Key takeaways
- Patterns do carry information, but not much. On daily charts the best of the twelve, the hammer, beat a matched random entry by 6.5 percentage points. Most beat it by under two.
- Raw win rate is close to meaningless. A hammer wins 48.8% and a hanging man 15.1%, but that gap is almost entirely the stop distance their shapes imply, not predictive power.
- After real spreads, almost nothing is profitable. Only 1 of 24 daily pattern-and-target combinations cleared its breakeven win rate. On 4-hour charts it was 0 of 24.
- The edges do not replicate. The hammer's 6.5-point daily edge falls to 0.6 points on 4-hour bars and stops being statistically significant.
- Statistically real is not the same as tradeable. Inside bars on 4-hour charts beat random across 99,295 signals with overwhelming significance — and still lose money after costs.
How this was measured
The method matters more than the numbers here, because the usual way of counting pattern win rates produces figures that cannot be compared to anything.
- Data. Dukascopy bid-side OHLC for 28 instruments — major and cross FX pairs, gold, silver, stock indices, and crypto — from 2010 to August 2026. 113,913 daily bars and 584,667 four-hour bars, 698,580 in total.
- Detection. Twelve patterns defined by objective arithmetic rules on open, high, low and close, with no discretion. Reversal patterns additionally require the prior five-bar trend to run the right way.
- The trade. Enter at the open of the bar after the pattern completes. Stop just beyond the pattern's extreme. Target at 1:1, then again at 1:2. Walk forward bar by bar for up to 20 bars until one level is touched.
- Ties go against you. When a single bar contains both the stop and the target we cannot know which came first, so it is scored a loss every time.
- Real costs. Spreads come from our own matched bid/ask measurement of the same feed — a real median, per instrument, not a round number. Longs buy the ask and exit on the bid; shorts sell the bid and cover at the ask.
- The control. This is the part that is usually missing. For every real signal we generate a control trade with the same direction and the same stop distance as a fraction of price, but at a randomly chosen bar — five control draws per signal, so the baseline is not itself noise. Whatever the pattern knows that random timing does not is the gap between the two columns.
Why the control changes everything
Run the naive version of this test and the hammer looks extraordinary while the hanging man looks like a disaster. The two patterns are the same shape. The only difference is which way you trade them.
A hammer has a long lower wick. Go long, put the stop under that wick, and your stop is far away, so ordinary noise rarely reaches it and the win rate is high. A hanging man is that identical candle in an uptrend, traded short with the stop just above a very short upper wick — a tight stop that noise reaches constantly. Hammer wins 48.8%, hanging man 15.1%, and virtually none of that gap is the candle telling you anything.
The matched control absorbs exactly that effect. It gives the hanging man a random entry with the same tight stop, which also wins about 15%. The remaining difference — nothing at all — is the pattern's real contribution.
Daily charts: 1:1 target
Breakeven at a 1:1 target is a 50% win rate. Anything below that loses money however good the edge column looks.
| Pattern | Signals | Win rate | Matched random | Edge | Significance | Beats breakeven? |
|---|---|---|---|---|---|---|
| Hammer | 729 | 48.8% | 42.4% | +6.5 pts | ** | No |
| Bullish Pin Bar | 6,143 | 47.0% | 43.8% | +3.2 pts | *** | No |
| Evening Star | 4,937 | 44.8% | 42.9% | +2.0 pts | * | No |
| Bullish Engulfing | 7,768 | 47.9% | 46.1% | +1.8 pts | ** | No |
| Morning Star | 5,059 | 47.7% | 46.0% | +1.7 pts | * | No |
| Doji | 5,770 | 37.6% | 36.1% | +1.6 pts | * | No |
| Bearish Engulfing | 8,918 | 45.3% | 44.2% | +1.1 pts | * | No |
| Bearish Pin Bar | 5,372 | 39.7% | 38.8% | +1.0 pts | not significant | No |
| Shooting Star | 734 | 39.1% | 38.2% | +0.9 pts | not significant | No |
| Inverted Hammer | 555 | 22.0% | 21.3% | +0.7 pts | not significant | No |
| Inside Bar | 20,759 | 30.9% | 30.4% | +0.5 pts | not significant | No |
| Hanging Man | 669 | 15.1% | 15.1% | +0.0 pts | not significant | No |
Daily charts: 1:2 target
Breakeven at 1:2 is 33.3%.
| Pattern | Signals | Win rate | Matched random | Edge | Significance | Beats breakeven? |
|---|---|---|---|---|---|---|
| Hammer | 712 | 34.6% | 29.7% | +4.8 pts | * | Yes |
| Shooting Star | 724 | 27.9% | 25.2% | +2.7 pts | not significant | No |
| Bullish Pin Bar | 6,027 | 33.2% | 30.7% | +2.4 pts | *** | No |
| Inverted Hammer | 555 | 18.9% | 16.6% | +2.3 pts | not significant | No |
| Morning Star | 4,839 | 32.4% | 30.8% | +1.6 pts | * | No |
| Bearish Engulfing | 8,577 | 30.3% | 28.8% | +1.6 pts | ** | No |
| Evening Star | 4,756 | 29.8% | 28.3% | +1.6 pts | * | No |
| Hanging Man | 669 | 13.0% | 12.0% | +1.0 pts | not significant | No |
| Bullish Engulfing | 7,358 | 32.2% | 31.3% | +0.9 pts | not significant | No |
| Bearish Pin Bar | 5,316 | 27.8% | 27.0% | +0.8 pts | not significant | No |
| Doji | 5,740 | 27.8% | 27.1% | +0.7 pts | not significant | No |
| Inside Bar | 20,708 | 24.1% | 23.6% | +0.5 pts | not significant | No |
Four-hour charts: the edges survive, the profits do not
The 4-hour sample is five times larger, so the statistics get much sharper — and much less flattering. Inside bars beat their control by 2.0 points across 99,295 signals, a result you would see by chance essentially never. It is still a losing trade at 32.4% against a 50% breakeven.
| Pattern | Signals | Win rate | Matched random | Edge | Significance | Beats breakeven? |
|---|---|---|---|---|---|---|
| Inverted Hammer | 3,529 | 20.3% | 18.1% | +2.2 pts | ** | No |
| Inside Bar | 99,295 | 32.4% | 30.4% | +2.0 pts | *** | No |
| Bullish Engulfing | 43,548 | 44.1% | 42.6% | +1.4 pts | *** | No |
| Morning Star | 23,808 | 43.2% | 41.8% | +1.4 pts | *** | No |
| Doji | 32,553 | 34.6% | 33.6% | +0.9 pts | ** | No |
| Bearish Pin Bar | 29,282 | 40.3% | 39.4% | +0.9 pts | ** | No |
| Bullish Pin Bar | 38,203 | 41.9% | 41.1% | +0.8 pts | ** | No |
| Hanging Man | 4,866 | 13.4% | 12.7% | +0.7 pts | not significant | No |
| Evening Star | 24,092 | 40.7% | 40.0% | +0.6 pts | not significant | No |
| Bearish Engulfing | 44,730 | 41.9% | 41.2% | +0.6 pts | * | No |
| Hammer | 4,918 | 42.2% | 41.6% | +0.6 pts | not significant | No |
| Shooting Star | 4,080 | 39.0% | 39.7% | -0.7 pts | not significant | No |
At a 1:2 target on 4-hour bars the picture is the same: real edges, no profits.
| Pattern | Signals | Win rate | Matched random | Edge | Significance | Beats breakeven? |
|---|---|---|---|---|---|---|
| Inverted Hammer | 3,527 | 16.7% | 14.7% | +2.0 pts | ** | No |
| Bullish Engulfing | 41,586 | 29.4% | 28.3% | +1.1 pts | *** | No |
| Morning Star | 22,915 | 29.3% | 28.2% | +1.1 pts | *** | No |
| Inside Bar | 98,901 | 23.5% | 22.5% | +1.0 pts | *** | No |
| Bearish Pin Bar | 28,912 | 27.8% | 26.8% | +1.0 pts | *** | No |
| Evening Star | 23,398 | 27.4% | 26.6% | +0.8 pts | * | No |
| Bearish Engulfing | 43,071 | 27.7% | 27.2% | +0.6 pts | * | No |
| Doji | 32,309 | 24.9% | 24.4% | +0.5 pts | not significant | No |
| Hanging Man | 4,866 | 10.6% | 10.3% | +0.4 pts | not significant | No |
| Bullish Pin Bar | 37,334 | 28.2% | 27.9% | +0.3 pts | not significant | No |
| Hammer | 4,775 | 27.9% | 28.1% | -0.2 pts | not significant | No |
| Shooting Star | 4,018 | 26.0% | 27.1% | -1.1 pts | not significant | No |
What this means if you trade patterns
The honest reading is not that candlestick patterns are worthless. They are not — a dozen of them beat a matched random entry, several with overwhelming statistical significance, and that is a genuine finding about market microstructure. The problem is the size of what they know.
- A pattern is not a strategy. An edge of one to six points on entry timing cannot carry a system by itself. It has to sit inside something that already has a reason to be in the trade.
- Your exits matter more than your entries. Moving from a 1:1 to a 1:2 target changed outcomes far more than swapping one pattern for another ever did.
- Costs are the same size as the edge. That is the whole game on lower timeframes. A one-point pattern edge is simply not visible underneath a spread.
- Test on your own instrument and timeframe. The hammer result on daily bars did not survive the move to 4-hour bars. Nothing here transfers automatically.
Limitations
Stated plainly, because a study without them is marketing.
- Bid-side data. Ask prices are reconstructed from a measured median spread rather than a tick-level ask feed, so cost is modelled rather than replayed.
- One entry rule, one stop rule, two targets. A pattern that fails at a fixed 1:1 might work with a trailing stop, a time-based exit, or a filter this test does not apply.
- Same-bar ties scored as losses. Deliberately conservative, and it pushes every win rate down slightly.
- Trades unresolved after 20 bars are excluded rather than closed at market. That is at most 6.9% of any row.
- Spread only — no commission and no swap. Adding those moves every row further into the red, not out of it.
Download this dataset
Every figure on this page as raw CSV and JSON, regenerated whenever the archive updates. Free for any use, including commercial, with attribution to FxBacktest.
- daily-charts-1-1-target.csvDaily charts: 1:1 target
- daily-charts-1-2-target.csvDaily charts: 1:2 target
- four-hour-charts-the-edges-survive.csvFour-hour charts: the edges survive, the profits do not
- four-hour-charts-the-edges-survive-2.csvFour-hour charts: the edges survive, the profits do not
- candlestick-pattern-win-rates.jsonAll tables plus metadata, one file
Cite or republish this data
This research is free to quote, screenshot, or republish — in an article, a video, a newsletter, or a course — as long as you credit FxBacktest with a link back to this page. Copy the attribution below.
Frequently asked
Do candlestick patterns actually work?
Barely, and by far less than most trading courses claim. Measured across 420,317 signals on 28 instruments, the strongest pattern on daily charts (the hammer) won 48.8% of 1:1 trades against 42.4% for a matched random entry - a real 6.5 point edge, but still below the 50% needed to break even at 1:1. On 4-hour charts none of the twelve was profitable after spreads at either 1:1 or 1:2.
Why compare patterns to a random entry instead of to 50%?
Because raw win rate is mostly decided by where the stop sits, not by the pattern. A hammer has a long lower wick, so a long entry stopped below it has a wide stop and a naturally high win rate. A hanging man is the same shape traded short, so its stop is tight and its win rate is naturally low. Comparing each to a random entry with the same direction and the same stop distance removes that artifact and leaves only the pattern's predictive content.
Which candlestick pattern has the highest win rate?
On daily charts the hammer won 48.8% at 1:1, the highest of the twelve tested - but that number is inflated by its wide stop. Against a matched random entry the genuine edge is 6.5 points. On 4-hour charts the hammer's edge falls to 0.6 points and stops being statistically significant, so the daily result does not replicate.
Are candlestick patterns profitable after spreads?
Almost never in this test. At 1:1 you need to win more than 50% of trades; at 1:2 more than 33.3%. With real measured spreads applied, 1 of 24 daily pattern-and-target combinations cleared breakeven, and 0 of 24 on 4-hour charts.
What data was used?
Dukascopy bid-side OHLC for 28 instruments - major and cross FX pairs, gold, silver, stock indices, and crypto - covering 2010 to August 2026. That is 113,913 daily bars and 584,667 four-hour bars, 698,580 in total. Spreads come from our own matched bid/ask measurement of the same feed.