Loss Given Default Distributions in Different Countries: The Modality Defines the Estimation Method

Abstract

Estimating the credit risk parameter loss given default (LGD) is important for banks from an internal risk management and a regulatory perspective. Several estimation approaches are common in the literature and in practice. However, it remains unclear which approach leads to the highest estimation accuracy. In this regard, existing comparative studies in the literature come to different conclusions. The differences can be attributed to the specific choice of loan portfolio and, thus, to the specific choice of the LGD distribution. Against this background, we examine the estimation accuracy of various LGD estimation methods, including traditional regression and advanced machine learning. Our analysis is based on international loan portfolios of 16 European countries, with a total of 26, 227 defaulted loans of small and medium enterprises. Using a cluster analysis, we assign country-specific loan portfolios to three relevant modality types of LGD distributions. For each of these three types, we empirically determine the estimation method with the highest estimation accuracy.

Keywords: Risk Management, Forecasting, Loss Given Default, Country-specific LGD distributions, Machine Learning, Global Credit Data

JEL Classification: C45, C46, G21, G28