Assessing credit quality from the equity market: can a structural approach forecast credit ratings?

• DOI: http://doi.org/10.1002/cjas.27
• Author: Yu Du,Wulin Suo
• Date: 01 September 2007
• Published date: 01 September 2007

Can J Adm Sci

Copyright © 2007 ASAC. Published by John Wiley & Sons, Ltd. 212 24(3), 212–228

Assessing Credit Quality from the

Equity Market: Can a Structural

Approach Forecast Credit Ratings?

Yu Du

RBC Financial Group

Wulin Suo*

Queen’s University

Canadian Journal of Administrative Sciences

Revue canadienne des sciences de l’administration

24: 212–228 (2007)

Published online 29 August 2007 in Wiley Interscience (www.interscience.wiley.com). DOI: 10.1002/CJAS.27

We are grateful for the comments and suggestions from the Editor, Jason Wei, and the two anonymous referees, as well as Edward Altman, Darrel

Duff‌i e, Kim Huynh, Louis Gagnon, Lew Johnson, Frank Milne and Lynnette Purda. Financial support from SSHRC and Queen’s School of Business

are gratefully acknowledged.

*Please address correspondence to: Wulin Suo, School of Business, Queen’s University, 99 University Avenue, Kingston, Ontario, Canada K7L 3N6.

Email: wsuo@business.queensu.ca

Abstract

We investigate the empirical performance of default

probability prediction based on Merton’s (1974) struc-

tural credit risk model. More specif‌i cally, we study if

distance-to-default is a suff‌i cient statistic for the equity

market information concerning the credit quality of the

debt-issuing f‌i rm. We show that a simple reduced form

model outperforms the Merton (1974) model for both

in-sample f‌i tting and out-of-sample predictability for

credit ratings, and that both can be greatly improved by

including the f‌i rm’s equity value as an additional vari-

able. Moreover, the empirical performance of this hybrid

model is very similar to that of the simple reduced form

model. As a result, we conclude that distant-to-default

alone does not adequately capture the f‌i rm’s credit

quality information from the equity market. Copyright ©

2007 ASAC. Published by John Wiley & Sons, Ltd.

JEL Classif‌i cation: G13, G33

Keywords: Credit rating, default probability, distance-

to-default, structural credit risk model

Résumé

Dans cet article, nous évaluons la performance empirique

de la prédiction de probabilité de non-remboursement à

partir du modèle de risque de crédit structurel de Merton

(1974). Plus particulièrement, nous cherchons à exami-

ner dans quelle mesure la “distance par rapport à la

défaillance” (distance-to-default) est une statistique suf-

f‌i sante pour l’information du marché d’actions concer-

nant la qualité du crédit de la compagnie émettrice. Nous

démontrons qu’un modèle simple de formule réduite

donne de meilleurs résultats à la fois pour l’agencement

échantillonné et pour la prédictibilité hors-échantillon

que le modèle de Merton (1974), et que ces deux résultats

peuvent être encore plus précis si la valeur comptable

de la f‌i rme est prise en compte comme variable supplé-

mentaire. Par ailleurs, la performance empirique de ce

modèle hybride est très semblable à celle du modèle

simple de forme réduite. Aussi concluons-nous que la

distance par rapport à la défaillance ne rend pas f‌i dèle-

ment compte de l’information concernant la qualité du

crédit provenant du marché d’actions. Copyright © 2007

ASAC. Published by John Wiley & Sons, Ltd.

Mots-clés : Cote de crédit; probabilité de défaut;

distance par rapport à la défaillance; modèle de risque

de crédit structurel

ASSESSING CREDIT QUALITY FROM THE EQUITY MARKET DU & SUO

Can J Adm Sci

Copyright © 2007 ASAC. Published by John Wiley & Sons, Ltd. 213 24(3), 212–228

The proper assessment of credit risk has always been

important to f‌i nancial institutions. Recently, f‌i nancial

institutions have devoted even more resources to the

development of tools that will adequately assess and

manage the credit risk in their portfolios because, under

the new Basel accord (Basel II) issued by the Basel Com-

mittee on Banking Supervision, regulatory capital will be

determined using a bank’s internal assessments of the

probabilities that its counterparts will default. One

popular approach to assessing credit risk involves the

Merton (1974) model, where debts issued by a company

are treated as derivatives written on the f‌i rm’s underlying

assets. This approach allows one to establish structural

relationships among the f‌i rm’s debt, equity, and the asset

value. Using equity market data, one can generate a

quantity that determines the default probability of a debt-

issuing f‌i rm. These types of models, usually referred to

as structural models, can thus establish relationships

between default probabilities and some credit risk

measures that are analytic functions of various relevant

factors. Such models have become increasingly popular

in recent years. In this paper, we investigate whether the

credit risk measures inferred from structural credit risk

models are adequate to ref‌l ect the credit risk information

contained in those individual factors or, more generally,

whether structural models can provide a better alternative

to assessing f‌i rms’ credit quality than those traditional

statistical methods.

Empirical studies on the performance of structural

models generally focus on the relationship between

default risk and corporate bonds yields (see Eom,

Helwege, & Huang, 2004 and the references therein).

Few empirical studies have been conducted on the rela-

tionship between the actual default frequency and the

theoretical default probability calculated from these

models. Notable exceptions include Hillegeist, Keating,

Cram, and Lundstedt (2004) and Vassalou and Xing

(2004). Hillegeist et al. f‌i nd that structural default prob-

ability measures contain relatively more information

than Altman’s Z-Score and Ohlson’s O-Score. Using

measures calculated from structural credit risk models as

proxies to f‌i rms’ default risks, Vassalou and Xing f‌i nd

that the Fama-French SMB and HML factors (see Fama

& French, 1996) do not proxy the default factors. More-

over, in practice, Moody’s KMV has been using the

Merton (1974) type of structural models to predict the

default probabilities of individual f‌i rms.

On the other hand, f‌i rms’ credit qualities have tradi-

tionally been linked to their credit ratings, which are

ordinal measures assigned by rating agencies to ref‌l ect

the debt-issuing f‌i rms’ ability to serve their debts. The

rating process includes quantitative as well as qualitative

and legal analysis. While similar approaches are used by

different rating agencies when assigning credit ratings,

and although some agencies are more forthcoming than

others in describing the procedures they follow in assign-

ing or reviewing a rating, they all use proprietary methods

to do so. For these reasons, various models have been

proposed to predict a f‌i rm’s credit rating changes based

on publicly available information. Research in this area

includes Blume, Lim, and MacKinlay (1998), Horrigan

(1966), Kaplan, and Urwitz (1979), Pogue and Soldofsky

(1969), and West (1973) among others. Despite the skep-

ticism from rating agencies, these models have been

fairly successful in explaining and predicting the ratings

of a large cross-section of corporate bonds.

Most of the studies concerning credit rating forecast-

ing have focused on either determining the factors affect-

ing a f‌i rm’s credit rating or building better econometric

techniques. Popular approaches include hazard rate

models, logistic type of models, and discriminant analy-

sis, where usually the chosen independent variables enter

the estimation process in a linear way. Little attention

has been paid to the possible structural interactions

among the independent variables and to the possible non-

linear effect of the covariates.

Although credit ratings are not designed to pinpoint

the precise probability of default, one should expect a

close relationship between these two measures if one

believes that credit ratings are sound measures of the

credit qualities of the debt-issuing f‌i rms. Default proba-

bilities derived from structural credit risk models, at least

prima facie, provide a promising alternative for predict-

ing credit rating changes. It is thus important to investi-

gate whether these models can indeed perform better

than the traditional statistical models in terms of in-

sample f‌i tting and out-of-sample prediction of credit

rating changes. More specif‌i cally, we are interested in

answering the following question: do the credit risk mea-

sures calculated from the structural models have better

explanatory and predictive power for a f‌i rm’s credit

quality, which we assume are represented by the f‌i rm’s

credit rating, than the linear combination of those vari-

ables appearing in the structural models?

In spite of the importance of this subject, few studies

have been conducted, with the exceptions of Geske and

Delianedis (1998); KMV (1998); Leland (2002); and

Vassalou and Xing (2003). Geske and Delianedis docu-

ment that the default probability calculated from struc-

tural models contains enough information that both the

rating migration and defaults are detected months before

the actual credit events. Vassalou and Xing employ the

default measure calculated from structural models and

study the equity returns following changes in credit

ratings. KMV compares the credit risk measure obtained

from the credit transition matrix provided by rating