A Comparison of Two Methods in Credit Scoring

The credit scoring methodology is increasingly used in companies and business field. This method once in the 1950 s introduced has developed and cover new branches today, such as mailing and fraud. This methodology is one of the most important to manage risks and estimate the probability of default of a client by borrowing money. There are several statistical methods to select the relevant characteristics for the scoring model such as linear regression models and classification tree and the goal in this study is to compare the two most used statistical methods on credit scoring: logistic regression - with categorized and raw data - and discriminant analysis, as also their pros and cons. The analysis was made using a dataset from METRO Cash&Carry enabling the application of the methods on real data. The logistic regression with a Scorecard example and the discriminant analysis have a similar performance on estimating the probability of default from a client, despite the methodologies being different the approaches are similar and the results shown a small difference between them.

Autorentext

Daniela Rodrigues Recchia, M.Sc. in Statistics: Master of Science in Statistics at the Technische Universität Dortmund, Germany. Bachelor in Statistics at the State University of Campinas, Brazil. Professional experience as Statistician and Risk Analyst.