ID: S023
Slug: overnight-only-capture-v1-stage2
Run date: 2026-07-17
Status: passed-s2 — conditional; the edge is regime-dependent (see Conditions & caveats)
Vehicle: liquid-ETF composite (equal-weight QQQ+IWM), pre-specified from Stage-1
Conditions & caveats
⚠ Prominent caveat — this is a BULL-REGIME edge. The overnight premium is +13.9%/yr (p=0.000) when SPY is above its 200-day SMA (simple moving average) and −15.0%/yr (p=0.08) when below. It bleeds hard in downtrends — 2008 −23%, 2022 −19%, 2016 −5% — because unhedged overnight gap risk materialises as overnight gap-downs in bear markets. The +5.9%/yr unconditional headline is a blend that stays net-positive only because bull regimes are ~75% of days. Managed-risk, not a robust all-weather edge. The obvious refinement — trade overnight only while SPY>200SMA — is a separate hypothesis (new identifier), not claimed here.
What we tested
The strategy is parameter-free (buy adjusted close, sell adjusted open, daily), so there is nothing to fit — every window is out-of-sample (OOS) by construction. Stage-2 therefore tests temporal and regime robustness of the thin Stage-1 edge over 2005→2026, at the frozen 1 bp/leg (basis-point-per-leg) friction (2× = 2 bp/leg).
What we found
| net bp/day | net ann | p | 2× friction | |
|---|---|---|---|---|
| bull — SPY > 200d-SMA (n=4,051) | +5.34 | +13.9% | 0.000 | +3.34 |
| bear — SPY < 200d-SMA (n=1,333) | −5.71 | −15.0% | 0.084 | −7.71 |
| overall @1 bp/leg | +2.61 | +5.9% | — | — |
| overall @2 bp/leg (2× stress) | +0.61 | +0.7% | — | — |
Purged combinatorial sub-period Sharpe (28 held-out combos): median +0.58, 96% positive, worst −0.04.
What we learned
Stage-2 did its job: it showed that a headline which looks unconditional is really a regime-conditional edge. The overnight premium is real and temporally consistent, but it is not weather-proof — the bear-market bleed is the load-bearing fact, and it now sits in the open rather than hidden inside a blended average.
What this doesn't tell us yet
Whether the edge survives at real auction fills traded forward — this remains a backtest at an assumed cost — and whether a regime filter (trading overnight only in up-trends) keeps the upside while cutting the bleed. That is a separate, not-yet-tested hypothesis. It also says nothing about capacity.
What happens next
Stage-3 forward-paper: track the composite's overnight net at realistic auction fills and report performance split by SPY regime, so the bear-market bleed is visible in real time, and let it motivate the regime-filtered (bull-only) sibling as a cleaner separate hypothesis. The next gate is the spec's frozen Stage-3 forward-paper gate. This gate was set when the result was adjudicated, not pre-registered.
Methodology appendix — gates, exact parameters, look-ahead audit — is visible to subscribers. See the plans →