Generated 2026-10-05 | D1 = flat when SPY < 200d SMA (exact production semantics); state map walk-forward (2010-01-01.., holdout >= 2023-01-01).
| variant | ret | sharpe | maxDD | downvol | CVaR5 | calmar |
|---|
| BASE | +970.5% | 0.97 | -34.1% | 16.6% | -2.08% | 28.42 |
| D1 | +437.7% | 0.87 | -26.9% | 13.6% | -1.72% | 16.29 |
| D1+state | +467.1% | 1.01 | -17.4% | 12.6% | -1.58% | 26.85 |
| variant | ret | sharpe | maxDD | downvol | CVaR5 | calmar |
|---|
| BASE | +236.6% | 1.49 | -26.0% | 18.3% | -2.48% | 9.08 |
| D1 | +158.6% | 1.28 | -21.5% | 17.2% | -2.40% | 7.37 |
| D1+state | +158.6% | 1.28 | -21.5% | 17.2% | -2.40% | 7.37 |
state-on-D1 block-placebo p = 0.000 (holdout map=M0_off); D1 base is the incumbent shadow candidate.
| feature | target | IC | IC OOS | partial vs state | q |
|---|
| corr63 | vol21 | +0.128 | -0.393 | +0.105 | 0.1 |
| corr63 | dd21 | -0.051 | +0.299 | -0.040 | 0.428 |
| corr63 | fwd21 | +0.087 | +0.171 | +0.086 | 0.171 |
| avg_dd | fwd21 | -0.192 | -0.149 | -0.193 | 0.000851 |
| n_above_ma | vol21 | -0.056 | -0.141 | -0.016 | 0.428 |
| n_in_dd | fwd21 | +0.207 | +0.135 | +0.212 | 0.000338 |
| pct_pos_20d | vol21 | +0.077 | -0.095 | +0.096 | 0.321 |
| pct_pos_5d | vol21 | +0.083 | -0.083 | +0.068 | 0.295 |
| pct_pos_20d | fwd21 | +0.033 | -0.069 | +0.023 | 0.558 |
| n_above_ma | dd21 | -0.055 | +0.069 | -0.054 | 0.428 |
| n_above_ma | fwd21 | -0.028 | -0.067 | -0.048 | 0.586 |
| pct_pos_5d | fwd21 | +0.052 | -0.067 | +0.038 | 0.428 |
| n_in_dd | dd21 | -0.148 | +0.051 | -0.105 | 0.0308 |
| pct_pos_20d | dd21 | -0.098 | +0.051 | -0.085 | 0.171 |
| pct_pos_5d | dd21 | -0.108 | +0.048 | -0.076 | 0.138 |
| avg_dd | dd21 | +0.193 | -0.043 | +0.138 | 0.00273 |
| n_in_dd | vol21 | +0.329 | -0.030 | +0.269 | 3.82e-07 |
| avg_dd | vol21 | -0.359 | -0.003 | -0.288 | 3.15e-08 |
| variant | ret | sharpe | maxDD | downvol | CVaR5 | calmar |
|---|
| BOOK | +785.5% | 0.97 | -32.3% | 15.0% | -1.82% | 24.33 |
| BOOK+cross_de-risk | +238.3% | 0.80 | -20.0% | 9.8% | -1.33% | 11.94 |
| variant | ret | sharpe | maxDD | downvol | CVaR5 | calmar |
|---|
| BOOK | +221.2% | 1.76 | -21.2% | 13.3% | -1.89% | 10.44 |
| BOOK+cross_de-risk | +140.5% | 1.71 | -15.1% | 10.2% | -1.46% | 9.32 |
- D1 does not transfer to the current baseline. On the frozen residual-momentum eqmom curve, D1 is WORSE than BASE in the walk-forward (Sharpe 0.87 vs 0.97) and on the holdout (1.28 vs 1.49) - the prior D1 win (SR 1.53 vs 1.10) was specific to the live101 12m raw-momentum config, not this one.
- **State-on-D1 helps the rolling walk-forward (D1+state 1.01 vs D1 0.87, block-placebo p=0.00) but NOT the untouched holdout: the map fitted on all pre-holdout data selected M0_off (M0_off), i.e. it chose to use no state overlay. The walk-forward uplift is therefore period-concentrated and does not survive the holdout selection.
- Cross-sleeve (§12): simultaneous sleeve deterioration predicts higher forward book vol (n_in_dd -> vol21 IC +0.33) but the simple de-risk overlay loses Sharpe (walk-forward 0.80 vs 0.97; holdout 1.71 vs 1.76) while cutting drawdown - the same de-risk-fails-to-pay pattern.
Recommendation: DO NOT PURSUE. Both incremental tests fail the untouched holdout. Separately, the incumbent D1 shadow does not replicate on the live residual-momentum baseline and should be re-validated on that config before any promotion.