How Chart Pattern Win Rate Statistics Are Calculated

What Does a Pattern Win Rate Mean?

A win rate is meaningful only when its sample and outcome rules are clear. Reaching a measured-move target, closing above an entry, and earning a profit after trading costs are different tests. TradingPatterns.io reports outcomes from detected patterns under the rules below, not audited brokerage returns.

Our Methodology: From Detection to Outcome

1. Record the Setup

Detections record the pattern type, ticker, timeframe, dates, confidence score, and available entry, target, and stop levels. The calculation uses the recorded levels; it does not simulate each reader's fills or discretionary exits.

2. Evaluate Target and Stop

The evaluator scans candles after the pattern's resolution anchor: the pattern-end date when available, otherwise the detection date. When a candle contains both target and stop, the stop takes precedence because candle data cannot establish which was touched first.

Target and stop outcomes use the recorded entry level to calculate a direction-adjusted return. A bearish detection therefore measures a downward move as favorable.

3. Resolve Timeouts

The evaluation window depends on the timeframe and candle limit. If neither level is reached before the horizon, the timeout return is measured from the first post-anchor candle's open to the last evaluated candle's close, adjusted for direction.

Positive timeout returns count as wins; zero or negative timeout returns count as losses. When no evaluation candles are available at the horizon, the resolver records a zero-return loss. Timeout returns consequently use a different price basis from target and stop returns.

This means the published win rate is not the percentage of patterns that reached their target. There is no separate partial-win category in this calculation.

4. Aggregate Eligible Outcomes

Win rate = completed outcomes / (completed + failed outcomes) × 100.

Only valid, resolved detections enter this aggregate. Forming detections that expire are excluded. The selected lookback uses pattern-end date, falling back to detection date. Average return includes both completed and failed outcomes; average days to completion applies to completed outcomes, including positive timeouts.

Corrections can invalidate detections or revise results. Histories are therefore a record of currently eligible detections, not an immutable ledger of every signal ever seen. Consult the public methodology when comparing figures over time.

Read the Sample Before Comparing Patterns

Check the pattern, timeframe, lookback, and number of resolved detections. A small sample can move sharply after a few outcomes. A large sample can still contain correlated detections from the same assets or market period; it is not automatically a set of independent trades.

Compare like-for-like windows on the track record and relevant pattern history pages. This article does not maintain a separate current database count or a verified historical performance snapshot.

Market Context

Earnings, market direction, and liquidity can affect price behavior. Investigate these in your own research; the public track record does not provide automatic bull-, bear-, and sideways-market segmentation.

Timeframe

Timeframes use different candle sizes and evaluation horizons. Compare their own samples rather than assuming weekly, daily, or intraday patterns have a universal ranking of win rates.

Confidence and Volume

The confidence score measures the detector's criteria, not calibrated odds. Volume is supporting evidence. Neither implies a fixed improvement in win rate without a documented comparison with matching samples and outcome rules.

Why Win Rate Alone Is Misleading

Consider two hypothetical examples, before costs, where R is the planned amount at risk:

| Example | Win rate | Average win | Average loss | Expected return per outcome |

|---|---|---|---|---|

| A | 70% | 1R | 1R | 0.70 × 1R − 0.30 × 1R = +0.40R |

| B | 50% | 3R | 1R | 0.50 × 3R − 0.50 × 1R = +1.00R |

Example B has a lower win rate but higher mathematical expectancy under these assumptions. These are illustrations, not measured returns for any named pattern. An intended target-to-stop ratio is also different from the average gains and losses actually realized.

What These Statistics Do Not Include

The model excludes commissions, spreads, slippage, borrow costs, and position sizing. It does not establish that an account could enter every detection, especially when signals overlap or prices gap. Positive modeled returns are not a guarantee of executable profit.

Keep a separate trading journal with actual fills, costs, and exits. Compare that journal with the detection history only after accounting for the different rules. Use the track record for the current published sample and the methodology for how it is scored.