More Power to You : Properties of a More Powerful Event Study Methodology

There are more than 500 event studies in the top finance journals, which can broadly be split into 2 subgroups depending on whether firms are all affected at the same time, like banks reacting to new regulation, or different firms are affected at different times, like mergers and acquisitions studies. For the former, cluster effects (contemporaneous covariances) have been modeled with GLS but studies have shown that embodying these effects leads to little practical impact. This may be the reason why no-one has looked at GLS modeling of variances (precision of estimated firm counterfactual model) for the latter subgroup. We use subsample analysis and analysis hypothetically reducing measured event effects and find that it is often the case that traditional tests will not reject the null when a GLS based test may (strongly) reject the null. The power of the former is poor. There are many published event studies where the null is not rejected. This may be because of the phenomenon being tested but it may also be because of the lack of power of the traditional estimators. Without rerunning the data from these tests using GLS there is no way to know if these published results are misleading