Calendar-Time Portfolio Tests

A classic event study estimates abnormal returns for each firm around its own event date, then averages across events and tests the average against zero. This works well when events are spread thinly through time. It breaks down when many firms experience similar events in overlapping windows — spinoffs during a hot M&A year, or dozens of firms missing earnings in the same reporting season — because the underlying event-window returns become cross-sectionally correlated, and a test that assumes independence across events badly understates the true standard error.

The calendar-time portfolio approach fixes this by changing what gets tested. Instead of one test per event, every firm currently inside its event window (say, the 36 months after an event) is pooled into a single portfolio, rebalanced monthly, and that portfolio's time series of monthly returns is regressed on a factor model like Fama-French three- or four-factor. The regression's intercept, or alpha, is the test statistic: it measures the average monthly abnormal return of "being a firm that recently had this event," using only one number per calendar month rather than one per event, so events happening at the same time can't inflate the sample size or fake statistical significance.

The cost is lower statistical power when event timing is genuinely clustered, since many overlapping events now contribute to the same monthly observation rather than each counting separately — a trade-off researchers accept because the alternative, treating clustered events as independent, produces test statistics that are simply wrong.