Not supported · Patterns · 2026-09-26; timeframe gap fill 2026-10-11
Are some harmonic patterns better than others on 15m?
What this means
On 15m, mean R ranged from about 0R (Bat) to +0.19R (Deep Crab), but none of the differences survived the multiple-testing check (best q = 0.28). Dropping the two weakest patterns would add under 0.01R per setup. On this evidence, all patterns perform about the same on 15m. The same holds on 1h, 4h and daily. The only nominal differences, on 4h, do not survive the multiple-testing correction, and Gartley on 4h is the same pattern that ranks worst on 1h.
The explanation is our best reading; the data shows the result but not the cause directly.
By timeframe
| Timeframe | Verdict | n | Result |
|---|---|---|---|
| 15-minute | Not supported | 16,500 | Deep Crab +0.19 to Bat -0.01; best q=0.276 |
| 1-hour | Not supported | 13,208 | pattern minus rest from -0.06 (Gartley) to +0.06 (Bat); all 7 CIs span 0; lowest q=0.42 gap fill G8, 2026-10-11 (trade logs) |
| 4-hour | Inconclusive | 3,990 | Gartley +0.28 (+0.04, +0.51) q=0.06, Shark -0.12 q=0.05, others span 0 gap fill G8, 2026-10-11 (trade logs) |
| Daily | Not supported | 2,098 | all 7 CIs span 0; lowest q=0.43 gap fill G8, 2026-10-11 (trade logs) |
- Hypothesis
- Pattern type (Bat, Butterfly, Crab, Cypher, Deep Crab, Gartley, Shark) changes win rate or mean R.
- Method
- 15m TradingView setups (16,500); per-pattern test with FDR; train/test split.
- Result
- Mean R from Deep Crab +0.193 (n=466) to Bat -0.006 (n=668); best q = 0.276. Dropping Bat + Butterfly adds +0.0075 (train) / +0.0086 (test). Gap fill (2026-10-11), each pattern vs all others with stock-clustered CIs: 1h all seven within -0.06 to +0.06R and every CI spans 0; 4h Gartley +0.28R (+0.04, +0.51) and Shark -0.12R (-0.22, -0.02) but q = 0.06 and 0.05 after correcting for the 82 gap-fill tests; daily all spans 0.
- How much to trust it
- High: TV fills, FDR-controlled
- Did it change the live rule?
- all patterns stay on
Scripts and data for this study will be linked here when the research code is published on GitHub.
Education and research only. Not financial advice. Backtests use today's index members (survivorship bias).