Understanding Credit Risk for Chinese Companies using Machine Learning: A Default-Based Approach
76 Pages Posted: 23 Nov 2020 Last revised: 12 Mar 2021
Date Written: March 12, 2021
In response to the recent elevated corporate credit risk environment in China’s credit market, we develop a probability of default (PD) measure for Chinese companies using actual corporate bond defaults by applying the Least Absolute Shrinkage and Selection Operator (LASSO) machine learning model. Our PD measure is applicable to publicly listed and also, importantly, to unlisted companies. Our measure’s bond default prediction accuracy outperforms models generated by alternative machine learning techniques and other prominent credit risk measures. Further analysis documents a large pricing effect of corporate default risk using our PD measure in primary and secondary bond markets. The pricing effect of default risk became more pronounced following two crucial market events in 2014 that raised market awareness of credit risk and is stronger for bonds likely traded by retail and foreign investors. In the cross section of bond and stock returns, we observe a positive distress risk premium after controlling for common risk factors. Finally, stocks of low PD firms outperformed those of high PD firms during the COVID-19 pandemic.
Keywords: Default risk; Chinese bond market; Z-score; Distress risk premium
JEL Classification: G12; G15; G23; G24
Suggested Citation: Suggested Citation