G1 gap-down bounce / G2 gap continuation (±1%, 9:30->15:50) / IM first-30m>0.5% held 15:30->16:00. 4bps round trip. Pooled bootstrap p vs zero.
| Variant | names | trade-days | Sharpe | maxDD | p |
|---|---|---|---|---|---|
| G1 | 61 | 8415 | 0.08 | -69.3% | 0.109 |
| G2 | 61 | 8922 | 0.01 | -95.2% | 0.444 |
| IM | 61 | 18995 | -1.12 | -100.0% | 1.000 |
Verdicts (2026-09-09)
All three FAIL. Scoreboard: 43 judged, 0 adopted, 1 watchlisted.
- G1 gap-down bounce — FAIL, nearest of the three. 38/61 names positive but pooled p=0.109 and maxDD -69.3%: gap-downs cluster in crashes, and bounces fail exactly when most needed. The tails own the strategy. No filter added (shock-filters are parked family).
- G2 gap continuation — FAIL. 0.01, p=0.444. Nothing.
- IM intraday momentum — FAIL decisively (-1.12, 2/61 positive). The Gao et al. first-to-last-half-hour effect does not replicate on these names at 1-min resolution long-only. Dead here.
The 1-minute well is now dry across two universes and four mechanisms (ORB, gap-bounce, gap-continuation, intraday momentum): everything either loses outright or shows an untradable sliver. The pilot cache stays for genuinely new questions, but the intraday trend/reversal family is done.