Post-earnings drift measured from the other end. Instead of comparing EPS with the consensus, use the market's own verdict: the announcement-day move already folds in guidance, margins and sales, so drift conditioned on it should be cleaner and last longer (Brandt, Kishore, Santa-Clara and Venkatachalam, 2008). A large opening gap on heavy volume marks a reaction the market actually took part in, and the slower holders who catch up over the next weeks are the counterparty.
event-driven / post-earnings drift
Earnings-reaction drift
The idea
How it was tested
Pre-registered on 2026-10-14 (commit b80e123), committed before running as one of a batch of three conjunctions, so the bar was a Bonferroni t of 2.13, plus Gate 8 and at least 0.30% alpha per trade. Signal: within 3 calendar days of earnings, an overnight gap above 5% and relative volume above 2 against the 20-day average. Universe: US stocks at $10 or more with $5M of daily dollar volume. Buy at the next open, largest gap first, hold 20 sessions with no stop or target, at the lab's neutral default sizing: $5,000, 1% risk per trade, 10 slots.
Where it died
It failed at the first gate, and in the wrong direction. Over 9,097 trades, Jensen alpha was −0.676% per trade with t −4.021, on top of a beta of 1.307. Buying the biggest earnings gaps did not ride a drift; it bought high-beta names just before they underperformed. Gate 8 was never run.
The negative result had a sequel. Its mirror image, shorting the same gaps, was tested later as the earnings gap fade short. With a stop added, the short had no alpha either: the gaps that kept running were exactly the trades the stop removed.
The record
Thesis
Post-earnings drift measured by the market's reaction to the announcement, not by the EPS surprise. The announcement-day return aggregates all the information (guidance, margins, sales), so drift conditioned on it should be stronger and more persistent than drift conditioned on the earnings surprise. A large opening gap with heavy volume marks a reaction the market actually participated in.
Specification
- signal
- within 3 calendar days of earnings; overnight gap > 5%; 20-day relative volume > 2
- universe
- US stocks, price ≥ $10, 20d dollar volume ≥ $5M
- entry
- next open
- exit
- 20 sessions, no stop or target
- ranking
- largest gap first
- sizing
- neutral default — $5,000, 1% risk per trade, 10 slots
- validation
- one of a batch of 3 pre-registered conjunctions; Bonferroni k=3 → t ≥ 2.13; plus Gate 8 and alpha ≥ 0.30% per trade
Gates
Pre-registrationpass
Was the hypothesis, its parameters and its pass/fail rule written down before any data was looked at?
Gate 7 — Jensen alphafail
After regressing out market exposure, is there alpha left?
t ≥ 2.13 (Bonferroni, batch of 3)
t (HAC)-4.02alpha / trade %-0.68beta+1.31trades9,097
Gate 8 — Capture-preserved alphanot evaluated
Does the subset of trades a real portfolio can actually take keep the alpha?
Not evaluated once Gate 7 fails.
Why it died
The opposite of the thesis: buying the upside earnings gap on heavy volume returns −0.68% alpha per trade over 20 sessions (t −4.02), on top of a beta of 1.31. The reaction-based drift does not appear in this universe and period; its reverse does.
What it taught
Distinct from the lab's four earnings-surprise drift variants (which had also failed) only in the trigger variable — and it failed the same way. Together with the relative-volume continuation test, it is the second independent measurement in which buying names that are drawing attention earns less afterwards.
Verdict history
- 2026-10-14Pre-registered with two other conjunctions, committed before running
- 2026-10-14Single run: Gate 7 t −4.02 No alpha