I use classification-based machine-learning methods to decompose 32 anomaly payoffsinto risk exposures and mispricing. The component driven by risk earns statistically insignificantreturns, despite its efficacy in explaining the time-series variation in anomaly payoffs.The mispricing component...