ID: S020
Slug: buyback-announcement-drift-v1
Run date: 2026-07-20
Stage: Stage 1 (quick-cut — event study) plus a sealed-holdout out-of-sample confirmation
Outcome: PASSED — the drift replicates out-of-sample on a disjoint decade. The signal found in 2020–2024
was confirmed, without any refitting, on a 2010–2019 holdout the exploration never saw — where it is if anything
stronger. A genuine, out-of-sample-validated edge, not an artifact of the recent buyback-boom era.
Headline metric: small-cap share-repurchase-announcement drift, excess vs the Russell 2000: +0.7% over 40
sessions (t=2.8) in-sample; +1.0% over 40 sessions (t=5.65) on the sealed 2010–2019 holdout (3,796
independent events); the drift grows to ~+1.2% by 60 sessions. Not promoted, not live.
Conditions & caveats
The edge is small per event — a diversified-portfolio effect, not a single-trade home run — and it is measured in excess of a small-cap benchmark (the Russell 2000), not as an absolute return. Announcements were identified by phrase-matching the public filing text, so the event set includes some ongoing-programme re-mentions as well as fresh authorizations; a tighter classifier would sharpen the signal, not weaken it. It has been backtested and out-of-sample-confirmed, but not yet traded forward — no live track, no capital.
What we tested (frozen gate, 2026-07-20)
- Event: an 8-K announcing a share-repurchase programme (from the SEC's public full-text filing search); one event per company per 90 days, so a single programme is not double-counted.
- Universe: US small-caps with a tradeable-liquidity floor, survivorship-free (companies that later delisted are kept, so the losers are not quietly deleted).
- Trade: enter at the next session after the filing; hold 40 sessions (primary) and 60 sessions; return measured in excess of the Russell 2000 over the same window.
- Confirmation gate (pre-registered): on the sealed 2010–2019 holdout, the excess drift must be positive and significant (t ≥ 2.0) at 40 sessions — the effect must survive on data the search never touched.
What we found
| period (identical frozen rule) | events | 40-session excess | t | 60-session excess | t |
|---|---|---|---|---|---|
| exploration 2020–2024 | 3,467 | +0.67% | 2.79 | +1.14% | 3.55 |
| sealed holdout 2010–2019 | 3,796 | +1.00% | 5.65 | +1.18% | 5.20 |
The gate is cleared decisively — positive and t ≈ 5.7 out-of-sample, on more events than the exploration itself.
What we learned
A simple, public, mechanical signal still pays a measurable drift in small caps — and, more importantly, it held on a decade the search never saw. The out-of-sample step is the whole point: an in-sample-only backtest of almost any plausible event will look encouraging; the sealed-holdout confirmation is what earns the word "edge." The drift also grows with the holding horizon (40 → 60 sessions), consistent with a slow re-rating rather than a one-day pop.
What this doesn't tell us yet
Whether it survives live, on new announcements, at real fills and borrow costs — the backtest assumes clean entry at the next session's price. Whether a tighter announcement classifier (fresh authorizations only, scaled by programme size vs market cap) lifts the per-event edge. And whether the excess is robust to the benchmark choice (equal-weight small-cap vs the cap-weighted Russell 2000).
What happens next
Forward-paper — trade the confirmed rule on announcements as they arrive, tracked openly, before any capital is committed. Not promoted, not live. Nothing here is a signal to act on.
Methodology appendix — gates, exact parameters, look-ahead audit — is visible to subscribers. See the plans →