universe: equities rows: 3950, dates: 79, trials K=160
1. 21-session OOS accuracy (expanding calibration)
width : QLIKE 0.7807 | log-MSE 2.712 | rank-corr +0.594 (n=2700) trail21 : QLIKE 0.7496 | log-MSE 2.576 | rank-corr +0.596 (n=2700) HAR(5/21/63) : QLIKE 0.7418 | log-MSE 2.538 | rank-corr +0.643 (n=2700) w+t21 : QLIKE 0.7540 | log-MSE 2.589 | rank-corr +0.637 (n=2700) HAR+w : QLIKE 0.7501 | log-MSE 2.570 | rank-corr +0.653 (n=2700) blend(w,t21) : QLIKE 0.7573 | log-MSE 2.602 | rank-corr +0.640 (n=2700)
2. multi-horizon (rank corr / QLIKE)
fwd 5d width : rank-corr +0.422 QLIKE 1.0959 (n=2750) fwd 5d trail21 : rank-corr +0.424 QLIKE 1.0713 (n=2750) fwd21d width : rank-corr +0.594 QLIKE 0.7807 (n=2700) fwd21d trail21 : rank-corr +0.596 QLIKE 0.7496 (n=2700) fwd63d width : rank-corr +0.652 QLIKE 0.7156 (n=2600) fwd63d trail21 : rank-corr +0.650 QLIKE 0.6892 (n=2600)
3. incremental (pooled OOS: width + trail21)
log-vol ~ log(width) + log(trail21): b_width +0.410 (t=+16.69), b_trail +0.355 (t=+16.68) b_width date-clustered 95% CI [+0.331,+0.487] (P(>0)=1.00)
4. per-date cross-sectional rank corr(predictor, fwd21)
width : mean +0.631 CI[+0.603,+0.659] (n=78, t=+55.0) trail21 : mean +0.626 CI[+0.591,+0.660] (n=78, t=+47.5) HAR-avg : mean +0.659 CI[+0.631,+0.688] (n=78, t=+55.2)
5. 21-session rank corr by prior trail21 tercile
low-vol (n=1300): width +0.312 | trail21 +0.270 | HAR +0.309 mid-vol (n=1300): width +0.290 | trail21 +0.169 | HAR +0.278 high-vol (n=1300): width +0.488 | trail21 +0.422 | HAR +0.505
6. sizing head-to-head (mom top-decile, 21-session fwd)
equal : mean +2.589% CI[+1.056%,+4.243%] maxDD -32.5% (n=78) inv_width : mean +2.128% CI[+0.726%,+3.596%] maxDD -30.3% (n=78) inv_trail21: mean +2.035% CI[+0.593%,+3.549%] maxDD -29.4% (n=78) inv_HAR : mean +2.032% CI[+0.638%,+3.544%] maxDD -30.1% (n=78)