Business Cycle and Realized Losses in the Consumer Credit Industry

Abstract

We study the determinants of the loss given default (LGD) of consumer credit. Exploiting a dataset including more than 6 million of Italian consumer loans from 2007 to 2019, we find that macroeconomic and social variables significantly enhance forecasting performance both at the individual and portfolio levels, by up to 10 percentage points in terms of R². This result is robust across forecasting exercises and model specifications. In particular, non-linear forecast combination schemes relying on neural networks are among the best performers in terms of mean absolute error, RMSE, R², and model confidence set in every considered exercise. The relationship between the expected LGD and the macro predictors unveiled by accumulated local effects plots confirms the intuition that lower real activity, increasing cost-of-debt to GDP ratio, and greater economic uncertainty are associated with a greater LGD for consumer credit.

Keywords: Credit Risk, Consumer Credit, Loss Given Default, Non-Performing Loans