Optimal Design Of Contrarian Investment Strategy For CSI 1000 Index

Contrarian investment strategy is a kind of active management strategy,which makes use of the "reversal effect" and " winner-loser effect " in the market,and considers that the investors in the market are not completely rational,that is,there are behavioral financial phenomena.The idea of contrarian investment is to look for stocks with sound fundamentals that investors are not paying attention to,or to look for stocks that deviate from their intrinsic value due to overreaction caused by unexpected events in the market.This thesis innovatively combines the machine learning algorithm of artificial intelligence with contrarian investment strategy to explore its effect on the China Securities 1000 Index,that is,several indicators that the contrarian investment strategy focuses on: Low price-to-book ratio,low price-to-earnings ratio,low price-to-cash ratio,high dividend and lower than the industry pricing index are brought into the machine learning algorithm to explore the result from investment.On the premise of preserving the value investing philosophy,this thesis looks for stocks with sound fundamentals,such as corporate management and earnings,but not over-hyped yet.Combined with machine learning,a front-end artificial intelligence algorithm,data mining,data analysis,and a series of processes such as investment decision judgment,it tries to use information technology to optimize the contrarian investment strategy,and to obtain higher and more stable excess returns.The stock pool in this thesis is the component stocks of the CSI 1000 index,mainly 1000 stocks with small market value but relatively good liquidity in China’s A-share market.Since these stocks are not often overly concerned,it is very suitable for the application of contrarian investment strategy.The time span is from 2007 to 2021,and the stocks with ST or suspension during the period are eliminated,the stocks listed less than 3 months are eliminated.According to the characteristics of sample stocks selected by the contrarian investment strategy,and then the data are preprocessed,the machine learning algorithm is used to carry out rolling training on the data in the sample,the importance of each data feature is output,and the back test fund curve is generated.The three groups of algorithms are compared and optimized,and the ideal investment scheme is finally obtained.Based on the above research process,the three machine learning algorithms can continuously and stably obtain excess returns in the CSI 1000 index.According to the total rate of return,the best learning effect is gradient boosted decision tree,followed by support vector machine algorithm,and finally is random forests algorithm.In terms of maximum drawdown,the random forests algorithm has the lowest maximum drawdown,followed by gradient boosting decision tree and support vector machine.According to Sharpe Ratio,the gradient boosting decision tree is slightly better than the support vector machine,and the random forests performs relatively poorly.At the same time,it also shows that the current CSI 1000 index market has not reached the semi-strong efficient market.