Panel 2020-01-01..2026-08-31 | long k lowest-z / short k highest-z, dollar-neutral | held [t,t+1] | IS<2024-01-01/OOS>= | 15bps/unit turnover | random-rank placebo x500.
| look | k | FULL net SR | IS net SR | OOS net SR | OOS gross SR | OOS ret | OOS DD | beta EW | avg turnover |
|---|
| 10 | 2 | -1.33 | -1.01 | -3.16 | 0.24 | -86.9% | -87.3% | -0.01 | 1.40 |
| 10 | 3 | -1.29 | -1.01 | -2.68 | 0.91 | -74.7% | -75.7% | -0.00 | 1.07 |
| 30 | 2 | -1.04 | -0.67 | -2.97 | 0.28 | -86.0% | -86.7% | -0.02 | 1.34 |
| 30 | 3 | -1.03 | -0.55 | -2.96 | 0.35 | -79.9% | -80.4% | -0.02 | 1.03 |
| 60 | 2 | -0.84 | -0.18 | -2.72 | 0.51 | -83.7% | -84.1% | -0.02 | 1.31 |
| 60 | 3 | -0.98 | -0.52 | -2.88 | 0.45 | -78.1% | -78.0% | -0.01 | 1.01 |
| variant | avg turnover | IS net SR | OOS net SR | OOS gross SR | OOS ret | placebo p |
|---|
| daily | 1.34 | -0.67 | -2.97 | 0.28 | -86.0% | 0.154 |
| every 5d | 0.29 | 0.12 | -1.82 | -1.11 | -69.3% | 0.000 |
- daily OOS obs net SR -2.97 vs null mean -3.66/p95 -2.56 -> p 0.154
- daily look 30 / k 2: OOS net SR -2.97, IS -0.67, placebo p 0.154, beta -0.02
- every-5d look 30 / k 2: OOS net SR -1.82, IS +0.12, placebo p 0.000, beta -0.04
- The gross book has a small positive OOS edge, but the premium z flips membership daily: turnover ~1.3x/day costs ~20bps/day, and even at a 5-day rebalance the selection does not clear the placebo. Market-neutral removed the beta but not the cost. REJECT; premium feed stays infrastructure.