the web's biggest credit risk modeling resource.

Credit Jobs

Home Glossary Links FAQ / About Site Guide Search

Submit Your Paper

In Rememberance: World Trade Center (WTC)

doi> search: A or B

Export citation to:
- Text (plain)
- BibTeX

Fledelius, Peter, David Lando, and Jens Perch Nielson, "Non-Parametric Analysis of Rating Transition and Default Data", Journal of Investment Management, Vol. 2, No. 2, (Q2 2004), pp. 71-85.

Abstract: The key purpose of rating systems is to provide a simple classification of default risk of bond issuers, counterparties, borrowers etc. A desirable feature of a rating system is of course that it is successful in ordering firms so that default rates are higher for lower rated firms. However, this ordering of credit risk is not sufficient for the role which ratings are bound to play in the future. A rating system will be put to use for risk management purposes and the transition probabilities and default probabilities associated with different ratings will have concrete implications for internal capital allocation decisions and for solvency requirements put forth by regulators. The accuracy of these decisions and requirements depends critically on a solid understanding of the statistical properties of the rating systems employed.

In this paper we use non-parametric techniques to document the dependencies on duration and previous state. This exercise serves two key purposes: First, we show that these effects can be more clearly demonstrated in a non-parametric setting where the specific modeling assumptions are few. For example, we find a remarkable pattern in the classes analyzed in the sense that for all the classes analyzed, the effect of whether the previous move was a downgrade or an upgrade vanishes after about 30 months since the last move but that it is significant up to that point in time. We also consider stratification of firms in a particular rating class according to the way in which the current rating class was reached, reinforcing the results reached, for example, in Lando and Skødeberg (2002). Again, we are able to quantify how long the effect persists. To do this requires a notion of significance and we base this on calculation of pointwise confidence intervals by a bootstrap method which we explain in the paper.

This paper is republished as Ch.4 in...