In short: a plain-language reference for the core event-study terms, from abnormal return to the Kolari-Pynnonen tests. Each entry links to the page that explains it in full. Use it as a quick lookup while you design, run, and report a study.
Jump to a group: Core quantities · Windows and timing · Expected-return models · Significance tests · Inference concepts · Study types · News analytics
Core quantities
- Abnormal return (AR)
- The part of a stock's return on a given day that is not explained by its normal, expected return. Computed as actual return minus expected return. The basic unit of every return event study. See methodology.
- Expected (normal) return
- The return a stock would have earned absent the event, predicted by an expected-return model from the estimation window. See expected-return models.
- Cumulative abnormal return (CAR)
- The sum of one firm's abnormal returns across the event window. Measures the total event impact for that firm. See a worked example.
- Average abnormal return (AAR)
- The abnormal return on a given event day averaged across all firms or events in the sample.
- Cumulative average abnormal return (CAAR)
- The average of firm-level CARs across the sample. The headline figure most event studies report and test. See interpreting CAAR.
- Alpha
- The intercept of a return model. In the market model it captures a stock's average return unrelated to the market.
- Beta
- The slope of a return model: how strongly a stock moves with its reference market. Estimated over the estimation window. See the market model.
Windows and timing
- Event date
- The day the market first learns of the event (day 0). Choosing it correctly is critical: information often leaks before the formal announcement.
- Estimation window
- The pre-event period used to estimate the expected-return model. A common choice is 120 trading days; a minimum of about 80 is typical. See the application blueprint.
- Event window
- The short period around the event date over which abnormal returns are measured and accumulated, often a few days (a 5-day window is common). Wider windows capture leakage and drift but add noise.
- Confounding event
- A second, unrelated event inside the event window that contaminates the abnormal return. Clean samples exclude confounded observations.
Expected-return models
- Market model (MM)
- The most widely used model (about 79% of studies). Regresses a stock's return on its reference market over the estimation window to get alpha and beta. See expected-return models.
- Market-adjusted model (MAM)
- The simplest model: expected return equals the market return (beta fixed at 1, alpha at 0). Useful when the estimation window is short or unavailable.
- Comparison-period mean-adjusted model (CPMAM)
- Expected return equals the stock's own mean return over the estimation period. Makes no use of the market.
- Scholes-Williams beta
- A beta estimator that corrects for non-synchronous (thin) trading, where a stock and its index do not trade at exactly the same instants.
- GARCH / EGARCH
- Models that let return variance change over time. Useful when volatility clusters, which violates the constant-variance assumption of the plain market model.
- Fama-French three-factor model
- Adds size (SMB) and value (HML) factors to the market factor, addressing known biases of the single-factor CAPM. EST also supports the four- and five-factor extensions.
- Fama-French-Carhart four-factor model
- The three-factor model plus a momentum (WML) factor.
- Fama-French five-factor model
- Adds profitability (RMW) and investment (CMA) factors to the three-factor model.
- CAPM
- The capital asset pricing model: expected return as a function of the risk-free rate and a single market beta. Rarely the benchmark of choice in modern event studies (under 1% of studies).
Significance tests
EST computes 16 tests with full formulas and null distributions. The full math is on significance tests; a plain-language chooser is in how to choose.
- Parametric test
- A test that assumes firm-level abnormal returns are normally distributed. Reliable for larger samples (roughly n > 30 to 50).
- Nonparametric test
- A test that makes no distributional assumption. Less affected by non-normality and outliers, so researchers usually run parametric and nonparametric tests side by side.
- T-test
- The simplest parametric test of whether an AR or CAR differs from zero. Sensitive to event-induced volatility and to departures from normality.
- Cross-sectional T (CSect T)
- Uses the spread of abnormal returns across firms. Robust to event-induced volatility but assumes firms are cross-sectionally independent.
- Crude dependence adjusted T (CDA T)
- Uses the time-series standard deviation of the average abnormal return to account for cross-sectional dependence.
- Patell Z
- Standardizes each abnormal return by its forecast-error-corrected standard deviation before aggregating (Patell, 1976). A workhorse test, but sensitive to cross-sectional correlation and event-induced volatility.
- Adjusted Patell Z (Kolari-Pynnonen, 2010)
- The Patell Z corrected for cross-sectional correlation among abnormal returns. One of the modern tests EST offers that most free tools do not.
- Standardized cross-sectional T (BMP)
- The Boehmer-Musumeci-Poulsen (1991) test: combines Patell-style standardization with a cross-sectional variance, so it handles event-induced volatility.
- Adjusted standardized cross-sectional T (Kolari-Pynnonen)
- The BMP test with an added cross-correlation adjustment.
- Skewness-corrected T (Hall, 1992)
- Corrects the t-statistic for skewness in the return distribution.
- Rank Z (Corrado, 1989)
- A nonparametric test built on the ranks of abnormal returns. Loses power for longer event windows.
- Generalized Rank T (Kolari-Pynnonen, 2011)
- A rank test that accounts for cross-sectional and serial correlation as well as event-induced volatility. The most robust nonparametric option EST provides.
- Generalized Rank Z
- A simpler companion to the Generalized Rank T, slightly less robust to cross-sectional correlation.
- Sign Z (Cowan, 1992)
- Counts positive abnormal returns against a 0.5 benchmark. Robust to skewness, weaker for long windows.
- Generalized Sign Z (Cowan, 1992)
- The sign test using the empirical fraction of positive abnormal returns from the estimation window instead of 0.5.
- Wilcoxon signed-rank
- A nonparametric test that uses both the sign and the magnitude of abnormal returns (Wilcoxon, 1945).
- Permutation test (Nguyen-Wolf, 2023)
- A resampling-based test robust to non-normality, at higher computational cost. One of the newest tests in the literature, available in EST.
Inference concepts
- Null hypothesis
- The default claim that the expected abnormal return (or CAAR) is zero. A test asks whether the data are extreme enough to reject it.
- p-value
- The probability of seeing a test statistic at least as extreme as the observed one if the null were true. Values below 0.05 are usually treated as solid evidence of an effect.
- Significance level
- The threshold (often 0.05) at which the null is rejected. It is the accepted rate of false positives.
- Two-sided test
- A test that looks for an effect in either direction (positive or negative abnormal return). The convention in event studies.
- Event-induced volatility
- The tendency of return variance to rise around an event. If ignored, it inflates test statistics and overstates significance; several tests (BMP, Generalized Rank T) are built to handle it.
- Cross-sectional correlation
- Correlation of abnormal returns across firms, typically when events cluster in calendar time or industry. It biases simple tests; the Kolari-Pynnonen adjustments correct for it.
- Joint test
- Every event study jointly tests the research hypothesis, the chosen return model, and the market-efficiency assumptions together. A rejection cannot, by itself, say which one failed.
Study types
- Return event study
- The standard design: measure abnormal returns around an event. Run it with the Abnormal Return Calculator.
- Volume event study
- Measures abnormal trading volume around an event, a signal of informed trading or attention. See volume event study.
- Long-run event study (BHAR)
- Measures persistent effects over months or years, usually with buy-and-hold abnormal returns or calendar-time portfolios. See long-run event study.
- Intraday event study
- Measures high-frequency reactions within the trading day. See intraday event studies.
- Reverse event study
- Looks at spillover effects of an event on firms that did not experience it, such as rivals or partners. See reverse event study.
News analytics
- News analytics (CATA)
- Turning news streams, filings, and transcripts into quantitative features (tone, sentiment, topics) that can feed an event study. EST's content-analysis pillar, which no free competitor offers. See news analytics.
- Content analysis
- Dictionary-based scoring of text to quantify themes, sentiment, and readability.
- Sequence analysis
- Tracking how the narrative around a firm changes over time, to detect shifts in framing or attention. See sequence analysis.