Modelling of Chinese corporate bond default – A machine learning approach

We apply machine learning techniques to construct a series of models of corporate bond defaults. By combining Chinese accounting information and corporate bond data from January 2012 to December 2019, we construct an out-of-sample forecast that significantly improves the identification rate of corporate bond defaults, with an area under the receiver operating characteristics curve greater than 90 percent. Our models are robust to different machine learning models, including stacking, boosting, and bagging ensembling models. Our models consider cross-sectional heterogeneity, such as different ownership structures, accessibility to external finance, industry heterogeneity, different sample periods, and government policy impact