Trading the foreign exchange market using reinforcement learning enhanced with the exploitation of lead-lag correlations

Abstract

This project aims to create a profitable trading strategy in the Intraday Foreign Exchange Market through the use of Reinforcement Learning for Quantitative Trading. Inspired by the success of the DeepScalper framework in trading Equities and Bonds, we adapt the framework, that is built on top of Deep Q-Learning to Foreign Exchange trading. In addition to using technical indicators to model for market state, features are created to enable the Reinforcement Learning agent to capture dynamic lead-lag correlations that appear amongst the major currency pairs. This novel approach to modelling has led to reasonably strong experimentation results, showing potential that Reinforcement Learning could act as a market time traveller, detecting subtle lead-lag correlations in Foreign Exchange and positioning ahead of the curve, hence generating a unique opportunity to trade the highly liquid but extraordinarily unpredictable Foreign Exchange market.