The benchmark effect in long-run event studies

An important step in long-horizon event studies is the choice of the benchmark that is used as a proxy for the expected return of the individual securities. Most existing studies use the rate of return of a stock market index. However, our simulation shows that such a procedure creates a significant bias in the means and in the test statistics. We simulate event studies by randomly choosing stocks out of a large database of historical rates of returns. The main results are: (1) Benchmarks should be selected by a simulation procedure before the design of a study is established in order to reduce potential bias. (2) Using reference portfolios that take account of return anomalies, and testing for long-run abnormal returns with bootstrapped skewness-adjusted t-statistics lead to more precise results. (3) Misspecification due to time-varying return effects can be analyzed by the construction of random samples that share the same time distribution as the real events.