Cooperative Multi-Agent Deep Reinforcement Learning for Forex Algorithmic Trading using Proximal Policy Optimization

This paper presents a novel multi-agent deep reinforcement learning (DRL) approach for automated trading in the foreign exchange (Forex) market, utilizing Proximal Policy Optimization (PPO). The proposed system employs two cooperative agents: one specializing in position entry and another in position exit. Both agents leverage Long Short-Term Memory (LSTM) neural networks to process temporal market data and technical indicators. The reward function is designed to encourage collaborative behavior, with rewards shared between agents upon trade closure. Furthermore, the implementation of Generalized Advantage Estimation (GAE) method enables the agents to balance immediate rewards with long-term consequences, enhancing the stability and efficiency of the learning process. This research contributes to the growing field of AI-driven Forex trading strategies, offering valuable insights into applying multi-agent reinforcement learning in dynamic financial markets. The proposed methodology shows efficiency in developing lucrative trading systems capable of navigating the complexities of the Forex market. Experiments conducted on EUR/USD data demonstrate the effectiveness of this approach. Across 150 runs on the test dataset, the framework achieved profits ranging from 3.5% to 6.9%, with a median return of 5.8%.