Stefan Nagel (Booth): Rerunning the history of momentum and reversal discoveries

Stefan Nagel: Rerunning the history of momentum and reversal discoveries

Abstract: We examine the cross-sectional predictability of stock returns with lags of past returns from the perspective of a Bayesian observer who (i) does not take a stand, a priori, whether momentum, reversals, or any other types of predictability patterns are present, but instead entertains a prior that allows for a very flexible relationship between returns and a large number of lags of past returns; (ii) estimates prior hyperparameters with an empirical Bayes approach that accounts for the possibility that investor learning can induce correlation ex-post between returns and lagged returns that is not present ex-ante; (iii) allows for downweighting of data in the distant past to allow for possible decay or emergence of predictability. Based on resulting posterior at every point in time, the observer would have discovered well-known anomalies such as momentum, long-term reversal, or the momentum echo effect, around the time, or earlier, than published studies, which suggest that these findings are not the product of data mining or an ex-post artifact of investor learning. But the anomalies temporarily disappear following the publication of these studies. Furthermore, short to medium-term momentum and long-term reversals have a tendency to appear, disappear, and reappear together.

Slides: Link to slides