An Asynchronous Advantage Actor-Critic Reinforcement Learning Method For Stock Selection And Portfolio Management

Computation finance has been a classical field that uses computer techniques to handle financial challenges. The most popular domains include financial forecast and portfolio management. They often involve large datasets with complex relations. Due to the special properties of computation finance problems, machine learning techniques, especially deep learning techniques, are widely used as the quantitative analysis tool. In this paper, we try to apply the state-of-art Asynchronous Advantage Actor-Critic algorithm to solve the portfolio management problem and design a standalone deep reinforcement learning model. In the simulated market environment with practical portfolio constrain settings, asset value managed by the proposed machine learning model largely outperforms S&P500 stock index in the test period.