In accordance with the International Financial Reporting Standard credit institutions are required to
assess and form a reserve for expected credit loss on loans issued. The expected credit loss is estimated based on probability of default of the borrower (PD), loss given default of the borrower and exposure at default. This article discusses the quantitative assessment of loss given default by a bank or financial institution by the case of a portfolio of car loan agreements for individuals using classical regression, estimated by OLS, and regression based on neural network architecture. Based on the results of the evaluated models, a cluster analysis of borrowers was carried out taking loss given default and probability of default as input.