Post-Earnings-Announcement Drift (PEAD): A Real, Decaying Anomaly With Numbers

PEAD is the finding that stock prices keep drifting in the direction of an earnings surprise for months after the announcement, instead of jumping to the new level immediately. It is one of the oldest documented anomalies (1968) and one of the best-replicated, but its magnitude has shrunk by roughly half to two-thirds since the 1980s original.

Origin and original magnitude (Ball-Brown 1968, Bernard-Thomas 1989)

Ball and Brown (1968) first showed that markets do not fully incorporate earnings-announcement information immediately (Wikipedia PEAD article). Foster, Olsen, and Shevlin (1984) replicated an effect of similar magnitude.

Bernard & Thomas (1989) established the canonical decile-sort design: a strategy long the highest decile of unexpected earnings and short the lowest decile, measured using Standardized Unexpected Earnings (SUE), which ranks firms by decile spread between highest and lowest surprise (Wikipedia Earnings Surprise article). The 1989 paper itself reported the spread's *consistency*, not a single quantified annualized return figure; the specific magnitude (~8-9% quarterly) was reported in Bernard & Thomas's 1990 follow-up, below (Wikipedia PEAD article distinguishes the two papers' findings). The paper has 3,622+ Google Scholar citations, among the most-cited results in empirical asset pricing.

Bernard & Thomas (1990) refined the estimate to ~8-9% quarterly abnormal returns (compounding to roughly 36-41% annualized before costs), over a 1974-1985 sample (48 quarters), and found the top-minus-bottom decile spread was positive in 41 of 48 quarters — a strong within-sample persistence check. They also documented the earnings-surprise autocorrelation structure that underlies the drift: first-lag correlation 0.34, second-lag 0.19, third-lag 0.06, fourth-lag -0.24 (a reversal at the four-quarter mark).

Proof regime

peer-reviewed, high-citation, multi-decade replicated — the strongest tier this wiki recognizes (see Proof Regimes: Peer-Reviewed vs Practitioner-Attested vs Unverified — how to grade the source of a claimed edge).

Effect size today vs the 1980s: documented decay

The magnitude has fallen substantially and this decay is itself part of what the effect is: from ~5% (1980s/1990s) to ~3% or lower by the late 2010s, attributed to increased arbitrage activity and to declining persistence of earnings themselves (Wikipedia PEAD article). This is a smaller, later-period estimate than the 1989/1990 papers and should not be conflated with them.

Recent large-sample evidence (Garfinkel, Hribar, and Hsiao 2024) reports 5.1% risk-adjusted three-month returns, translating to over 20% annually — closer to the original magnitude than the "3% or lower" decay estimate, which the source material does not reconcile; treat the spread across studies (8.76% to 43.08% annual, depending on methodology and sample, per Wikipedia) as evidence that PEAD's measured size is highly sensitive to sample period and construction, not a single stable number. This is the pattern this wiki tracks under Out-of-Sample vs Post-Publication Decay: The Two Numbers That Tell You If a Premium Is Real: an anomaly can decay in aggregate while still showing up large in some slices.

Two mechanism-level 2020s studies confirm the effect persists structurally rather than as noise: - A 2020 machine-learning study (XGBoost, Russell 1000, 1997-2018, arXiv 2009.03094) finds drift direction is "driven by different factors for stocks from different industrial sectors and in different quarters" — i.e., PEAD is not one uniform mechanism, which complicates building a single static rule around it. - A StockTwits-based study (2010-2021, arXiv 2512.00280) finds long-horizon retail investors show "larger initial reactions and pronounced Post-Earnings Announcement Drift," with a zero-cost strategy yielding 0.43% per month risk-adjusted alpha (~5.2% annualized) — a small but statistically distinguishable modern-era number, roughly a fifth of the 1989 estimate (about 4.8x smaller, not a full order of magnitude). - Higher manager non-response rates on earnings calls correlate with "greater post-earnings announcement drift, higher return volatility, increased trading volume, and wider bid-ask spreads" (arXiv 2505.18419) — a candidate mechanism (information incompleteness) and a warning that the highest-drift names are also the highest-cost names to trade.

Transaction cost sensitivity

The holding period convention is 60 trading days (~3 months), which is short enough that turnover from quarterly rebalancing is a real cost consideration, not a rounding error. The original 25%-annualized (Bernard & Thomas 1989) figure is explicitly before transaction costs and provides a large buffer relative to typical round-trip equity costs; the modern ~3%-or-lower estimate does not have the same buffer and is "likely more sensitive to costs than the historical 5% effect" (Wikipedia PEAD article, decay discussion — this is the source's own framing, not an independently derived cost model).

UNVERIFIED / gap in these notes

no source found gives a specific breakeven cost (in basis points) at which PEAD profitability disappears. The one adjacent, concrete signal is that periods of high drift also show "wider bid-ask spreads" (arXiv 2505.18419), meaning the trades with the largest apparent edge are also getting more expensive to execute — the edge and the cost may be rising together, which a naive gross-return backtest would not show. See Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one for how this class of gap is generally handled, and Computing a Strategy's Transaction-Cost Threshold — the four-step check that decides whether a documented edge is tradable for the general method to close it — applying that method to PEAD specifically is unfinished, not because the technique failed but because the source material for this page did not carry the number.

What does NOT work

- Treating the 1989 magnitude (25% annualized, before costs) as the number to expect today: contemporary estimates cluster far lower (0.43%/month ≈ ~5%/year in the 2010-2021 retail-flow study; "3% or lower" by the late 2010s per Wikipedia's decay discussion). Anyone building on this page should use the 2020s numbers, not the 1989 headline figure, and should still expect them to erode further under costs not yet quantified in these sources. - Assuming SUE is one static rule: the ML evidence (arXiv 2009.03094) shows the drift's drivers differ by sector and by quarter, so a single fixed cross-sectional rule calibrated on one period/sector mix will not necessarily transfer.

Related

- Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one — this page documents that PEAD's headline return is pre-cost and that high-drift names carry wider spreads, but stops short of a breakeven number; that page is where the general accounting method for closing this gap lives. - Out-of-Sample vs Post-Publication Decay: The Two Numbers That Tell You If a Premium Is Real — PEAD is a textbook case of the decay this concept page covers: ~5% (1980s/90s) to ~3%-or-lower (late 2010s), and the objective's requirement to report post-publication persistence is answered here with the actual shrinking numbers rather than a "still works" claim taken at face value. - Short-Term (1-Week/1-Month) Reversal — Large in 1962-1986, Roughly Zero Unconditionally After 2000 — a contrasting short-horizon equity anomaly whose net-of-cost survival is far more doubtful; useful comparison for how differently two anomalies with similar-sounding holding periods can hold up. - What Counts as an Edge Here: The Evidence Bar This Wiki Applies to Every Technique — PEAD clears this wiki's bar on effect size, proof regime, and cross-decade persistence, but is only partially verified on cost survival: the worked example there can use PEAD as the case where three of four requirements are documented and one is an open gap.