Research On Default Risk Evaluation Of Credit Bonds Based On Heterogeneous Information Network

At the initial stage of the development of China’s bond market,there has always been an implicit rule of "rigid payment" in the market.However,when the"11 Chaori" bond defaulted in 2014,the "zero default" of China’s bonds became history.Especially in recent years,with China’s industrial restructuring,domestic and international political turmoil,and the impact of the COVID-19 since 2020,the global economic situation has turned sharply downward,and bond defaults have occurred frequently,Among them,there are also AAA events default of state-owned enterprises,such as Yongmei Event and Huachen Event.Especially in 2021,the credit risk of state-owned enterprises accelerated to release,and the proportion of new default subjects of state-owned enterprises exceeded that private enterprises for the first time,leaping to the first place,causing market concern.Under the background of accelerated exposure of bond default risk,China’s risk identification of bonds and supervision of bond market have been controversial.The traditional bond risk identification often uses the bond’s information and the financial information of the issuing companies.In recent years,some scholars have further considered the relevance of various enterprises in the financial system and introduced the perspective of complex network to describe the contagion of risk among the issuers.However,due to the one-to-many relationship between issuers and bonds,as well as the different sizes,types and maturities of different bonds,it is difficult to accurately describe the risks of specific bonds based on the complex network of issuers only.In order to better identify the default risk of credit bonds,on the basis of heterogeneous information network technology,this paper designs and constructs a weighted heterogeneous information network with multiple node relationships by using the issuance relationship between the bonds and the issuers and the shareholder relationship between the issuers,on the basis of this network,excavates the relevant features that reflect the risk contagion and the complex correlation between the nodes.In view of the imbalance of positive and negative samples of credit debt data,this paper proposes the Optuna-Imbalance-Xgboost algorithm based on the Imbalance-Xgboost algorithm,and combines the heterogeneous information network and related features proposed in this paper to effectively improve the recognition ability of credit debt default.According to the existing literature and relevant research,this paper selects 4 basic information indicators of bond issuance and 15 financial indicators of issuing companies in T-1 year,and extracts 5 network features from heterogeneous information networks.Using the real data of credit bonds in 2017-2021,this paper trains and tests model under the framework of random forest and Xgboost algorithm,Combined with the model evaluation indicators and the importance of features,the effectiveness of the network and related features established in this paper is verified.Further,based on the basic indicators,financial indicators and network indicators of bonds,this paper compares Xgboost model,Imbalance-Xgboost model and Optuna-Imbalance-Xgboost model to verify the effectiveness of the model proposed in this paper.Finally,the interpretability of Optuna-Imbalance-Xgboost is enhanced by combining the SHAP interpretability method and the validity of heterogeneous information network features is verified here.Through theoretical and empirical research,this paper draws the following conclusions:(1)The features extracted from the heterogeneous information network established in this paper can significantly improve the performance of the bond default risk assessment model.(2)Compared with Xgboost algorithm,the proposed Optuna-Imbalance-Xgboost algorithm can significantly improve the performance of the model.(3)From the analysis of feature importance and SHAP interpretability,it can be seen that although network features can play an important role in the identification of bond default,the traditional financial features are important indicators for the identification of credit bond default.