Earnings Announcements

The earnings-announcement event study is the founding application of the event study methodology and remains the canonical test bed of empirical capital-markets research. The question it answers is direct: when a firm reports its quarterly or annual earnings, does the new number carry information the market did not already have, and how fast does the share price absorb it? Because prices in a reasonably efficient market impound public information quickly, the abnormal return earned in a short window around the report date measures the information content of the earnings surprise, while the abnormal return earned over the following weeks measures how completely the market reacted. Earnings announcements are special because they generate two economically distinct signals from one event: a sharp immediate reaction and a slow, predictable continuation called the post-earnings-announcement drift (PEAD). That makes the event study the right tool, and it makes this event type the standard teaching example for the methodology itself.

The seminal arc runs from Ball and Brown (1968), the first event study to show that accounting earnings have information content, through the standardized-unexpected-earnings design of Foster, Olsen and Shevlin (1984), to the definitive characterization of drift by Bernard and Thomas (1989, 1990), surveyed authoritatively by Kothari (2001). The stylized facts these papers establish are summarized next.

What the research shows

Three robust facts define this literature: earnings have information content, the immediate reaction is incomplete, and cumulative abnormal returns drift in the direction of the surprise for weeks afterward.

Information content: prices react to the sign of the surprise

Ball and Brown (1968) established the foundational result on a sample of 261 NYSE firms (1957 to 1965): firms with positive earnings forecast errors earn positive abnormal returns and firms with negative errors earn negative abnormal returns. Strikingly, roughly 85% to 90% of the information in annual earnings is already anticipated by the market before the release, as prices drift toward the eventual good or bad news over the preceding twelve months. A measurable reaction nonetheless occurs in the announcement month, and a residual drift continues for up to a year afterward, the first documentation of what became PEAD. The directional "sign match" (positive surprise to positive return, negative surprise to negative return) is the bedrock information-content finding on which everything else builds.

Measuring the surprise: standardized unexpected earnings (SUE)

The independent variable in this literature is the earnings surprise, almost always operationalized as standardized unexpected earnings (SUE): actual earnings minus expected earnings, scaled by a measure of dispersion. There are two standard ways to form the expectation. The time-series approach treats quarterly earnings as a seasonal random walk and computes the surprise as the change from the same quarter one year earlier, scaled by the standard deviation of past seasonal differences (Foster, Olsen and Shevlin, 1984). The analyst-based approach defines the surprise as actual earnings minus the latest median I/B/E/S consensus forecast, scaled by price or by forecast dispersion. The choice matters: Livnat and Mendenhall (2006) show that analyst-based SUE generates materially larger drift than time-series SUE, and that a two-way sort on both yields the strongest effect. Firms are then sorted each quarter into SUE deciles, and the headline statistic is the abnormal return of the long-top, short-bottom hedge portfolio, evaluated for monotonicity across deciles.

The immediate reaction: short and sharp

Most of the immediate price response occurs in a tight three-day window around the announcement. A (-1,+1) or (0,+1) cumulative abnormal return (CAR) is the standard measure of the immediate information content, with the highest-surprise decile earning announcement-window CARs on the order of a few percent and the lowest-surprise decile a symmetric negative reaction. This short-window reaction is robust and is what most one-shot event studies would stop at. The point of an earnings study is that the reaction does not end there.

Post-earnings-announcement drift (PEAD)

The defining phenomenon of this event type is that cumulative abnormal returns continue to drift in the direction of the surprise for weeks to a quarter after the announcement. Bernard and Thomas (1989) provide the definitive characterization: the high-minus-low SUE decile spread earned roughly 4.2% over the 60 trading days following the announcement, and the spread was positive in 41 of 48 quarters (1974 to 1985), including 11 of 16 quarters in which the market itself fell. That a strategy can be profitable in down markets makes a risk-premium explanation implausible, so PEAD is read as a delayed price response rather than compensation for risk. The effect is monotonic in SUE (higher surprise, larger subsequent drift), which is the hallmark used to validate it, and it is concentrated in small firms.

Bernard and Thomas (1990) pinned down the mechanism, the "naive-expectations" hypothesis. Three-day abnormal returns around the next four quarterly announcements are predictable from the current quarter's surprise: positive (and declining) around quarters t+1, t+2 and t+3, then negative around t+4, exactly mirroring the autocorrelation structure of seasonally-differenced earnings (lag-1 autocorrelation of about +0.34, declining to about -0.24 at lag 4). About 25% to 30% of the total drift is earned in the short windows around those subsequent announcements, which are only a small fraction of trading days. The interpretation is that investors price earnings as if they follow a seasonal random walk and fail to anticipate the predictable autocorrelation in quarterly earnings, a direct mispricing rather than a risk story. Kothari (2001), surveying more than a thousand papers, frames the short-window information-content test against the long-window drift test and identifies PEAD as the most serious surviving challenge to semi-strong market efficiency.

Cross-sectional drivers and limits to arbitrage

The drift is systematically larger for small-cap, illiquid, low-volume, low-analyst-coverage and low-institutional-ownership stocks, the very firms where trading frictions and limits to arbitrage prevent fast correction. Chordia, Goyal, Sadka, Sadka and Shivakumar (2009) show that PEAD concentrates in illiquid, high-spread names and that transaction costs consume roughly 70% to 100% of the gross long-short profits, so the anomaly can persist without being a free lunch, consistent with a transaction-cost-bounded form of efficiency. This is why the drift is sometimes called an "anomalous anomaly": it is unusually robust, survives risk adjustment (CAPM and Fama-French factors), is positive in most quarters including down markets, and is documented internationally.

The decline of the drift over time

The drift has weakened over recent decades. The high-minus-low SUE spread fell from roughly 5% in the 1980s and 1990s toward 3% or less by the late 2010s; PEAD began disappearing from non-microcap stocks around 2001 and was near zero for large caps by about 2006. The usual drivers cited are decimalization, Reg NMS and high-frequency trading, more arbitrage capital, and a richer information environment. Whether the decline reflects arbitrage or a fall in earnings persistence remains debated, and several studies find the drift survives in the hardest-to-arbitrage segments. For a modern study this is essential context: split the sample pre- and post-decimalization, report large-cap and small-cap subsamples separately, and do not assume the pre-2001 magnitudes hold today.

International evidence and joint signals

A substantial cross-country literature examines how the earnings-return relation varies with institutions. Many cross-country studies cannot determine whether investors actually use the information to price securities, so more recent work emphasizes investor reactions directly and the structural factors that drive differences in information content. DeFond, Hung and Trezevant (2007), examining more than 50,000 announcements across 26 countries, find that earnings announcements are more informative in countries with higher earnings quality and better-enforced insider-trading laws, evidence that the information content of earnings is conditioned by the reporting and investor-protection environment. A related design challenge is that earnings are rarely released in isolation: Anderson (2007) studies the joint release of earnings and dividends and shows the difficulty of isolating the dividend signal when both are made public at once, a confounding-event problem that recurs throughout this literature.

How to run this kind of event study

The general workflow (estimation window, expected-return model, event window, abnormal returns, significance testing) is covered in our Introduction to Event Study Methodology and the step-by-step Event Study Application Blueprint. Earnings studies add several event-type-specific design choices that matter a great deal in practice.

Event-date identification: use the press-release date

Use the actual press-release date, not the SEC filing date and not the fiscal-period end. The Compustat report date (RDQ) can lag the true release by one to five days, so best practice cross-checks Compustat RDQ against I/B/E/S and requires agreement within one calendar day, otherwise the announcement-window reaction is mismeasured. Day-0 timing is acute here: announcements made after the market close should be coded as the next trading day, which is the rationale for the (0,+1) convention. Getting the timestamp right is the single most important data-quality decision in an earnings study. Our Event Date Identifier (EDI) is built for exactly this problem.

The two-window design

Earnings studies are distinctive in requiring two windows by design. A short window, typically (-1,+1) trading days, measures the immediate information content. A separate long window, running from the day after the announcement through roughly +60 trading days (or until the next quarterly announcement, about +1 to +63), measures the drift. The long window deliberately starts after the immediate reaction so the two effects are not conflated. State both windows explicitly and report them separately, because they measure economically distinct quantities. The estimation window should provide sufficient pre-event data and end before the announcement.

SUE deciles and the hedge portfolio

Define the surprise explicitly and report robustness to the choice (time-series seasonal random walk versus analyst consensus error). Sort firms each quarter into SUE deciles or quintiles, then form a zero-investment portfolio that is long the top decile and short the bottom decile, held over the long window. The test statistic is the abnormal return of that hedge portfolio, and the monotonicity of returns across deciles is the validation that the effect is real and not driven by the extremes alone. Report both equal- and value-weighted hedge returns, since PEAD concentrates in small illiquid stocks and value-weighting sharply attenuates it.

Confounders, clustering and benchmark choice

Three confounders are acute for earnings. First, concurrent disclosures: earnings are often released jointly with dividend changes, guidance, buybacks or management commentary, so isolate the earnings signal by excluding firms with other major disclosures in the announcement window or by partitioning on whether the joint signal confirms or contradicts the earnings news (the Anderson dividend problem above). Second, announcement clustering: earnings bunch in calendar quarters, inducing cross-sectional correlation in residuals that inflates t-statistics, which calls for calendar-time portfolio regressions or standardized cross-sectional tests such as the BMP (Boehmer, Musumeci and Poulsen) statistic. Third, overlapping windows across a firm's successive quarters. On the model: market-model or characteristic-matched benchmarks are fine for the short window, but for the 60-to-90-day drift, size and book-to-market matching or Fama-French (and momentum) adjustment reduce bad-model bias; see Expected Return Models and the full battery in Significance Tests. CAR is standard for the drift; buy-and-hold abnormal returns (BHAR) are more sensitive to benchmark and skewness problems over long horizons (Kothari, 2001).

Statistical versus exploitable drift

Finally, distinguish a statistically significant drift from an economically exploitable one. Because realistic bid-ask spreads and transaction costs can consume 70% to 100% of the gross long-short profits (Chordia et al., 2009), report hedge returns net of trading costs and beware look-ahead bias from restated Compustat figures by using point-in-time data. A volume reaction (the Beaver information-content-via-volume result) is also worth examining: trading volume spikes at earnings even when the directional price effect is ambiguous.

Run it with our tools

The applications on this site implement the workflow above end to end:

• Abnormal Return Calculator (ARC) is the core tool for both windows. Supply an event file keyed to the press-release date, choose your estimation and event windows (a short (-1,+1) for the immediate reaction and a separate post-event window for the drift), pick an expected-return model (Market Model, CAPM, Fama-French 3-factor, Fama-French 5-factor, Carhart 4-factor, or comparison-period mean), and run the battery of parametric and non-parametric significance tests built to handle clustered, calendar-bunched earnings samples. Group your events by SUE decile to reproduce the long-top, short-bottom hedge comparison and check monotonicity.

• Event Date Identifier (EDI) helps pin down the true press-release date and reconcile RDQ versus I/B/E/S timing, the single most important design decision in an earnings study.

• Abnormal Volume Calculator (AVC) measures the trading-volume reaction (the Beaver effect), which corroborates that information arrived even when the price move is small. Abnormal Volatility Calculator (AVyC) captures the volatility response around the announcement.

• News Analytics (CATA) lets you build a surprise or tone signal directly from the earnings press release or call transcript instead of a numeric SUE, then use that sentiment as the sorting variable. This connects the content of the disclosure (see Henry, 2008) to a concrete abnormal-return workflow.

Related use cases

Earnings announcements are one of several corporate events analyzed with this methodology, and they share design questions with the others. See the closely related pages on Mergers and Acquisitions, Divestitures, Alliances and Joint Ventures, and Competitive Dynamics; the methodologically adjacent Litigation application; the use of drift signals in Investment Strategies; and the broader Comparative Event-Type Analyses. For the full catalogue, return to the Practical Applications overview.