| For the design of quantitative investment strategy,considering that no style factor can produce stable alpha income for a long time,a multi factor dynamic position adjustment investment strategy based on XGBoost model is designed.Firstly,based on the factors in the factor analysis of wind quantitative platform,the factor library is preliminarily constructed;Secondly,through data preprocessing and factor validity test,seven typical factors such as return on net assets are selected as stock selection multi factors;Then,XGBoost model is used to analyze the importance of factor characteristics,factor IC prediction and single factor timing analysis;Then,the multi factor dynamic position adjustment demonstration is carried out by synthesizing factors.Taking China Securities 500 component stocks as the stock pool and taking January 4,2012 to December 31,2021 as the sample research period,the factor equal weight strategy,the dynamic position adjustment strategy based on XGBoost fixed training and the dynamic position adjustment strategy based on XGBoost rolling training are used for back testing respectively,and the net worth curves outside the sample of the three strategies are obtained;Then,in order to reduce the erosion of transaction costs on returns,consider the multi factor dynamic position adjustment strategy to limit the turnover rate;Finally,the back test results of XGBoost dynamic warehouse adjustment rolling model strategy,factor equal weight and dynamic warehouse adjustment fixed model strategy are compared.The back test results show that the middle-aged return rate of XGBoost rolling training dynamic position adjustment strategy in the four-year back test is 22.59%,which is higher than 10.81% of the factor equal weight strategy.At the same time,the cumulative return rate is as high as 109.42% and the information ratio is 1.85.In the loose years of market economy,the multi factor strategy based on rolling training of XGBoost model can obtain higher excess return;In the year of drastic changes in market style,this strategy can also respond quickly,control the pullback to the minimum level,and obtain better performance than the factor equal weight strategy.The advantages of the design of the multi factor stock selection strategy based on XGBoost model are as follows: Firstly,in the selection of factors,the strategy not only considers the style factors based on financial and technical aspects,but also introduces market variables such as the Monthly rise and fall of CSI 300 index and macro variables such as CPI year-on-year,so as to enhance the prediction ability of style factors;Secondly,the comprehensive use of regression method,factor IC method and factor cumulative return method to test factors,and the factor effectiveness test method is more systematic and diverse.Finally,in order to reduce the transaction cost,the strategy constructs a stock buffer pool to limit the turnover rate of the strategy and enhance the practicability of the strategy.The empirical results show that when the number of shares in the stock buffer pool accounts for about 10%,the erosion of the transaction cost to the investment income can be effectively reduced.The design and application of this strategy provides a new practical perspective for multi factor stock selection,and has certain guiding significance for optimizing investors’ asset allocation and improving investment returns. |
