Deep Deterministic Policy Gradient based portfolio management system

Abstract

The Deep Reinforcement Learning Process (DRL) has made considerable progress since its inception towards the production of various independent systems. Nowadays, deep learning gives Reinforcement Learning (RL) the ability to conform to several dysfunctions that were previously difficult, if not impossible to resolve. In this paper, we explore the power of Deep Reinforcement Learning in order to solve the problem of portfolio management which is a process that allows the selection, prioritization and management of funds by following a defined policy. Our objective is to obtain an optimal strategy to maximize the expected return while minimizing the costs. To do this, we use the Deep Deterministic Policy Gradient which is an off-policy algorithm adapted for environments with continuous action spaces. The obtained results demonstrate that the model achieves a higher rate of return than the strategy of “Uniform Buy and Hold” stocks and the strategy of “Buy Best Stock in last month”.