Base OOS (>= 2016-01-01): Sharpe 1.10, maxDD -35.5%. 2000 permutations of each overlay's exposure series (same marginal exposure, random days).
| overlay | avg exposure | OOS Sharpe | placebo p(SR) | OOS maxDD | null maxDD | placebo p(DD) |
|---|---|---|---|---|---|---|
| D1 SPY-200d gate | 0.83 | 1.07 (null 1.00) | 0.277 | -23.7% (null -33.4%) | -33.4% | 0.017 |
| B5 vol-manage 12% | 0.55 | 1.03 (null 1.07) | 0.686 | -18.5% (null -20.8%) | -20.8% | 0.193 |
| C7 stress SPY hedge | 0.11 | 1.16 (null 1.08) | 0.061 | -27.9% (null -32.9%) | -32.9% | 0.053 |
Base: 1.10 / -35.5% (no overlay).
Reading
- p(SR) = P(random-timing exposure >= observed Sharpe); p(DD) = P(random timing achieves an equal-or-shallower drawdown).
- If p is not small, the overlay's benefit is exposure reduction (beta/vol), not timing skill — the classic overlay placebo trap.
- D1 note: the prior report's D1 Sharpe 1.53 gated the return SAME day (SPY close[t] on return[t] = lookahead). With a correct shift(1), D1 is a genuine DD reducer (p(DD)=0.017) but Sharpe is ~flat vs base — a risk tool, not return alpha.
- C7 is the only overlay that improves BOTH Sharpe and DD beyond random timing (p≈0.05 each) — the best shadow candidate. Time 12s.