Publication detail

Bayesian default probability models

Author(s): Petra Andrlíková MSc.,
Type: IES Working Papers
Year: 2014
Number: 14
Published in: IES Working Papers 14/2014
Publishing place: Prague
Keywords: default probability, bayesian analysis, logistic regression, goodness-of-fit
JEL codes: C11, C51, C52, G10
Suggested Citation: Andrlíková P. (2014). “Bayesian default probability models” IES Working Paper 14/2014. IES FSV. Charles University.
Abstract: This paper proposes a methodology for default probability estimation for low default portfolios, where the statistical inference may become troublesome. The author suggests using logistic regression models with the Bayesian estimation of parameters. The piecewise logistic regression model and Box-Cox transformation of credit risk score is used to derive the estimates of probability of default, which extends the work by Neagu et al. (2009). The paper shows that the Bayesian models are more accurate in statistical terms, which is evaluated based on Hosmer-Lemeshow goodness of fit test, Hosmer et al. (2013).
Downloadable: WP_2014_14_Andrlikova


McKinsey & Company
Moneta Money Bank