Stocks that are already rising in the days before their earnings report might keep rising into it: leaked information, upward revisions, funds positioning ahead of an expected beat. Exiting before the announcement avoids the event itself.
It was included on purpose as the weakest of a set of four event strategies. The literature is thin and mixed, and Reg FD (2000) was written to shut down exactly this kind of leak. It served as a planted control: if it had survived alongside the others, the engine would have been suspect.
How it was tested
From the lab's early phase, May 2026, without pre-registration. Signal: 5 to 10 days before earnings, a 20-day momentum rank above 70 and an RSI(14) rank between 50 and 90. Universe: US stocks at $15 or more with $10M of daily dollar volume. Buy at the next open, hold 3 sessions, always out before the report, with a 1.5×ATR stop and a 2.5×ATR target. At most 20 trades a day, filled alphabetically. Judged on a Sharpe split, 2014-2019 against 2020-2025.
Where it died
The first warning came before the real test. A smoke test on 2024 alone showed a Sharpe of 0.92, which is the kind of number a weak thesis should not produce. It was flagged as suspicious instead of celebrated.
The full run settled it. The Sharpe was −0.128 in 2014-2019, on 14,919 trades, and 0.250 in 2020-2025: a negative design period and a held-out half that looks like noise, not recovery. The retroactive check against SPY on 2026-05-27, with beta assumed to be 1, was the worst of the set at t −7.22. The control died as expected.
The record
Thesis
Stocks already rising in the days before their earnings report keep rising into the event — information leakage, upward analyst revisions and institutional positioning ahead of an expected beat. Exit before the announcement to avoid the event itself.
Prior: Included on purpose as the weakest hypothesis of a set of four event strategies: thin and mixed literature, and a mechanism that Reg FD (2000) was written to shut down. If it had survived alongside the others, the lab would have suspected the engine.
So, E. & Wang, S. (2014). News-Driven Return Reversals. Journal of Financial Economics.
Kim, O. & Verrecchia, R. (1991, 1997). Models of pre-announcement information.
Specification
signal
5 to 10 days before earnings; 20-day momentum rank > 70; RSI(14) rank between 50 and 90
universe
US stocks, price ≥ $15, dollar volume ≥ $10M
entry
next open
exit
3 sessions (always before the report), 1.5×ATR stop, 2.5×ATR target
daily cap
20 trades/day, filled in alphabetical order (no ranking)
caveat
From the lab's early phase: not pre-registered, judged on an in-sample / held-out Sharpe split.
Gates
Raw significance (legacy)fail
Are the strategy's raw returns statistically different from zero?
positive in-sample Sharpe, held-out degradation better than −50%
A held-out Sharpe test, not a t-stat — the instrument of May 2026.
Market-adjusted alpha (legacy)fail
Does the strategy beat SPY over the same holding windows, assuming beta = 1?
alpha t ≥ 1.65
alpha t-7.22
Retroactive check, beta assumed = 1; not re-derived with Jensen alpha.
Why it died
Negative Sharpe in the design period (−0.128) and a ~50% win rate in both halves; the positive held-out Sharpe (0.250) is noise, not a recovery. The retroactive alpha check against SPY was the worst of the set, t −7.22. The signal is effectively "buy a rising stock", with nothing specific to the earnings calendar.
What it taught
A planted weak hypothesis is a cheap calibration of the whole pipeline: it was expected to die, and it did. A single-year smoke test on 2024 had shown Sharpe 0.92 — exactly the kind of number that is too good for a weak thesis.
Verdict history
2026-05-09Smoke test on 2024 alone: Sharpe 0.92 — flagged as suspicious
2026-05-09Held-out test: Sharpe −0.128 in 2014-2019, +0.250 in 2020-2025 No alpha
2026-05-27Retroactive alpha-vs-SPY check: t −7.22 No alpha