Research On Stock Pairs Trading Algorithm Based On Clustering

The stock pairs trading strategy based on statistical arbitrage is a classical quantitative strategy,which is independent of the market trend.The stock pairs trading strategy has become promising since the official launch of securities margin trading,which not only perfects the Chinese securities market,but also makes the stock pairs trading strategy more profitable.However,its characteristics restrict the traditional stock pairs trading strategy.And there is a strong prior relationship between two homogeneous stocks in traditional pairs trading algorithm,which makes only a few stocks available for the strategy.Aiming to address the above problems,this paper improves the stock pairs trading algorithm via machine learning algorithms.Main research is as follows:(1)A multi-stocks pairs trading algorithm based on AdaBoost-ElasticNet is proposed.We selected stocks from different industries by clustering to break the industry barrier,then we used the AdaBoost-ElasticNet algorithm to find the quantitative relation in multi stocks.The backtest results show that the proposed algorithm reaches the cumulative return of 80.48% with a lower max-drawdown ration and a higher sharpe ratio during the period from April 16,2016 to April 16,2018.(2)According to K-means++ algorithm and the financial factor data,a virtual market index is constructed.And based on it and generalized linear algorithm,a multistocks pairs trading algorithm is presented.Experimental results show that the proposed algorithm reaches the cumulative return of 31.55% during the period from August 3,2018 to December 31,2018,and the XGBoost based on linear model is the best.