We examine how the return predictability of deep learning models varies with stocks’ vulnerability to investors’ behavioral biases. Using an extensive list of anomaly variables, we find that the long-short strategy based on deep learning signals generates greater returns for stocks more vulnerable to behavioral biases: stocks that are small, young, illiquid, unprofitable, volatile, non-dividend-paying, close to default or extreme growth, far from the 52-week high, and lottery-like. Such performance of deep learning models becomes more pronounced for stocks held by less sophisticated investors. These results suggest that deep learning models accommodating time-varying nonlinear factor exposures are useful in capturing mispricing induced by behavioral biases. 

JEL classification: To be included

Keywords: Deep Learning, Behavioral Biases, Empirical Asset Pricing​