Problems in Measuring Contrarian Performance with Jensen's Alpha : A Reconciliation of Real-Time and Event-Time Tests

Much of the early research on contrarian performance adjusts for risk using a Jensen's alpha estimated with in-sample returns in either real time or event time. The use of in-sample returns means that beta is estimated with the same observation period returns it is intended to risk adjust. This results in attributing systematic sources of overreaction to beta-risk and understates contrarian performance. Several later papers find mixed results when they attempt to correct for this problem using alternative Jensen's alpha-based methods, the relative merits of which are not well understood. In this paper, we identify the sources of bias that arise from these alternative methods. Our analysis indicates that the methods are all biased against detecting contrarian performance. The event-time Jensen's alpha is the most heavily biased, regardless of whether its beta is estimated in-sample, or conditional on formation period market returns. A real-time Jensen's alpha conditioned on formation period returns is the least biased against contrarian performance. We reject a joint hypothesis of the CAPM and market efficiency because this conditional real-time alpha detects significant long-horizon contrarian performance