Tests for cumulative abnormal returns over long periods: Simulation evidence

Abstract

The usual test of cumulative abnormal returns for multiple-day periods assumes that abnormal returns are serially independent. The assumption imparts an upward bias to test statistics even when raw returns are serially independent. In simulation, the usual test rejects true null hypotheses too frequently in the longest cumulation periods and in shorter periods when events are clustered in calendar time. Excessively frequent rejection implies that the nominal significance level understates the actual significance level. A corrected statistic, derived without the serial independence assumption, rejects true null hypotheses with a frequency less than or equal to the nominal significance level. However, the corrected test is not very powerful in the longest event periods.