Comparative Event Type Analyses

Most event-study work asks how the market reacts to one kind of corporate news. Comparative event-type analysis turns the question around and asks across kinds: among the many events a firm can announce (a merger, an earnings surprise, a buyback, a dividend initiation, a stock split, a product recall, a data breach), which ones move prices the most, in which direction, by how much, and how persistently? Because the event study reduces every announcement to a common, comparable currency (the abnormal return earned around a precisely dated event), it is the natural instrument for building a cross-event hierarchy and for taking a meta-analytic reading of the entire literature. This page is the comparative hub of the use-case cluster: it ranks the events, explains the sign and magnitude patterns, shows how to design a study that compares event types fairly, and links out to the individual event-type pages where each application is treated in depth.

The empirical record here is unusually clean. Event studies are the dominant tool in empirical corporate finance (a census of five top journals counted 565 event-study papers between 1974 and 2000, all built on the template of the founding stock-split study), and decades of replication have produced stable, citable magnitudes per event class. The result is a ranking that is robust enough to teach from, provided one respects two cross-cutting regularities (sign asymmetry and differences in the speed of adjustment) and one design discipline (holding the estimation method constant across event types). We take each in turn.

What the research shows: a cross-event hierarchy

Pooling the announcement-window evidence across event types yields a remarkably consistent ordering. Mergers and acquisitions dominate everything else at the single-firm level: target shareholders earn the largest reaction of any common corporate event, on the order of +16% to +30% over a short window, while acquirers, on average, roughly break even (Andrade, Mitchell and Stafford, 2001). One rung down sit the capital-distribution and signaling events: share-buyback announcements and dividend initiations each draw roughly +3% to +4% (Ikenberry, Lakonishok and Vermaelen, 1995; Michaely, Thaler and Womack, 1995). Stock splits and dividend increases are positive but smaller (~+1% to +3%), and earnings surprises scale with the surprise itself rather than carrying a fixed label-specific size (~+1% to +3% for good news, -1% to -5% for bad). On the negative side, reputational and operational shocks cluster between roughly -1% and -8%: product recalls at the larger end (-2% to -8%), earnings restatements, regulatory rejections and customer losses in the middle (-1% to -5%), data breaches similar (-1% to -5%), and ESG downgrades smaller (~-0.5% to -2%).

The single most authoritative cross-event source is Neuhierl, Scherbina and Schlusche (2013), which classifies a comprehensive corpus of corporate press releases into topic categories and ranks the market reactions directly. The strongest positive reactions come from positive earnings pre-announcements, share buybacks, regulatory (FDA) approvals, special dividends and spin-off intentions; the strongest negative reactions come from negative earnings pre-announcements, regulatory rejections, customer losses, product defects and earnings restatements. The same paper documents two corroborating facts that any comparative reading should carry: news of almost every type is followed by a near-universal decline in bid-ask spreads (announcements reduce information asymmetry, as disclosure regulation intends), and post-announcement return volatility tends to rise, because fresh news weakens investors' priors and makes prices more responsive to subsequent information.

Table 1: Announcement-window abnormal-return benchmarks by event type (rough cross-study consensus)

Event type	Typical announcement-window CAR	Sign / notes	Representative source
M&A: target	+16% to +30%	Large positive; higher in tender offers and contested deals	Andrade, Mitchell and Stafford (2001)
M&A: acquirer	~0% (often -1% to -3%)	Near zero on average; negative for all-stock bidders	Andrade, Mitchell and Stafford (2001)
M&A: combined entity	~+1.5% to +3%	Modest net synergy; gains accrue to the target	Andrade, Mitchell and Stafford (2001)
Share buyback (open-market)	~+3% to +4%	Positive; larger for infrequent repurchasers	Ikenberry, Lakonishok and Vermaelen (1995)
Dividend initiation	~+3% to +4%	Positive (signaling); omissions strongly negative (~-7%)	Michaely, Thaler and Womack (1995)
Stock split	~+1% to +3%	Positive; near-instant adjustment at announcement	Fama, Fisher, Jensen and Roll (1969)
Earnings surprise	+1% to +3% (good); -1% to -5% (bad)	Scales with the surprise; drifts ~60 days after	Ball and Brown (1968); Bernard and Thomas (1989)
Product recall / defect	-2% to -8%	Negative; reputational and operational	Neuhierl, Scherbina and Schlusche (2013)
Restatement / customer loss / regulatory rejection	-1% to -5%	Negative; among the largest negative categories	Neuhierl, Scherbina and Schlusche (2013)
Data breach	-1% to -5%	Negative over ~3 days	Neuhierl, Scherbina and Schlusche (2013)

Magnitudes are rough, sample-dependent consensus ranges intended for orientation, not point estimates. Windows differ across the underlying studies; see the methodology section on why raw cross-event magnitudes must be standardized before they are strictly comparable.

Sign and magnitude patterns: who wins, who loses, and by how much

The hierarchy is not just a list of sizes; it has a structure. Three patterns recur across event types and are the substance of a comparative reading.

Sign depends on the event and its direction, not on a uniform rule. M&A is the canonical illustration: the same transaction is strongly positive for the target and roughly zero (or negative) for the acquirer, so the wealth created is overwhelmingly transferred rather than newly generated, with only a modest positive combined return (Andrade, Mitchell and Stafford, 2001). Within a single event class the sign can be sharply asymmetric: dividend initiations earn about +3% but dividend omissions draw roughly -7%, a larger absolute reaction than the good-news case (Michaely, Thaler and Womack, 1995).

Bad news tends to be bigger and stickier than good news. Negative-news categories often produce larger absolute reactions and more post-event drift than comparable positive news. In the press-release taxonomy of Neuhierl, Scherbina and Schlusche (2013), the most negative categories (negative earnings pre-announcements, regulatory rejections, product defects, restatements) rival the largest positive ones in absolute size, and the dividend-omission drift (~-11% over the following year) exceeds the earnings drift in magnitude (Michaely, Thaler and Womack, 1995).

Magnitude scales with surprise, not with the label. Earnings-announcement CARs are monotonic in the size and sign of the surprise, not fixed per event type (Ball and Brown, 1968; MacKinlay, 1997). The same logic applies to M&A bidders, whose near-zero average CAR is partly an artifact of anticipation: the market prices in likely deals in advance, so the announcement captures only the surprise component. A raw cross-event average that ignores surprise will therefore understate the reaction to genuinely unexpected events and can mislead a ranking.

Drift and the speed of adjustment differ by event type

Market efficiency is not uniform across event classes, and the post-event story is itself a comparative dimension. Stock-split information is impounded essentially instantaneously: the founding study of Fama, Fisher, Jensen and Roll (1969) found a pre-split run-up followed by near-instant adjustment at announcement and random returns afterwards, the original evidence for semi-strong efficiency. Earnings information, by contrast, drifts: cumulative abnormal returns continue in the direction of the surprise for roughly 60 trading days after the announcement, the post-earnings-announcement drift (PEAD) first noted by Ball and Brown (1968) and quantified by Bernard and Thomas (1989). Other events drift on their own timetables: dividend-omission drift runs to about -11% over twelve months (Michaely, Thaler and Womack, 1995), stock-split and buyback announcements are each followed by positive longer-run abnormal returns in several samples (Ikenberry, Lakonishok and Vermaelen, 1995), and media tone predicts short-run pressure followed by reversal (Tetlock, 2007). Comparing post-event behavior across types is as informative as comparing the announcement jump, but it must be read with the long-horizon caveat below.

Cross-sectional drivers of within-type variation

Each event type has its own dispersion, and the same handful of firm- and deal-level variables move abnormal returns within almost every category: firm size (smaller firms react more), the pre-event information environment and analyst coverage, the method of payment (for M&A), business relatedness, leverage and profitability. Buybacks add a frequency nuance, with infrequent repurchasers earning markedly larger reactions than serial repurchasers. The lesson for a meta-analytic synthesis is that the average per event type hides systematic, codable variation: a comparison that does not condition on these moderators is comparing mixtures, not effects.

A meta-analytic perspective

Synthesizing abnormal returns across many studies of one event type is itself a discipline. The cleanest illustration is M&A: a meta-analysis of post-acquisition performance found that acquirer performance does not improve on average and is modestly negative, that the commonly studied moderators explain little of the variance, and that cross-study heterogeneity is large (King, Dalton, Daily and Covin, 2004). Two practical rules follow. First, code the moderators (sample period, region, method of payment, expected-return model, event window) and test for heterogeneity rather than reporting a single pooled average; magnitudes are sample-period-conditional, and a stylized fact estimated on pre-crisis data need not hold later. Second, distinguish the well-behaved short-window evidence from the fragile long-horizon evidence: short-window announcement CARs are statistically reliable, whereas long-horizon abnormal returns are model-sensitive and prone to the bad-model problem (Kothari and Warner, 2007). Report dispersion, not just a point estimate.

A worked cross-event example

A single recent open-access study demonstrates the whole comparative method end to end. Studying 657 events across 217 firms, it estimated (-1,+1) CARs of about +6.45% for share repurchases, +5.05% for M&A targets, +4.23% for cash dividends, +2.88% for stock dividends and -3.86% for M&A acquirers (all significant), then used a pairwise "difference indicator" to formally rank reaction strength: buybacks strongest and longest-lived, stock dividends weakest, cash dividends shortest, acquirers negative, targets positive. A three-year follow-up found the market reaction was biased (over- or under-stated relative to realized outcomes) for M&A and stock dividends but accurate for cash dividends and repurchases, with about a one-day pre-announcement leakage in every event type (Luu, 2024). This is exactly the design the rest of this page teaches: one estimation method, common windows, standardized returns, and a formal test on the differences between event types.

How to design a comparable cross-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. What makes a study comparative is an extra discipline: every design choice that differs between event types confounds the "event effect" with a "method effect," so the comparison is only as credible as the things you hold constant.

Hold the estimation design constant

Use the same normal-return model, the same estimation-window length, the same event windows and the same test statistics across every event type. The market model is the usual common denominator for short windows, where model choice barely matters; report robustness with the Fama-French three- or five-factor or Carhart four-factor model for cross-sections tilted by size, value or momentum (see Expected Return Models). Differing windows or models across event types is the most common way a cross-event ranking goes wrong, because the apparent hierarchy then partly reflects the analyst's choices rather than the market's.

Standardize abnormal returns so volatility is not mistaken for effect

Raw CARs are not directly comparable across event types because the underlying firms differ in volatility: a small-cap M&A target is far noisier than a large-cap earnings announcer. Express each event's reaction in its own volatility units using standardized abnormal returns (the Patell statistic, or the standardized cross-sectional BMP statistic of Boehmer, Musumeci and Poulsen, 1991), and only then compare. A pooled regression of standardized CARs on event-type dummies, or a pairwise difference test, gives a formal ranking rather than an eyeballing of point estimates.

Account for differing statistical power

A smaller measured CAR can reflect lower power, not a smaller true effect. Power is high when the event date is precise and the firms are large and low-volatility (earnings, M&A), and low when the date is fuzzy or the firms are small and volatile. Kothari and Warner (2007) make the point vividly: a 10% abnormal return concentrated on a single known day is detected with as few as six stocks, whereas the same 10% spread over six months needs about 200 stocks to reach only 65% power. Report sample volatility, and ideally a minimum-detectable-effect note, per event type so that "small" reactions are not over-interpreted.

Identify the event date correctly for each type

Event-date identification is the make-or-break step and it differs by event type: use the first public disclosure (the press-release timestamp, the 8-K, the wire dissemination), not the completion or effective date. For M&A use the announcement date, not closing; for dividends the declaration date; for splits the announcement, not the ex-date. A one-day pre-announcement run-up is common across event types, which both argues for symmetric windows such as (-1,+1) and makes precise dating essential. Our Event Date Identifier (EDI) is built for this, and News Analytics (CATA) can screen the surrounding news flow for the true first disclosure.

Screen confounders and correct for clustering

Two threats are especially acute when pooling many same-type events. Confounding events (an earnings release alongside a buyback, a guidance update inside a merger window) are the chief reason a CAR cannot be attributed to a specific event type; screen each event and exclude or flag overlaps. Event-date clustering and cross-sectional correlation arise because some event types cluster by calendar (earnings season) or by industry and merger wave; even low cross-sectional correlation severely over-rejects the null of zero average CAR. Use cross-correlation-robust tests (the adjusted statistic of Kolari and Pynnonen, 2010) or calendar-time portfolios, alongside the variance-robust BMP test and non-parametric rank and sign tests; see our overview of Significance Tests.

Separate the short-horizon ranking from the long-horizon comparison

Do not mix the two in one table. Announcement-window CARs are robust and comparable across events and are the right basis for the magnitude hierarchy. Long-run (twelve-month-plus) abnormal returns, used to compare drift across event types, are low-power and benchmark-sensitive; report them with calendar-time portfolio methods (or buy-and-hold abnormal returns) and explicit caveats, never as if they carried the same authority as the short-window numbers (Kothari and Warner, 2007).

Align the samples

Firms that do M&A, splits or IPOs differ systematically in size, industry and volatility from the broad market, which both biases benchmarks and changes power. Align periods, exchanges and liquidity filters across event types, control for size and industry, and remember that thin trading and non-synchronous prices bias small-firm abnormal returns, which matters because small firms dominate some event categories. For earnings and guidance, condition on the surprise, or state explicitly that the reported magnitudes are surprise-weighted averages.

Run it with our tools

The applications on this site implement the comparative workflow end to end. The decisive practical trick is to build one combined event file in which each row is tagged with an event-type label, run every type through the same calculator with identical estimation and event windows, export the standardized CARs, and then run event-type dummies in the cross-section.

• Abnormal Return Calculator (ARC) is the core tool for the comparison. Supply your tagged event file, choose one estimation window and one set of event windows for all types, pick a single expected-return model (Market Model, CAPM, Fama-French 3-factor, Fama-French 5-factor, Carhart 4-factor, or comparison-period mean), and run the full battery of more than a dozen parametric and non-parametric tests, including the standardized BMP test and the Kolari-Pynnonen cross-correlation adjustment needed for clustered samples. ARC exports per-event standardized CARs ready for the second-stage event-type regression.

• Event Date Identifier (EDI) pins down the true first-disclosure date per event type, the one design choice that must be right for every category before the comparison is meaningful.

• News Analytics (CATA) classifies the surrounding news flow and screens for confounders and for the tone of coverage, which lets you treat news type and tone as event dimensions in their own right (Tetlock, 2007; Neuhierl, Scherbina and Schlusche, 2013).

• Abnormal Volume Calculator (AVC) and Abnormal Volatility Calculator (AVyC) extend the comparison beyond returns: trading-volume and volatility reactions confirm that information actually arrived and let you compare the liquidity and volatility footprint of different event types, including the post-announcement spread declines and volatility increases documented by Neuhierl, Scherbina and Schlusche.

Related use cases

This page is the comparative hub; each event type it ranks has its own dedicated application. See Mergers and Acquisitions (the largest single-firm reaction), Earnings Announcements (the canonical surprise-scaled, drifting event), and the related corporate-event pages on Divestitures, Alliances and Joint Ventures and Competitive Dynamics; the Litigation application covers reputational and legal shocks on the negative side. For the news-classification and sentiment dimension introduced above, see News Analytics (CATA). For the full catalogue of applications, return to the Practical Applications overview.