Anticipating consumer credit recovery failure: classification approaches
This study offers an advanced method of assessing credit for non-performing consumer loans, which may represent a new investment opportunity in the post-pandemic era. Our results, based both on a single account-level dataset and machine learning techniques, imply that the artificial neural network algorithm with demographic and account-related variables performs best in terms of prediction of consumer credit collection failure within 24 months. We also find that the main determinants of these failures are the total amount of overdue debt, the age of the applicant and the maximum length of the delay period. A forecasting model using the random forest algorithm can also be improved by using additional information that is determined after a debtor requests the credit recovery program. Our results have practical implications for banks, financial institutions, and investors who need to manage and value NPLs.