Research On The Logistic Regression Method With Punishment And Its Application In The Financial Early-warning Of Enterprises

With the development of the information technology,the data has infiltrated every industry and functional areas and has gradually become an important factor of production.The variable selection is an important way to extract valid information.Since 1996,Tibshirani proposed Lasso method,which has greatly promoted the study of variable selection method.There have been many variable selection methods,such as Lasso,Ridge regression,Elastic-Net,SCAD,MCP and so on.These methods impose penalty functions on the basis of least squares to achieve variable selection.The purpose of this paper is to compare the application effects of these five methods through numerical simulation and empirical analysis.Comparing the results,it can be found the predictive accuracy of variable selection models is higher than all-variable Logistic model.The Lasso and SCAD methods have strong applicability in different data structures,high regression accuracy and small coefficient estimation error can be guaranteed.Ridge regression and Elastic-Net tend to retain more variables,although the prediction accuracy is higher but the coefficient estimation error is larger.The MCP method is suitable for the strong correlation of variables.The empirical study of the financial data of listed companies in Shanghai and Shenzhen A manufacturing industry shows that the penalty function of SCAD and Lasso have good forecasting effect and economic explanation ability and can identify the important factors of profitability,growth ability,management ability and capital structure.