Nonlinear relationships in soybean commodities Pairs trading-test by deep reinforcement learning.

• Fills a gap in the use of DRL in futures trading. • Demonstrates the ability of DRL to gain profits from the nonlinear relationship between assets and outperform traditional linear methods in dealing with trading tasks. • Provides a quantitative investment strategy for reference. • For the whole futures market, the application of pairs trading strategy in China's soybean futures market is conducive to establish a more appropriate and effective price discovery mechanism. The pairs trading strategy involves selecting two highly correlated securities to profit from mean reversion. However, the traditional simple threshold method is subjective, random, and ignores nonlinear relationships. This paper proposes a new cointegration deep reinforcement learning (DRL) pairs trading model applied to Dalian Commodity Exchange futures to capture nonlinear relationships and gain profits. The CA-DRL model outperforms other models in terms of efficiency and performance. [ABSTRACT FROM AUTHOR]