Edge Finder
Where your record has been strongest and weakest. Descriptive statistics about the past with their uncertainty attached, not forecasts.
Strongest signals
47 buckets tested · Bonferroni α = 0.0011- Setup / strategy tag: Opening Range Breakout produced positive expectancy of +$234 per trade across 50 trades (95% CI +$134 to +$335), win rate 70%, profit factor 4.443.Holds up after correcting for the 47 buckets tested.
- Time of day (30 min): 09:30–10:00 produced positive expectancy of +$229 per trade across 48 trades (95% CI +$129 to +$329), win rate 71%, profit factor 4.362.Holds up after correcting for the 47 buckets tested.
- Volatility regime: Low volatility produced negative expectancy of −$166 per trade across 65 trades (95% CI −$315 to −$17), win rate 37%, profit factor 0.416.Does not survive correction for the 47 buckets tested; could be a multiple-comparisons artefact.
- Planned risk/reward: R:R 1–2 produced negative expectancy of −$125 per trade across 112 trades (95% CI −$266 to +$16), win rate 36%, profit factor 0.624.Confidence interval includes zero; not distinguishable from noise.
Explore
Bars: mean net P&L per trade. Whiskers: 95% confidence interval. Faded bars have fewer than 30 trades.
| Bucket | Trades | Win rate (95% CI) | Avg win | Avg loss | Expectancy (95% CI) | Profit factor | Max DD | Edge score | Evidence |
|---|---|---|---|---|---|---|---|---|---|
| 09:30–10:00 | 48 | 71% (57%–82%) | +$419 | −$233 | +$229 (129 to 329) | 4.362 | −$505 | 96 | holds after correction |
| 10:00–10:30 | 2 | 50% (9%–91%) | +$873 | −$137 | +$368 (-6049 to 6785) | 6.384 | −$137 | 52 | noise-compatiblen = 2: too few trades to distinguish skill from luck. |
| 10:30–11:00 | 8 | 38% (14%–69%) | +$661 | −$511 | −$71 (-782 to 640) | 0.776 | −$2,553 | 48 | noise-compatiblen = 8: too few trades to distinguish skill from luck. |
| 11:00–11:30 | 19 | 42% (23%–64%) | +$435 | −$399 | −$48 (-312 to 217) | 0.794 | −$2,032 | 45 | noise-compatiblen = 19: small sample; expect wide error bars. |
| 11:30–12:00 | 25 | 32% (17%–52%) | +$837 | −$627 | −$159 (-535 to 218) | 0.628 | −$5,246 | 38 | noise-compatiblen = 25: small sample; expect wide error bars. |
| 12:00–12:30 | 23 | 48% (29%–67%) | +$783 | −$920 | −$105 (-607 to 397) | 0.781 | −$6,226 | 44 | noise-compatiblen = 23: small sample; expect wide error bars. |
| 12:30–13:00 | 19 | 42% (23%–64%) | +$637 | −$449 | +$8 (-362 to 379) | 1.033 | −$3,461 | 51 | noise-compatiblen = 19: small sample; expect wide error bars. |
| 13:00–13:30 | 20 | 35% (18%–57%) | +$1,237 | −$596 | +$46 (-444 to 535) | 1.118 | −$4,334 | 52 | noise-compatiblen = 20: small sample; expect wide error bars. |
| 13:30–14:00 | 11 | 27% (10%–57%) | +$635 | −$937 | −$508 (-1275 to 259) | 0.254 | −$5,990 | 37 | noise-compatiblen = 11: small sample; expect wide error bars. |
| 14:00–14:30 | 3 | 0% (0%–56%) | n/a | −$421 | −$421 (-1164 to 323) | 0 | −$1,263 | 42 | noise-compatiblen = 3: too few trades to distinguish skill from luck. |
| 15:00–15:30 | 3 | 33% (6%–79%) | +$70 | −$178 | −$96 (-451 to 260) | 0.195 | −$357 | 46 | noise-compatiblen = 3: too few trades to distinguish skill from luck. |
| 15:30–16:00 | 13 | 23% (8%–50%) | +$209 | −$151 | −$68 (-166 to 31) | 0.415 | −$1,008 | 36 | noise-compatiblen = 13: small sample; expect wide error bars. |
Methodology and limits
Edge Score (0–100, 50 = no evidence either way) = 50 + 50 · tanh(t · w / 2), where t is expectancy divided by its standard error and w = n / (n + 20) shrinks small samples toward 50. It ranks how consistently a bucket made money in the past. It is not a probability of future profit.
Win-rate intervals are Wilson score intervals. Expectancy intervals use Student's t and assume trades are independent, which streaky behaviour violates; real uncertainty is wider.
Multiple comparisons. Slicing 47 ways guarantees some buckets look good by chance. “Holds after correction” means the bucket survives a Bonferroni adjustment across all of them.
Selection bias: setup tags added after the fact, or only journaling memorable trades, inflate results. Past regimes may not recur. Treat any edge here as a hypothesis to test forward on small size.