Statistical and Optimization Methods in Credit Scoring

Abstract

Introduction. The role of credit scoring in the work of financial institutions is difficult to overestimate. Accurate and efficient scorecards allow lenders to assess risks correctly and monitor their investments. Such cards should be based on reliable statistical data about previous and current customers using statistical analysis methods. Over the years of its development, the toolkit of credit scoring has also been supplemented with non-statistical methods based on the use of optimization procedures, decision trees, intelligent databases and knowledge bases, building network models, etc. Given the wide range of available methods, there is a need for their classification and application analysis. The purpose of the article is to provide a brief description of all relevant statistical and non-statistical methods that allow solving credit scoring tasks in modern formulations. To reveal the features of using the methods described and conduct their comparison. Results. Statistical methods allow to investigate the significance of all the factors included in the model, as well as to obtain a set of statistical estimates that help to assess the quality of the model. Thus, these methods allow to build an optimal and reliable model. Non-statistical methods allow you to add arbitrary restrictions to the model, automatically detect and process interactions between characteristics, and solve problems with a large number of applicants and their characteristics, which is facilitated due to the development of computational methods. Conclusions. Modern mathematical methods allow to solve credit scoring tasks effectively, among which one of the main ones is the binary and multigroup classification. The choice of the optimal method depends on the type of system (static or dynamic), the creditor's computing capabilities and the importance of the results interpretation. Keywords: credit scoring, statistical methods, mathematical programming, neural networks, genetic algorithms.