liu.seSearch for publications in DiVA
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • oxford
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Predicting Bankruptcy Risk: A Gaussian Process Classifciation Model
Linköping University, Department of Computer and Information Science.
2015 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
Abstract [en]

This thesis develops a Gaussian processes model for bankruptcy risk classification and prediction in a Bayesian framework. Gaussian processes and linear logistic models are discriminative methods used for classification and prediction purposes. The Gaussian processes model is a much more flexible model than the linear logistic model with smoothness encoded in the kernel with the potential to improve the modeling of the highly nonlinear relationships between accounting ratios and bankruptcy risk.

We compare the linear logistic regression with the Gaussian process classification model in the context of bankruptcy prediction. The posterior distributions of the GPs are non-Gaussian, and we investigate the effectiveness of the Laplace approximation and the expectation propagation approximation across several different kernels for the Gaussian process. The approximate methods are compared to the gold standard of Markov Chain Monte Carlo (MCMC) sampling from the posterior.

The dataset is an unbalanced panel consisting of 21846 yearly observations for about 2000 corporate firms in Sweden recorded between 1991−2008. We used 5000 observations to train the models and the rest for evaluating the predictions. We find that the choice of covariance kernel affects the GP model’s performance and we find support for the squared exponential covariance function (SEXP) as an optimal kernel.

The empirical evidence suggests that a multivariate Gaussian processes classifier with squared exponential kernel can effectively improve bankruptcy risk prediction with high accuracy (90.19 percent) compared to the linear logistic model (83.25 percent).

Place, publisher, year, edition, pages
2015. , 64 p.
National Category
Computer and Information Science
Identifiers
URN: urn:nbn:se:liu:diva-119120ISRN: LIU-IDA/STAT-A--15/001—SEOAI: oai:DiVA.org:liu-119120DiVA: diva2:824142
Subject / course
Statistics
Presentation
2015-06-03, Von Neumann, IDA, Linköping, 10:45 (English)
Supervisors
Examiners
Available from: 2015-06-23 Created: 2015-06-09 Last updated: 2015-06-23Bibliographically approved

Open Access in DiVA

fulltext(2317 kB)341 downloads
File information
File name FULLTEXT01.pdfFile size 2317 kBChecksum SHA-512
6ead3015782cc6b0d999025e2315795cb59dbf5b4af32dee178747bd6c50c7991ba20b4247ddcad0d63402a91e689717f9a84681f5bd0d2a1effda9d9722df8c
Type fulltextMimetype application/pdf

Search in DiVA

By author/editor
Seidu, Mohammed Nazib
By organisation
Department of Computer and Information Science
Computer and Information Science

Search outside of DiVA

GoogleGoogle Scholar
Total: 341 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 554 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • oxford
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf