Research On Barra Factor Stock Selecting Model Based On Logistic Regression

Quantitative investment is a new type of investment method.Compared with traditional investment methods,it can quickly acquire and process large amounts of data.At the same time,it also has the characteristics of being insensitive to emotions,maintaining analytical methods,and consistent investment logic.China's quantitative investment is still in its infancy,and the relevant theories are not yet mature.Therefore,it has great development space and research value.The multi-factor stock selection strategy is one of the most widely applied and researched strategies in China's quantitative investment field,and has the advantages of being flexible and adaptable to changing market conditions.The core of multi-factor stock picking strategy lies in the construction of multi-factor model.The construction of model includes two parts,factor selection and model selection.The choice of factor lies in the determination of explanatory variables.The choice of model lies in the determination of the integration of explanatory variables.However,the vast majority of scholars now focus their research on factors that can better explain the rate of return.On the model selection,the multiple linear regression model is used by default.This method actually ignores the research on model level.The complete study of multi-factor models in the broad sense should include both model research and factor research.Based on this,this paper studies the multi-factor selection model from the factor level and the model level respectively.At the factor level,this paper first studies the basic principles,assumptions,and model conclusions of the linear multi-factor model and the Logistic multi-factor model.Starting from the assumptions,the rationality and completeness of factors in the Fama-French three-factor model,the Fama-French five-factor model,and the Barra risk model were analyzed and compared and determine the 10 risk factors and 28 industry factors in the Barra risk model as the model's alternative factors;secondly,the paper analyzes the construction method,theoretical connotation and data sources of the Barra factor,constructs 38 Barra factors,including industry factors,and uses the simple locator method and industry-neutralized quartile method to empirically test the ability of single factor predicting the yield of factors.At the model level,this paper firstly constructs Barra's factor selection model based on Logistic regression,and tests and analyzes the merits of the model's stock selection ability in the ideal environment without transaction costs and in the actual environment with transaction costs and researches and analyzes the differences of the stock selection ability under the two kinds of environments;Secondly,this article compares the Barra factor selection model based on Logistic regression.On the one hand,the paper compare the stock selection model that does not include industry factors and the stock selection model that incorporates industry factors.On the other hand,the paper compare the stock picking ability of Barra's factor selection model based on multiple linear regression and Barra's factor selection model based on Logistic regression.By using the daily data from January 1,2013 to December 31,2017 for back analysis,this paper finds that 10 risk factors in the Barra China stock market model can effectively predict the yield.Including Beta,Momentum,Btop,the higher the Liquidity,Earnyield,Growth,and Leverage factors,the higher the short-term return rate of the company,and the higher the Size,Resvol,and Nsize factors,the lower the short-term return rate of the company;A stock selection strategy based on Logistic regression constructed using 10 Barra risk factors and 28 industry factors can achieve an absolute return that exceeds the benchmark of the CSI 300 Index.The strategy can achieve 60%annualized yield and 1.64 sharp rate under the ideal environment without consideration of transaction costs.After considering the transaction costs,the strategy returns decreased greatly,but still can obtain 17%annualized rate of return and 0.46 of the Sharpe rate and both indicators are better than the benchmark of the CSI 300 index.The strategy has theoretical research value and practical application value;the stock selection model incorporating industry factors is superior to the stock selection model that does not include industry factors in terms of profitability or stability.The Barra factor selection model based on Logistic regression is superior to the Barra factor selection model based on linear regression both in the profitability and stability of the model.So Barra's factor selection model based on Logistic regression has relatively better stock selection ability.Barra factor selection model based on Logistic regression has some limitations.On the one hand,because the explanatory variable and the explanatory variable in Logistic regression model are not in a one-to-one correspondence,it is impossible to further study and analyze the contribution of a single factor to the rate of return,and it is impossible to accurately analyze the return rate of attribution.On the other hand,the model has some reflections.Although the technical factors of short-term market sentiment can flexibly capture market anomalies,frequent transactions also increase the cost of the strategy,as well as the volatility and downside risk of the strategy.How to further analyze the specific source of the model revenue and reduce the volatility and downside risks of strategy need further research and analysis.