Machine Learning CPR for Stock Anomalies?

Much prior research indicates that most stock anomalies fail to deliver due to data snooping in their discovery, post-publication market adaptation and, especially, implementation costs. In their March 2026 paper entitled "Reviving Anomalies", Heiner Beckmeyer, Florian Berg, Timo Wiedemann and Jonas Wortmann describe and test a framework to address the poor performance of simple long-short portfolios by double-sorting based first on anomaly rules and then on expected next-month net returns of anomaly stocks. They employ machine learning return forecasts based on  153 firm/stock characteristics to compute expected returns. They quantify expected trading frictions with impact of trading scaled by fund size (micro, small, medium and large). Using data for the 153 firm/stock characteristics and return data for a broad sample of U.S. stocks during January 2004 to December 2023, they find that: