら売り | 同じ20・2σでも、タッチを逆張りするか、終値抜けについていくかで成績は反対になります。シグマだけを比較するときも、**先に売買の向きを固定する**必要があります。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01〜2025-12-31、比較用に2024年、2025年前半・後半 | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 2025年の本数 | 15分足24,903本、1時間足6,226本、4時間足1,610本 | | 期間 | 10・14・20・25・30・40・50の7種 | | 偏差 | 0.5〜4σを0.25刻み、15種 | | 逆張り買い | 安値が下バンドに触れた足の終値で買い、反対側の1σバンドに触れた足の終値で決済 | | 逆張り売り | 高値が上バンドに触れた足の終値で売り、反対側の1σバンドに触れた足の終値で決済 | | 順張り | 終値が上バンドを抜けたら買い、下バンドを抜けたら売り。反対側を終値で抜けたらドテン | | 組み合わせ | 7期間×15偏差×3売買ルール＝315設定を、3時間足×4期間で実行。合計3,780回 | | コスト | スプレッド0.3pips、スリッページ0、0.1ロット | | 約定 | シグナル足の終値。決済後の反対方向へのドテンは同じ足、同方向への再エントリーは次の足以降 | | 損切り・利確 | 主検証では使わない。別セクションで追加測定 | | 測定 | Formiqのバックテスト機能で実行し、全約定価格からpipsを再集計 | 逆張りの決済を反対側の1σに固定したのは、入口のシグマだけを比較するためです。出口まで同時に変えると、「入口が良かったのか、遠い出口が良かったのか」を分けられません。 ## 何シグマが一番いいか 1時間足では0.5〜1.25σが7期間すべて両年黒字で、最も安定していました。各偏差を7期間で平均した順張りの結果です。「両年黒字」は、その偏差に含まれる7期間のうち、2024年と2025年の両方でプラスだった数です。 | 偏差 | 2025年の平均損益 | 2024年の平均損益 | 2025年の平均取引数 | 両年黒字 | |---:|---:|---:|---:|---:| | 0.5σ | +1,069.1 | +1,430.3 | 363.4 | 7 / 7 | | 0.75σ | +1,113.0 | +1,217.8 | 301.1 | 7 / 7 | | 1σ | +1,081.9 | +1,568.8 | 260.4 | 7 / 7 | | 1.25σ | +999.7 | +1,088.6 | 222.4 | 7 / 7 | | 1.5σ | +1,368.1 | +693.0 | 188.7 | 5 / 7 | | 1.75σ | +1,850.2 | +590.2 | 151.7 | 5 / 7 | | 2σ | +979.6 | +742.2 | 129.0 | 6 / 7 | | 2.25σ | −18.9 | +1,530.9 | 97.7 | 3 / 7 | | 2.5σ | −812.6 | +895.6 | 65.0 | 2 / 7 | | 2.75σ | −1,133.3 | +990.7 | 40.0 | 1 / 7 | | 3σ | −124.9 | +462.2 | 21.7 | 3 / 7 | | 3.25σ | +772.2 | +114.0 | 12.7 | 2 / 7 | | 3.5σ | +509.1 | +210.8 | 6.3 | 4 / 7 | | 3.75σ | +329.7 | +697.9 | 4.1 | 4 / 7 | | 4σ | +394.5 | +989.9 | 2.3 | 4 / 7 | 0.5〜1.25σは7期間すべてが両年黒字です。中でも**1σは2024年+1,568.8pips、2025年+1,081.9pips**でした。2σも7期間中6期間が残っているため、定番設定が否定されたわけではありません。ただし、1.5〜2σは期間による差が増えます。 3.5σ以上が4 / 7まで戻っているように見える点には注意が必要です。2025年の年平均取引数は3.5σで6.3回、3.75σで4.1回、4σで2.3回です。**数回の大きなトレードが当たった結果と、設定が安定していることは別**です。 時間足を変えると、順張り105設定の全体像も変わりました。 | 時間足 | 年 | 黒字設定 | 損益中央値 | 両年とも黒字 | |---|---:|---:|---:|---:| | 15分足 | 2025 | 55 / 105 | +138.5pips | 44 / 105 | | 15分足 | 2024 | 65 / 105 | +404.3pips | 44 / 105 | | 1時間足 | 2025 | 75 / 105 | +672.9pips | 67 / 105 | | 1時間足 | 2024 | 86 / 105 | +902.5pips | 67 / 105 | | 4時間足 | 2025 | 32 / 105 | −270.8pips | 27 / 105 | | 4時間足 | 2024 | 84 / 105 | +926.6pips | 27 / 105 | 1時間足は105設定中67設定が両年黒字でした。4時間足は2024年に84設定が黒字でも、2025年には32設定へ減っています。この記事の答えを1時間足に限定しているのは、この差があるためです。 ## 定番の期間20ならどのσか 期間20だけを見ると1.75σが両年とも1σと2σを上回りました。期間を固定して偏差だけを変えると、取引数と1回あたりの重みが見えます。 | 偏差 | 年 | 取引数 | 勝率 | PF | 損益 | |---:|---:|---:|---:|---:|---:| | 1σ | 2025 | 277回 | 38.27% | 1.137 | +1,052.5pips | | 1σ | 2024 | 280回 | 34.29% | 1.281 | +2,071.7pips | | 1.75σ | 2025 | 158回 | 44.94% | 1.348 | +1,846.0pips | | 1.75σ | 2024 | 169回 | 40.24% | 1.407 | +2,331.5pips | | 2σ | 2025 | 132回 | 43.18% | 1.235 | +1,177.5pips | | 2σ | 2024 | 137回 | 41.61% | 1.270 | +1,434.6pips | | 2.75σ | 2025 | 48回 | 41.67% | 0.703 | −955.6pips | | 2.75σ | 2024 | 48回 | 50.00% | 1.219 | +791.9pips | | 3σ | 2025 | 24回 | 37.50% | 1.017 | +33.7pips | | 3σ | 2024 | 34回 | 44.12% | 1.255 | +780.1pips | ただし、期間7種で平均した2024年の損益は、1.75σが+590.2pips、1σが+1,568.8pipsです。**期間20を先に選んだ結果と、期間を変えて比較した結果は分けて読む必要があります。** ## 逆張りなら何シグマがいいか 逆張りに一番良い値はありませんでした。1時間足の105設定を、逆張り買い・逆張り売り・順張りに分けて比べます。 | 売買ルール | 2025年黒字 | 中央値 | 2024年黒字 | 中央値 | 両年黒字 | |---|---:|---:|---:|---:|---:| | 逆張り買い | 43 / 105 | −114.4pips | 48 / 105 | −20.5pips | 23 / 105 | | 逆張り売り | 11 / 105 | −483.9pips | 25 / 105 | −995.2pips | 1 / 105 | | 順張り | 75 / 105 | +672.9pips | 86 / 105 | +902.5pips | 67 / 105 | 逆張り買いは、偏差を広げると平均損失が小さくなる傾向がありました。たとえば偏差1σの7期間平均は2025年−811.6pips、2024年−111.6pipsです。2.75σでは+230.2pipsと−45.3pips、3σでは+35.6pipsと−12.7pipsでした。 2025年の逆張り買いでは2.75σが+230.2pipsでしたが、2024年は−45.3pipsです。3σも2025年+35.6pips、2024年−12.7pipsでした。売り逆張りは、両年黒字が105設定中1設定だけです。**広いシグマは取引を減らし、損失を小さく見せただけの可能性があります。** ## 去年よかったσは今年も勝てるか 1年だけで最大値を選ぶと翌年に縮み、両年で残った1σのほうが安定していました。偏差だけで2025年の最大値を選ぶと、順張りは1.75σで7期間平均+1,850.2pipsです。しかし2024年では+590.2pipsでした。2024年に平均+1,568.8pipsだった1σは、2025年にも+1,081.9pipsでした。 | 選択年 | 偏差 | 選択年の平均損益 | 確認年 | 確認年の平均損益 | |---:|---:|---:|---:|---:| | 2025 | 1.75σ | +1,850.2pips | 2024 | +590.2pips | | 2024 | 1σ | +1,568.8pips | 2025 | +1,081.9pips | 期間と偏差を同時に選ぶと、2025年に+2,753.0pipsだった期間10・1.75σは、2024年に+686.0pipsでした。2024年に+2,793.5pipsだった期間25・2.25σは、2025年に+231.3pipsです。そこで、1年だけの最大値ではなく「近い偏差の複数設定が両年で黒字か」を優先しました。 ## 勝率で選べるか 選べません。1時間足・2025年の全105設定を平均すると、逆張り買いの勝率は61.9%、逆張り売りは61.2%、順張りは41.0%でした。 | 売買ルール | 平均勝率 | 平均勝ち幅 | 平均負け幅 | |---|---:|---:|---:| | 逆張り買い | 61.9% | +39.2pips | −65.9pips | | 逆張り売り | 61.2% | +40.2pips | −84.8pips | | 順張り | 41.0% | +188.8pips | −93.0pips | 逆張りは当たる回数が多くても、負け幅が勝ち幅より大きい形です。順張りは勝率が低くても、平均勝ち幅が平均負け幅の約2倍あります。シグマを広げて勝率が上がったかだけを見ると、この差を落とします。 ## フィルター、決済、コスト 安定性の基準にした期間20・1σの順張りへ、よく使われる条件を追加しました。 | 条件 | 年 | 取引回数 | 年間損益 | |---|---:|---:|---:| | そのまま | 2025年 | 277回 | +1,052.5pips | | そのまま | 2024年 | 280回 | +2,071.7pips | | ADX 20以上 | 2025年 | 188回 | −1,033.2pips | | ADX 20以上 | 2024年 | 185回 | −3.9pips | | ADX 25以上 | 2025年 | 134回 | −1,572.2pips | | ADX 25以上 | 2024年 | 126回 | +218.7pips | | ADX 30以上 | 2025年 | 94回 | −972.7pips | | ADX 30以上 | 2024年 | 88回 | −231.8pips | | 東京時間 | 2025年 | 164回 | +519.4pips | | 東京時間 | 2024年 | 162回 | +2,139.6pips | | ロンドン・NY時間 | 2025年 | 207回 | +1,073.5pips | | ロンドン・NY時間 | 2024年 | 218回 | +2,154.3pips | ADXは3水準とも両年で基準を上回りませんでした。ロンドン・NY時間だけは、2025年+1,052.5から+1,073.5pips、2024年+2,071.7から+2,154.3pipsへ、わずかながら両年で改善しています。ただし時間帯まで含めて一番良い設定と断定するには、別の通貨ペアと年が必要です。 | 決済ルール | 年 | 年間損益 | |---|---:|---:| | 反対側の1σだけ | 2025年 | +1,052.5pips | | 反対側の1σだけ | 2024年 | +2,071.7pips | | 損切り30・利確60 | 2025年 | +1,687.6pips | | 損切り30・利確60 | 2024年 | +1,467.1pips | | 損切り50・利確100 | 2025年 | +1,157.3pips | | 損切り50・利確100 | 2024年 | +1,896.2pips | | 損切り100・利確200 | 2025年 | +1,340.5pips | | 損切り100・利確200 | 2024年 | +1,987.6pips | | 24本で時間決済 | 2025年 | +989.1pips | | 24本で時間決済 | 2024年 | +1,559.0pips | 固定の損切り・利確はエントリー約定価格から指定pips離れた約定水準です。到達時はその水準どおりに決済し、決済側のコストをもう一度差し引きません。 固定の損切り・利確は2025年を改善しても2024年を下げ、両年で基準を上回るものはありませんでした。 スプレッドを変えると、2025年の期間20・1σは0pipsで+1,135.6pips、0.3pipsで+1,052.5pips、1pipsで+858.6pips、2pipsで+581.6pips、3pipsで+304.6pipsでした。単純計算の損益分岐は**約4.10pips**です。1σは取引数が多いため、低コストであることも設定の一部です。 [ボリンジャーバンドの期間は何本がいいか](/ja/blog/bollinger-band-period-settings)では、期間5〜200を比較し、いちばん良いバー数が時間足によって変わるかを測っています。逆張りと順張りそのものを比較した結果は[ボリンジャーバンドの315通り検証](/ja/blog/bollinger-band-settings)にまとめています。[RSIの期間と30・70を比べた検証](/ja/blog/rsi-settings)も、逆張りの高い勝率が高い損益を意味しませんでした。バンド幅を価格に対する割合で置く方法は[移動平均線乖離率の検証](/ja/blog/ma-disparity-settings)、条件を自分の通貨ペアで組み直す手順は[プログラミングなしのバックテスト](/ja/blog/backtest-without-coding)で確認できます。 ## 補足 - 米ドル／円だけを対象にし、比較した独立年は2024年と2025年の2年です。他の通貨ペアや期間へそのまま移る保証はありません - 期間と偏差の候補から最大値を選ぶ行為には、多重比較による上振れが含まれます。確認年も完全に未使用の将来データではありません - 3σ以上は取引数が少なく、PFと勝率の幅が大きくなります。4σの年間平均2.3回を、260.4回ある1σと同じ確かさでは比べられません - バンドタッチは高値・安値、ブレイクは終値で判定しています。ヒゲ抜けをブレイクに含めるルールでは結果が変わります - 約定は足の終値、スプレッドは常時固定です。実際の約定価格と変動スプレッドは再現していません - 逆張りの主検証は入口偏差だけを比べるため、決済を反対側の1σへ固定しました。出口も最適化する検証とは目的が異なります --- ### CCI「±100で逆張り」は勝てる？2,448通り検証した結果 URL: https://formiq.jp/ja/blog/cci-settings Language: ja Published: 2026-08-27 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: CCI, オシレーター, 逆張り, 順張り, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/cci-settings CCI（Commodity Channel Index）は、価格がその平均からどれだけ離れているかを、離れ方の平均で割って目盛りにしたオシレーターです。**±100の2本の線が引かれていて、そこを超えたら「行き過ぎ」**という説明とセットで紹介されます。 ただし、その先の使い方は2つに割れています。**±100の外側で反対方向に入る逆張り**と、**±100を抜けた方向に入る順張り**です。同じ線を見て、正反対のことをします。どちらが正しいのかは、指標の説明からは決まりません。 もう一つ、あまり触れられない前提があります。**±100という線は「珍しい」ことになっている**、という前提です。CCIの式に入っている0.015という定数は、読み取り値のおおむね7〜8割が±100の内側に収まるように選ばれた、と説明されます。逆に言えば、外側は2〜3割のはずです。 そこから測りました。**±100の外側は、年の約4割でした。** ## CCIは何を測っているのか 計算は3段です。 1. **典型価格** =（高値＋安値＋終値）÷3 2. **平均偏差** = 直近N本の典型価格が、その平均からどれだけ離れているかの平均 3. **CCI** =（典型価格 − 平均）÷（0.015 × 平均偏差） 分子は「平均からどれだけ離れたか」、分母は「ふだんどれくらい離れるものか」です。**ふだんの離れ方の何倍離れているかを測っている**、と読めます。 ここに恣意的な選択が3つ入っています。典型価格を使うこと、標準偏差ではなく平均偏差を使うこと、そして0.015という定数です。順に測りました。 ### 定数0.015の変更は、判定水準の変更と同じ 0.015は割る数なので、変えると線全体が拡大縮小するだけです。**別の定数にすることは、別の水準で読むことと同じ**であって、別の指標にはなりません。 | 定数 | 0.015で読むと | 最大誤差 | |---|---|---| | 0.010 | 水準66.7 | 1.1e-13 | | 0.015 | 水準100 | 0 | | 0.020 | 水準133.3 | 5.7e-14 | | 0.030 | 水準200 | 0 | 誤差は浮動小数点の丸めの桁です。**定数と判定水準を同じ比率で変えても、売買条件は変わりません。** ### 期間がCCIの上限を決める 平均からのずれは合計すると必ず0になるので、プラス側の合計とマイナス側の合計は等しくなります。1本の足が持てるのはプラス側の全部までで、そこから**|CCI| ≤ 期間 ÷ 0.03** という上限が出ます。 | 期間 | 上限 | 2025年の実測最大（1時間足） | |---|---|---| | 7 | 233.3 | 233.3 | | 10 | 333.3 | 333.3 | | 14 | 466.7 | 466.7 | | 20 | 666.7 | 507.2 | | 30 | 1000.0 | 596.4 | | 50 | 1666.7 | 592.1 | **上限は理屈だけの話ではなく、実際に触ります。** 時間足3種×期間6種の18通りのうち9通りで、その年の最大値が上限とぴったり一致しました。 実務上の意味は一つです。**期間を短くすると、水準を上げても届かなくなります。** このスイープでも、期間7で±250を待つ設定は、時間足と期間の12区分すべてで7通りずつが1回も発動せず、取引数0で終わりました。CCI期間によって到達可能な水準の上限が決まるため、期間と水準は独立して選べません。 ### 平均偏差か標準偏差か 分母を標準偏差に差し替えて比べました（1時間足・2025年）。 | 期間 | 相関 | ±100の内外が食い違う足 | ゼロの上下が食い違う足 | |---|---|---|---| | 7 | 0.9961 | 9.73% | 0% | | 14 | 0.9950 | 9.11% | 0% | | 50 | 0.9957 | 9.57% | 0% | **ゼロの上下は1本も食い違いません。** 分母は絶対に負にならないので、どんな散らばりの測り方を使っても符号は変わらないからです。分母が動かせるのは目盛りだけで、±100の内外は1割の足で入れ替わります。 ## CCIの設定と読み方 主要な環境には最初から入っています。CCIは**価格チャートの下に別枠で表示される**タイプで、ローソク足には重なりません。 | 環境 | 追加する場所 | |---|---| | MT4 / MT5 | 挿入 → インディケータ → オシレーター → Commodity Channel Index | | TradingView | インジケーター検索から Commodity Channel Index | | ブラウザ（Formiq） | インジケーター一覧から「CCI」 | ### 設定ダイアログに並ぶ項目 | 項目 | 初期設定 | 意味 | |---|---|---| | Period（期間） | 14 | 何本ぶんの価格を使うか。20を使う流派もある | | Apply to（適用価格） | Typical price (HLC/3) | 高値・安値・終値の平均。終値ではない | | Levels（水準線） | −100 と +100 | 枠の中に引かれる2本の横線。売買の判断に使う | **MT4で引かれる水準線は初期状態では0本のことがあります。** その場合は「レベル表示」タブで −100 と 100 を手で追加します。この2本が無いと、この記事の話は画面上で確認できません。 ### 画面に出るもの 別枠の中を、1本の線が上下します。**RSIやストキャスティクスと違って上限も下限もありません。** 0を挟んで、大きく上に行くこともあれば下に行くこともあります。 | 線の位置 | 一般的な読み方 | |---|---| | +100より上 | 買われすぎ、または上昇の勢いが強い | | −100〜+100 | 中立。何もしない | | −100より下 | 売られすぎ、または下降の勢いが強い | | 0を上下に抜ける | 上向き／下向きの転換 | 「または」と書いたのは、±100の外側から戻る動きを狙う逆張りと、±100を抜けた方向へ売買する順張りがあるからです。この記事では、逆張りと±100の水準抜け順張りを別々に測っています。 ### 売買のルールとして使われている形 | 呼び方 | 入り方 | 考え方 | |---|---|---| | 逆張り | −100より下で買い、+100より上で売り | 行き過ぎたから戻る | | 順張り（±100の抜け） | +100を上に抜けたら買い、−100を下に抜けたら売り | 抜けたということは勢いがある | | ゼロライン | 0を上に抜けたら買い、下に抜けたら売り | 向きが変わった | **同じ±100の線を見て、逆張りと順張りは正反対の売買をします。** どちらが正しいかは指標の説明からは決まらないので、この記事では両方を同じ条件で回して比べました。 ### 期間と水準は独立して設定できない 期間と水準の2つに見えますが、後半で書くとおり**期間は水準の上限も決めています**（期間7のCCIは±233.3までしか動きません）。適用価格を終値に変えると別の線になり、±100の内外が食い違う足が期間7で14.68%、期間50で5.00%出ます。無視できる差ではありません。 ## 検証の条件 数字を見る前に、何をどう測ったかを書いておきます。この表と同じ設定を入れれば、同じ結果が出ます。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 読み方 | 逆張り（±水準の外側で反対方向）／ ±水準の抜け（抜けた方向）／ ゼロライン（水準を上下に抜ける） | | 期間 | 7・10・14・20・30・50 | | 水準 | 逆張り：入口100・150・200・250 × 出口0・50・100／ 抜け：50・100・150・200・250／ ゼロライン：−100・−50・0・50・100 | | 組み合わせ | 1区分あたり204通り。時間足3種 × 期間4区分で合計2,448通り | | 決済 | 反対シグナルまで。逆張りだけは出口の水準を別に指定する | | 損切り・利確 | 使わない（別セクションで測定） | | スプレッド | 0.3pips固定。約定は終値 | | ロット | 0.1ロット | 逆張りは買いと売りのどちらか一方しか作りません（「−100より下は買い」は買いしか作らない）ので、**買い逆張りと売り逆張りを別々の売買ルールとして回しました。** 抜けとゼロラインは対称なので、エントリーと決済に同じルールセットを使います。 ## ±100を抜けた先に何があるか バックテストの前に、値動きだけを見ました。売買ルールも決済も挟まないので、**「±100が何か言っているのか」だけ**が残ります。 期間14のCCIが±100を抜けた足、そして±100の内側に戻った足を全部拾い、そこから10本・20本・50本先の値幅を測ります。ただし生の値幅では意味がありません。2024年のドル円は年間で1,632pips上げているので、**どの足から測っても買い方向の数字は大きくなる**からです。そこで同じ区間の全足の平均値幅を引き、**平均的な足と比べてどれだけ余分に動いたか**を見ます。 | 時間足 | 抜けた直後（10 / 20 / 50本先） | 内側に戻った直後 | 件数 | |---|---|---|---| | 15分足 | +0.46 / +0.93 / +2.04 | −0.86 / −1.01 / −1.92 | 5,776 | | 1時間足 | +2.51 / +3.93 / +2.89 | −2.61 / −3.06 / −3.17 | 1,396 | | 4時間足 | −3.89 / −3.14 / −9.79 | +2.91 / +5.02 / +12.50 | 346 | 2024年と2025年、上抜けと下抜けを合わせた平均です。**15分足と1時間足では、抜けた直後は平均より伸び、戻った直後は伸びません。** 年・時間足・先読み本数で分けた24通りのうち、抜けが平均を上回ったのは22通り、戻りが下回ったのは24通り全部でした。 **4時間足では、±100を抜けた後の値幅が平均を下回り、±100の内側へ戻った後の値幅が平均を上回りました。** 年・先読み本数で分けた12通りのうち、この関係になったのはそれぞれ6通りと5通りで、対象も346件しかありません。 つまり±100は「行き過ぎ」ではなく、**「まだ続く」の側に寄っています**。少なくとも15分足と1時間足では。 ## 逆張りと順張り、どちらで使うか 逆張りは6通りすべてで赤字、抜けとゼロラインは黒字が多数でした。同じ期間14・水準100を、3通りに読ませます。逆張りは買い側と売り側の合計です。 | 時間足 | 年 | 逆張り | ±100の抜け | ゼロライン | |---|---|---|---|---| | 15分足 | 2025 | −1,175.2 | **+643.4** | −1,136.4 | | 15分足 | 2024 | −805.2 | −203.2 | −910.7 | | 1時間足 | 2025 | −1,193.7 | **+424.8** | **+2,294.4** | | 1時間足 | 2024 | −2,249.8 | **+1,839.9** | **+2,579.8** | | 4時間足 | 2025 | −156.2 | **+577.5** | **+569.5** | | 4時間足 | 2024 | −995.8 | **+863.8** | **+1,363.3** | **逆張りは6通りすべてで赤字**、抜けは6通り中5通り、ゼロラインは6通り中4通りで黒字でした。値動きだけを見て出た結論と、売買した結果は一致しています。 取引数も添えておきます。1時間足2025年で、逆張りは買い210回・売り212回、抜けは321回、ゼロラインは693回です。 ### 逆張りの中身 逆張りが負ける形はいつも同じでした。**勝率は高く、勝ちが小さく、負けが大きい。** | 読み方 | 時間足 | 平均勝率 | 平均利益 | 平均損失 | 2025年に黒字だった設定 | |---|---|---|---|---|---| | 逆張り・買い | 1時間足 | 59.9% | +44.4 | −68.7 | 27/69 | | 逆張り・売り | 1時間足 | 56.7% | +44.0 | −83.1 | 2/69 | | ±100の抜け | 1時間足 | 40.1% | +101.6 | −71.8 | 20/29 | | ゼロライン | 1時間足 | 31.2% | +59.5 | −23.3 | 27/30 | 売買の生きている設定だけの平均です。**勝率で並べると順序が完全に逆になります。** これはこのシリーズで18本目の同じ結果です。 逆張りの売り側はとくにひどく、**15分足では2025年に黒字の設定が69通り中0通り**でした。2024年のドル円が1,632pips上げた年だったことを差し引いても、2025年は年間−56pipsの横ばいです。相場の向きだけでは説明が付きません。 ## 2年とも勝てた組み合わせはどれか 1時間足のゼロラインが22/30で最多でした。片方の年だけの結果は選び方の問題になるので、2024年と2025年の両方で黒字だった設定を数えます。 | 読み方 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---| | 逆張り・買い | 2/72 | 8/72 | 22/72 | | 逆張り・売り | 0/72 | 0/72 | 1/72 | | ±100の抜け | 17/30 | 20/30 | 11/30 | | ゼロライン | 6/30 | 22/30 | 13/30 | **1時間足のゼロラインが22/30で最も残りました。** 逆張りの売りは216通り中1通りです。 4時間足は逆張りの買いが22/72と、逆張りの中では唯一まともな数字を出しています。ただし2025年だけを見ると56/69が黒字で、2024年と重ねると22通りまで落ちます。**片方の年で6割が黒字になる読み方は、翌年に残るとは限りません。** ## 期間はいくつがよいか 売買ルールごとに答えが違い、期間だけでは決まりませんでした。期間別の成績を、ゼロライン、±100の抜け、逆張りに分けて示します。同じ期間でも、売買ルールによって年間損益と黒字設定数が異なるためです。 **ゼロライン・15分足（期間別、2025年／2024年に黒字だった数）** | 期間 | 2025年の平均 | 平均取引数 | 黒字 2025 | 黒字 2024 | |---|---|---|---|---| | 7 | −1,121 | 4,551 | 0/5 | 0/5 | | 14 | −730 | 3,030 | 0/5 | 0/5 | | 30 | +328 | 1,956 | 3/5 | 3/5 | | 50 | **+1,255** | 1,419 | 5/5 | 4/5 | **15分足のゼロクロスは、期間を伸ばすほど良くなります。** 理由は取引数です。期間7・水準0は年4,533回売買していて、スプレッド0.3pipsだけで1,359.9pipsを払います。実際の損益は−1,699.3pipsなので、**取引しなければ払わずに済んだぶんが損失の8割**です。 同じ設定を期間50にすると1,469回まで減り、2025年は+614.1pips、2024年は+189.5pipsになります。 **±100の抜け・1時間足（水準別）** | 水準 | 2025年の平均 | 平均取引数 | 両方の年で黒字 | |---|---|---|---| | ±50 | +1,551 | 404 | 6/6 | | ±100 | +1,395 | 292 | 6/6 | | ±150 | +1,519 | 182 | 6/6 | | ±200 | −936 | 85 | 2/6 | | ±250 | −675 | 29 | 0/6 | ±50から±150までは平らで、**±200から先で崩れます。** ここも取引数の問題で、±250は年29回しか発動しません。 **ゼロライン・1時間足（水準別）** | 水準 | 2025年の平均 | 2024年の平均 | 両方の年で黒字 | |---|---|---|---| | −100 | +166 | +1,892 | 3/6 | | −50 | +1,325 | +2,309 | 6/6 | | 0 | +1,548 | +1,817 | 6/6 | | +50 | +1,763 | +653 | 4/6 | | +100 | +1,710 | −425 | 3/6 | **ゼロラインをゼロから動かす理由は見つかりませんでした。** −50と0だけが両方の年で6/6です。+50と+100は2025年のほうが良く見えますが、2024年で落ちます。 ## 去年よかった読み方は今年も勝てるか 4通り中2通りしか、当てた年の中央値を上回りませんでした。このシリーズの中心にある検定です。 | 読み方 | 時間足 | 2024年の設定 | 2024年の損益 | 2025年の損益 | 2025年の中央値 | |---|---|---|---:|---:|---:| | ±100の抜け | 1時間足 | 20@±100 | +3,393.7pips | +1,048.5pips | +1,113.0pips | | ゼロライン | 1時間足 | 30@−50 | +3,046.7pips | +1,729.0pips | +1,294.1pips | | ±100の抜け | 15分足 | 30@±50 | +3,250.8pips | +1,717.2pips | +549.8pips | | 逆張り・買い | 4時間足 | 10/200→100 | +946.1pips | +176.7pips | +387.5pips | 4通り中2通りが2025年の中央値を上回りました。1時間足の抜けは中央値+1,113.0pipsに対して+1,048.5pips、4時間足の逆張りは中央値+387.5pipsに対して+176.7pipsでした。 なお4時間足の抜けで2025年にいちばん良かったのは水準±250・期間50で+1,777.3pipsですが、**取引数は2回**です。この手の数字を設定選びの根拠にはできません。 ## ゼロラインは何を意味するのか 移動平均線クロスそのものです。 CCIの分母は 0.015 × 平均偏差 で、平均偏差は絶対に負になりません。したがって**CCIの符号は（典型価格 − その単純移動平均）の符号と完全に一致します。** 実測でも、期間7から50まで、符号の不一致は0件、クロスした足の不一致も0件でした。 つまり**CCIのゼロクロスは、価格が自分の移動平均を抜けたことを別の目盛りで見たもの**です。この記事でいちばん成績のよかった読み方は、CCIでなくても再現できます。 では、CCIであることに意味はないのか。**典型価格を使う点だけは残ります。** 同じルールを終値で読ませたもの（SMA(1)とSMA(N)のクロス。SMA(1)は終値そのもの）と、36通りで突き合わせました。 | | CCIのゼロクロス | 終値の移動平均線クロス | |---|---|---| | 取引数（1時間足・2025年・期間14） | 693 | 885 | | 勝率 | 30.45% | 26.44% | | 損益 | +2,294.4 | +1,801.6 | 36通り全部でこうなりました。 - **取引数が少ないのは36通り中36通り** - **勝率が高いのも36通り中36通り**（差は1.5〜8.0ポイント） - **損益で上回ったのは36通り中25通り** **（高値＋安値＋終値）÷3は、だましを減らします。** 1本の足の終値だけを見ると、ヒゲの先で平均線を抜けて戻る足を数えてしまう。典型価格はその足の中身を平均するので、抜けたことにならない。これは36通り全部で一貫していて、例外がありません。 ただし**減った取引が良い取引だったこともあります。** 典型価格が損益で負けた11通りでは、除外した取引を残したほうが利益につながりました。取引を減らす効果は確実でも、損益が改善するとは限りません。 ## フィルター・損切り利確・コスト ### ADXフィルターは順張りを壊す トレンドの強さで絞れば順張りが良くなりそうに見えますが、逆でした。 | 読み方 | 時間足 | 年 | ADX条件 | 取引数 | 損益 | |---|---|---:|---|---:|---:| | ±100の抜け | 15分足 | 2025 | なし | 1,298回 | +643.4pips | | ±100の抜け | 15分足 | 2025 | 20以上 | 854回 | −183.8pips | | ±100の抜け | 15分足 | 2025 | 25以上 | 577回 | −618.9pips | | ±100の抜け | 15分足 | 2025 | 30以上 | 358回 | −229.5pips | | ゼロライン | 1時間足 | 2025 | なし | 693回 | +2,294.4pips | | ゼロライン | 1時間足 | 2025 | 20以上 | 399回 | +786.0pips | | ゼロライン | 1時間足 | 2025 | 25以上 | 285回 | +617.2pips | | ゼロライン | 1時間足 | 2025 | 30以上 | 179回 | −19.7pips | **強い順張りに強さのフィルターを足すと、条件が重複します。** GMMAの検証でも同じことが起きました。これで3回目です。 4時間足の逆張りだけはADXで良くなりますが（2025年 +796.2 → +1,091.2、ただし16回）、**2024年は同じフィルターで+26.6 → −527.9** です。年をまたぎませんでした。 ### 時間帯 東京時間（UTC 0〜8）は±100の抜けに効きました。15分足2025年 +643.4 → +877.0、2024年 −203.2 → **+1,718.7**。1時間足も2025年 +424.8 → +1,146.8。ただし2024年の1時間足は +1,839.9 → +1,746.0 で、4通り中3通りです。ゼロラインには効きませんでした（1時間足2025年 +2,294.4 → +571.3）。 ### 損切り・利確 ゼロライン・1時間足で、損切り100・利確200を足すと2025年は+2,811.4まで伸びますが、2024年は+2,095.4に落ちます（ベース+2,579.8）。時間決済24本も2025年は+3,588.2、2024年は+2,303.0です。**どれも片方の年でしか良くなりません。** ### コスト 損益はスプレッドに対してきれいに直線でした。**失ったpips＝取引回数×スプレッド**が、4つの設定すべてで誤差なく成立します。 | 設定 | 取引数 | スプレッド0 | 0.3 | 1.0 | 損益分岐 | |---|---|---|---|---|---| | ±100の抜け・15分足 | 1,298 | +1,032.8 | +643.4 | −265.2 | 0.80pips | | ±100の抜け・1時間足 | 321 | +521.1 | +424.8 | +200.1 | 1.62pips | | ゼロライン・1時間足 | 693 | +2,502.2 | +2,294.4 | +1,809.3 | **3.61pips** | **1時間足のゼロクロスの損益分岐3.61pipsは、このシリーズで最も高い数字です。** 15分足の抜けは0.80pipsしかなく、スプレッドが1pips開く環境では成立しません。 ## 関連記事 - [RSIの期間と水準](/ja/blog/rsi-settings)：設定水準に到達したら売買する逆張りと、水準を抜けた方向へ売買する順張りを分けて測っています - [ボリンジャーバンドの偏差](/ja/blog/bollinger-band-settings)：バンドを「触れたら逆張り」と「抜けたら順張り」に分けた最初の記事です - [GMMAの本数](/ja/blog/gmma-settings)：トレンド系にADXフィルターを足すと壊れる例が、こちらにもあります - [アリゲーターの前方シフト](/ja/blog/alligator-settings)：式の中の恣意的な選択を1つずつ差し替える型を、時間軸のずらしに当てています - [移動平均線クロスの期間](/ja/blog/moving-average-cross-settings)：この記事のゼロラインが行き着く先です - [サイコロジカルラインの75%](/ja/blog/psychological-line-settings)：同じ「その指標である必要があるのか」を、値幅を完全に捨てた指標で測っています - [RCI「−80で買い」は本当に勝てる？2,484通り検証](/ja/blog/rci-settings)：この記事と同じ5つの読み方のうち4つを、順位相関の指標で比べています - [平均足の設定](/ja/blog/heikin-ashi-settings)：平均足の色も、結局は価格と移動平均のクロスでした - [移動平均線乖離率の±5%](/ja/blog/ma-disparity-settings)：平均からの距離を、割る数を置かずに割合のまま読んだ場合です ## 補足 - **通貨ペアはドル円だけ、期間は2年だけ**です。CCIは商品先物のために作られた指標なので、対象が変われば±100の外にいる時間の割合そのものが変わります - **スプレッドは0.3pips固定**です。実際は指標発表時に開きます。損益分岐0.80pipsの設定は、この前提が崩れると成立しません - **決済は反対シグナルまで**です。損切りと利確を足した場合は別セクションの数字になり、片方の年でしか良くなりませんでした - **4時間足の数字は標本が小さい**（±100の抜けで年40回前後、逆張りで20回前後）。4時間足だけが値動きの計測でも逆を向きましたが、12通り中6通りなので判定できていません - **「典型価格が効く」は取引数と勝率でだけ一貫**していて、損益では36通り中25通りです。ここを「効く」と読むかは、何を見るかによります --- ### パラボリックSARの加速因子と時間足を36通り検証 URL: https://formiq.jp/ja/blog/parabolic-sar-settings Language: ja Published: 2026-08-27 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: パラボリックSAR, Parabolic SAR, インジケーター, トレンド, 順張り, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/parabolic-sar-settings パラボリックSAR（Parabolic Stop and Reverse）は、ローソク足の上下に点を置き、値動きが続くほど点を価格へ近づけるトレンド指標です。点が価格の下から上へ移れば上昇から下降への転換、上から下へ移れば下降から上昇への転換と読みます。 設定は**増分0.02、上限0.20**が広く使われています。小数が2つ並ぶため、どちらを動かすと何が変わるのかは分かりにくいところです。そこで増分6種と上限6種を組み合わせ、ドル円の15分足・1時間足・4時間足で36通りを測りました。 **1時間足は36通り中23通りが2024年と2025年の両方で黒字**でした。ただし初期設定の0.02・0.20は、4時間足で2025年に+637.7pips、2024年に−108.2pipsです。1時間足でも2025年前半は−1,032.5pips、後半は+1,847.7pipsでした。 ## パラボリックSARは何を測っているのか SARの次の点は、現在の点をトレンド中の極値へ少しずつ近づけて作ります。 **次のSAR = 現在のSAR + 加速係数 ×（極値 − 現在のSAR）** 上昇中の極値はその上昇が始まってからの最高値、下降中は最安値です。新しい高値または安値を作るたびに加速係数を増やし、設定した上限で止めます。点がローソク足に追いついて反対側へ移ると、向きが切り替わります。 初期設定の0.02・0.20なら、加速係数は0.02から始まります。極値を更新するたびに0.04、0.06と増え、9回の更新で0.20です。その後は上限0.20のまま点が価格を追います。 この仕組みにより、増分と上限では変更したときの影響が異なります。 - **増分**：最初の追いつく速さと、極値更新後に加速する幅を同時に変える - **上限**：長く続いた値動きの後半で、点がどこまで速く追いつくかを制限する 点は損切り価格の候補としても紹介されますが、この記事で測るのは**点が反対側へ切り替わった足の終値で売買するルール**です。点そのものの価格で約定する検証ではありません。 ## パラボリックSARの設定と読み方 追加すると、**ローソク足に重なる1列の点**が表示されます。価格の下にある点と上にある点が交互に続き、下の別枠や水準線は使いません。 | 環境 | 追加する場所 | |---|---| | MT4 / MT5 | 挿入 → インディケータ → トレンド → Parabolic SAR | | TradingView | インジケーター検索から Parabolic SAR | | ブラウザ（Formiq） | インジケーター一覧から「パラボリックSAR」 | ### 設定ダイアログに並ぶ2項目 | 項目 | 初期設定 | 意味 | |---|---:|---| | Step（増分） | 0.02 | 加速係数の初期値であり、極値を更新するたびに足す値 | | Maximum（上限） | 0.20 | 加速係数がこれ以上は大きくならない上限 | 0.02は価格の2%ではありません。**現在の点から極値まで残っている距離の2%を進む**という意味です。上限0.20なら、最大でも残り距離の20%ずつ進みます。 ### 画面に出るもの 表示されるのはSARの点1系列だけです。一般には点が価格の下なら上昇、価格の上なら下降と読みます。色は環境ごとに変えられますが、計算に色は関係しません。RSIの30・70のような水準線もありません。 ### 売買ルールにすると反転だけを見る この記事が測るルールは次の1つです。 | 点の切り替わり | 売買 | |---|---| | 価格の上から下へ移る | 買い。次に上へ移るまで保有 | | 価格の下から上へ移る | 売り。次に下へ移るまで保有 | Stop and Reverseという名前どおり、反対のシグナルで現在のポジションを閉じ、その向きへ持ち替えます。買いと売りは同じ形で起きます。 ### 増分と上限のどちらを変更すると結果が大きく変わるか 早く反応させたいなら増分を大きくします。ただし点が価格へ早く近づくため、転換判定と売買回数も増えます。上限は長く続くトレンドの後半だけに効くため、増分ほど直接には回数を動かしません。 実測でも、1時間足2025年の平均取引数は増分0.005で215.3回、0.05で777.7回でした。一方、上限は0.30で608.8回、0.50で609.8回です。**回数を動かしていたのは主に増分**でした。 ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31。比較用に2024年も同条件。2025年は前半・後半にも分割 | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 増分 | 0.005 / 0.01 / **0.02** / 0.03 / 0.04 / 0.05 | | 上限 | 0.05 / 0.10 / 0.15 / **0.20** / 0.30 / 0.50 | | 組み合わせ | 増分6種 × 上限6種 = 36通り。3時間足 × 4期間区分で432回実行 | | 買い | SARの点が価格の上から下へ切り替わった足の終値 | | 売り | SARの点が価格の下から上へ切り替わった足の終値 | | 決済 | 反対側への切り替わりまで。エントリーと決済に同じルールセットを使用 | | 損切り・利確 | 使わない。別の節で追加検証 | | スプレッド | 0.3pips固定。スリッページ0 | | ロット | 0.1ロット | | 測定 | Formiqのバックテスト機能で実行し、各取引の約定価格からpipsを再計算 | ## どの時間足で使えるか 1時間足で23通りが両年黒字でした。 | 時間足 | 2025年に黒字 | 2025年の中央値 | 2024年に黒字 | 2024年の中央値 | 両方の年で黒字 | |---|---:|---:|---:|---:|---:| | 15分足 | 11/36 | −433.3pips | 21/36 | +248.8pips | 11/36 | | **1時間足** | **29/36** | **+928.8pips** | **30/36** | **+1,385.1pips** | **23/36** | | 4時間足 | 8/36 | −372.4pips | 32/36 | +679.4pips | 4/36 | 1時間足は黒字数と中央値の両方で最も揃いました。15分足は取引回数が多く、初期設定で年2,000回を超えます。4時間足は2024年に32通りが黒字なのに、2025年は8通りまで減りました。 1時間足でも、2024年に年間損益が最大だった0.03・0.10は、2024年の+3,910.5pipsから2025年の+981.4pipsへ減りました。黒字設定数が多いことと、個別設定の年間損益が同じ水準で続くことは別です。 ## 増分と上限は何を変えるか 増分は取引回数を3.6倍にしました。上限6種を平均し、増分だけで1時間足の成績をまとめます。 | 増分 | 2025年 平均回数 | 2025年 平均損益 | 2024年 平均損益 | |---:|---:|---:|---:| | 0.005 | 215.3回 | +1,240.3pips | −1,138.4pips | | 0.01 | 342.5回 | −144.9pips | +1,535.3pips | | **0.02** | 504.2回 | +573.4pips | +2,333.2pips | | 0.03 | 611.5回 | +1,246.6pips | +2,335.3pips | | 0.04 | 701.7回 | +597.2pips | +1,749.9pips | | 0.05 | 777.7回 | +1,459.8pips | +923.4pips | 増分を0.005から0.05へ10倍にすると、平均回数は215.3回から777.7回へ**3.6倍**になりました。ところが損益は単調に増えません。2025年に+1,240.3pipsだった0.005は、2024年に−1,138.4pipsです。 上限は0.05から0.10へ上げると、2025年の平均回数が316.3回から479.8回へ増えます。その後は鈍り、0.30で608.8回、0.50で609.8回でした。**上限を高くし続けても、実際の加速係数がそこへ届かなければ結果はほとんど変わりません。** ## 初期設定の0.02・0.20のままでいいか 初期設定がどこに位置するかを見ます。 | 時間足 | 年 | 取引回数 | 勝率 | PF | 年間損益 | |---|---:|---:|---:|---:|---:| | 15分足 | 2025年 | 2,290回 | 37.51% | 0.984 | −375.3pips | | 15分足 | 2024年 | 2,107回 | 39.16% | 1.040 | +870.8pips | | **1時間足** | **2025年** | **547回** | **40.04%** | **1.074** | **+886.8pips** | | **1時間足** | **2024年** | **531回** | **38.79%** | **1.240** | **+2,683.6pips** | | 4時間足 | 2025年 | 138回 | 46.38% | 1.110 | +637.7pips | | 4時間足 | 2024年 | 130回 | 35.38% | 0.982 | −108.2pips | 初期設定が両方の年で黒字だったのは1時間足だけです。年間損益は2025年の+886.8pipsから2024年の+2,683.6pipsまで開きました。 4時間足の初期設定は2025年に+637.7pips、2024年に−108.2pipsでした。片方の年だけでは、初期設定を使うかどうかを決められません。 ## 去年よかった設定は今年も通用するか 片方の年にいちばん良かった設定を、もう片方の年へそのまま移します。 | 時間足 | 選定年 | 設定 | 選定年の年間損益 | もう片方の年の年間損益 | |---|---:|---|---:|---:| | 15分足 | 2025年 | 0.02/0.05 | +1,147.3pips | 2024年：+350.9pips | | 15分足 | 2024年 | 0.005/0.05 | +2,852.8pips | 2025年：+678.8pips | | 1時間足 | 2025年 | 0.05/0.10 | +2,062.7pips | 2024年：+2,969.8pips | | 1時間足 | 2024年 | 0.03/0.10 | +3,910.5pips | 2025年：+981.4pips | | 4時間足 | 2025年 | 0.02/0.20 | +637.7pips | 2024年：−108.2pips | | 4時間足 | 2024年 | 0.005/0.05 | +2,369.4pips | 2025年：−452.2pips | 15分足と1時間足では、選定した設定がもう片方の年でも黒字でした。一方、4時間足ではどちらの選び方ももう片方の年に赤字です。**1つの設定が両年黒字だったことと、他の設定でも同じ結果になることは別です。** 年内でも同じです。1時間足の初期設定は2025年前半に281回・−1,032.5pips、後半は266回・+1,847.7pipsでした。年間+886.8pipsは、符号の違う2つの半年を合計した数字です。 ## 勝率4割で黒字になるのはなぜか 初期設定の1時間足2025年は547回、勝率40.04%、PF 1.074で+886.8pipsでした。勝ち取引の平均は+58.71pips、負け取引は−36.49pipsです。**4割しか勝たなくても、勝ち幅が負け幅より大きいため黒字**になっています。 初期設定は2024年に勝率38.79%で+2,683.6pips、2025年に勝率40.04%で+886.8pipsでした。勝率が1.25ポイント高かった2025年のほうが、年間損益は1,796.8pips小さくなっています。 ## ADXを足すと改善するか 1時間足の初期設定に、エントリー時だけフィルターを追加します。 | 条件 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | SARのみ | 2025年 | 547回 | 40.04% | +886.8pips | | SARのみ | 2024年 | 531回 | 38.79% | +2,683.6pips | | ADX 20以上 | 2025年 | 357回 | 38.66% | −814.5pips | | ADX 20以上 | 2024年 | 351回 | 36.47% | +979.4pips | | ADX 25以上 | 2025年 | 263回 | 41.83% | −601.6pips | | ADX 25以上 | 2024年 | 243回 | 38.27% | +140.2pips | | ADX 30以上 | 2025年 | 178回 | 38.20% | −906.7pips | | ADX 30以上 | 2024年 | 158回 | 38.61% | +431.5pips | | 東京時間だけ | 2025年 | 207回 | 42.51% | +158.2pips | | 東京時間だけ | 2024年 | 202回 | 38.12% | +585.2pips | | ロンドン・NY時間 | 2025年 | 307回 | 38.76% | +982.7pips | | ロンドン・NY時間 | 2024年 | 311回 | 40.84% | +2,465.3pips | ADXは3水準すべてで2025年を赤字にしました。2024年も基準の+2,683.6pipsから大きく下がっています。トレンド転換を読むSARに、トレンドの強さを条件として重ねても改善しませんでした。 ロンドン・NY時間は2025年を+982.7pipsへ少し増やしますが、2024年は+2,465.3pipsへ減らします。両方の年で基準を上回るフィルターはありません。 ## 損切りや時間決済で改善するか | 決済 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | 反対シグナルまで | 2025年 | 547回 | 40.04% | +886.8pips | | 反対シグナルまで | 2024年 | 531回 | 38.79% | +2,683.6pips | | 損切り30・利確60 | 2025年 | 459回 | 38.13% | +1,317.5pips | | 損切り30・利確60 | 2024年 | 445回 | 36.18% | +823.0pips | | 損切り50・利確100 | 2025年 | 471回 | 39.49% | +1,047.9pips | | 損切り50・利確100 | 2024年 | 464回 | 37.07% | +970.2pips | | 損切り100・利確200 | 2025年 | 541回 | 40.30% | +1,600.4pips | | 損切り100・利確200 | 2024年 | 515回 | 38.64% | +2,302.3pips | | 損切り50・利確50 | 2025年 | 470回 | 44.89% | +1,092.1pips | | 損切り50・利確50 | 2024年 | 468回 | 41.88% | +138.0pips | | 24本で時間決済 | 2025年 | 539回 | 40.07% | +1,187.7pips | | 24本で時間決済 | 2024年 | 522回 | 38.89% | +2,153.1pips | 追加した5つの決済はすべて両方の年で黒字でした。ただし、基準を両方の年で上回るものはありません。損切り100・利確200は2025年を+1,600.4pipsへ増やしましたが、2024年は+2,302.3pipsへ減らしました。 ## スプレッドはどこまで耐えられるか 1時間足・初期設定・2025年のスプレッドだけを変えます。 | スプレッド | 取引数 | 損益 | |---:|---:|---:| | 0.0pips | 547回 | +1,050.9pips | | 0.3pips | 547回 | +886.8pips | | 0.6pips | 547回 | +722.7pips | | 1.0pips | 547回 | +503.9pips | | 1.5pips | 547回 | +230.4pips | | 2.0pips | 547回 | −43.1pips | | 3.0pips | 547回 | −590.1pips | 失う額は取引回数×スプレッドに一致します。スプレッド0の+1,050.9pipsを547回で割った**損益分岐スプレッドは1.92pips**です。 15分足の初期設定は年2,290回売買して−375.3pipsでした。小さな時間足で点が頻繁に切り替わるほど、増分の違いより先にコストが効きます。 ## 関連記事 - [スーパートレンドの設定検証](/ja/blog/supertrend-settings)：同じトレンド転換系で、期間と倍率のどちらが効くかを36通り測定 - [アリゲーターの使い方と設定](/ja/blog/alligator-settings)：点ではなく3本の線でトレンドを読む指標 - [移動平均線クロスの設定検証](/ja/blog/moving-average-cross-settings)：反転まで保有する同じ決済ルールで期間を比較 - [プログラミングなしでバックテストする方法](/ja/blog/backtest-without-coding)：この記事と同じ条件を画面で組む手順 - [ドル円の設定は他の通貨ペアで通用する？3,024通り検証](/ja/blog/settings-across-pairs)：この設定は2023年の15分足でドル円1位でしたが、ユーロドル24位・ポンドドル34位・ユーロ円36位でした ## 補足 - 通貨ペアは米ドル／円だけ、期間は2024年と2025年の2年だけです - パラボリックSARは初期方向、極値を更新する順序、直前2本の高安による補正で実装差が出ます。この数値はFormiqの実装によるものです - 点の価格に逆指値を置くのではなく、点が切り替わった足の終値で約定しています - スプレッドは全期間0.3pips固定、スリッページは0です - 36通りは一般的な初期値の周辺を比べる格子であり、すべての小数を試したものではありません - 4時間足の初期設定は年間130〜138回です。勝率とPFは、1時間足や15分足より少ない取引回数から計算しています --- ### フィボナッチ「61.8%で反発」は本当？FXで2年分検証した結果 URL: https://formiq.jp/ja/blog/fibonacci-retracement-levels Language: ja Published: 2026-08-26 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: フィボナッチ, リトレースメント, 押し目買い, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/fibonacci-retracement-levels フィボナッチ・リトレースメントは、直近の安値から高値（またはその逆）に線を引き、その値幅の23.6%・38.2%・50%・61.8%・78.6%の位置に水平線を並べる描画ツールです。**押し目や戻りがその水準で止まりやすい**という前提で使われます。 では、その5つのうちどれが一番よく止まるのか。実測しました。 ただし5水準だけを比べれば、その中のどれかが必ず最高値になります。そこで押し戻しの深さを10.0%から90.0%まで0.2%刻みにした**401水準**をすべて測り、フィボナッチ比率の反応率が隣接する水準より高いかを確認しました。対象はドル円の2024年と2025年、15分足・1時間足・4時間足で自動検出したスイング6,266本です。 **押し目がその水準で終わった割合は、深さ30%から先では3.5〜4.4%のあいだで平らでした。** フィボナッチ比率5つを前後3%の水準と比べると差は平均−0.037ポイントで、比率でない6水準（+0.046ポイント）と区別が付きません。**効いていたのは「深さ」であって、「数字」ではありませんでした。** ## フィボナッチ・リトレースメントとは 計算は引き算と掛け算だけです。スイングの安値をL、高値をHとすると、深さd%の水準は `H −（H − L）× d/100` に置かれます。61.8%なら、高値から値幅の61.8%だけ下がった価格です。 つまりこのツールが測っているのは、**直近の一方向の動きに対する、戻りの割合**です。値幅そのものではありません。この検証で自動検出したスイングの中央値は15分足で38.4pips、1時間足で85.8pips、4時間足で178.7pips（いずれも2025年）でした。同じ「61.8%」でも、4時間足では110pips戻ってようやく届く水準になります。 設定する項目は、**どの深さを見るか**と、**どの2点を起点と終点にするか**です。この記事では、スイングの検出本数（前後何本より高い足をスイング高値とするか）を3・5・8・13・21本の5通りで変えると、到達率が10ポイント以上変わりました。比率を選ぶ前に、どの条件でスイングを検出するかを決める必要があります。 ### 押し戻しの半分以上は、起点まで戻る 深さの話をする前に、押し戻しがそもそもどう終わるのかを数えました。スイング6,266本の内訳です。 | 終わり方 | 本数 | 割合 | |---|---|---| | 途中で反転して、スイングの終点を更新した | 2,955 | 47.2% | | 起点まで完全に戻され、スイングが消えた | 3,295 | 52.6% | | 200本たっても決着しなかった | 16 | 0.3% | **半分強のスイングは、100%戻されて無くなります。** 「押し目がどこで止まるか」を問う前に、止まらないほうがわずかに多い、というのがこのデータの前提です。 ## フィボナッチをチャートに引く方法 どのプラットフォームにも描画ツールとして入っています。 | 環境 | 手順 | |---|---| | MT4 / MT5 | 挿入メニュー → フィボナッチ → リトレースメント を選び、安値から高値へドラッグする | | TradingView | 描画ツールの一覧からフィボナッチ・リトレースメントを選ぶ | | ブラウザ（Formiq） | 描画ツールに入っている。バックテスト側では直近のスイングを自動で検出して水準を条件にできる | 手で引く場合と自動検出の違いは、**どの2点を選ぶかを人が決めるかどうか**だけです。この記事は自動検出で測っています。前後N本より高い足をスイング高値、安い足をスイング安値とし、高値と安値が交互に並ぶように整理して、直近の1区間を1本のスイングとして扱いました。 **この方式には確定の遅れがあります。** ある足がスイング高値だと分かるのは、その後N本が高値を更新しなかったと確認できたときです。つまり線が引けるのは、天井を打ってからN本あとになります。この遅れは後のセクションで数字にしています。 ## 検証の条件 数字を見る前に、何をどう測ったかを書いておきます。この表と同じ設定を入れれば、同じ結果が出ます。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | スイング検出 | 前後5本より高い（安い）足を高値（安値）とする。高値と安値が交互になるよう整理し、直近の1区間を使う。感度検証として前後3・8・13・21本でも実行 | | スイング本数 | 6,266本（15分足4,752 / 1時間足1,176 / 4時間足338。2年合計） | | 深さの刻み | 10.0%〜90.0%を0.2%刻み＝401水準。23.6・38.2・50・61.8・78.6はこの刻みにちょうど乗る | | 反転率 | その深さに到達したスイングのうち、そこから5%以内で押し戻しが終わったものの割合。感度検証として2.5%と10%でも計算 | | 売買の検証 | 9水準（フィボナッチ5＋比率でない4）× スイング検出5通り × トリガー2種＝90通りを、時間足3種×期間4区分で実行 | | 損切り・利確 | 使わない（別セクションで測定） | | スプレッド | 0.3pips（スリッページ0、0.1ロット）。決済と新規の両方に掛かる | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | 比率でない4水準（30・45・70・80%）を混ぜているのが、この検証の要点です。**フィボナッチ比率に固有の効果があるなら、隣り合う比率でない水準との差として出てくるはずだから**です。 ## どの深さまで価格は戻るのか 押し戻しが各水準の深さまで届いた割合を、スイング6,266本で数えました。 | 深さ | 到達率 | 到達本数 | |---|---|---| | 23.6% | 95.9% | 6,012 | | 30% | 92.5% | 5,798 | | 38.2% | 87.2% | 5,462 | | 45% | 82.3% | 5,157 | | 50% | 78.8% | 4,936 | | 55% | 75.4% | 4,725 | | 61.8% | 71.4% | 4,474 | | 70% | 66.8% | 4,184 | | 78.6% | 62.3% | 3,902 | | 80% | 61.5% | 3,855 | **きれいに単調です。** 深いほど届きにくく、フィボナッチ比率のところで折れ曲がったりはしません。23.6%はほぼ必ず届き（95.9%）、78.6%は3本に2本届きません。 この表だけで言えることがひとつあります。**浅い水準を「効いた」と感じやすいのは、単にそこへ行く回数が多いからです。** 23.6%に線を引いておけば、20本中19本はその線に触ります。触れば記憶に残ります。 ## 一番反応する数字はどれか 最も高いのは45%で、フィボナッチ比率ではありませんでした。到達した回数で割って、**その水準に届いたスイングのうち、そこから5%以内で押し戻しが終わった割合**を出しました。「反応」を任意のpips幅で定義せず、水準へ到達したか、押し戻しがどこで終わったかだけで判定します。 | 深さ | 反転率 | 95%区間 | 反転本数／到達本数 | |---|---|---|---| | 20% | 2.05% | ±0.36 | 125 / 6,096 | | 23.6% | 2.73% | ±0.41 | 164 / 6,012 | | 30% | 3.55% | ±0.48 | 206 / 5,798 | | 38.2% | 3.94% | ±0.52 | 215 / 5,462 | | 45% | 4.29% | ±0.55 | 221 / 5,157 | | 50% | 4.27% | ±0.56 | 211 / 4,936 | | 55% | 3.92% | ±0.55 | 185 / 4,725 | | 61.8% | 4.22% | ±0.59 | 189 / 4,474 | | 70% | 3.78% | ±0.58 | 158 / 4,184 | | 78.6% | 3.95% | ±0.61 | 154 / 3,902 | | 80% | 3.76% | ±0.60 | 145 / 3,855 | **一番高いのは45%（4.29%）で、フィボナッチ比率ではありません。** 次が50%（4.27%）、その次が61.8%（4.22%）です。 ただし、この小さな差から優劣は決められません。**38.2%以上のどの水準も、互いの95%区間の中に収まっています。** 45%の4.29%と78.6%の3.95%の差は0.34ポイントで、それぞれの区間の幅（±0.55と±0.61）より小さい。この検証で言えるのは「38.2%より深ければどこも同じくらい」までです。 構造が出ているのは浅いほうだけです。**20%と23.6%は明確に低い**（2.05%と2.73%）。値幅の2割しか戻していない段階で押し目が終わることは、めったにありません。 冒頭の図がこの表の全体像です。401水準ぶんを並べると、反応率は10%から30%にかけて上がり、30%から90%までは3.5〜4.4%の範囲に収まります。**5つのフィボナッチ比率は、いずれも隣接する水準より高い反応率を示しませんでした。** 判定幅を変えても同じでした。 | 判定幅 | 23.6% | 38.2% | 50% | 61.8% | 78.6% | |---|---|---|---|---|---| | ±2.5% | 1.1% | 2.3% | 2.0% | 2.1% | 1.9% | | ±5% | 2.7% | 3.9% | 4.3% | 4.2% | 3.9% | | ±10% | 6.0% | 8.4% | 8.0% | 8.0% | 7.8% | ## 61.8%は特別な水準か すぐ隣の深さと比べると、比率に固有の効果は出ませんでした。「深いほど反転しやすい」なら、深さの効果と数字の効果を分けなければいけません。そこで各水準の反転率を、**前後3%以内にある水準（ただし前後0.8%は除く）の平均**と比べました。深さがほぼ同じで数字だけが違う相手との比較です。 12通り（3時間足×2年×上下の向き）それぞれで計算した差の平均です。 | 水準 | 隣との差 | 上回った回数 | |---|---|---| | 20% | −0.078ポイント | 6 / 12 | | 23.6%（比率） | −0.196ポイント | 5 / 12 | | 30% | +0.160ポイント | 6 / 12 | | 38.2%（比率） | −0.077ポイント | 6 / 12 | | 45% | +0.033ポイント | 5 / 12 | | 50%（比率） | −0.105ポイント | 7 / 12 | | 55% | −0.071ポイント | 4 / 12 | | 61.8%（比率） | −0.040ポイント | 6 / 12 | | 70% | +0.091ポイント | 8 / 12 | | 78.6%（比率） | +0.234ポイント | 7 / 12 | | 80% | +0.144ポイント | 7 / 12 | **フィボナッチ比率5つは平均−0.037ポイントで、60通り中31通りで隣を上回りました。比率でない6水準は+0.046ポイントで72通り中36通りです。** どちらも「半分の確率で隣より上」で、0と区別が付きません。 一番分かりやすい形で書きます。**61.8%の反転率は4.22%、隣の62.0%は4.30%です。** 61.0%も4.22%。38.2%は3.94%で、37.0%が3.92%、39.0%が3.97%、40.0%が4.07%。**線を1つ横にずらしても、何も起きません。** この検証の範囲では、**フィボナッチ比率が他の深さより価格を止めるという証拠は出ませんでした。** ## 線を引く前に価格は通り過ぎるか 浅い水準は通り過ぎています。実務上は、スイングが確定するまでに各水準を通過した割合も無視できません。 スイング高値が確定するのは、天井から5本あと（この検証の初期設定）です。線が引けるのはそれ以降。ところが**その5本のあいだに、価格はもう浅い水準を通り過ぎています。** 12通りの平均で、確定前にすでに水準を通過していたスイングの割合です。 | 深さ | 確定前に通過していた割合 | |---|---| | 23.6% | 94.1% | | 30% | 88.8% | | 38.2% | 81.3% | | 45% | 73.4% | | 50% | 67.9% | | 61.8% | 52.5% | | 70% | 43.4% | | 78.6% | 35.0% | | 80% | 34.0% | **23.6%の水準は、100本中94本で「線を引いたときにはもう抜けている」水準でした。** 38.2%でも81.3%です。61.8%でようやく半分（52.5%）になります。 手で引く場合も事情は同じです。天井が天井だと分かるのは、下げてからです。**浅いフィボナッチ水準で押し目買いを狙うという発想は、それが押し目だと分かる前に注文を置けることを前提にしています。** 自動検出でも裁量でも、そこは変わりません。 ## 水準で売買すると勝てるのか ここまでは価格の動きの話です。次は、その水準で実際に売買したらどうなるかを測りました。よく紹介される使い方は2つあります。 - **水準にタッチしたら、元のトレンド方向に入る**（上昇スイングなら押し目買い） - **水準を終値で抜けたら、抜けた方向に入る**（上昇スイングの61.8%を割ったら売り） 前者は水準で止まることに賭け、後者は止まらないことに賭けます。同じ線から正反対のルールが出てきます。9水準×スイング検出5通り×2トリガーで、時間足3種・期間4区分を回しました。 2024年と2025年の**両方**で黒字だった設定の数です（各45通り）。 | 時間足 | 押し目買い（タッチ） | 抜けに乗る（ブレイク） | |---|---|---| | 15分足 | 2 / 45 | 20 / 45 | | 1時間足 | 6 / 45 | 14 / 45 | | 4時間足 | 4 / 45 | 3 / 45 | **押し目買いはほとんど残りません。** 15分足では45通り中2通り（78.6%と80%、いずれもスイング検出21本）だけです。中央値は2025年が−1,095.7pips、2024年が−2,153.6pipsでした。 **抜けに乗る側は15分足で20通りが両方の年に残りました。** 中央値は2025年+631.8pips、2024年+512.2pipsです。この検証で最も安定していたのはここです。 ここでもフィボナッチ比率に優位はありません。押し目買いの中央値はフィボナッチ5水準が−742.4pips（150通り）、比率でない4水準が−668.5pips（120通り）。抜けに乗る側はフィボナッチが+118.4pips（150通り）、比率でないほうが+222.4pips（120通り）で、**どちらも比率でないほうがわずかに上**です。 同じ時間足・同じ年・同じトリガー・同じスイング検出で、フィボナッチ水準とすぐ隣の比率でない水準を1対1で比べると、**フィボナッチが勝ったのは300回中143回**でした。 ## 押し目買いとブレイク、どちらが勝てるか 勝率は押し目買いが18ポイント高く、年間損益はブレイクが835pips上でした。取引20回以上の設定について、平均勝率と年間損益の中央値を比べます。 | 売買ルール | 平均勝率 | 年間損益の中央値 | 設定数 | |---|---|---|---| | 押し目買い（タッチ） | 61.17% | −629.1pips | 256 | | 抜けに乗る（ブレイク） | 43.23% | +206.1pips | 217 | **勝率が18ポイント高いほうが、835pips負けています。** 1時間足の初期設定（61.8%・スイング検出5本）を見ると、2025年の押し目買いは292回・勝率65.75%で+921.1pipsでした。平均利益43.69pipsに対して平均損失は−74.68pipsで、1回の損失を取り戻すには平均利益約1.7回分が必要です。抜けに乗る側は112回・勝率35.71%ですが、平均利益110.23pipsが平均損失−60.07pipsの約1.8倍あります。 同じ現象は[ボリンジャーバンド](/ja/blog/bollinger-band-settings)でも[RSI](/ja/blog/rsi-settings)でも出ました。**勝率だけで指標や水準を選ぶと、年間損益の良し悪しを取り違えます。** ## 去年よかった設定を今年に持ち込むと そのまま持ち込めたのは半分でした。2024年の年間損益が最も大きかった設定を、そのまま2025年に当てます。 | 時間足 | トリガー | 2024年に選んだ設定 | 2024年の取引回数 | 2024年の年間損益 | 2025年の年間損益 | |---|---|---|---:|---:|---:| | 15分足 | タッチ | 50%・検出3本 | 2,491回 | +445.8pips | −846.4pips | | 15分足 | ブレイク | 80%・検出3本 | 491回 | +3,922.6pips | −1,410.7pips | | 1時間足 | タッチ | 30%・検出21本 | 150回 | +1,909.5pips | +320.2pips | | 1時間足 | ブレイク | 80%・検出13本 | 35回 | +3,335.4pips | −251.4pips | | 4時間足 | タッチ | 23.6%・検出8本 | 94回 | +2,891.4pips | +692.2pips | | 4時間足 | ブレイク | 23.6%・検出21本 | 7回 | +3,775.7pips | +148.7pips | 6設定のうち、2025年も黒字だったのは3設定です。15分足のブレイクは+3,922.6pipsから−1,410.7pipsへ反転しました。4時間足のブレイクは2024年に7回しか取引しておらず、+3,775.7pipsを安定した成績とは判断できません。 ## フィルターと損切りを足す 1時間足・61.8%・スイング検出5本に、よく紹介される条件を足しました。 | 条件 | トリガー | 年 | 取引回数 | 年間損益 | |---|---|---:|---:|---:| | そのまま | タッチ | 2025年 | 292回 | +921.1pips | | そのまま | タッチ | 2024年 | 255回 | −1,846.5pips | | そのまま | ブレイク | 2025年 | 112回 | +84.6pips | | そのまま | ブレイク | 2024年 | 101回 | −622.8pips | | ADX 20以上 | タッチ | 2025年 | 76回 | +149.9pips | | ADX 20以上 | タッチ | 2024年 | 64回 | +642.8pips | | ADX 20以上 | ブレイク | 2025年 | 29回 | −887.4pips | | ADX 20以上 | ブレイク | 2024年 | 25回 | +218.8pips | | ADX 25以上 | タッチ | 2025年 | 74回 | +3.1pips | | ADX 25以上 | タッチ | 2024年 | 69回 | −713.9pips | | ADX 25以上 | ブレイク | 2025年 | 28回 | −536.8pips | | ADX 25以上 | ブレイク | 2024年 | 25回 | +282.6pips | | ADX 30以上 | タッチ | 2025年 | 66回 | −308.9pips | | ADX 30以上 | タッチ | 2024年 | 53回 | −785.0pips | | ADX 30以上 | ブレイク | 2025年 | 25回 | −1,018.2pips | | ADX 30以上 | ブレイク | 2024年 | 16回 | +1,068.7pips | | ロンドン・NY時間 | タッチ | 2025年 | 250回 | −58.4pips | | ロンドン・NY時間 | タッチ | 2024年 | 223回 | −2,132.8pips | | ロンドン・NY時間 | ブレイク | 2025年 | 82回 | −515.6pips | | ロンドン・NY時間 | ブレイク | 2024年 | 79回 | −1,194.6pips | | 東京時間 | タッチ | 2025年 | 185回 | −611.2pips | | 東京時間 | タッチ | 2024年 | 178回 | −811.2pips | | 東京時間 | ブレイク | 2025年 | 58回 | −108.4pips | | 東京時間 | ブレイク | 2024年 | 49回 | +176.8pips | **ADXフィルターは年で反対を向きました。** タッチ側は2025年に+921.1が+3.1へ落ち（ADX 25以上）、2024年は−1,846.5が+642.8へ改善します（ADX 20以上）。**片方の年だけ見れば「効く」とも「効かない」とも書けてしまいます。** 時間帯フィルターは、4つのうち3つで基準を下回りました。ロンドン・NY時間はどの列でも悪化しています。 損切りと利確も足しました。 | 決済ルール | トリガー | 年 | 取引回数 | 年間損益 | |---|---|---:|---:|---:| | 反対シグナルまで | タッチ | 2025年 | 292回 | +921.1pips | | 反対シグナルまで | タッチ | 2024年 | 255回 | −1,846.5pips | | 反対シグナルまで | ブレイク | 2025年 | 112回 | +84.6pips | | 反対シグナルまで | ブレイク | 2024年 | 101回 | −622.8pips | | 損切り30・利確60 | タッチ | 2025年 | 743回 | −1,145.5pips | | 損切り30・利確60 | タッチ | 2024年 | 650回 | −568.6pips | | 損切り30・利確60 | ブレイク | 2025年 | 182回 | +170.7pips | | 損切り30・利確60 | ブレイク | 2024年 | 163回 | +164.7pips | | 損切り50・利確100 | タッチ | 2025年 | 525回 | +436.7pips | | 損切り50・利確100 | タッチ | 2024年 | 492回 | −1,968.8pips | | 損切り50・利確100 | ブレイク | 2025年 | 154回 | +303.6pips | | 損切り50・利確100 | ブレイク | 2024年 | 141回 | −64.5pips | | 損切り100・利確200 | タッチ | 2025年 | 381回 | +1,269.4pips | | 損切り100・利確200 | タッチ | 2024年 | 351回 | −2,208.7pips | | 損切り100・利確200 | ブレイク | 2025年 | 130回 | +29.8pips | | 損切り100・利確200 | ブレイク | 2024年 | 119回 | +441.9pips | | 24本で時間決済 | タッチ | 2025年 | 382回 | +923.2pips | | 24本で時間決済 | タッチ | 2024年 | 338回 | −1,865.9pips | | 24本で時間決済 | ブレイク | 2025年 | 162回 | +303.3pips | | 24本で時間決済 | ブレイク | 2024年 | 145回 | −28.6pips | **ブレイク側は損切り30／利確60が両方の年で黒字になった唯一の組み合わせ**でした（+170.7と+164.7）。タッチ側はどの決済でも2024年が赤字のままです。 ## コストは取引回数×スプレッド スプレッドだけを変えて2025年の1時間足を回し直します。 | スプレッド | タッチ 61.8%（292回） | ブレイク 61.8%（112回） | |---|---|---| | 0.0pips | +1,008.7 | +118.2 | | 0.3pips | +921.1 | +84.6 | | 0.6pips | +833.5 | +51.0 | | 1.0pips | +716.7 | +6.2 | | 1.5pips | +570.7 | −49.8 | | 2.0pips | +424.7 | −105.8 | 0から2.0pipsで失ったのは、タッチ側が584.0pips（292回×2.0）、ブレイク側が224.0pips（112回×2.0）。**どちらも取引回数×スプレッドと誤差なく一致します。** このシリーズの16本すべてで成立している関係です。 ブレイク側は1.0pipsを超えたあたりで水面下に沈みます。**112回しか取引しない設定でも、利益が118.2pipsしかなければスプレッドは十分に重い**ということです。 「61.8%は特別だから効く」という説明は、この検証の範囲では支持できませんでした。ただし**深い水準ほど反転しやすいこと自体は成立している**ので、フィボナッチを深さの目盛りとして使うことまで否定する結果ではありません。 同じ「逆張りは勝率が高く負ける」形は[ボリンジャーバンド](/ja/blog/bollinger-band-settings)と[RSI](/ja/blog/rsi-settings)でも測っています。[移動平均線クロス](/ja/blog/moving-average-cross-settings)は水準ではなく交差で同じ問いを見ています。「本数を増やせば見えるものがある」という主張を同じやり方で崩したのが[GMMA](/ja/blog/gmma-settings)です。[プログラミングなしで条件を組んで検証する](/ja/blog/backtest-without-coding)なら、この記事の表と同じことを自分の通貨ペア・自分の期間で確かめられます。 関連して、[アリゲーター](/ja/blog/alligator-settings)の13・8・5という期間もフィボナッチ数です。フィボナッチ数を含まない4組と、トリガー・時間足・ずらし・年をそろえて比べると、13・8・5の年間損益が上回ったのは288通り中136通りでした。 描画ツールを条件に昇格させて測る形は、[フェアバリューギャップ](/ja/blog/fair-value-gap-settings)でもう一度使いました。あちらは対照群として「ギャップを作らなかった足」を並べたところ、埋まる割合がほとんど同じでした。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです。他の通貨ペアや他の年で同じ傾向になる保証はありません - **スイングの取り方は自動検出です。** 手で引くフィボナッチは、どの波に引くかを人が選びます。その選び方が結果を変える可能性はこの検証では測れていません。感度検証として検出幅を5通り試し、反転率の順序は変わりませんでした - **反転率の判定幅（5%）は任意です。** 2.5%と10%でも計算し、水準どうしの関係は変わりませんでした - 反転率は「押し戻しがそこで終わったか」を測っています。「一時的に止まって、また下げた」動きは反転として数えていません - 401水準は互いに重なりを持つため、隣り合う水準の数字は独立ではありません。最大値を、独立な401回の試行から得た値としては読めません - エントリーも決済も足の終値で行っています。実際の約定はここからずれます - スプレッドは全期間0.3pips固定として計算しています。実際のスプレッドは時間帯と指標発表で変動します --- ### GMMAは移動平均線2本と何が違う？216通り検証 URL: https://formiq.jp/ja/blog/gmma-settings Language: ja Published: 2026-08-26 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: GMMA, 移動平均線, リボン, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/gmma-settings GMMA（Guppy Multiple Moving Average）は、期間の違う指数移動平均線を12本重ねて表示するインジケーターです。**短期6本（3・5・8・10・12・15）が短期売買のトレーダー、長期6本（30・35・40・45・50・60）が長期保有の投資家**を表す、という説明とセットで紹介されます。 主張の中身は「速い平均が遅い平均を抜けた」ではありません。それは2本の線でも分かります。GMMAが主張しているのは、**12本にすることで初めて見えるものがある**という点です。2つの集団が同じ方向を向いているのか、迷っているのか。 そこを直接測りました。**各リボンの平均期間を9と45に固定したまま、本数だけ6本から1本まで減らします。** 1本ずつにすればリボンは消えて、ただのEMA9とEMA45のクロスになります。速さは同じで、扇の厚みだけが変わる。差が出れば、それが扇の価値です。 **12本に増やす効果は確認できませんでした。** 1本が6本を36通り中21通りで上回り、中央値は1本が+1,313.5pips、6本が+1,371.5pipsでした。 ## GMMAは何を測っているのか 計算は指数移動平均線そのもので、GMMA固有の式はありません。あるのは**どの期間を何本並べるか**という選び方だけです。 この検証で使ったリボンは、平均期間を保ったまま本数を変えています。 | 本数 | 短期リボン | 長期リボン | |---|---|---| | 6（本来のGMMA） | 3・5・8・10・12・15 | 30・35・40・45・50・60 | | 5 | 3・6・9・12・15 | 30・38・45・53・60 | | 4 | 3・7・11・15 | 30・40・50・60 | | 3 | 3・9・15 | 30・45・60 | | 2 | 3・15 | 30・60 | | 1 | 9 | 45 | 短期リボンの平均期間はどの行でも9.0（6本の行だけ8.83）、長期は45.0（6本は43.33）です。**つまり1本の行と6本の行は、ほぼ同じ速さの平均を見ています。** この設計にしたのは、本数を減らすときに端から削ると速さまで変わってしまうからです。3・5・8だけを残せば平均期間は5.3になり、それは「扇が薄いから負けた」のか「速すぎて負けた」のか区別が付かなくなります。 ### 1本ずつにすると、EMA9とEMA45のクロスになる リボンが1本ずつになると、GMMAの「クロス」トリガーは短期側の平均と長期側の平均を比べるだけになります。つまり**EMA9とEMA45のゴールデンクロス／デッドクロス**です。 検証でも一致しました。2025年の1時間足で、1本ずつのGMMAは184回・+393.5pips。EMA9とEMA45のクロスを別に組んで回しても184回・+393.5pipsで、1回も違いません。**この記事の「本数を減らす」は、最後にはよく知られた2本のクロスへ行き着きます。** ## GMMAをチャートに表示する方法 12本の移動平均線なので、標準搭載でない環境でも自力で並べられます。 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準では入っていない。移動平均線を12本重ねて期間と色を設定するか、配布されているGMMAのファイルを探してインストールする | | TradingView | インジケーター検索から Guppy Multiple Moving Average を追加する | | ブラウザ（Formiq） | 最初から入っている。バックテスト側ではトリガー3種と各リボンの本数を条件にできる | **手で12本並べる場合、期間の選び方は自由です。** この記事の表のとおり、平均期間を保ったまま本数を減らしても損益はほとんど動きませんでした。12本を並べる手間に見合う差は、この検証では見つかっていません。 ## 検証の条件 数字を見る前に、何をどう測ったかを書いておきます。この表と同じ設定を入れれば、同じ結果が出ます。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | トリガー | クロス（2つの平均が交差）／ 拡大（平均の差が開く）／ 全上抜け（短期6本すべてが長期6本すべての上） | | リボンの本数 | 1・2・3・4・5・6本。平均期間は9と45に固定 | | 組み合わせ | トリガー3種 × 本数6種 = 18通り。時間足3種 × 期間4区分で合計216通り | | 決済 | 反対シグナルまで。エントリーと決済に同じルールセットを使う | | 損切り・利確 | 使わない（別セクションで測定） | | スプレッド | 0.3pips（スリッページ0、0.1ロット）。決済と新規の両方に掛かる | | リボンの状態 | 取引ごとに、注文を出した足のリボン間の距離・短期リボンの幅・長期リボンの幅・長期リボンが順番に並んでいるかを記録（5,217件） | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | ## 12本の線は必要か 1本まで減らしても損益は変わりませんでした。時間足ごとに、本数を変えたときの平均損益（3トリガー×2年）です。 | 本数 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---| | 1 | +1,719 | +680 | +487 | | 2 | +1,830 | +441 | −1 | | 3 | +1,797 | +617 | +70 | | 4 | +1,742 | +538 | +86 | | 5 | +1,749 | +560 | +65 | | 6 | +1,880 | +645 | +138 | **上がっていく列がありません。** 15分足は1,719から1,880のあいだで往復し、1時間足も441から680のあいだです。 同じ時間足・同じ年・同じトリガーで1本と6本を1対1で比べると、**1本が勝ったのは36通り中21通り**、差の平均は1本側へ+124.9pipsでした。 中央値で見ても、1本ずつが+1,313.5pips、6本ずつが+1,371.5pips。**12本を並べたことで増えた分は、この検証では測定できる大きさになりませんでした。** 冒頭の図がこの表です。4時間足だけ1本の点が飛び出していますが、これは4時間足が2年で正反対になったせいで（後述）、扇の効果ではありません。 ## 使えた時間足はどれか 15分足だけです。時間足ごとに、2024年と2025年の**両方**で黒字だった設定の数です（各18通り）。 | 時間足 | 年 | 黒字設定 | 損益中央値 | 両年とも黒字 | |---|---:|---:|---:|---:| | 15分足 | 2025 | 18 / 18 | +1,703.6pips | **18 / 18** | | 15分足 | 2024 | 18 / 18 | +1,917.9pips | **18 / 18** | | 1時間足 | 2025 | 14 / 18 | +221.9pips | 14 / 18 | | 1時間足 | 2024 | 18 / 18 | +1,067.7pips | 14 / 18 | | 4時間足 | 2025 | 0 / 18 | −1,719.0pips | 0 / 18 | | 4時間足 | 2024 | 18 / 18 | +2,100.2pips | 0 / 18 | **15分足は18通りすべてが両方の年で黒字でした。** このシリーズ17本のなかで最も揃った結果です。 中身を見ると典型的なトレンドフォローの形です。初期設定（クロス・6本）は2025年に831回・勝率27.32%で+1,703.6pips。平均利益55.02pipsに対し平均損失−17.86pipsで、**3回外して1回当てれば釣り合う比率**になっています。1トレードの保有は平均42本＝10時間半で、スキャルピングではありません。 **4時間足は2年で完全に反転しました。** 2024年は18通り全部が黒字（中央値+2,100.2）、2025年は18通り全部が赤字（中央値−1,719.0）です。クロス・2本は2024年の+2,334.6pipsから、2025年の−2,419.9pipsへ変わりました。 ## 3つのトリガーはどう違うか 同じリボンを3通りに読みます。両方の年をまとめた平均です。 | トリガー | 平均勝率 | 損益の中央値 | 平均取引回数 | 平均保有本数 | |---|---|---|---|---| | クロス | 26.52% | +1,468.4pips | 364 | 42.7 | | 拡大 | 29.14% | +1,313.5pips | 321 | 48.3 | | 全上抜け | 32.76% | +1,345.4pips | 209 | 76.1 | **全上抜けは取引回数が最も少なく（クロスの57%）、勝率が最も高く（クロスより6.2ポイント上）、保有が最も長い**という、条件を厳しくしたときに出る形をそのまま示しています。損益の中央値ではクロスが最も高く、3つの差は150pips以内です。 ## リボンが広がったら強いのか 勝率は上がりました。GMMAが「12本の絵で分かる」とされる内容を測るため、全トリガーの取引5,217件を、**注文を出した足のリボン間の距離**で四分位に分けました。 | 時間足 | 第1（最も狭い） | 第2 | 第3 | 第4（最も広い） | |---|---|---|---|---| | 15分足（各995件） | 24.32% | 28.04% | 29.05% | **32.96%** | | 1時間足（各239件） | 22.18% | 27.20% | 35.98% | **35.98%** | | 4時間足（各69件） | 21.74% | 15.94% | 30.43% | **34.78%** | **どの時間足でも、最も広い区分の勝率が最も高くなりました。** 15分足では8.64ポイント、1時間足で13.80ポイント、4時間足で13.04ポイントの差です。検証したGMMA固有の主張のうち、成立したのはリボン間の距離と勝率の関係だけでした。 ただし勝率と損益は別でした。1回あたりの平均損益はこうなります。 | 時間足 | 第1 | 第2 | 第3 | 第4 | |---|---|---|---|---| | 15分足 | +3.96pips | +2.58 | +2.85 | **+2.06** | | 1時間足 | −3.68pips | +1.83 | +7.08 | **+12.15** | | 4時間足 | −5.62pips | −34.82 | +24.73 | **+34.48** | **15分足だけ逆を向いています。** 勝率は上がるのに1回あたりの利益は3.96pipsから2.06pipsへ下がる。リボンが離れきったところで入ると当たりやすいが、その時点で動きの大半は終わっている、という読み方になります。1時間足と4時間足では勝率と損益が同じ方向を向きました。 ## 長期リボンの並びは使えるのか 判断材料になりませんでした。GMMAの説明でよく出てくるのが「**長期リボンが順番に並んでいれば本物のトレンド**」という読み方です。長期6本が30<35<40<45<50<60の順に整列していれば投資家も同意している、という理屈です。 長期リボンが順番に並んでいた足の割合を数えました。 | 時間足・年 | 検証本数 | 長期リボンが整列していた割合 | |---|---|---| | 15分足 2025年 | 24,903 | 85.18% | | 15分足 2024年 | 24,999 | 85.39% | | 1時間足 2025年 | 6,226 | 86.35% | | 1時間足 2024年 | 6,250 | 85.74% | | 4時間足 2025年 | 1,610 | 85.09% | | 4時間足 2024年 | 1,616 | 90.47% | **8割5分の足で整列しています。** 6本の指数移動平均線は互いに近い期間なので、よほど激しく折り返さないかぎり順番は崩れません。**1年のうち85%で成立している条件は、絞り込みとして機能しません。** 取引を整列の有無で分けても、方向が揃いませんでした。 | 時間足 | 整列していた | 整列していなかった | |---|---|---| | 15分足 | 3,020件・勝率28.48%・平均+1.70pips | 961件・勝率28.93%・平均**+6.39pips** | | 1時間足 | 742件・勝率27.49%・平均−0.81pips | 216件・勝率39.81%・平均**+20.71pips** | | 4時間足 | 225件・勝率24.44%・平均**+5.78pips** | 53件・勝率30.19%・平均−8.90pips | **3つのうち2つで、整列していないときのほうが良い結果でした。** 1時間足では勝率が12.32ポイント高く、1回あたり21.52pips多い。4時間足だけ逆を向きますが、そちらは53件しかありません。 「長期リボンが整列していたら本物」という読み方は、この検証では支持できませんでした。 ## 収縮のあとは大きく動くのか 動きませんでした。もうひとつの定番が「**リボンが収縮したら次の大きな動きが近い**」です。これは売買ルールではなく価格についての主張なので、バックテストを使わずに直接測れます。 短期リボンと長期リボンの幅を足して、**最も狭い1割の足**を「収縮」とし、そのあと24本の値幅（絶対値）を、それ以外の足のあとと比べました。 | 時間足・年 | 収縮後 | それ以外 | |---|---|---| | 15分足 2025年 | 30.61pips（2,491本） | 32.71pips（22,388本） | | 15分足 2024年 | 23.56pips（2,500本） | 32.94pips（22,499本） | | 1時間足 2025年 | 68.27pips（618本） | 68.41pips（5,584本） | | 1時間足 2024年 | 47.60pips（626本） | 73.69pips（5,624本） | | 4時間足 2025年 | 85.83pips（147本） | 134.22pips（1,439本） | | 4時間足 2024年 | 153.56pips（162本） | 145.58pips（1,454本） | **6通りのうち5通りで、収縮後のほうが動きが小さくなりました。** 先読みを6本・12本・24本・48本の4通りに広げても同じで、**24通り中21通りで収縮後のほうが小さい**という結果です。 静かな相場のあとは静かなまま、というのは値動きの一般的な性質で、GMMAに限った話ではありません。**「収縮＝溜めている」という読み方は、この検証の範囲では逆でした。** ひとつだけ方向性が見えたのは、値幅ではなく**向き**です。収縮後の24本の動きを「リボンが既に向いていた側」で測ると、15分足では収縮後が+6.31pipsで、それ以外の+0.24pipsより大きくなります。動きは小さいが、既にある傾きは維持されやすい、という読み方はできます。ただし1時間足2024年は−8.50pipsで反対を向いており、年をまたいで成立してはいません。 ## 去年いちばんの設定は今年も勝てるか 15分足は黒字のまま、4時間足は符号ごと反転しました。 | 時間足 | トリガー | 本数 | 2024年の取引数 | 2024年の損益 | 2025年の損益 | |---|---|---:|---:|---:|---:| | 15分足 | 全上抜け | 2本 | 436回 | +2,304.6pips | +1,404.1pips | | 1時間足 | クロス | 1本 | 173回 | +1,209.7pips | +393.5pips | | 4時間足 | クロス | 2本 | 41回 | +2,334.6pips | −2,419.9pips | 15分足は18設定すべてが両年黒字で、この設定も2025年に+1,404.1pipsを残しました。4時間足は同じ設定が+2,334.6pipsから−2,419.9pipsへ反転しています。 ## フィルターと損切りを足す 15分足のクロス・6本に、よく紹介される条件を足しました。左が取引回数、右が損益です。 | 条件 | 年 | 取引数 | 損益 | |---|---:|---:|---:| | そのまま | 2025 | 831回 | +1,703.6pips | | そのまま | 2024 | 821回 | +2,067.4pips | | ADX 20以上 | 2025 | 113回 | −769.3pips | | ADX 20以上 | 2024 | 120回 | +735.1pips | | ADX 25以上 | 2025 | 68回 | −474.8pips | | ADX 25以上 | 2024 | 75回 | −352.6pips | | ADX 30以上 | 2025 | 20回 | −199.6pips | | ADX 30以上 | 2024 | 30回 | −335.3pips | | ロンドン・NY時間 | 2025 | 441回 | +1,524.9pips | | ロンドン・NY時間 | 2024 | 435回 | +382.4pips | | 東京時間 | 2025 | 321回 | +799.8pips | | 東京時間 | 2024 | 310回 | +1,186.2pips | **ADXフィルターはこの検証で最も壊した条件でした。** 831回が113回まで減り、+1,703.6pipsが−769.3pipsになります。ADX 25以上では両方の年が赤字です。**トレンド指標にトレンドフィルターを足して悪化する**という結果は、GMMAが既にトレンドの条件を持っているぶん、絞り込みが重複していると読めます。 時間帯フィルターはどちらも基準を下回りましたが、赤字にはなりません。 損切りと利確も足しました。 | 決済ルール | 年 | 取引数 | 損益 | |---|---:|---:|---:| | 反対シグナルまで | 2025 | 831回 | +1,703.6pips | | 反対シグナルまで | 2024 | 821回 | +2,067.4pips | | 損切り20・利確40 | 2025 | 730回 | +1,115.7pips | | 損切り20・利確40 | 2024 | 731回 | +335.8pips | | 損切り30・利確60 | 2025 | 768回 | +2,039.9pips | | 損切り30・利確60 | 2024 | 757回 | +1,020.0pips | | 損切り50・利確100 | 2025 | 813回 | +1,512.3pips | | 損切り50・利確100 | 2024 | 796回 | +757.3pips | | 48本で時間決済 | 2025 | 827回 | +2,036.1pips | | 48本で時間決済 | 2024 | 815回 | +1,409.4pips | **両方の年で基準を上回った決済はありませんでした。** 損切り30／利確60と48本の時間決済は2025年を伸ばしますが、どちらも2024年を1,000pips前後削ります。**平均利益55pipsに対し平均損失18pipsという形を、固定幅の決済が壊しています。** ## コストは回数で決まるのか スプレッドだけを変えて2025年の15分足を回し直します。 | スプレッド | 損益 | |---|---| | 0.0pips | +1,952.9 | | 0.3pips | +1,703.6 | | 0.6pips | +1,454.3 | | 1.0pips | +1,121.9 | | 1.5pips | +706.4 | | 2.0pips | +290.9 | | 3.0pips | −540.1 | 取引回数は831回で固定です。0から3.0pipsで失ったのは2,493.0pipsで、**831回×3.0pipsと誤差なく一致します。** このシリーズ17本すべてで成立している関係です。 **水面下に沈むのは2.35pipsあたり**でした（2.0pipsで+290.9、3.0pipsで−540.1の間）。15分足で年831回売買しても、実際のドル円のスプレッドなら余裕があります。このシリーズでは、15分足がコスト負けしなかった最初の指標です。理由は保有時間で、平均42本＝10時間半持つので、1回あたりの利益がスプレッドの何十倍もあります。 「12本並べることで見えるものがある」という部分は支持できませんでしたが、**リボン間の距離を勝率の目安として使う**ところは残りました。 線の本数ではなく交差そのものを測ったのが[移動平均線クロスの記事](/ja/blog/moving-average-cross-settings)です。3本すべての並びを条件にした[パーフェクトオーダー](/ja/blog/perfect-order-settings)では、時間足ごとに残る計算方法が変わりました。同じ「勝率が低いほうが勝つ」形は[ボリンジャーバンド](/ja/blog/bollinger-band-settings)と[RSI](/ja/blog/rsi-settings)でも出ました。水準そのものを疑う型は[フィボナッチ](/ja/blog/fibonacci-retracement-levels)で作っています。トレンド系にトレンドフィルターを足して壊れる例は[CCI](/ja/blog/cci-settings)でも再現しました。「85%の足で成立している状態は条件にならない」は[アリゲーター](/ja/blog/alligator-settings)でもそのまま出ています（3本が順番に並ぶのは年の8割）。同じ数え方を突き詰めると[フェアバリューギャップ](/ja/blog/fair-value-gap-settings)になり、そちらは「ギャップは99%埋まる」が普通の足でも99%だったという結果でした。[プログラミングなしで条件を組んで検証する](/ja/blog/backtest-without-coding)なら、この記事の表と同じことを自分の通貨ペア・自分の期間で確かめられます。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです。他の通貨ペアや他の年で同じ傾向になる保証はありません - **2024年は+1,632pips上昇し、2025年は−56pipsで終わりました。** 年間の高値と安値の差はそれぞれ2,237pipsと1,900pipsです。15分足の結果は、この値動きが異なる2年間だけで確認したものです - **本数を減らす方法は1通りに決めています。** 端点（3〜15と30〜60）を保ったまま等間隔に間引く形で、平均期間が動かないようにしました。別の減らし方（端から削る、中央だけ残す）では違う結果になり得ます - 6本の行だけ本来のGMMAの期間をそのまま使っているため、平均期間が8.83と43.33で、他の行の9.0と45.0からわずかにずれています - リボンの状態は注文を出した足の1本前で読んでいます。その取引自身の値動きが指標に混ざらないようにするためです - 「収縮」は各期間の中で最も狭い1割という相対的な定義です。絶対的な幅で切ると別の結果になり得ます - エントリーも決済も足の終値で行っています。実際の約定はここからずれます - スプレッドは全期間0.3pips固定として計算しています。実際のスプレッドは時間帯と指標発表で変動します - [時間帯によって値動きの大きさは変わります](/ja/data)。時間フィルターの結果はその影響を受けています --- ### ボリンジャーバンド「2σで逆張り」は勝てる？315通り検証の結果 URL: https://formiq.jp/ja/blog/bollinger-band-settings Language: ja Published: 2026-08-25 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ボリンジャーバンド, 逆張り, 順張り, インジケーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/bollinger-band-settings ボリンジャーバンドは移動平均線の上下に、値動きのばらつき（標準偏差）で決めた幅を足した3本の線です。価格の大半がこの幅に収まるので、**バンドに触れたら反対方向に戻ることを狙う（逆張り）**か、**バンドを抜けたらその方向に伸びることを狙う（順張り）**か、正反対の2つの使い方があります。 バンドタッチ逆張りと終値ブレイク順張りのどちらが利益を残すかは、指標の説明だけでは決まりません。そこで、同じ通貨ペア・同じ期間・同じコストで2つの売買ルールを測りました。 ドル円の2025年（1月1日〜12月31日）で、期間7種×偏差5種の組み合わせに決済位置4種を掛けた**逆張り140通りを買い側と売り側それぞれ**、**順張り35通り**、合わせて315通りをバックテストしています。時間足は15分足・1時間足・4時間足の3つ、比較のために2024年でも同じ315通りを回しました。 **1時間足で、平均勝率60.4%の逆張り（買い）は140通り中25通りしか2024年と2025年の両方で黒字にならず、平均勝率40.4%の順張りは35通り中23通りが黒字**でした。上バンドを売る逆張りに至っては、15分足でも1時間足でも**両方の年で黒字になった設定はゼロ**です。勝率で並べると順序が逆になります。 ## ボリンジャーバンドは何を測っているのか 真ん中の線は単純移動平均です（**期間**）。上下のバンドは、その期間の終値のばらつきを標準偏差で測り、それを何倍したところに引くかで決まります（**偏差**）。 よく「1σに約68%、2σに約95%が収まる」と説明されますが、それは値動きが正規分布に従う場合の数字です。実際のドル円で数えると違いました。期間20のバンドに終値が収まった割合は、2025年の1時間足6,226本で次のとおりです。 | 偏差 | 終値が中に収まった割合 | |---|---| | 1σ | 45.0% | | 1.5σ | 70.4% | | 2σ | 87.1% | | 2.5σ | 95.1% | | 3σ | 99.1% | **教科書の数字より、どれも内側が薄くなります。** 2σで約87%、つまり8本に1本は外に出ます。「めったに出ない」という前提で逆張りを組むと、想定より頻繁に外を歩かれることになります。 バンドそのものは「今の値動きの荒さに合わせて伸び縮みする物差し」です。荒れているときは広がり、静かなときは縮みます。同じ2σでも、そのときの相場によって実際の値幅はまったく違います。 設定するパラメータは期間と偏差です。逆張りでは決済に使う偏差も設定します。 | パラメータ | 値を上げると何が起きるか | |---|---| | 期間 | 真ん中の線が鈍くなる。バンドに触れる回数が減る | | 偏差 | バンドが外側に離れる。触れる／抜ける回数が減る | | 決済の偏差（逆張りのみ） | 反対側のバンドを遠くに置く＝利確を引っ張る。回数が減り、1回が長くなる | 実際の数字で確かめると、逆張り買い・1時間足・2025年の平均取引回数は、期間10のとき319回、期間50のとき53回でした。偏差では1σのとき213回、3σのとき77回です。期間と偏差を大きくすると取引回数が減り、その回数に比例して支払うスプレッドの合計も減ります。 ## チャートに表示する方法 ボリンジャーバンドはどのプラットフォームにも標準で入っています。 | 環境 | 手順 | |---|---| | MT4 / MT5 | ナビゲータのインジケータ一覧から Bollinger Bands を選び、期間と偏差を設定する | | TradingView | インジケーター検索から Bollinger Bands を追加し、期間と偏差を設定する | | ブラウザ（Formiq） | 最初から入っている。バックテスト側は逆張り（バンドタッチ）と順張り（バンドブレイク）をトリガーで切り替える | 表示そのものはどこでも同じですが、**売買ルールとして組むときに「触れたら」なのか「抜けたら」なのかを決める必要があります。** この記事では、バンドに触れたら反対方向へ売買する逆張りと、終値がバンドを抜けた方向へ売買する順張りを別々に測っています。 ## 検証の条件 数字を見る前に、何をどう測ったかを書いておきます。この表と同じ設定を入れれば、同じ結果が出ます。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 2025年で15分足24,903本、1時間足6,226本、4時間足1,610本 | | 逆張り・買い | 安値が下バンドに触れた足の終値で買い、高値が上バンドに触れた足の終値で決済 | | 逆張り・売り | 高値が上バンドに触れた足の終値で売り、安値が下バンドに触れた足の終値で決済 | | 順張り | 終値が上バンドを上抜けたら買い、下バンドを下抜けたら売り。反対側のバンドを終値で抜けたら同じ足でドテン | | 損切り・利確 | 使わない（別セクションで測定） | | スプレッド | 0.3pips（スリッページ0、0.1ロット）。決済と新規の両方に掛かる | | 逆張りの組み合わせ | 期間7種（10・14・20・25・30・40・50）× 偏差5種（1・1.5・2・2.5・3σ）× 決済の偏差4種（0.5・1・1.5・2σ）＝140通り。買い側と売り側で別々に実行 | | 順張りの組み合わせ | 期間7種 × 偏差5種＝35通り | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | 買い逆張りと売り逆張りを別々に集計するのは、バンドタッチが買いと売りのどちらか一方しか作らないからです。下バンドを買う設定と上バンドを売る設定は、まったく別の成績になります。 ## 逆張りと順張りはどちらが勝てるか 勝率が高いのは逆張りですが、年間損益が残ったのは順張りでした。買い逆張り、売り逆張り、終値ブレイク順張りの成績を、時間足ごとに並べます。 | 売買ルール | 時間足 | 2025年の黒字数 | 2025年の中央値 | 2024年の黒字数 | 2024年の中央値 | 両方の年で黒字 | 平均勝率 | |---|---|---|---|---|---|---|---| | 逆張り・買い | 15分足 | 22 / 140 | −503pips | 44 / 140 | −223pips | 7 / 140 | 64.9% | | 逆張り・買い | 1時間足 | 39 / 140 | −252pips | 82 / 140 | +106pips | 25 / 140 | 60.4% | | 逆張り・買い | 4時間足 | 104 / 140 | +511pips | 36 / 140 | −308pips | 15 / 140 | 69.5% | | 逆張り・売り | 15分足 | 4 / 140 | −661pips | 2 / 140 | −1,376pips | 0 / 140 | 63.9% | | 逆張り・売り | 1時間足 | 21 / 140 | −489pips | 15 / 140 | −1,023pips | 0 / 140 | 60.8% | | 逆張り・売り | 4時間足 | 99 / 140 | +526pips | 6 / 140 | −1,247pips | 1 / 140 | 63.4% | | 順張り | 15分足 | 25 / 35 | +432pips | 18 / 35 | +33pips | 16 / 35 | 38.3% | | 順張り | 1時間足 | 25 / 35 | +789pips | 30 / 35 | +838pips | 23 / 35 | 40.4% | | 順張り | 4時間足 | 12 / 35 | −1,025pips | 28 / 35 | +861pips | 9 / 35 | 36.3% | （平均勝率は2025年の全設定の平均。取引が1回も出なかった設定は勝率の集計から除いています） 冒頭の図が、この表の1時間足を1点ずつ描いたものです。横軸が勝率、縦軸が年間の損益。**赤と紫（逆張り）は右下に、青（順張り）は左上に集まります。** よく当たるほうが負けている、という形がそのまま見えます。 **上バンドを売る逆張りは、15分足でも1時間足でも両方の年で黒字になった設定が1つもありません**（140通り中0通り）。4時間足でようやく1通りです。2024年のドル円が年間1,632pips上昇したので、上バンドを売り続ける行為が上昇に轢かれ続けた、という説明がつきます。 いっぽう順張りは、1時間足で35通り中23通りが両方の年で黒字でした。 ## なぜ勝率が高いほうが負けるのか 1回あたりの勝ち幅と負け幅が非対称だからです。 2025年の全設定を平均すると、こうなります。 | 売買ルール | 時間足 | 平均勝ち幅 | 平均負け幅 | |---|---|---|---| | 逆張り・買い | 1時間足 | +41.3pips | −68.6pips | | 逆張り・売り | 1時間足 | +45.1pips | −83.8pips | | 順張り | 1時間足 | +123.8pips | −80.9pips | 逆張りは、バンドに触れて戻れば小さく勝ちますが、戻らずにそのまま走られたときに大きく負けます。反対側のバンドまで待つ決済ルールなので、外したときの損失が勝ちの1.7〜1.9倍になります。**勝率6割でも、勝ち幅が負け幅の6割しかなければ合計は負けます。** 順張りはその逆です。抜けてすぐ戻される回数のほうが多い（勝率40%）かわりに、伸びたときの1回が大きくなります。 15分足と4時間足でも関係はおおむね同じ向きでした。逆張り買いは15分足で+20.8／−40.6、4時間足で+84.9／−140.9。順張りは15分足で+66.5／−42.5、4時間足で+286.1／−136.4です。 例外が1つあります。**4時間足の逆張り売りだけは、平均勝ち幅+99.3pipsが平均負け幅−94.5pipsを上回りました。** ただしこの組み合わせも、両方の年で黒字だったのは140通り中1通りだけです。幅の関係が有利でも、2024年の上昇に轢かれた分を取り返せていません。 ## 定番の20・2σはどうだったか いちばんよく見かける「期間20・偏差2σ」を、バンドタッチ逆張りと終値ブレイク順張りで比べます。 | 使い方 | 時間足 | 2025年 | 2024年 | |---|---|---|---| | 逆張り・買い（決済1σ） | 15分足 | 667回・勝率65.07%・−411.1pips | 642回・勝率67.76%・−613.3pips | | 逆張り・買い（決済1σ） | 1時間足 | 167回・勝率63.47%・+71.0pips | 157回・勝率68.79%・−597.6pips | | 逆張り・買い（決済1σ） | 4時間足 | 40回・勝率77.50%・+1,030.3pips | 34回・勝率67.65%・−389.0pips | | 順張り | 15分足 | 570回・勝率37.72%・+292.5pips | 566回・勝率35.87%・+344.4pips | | 順張り | 1時間足 | 132回・勝率43.18%・+1,177.5pips | 137回・勝率41.61%・+1,434.6pips | | 順張り | 4時間足 | 41回・勝率19.51%・−2,126.7pips | 34回・勝率35.29%・+775.5pips | **両方の年で黒字だったのは、15分足と1時間足の順張りだけ**です。同じ20・2σを逆張りに使うと、15分足は両年赤字、1時間足は2025年に小幅黒字でも2024年に赤字でした。 4時間足は逆張りと順張りのどちらも年で符号が入れ替わります。4時間足の順張りは2025年に41回で−2,126.7pips、勝率19.51%。この件数と勝率で語れることは多くありません。 ## 期間と偏差はどう選ぶか 逆張りで残ったのは、めったに触れない長い期間と大きい偏差だけでした。1時間足で、期間または偏差を1項目ずつ変え、残りの全設定で平均を取ります。数字は「その値を持つ設定のうち、2024年と2025年の両方で黒字だった数」です。 **順張りの期間** | 期間 | 2025年の平均 | 2024年の平均 | 両方の年で黒字 | |---|---|---|---| | 10 | +520pips | +732pips | 2 / 5 | | 14 | +243pips | +1,221pips | 3 / 5 | | 20 | +885pips | +1,489pips | 5 / 5 | | 25 | +596pips | +870pips | 5 / 5 | | 30 | +804pips | +287pips | 2 / 5 | | 40 | +452pips | +803pips | 4 / 5 | | 50 | −12pips | +704pips | 2 / 5 | **期間20と25は5通り中5通りが両方の年で黒字**でした。定番の20が順張り側では実際に上位にいます。 **順張りの偏差** | 偏差 | 2025年の平均 | 2024年の平均 | 両方の年で黒字 | |---|---|---|---| | 1σ | +1,082pips | +1,569pips | 7 / 7 | | 1.5σ | +1,368pips | +693pips | 5 / 7 | | 2σ | +980pips | +742pips | 6 / 7 | | 2.5σ | −813pips | +896pips | 2 / 7 | | 3σ | −125pips | +462pips | 3 / 7 | **偏差1σは7通り全部が両方の年で黒字**です。バンドを狭くするほどブレイクの回数が増え、標本が増えます。逆に2.5σ以上はブレイク自体が珍しくなり、年による振れが大きくなりました。 **逆張りの偏差**（買い側） | 偏差 | 2025年の平均 | 2024年の平均 | 両方の年で黒字 | |---|---|---|---| | 1σ | −741pips | −27pips | 1 / 28 | | 1.5σ | −532pips | +5pips | 4 / 28 | | 2σ | −252pips | +83pips | 6 / 28 | | 2.5σ | −44pips | +27pips | 6 / 28 | | 3σ | −14pips | +107pips | 8 / 28 | 逆張りは順張りと反対で、**バンドを広げるほどマシになります**。ただし「マシ」であって、2025年はどの偏差でも平均がマイナスです。偏差1σの逆張りは28通り中1通りしか両方の年で黒字になりませんでした。 期間についても同じ向きで、1時間足の逆張り買いは期間50が20通り中7通り、期間20は20通り中0通りです。**逆張りで残るのは「めったに触れない、遅い設定」だけ**でした。 ## 去年トップだった設定は今年どうか 逆張りの首位は翌年に中央値も下回り、順張りの首位は中央値を上回りました。2024年に年間損益が最も大きかった設定を、同じ条件のまま2025年で検証します。 | 売買ルール | 設定 | 選定年 | 選定年の年間損益 | 取引回数 | 検証年 | 検証年の年間損益 | 検証年の中央値 | |---|---|---:|---:|---:|---:|---:|---:| | 逆張り・買い（1時間足） | 期間40・2σ→1.5σ | 2024年 | +1,029.9pips | 67回 | 2025年 | −268.0pips | −251.7pips | | 順張り（1時間足） | 期間20・1.5σ | 2024年 | +2,637.8pips | 195回 | 2025年 | +1,488.1pips | +788.6pips | 逆張りの期間40・2σ→1.5σは2025年に−268.0pipsとなり、同年の中央値−251.7pipsも下回りました。順張りの期間20・1.5σは2025年も+1,488.1pipsで、同年の中央値+788.6pipsを上回りました。 逆向きに見ると、期間10・1.5σの順張りは2025年に363回で+1,838.4pipsでしたが、2024年は−156.5pipsです。**順張りでも、片方の年の年間損益だけでは設定を選べません。** ## フィルターと損切り利確 1時間足の2つの代表設定に、よく紹介される条件を足しました。 | 売買ルール | 条件 | 年 | 取引回数 | 年間損益 | |---|---|---:|---:|---:| | 逆張り買い20/2σ→1σ | そのまま | 2025年 | 167回 | +71.0pips | | 逆張り買い20/2σ→1σ | そのまま | 2024年 | 157回 | −597.6pips | | 逆張り買い20/2σ→1σ | ADX 20以上 | 2025年 | 120回 | +385.6pips | | 逆張り買い20/2σ→1σ | ADX 20以上 | 2024年 | 111回 | −728.0pips | | 逆張り買い20/2σ→1σ | ADX 25以上 | 2025年 | 97回 | +735.5pips | | 逆張り買い20/2σ→1σ | ADX 25以上 | 2024年 | 75回 | −1,055.9pips | | 逆張り買い20/2σ→1σ | ADX 30以上 | 2025年 | 73回 | +836.7pips | | 逆張り買い20/2σ→1σ | ADX 30以上 | 2024年 | 53回 | −879.3pips | | 逆張り買い20/2σ→1σ | ロンドン・NY時間 | 2025年 | 140回 | +309.8pips | | 逆張り買い20/2σ→1σ | ロンドン・NY時間 | 2024年 | 131回 | −691.5pips | | 逆張り買い20/2σ→1σ | 東京時間 | 2025年 | 91回 | +2.2pips | | 逆張り買い20/2σ→1σ | 東京時間 | 2024年 | 81回 | −727.9pips | | 順張り20/2σ | そのまま | 2025年 | 132回 | +1,177.5pips | | 順張り20/2σ | そのまま | 2024年 | 137回 | +1,434.6pips | | 順張り20/2σ | ADX 20以上 | 2025年 | 94回 | +703.6pips | | 順張り20/2σ | ADX 20以上 | 2024年 | 106回 | −322.6pips | | 順張り20/2σ | ADX 25以上 | 2025年 | 69回 | −199.2pips | | 順張り20/2σ | ADX 25以上 | 2024年 | 74回 | −120.3pips | | 順張り20/2σ | ADX 30以上 | 2025年 | 52回 | −445.6pips | | 順張り20/2σ | ADX 30以上 | 2024年 | 49回 | +263.3pips | | 順張り20/2σ | ロンドン・NY時間 | 2025年 | 114回 | −54.1pips | | 順張り20/2σ | ロンドン・NY時間 | 2024年 | 113回 | +1,599.8pips | | 順張り20/2σ | 東京時間 | 2025年 | 84回 | +78.2pips | | 順張り20/2σ | 東京時間 | 2024年 | 79回 | +449.2pips | ADXフィルターは逆張りの2025年を黒字に変えますが、2024年には逆に働いて赤字を広げます。順張りに足すと、**2025年も2024年も基準を下回りました。** トレンドの強さで絞ればトレンド追随がよくなる、とはなりませんでした。 損切りと利確も足しました。 | 売買ルール | 決済ルール | 年 | 年間損益 | |---|---|---:|---:| | 逆張り買い20/2σ→1σ | そのまま | 2025年 | +71.0pips | | 逆張り買い20/2σ→1σ | そのまま | 2024年 | −597.6pips | | 逆張り買い20/2σ→1σ | 損切り30・利確60 | 2025年 | −230.9pips | | 逆張り買い20/2σ→1σ | 損切り30・利確60 | 2024年 | −530.5pips | | 逆張り買い20/2σ→1σ | 損切り50・利確100 | 2025年 | −197.8pips | | 逆張り買い20/2σ→1σ | 損切り50・利確100 | 2024年 | −531.7pips | | 逆張り買い20/2σ→1σ | 損切り100・利確200 | 2025年 | −179.3pips | | 逆張り買い20/2σ→1σ | 損切り100・利確200 | 2024年 | +41.2pips | | 逆張り買い20/2σ→1σ | 24本で時間決済 | 2025年 | +303.4pips | | 逆張り買い20/2σ→1σ | 24本で時間決済 | 2024年 | −635.2pips | | 順張り20/2σ | そのまま | 2025年 | +1,177.5pips | | 順張り20/2σ | そのまま | 2024年 | +1,434.6pips | | 順張り20/2σ | 損切り30・利確60 | 2025年 | +193.6pips | | 順張り20/2σ | 損切り30・利確60 | 2024年 | +231.7pips | | 順張り20/2σ | 損切り50・利確100 | 2025年 | +54.5pips | | 順張り20/2σ | 損切り50・利確100 | 2024年 | +1,121.3pips | | 順張り20/2σ | 損切り100・利確200 | 2025年 | +1,328.6pips | | 順張り20/2σ | 損切り100・利確200 | 2024年 | +1,128.8pips | | 順張り20/2σ | 24本で時間決済 | 2025年 | +1,413.8pips | | 順張り20/2σ | 24本で時間決済 | 2024年 | +1,536.2pips | 固定の損切り・利確は、エントリー約定価格から指定pips離れた**約定水準**として扱っています。たとえば利確60pipsなら、その水準へ到達した取引はちょうど+60pipsで決済し、さらに決済側のコストを重ねて差し引きません。 **24本の時間決済だけが、順張りで両方の年とも基準を上回りました**（+1,177.5→+1,413.8、+1,434.6→+1,536.2）。損切りを浅くするほど順張りの成績は落ちます。抜けた方向に伸びるのを待つ仕組みなので、途中で切ると本体が消えるためです。 逆張りには、両方の年で効いた決済ルールがありませんでした。 ## スプレッドで利益はどれだけ減るか 逆張りでは約4割が消えました。スプレッドだけを変えて2025年を回し直します。 | スプレッド | 逆張り買い 20/2σ→1σ（167回） | 順張り 20/2σ（132回） | |---|---|---| | 0.0pips | +121.1 | +1,217.1 | | 0.3pips | +71.0 | +1,177.5 | | 0.6pips | +20.9 | +1,137.9 | | 1.0pips | −45.9 | +1,085.1 | | 1.5pips | −129.4 | +1,019.1 | | 2.0pips | −212.9 | +953.1 | **逆張りはスプレッドが0なら+121.1pips、0.3pipsを払うと+71.0pips**です。コストは利益を約4割削りました。失った50.1pipsは167回×0.3pipsと一致します。 順張りは2.0pipsでも+953.1pipsが残ります。0から2.0pipsで失うのは264.0pips（132回×2.0）で、こちらも一致しました。この関係はこのシリーズの[QQE](/ja/blog/qqe-settings)や[移動平均線クロス](/ja/blog/moving-average-cross-settings)でも誤差なく成立しています。 勝率の高さを根拠に逆張りを選ぶ、という判断はこの検証の範囲では支持できませんでした。[偏差を0.5〜4σまで0.25刻みで比べた記事](/ja/blog/bollinger-band-sigma-settings)では、「何シグマが一番いいのか」を順張りと逆張りに分けて掘り下げています。[BBWの125本安値を検証した記事](/ja/blog/bollinger-bandwidth-settings)は、上下バンドの距離をスクイーズ条件として使います。[ウィリアムズ%Rでも、高勝率の設定ほど年間損益が良いとは限りませんでした](/ja/blog/williams-r-settings)。[RSIの逆張りと水準抜け順張りを比べた記事](/ja/blog/rsi-settings)では、平均勝率が逆張り65.3%、順張り23.7%でした。バンドのタッチをそのRSIの絞り込みに使った場合は[RSIとボリンジャーバンドの組み合わせ](/ja/blog/rsi-bollinger-combo)にあります。線を1本引いてそこで止まるかを問う売買ルールは[フィボナッチ・リトレースメント](/ja/blog/fibonacci-retracement-levels)でも測り、水準を401通りに刻んで「その数字でなければいけないのか」まで検証しています。バンドの幅を標準偏差ではなく価格に対する割合で決めた場合は[移動平均線乖離率](/ja/blog/ma-disparity-settings)で、そちらは同じ2本の線を終値だけで判定しています。[プログラミングなしで条件を組んで検証する](/ja/blog/backtest-without-coding)なら、この記事の表と同じことを自分の通貨ペア・自分の期間で確かめられます。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです。他の通貨ペアや他の年で同じ傾向になる保証はありません - 逆張りの決済は「反対側のバンドに触れるまで」です。損切りを置かない形なので、外したときの負け幅がそのまま出ています。損切りを付けた場合は別のセクションで測りました - 順張りは終値で判定しています。ヒゲでバンドを抜けて終値では戻った足は、シグナルになりません - エントリーも決済も足の終値で行っています。実際の約定はここからずれます - スプレッドは全期間0.3pips固定として計算しています。実際のスプレッドは時間帯と指標発表で変動します - 4時間足は設定によって年4〜153回（逆張り買い）と幅があり、順張りには1回も取引が出なかった設定が1つあります。件数の少ない設定の勝率やPFは、そもそも幅の広い数字です --- ### 移動平均線のゴールデンクロスは勝てる？280通り検証した結果 URL: https://formiq.jp/ja/blog/moving-average-cross-settings Language: ja Published: 2026-08-25 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: 移動平均線, ゴールデンクロス, インジケーター, パラメーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/moving-average-cross-settings ゴールデンクロスとデッドクロスを使う基本ルールは、移動平均線を2本使い、短期線が長期線を上抜けたら買い、下抜けたら売るというものです。 この記事では、期間、計算方法、時間足を変えたときに結果がどう変わるかを検証します。以降、短期線が10、長期線が20の組み合わせを「10/20」と表記します。 短期線と長期線の組み合わせ56通りに、5つの計算方法を掛けた280通りを、ドル円の15分足・1時間足・4時間足でバックテストしました。対象期間は2024年と2025年です。 2025年に年間損益が最も大きかったのは、15分足の12/50・加重移動平均（WMA）でした。751回の取引で+2,978.0pips、勝率33.95%です。ただし、4時間足では黒字設定が2024年の243/280から、2025年には60/280へ減りました。**特定の期間を選ぶことより、時間足と相場が一方向に動いたかどうかの影響が大きい結果でした。** ## 移動平均線クロスは何を測っているのか 移動平均は、直近n本の価格をならした線です。ならす本数（期間）を増やすほど価格から離れ、向きが変わるまでの時間が長くなります。 2本の上下が入れ替わるのは、直近の価格が以前の平均から離れて動いたときです。そのため、クロスは値動きの方向が変わった可能性を示します。 設定項目は3つです。 | 設定項目 | 数値や種類を変えたときの影響 | |---|---| | 短期線の期間 | 短期線の反応が遅くなり、クロスが減る | | 長期線の期間 | 長期線の反応が遅くなり、保有時間が長くなる | | 計算方法 | 同じ期間でも価格との距離とクロス回数が変わる | 1時間足・2025年の平均取引回数は、長期線20で398回、長期線200で108回でした。計算方法別では、ハル移動平均（HMA）が455回、平滑移動平均（SMMA）が86回です。期間と計算方法は、どちらも取引回数を変えます。 ## チャートに表示する方法 移動平均線は価格チャートに重ねて表示します。短期線と長期線を1本ずつ追加し、それぞれの期間と計算方法を設定します。 | 環境 | 手順 | |---|---| | MT4 / MT5 | Moving Averageを2本追加する。SMA・EMA・SMMA・線形加重を選べる | | TradingView | Moving Averageを2本追加し、それぞれ期間と計算方法を設定する | | ブラウザ（Formiq） | 移動平均線を2本追加する。SMA・EMA・WMA・SMMA・HMAを選べる | この記事では、単純移動平均（SMA）、指数移動平均（EMA）、加重移動平均（WMA）、平滑移動平均（SMMA）、ハル移動平均（HMA）の5種類を比べます。HMAはMT4とMT5の標準機能には含まれません。 ## 検証の条件 検証条件は次のとおりです。 | 項目 | 値 | |---|---| | 通貨ペア | ドル円 | | 期間 | 2025年1月1日〜12月31日（比較用に2024年も同条件） | | 時間足 | 15分足・1時間足・4時間足 | | 検証本数 | 2025年で15分足24,903本、1時間足6,226本、4時間足1,610本 | | 買い | 短期線が長期線を上抜けた足の終値 | | 売り | 短期線が長期線を下抜けた足の終値 | | 決済 | 反対側のクロス。同じ足の終値でそのまま反対方向へ建て直す（ドテン）。損切り・利確・時間決済は使わない | | スプレッド | 0.3pips（スリッページ0、0.1ロット）。決済と新規の両方に掛かる | | 組み合わせ | 短期線8種（5・8・10・12・15・20・25・50）×長期線8種（20・25・30・40・50・75・100・200）のうち、短期線が短い56通り。5つの計算方法を掛けて計280通り | | 測定 | Formiqのバックテスト機能で実行し、1取引ずつの約定価格からpipsを再集計 | 基本検証では短期線と長期線に同じ計算方法を使います。2本で異なる計算方法を使う場合は、後半で分けて検証します。 損切りを置かず反対クロスまで持つのは、移動平均クロス以外の要素を混ぜないためです。損切りを付けた場合は「フィルターと損切り利確」で別に測っています。 ## どの時間足で使うのがいいか 4時間足は年による差が最も大きく、両年黒字は280通り中32通りでした。同じ280通りを時間足ごとに比べます。 | 時間足 | 2025年に黒字だった数 | 2025年の中央値 | 2024年に黒字だった数 | 2024年の中央値 | 両方の年で黒字 | |---|---|---|---|---|---| | 15分足 | 231/280 | +810pips | 234/280 | +1,420pips | 205/280 | | 1時間足 | 188/280 | +445pips | 259/280 | +1,671pips | 172/280 | | 4時間足 | 60/280 | −1,153pips | 243/280 | +2,151pips | 32/280 | 4時間足は、2024年には243/280が黒字でしたが、2025年は60/280に減りました。年間損益の中央値も+2,151pipsから−1,153pipsへ反転し、両年とも黒字だったのは32/280です。 15分足では205/280、1時間足では172/280が両年とも黒字でした。この2年間では、4時間足より年をまたいで黒字を保った設定が多くなっています。 2024年と2025年では、価格の進み方が違いました。2024年は年初140.873から年末157.197まで1,632pips上昇しました。2025年は年初157.227、年末156.670で、年間の差は−56pipsです。高値と安値の差は2024年が2,237pips、2025年が1,900pipsで大きく変わりません。**2024年は上昇方向が続き、2025年は大きく動いても年末には年初付近へ戻りました。** 移動平均線クロスは一方向の値動きが続く場面で利益を伸ばすため、この違いが4時間足の結果に強く表れました。 ## 定番の組み合わせはどうだったか よく使われる組み合わせの年間損益を比べます。 | 組み合わせ | 時間足 | 2025年 | 2024年 | |---|---|---:|---:| | 5/20・SMA | 15分足 | +264.5pips | +646.1pips | | 5/20・SMA | 1時間足 | −484.9pips | +3,744.7pips | | 10/20・SMA | 15分足 | +1,015.0pips | +1,600.9pips | | 10/20・SMA | 1時間足 | +377.4pips | +2,888.4pips | | 12/25・EMA | 15分足 | +1,436.6pips | +2,102.1pips | | 12/25・EMA | 1時間足 | +970.9pips | +440.6pips | | 20/50・SMA | 15分足 | +1,774.9pips | +1,464.8pips | | 20/50・SMA | 1時間足 | −220.5pips | −34.6pips | | 25/75・SMA | 15分足 | −1,023.1pips | +3,613.7pips | | 25/75・SMA | 1時間足 | −56.8pips | −253.1pips | | 50/200・SMA | 15分足 | +629.3pips | +301.2pips | | 50/200・SMA | 1時間足 | −1,340.4pips | +2,435.4pips | 25/75・SMAは、15分足で2024年に+3,613.7pips、2025年に−1,023.1pipsでした。5/20・SMAも、1時間足で+3,744.7pipsから−484.9pipsへ反転しています。**よく知られた期間でも、2年続けて同じ結果にはなりませんでした。** 初期設定の10/20・SMAは、15分足では2年とも黒字でした。1時間足も黒字ですが、4時間足では2025年に−1,336.2pips、2024年に−1,245.7pipsとなり、両年とも赤字です。 ## 50/200は有効だったか 取引回数が少なく、勝率だけでは有効と判断できませんでした。よく知られた50/200・SMAを時間足ごとに比べます。 | 時間足 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | 15分足 | 2025年 | 147回 | 38.10% | +629.3pips | | 15分足 | 2024年 | 147回 | 29.93% | +301.2pips | | 1時間足 | 2025年 | 47回 | 31.91% | −1,340.4pips | | 1時間足 | 2024年 | 27回 | 55.56% | +2,435.4pips | | 4時間足 | 2025年 | 14回 | 50.00% | +173.6pips | | 4時間足 | 2024年 | 7回 | 71.43% | +1,357.5pips | 4時間足・2024年の勝率71.43%は、7回中5回の勝ちで出た数字です。取引回数が少ないため、勝率だけで50/200が有効とは判断できません。1時間足では、2024年の+2,435.4pipsから2025年の−1,340.4pipsへ反転しています。 ## 長期線は長いほどいいのか 伸ばすほど年による差が広がりました。1時間足について、長期線の期間ごとに結果をまとめます。 | 長期の期間 | 2025年の平均損益 | 2024年の平均損益 | 両方の年で黒字 | |---|---|---|---| | 20 | +736pips | +1,899pips | 20/25 | | 25 | +992pips | +1,798pips | 30/30 | | 30 | +822pips | +1,474pips | 31/35 | | 40 | +709pips | +1,388pips | 26/35 | | 50 | +567pips | +1,468pips | 27/35 | | 75 | −129pips | +1,293pips | 17/40 | | 100 | −158pips | +1,679pips | 17/40 | | 200 | −1,179pips | +1,964pips | 4/40 | 長期線25では30/30が両年とも黒字でしたが、長期線200では4/40です。長期線200の平均損益は、2024年の+1,964pipsから2025年の−1,179pipsへ反転しました。**長期線を伸ばすほど、年による差が大きくなっています。** 15分足の長期線200は、両年とも黒字だった設定が24/40でした。長期線の影響は1時間足より小さく、時間足によっても結果が変わります。 ## 去年よかった設定は今年も再現するか 再現しませんでした。2024年に年間損益が最も大きかった設定を、同じ条件で2025年にも使います。 | 時間足 | 2024年にいちばん良かった設定 | 2024年の年間損益 | 2025年の年間損益 | 2025年の中央値 | |---|---|---|---|---| | 15分足 | 20/75・SMA | +3,995.2pips（387回） | −234.6pips | +809.8pips | | 1時間足 | 5/25・SMA | +3,828.5pips（309回） | +583.3pips | +445.0pips | | 4時間足 | 8/50・WMA | +3,378.8pips（44回） | −1,367.2pips | −1,152.9pips | 15分足と4時間足では、2024年にいちばん良かった設定が2025年に赤字となり、2025年の中央値も下回りました。1時間足は黒字でしたが、中央値を少し上回る程度です。**前の年の成績で設定を選んでも、翌年の好成績は再現しませんでした。** ## 平均の種類はどれがいいか 時間足ごとに違い、15分足はEMA、1時間足はHMA、4時間足はSMMAが最多でした。2024年と2025年の両方で黒字だった設定数を、計算方法別に数えます。各計算方法につき56通りです。 | 計算方法 | 15分足 | 1時間足 | 4時間足 | |---|---|---|---| | SMA | 48/56 | 28/56 | 2/56 | | EMA | 52/56 | 34/56 | 2/56 | | WMA | 49/56 | 39/56 | 8/56 | | SMMA | 43/56 | 22/56 | 13/56 | | HMA | 13/56 | 49/56 | 7/56 | 黒字設定が最も多かったのは、15分足ではEMAの52/56、1時間足ではHMAの49/56、4時間足ではSMMAの13/56でした。HMAは15分足では13/56に減り、平均取引回数は1,852回と、他の4種類の324〜875回を上回りました。計算方法を選ぶときは、黒字設定数と取引回数を時間足ごとに確認する必要があります。 ### SMAとEMAを同じ期間で比べると SMAは直近n本を同じ重さで平均し、n本より古い終値は計算から外します。EMAは新しい終値ほど重くし、古い終値の影響を少しずつ減らします。この違いから「EMAのほうが速い」と説明されますが、速く反応することと損益が良くなることは別です。 そこで、短期と長期の期間、時間足、検証期間をそろえ、SMAとEMAだけを入れ替えて672通りを1対1で比べました。 年間損益がEMAのほうが大きかったのは672通り中318通り、SMAは354通りでした。EMAの取引回数が少なかったのは405通りです。EMAは取引回数を減らす傾向がありましたが、年間損益ではSMAとほぼ同数でした。 | 時間足 | EMAで両年黒字 | SMAで両年黒字 | 2024年に上回った種類が2025年も上回った数 | |---|---:|---:|---:| | 15分足 | 52 / 56 | 48 / 56 | 19 / 56 | | 1時間足 | 34 / 56 | 28 / 56 | 26 / 56 | | 4時間足 | 2 / 56 | 2 / 56 | 25 / 56 | 両年黒字の設定数は15分足と1時間足でEMAがやや多く、4時間足は同数でした。一方、前年に年間損益が大きかった種類が翌年も上回った数は、3時間足とも半数未満です。15分足の10/20も、2025年にはEMA、2024年にはSMAが上回りました。**前年の年間損益だけでSMAかEMAを固定する根拠は見つかりませんでした。** 重みの配り方、反応速度、初期履歴の影響まで含めた比較は、[SMAとEMAのどちらを使うべきか](/ja/blog/sma-vs-ema)で詳しく検証しています。 ## 2本で違う種類を混ぜるとどうなるか 期間の長いほうが速く動き、短期線と長期線の役割が逆転することがあります。短期線と長期線に異なる計算方法を使う25通りを、1時間足で検証しました。次の図は、各移動平均線と終値の平均距離です。 期間20のHMAは終値から平均18.7pips、期間10のSMAは22.1pips離れていました。期間が長いHMAのほうが価格に近いため、10/20で長期線だけをHMAにすると、長期線が短期線より速く動きます。 2024年の1時間足では、短期線SMA・長期線HMAの10/20が385回で−3,393.4pips、短期線HMA・長期線SMAは397回で+3,739.2pipsでした。**異なる計算方法を使う場合は、期間だけで短期線と長期線を判断できません。** ## 勝率を上げると利益は増えるのか 15分足で勝率が最も高かった15/20・HMAは、2,680回の取引で勝率44.22%、年間損益−755.1pipsでした。一方、年間損益が最も大きかった12/50・WMAの勝率は33.95%です。平均利益+56.74pipsに対して平均損失は−23.17pipsで、低い勝率を1回あたりの利益幅が補っています。 4時間足には勝率100%の設定もありましたが、取引は2回だけです。**勝率だけでは判断せず、取引回数と年間損益も確認する必要があります。** ## フィルターと損切り利確 15分足の10/20・SMAにフィルターを加えます。 | 条件 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | 10/20・SMA | 2025年 | 1,402回 | 37.52% | +1,015.0pips | | 10/20・SMA | 2024年 | 1,397回 | 38.08% | +1,600.9pips | | ＋ADX 20以上 | 2025年 | 674回 | 35.31% | −495.4pips | | ＋ADX 20以上 | 2024年 | 705回 | 37.02% | +923.1pips | | ＋ADX 25以上 | 2025年 | 354回 | 34.75% | −271.0pips | | ＋ADX 25以上 | 2024年 | 400回 | 40.25% | +1,847.9pips | | ＋ADX 30以上 | 2025年 | 170回 | 34.71% | +195.7pips | | ＋ADX 30以上 | 2024年 | 217回 | 37.33% | +703.8pips | | ＋ロンドン・NY時間（UTC 7〜21） | 2025年 | 802回 | 37.16% | +846.9pips | | ＋ロンドン・NY時間（UTC 7〜21） | 2024年 | 797回 | 37.77% | +41.1pips | | ＋東京時間（UTC 0〜8） | 2025年 | 460回 | 38.91% | +309.7pips | | ＋東京時間（UTC 0〜8） | 2024年 | 443回 | 39.73% | +2,194.9pips | ADX 25以上は、2024年には年間損益を+1,600.9pipsから+1,847.9pipsへ増やしましたが、2025年は+1,015.0pipsから−271.0pipsへ悪化させました。東京時間への限定も、2024年は改善し、2025年は悪化しています。**どちらのフィルターも2年続けて改善しませんでした。** 15分足の2設定には、損切り・利確・時間決済も加えました。 | 短期／長期 | 種類 | 年 | 決済ルール | 年間損益 | |---|---|---:|---|---:| | 10／20 | SMA | 2025年 | 反対クロスのみ | +1,015.0pips | | 10／20 | SMA | 2024年 | 反対クロスのみ | +1,600.9pips | | 10／20 | SMA | 2025年 | 損切り30・利確60 | +260.0pips | | 10／20 | SMA | 2024年 | 損切り30・利確60 | +187.9pips | | 10／20 | SMA | 2025年 | 損切り50・利確100 | +535.1pips | | 10／20 | SMA | 2024年 | 損切り50・利確100 | +961.9pips | | 10／20 | SMA | 2025年 | 損切り30・利確90 | +615.6pips | | 10／20 | SMA | 2024年 | 損切り30・利確90 | +1,167.5pips | | 10／20 | SMA | 2025年 | 24本で時間決済 | +621.7pips | | 10／20 | SMA | 2024年 | 24本で時間決済 | +273.0pips | | 12／50 | WMA | 2025年 | 反対クロスのみ | +2,978.0pips | | 12／50 | WMA | 2024年 | 反対クロスのみ | +2,103.7pips | | 12／50 | WMA | 2025年 | 損切り30・利確60 | +1,865.5pips | | 12／50 | WMA | 2024年 | 損切り30・利確60 | +599.0pips | | 12／50 | WMA | 2025年 | 損切り50・利確100 | +2,498.5pips | | 12／50 | WMA | 2024年 | 損切り50・利確100 | +774.7pips | | 12／50 | WMA | 2025年 | 損切り30・利確90 | +2,314.2pips | | 12／50 | WMA | 2024年 | 損切り30・利確90 | +1,405.1pips | | 12／50 | WMA | 2025年 | 24本で時間決済 | +641.2pips | | 12／50 | WMA | 2024年 | 24本で時間決済 | −108.5pips | どちらの設定も、2024年・2025年ともに「反対クロスのみ」の年間損益が最も大きくなりました。 ただし1時間足の10/20・SMAでは、損切り30・利確90が2025年に+1,260.9pipsとなり、反対クロスのみの+377.4pipsを上回りました。同じ決済ルールは2024年に下回っており、時間足と年が変わると結果も変わります。 ## スプレッドはいくら持っていくか スプレッドだけを変えて2025年を回し直すと、失う額はきれいに取引回数に比例します。 | 時間足 | 短期／長期 | 種類 | 取引回数 | スプレッド | 年間損益 | |---|---|---|---:|---:|---:| | 15分足 | 10／20 | SMA | 1,402回 | 0.0pips | +1,435.6pips | | 15分足 | 10／20 | SMA | 1,402回 | 0.3pips | +1,015.0pips | | 15分足 | 10／20 | SMA | 1,402回 | 0.6pips | +594.4pips | | 15分足 | 10／20 | SMA | 1,402回 | 1.0pips | +33.6pips | | 15分足 | 10／20 | SMA | 1,402回 | 1.5pips | −667.4pips | | 15分足 | 10／20 | SMA | 1,402回 | 2.0pips | −1,368.4pips | | 15分足 | 12／50 | WMA | 751回 | 0.0pips | +3,203.3pips | | 15分足 | 12／50 | WMA | 751回 | 0.3pips | +2,978.0pips | | 15分足 | 12／50 | WMA | 751回 | 0.6pips | +2,752.7pips | | 15分足 | 12／50 | WMA | 751回 | 1.0pips | +2,452.3pips | | 15分足 | 12／50 | WMA | 751回 | 1.5pips | +2,076.8pips | | 15分足 | 12／50 | WMA | 751回 | 2.0pips | +1,701.3pips | | 1時間足 | 10／20 | SMA | 358回 | 0.0pips | +484.8pips | | 1時間足 | 10／20 | SMA | 358回 | 0.3pips | +377.4pips | | 1時間足 | 10／20 | SMA | 358回 | 0.6pips | +270.0pips | | 1時間足 | 10／20 | SMA | 358回 | 1.0pips | +126.8pips | | 1時間足 | 10／20 | SMA | 358回 | 1.5pips | −52.2pips | | 1時間足 | 10／20 | SMA | 358回 | 2.0pips | −231.2pips | | 4時間足 | 50／200 | SMA | 14回 | 0.0pips | +177.7pips | | 4時間足 | 50／200 | SMA | 14回 | 0.3pips | +173.6pips | | 4時間足 | 50／200 | SMA | 14回 | 0.6pips | +169.4pips | | 4時間足 | 50／200 | SMA | 14回 | 1.0pips | +163.8pips | | 4時間足 | 50／200 | SMA | 14回 | 1.5pips | +156.8pips | | 4時間足 | 50／200 | SMA | 14回 | 2.0pips | +149.8pips | スプレッドを0から2.0pipsへ広げると、15分足の10/20・SMAは2,804.0pips悪化しました。これは1,402回×2.0pipsと一致します。**スプレッドによる損失は、取引回数に比例します。** 15分足の10/20・SMAは、スプレッド1.0pipsで年間損益がほぼゼロになり、1.5pipsでは赤字です。取引回数が少ない12/50・WMAは、2.0pipsでも+1,701.3pipsが残りました。短い時間足では、期間だけでなく実際のスプレッドも確認する必要があります。 ## 関連記事 - [SMAとEMAのどちらを使うべきか](/ja/blog/sma-vs-ema) - [GMMAの設定を検証](/ja/blog/gmma-settings) - [移動平均線乖離率の設定を検証](/ja/blog/ma-disparity-settings) - [DMI/ADXの期間と水準を検証](/ja/blog/dmi-adx-settings) ## 補足 - 対象はドル円の2024年と2025年です。他の通貨ペアや期間は検証していません - エントリーと決済は足の終値で計算しています。実際の約定価格とはずれる場合があります - スプレッドは全期間0.3pips固定です。実際のスプレッドは時間帯や指標発表によって変動します - 長期線200は、検証開始前の200本も使って計算しています。過去データが少ないチャートでは同じ線にならない場合があります --- ### RSI「30以下で買い」は本当に勝てる？330通り検証した結果 URL: https://formiq.jp/ja/blog/rsi-settings Language: ja Published: 2026-08-25 Updated: 2026-09-02 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: RSI, 逆張り, 順張り, インジケーター, バックテスト, ドル円 Evidence: https://formiq.jp/ja/evidence/rsi-settings Evidence JSON: https://formiq.jp/ja/evidence/rsi-settings/data.json Also available in: en — https://formiq.jp/blog/rsi-settings RSI（相対力指数）は、直近N本の値上がり幅と値下がり幅を比べて0〜100の数字にした指標です。**30を下回ったら売られすぎ、70を上回ったら買われすぎ**という読み方が一番よく紹介されます。もうひとつ、**50を境に上なら強い・下なら弱いと読む**使い方もあります。 RSIが30以下なら買い、70以上なら売る読み方は逆張りです。RSIが設定水準を上抜けたら買い、下抜けたら売る読み方は順張りです。同じ通貨ペア・同じ期間・同じコストで、逆張りと水準抜け順張りを測りました。 ドル円の2025年（1月1日〜12月31日）で、期間6種×エントリー水準5種×決済水準5種の**逆張り150通りを買い側と売り側それぞれ**、期間6種×水準5種の**順張り30通り**、合わせて330通りをバックテストしています。時間足は15分足・1時間足・4時間足の3つ、比較のために2024年でも同じ330通りを回しました。 **1時間足で、平均勝率65.3%の逆張り（買い）は150通り中31通りしか2024年と2025年の両方で黒字にならず、平均勝率23.7%の順張りは30通り中17通りが黒字**でした。順張り側の勝率は、どの設定を取っても3割に届いていません。 ## RSIは何を測っているのか 計算は単純です。直近N本のうち、前の足より上がった分の平均と、下がった分の平均を出し、上がった分が全体の何割かを百分率にします。計算で設定する項目は**期間**（N）です。 - 上がった足しかなければ100 - 下がった足しかなければ0 - 上下が釣り合えば50 つまり50は「上げ幅と下げ幅が拮抗している」点で、そこから離れるほど片方向に偏っている、という意味になります。「買われすぎ」「売られすぎ」という言葉は解釈であって、計算そのものは偏りの度合いを測っているだけです。 ### 期間を伸ばすと、30には届かなくなる ここが実務上いちばん効いてきます。平均を取る本数が増えると、上げと下げが混ざって数字は50に寄ります。2025年の1時間足6,226本で、RSIが計算できた足を数えました。 | 期間 | 30を下回った割合 | 70を上回った割合 | 1年間の最小値 | 最大値 | |---|---|---|---|---| | 7 | 13.97% | 14.13% | 1.0 | 97.0 | | 9 | 10.81% | 11.20% | 2.4 | 94.5 | | 14 | 5.73% | 6.54% | 8.0 | 89.5 | | 21 | 1.80% | 4.17% | 16.4 | 84.7 | | 30 | 0.39% | 2.03% | 24.9 | 80.8 | | 50 | 0.00% | 0.32% | 31.2 | 74.7 | **期間50のRSIは、2025年の1年を通して一度も30を下回りませんでした。** 最小値は31.2です。「期間を長くして、だましの少ない売られすぎを拾う」という組み方は、拾う対象がそもそも存在しない、ということになります。4時間足ではさらに極端で、期間30も期間50も30を下回った足は0本でした。 この検証でも、4時間足の逆張り買い150通りのうち**50通りは1年間で1回もシグナルが出ていません**。1時間足でも20通りがゼロです。 ## チャートに表示する方法 RSIはどのプラットフォームにも標準で入っています。 | 環境 | 手順 | |---|---| | MT4 / MT5 | ナビゲータのインジケータ一覧から Relative Strength Index を選び、期間を設定する | | TradingView | インジケーター検索から RSI を追加し、期間を設定する | | ブラウザ（Formiq） | 最初から入っている。バックテスト側は逆張り（レベル判定）と順張り（レベル抜け）をトリガーで切り替える | 表示はどこでも同じですが、**売買ルールとして組むときに「水準に到達したら」なのか「水準を抜けたら」なのかを決める必要があります。** この記事では、RSIが設定した水準に到達したら売買する逆張りと、設定した水準を上抜けたら買い・下抜けたら売る順張りを、別々の売買ルールとして測っています。 ## 検証の条件 数字を見る前に、何をどう測ったかを書いておきます。この表と同じ設定を入れれば、同じ結果が出ます。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 2025年で15分足24,903本、1時間足6,226本、4時間足1,610本 | | 逆張り・買い | RSIがエントリー水準を下回った足の終値で買い、決済水準を上回った足の終値で決済 | | 逆張り・売り | 買い側を50で折り返した鏡。RSIが「100−エントリー水準」を上回ったら売り、「100−決済水準」を下回ったら決済 | | 順張り | RSIが水準を上抜けたら買い、下抜けたら売り。反対向きに抜けたら同じ足でドテン | | 損切り・利確 | 使わない（別セクションで測定） | | スプレッド | 0.3pips（スリッページ0、0.1ロット）。決済と新規の両方に掛かる | | 逆張りの組み合わせ | 期間6種（7・9・14・21・30・50）× エントリー水準5種（20・25・30・35・40）× 決済水準5種（50・55・60・65・70）＝150通り。買い側と売り側で別々に実行 | | 順張りの組み合わせ | 期間6種 × 水準5種（40・45・50・55・60）＝30通り | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | 逆張りの買いと売りを別々に集計するのは、水準判定が買いと売りのどちらか一方しか作らないからです。「30を割ったら買う」と「70を超えたら売る」は、まったく別の成績になります。 ## 逆張りと順張り、どちらが勝てるか 勝率の高い逆張りより、勝率の低い順張りに利益が残りました。逆張り買い、逆張り売り、RSIが設定水準を抜けた方向へ売買する順張りの成績を、時間足ごとに並べます。 | 売買ルール | 時間足 | 2025年の黒字数 | 2025年の中央値 | 2024年の黒字数 | 2024年の中央値 | 両方の年で黒字 | 平均勝率 | |---|---|---|---|---|---|---|---| | 逆張り・買い | 15分足 | 77 / 150 | +5pips | 67 / 150 | −45pips | 29 / 150 | 64.6% | | 逆張り・買い | 1時間足 | 78 / 150 | +11pips | 58 / 150 | −65pips | 31 / 150 | 65.3% | | 逆張り・買い | 4時間足 | 81 / 150 | +95pips | 36 / 150 | −211pips | 13 / 150 | 67.6% | | 逆張り・売り | 15分足 | 42 / 150 | −264pips | 33 / 150 | −367pips | 5 / 150 | 57.7% | | 逆張り・売り | 1時間足 | 48 / 150 | −137pips | 23 / 150 | −727pips | 9 / 150 | 61.2% | | 逆張り・売り | 4時間足 | 84 / 150 | +272pips | 17 / 150 | −1,075pips | 1 / 150 | 61.0% | | 順張り | 15分足 | 8 / 30 | −347pips | 12 / 30 | −408pips | 4 / 30 | 22.3% | | 順張り | 1時間足 | 25 / 30 | +830pips | 22 / 30 | +846pips | 17 / 30 | 23.7% | | 順張り | 4時間足 | 7 / 30 | −885pips | 28 / 30 | +1,388pips | 6 / 30 | 23.0% | （平均勝率は2025年のうち取引が1回以上出た設定の平均。中央値はその時間足の全設定の損益の中央値） 冒頭の図が、この表の1時間足を1点ずつ描いたものです（年10回に満たない設定は除いています）。横軸が勝率、縦軸が年間の損益。**赤と紫（逆張り）は勝率40〜90%の帯に広く散らばり、青（順張り）は勝率17〜30%の細い帯に固まっています。** それでも中央値は青が+830pips、赤が+11pips、紫が−137pipsです。 **1時間足の順張りは、2025年に30通り中25通り、2024年に30通り中22通りが黒字**でした。両方の年で黒字は17通りです。このシリーズで測った指標の中では高い残存率になります。 いっぽう逆張りは、買い側で31通り、売り側で9通りでした。売り側が弱いのは、2024年のドル円が年間1,632pips上昇したためです。上げ相場で「買われすぎ」を売り続けると、そのぶん轢かれます。 ## RSIの期間は何がいいか 1つには決まらず、期間を伸ばすと逆張りと順張りで結果が分かれました。この検証でいちばんはっきり出たのは、**期間の効き方**です。 1時間足で、期間ごとの「2024年と2025年の両方で黒字だった数」を並べます。 | 期間 | 逆張り・買い | 順張り | |---|---|---| | 7 | 2 / 25 | 4 / 5 | | 9 | 3 / 25 | 5 / 5 | | 14 | 7 / 25 | 3 / 5 | | 21 | 7 / 25 | 2 / 5 | | 30 | 8 / 25 | 2 / 5 | | 50 | 4 / 25 | 1 / 5 | **買い逆張りは期間30、水準抜け順張りは期間9で、両年黒字の設定数が最も多くなりました。** 買い逆張りは期間30が25通り中8通り、水準抜け順張りは期間9が5通り中5通りです。期間50では、買い逆張りが4通り、水準抜け順張りが1通りまで減りました。 買い逆張りでは、両年黒字の設定数が期間7の2通りから期間30の8通りまで増えた後、期間50では4通りへ減りました。期間を長くするとRSIが30以下になる回数が減り、期間50では2025年の1時間足で一度も30を下回っていません。期間7から30にかけて両年黒字の設定が増えた理由は、この集計だけでは特定できません。確実に言えるのは、期間30から50に伸ばすと、両年黒字の設定数が8通りから4通りへ減ったことです。 水準抜け順張りでは、期間9が5水準すべてで両年黒字でした。期間を14、21、30、50へ伸ばすにつれて、両年黒字の水準は3、2、2、1通りへ減りました。期間9は1時間足で年775回、期間50は231回です。取引回数が減るほど、数回の大きな勝ち負けが年間損益に占める割合が大きくなります。 水準のほうも見ておきます。 | 逆張り・エントリー水準 | 両方の年で黒字 | |---|---| | 20 | 6 / 30 | | 25 | 13 / 30 | | 30 | 4 / 30 | | 35 | 2 / 30 | | 40 | 6 / 30 | | 順張り・水準 | 両方の年で黒字 | |---|---| | 40 | 3 / 6 | | 45 | 5 / 6 | | 50 | 4 / 6 | | 55 | 4 / 6 | | 60 | 1 / 6 | 逆張りの決済水準は、70に置いたときが最も残りました（9 / 30。50に置くと6 / 30、65で5 / 30）。**「30で買って70で売る」という組み合わせの後半だけは、この検証の範囲で妥当**でした。 ## 定番の14・30/70はどうだったか いちばんよく見かける逆張り設定「期間14、30で買い・70で決済」と、期間14・水準50の順張りを比べます。 | 使い方 | 時間足 | 2025年 | 2024年 | |---|---|---|---| | 逆張り・買い（30→70） | 15分足 | 105回・勝率63.81%・+332.1pips | 93回・勝率75.27%・+9.0pips | | 逆張り・買い（30→70） | 1時間足 | 25回・勝率60.00%・−250.7pips | 22回・勝率63.64%・+26.1pips | | 逆張り・買い（30→70） | 4時間足 | 9回・勝率66.67%・+1,148.4pips | 7回・勝率85.71%・+209.2pips | | 逆張り・売り（70→30） | 15分足 | 106回・勝率62.26%・+435.5pips | 94回・勝率50.00%・−1,545.0pips | | 逆張り・売り（70→30） | 1時間足 | 25回・勝率64.00%・−256.1pips | 22回・勝率40.91%・−1,429.7pips | | 逆張り・売り（70→30） | 4時間足 | 9回・勝率66.67%・+1,059.2pips | 8回・勝率50.00%・−1,150.7pips | | 順張り（50ライン） | 15分足 | 2,840回・勝率22.64%・−265.7pips | 2,820回・勝率21.99%・+94.9pips | | 順張り（50ライン） | 1時間足 | 638回・勝率24.14%・+1,272.0pips | 681回・勝率22.17%・+666.0pips | | 順張り（50ライン） | 4時間足 | 164回・勝率27.44%・−790.1pips | 141回・勝率23.40%・+1,798.5pips | **両方の年で黒字だったのは、15分足と4時間足の逆張り買い、そして1時間足の順張りだけ**です。 数字の中で目を引くのは2024年の15分足です。**93回のうち75.27%が勝ちトレードで、1年の合計は+9.0pips**でした。4回に3回当てて、残ったのが9pipsです。当たった回数が損益とほとんど関係していない、ということがそのまま出ています。 1時間足の逆張りは、年25回しかシグナルが出ません。**この件数で語れることは多くありません。** 勝率60%と書けば強そうに見えますが、25回のうち15回勝って10回負けた、という意味です。 順張りの1時間足は638回で+1,272.0pips、翌年は681回で+666.0pipsでした。回数が桁違いに多く、そのぶん年による振れが小さくなっています。 ## なぜ勝率が高いほうが負けるのか 1回あたりの勝ち幅と負け幅が非対称だからです。2025年の全設定を平均するとこうなります。 | 売買ルール | 時間足 | 平均勝ち幅 | 平均負け幅 | 平均保有本数 | |---|---|---|---|---| | 逆張り・買い | 1時間足 | +88.6pips | −120.8pips | 188.9本 | | 逆張り・売り | 1時間足 | +80.7pips | −129.6pips | 273.4本 | | 順張り | 1時間足 | +71.5pips | −20.5pips | 21.6本 | **順張りは、勝ち幅が負け幅の3.5倍あります。** 勝率23.7%でも、0.237×71.5 − 0.763×20.5 は正の値になります。抜けた方向に伸びれば持ち続け、反対に抜けたら同じ足でドテンするので、外れたときの傷が浅くなります。 逆張りは反対です。勝ち幅が負け幅の0.73倍しかありません。決済水準まで戻れば小さく勝ちますが、戻らずに走られたときの損失が大きくなります。保有本数の差も極端で、逆張りは平均189本（1時間足で約8日）、順張りは21.6本（約1日）でした。 15分足と4時間足でも関係は同じ向きです。逆張り買いは15分足で+49.6／−73.1、4時間足で+126.2／−196.2。順張りは15分足で+36.8／−10.4、4時間足で+120.1／−42.8でした。 ## 去年勝てた設定は今年どうなるか 逆張りの首位は翌年に赤字、順張りの首位は減っても黒字でした。2024年に年間損益が最も大きかった設定を、同じ条件のまま2025年で検証します。1時間足です。 | 売買ルール | 設定 | 選定年 | 選定年の年間損益 | 取引回数 | 検証年 | 検証年の年間損益 | 検証年の中央値 | |---|---|---:|---:|---:|---:|---:|---:| | 逆張り・買い | 期間14・40→70 | 2024年 | +1,139.8pips | 44回 | 2025年 | −798.8pips | +11.2pips | | 順張り | 期間9・水準45 | 2024年 | +3,601.3pips | 725回 | 2025年 | +1,254.9pips | +830.3pips | 逆張り・買いの期間14・40→70は、2024年に勝率77.27%・44回で+1,139.8pipsでしたが、2025年は−798.8pipsです。同年の150設定の中央値+11.2pipsも下回りました。 順張りの期間9・水準45は、2024年の+3,601.3pipsから2025年の+1,254.9pipsへ減りましたが、2025年の中央値+830.3pipsは上回りました。逆向きに見ると、2025年に+1,815.4pipsだった期間9・水準40は2024年に+1,003.6pipsです。個別設定の年間損益は変わっても、この2設定は両年とも黒字でした。 ## フィルターと損切り利確 1時間足の逆張り買い（期間14・30→70）と50ライン順張り（期間14・水準50）に、よく紹介される条件を足しました。 | 売買ルール | 条件 | 年 | 取引回数 | 年間損益 | |---|---|---:|---:|---:| | 逆張り買い14・30→70 | そのまま | 2025年 | 25回 | −250.7pips | | 逆張り買い14・30→70 | そのまま | 2024年 | 22回 | +26.1pips | | 逆張り買い14・30→70 | ADX 20以上 | 2025年 | 25回 | −225.0pips | | 逆張り買い14・30→70 | ADX 20以上 | 2024年 | 21回 | −99.2pips | | 逆張り買い14・30→70 | ADX 25以上 | 2025年 | 25回 | −58.2pips | | 逆張り買い14・30→70 | ADX 25以上 | 2024年 | 20回 | −164.8pips | | 逆張り買い14・30→70 | ADX 30以上 | 2025年 | 22回 | −200.8pips | | 逆張り買い14・30→70 | ADX 30以上 | 2024年 | 16回 | −403.2pips | | 逆張り買い14・30→70 | ロンドン・NY時間 | 2025年 | 22回 | −561.0pips | | 逆張り買い14・30→70 | ロンドン・NY時間 | 2024年 | 20回 | +100.2pips | | 逆張り買い14・30→70 | 東京時間 | 2025年 | 21回 | −113.6pips | | 逆張り買い14・30→70 | 東京時間 | 2024年 | 18回 | +53.1pips | | 順張り14・50 | そのまま | 2025年 | 638回 | +1,272.0pips | | 順張り14・50 | そのまま | 2024年 | 681回 | +666.0pips | | 順張り14・50 | ADX 20以上 | 2025年 | 290回 | +389.4pips | | 順張り14・50 | ADX 20以上 | 2024年 | 334回 | −827.0pips | | 順張り14・50 | ADX 25以上 | 2025年 | 163回 | +710.9pips | | 順張り14・50 | ADX 25以上 | 2024年 | 165回 | +94.6pips | | 順張り14・50 | ADX 30以上 | 2025年 | 79回 | +420.8pips | | 順張り14・50 | ADX 30以上 | 2024年 | 89回 | −152.8pips | | 順張り14・50 | ロンドン・NY時間 | 2025年 | 370回 | +1,669.9pips | | 順張り14・50 | ロンドン・NY時間 | 2024年 | 400回 | −117.9pips | | 順張り14・50 | 東京時間 | 2025年 | 247回 | +284.3pips | | 順張り14・50 | 東京時間 | 2024年 | 236回 | +1,254.5pips | **ADXフィルターは、順張りの両方の年で基準を下回りました。** 2025年は+1,272.0が+710.9へ、2024年は+666.0が+94.6へ（いずれもADX 25以上）。トレンドの強さで絞ればトレンド追随がよくなる、とはなりませんでした。取引ごとの効率は上がります（PFは1.146→1.322）が、残る額は減ります。 時間帯フィルターは年で反対を向きました。ロンドン・NY時間は2025年に+1,669.9まで伸びますが、2024年は−117.9で赤字に落ちます。東京時間はその逆です。**片方の年だけ見て選ぶと、翌年に反対の結果が出ます。** 損切りと利確も足しました。 | 売買ルール | 決済ルール | 年 | 年間損益 | |---|---|---:|---:| | 逆張り買い14・30→70 | そのまま | 2025年 | −250.7pips | | 逆張り買い14・30→70 | そのまま | 2024年 | +26.1pips | | 逆張り買い14・30→70 | 損切り30・利確60 | 2025年 | −360.0pips | | 逆張り買い14・30→70 | 損切り30・利確60 | 2024年 | +90.0pips | | 逆張り買い14・30→70 | 損切り50・利確100 | 2025年 | −243.6pips | | 逆張り買い14・30→70 | 損切り50・利確100 | 2024年 | +189.2pips | | 逆張り買い14・30→70 | 損切り100・利確200 | 2025年 | −513.0pips | | 逆張り買い14・30→70 | 損切り100・利確200 | 2024年 | −731.2pips | | 逆張り買い14・30→70 | 24本で時間決済 | 2025年 | −536.7pips | | 逆張り買い14・30→70 | 24本で時間決済 | 2024年 | −878.1pips | | 順張り14・50 | そのまま | 2025年 | +1,272.0pips | | 順張り14・50 | そのまま | 2024年 | +666.0pips | | 順張り14・50 | 損切り30・利確60 | 2025年 | +1,427.1pips | | 順張り14・50 | 損切り30・利確60 | 2024年 | −312.7pips | | 順張り14・50 | 損切り50・利確100 | 2025年 | +1,964.5pips | | 順張り14・50 | 損切り50・利確100 | 2024年 | −300.1pips | | 順張り14・50 | 損切り100・利確200 | 2025年 | +2,122.6pips | | 順張り14・50 | 損切り100・利確200 | 2024年 | +761.1pips | | 順張り14・50 | 24本で時間決済 | 2025年 | +2,960.5pips | | 順張り14・50 | 24本で時間決済 | 2024年 | −84.7pips | **順張りで両方の年とも基準を上回ったのは、損切り100／利確200だけ**でした（+1,272.0→+2,122.6、+666.0→+761.1）。浅い損切りは2025年には効き、2024年には赤字に落としています。24本の時間決済は2025年に+2,960.5で一番良かったのですが、2024年は−84.7で基準を割りました。 逆張りには、両方の年で基準を上回った決済ルールがありませんでした。 ## スプレッドで負けているのか 順張りは当てはまり、逆張りは当てはまりませんでした。スプレッドだけを変えて2025年を回し直します。 | スプレッド | 逆張り買い 14・30→70（25回） | 順張り 14・50（638回） | |---|---|---| | 0.0pips | −243.2 | +1,463.4 | | 0.3pips | −250.7 | +1,272.0 | | 0.6pips | −258.2 | +1,080.6 | | 1.0pips | −268.2 | +825.4 | | 1.5pips | −280.7 | +506.4 | | 2.0pips | −293.2 | +187.4 | **逆張りはスプレッドを0にしても−243.2pipsのままです。** 0から2.0pipsで失うのは50.0pips（25回×2.0）にすぎません。このシリーズで繰り返し出てきた「回数×スプレッドで負けている」という説明が、ここでは当てはまりません。**逆張りが負けているのはコストのせいではなく、トレードそのもののせいです。** 50ライン順張りは取引回数が多いため、コストの比重が大きくなります。0.3pipsで払う191.4pips（638回×0.3）は、スプレッド0の利益+1,463.4pipsの13%です。2.0pipsまで上がると1,276.0pipsを払い、残りは+187.4pipsまで細ります。**この関係はこのシリーズの[ボリンジャーバンド](/ja/blog/bollinger-band-settings)や[移動平均線クロス](/ja/blog/moving-average-cross-settings)でも誤差なく成立しています。** 勝率の高さを根拠にRSIの逆張りを選ぶ、という判断はこの検証の範囲では支持できませんでした。[RSIが70以上・30以下に何本続くかを測った検証](/ja/blog/rsi-overbought-oversold-sticking)では、水準へ入った次の足も同じ範囲に残る確率を調べています。[ウィリアムズ%Rでも、高勝率の設定ほど年間損益が良いとは限りませんでした](/ja/blog/williams-r-settings)。[ボリンジャーバンド](/ja/blog/bollinger-band-settings)では、バンドタッチ逆張りと終値ブレイク順張りを同じ条件で比較しています。その2つを重ねた場合は[RSIとボリンジャーバンドの組み合わせ](/ja/blog/rsi-bollinger-combo)で、同じ本数をランダムに間引いた場合と比べています。RSIを平滑化した[QQE](/ja/blog/qqe-settings)や、RSIをさらにストキャスティクスに掛けた[ストキャスRSI](/ja/blog/stoch-rsi-settings)も同じ手順で測りました。価格そのものに水準を引く[フィボナッチ・リトレースメント](/ja/blog/fibonacci-retracement-levels)でも、水準に到達したら売買するルールと、水準を抜けた方向へ売買するルールを比較しています。上下に枠のない[CCI](/ja/blog/cci-settings)の逆張りは6通りすべてが赤字でした。[プログラミングなしで条件を組んで検証する](/ja/blog/backtest-without-coding)なら、この記事の表と同じことを自分の通貨ペア・自分の期間で確かめられます。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです。他の通貨ペアや他の年で同じ傾向になる保証はありません - 逆張りの決済は「決済水準に届くまで」です。損切りを置かない形なので、外したときの負け幅がそのまま出ています。損切りを付けた場合は別のセクションで測りました - **逆張りは取引回数が少なすぎる設定を多く含みます。** 1時間足の中央値は年29回、4時間足では150通り中50通りが1回も取引していません。件数の少ない設定の勝率やPFは、そもそも幅の広い数字です - エントリーも決済も足の終値で行っています。実際の約定はここからずれます - スプレッドは全期間0.3pips固定として計算しています。実際のスプレッドは時間帯と指標発表で変動します --- ### ACオシレーターの時間足別成績：27通り検証 URL: https://formiq.jp/ja/blog/ac-oscillator-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ACオシレーター, アクセラレーター, インジケーター, パラメーター, ドル円 Also available in: en — https://formiq.jp/blog/ac-oscillator-settings ACオシレーター（アクセラレーター・オシレーター）は、オーサムオシレーターの「加速度」を見る指標です。勢いそのものではなく、勢いが増しているか衰えているかを棒グラフで表します。 ドル円の2025年（1月1日〜12月31日）で、短期・長期・平滑化の3つのパラメータを27通り組み合わせて検証しました。時間足は15分足・1時間足・4時間足、比較のために2024年でも同じ27通りを回しています。 **1時間足では27通り中24通りが、2025年と2024年の両方で黒字でした。** 一方、15分足では0通りです。 ## ACオシレーターはどの時間足で使えるか 黒字が残った時間足を並べます。 | 時間足 | 年 | 黒字の設定数 | 年間損益の中央値 | 年間の取引回数（中央値） | |---|---:|---:|---:|---:| | 15分足 | 2025年 | 7 / 27 | −437pips | 2,934回 | | 15分足 | 2024年 | 0 / 27 | −1,707pips | 対象外 | | 1時間足 | 2025年 | 26 / 27 | +1,292pips | 711回 | | 1時間足 | 2024年 | 25 / 27 | +1,814pips | 対象外 | | 4時間足 | 2025年 | 24 / 27 | +658pips | 184回 | | 4時間足 | 2024年 | 16 / 27 | +274pips | 対象外 | 15分足は年間2,930回売買します。この回数だとスプレッド0.3pipsでも年間880pips相当の負担になり、それを超える利益が残りませんでした。**2024年は27通り全部が赤字**で、両方の年で黒字だった設定はひとつもありません。 両方の年で黒字だった設定は、1時間足が24/27、4時間足が16/27です。4時間足は2025年に24/27まで増えましたが、2024年は16/27でした。 ## 一番良かった設定は前年も勝てたか 2025年に最大だった設定を、2024年へそのまま当てます。 | 設定 | 年 | 年間損益 | |---|---:|---:| | 初期設定 5/34/5 | 2025年 | +2,139.4pips | | 初期設定 5/34/5 | 2024年 | +1,436.8pips | | 2025年に最大だった3/34/3 | 2025年 | +3,359.1pips | | 2025年に最大だった3/34/3 | 2024年 | +1,108.6pips | | 2024年に最大だった3/55/8 | 2025年 | +1,031.6pips | | 2024年に最大だった3/55/8 | 2024年 | +3,400.9pips | | 2025年に最小だった8/21/3 | 2025年 | −175.0pips | | 2025年に最小だった8/21/3 | 2024年 | +2,716.1pips | 2025年に+3,359.1pipsだった3/34/3は、2024年には+1,108.6pipsまで減りました。逆に2025年に−175.0pipsだった8/21/3は、2024年には+2,716.1pipsです。**1年の損益だけでは、翌年の設定を選べませんでした。** そのなかで初期設定の5/34/5は2025年+2,139.4pips、2024年+1,436.8pipsと、**どちらの年も黒字**でした。 1時間足の取引回数は434回から1,208回でした。短期側を小さくすると回数が増えます。 ## 初期設定のままでも使えるか 4時間足の初期設定は2025年に+2,192.3pipsでした。 | 設定 | 年 | 年間損益 | |---|---:|---:| | 初期設定 5/34/5 | 2025年 | +2,192.3pips | | 初期設定 5/34/5 | 2024年 | +274.2pips | | 2024年に最大だった5/21/3 | 2025年 | +1,988.2pips | | 2024年に最大だった5/21/3 | 2024年 | +2,053.6pips | | 2025年に最小だった8/55/3 | 2025年 | −901.1pips | | 2025年に最小だった8/55/3 | 2024年 | −1,227.5pips | 4時間足では、初期設定の年間損益が2025年に試した27設定の中で最大でした。取引回数は190回、勝率は46.84%です。1時間足の711回・41.49%より標本が少ないため、この勝率差だけで時間足の性質とは判断できません。 ## 勝率で設定を選べるか 1時間足・2025年の27通りで、勝率は38.34%〜43.79%、中央値41.0%でした。範囲が5.45ポイントと狭い一方、年間損益は−175.0〜+3,359.1pipsに分かれています。**勝率だけでは年間損益を選別できません。** ゼロラインを抜けた方向に持ち続ける形なので、[OsMA](/ja/blog/osma-settings)や[スーパートレンド](/ja/blog/supertrend-settings)と同じく勝率は50%を下回ります。負けは小さく何度も出て、勝ちが伸びる形です。 ## チャートに表示する方法 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準で入っている。ナビゲーターの「Bill Williams」から Accelerator Oscillator を選ぶ | | TradingView | インジケーター検索で「Accelerator Oscillator」を選ぶ | | ブラウザ（Formiq） | インジケーター一覧から選ぶ。短期・長期・平滑化を設定から変更できる | ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い | ACがゼロを上抜けた足の終値 | | 売り | ゼロを下抜けた足の終値 | | 決済 | 反対側へのゼロ抜けのみ。損切り・利確・時間決済は使わない | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した組み合わせ | 短期3種（3・5・8）× 長期3種（21・34・55）× 平滑化3種（3・5・8）＝ 27通り | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | 同じ方法で調べた[OsMA](/ja/blog/osma-settings)・[QQE](/ja/blog/qqe-settings)・[アルーン](/ja/blog/aroon-settings)・[RVI](/ja/blog/rvi-settings)・[スーパートレンド](/ja/blog/supertrend-settings)・[ウィリアムズ%R](/ja/blog/williams-r-settings)・[MFI](/ja/blog/mfi-settings)の記事もあります。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです - 試した組み合わせは27通りで、他の指標の検証（54〜240通り）より粗い格子です - ACオシレーターの計算はAOの実装（中値の取り方、移動平均の種類）を引き継ぎます。この数値はFormiqの実装によるものです - エントリーも決済も足の終値で行っています - スプレッドは全期間0.3pips固定です。15分足の結果はこの前提に強く依存します - 4時間足は年間61〜161回です。この件数の勝率やPFは幅の広い数字です --- ### アルーンの期間と時間足：84通りの設定を検証 URL: https://formiq.jp/ja/blog/aroon-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: アルーン, Aroon, インジケーター, パラメーター, ドル円 Also available in: en — https://formiq.jp/blog/aroon-settings アルーン（Aroon）は、直近の高値と安値が「何本前についたか」だけを見る指標です。値動きの大きさも、勢いも見ません。時間しか見ない。 その素直さが効くのか、それとも足りないのか。ドル円の2025年（1月1日〜12月31日）で、期間・判定方法・水準を組み合わせた84通りをひとつずつバックテストしました。時間足は15分足・1時間足・4時間足の3つ、比較のために2024年でも同じ84通りを回しています。 結果を先に書きます。**アルーン単体は、この2年のドル円ではかなり弱い指標でした。** 1時間足・2025年で黒字になったのは84通り中43通り、損益の中央値は+6pipsです。ほぼコイン投げでした。 さらに、**4時間足で2024年に+4,091.3pipsだった期間40・水準70は、2025年に−1,664.5pipsでした。** 4時間足全体でも、黒字設定は2024年の61通りから2025年の17通りへ減っています。 ## アルーンが測っているもの アルーンは2本の線でできています。 - **アルーン・アップ**：直近N本の中で、最高値をつけてから何本経ったか。今日が最高値なら100、N本前が最高値なら0 - **アルーン・ダウン**：同じことを最安値で 上げ相場では高値が更新され続けるのでアップが100付近に張り付き、ダウンは下がります。下げ相場では逆です。**どちらも「時間」しか測っていません。** 値幅もボラティリティも入りません。 読み方は2つあります。 | 判定方法 | 買いのタイミング | 特徴 | |---|---|---| | **クロス** | アップがダウンを上抜けた足 | 「その瞬間」に1回だけ発生する出来事 | | **水準** | アップが基準値以上・ダウンが基準値未満 | 条件を満たしている「状態」。満たしている間ずっと成立する | 同じ2本の線を使っても、クロスは線が交差した瞬間だけ売買し、水準判定は条件を満たしている間ずっと成立します。そこで、クロスと水準判定を別々の売買ルールとして検証しました。 設定するパラメータは3つです。 | パラメータ | 値を上げると何が起きるか | |---|---| | 期間 | 高値・安値を探す範囲が広がる。線の反転が減り、シグナルが減る | | 判定方法 | クロスか水準か（数値ではなく選択） | | 水準 | 水準判定のときの基準値。上げるほど条件が厳しくなり、シグナルが減る | ## チャートに表示する方法 | 環境 | 手順 | |---|---| | MT4 | 標準では入っていない。配布されている .ex4／.mq4 を `MQL4/Indicators` に置いて再起動する | | MT5 | 標準で入っている。ナビゲーターの「Oscillators」から追加する | | TradingView | インジケーター検索で「Aroon」を選ぶ | | ブラウザ（Formiq） | インジケーター一覧から選ぶ。期間は設定から変更できる | ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い（クロス） | アップがダウンを上抜けた足の終値 | | 買い（水準） | アップが基準値以上かつダウンが基準値未満 | | 売り | 上記の反対 | | 決済 | 反対側の判定成立のみ。損切り・利確・時間決済は使わない。反対側が成立した足で同時にドテンする | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した組み合わせ | 期間12種（7〜75）×〔クロス、または水準6種（50〜100）〕＝ 84通り。時間足ごと・年ごとに実行 | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | ## どの時間足で使えるか | 時間足 | 年 | 黒字設定 | 損益中央値 | 両年とも黒字 | |---|---:|---:|---:|---:| | 15分足 | 2025 | 63 / 84 | +452pips | 48 / 84 | | 15分足 | 2024 | 63 / 84 | +1,955pips | 48 / 84 | | 1時間足 | 2025 | 43 / 84 | +6pips | 33 / 84 | | 1時間足 | 2024 | 63 / 84 | +1,429pips | 33 / 84 | | 4時間足 | 2025 | 17 / 84 | −1,207pips | 3 / 84 | | 4時間足 | 2024 | 61 / 84 | +1,649pips | 3 / 84 | 4時間足の行を2回見てください。2024年は84通り中61通りが黒字で中央値+1,649pips、**2025年は17通りしか黒字にならず中央値−1,207pips**です。1年で1位から最下位へ変わり、**両方の年で黒字だったのは84通り中3通りだけ**でした。 **時間足を下げるほど結果が良くなる**のがアルーンの特徴です。15分足は両方の年で黒字が多く（48/84）、両方の年の中央値もプラスでした。この点は他の多くの指標と逆で、あとのコストの節でその理由の一部が出てきます。 ## 4時間足の好成績は本物か 取引回数が少なく、そのまま受け取れる数字ではありませんでした。2024年の4時間足で最も成績が良かったのは、期間40・水準70の設定でした。 | 年 | トレード回数 | 勝率 | PF | 損益 | |---:|---:|---:|---:|---:| | 2024 | 27回 | 62.96% | 4.335 | **+4,091.3pips** | | 2025 | 40回 | 25.00% | 0.533 | **−1,664.5pips** | 勝率63.0%、PF 4.3だった設定が、翌年は勝率25.0%、PF 0.533、−1,664.5pipsまで悪化しました。 そして、これは1つの設定に起きた事故ではありません。 この記事の冒頭の図は、横軸に2024年、縦軸に2025年の損益を置き、84設定を1点ずつ示しています。4時間足では2024年の黒字設定61通りに対し、2025年の黒字設定は17通り、両年とも黒字は3通りでした。 見落としてはいけない数字がもうひとつあります。**この27回という取引回数です。** 27回なら、1回の勝ち負けで勝率が3.7ポイント動きます。4時間足の84通りの取引回数は年間15〜188回で、多くはこの水準です。勝率63.0%を「そういう性質の設定だ」と読むには足りません。 ## 期間はいくつがいいか 2年とも黒字だった期間を並べます。1時間足のクロス判定は、期間だけが変数になります（水準を使わないため）。12通りの期間を両方の年で並べるとこうです。 | 期間 | 2025年 | 2024年 | |---|---|---| | **7** | **+638pips** | **+1,136pips** | | **9** | **+360pips** | **+1,589pips** | | **12** | **+47pips** | **+1,620pips** | | 14 | −325pips | +2,014pips | | 18 | −1,682pips | +1,781pips | | 20 | −621pips | +2,066pips | | **25** | **+381pips** | **+578pips** | | 30 | +220pips | −944pips | | 40 | +1,042pips | −1,802pips | | **50** | **+37pips** | **+332pips** | | 60 | −922pips | −215pips | | 75 | −1,857pips | +1,696pips | 12通りのうち、年をまたいで符号が変わったのは6通り。両方の年で赤字だったのは1通り（期間60）。そして**両方の年で黒字だったのは期間7・9・12・25・50の5通り**です。 **残った5通りは、期間12以下の短い側と、25・50に分かれます。** その中で期間7は、2025年に+637.8pips、2024年に+1,135.9pipsでした。 よく紹介される期間25も両方の年で黒字です。ただし**5通りの中での位置は年で入れ替わります**。2025年は期間7に次ぐ2番目ですが、2024年は下から2番目でした。 | 年 | トレード回数 | 勝率 | PF | 損益 | 最大ドローダウン | |---:|---:|---:|---:|---:|---:| | 2025 | 228回 | 41.23% | 1.056 | +380.6pips | 1,530.0pips | | 2024 | 237回 | 36.29% | 1.075 | +578.4pips | 1,440.8pips | PFは両方の年とも1.06前後、最大ドローダウンは2025年で利益の4.0倍、2024年で2.5倍あります。**どちらの年も黒字ではありますが、この曲線に資金を預ける根拠としては薄い数字**です。84通りを見渡しても、それ以上のものはほとんど見つかりませんでした。 ## 勝率だけで設定を選べるか 1時間足・2025年で最も勝率が高かったのは期間30・水準90で、勝率49.62%でした。ただし、同じ設定の2024年は勝率36.43%、損益+335.4pipsです。**1年の勝率だけでは、翌年の結果は判断できません。** 勝率の幅は32.50%〜49.62%でした。 ## フィルターで改善するか 期間25・クロス判定に足して測りました。 | 条件 | 年 | 取引数 | 勝率 | 損益 | |---|---:|---:|---:|---:| | アルーンのみ | 2025 | 228回 | 41.23% | +380.6pips | | アルーンのみ | 2024 | 237回 | 36.29% | +578.4pips | | ADX 20以上 | 2025 | 108回 | 40.74% | +46.0pips | | ADX 20以上 | 2024 | 123回 | 33.33% | +38.8pips | | ADX 25以上 | 2025 | 57回 | 36.84% | −698.7pips | | ADX 25以上 | 2024 | 66回 | 37.88% | −465.7pips | | ADX 30以上 | 2025 | 30回 | 40.00% | −246.3pips | | ADX 30以上 | 2024 | 33回 | 39.39% | +59.5pips | | 東京時間のみ（UTC 0〜8） | 2025 | 71回 | 47.89% | +696.3pips | | 東京時間のみ（UTC 0〜8） | 2024 | 75回 | 37.33% | −80.1pips | | ロンドン・NY時間のみ（UTC 7〜21） | 2025 | 149回 | 36.91% | −1,076.4pips | | ロンドン・NY時間のみ（UTC 7〜21） | 2024 | 162回 | 37.04% | +849.7pips | **ADXのフィルターは、どの水準でも両方の年で損益を減らしました。** ADX 25以上は両方の年とも赤字に落ちています。トレンドの強さで絞るという発想そのものが、この組み合わせでは機能しませんでした。 時間帯フィルターは前回と同じ挙動です。東京時間は2025年に効き、ロンドン・NY時間は2024年に効きました。符号ごと入れ替わります。 ## 損切りと利確で改善するか | 決済ルール | 年 | 取引数 | 勝率 | 損益 | |---|---:|---:|---:|---:| | 反対の判定のみ | 2025 | 228回 | 41.23% | +380.6pips | | 反対の判定のみ | 2024 | 237回 | 36.29% | +578.4pips | | 損切り30・利確60 | 2025 | 212回 | 35.38% | +217.9pips | | 損切り30・利確60 | 2024 | 228回 | 39.47% | +1,084.2pips | | 損切り50・利確100 | 2025 | 213回 | 40.85% | +1,703.7pips | | 損切り50・利確100 | 2024 | 223回 | 37.22% | +1,058.6pips | | 損切り100・利確200 | 2025 | 213回 | 41.78% | +1,716.4pips | | 損切り100・利確200 | 2024 | 223回 | 38.12% | +1,888.2pips | | 損切り50・利確50 | 2025 | 212回 | 50.94% | +845.4pips | | 損切り50・利確50 | 2024 | 230回 | 48.26% | +795.2pips | | 24本で時間決済 | 2025 | 220回 | 48.18% | +2,100.2pips | | 24本で時間決済 | 2024 | 240回 | 42.50% | +1,324.0pips | ここはこの記事で唯一、条件を足して素直に良くなった場所です。**5つのうち4つが両方の年で「反対の判定のみ」を上回りました。** 特に損切り100／利確200は+1,716.4／+1,888.2pips、24本の時間決済は+2,100.2／+1,324.0pipsです。 裏を返すと、**アルーンの「反対の判定を待つ」という決済がそれだけ弱い**ということでもあります。反対側の判定が出るまで持ち続けると、伸びたぶんを返してしまう。損切り50／利確50は勝率を両方の年で50%前後に乗せています（50.94%／48.26%）。 ## スプレッドはどれだけ効くか 回数の多い設定ほどスプレッドで決まる、という前回の結論はここでも成り立ちました。 | 時間足 | 期間 | 判定 | 取引回数 | スプレッド | 年間損益 | |---|---:|---|---:|---:|---:| | 1時間足 | 30 | 水準100 | 121回 | 0.0pips | +2,380.0pips | | 1時間足 | 30 | 水準100 | 121回 | 0.3pips | +2,343.7pips | | 1時間足 | 30 | 水準100 | 121回 | 0.6pips | +2,307.4pips | | 1時間足 | 30 | 水準100 | 121回 | 1.0pips | +2,259.0pips | | 1時間足 | 30 | 水準100 | 121回 | 1.5pips | +2,198.5pips | | 1時間足 | 30 | 水準100 | 121回 | 2.0pips | +2,138.0pips | | 15分足 | 25 | クロス | 1,031回 | 0.0pips | +1,101.1pips | | 15分足 | 25 | クロス | 1,031回 | 0.3pips | +791.9pips | | 15分足 | 25 | クロス | 1,031回 | 0.6pips | +482.6pips | | 15分足 | 25 | クロス | 1,031回 | 1.0pips | +70.2pips | | 15分足 | 25 | クロス | 1,031回 | 1.5pips | −445.3pips | | 15分足 | 25 | クロス | 1,031回 | 2.0pips | −960.8pips | 1時間足の121回の設定は0から2.0pipsまで動かしても242pipsしか減りません。15分足の1,031回の設定は2,062pips減り、**1.5pipsで赤字に転落**します。 15分足がアルーンで唯一「両方の年で黒字が多かった」時間足だったことを思い出してください。その優位は取引回数の多さの上に乗っていて、スプレッドの前提が変わると消えます。同じ関係は他の指標でも成り立ちます（[RVI](/ja/blog/rvi-settings)・[スーパートレンド](/ja/blog/supertrend-settings)）。 アルーンを使うなら、単体で判断させるより、他の条件と組み合わせたときの1要素として置くほうが結果に合っています。決済を「反対の判定待ち」から損切り・利確に替えるだけで両方の年とも改善した点も、そのことを示しています。[プログラミングなしで条件を組む](/ja/blog/backtest-without-coding)なら、アルーンに何を足すと変わるのかを自分の通貨ペアで確かめられます。同じ方法で[QQEを200通り検証した記事](/ja/blog/qqe-settings)もあります。指標が違うと結論がどう変わるかの比較になります。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです - 4時間足は取引回数が年間15〜188回しかありません。この件数の勝率やPFは、そもそも幅の広い数字です - アルーンの計算は実装による差が小さい指標ですが、判定ロジック（クロスの定義、水準の等号の扱い）はツールによって違います。この数値はFormiqの実装によるものです - エントリーも決済も足の終値で行っています - スプレッドは全期間0.3pips固定です。実際には時間帯と指標発表で変動します - [時間帯によって値動きの大きさは変わります](/ja/data)。時間フィルターの結果はその影響を受けています --- ### MFIの逆張りはFXで何回売買できる？240通り検証 URL: https://formiq.jp/ja/blog/mfi-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: MFI, マネーフローインデックス, インジケーター, 出来高, ドル円 Also available in: en — https://formiq.jp/blog/mfi-settings MFI（マネーフローインデックス）は「出来高で重み付けしたRSI」と説明されます。20を下回れば売られすぎ、80を上回れば買われすぎ。RSIと同じ読み方に、出来高という裏付けが加わる。そういう位置づけです。 ドル円の2025年（1月1日〜12月31日）で、期間6通り × エントリー水準5通り × 決済水準4通り × 買い/売りの2方向、合わせて240通りを検証しました。時間足は15分足・1時間足・4時間足、比較のために2024年でも同じ240通りを回しています。 **最初に分かったのは、成績の良し悪し以前の問題でした。この指標は、FXの上位足ではほとんど売買しません。** 4時間足では、240通りの設定のうち**48通りが年間1度も売買せず**、137通りが年10回未満です。1時間足でも20通りがゼロ、81通りが10回未満。「MFIの最適な期間は」という議論をする前に、**そもそも判断材料になる回数が出ていない**設定がこれだけあります。 ## MFIはFXで使えるのか MFIが20を割るには、直近N本のあいだ、下落した足の出来高付き合計が上昇した足のそれを大きく上回る必要があります。RSIより条件が重なるぶん、極端な水準に届きにくくなります。 そのうえ時間足を上げると足の本数が減るので、条件が揃う機会はさらに減ります。4時間足の1年は約1,610本しかありません。ここで期間28・水準10のような設定を選べば、年に数回も条件が揃わない、ということが普通に起きます。 | 時間足 | 年間の取引回数（240通りの中央値） | 0回だった設定 | 10回未満だった設定 | |---|---|---|---| | 15分足 | 80回 | 8 / 240 | 54 / 240 | | 1時間足 | 23回 | 20 / 240 | 81 / 240 | | 4時間足 | 5回 | 48 / 240 | 137 / 240 | （2025年、買い側と売り側の合計） ## 出来高の前提はFXで成り立つか もうひとつ、この指標を読むうえで無視できない点があります。 MFIの計算には出来高が入ります。株式なら取引所に集約された出来高があり、「いくら分の資金が動いたか」に近い意味を持ちます。**FXにはその集中した出来高が存在しません。** チャートに出るのは、その配信元が観測した範囲の数字です。 だからMFIをFXで使うと、「出来高で重み付けしたRSI」という説明の重み付けの部分が、株式ほど強い根拠を持ちません。この検証の数字も、その前提の上に乗っています。 ## MFIの表示と設定 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準で入っている。ナビゲーターの「Oscillators」から Money Flow Index を選ぶ | | TradingView | インジケーター検索で「Money Flow Index」を選ぶ | | ブラウザ（Formiq） | インジケーター一覧から選ぶ。期間は設定から変更できる | ## 検証の条件 この指標のルールは、**買いと売りのどちらか一方しか作りません**。「20を下回る」は買いの条件にしかならず、「80を上回る」は売りの条件にしかなりません。買い逆張りと売り逆張りを別々の売買ルールとして検証しています。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い側 | MFIがエントリー水準（10〜30）を下回ったら買い、決済水準（50〜90）を上回ったら決済 | | 売り側 | 上記を反転させたもの（70〜90で売り、10〜50で決済） | | 損切り・利確 | 使わない | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した組み合わせ | 期間6種（7〜40）× エントリー5種 × 決済4種 × 2方向 ＝ 240通り。時間足ごと・年ごとに実行 | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | ## 定番の14・20/80はどうだったか 期間14、20で買って80で決済。最もよく紹介される形です。 | 時間足 | 売買方向 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---|---:|---:|---:|---:| | 15分足 | 買い | 2025年 | 145回 | 61.38% | +1,379.1pips | | 15分足 | 買い | 2024年 | 113回 | 62.83% | −654.2pips | | 15分足 | 売り | 2025年 | 146回 | 60.27% | +1,426.3pips | | 15分足 | 売り | 2024年 | 113回 | 50.44% | −2,236.8pips | | 1時間足 | 買い | 2025年 | 37回 | 56.76% | −108.0pips | | 1時間足 | 買い | 2024年 | 27回 | 77.78% | +245.4pips | | 4時間足 | 買い | 2025年 | 9回 | 77.78% | +570.4pips | | 4時間足 | 買い | 2024年 | 6回 | 50.00% | −277.0pips | 15分足の買い側は、勝率が2025年61.38%、2024年62.83%と近い一方、年間損益は+1,379.1pipsから−654.2pipsへ反転しました。 点ひとつが120通りの設定それぞれです。横軸が2024年の損益、縦軸が2025年の損益。赤い点が定番の設定で、左上にあります。2024年に大きく負けて、2025年に大きく勝った位置です。 散布図にも、2024年の損益が大きいほど2025年も大きくなる傾向は見えません。定番設定のように、同じ条件で符号が変わる例もあります。 ## 買い側と売り側で違うのか 結果は分かれました。 | 時間足 | 売買方向 | 年 | 黒字の設定数 | 年間損益の中央値 | 両方の年で黒字 | |---|---|---:|---:|---:|---:| | 15分足 | 買い | 2025年 | 74 / 120 | +35pips | 50 / 120 | | 15分足 | 買い | 2024年 | 77 / 120 | +142pips | 50 / 120 | | 15分足 | 売り | 2025年 | 80 / 120 | +97pips | 7 / 120 | | 15分足 | 売り | 2024年 | 11 / 120 | −1,195pips | 7 / 120 | | 1時間足 | 買い | 2025年 | 46 / 120 | −31pips | 31 / 120 | | 1時間足 | 買い | 2024年 | 56 / 120 | 0pips | 31 / 120 | | 1時間足 | 売り | 2025年 | 37 / 120 | −186pips | 3 / 120 | | 1時間足 | 売り | 2024年 | 5 / 120 | −1,214pips | 3 / 120 | | 4時間足 | 買い | 2025年 | 77 / 120 | +156pips | 53 / 120 | | 4時間足 | 買い | 2024年 | 65 / 120 | +84pips | 53 / 120 | | 4時間足 | 売り | 2025年 | 64 / 120 | +33pips | 7 / 120 | | 4時間足 | 売り | 2024年 | 20 / 120 | −603pips | 7 / 120 | **売り側は3つの時間足すべてで2024年に崩れます。** 2024年のドル円は年間1,632pips上昇しました。買われすぎを売り続ける形は、上昇相場では成立しません。[ウィリアムズ%Rでも同じことが起きています](/ja/blog/williams-r-settings)。 買い側は中央値がゼロ付近をうろうろしています。15分足で+35／+142pips、1時間足で−31／0pips。**大きく勝ちも負けもしない**、というのがこの指標の素の姿でした。 ## 回数が足りている設定はどうか 年20回以上売買した設定に絞ると、数字はこうなります（1時間足・買い側・2025年、65通り）。 - 勝率の範囲: 47.5%〜72.73% - 損益の中央値: **−304pips** - 黒字だったもの: 65通り中13通り 勝率は高く出ます。損益は伴いません。ウィリアムズ%Rの記事で見た構造と同じで、**逆張りの決済ルールが利益に上限をつける**からです。 なお、回数の少ない設定を除かずに集計すると、勝率が0%や100%の行が混ざります。年に1〜2回しか売買していない設定の勝率は、勝率という言葉が意味を持つ数字ではありません。**この記事で勝率を語るときは、必ず取引回数と一緒に見てください。** MFIをFXで単体の売買根拠にするのは、この数字では支持できません。使うなら、回数が確保できる時間足で、方向の判断を別に持ったうえでの補助にとどまるはずです。 同じ方法で調べた[ウィリアムズ%R](/ja/blog/williams-r-settings)・[QQE](/ja/blog/qqe-settings)・[アルーン](/ja/blog/aroon-settings)・[RVI](/ja/blog/rvi-settings)・[スーパートレンド](/ja/blog/supertrend-settings)の記事もあります。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです - **出来高はFXでは市場全体の数字ではありません。** MFIの計算はこの前提の上に成り立っています - エントリーも決済も足の終値で行っています - 取引回数が極端に少ない設定を集計に含めています（除外した場合の数字は本文に別途記載しました） --- ### OsMAの時間足と設定：54通りの損益を比較 URL: https://formiq.jp/ja/blog/osma-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: OsMA, MACD, インジケーター, パラメーター, ドル円 Also available in: en — https://formiq.jp/blog/osma-settings OsMA（Moving Average of Oscillator）は、MACD線とシグナル線の差を棒グラフにしたものです。MACDのヒストグラムと同じ計算で、ゼロより上ならMACD線がシグナル線の上にいる、という意味になります。 ドル円の2025年（1月1日〜12月31日）で、短期・長期・シグナルの3つのパラメータを54通り組み合わせて検証しました。時間足は15分足・1時間足・4時間足、比較のために2024年でも同じ54通りを回しています。 **1時間足では、54通り全部が2025年と2024年の両方で黒字になりました。** この検証シリーズで6つの指標を同じ方法で調べていますが、片方の年ですら全通り黒字になった例は他にありません。ただしこれは「OsMAが強い」という話では終わりません。**同じ54通りを15分足でやると、両方の年で黒字だったのは13通りだけ**です。 ## OsMAはどの時間足がいいか 機能したのは1時間足だけでした。 | 時間足 | 年 | 黒字の設定数 | 年間損益の中央値 | 年間の取引回数（中央値） | |---|---:|---:|---:|---:| | 15分足 | 2025年 | 39 / 54 | +433pips | 1,943回 | | 15分足 | 2024年 | 13 / 54 | −968pips | 対象外 | | 1時間足 | 2025年 | 54 / 54 | +1,874pips | 460回 | | 1時間足 | 2024年 | 54 / 54 | +2,135pips | 対象外 | | 4時間足 | 2025年 | 43 / 54 | +590pips | 109回 | | 4時間足 | 2024年 | 29 / 54 | +97pips | 対象外 | 15分足の2024年は54通り中13通りしか黒字になっていません。年間1,940回の売買にスプレッドがかかると、それだけで大きな負担になります。[この関係はどの指標でも同じでした](/ja/blog/rvi-settings)。 4時間足は年110回まで減りますが、こちらは逆に標本が薄くなります。両方の年で黒字だったのは22/54で、1時間足より落ちます。 ## 一番良い設定はどれか 1時間足では、どの設定を選んでも黒字でした。 | 設定 | 年 | 年間損益 | |---|---:|---:| | 初期設定 12/26/9 | 2025年 | +2,223.6pips | | 初期設定 12/26/9 | 2024年 | +3,152.1pips | | 2025年に最大だった6/18/9 | 2025年 | +3,133.1pips | | 2025年に最大だった6/18/9 | 2024年 | +2,453.6pips | | 2025年に最小だった24/26/15 | 2025年 | +575.2pips | | 2025年に最小だった24/26/15 | 2024年 | +2,912.7pips | 注目すべきは3行目です。**1時間足で最も成績が悪かった設定でも、2025年に+575.2pips、2024年に+2,912.7pipsの黒字**でした。パラメータの選び方で結果が黒字か赤字かに分かれる、という状況がそもそも起きていません。 設定ごとの大小は年によって入れ替わりましたが、54設定すべてが両年で黒字だったため、黒字・赤字の判断は変わりませんでした。 1時間足の取引回数は260回から831回の範囲でした。パラメータを変えると頻度は動きますが、どの頻度でも黒字が残っています。 ## 勝率36.9%でも勝てるのはなぜか 1時間足・2025年の54通りで、勝率は34.02%〜39.89%、中央値36.9%でした。勝率の幅は5.87ポイントですが、年間損益は+575.2〜+3,133.1pipsまで開いています。 ゼロラインを抜けた方向に持ち続ける形なので、負けは小さく何度も出て、勝ちが大きく伸びます。[ウィリアムズ%Rの記事で見た形](/ja/blog/williams-r-settings)のちょうど裏返しです。**勝率だけでは年間損益を比較できません。** ## チャートに表示する方法 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準で入っている。ナビゲーターの「Oscillators」から Moving Average of Oscillator を選ぶ | | TradingView | MACDを追加してヒストグラムを見る（同じ計算） | | ブラウザ（Formiq） | インジケーター一覧から選ぶ。短期・長期・シグナルを設定から変更できる | ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い | OsMAがゼロを上抜けた足の終値 | | 売り | ゼロを下抜けた足の終値 | | 決済 | 反対側へのゼロ抜けのみ。損切り・利確・時間決済は使わない。反対側へ抜けた足で同時にドテンする | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した組み合わせ | 短期5種（6〜24）× 長期4種（18〜60）× シグナル3種（5〜15）のうち短期<長期の54通り | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | ## 4時間足でも設定を選べるか 年ごとに最大の設定が入れ替わります。1時間足の安定ぶりと対照的に、4時間足は他の指標と同じ挙動を見せました。 | 設定 | 年 | 年間損益 | |---|---:|---:| | 2025年に最大だった9/18/9 | 2025年 | +2,184.6pips | | 2025年に最大だった9/18/9 | 2024年 | −1,052.9pips | | 2024年に最大だった24/60/15 | 2025年 | −1,525.0pips | | 2024年に最大だった24/60/15 | 2024年 | +1,871.7pips | 2024年に+1,871.7pipsだった24/60/15は、2025年に−1,525.0pipsへ反転しました。2025年に+2,184.6pipsだった9/18/9も、2024年は−1,052.9pipsです。取引回数の中央値は年109回で、1年の結果だけでは次の年の設定を選べませんでした。 同じ指標でも、時間足を変えると「設定を選ぶ意味がある/ない」の性質まで変わる、ということです。 同じ方法で調べた[QQE](/ja/blog/qqe-settings)・[アルーン](/ja/blog/aroon-settings)・[RVI](/ja/blog/rvi-settings)・[スーパートレンド](/ja/blog/supertrend-settings)・[ウィリアムズ%R](/ja/blog/williams-r-settings)・[MFI](/ja/blog/mfi-settings)の記事もあります。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです - OsMAの計算はMACDの実装差（EMAの初期化方法など）をそのまま引き継ぎます。この数値はFormiqの実装によるものです - エントリーも決済も足の終値で行っています - スプレッドは全期間0.3pips固定です。15分足の結果はこの前提に強く依存します - 4時間足は年間69〜220回です。この件数の勝率やPFは幅の広い数字です --- ### QQEの設定を比較：200通りの損益と売買回数 URL: https://formiq.jp/ja/blog/qqe-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: QQE, インジケーター, パラメーター, バックテスト, ドル円 Also available in: en — https://formiq.jp/blog/qqe-settings QQEの設定を検索すると、たいてい「RSI期間14、平滑化5、factor 4.236」という3つの数字が出てきます。多くのツールの初期設定もこれです。 では、その3つは本当に一番いい数字なのか。ドル円の2025年（1月1日〜12月31日）で、200通りのパラメータをひとつずつバックテストして確かめました。時間足も15分足・1時間足・4時間足の3つで同じことをやり、比較のために2024年でも同じ200通りを回しています。 **2025年のドル円で年間損益が最も大きかったのはRSI期間17／平滑化3／factor 4.236で、465回の取引で+3,341.0pips**でした。一方、2024年に+4,027.1pipsだったRSI期間10／平滑化8／factor 5.236は、2025年には+1,186.1pipsとなり、同年の中央値+1,822.3pipsを下回りました。1年の年間損益だけで設定を決めると、翌年には中央値を下回ることがあります。 ## QQEは何を測っているのか QQE（Quantitative Qualitative Estimation）は、RSIを直接読む代わりに、RSIを平滑化した線と、その線自身の変動幅から作ったバンドを比べる指標です。 1. RSIを計算する（**RSI期間**） 2. RSIをEMAで平滑化する（**平滑化＝SF**） 3. 平滑化したRSIの1本ごとの変化量を取り、それも平滑化する 4. その値に係数を掛けてバンドの幅にする（**factor**、初期設定は4.236） 5. 平滑化RSIがバンドを抜けた側でトレンドの向きを判定する 設定するパラメータはRSI期間、平滑化、factorの3つです。 | パラメータ | 値を上げると何が起きるか | |---|---| | RSI期間 | 元のRSIが鈍くなる。だましが減り、反応が遅れる | | 平滑化（SF） | 線が滑らかになる。シグナルの回数が減る | | factor | バンドが広がる。シグナルの回数が減り、1回が長くなる | 実際の数字で確かめると、1時間足・2025年で平均取引回数は平滑化2のとき689回、平滑化12のとき369回でした。factorを2から5.236に上げると656回から391回に減ります。RSI期間、平滑化、factorを大きくすると、いずれも取引回数が減り、その回数に比例して支払うスプレッドの合計も減ります。 ## チャートに表示する方法 QQEはどのプラットフォームでも標準搭載というわけではありません。使っている環境ごとに手順が違います。 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準では入っていない。配布されている .ex4／.mq4 を入手し、データフォルダの `MQL4/Indicators`（MT5は `MQL5/Indicators`）に置いて再起動する | | TradingView | インジケーター検索で「QQE」と入力し、コミュニティスクリプトから追加する | | ブラウザ（Formiq） | 最初から入っている。インジケーター一覧から選ぶ | ひとつ注意点があります。**QQEは実装によって値が違います。** 平滑化にEMAを使うかWilderの平滑化を使うか、バンドの追従ルールをどう書くかで線がずれます。この記事の数値はFormiqの実装によるもので、別のスクリプトで同じパラメータを入れても同じ売買にはならない場合があります。 ## 検証の条件 数字を見る前に、何をどう測ったかを書いておきます。この表と同じ設定を入れれば、同じ結果が出ます。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い | QQEラインがシグナル線を上抜けた足の終値 | | 売り | QQEラインがシグナル線を下抜けた足の終値 | | 決済 | 反対側のクロスのみ。損切り・利確・時間決済は使わない。反対クロスが出た足で同時にドテンする | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した組み合わせ | RSI期間8種（6〜25）× 平滑化5種（2〜12）× factor 5種（2〜5.236）＝ 200通り。時間足ごと・年ごとに実行 | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | 損切りを置かず反対シグナルまで持つのは、QQE以外の要素を混ぜないためです。損切りを付けた場合は後の「損切りと利確を付けるとどうなるか」で別に測っています。 ## 初期設定のままでも大丈夫か 初期設定の14／5／4.236は、両方の年で200設定の中央値を上回りました。1時間足の200通りで年間損益が最大だったRSI期間17／平滑化3／factor 4.236は、2025年に465回の取引で+3,341.0pips、勝率42.37%、PF 1.348でした。2024年の年間損益は+2,553.6pipsです。 初期設定の14／5／4.236は、2025年に+2,182.3pips、2024年に+2,309.7pipsでした。同じ200設定の中央値はそれぞれ+1,822.3pipsと+2,088pipsです。最大値だけでなく、初期設定と中央値を並べると、設定差の大きさを同じ基準で読めます。 負けた設定も見ておきます。2025年に最も悪かったのはRSI期間6／平滑化2／factor 4.236の−1,516.0pipsで、2024年も−580.9pipsでした。**1時間足の200通りのうち、両方の年で赤字だったのはこの1通りだけ**です。RSI期間を最短にして平滑化も最小にする、という「いちばん敏感な設定」がそれに当たります。 ## 一番良い設定はどれか 前年にいちばん稼いだ設定は、翌年に中央値を636.2pips下回りました。 **2024年に年間損益+4,027.1pipsだったRSI期間10／平滑化8／factor 5.236を2025年に使うと、+1,186.1pipsでした。** 2025年の200設定の中央値は+1,822.3pipsです。前年の年間損益だけで選んだ設定は、翌年に中央値を636.2pips下回りました。 冒頭の図は200設定について、横軸に2024年の年間損益、縦軸に2025年の年間損益を置いています。前年の年間損益が翌年にもそのまま残るなら点は右上がりにまとまりますが、実際には広く散らばっています。 念のため書いておくと、これは「QQEが機能しなかった」という話ではありません。1時間足では200通り中194通りが2025年に黒字、188通りが2024年に黒字で、**両方の年で黒字だったものが183通り**あります。指標そのものはこの2年間のドル円で機能しています。機能していないのは「一番良い数値を探す」という作業のほうです。 ## どの時間足が向いているか 両年黒字は1時間足で200通り中183通り、15分足では3通りでした。同じ200通りを、時間足だけ変えて回します。 | 時間足 | 年 | 黒字設定 | 年間損益の中央値 | 両年黒字 | |---|---:|---:|---:|---:| | 15分足 | 2025年 | 78 / 200 | −305pips | 3 / 200 | | 15分足 | 2024年 | 4 / 200 | −2,031pips | 3 / 200 | | 1時間足 | 2025年 | 194 / 200 | +1,822pips | 183 / 200 | | 1時間足 | 2024年 | 188 / 200 | +2,088pips | 183 / 200 | | 4時間足 | 2025年 | 197 / 200 | +1,385pips | 136 / 200 | | 4時間足 | 2024年 | 138 / 200 | +642pips | 136 / 200 | 15分足は2024年に**200通りのうち4通りしか黒字になりませんでした**。その4通りはすべてRSI期間20以上・平滑化8以上、つまりこの表の中で最も鈍い側に固まっています。うち3通りは2025年も黒字でした。 4時間足は2025年だけ見れば1時間足より黒字の数が多いのですが、2024年には138通りまで落ちます。両方の年で黒字だった数では1時間足（183通り）が最も多くなりました。 同じ設定を時間足だけ変えて並べるとこうです。初期設定の14／5／4.236の場合： | 時間足 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | 15分足 | 2025年 | 1,710回 | 37.37% | −460.6pips | | 15分足 | 2024年 | 1,739回 | 38.59% | −1,008.1pips | | 1時間足 | 2025年 | 415回 | 42.89% | +2,182.3pips | | 1時間足 | 2024年 | 407回 | 43.73% | +2,309.7pips | | 4時間足 | 2025年 | 105回 | 49.52% | +1,855.6pips | | 4時間足 | 2024年 | 107回 | 32.71% | +130.8pips | パラメータは一切変えていません。動かしたのは時間足だけです。 ## 15分足はコストのせいで負けたのか 半分はそうで、半分は違いました。ここは分けて見る必要があります。 スプレッドを0にした場合の損益（＝グロス）と、0.3pipsを払った場合を並べます。中央値は200通りぶんの中央値です。 | 時間足 | 年 | グロス | ネット | |---|---:|---:|---:| | 15分足 | 2025年 | +181pips | −305pips | | 15分足 | 2024年 | −1,251pips | −2,031pips | | 1時間足 | 2025年 | +1,976pips | +1,822pips | | 1時間足 | 2024年 | +2,257pips | +2,088pips | | 4時間足 | 2025年 | +1,428pips | +1,385pips | | 4時間足 | 2024年 | +681pips | +642pips | **2025年の15分足は、コストを引く前ならプラスでした。** グロスの中央値は+181pips、黒字だった設定は200通り中116通り。そこに0.3pipsを掛けると中央値は−305pipsに落ち、黒字は78通りに減ります。**中央値がゼロを割るのはスプレッド0.12pips付近**です。実在するどのスプレッドよりも手前で沈みます。 いっぽう**2024年の15分足はグロスの時点で−1,251pips**でした。この年はコストを取り除いても負けています。「15分足が負けるのはコストのせい」という説明は、2025年には当てはまり、2024年には当てはまりません。 コストそのものの効き方は素直です。スプレッドを0から2.0pipsまで動かしたときの中央値は、15分足が+181→−3,381pips、1時間足が+1,976→+969pips、4時間足が+1,428→+1,140pips。**1.5pipsの時点で15分足は200通り全部が赤字になり、4時間足は196通りが黒字のまま**でした。失う額は取引回数にそのまま比例します。 これはQQEに限った話ではなく、シグナルの回数が多い設定すべてに効きます。同じ条件で年1,094回売買する[RVI](/ja/blog/rvi-settings)は0→2.0pipsで2,188pips、年152回の[スーパートレンド](/ja/blog/supertrend-settings)は304pipsを失いました。どちらも回数×スプレッドと完全に一致します。自分の口座のスプレッドを入れて確かめる価値があるのは、パラメータよりむしろこちらです。 ## 2年とも同じ向きだったのはどれか 平滑化とRSI期間の2つだけでした。RSI期間、平滑化、factorを1項目ずつ変え、残りの項目の全設定で平均を取ります。2024年と2025年の両方で年間損益の大小関係が一致したのは、平滑化とRSI期間でした。 | 平滑化 | 2025年の平均損益 | 2024年の平均損益 | |---|---|---| | 平滑化 2 | +923pips | +563pips | | 平滑化 3 | +1,807pips | +1,428pips | | 平滑化 5 | +2,167pips | +2,024pips | | 平滑化 8 | +1,895pips | +2,691pips | | 平滑化 12 | +1,431pips | +2,793pips | **平滑化2は両方の年で最下位**でした。他の値の半分以下です。 | RSI期間 | 2025年の平均損益 | 2024年の平均損益 | |---|---|---| | 6 | +1,173pips | +1,266pips | | 10 | +1,645pips | +1,876pips | | 14 | +1,736pips | +1,937pips | | 20 | +1,790pips | +2,177pips | | 25 | +1,807pips | +2,206pips | **RSI期間も、6のような短さが両方の年で最下位**でした。2024年は25まで単調に良くなり、2025年は17でほぼ頭打ちになります。どちらも「短くしすぎない」という同じ話です。 一方、factorは年によって平均損益が最大となる値が異なりました。2025年は2.618（+1,840pips）、2024年は4.236（+2,255pips）です。factorを2から5.236へ上げると、勝率は2025年で39.8%から43.5%、2024年で38.8%から43.2%へ上がり、取引回数は656回から391回へ減りました。factorを大きくすると勝率は上がりましたが、年間損益が最大になる値は両年で一致しませんでした。 ## 勝率を上げると利益は増えるのか 増えませんでした。 1時間足・2025年の200設定で、勝率は37.09%〜47.32%でした。次のフィルター比較では、勝率を上げても年間損益が増えない例を取引回数と一緒に確認します。 では、よく紹介されるフィルターを足すとどうなるか。初期設定の14／5／4.236に足して測りました。 | 条件 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | QQEのみ | 2025年 | 415回 | 42.89% | +2,182.3pips | | QQEのみ | 2024年 | 407回 | 43.73% | +2,309.7pips | | ADX 20以上 | 2025年 | 261回 | 41.76% | +830.9pips | | ADX 20以上 | 2024年 | 248回 | 44.76% | +1,652.9pips | | ADX 25以上 | 2025年 | 198回 | 43.43% | +774.9pips | | ADX 25以上 | 2024年 | 175回 | 45.14% | −257.4pips | | ADX 30以上 | 2025年 | 128回 | 42.19% | +582.1pips | | ADX 30以上 | 2024年 | 122回 | 42.62% | −70.9pips | | ADX 20以上＋DIの向き一致 | 2025年 | 91回 | 35.16% | −408.4pips | | ADX 20以上＋DIの向き一致 | 2024年 | 83回 | 43.37% | +686.5pips | | 東京時間のみ（UTC 0〜8） | 2025年 | 167回 | 47.90% | +1,281.0pips | | 東京時間のみ（UTC 0〜8） | 2024年 | 151回 | 47.68% | +702.6pips | | ロンドン・NY時間のみ（UTC 7〜21） | 2025年 | 223回 | 38.57% | +240.4pips | | ロンドン・NY時間のみ（UTC 7〜21） | 2024年 | 222回 | 44.14% | +1,582.7pips | **この6つのうち、どちらの年でもフィルターなしを上回ったものはひとつもありません。** ADXを厳しくするほど取引が減り、残る額も減ります。ADX 25以上は2024年には−257.4pipsまで落ちて符号が変わりました。 勝率だけを見れば話は違います。東京時間に絞ると勝率は42.89%から47.90%に上がり、ADX 25以上でも43.43%に上がります。**勝率は上げられます。上げたぶん取引が減り、合計は減ります。** 勝率と損益が別のものだという、この記事の前半と同じ話がここでも出ています。 ## 損切りと利確を付けるとどうなるか 同じ現象がここでも起きます。RSI期間17／平滑化3／factor 4.236に損切りと利確を足した結果です（反対クロスでの決済は残したまま）。 | 決済ルール | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | 反対クロスのみ | 2025年 | 465回 | 42.37% | +3,341.0pips | | 反対クロスのみ | 2024年 | 479回 | 40.71% | +2,553.6pips | | 損切り50・利確100 | 2025年 | 420回 | 41.67% | +2,674.1pips | | 損切り50・利確100 | 2024年 | 429回 | 38.93% | +1,487.7pips | | 損切り50・利確50 | 2025年 | 425回 | 48.24% | +2,300.5pips | | 損切り50・利確50 | 2024年 | 432回 | 43.06% | +85.7pips | | 損切り100・利確200 | 2025年 | 459回 | 42.48% | +3,182.5pips | | 損切り100・利確200 | 2024年 | 469回 | 39.66% | +1,108.7pips | | 24本で時間決済 | 2025年 | 457回 | 43.11% | +3,519.5pips | | 24本で時間決済 | 2024年 | 469回 | 41.15% | +2,609.5pips | **損切りと利確を足した3つは、どれも両方の年で反対クロスのみを下回りました。** 損切り50／利確50は2025年の勝率を48.24%まで押し上げますが、合計は+2,300.5pipsに落ち、2024年に至っては+85.7pipsでほぼゼロです。 例外は時間決済（24本で強制手仕舞い）で、**2025年も2024年も反対クロスのみをわずかに上回りました**（+3,341.0→+3,519.5、+2,553.6→+2,609.5）。追加して両方の年とも改善した条件は、時間決済だけです。 QQEの数値をひとつに絞る作業には、この検証の範囲では意味が見つかりませんでした。素のRSIを同じ手順で測ると、[逆張りと順張りで期間の良し悪しが反対を向きました](/ja/blog/rsi-settings)。代わりに効いていたのは、時間足の選択と、極端な値を避けることと、コストです。[プログラミングなしで条件を組んで検証する](/ja/blog/backtest-without-coding)なら、この記事の表と同じことを自分の通貨ペア・自分の期間で回して確かめられます。 ## 補足 - 対象は米ドル／円の2024年と2025年です。他の通貨ペアや期間は含みません - QQEの実装はスクリプトによって差があります。この数値はFormiqの実装によるもので、他のスクリプトの同じパラメータとは一致しないことがあります - エントリーも決済も足の終値で行っています。実際の約定はここからずれます - スプレッドは全期間0.3pips固定として計算しています。実際のスプレッドは時間帯と指標発表で変動します - 4時間足は設定によって年67〜257回です。取引回数が少ない設定の勝率とPFは変動しやすくなります --- ### RVIの期間とスプレッドの影響をドル円2年間で検証 URL: https://formiq.jp/ja/blog/rvi-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: RVI, インジケーター, パラメーター, スプレッド, ドル円 Also available in: en — https://formiq.jp/blog/rvi-settings RVI（Relative Vigor Index）は「終値が始値からどれだけ離れたか」を「その足の高値と安値の幅」で割った指標です。上げ相場では終値が高値寄りで引けるはず、という発想でできています。 期間を変えたらどうなるのか。ドル円の2025年（1月1日〜12月31日）で、期間4から55まで10通りをバックテストしました。時間足は15分足・1時間足・4時間足、比較のために2024年でも同じ10通りを回しています。 結果を先に書きます。**期間を4から55へ約14倍に広げても、取引回数は年1,336回から1,078回へ減っただけでした。** RVI期間を大きく変えても、1時間足の年間取引回数は3割以上変わりませんでした。 そして年間1,000回を超える売買回数は、それ自体が損益を決める要素になります。この記事の後半はその話です。 ## RVIは何を測っているのか 1本の足について、まず2つの量を計算します。 - **分子**：終値 − 始値（直近4本を1:2:2:1で加重平均） - **分母**：高値 − 安値（同じく加重平均） 分子を分母で割ったものがRVIです。値は −1 から +1 のあたりを動きます。終値が高値付近で引ける足が続けばプラス側、安値付近で引ける足が続けばマイナス側に寄ります。 このRVIを**期間N本ぶん合計**したものが表示される線で、**期間はここにしか効きません**。 そしてもう1本、シグナル線があります。こちらはRVI本体を直近4本で1:2:2:1に加重平均したものです。**期間の指定はここには入りません。** 売買はこの2本の交差で判定します。つまり、**片方は期間で滑らかにできますが、もう片方は常に「直近4本」でそれを追いかけてきます。** 2本が離れ続けることがないので、交差の頻度は期間ではなくシグナル線の作り方で決まります。 これは机上の話ではなく、実測でそのとおりになりました。 ## チャートに表示する方法 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準で入っている。ナビゲーターの「Oscillators」から Relative Vigor Index を選ぶ | | TradingView | インジケーター検索で「Relative Vigor Index」を選ぶ | | ブラウザ（Formiq） | インジケーター一覧から選ぶ。期間は設定から変更できる | ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い | RVIがシグナル線を上抜けた足の終値 | | 売り | RVIがシグナル線を下抜けた足の終値 | | 決済 | 反対側のクロスのみ。損切り・利確・時間決済は使わない。反対クロスが出た足で同時にドテンする | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した期間 | 4 / 6 / 8 / 10 / 14 / 18 / 24 / 30 / 40 / 55 の10通り。時間足ごと・年ごとに実行 | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | ## 期間を長くすると売買は減るか 減りません。損益は変わるのに、取引回数はほとんど動きませんでした。1時間足の10通りを並べます。冒頭の図が、この表を描いたものです。 | 期間 | 年 | 取引回数 | 年間損益 | |---:|---:|---:|---:| | 4 | 2025年 | 1,336 | +511.0pips | | 4 | 2024年 | 1,325 | +944.8pips | | 6 | 2025年 | 1,110 | +38.6pips | | 6 | 2024年 | 1,155 | −1,250.0pips | | 8 | 2025年 | 1,048 | +272.4pips | | 8 | 2024年 | 1,153 | −1,545.5pips | | 10 | 2025年 | 1,094 | +2,233.9pips | | 10 | 2024年 | 1,144 | +184.6pips | | 14 | 2025年 | 1,075 | +1,853.1pips | | 14 | 2024年 | 1,126 | +1,910.1pips | | 18 | 2025年 | 1,043 | −183.8pips | | 18 | 2024年 | 1,114 | −831.9pips | | 24 | 2025年 | 1,081 | +2,810.0pips | | 24 | 2024年 | 1,099 | −991.4pips | | 30 | 2025年 | 1,081 | +1,489.3pips | | 30 | 2024年 | 1,086 | −430.2pips | | 40 | 2025年 | 1,094 | +2,764.9pips | | 40 | 2024年 | 1,089 | +289.0pips | | 55 | 2025年 | 1,078 | +3,071.8pips | | 55 | 2024年 | 1,109 | +606.5pips | 回数の列を縦に読んでください。**1,043から1,336の間に全部収まっています。** 期間を4から55へ、14倍近く広げても、年間の売買回数は3割も動きません。 損益の列は逆で、−1,545から+3,072まで4,600pips以上の幅があります。しかも2025年と2024年で符号がばらばらです。期間18は両方の年で赤字、期間8は2025年に+272pipsで2024年に−1,546pips。 つまり、**期間を変えても取引頻度はほぼ同じなのに、年間損益は大きく変わりました。** 過去1年の損益だけで期間を選ぶ根拠としては弱い結果です。 両方の年で黒字だったのは10通り中5通り（期間4・10・14・40・55）でした。期間14は+1,853.1／+1,910.1pips、初期設定の期間10は+2,233.9／+184.6pipsです。 ## RVIは15分足で使えるか 使えません。時間足ごとの結果を並べます。 | 時間足 | 年 | 黒字の期間数 | 年間損益の中央値 | 年間の取引回数（中央値） | |---|---:|---:|---:|---:| | 15分足 | 2025年 | 2 / 10 | −683pips | 約4,400回 | | 15分足 | 2024年 | 0 / 10 | −1,510pips | 対象外 | | 1時間足 | 2025年 | 9 / 10 | +1,671pips | 約1,080回 | | 1時間足 | 2024年 | 5 / 10 | −123pips | 対象外 | | 4時間足 | 2025年 | 5 / 10 | +15pips | 約275回 | | 4時間足 | 2024年 | 10 / 10 | +927pips | 対象外 | 1時間足と4時間足は年をまたぐと入れ替わります。2025年は1時間足が良く、2024年は4時間足が10通り全部黒字です。 **15分足は2024年に10通りすべてが赤字**で、両方の年で黒字だった設定はひとつもありません。年4,400回という回数が理由です。 ## スプレッドで結果は変わるか コストが損益を決めていました。同じ2025年、同じ設定で、スプレッドだけを変えます。 | スプレッド | 時間足 | 期間 | 取引回数 | 年間損益 | |---:|---|---:|---:|---:| | 0.0pips | 1時間足 | 10 | 1,094回 | +2,562.1pips | | 0.3pips | 1時間足 | 10 | 1,094回 | +2,233.9pips | | 0.6pips | 1時間足 | 10 | 1,094回 | +1,905.7pips | | 1.0pips | 1時間足 | 10 | 1,094回 | +1,468.1pips | | 1.5pips | 1時間足 | 10 | 1,094回 | +921.1pips | | 2.0pips | 1時間足 | 10 | 1,094回 | +374.1pips | | 0.0pips | 15分足 | 10 | 4,375回 | +396.5pips | | 0.3pips | 15分足 | 10 | 4,375回 | −916.0pips | | 0.6pips | 15分足 | 10 | 4,375回 | −2,228.5pips | | 1.0pips | 15分足 | 10 | 4,375回 | −3,978.5pips | | 1.5pips | 15分足 | 10 | 4,375回 | −6,166.0pips | | 2.0pips | 15分足 | 10 | 4,375回 | −8,353.5pips | 15分足は**スプレッドが0のときですら+396.5pipsしかありません。** そこから0.3pipsを引くだけで赤字になり、2.0pipsでは−8,353.5pipsです。 1時間足も無傷ではありません。0から2.0pipsで**2,188pipsを失い**、+2,562.1pipsだった利益が+374.1pipsまで削られます。 この数字には規則があります。 同じドル円・同じ2025年・同じ1時間足で、指標だけを変えた4本を並べたものです。 | 指標 | 年間の取引回数 | 0→2.0pipsで失うpips | |---|---|---| | スーパートレンド 10/3 | 152回 | 304pips | | アルーン 25・クロス | 228回 | 456pips | | QQE 17/3/4.236 | 465回 | 930pips | | **RVI 10** | **1,094回** | **2,188pips** | すべて**取引回数 × 2.0pips**にぴったり一致します。当たり前の算数ですが、当たり前だからこそ確実に効きます。**指標を選ぶことは、支払うコストの総額を選ぶことでもあります。** RVIはこの4つの中で最も回数が多く、したがって最もスプレッドに弱い指標でした。 ## 勝率で期間を選べるか 選べません。1時間足・2025年の10通りで、勝率は41.61%〜44.94%でした。一方、年間損益は−183.8〜+3,071.8pipsまで開いています。 フィルターを足すとどうなるか。期間10に対して測りました。 | 条件 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | RVIのみ | 2025年 | 1,094回 | 42.23% | +2,233.9pips | | RVIのみ | 2024年 | 1,144回 | 42.05% | +184.6pips | | ADX 20以上 | 2025年 | 701回 | 43.08% | +1,431.2pips | | ADX 20以上 | 2024年 | 713回 | 43.34% | −385.6pips | | ADX 25以上 | 2025年 | 488回 | 44.06% | +1,691.3pips | | ADX 25以上 | 2024年 | 500回 | 41.80% | −449.7pips | | ADX 30以上 | 2025年 | 313回 | 44.41% | +915.3pips | | ADX 30以上 | 2024年 | 324回 | 41.98% | +125.5pips | | 東京時間のみ | 2025年 | 369回 | 41.46% | +315.0pips | | 東京時間のみ | 2024年 | 389回 | 40.87% | +640.1pips | | ロンドン・NY時間のみ | 2025年 | 636回 | 42.45% | +1,876.6pips | | ロンドン・NY時間のみ | 2024年 | 643回 | 43.86% | +741.4pips | ADX 20以上と25以上は2025年に勝率を1〜2ポイント上げますが、2024年には損益を+184.6pipsから赤字に落とします。**どの条件も2025年の基準（+2,233.9pips）を上回りませんでした。** ただし**符号だけを見ると、両方の年で黒字を保った条件が3つあります**。ADX 30以上（+915.3／+125.5pips）、東京時間（+315.0／+640.1pips）、ロンドン・NY時間（+1,876.6／+741.4pips）です。いずれも回数を4〜7割に減らしています。回数の多さがこの指標の弱点である以上、回数を減らして黒字が残る絞り方には意味があります。 ## 損切りと利確は効くのか | 決済ルール | 年 | 取引回数 | 年間損益 | |---|---:|---:|---:| | 反対クロスのみ | 2025年 | 1,094回 | +2,233.9pips | | 反対クロスのみ | 2024年 | 1,144回 | +184.6pips | | 損切り30・利確60 | 2025年 | 980回 | +1,371.0pips | | 損切り30・利確60 | 2024年 | 1,044回 | −1,188.5pips | | 損切り50・利確100 | 2025年 | 1,031回 | +1,144.0pips | | 損切り50・利確100 | 2024年 | 1,080回 | −972.7pips | | 損切り100・利確200 | 2025年 | 1,082回 | +2,618.3pips | | 損切り100・利確200 | 2024年 | 1,131回 | −332.6pips | | 損切り50・利確50 | 2025年 | 1,027回 | +901.5pips | | 損切り50・利確50 | 2024年 | 1,076回 | −1,658.6pips | | 24本で時間決済 | 2025年 | 1,094回 | +2,222.0pips | | 24本で時間決済 | 2024年 | 1,144回 | +184.6pips | **損切りと利確を付けた4通りすべてが、2024年には赤字**でした。2025年に最も良かった損切り100／利確200は、2024年に−332.6pipsです。 RVIは平均8.0本で決済まで到達する短いトレードなので、そこに固定幅の損切りと利確を重ねると、指標が出そうとしたシグナルより先に価格の幅が決めてしまいます。**24本の時間決済がほぼ何も変えていない**（2024年は1pipも動かず、2025年も−11.9pips）のは同じ理由です。8本で終わるトレードに24本の期限を付けても、ほとんど届きません。反対クロスまで持つ形のほうが、両方の年で安定していました。 同じ方法で調べた他の指標との比較として、[QQEを200通り検証した記事](/ja/blog/qqe-settings)と[アルーンを84通り検証した記事](/ja/blog/aroon-settings)、[スーパートレンドを36通り検証した記事](/ja/blog/supertrend-settings)があります。指標ごとに結論が変わる様子が読み取れます。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです - RVIの実装はツールによる差が小さい指標ですが、シグナル線の作り方には流儀があります。この数値はFormiqの実装によるものです - エントリーも決済も足の終値で行っています - スプレッドは全期間固定として計算しています。実際には時間帯と指標発表で変動し、回数の多い設定ほどその影響を受けます - 4時間足は年間約275回で、10通りの幅も271〜333回と狭く、期間による差はここでも出ませんでした --- ### スクイーズモメンタムの倍率と時間足を45通り検証 URL: https://formiq.jp/ja/blog/squeeze-momentum-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: スクイーズモメンタム, Squeeze, インジケーター, パラメーター, ドル円 Also available in: en — https://formiq.jp/blog/squeeze-momentum-settings スクイーズモメンタムは、ボリンジャーバンドがケルトナーチャネルの内側に収まる「スクイーズ」状態を検出し、それが解けた方向に乗る指標です。値動きが縮んだあとの拡大を狙う、という発想でできています。 ドル円の2025年（1月1日〜12月31日）で、期間・ボリンジャーの倍率・ケルトナーの倍率の3つを45通り組み合わせて検証しました。時間足は15分足・1時間足・4時間足、比較のために2024年でも同じ45通りを回しています。 **4時間足では、黒字設定が2024年の26/45から2025年の5/45へ減りました。** 年間損益の中央値も+251pipsから−461pipsへ変わり、両年とも黒字だったのは1/45です。 ## どの時間足で使えるか | 時間足 | 2025年に黒字 | 2025年の中央値 | 2024年に黒字 | 2024年の中央値 | 両方の年で黒字 | 年間の取引回数 | |---|---|---|---|---|---|---| | **15分足** | 37 / 45 | +273pips | 36 / 45 | +468pips | **29 / 45** | 約366回 | | 1時間足 | 30 / 45 | +168pips | 36 / 45 | +814pips | 24 / 45 | 約81回 | | 4時間足 | **5 / 45** | −461pips | 26 / 45 | +251pips | **1 / 45** | 約21回 | 両年とも黒字だった設定は15分足が29/45で、1時間足の24/45と4時間足の1/45を上回りました。 理由は取引回数です。スクイーズが解ける瞬間だけを狙うので、そもそもシグナルが少ない指標です。1時間足で年81回、4時間足では年21回まで落ちます。**回数が少なすぎて、数字が安定しない側の問題**が出ました。 15分足でも年366回で、他の指標の15分足（1,900〜4,900回）と比べると圧倒的に少なくなります。コストの負担が軽いまま、標本だけが確保できる時間足でした。 ## 去年よかった設定は今年も勝てるか 1時間足では黒字のまま残りました。 | 設定 | 2025年の年間損益 | 2024年の年間損益 | |---|---:|---:| | 初期設定 20/2/1.5 | +398.3pips | +1,530.5pips | | 10/2/1 | +1,234.2pips | +1,536.6pips | | 20/2.5/2 | +624.7pips | +2,732.0pips | | 14/1.5/1.5 | −1,152.8pips | +2,029.9pips | 10/2/1は2025年に225回の取引で+1,234.2pips、2024年も+1,536.6pipsでした。20/2.5/2は2024年の+2,732.0pipsから2025年の+624.7pipsへ減りました。 初期設定は両方の年で黒字ですが、2025年の+398.3pipsは10/2/1の+1,234.2pipsを835.9pips下回りました。 ## 4時間足の初期設定はどうか 23回のうち4回しか勝てませんでした。 | 設定 | 2025年の年間損益 | 2024年の年間損益 | |---|---:|---:| | 初期設定 20/2/1.5 | −967.3pips | +5.6pips | | 20/2.5/1.5 | +295.5pips | −188.9pips | | 10/1.5/1.5 | −1,147.7pips | +1,683.0pips | | 20/1.5/2 | −2,399.9pips | +1,351.4pips | 初期設定は2025年に**23回のトレードで勝率17.39%**、−967.3pipsです。23回のうち4回しか勝てなかった、という意味でしかありません。 2025年に+295.5pipsだった20/2.5/1.5は年11回の取引で、勝率54.55%、PF 2.507でした。11回中6回の勝ちで出た数値であり、2024年の年間損益は−188.9pipsです。 4時間足の取引回数は年0〜80回で、**45通りのうち8通りは1度も売買していません。** 両方の年で黒字だったのは45通り中1通りです。 ## 3つの設定は何を変えるか 期間が値そのものを決め、倍率2つは「いつ入るか」しか決めません。効き方が違います。 - **期間**：モメンタムの計算に使います。値そのものを決めます - **ボリンジャーの倍率・ケルトナーの倍率**：スクイーズ判定にしか使いません。モメンタムの値は変わりません つまり倍率2つは「いつエントリーするか」を決め、期間は「どちらに向かうか」を決めます。倍率を締める（ボリンジャーを広く、ケルトナーを狭く）ほどスクイーズが成立しにくくなり、シグナルが減ります。 1時間足・2025年で、取引回数は0回から319回まで動きました。倍率を締めすぎると、45通りのうち3通りは1年間まったく売買しません。 ## 勝率はどのくらいか 1時間足・2025年の45設定で、勝率の中央値は41.2%でした。4時間足では、年10回以上売買した29設定に限っても勝率が16.67%〜54.55%まで開きます。**平均21回という取引回数では、勝率の高低だけで年間損益を判断できません。** ## スクイーズモメンタムの表示と設定 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準では入っていない。配布されている .ex4／.mq4 を `MQL4/Indicators`（MT5は `MQL5/Indicators`）に置いて再起動する | | TradingView | インジケーター検索で「Squeeze Momentum」を選ぶ | | ブラウザ（Formiq） | インジケーター一覧から選ぶ。期間とバンド幅の倍率2つを設定から変更できる | ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い | スクイーズが解けた足で、モメンタムがプラスなら買い | | 売り | スクイーズが解けた足で、モメンタムがマイナスなら売り | | 決済 | モメンタムがゼロを反対側へ抜けたとき。損切り・利確・時間決済は使わない | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した組み合わせ | 期間5種（10〜40）× ボリンジャー倍率3種（1.5・2・2.5）× ケルトナー倍率3種（1・1.5・2）＝ 45通り | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | 同じ方法で調べた[OsMA](/ja/blog/osma-settings)・[ACオシレーター](/ja/blog/ac-oscillator-settings)・[ストキャスRSI](/ja/blog/stoch-rsi-settings)・[STC](/ja/blog/stc-settings)・[ウェーブトレンド](/ja/blog/wavetrend-settings)・[QQE](/ja/blog/qqe-settings)・[アルーン](/ja/blog/aroon-settings)・[RVI](/ja/blog/rvi-settings)・[スーパートレンド](/ja/blog/supertrend-settings)・[ウィリアムズ%R](/ja/blog/williams-r-settings)・[MFI](/ja/blog/mfi-settings)の記事もあります。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです - スクイーズモメンタムの実装はスクリプトによって差があります（モメンタムの線形回帰の取り方、バンドの基準線）。この数値はFormiqの実装によるものです - エントリーはスクイーズが解けた足に限定しています。スクイーズ中のモメンタム転換は拾いません - エントリーも決済も足の終値で行っています - スプレッドは全期間0.3pips固定です - 4時間足は年間0〜80回で、45通り中8通りは1度も売買していません。この件数の勝率やPFは幅の広い数字です --- ### STCの期間と売買水準：72通りの設定を比較 URL: https://formiq.jp/ja/blog/stc-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: STC, シャフトレンドサイクル, インジケーター, パラメーター, ドル円 Also available in: en — https://formiq.jp/blog/stc-settings STC（シャフ・トレンド・サイクル）は、MACDにストキャスティクスを二重にかけて0〜100の範囲に収めた指標です。MACDより反応が速く、レンジ相場での往復を減らすことを狙って作られています。 ドル円の2025年（1月1日〜12月31日）で、短期・長期・ストキャス期間・判定水準の4つを72通り組み合わせて検証しました。時間足は15分足・1時間足・4時間足、比較のために2024年でも同じ72通りを回しています。 両方の年で黒字だった設定は、15分足10/72、1時間足49/72、4時間足36/72でした。4時間足の10/30/5・水準25は年間損益が2025年に+3,399.7pips、2024年に+3,014.6pipsでしたが、同じ4時間足の初期設定は2024年に−593.4pipsです。 ## 初期設定の23/50/10はどうか | 設定 | 2025年の年間損益 | 2024年の年間損益 | |---|---:|---:| | 初期設定 23/50/10・水準25 | +1,718.3pips | +1,721.7pips | | 23/80/10・水準40 | +3,024.7pips | +2,730.4pips | | 23/80/20・水準10 | −1,670.8pips | +713.3pips | **初期設定の2年の差は3.4pips**（+1,718.3と+1,721.7）です。ただし、23/80/10・水準40の年間損益は2025年に+3,024.7pipsで、初期設定より1,306.4pips大きくなりました。 「初期設定が両年黒字だった」と「初期設定の年間損益が最大だった」は別の話です。 ## 一番良かった設定はどれか 4時間足の10/30/5・水準25は、年間損益が2025年に+3,399.7pips、2024年に+3,014.6pipsとなり、検証した72設定の中で両年とも最大でした。 ただし、確認したのはドル円の2年間です。この1設定が両年とも最大だった事実だけでは、別の年や通貨ペアでも同じ設定を選ぶ根拠にはなりません。 判断材料として、この設定の中身を見ておきます。 | 指標 | 2025年 | 2024年 | |---|---|---| | トレード回数 | 164回 | 165回 | | 勝率 | 50.00% | 40.61% | | プロフィットファクター | 1.657 | 1.575 | | 損益 | +3,399.7pips | +3,014.6pips | 年160回台という回数は、4時間足としては多いほうです。15回や27回でPFが跳ね上がったケース（[アルーン](/ja/blog/aroon-settings)・[スーパートレンド](/ja/blog/supertrend-settings)の4時間足）とは違い、標本の薄さで説明できる数字ではありません。 ## STCはどの時間足で使えるか 両年とも黒字だった設定が最も多いのは、1時間足の49/72でした。 | 時間足 | 2025年に黒字 | 2025年の中央値 | 2024年に黒字 | 2024年の中央値 | 両方の年で黒字 | 年間の取引回数 | |---|---|---|---|---|---|---| | 15分足 | 33 / 72 | −54pips | 19 / 72 | −1,244pips | 10 / 72 | 約1,400回 | | **1時間足** | 62 / 72 | +1,198pips | 59 / 72 | +1,234pips | **49 / 72** | 約340回 | | 4時間足 | 54 / 72 | +925pips | 52 / 72 | +642pips | 36 / 72 | 約90回 | 15分足は年1,400回。ここでもコストが効いて、両方の年で黒字だったのは10/72にとどまります。 4時間足は年90回まで落ちて、両方の年で黒字が36/72。1時間足の49/72より低い数字です。 ## 4時間足でも初期設定は使えるか | 設定 | 2025年の年間損益 | 2024年の年間損益 | |---|---:|---:| | 初期設定 23/50/10・水準25 | +1,722.8pips | −593.4pips | | 10/30/5・水準25 | +3,399.7pips | +3,014.6pips | | 40/80/20・水準40 | −875.6pips | +1,367.1pips | 4時間足の初期設定は2025年に+1,722.8pips、2024年に−593.4pipsでした。**同じ初期設定が、1時間足では2年とも約+1,720pips、4時間足では片方の年だけ赤字になりました。** 4時間足の勝率は37.50%〜64.29%と幅が広く、年90回という回数の薄さがそのまま出ています。 ## 勝率はどこまで開くか 2025年の72設定で、1時間足の勝率は32.98%〜52.40%、中央値は42.9%でした。4時間足では37.50%〜64.29%まで開きます。4時間足の取引回数は年間約90回なので、勝率が高い設定も取引回数と年間損益を一緒に確認する必要があります。 ## チャートに表示する方法 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準では入っていない。配布されている .ex4／.mq4 を `MQL4/Indicators`（MT5は `MQL5/Indicators`）に置いて再起動する | | TradingView | インジケーター検索で「Schaff Trend Cycle」を選ぶ | | ブラウザ（Formiq） | インジケーター一覧から選ぶ。短期・長期・ストキャス期間を設定から変更できる | ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い | STCが判定水準（10・25・40）を上抜けた足の終値 | | 売り | STCが「100 − 判定水準」を下抜けた足の終値 | | 決済 | 反対側の水準抜けのみ。損切り・利確・時間決済は使わない | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した組み合わせ | 短期3種（10・23・40）× 長期3種（30・50・80）× ストキャス期間3種（5・10・20）× 水準3種のうち短期<長期の72通り | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | 同じ方法で調べた[OsMA](/ja/blog/osma-settings)・[ACオシレーター](/ja/blog/ac-oscillator-settings)・[ストキャスRSI](/ja/blog/stoch-rsi-settings)・[QQE](/ja/blog/qqe-settings)・[アルーン](/ja/blog/aroon-settings)・[RVI](/ja/blog/rvi-settings)・[スーパートレンド](/ja/blog/supertrend-settings)・[ウィリアムズ%R](/ja/blog/williams-r-settings)・[MFI](/ja/blog/mfi-settings)の記事もあります。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです - STCの実装はスクリプトによって差があります（二重ストキャスの平滑化係数など）。この数値はFormiqの実装によるものです - エントリーも決済も足の終値で行っています - スプレッドは全期間0.3pips固定です - 4時間足は年間57〜172回です。この件数の勝率やPFは幅の広い数字です --- ### ストキャスRSIのクロスと時間足を36通り検証 URL: https://formiq.jp/ja/blog/stoch-rsi-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ストキャスRSI, Stoch RSI, インジケーター, パラメーター, ドル円 Also available in: en — https://formiq.jp/blog/stoch-rsi-settings ストキャスRSIは、RSIにストキャスティクスをもう一段かけた指標です。RSIの数値そのものではなく、直近N本のRSIの高安レンジの中でいまのRSIがどこにいるかを見ます。二重に加工するぶん反応が速くなります。 ドル円の2025年（1月1日〜12月31日）で、RSI期間・ストキャス期間・%K平滑化・%D平滑化の4つを36通り組み合わせて検証しました。時間足は15分足・1時間足・4時間足、比較のために2024年でも同じ36通りを回しています。 **1時間足では、両方の年で黒字だった設定が36通り中3通りしかありません。** 2024年に黒字だったのは7/36です。 その一方で、**4時間足では30通りが両方の年で黒字**でした。同じ指標・同じパラメータで、時間足を変えるだけで結果が入れ替わっています。 ## 時間足でどう変わるか 結果が正反対になりました。 | 時間足 | 年 | 黒字の設定数 | 年間損益の中央値 | 年間の取引回数（中央値） | |---|---:|---:|---:|---:| | 15分足 | 2025年 | 0 / 36 | −1,139pips | 約4,850回 | | 15分足 | 2024年 | 3 / 36 | −1,205pips | 対象外 | | 1時間足 | 2025年 | 20 / 36 | +162pips | 約1,230回 | | 1時間足 | 2024年 | 7 / 36 | −1,008pips | 対象外 | | 4時間足 | 2025年 | 30 / 36 | +1,261pips | 約300回 | | 4時間足 | 2024年 | 36 / 36 | +2,534pips | 対象外 | 理由は取引回数にあります。 この指標は**二重に加工されているぶん、シグナルが非常に多く出ます**。15分足では年間約4,850回。同じ条件で[QQE](/ja/blog/qqe-settings)は年465回、[スーパートレンド](/ja/blog/supertrend-settings)は年152回でした。 スプレッドで失う額は取引回数にそのまま比例します。年4,850回なら0.3pipsでも年間1,455pips相当の負担です。1時間足の1,230回でも369pips。**4時間足の300回まで落として、ようやく回数がコストに見合う水準になりました。** ## 1時間足では設定を変えるべきか 変えても報われませんでした。 | 設定 | 年 | 年間損益 | |---|---:|---:| | 初期設定 14/14/3/3 | 2025年 | +764.5pips | | 初期設定 14/14/3/3 | 2024年 | −1,811.3pips | | 2025年に最大だった14/21/3/5 | 2025年 | +1,560.4pips | | 2025年に最大だった14/21/3/5 | 2024年 | −2,011.2pips | | 2024年に最大だった14/7/5/3 | 2025年 | −2,200.9pips | | 2024年に最大だった14/7/5/3 | 2024年 | +858.4pips | | 2025年に最小だった7/7/5/5 | 2025年 | −3,221.7pips | | 2025年に最小だった7/7/5/5 | 2024年 | −1,009.9pips | 2024年に+858.4pipsだった14/7/5/3は、2025年に−2,200.9pipsへ反転しました。逆に2025年に+1,560.4pipsだった14/21/3/5は、2024年に−2,011.2pipsです。**片方の年で選んだ設定は、もう片方の年でどちらも赤字でした。** ## 4時間足なら設定を選べるか 年ごとに最大の設定が変わります。 | 設定 | 年 | 年間損益 | |---|---:|---:| | 初期設定 14/14/3/3 | 2025年 | +1,437.3pips | | 初期設定 14/14/3/3 | 2024年 | +2,430.4pips | | 2025年に最大だった21/21/3/3 | 2025年 | +2,333.0pips | | 2025年に最大だった21/21/3/3 | 2024年 | +2,205.8pips | | 2024年に最大だった7/21/5/5 | 2025年 | −863.4pips | | 2024年に最大だった7/21/5/5 | 2024年 | +3,799.1pips | | 2025年に最小だった7/14/5/5 | 2025年 | −1,389.5pips | | 2025年に最小だった7/14/5/5 | 2024年 | +1,918.2pips | 2024年に+3,799.1pipsだった7/21/5/5は、2025年に−863.4pipsへ反転しました。2025年に最大だった21/21/3/3は、両方の年で+2,200pipsを超えています。 2024年は36通り全部が黒字でした。初期設定の14/14/3/3は、**両方の年で+1,400pips以上**を残しています。 ## 勝率は損益と結びつくか 無関係でした。1時間足・2025年の36通りで、勝率は38.17%〜41.84%、中央値39.8%でした。3.67ポイントの狭い範囲でも、年間損益は−3,221.7〜+1,560.4pipsに分かれています。 4時間足では勝率が36.72%〜45.85%まで広がりました。勝率の範囲だけで設定の年間損益は判断できません。 ## ストキャスRSIの表示と設定 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準では入っていない。配布されている .ex4／.mq4 を `MQL4/Indicators`（MT5は `MQL5/Indicators`）に置いて再起動する | | TradingView | インジケーター検索で「Stochastic RSI」を選ぶ | | ブラウザ（Formiq） | インジケーター一覧から選ぶ。RSI期間・ストキャス期間・%K・%Dを設定から変更できる | ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い | %Kが%Dを上抜けた足の終値 | | 売り | %Kが%Dを下抜けた足の終値 | | 決済 | 反対側のクロスのみ。損切り・利確・時間決済は使わない | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した組み合わせ | RSI期間3種（7・14・21）× ストキャス期間3種（7・14・21）× %K 2種（3・5）× %D 2種（3・5）＝ 36通り | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | この指標を使うなら、時間足を上げて回数を減らすことが前提条件になります。[同じことがRVIでも起きています](/ja/blog/rvi-settings)。元になっている[RSI](/ja/blog/rsi-settings)は、加工しないぶんシグナルが桁違いに少なくなります。ほかに[OsMA](/ja/blog/osma-settings)・[ACオシレーター](/ja/blog/ac-oscillator-settings)・[QQE](/ja/blog/qqe-settings)・[アルーン](/ja/blog/aroon-settings)・[スーパートレンド](/ja/blog/supertrend-settings)・[ウィリアムズ%R](/ja/blog/williams-r-settings)・[MFI](/ja/blog/mfi-settings)の検証記事があります。 ## 関連記事 - [ストキャスティクスの期間・スロー化・勝率](/ja/blog/stochastic-settings)：価格の高安幅を使う水準読みとクロスを比較 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです - ストキャスRSIの実装はスクリプトによって差があります（RSIの平滑化方法、%Kの取り方）。この数値はFormiqの実装によるものです - 判定は%Kと%Dのクロスのみで、20/80のような水準判定は試していません。水準で絞れば回数は減ります - エントリーも決済も足の終値で行っています - スプレッドは全期間0.3pips固定です。回数の多いこの指標では、この前提が結果を大きく左右します - 4時間足は年間230〜397回です --- ### スーパートレンドの期間と倍率：36通りの設定を検証 URL: https://formiq.jp/ja/blog/supertrend-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: スーパートレンド, Supertrend, インジケーター, パラメーター, ドル円 Also available in: en — https://formiq.jp/blog/supertrend-settings スーパートレンドは「10と3」という2つの数字とセットで紹介されます。10はATR期間、3はATRへ掛ける倍率です。設定画面ではどちらも変更できますが、売買結果への影響は同じではありません。 実際に両方を動かしてみました。ドル円の2025年（1月1日〜12月31日）で、期間6通り × 倍率6通りの36通りをバックテストし、時間足を3つ、比較のために2024年でも同じ36通りを回しています。 結果を先に書きます。**期間はほとんど何もしていませんでした。** 1時間足で期間を5から30まで6倍に広げても、年間の取引回数は205回から212回の間です。損益の平均も406pipsの幅しか動きません。 動かしていたのは倍率のほうでした。1.5から5に変えると、取引回数は408回から80回へ5倍の幅で動きます。 ## スーパートレンドが計算していること 1. **ATR（平均的な値幅）を期間N本ぶん求める** ← ここが「期間」 2. 中値（高値と安値の平均）から、ATR × 倍率だけ上下に離したバンドを引く ← ここが「倍率」 3. 価格がバンドを抜けた側でトレンドの向きを判定し、線がその反対側に付いてくる ATR期間と倍率では、バンド幅と取引回数への影響が異なります。 - **期間**を変えると、ATRの計算に使う本数が変わります。ただしドル円のATRは期間5でも期間30でも近い値になるので、**バンドの幅はあまり変わりません** - **倍率**を変えると、そのATRに掛ける数がそのまま変わります。**バンドの幅が直接変わります** バンドの幅が変われば、価格が抜ける頻度が変わります。だから倍率だけが取引回数を動かします。 実測でもそのとおりになりました。冒頭の図がその比較です。 | 比較順 | 期間を変えた場合の年間回数 | 倍率を変えた場合の年間回数 | |---|---|---| | 1番目 | 期間5 → 212回 | 倍率1.5 → 408回 | | 2番目 | 期間7 → 207回 | 倍率2 → 291回 | | 3番目 | 期間10 → 205回 | 倍率2.5 → 201回 | | 4番目 | 期間14 → 205回 | 倍率3 → 159回 | | 5番目 | 期間20 → 206回 | 倍率4 → 104回 | | 6番目 | 期間30 → 207回 | 倍率5 → 80回 | （1時間足・2025年。期間の列は倍率6通り、倍率の列は期間6通りで平均した値） 期間の列は205〜212に収まり、倍率の列は80〜408です。 ## チャートに表示する方法 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準では入っていない。配布されている .ex4／.mq4 を `MQL4/Indicators`（MT5は `MQL5/Indicators`）に置いて再起動する | | TradingView | インジケーター検索で「Supertrend」を選ぶ | | ブラウザ（Formiq） | インジケーター一覧から選ぶ。期間と倍率を設定から変更できる | ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い | スーパートレンドが上向きに転換した足の終値 | | 売り | 下向きに転換した足の終値 | | 決済 | 反対側への転換のみ。損切り・利確・時間決済は使わない | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した組み合わせ | 期間6種（5〜30）× 倍率6種（1.5〜5）＝ 36通り。時間足ごと・年ごとに実行 | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | ## 期間と倍率のどちらが重要か 結果を動かしたのは倍率のほうでした。1時間足の36通りを、2025年と2024年で並べます。 数字で見ると次のようになります（1時間足・2025年・単位はpips）。 | 期間＼倍率 | 1.5 | 2 | 2.5 | 3 | 4 | 5 | |---|---|---|---|---|---|---| | 5 | −162 | +370 | **+1,927** | +1,009 | +1,132 | +562 | | 7 | −99 | +342 | +1,091 | +1,068 | +797 | +267 | | 10 | −56 | −188 | +726 | **+1,122** | +1,011 | +292 | | 14 | −122 | +497 | +911 | +674 | +1,567 | +478 | | 20 | +396 | −164 | +889 | −142 | +1,311 | +109 | | 30 | +1,024 | −349 | +1,817 | −297 | +1,292 | +52 | 縦（期間）に読んでも規則が出てきません。横（倍率）に読むと、**倍率1.5の列は6通り中4通りが赤字**で、2.5と4の列が良い、という傾向が見えます。 2024年は同じ表がこうなります。 | 期間＼倍率 | 1.5 | 2 | 2.5 | 3 | 4 | 5 | |---|---|---|---|---|---|---| | 5 | +347 | +1,722 | +1,091 | +840 | +1,224 | +1,408 | | 7 | +1,422 | +1,706 | +319 | +570 | +1,425 | +1,428 | | 10 | +1,575 | +1,401 | −269 | **+1,145** | +1,956 | +2,075 | | 14 | +1,413 | −497 | −257 | +1,531 | +1,236 | +2,232 | | 20 | +1,775 | −856 | −165 | +172 | +966 | +1,749 | | 30 | +2,187 | −421 | −258 | −242 | −4 | +2,664 | 2024年は**倍率2.5の列が平均で最も悪く（+77pips）、5の列が最も良い（+1,926pips）**。2025年はその2.5が1位（+1,227pips）で、5は下から3番目（+293pips）です。**同じ倍率が、片方の年で1位・もう片方で最下位**になっています。 | 倍率 | 年 | 平均年間損益 | 年間の取引回数 | |---:|---:|---:|---:| | 1.5 | 2025年 | +164pips | 408回 | | 1.5 | 2024年 | +1,453pips | 対象外 | | 2 | 2025年 | +85pips | 290回 | | 2 | 2024年 | +509pips | 対象外 | | 2.5 | 2025年 | +1,227pips | 201回 | | 2.5 | 2024年 | +77pips | 対象外 | | 3 | 2025年 | +572pips | 159回 | | 3 | 2024年 | +669pips | 対象外 | | 4 | 2025年 | +1,185pips | 104回 | | 4 | 2024年 | +1,134pips | 対象外 | | 5 | 2025年 | +293pips | 80回 | | 5 | 2024年 | +1,926pips | 対象外 | ## 初期設定の10・3のままでいいか | 時間足 | 年 | 取引回数 | 年間損益 | |---|---:|---:|---:| | 15分足 | 2025年 | 659回 | +1,064.3pips | | 15分足 | 2024年 | 613回 | +1,849.0pips | | 1時間足 | 2025年 | 152回 | +1,121.8pips | | 1時間足 | 2024年 | 155回 | +1,145.0pips | | 4時間足 | 2025年 | 42回 | −99.1pips | | 4時間足 | 2024年 | 34回 | +2,960.5pips | 15分足と1時間足では、初期設定は**どちらの年も黒字**でした。この検証シリーズで扱った指標のうち、初期設定がこれだけ安定していたのはスーパートレンドだけです。 一方で、2024年に+2,663.7pipsだった期間30・倍率5は、2025年に+52.2pipsまで減りました。片方の年で年間損益が最大だった設定が、翌年も最大になるとは限りません。 ## どの時間足がいいか | 時間足 | 年 | 黒字の設定数 | 両方の年で黒字 | |---|---:|---:|---:| | 15分足 | 2025年 | 26 / 36 | 21 / 36 | | 15分足 | 2024年 | 26 / 36 | 21 / 36 | | 1時間足 | 2025年 | 27 / 36 | 21 / 36 | | 1時間足 | 2024年 | 27 / 36 | 21 / 36 | | 4時間足 | 2025年 | 10 / 36 | 10 / 36 | | 4時間足 | 2024年 | 36 / 36 | 10 / 36 | 4時間足だけが崩れます。2024年は36通りが36通りとも黒字だったのに、2025年は10通りしか残りませんでした。そして崩れ方には理由があります。 2024年の4時間足で最も成績が良かったのは期間30・倍率5でした。 | 年 | 取引回数 | 勝率 | プロフィットファクター | 年間損益 | |---:|---:|---:|---:|---:| | 2024年 | 15回 | 60.00% | 5.038 | +3,556.4pips | | 2025年 | 28回 | 21.43% | 0.337 | −2,235.9pips | **15回**です。プロフィットファクター5.04という数字は、15回のうち9勝したという以上の意味を持ちません。翌年は−2,235.9pipsでした。 4時間足の36通りは年間26〜110回しかトレードしません。倍率を上げるほど回数が減り、倍率5の行は年間26〜29回まで落ちます。**倍率5・4時間足の組み合わせは構造的に「数字が大きく振れる設定」**です。良く見えたときほど疑うべき領域です。 ## スプレッドにどれだけ強いか 回数が少ないことは、この指標の実用上の利点でもあります。 | 指標（1時間足・2025年） | 年間の取引回数 | スプレッド0→2.0pipsで失うpips | |---|---|---| | **スーパートレンド 10/3** | **152回** | **304pips** | | アルーン 25・クロス | 228回 | 456pips | | QQE 17/3/4.236 | 465回 | 930pips | | RVI 10 | 1,094回 | 2,188pips | 失う額は、どの指標でも**取引回数 × スプレッド**にぴったり一致します。スーパートレンドは同じ条件で最も回数が少ないため、コストの影響が最小でした。 | スプレッド | 時間足 | 設定 | 取引回数 | 年間損益 | |---:|---|---|---:|---:| | 0.0pips | 1時間足 | 10/3 | 152回 | +1,167.4pips | | 0.3pips | 1時間足 | 10/3 | 152回 | +1,121.8pips | | 1.0pips | 1時間足 | 10/3 | 152回 | +1,015.4pips | | 2.0pips | 1時間足 | 10/3 | 152回 | +863.4pips | | 0.0pips | 15分足 | 10/3 | 659回 | +1,262.0pips | | 0.3pips | 15分足 | 10/3 | 659回 | +1,064.3pips | | 1.0pips | 15分足 | 10/3 | 659回 | +603.0pips | | 2.0pips | 15分足 | 10/3 | 659回 | −56.0pips | 1時間足は2.0pipsのスプレッドでも+863.4pipsを残します。15分足は2.0pipsでちょうどゼロを割りました。 ## 勝率で設定を選べるか 倍率が大きいほど勝率も年間損益も上がりますが、年による差が残ります。1時間足・2025年では、倍率が大きいほど取引回数が減り、勝率と年間損益が大きい設定が増えました。ただし倍率5の平均年間損益は2025年+293pips、2024年+1,926pipsで、年による差があります。 フィルターを初期設定に足すとこうなります。 | 条件 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | スーパートレンドのみ | 2025年 | 152回 | 44.08% | +1,121.8pips | | スーパートレンドのみ | 2024年 | 155回 | 39.35% | +1,145.0pips | | ADX 20以上 | 2025年 | 81回 | 45.68% | +673.0pips | | ADX 20以上 | 2024年 | 80回 | 36.25% | +825.9pips | | ADX 25以上 | 2025年 | 54回 | 44.44% | +362.8pips | | ADX 25以上 | 2024年 | 47回 | 38.30% | −318.9pips | | ADX 30以上 | 2025年 | 32回 | 50.00% | +243.2pips | | ADX 30以上 | 2024年 | 31回 | 35.48% | −328.6pips | | 東京時間のみ | 2025年 | 48回 | 45.83% | +582.1pips | | 東京時間のみ | 2024年 | 55回 | 38.18% | −79.3pips | | ロンドン・NY時間のみ | 2025年 | 103回 | 43.69% | +571.7pips | | ロンドン・NY時間のみ | 2024年 | 100回 | 42.00% | +1,378.7pips | ADXは両方の年で損益を下げました。トレンドの強さで絞るフィルターを、すでにトレンドフォローである指標に重ねても、削るのは利益のほうだったということです。ADX 25以上と30以上は2024年を赤字に落としています。 **両方の年で黒字を保ったのは、ADX 20以上（+673.0／+825.9pips）とロンドン・NY時間（+571.7／+1,378.7pips）の2つ**です。ただしどちらも2025年は基準の+1,121.8pipsを下回ります。符号は守るが額は増やさない、という他の指標と同じ挙動でした。 ## 損切りと利確 | 決済ルール | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | 反対への転換のみ | 2025年 | 152回 | 44.08% | +1,121.8pips | | 反対への転換のみ | 2024年 | 155回 | 39.35% | +1,145.0pips | | 損切り30・利確60 | 2025年 | 139回 | 37.41% | +486.1pips | | 損切り30・利確60 | 2024年 | 146回 | 34.25% | +107.3pips | | 損切り50・利確100 | 2025年 | 136回 | 39.71% | +1,103.2pips | | 損切り50・利確100 | 2024年 | 141回 | 36.17% | +238.8pips | | 損切り100・利確200 | 2025年 | 144回 | 45.83% | +2,952.4pips | | 損切り100・利確200 | 2024年 | 141回 | 39.72% | +1,291.8pips | | 損切り50・利確50 | 2025年 | 138回 | 55.07% | +817.2pips | | 損切り50・利確50 | 2024年 | 145回 | 51.72% | +524.0pips | | 24本で時間決済 | 2025年 | 149回 | 52.35% | +2,035.9pips | | 24本で時間決済 | 2024年 | 152回 | 46.05% | +942.9pips | **6通りすべてが両方の年で黒字**です。これはこのシリーズで初めてでした。さらに**損切り100／利確200は両方の年で「反対への転換のみ」を上回ります**（+2,952.4／+1,291.8pips）。広い損切りと広い利確を置くと、転換を待つより良い結果になりました。 損切り50／利確50は勝率を両方の年で50%台に乗せます（55.07%／51.72%）。損益は増えませんが、勝率の見た目は安定します。 同じATRを単体で測った[ATRの期間と比較本数](/ja/blog/atr-settings)では、期間14が売買ルールとして一番良いかを検証しています。同じ方法で調べた[パラボリックSAR](/ja/blog/parabolic-sar-settings)・[QQE](/ja/blog/qqe-settings)・[アルーン](/ja/blog/aroon-settings)・[RVI](/ja/blog/rvi-settings)・[移動平均線クロス](/ja/blog/moving-average-cross-settings)の記事もあります。[プログラミングなしで条件を組む](/ja/blog/backtest-without-coding)なら、この表と同じことを自分の通貨ペアで確かめられます。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです - スーパートレンドの実装はATRの計算方法（単純平均かWilderか）とバンドの追従ルールで差が出ます。この数値はFormiqの実装によるものです - エントリーも決済も足の終値で行っています - スプレッドは全期間0.3pips固定として計算しています - 4時間足は年間8〜55回です。この件数の勝率やPFは、そもそも幅の広い数字です --- ### ウェーブトレンドの設定と時間足を30通り検証 URL: https://formiq.jp/ja/blog/wavetrend-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ウェーブトレンド, WaveTrend, インジケーター, パラメーター, ドル円 Also available in: en — https://formiq.jp/blog/wavetrend-settings ウェーブトレンドは、中値がその移動平均からどれだけ離れているかを、平均的な乖離幅で割って標準化した指標です。標準化した値から2本の線を作り、その交差で売買を判定します。 ドル円の2025年（1月1日〜12月31日）で、チャネル期間と平均期間の2つを30通り組み合わせて検証しました。時間足は15分足・1時間足・4時間足、比較のために2024年でも同じ30通りを回しています。 **1時間足では2025年に30通り全部が黒字、4時間足でも2024年に30通り全部が黒字**でした。そして**15分足では両方の年で黒字だった設定がゼロ**です。 ## どの時間足で使えるか 15分足では黒字になった設定がありません。 | 時間足 | 年 | 黒字の設定数 | 年間損益の中央値 | 年間の取引回数（中央値） | |---|---:|---:|---:|---:| | 15分足 | 2025年 | 2 / 30 | −2,142pips | 3,598回 | | 15分足 | 2024年 | 1 / 30 | −953pips | 対象外 | | 1時間足 | 2025年 | 30 / 30 | +2,466pips | 838回 | | 1時間足 | 2024年 | 24 / 30 | +1,185pips | 対象外 | | 4時間足 | 2025年 | 29 / 30 | +1,318pips | 224回 | | 4時間足 | 2024年 | 30 / 30 | +1,566pips | 対象外 | 15分足は年3,600回。**2025年は30通り中2通り、2024年は1通りしか黒字にならず、両方の年で黒字だった設定はありません。** 0.3pipsのスプレッドでも年間1,080pips相当の負担になり、それを超える利益が残りませんでした。 1時間足と4時間足は対照的に安定しています。4時間足は両方の年で黒字が29/30。この検証シリーズで扱った9つの指標の中でも上位の安定度でした。 ## 初期設定の10/21のままでいいか 片方の年だけでは判断できませんでした。 | 設定 | 年 | 年間損益 | |---|---:|---:| | 初期設定 10/21 | 2025年 | +3,283.4pips | | 初期設定 10/21 | 2024年 | +1,076.6pips | | 2025年に最大だった20/15 | 2025年 | +3,496.9pips | | 2025年に最大だった20/15 | 2024年 | +606.9pips | | 2024年に最大だった30/30 | 2025年 | +2,248.7pips | | 2024年に最大だった30/30 | 2024年 | +2,938.9pips | | 2025年に最小だった5/10 | 2025年 | +1,339.6pips | | 2025年に最小だった5/10 | 2024年 | −1,266.5pips | 2025年の差は213.5pipsですが、2024年は初期設定のほうが469.7pips大きい結果でした。**片方の年だけで20/15へ変える根拠にはなりません。** 2025年に年間損益が最も小さかった5/10でも+1,339.6pipsの黒字です。一方、同じ設定は2024年に−1,266.5pipsでした。 ## 4時間足でも同じことが言えるか 4時間足の初期設定も両年黒字でした。 | 設定 | 年 | 年間損益 | |---|---:|---:| | 初期設定 10/21 | 2025年 | +1,262.2pips | | 初期設定 10/21 | 2024年 | +1,373.5pips | | 2025年に最大だった20/21 | 2025年 | +2,476.8pips | | 2025年に最大だった20/21 | 2024年 | +506.1pips | | 2024年に最大だった5/10 | 2025年 | +1,232.4pips | | 2024年に最大だった5/10 | 2024年 | +3,321.3pips | | 2025年に最小だった30/45 | 2025年 | −98.2pips | | 2025年に最小だった30/45 | 2024年 | +1,009.8pips | 4時間足の初期設定は、2025年+1,262.2pips、2024年+1,373.5pipsで両年黒字でした。ただし30/45は2025年に−98.2pips、2024年に+1,009.8pipsと符号が変わっています。 ## 勝率で設定を選べるか 選べません。1時間足・2025年の30通りで、勝率は33.48%〜39.45%、中央値36.9%でした。勝率の幅は5.97ポイントですが、年間損益は+1,339.6〜+3,496.9pipsに分かれています。 15分足では30設定中28設定が2025年に赤字でした。勝率を上げることより、年3,598回という取引回数とコストを先に見る必要があります。 ## ウェーブトレンドの表示と設定 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準では入っていない。配布されている .ex4／.mq4 を `MQL4/Indicators`（MT5は `MQL5/Indicators`）に置いて再起動する | | TradingView | インジケーター検索で「WaveTrend」を選ぶ | | ブラウザ（Formiq） | インジケーター一覧から選ぶ。チャネル期間と平均期間を設定から変更できる | ## 検証の条件 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い | wt1がwt2を上抜けた足の終値 | | 売り | wt1がwt2を下抜けた足の終値 | | 決済 | 反対側のクロスのみ。損切り・利確・時間決済は使わない | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した組み合わせ | チャネル期間6種（5〜30）× 平均期間5種（10〜45）＝ 30通り | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | 同じ方法で調べた[OsMA](/ja/blog/osma-settings)・[ACオシレーター](/ja/blog/ac-oscillator-settings)・[ストキャスRSI](/ja/blog/stoch-rsi-settings)・[STC](/ja/blog/stc-settings)・[QQE](/ja/blog/qqe-settings)・[アルーン](/ja/blog/aroon-settings)・[RVI](/ja/blog/rvi-settings)・[スーパートレンド](/ja/blog/supertrend-settings)・[ウィリアムズ%R](/ja/blog/williams-r-settings)・[MFI](/ja/blog/mfi-settings)の記事もあります。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです - ウェーブトレンドの実装はスクリプトによって差があります（乖離の標準化に使う係数、平滑化の種類）。この数値はFormiqの実装によるものです - 判定は2本の線の交差のみで、±60のような過熱水準は使っていません - エントリーも決済も足の終値で行っています - スプレッドは全期間0.3pips固定です。15分足の結果はこの前提に強く依存します - 4時間足は年間147〜325回です --- ### ウィリアムズ%Rの逆張り設定を240通り比較 URL: https://formiq.jp/ja/blog/williams-r-settings Language: ja Published: 2026-08-24 Updated: 2026-08-30 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: ウィリアムズ%R, Williams %R, インジケーター, 勝率, ドル円 Also available in: en — https://formiq.jp/blog/williams-r-settings ウィリアムズ%Rは直近N本の高値・安値のレンジの中で、いまの終値がどこにいるかを 0 から −100 で表す指標です。−80より下なら売られすぎ、−20より上なら買われすぎ、と紹介されます。 この使い方をそのまま検証しました。ドル円の2025年（1月1日〜12月31日）で、期間6通り × エントリー水準5通り × 決済水準4通り × 買い/売りの2方向、合わせて240通りです。時間足は15分足・1時間足・4時間足、比較のために2024年でも同じ240通りを回しています。 結果を先に書きます。**勝率は高く出ました。そして損益はマイナスでした。** 1時間足の買い側120通りで、勝率の中央値は59.0%。**それでも黒字になったのは120通り中10通り**です。定番の「期間14・−80で買い・−20で決済」は、2024年に**勝率68.42%を記録しながら−569.3pips**でした。 点ひとつが1つの設定です。横軸が勝率、縦軸が年間の損益。勝率は49%から67%まで散らばっていますが、点の集団はほぼ全部ゼロ線の下にあります。 ## 勝率68%なのになぜ負けるのか 勝ちトレードの平均より、負けトレードの平均損失が大きいためです。 | 時間足 | 年 | 設定 | 勝率 | 勝ちの平均 | 負けの平均 | 年間損益 | |---|---:|---|---:|---:|---:|---:| | 1時間足 | 2025年 | 期間14・−80/−20 | 58.11% | +39.8pips | −72.7pips | −1,082.8pips | | 1時間足 | 2024年 | 期間14・−80/−20 | 68.42% | +44.9pips | −110.8pips | −569.3pips | | 1時間足 | 2025年 | 勝率67.05%の設定 | 67.05% | +28.3pips | −74.1pips | −953.7pips | | 4時間足 | 2024年 | 期間14・−80/−20 | 79.41% | +84.9pips | −267.0pips | +423.5pips | 4時間足・2024年の定番設定は**勝率79.41%でも年間損益が+423.5pips**でした。平均損失−267.0pipsが平均利益+84.9pipsの3.1倍あるため、約2割の負けで利益の大半を失っています。 勝率が高くても損失が大きくなるのは、−80以下で買い、−20まで戻るのを待って決済するためです。**利益は−20まで戻った値幅に限られます。** 一方、買った後も下落が続くと決済条件が成立せず、損失が平均利益より大きくなります。 この売買ルールでは、高い勝率が大きな年間損益にはつながりませんでした。 **1時間足・2025年の120通りで、勝率70%を超えた設定はひとつもありません**（最高67.05%）。その設定も年間損益は−953.7pipsでした。 ## ウィリアムズ%Rの表示と設定 | 環境 | 手順 | |---|---| | MT4 / MT5 | 標準で入っている。ナビゲーターの「Oscillators」から Williams' Percent Range を選ぶ | | TradingView | インジケーター検索で「Williams %R」を選ぶ | | ブラウザ（Formiq） | インジケーター一覧から選ぶ。期間は設定から変更できる | ## 検証の条件 この指標のルールは、**買いと売りのどちらか一方しか作りません**。「−80より下」は買いの条件にしかならず、「−20より上」は売りの条件にしかなりません。そのため買い逆張りと売り逆張りを別々の売買ルールとして検証しています。 | 項目 | 値 | |---|---| | 通貨ペア | 米ドル／円 | | 期間 | 2025-01-01 〜 2025-12-31（比較用に2024年も同条件） | | 時間足 | 15分足 / 1時間足 / 4時間足 | | 検証本数 | 1時間足で6,226本（2025年）。15分足24,903本、4時間足1,610本 | | 買い側 | %Rがエントリー水準（−95〜−70）を下回ったら買い、決済水準（−50〜−10）を上回ったら決済 | | 売り側 | 上記を0側で反転させたもの（−5〜−30で売り、−50〜−90で決済） | | 損切り・利確 | 使わない（別途「損切りを付けたら」で測定） | | スプレッド | 0.3pips（スリッページ0、0.1ロット） | | 試した組み合わせ | 期間6種（7〜40）× エントリー5種 × 決済4種 × 2方向 ＝ 240通り。時間足ごと・年ごとに実行 | | 測定 | Formiqのバックテスト機能で実行し、1トレードずつの約定価格からpipsを再集計 | ## 買い側と売り側で違うのか 結果はまったく違いました。 | 時間足 | 売買方向 | 年 | 黒字の設定数 | 年間損益の中央値 | 勝率の中央値 | |---|---|---:|---:|---:|---:| | 15分足 | 買い | 2025年 | 3 / 120 | −661pips | 64.3% | | 15分足 | 買い | 2024年 | 54 / 120 | −226pips | 対象外 | | 15分足 | 売り | 2025年 | 7 / 120 | −777pips | 63.3% | | 15分足 | 売り | 2024年 | 2 / 120 | −1,169pips | 対象外 | | 1時間足 | 買い | 2025年 | 10 / 120 | −720pips | 59.0% | | 1時間足 | 買い | 2024年 | 47 / 120 | −201pips | 対象外 | | 1時間足 | 売り | 2025年 | 10 / 120 | −633pips | 59.3% | | 1時間足 | 売り | 2024年 | 0 / 120 | −1,442pips | 対象外 | | 4時間足 | 買い | 2025年 | 89 / 120 | +572pips | 64.1% | | 4時間足 | 買い | 2024年 | 71 / 120 | +123pips | 対象外 | | 4時間足 | 売り | 2025年 | 78 / 120 | +333pips | 57.6% | | 4時間足 | 売り | 2024年 | 0 / 120 | −1,396pips | 対象外 | 読みどころが2つあります。 **ひとつめ。売り側は2024年に全滅しました。** 1時間足・4時間足とも120通り中0通りです。2024年のドル円は年間で1,632pips上昇しています。**上昇相場で「買われすぎ」を売り続ければ、こうなります。** 逆張り指標は相場の方向とセットでしか評価できない、ということが数字で出ています。 **ふたつめ。4時間足の買い側だけが機能しました。** 2025年に89通り、2024年に71通り、両方の年で黒字だったのは40通りです。この記事で唯一、使える形が見つかった組み合わせでした。 4時間足の買い側で両方の年に黒字だったのは40/120です。片方の年だけ黒字だった設定も多く、1年の損益だけで水準と期間を決められませんでした。 ## どの時間足で使えるか 機能したのは4時間足の買い側だけでした。 | 期間 | 年 | 平均年間損益 | 平均勝率 | |---:|---:|---:|---:| | 7 | 2025年 | −83pips | 57.1% | | 7 | 2024年 | +549pips | 73.3% | | 9 | 2025年 | −52pips | 56.7% | | 9 | 2024年 | +520pips | 77.3% | | 14 | 2025年 | +203pips | 58.9% | | 14 | 2024年 | +197pips | 74.1% | | 21 | 2025年 | +711pips | 66.8% | | 21 | 2024年 | −138pips | 66.4% | | 28 | 2025年 | +1,143pips | 70.7% | | 28 | 2024年 | −215pips | 65.1% | | 40 | 2025年 | +935pips | 76.5% | | 40 | 2024年 | −147pips | 63.4% | 期間の効き方が真逆です。2025年は長い期間（28・40）が良く、2024年は短い期間（7・9）が良い。**両方の年で黒字だったのは期間14だけ**（+203／+197pips）でした。 エントリー水準は、深くするほど2025年の成績が上がります（−70で+135pips、−95で+757pips）。ただし2024年ではその差がほぼ消えます（+143／+71pips）。決済水準は両方の年で目立った傾向がありませんでした。 ## 損切りを付けるとどうなるか 「負けが大きいなら損切りを入れればいい」と考えるのが自然です。試しました。 | 決済ルール | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | 損切りなし | 2025年 | 148回 | 58.11% | −1,082.8pips | | 損切りなし | 2024年 | 133回 | 68.42% | −569.3pips | | 損切り30・利確60 | 2025年 | 376回 | 33.51% | −1,501.2pips | | 損切り30・利確60 | 2024年 | 297回 | 37.04% | −667.1pips | | 損切り50・利確100 | 2025年 | 281回 | 43.77% | −1,788.9pips | | 損切り50・利確100 | 2024年 | 242回 | 45.45% | −900.4pips | | 損切り20・利確40 | 2025年 | 476回 | 29.62% | −1,331.7pips | | 損切り20・利確40 | 2024年 | 370回 | 32.43% | −709.5pips | | 損切り50・利確50 | 2025年 | 304回 | 49.67% | −1,234.4pips | | 損切り50・利確50 | 2024年 | 260回 | 50.77% | −889.9pips | | 24本で時間決済 | 2025年 | 184回 | 58.15% | −989.6pips | | 24本で時間決済 | 2024年 | 152回 | 65.13% | −729.6pips | **どれも赤字のままです。** 損切りと利確を置いた4通りは、両方の年とも損切りなしより悪くなりました。24本の時間決済だけは2025年に−989.6pipsとわずかにマシですが、2024年は−729.6pipsで悪化しており、どちらにしても赤字から出ていません。 負けの平均は確かに小さくなります（−110.8pips → −30.0pips）。ところが勝率が68.42%から37.04%へ落ちて、差し引きで悪化します。損切りは「大きな負け」を「たくさんの小さな負け」に置き換えただけでした。 この形の負けが大きいのは、決済条件が来るまで待つという設計そのものから来ています。損切り幅を変えても、その設計は変わりません。 ## スプレッドのせいで負けたのか | スプレッド | 1時間足・買い・14・−80/−20（148回） | |---|---| | 0.0pips | −1,038.4 | | 0.3pips | −1,082.8 | | 1.0pips | −1,186.4 | | 2.0pips | −1,334.4 | スプレッドを0にしても−1,038.4pipsです。**この赤字はコストのせいではありません。** [他の指標では回数の多さがコストで効いてきました](/ja/blog/rvi-settings)が、ウィリアムズ%Rの1時間足は年148回で、スプレッドを0.3pipsから0へ下げても改善するのは44.4pipsだけです。1時間足・2024年の定番設定では、平均利益44.86pipsに対して平均損失が−110.75pipsだったため、コストを除いても赤字が残りました。 ## 4時間足はさらに絞り込めるか 唯一機能した4時間足の買い側で、絞り込みを試しました。 | 条件 | 年 | 取引回数 | 勝率 | 年間損益 | |---|---:|---:|---:|---:| | 期間14・−80/−20 | 2025年 | 36回 | 61.11% | +168.1pips | | 期間14・−80/−20 | 2024年 | 34回 | 79.41% | +423.5pips | | 期間28・−80/−20 | 2025年 | 24回 | 70.83% | +912.1pips | | 期間28・−80/−20 | 2024年 | 15回 | 60.00% | −785.0pips | | ロンドン・NY時間のみ | 2025年 | 36回 | 66.67% | +609.5pips | | ロンドン・NY時間のみ | 2024年 | 28回 | 75.00% | +16.8pips | | 東京時間のみ | 2025年 | 30回 | 63.33% | +363.0pips | | 東京時間のみ | 2024年 | 23回 | 69.57% | −1.0pips | 時間帯で絞ると2025年は改善しますが、2024年はどちらもほぼゼロになります。**両方の年で黒字を保ったのは、絞り込まない期間14・−80/−20だけ**でした。ただし年15〜36回という件数なので、この数字自体が細い根拠です。 単体で判断させるより、方向の判断を別に持ったうえでタイミングの一要素として使うほうが、この数字には合っています。同じ方法で調べた[QQE](/ja/blog/qqe-settings)・[アルーン](/ja/blog/aroon-settings)・[RVI](/ja/blog/rvi-settings)・[スーパートレンド](/ja/blog/supertrend-settings)の記事もあります。トレンドフォロー側の4本とは勝率の出方が正反対になる点が読み比べどころです。[ボリンジャーバンド](/ja/blog/bollinger-band-settings)では、同じ指標の中で逆張りと順張りを並べて同じ逆転を測っています。 ## 補足 - 通貨ペアは米ドル／円のみ、期間は2024年と2025年の2年だけです - ウィリアムズ%Rの計算はツールによる差がほとんどありませんが、水準の等号の扱いには流儀があります。この数値はFormiqの実装によるものです - エントリーも決済も足の終値で行っています - スプレッドは全期間0.3pips固定、スリッページは0として計算しています - 4時間足は年間9〜92回です。この件数の勝率は幅の広い数字です --- ### FXリプレイ練習の進め方：売買ルールと記録の見直し方 URL: https://formiq.jp/ja/blog/1000-replay-trades Language: ja Published: 2026-07-25 Updated: 2026-09-06 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: リプレイ練習, FX練習, トレード記録, メンタル Also available in: en — https://formiq.jp/blog/1000-replay-trades FXのリプレイ練習では、先のローソク足を隠した状態で売買を判断し、その後の値動きと照合します。練習前に売買条件を決め、取引後に条件を守れたかを記録すると、損益と操作上のミスを分けて振り返れます。 ## 練習するルールと期間を決める 通貨ペア、時間足、練習する期間を選びます。エントリー条件に加えて、損切りと利確の条件も開始前に書いておきます。「反転しそう」のような判断なら、どのローソク足の形や価格を見て反転と判断するのかまで具体化します。 途中でルールを変えた場合は、変更前と変更後の取引を分けて集計します。異なる条件の取引を合算すると、どの変更が結果に関係したかを追いにくくなります。 ## 次の足を見る前に判断する 手動モードでは、現在の足までを見て、注文するか見送るかを決めてから次へ進めます。注文する場合は、根拠と決済条件を先に記録します。 見送った場面にも「価格が水平線に届いていない」「決めた条件が揃っていない」などの理由を残します。あとで価格が上がっていても、当初の条件を満たしていなければルールどおりの見送りです。 ## 損益とルール違反を別々に記録する 取引ごとに、日時、通貨ペア、時間足、売買方向、エントリーと決済の価格、損益、判断の理由を残します。予定外に損切りを動かした場合などは、どこで計画から外れたかも書きます。 Formiqのトレード記録にはメモを追加できます。チャートを見直すときは、負けた取引だけでなく、ルールを破って利益が出た取引も確認します。利益が出たことだけでは、その操作が再現できるかは分かりません。 ## 勝率と平均損益を合わせて見る 勝率だけでは、取引全体の損益は分かりません。勝った取引の平均利益、負けた取引の平均損失、取引回数を合わせて確認します。利益の大半を少数の取引が占めていないかも見ます。 通貨ペアや時間帯で集計を分けると、利益と損失がどこに集中したかを確認できます。ただし、分けたあとの件数が少ない場合は、時間帯の良し悪しを断定せず、別の期間でも確認します。[時間帯別の変動幅](/ja/data)は値動きの比較資料であり、自分の売買ルールの利益を証明するものではありません。 ## 覚えていない相場で確認する 同じ区間を繰り返すと、次の値動きを覚えてしまうことがあります。操作の練習には使えますが、売買判断を評価する場合は、まだ見ていない期間を選びます。 ルールを変更したら、変更に使った期間とは別の期間で試し、損益とルールを守れた割合を分けて比べます。練習回数だけを増やしても、利益の出る手法になったとは判断できません。 ## Formiqで練習を始める [チャート](/chart)で通貨ペアと時間足を選び、リプレイを開始します。終了後はトレード記録を開き、判断の理由と結果を見直します。無料プランの利用範囲は[料金ページ](/ja/pricing)で確認できます。 リプレイの売買はシミュレーションです。実際の資金を失う状況で同じ判断ができるか、実際の注文が同じ価格で約定するかは、この練習だけでは確認できません。 --- ### FXバックテストのやり方：プログラミング不要で手法を検証 URL: https://formiq.jp/ja/blog/backtest-without-coding Language: ja Published: 2026-07-25 Updated: 2026-09-06 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: バックテスト, ノーコード, 戦略検証, 初心者 Also available in: en — https://formiq.jp/blog/backtest-without-coding Formiqのバックテストでは、インジケーターの条件を選択して売買ルールを作れます。通貨ペア、時間足、期間、売買コストを揃えて実行し、損益や個々の取引を確認します。 ## エントリーと決済を設定する たとえば、次の条件を入力します。これは設定方法を説明する例で、利益を確認した手法ではありません。 | 項目 | 設定例 | |---|---| | 買い条件 | EMA(9)がEMA(21)を上抜け、かつRSI(14)が40より大きい | | 損切り | 30pips | | 利確 | 60pips | 条件を選んだら、通貨ペアと時間足、対象期間、スプレッドを確認して実行します。同じルールを比較する場合は、変更する項目以外の条件を揃えます。 ## 結果画面で確認する項目 総損益に加えて、勝率、平均利益と平均損失、プロフィットファクター、最大ドローダウン、取引回数を確認します。勝率が高くても、負けたときの損失が大きければ総損益はマイナスになります。 プロフィットファクターは総利益を総損失の絶対値で割った値です。取引が少ない場合や、利益が一度の大きな値動きに集中している場合は、値が高くても安定性を判断できません。 ## 売買位置をチャートで見る 結果の取引をチャートで確認し、意図した条件でエントリー・決済しているかを見ます。損益の集計だけでは、条件の入力間違いや特定の期間への取引の集中を見落とすことがあります。 損切りと利確の両方に届く足など、ローソク足の中の値動きの順序が影響する取引にも注意します。バックテストの約定方法によって結果が変わるためです。 ## 設定を選ぶ期間と確認する期間を分ける 同じ期間でパラメータを変え続け、最も利益が出た設定だけを選ぶと、その期間に偶然合った条件を採用する可能性が高まります。設定を選んだあとに、変更せず別の期間で実行します。 [QQEの検証](/ja/blog/qqe-settings)では、ある年に選んだ設定を別の年にも当て、同じ設定がどの程度の損益を出したかを比較しています。別期間で損益が悪化した場合は、その結果も保存します。 ## 選択式で表現できる条件を確認する 手法ビルダーで作れるのは、用意された条件と決済方法の組み合わせです。独自の外部データや細かな注文管理が必要な場合は、条件一覧で表現できるかを確認します。近そうな条件に置き換えた場合は、元の手法を検証したことにはなりません。 ## 利用回数とデータの範囲 無料プランの実行上限は[料金ページ](/ja/pricing)に掲載しています。過去データの対象範囲は、チャートで選択できる期間を確認してください。 バックテストは過去の価格と約定条件から計算した結果です。将来の利益や実際の約定価格を保証するものではありません。[チャートで条件を設定する](/chart)。 --- ### iPhoneでFXを練習する方法：リプレイと記録の手順 URL: https://formiq.jp/ja/blog/forex-practice-iphone-free Language: ja Published: 2026-07-25 Updated: 2026-09-06 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: iPhone, 無料, スマホ, FX練習 Also available in: en — https://formiq.jp/blog/forex-practice-iphone-free iPhoneでは、FormiqのiOSアプリを使って過去チャートのリプレイ練習ができます。通貨ペアと時間足を選び、エントリー・決済の条件を決めてから開始します。 ## アプリを開いて練習する相場を選ぶ [App Store](https://apps.apple.com/jp/app/formiq-fx-backtest-tester/id6759964075)からアプリを入れ、ログインします。通貨ペア、時間足、過去の期間を選び、リプレイを開始してください。 はじめは足を手動で進め、注文と決済の操作を確認します。操作に慣れるために繰り返す区間と、売買判断を試すための未見の区間は分けて使います。 ## 価格と注文位置を読み取れる表示にする インジケーターを重ねたり、サブ画面を増やしたりすると、価格チャートに使える高さが小さくなります。ローソク足や注文位置が読みにくい場合は、表示を減らすかズームを調整します。 表示を減らすために売買条件まで変えると、別の手法の練習になります。判断に複数の指標が必要なら、PCやiPadなど画面の広い端末で確認します。 ## 注文前に理由を決めて、取引後に残す エントリー前に、どの条件を満たしたかと損切り・利確の位置を確認します。決済後は、予定した条件を守れたかをメモします。損益だけでは、ルールどおりの取引だったかは分かりません。 短い時間で終える場合も、振り返りの時間を残しておきます。[練習記録の見直し方](/ja/blog/1000-replay-trades)では、見送りも含めた記録方法を説明しています。 ## PCで履歴を見直す 同じアカウントを使うと、チャート設定やトレード記録をブラウザ版と共有できます。端末を切り替えたら、直前の記録が反映されているか確認してから振り返ります。 複数の指標や長い期間を並べて見る作業は、画面の広い端末で行うと価格と履歴を確認しやすくなります。アプリとブラウザ版で必要な機能が使えるかは、それぞれの画面で確認してください。 ## 無料の範囲を確認する 利用できる機能と回数の上限は[料金ページ](/ja/pricing)とアプリ内のプラン表示で確認できます。購入前に、練習に必要な表示と操作を試してください。 リプレイは過去データを使ったシミュレーションです。リアルタイムの注文や実際の資金の取引には使いません。 --- ### 無料のFX検証ソフトの選び方：機能と利用制限を確認 URL: https://formiq.jp/ja/blog/forex-tester-free Language: ja Published: 2026-07-25 Updated: 2026-09-06 Author: 斉木勇一 — https://formiq.jp/ja/about/yuichi-saiki Tags: FX検証ソフト, 無料, リプレイ練習, バックテスト Also available in: en — https://formiq.jp/blog/forex-tester-free 無料のFX検証ソフトを選ぶときは、リプレイで自分の売買判断を練習したいのか、決めたルールを自動でバックテストしたいのかを先に分けます。必要な機能と、無料で利用できる範囲が変わるためです。 ## リプレイとバックテストの用途を分ける リプレイでは、未来のローソク足を隠し、価格を進めながら自分でエントリーと決済を判断します。バックテストでは、決めた条件に従って過去の取引を計算します。 裁量判断を練習するなら、足を進める操作と売買記録を確認します。ルールを比較するなら、その条件を入力できるか、スプレッドや決済方法を指定できるかを確認します。 ## 無料で使える範囲を調べる | 確認項目 | 確かめること | |---|---| | 利用期限 | 期限付きの体験版か、継続利用できる無料プランか | | 過去データ | 必要な通貨ペア・時間足・期間が含まれるか | | 利用回数 | リプレイやバックテストに上限があるか | | 条件と表示 | 必要な売買条件とインジケーターを使えるか | | 記録 | 取引履歴を保存し、判断の理由をメモできるか | 「無料」という表示だけでは、練習に必要な範囲を使えるかは判断できません。各社の条件は[FX検証ツールの比較](/ja/compare/forex-backtesting-tools)から確認できます。 ## Formiqで過去チャートを動かす Formiqはブラウザで使えます。[チャート](/chart)を開いて通貨ペアと時間足を選び、リプレイを開始します。自動で再生する方法と、手動で足を進める方法があります。 注文前にエントリーと決済の条件を決め、終了後にトレード記録へ判断の理由を残します。操作と振り返りの手順は[リプレイ練習の進め方](/ja/blog/1000-replay-trades)で説明しています。 ## 条件を入力して検証する Formiqのバックテストでは、インジケーターの条件と損切り・利確を設定して実行します。結果の損益だけでなく、取引数と売買位置も確認します。 選択できる条件で自分のルールを表現できるかを先に確認してください。[ノーコードのバックテスト手順](/ja/blog/backtest-without-coding)に設定例を載せています。 ## 有料プランを検討する前に Formiqの無料プランには、インジケーターの同時表示数やバックテストの実行回数などに上限があります。[料金ページ](/ja/pricing)で現在の範囲を確認し、自分の練習に必要な操作を試してください。 使用するデータは過去の価格です。リプレイでの売買はシミュレーションで、実際の資金は動きません。 ---