In this paper we study the pricing of credit risk as reflected in the market for credit default swaps (CDS) between 2003 and 2008. This market has newly emerged as the reference for credit risk pricing because of its use of standardized contract specifications and has achieved a higher level of liquidity than typically prevails in the markets for the underlying notes and bonds of the named corporate issuers. We initiate our exploration by studying a particular case which allows us to set out some of the issues of CDS pricing in a simple way. We show that for the purposes of accounting for relatively short-term changes of CDS spreads, an approach based on the structural (or firm-value based) models of credit risk faces an important obstacle in that reliable information about the firm's liabilities required to calculate the \distance to default" are available only quarterly or in some cases annually. Thus structural models account for short-term movements in credit spreads largely by changes in the issuer's equity price. In the case studied we show the e®ect of equity returns in explaining weekly changes of spreads is insignificant and of the wrong sign. In examination of particular episodes when the CDS spread was particularly delinked from the firm's equity series, we find that a likely explanation is changes in expectations about the firm's planned capital market operations. Since these are hard to capture in an observed proxy variable, we argued that this motivates the use of latent variable models that have recently been employed in the credit risk literature. We further see that movements in the CDS spreads for the particlar name chosen are highly correlated with an index of CDS spreads for industrial Blue-chip names. Building on these observations we then consider CDS pricing for a panel of firms for which CDS contracts were traded between between September 2003 and through January 2008. To facilitate comparison we have drawn our sample from two sectors, energy and media, from North America and Europe. Overall we have 41 firms across four subsamples allowing us to make two-way comparisons (across sectors and regions). Our estimates of a linear model show a strong positive association between spread changes on individual names and a broad-based index of CDS price changes. In contrast, the association with equity prices is very weak, generally statistically insignificant, and often of the wrong sign. These results are robust to inclusion of firm fixed or random e®ects. We find a negative autocorrelation of residuals in these panel estimates which we interpret as evidence of mean reversion in unobserved risk factors. All these results are consistent across our four subsets, i.e., they hold for North American Energy and Media and European Energy and Media. We pursue our study by exploring a latent variable model recently introduced in the literature which assumes that defaults on a name follow a jump process where the log intensity of arrivals of defaults itself follows an Ornstein-Uhlenbeck process. After developing a continuous time model of CDS pricing with this underlying stochastic process, we estimate our model for our 41 firms individually, applying no restrictions across firms. Our results are rather mixed in the sense that some firms do seem have mean reverting default intensities and others do not. Overall the evidence of mean reversion is stronger in our study than that found previously. The estimated models are then used to produce an implied time-series of instan- taneous default intensities for our 41 firms observed at weekly intervals. We carry out a principal components analysis of the panels of default intensities for our four sector-region combinations. In all cases a very high fraction of weekly variations in the implied default intensity is explained by a single common factor. We find that the implied common factor for each subsample is highly correlated with the default intensity implied by the index of CDS spreads on Blue-chip names. This is strong evidence con¯rming the presence of a general credit risk factor whose existence has been proposed in a number of recent contributions. We then ask what our estimates of default intensities derived from CDS prices imply for the market price of default risk. In order to answer this question we need to compare our estimates of the risk neutral intensity process with estimates of the statistical default process. We argue that recent studies which have used the Moodys- KMV EDF (estimated default frequencies) are essentially confounding information about the risk-neutral and statistical default distributions. Other estimates based on ratings su®er from the well-know problem of inertia in ratings changes. We therefore employ proxies for the statistical default intensities derived from a large panel data set of North American firms using firm accounting variables as well as macro covariates. Speci¯cally, we use the estimates recently derived by Zhou (2007) who employs a methodology similar to Shumway (2001) and Campbell et al (2005) but corrects for possible sample selection bias induced by the earlier studies' treatment of missingobservations. These estimates are implemented for our North American ¯rms only. Our results show that in both the energy and media subsamples risk-neutral intensities are much more variable than statistical intensities. A high proportion of observed variation in both kinds of intensities is accounted for by firm level di®erences. There is a high positive correlation between risk neutral and statistical default intensities. We then combine estimates to ¯nd the implied market price of risk measured as the natural logarithm of the ratio of risk-neutral intensity and statistical intensity of default. We show that a relatively high fraction of the observed variation of this market price of default risk can be accounted for by a common time variation. In order to identify this factor, we explore a linear model of the market price of default risk using as observed covariates macro indicators, firm indicators and indicators of equity market and credit market conditions. Our estimates show a strong association between that credit market conditions and the market price of risk. The estimated coe±cients have the correct signs. These are robust findings across a variety of alternative proxies for credit market conditions and across our two subsamples. In contrast equity market risk factors and general business conditions do not always have coefficient estimates of the right sign and are not always significant. However, there is some evidence that changes in the value ¯rm premium are partially correlated with changes in the pricing of default risk. Overall, our results provide evidence of the partial segmentation of credit markets.
