Investing in Disappearing Anomalies

We argue that anomalies may experience prolonged decay after discovery and propose a Bayesian framework to study how that impacts portfolio decisions. Using the January effect and short-term index autocorrelations as examples of disappearing anomalies, we find that prolonged decay is empirically important, particularly for small-cap anomalies. Papers that document new anomalies without accounting for such decay may actually underestimate the original strength of the anomaly and imply an overstated level of the anomaly out of sample. We show that allowing for potential decay in the context of portfolio choice leads to out-of-sample outperformance relative to other approaches.