Download PDFOpen PDF in browser

Deep Reinforcement Learning for Portfolio Management

11 pagesPublished: November 24, 2022

Abstract

This paper discussed how to build deep reinforcement learning (DRL) agents to determine the allocation of money for assets in a portfolio so that the maximum return can be gained. The policy gradient method from reinforcement learning and convolutional neural network/recurrent neural network/convolutional neural network concatenated with the recurrent neural network from deep learning are combined together to build the agents. With the proposed models, three types of portfolios are tested: stocks portfolio which has a positive influence due to the Covid-19, stocks portfolio which has a negative influence due to the Covid-19, and portfolio of stocks combined with cryptocurrency which are randomly selected. The performance of our DRL agents was compared with that of equal-weighted agent and all the money fully invested on one stock agents. All of our DRL agents showed the best performance on the randomly selected portfolio, which has an overall stable up-ticking trend. In addition, the performance of linear regression model was also tested with the random selected portfolio, and it shows a poor result compared to other agents.

Keyphrases: deep reinforcement learning, policy gradient, portfolio management

In: Yan Shi, Gongzhu Hu, Krishna Kambhampaty and Takaaki Goto (editors). Proceedings of 35th International Conference on Computer Applications in Industry and Engineering, vol 89, pages 41-51.

BibTeX entry
@inproceedings{CAINE2022:Deep_Reinforcement_Learning_Portfolio,
  author    = {Yue Ma and Ziping Liu and Chuck McAllister},
  title     = {Deep Reinforcement Learning for Portfolio Management},
  booktitle = {Proceedings of 35th International Conference on Computer Applications in Industry and Engineering},
  editor    = {Yan Shi and Gongzhu Hu and Krishna Kambhampaty and Takaaki Goto},
  series    = {EPiC Series in Computing},
  volume    = {89},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2398-7340},
  url       = {/publications/paper/nlqD},
  doi       = {10.29007/w2m3},
  pages     = {41-51},
  year      = {2022}}
Download PDFOpen PDF in browser