DefaultRisk.com the web's biggest credit risk modeling resource. |  |
|
|
| Evidence on the Incompleteness of Merton-type Structural Models for Default Prediction by Roger M. Stein of Moody's|KMV February 9, 2005 Abstract: One may argue that structural models of default incorporating equity market information, such as the Merton (1974) model, are complete in the sense that univariate (single-factor) models are thought to be sufficient to capture all significant aspects of the future prospects for a firm. This assertion is testable empirically. In this short paper we provide some evidence that unmodified, Merton-type models are not, in fact, complete in the sense that additional information provides better discrimination between defaulters and non-defaulters even when conditioned on Merton-based variables. Using Moody's extensive database of corporate defaults, we first show heuristically that partitioning a standard Merton model by a second variable provides more information about default. We then show that econometric tests of significance refute the assertion that additional information does not help explain default. Finally, we show that even a simple regression-based multi-factor model appears to outperform its single-factor (basic Merton-only) counterpart in rigorous (out-of-sample and out-of-time) validation. This suggests merit to exploring enhancements to the Merton framework such as, for example, those introduced by the Vasicek-Kealhofer model. JEL Classification: C52, G20, G33. Books Referenced in this paper: (what is this?)
|
