An Empirical Study Of The Enhanced Model Of The CSI 500 Index Based On Barra And Quantile Regression

In recent years,the scale of index-enhanced fund management has increased year by year.Index enhancement,as an organic combination of active and passive investment,has the advantages of relatively transparent strategies,relatively stable styles,strong risk controllability,and high possibility of obtaining excess returns.Fama-French model and Barra model,as common multi-factor models,are scientific methods for constructing index enhancement strategies,and are widely used in actual investment.Compared to the Fama-French model,the Barra model’s factor exposure data is calculated using financial indicators,with a focus on cross-section regression,which is theoretically more accurate;and it builds a pure factor combination,so it is easier to control risk.However,the current research on the index enhancement of the Barra model has many shortcomings.For example,when designing the factor structure,only focus on the explanatory power of the model sample,and ignore the stability,persistence,and correlation between the factors;The research on prediction models is insufficient;the solution of the factor rotation phenomenon is insufficient.Based on this,based on the Barra CNE5 model introduced by MSCI,this paper uses the combination optimization method to study the index enhancement model of the CSI 500 index stocks.First,this paper builds an index enhanced benchmark model.The benchmark model uses the Barra model as the risk model and the linear regression as the return forecasting model.The index enhancement effect is achieved through the designed optimization equation.The Barra model includes 1 country factor,29 industry factors,and 10 style factors.First,the single factor validity test is carried out.The results show that the significance,volatility,stability,and correlation between the factors meet the model establishment requirements.Then use the least squares method with weights and constraints to perform cross-section regression on factor returns.Finally,when estimating the covariance matrix between stock returns,the half-life weighting,Newey West method,and systematic bias adjustment were used to adjust the factor covariance matrix and specific return variance matrix.In addition,the eigenvalue adjustment method is used to adjust the factor covariance matrix.The study found that compared with the CSI 500 Index,the index enhancement effect of the benchmark model is not obvious,which is similar to index tracking.Secondly,this paper builds an index enhanced optimization model based on the benchmark model.There are three main improvements: First,compared with conventional momentum factors,trait momentum factors have less volatility and lower risk exposure to other style factors(Lu S and Lu C,2018).So he momentum factor in the model is replaced by the trait momentum factor.Second,because the stock factor data has the characteristics of spikes and thick tails,and the quantile regression does not require the assumption of residual normality and homoscedasticity compared to linear regression.The quantile regression for such data is more robust,so this paper replaces the linear regression model of the income forecast model with a quantile regression model,specifically: construct a quantile regression model based on 8 factors: scale,beta,idiosyncratic momentum,residual volatility,equity-to-market value ratio,liquidity,profitability,and leverage,as a profit prediction model.Third,the factor exposure method of dynamic control combination optimization equations is used to solve the problem of factor rotation.The study found:(1)The fitting effect of the improved risk model is better,the average value of the adjusted R-square is increased from 0.225 to 0.255,and the risk estimation is more accurate.The deviation statistics of the expected volatility and the actual volatility have been reduced from 1.18 to 1.12.The correlation coefficient increased from 0.73 to 0.77;(2)The period from August 1,2017 to June 28,2019 was selected as out-of-sample interval.Compared with the CSI 500 Index,the index was enhanced in the case of monthly position adjustment.The annualized excess return achieved by the optimization model is 13.86%,the information ratio is 2.13,the annualized tracking error is only 6.52%,and stable excess returns are achieved in each year.This fully demonstrates that the index enhanced optimization model in this paper is more successful;(3)The performance of the optimization model significantly exceeds the benchmark model,which validates the effectiveness of the three improvements of momentum factor replacement,quantile regression model,and dynamic control factor exposure.The research in this paper successfully applies the structured risk model and quantile regression model to the field of index enhancement,which can provide theoretical support for index enhancement fund investors.From a practical perspective,the index enhancement optimization model proposed in this article can be used as a reference for index enhanced fund investors.