Modeling The Probability Of Loan Default Via Logistic Regression And Survival Analysis Techniques

Author

Rebecca Tafadzwa Mbengi

Title

Modeling The Probability Of Loan Default Via Logistic Regression And Survival Analysis Techniques

Abstract

Prediction of loan defaults is an essential component of credit risk assessment, which informs the decision-making processes of financial institutions. This empirical research project aimed to develop a predictive model for estimating the default risk of a loan portfolio by analyzing historical loan data and borrower characteristics. The researcher utilized logistic regression and survival analysis methods to analyze a vast dataset of loan portfolios obtained from KCI Management Consultants. The study's results demonstrated that both logistic regression and survival analysis strategies have relatively similar performance based on the Receiver Operating Characteristic assessment. However, the survival model outperformed the logistic regression method in accurately predicting defaulted and non-defaulted loan portfolios. This suggests that survival analysis strategy offers a viable alternative to the conventional logistic regression method used in assessing credit risk in the MFI sector. The research project also revealed that survival analysis provides several advantages for credit risk management and capital management. By modeling the time to default, survival analysis can help identify key risk indicators before they impact credit markets and provide insights into the relationship between default and borrower characteristics. The study confirmed the importance of utilizing empirical approaches to credit risk management and showcased the advantages of employing survival analysis over logistic regression as a predictive model for loan default risk. These insights provide valuable information for financial institutions looking for a more accurate and effective way of addressing credit risk management.

Date

June 2024

Publisher

BUSE

Keywords

Modeling

Probability

Loan Default

Logistic Regression

Supervisor

Ms J. Pagan'a