Evidence by cohort
Signal performance laboratory
Fixed-horizon outcomes grouped without removing failures. Medians reduce outlier distortion; MAE and drawdown expose risk that a headline win rate can hide.
Signal set
Sample
154
Win rate
20.78%
Median return
-4.99%
Median excess
-1.02%
Median MFE
+3.67%
Median MAE
-7.80%
Cohort drawdown
-39.40%
Evidence
30+ sample
8 Sept 2026 to 22 Sept 2026Regime benchmark: Nifty 500
| Market cap | Sample | Win rate | Median return | Median excess | MFE | MAE | Drawdown | Evidence |
|---|---|---|---|---|---|---|---|---|
| Large | 76 | 18.42% | -5.24% | -1.28% | +4.47% | -7.88% | -42.27% | 30+ sample |
| Small | 51 | 29.41% | -3.02% | +0.59% | +4.05% | -7.17% | -31.52% | 30+ sample |
| Mid | 27 | 11.11% | -6.75% | -2.18% | +1.91% | -8.22% | -60.18% | Early |
Returns: Median is used for return, excess return, MFE, and MAE to reduce outlier distortion. Return greater than 0% at the selected fixed horizon.
Drawdown: Signals from the same publication date are equal-weighted as one cohort, then cohort returns are compounded chronologically.
Market regime: Bullish, bearish, or sideways uses the benchmark close versus its trailing 50-session average and trailing 20-session return, using only data available by publication day.
Sample discipline: Groups below 10 outcomes are insufficient; 10-29 are early evidence; 30 or more are established samples, not proof of future performance.