The 17th Academic Luncheon of the Digital Technology and Finance

Abstract：

Pairs trading is an important strategy in hedge funds. The arbitrage opportunities of this strategy have shrunk over recent years due to its broad application and the constraints of conventional approaches. This project aims to address the limitation of conventional pairs trading strategies that can only use the linear correlation， deviation and regression between asset prices. A novel quantitative trading approach is proposed in this project based on a Deep Reinforcement Learning algorithm framework. The main research contents are as follows: (1) using genetic algorithms to optimize the trading threshold in traditional pairs trading strategies to examine the full potential of conventional pairs trading strategies; (2) introducing a Deep Reinforcement Learning-pairs trading model to solve four key issues: reward function design, Deep Reinforcement Learning brain network structure design, invalid training problem, and computational complexity problem; (3) conduct the empirical analyses for different commodity pairs in both domestic market and global market. Meanwhile, this project solves the problem of multi-asset pairing and investigates the impact of pre-selection on the Deep Reinforcement Learning model.

Speaker Introduction：

Xiangyu Zong holds a PhD from Adam Smith Business School, University of Glasgow. He is engaged in the research of machine learning, complex system theory and quantitative trading. He has published papers on machine learning and complex systems theory in international journals such as Energy Economics, Economic Modelling, Finance Research Letters, and acted as an anonymous reviewer for several international first-class journals.