Research On Default Risk Measurement Of Credit Debt Based On KMV Model And Machine Learning

In 2014,the "11 Chaori Bond" had a substantial default.Since then,the "rigid payment belief" of my country’s credit bonds has been shattered,by the end of 2021,the scale of defaulted bonds has increased by 85 times in the past seven years,which has attracted great attention from the market,and the types of defaulting entities have also undergone significant changes,from initially privately dominated to high-rated state-owned enterprises defaulting.Combined with the turbulent international situation and the negative impact of COVID-19 on the economy,the domestic economy will still face downward pressure in the future.Against this background,this paper attempts to use the default distance output by the KMV model as the feature index of the machine learning model,using the dynamic advantages of the KMV model to improve the performance of the machine learning model,so as to study and measure the default risk of credit debt,and compare the prediction accuracy of three machine learning models.This paper firstly sorts out the domestic and foreign researches on the default risk of credit bonds,and analyzes the current situation of the default of credit bonds in my country.Secondly,the 2019-2021 A-share listed companies in the non-financial industry are selected as the research objects,and the "ST" and real default samples are classified as the risk sample group,and the rest are the control group.This paper takes one year as the forecast period,collects and organizes the stock market data of the previous year of the sample from 2019 to 2021,and empirically tests the applicability of the KMV model.Then,the default distance is combined with the financial data and macroeconomic data of the corresponding year to construct a characteristic indicator system.After preprocessing the samples and data,the data set is divided into training set and test set,and the SMOTE algorithm is used to balance the sample categories.By optimizing the experimental settings such as model parameters,three machine learning models of support vector machine,random forest and BP neural network are simulated and trained respectively.Finally,the performance of the three machine learning models on the test set is compared,and an ablation experiment is designed to verify the effectiveness of the default distance indicator.The experimental results show that:(1)The default distance output by the KMV model is significantly related to the default risk,that is,the default distance can be used as a powerful indicator to identify the default risk.(2)The prediction ability of BP neural network is better than that of support vector machine and random forest,and its prediction accuracy on the test set reaches 80.7%,which is a relatively more suitable model for the measurement of default risk of listed companies in my country.(3)After excluding the default distance index,the performance of the three models all declined,indicating that the default distance index with dynamic advantages has made a certain contribution to the discriminant accuracy of the model.