Agenda

Titolo completo: Q-Learning and SARSA: Machine learning-based stochastic control approaches for financial trading

Abstract: In this paper, we propose some novel automated financial trading systems (FTSs) based on two different machine learning techniques, respectively known as Q-Learning (QL) and State-Action-Reward-State-Action (SARSA). Both QL and SARSA belong to the family of the Reinforcement Learning methodologies. Such methodologies real-time optimize their undertaking accordingly to the responses they get from the environment in which they operate. In particular: first, we introduce the basic aspects of RL, QL and SARSA which are of interest for our purposes; then, we present the novel and differently configurated FTSs based on QL and SARSA we propose; finally, we apply these FTSs to eight time series of daily closing stock returns from the Italian stock market, all belonging to the FTSE MIB basket. The results we achieve are generally satisfactory.