Research
Do chart patterns work? We tested 179 patterns on six markets
Only 3 of the 174 patterns whose detection we could check beat chance on all markets pooled, on daily bars. 4 more beat it on a single market and timeframe, and 10 did worse than chance. For almost all the others, we could not tell what followed the pattern apart from what follows a random bar.
Updated November 5, 2026, from the latest run of our test bench. Historical statistics before fees and slippage. Not investment advice: past results do not predict future ones.
- Patterns tested
- 179
- Occurrences detected
- 52,032,635
- Assets, six markets
- 2,368
- Beat chance everywhere
- 3
The short answer
Each pattern is a written rule run on full price history. What happened in the 10 bars after it is compared with the rest of the market, and a simple 2:1 reward-to-risk trade on it is compared with the same trade on random dates. A result only counts once it survives a correction for the thousands of tests we run.
| Verdict | Patterns |
|---|---|
| Beat chance on the headline sample | 3 |
| Beat chance on one market and timeframe only | 4 |
| No measurable edge | 153 |
| Worse than chance | 10 |
| Not enough data to judge | 4 |
| Not judged: detection not verified | 4 |
| Never found on our data | 1 |
The patterns that beat chance
All markets pooled, daily bars
- Bearish stalled (deliberation): +0.47% excess return over the rest of the market after 10 bars
- Bearish 5-0: +0.34% excess return over the rest of the market after 10 bars
- Bearish harami: +0.18% excess return over the rest of the market after 10 bars
On one market and timeframe only
- Bearish butterfly: +0.98% excess return over the rest of the market after 10 bars (Crypto, 4-hour)
- Diamond top: +0.34% excess return over the rest of the market after 10 bars (Crypto, 1-hour, downward breakouts)
- Bearish ABCD: +0.20% excess return over the rest of the market after 10 bars (US stocks, daily)
- Bearish harami cross: +0.17% excess return over the rest of the market after 10 bars (US stocks, daily)
Before fees. A small edge per trade can still vanish after costs on short timeframes.
The patterns that did worse than chance
On their headline sample, these 10 patterns did worse than chance after correction, on the 10-bar move (against the direction they are taught) or on the 2:1 trade (against the same trade on random dates):
- Bearish J-hook
- Black candle
- Black closing marubozu
- Black marubozu
- Head and shoulders
- Long black line
- Ugly double bottom
- V bottom
- White candle
- White closing marubozu
What textbooks teach against what we measured
- Direction. Of 160 patterns with a taught direction, the measured move confirmed it for 3, contradicted it for 11, and could not tell for 146.
- Targets. For 69 of 84 chart patterns, the taught target is reached clearly less than half the time before the pattern fails (all markets pooled, daily).
- Trading them. Across 2,212 combinations of pattern, market and timeframe with at least 30 trades, a 2:1 trade on the pattern beat the same trade on random dates in none, and did worse in 40.
Market by market
The same patterns on each market, timeframe and breakout side, counting only the 2,325 combinations with at least 30 occurrences (all markets pooled included in that total, not in the table below):
| Market | Combinations | Beat chance | Worse than chance |
|---|---|---|---|
| Crypto | 483 | 3 | 38 |
| Forex | 378 | 0 | 10 |
| US stocks | 339 | 4 | 10 |
| ETFs | 206 | 0 | 5 |
| Indices | 171 | 0 | 14 |
| Commodities | 140 | 0 | 1 |
10-bar move against the rest of the market, after correction for the number of tests.
How we tested
Full history, strict geometric rules applied only to what was known at the time, a comparison with random bars of the same assets, and a correction for the number of tests. The whole pipeline also runs on random-walk prices, where no pattern can work, and has to pass before any number is published. The full method is on the home page, and every pattern has its own page with its rule, its real examples and its numbers by market.
Limits
- Stocks: survivorship bias not quantified (delisted stocks missing). Bullish rebounds on stocks are overstated by an unknown amount.
- Recall of the detector is low by design (narrow ratio bands or template): occurrences are undercounted, figures per occurrence remain valid.
- Prior trend and drawn context are read by rule, not as a chart reader would judge them.
- Commodities: unadjusted futures rolls without a detectable jump stay in the data.
- Forex intraday: results include the weekend gap, a real market risk.
- Detection was checked by a blind reading by a model, not by a human panel.
Data, citation and cheat sheets
Every number on this page comes from our open data: JSON files and a read-only MCP server (https://patternswizard.com/mcp/) that AI assistants can query. Quote it with a link to this page, credited as “PatternsWizard (https://patternswizard.com/), pattern statistics”.
On one printable page, with every pattern's verdict: the candlestick patterns cheat sheet and the chart patterns cheat sheet.