COMPARATIVE ANALYSIS OF RISK IDENTIFICATION MODELS 

Moiseev Aleksandr Vladimirovich, Candidate of physical and mathematical sciences, head of sub-department of applied mathematics and operations research in economics, Penza State Technological University (1a Baydukova passage, Penza, Russia), amoiseev@pgta.ru
Popravko Evgeniy Aleksandrovich, Postgraduate student, Penza State University (40 Krasnaya street, Penza, Russia), popravko@prideautomatics.com
Fedotov Nikolay Gavrilovich, Doctor of engineering sciences, professor, head of sub-department of economic cybernetics, Penza State University (40 Krasnaya street, Penza, Russia), fedotov@pgu.ru 

004.67,004.62 

Background. The interest to models of risk identification relates to the desire to automate management decision making inconditions of risks when the account of the averaged end result is insufficient. At the present time there are several approaches to risk assessment. In every particular case it requires comparison of models by quality of identification. The study is aimed at consideration of model comparison procedure. In the work the comparison of various models is carried out for concrete results obtained on the basis of real statistics of a bank using the algorithm suggested by the authors.
Materials and methods. Model building was carried out using SPSS and Statistica applied programs. The article considers models of discriminant analysis, logit model and probit model. To characterize the quality of identification the authors determine the values of first and second type errors probability. To compare the models the ROC-curve is involved. To form the final conclusion about the built model quality the researchers apply examining sample.
Results. Combined analysis of the built models on training and examining samples showed high efficiency of the model of discriminant analysis for placing potential borrowers into either of two groups. The given model is characterized by high level of client’s creditworthiness forecasting, as well as by high quality of borrower’s default identification. The application of results of the discriminant analysis in the study algorithm allows forecasting insolvency of borrowers and may serve as a criterion for high risk group formation of bank’s certain clients. Conclusions. The considered procedure of identification model quality comparison enables to increase quality of information support for decision making inconditions of risks. 

Recognition System Risk, Credit risk, discriminant analysis, logit model, probit model. 

1. Ayvazyan S. A., Mkhitaryan B. C. Prikladnaya statistika. Osnovy ekonometriki: uchebnik dlya vuzov: v 2 t. [Applied statistics. Basic econometrics: textbook fro universities: in 2 volumes]. Moscow: YuNITI, 2001, 1008 p.
2. Bankovskie riski: ucheb. posobie [Bank risks: tutorial]. Ed. O. I. Lavrushin, N. I. Valentseva. Moscow: KNORUS, 2007, 232 p.
3. Fries C. Mathematical finance: theory, modeling, implementation. New Jersey: Wiley, 2007, 520 p.