Blanco - Brennan - Marsh Journal of Finance 2005
Blanco - Brennan - Marsh Journal of Finance 2005
Blanco - Brennan - Marsh Journal of Finance 2005
An Empirical Analysis of the Dynamic Relation between Investment-Grade Bonds and Credit
Default Swaps
Author(s): Roberto Blanco, Simon Brennan and Ian W. Marsh
Source: The Journal of Finance, Vol. 60, No. 5 (Oct., 2005), pp. 2255-2281
Published by: Wiley for the American Finance Association
Stable URL: http://www.jstor.org/stable/3694748 .
Accessed: 21/11/2013 10:20
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Wiley and American Finance Association are collaborating with JSTOR to digitize, preserve and extend access
to The Journal of Finance.
http://www.jstor.org
ABSTRACT
We test the theoretical equivalence of credit default swap (CDS) prices and credit
spreads derived by Duffle (1999), finding support for the parity relation as an equi-
librium condition. We also find two forms of deviation from parity. First, for three
firms, CDS prices are substantially higher than credit spreads for long periods of
time, arising from combinations of imperfections in the contract specification of CDSs
and measurement errors in computing the credit spread. Second, we find short-lived
deviations from parity for all other companies due to a lead for CDS prices over credit
spreads in the price discovery process.
*Blanco is at the Banco de Espafia and was on secondment to the Bank of England while this
paper was being written, Brennan is at the Bank of England, and Marsh is at Cass Business School,
and the Cambridge Endowment for Research in Finance. He was on leave of absence at the Bank
of England when this paper was being written. We would like to thank Bill Allen, Eva Catarineu,
Gordon Gemmill, Charles Goodhart, Andrew Haldane, Simon Hayes, Kevin James, David Rule,
Hyun Shin, Michela Vecchi, Geoffrey Wood, seminar participants at the Bank of England, Banco
de Espafia, Western Finance Association 2003, Foro de Finanzas 2003, a referee, and the editor
for useful comments. Karen Goff and Andrew Paterson provided very able research assistance.
CreditTrade and J.P. Morgan Securities kindly allowed us to use their credit default swap data.
Numerous people at Banc of America Securities, Bloomberg, BNP Paribas, CreditTrade, Deutsche
Bank, and J.P. Morgan answered our questions and corrected our misunderstandings. They know
whom they are and that we are very grateful. This paper represents the views and analysis of the
authors and should not be thought to represent those of the Bank of England, Monetary Policy
Committee members, or the Banco de Espafia. Any errors and omissions are our own.
2255
1 Other basic credit derivatives include total return swaps, where the return from one asset or
group of assets is swapped for the return on another, and credit spread options, which are options
on the spread between the yield earned on two assets.
2 The economic effect of a CDS is similar to that of an insurance contract. The legal distinction
comes from the fact that it is not necessary to hold an insured asset (e.g., the underlying bond or
loan) in order to claim "compensation" under a CDS. Speculators can take long (short) positions in
credit risk by selling (buying) protection without needing to trade the cash instrument.
with the fact that CDSs are a cleaner indicator than bond spreads, our findings
suggest that CDS prices are useful indicators for analysts interested in mea-
suring credit risk.3
The paper is organized as follows. The following section describes the CDS
market and the relation between CDS prices and credit spreads. Section II
describes the data. Section III investigates empirically the short- and long-
term relations between CDS prices and spreads. Section IV contains concluding
comments.
3 Unlike bonds, CDSs have a constant maturity, the underlying instrument is always par valued,
This is the relation used in the empirical analysis that follows, although
we recognize that the arbitrage is only perfect in some instances. Duffie (1999)
shows that the spread on a par risky floating-rate note over a risk-free floating-
rate note exactly equals the CDS price. Unfortunately, floating-rate notes
are rare. The spread on a par fixed-coupon risky bond over the par fixed-
coupon risk-free bond exactly equals the CDS price if the payment dates on
the CDS and bond coincide, and recovery on default is a constant fraction
of face value (Houweling and Vorst (2002)). Alternatively, with a flat risk-
free curve and constant interest rates, the arbitrage is perfect if the payout
from a CDS on default is the sum of the principal amount plus accrued in-
terest on a risky par yield bond times one minus the recovery rate (Hull
and White (2000a)). As noted above, however, the payout from a CDS usu-
ally equals the principal amount minus the recovery rate times the sum of
principal and accrued interest on the reference obligation. Nevertheless, the
referenced papers show that the arbitrage is reasonably accurate (within 5-
10 basis points) for assets trading close to par when interest rates are not
high and yield curves are relatively flat, as was the case in our sample
period.
Three other considerations are relevant. First, as noted above, physically
settled CDS prices, especially for European entities, may contain CTD op-
tions. Other things equal, this will lead to CDS prices being greater than
the credit spread. Unfortunately, it is impossible to value this option analyti-
cally since there is no benchmark for the post-default behavior of deliverable
bonds, so we cannot simply subtract its value from the CDS price. Second,
the arbitrage relation that should keep the two prices together may rely on
short selling the cash bond. This is not always costless and indeed is some-
times not even possible in illiquid corporate bond markets. If the repo cost
of shorting the cash bond is significant, then the credit spread we have com-
puted (bond yield minus risk-free rate) underestimates the true credit spread
(bond yield minus risk-free rate plus repo cost). Again, the CDS price will
tend to be greater than the measured credit spread (Duffle (1999)). Quanti-
fying these two factors is difficult in the absence of reliable repo cost data or
a valuation model for the option. However, since both the repo cost and the
option value are bounded at zero, we can say that the CDS price is an up-
per limit on the price of credit risk while the credit spread provides a lower
limit.
Third, liquidity premia exist in both the cash bond and CDS markets.
The cash bond market is often described as relatively illiquid, particularly
outside the United States. Movements in liquidity premia may explain a
large proportion of the total variation in credit spreads (Collin-Dufresne,
Goldstein, and Martin (2001)). The CDS market is still relatively small
despite its rapid recent growth and so demand/supply imbalances can of-
ten cause short-term price movements unrelated to default expectations.
We make strenuous efforts to reduce the importance of liquidity premia
for the reference entities that we examine, as detailed in the following
section.
4 J.P. Morganwas the most active trader in the Cossin et al. (2002) CDS transactions database.
5 See also Hull, Predescu, and White (2003) and He (2002) for further elaborationon this point.
Table I
Descriptive Statistics
This table lists the reference entities in our sample, together with basic descriptive information
and the number of observations in the CDS and credit spread series. See Section II of the paper for
details on the criteria for inclusion in the sample.
Observations
350
300
250
S200
150
100
50
02/01/01 02/03/01 02/05/01 02/07/01 02/09/01 02/11/01 02/01/02 02/03/02 02/05/02
CDS - Creditspread
Figure 1. CDS price and credit spread over swaps for Ford.
reference rate should equal the CDS price of the same maturity. Define the ba-
sis to be the difference between the time t CDS price, PCDS,t,and the credit
spread, pcs,t:
basiSswaps swaps swaps
basistwaps - PCDS,t - Pcs,t = PCDSt
--Y--Xt
basisovt sPCDS,t
-=PS = PCDS,t
- -
Xgovt),
where 9 denotes the interpolated 5-year yield on the(Ytrisky bond, xswapsdenotes
the 5-year swap rate, and xgovt is the 5-year government bond yield. Figure 1
gives a representative plot of daily CDS prices and credit spreads over swaps for
Ford Motor Credit Corp. The difference is small and not always of the same sign.
In the first panel of Table II, we give the average basis and the average
absolute basis for each of our 33 reference entities, using both swap rates and
government bond yields as candidate reference rates. The mean basis across
all companies and time periods is -41 basis points using 5-year government
bond yields and +6 basis points if 5-year swap rates are used. Similarly, the
mean average absolute basis falls from 46 basis points over government bonds
to 15 basis points over swaps.6 Using median values does not alter the story.
These results are consistent with Houweling and Vorst (2002) who find an
average absolute pricing error of around 11 basis points when using swap rates
6To put these numbers in perspective, we find that the average bid-ask spread in the CDS
(indicative) quotes across all reference entities and over the full sample period was 12 basis points,
and the average spread in the cash bond market was around 9.5 basis points.
Table II
Discrepancies in the Average Pricing of Credit Risk in CDS
and Cash Bond Markets
Panel A provides descriptive statistics of the basis, defined to be the difference between the CDS
price and the credit spread, for each reference entity and expressed in basis points. The credit
spread is calculated as the difference between the interpolated 5-year yield on the risky bonds and
either the 5-year Treasury bond rate or the 5-year swap rate. Panel B provides summary statistics
for groups of bonds according to rating and nationality.
Panel A
(continued)
Table II-Continued
Panel B
and 33 basis points when using Treasury yields for bonds rated A and AA. Panel
B of Table II gives mean average basis and mean average absolute basis with
the data split by credit rating and location. The mean average absolute basis
over swaps rises as credit quality (proxied by rating) declines, a finding also
emphasized by Houweling and Vorst (2002). Similarly, the basis over swaps is
higher for European corporates than U.S. corporates (partly because the av-
erage rating of the European corporates is lower). We compute credit spreads
using swap rates rather than government bonds as the proxy for default-free
interest rates in our subsequent analysis.
The previous results suggest that the theoretical relation linking credit
spreads over the risk-free rate to CDS prices holds reasonably well on average
for most reference entities (and especially for U.S. firms). However, for some
reference entities, the average basis over swaps is meaningfully greater than
zero. The two extreme cases are France Telecom (64 basis points) and Fiat (45
basis points), with the former plotted in Figure 2. Traders indicate that large
and persistent positive bases such as these are due to the presence of the two
imperfections noted above-nonzero repo costs in the bond market leading to an
underestimated true credit spread, and the CTD option that inflates the CDS
price. J. P. Morgan (2002) illustrates the importance of including repo costs for
19 bonds with a large basis on August 16, 2002 (unfortunately just after our
sample ends). In their sample, a France Telecom 8-year bond had the highest
basis on that day (186 basis points), but it was impossible to borrowthis bond
on repo, making the true credit spread impossible to calculate.
The average basis for the remaining 18 bonds in J.P. Morgan's sample was
103 basis points, but once the repo cost was added to the credit spread over
swaps, the average repo-adjustedbasis fell to 13 basis points. High basis levels
remain for some bonds even after including repo costs. For example, the eight
European reference entities on the J.P. Morgan list had an average basis of 96
basis points and an average repo-adjustedbasis of 29 basis points.7 This rather
7 The U.S. entities had an average basis of 109 basis points and an average repo-adjusted basis
of -0.5 basis points, consistent with the hypothesis that CTD options are less important in this
jurisdiction.
500
450
400
350
300
C.
250
200
150
100
50
0
02/01/01 02/03/01 02/05/01 02/07/01 02/09/01 02/11/01 02/01/02 02/03/02 02/05/02
- CDS - Creditspread
Figure 2. CDS price and credit spread over swaps for France Telecom.
large residual is likely to be due to the CTD option. A natural experiment il-
lustrates this point. The average basis for Fiat from the start of our sample to
December 10, 2001, is just 8.8 basis points, suggesting no large repo costs or
valuable CTD option. On December 11, 2001, Fiat issued a bond convertible
into the stock of General Motors that traded at a substantial discount to ex-
isting straight Fiat bonds. If restructuring was thought possible for Fiat (and
press reports around the time suggest it was), this would increase the value
of the CTD option since this was a deliverable bond under prevailing ISDA
documentation. Immediately following the issue, the basis jumped to 50 basis
points, due almost entirely to the increase in the CDS price. Since we have no
evidence that Fiat's extant straight bonds went special after the issue, we as-
cribe this jump in the basis to the newly emerged CTD option value.8 Figure 3
illustrates the behavior of the CDS price, credit spread, and basis around this
time.
A more formal test of the equivalence of the price of credit risk across the
two markets over time can be specified in terms of transitory and permanent
price movements. Suppose that the unobservable efficient price of credit risk,
mt, follows a random walk,
I
The basis also jumped in subsequent months when Fiat was affected by rating agency actions
and equity issuance likely to have altered the valuation of the option.
200 80
190
60
180
170
40
160
.SE
01 20
140 -2
,L
130
120
-20
110
100 _ -40
F C Ba si
0o a CaS a Cedi a 0HCS
oa
-ODscpread Crdi -Bas cO
--CDS Credit spread ------ Basis (RHS}
Figure 3. Fiat's convertible bond issue and the value of the CTD option.
where ut is independently and identically distributed with zero mean and con-
stant variance. The observed price in each marketj at time t, pj,t, is equal to this
efficient price plus a component containing microstructural noise, sj,t, assumed
to be transient, plus a component reflecting other possibly nontransient factors
included in the observed price, dj,t,
B. Price Discovery
The analysis in the previous section concentrates on the long-run equilibrium
behavior of the series. In the rest of the paper,we analyze the dynamic behavior
of CDS prices and credit spreads with a focus on lead-lag relations. One of the
objectives of the paper is to evaluate the information content of indicators of
the price of credit risk, so it is important to know which market provides more
timely information.
One of the key functions of financial markets is price discovery, defined by
Lehmann (2002) to be the efficient and timely incorporationof the information
implicit in investor trading into market prices. When there is only one location
for trading an asset, by definition all price discovery takes place in that market
place. When closely related assets trade in different locations, order flow is
fragmented and price discovery is split among markets. We have demonstrated
9 The presence of a cointegrating vector is suggested for all 16 U.S. companies. Of these, three
reject the restriction of a stationary basis at the 5%level but none reject at the 1%level.
Table III
The Long-Run Relation between the Price of Credit Risk in CDS
and Bond Markets
The first two columns of Panel A present Johansen trace test statistics for the number of cointegrating
relations between the CDS price and the credit spread over swap rates. A constant is included in the
long-term relation, and the number of lags in the underlying vector autoregression is optimized using the
AIC for each entity. The third and fourth columns give test statistics for restrictions on the cointegrating
space for those entities for which a cointegrating vector appears to be present. The first restriction is that
the CDS price minus the credit spread over swaps is constant, and is distributed as chi-squared with one
degree of freedom. The second restriction is that the CDS price equals the credit spread over swaps, and
is distributed as chi-squared with two degrees of freedom. Panel B reports similar tests for Fiat over a
restricted sample period. Rejection of the null at the 10%,5%, or 1%level is indicated by a superscript *,
**,or ***,respectively. NA indicates that tests of restrictions on the cointegrating space are not possible
in the absence of a significant cointegrating vector.
Number of Cointegrating
Vectors Restrictions on Vector
that both the cash bond and CDS markets usually appear to price credit risk
equally on average. CDS prices and credit spreads are cointegrated I(1) vari-
ables for most of our sample of companies and the commonfactor can be viewed
as the implicit efficient price of credit risk. In this section, we examine which
of the two markets is more important for price discovery of credit risk.
The appropriate method to investigate the mechanics of price discovery is
not clear. The two popular common factor models due to Hasbrouck (1995)
and Gonzalo and Granger (1995) both rely on vector error-correctionmodels
(VECM)of market prices. Hasbrouck'smodel of "informationshares" assumes
that price volatility reflects new information, and thus the market that con-
tributes most to the variance of the innovations to the common factor is also
presumed to contributemost to price discovery.Gonzaloand Granger'sapproach
decomposesthe commonfactor itself, and, ignoring the correlationbetween the
markets, attributes superior price discovery to the market that adjusts least
to price movements in the other market. When price-change innovations are
correlated, Hasbrouck'sapproachcan only provide upper and lower bounds on
the information shares of each market. However, Baillie et al. (2002) argue
that the average of these bounds provides a sensible estimate of the markets'
roles in the discovery of the efficient price. Since neither method is considered
universally superior,we report both.
To compute the measures of the contributions to price discovery,it is neces-
sary first to estimate the following VECM:
and
where sit and 82t are i.i.d. shocks. If the cash bond market is contributing
significantly to the discovery of the price of credit risk, then i- will be nega-
tive and statistically significant as the CDS market adjusts to incorporatethis
information. Similarly, if the CDS market is an important venue for price dis-
covery,then -2 will be positive and statistically significant. If both coefficients
are significant, then both markets contribute to price discovery.The existence
of cointegration means that at least one market has to adjust by the Granger
representation theorem (Engle and Granger (1987)). That market is inefficient
since the price reacts to publicly available information.
Manipulations of the relative magnitudes of the - coefficients reveal which of
the two markets leads in terms of price discovery. The contributions of market 1
(the CDS market) to price discovery are defined by the following expressions:
(22
HAS 1 HAS1
2-2oI2+212 HAS2 2
1X20212
X21 -2X120r12 -2l 2- -
2X1 a12 +-
?Xa?102'
and
GG = (5)
X2 - X1
where HAS1 and HAS2 give the two bounds of Hasbrouck's measures and GG
stands for the Gonzalo and Granger measure. The covariance matrix of sit and
82t is represented by the terms a1, 12, and a2. The price discovery statistics
are reported in Panel A of Table IV for those entities for which cointegration
is present between CDS prices and credit spreads. Where appropriate, the re-
strictions that a, equals unity and ao equals zero are imposed.
In 25 of the 27 cases, X2 is significantly positive, indicating that the CDS
market contributes to price discovery. The cash bond market appears to have
a significant role to play in only eight cases. Of these eight, the cash market is
the source of all information in only one (United Utilities). In five cases, while
both cash and derivatives markets contribute significantly, the CDS market
is dominant (defined as both the Hasbrouck lower bound and the Gonzalo-
Granger measures suggesting more than 50% of the discovery occurs in the
CDS market), and in the remaining two cases the price discovery measures
give conflicting signals. On average, the CDS market contributes around 80%
of price discovery.10 Since the prices are measured asynchronously in the bond
and CDS markets, we re-compute the price discovery measures with the CDS
prices lagged by 1 day to favor deliberately the bond market. Obviously, the
CDS market's contribution to price discovery is lower in this case, but it still
remains the main forum for price discovery.
Cointegration is rejected for a small subset of our reference entities and hence
the VECM representation is not valid. We believe that rejection is due to the
presence of a substantial CTD option in the CDS price and/or binding short-
sales constraints in the cash bond market, meaning that we are markedly mis-
measuring the credit spread. Since we cannot price the option or more accu-
rately measure the spread, we rely on the simpler concept of Granger causality
in VAR-in-differences to test for price leadership in these cases. These results
are given in Panel B of Table IV. CDS prices Granger-cause credit spreads
for four of the six entities. For the other two entities, there is no causation
in either direction, while credit spreads cause CDS prices for three entities
(indicating bi-directional causality). With the exception of France Telecom,
the sum of the coefficients on lagged CDS prices is noticeably greater than
for lagged spreads, suggesting that the economic importance of CDS prices is
greater.
10 In three cases, the Gonzalo-Granger measure produces a statistic greater than one, which is
difficult to interpret. In computing the average value, we replace these numbers by unity.
Table IV
Contributions to Price Discovery
Panel A reports various measures of the contribution to the credit price discovery process made by
CDS prices for those reference entities for which the results in Table III indicate a long-run relation
between CDS prices and credit spreads exists. The measures are based on the two regressions:
p p
ApCDS,t
= 1(PCDS,t-1 - a0 1PCS,t-1) j + +
lAPCDS,t-ij Lij APCS,t-j 81t
-- -+ j=1 j=1
and
p p
=
2j APCDS,t- j
- - +
APcs,t 12(PCDS,t -1 O0 lPCS,t-1) + 622j Apcs,t- +j 2t
j=1 j=1
Where appropriate, according to the results in Table III, the restriction that ao equals zero and/or
equals unity are imposed. The Hasbrouck measure provides upper and lower bounds to the price
a
discovery contribution made in the CDS market. The table also reports the midpoint of this range. The
final column reports the Granger-Gonzalo measure. Panel B reports Granger causality test results
for those reference entities for which the results in Table III suggest no long-term relation between
CDS prices and credit spreads.
Panel A
Hasbrouck
(continued)
Table IV-Continued
Panel B
CDS Price
(1) (2) (3) (4) (5)
CDS Price
decrease in the equity price of a firm with a CDS price of 250 basis points
(the price quoted for Ford in early 2002) is associated with a 20-basis point
jump in CDS prices but just an additional 9 basis points on the credit spread.
However,the arbitrage-basedequivalence of CDS prices and credit spreads im-
plies that both are equally sensitive to firm-specific factors in the long run.
This is brought about through the lagged adjustment of the credit spread to
the CDS price, captured in the regressions by the large and significant lagged
basis term. Collin-Dufresne et al. (2001) note the sensitivity of credit spreads
to macro-variables, and question the validity of structural models of default
that focus on firm-specific variables. Our findings suggest that CDS prices
react more to firm-specific variables and that credit spreads react to lagged
changes in CDS prices; thus, our results lend some support to the structural
models.
Third, the maximum explanatory power that we are able to generate still
leaves three-quarters of the variance in both CDS prices and credit spreads un-
explained. This correspondsclosely to the proportionsfound by Collin-Dufresne
et al. (2001) in their analysis of monthly changes in credit spreads. They find
that the residual terms from their regressions are highly cross-correlated,sug-
gesting the existence of an unidentified common systematic factor, and they
suggest that credit spreads may be largely driven by market-wide demand and
supply shocks. Principal components analysis of portfolios of the residuals of
the regressions presented in Table V supports similar conclusions. Irrespective
of the formation of the portfolios,the first principal component explains a large
and essentially identical proportion of the variation of the residuals in both
CDS and credit spread equations, with approximately equal weighting on each
portfolio. As with Collin-Dufresne et al. (2001), the regressions appear to be
missing a common factor. This factor is common across reference entities and
across both cash and credit derivative markets.12
IV. Discussion
Why do we find such strong evidence that CDS prices lead credit spreads?13
Price discovery will occur in the market in which informed traders transact
firm-specific stock price and volatility increased. This increase in the value of convertibility would
raise the price of the bond and thereby reduce the value of the delivery option in the CDS price.
However, the coefficient on firm volatility should then be negative, as opposed to the positive
coefficient that we find. Further, we obtain quantitatively similar results when we only consider
U.S. entities for which the CTD option is less valuable.
12
Residuals of the regressions reported in Table G, panel C, are collected. The 32 reference
entities are repeatedly arbitrarily grouped into eight portfolios, taking simple averages of the
residuals for both CDS and credit-spread regressions. Principal components analysis is performed
on both sets of portfolios for the various groupings. The first principal component explains between
46% and 61% of the variation in the portfolio residuals, depending on the grouping of the reference
entities. Detailed results are available on request.
13 Longstaff, Mithal, and Neis (2003) also find that CDS prices lead credit spreads using simple
differenced VARs. Hull et al. (2003) find that CDS prices lead credit ratings.
most. The CDS market, as we noted above, benefits from being the easiest
place in which to trade credit risk. Its synthetic nature means that it does
not suffer from the short-sales constraints observed in the cash bond market,
and buying (or selling) relatively large quantities of credit risk is possible. The
standard CDS contract size is $10 million, while Schultz (1998) reports the
average cash market trade size is $1.5 million. Additionally, the participants
in the cash and credit derivatives markets are likely to be different. There is
no counterparty risk (beyond settlement risk) when trading a cash bond. CDS
trading, on the other hand, does entail taking on counterparty risk and for
this reason is usually restricted to institutions of relatively high credit rating.
Perhaps more importantly, the CDS market is the forum for trading credit risk,
whereas the cash market trades bond credit risk. Participants hedging loan
and counterparty exposures are able to do so in the CDS market. According
to some market participants, it is this concentration of liquidity from different
pools that means the CDS market leads the bond market.
Given CDS prices and credit spreads are linked by an arbitrage relation,
how can the markets persist in pricing credit risk differently? Our answer is in
several parts. First, in the absence of data on transactions costs, we cannot be
sure that the discrepancies are large enough to be profitable to arbitrageurs.
Second, the arbitrage relation is only approximate as noted above and we are
using a synthetic 5-year bond spread that is not traded in the market. Third,
we do not measure the repo costs of shorting the bond. It is possible that when
the credit quality of an entity declines, the repo market price increases such
that the arbitrage gap is closed. It could be argued that we have only partially
captured the price contribution from the cash market by ignoring the repo cost.
However, since repos are not traded for terms in excess of 1 year, let alone the
5 years necessary in our construct, the repo market cannot contribute toward
the discovery of the price of 5-year credit risk. Furthermore, even if the holder
of a bond sees mispricing in the CDS market, there are two reasons why he
cannot arbitrage the discrepancy-fund managers are often not permitted to
trade CDS contracts either by national law or mandate, and the notional size of
the CDS contract is so large that the cash bond holding is unlikely to be large
enough (see Dhillon, 2002).
V. Concluding Comments
This paper is a contribution to the relatively small empirical literature on
credit derivatives and, to our knowledge, is the first to examine CDS prices in
a time-series framework. It reports two major findings.
First, the theoretical relation equating CDS prices to credit spreads forms a
valid equilibrium relation for all of the U.S. and some of the European firms
examined. In the three cases in which this relation most clearly fails, CDS
prices are substantially greater than credit spreads. Two factors are at work:
(a) contract specifications, particularly in Europe, mean that a CDS price also
contains a CTD option, and so is an upper bound on the true price of credit risk;
and, (b) the credit spread used in this paper (bond yield minus risk-free rate
proxied by the swap rate) understates the true credit spread in the presence of
repo costs, and so forms a lower bound on the true price of credit risk.
Second, the CDS market leads the bond market in determining the price
of credit risk. For the 27 firms for which the equilibrium relation holds, the
CDS market contributes on average around 80% of price discovery. In four of
the remaining six cases, CDS prices Granger-causecredit spreads, suggesting
price leadership. When examining the determinants of changes in the pricing of
credit risk in the two markets, we find that macro-variables(interest rates, term
structure, equity market returns, and equity market implied volatilities) have
a larger immediate impact on credit spreads than on CDS prices. Conversely,
firm-specific equity returns and implied volatilities have a greater immediate
effect on CDS prices than on credit spreads. However, the equilibrium equiva-
lence of CDS prices and credit spreads implies that both are equally sensitive
to these variables in the long run, and we find that this is achieved through
the lagged adjustment of the credit spreads to the CDS prices, confirming the
price discovery results. We argue that price discovery occurs in the CDS market
because of (micro)structuralfactors that make it the most convenient location
for the trading of credit risk, and because there are different participants in
the cash and derivative markets who trade for different reasons.
This study leaves several avenues open to further analysis. Most obviously,
since the credit derivatives market is still small and developing, these results
are not necessarily representative of the period before or after our relatively
short span of data. Second, we have only analyzed investment-grade corporate
reference entities, although there are several sovereigns with very liquid CDS
and bond markets. Similarly, we have not considered speculative-grade cor-
porate entities, primarily because their bonds typically trade well below par,
particularly in the case of fallen angels, which weakens the arbitrage rela-
tion that underpins much of our analysis. Finally, a microstructural analysis of
price discovery across credit-risk-sensitive information releases would further
illuminate the price discovery process.
REFERENCES
Alexander, Gordon, Amy K. Edwards, and Michael G. Ferri, 1998, Trading volume and liquid-
ity in Nasdaq's high-yield bond market, Unpublished manuscript, Securities and Exchange
Commission, Washington, DC.
Baillie, Richard T., G. Geoffrey Booth, Yiuman Tse, and Tatyana Zabotina, 2002, Price discovery
and common factor models, Journal of Financial Markets 5, 309-321.
Black, Fischer, and John C. Cox, 1976, Valuing corporate securities: Some effects of bond indentures
provisions, Journal of Finance 31, 351-367.
Blanco, Roberto, Simon Brennan, and Ian W. Marsh, 2004, An empirical analysis of the dynamic
relationship between investment-grade bonds and credit default swaps, Working paper, Bank
of England.
Blume, Marshall, Felix Lim, and A. Craig MacKinlay, 1998, The declining credit quality of U.S.
corporate debt, Journal of Finance 53, 1389-1413.
British Bankers Association, 2002, BBA Credit Derivatives Report 2001/2002.
Collin-Dufresne, Pierre, Robert S. Goldstein, and J. Spencer Martin, 2001, The determinants of
credit spread changes, Journal of Finance 56, 2177-2207.
Collin-Dufresne, Pierre, and Bruno Solnik, 2001, On the term structure of default premia in the
swap and LIBOR markets, Journal of Finance 56, 1095-1115.
Cossin, Didier, Tomas Hricko, Daniel Aunon-Nerin, and Zhijiang Huang, 2002, Exploring for the
determinants of credit risk in credit default swap transaction data: Is fixed-income markets in-
formation sufficient to evaluate credit risk?, Unpublished manuscript, University of Lausanne.
Das, S. R., 1995, Credit risk derivatives, Journal of Derivatives 2, 7-23.
Das, S. R., and R. K. Sundaram, 1998, A direct approach to arbitrage-free pricing of credit deriva-
tives, NBER Working paper no. 6635.
Dhillon, Haldeep, 2002, From small beginings ..., Credit 3, 24-26.
Duffie, Darrell, 1999, Credit swap valuation, Financial Analysts Journal 55, 73-87.
Duffie, Darrell, and Ming Huang, 1996, Swap rates and credit quality, Journal of Finance 51,
921-949.
Duffie, Darrell, and Ken Singleton, 1999, Modeling term structures of defaultable bonds, Review
of Financial Studies 12, 687-720.
Engle, Robert F., and Clive W. J. Granger, 1987, Cointegration and error-correction representation,
estimation and testing, Econometrica 55, 251-276.
Geske, Robert, 1977, The valuation of corporate liabilities as compound options, Journal of Finan-
cial and Quantitative Analysis 12, 541-552.
Gonzalo, Jesus, and Clive W. J. Granger, 1995, Estimation of common long-memory components in
cointegrated systems, Journal of Business and Economic Statistics 13, 27-35.
Hakkio, Craig, and Mark Rush, 1991, Cointegration: How short is the long-run? Journal of Inter-
national Money and Finance 10, 571-581.
Hasbrouck, Joel, 1995, One security, many markets: Determining the contributions to price discov-
ery, Journal of Finance 50, 1175-1199.
He, Hua, 2002, Modeling term structure of swap spreads, Unpublished manuscript, Yale University.
Hong, Gwangheon, and Arthur Warga, 2000, An empirical study of bond market transactions,
Financial Analysts Journal 56, 32-46.
Houweling, Patrick, and Ton Vorst, 2002, An empirical comparison of default swap pricing models,
Unpublished manuscript, Erasmus University Rotterdam.
Hull, John C., Mirela Predescu, and Alan White, 2003, The relationship between credit default swap
spreads, bond yields, and credit rating announcements, Unpublished manuscript, University
of Toronto.
Hull, John C., and Alan White, 2000a, Valuing credit default swaps I: No counterparty default risk,
Journal of Derivatives 8, 29-40.
Hull, John C., and Alan White, 2000b, Valuing credit default swaps II: Modeling default correla-
tions, Journal of Derivatives 8, 12-22.
Jarrow, Robert A., David Lando, and Stuart Turnbull, 1997, A Markov model for the term structure
of credit spreads, Review of Financial Studies 10, 481-523.
Jarrow, Robert A., and Stuart Turnbull, 1995, Pricing derivatives on financial securities subject to
credit risk, Journal of Finance 50, 53-86.
Lando, David, 1997, Modelling bonds and derivatives with default risk, in M. Dempster,
and S. Pliska, eds.: Mathematics of Derivative Securities (Cambridge University Press,
Cambridge).
Lehmann, Bruce N., 2002, Some desiderata for the measurement of price discovery across markets,
Journal of Financial Markets 5, 259-276.
Longstaff, Francis A., Sanjay Mithal, and Eric Neis, 2003, The credit-default swap market: Is credit
protection priced correctly? Unpublished manuscript, UCLA.
Longstaff, Francis A., and Eduardo Schwartz, 1995, A simple approach to valuing risky fixed and
floating rate debt, Journal of Finance 50, 789-821.
McCauley, Robert N., 2002, International market implications of declining treasury debt, Journal
of Money, Credit, and Banking 34, 952-966.
Merton, Robert C., 1974, On the pricing of corporate debt: The risk structure of interest rates,
Journal of Finance 29, 449-470.