Evaluating the Predictive Power of Debt Register Data in Consumer Loan Defaults: A Pre- and Post-Implementation Machine Learning Approach

Abstract

The introduction of the Norwegian Debt Register in 2019 marked a significant milestone in the landscape of consumer lending in Norway. This centralized database provides lenders with real- time, comprehensive insights into borrowers’ unsecured debt obligations, potentially enhancing credit risk assessment practices. This thesis investigates the impact of incorporating debt register data on the accuracy of machine learning credit risk models in predicting loan defaults.

Utilizing a dataset from Resurs Bank spanning from 2015 to 2022, we develop and compare various predictive models, including logistic regression, decision trees, random forests, XGBoost, support vector machines, and neural networks. We evaluate model performance using metrics such as accuracy, sensitivity, specificity, and area under the ROC curve (AUC), while employing techniques like SMOTE to address class imbalance.

Our findings suggest that the inclusion of debt register data provides a marginal improvement in the models’ ability to predict defaults, particularly in terms of sensitivity. However, the overall impact is less substantial than anticipated, likely due to the comprehensiveness of existing application data and the inherent challenges in predicting defaults on accepted loans. Nevertheless, our analysis of monthly default rates and application acceptance rates reveals notable trends following the debt register’s implementation, suggesting more stringent lending practices and improved risk assessment processes.

This study contributes to the understanding of how debt register data and machine learning can enhance credit risk assessment in the Norwegian consumer loan market. While the findings high- light the potential benefits and limitations of incorporating debt register data, they also underscore the need for further research to explore additional data sources, investigate long-term impacts, and assess generalizability across different financial institutions and markets.