Generated 2026-10-05 | frozen sleeves, walk-forward from 2010-01-01, untouched holdout >= 2023-01-01 | monthly re-selected sizing map (chosen on past only, >= 90% of base return) | PIT buckets | overlay cost 5bps on |dmult|.
eqmom -- walk-forward OOS
| overlay | ret | sharpe | maxDD | downvol | CVaR5 | calmar |
|---|---|---|---|---|---|---|
| BASE | +970.5% | 0.97 | -34.1% | 16.6% | -2.08% | 28.42 |
| breadth | +420.7% | 0.84 | -34.1% | 15.0% | -1.68% | 12.32 |
| dispersion | +292.4% | 0.82 | -21.3% | 12.0% | -1.60% | 13.74 |
| concentration | +414.3% | 0.90 | -26.4% | 13.8% | -1.63% | 15.69 |
| combined | +529.4% | 0.89 | -34.1% | 15.4% | -1.85% | 15.51 |
eqmom -- HOLDOUT
| overlay | ret | sharpe | maxDD | downvol | CVaR5 | calmar |
|---|---|---|---|---|---|---|
| BASE | +236.6% | 1.49 | -26.0% | 18.3% | -2.48% | 9.08 |
| breadth | +236.6% | 1.49 | -26.0% | 18.3% | -2.48% | 9.08 |
| dispersion | +217.7% | 1.56 | -21.4% | 15.9% | -2.16% | 10.18 |
| concentration | +236.6% | 1.49 | -26.0% | 18.3% | -2.48% | 9.08 |
| combined | +236.6% | 1.49 | -26.0% | 18.3% | -2.48% | 9.08 |
flagship -- walk-forward OOS
| overlay | ret | sharpe | maxDD | downvol | CVaR5 | calmar |
|---|---|---|---|---|---|---|
| BASE | +583.1% | 0.84 | -30.8% | 15.0% | -1.88% | 18.93 |
| breadth | +476.9% | 0.83 | -30.8% | 15.0% | -1.88% | 15.48 |
| dispersion | +464.6% | 0.82 | -30.8% | 15.0% | -1.88% | 15.08 |
| concentration | +385.0% | 0.78 | -30.8% | 14.8% | -1.76% | 12.50 |
| combined | +435.3% | 0.81 | -30.8% | 14.6% | -1.82% | 14.13 |
flagship -- HOLDOUT
| overlay | ret | sharpe | maxDD | downvol | CVaR5 | calmar |
|---|---|---|---|---|---|---|
| BASE | +193.8% | 1.62 | -18.3% | 13.1% | -1.57% | 10.58 |
| breadth | +193.8% | 1.62 | -18.3% | 13.1% | -1.57% | 10.58 |
| dispersion | +193.8% | 1.62 | -18.3% | 13.1% | -1.57% | 10.58 |
| concentration | +193.8% | 1.62 | -18.3% | 13.1% | -1.57% | 10.58 |
| combined | +193.8% | 1.62 | -18.3% | 13.1% | -1.57% | 10.58 |
Parameter perturbation (combined overlay, eqmom walk-forward)
| look | step | dSharpe | dMaxDD | dRet |
|---|---|---|---|---|
| 504.0 | 63.0 | -0.05 | +4.2% | -518.5% |
| 504.0 | 126.0 | -0.05 | +4.2% | -530.1% |
| 504.0 | 252.0 | -0.06 | +4.2% | -549.6% |
| 756.0 | 63.0 | -0.04 | -0.0% | -389.7% |
| 756.0 | 126.0 | -0.08 | -0.0% | -441.1% |
| 756.0 | 252.0 | -0.07 | -0.0% | -431.2% |
| 1008.0 | 63.0 | -0.02 | +6.6% | -438.5% |
| 1008.0 | 126.0 | -0.03 | +6.6% | -449.3% |
| 1008.0 | 252.0 | -0.03 | +6.6% | -453.5% |
Placebo (block-permuted state buckets, 100 reps) p = 0.920
Phase-3 verdict
-
eqmom: walk-forward best overlay has Sharpe 0.90 vs base 0.97 and maxDD -26.4% vs -34.1%; holdout best is 1.56 vs base 1.49.
-
flagship: walk-forward best overlay has Sharpe 0.83 vs base 0.84 and maxDD -30.8% vs -30.8%; holdout best is 1.62 vs base 1.62.
-
Holdout cells meeting a naive gate (better maxDD AND Sharpe within 0.10 of base): eqmom/dispersion (Sharpe 1.56 vs 1.49, maxDD -21.4% vs -26.0%). These are single cells among ~40 tested with no multiple-testing correction, and their walk-forward counterparts are inferior; they do NOT meet the promotion bar.
-
The walk-forward overlay systematically REDUCES Sharpe and gives back large return for at best a comparable drawdown; the parameter perturbation shows the drawdown effect is unstable (flips sign with look/step), and the block-placebo p=0.92 shows the observed overlay Sharpe change is no better than a permuted/random state assignment.
-
The state features are genuine forward-RISK predictors (Phase 1/2) but do NOT produce a promotable exposure overlay: the frozen sleeves already carry their own risk controls (vol target, stops, credit/VIX gates), so an exogenous de-risk mostly subtracts return.
Recommendation: DO NOT PURSUE the overlay. Equity-state is informative as a risk monitor / research diagnostic, not as a tradeable sizing signal for these sleeves. No production change.