Interacting biases, non-normal return distributions and the performance of tests for long-horizon event studies

Abstract

We report simulations of one-, three-, and five-year abnormal buy-and-hold stock return tests. Using benchmark portfolios purged of new-listings and rebalancing biases, we find severe misspecification of most tests, due in part to skewness. Control-firm matching also results in misspecification, particularly in large samples. We document a negative relation between skewness bias and sample size, and an overlapping-horizons bias. Both biases become more severe as the holding period lengthens. The biases interact such that tests can be well-specified in one situation but not another. A two groups test using winsorized abnormal returns yields correct specification and considerable power in many situations.