Srovnání skóringových modelů při odhadu pravděpodobnosti úpadku bank v USA

Abstract

The aim of thesis is to estimate, verify and compare models for estimation the probability of default of U.S. commercial banks and then these models applied to the control sample to determine the most appropriate model.

The theoretical part deals with credit risk characteristics and description of the basic types of models for predicting default. Three types of credit scoring models are processed in more detail in this thesis: logit model, probit model and linear discriminant analysis. The practical part describes the application of presented theoretical-methodological starting points for the sample banks.

There are three credit-scoring models set up in total. Logit and probit models have proved to be very similar, LDA model has the lowest explanatory power of the estimated models (78.44 %). The estimated models have been applied to a control sample of banks. Logit model seems as the best model for predicting default in this control sample. This model has three parameters: the logarithm of total assets (LTA), return on average equity (ROAE) and ratio problem loans to gross loans (PL GL).