Credit Risk Prediction Model Based On Logistic Spline Regression

Credit risk forecasting is based on inferring corporate default status based on corporate financial and non-financial data.The company's credit risk forecasting provides important decision-making basis for stock investors,corporate bond investors,commercial bank loans,and commercial credit management of accounts receivable and payables between companies.This paper studies the credit risk prediction model based on logical spline regression.The first chapter of this paper is the introduction,the second chapter is the theoretical basis,the third chapter is based on the establishment of the optimal combination of indicators and the logical spline regression model,the fourth chapter is the empirical analysis,and the fifth chapter is the conclusion.The main focus of this paper is the determination of the number of spline function nodes.The difference in the number of nodes of the spline function will result in different credit risk prediction accuracy of the model.The number of nodes in the random selection of the sampling function will cause the prediction accuracy of the model to be too low.The second is the control of the over-fitting of the spline function.If you do not impose any constraint control on the spline function,it will lead to "over-fitting" of the model,and the robustness of the model is poor.The third is the determination of the model prediction period.The forecast period is a reflection of the duration of the forecasting ability.The innovation and characteristics of this paper: First,the model established in this paper predicts the prediction accuracy of the default state of t+s years,higher than the logical spline regression model without penalty terms and includes neural network model,decision tree,K-nearest neighbor,linear discriminant,the accuracy of the seven typical prediction models,the logistic regression model,the Gaussian na?ve Bayesian and the support vector machine model,and the data predicted by t years to predict t + s(s = 1,2,...,5)years The effect of the post-business default status.Secondly,the generalized cross-validation(GCV)method is used to determine the optimal smoothing parameters and the number of nodes of the spline function,and a high-precision credit risk prediction model is established.The third is to combine the spline function with logistic regression to establish a model,and introduce the penalty term of the spline function to control the balance between the goodness of fit of the model and the smoothness of the fitted curve,avoiding the phenomenon of over-fitting and ensuring The model is better robust.The study found that the regional characteristics of Chinese listed companies' cred it are: from high to low,the credit qualifications are East China,Central South,NorthChina,Southwest China,Northeast China and Northwest China.The industry characte ristics of Chinese listed companies' credit are: “Transportation,warehousing and postal services” have the best credit qualifications,“wholesale and retail” credits are in the middle,and “real estate” industry has the worst credit qualifications.