MIE.08 – Forex Trading Using Reinforcement Learning

• Abby Hellinga
• Joseph Liotine
• George Markopoulos
• Sriman Poladi
• Nicole Poplawski

Financial institutions across the globe struggle with the unpredictability of volatility when predicting currency movement. To address this challenge, machine learning and its derivatives are leading the way in forex market forecasting. Forex is defined to be the foreign exchange market where global currencies are traded in pairs. Algorithmic trading has become the standard in the financial market, with most algorithms relying on rule-based expert systems that need manual updating to changing market conditions. Machine learning is the next natural step in algorithmic trading because it can directly learn market patterns and behaviors from historical trading data to make accurate trading decisions. This project proposes a complete, end-to-end, system for automated, low-frequency quantitative trading in the Forex markets using machine learning. The system aims to reduce Profit and Loss (PNL) volatility by identifying a small number of daily hedges that decrease Mark-to-Market (MTM) losses, decrease MTM volatility, maximize Sharpe Ratio, and are 50k notional or higher. The system utilizes reinforcement learning using proximal policy optimization (PPO) on a multilayer perceptron neural network (MLP) and compares it to linear regression and hierarchical risk parity (HRP). These models have been shown in academic research to be among the best and most consistent performers for Forex trading. The proposed system represents a promising approach to automating Forex trading and achieving more consistent and profitable results.