Deep reinforcement learning strategy in automated trading systems

Abstract

Every day, a considerable amount of money is traded in the form of derivatives in global financial markets. Artificial intelligence (AI) software is now the main competitor of traditional investors; it allows market players to anticipate price variations. Adopting a strategy in the stock market using machine learning is essential to make a profit when the markets are depressed. Thus, this study proposes a new approach to train the model of an intelligent trading agent based on deep reinforcement learning (DRL), highlighting the dilemma of the exploitation and exploration phase during the training model as well as its impact on the model performance on a separate or combined portfolio asset. This study shows a clear improvement by backtesting a portfolio composed of three stock market indices, illustrating the advantages and gains obtained by deep reinforcement learning strategy in financial markets.