An Evaluation of Testing Procedures for Long Horizon Event Studies

We conduct a comprehensive simulation study to evaluate testing procedures for long horizon event studies. The simulation results raise the following concerns about some popular practices: (1) using the four-factor model that includes the Fama-French three factors and a momentum-related factor causes serious over rejection of the null hypothesis; (2) using reference portfolios as benchmark tends to overestimate event firms' long-term returns; and (3) the computation-intensive bootstrap test has low power for long event horizons. Moreover, unless the number of event firms in a study is very large, all testing procedures suffer substantial loss of power quickly as event horizon increases, especially for samples of small firms. Of particular interest, the combination of the nonparametric sign test with a single firm benchmark shows the best performance consistently in our simulations.