Free tool
Candlestick Pattern Scanner
Detects all 14 patterns from Nison's book on real price data — then does the thing no pattern scanner does: benchmarks each one against a random entry with identical exits, so you can see whether the pattern is doing any work at all. Defaults to all five symbols pooled, because single-symbol samples are too thin to read.
Signal confirmed on the close, filled at the next bar's open. 10 bps round-trip costs. When stop and target are both touched in one bar, the stop is assumed to fill first.
Every pattern, versus a random entry
| Pattern | Side | Signals | Win rate | Avg return | Random | Edge | Read |
|---|
How to read the Edge column. It is the pattern's average return minus what random entries earned using the same exits on the same bars. Positive means the pattern's timing added something. Near zero means you were paid by the market, not the pattern.
Anything under 20 signals is flagged thin. Treat those as unreadable — in our full study the median cell had just 10 trades, and cells that small produce impressive-looking percentages that mean nothing.
Use your own data
Paste OHLC as CSV — date,open,high,low,close, one row per bar, oldest first. Header row optional.
To wire this to your own feed, replace the DATA object with a fetch that returns
{t:[],o:[],h:[],l:[],c:[]}. Everything below that line is pure client-side computation —
no backend required.
What this tool will probably show you
Run it across a few symbols and the pattern will be familiar: bullish patterns post decent-looking win rates, and the random control posts almost the same ones. That gap — usually close to nothing — is the whole story. We measured it properly across 2,022 trades and ten years, and 0 of 14 patterns showed a significant edge.
This tool runs on a few hundred bars of one symbol, so treat it as a demonstration of the method rather than evidence about any pattern. The point is the comparison, not the number. The full study is here →
For educational purposes only. Results are computed on historical data with modelled costs and are not indicative of future performance. Not investment advice.