The statistical truth about pattern win rates

Why the textbook win rates do not survive contact with reality

Open any pattern textbook and you will find win-rate statistics. Head and shoulders works 65 percent of the time. Cup and handle works 75 percent of the time. Double bottoms work 78 percent of the time. The numbers vary by author but the structure is always the same. A shape gets a percentage attached, and beginners come away thinking the pattern has reliable predictive power. The numbers do not survive scrutiny.

Here are the methodology problems baked into most pattern statistics. First, selection bias. The researcher decides what counts as a clean pattern after looking at the chart, which means messy or failed patterns are often quietly excluded. Second, survivorship bias. Patterns that did not result in clean follow-through are sometimes dismissed as 'not really patterns' rather than counted as failures. Third, look-ahead bias. Patterns are identified using information that would not have been available at the moment a trader was considering acting on them. Each of these problems inflates the win rate.

When researchers apply strict mechanical pattern definitions to large, out-of-sample datasets, reliability figures generally come back closer to coin flip territory. That sounds discouraging, but it is actually useful. It tells you the shape alone is not where the edge lives. Pattern trading can still be profitable when the patterns are used as a filter inside a larger trading system, rather than as a standalone signal.

Where real pattern edge comes from. First, alignment with the trend on the higher timeframe. A bullish pattern in a daily uptrend has materially better odds than the same pattern against the daily trend. Second, location on the higher timeframe. A pattern forming at meaningful structure on the daily or weekly chart has more weight than the same shape forming in random middle-of-range price action. Third, volume behavior. A breakout on rising volume tends to follow through more reliably than a breakout on dry volume. Fourth, asymmetric risk-reward. You do not need to be right 70 percent of the time if your winners are two or three times larger than your losers.

This is the chapter's quiet thesis. Classical chart patterns are useful, but only as one input among several. Treat the shape as a hypothesis. Confirm with trend, location, and volume. Manage risk so you are not relying on a high win rate. That framing keeps you out of the trap of believing the textbook numbers and protects you from the disappointment that follows when those numbers fail to show up in your own results.

Recap: published pattern win rates are inflated by selection, survivorship, and look-ahead bias. Real out-of-sample reliability is closer to coin flip. Edge comes from trend alignment, location, volume, and asymmetric risk-reward, not from the shape alone.