The University of Southampton
University of Southampton Institutional Repository

Applications of machine learning in consumer credit risk modelling

Applications of machine learning in consumer credit risk modelling
Applications of machine learning in consumer credit risk modelling
This thesis investigates three separate type of prediction problems, in differing contexts, with a common theme of experimental comparison of standard methods with more advanced machine learning methods. The objective is evaluation of the predictive power of machine learning methods through experiments on real world data. The first paper is an application of machine learning classification methods to predict mortgage arrears. It finds that both machine learning and a flexible statistical model outperform standard approaches. This can help identification of important predictive factors for the management of loan arrears within banks and loan servicers. The second paper applies both regression and classification methods to prediction of Peer to Peer (P2P) loan returns and default using different types of information.The main findings are that linear methods perform well on several (but not all) criteria; whether machine learning ensemble methods perform better than individual methods depends on the performance measure used to assess them. Use of alternative text-based information does not improve predictive outcomes. As a consequence, investors can be more informed about investments in this market. The third uses survival analysis to predict time to sale of property collateral used for mortgage loans. When property sales occur, as separate set of statistical and machine-learning models are used to predict the haircut or discount between the indexed property valuation at the point of sale and the actual transaction price. Random survival forests worked well to predict the time to sale; while deep learning, random forests, and neural network regression methods performed best predicting the discount. Based on predictive models for these two parameters, a sensitivity analysis illustrated how predictive modelling of these parameters produces more conservative (i.e., higher) loss estimates than one current industry approach.
University of Southampton
Fitzpatrick, Trevor
b3d78774-8c4d-4f7d-875c-8483843da9ef
Fitzpatrick, Trevor
b3d78774-8c4d-4f7d-875c-8483843da9ef
Mues, Christophe
07438e46-bad6-48ba-8f56-f945bc2ff934

Fitzpatrick, Trevor (2020) Applications of machine learning in consumer credit risk modelling. University of Southampton, Doctoral Thesis, 134pp.

Record type: Thesis (Doctoral)

Abstract

This thesis investigates three separate type of prediction problems, in differing contexts, with a common theme of experimental comparison of standard methods with more advanced machine learning methods. The objective is evaluation of the predictive power of machine learning methods through experiments on real world data. The first paper is an application of machine learning classification methods to predict mortgage arrears. It finds that both machine learning and a flexible statistical model outperform standard approaches. This can help identification of important predictive factors for the management of loan arrears within banks and loan servicers. The second paper applies both regression and classification methods to prediction of Peer to Peer (P2P) loan returns and default using different types of information.The main findings are that linear methods perform well on several (but not all) criteria; whether machine learning ensemble methods perform better than individual methods depends on the performance measure used to assess them. Use of alternative text-based information does not improve predictive outcomes. As a consequence, investors can be more informed about investments in this market. The third uses survival analysis to predict time to sale of property collateral used for mortgage loans. When property sales occur, as separate set of statistical and machine-learning models are used to predict the haircut or discount between the indexed property valuation at the point of sale and the actual transaction price. Random survival forests worked well to predict the time to sale; while deep learning, random forests, and neural network regression methods performed best predicting the discount. Based on predictive models for these two parameters, a sensitivity analysis illustrated how predictive modelling of these parameters produces more conservative (i.e., higher) loss estimates than one current industry approach.

Text
thesis_final_electronic - Version of Record
Available under License University of Southampton Thesis Licence.
Download (1MB)
Text
Final_Permission to deposit thesis form_both_signed.pdf
Restricted to Repository staff only
Available under License University of Southampton Thesis Licence.

More information

Published date: June 2020

Identifiers

Local EPrints ID: 447638
URI: http://eprints.soton.ac.uk/id/eprint/447638
PURE UUID: ce1663ce-413e-4901-96b8-11b7374a541b
ORCID for Christophe Mues: ORCID iD orcid.org/0000-0002-6289-5490

Catalogue record

Date deposited: 17 Mar 2021 17:33
Last modified: 01 Jun 2022 04:01

Export record

Contributors

Author: Trevor Fitzpatrick
Thesis advisor: Christophe Mues ORCID iD

Download statistics

Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.

View more statistics

Atom RSS 1.0 RSS 2.0

Contact ePrints Soton: eprints@soton.ac.uk

ePrints Soton supports OAI 2.0 with a base URL of http://eprints.soton.ac.uk/cgi/oai2

This repository has been built using EPrints software, developed at the University of Southampton, but available to everyone to use.

We use cookies to ensure that we give you the best experience on our website. If you continue without changing your settings, we will assume that you are happy to receive cookies on the University of Southampton website.

×