Estimation of the Loss Given Default of Corporate Borrowers

Lending to corporate borrowers occupies most of the loan portfolio of banking organizations. That is why the reduction of credit risks in the event of a borrower in default is one of the priority areas for the development of analytical models in the banking sector. The three most significant components of credit risk assessment are components: PD (Probability of Default), LGD (Loss Given Default), EAD (Exposure-At-Default). Given the current economic crisis associated with the coronovirus infection pandemic, calculating the level of default losses is an important topic, as it helps to identify negative trends in the company's financial activities when issuing loans at the stage of financial monitoring of borrowers. These models will allow the bank to quickly take measures to resolve potentially bad debt of customers. The main methods and existing models for assessing the level of losses during default were examined in the course of this research, their advantages and disadvantages were analyzed, new methods for assessing this indicator were proposed, two models were built: estimates of the likelihood of solvency scenarios of borrowers after default (bankruptcy or recovery) and assessment default loss level using current statistics. The resulting models are tested for compliance with the economic sense. As statistical approaches to the development of such models, classification trees and regression trees, respectively, were used. The approaches used in the study for the indicated purposes carry scientific novelty, in particular, in the Russian and foreign literature, similar practical studies and results on Russian corporate borrowers are absent. Additionally, random forest and logistic regression models were developed and it was justified why they did not give a better result than the one obtained when building the final models. The results can be applied in the banking sector to improve existing models for assessing the creditworthiness of a client and making decisions on his lending. In total, the work presents 41 pages, taking into account the title and applications. 17 images were used for tables and graphs, and for some research results. 16 sources have been used.