POE: A General Portfolio Optimization Environment for FinRL

Abstract

Portfolio optimization is a common task in financial markets in which a manager rebalances the invested assets in the portfolio periodically aiming to make a profit, minimize losses and maximize long-term returns. Due to their great adaptability, Reinforcement Learning (RL) techniques are considered convenient for this task but, despite RL’s great results, there is a lack of standardization related to simulation environments. In this paper, we present an RL environment for the portfolio optimization problem based on state-of-the-art mathematical formulations. The environment aims to be easy-to-use, very customizable, and have integrations with modern RL frameworks.