Abstract

The basis of this analysis is a model presented at the ACM International Conference in New York on AI in Finance in October 2020 (Yang et al., 2020). The authors claim that the introduced deep reinforcement learning ensemble model outperforms the Dow Jones Industrial Average Index, and the three individual algorithms that form the ensemble in terms of the risk-adjusted returns measured by the Sharpe ratio. Furthermore, it is claimed that the ensemble is more robust and reliable than the individual agents (Yang et al., 2020). We evaluate these claims for statistical significance. As some weaknesses of the model become evident, we suggest a work-around and show the results with the suggested alteration. Finally, we combine all the findings and present an alternative model.

Keywords: Deep reinforcement learning, Dow Jones, Algorithms, Automated stock trading