Question

Compare and contrast benchmark spreads and CDS prices as market-based measures of default risk.

Answer 1

The prices of or spread on credit default swaps (CDS) theoretically represent the pure credit risk of a firm. Callen, Livnat and Segal (2007) note that although the CDS premium is related to credit ratings issued by the rating agencies, rather wide variation in CDS spreads are observed for firms having a given rating. Following the recent subprime debacle, rating agencies have come under much scrutiny due to their role in the mispricing of credit risk and questions regarding the validity of the ratings that they issue are being questioned. This paper investigates the relationship between CDS spreads and credit ratings to help explain how market participants perceive and price credit risk. Using daily data obtained from Bloomberg on 391 five-year CDS contracts over the period 2003 to 2008, we model the credit default spreads as well as the variation between CDS spreads and credit ratings. Empirical results indicate that after controlling for market returns, market volatility and interest rates, CDS spreads increase with the subordination of the debt instrument, the put-implied volatility or deteriorating credit quality of the reference entity. We construct a credit quality variable derived from the quintiles of daily CDS spreads. Empirical results reveal statistically significant differences between credit ratings and our spread based credit quality variable. Observed discrepancies can be partly explained by stock market returns, levels of the VIX index, short-term and long-term interest rates as well as credit quality. However, empirical results indicate that a substantial share of the difference between credit ratings and CDS spreads cannot be attributed to either market or reference entity related variables.

Credit risk makes up perhaps the largest risk an investor bares when buying a bond,,

which theoretically effects all bonds with the exception of certain entities that are effectively

default remote (such as the U.S. government). Credit risk is defined as the uncertainty associated

with potential loss of either principle or interest on a fixed income obligation, and can be

decomposed into the probability of loss and the loss given default. In many cases credit risk is

synonymous with default risk, in that default is associated with the inability or unwillingness of a

borrower to make payments. However, the concept of credit risk is broader, in that in the event

that a borrower suffers severe credit deterioration (e.g., a multi-notch downgrade or steep decline

in the price of debt), then is becomes likely that the lender will not receive any future anticipated

cash flows and a loss may have to be recognized. Indeed, even in the case of marketable bank

debt, a borrower may be deemed unlikely to pay even if the loan is performing. It is for these

reasons market participants use derivatives to mitigate such risks using CDS.

According to the policy and guidelines issued by the Nationally Recognized Statistical

Rating Organizations (NRSROs)29 at any given time credit rating on an issue of debt reflects its

relative credit quality. This has the interpretation that a credit rating embodies information on

probability of default (PD) relative to a cohort, potentially allowing for a standard comparison of

likelihood of default and severity of loss in the event of default. Therefore, ratings represent an

opinion of the rating agency regarding potential loss, a firm’s capacity to pay back all its sources

of financing, as well as the recovery of a particular in the event of default (Micu, Remolona, and

Wooldridge 2006). However, there are some subtle differences between the rating agencies

(Standard and Poor’s (S&P), Moody’s and Fitch), as well as the type of ratings that they issue.

S&P has historically issued primarily a senior unsecured debt rating, presumably a ranking or

pure default risk, and in cases where there is subordinated debt a separate rating that may be

worse to reflect the greater recovery risk. On the other hand, Moody’s claims to have always

issues a debt specific expected loss (EL) based rating, such that every debt of an issuer could

have different ratings reflective of varying expected LGDs amongst different parts of the capital

structure; and in cases where there is senior unsecured debt, a rating that reflects more purely the

PD.

However, ratings typically are issued for a firm and not for individual firm debt, since in

many cases all of a firm’s outstanding debt will have the same rating; this is typically the case of

a simple capital structure, such as a firm issuing mostly pari pasu bonds, as opposed to complex

capital structures of the very largest firms. The agencies claim that only if a fundamental change

occurs in a borrower’s creditworthiness will they modify the firm’s relative credit quality,

implying that they would not be reacting to systematic events that affect all firms equally but do

not impact relative credit quality30.

Variables Affecting CDS Spreads

While under certain ideal conditions the theoretical the CDS spread should represent pure

credit risk, in practice there are many cases in which this does not hold true. First, one may

would argue that both systematic and unsystematic factors independently influence spread levels

(e.g., interest rates), and pure measures of credit risk should only be influenced systematical

factors to the extent that fundamental credit quality is a function of such. For example to the

extent that worsening macroeconomic conditions effect the risk aversion of CDS market

participants, this may affect spreads above and beyond the detrimental impact on the credit

quality of the reference entity (i.e., greater PD). Other variables affecting spread may be said to

lay somewhere between systematic and unsystematic factors; a prime example being liquidity, in

that we can think of systematic as well as idiosyncratic notions of liquidity32

Five of the most common variables found to affect CDS spread include the equity

market’s implied volatility, industry, leverage of the reference entity, the risk-free rate, and

liquidity of the CDS contract. Scneider, Sogner and Veza (2007) find evidence that equity

market volatility as measured by the VIX index is positively correlated with long term and short

term default factors that directly influence the valuation of the CDS.

The main empirical results of this study are shown in Tables 1 through 3. In Table 1 we

present our leading two models for the logarithm of the CDS spread, LCDS. Table 2 tabulates

the results of the models for the CDS Category (CDSC) and the Rating Category (RTGC).

CDSC is oriented such that 1 is lowest daily quintile of the CDS spread (i.e. best in terms of the

CDS credit signal), and 5 is the highest quintile of the CDS spread (i.e., the worst CDS credit

signal). RTGC represents a simple re-coding of the S&P rating, oriented in the same manner as

CDSC, such that 1 is the best rating and 5 is the worst rating. In Table 3 can be found the

results of models for CDSC and RTGC agreement (AGREE), as well as for the distance between

the CDSC and RTGD (DIST). AGREE is an indicator that takes the value 1 if TGC and CDSC

are the same and 0 if they are different. DIST is CDSC minus RTGC, such that if this is positive

(negative), then the CDS credit signal is worse (better), or that the issue is priced more cheaply

(richly) in the CDS market as relative to that implied by the agency rating 34. All models are

cross-sectional time series regression models estimated using the method generalized least

squared (GLS). Across all models, we note that in all cases all coefficient estimates are

statistically significant at the 1% level, and overall goodness of fit measures compare favorably

with what has been seen in the literature.

In the LCDS regression of Table 1, where we show our two leading models, we see that

our set of covariates can explain anywhere from 29% to 40% of the variation in CDS spreads.

However, note that Model 1, having market capitalization in lieu of trading volume in Model 2,

is slightly better by this metric than Model 2. Model 1 has an r-squared between (across) cross-

sectional groups of 39.6% (37.8%), while the respective statistics for Model 2 are 32.0%

(29.0%)