nyasekai

Graveyard / Earnings / US stocks

Post-earnings announcement drift

Do stocks keep drifting after a good earnings surprise?

Family
Earnings
Verdict
Dead
Cause
No edge
Buried
14 October 2026

The idea

Stocks that beat earnings expectations tend to keep rising for weeks, because the market adjusts slowly. Buy the positive surprises, short the negative ones, hedge the market.

How we tested it

DataTest 1: 164 large caps, 7,568 earnings events, trade the sign of the surprise or the announcement-day return from the reaction-day close, hedged with SPY, 1 to 5 sessions. Test 2: about 3,300 US stocks, surprise measured against a seasonal random walk on reported EPS (no analyst consensus is available for free), deciles, entry at the open after the report, holds of 5, 20, 40 and 60 sessions, long-short and long-only.
PeriodTest 1: 2010 to 2026. Test 2: June 2020 to October 2026, in-sample before 2023, out-of-sample from January 2023.
CostsTest 1: 0.22%. Test 2: 10 bp per side, 1 bp on the hedge, 0.5% a year borrow, with a 1.5x stress view.
HoldoutTest 2 required positive in-sample and out-of-sample results, a pass against a shuffled-date null, and at least 60% of years positive. 48 cells.
ControlTest 2 moved each entry to a random session of the same stock (300 repetitions).

Result

Test 1 variants passing0 of 4
Test 1 gross, 5 days+3 to +7 bp
Test 1 net of cost−15 to −19 bp (t −2.9 to −6.6)
Test 1 profit factor0.76 to 0.89
Test 2 cells passing0 of 48
Test 2 best null p value0.3
Long-only top decile vs SPY, out-of-sample−65 to −354 bp, every hold
Small and mid caps, 40 to 60 day long-short, out-of-sample+77 to +119 bp, but in-sample −27 to −98 bp

Why it failed

In large caps the drift is gone: gross +3 to +7 bp is less than one round trip. In small caps, the only positive out-of-sample cells flip sign between periods, which is a change of market regime, not a drift. The 2020-2022 short leg did not pay and the 2023-2026 short leg did. Current tickers were used, which flatters longs and hurts shorts.

What would make it worth retesting

Not for large caps. For small caps only with real trading costs and a proper analyst-consensus surprise measure.

Past results of a backtest say nothing certain about the future.