Heikin Ashi Strategy: 960 Tests and a Price That Does Not Exist

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.

priced at the real close (−523 pips)priced at the Heikin Ashi close (+34,120 pips)USD/JPY 2025, 15-minute, colour flip, 6,107 trades, zero spread

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:

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:

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

This is the sharpest result in the article.

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

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

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.

The same trades, priced two ways

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.

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

The mechanism is plain. Every trade fills twice, and each fill collects the gap between the Heikin Ashi close and the real one. As the previous section showed, that gap leans cheap for buys and dear for sells.

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.

The smoothing is real

Heikin Ashi is sold on cutting noise and holding a trend. Measured, that part is true.

Heikin Ashi (1,548 runs)ordinary candles (3,185 runs)USD/JPY 2025, hourly bars

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:

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

From here on, fills come from the real candles. Standard variant throughout.

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 — on a system that 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). That is the twenty-first consecutive article with the same shape: the reversal reading wins more often and finishes behind.

What survived both years

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.

The run-length ladder

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).

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.

4-hour bars go the other way — waiting three bars or more drops to 0/8. Whether patience pays is a property of the timeframe, not of the reading.

The four calculation variants

Standard, better (average two bars, then transform), smoothed and double smoothed. What changes first:

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):

Smoothed (3) was the only variant in the top group in both years. Smoothed (10) made +267 in 2025 and +3,200 in 2024. On 15-minute bars only double smoothed (10) cleared both years (+1,706 / +1,697), and it does that by cutting the trades to 1,198.

Out of sample: last year's best, applied to this year

Two of three landed in the bottom fifth. Only the 4-hour pick beat its median. Hourly bars have 50 of 80 settings ahead in both years, and picking last year's best out of that pool still drew 71st of 80 — the same result twenty previous articles have found.

Rank correlations across all 80:

Only 15-minute bars show a positive correlation, and 51 of its 80 settings lost money in 2025 — so what is stable there is the ordering of the losses.

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: same shape, no transform. Compared across 48 cells of three timeframes, two years and eight variants.

M15H1H4above the line = the transform won (31 of 48)

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

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) pointed opposite ways in the two years: the hourly colour flip fell from +1,153.3 to +646.5 in 2025 and rose from +101.3 to +1,558.7 in 2024. London and New York hours hurt in both (+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 +1,047.2 in 2025 (baseline +1,153.3) and +15.3 in 2024 (baseline +101.3). Nothing here counts as an improvement. A 30-pip stop with a 60-pip target dropped 2025 to −249.6.

Cost

Net pips were exactly linear in the spread — pips lost = trades × spread held without error on all five settings.

The readings that trade less break even later. The 15-minute colour flip loses at zero spread, so cost is not its problem.

What this test supports

1. The Heikin Ashi close is not a fill price. It sits a median 2.22 to 9.26 pips from the real close and, on green bars, lands on the cheap side 63.7% to 67.3% of the time. Pricing the same 6,107 trades both ways gives −523.1 and +34,120 pips. Across 36 cells the Heikin Ashi price was kinder in 35, and turned a loss into a profit in 15
2. The invented profit is roughly twice the gap, times the trade count. Once at each end of every trade — measured at 0.86 to 0.98 times that
3. The smoothing is real. One colour lasts 1.95 bars on ordinary candles and 4.02 on Heikin Ashi, and colour changes halve. All six cells agree
4. The colour is a moving average crossover. (O+H+L+C)/4 against its own 3-period EMA, with zero disagreements across six cells. Run on ordinary candles, the transform cuts trades in 48 of 48 and lifts the win rate in 46, and finishes ahead on net pips in 31
5. With honest fills, only hourly bars worked. 50 of 80 settings ahead in both years, against 16 on 15-minute and 23 on 4-hour

Related reading

• 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 — where the zero line turned out to be a moving average crossover
• Moving average cross settings — where this article's control ends up
• GMMA line count — the first "a condition that holds 85% of the time separates nothing"
• Renko chart settings — the second indicator here that rewrites the candles. Its close is a real price the bar has already left

Limits of this test

• 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
• 2024 rose 1,632 pips and 2025 fell 56 — completely different market characters. Hourly bars surviving both is two years, one of each
• 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