Data-mined Anomalies and the Expected Market Return

Based on a theoretical framework for mispricing correction persistence, we propose a two-stage anomaly selection approach to predict market returns. The first stage screens the t-statistic of anomaly returns, and the second stage estimates the slope coefficient of predictive regression to select the most promising anomalies for predicting the market. The selected data-mined anomalies from a universe of several thousand signals exhibit strong and persistent mispricing correction dynamics. We show that aggregate returns of long-short portfolios constructed from these data-mined anomalies are significantly linked to the predictability of aggregate excess market returns, delivering statistically significant out-of-sample predictions of market excess returns and surpassing the predictive power of published anomalies.