Explainable Deep Reinforcement Learning for Portfolio Management: An Empirical Approach

Deep reinforcement learning (DRL) has been widely studied in the portfolio management task. However, it is challenging to understand a DRL-based trading strategy because of black-box nature deep neural networks. In this paper, we propose an empirical approach explain strategies DRL agents for First, use linear model hindsight as reference model, which finds best weights by assuming knowing actual stock returns foresight. particular, coefficients feature weights. Secondly, agents, integrated gradients define weights, are between reward and features under regression model. Thirdly, study prediction power two cases, single-step multi-step prediction. quantify calculating correlations agent similarly machine methods. Finally, evaluate task on Dow Jones 30 constituent stocks during 01/01/2009 09/01/2021. Our empirically reveals that exhibits stronger than