Post-earnings drift, short side. After a material miss, analysts cut their estimates over weeks and institutions trim positions gradually, so the price keeps sliding after the report (Bernard and Thomas, 1989). The trigger is the surprise itself, not the price drop: the target is the misses the market under-reacted to on the first day, where the drift should still have room to run.
event-driven / post-earnings drift
Earnings miss drift short
The idea
How it was tested
Pre-registered on 2026-06-03 as the surprise-only short of the drift family. Signal: EPS more than 10% below consensus, taken the day after the report. Universe: US stocks at $10 or more with at least $5M of daily dollar volume. Short at the next open, worst miss first, out after 20 sessions or at a 2.5×ATR stop or 4×ATR target. One run.
Where it died
The June run looked like a partial win. Alpha t 5.28 and a Gate 8 pass said the drift was real; the gross return of +0.041% per trade said it was too thin to pay, and net was negative at every capital level. Verdict: friction. That t assumed a beta of 1.
On 2026-10-30, after an audit of the lab's methods, it was re-measured on the full history with Jensen alpha and a standard error clustered by calendar date, against a threshold of 1.96 because two repaired verdicts were measured together. Beta came out at 1.165. Net of a 20%/yr borrow cost, alpha was −0.428% per trade with t −1.233; with free borrow it would have been +2.15. Gate 8 failed (t −0.957), both halves were negative (−2.252 and −0.428), and each trade lost about $5.5 after costs. The verdict moved from friction to no alpha, and the June reading that the thesis was vindicated was withdrawn.
The record
Thesis
Investors underreact to earnings surprises: after a material EPS miss, analysts revise down and institutions rebalance over weeks, so the price keeps drifting down. Triggering on the surprise itself (not on the price drop) aims at the misses where the first-day reaction was too small.
Specification
- signal
- EPS surprise < −10% vs consensus, signal the day after the report
- universe
- US stocks, price ≥ $10, dollar volume ≥ $5M
- entry
- next open, short
- exit
- 20 sessions, 2.5×ATR stop, 4×ATR target
- ranking
- worst miss first
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 ≥ 1.96 (Bonferroni, 2 repairs measured together), borrow 20%/yr
t (calendar SE)-1.23alpha / trade %-0.43beta+1.17trades6,906
Re-measured 2026-10-30 after the method audit (full history, calendar standard error). With zero borrow cost t is +2.15 — the alpha verdict hinges on a borrow rate the lab has never measured.
Gate 8 — Capture-preserved alphafail
Does the subset of trades a real portfolio can actually take keep the alpha?
t_cap ≥ 1.65·√(n_cap/n_eng) and ratio ≥ 0.50
t captured-0.96capture ratio+0.00
Gate F — Frictionfail
Is the alpha per trade large enough to pay for its own execution?
alpha/trade ≥ 0.30% and net $/trade > 0 at the declared sizing
alpha / trade %-0.43net usd per trade-5.52
Stabilityfail
Is the alpha present in both halves of history?
t > 0 in both halves
t 2014 2019-2.25t 2020 2026-0.43
Why it died
The June verdict said the alpha was real (t 5.28) but too gentle to pay. That t assumed beta = 1. Measured properly in October, beta is 1.165 and the alpha is not there once a 20% borrow cost is charged; with no borrow cost it would barely pass. Either way the trade loses money at small size: about $5.5 per trade, because a fixed commission on a small ticket swamps an edge of a few hundredths of a percent.
What it taught
It was archived for the wrong reason — a gross-return rule that the lab's own surviving strategy would also fail — and the right reason turned out to be the same answer. Repairing a verdict is worth it even when it does not flip: the June claim that the thesis was "vindicated" did not survive.
Verdict history
- 2026-06-03Pre-registered: short side of PEAD, surprise-only trigger
- 2026-06-03Single run: alpha t 5.28 and Gate 8 pass, but gross +0.041%/trade and negative net at every capital level Eaten by friction
- 2026-10-30Re-measured after the method audit: Jensen t −1.23 at 20% borrow (+2.15 at 0%), beta 1.17; fails Gates 7, 8, F and stability No alpha