Pairs trading strategy optimization using the reinforcement learning method: a cointegration approach

Recent studies show that the popularity of the pairs trading strategy has been growing and it may pose a problem as the opportunities to trade become much smaller. Therefore, the optimization of pairs trading strategy has gained widespread attention among high-frequency traders. In this paper, using reinforcement learning, we examine the optimum level of pairs trading specifications over time. More specifically, the reinforcement learning agent chooses the optimum level of parameters of pairs trading to maximize the objective function. Results are obtained by applying a combination of the reinforcement learning method and cointegration approach. We find that boosting pairs trading specifications by using the proposed approach significantly overperform the previous methods. Empirical results based on the comprehensive intraday data which are obtained from S&P500 constituent stocks confirm the efficiently of our proposed method.