Equity Trading at the Brazilian Stock Market Using a Q-Learning Based System

This work aims to show how an intelligent agent trained by the Q-Learning algorithm can achieve satisfactory results, using non-linear, non-stationary and noisy financial data. In order to evaluate the system performance, a stock market simulator was developed. The agent outperformed the Buy & Hold and two Technical Analysis strategies in most cases, using technical indicators as input data and a modified version of daily return as the reward function.