Recently, the Department of Management Sciences of the National Natural Science Foundation of China (NSFC) organized the 2024 Project Conclusion Performance Evaluation Meeting. The National Natural Science Foundation project "Study on Credit Risk Measurement of Corporate Bonds Based on Bayesian Generalized Linear Mixed Models" (Project No.: 71901230), undertaken by Associate Professor Yao Xiao of our school, was rated "Excellent" at the meeting.
The measurement of corporate credit risk is not only related to the interests of investors but also affects the stability of financial markets. This project constructs a credit risk prediction framework by combining machine learning and econometric models, explores influential factors with information value, and improves the effectiveness of credit risk prediction. Specifically, the research work of this project mainly includes the following aspects:
Based on the detailed collection of corporate default information, it in-depth explores the influencing factors of different degrees of default events of listed companies. Starting from the imbalance of samples, a default risk model for corporate bonds is established, and the generative adversarial network technology based on Wasserstein distance is used to rebalance the sample proportion, thereby effectively improving the prediction effect of the default risk model.
It explores the application of recovery information in the loan default loss rate. By constructing a hierarchical learning framework with recovery information as an intermediate variable, the connection between loan application characteristics and default loss rate is established, so that both application characteristics and recovery information can be used to improve the prediction effect of default loss rate.
It expands the research on predicting the expected loss and return of loan credit risk. By using various research methods such as survival analysis and machine learning to predict the prepayment risk and default risk of loans, it can comprehensively describe the expected loss and return of loans and apply them to loan pricing.
It uses a variety of different text analysis and deep learning methods to extract the text tone from corporate annual reports and apply it to predict the credit risk of listed companies.
It further explores the influencing factors of corporate bond issuance spreads from multi-channel information disclosure and evaluates the credit risk of enterprises from the perspective of financing costs.
The project research finds that the credit risk and debt financing cost of enterprises are not only affected by financial factors but also reflected through other forms of characteristics such as corporate text information. Moreover, the evaluation of corporate credit risk needs to consider not only default risk but also recovery risk and prepayment risk to comprehensively assess the expected loss. This research can not only help market participants better understand and evaluate the credit risk of enterprises but also highlight the important role of information disclosure in risk management and financial product pricing. It also helps regulatory authorities to further standardize and improve the information disclosure mechanism of enterprises and promote the healthy and sustainable development of the bond market.
Relying on the support of the foundation, the project has carried out a series of research around the research objectives and published a total of 10 journal papers in domestic and foreign authoritative journals. These include international authoritative journals such as European Journal of Operational Research, Annuals of Operations Research, International Journal of Forecasting, Knowledge-Based Systems, International Review of Financial Analysis, and Economic Modelling, as well as important domestic journals such as Journal of Management Sciences in China, Systems Engineering - Theory & Practice, and Economic Theory and Business Management.
