Study On Bayesian Elastic Net Quantile Regression For Panel Data Models And Its Applications

Panel Data,as a two-dimensional data type that combines cross-sectional data with time-series data,can provide more sample information and has a wide range of applications in many fields such as economics,management,biology and so on.The traditional panel data model is actually a conditional mean model,which can only describe the mean information of the dependent variable,while ignoring other information.The quantile regression method considers the influence of the independent variable on the dependent variable at different quantile points.Compared with the conditional mean model,it not only has better robustness,but also can measure the influence of the independent variable on the tail of the dependent variable,providing far richer information.Therefore,more and more scholars adopt panel data quantile regression model,and how to automatically select important explanatory variables while estimating the parameters of the model has been one of the hot issues worth exploring.Based on the panel data quartile regression model,this paper proposes a Bayesian Elastic Net quantile regression model(BQR.EN)for panel data using Elastic Net,a penalty method for dealing with high-dimensional data and highly correlated data.We derive the full condition posterior distribution density function of each parameter,and construct the Gibbs sampling.In the simulation experiment,Firstly,we compared the BALQR model,the BLQR model and the BQR model in many cases,and then the BQR.EN model proposed in this paper was used to estimate the parameters under different disturbance hypotheses and different sample sizes.The comparison shows that the BQR.EN model has obvious advantages in dealing with high-dimensional data and highly correlated data,and to some extent makes up for the deficiency of the BLQR model based on Lasso penalty and the BALQR model based on adaptive Lasso penalty;A good estimation result can also be obtained even if the actual distribution of the item does not conform to the ALD distribution or the sample size is small.In empirical analysis,in order to explore the influencing factors of EVA of listed companies in the Internet Finance field,this paper constructs a Bayesian Elastic Net quantile regression model of panel data,demonstrating the ability of the new method to estimate parameters and select variables in practical problems.The empirical results show that profitability,innovation ability and company size have the greatest impact on the economic value added of listed companies in the Internet Finance field.