Incomplete temporal overlap and cross-sectional independence in event studies

Abstract

In the event study literature, estimates of security abnormal returns are considered independent whenever securities have different event dates, i.e. in the absence of ‘event clustering’. Nonetheless, there are three sources of cross-sectional correlations in estimated abnormal returns even when no two securities have a common event date. First, the estimation interval (for market model parameters) may overlap; second, the event date for one security may overlap the estimation interval for another; third, event windows longer than a one (or two) day announcement may overlap. In this article, analytical and simulations methods are used to assess the influence of these partial overlaps. Simulations reveal that for short event windows (≤11 days, with 300 days in the estimation interval) these partial overlaps do not create any measurable bias, even when 50 separate events are contained within 125 trading days. However, there is potential for bias in ‘long horizon’ event studies with nearly clustered event dates.