Shepeleva Marina

 The Faculty of Computer Information Technologies and Automatics

 The speciality of Automated Systems of Control (ASC)

group ASC-01-B

    Theme of master's work: "Development of expert     system of assessment credit risks of crediting natural person"

        supervisor: Ph.D. of Computer Science Zhukova Tamara Porfirievna

    m_shepeleva@mail.ru                                                   

The author's abstract

 Crediting is the most profitable and most risky part of bank operations in this time. Unredeemed credits, especially large, can lead the bank to bankruptcy, and depends of it position in economy, to a lot of bankruptcies connected with it enterprises, banks and private persons. That is why management of credit risk is a necessary part of strategy and tactics of a survival and development of any bank. Crediting always was and remains priority economic function of banks.

Insufficient level of development of theoretical and methodological questions of the risk’s analysis of credit operations in system of the analysis of bank activity cause a choice of master’s work theme  and show its urgency.

The purpose of the given research is development of expert system of economically proved model of an estimation and regulation of credit risk with the purpose of satisfaction of interests of the bank connected with minimization of risk of a credit portfolio of bank and increase of its quality.

As object of research the theoretical and methodological toolkit of an estimation and regulation of risk of a credit portfolio of bank acts. In developed system will be used scoring model including two methods: neural networks and classification trees of decisions. In a combination of these two methods in scoring models scientific novelty also will be shown. Depending on type of used entrance data about the potential borrower allocate following types of scoring:

- Credit scoring (scoring according to the application, application scoring) - decision-making on delivery of the credit to new clients according to, specified in the application.

- Behavioral scoring (behavior scoring) - a dynamic estimation of a condition of credit status of the existing borrower, based on data about history of transaction under its accounts. By results of an estimation the current limit of the credit for the borrower can be defined; the measures accepted in case of a delay of payments; Marketing courses which can be directed on the client. The system of scoring can be used not only at a stage of sale of a credit product, but also at its designing as with its help it is possible to define to analyze credit status of group of potential borrowers, for which group the product is projected, and, having allocated the basic qualities of borrowers promoting decrease of risk to direct the basic marketing efforts to such borrowers. Now for credit scoring methods of statistics are used: the discriminated analysis, linear regress, logistical regress, trees of classification; researches of operations: linear programming, nonlinear optimization and an artificial intellect: neural networks, expert systems, genetic algorithms, methods of the nearest neighbors, Bayesian networks, logical probabilistic methods. The specified methods can be applied as separately, and in various combinations.

It is the most rational to use scoring model including two methods: neural networks and classification trees of decisions. By means of neural networks the analysis of credit history of the last years will be spent and on the basis of the received results recommendations and preferences will stand out at delivery of credit products. Than with help of classification method of trees of decisions on the basis of entering parameters of system, namely the questionnaires filled by the borrower the classification model which on an output will carry the borrower to the certain class will be under construction, in conformity to which will make the decision on delivery of the credit. Before bank following problems concerning credit policy cost: to minimize manual processing operations by employees of bank - costs on the personnel and minimization of operational risks; to minimize manual registration by the client of documents - time costs of service of the client and creation for the client of comfort at service; to minimize losses on credit risks; to receive expert system of an automatic estimation of credit risks on the basis of scientific and flexible methods; For the decision of all problems it is necessary to choose such model of an estimation which is optimum, it is adaptive changed at any macroeconomic conditions, considered features business of processes, would estimate automatically credit applications and would consider an economic situation in the local market. The model according to all these parameters is s scoring, including various methods of an estimation of credit risks. The optimality of the given model will depend on many factors. For an estimation of credit status of physical persons the most productive and optimum is application scoring models in a combination of two methods - neural networks and trees of decisions.

Thus, by consideration of various methods of an estimation of credit risks and credit status of the borrower and their mathematical models, it has been found out, that one of the best and perspective methods of an estimation of risks and credit status is scoring method.

In Ukraine introduction of scoring should be carried out gradually. For the beginning it is possible to make the automated expert system of a tentative estimation of borrowers which will automatically eliminate obviously "bad" risks, and on consideration of credit committee to offer risks "good" and "boundary". As an initial material for scoring the various information on the last clients on the basis of which classifications by means of various methods the forecast about credit status of the future borrowers is done is used. As mathematical model for scoring method the best, at an estimation of credit status of physical persons are neural networks and trees of decisions. The given methods will most quickly and precisely make an estimation of factors and will give out result - distribution of borrowers on categories of risk. So, on the basis of the considered materials in the given work the system using methods of trees of decisions and neural networks will be developed expert scoring. The given system will allow bank workers to make quickly of the decision on crediting, to adjust volumes of crediting depending on a situation in the market and to define an optimum parity between profitableness of credit operations and a risk level.