Skip to main content
Login | Suomeksi | På svenska | In English

Credit risk scorecard estimation by logistic regression

Show full item record

Title: Credit risk scorecard estimation by logistic regression
Author(s): Peussa, Aleksandr
Contributor: University of Helsinki, Faculty of Science, Department of Mathematics and Statistics
Discipline: Statistics
Language: English
Acceptance year: 2016
The major concern of lenders is to answer the next question: 'Who we lend to?' Until 1970s the traditional schema was used to answer this question. Traditional credit assessment relied on 'gut feel', which means that a bank clerk or manager analyses a borrower's character, collateral and ability to repay. Also, some recommendations from the borrower's employer or previous lender are used. The alternative approach is credit scoring, which is a new way to approach a customer. Credit scoring is one of the most successful applications of statistics in finance and banking industry today. It lowers the cost and time of application processing and gives flexibility in making trade off between risk and sales for financial institution. Credit scorecards are essential instruments in credit scoring. They are based on the past performance of customers with characteristics similar to a new customer. So, the purpose of a credit scorecard is to predict risk, not to explain reasons behind it. The purpose of this work is to review credit scoring and its applications both theoretically and empirically, and to end up with the best combination of variables used for default risk forecasting. The first part of the thesis is focused on theoretical aspects of credit scoring - statistical method for scorecard estimation and measuring scorecard's performance. Firstly, I explain the definition of the scorecard and underlying terminology. Then I review the general approaches for scorecard estimation and demonstrate that logistic regression is the most appropriate approach. Next, I describe methods used for measuring the performance of the estimated scorecard and show that scoring systems would be ranked in the same order of discriminatory power regardless the measure used. The goal of the second part is empirical analysis, where I apply the theoretical background discussed in the first part of the master's thesis to a data set from a consumer credit bank, which includes variables obtained from the application forms and from credit bureau data, and extracted from social security numbers. The major finding of the thesis is that that the estimated statistical model is found to perform much better than a non-statistical model based on rational expectations and managers' experience. This means that banks and financial institutions should benefit from the introduction of the statistical approach employed in the thesis.

Files in this item

Files Size Format View
gradu_Aleksandr_Peussa.pdf 263.8Kb PDF

This item appears in the following Collection(s)

Show full item record