28 Pages Posted: 27 Jul 2001
Date Written: August 2000
A variety of test statistics have been employed in the finance and accounting literatures for the purpose of conducting hypothesis tests in event studies. This paper begins by formally deriving the result that these statistics do not follow their conventionally assumed asymptotic distribution even for large samples of firms. Test statistics exhibit a statistically significant bias to size in practice, a result that I document extensively. This bias arises partially due to commonly observed stock return traits which violate conditions underlying event study methods. In this paper, I develop two alternatives. The first involves a simple normalization of conventional test statistics and allows for the statistics to follow an asymptotic standard normal distribution. The second approach augments the simple normalization with bootstrap resampling. These alternatives demonstrate remarkable robustness to heteroskedasticity, autocorrelation, non-normality, and event-period model changes, even in small samples, and they are useful for event studies with non-clustered events.
Suggested Citation: Suggested Citation