Edges
Chart Pattern Win Rates: Real Accuracy Measured Across Timeframes
The headline you see on every trading site — "Bullish Engulfing: 72% win rate" — is almost certainly wrong. Not because the pattern is unreliable, but because that number was measured on daily data, in a different market regime, and then applied uncritically to intraday charts where it performs near random. Oyamori measures chart pattern win rates against real price data across 103 patterns and multiple timeframes. What the data shows challenges most of the conventional wisdom.
What Chart Pattern Win Rate Actually Means
A pattern's win rate is the percentage of times price moved in the pattern's predicted direction on the very next bar after the pattern appeared.
The definition is strict by design:
- Bullish pattern at bar N → win if the next bar closes higher
- Bearish pattern at bar N → win if the next bar closes lower
- Neutral patterns (doji, spinning top without context) → excluded entirely — no directional claim
This 1-bar forward definition is the standard for rigorous pattern research. A bullish pattern that takes three sessions to "eventually" work is not reliably predictive — it is noise with a delayed story attached. If a pattern has genuine edge, the next bar should reflect it.
50% is the coin-flip baseline. Any pattern consistently measuring below 55% across a large sample is effectively random noise at that timeframe, regardless of how it looks on a chart.
Why Most Published Win Rate Numbers Are Wrong
The 70–80% figures cited for classic candlestick patterns typically come from:
- Studies run on daily data, then applied to intraday charts where the same pattern behaves very differently
- Small sample sizes — 20 to 50 occurrences across a few months rarely produces statistically reliable estimates
- Bull market datasets where bullish patterns appeared reliable because the market was trending up, not because the patterns themselves provided edge
- No control for session time, volume regime, or broader trend context
The mismatch between cited win rates and intraday reality is dramatic. Consider bull engulfing — one of the most widely cited patterns, often credited with a 65–72% win rate. Across real 15-minute bars, Oyamori measured 48% — near random. The same pattern at the daily timeframe: 50%. The textbook number was never true at 15m. It was a daily number misapplied downward.
This matters because most retail traders work from 15m and 1h charts. A pattern reference listing "72% win rate" for a pattern that actually delivers 48% intraday gives traders false confidence in signals that have no consistent edge at their actual timeframe.
Oyamori's Measurement Methodology
Oyamori builds and maintains its win rate data using the following approach.
RTH bars only. All measurements use Regular Trading Hours bars. After-hours data is excluded because pattern behavior during pre-market and overnight sessions is structurally different — thinner liquidity, gap fills, and institutional positioning create a population that does not represent how patterns behave when a trader would actually act on them.
Multiple instruments. Measurements are taken across several instruments covering different volatility regimes — high-volume growth names, large-cap liquid instruments, and broad market ETF data. This prevents a single trending stock from inflating the results for bullish patterns.
Minimum 10 sample threshold. A pattern at a given timeframe must appear at least 10 times before a win rate is recorded. Below that threshold, the entry is flagged and the prior estimate is retained. This prevents small-sample flukes from entering the dataset.
Quarterly refresh. Win rates are updated every quarter, or sooner after a significant regime shift:
- Bull market transitioning to sideways or bear conditions
- After new patterns are added to the detection library
- When signal confidence outputs begin feeling systematically misaligned with actual outcomes
The quarterly cadence matters because reliability shifts with market regime. Bearish patterns perform differently in a sustained bull trend than in a choppy or declining environment. Refreshing from raw data ensures the numbers reflect current structure rather than a snapshot from a different market era.
The Timeframe Effect: The Most Important Finding
The same pattern produces measurably different win rates depending on which timeframe you trade. The magnitude of this difference is the most important thing to understand about pattern win rates.
| Timeframe | Typical WR range | Why |
|---|---|---|
| 15m | 42–65% | Bar-to-bar noise dominates; macro structure not visible in individual bars |
| 1h | 50–72% | Enough resolution to reflect meaningful price structure |
| 4h | 55–75% | Multi-session context visible; fewer false-positive structures |
| Daily | 55–83% | Full sessions, institutional participation, macro structure dominant |
Typical win rate degradation moving one timeframe step lower:
- Volume patterns: approximately −3% per step down
- Single-bar candlesticks: approximately −5% per step
- Multi-bar candlesticks: approximately −6% per step
- Chart and structure patterns: not meaningful below 1h — too few bars form the required structure
- EMA cloud signals: not meaningful below 5m — crossovers at 1m frequency become near-random
What Oyamori's Data Shows: Notable Findings
The most instructive findings are the cases where measured win rates diverged significantly from the numbers most references cite — both lower and higher.
Patterns That Underperform Their Reputation
Double Top at daily delivering 43% is the most striking result. The pattern is real — it correctly identifies structural resistance. But as a 1-bar forward predictor, it is no better than a coin flip. This makes sense in hindsight: price commonly tests a double top level multiple times before committing in either direction. The pattern marks a zone, not a timing signal.
Bear Engulfing at 4h measuring 25% reflects market regime, not pattern failure. In a sustained bull market, bearish reversal signals fire frequently and resolve upward just as frequently. Regime matters.
Patterns That Beat Expectations
Hammer at 15m measuring 64% is consistently underestimated in most pattern references, which treat it primarily as a daily reversal signal. The intraday version, particularly after a short downmove within session, carries genuine edge.
Bull Counterattack at 70% appears in few mainstream references but shows up reliably in the data. It requires a specific two-bar structure — a strong down bar followed by a recovery to approximately the same close — that identifies exhausted selling.
Win Rate as a Tie-Breaker, Not a Gate
The most practical takeaway from Oyamori's measurement work is this: at 15m, win rate is a context-dependent quality score, not a standalone signal.
Most 15m patterns produce measured WRs between 42% and 54%. This does not make them useless — it means they require supporting context to produce consistent edge. On their own, they are statistical noise. With supporting context, the same patterns can be highly actionable.
Oyamori charts display a confluence score (0–7) alongside each pattern detection, accounting for:
- EMA cloud alignment and trend direction
- VWAP position (above or below)
- Pattern quality tier
- Session timing (early vs. late session behavior)
- S/R zone proximity
A 48% pattern with a confluence score of 6 is more actionable than a 65% pattern with a score of 1. The win rate reflects base rate. The confluence score reflects whether this particular occurrence has the supporting context that separates signal from noise.
At higher timeframes (1h, 4h, daily), win rate carries more weight because there is less noise and macro structure already provides much of the confluence. At 15m, use win rate to select between patterns of similar confluence — not as a go/no-go gate on its own.
Keeping Win Rates Current: Quarterly Refresh
Markets change regime, and pattern reliability changes with them. The same pattern that delivers 64% in a trending bull market may drift toward 52% in a sideways or volatile market where breakout patterns fail repeatedly.
Oyamori refreshes win rate data quarterly from raw bar data — no blending with prior quarter results, no smoothing. If the market changed, the measured WRs change with it. The refresh process runs the full measurement pipeline fresh: new bar data, full detection pass across all patterns, forward outcome scoring, and update of any pattern × TF combination with at least 10 new samples.
When the AI Read confidence output in Oyamori charts reads lower than expected on a pattern you associate with a high win rate, this is often the system correctly reflecting a regime shift. The data is live, not locked to the era when the textbook was written.
Summary: What to Take Away
A chart pattern's win rate is only meaningful in context — the right timeframe, the right market regime, and with supporting confluence. The numbers from published textbooks are not measured from the same environment you are trading. Oyamori's approach is to measure continuously from real bars, keep the data current, and surface that accuracy as a quality signal inside the chart rather than as a headline number taken on faith.
The patterns that matter most at 15m are not the ones with the highest theoretical win rates — they are the ones with measurable edge at 15m combined with the confluence context that converts a base rate into a live signal.