Reinforcement learning in Portfolio Management and its interpretation

Since all machine learning methods commonly in use today are viewed as black boxes, the goal of this paper is to make one of these meth- ods transparent in the context of Portfolio management. I interpret the strategies implied by reinforcement learning (RL) and relate them to the strategies implied by academic portfolio advice with the help of their classical portfolio (CP) management models. Because RL is ac- tually approximate dynamic programming (DP), it is perfectly suited for the volatile DP environment of portfolio management compared to other machine learning methods in use. In terms of performance, this RL method is able to: 1) achieve the same average terminal wealth of 1.33, which is an increase of 33% in portfolio value over ve years, as the CP model with a low risk-aversion 2) diminish the standard devia- tion of the terminal wealth by 30% from 0.35 to 0.25 3) have a lower turnover than the same CP model by three percent. This can mostly be explained by the conservative investing of the reinforcement learning method overall.