Bottom line: on ES (S&P 500 futures) no candlestick pattern makes money traded on its own. Intraday reversal patterns (Brooks reversal bars, hammer, shooting star, engulfing) win 2–4 percentage points more than the same stop placed anyway, which is statistically real, but their 1R win rate is still only 47%–51% and slippage and fees take it away. Daily samples are too small for anything to be significant. Candles are for reading the market and placing stops; what decides the result is location (key levels, range edges) and filters.
1. Classic candlestick patterns (continuous ES, 2010–2026)
Data: continuous ES 1-minute bars (June 2010 to June 2026) built into 5-minute, 15-minute and 1-hour bars, plus daily bars. Rules: enter at the next bar's open after the pattern closes; stop one tick beyond the pattern's extreme; target 1R (1:1); whichever comes first within 20 bars, a bar touching both counts as the stop; no costs. Baseline: the same trade on every bar with the stop beyond the last 1–3 bars, grouped into 10 buckets of stop size / ATR, taken at the bucket of each pattern's own stop: what that stop would have won without the pattern. Each cell: win rate (share of trades that hit the target first); in brackets, the pattern's win rate minus the baseline's, in percentage points (negative = worse than the baseline); bold = that difference is more than 3 times its own standard error, i.e. statistically significant rather than luck, though not necessarily large. Hover a cell for the sample count.
| Pattern | Side | 5 min | 15 min | 1 hour | Daily |
|---|---|---|---|---|---|
| Bull reversal bar (Brooks, after a decline) | Long | 49.6% (+2.1) | 49.8% (+2.4) | 49.7% (+2.0) | 48.6% (-1.2) |
| Bear reversal bar (Brooks, after a rally) | Short | 48.9% (+2.2) | 48.0% (+2.4) | 47.5% (+3.3) | 42.1% (+2.5) |
| Hammer (after a decline) | Long | 49.9% (+2.8) | 49.8% (+3.4) | 50.4% (+4.0) | 46.6% (-2.2) |
| Shooting star (after a rally) | Short | 49.0% (+1.4) | 48.0% (+2.2) | 46.9% (+3.7) | 41.6% (+4.2) |
| Bullish engulfing | Long | 50.4% (+1.5) | 50.9% (+1.8) | 50.7% (+1.5) | 53.2% (-0.3) |
| Bearish engulfing | Short | 49.5% (+1.5) | 48.3% (+1.0) | 47.9% (+1.7) | 47.2% (+3.1) |
| Piercing line | Long | 49.9% (+1.0) | 50.2% (+1.4) | 50.4% (+1.2) | – |
| Dark cloud cover | Short | 50.2% (+2.3) | 48.4% (+1.5) | 47.6% (+2.0) | 50.0% (+6.4) |
| Morning star | Long | 50.8% (+1.3) | 51.3% (+1.5) | 52.0% (+2.3) | 54.8% (+0.1) |
| Evening star | Short | 49.4% (+0.8) | 48.9% (+1.0) | 50.5% (+3.5) | 50.0% (+4.2) |
| Bull outside bar | Long | 48.9% (+1.1) | 48.9% (+0.5) | 49.3% (-0.1) | 47.4% (-5.3) |
| Bear outside bar | Short | 48.8% (+1.8) | 48.0% (+1.0) | 46.0% (-0.8) | 44.8% (+1.3) |
| Bull trend bar (Brooks, big body) | Long | 49.8% (+2.2) | 50.5% (+1.5) | 51.9% (+0.7) | 49.4% (-4.0) |
| Bear trend bar (Brooks, big body) | Short | 49.6% (+2.8) | 49.3% (+1.7) | 49.8% (+0.8) | 40.7% (-1.4) |
| Three white soldiers | Long | 50.2% (+0.8) | 49.9% (-0.2) | 49.0% (-1.2) | 50.5% (-3.8) |
| Three black crows | Short | 50.0% (+1.5) | 50.6% (+1.7) | 49.6% (+0.5) | 44.7% (+4.2) |
| Doji after a decline | Long | 45.7% (+1.3) | 44.8% (+2.4) | 43.3% (+2.1) | 36.6% (-4.9) |
| Doji after a rally | Short | 44.9% (+1.6) | 42.6% (+2.2) | 39.2% (+2.0) | 25.8% (-4.0) |
| Inside bar, buy over it | Long | 41.0% (-0.1) | 40.7% (+1.2) | 40.3% (+1.6) | 27.8% (-1.6) |
- Intraday, every pattern's 1R win rate is between 39% and 52%: about a coin flip.
- Against the same stop placed anyway, reversal patterns add 2–4 points (hammer 1h +4.0, Brooks bear reversal bar 1h +3.3) over tens of thousands of cases, which is reliable but too small: the median 5-minute hammer stop is 1.25 points, so a 0.5-point cost is 0.4R per trade.
- Dojis and inside bars have the lowest win rates (about 40%–46%): the stop is so close that it is hit first.
- Daily patterns occur tens to a few hundred times each and none of the lifts exceeds 3 standard errors; morning star 54.8% and bullish engulfing 53.2% look high on 33 and 179 cases.
2. Reading failed breakdowns the Al Brooks way (ES 5-second bars, 2024-01 to 2026-05)
After a flush through a major level (prior-day or overnight high / low), the bars of 1, 2, 3, 5 and 15 minutes were read the Brooks way: the breakout bar and its follow-through, the reversal (signal) bar, a stop entry beyond it, second entries, the next timeframe up. Target 1.5R, stop beyond the flush extreme, 0.5 point cost per trade, 4 hours to play out.
| Read | Trades | 1.5R reached | Mean R per trade (after costs) |
|---|---|---|---|
| FBD confirmation (100 s held) | 5,238 | 31.7% | -0.055 |
| Bar entry, all timeframes | 19,273 | 34.9% | -0.114 |
| Bar entry 1m | 5,299 | 37.0% | -0.142 |
| Bar entry 2m | 4,404 | 36.6% | -0.111 |
| Bar entry 3m | 3,915 | 35.5% | -0.103 |
| Bar entry 5m | 3,374 | 33.8% | -0.103 |
| Bar entry 15m | 2,281 | 27.6% | -0.093 |
| Signal bar strong | 12,491 | 34.7% | -0.090 |
| Signal bar ordinary | 6,782 | 35.4% | -0.158 |
| Signal bar closed back through the level | 4,714 | 32.3% | -0.136 |
| Signal bar not back yet | 14,559 | 35.8% | -0.107 |
| Breakout bar strong + follow-through | 3,046 | 35.1% | -0.093 |
| Breakout bar failing | 8,446 | 34.1% | -0.147 |
| No follow-through | 9,968 | 35.1% | -0.111 |
| Second entry | 14,236 | 34.5% | -0.107 |
| First entry | 5,037 | 36.2% | -0.134 |
| Next timeframe agrees | 5,055 | 35.5% | -0.125 |
| Big signal bar | 1,773 | 28.0% | -0.070 |
| FBD at a poor low / high | 440 | 29.5% | -0.153 |
| FBD, 1m bounce already running | 1,009 | 37.2% | -0.156 |
| FBD, neither | 3,789 | 30.5% | -0.016 |
| PA break and retest | 834 | 32.6% | -0.054 |
| Range-day H/L at the edge | 1,195 | 31.5% | -0.089 |
- No timeframe's bar entry beat the plain confirmation (back above the level and held 100 seconds); all were negative after costs, and the A / B / C grades did not rank the outcomes.
- The one read that held in both halves of the data: a strong signal bar beat an ordinary one by 0.05–0.08R, still negative.
- What helped were two filters: at a poor low / high (Market Profile: two or more 30-minute periods at the extreme, no tail) the failed breakdown made -0.15R a trade; when the 1-minute bounce had already run at the confirmation (a late entry) -0.16R; neither, -0.02R, about break-even.
- Gold (9 months of 2025–2026) agreed: no timeframe's bar entry was clearly positive.
For research only, not investment advice. The replays used automatic levels only, not hand-drawn ones; other markets and other levels may differ.