SGL Threshold Regression Model And Its Application In Stock Analysis

With the arrival of the big data era,the data sources in the social economy are more diversified.In the stock market,the influencing factors of stock returns have always been a hot issue in the study.According to the relevant literature of scholars both at home and abroad,the previous studies on stock returns mainly studied the macroeconomic and microeconomic indicators respectively.However,with the diversification of data sources,the number of economic indicators available for stock return research soars.In the current study,however,the industry is also keen to find out those factors that have a significant impact on yield.In addition,the research on stock returns should not be apart from macroeconomic indicators and microeconomic indicators,and some studies have shown that macroeconomic indicators have a threshold effect on the stock return.This paper proposes a threshold regression model with Sparse Group LASSO penalty,which can not only estimate the threshold of the regression model but also make variable selection of explanatory variables.In order to correctly estimate the threshold of the regression model,we use the method of smoothing least squares to solve the extremum of the objective function by Newton-Raphson method and obtain the estimated threshold.In order to judge whether the estimated threshold value is a true threshold value or not,Sparse Group LASSO penalty is added to the objective function to select the variables so that the judgment of the true threshold value is converted to the variable selection problem.In order to test the parameter estimation of SGL threshold regression model and the validity of model fitting,Monte Carlo simulation method is used to estimate the parameters and model fitting of the first-order SGL threshold regression model,the second-order SGL threshold regression model and other control models comparing.In the Monte Carlo simulation,the second-order SGL threshold regression model has better estimation accuracy and robustness than the other four models,and the accuracy of the estimation of the key parameters and real threshold can reach 100%.The model fitting effect of the second-order SGL threshold regression with the small sample size n = 100 is also better than the other models,and the overall fitting of the SCAD threshold regression model is better when the sample size is larger.This article applies the SGL threshold regression model to the stock market to study the influencing factors on the yield of Shanghai-Shenzhen 300,China Securities 500 and China Securities 800 Index.The 67 explanatory variables are divided into three categories: corporate financial indicators,fundamental indicators and style factor indicators,The threshold variable for the national consumer price index.Empirical,the second-order SGL threshold regression model model is better than the other four sets of control model,the results show that the CSI 300 index has two true threshold rate of return: CPI = 102.244 and CPI = 103.723;CSI 500 Index The yield also has two true thresholds: CPI = 102.026 and CPI = 104.037.There are also two true thresholds for the CSI 800 index return: CPI = 102.026 and CPI = 104.03.The result expands the existing research that only one threshold of stock returns CPI = 102,found in the vicinity of CPI = 104 will also have a threshold effect on stock returns.