Portfolio Management with Reinforcement Learning

Abstract

Portfolio management is a crucial trading task for investment companies in the market. In this work, reinforcement learning (RL) incorporating the transformer structure is combined with deep learning (DL) to build an automated portfolio management model. The proposed method uses the Sharpe ratio along with transaction cost as the reward and build an environment that contains the whole A-share market to train the RL agent. The result demonstrates that the trained strategy outperforms The Shanghai Composite Index.

Tasks

Reproductions