A Study On The Default Risk Of Credit Debt Of Listed Companies In China Based On GWO-KMV-XGBoost Model

Since 1981,China’s bond market has continued to develop.The types of bond issuers have been increasing,expanding from state-owned enterprises and financial institutions to private enterprises and foreign-funded enterprises;the amount of bonds issued,too,has increased from a few hundred million initially to40 trillion yuan.In 2014,the first open market bond default occurred in China’s bond market,breaking the belief of "rigid payment" in China’s bond market and normalising the default of credit bonds.How to prevent bond default risk and credit assessment of bonds has become the focus of researchers.As an important part of corporate credit rating,listed companies are related to the smooth development of China’s economy and the fundamental interests of investors.However,China’s credit rating started late,the system of credit rating market is not perfect,and the rating methods of rating agencies are relatively backward.A credit rating method suitable for listed companies is of great significance to promote the development of the rating market.Through literature review,this paper finds that the field of credit bond assessment mainly suffers from the difficulties of small default samples,data imbalance and the difficulty for a single model to meet the requirements of high prediction accuracy and interpretability at the same time.Based on this,this paper constructs a hybrid GWO-KMV-XGBoost model to cope with the above problems and assess the default risk of China’s bond market.Due to the small default sample size,previous studies usually use "whether or not the bond is marked ST" as the prediction label.In order to improve the ability of the model to identify the real default sample,this paper selects the real defaulted bonds of listed companies in China from 2014 to 2022 as the research object,and selects non-defaulted companies as the control group according to the principles of the same industry and similar company size,etc.Considering the lag of the data,this paper uses the data of listed companies for T-2 years for model construction.Firstly,the KMV model was constructed and the default distance was calculated by inputting key indicators such as equity market value,equity market value volatility and risk-free interest rate,and a significance test was conducted on the default distance and it was found that the default distance was significantly different between defaulting and non-defaulting companies.In order to enhance the prediction accuracy and interpretability of the model,this paper adds default distance as a characteristic variable to the XGBoost model and constructs the GWO-KMV-XGBoost model using the grey wolf optimisation algorithm with high accuracy and stability.Based on the analysis of bond default factors,this paper selects 24 financial indicators and 3 bond own attribute indicators,combined with the default distance calculated by the KMV model,to measure the credit risk of listed companies in China,taking into full consideration the bond own attributes,the profitability,cash flow,capital structure,solvency,operating capacity and growth capacity of the enterprise.After independent sample t-test and Lasso＿CV feature screening,13 indicators were finally screened out.Considering the unbalanced sample size,the SMOTE method was used to oversample the defaulted company samples in the training set to equalise the sample size.Finally,the GWO-KMV-XGBoost model was used to build the credit evaluation model,and compared with the Grid Search CV-based optimised XGBoost model,the Grid Search CV-based KMV-XGBoost model and the SVM model,and found that the GWO-KMV-XGBoost model constructed in this paper The GWO-KMV-XGBoost model was found to be the best,and the model can measure the default risk of credit debt of listed companies in China.