One of the oldest documented anomalies in finance: stocks that beat earnings and get bought on the news
keep drifting up for weeks. We replicated it independently on ~930 earnings events over four years.
Result
| Configuration | Trades | Win rate | Avg/trade | Profit factor |
|---|---|---|---|---|
| Beat + positive reaction day, long only, 60-day hold | 420 | 59% | +6.9% | 2.89 |
| Long + short version | 504 | 57% | +5.2% | 2.14 (short side bleeds) |
| + big-surprise filter | 269 | 61% | +7.0% | 2.89 (filter adds nothing) |
The literature replicated cleanly: the long side is the edge; shorting the misses loses (as published);
extra filters don’t stack. A companion finding — selling puts on names that just beat — was the best
single-name put trigger we ever tested (87% win rate), because the gap risk is behind you the day
after earnings.
Where it stands
Adopted into the fund’s original build as an entry trigger, and it shaped two rules that outlived that
build: positions are multi-week holds (the drift takes 60 trading days, not 6), and earnings dates gate our
entries to this day. The v3 rework consolidated stock selection under the fundamental score, but PEAD remains
a validated trigger in the research library and its fingerprints are all over the current rulebook.
Update (Nov 9): the clean re-test killed it
The August test had two flaws we could not fix at the time: a small hand-picked universe (survivorship bias)
and only four years of data. In November we re-ran the idea properly — 26,173 filing events on
point-in-time S&P 500 membership, 2010–2026, real SEC filing dates, forward returns measured against the
eligible universe with month-clustered statistics. The result reversed: both extremes of the
announcement-reaction underperform the market over the following 3–6 months (top-reaction quintile
−1.7% over 6 months, t −4.2; bottom quintile −3.1%, t −6.8), and every
quality/growth interaction we tried was negative too. There is no drift left to harvest long-only in large
caps — the anomaly has been arbitraged away in this universe, exactly what the post-publication-decay
literature predicts. The verdict badge above now reflects the honest answer, and the earlier numbers stand
as a documented example of how survivorship bias manufactures an edge. What survives: the multi-week-hold
lesson and the earnings-date entry gate, both of which remain in the live rulebook.