Research On Real Estate Bond Default Risk Based On LightGBM

Due to the rapid development of China’s bond market,bond defaults occur frequently,and their number and scale have increased dramatically.Since 2019,with the prevention and resolution of credit risks in my country’s bond market,the risk of default in my country’s bond market has been alleviated to a certain extent.However,considering that real estate companies have the characteristics of sufficient cash flow,frequent policy regulation,and high difficulty in overall refinancing,they have natural characteristics of high credit risk.Due to the fierce competition in my country’s real estate industry and the continuous increase in the concentration of real estate companies,coupled with changes in the real estate financing environment and the adjustment of the development pattern of the regional real estate market,the problem of real estate bond defaults has become more and more serious,and the number and amount of them will continue to rise.In 2022 Both the default amount and quantity of real estate bonds exceeded 60%of the annual total.Real estate is an asset with dual attributes of a commodity and an investment product,so in the event of a default,it often brings greater losses than other industries.In addition,the operation and transactions of the real estate market require multi-faceted cooperation and coordination.Once a default event occurs,its impact may spread to the entire market and have a negative impact on the entire economy.Therefore,the research on the default risk of real estate bonds is very important.The occurrence of real estate bond default events is usually caused by various factors,usually external economic factors and corporate factors.Due to the special attributes of the real estate industry,the previous bond default risk early warning model can no longer meet the existing needs.Based on the characteristics of the real estate industry and the analysis of the reasons for the default of real estate bonds,this paper selects 20 indicators as explanatory variables from three aspects:the external environment,the enterprise itself and the bond project,and whether the default is the explained variable.Due to the good performance of the decision tree model in classification problems,three models of LightGBM,Random Forest,and XGBoost were selected for research,and compared with the logistic regression model.This paper takes the maturing bonds issued by my country’s real estate companies as the research object,selects 102 bonds that have substantially defaulted in the real estate industry as default samples,and selects 612 non-default bonds as control samples in a ratio of 1:6.External indicator data and financial data select the data of the year before the default date or the due date as sample data for default prediction.The total sample is divided into training set and test set according to the ratio of 8:2,and stratified sampling is set at the same time,that is,the ratio of default samples to non-default samples in training set and test set is 1:6.In order to solve the problem of unbalanced sample distribution,Adasyn oversampling is used to rebalance the training set to improve the learning effect of the model on default samples.By constructing LightGBM,random forest,XGBoost and logistic regression models to predict the default risk of real estate bonds,the results show that the LightGBM model performs best in predicting the default risk of real estate bonds.Combined with the conclusions obtained,the research on the default risk of real estate bonds based on the LightGBM model is beneficial to the bond regulatory authorities to strengthen the early warning and risk control capabilities of real estate corporate bond defaults,which has played a role in promoting the smooth operation of real estate companies and has a positive effect on promoting the development of my country’s bond market.Certain reference value.