Picking Winners in Factorland: A Machine Learning Approach to Predicting Factor Returns

The article examines the cross-sectional predictability of factor returns. Using machine learning techniques, the authors extract information from a comprehensive set of factor characteristics measuring their past returns, risks, and spreads. Applying this method to a repertoire of US stock market anomalies, they find robust predictability in their performance. The decile of factors with the highest expected return outperforms those with the worst outlook by up to 1.39% per month. The alphas are robust to many considerations but gradually decline over time. Predictability is mainly driven by factor momentum, which captures most of the cross-sectional variation in anomaly returns.