Q-Learning-based financial trading systems with applications

The design of financial trading systems (FTSs) is a subject of high interest both for the academic environment and for the professional one due to the promises by machine learning methodologies. In this paper we consider the Reinforcement Learning-based policy evaluation approach known as Q-Learning algorithm (QLa). QLa is an algorithm which real-time optimizes its behavior in relation to the responses it gets from the environment in which it operates. In particular: first we introduce the essential aspects of QLa which are of interest for our purposes; second we present some original FTSs based on differently configured QLas; then we apply such FTSs to an artificial time series of daily stock prices and to six real ones from the Italian stock market belonging to the FTSE MIB basket. The results we achieve are generally satisfactory.