Pre-registered 2026-10-23 · 2 runs · data snapshot 2026-09-19T090800Z
The idea
The same attention story, aimed at its loudest moment. A stock that gaps up more than 5% overnight right after earnings, on twice its usual volume, is on every screener. Retail buying on that attention should overshoot, and a short at the next open should collect the fade.
There was a second reason to try it. The lab had already measured the long version of the same signal, buying the earnings gap, and it lost money with t −4.02. If buying loses that reliably, shorting looked like it should win.
How it was tested
Tested at two horizons. A 20-session version was pre-registered on 2026-10-14 (commit d36a1c8). The 2-session version followed on 2026-10-23, declared expected to fail because the horizon change came after the first result; the family threshold counted both runs (k = 4, t ≥ 2.24). Signal: within 3 days of earnings, overnight gap above 5%, relative volume above 2, price at least $10, $5M of daily dollar volume. Short at the next open, out after 2 sessions or at a 3×ATR stop, largest gaps first, up to 20 a day. Borrow at 20%/yr, 5 bp slippage per side, per-share commission.
Where it died
The 20-session version had no alpha even before borrow: t +1.79 with free borrow, −6.60 at 20%/yr, with both halves negative. The 2-session version did no better. Gate 7 at t −0.83 and −0.057% per trade, with a beta of 1.27: shorting these names is mostly a bet against a stock that moves more than the market. 2014-2019 was negative (t −2.21), and the trade lost $1.47 after costs.
The only gate it passed was overlap: 9.3% of its trades coincide with the lab's surviving short. That pass means nothing here; there was no alpha to be redundant with. The family is closed at both horizons.
The record
Thesis
A large overnight gap up on high volume right after an earnings report draws retail attention buying that overshoots; shorting the gap captures the fade as the attention wears off.
Prior: Expected to fail at 2 sessions: the 20-session version had no alpha even before borrow was charged, and the horizon change was chosen after seeing that result.
Barber, B. & Odean, T. (2008). All That Glitters: The Effect of Attention and News on the Buying Behavior of Individual and Institutional Investors. Review of Financial Studies 21(2).
Specification
signal
within 3 days of earnings, overnight gap > 5%, relative volume (vs 20-day average) > 2
universe
US stocks, price ≥ $10, 20d dollar volume ≥ $5M
entry
next open, short
exit
2 sessions or 3×ATR stop, no target
ranking
overnight gap, largest first, max 20 new positions/day
costs
borrow 20%/yr, 5 bp slippage per side, tiered per-share commission computed trade by trade
caveat
Mirror image of a long already measured in the lab (earnings reaction drift, t −4.02), with a stop added.
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 ≥ 2.24 (Bonferroni, k = 4 including the 20-session runs), net of 20%/yr borrow
t (HAC)-0.83alpha / trade %-0.06beta+1.27trades9,101
Gate 8 — Capture-preserved alphafail
Does the subset of trades a real portfolio can actually take keep the alpha?
scaled t_cap and ratio ≥ 0.50
t captured-1.38capture 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 small-account sizing
alpha / trade %-0.06net usd per trade-1.47
Stabilityfail
Is the alpha present in both halves of history?
t > 0 in both halves
t 2014 2019-2.21t 2020 2026+0.26
Own alphapass
Is the alpha new, or already captured by a strategy that passed?
≤ 30% of trades overlap the lab's surviving short
overlap pct+9.30
Why it died
No alpha at either horizon. At 20 sessions: t −6.60 at 20%/yr borrow and only +1.79 even with free borrow. At 2 sessions: t −0.83, beta 1.27, negative in 2014-2019. The long it mirrors had t −4.02, but the short does not inherit it: earnings gaps that keep running jump the short's stop.
What it taught
A short is not the negative of a long. Adding the stop that any sane short needs changes which trades exist, and here it removed exactly the trades that made the long lose. "The long side has significant negative alpha" was not evidence for the short.
Verdict history
2026-10-1420-session version, pre-registered (commit d36a1c8): t −6.60 at 20%/yr borrow, alpha −1.04%/trade, both halves negative No alpha
2026-10-232-session version: t −0.83, fails every gate except overlap; the family is closed at both horizons No alpha