Bayesian Analysis of a Single-Firm Event Study

We evaluate event studies of single firms using Bayesian methodologies. Event studies determine the effect of a public event on firms’ stock returns. Securities litigation and antitrust investigations frequently use single-firm event studies to respectively determine loss causation\damages and anticompetitive behavior. The standard approach for inference assumes normally distributed returns; however, theoretical and applied evidence suggest normality does not necessarily hold, resulting in misleading results from inference. Assumptions of normality can be relaxed using Bayesian statistics. Using data from ten firms, eight of which were involved in securities litigation, we analyze various Bayesian models and methods of inference to compare to the classical, standard method. We also incorporate a flexible Bayesian model, replacing parametric likelihood functions with the empirical distribution function, allowing for a Bayesian event study analysis without parametric assumptions. Our results suggest that using a Bayesian event study with no assumption of normality can result in outcomes of loss causation and damages conflicting those determined by courts using the potentially invalid, standard methods. ∗Department of Economics, Oberlin College, email: [email protected]. I thank Katherine Barnes, Price Fishback, Jonah Gelbach, Keisuke Hirano, and Stanley Reynolds for helpful comments and discussions.