Misspecified Copulas in Credit Risk Models: How Good is Gaussian?

Posted: 8 Nov 2005

See all articles by Alfred Hamerle

Alfred Hamerle

University of Regensburg - Faculty of Business, Economics & Information Systems

Daniel Roesch

University of Regensburg

Multiple version iconThere are 2 versions of this paper

Abstract

In addition to "classical" approaches, such as the Gaussian CreditMetrics or Basel II model, the use of other copulas has recently been proposed in the area of credit risk for modeling loss distributions, particularly T copulas which lead to fatter tails ceteris paribus. As an amendment to recent research this paper shows some estimation results when the copula in a default-mode framework using a latent variable distribution is misspecified. It turns out that parameter estimates may be biased, but that the resulting forecast for the loss distribution may still be adequate. We also compare the performance of the true and misspecified models with respect to estimation risk. Finally, we demonstrate the ideas using rating agencies data and show a simple way of dealing with estimation risk in practice. Overall, our findings on the robustness of the Gaussian copula considerably reduce model risk in practical applications.

Keywords: Gaussian CreditMetrics, Basel II model, copulas, credit risk, loss distributions, fatter tails ceteris paribus, latent variable distribution

Suggested Citation

Hamerle, Alfred and Roesch, Daniel, Misspecified Copulas in Credit Risk Models: How Good is Gaussian?. Journal of Risk, Vol. 8, No. 1, Fall 2005, Available at SSRN: https://ssrn.com/abstract=839765

Alfred Hamerle (Contact Author)

University of Regensburg - Faculty of Business, Economics & Information Systems ( email )

Universitstrasse 31
Regensberg D-93053
Germany

Daniel Roesch

University of Regensburg ( email )

Chair of Statistics and Risk Management
Faculty of Business, Economics and BIS
Regensburg, 93040
Germany

HOME PAGE: http://www-risk.ur.de/

Here is the Coronavirus
related research on SSRN

Paper statistics

Abstract Views
1,507
PlumX Metrics