Pairs Trading Strategy Optimization Using Proximal Policy Optimization Algorithms

Yi-Feng Chen, Wen-Yueh Shih, Hsu-Chao Lai, Hao-Chun Chang, Jiun-Long Huang

Abstract: Pairs trading is a market-neutral quantitative trading strategy which exploits historically correlated stock prices by forming pairs with weighted long and short positions. A pair of opened offsetting positions can profit regardless of positive or negative price trend. Positions are opened when the spread exceeds the trading boundary, and closed when the spread reverts back to the historical mean. In this paper, we adopt proximal policy optimization, which is a deep reinforcement learning algorithm, to determine the trading boundaries as well as stop loss boundaries for maximizing the profit in pairs trading. Besides, we propose to utilize a demonstration butter to pre-train the model for better training efficacy. The experimental results manifest that the proposed method outperforms state-of-the-art strategies in terms of investment return and investment risk in both the Taiwan stock market and the United States stock market.