Daria D. Frolova Loss Given Default Modelling for a Bank’s Retail Portfolio of Unsecured Loans

DOI: 10.38050/2078-3809-2025-17-4-136-150

Abstract

Proper modelling of loss given default (LGD) is an important task for commercial banks. The relevance of the study is due to the active expansion of retail lending over the past 5 years. The increase in the volume of the retail portfolio from 17.6 to 36.9 trillion rubles from the 1st of January until the 1st of January 2025 contributed to an increase in the credit risk of Russian commercial banks. Additional difficulties arise due to the high level of the Bank of Russia's key interest rate and the tightening of macroprudential policy in the segment of unsecured retail loans. The paper describes the main approaches to LGD modelling. Based on data on unsecured consumer loans from one of the Russian banks for the period from December 2020 to March 2025, which contain more than 700,000 observations, a two-stage LGD assessment methodology has been proposed. The approach includes vintage analysis and the use of machine learning models (XGBoost and LightGBM) to assess and predict LGD considering new macroeconomic factors: the USD/RUB exchange rate on the Forex market and the average monthly salary of Russian employees.

Keywords: Loss Given Default, retail portfolio, commercial banks, unsecured retail loans, modelling, vintage analysis, machine learning.

JEL: C53, G17, G21.

For citation: Frolova, D.D. (2025) Loss Given Default Modelling for a Bank’s Retail Portfolio of Unsecured Loans. Scientific Research of Faculty of Economics. Electronic Journal, vol. 17, no. 4, pp. 136-150. DOI: 10.38050/2078-3809-2025-17-4-136-150.

Download