Foreign exchange trading: a risk-averse batch reinforcement learning approach

Automated Trading Systems' impact on financial markets is ever growing, particularly on the intraday Foreign Exchange market. Historically, the FX trading systems are based on advanced statistical methods and technical analysis able to extract trading signals from financial data. In this work, we explore how to find a trading strategy via Reinforcement Learning by means of a state-of-the-art batch algorithm, Fitted Q-Iteration. Furthermore, we include a Multi-Objective formulation of the problem to keep the risk of noisy profits under control. We show that the algorithm is able to detect favorable temporal patterns, which are used by the agent to maximize the return. Finally, we show that as risk aversion increases, the resulting policies become smoother, as the portfolio positions are held for longer periods.