Deep reinforcement learning for portfolio management of markets with a dynamic number of assets

作者：

Highlights：

• Formulation of a trading method for markets with a dynamic number of assets.

• Unseen assets are easily integrated without changing or retraining the network.

• Optimal transactions are computed for markets with transaction costs.

• The method outperforms the baselines on a cryptocurrency market database.

摘要

•Formulation of a trading method for markets with a dynamic number of assets.•Unseen assets are easily integrated without changing or retraining the network.•Optimal transactions are computed for markets with transaction costs.•The method outperforms the baselines on a cryptocurrency market database.

论文关键词：Reinforcement learning,Deep learning,Portfolio management,Transaction costs,Multiple assets

论文评审过程：Received 1 April 2020, Revised 10 September 2020, Accepted 11 September 2020, Available online 16 September 2020, Version of Record 23 September 2020.

论文官网地址：https://doi.org/10.1016/j.eswa.2020.114002