基于强化学习的配对交易投资策略实证研究  

An Empirical Study on Paired Trading Investment Strategy Based on Reinforcement Learning

作　　者：黄圳峰 Huang Zhenfeng(School of Economics and Management,North University of Technology,Beijing 100043)

机构地区：[1]北方工业大学经济管理学院(北京市石景山区),北京100043

出　　处：《现代计算机》2021年第30期17-23,31,共8页Modern Computer

摘　　要：针对配对交易策略目前存在的套利空间小、投资收益低等问题,本文基于强化学习算法构建配对交易策略,并以2010-2016年期间美国公共事业股的收盘价作为研究对象,验证配对交易策略的投资绩效。研究结果表明,相较于传统的配对交易策略,基于强化学习算法的配对交易策略避免了经验参数对于交易结果的不利影响,可以更好的捕捉潜在的交易机会,在夏普比率、年化收益率等指标上表现更加优异,因而将强化学习算法引入配对交易当中可以为投资者提供一种有效的套利手段和风控工具。In view of the existing problems of small arbitrage space and low investment returns of paired trading strategy,this pa⁃per builds a paired trading strategy based on reinforcement learning algorithm,and takes the closing price of US public utility stocks from 2010 to 2016 as the research object to verify the investment performance of the paired trading strategy.The research results show that compared with the traditional paired trading strategy,the paired trading strategy based on reinforcement learning algorithm avoids the adverse effects of empirical parameters on trading results,can better capture potential trading opportunities,and has better perfor⁃mance in Sharpe ratio,annualized return rate and other indicators.Therefore,the introduction of reinforcement learning algorithm into paired trading can provide investors with an effective arbitrage method and a risk control tool.

关 键 词：配对交易 DQN算法 统计套利 

分 类 号：F831.51[经济管理—金融学]