Evaluating Long-Horizon Event Study Methodology

We describe the fundamental issues that long-horizon event studies face in choosing the proper research methodology, and summarize findings from existing simulation studies about the performance of commonly used methods. We document in detail how to implement a simulation study and report findings from our own study that focuses on large-size samples. The findings have important implications for future research. In our simulation study, we examine the performance of more than twenty different testing procedures, which can be broadly classified into two categories: The buy-and-hold benchmark approach and the calendar-time portfolio approach. The first approach uses a benchmark to measure the abnormal buy-and-hold return for every event firm, and tests the null hypothesis that the average abnormal return is zero. We investigate the performance of five ways of choosing the benchmark and four test statistics including the standard t-test, the Johnson's skewness-adjusted t-test, the bootstrapped Johnson's skewness-adjusted t-test, and the Fisher's sign test. The second approach forms a portfolio in each calendar month consisting of firms that have had an event within a certain time period prior to the month, and tests the null hypothesis that the intercept is zero in the regression of monthly calendar-time portfolio returns against the factors in an asset-pricing model. We implement this approach with both the Fama-French three-factor model and the four-factor model with an additional momentum factor, and with both the ordinary least-squares and weighted least-squares estimation methods. We find that the combination of the sign test and the benchmark with a single most correlated firm provides the best overall performance for various sample sizes and long horizons. Furthermore, the Fama-French three-factor model is a better asset pricing model for monthly returns of calendar-time portfolios than the four-factor model, as the latter leads to serious overrejection of the null hypothesis