All posts
Williams %Rindicatorswin ratebacktestingUSDJPY

Williams %R won 68% of trades and lost money

240 Williams %R settings backtested on USD/JPY across 2025 and 2024: win rates above 60% with negative results, and the win-to-loss size gap behind it.

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.

−1500−1000−5000+50050%55%60%65%70%win rate →net pips for the year
the classic 14, −80 / −20highest win rate, 67.0%one dot per setting, 120 of them — 10 finished positive
Win rates run from 49% to 67% and almost the whole cloud sits below the zero line. Winning more often did not mean making money.

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 winning often still loses

The arithmetic is visible.

average pips per trade0100200H1, 2025 · 14, −80/−2058.1% · -1083pH1, 2024 · 14, −80/−2068.4% · -569pH1, 2025 · highest win rate67% · -954pH4, 2024 · 14, −80/−2079.4% · +424p
average winaverage losswin rate and net result in brackets
A 79% win rate made 424 pips because the losing fifth of trades were three times the size of the winners.
SettingWin rateAverage winAverage lossNet for the year
H1, 2025 · 14, −80/−2058.11%+39.8 pips−72.7 pips−1,082.8 pips
H1, 2024 · 14, −80/−2068.42%+44.9 pips−110.8 pips−569.3 pips
H1, 2025 · highest win rate67.05%+28.3 pips−74.1 pips−953.7 pips
H4, 2024 · 14, −80/−2079.41%+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.

That is not bad luck; it is the shape of the strategy. Buy oversold, close on a bounce to −20: the gain is capped at "how far it bounced." When the oversold reading keeps going, no exit arrives — the trade either eventually bounces or runs. Small wins, frequent; large losses, occasional.

The high win rate is a consequence of that shape, not evidence of an edge.

None of the 120 hourly settings in 2025 reached a 70% win rate (the maximum was 67.05%), and that 67.05% setting ranked 86th of 120 on profit.

Adding Williams %R to a chart

PlatformHow to add it
MT4 / MT5Built in — Navigator, Oscillators, Williams' Percent Range
TradingViewIn 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. So the two directions are separate systems, tested separately.

ItemValue
PairUSD/JPY
Period2025-01-01 to 2025-12-31 (2024 run identically for comparison)
Timeframes15-minute / 1-hour / 4-hour
Bars tested6,226 hourly bars in 2025; 24,903 on 15-minute, 1,610 on 4-hour
Long sideBuy when %R falls below the entry level (−95 to −70), close when it rises above the exit level (−50 to −10)
Short sideThe mirror of that on the 0 side (sell at −5 to −30, close at −50 to −90)
Stops and targetsNone, except in the section that measures them
Costs0.3 pip spread, zero slippage, 0.1 lot
Combinations6 periods (7–40) × 5 entry levels × 4 exit levels × 2 directions = 240, per timeframe and per year
MethodExecuted in Formiq's backtester; pips recomputed from each trade's fill prices

By timeframe and direction

Timeframe · sideProfitable 20252025 medianProfitable 20242024 median2025 median win
15-minute long3 / 120−661 pips54 / 120−226 pips64.3%
15-minute short7 / 120−777 pips2 / 120−1,170 pips63.3%
1-hour long10 / 120−720 pips47 / 120−201 pips59.0%
1-hour short10 / 120−633 pips0 / 120−1,442 pips59.3%
4-hour long89 / 120+572 pips71 / 120+123 pips64.1%
4-hour short78 / 120+333 pips0 / 120−1,396 pips57.6%

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.

Even there, the rank correlation between the two years is −0.611. The settings that led one year trailed the next.

Inside the four-hour long side

Period2025 average2025 win2024 average2024 win
7−83 pips57.1%+549 pips73.3%
9−52 pips56.7%+520 pips77.3%
14+203 pips58.9%+197 pips74.1%
21+711 pips66.8%−138 pips66.4%
28+1,143 pips70.7%−215 pips65.1%
40+935 pips76.5%−148 pips63.4%

The period works in opposite directions in the two years — long periods in 2025, short ones in 2024. Only period 14 was positive in both (+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 (H1 long, 14, −80/−20)20252024
No stop148 trades, 58.11% win, −1,082.8 pips133 trades, 68.42% win, −569.3 pips
Stop 30 / target 60376 trades, 33.51% win, −1,547.6 pips297 trades, 37.04% win, −702.6 pips
Stop 50 / target 100281 trades, 43.77% win, −1,813.9 pips242 trades, 45.45% win, −921.5 pips
Stop 20 / target 40476 trades, 29.62% win, −1,399.6 pips370 trades, 32.43% win, −760.5 pips
Stop 50 / target 50304 trades, 49.67% win, −1,271.1 pips260 trades, 50.77% win, −920.7 pips
Time exit after 24 bars184 trades, 58.15% win, −989.6 pips152 trades, 65.13% win, −729.6 pips

Every one lost, and every one lost more than doing nothing.

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.

Cost is not the problem here

SpreadH1 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, but 148 trades a year only moves the total by 296 pips across the whole range. What is losing here is the structure.

What helped on four-hour bars

Narrowing the one cell that worked:

Condition20252024
Period 14, −80/−2036 trades, 61.11% win, +168.1 pips34 trades, 79.41% win, +423.5 pips
Period 28, −80/−2024 trades, 70.83% win, +912.1 pips15 trades, 60.00% win, −785.0 pips
London/NY hours only36 trades, 66.67% win, +609.5 pips28 trades, 75.00% win, +16.8 pips
Tokyo hours only30 trades, 63.33% win, +363.0 pips23 trades, 69.57% win, −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 — on 15 to 36 trades a year, which is thin evidence in itself.

What this test supports

  1. The win rate comes out high and does not translate into profit. Median 59.0% on hourly bars in 2025, with 10 of 120 settings positive
  2. The cause is the size gap. Average wins under half the average losses; a 79.41% win rate produced 423.5 pips
  3. A stop does not repair it. All five variants did worse than no stop, in both years
  4. Counter-trend results depend on the trend. In 2024's rising market, zero of 120 short-side settings made money
  5. Only the four-hour long side was usable — 40 of 120 positive in both years, and even there the rank correlation is −0.611

Used as one input alongside a separate judgement about direction, rather than as a standalone system, is what these numbers support. The same method applied to QQE, Aroon, RVI and Supertrend — where the win rates come out the other way round. Bollinger Bands measure the same inversion inside a single indicator, fading the band against following it.

Limits of this test

  • One pair, two years
  • 2024 rose 1,632 pips; 2025 finished 56 pips lower inside a 1,900-pip range. For a counter-trend indicator that difference is decisive, and the short side's shutout is a direct reflection of what 2024 did
  • 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
  • Four-hour settings trade 9 to 92 times a year — thin for reading win rates
  • Volatility varies by hour of the day, which the session-filter rows are exposed to

Questions people ask

What are the best Williams %R settings?
Of the 120 long-side settings on USD/JPY hourly bars in 2025, only 10 finished positive. The best was period 40, entering at −95 and exiting at −30 (50 trades, +499.1 pips), which ranked 22nd of 120 in 2024. The textbook setup — period 14, buy below −80, exit above −20 — lost 1,082.8 pips in 2025 and 569.3 pips in 2024.
What win rate does Williams %R produce?
A high one. Across the 120 long-side hourly settings in 2025 the median win rate was 59.0%, ranging from 48.57% to 67.05%. On four-hour bars the median was 64.1%, and the textbook setup reached 79.41% in 2024. The results over the same settings were still negative.
Why does a high win rate still lose money?
Because the losses are larger than the wins. The textbook hourly setup won 68.42% of its trades in 2024, but the average win was 44.86 pips against an average loss of 110.75 — a 2.5-fold gap. Buying oversold and closing on a small bounce caps the upside; the trades that never bounce are the ones that run.
Which timeframe works for Williams %R?
Only the four-hour long side in this test: 89 of 120 settings profitable in 2025 and 71 in 2024. Hourly managed 10 in 2025 and 15-minute managed 3. Even the four-hour cell had a rank correlation of −0.611 between the years, so the settings that led one year trailed the next.
Is Williams %R built into MetaTrader?
Yes. Both MT4 and MT5 ship it as Williams' Percent Range under Oscillators. TradingView lists it as "Williams %R". Formiq's charts include it with an adjustable period.

Formiq is a free browser-based FX terminal with replay practice and no-code backtesting. Open the chart or see what the free plan includes.