Technical University of Crete (TUC)
Accessing bankruptcy risk for financial institutions: methodological framework and predictive modelling
Abstract
dc:descriptionThis thesis is a comprehensive and complete research on bank failure prediction, as it examines various modeling aspects for obtaining improved results. The analysis is based on a comprehensive dataset of approximately 60,000 observations over an extensive period of nine years (2005-2014), and it examines different prediction horizons, for up to three years prior to failure. We explore whether the addition of variables related to the diversification of the banks’ activities, along with local effects, improves the predictability of the models. Seven popular and widely used machine-learning techniques are compared (logistic regression, support vector machines with linear and radial kernels, naïve Bayes, extreme gradient boosting, random forests and artificial neural networks) and three different classification performance metrics are calculated (AUROC, H-measure, and Kolmogorov-Smirnov metric). In order to ensure the robustness of the results, bootstrap testing is used. The results show that mid- and long-range predictions improve significantly with the addition of diversification variables. Local effects exist and further improve the results while support vector machines along with gradient boosting and random forests outperform the traditional models with the differences increasing over longer prediction horizons.
Degree
thesis:*- Grantor dc:publisher
- Technical University of Crete (TUC)
- Year dc:date
- 2019
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Manthoulis, Georgios
- Μανθούλης, Γεώργιος
Subjects
dc:subject × 10Rights
dc:rights- Statement dc:rights
-
- BY
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/46616