Account-level analytic hierarchical mixing modeling for credit risk of Chinese Government financing vehicle portfolios

Traditional credit risk measurement models, requiring fair amounts of default debts, have trouble in measuring the true default probability of Government financing vehicle (GFVs) loans in China. In this study, an analytic hierarchical mixing model (AHMM) was proposed to estimate the real states of Chinese GFVs with little default observations. AHMM outputs abstract risk indices for each loan based on the account-level GFV data, mapped to the probability of default by a calibration curve for loan decision, dynamic risk control, and stress testing. Furthermore, we also applied municipal bond data from the U.S. to AHMM and found that the accuracy ratio was 0.89 for the U.S. data and 0.84 for the Chinese data. The fitting error range based on U.S. data is [− 0.07 0.117], which is significantly lower than the Chinese GFV data, [− 0.10 0.2115]. Thus, although AHMM could be a credit risk model with few default observations, it works better on data with more default observations. The methodology in this study can be used on aggregate data to evaluate the entire Chinese GFV portfolio and thus bring a clear sovereign solvency picture to regulators and investors.