Main _ADJ panel + earnings event windows. Observed = full-window net Sharpe at 10bps; null = within-date score permutation, N=200. Rule: REJECT if placebo p>=0.05; else SIGNAL.
| Candidate | Observed SR | null mean | placebo p | corr vs PEAD0 | verdict |
|---|---|---|---|---|---|
| PEAD0 plain-pead | 0.64 | 0.63 | 0.395 | nan | REJECT |
| H9 crowded-PEAD | 0.49 | 0.48 | 0.345 | 0.88 | REJECT |
| H10 preann-drift | 0.73 | 0.62 | 0.020 | 0.72 | SIGNAL |
| H11 earn-gap-fade | -0.15 | -0.27 | 0.250 | -0.00 | REJECT |
Size calibration (random scores on the H10 event mask)
K=4 masked-noise signals -> p: 0.15, 0.72, 0.52, 0.82; fraction p<0.05 = 0%.
Reading
- SIGNAL + low corr vs PEAD0 = a distinct event edge; SIGNAL + high corr = shared exposure (PEAD), not incremental.
- REJECT = the event-window selection adds nothing over random names in the same windows.