Paired Trading Strategy Optimization Using the Reinforcement Learning Method: Intraday Data of Tehran Stock Exchange

Objective: Paired trading is among the most well-known and oldest algorithmic trading systems. The efficiency and profitability of this system have been demonstrated in many studies conducted so far in financial markets. Paired trading is principally based on long-run equilibrium relationships or reverting to the mean characteristic. In recent years, a large number of studies have been conducted on algorithmic trading using machine learning. Methods: In this research, the reinforcement learning method - an appropriate method for modeling and optimizing problems involving different long-run relationships - was used in order to select appropriate trading thresholds and time windows for the purpose of maximizing efficiency and minimizing negative risks in paired trading through adopting the co-integration approach. Results are obtained by applying a combination of reinforcement learning method and co-integration approach in paired trading. Results: Empirical results based on the intraday dataof paired stocks showed that the reinforcement learning method used to design trading systems in paired trading had significant advantages over the other methods in previous works. Conclusion: A pair trading strategy with the proposed algorithm can be used as a neutral market strategy in all market conditions, including prosperity and recession, by investors and individual and institutionaltraders.Also, for future research, it is possible to consider transaction costs in a pair trading strategy.