Risk-sensitive policies for portfolio management

作者：

Highlights：

• Risk-sensitive policies can avoid the effect of the crisis in the financial market.

• The hierarchical deep learning algorithm is developed to control the portfolio risk.

• The distributional learning method is proposed for risk-averse trading policies.

• The study shows our proposed models effectively protect investors from a large loss.

摘要

•Risk-sensitive policies can avoid the effect of the crisis in the financial market.•The hierarchical deep learning algorithm is developed to control the portfolio risk.•The distributional learning method is proposed for risk-averse trading policies.•The study shows our proposed models effectively protect investors from a large loss.

论文关键词：Portfolio management,Portfolio optimization,Reinforcement learning,Worst-case scenario,Hierarchical DDPG,Distributional DDPG

论文评审过程：Received 22 July 2021, Revised 28 February 2022, Accepted 28 February 2022, Available online 12 March 2022, Version of Record 24 March 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.116807