_Events: 692 raw, 232 mapped to the ADJ panel (219 adds, 13 dels). The panel is the current ~503-name universe, NOT delisting-inclusive, so most removals (delisted/acquired) are unmeasurable and additions are survivor-biased.
_Monthly PIT snapshots; forward returns from the snapshot close; excess vs equal-weight ADJ universe. Bootstrap p is one-sided P(mean >= 0). IS<2016<=OOS. Reversal story: additions<0, removals>0.
| kind | horizon | n | mean excess | boot p | IS | OOS | n_oos |
|---|---|---|---|---|---|---|---|
| add | 21d | 214 | -0.36% | 0.3070 | -0.62% | -0.26% | 152 |
| add | 63d | 214 | +0.71% | 0.6864 | -1.06% | +1.43% | 152 |
| add | 126d | 213 | +1.99% | 0.6522 | -0.40% | +2.97% | 151 |
| del | 21d | 9 | +10.05% | nan | +5.60% | +15.62% | 4 |
| del | 63d | 9 | +14.47% | nan | +7.60% | +23.07% | 4 |
| del | 126d | 9 | +11.02% | nan | +5.19% | +18.32% | 4 |
Additions @63d: excess +0.71%, random-symbol-on-same-date placebo p = 0.772 (reversal needs p<0.05 and obs<0).
Verdict
NOT TESTABLE (data-limited) — with the free/local panel the removal leg is unmeasurable (only 13 of 346 removed names have prices) and the addition leg is survivor-biased upward, which works against the reversal hypothesis. Among the survivors that can be measured there is no reversal (63d excess +0.71%, boot p 0.68, placebo p 0.77). A real test needs a delisting-inclusive price panel; free delisted history is not available (Yahoo 404s, Stooq blocks). Not adoptable; reopen only with that panel.
Reading
- PRIMARY (63d): additions reversal if mean excess < 0 with boot p<0.05 AND IS/OOS same sign; deletions mirror.
- This is post-effective drift only — the announcement run-up is invisible at monthly resolution, so a null here does not refute the short-horizon index effect, only its tradeable tail.
Time 9s.