Использование инструментария Data Mining в управлении кредитными рисками

Abstract

Обґрунтовано необхідність побудови моделі з бінарною залежною змінною для оцінювання і прогнозування кредитоспроможності фізичних осіб – потенційних позичальників банку – з метою зниження рівня кредитного ризику. Виконано оцінювання параметрів logit-моделі методом максимальної правдоподібності з використанням пакета Statistica. Запропоновано процедури оцінювання якості моделі. Ключові слова: кредитний ризик, Data-Mining, кредитоспроможність, логiт-модель.

Обоснована необходимость построения модели с бинарной зависимой переменной для оценивания и прогнозирования кредитоспособности физических лиц – потенциальных заемщиков банка – с целью снижения уровня кредитного риска. Выполнено оценивание параметров logit-модели методом максимального правдоподобия с использованием пакета Statistica. Предложены процедуры оценивания качества модели. Ключевые слова: кредитный риск, Data-Mining, кредитоспособность, логит-модель.

Credit activity determines the effectiveness of the functioning of the bank, as a significant part of the bank income comes from lending operations. This lending is always associated with risk. NPLs could lead to the bankruptcy of the bank and this may lead to the bankruptcy of its related companies. Therefore, the problem of effective management of credit risk is a necessary part of the strategy and tactics of survival and growth for every commercial bank. The purpose of this work is to show the usage of advanced mathematical methods and IT-technologies as to assess the creditworthiness of individuals - potential borrowers. The article proves the necessity of building a model with a binary dependent variable to estimate and predict creditworthiness of potential borrowers in order to reduce the level of credit risk. The research was performed in accordance with the materials of the retail lending of a bank and the logistic model of creditworthiness diagnostics of a potential clientwas built on this basis. In this model the dependent variable is a binary variable reflecting the status of the client. The dependent variable will be zero if the loan is problematic, and otherwise will be equal to 1. The value that ranges from 0 to 1 would indicate the probability of loan default or other problems concerning the recovery of a debt. The parameter estimation was made with the help of logit-models that uses maximum likelihood method. In this research theStatistica software was used – the package for data analysis, data management, statistics, data mining, and data visualization procedures. The procedures of estimating the quality of the model were also proposed. With the help of the model it is possible to determine the percentage of trustworthy borrowers and the percentage of unscrupulous borrowers. Keywords: credit risk, data mining, creditworthiness, logit.