Learning FX trading strategies with FQI and persistent actions

Automated Trading Systems are constantly increasing their impact on financial markets, but learning from historical data, detecting interesting patterns and producing profitable strategies are still challenging objectives for autonomous agents. This holds true especially in the intraday Foreign Exchange market, where prices are heavily affected by random noise and high non-stationarity. In this volatile market, opportunities are present at many time-scales, but not all of them can be easily learnt. The signal-to-noise ratio has, indeed, a critical impact on the ability of autonomous agents to learn effectively. In this paper, we formulate multi-currency trading as a Markov Decision Process and we train an agent via Fitted-Q Iteration, a Reinforcement Learning value-based algorithm. Focusing on a three-currencies framework, we study the importance of tuning the control frequency, in order to obtain effective trading policies. We backtest the developed approaches on real data from the FX market considering two currency triplets, comparing results employing either a single pair or both ones at the same time.