- Label: 21d demeaned forward return; exposure ∝ RL long-prob
- Base: residual-momentum top-50, H21, 10bps
- Split: IS < 2016-01-01 (RL train), OOS >=
| Variant | Seg | Ret | Sharpe | MaxDD | days |
|---|---|---|---|---|---|
| base EW top-50 | IS | +220.4% | 0.72 | -66.0% | 2363 |
| base EW top-50 | OOS | +1272.3% | 1.47 | -34.4% | 2667 |
| base x RL conviction | IS | +0.0% | 0.00 | 0.0% | 2363 |
| base x RL conviction | OOS | +4344.6% | 1.68 | -33.5% | 2667 |
Controls
- Paired bootstrap OOS (RL - base): p = 0.000
- Weight-shuffle placebo: mean OOS SR 1.41, p(SR >= RL 1.68) = 0.005
- RL long-prob mean OOS 0.157