Optimizing Advantage Actor-Critic with Policy Gradient and Deep Q-learning to Maximize Profit in Forex Trading Prediction

Financial markets, such as forex markets, are inherently complex, with high volatility, noise, trends, and market shocks. Therefore, designing effective trading strategies is crucial for maximizing investment returns. The implementation of deep reinforcement learning (DRL) in financial trading has become an attractive research topic in recent years. This study proposes Advantage Actor-Critic-based algorithms (A2C) algorithms to learn forex trading strategies using deep reinforcement schemes. The ensemble trading strategy was obtained by combining the best features of the two algorithms' Policy Gradient as Actor and Q-learning as Critic, resulting in a robust strategy that adapts to different market situations. Our experiment on the EURUSD currency on the 4 hour candlestick timeframe using five years of data history was divided into three parts for training, validating, and testing. The A2C agent learns to buy, sell, and hold a trade with the goals of maximizing profits and reducing losses. Our proposed method provides an effective and robust solution to various financial trading problems using deep reinforcement learning. Our proposed methods provide effective and robust solutions to various financial trading problems using deep reinforcement learning. The evaluation results improved from each training, where the Cumulative Return evaluation increased from 29.65 to 112.74, the Maximum Drawdown reduced from 36.18% to 4.48%, the Sharp Ratio increased from 0.94 to 2.47%, and the AHPR increased from 0.19% to 3.56%. This indicates that longer training data result in higher cumulative rewards and an increase in total profit.