Exploring the Forex Market with the Reinforcement Learning Agent Deep-Q-Network

The Forex market is the largest in the world. Such a competitive environment, in conjunction with the Efficient Market Hypothesis (EMH), often leads people to doubt the feasibility of generating profits from it. Conversely, the Adaptive Market Hypothesis (AMH) suggests that investors can sometimes behave irrationally, creating opportunities for profitable strategies. In line with the latter hypothesis, many investors seek new opportunities, employing either fundamental or technical analysis trading strategies. However, with the advent of artificial intelligence (AI), new automated trading systems have emerged. One of the most promising machine learning models is known as Reinforcement Learning (RL), where an agent acquires knowledge through interactions with an environment, aiming to optimize a specific reward function. In this context, a Deep-Q-Network (DQN) agent has been implemented as an online-learning automated trading system. The DQN agent was tested on 30-minute timeframe EUR/USD data, spanning from January 2019 to December 2022. This evaluation encompassed three distinct reward functions and included a comparative analysis with traditional trading strategies, resulting in promising findings.