Optimizing Foreign Exchange Trading Performance Through Reinforcement Machine Learning Framework

The ever-changing financial market of foreign exchange attracts many traders. Traders must make wise decisions to avoid significant losses when buying and selling currencies. This project intends to reduce the chance of suffering from loss by providing a trading strategy. The research on developing a trading strategy specifically for the foreign exchange market is still lacking due to the limitation in selecting the best model to create a trading strategy, which is still a working area. Even with current research on trading strategy, it tends not to work overtime due to unpredictable market trends. Therefore, this paper proposed three models using the algorithms A2C, PPO & DQN to find the best strategy in foreign exchange trading, analyze the impact of individual features on the trading strategy and identify the most influential features to develop the best trading strategy using reinforcement learning and finally evaluate the performance on unseen data using Sharpe Ratio, Sortino Ratio, Omega Ratio, Profit & Loss (%), Maximum Drawdown (%) and Cumulative Score. The experiment result showed that the PPO algorithm performed best on 2 of the currency pairs which is GBP/USD and USD/JPY, with a Sharpe Ratio of 0.23 and 0.70, respectively, and a Profit & Loss of 7.4% and 16.78%, respectively, when tested on unseen data. Meanwhile, when tested on unseen data, the A2C model performed the best on the EUR/USD currency pair with a Sharpe Ratio of 0.16 and a Profit & Loss of 3.34%.