| This paper studies the logistic credit scoring model about the variables selection. Variable selection is not only a problem to be considered in the process of modeling establishment, also a very important issue in the statistical study. The model need to ensure the variables selected should reflect the characteristics of the object to be studied, and ensure that the results obtained with a better fitting and predictability.The main important work of this paper:First Systematical introduced variable selection methods:subset selection method and the coefficient compression method. Subset selection method includes AIC, BIC and stepwise regression. The coefficient compression method includes ridge regressionã€LASSO and adaptive LASSO.Second Focusing on the adaptive LASSO variable selection of Logistic model and parameter estimation, the core is to discuss adaptive LASSO penalty variable term weighting with the value of variable information, and parameter estimators obtained ORACLE nature.Third The empirical analysis mainly through the Shenzhen Stock Exchange 148 SMEs and 30 credit metrics system variable to compute value of information as a penalty term weight of the adaptive LASSO variable selection, then establish logistic credit scoring model, using ratios to explain the occurrence of regression coefficients (?), and through economic sense test that variable model includes the ability to cover the basic indicators of enterprises. Finally, by comparing stepwise regression and logistic LASSO two variable selection method established credit scoring models,using ROC curve, KS value, WGRP and CIER indicators to compare and verify the ability to distinguish between risk estimation accuracy of three models, obtain adaptive LASSO establish logistic model is better than other models of the two methods, but also shows the adaptive LASSO variable contains the value of variable information selection method has good performance in the empirical test. |
PDF Full Text Request