ANALYSIS OF STATISTICAL CREDIT RISK ESTIMATION MODELS EFFICIENCY

Abstract

Banks always seek to reduce potential loss due to crediting not reliable clients. So they must be able to to estimate credit risk of each client properly. One of possible instruments for credit risk measure is use of internal credit rating models in banks. Bank must determine significant attributes and select methods for credit risk estimation so that possibility of false decisions would be minimized. In most cases more than one model can be constructed and there is a problem which model to use in bank’s activity. In practice the best model must be used so for the comparison of different models certain rates of models efficiency must be calculated. The purpose of paper is to define the indicators of statistical credit risk estimation models efficiency and to analyze the efficiency of created credit risk estimation models. In this paper rates of credit risk estimation models were described: correct classification and misclassification rates, false negative and false positive rates, model sensivity and specificity, F-measure, ROC analysis. Analysis of scientific publications about credit risk estimation models has shown that the most efficient of the most commonly used methods are logistic regression and artificial neural networks. Also three artificial neural networks models (multilayer perceptrons) were constructed. The most efficient model analyses data about clients of 3 years. Models efficiency rates allowed to estimate risk of client misclassification and other characteristics. Also they help make a decision which model to use in practice.