On clustering for credit risk management : applications to loan selection and default prediction
(2026)
Files
PhD_Thesis.pdf
Open Access - Adobe PDF
- 3.23 MB
Details
- Authors
- Supervisors
- Abstract
- (en) This thesis investigates how clustering can improve credit risk management, both in optimization and prediction tasks. The first chapter studies the selection of loans to be included in collateralized loan obligations and proposes a clustering-based approximation to reduce the dimensionality of the decision space and mitigate the computational burden of the mixed-integer nonlinear optimization problem. The second chapter introduces clagging, a new ensemble learning strategy that fits predictive models across clusters of the training set and aggregates their forecasts using distances to cluster centroids. The third chapter applies this methodology to the design of an Early Warning System for corporate clients of a systemic European bank. In this setting, clustering is used to address client heterogeneity, while time heterogeneity is handled by rescaling model outputs with macroeconomic forecasts through Bayes’ theorem. Overall, the thesis shows that clustering can serve as a practical and effective tool to enhance loan selection, predictive modeling, and default risk monitoring in finance.
- Affiliations
Citations
Germain, A. (2026). On clustering for credit risk management : applications to loan selection and default prediction. https://hdl.handle.net/2078.5/278799