Generated 2026-10-05 | spec 13D/E/F. Headline relations re-estimated across universes, regimes and rolling windows.
| pair | full | large100 | halfA | halfB | sign-stable |
|---|
| avg_corr20->momls_fwd21 | -0.094 | -0.119 | -0.060 | -0.102 | YES |
| breadth_sma200->momls_fwd21 | +0.072 | +0.072 | +0.108 | +0.046 | YES |
| breadth_sma200->spy_vol_fwd21 | -0.145 | -0.121 | -0.138 | -0.151 | YES |
| corr_spy60->flagship_mdd_fwd21 | +0.051 | +0.018 | +0.049 | +0.052 | YES |
| corr_spy60->spy_vol_fwd21 | +0.059 | +0.034 | +0.063 | +0.056 | YES |
| disp20->spy_vol_fwd21 | +0.245 | +0.260 | +0.222 | +0.247 | YES |
| top10_dvwt->eqmom_vol_fwd21 | +0.324 | +0.324 | +0.218 | +0.319 | YES |
| pair | high_vol | low_vol | bull | bear |
|---|
| avg_corr20->momls_fwd21 | -0.093 | -0.137 | -0.078 | -0.049 |
| breadth_sma200->momls_fwd21 | +0.154 | -0.038 | -0.049 | +0.245 |
| breadth_sma200->spy_vol_fwd21 | -0.372 | +0.019 | -0.067 | -0.050 |
| corr_spy60->flagship_mdd_fwd21 | +0.050 | +0.026 | +0.086 | -0.168 |
| corr_spy60->spy_vol_fwd21 | +0.033 | +0.022 | +0.013 | -0.052 |
| disp20->spy_vol_fwd21 | +0.265 | +0.198 | +0.179 | +0.409 |
| top10_dvwt->eqmom_vol_fwd21 | +0.209 | +0.382 | +0.371 | +0.161 |
| pair | windows | mean | min | max | frac same sign |
|---|
| breadth_sma200->spy_vol_fwd21 | 68 | -0.447 | -0.744 | -0.154 | 100% |
| disp20->spy_vol_fwd21 | 68 | +0.285 | -0.010 | +0.774 | 99% |
| corr_spy60->spy_vol_fwd21 | 67 | +0.319 | -0.311 | +0.636 | 91% |
| top10_dvwt->eqmom_vol_fwd21 | 68 | +0.140 | -0.180 | +0.493 | 84% |
| corr_spy60->flagship_mdd_fwd21 | 67 | -0.155 | -0.364 | +0.288 | 84% |
| avg_corr20->momls_fwd21 | 68 | -0.138 | -0.408 | +0.072 | 96% |
| breadth_sma200->momls_fwd21 | 68 | +0.130 | -0.117 | +0.367 | 85% |
- 7/7 headline pairs keep the same sign across all four universes (full, large100, two random halves).
- Regime table shows whether the relations are concentrated in high-vol/bear windows (a concern for a volatility-based overlay) or hold unconditionally.
- Rolling-window 'frac same sign' near 100% means the relation is stable through time, not driven by one episode.
The risk-prediction relations are expected to be the robust part (vol clustering); the momentum-spread/correlation relation is the one to watch for instability.