Post-Earnings Announcement Drift: When a Beat Sells Off

Post-earnings announcement drift is the empirical tendency of a stock to keep moving in the direction of its earnings surprise for weeks after the report. Firms that surprise to the upside keep outperforming; firms that surprise to the downside keep underperforming. Ball and Brown (1968) first documented that a large fraction of the price adjustment to annual earnings occurs after the announcement date rather than on it. Foster, Olsen, and Shevlin (1984) established the effect on quarterly data and tied its magnitude to the size of the surprise, and Bernard and Thomas (1989, 1990) pinned down the mechanism: investors treat seasonally-differenced quarterly earnings as though they follow a random walk, when in fact the surprises are positively autocorrelated across the next three quarters and negatively correlated at the fourth lag.

That misspecification is the engine. If a firm beats this quarter, the beat itself raises the probability of a beat next quarter, but a market that models earnings as a random walk does not fully price that implication on the day. Price then catches up as the subsequent quarters confirm the pattern — a delayed response to information already in hand, not a risk premium. Bernard and Thomas (1990) show the drift concentrates around the next earnings date, exactly where the naive random-walk forecast is falsified, which is difficult to reconcile with a risk-based story.

PEAD is a persistent under-reaction, and its survival is the interesting part: the anomaly is transparent, published for decades, and still detectable, which points to frictions rather than a pricing error that smart money simply hasn't noticed.

The surprise is quantified as standardized unexpected earnings (SUE): reported earnings minus expected earnings, divided by the standard deviation of that forecast error. Expected earnings come either from a seasonal-random-walk time-series model (this quarter versus the same quarter a year ago, adjusted for drift) or from the consensus of analyst estimates; Livnat and Mendenhall (2006) show the analyst-based SUE produces a larger and more reliable drift, because analyst forecasts better approximate the market's true expectation than a mechanical time-series model does. Scaling by the dispersion of the error is what makes surprises comparable across firms with different earnings volatility — a $0.05 beat is a large surprise for a utility and noise for a semiconductor name.

The empirical anchor is the SUE-decile sort. Bernard and Thomas (1989) rank firms into ten portfolios by SUE each quarter, go long the highest decile and short the lowest, and hold for sixty trading days after the announcement. That strategy generated an abnormal return on the order of 4 percentage points per quarter — roughly 18% annualized before costs in their sample — with returns rising monotonically from the bottom decile to the top. The drift is not a knife-edge effect concentrated in a handful of names; it is a smooth, ordered relationship between surprise magnitude and subsequent return.

This framework also sharpens what an overshoot is. In a clean PEAD episode, the sign of the day-one reaction matches the sign of SUE, and the drift extends that same-signed move. When the day-one reaction has the opposite sign to SUE — a positive surprise met by a sharp decline — the standard drift model does not apply, and the move is being driven by something outside reported earnings. Abbott's SUE on 17 July was positive; its price reaction was sharply negative. That divergence is the tell that the print was being priced on guidance and positioning, not on the earnings line.

A single earnings-day return compresses at least three distinct pieces of information into one number, and they need not agree in sign. First is the reported result versus the pre-announcement expectation — the SUE, the object PEAD is about. Second is the forward guidance and management commentary, which revises the expected path of future earnings and often carries more weight than the trailing quarter. Third is positioning: how the stock was set up going in, whether the buy-side had already discounted a strong quarter, and what the options market had priced for the move. The close aggregates all three against what was already in the price.

The guidance channel is why a beat can sell off. A firm can clear the consensus on revenue and EPS yet reaffirm full-year guidance where the market had positioned for an increase. The reported numbers are good; the news relative to expectations is negative, because the marginal information — the thing not already discounted — was the forward outlook, and it came in below the whispered bar. This is the mechanism behind 'buy the rumor, sell the news': the quarter was anticipated and pre-traded, so the absence of an upgrade becomes the surprise.

That produces a recognizable and, importantly, a diagnosable setup: positive SUE, negative price reaction, with the decline traceable to an expectations gap on guidance rather than to any deterioration in run-rate earnings power. The open question is whether the day-one move initiates a genuine drift lower — the market correctly reading a problem the headline masked — or overshoots a level the fundamentals still support and then reverts.

PEAD is under-reaction that continues. Its behavioral counterpart is overreaction that reverses (De Bondt and Thaler, 1985), and the two imply opposite positions, so discriminating between them is the entire exercise rather than a refinement of it. The useful question is what the day-one move is pricing. If the reaction reflects a real change in the earnings trajectory, continuation is the base case and the drift extends the move. If it reflects a sentiment overshoot on a salient but secondary factor — a guidance non-raise, one soft segment, a headline unrelated to run-rate earnings — while SUE is positive, a short-horizon reversion toward the fundamentals is the more likely path.

Chan, Jegadeesh, and Lakonishok (1996) sharpen the distinction empirically: past return momentum and past earnings surprise each predict future returns after controlling for the other, and both reflect gradual response to news rather than the correction of an overreaction. Their evidence says the durable, same-signed continuation belongs to genuine earnings surprises; a move whose sign contradicts the surprise sits outside that pattern and is a candidate for reversion, not drift.

The tape leaves fingerprints. A climactic, high-volume gap that overshoots a defensible fundamental level, with no change to the earnings trajectory and no downward revision to forward numbers, has the signature of a positioning flush rather than a re-rating. Volume five times the run-rate is consistent with forced or crowded selling exhausting itself, not with long-horizon holders methodically re-underwriting the franchise lower. None of this guarantees reversion; it identifies the configuration in which the odds favor it, provided a hard invalidation separates the overshoot thesis from a genuine drift lower.

A textbook-efficient market would arbitrage a published, decades-old drift to zero. It has not, and the literature explains why through three reinforcing frictions. Limited attention: investors under-weight earnings information, and the drift is larger precisely when attention is scarce — for firms with low institutional following and on days crowded with other announcements. The random-walk misspecification Bernard and Thomas (1990) document is itself a form of inattention to the autocorrelation structure of earnings.

Transaction costs and liquidity gate the arbitrage. Ng, Rusticus, and Verdi (2008) show the drift is concentrated in high-transaction-cost firms and that trading costs materially erode the paper profits of a SUE strategy — the anomaly is largest exactly where it is most expensive to exploit. Sadka (2006) goes further, arguing that a portion of the PEAD return is compensation for liquidity risk: the strategy loads on unexpected variation in market-wide liquidity, so part of what looks like free alpha is a premium for bearing that exposure. Mendenhall (2004) rounds this out by showing the drift is strongest for stocks with high arbitrage risk — those without close substitutes, where an arbitrageur cannot hedge the idiosyncratic component and therefore demands a wider margin before acting.

The synthesis is that PEAD persists because the frictions that would close it — attention, cost, liquidity risk, un-hedgeable idiosyncratic risk — are largest in the same names where the drift is largest. For a liquid mega-cap like Abbott, those frictions are small, which is one more reason the July 2025 episode is better read as a fast overshoot-and-revert in an efficiently-traded name than as a slow, friction-protected drift.

Abbott reported second-quarter 2025 results before the open on 17 July. The quarter was strong on the numbers: sales up 7.4% year over year, adjusted EPS of $1.26, up roughly 10.5%, with organic growth in the mid-to-high single digits and margin expansion — a beat-to-in-line print from a high-quality compounder, and a positive SUE.

The stock fell 8.5%, closing at $120.51 against the prior night's $131.74, on about 27.6 million shares — roughly five times its recent run-rate near 5 million. The cause was not the quarter but the outlook: Abbott reaffirmed full-year adjusted EPS guidance of $5.10-5.20 rather than raising it, and the next-quarter guide sat a touch below the Street. Into a stock that had drifted lower for two weeks going into the print, a strong quarter without a raise was read as a disappointment, and the reaction overshot the fundamentals.

Read retrospectively, the diagnosis follows directly: positive SUE, negative reaction, an expectations gap sourced to guidance rather than results, and a high-volume, climactic gap that undercut a defensible level with no deterioration in the earnings trajectory. Sign of reaction opposite to sign of surprise — the marker that this was not a drift setup but an overshoot candidate.

The retrospective classification, written in June 2026, is a BUY referenced to the earnings-day close of $120.51 on 17 July, with an estimated return band of +3.9% to +6.8% over 5-12 trading sessions and an invalidation at a daily close below the $119.77 earnings-day low. Each parameter is derived, not asserted.

Entry at the earnings-day close, rather than intraday, requires the overshoot to be a settled mark: a climactic gap that has printed its low and closed off it, so the classification fades a completed flush rather than a falling knife. The horizon is short by construction. Reversion of a positioning overshoot is a days-to-two-weeks phenomenon; hold it longer and you drift into the window where the genuine earnings surprise — the positive SUE — reasserts as PEAD, and where the next scheduled catalyst can dominate. The 5-12 session band is that reversion window, not an open-ended target.

The return band is scaled to the gap. An 8.5% overshoot that need not fully retrace to be monetized implies a partial recovery toward the pre-gap reference; +3.9% to +6.8% captures roughly the lower-half-to-two-thirds of the drop, which is the tradable, higher-confidence portion rather than a call for a complete round-trip. The invalidation is placed at the earnings-day low because a close beneath it changes the interpretation: it says the selling was not a positioning flush that exhausted itself but the leading edge of a market pricing a real fundamental problem — the overshoot thesis is falsified, and the correct read flips to a genuine drift lower.

The invalidation distance defines the risk unit. Entry at $120.51 against a close-based stop at $119.77 is a defined-risk structure of about 0.6% to the invalidation trigger, against a projected +3.9% to +6.8% band — a favorable asymmetry, but one that only holds if position size is set to the distance to invalidation rather than to conviction. A tight invalidation permits a larger position for the same dollar risk; it does not license a larger dollar risk. The discipline is that the stop distance, not the appeal of the thesis, sizes the trade.

Two things would have confirmed a genuine drift lower and voided the fade. A close below $119.77 is the mechanical trigger this classification sets, but the fundamental version is a downward revision to the forward earnings path: a guidance cut rather than a reaffirmation, a deteriorating segment that changes run-rate earnings power, or continuation selling on rising rather than exhausting volume. Any of those would have re-signed the episode from overshoot to negative drift — the case where the market's day-one read was correct and the beat masked a real problem. Distinguishing that path from a positioning flush before the fact is the judgment this kind of classification is built on; here it is applied retrospectively.

Because the reversion window is finite, the position carries a time stop as well as a price stop. If the band does not begin to fill within the 5-12 session horizon, the reversion premise decays toward noise regardless of price, and the classification steps down rather than converting silently into an open-ended directional bet.

Price stabilized immediately and drifted higher. Abbott recovered to roughly $123.67 the next session, reclaimed the $125-126 area within a week, and reached a window high of $128.55 on 30 July — up 6.7% from the $120.51 earnings-day close at about session nine, inside the classification's +3.9% to +6.8% band and tagging its upper half. The $119.77 earnings-day low was never threatened; the invalidation did not trigger.

Two forces operated in the same direction and are not contradictory: the market overshot on the secondary factor — the guidance non-raise — and then, as that overshoot worked off, under-reacted to the primary one, the genuine positive surprise, so the net path was a reversion up toward the fundamentals. The dominant effect over this window was the overshoot correction; a residual PEAD-style drift on the positive SUE was consistent with, but not separable from, the same recovery.

The claim here is bounded. This is not evidence that a strong stock always bounces after an earnings-day drop; plenty of post-earnings declines are the start of a real drift lower, because the market correctly read a deterioration the headline masked. The edge is not 'buy every beat that falls.' It is the ability to distinguish a guidance-driven overshoot on an intact franchise from a genuine, continuing negative surprise, and to fix the invalidation that separates them before entering. 'The company beat but the stock fell' is where the analysis starts, not where it ends — and this was a short-horizon reversion of a positioning overshoot, not a view on Abbott's full-year trajectory or its longer-term fundamentals.

Post-earnings announcement drift (PEAD) — the tendency of a stock to keep moving in the direction of its earnings surprise for roughly one to three months after the report, driven by under-reaction to the news.

Standardized unexpected earnings (SUE) — reported earnings minus expected earnings, scaled by the standard deviation of the forecast error; the standard measure of surprise magnitude, comparable across firms of different earnings volatility.

SUE-decile sort — ranking firms into ten portfolios by SUE and going long the top, short the bottom; the canonical way the drift is measured, and monotonic in surprise size.

Under-reaction vs overreaction — under-reaction produces continuation (drift, same sign as the surprise); overreaction produces reversal (opposite sign). A single post-earnings move can contain both.

Guidance gap / expectations gap — a strong quarter paired with forward guidance that disappoints relative to what was priced in; the marginal information that moves price when the reported numbers were already discounted.

Limits to arbitrage — the frictions (transaction costs, liquidity risk, un-hedgeable idiosyncratic risk, arbitrage risk) that let a published anomaly like PEAD persist rather than being traded to zero.

Invalidation level — a price set at classification (here the earnings-day low) that, once broken on a closing basis, falsifies the overshoot thesis and re-signs the episode as a genuine drift lower.