Journal of Financial Economics 103 (2012) 280–293

Contents lists available at SciVerse ScienceDirect

Journal of Financial Economics

journal homepage: www.elsevier.com/locate/jfec

Counterparty credit risk and the credit default swap market$

Navneet Arora a, Priyank Gandhi b, Francis A. Longstaff b,c,n
a
American Century Investments, United States
b
UCLA Anderson School, United States
c
NBER, United States

a r t i c l e in f o abstract

Article history: Counterparty credit risk has become one of the highest-profile risks facing participants
Received 19 February 2010 in the financial markets. Despite this, relatively little is known about how counterparty
Received in revised form credit risk is actually priced. We examine this issue using an extensive proprietary data
8 April 2011
set of contemporaneous CDS transaction prices and quotes by 14 different CDS dealers
Accepted 9 June 2011
selling credit protection on the same underlying firm. This unique cross-sectional data
Available online 19 October 2011
set allows us to identify directly how dealers’ credit risk affects the prices of these
JEL classification: controversial credit derivatives. We find that counterparty credit risk is priced in the
G12 CDS market. The magnitude of the effect, however, is vanishingly small and is
G13
consistent with a market structure in which participants require collateralization of
G24
swap liabilities by counterparties.
& 2011 Elsevier B.V. All rights reserved.
Keywords:
Counterparty credit risk
Credit default swaps
Collateralization

1. Introduction and swap obligations.1 Fears of systemic defaults were so

extreme in the aftermath of the Lehman bankruptcy that
During the past several years, counterparty credit risk Euro-denominated CDS contracts on the U.S. Treasury
has emerged as one of the most important factors driving were quoted at spreads as high as 100 basis points.
financial markets and contributing to the global credit crisis. Despite the significance of counterparty credit risk in
Concerns about counterparty credit risk were significantly the financial markets, however, there has been relatively
heightened in early 2008 by the collapse of Bear Stearns, but little empirical research about how it affects the prices of
then skyrocketed later in the year when Lehman Brothers contracts and derivatives in which counterparties may
declared Chapter 11 bankruptcy and defaulted on its debt default. This is particularly true for the $57.3 trillion
notional credit default swap (CDS) market in which
defaultable counterparties sell credit protection (essen-
tially insurance) to other counterparties.2 The CDS markets
$
The authors are grateful for the comments of Darrell Duffie, Chris
have been the focus of much attention recently because it
Jones, Peter Knez, Peter Meindl, Derek Schaeffer, Victor Wong, and was AIG’s massive losses on credit default swap positions
seminar participants at the 2010 NYU Moody’s Conference, the 2010
Moody’s Risk Practitioner Conference in San Francisco, the 2010 SIAM
1
conference on Financial Mathematics and Engineering, and the 2010 Lehman Brothers filed for Chapter 11 bankruptcy on September
USC/UCLA Financial Research Conference. We are particularly grateful 15, 2008. During the same month, American International Group (AIG),
for the comments and advice of the editor Bill Schwert and an Merrill Lynch, Fannie Mae, and Freddie Mac also failed or were placed
anonymous referee. All errors are our responsibility. under conservatorship by the U.S. government.
n 2
Corresponding author at: UCLA Anderson School, United States. The size of the CDS market as of June 30, 2008 comes from
E-mail address: [email protected] (F.A. Longstaff). estimates reported by the Bank for International Settlements.

0304-405X/$ - see front matter & 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.jfineco.2011.10.001
 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293 281

that led to the Treasury’s $182.5 billion bailout of AIG. dealers’ credit risk to be more evident in selling credit
Furthermore, concerns about the extent of counterparty protection on other financial firms. Surprisingly, we find
credit risk in the CDS market underlie recent proposals to that counterparty credit risk is priced in the CDS spreads
create a central clearinghouse for CDS transactions.3 of all firms in the sample except for the financials.
This paper uses a unique proprietary data set to These results have many implications for current pro-
examine how counterparty credit risk affects the pricing posals to regulate the CDS market. As one example, they
of CDS contracts. Specifically, this data set includes con- argue that market participants may view current CDS risk
temporaneous CDS transaction prices and quotations mitigation techniques such as the overcollateralization of
provided by 14 large CDS dealers for selling protection swap liabilities and bilateral netting as largely successful in
on the same set of underlying reference firms. Thus, we addressing counterparty credit risk concerns. Thus, propo-
can use this cross-sectional data to measure directly how sals to create a central CDS exchange may not actually be
a CDS dealer’s counterparty credit risk affects the prices at effective in reducing counterparty credit risk further.
which the dealer can sell credit protection. A key aspect of This paper contributes to an extensive literature on the
the data set is that it includes most of 2008, a period effect of counterparty credit risk on derivatives valuation.
during which fears of counterparty defaults in the CDS Important research in this area includes Cooper and Mello
market reached historical highs. Thus, this data set (1991), Sorensen and Bollier (1994), Duffie and Huang
provides an ideal sample for studying the effects of (1996), Jarrow and Yu (2001), Hull and White (2001),
counterparty credit risk on prices in derivatives markets. Longstaff (2004, 2010), and many others. The paper most
Four key results emerge from the empirical analysis. closely related to our paper is Duffie and Zhu (2009) who
First, we find that there is a significant relation between study whether the introduction of a central clearing
the credit risk of the dealer and the prices at which the counterparty into the CDS market could improve on
dealer can sell credit protection. As would be expected, existing credit mitigation mechanisms such as bilateral
the higher the dealer’s credit risk, the lower is the price netting. They show that a central clearing counterparty
that the dealer can charge for selling credit protection. might actually increase the amount of credit risk in the
This confirms that prices in the CDS market respond market. Thus, our empirical results support and comple-
rationally to the perceived counterparty risk of dealers ment the theoretical analysis provided in Duffie and Zhu.
selling credit protection. The remainder of this paper is organized as follows.
Second, although there is a significant relation Section 2 provides a brief introduction to the CDS market.
between dealer credit risk and the cost of credit protec- Section 3 discusses counterparty credit risk in the context
tion, we show that the effect on CDS spreads is vanish- of the CDS markets. Section 4 describes the data. Section 5
ingly small. In particular, an increase in the dealer’s credit examines the effects of dealers’ credit risk on spreads in
spread of 645 basis points only translates into a one- the CDS market. Section 6 summarizes the results and
basis-point decline on average in the dealer’s spread for presents concluding remarks.
selling credit protection. This small effect is an order of
magnitude smaller than what would be expected if swap 2. The credit default swap market
liabilities were uncollateralized. In contrast, the size of
the pricing effect is consistent with the standard practice In this section, we review briefly the basic features of a
among dealers of having their counterparties fully col- typical CDS contract. We then discuss the institutional
lateralize swap liabilities. structure of the CDS market.
Third, the Lehman bankruptcy in September 2008 was
a major counterparty credit event in the financial mar-
kets. Accordingly, we examine how the pricing of counter- 2.1. CDS contracts
party credit risk was affected by this event. We find that
counterparty credit risk was priced prior to the Lehman A CDS contract is best thought of as a simple insurance
bankruptcy. After the Lehman event, the point estimate of contract on the event that a specific firm or entity defaults
the effect increases but remains very small in economic on its debt. As an example, imagine that counterparty A
terms. The increase is significant at the 10% level (but not buys credit protection on Amgen from counterparty B by
at the 5% level). paying a fixed spread of, say, 225 basis points per year for
Fourth, we study whether the pricing of counterparty a term of five years. If Amgen does not default during this
credit risk varies across industries. In theory, the default period of time, then B does not make any payments to A. If
correlation between the firm underlying the CDS contract there is a default by Amgen, however, then B pays A the
and the CDS dealer selling protection on that firm should difference between the par value of the bond and the
affect the pricing. Clearly, to take an extreme example, no post-default value (typically determined by a simple
investor would be willing to buy credit protection on auction mechanism) of a specific Amgen bond. In essence,
Citigroup from Citigroup itself. Similarly, to take a less the protection buyer is able to put the bond back to the
extreme example, we might expect the pricing of CDS protection seller at par in the event of a default. Thus, the
CDS contract ‘‘insures’’ counterparty A against the loss of
value associated with default by Amgen.4
3
For example, see the speech by Federal Reserve Board Chairman
Ben S. Bernanke at the Council on Foreign Relations on March 10, 2009.
4
For an in-depth discussion of the economics of CDS clearinghouse For a detailed description of CDS contracts, see Longstaff, Mithal,
mechanisms, see Duffie and Zhu (2009). and Neis (2005).
282 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293

2.2. The structure of the CDS market Second, even if the reference firm underlying the CDS
contract does not default, a participant in the CDS market
Like interest rate swaps and other fixed income deri- could still experience a substantial loss in the event that
vatives, CDS contracts are traded in the over-the-counter the counterparty to the contract entered financial distress.
market between large financial institutions. During the The reason for this is that while CDS contracts initially
past 10 years, CDS contracts have become one of the have value of zero when they are executed, their mark-to-
largest financial products in the fixed-income markets. As market values may diverge significantly from zero over
of June 30, 2008, the total notional amount of CDS time as credit spreads evolve. Specifically, consider the
contracts outstanding was $57.325 trillion. Of this case where counterparty A has an uncollateralized mark-
notional, $33.083 trillion is with dealers, $13.683 trillion to-market liability of X to counterparty B. If counterparty
with banks, $0.398 trillion with insurance companies, A were to enter bankruptcy, thereby canceling the CDS
$9.215 trillion with other financial institutions, and contract and making the liability immediately due and
$0.944 trillion with nonfinancial customers.5 payable, then counterparty B’s only recourse would be to
Early in the development of the CDS market, partici- attempt to collect its receivable of X from the bankruptcy
pants recognized the advantages of having a standardized estate. As such, counterparty B would become a general
process for initiating, documenting, and closing out CDS unsecured creditor of counterparty A. Given that the debt
contracts. The chartering of the International Swaps and and swap liabilities of Lehman Brothers were settled at
Derivatives Association (ISDA) in 1985 led to the devel- only 8.625 cents on the dollar, this could result in
opment of a common framework which could then be counterparty B suffering substantial losses from the
used by institutions as a uniform basis for their swap and default of counterparty A.7
derivative transactions with each other. Currently, ISDA A third way in which a market participant could suffer
has 830 member institutions. These institutions include losses through the bankruptcy of a counterparty is
virtually every participant in the swap and derivatives through the collateral channel. Specifically, consider the
markets. As the central organization of the privately case where counterparty A posts collateral with counter-
negotiated derivatives industry, ISDA performs many party B, say, because counterparty B is counterparty A’s
functions such as producing legal opinions on the enfor- prime broker. Now imagine that the collateral is either
ceability of netting and collateral arrangements, advan- not segregated from counterparty B’s general assets (as
cing the understanding and treatment of derivatives and was very typical prior to the Lehman default), or that
risk management from public policy and regulatory capi- counterparty B rehypothecates counterparty A’s collateral
tal perspectives, and developing uniform standards and (also very common prior to the Lehman default). In this
guidelines for the derivatives industry.6 context, a rehypothecation of collateral is the situation in
which counterparty B transfers counterparty A’s collateral
to a third party (without transferring title to the collat-
3. Counterparty credit risk
eral) in order to obtain a loan from the third party.
Buhlman and Lane (2009) argue that under certain cir-
In this section, we first review some of the sources of
cumstances, the rehypothecated securities become part of
counterparty credit risk in the CDS market. We then
the bankruptcy estate. Thus, if counterparty B filed for
discuss ways in which the industry has attempted to
bankruptcy after rehypothecating counterparty A’s collat-
mitigate the risk of losses stemming from the default of a
eral, or if counterparty A’s collateral was not legally
counterparty to a CDS contract.
segregated, then counterparty A would become a general
unsecured creditor of counterparty B for the amount of
3.1. Sources of counterparty credit risk the collateral, again resulting in large potential losses. An
even more precarious situation would be when the
There are at least three ways in which a participant in rehypothecated collateral itself was seized and sold by
the CDS market may suffer losses when their counter- the third party in response to counterparty B’s default on
party enters into financial distress. First, consider the case the loan obtained using the rehypothecated securities as
in which a market participant buys credit protection on a collateral. Observe that because of this collateral channel,
reference firm from a protection seller. If the reference counterparty A could suffer significant credit losses from
firm underlying the CDS contract defaults, the protection counterparty B’s bankruptcy, even if counterparty B does
buyer is then owed a payment from the counterparty. If not actually have a mark-to-market liability to counter-
the default was unanticipated, however, then the protec- party A stemming from the CDS contract.
tion seller could suddenly be faced with a large loss. If the
loss was severe enough, then the protection seller could 3.2. Mitigating counterparty credit risk
potentially be driven into financial distress. Thus, the
protection buyer might not receive the promised protec- One of the most important ways in which the CDS
tion payment. market attempts to mitigate counterparty credit risk is

5 7
Data obtained from Table 4 of OTC Derivatives Market Activity for The settlement amount was based on the October 10, 2008
the First Half of 2008, Bank for International Settlements. Lehman Brothers credit auction administered by Creditex and Markit
6
This discussion draws on the information about ISDA provided on and participated in by 14 major Wall Street dealers. See the Lehman
its Web site www.isda.org. auction protocol and auction results provided by ISDA.
 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293 283

through the market infrastructure provided by ISDA. In the contract. In light of the potential credit risk, full
particular, ISDA has developed specific legal frameworks collateralization of CDS liabilities has become the market
for standardized master agreements, credit support standard. For example, the ISDA Margin Survey (2009)
annexes, and auction, closeout, credit support, and nova- reports that 74% of CDS contracts executed during 2008
tion protocols. These ISDA frameworks are widely used by were subject to collateral agreements and that the esti-
market participants and serve to significantly reduce the mated amount of collateral in use at the end of 2008 was
potential losses arising from the default of a counterparty approximately $4.0 trillion. Typically, collateral is posted
in a swap or derivative contract.8 in the form of cash or government securities. Participants
Master agreements are encompassing contracts between in the Margin Survey indicate that approximately 80% of
two counterparties that detail all aspects of how swap and the ISDA credit support agreements are bilateral, implying
derivative contracts are to be executed, confirmed, docu- two-way transfers of collateral between counterparties. Of
mented, settled, etc. Once signed, all subsequent swaps and the 20 largest respondents to the survey (all large CDS
derivative transactions become part of the original master dealers), 50% of their collateral agreements are with hedge
swap agreement, thereby eliminating the need to have funds and institutional investors, 15% are with corpora-
separate contracts for each transaction. An important tions, 13% are with banks, and 21% are with others.
advantage of this structure is that it allows all contracts The data set used in this study represents the CDS
between two counterparties to be netted in the event of a spreads at which the largest Wall Street dealers actually
default by one of the counterparties. This netting feature sell, or are willing to sell, credit protection. Both discus-
implies that when default occurs, the market value of all sions with CDS traders and margin survey evidence
contracts between counterparties A and B are aggregated indicate that the standard practice by these dealers is to
into a net amount, leaving one of the two counterparties require full collateralization of swap liabilities by both
with a net liability to the other. Without this feature, counterparties to a CDS contract. In fact, the CDS traders
counterparties might have incentives to demand payment we spoke with reported that the large Wall Street dealers
on contracts on which they have a receivable, but repudiate they trade with typically require that their non-dealer
contracts on which they have a liability to the defaulting counterparties overcollateralize their CDS liabilities
counterparty. slightly. This is consistent with the ISDA Margin Survey
Credit support annexes are standardized agreements (2009) that documents that the 20 largest firms
between counterparties governing how credit risk mitiga- accounted for 93% of all collateral received, but only 89%
tion mechanisms are to be structured. For example, a of all collateral delivered, suggesting that there was a net
specific type of credit risk mitigation mechanism is the inflow of collateral to the largest CDS dealers. Further-
use of margin calls in which counterparty A demands more, the degree of overcollateralization required can
collateral from counterparty B to cover the amount of vary over time. As an example, one reason for the liquidity
counterparty B’s net liability to counterparty A. The credit problems at AIG that led to emergency loans by the
support annex specifies details such as the nature and Federal Reserve was that AIG would have been required
type of collateral to be provided, the minimum collateral to post additional collateral to CDS counterparties if AIG’s
transfer amount, how the collateral amount is to be credit rating had downgraded further.9
calculated, etc. At first glance, the market standard of full collaterali-
ISDA protocols specify exactly how changes to master zation seems to suggest that there may be little risk of a
swap agreements and credit support annexes can be loss from the default of a Wall Street credit protection
modified. These types of modifications are needed from seller. This follows since the protection buyer holds
time to time to reflect changes in the nature of the collateral in the amount of the protection seller’s CDS
markets. For example, the increasing tendency among liability. In actuality, however, the Wall Street practice of
market participants to closeout positions through nova- requiring non-dealer protection buyers to slightly over-
tion rather than by offsetting positions motivated the collateralize their liabilities actually creates a subtle
development of the 2006 ISDA Novation Protocol II. counterparty credit risk. To illustrate this, imagine that a
Similarly, the creation of a standardized auction mechan- protection buyer has a mark-to-market liability to the
ism for settling CDS contracts on defaulting firms moti- protection seller of $15 per $100 notional amount.
vated the creation of the 2005–2009 ISDA auction Furthermore, imagine that the protection seller requires
protocols and the 2009 ISDA closeout amount protocol. the protection buyer to post $17 in collateral. Now
An important second way in which counterparty credit consider what occurs if the protection seller defaults.
risk is minimized is through the use of collateralization. The bankruptcy estate of the protection seller uses $15
Recall that the value of a CDS contract can diverge of the protection buyer’s collateral to offset the $15 mark-
significantly from zero as the credit risk of the reference to-market liability. Rather than returning the additional
firm underlying the contract varies over time. As a result, $2 of collateral, however, this additional capital becomes
each counterparty could have a significant mark-to-mar- part of the bankruptcy estate. This implies that the
ket liability to the other at some point during the life of protection buyer is now an unsecured creditor in the

8 9
Bliss and Kaufman (2006) provide an excellent discussion of the For example, see the speech by Federal Reserve Chairman Ben S.
role of ISDA and of netting, collateral, and closeout provisions in Bernanke before the Committee on Financial Services, U.S. House of
mitigating systemic credit risk. Representatives, on March 24, 2009.
284 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293

amount of the $2 excess collateral. Thus, in this situation, included in this file. Thus, all 14 of these dealers sold
the protection buyer could suffer a significant loss even credit protection to the asset management firm during the
though the buyer actually owed the defaulting counter- sample period. Of these transactions, however, most
party on the CDS contract. involve either firms that are not in the CDX index, or
This scenario is far from hypothetical. In actuality, a contracts with maturities significantly different from five
number of firms experienced major losses on swap con- years. Screening out these trades results in a sample of
tracts in the wake of the Lehman bankruptcy because of several hundred observations.
their net exposure (swap liability and offsetting collateral) To augment the sample, we also include quotes pro-
to Lehman.10 vided directly to the firm by the CDS dealers selling
protection on the firms in the CDX index. As described
4. The data above, these quotes represent firm offers to sell protection
and there can be sanctions for dealers who fail to honor
Fixed-income securities and contracts are traded pri- their quotes. For example, if the asset management firm
marily in over-the-counter markets. For example, Treas- finds that a dealer is often not willing to execute new
ury bonds, agency bonds, sovereign debt, corporate bonds, trades (or unwind existing trades) at quoted prices, then
mortgage-backed securities, bank loans, interest rate that dealer could be dropped from the list of dealers that
swaps, and CDS contracts are all traded in over-the- the firm’s traders are willing to do business with. Given
counter markets. Because of the inherent decentralized the large size of the asset management firm providing the
nature of these markets, however, actual transaction data, the major CDS dealers included in the study have
prices are difficult to observe. This is why most of the strong incentives to provide actionable quotes.
empirical research in the financial literature about fixed- There are a number of clear indications that the deal-
income markets has typically been based on the quotation ers respond to these incentives and provide reliable
data available to participants in these markets. quotes. First, the dealers included in the study frequently
We were fortunate to be given access to an extensive update their quotes throughout the trading day. The total
proprietary data set of CDS prices by one of the largest number of quotations records in the data set for firms in
fixed-income asset management firms in the financial the CDX index is 673,060. This implies an average of 2.19
markets. A unique feature of this data set is that it contains quotations per day per dealer for each of the firms in the
both actual CDS transaction prices for contracts entered sample. Thus, quotes are clearly being refreshed through-
into by this firm as well as actionable quotations provided out the trading day. Second, the fact that all 14 of the CDS
to the firm by a variety of CDS dealers. These quotations dealers sold protection to the asset management firm
are actionable in the sense that the dealers are keenly during the sample period suggests that each was active in
aware that the firm expects to be able to trade (and often providing competitive and actionable quotes during this
does) at the prices quoted by the dealers (and there are period. Third, we compare our sample of transaction
implicit sanctions imposed on dealers who do not honor prices directly to the quotes available in the market on
their quotations). Thus, these quotations should more the same day. This comparison is necessarily a little noisy
closely represent actual market prices than the indicative since the transaction prices are not time-stamped within
quotes typically used in the fixed-income literature. the day, and we are comparing them to quotes available
In this paper, we study the spreads associated with in the market at roughly 11:30 AM. Despite this, however,
contracts in which 14 major CDS dealers sell five-year the average transaction price is only 0.26 basis points
credit protection to the fixed-income asset management below the minimum quote available in the market. The
firm on the 125 individual firms in the widely followed standard deviation of the difference is 5.87 basis points
CDX index. The sample period for the study is March 31, and the difference between the mean transaction price
2008 to January 20, 2009. This period covers the turbulent and minimum quote is not statistically significant.
Fall 2008 period in which Fannie Mae, Freddie Mac, As mentioned, dealers frequently update their quota-
Lehman Brothers, AIG, etc. entered into financial distress tions throughout the day to insure that they are current.
and counterparty credit fears reached their peak. Thus, Since our objective is to study whether the cross-sectional
this sample period is ideally suited for studying the effects dispersion in dealer prices is related to counterparty
of counterparty credit risk on financial markets. credit risk, it is important that we focus on dealer prices
The transactions data in the sample are taken from a that are as close to contemporaneous as possible. To this
file recording the spreads on actual CDS contracts exe- end, we extract quotes from the data set in the following
cuted by the firm in which the firm is buying credit way. First, we select 11:30 AM as the reference time. For
protection. There are roughly 1,000 transactions in this each of the 14 CDS dealers, we then include the quote
file. The average transaction size is $6.5 million and the with time-stamp nearest to 11:30 AM, but within 15
average maturity of these contracts is 4.9 years. All 14 of minutes (from 11:15 to 11:45 AM). In many cases, of
the major CDS dealers to be studied in this paper are course, there may not be a quote within this 30-minute
period. Thus, we will generally have fewer than 14 prices
or quotes available for each firm each day. For a firm to be
10
From the October 7, 2008 Financial Times: ‘‘The exact amount of included in the sample for a particular day, we require
any claim is determined by the difference between the value of the
collateral and the cost of replacing the contract.. . . Moreover, many
that there be two or more prices or quotes for that firm.
counterparties to Lehman who believe it owes them money have joined We repeat this process for all days and firms in the
the ranks of unsecured creditors.’’ sample.
 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293 285

This algorithm results in a set of 13,383 observation from the Bloomberg system and reflect the market’s percep-
vectors of synchronous prices or quotations by multiple CDS tion of the counterparty credit risk of the dealers selling
dealers for selling protection on a common underlying credit protection to the asset management firm.
reference firm. Since there are 212 trading days in the Table 2 reports summary statistics for the CDS spreads
sample period, this implies that we have data for multiple for these dealers. As shown, the average CDS spread
CDS dealers for an average of 63.13 firms each day. Table 1 ranges from a low of 59.40 basis points for BNP Paribas
presents summary statistics for the data. As shown, the to a high of 355.10 basis points for Morgan Stanley. Note
number of synchronous quotes ranges from two to nine. On that CDS data for Lehman Brothers and Merrill Lynch are
average, an observation includes 3.073 dealer quotes for the included in the data set even though these firms either
reference firm for that day. Table 1 also shows that the went bankrupt or merged during the sample period. The
variation in the quotes provided by the various dealers is reason for including these firms is that both were actively
relatively modest. For most of the observations, the range of making markets in selling credit protection through much
CDS quotations is only on the order of two to three basis of the sample period. Thus, their spreads may be particu-
points, and the median range is three basis points. larly informative about the impact of perceived counter-
In addition to the prices and quotes provided by the party credit risk on CDS spreads.
dealers selling protection, we also need a measure of
the counterparty credit risk of the dealers themselves. To 5. Empirical analysis
this end, we obtain daily midmarket five-year CDS quotes
referencing each of the 14 major CDS dealers in the study. In this section, we begin by briefly describing the
The midmarket spreads for these CDS contracts are obtained methodology used in the empirical analysis. We then test

Table 1
The distribution of dealer prices and quotations.
This table provides summary statistics for the distribution of dealer prices or quotations for CDS contracts referencing the firms in the CDX index. The
panel on the left summarizes the distribution in terms of the number of dealer prices and quotations on a given day for a CDS contract referencing a
specific firm. The panel on the right summarizes the distribution in terms of the range R of prices and quotations (measured in basis points) provided by
dealers on a given day for a CDS contract on a specific reference firm. Only days on which two or more simultaneous prices or quotations are available for
a specific firm are included in the sample as an observation. The sample period is March 31, 2008 to January 20, 2009.

Number Observations Percentage Range Observations Percentage

2 4907 36.66 0 1175 8.78

3 4518 33.78 0oRr1 1952 14.59
4 2566 19.17 1oRr2 2298 17.17
5 1012 7.56 2oRr3 1925 14.38
6 267 1.99 3oRr4 1065 7.96
7 84 0.62 4oRr5 1800 13.44
8 21 0.16 5 o R r 10 2209 16.51
9 8 0.06 10 o R r 20 748 5.59
20 o R 211 1.58

Total 13,383 100.00 Total 13,383 100.00

Table 2
Summary statistics for CDS contracts referencing dealers.
This table provides summary statistics for the CDS spreads (in basis points) for contracts referencing the dealers listed below. The spreads are based on
daily observations obtained from the Bloomberg system. N denotes the number of days on which Bloomberg quotes are available for the indicated dealer.
The sample period is March 31, 2008 to January 20, 2009.

Standard
Dealer Mean deviation Minimum Median Maximum N

Barclays 122.65 43.33 53.27 122.17 261.12 212

BNP Paribas 59.40 13.29 34.24 59.08 107.21 212
Bank of America 121.60 35.77 61.97 119.75 206.85 209
Citigroup 180.67 71.13 87.55 162.90 460.54 207
Credit Suisse 111.66 37.20 57.59 101.40 194.22 212
Deutsche Bank 96.88 29.70 51.92 90.11 172.00 212
Goldman Sachs 230.58 110.62 79.83 232.69 545.14 177
HSBC 75.41 21.94 41.84 67.59 128.30 212
JP Morgan 110.86 27.96 62.54 107.68 196.34 209
Lehman 291.79 89.01 154.04 285.12 641.91 84
Merrill Lynch 243.19 71.34 114.35 218.43 472.72 193
Morgan Stanley 355.10 236.22 108.06 244.98 1360.00 187
Royal Bank of Scotland 116.45 45.16 55.17 110.69 304.89 212
UBS 139.09 56.81 55.45 126.24 320.80 212
286 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293

whether counterparty credit risk is reflected in the prices Table 3

of CDS contracts. Finally, we study whether the pricing of Results from the regression of CDS spreads on the CDS spread of the
corresponding dealer.
counterparty credit risk by dealers varies by industry as
This table reports the results from the regressions of CDS prices or
would be implied by a correlation-based credit model. quotations for the firms in the CDX Index on the CDS spread of the dealer
providing the CDS price or quotation. The sample period is March 31,
2008 to January 20, 2009. Regression specification II includes a dummy
5.1. Methodology variable IL that takes value one for the post-Lehman period beginning
September 15, 2008, and zero otherwise. The t-statistics are based on
For each reference firm and for each date t in the the White (1980) heteroskedasticity-consistent estimate of the covar-
iance matrix. The superscript nn denotes significance at the 5% level; the
sample, we have simultaneous prices from multiple CDS
superscript n denotes significance at the 10% level.
dealers for selling five-year credit protection on that firm.
Thus, we can test directly whether counterparty credit risk I: CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ Ei,j,t ,
is priced by a straightforward regression of the price of
II : CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ gIL,t Spreadj,t1 þ Ei,j,t :
protection sold or quoted by a dealer for a reference firm
on the price of protection for the dealer itself providing
that quotation. In this panel regression framework, we Regression specification II
allow for reference-firm-specific date fixed effects. Speci- Regression specification I with post-Lehman dummy
fically, we estimate the following regression:
Variable Coefficient t-Statistic Coefficient t-Statistic
CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ Ei,j,t , ð1Þ
Spread  0.001548  7.31nn  0.000991  3.73nn
where CDSi,j,t denotes the CDS spread for credit protection IL Spread  0.000713  1.92n
on reference firm i sold or quoted by dealer j at date t, ai,t
N 41,122 41,122
is a fixed effect parameter specific to firm i at time t, and
Spreadj,t1 is the CDS spread for dealer j as of the end of
the previous day.11 Under the null hypothesis that coun- CDS market. Furthermore, the sign of the coefficient is
terparty credit risk is not priced, the slope coefficient b is negative, consistent with economic intuition.
zero. The t-statistics for b reported in the tables are based We acknowledge, however, that we cannot completely
on the White (1980) heteroskedastic-consistent estimate rule out the possibility that the relation between CDS
of the covariance matrix. spreads and the credit risk of protection sellers may
As shown in Table 1, there are a total of 13,383 actually be due to some other factor that is correlated
observation vectors in the sample. On average, each with dealer spreads.12 For example, since CDS contracts
observation vector consists of 3.073 distinct quotations are traded in over-the-counter markets, the search costs
for selling credit protection on the reference firm, giving a associated with finding trading partners could play a role
total of 41,122 observations collectively. Thus, there are in determining equilibrium CDS spreads (see Duffie,
339.85 observations on average for each of the 121 Gârleanu, and Pedersen, 2002, 2005, 2008 and others). If
reference firms in the sample. these search costs were inversely related to dealer CDS
spreads, then they could potentially affect CDS spreads in
5.2. Is counterparty credit risk priced? a way consistent with the results reported in Table 3. We
will explore some of these possibilities in a later section
Although a formal model of the relation between a on robustness.
dealer’s credit risk and the price at which the dealer could
sell credit protection could be developed, the underlying 5.3. Why is the effect so small?
economics of the transaction makes it clear that there
should be a negative relation between the two. Specifi- Although statistically very significant, the slope coeffi-
cally, as the credit risk of a protection seller increases, the cient is relatively small in economic terms. In particular,
value of the protection being sold is diminished and the value of 0.001548 implies that the credit spread of a
market participants would not be willing to pay as much CDS dealer would have to increase by nearly 645 basis
for it. Thus, if counterparty credit risk is priced in the points to result in a one-basis-point decline in the price of
market, the slope coefficient b in the regressions should credit protection. As shown in Table 2, credit protection
be negative. on most of CDS dealers in the sample never even reached
Table 3 reports the results from estimating the regres- 645 basis points during the period under study. These
sion in Eq. (1) (which is designated specification I). The results are consistent with the results in Table 1 suggest-
slope coefficient b is  0.001548 with a t-statistic of ing that the cross-sectional variation in the dealers’
 7.31. Thus, the empirical results strongly support the quotes for selling credit protection on a specific reference
hypothesis that counterparty credit risk is priced in the firm is only on the order of several basis points.
A number of papers have explored the theoretical
11
We use the dealer’s spread as of t1 rather than t since the dealer magnitude of counterparty credit risk on the pricing of
data are as of the end of the day while the CDS quotation data are taken interest rate swaps. Important examples of this literature
from a narrow timeframe centered at 11:30 AM. Thus, using the dealer’s
spread as of the end of day t1 avoids using ex post data in the
12
regression. We are grateful to the referee for raising this issue.
 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293 287

include Cooper and Mello (1991), Sorensen and Bollier models that abstract from the collateralization of CDS
(1994), and Duffie and Huang (1996). Typically, these contracts. Rather, the small size of the pricing effect
papers find that since the notional amount is not exchanged appears more consistent with the standard market prac-
in an interest rate swap, the effect of counterparty credit tice of full collateralization, or even overcollateralization,
risk on an interest rate swap is very small, often only a basis of CDS contract liabilities.
point or two.
Unlike an interest rate swap, however, a CDS contract 5.4. Did pricing of counterparty credit risk change?
could involve a very large payment by the protection
seller to the protection buyer. For example, sellers of The discussion above suggests that the Lehman bank-
protection on Lehman Brothers were required to pay ruptcy event may have forced market participants to
$91.375 per $100 notional to settle their obligations to reevaluate the risks inherent in even fully collateralized
protection buyers. Thus, the results from the interest rate counterparty relationships. If so, then the pricing of
swap literature may not necessarily be directly applicable counterparty credit after the Lehman bankruptcy might
to the CDS market. differ from the pricing in the CDS market previous to the
A few recent papers have focused on the theoretical bankruptcy. To explore this possibility, we reestimate the
impact of counterparty credit risk on the pricing of CDS regression described above using a dummy slope coeffi-
contracts. Important examples of these papers include cient for the post-Lehman period. Specifically, we esti-
Jarrow and Yu (2001), Hull and White (2001), Brigo and mate the regression
Pallavicini (2006), Kraft and Steffensen (2007), Segoviano CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ g IL,t Spreadj,t1 þ Ei,j,t , ð2Þ
and Singh (2008), and Blanchet-Scalliet and Patras (2008).
In general, estimates of the size of the effect of counter- where IL is a dummy variable that takes value one for the
party credit risk in this literature tend to be orders of post-Lehman period beginning September 15, 2008, and
magnitude larger than those in the literature for interest zero otherwise. Table 3 also reports the results from this
rate swaps. For example, estimates of the potential size of regression (which is designated specification II). Note that
the pricing effect range from 7.0 basis points in Kraft and in this specification, the coefficient b represents the
Steffensen to more than 20 basis points in Hull and White, regression slope during the pre-Lehman period, while
depending on assumptions about the default correlations the coefficient g measures the change in the slope after
of the protection seller and the underlying reference firm. the Lehman bankruptcy. Thus, we can test for whether
Thus, this literature tends to imply counterparty credit there was a significant change in the pricing of counter-
risk pricing effects many times larger than those we find party credit risk after the Lehman bankruptcy by simply
in the data. testing whether g is statistically significant. The regres-
It is crucial to recognize, however, that this literature sion slope during the post-Lehman period can be obtained
focuses almost exclusively on the case in which CDS by simply summing the pre-Lehman slope coefficient b
contract liabilities are not collateralized. As was discussed and the post-Lehman change in the slope coefficient g.
earlier, the standard market practice during the sample The results provide some support for the hypothesis that
period would be to require full collateralization by both the pricing of counterparty credit risk changed after the
counterparties to a CDS contract. This would be particu- Lehman bankruptcy. Specifically, the pre-Lehman slope
larly true for CDS contracts in which one counterparty coefficient is  0.000991 and has a t-statistic of 3.73.
was a large Wall Street CDS dealer. After the Lehman bankruptcy, the change in the slope
In theory, full collateralization of CDS contract liabil- coefficient is  0.000713, making the pricing of counter-
ities would appear to imply that there should be no party credit risk in the post-Lehman period roughly twice
pricing of counterparty credit risk in CDS contracts. In as large as in the pre-Lehman period. The t-statistic for the
reality, however, there are several reasons why there change, however, is only  1.92. Thus, the change is
might still be a small pricing effect even if counterparties significant at the 10% level, but not the 5% level.
require full collateralization. First, as became clear after
the Lehman bankruptcy, counterparties who post collat- 5.5. Robustness of the results
eral in excess of their liabilities risk becoming unsecured
creditors of a defaulting counterparty for the amount of To provide some robustness checks for these results,
the excess collateral. As discussed earlier, however, Wall we also estimate several alternative specifications. In the
Street CDS dealers often require a small amount of over- first of these, we include the total number of trades
collateralization from their counterparties (typically on executed by each dealer each day as a control for trading
the order of several percent) thus creating the possibility activity. Specifically, we estimate the following regression
of a slight credit loss (ironically, however, only when the specifications:
counterparty owes the bankrupt firm money). Second, the CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ Z Volumej,t þ Ei,j,t , ð3Þ
Lehman bankruptcy also showed that there were a num-
ber of legal pitfalls that many market participants had not CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ gIL,t Spreadj,t1
previously appreciated. These include the risk of unse-
þ Z Volumej,t þ Ei,j,t , ð4Þ
gregated margin accounts or the disposition of rehypothe-
cated collateral. where Volumej,t denotes the total number of trades
In summary, the size of the counterparty pricing effect executed by dealer j on date t. Table 4 reports the results
in the CDS market appears too small to be explained by from the regressions.
288 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293

Table 4 Table 5
Results from the regression of CDS spreads on the CDS spread of the Results from the regression of CDS spreads on the CDS spread of the
corresponding dealer with control for dealer trading volume. corresponding dealer with fixed effects for individual dealers.
This table reports the results from the regressions of CDS prices or This table reports the results from the regression of CDS prices or
quotations for the firms in the CDX index on the CDS spread of the dealer quotations for the firms in the CDX index on the CDS spread of the dealer
providing the CDS price or quotation and on the total number of trades providing the CDS price or quotation. The regression also includes a
executed by the dealer in all CDX index firms that day as a control separate fixed effect dummy variable for each dealer (except for the
variable (denoted as volume). The sample period is March 31, 2008 to dealer with the largest number of quotes, arbitrarily designated dealer
January 20, 2009. Regression specification II includes a dummy variable 14). The sample period is March 31, 2008 to January 20, 2009. Regres-
IL that takes value one for the post-Lehman period beginning September sion specification II includes a dummy variable IL that takes value one for
15, 2008, and zero otherwise. The t-statistics are based on the White the post-Lehman period beginning September 15, 2008, and zero
(1980) heteroskedasticity-consistent estimate of the covariance matrix. otherwise. The t-statistics are based on the White (1980) heteroskedas-
The superscript nn denotes significance at the 5% level; the superscript n ticity-consistent estimate of the covariance matrix. The superscript nn
denotes significance at the 10% level. denotes significance at the 5% level; the superscript n denotes signifi-
cance at the 10% level.
I: CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ Z Volumej,t þ Ei,j,t ,
X
13

II : CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ gIL,t Spreadj,t1 þ Z Volumej,t þ Ei,j,t : I: CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ dj Ij þ Ei,j,t ,
j¼1

II : CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ gIL,t Spreadj,t1

Regression specification II
X
13 X
13
Regression specification I with post-Lehman dummy þ dj Ij þ Zj Ij IL þ Ei,j,t :
j¼1 j¼1
Variable Coefficient t-Statistic Coefficient t-Statistic

nn
Spread  0.001548  7.30  0.000990  3.73nn Regression specification II
IL Spread  0.000714  1.92n Regression specification I with post-Lehman dummy
Volume 0.008122 0.12 0.009988 0.14
Variable Coefficient t-Statistic Coefficient t-Statistic
N 41122 41122
nn
Spread  0.001338  4.49  0.001786  2.35nn
I1  1.4154  3.87nn  0.1130  0.23
I2 0.6574 4.17nn 0.7774 4.34nn
Even after controlling for dealer trading activity, I3 0.1707 1.56 0.1923 1.88n
Table 4 shows the regression coefficients and t-statistics I4 0.4062 4.95nn 0.5837 7.50nn
for the dealers’ CDS spreads are virtually the same as they I5 0.2106 1.95n 0.0086 0.09
I6 0.0326 0.64  0.0461  0.82
are in Table 3. Thus, the results provide evidence that the
I7 0.4728 2.28nn 0.4227 2.07nn
dealer spread is not simply proxying for dealer liquidity I8 0.6006 6.03nn 0.2026 2.28nn
effects. I9  0.1701  1.66n  0.1136  0.82
As another robustness check, we reestimate the I10 0.1041 1.49 0.3960 3.75nn
regressions in Table 3, but with dummy variables for I11 0.1862 3.60nn 0.1982 3.05nn
I12 0.9453 6.96nn 0.6462 3.74nn
individual dealers. This specification controls for dealer
I13 0.1922 1.64 0.0659 0.65
fixed effects. Thus, the relation between CDS spreads for
IL Spread 0.000347 0.36
the firms in the CDX index and dealer CDS spreads is
I1 IL  1.4112  2.21nn
identified using only the times-series variation in spreads. I2 IL  1.0839  2.78nn
The regressions estimated are I3 IL  0.0857  0.30
I4 IL  0.7415  2.90nn
X
13
CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ dj Ij þ Ei,j,t , ð5Þ I5 IL 0.4342 1.47
I6 IL 0.7280 2.58nn
j¼1
I7 IL 0.5204 0.32
I8 IL 1.6748 4.68nn
CDSi,j,t ¼ ai,t þ b Spreadj,t1 þ gIL,t Spreadj,t1 I9 IL – –
X
13 X
13 I10 IL  1.1010  4.79nn
þ dj Ij þ Zj Ij IL þ Ei,j,t , ð6Þ I11 IL 0.0423 0.17
j¼1 j¼1 I12 IL 0.6155 2.08nn
I13 IL 2.6544 1.87n
where Ij is the dummy variable for the j-th dealer. Note
that we only include 13 dealer dummies rather than all N 41,122 41,122
14. This is because inclusion of all 14 dummies results in a
collinearity with the firm and date fixed effects. Thus, the
regression coefficients for dealer dummies have the inter- corresponding estimate in Table 3. The t-statistic for
pretation of the marginal effect relative to that of the dealer CDS spread in this regression is  4.49. In the
omitted dealer, which is chosen to be the dealer with the second specification with the post-Lehman dummy vari-
highest trading activity throughout the sample period. able, the CDS spread of the dealer is again significantly
The results from these regressions are reported in Table 5. negative during the pre-Lehman period, and there is no
The results indicate that the previous results are significant change in the variable after the Lehman bank-
robust to the inclusion of dealer fixed effects. The coeffi- ruptcy. This again provides support for the result that
cient for dealer CDS spread is  0.001338 for the first dealer credit risk is priced in the market, although the
specification, which is only slightly less than the effect is very small.
 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293 289

The coefficients for the individual dealer dummy vari- Table 6

ables are also interesting. Although many of the coeffi- Results from regression of CDS spreads on the CDS spreads of the
corresponding dealer interacted with sector dummy variables for the
cients in the first specification are significant, almost all of
underlying firms.
them are much less than one basis point in magnitude. This table reports the results from the regression of CDS prices or
The same is also true for the pre-Lehman coefficients for quotations for the firms in the CDX index on the CDS spread of the dealer
the second specification. On the other hand, the results providing the CDS price or quotation interacted with five sector dummy
indicate that a number of the coefficients change in the variables where the dummy variables take value one if firm i is in the
consumer, energy, financial, industrial, or technology sectors, respec-
post-Lehman period by one or more basis points. These tively, and zero otherwise. The sample period is March 31, 2008 to
changes, however, are essentially equally divided January 20, 2009. Regression specification II includes a dummy variable
between positive and negative values. Thus, these results IL that takes value one for the post-Lehman period beginning September
provide some evidence of greater heterogeneity in dealer 15, 2008, and zero otherwise. The t-statistics are based on the White
(1980) heteroskedasticity-consistent estimate of the covariance matrix.
fixed effects in the post-Lehman period.13
The superscript nn denotes significance at the 5% level; the superscript n
denotes significance at the 10% level.

5.6. Are there differences across firms? X

5
I: CDSi,j,t ¼ ai,t þ bk ISectork Spreadj,t1 þ Ei,j,t ,
k¼1
A number of recent papers have emphasized the role
that the default correlation between the protection seller X
5
II : CDSi,j,t ¼ ai,t þ bk ISectork Spreadj,t1
and the reference firm should play in determining CDS k¼1

spreads. To illustrate the importance of correlation, let us X

5
þ gk ISectork IL Spreadj,t1 þ Ei,j,t :
take it to an extreme and imagine that Citigroup is willing k¼1
to sell credit protection against the event that Citigroup
itself defaults. Clearly, no one would be willing to pay
Citigroup for this credit protection.14 Similarly, a financial Regression specification II
Regression with post-Lehman
institution selling credit protection on another financial specification I dummy
institution might not be able to charge as much as a
nonfinancial seller might.15 Variable Coefficient t-Statistic Coefficient t-Statistic
To explore the effects of correlation on the price of
IConsumer Spread  0.001161  4.83nn  0.000015  0.04
credit protection, we do the following. First, we classify IEnergy Spread  0.002313  7.25nn  0.002253  5.14nn
the firms in the CDX index that are in our sample into one IFinancial Spread 0.001097 0.77  0.000910  0.67
of five broad industry sectors or categories: consumer, IIndustrial Spread  0.001324  3.61nn  0.001245  2.42nn
ITechnology Spread  0.002553  5.41nn  0.003173  4.69nn
energy, financials, industrials, and technology. We then
reestimate the regressions using the following specifica- IConsumer IL Spread  0.001719  3.65nn
IEnergy IL Spread  0.000079  0.09
tions:
IFinancial IL Spread 0.003183 1.27
X
5 IIndustrial IL Spread  0.000096  0.14
CDSi,j,t ¼ ai,t þ bk ISectork Spreadj,t1 þ Ei,j,t , ð7Þ ITechnology IL Spread 0.000674 0.80
k¼1
N 41,122 41,122

X
5
CDSi,j,t ¼ ai,t þ bk ISectork Spreadj,t1
k¼1
The most puzzling result, however, is that for the
X
5
þ gk ISectork IL Spreadj,t1 þ Ei,j,t , ð8Þ financial sector. As described above, the correlation argu-
k¼1 ment suggests that the counterparty credit risk for the
CDS dealers should be most evident when they are selling
where ISectork are dummy variables that take value one if protection on firms in the financial industry. In contrast to
firm i is in sector k, and zero otherwise. The regression this intuition, however, the results show that the CDS
results are reported in Table 6. dealers’ counterparty credit risk is not priced in the
As shown in the first specification, counterparty credit spreads of CDS contracts on financial firms. Furthermore,
risk is priced for the consumer, energy, industrial, and likelihood ratio tests strongly reject the hypotheses that
technology firms in the sample. The t-statistics for the the slope coefficient for the financial sector is equal to
corresponding coefficients are  4.83,  7.25,  3.61, and that of the consumer, energy, industrial, and technology
5.41, respectively. These results are clearly consistent sectors, with p-values of 0.00026, 0.00000, 0.00012, and
with the previous results. 0.00000, respectively. Thus, the pricing of counterparty
credit risk for financial firms is significantly different from
13
We are grateful to the referee for suggesting the robustness that of the other four categories of firms in the sample. In
checks discussed in this section. summary, far from being the most sensitive to counter-
14
It is interesting to note, however, that a number of European party credit risk, financial firms in the CDX index repre-
banks sell credit protection on the iTraxx index which includes these sent the only category in the sample for which
banks as index components.
15
Examples of recent papers discussing the role of correlation in the
counterparty credit risk is not priced.
pricing of CDS contracts include Hull and White (2001), Jarrow and Yu These patterns are repeated in the second specifica-
(2001), Longstaff, Mithal, and Neis (2005), Yu (2007), and many others. tion. As shown, counterparty credit risk is significantly
290 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293

priced for the energy, industrial, and technology firms In this model, we take the perspective of the protection
during the pre-Lehman period. Furthermore, there is no buyer and model the losses arising from the default of the
significant change in how counterparty credit risk is protection seller. To model default, we use the reduced-
priced for these firms in the post-Lehman period. Coun- form framework of Duffie and Singleton (1997, 1999) in
terparty credit risk for firms in the consumer sector is not which the default of a firm is triggered by the realization
priced during the pre-Lehman period, but there is a of a jump process. Let lt and nt denote the risk-neutral
significant change in pricing for these firms after the intensity processes of the firm underlying the CDS con-
Lehman event. The results also show that counterparty tract and the firm selling credit protection (the CDS
credit risk for the financial firms is not priced in the pre- counterparty), respectively. The risk-neutral dynamics
Lehman period, and that there is no significant change in for these intensity processes are given by,
this relation after the Lehman event. pffiffiffi
What factors might help account for the evidence that dl ¼ ðablÞdt þ s l dZ l , ð9Þ
counterparty credit risk is not priced for the financial
firms? First of all, the financial firms in the CDX index pffiffiffi
consist primarily of insurance firms, industrial lenders, dn ¼ ðmgnÞdt þs n dZ n , ð10Þ
consumer finance firms, and real estate companies. Thus,
it is possible that the default risk of these firms in the CDX where a, b, s, m, g, and s are constant parameters, and Corr
index may actually be much less correlated with that of ðdZ l ,dZ n Þ ¼ x. Given this model, the marginal distribution
the CDS dealers than one might expect based on their for the default time of the underlying firm has a hazard
designation as financials. Second, counterparty credit risk function equal to the realized path of the intensity (see
might not be priced in the cost of selling protection on the Lando, 1998), and similarly for the firm selling default
large financial firms in the CDX index if the market protection. Modeling the simultaneous distribution of
believed that the CDS dealers would not fail when the defaults would require a specification of the probability
large financial firms in the CDX index became vulnerable of simultaneous defaults. We will specify the joint dis-
to default. Thus, this possibility suggests that there might tribution of defaults in our discrete-time simulation.
be a state-contingent aspect to the default risk of CDS Following Gregory (2010), we distinguish between
dealers. Finally, it is important to acknowledge that there three types of default scenarios. The first is the case in
is actually little empirical evidence in the literature about which the underlying firm defaults but not the counter-
default correlations. Thus, while intuition suggests that party. In this case, the protection buyer receives the
the default correlation between financial firms should be protection payment from the protection seller and does
higher than the default correlation between financial and not suffer any counterparty credit losses.
nonfinancial firms, there is no direct empirical evidence The second case is when the counterparty defaults, but
supporting this intuition. For this reason, the analysis in the underlying firm does not. For simplicity, we assume
this section should be viewed more as an exploratory that both counterparties are required to post full collat-
investigation, rather than as a test rejecting specific eral daily for CDS liabilities, where the mark-to-market
empirical hypotheses about default correlations. liability is computed under the assumption that both
counterparties are default free.17 In addition, we assume
that there is zero recovery of uncollateralized liabilities in
6. Comparison to model-implied values
the event that the protection seller defaults.18 Given the
square-root dynamics in Eq. (9), the value of a CDS
The empirical results demonstrate that counterparty
contract can be obtained directly from the CDS valuation
credit risk is priced by the market, but that the size of the
model in Longstaff, Mithal, and Neis (2005, pp. 2221–
effect is very small. A natural question to ask is whether
2222). There are now two ways in which a protection
these empirical results can be reconciled with those
buyer can suffer a loss when the protection seller defaults.
implied by theoretical models of counterparty credit
If the mark-to-market value is positive, but the collateral
risk.16
posted the previous day (which equals the previous day’s
There is a large and rapidly growing literature on the
mark-to-market value of the CDS contract) is insufficient,
valuation of counterparty credit risk in CDS contracts
then the buyer’s loss is the difference between the two. As
which is far too extensive for us to review fully here.
discussed earlier, however, the buyer can also lose from a
Gregory (2010) provides an excellent summary of the
counterparty default when he owes the counterparty on
literature and discusses a number of the modeling
the CDS contract and the amount of collateral posted with
approaches that have been applied to the problem of
the defaulting protection seller exceeds the amount of the
valuing counterparty credit risk. In this section, we
buyer’s liability. In this situation, the excess collateral
compare our empirical results with those implied by a
becomes part of the bankruptcy estate and represents the
simple simulation-based model of the effects of counter-
protection buyer’s loss. Note that the loss of excess
party credit risk. A key feature of this framework is that
collateral does not occur when CDS liabilities are
it allows us to quantify the size of the effect when
CDS counterparties collateralize their mark-to-market
liabilities. 17
This assumption greatly simplifies the analysis but has virtually
no effect on the total amount of collateral required.
18
This is consistent with the Lehman default in which CDS contracts
16
We are grateful to the referee for raising this issue. referencing Lehman were settled at 8.625 cents on the dollar.
 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293 291

uncollateralized. Thus, there are states in which a protec- the longer-term properties of the CDX index.20 We also
tion buyer may be worse off with full bilateral collater- assume that the spread correlation parameter x takes on
alization of CDS liabilities. values of 2%, 6%, or 10%. Similarly, we assume that the
The third case occurs when both the underlying firm default correlation r takes on values of 2%, 6%, or 10%.
and the counterparty default at the same time. We will These values essentially bracket the default correlations
make the assumption that joint default occurs if both the reported by Longstaff and Rajan (2008) implied from the
firm and the counterparty default within a two-business- prices of CDX index tranches and the CDS spreads for the
day timeframe. This assumption reflects the reality that a constituents of the CDX index.21
discrete period of time is required operationally to post Table 7 reports the estimated basis-point cost of
collateral and settle trades. With collateralization, the counterparty default for a range of scenarios. Specifically,
protection buyer’s loss is the difference between the loss we compute the cost of events in which only the counter-
on the underlying firm and the amount of collateral held. party defaults, the cost of joint events in which both the
Again, since the buyer may have posted collateral with underlying firm and the counterparty default, and the
the defaulting counterparty, the buyer could actually be total of these two costs. The default intensity for the
worse off in some states in this joint default scenario than underlying firm takes values of 100 or 300 basis points,
without collateralization. essentially bracketing the CDX index values during the
Since we are simulating changes in the intensity sample period. Similarly, the default intensity for the
processes and the realization of defaults at each time counterparty selling protection takes values of 100, 300,
step, we only need to specify local or one-step joint and 500 basis points, again paralleling the behavior of
probabilities to simulate joint default events. In particu- broker CDS spreads during the sample period. The results
lar, conditional on no default having occurred before time are based on 100,000 simulations for a five-year CDS
t, the marginal probability of the underlying firm default- contract. The details on how the joint distribution of
ing between time t and t þ Dt is lt Dt. Similarly, the defaults is simulated are described in the Appendix.
marginal probability of the firm selling credit protection The results in Table 7 imply counterparty credit risk
defaulting between time t and t þ Dt is nt Dt. Let a, b, c, and pricing effects that are very consistent with those docu-
d denote the joint probabilities that neither firm defaults, mented in previous sections of this paper. For example, a
that only the underlying firm defaults, that only the firm 400-basis-point increase in the CDS spread of the protection
selling credit protection defaults, and that both firms seller from 100 to 500 basis points maps into an increase in
default between time t and t þ D, respectively. The Appen- counterparty credit costs of roughly 0.5, 1.0, and 2.0 basis
dix shows that these joint probabilities are completely points in the cases where the default correlation is 2%, 6%,
determined by the two marginal probabilities and a and 10%, respectively. Thus, the empirical estimates of the
default correlation parameter r. Thus, we are in essence size of the effect of counterparty credit risk on CDS spreads
assuming that the local joint distribution of default events given in this paper harmonize well with those implied by a
is given by a simple multinomial distribution. Further- model in which average default correlations are in the range
more, this approach explicitly allows for correlated of, say, zero to 4%.
defaults to occur. Given these joint probabilities, we
simulate the model in steps of Dt and sample the four 7. Conclusion
joint events based on their multinomial probabilities. We
repeat this process at each time step along a simulated We examine the extent to which the credit risk of a
path until the first default occurs.19 dealer offering to sell credit protection is reflected in the
Turning to the issue of calibration, it is important to prices at which the dealer can sell protection. We find
stress that our objective is simply to provide general strong evidence that counterparty credit risk is priced in
estimates of the size of counterparty default effects rather the market; the higher the credit risk of a dealer, the
than to model specific contracts. As such, we adopt a lower is the price at which the dealer can sell credit
generic parameterization and estimate counterparty protection in the market. The magnitude of the effect,
default costs under a broad range of assumptions about however, is extremely small. In particular, an increase in
default intensities and correlations. The average value of the credit spread of a dealer of about 645 basis points
the CDX index during the sample period is 95 basis points, maps into only a one-basis-point decline in the price of
while the average CDS spread for the dealers during the credit protection.
same period is 145 basis points. These values, of course, The price of counterparty credit risk appears to be too
are high by historical standards but they do provide a small to be explained by models that assume that CDS
realistic benchmark for the calibration of the risk-neutral liabilities are unsecured. The pricing of counterparty
intensity processes. Accordingly, we parameterize the credit risk, however, seems consistent with the standard
long-run values of lt and nt to be 100 and 150 basis market practice of requiring full collateralization, or even
points, respectively. Furthermore, we assume b ¼ g ¼ 0:50 the overcollateralization of CDS liabilities. These results
and s ¼ s ¼ 0:20. These parameters are consistent with
20
Specifically, the moments of normalized monthly changes in the
19
Note that the limiting distribution of this multinomial distribu- CDX index from 2004 to 2009 imply b ¼ 0:54 and s ¼ 0:18.
21
tion would likely be of the form of a bivariate exponential distribution as For a few of the 100,000 simulated paths, we assume a smaller
the number of time steps increases (see Johnson and Kotz, 1972). We are value of r to insure that simulated joint default probabilities remain
grateful to the referee for this insight. positive. See the discussion in the Appendix.
292 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293

Table 7
Basis point cost of CDS counterparty credit risk
This table reports the basis point cost to the protection buyer from the potential default of the protection seller. The central panel reports the costs
when CDS liabilities are not collateralized; the right panel reports the costs when CDS liabilities are collateralized. CP default cost denotes the cost of
events where only the counterparty defaults. Joint default cost denotes the cost of events where both the underlying firm and the counterparty default
together. Total cost denotes the sum of the costs of the two types of events. The parameter r denotes the default correlation between the underlying firm
and the counterparty. The parameters l and Z denote the basis point default intensities for the underlying firm and the counterparty, respectively.

Parameters Uncollateralized Collateralized

CP Joint CP Joint
default default Total default default Total
r l Z cost cost cost cost cost cost

0.02 100 100 0.69 0.89 1.58 0.07 0.88 0.95

300 0.98 1.02 2.00 0.10 1.00 1.10
500 1.31 1.21 2.52 0.14 1.18 1.32

300 100 0.68 1.16 1.84 0.08 1.14 1.22

300 1.16 1.50 2.66 0.13 1.48 1.61
500 1.62 1.56 3.18 0.18 1.53 1.71

0.06 100 100 0.68 2.50 3.18 0.07 2.43 2.50

300 0.99 3.35 4.34 0.10 3.27 3.37
500 1.32 3.52 4.83 0.13 3.43 3.56

300 100 0.69 3.24 3.93 0.08 3.18 3.26

300 1.16 4.15 5.31 0.13 4.07 4.20
500 1.63 4.48 6.11 0.18 4.38 4.56

0.10 100 100 0.71 3.82 4.53 0.07 3.73 3.80

300 1.02 5.17 6.19 0.10 5.04 5.14
500 1.35 5.99 7.34 0.13 5.87 6.00

300 100 0.69 5.11 5.80 0.08 5.00 5.08

300 1.15 6.89 8.04 0.13 6.75 6.88
500 1.65 7.70 9.35 0.18 7.55 7.73

also have implications for current proposals about two-day window. Let b denote the probability that the
restructuring derivatives markets. For example, since underlying firm defaults during the two-day window, but
market participants appear to price counterparty credit the counterparty does not. Let c denote the probability
risk as if it were only a relatively minor concern, this that the counterparty defaults during the two-day win-
suggests that attempts to mitigate counterparty credit dow, but the underlying firm does not. Finally, let d
risk through alternative approaches, such as the creation denote the probability that both the underlying firm and
of a central clearinghouse for CDS contracts, may not be as the counterparty default during the two-day window. It is
effective as might be anticipated. This implication paral- easily shown that the correlation r between I1 and I2 is
lels and complements the conclusions in the recent paper given by
by Duffie and Zhu (2009). dp1 p2
Corr½I1 ,I2  ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ðA:1Þ
Appendix A ðp1 p21 Þðp2 p22 Þ

Solving this expression for d gives,

To simulate correlated defaults in the model presented pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
in Section 6, we do the following. First, we define the d ¼ r p1 p2 ð1p1 Þð1p2 Þ þ p1 p2 : ðA:2Þ
discretization interval for the simulation to be two days;
Since the marginal probabilities of default are p1 and p2 ,
Dt ¼ 2=260 (there are approximately 260 trading days per
and since the total probability must equal one, we have
year). Let I1 denote a random binomial variable that takes
value one if the underlying firm defaults during the two- a ¼ 1bcd, ðA:3Þ
day window, and zero otherwise. Similarly, let I2 denote a
random binomial variable that takes value one if the b ¼ p1 d, ðA:4Þ
counterparty defaults during the two-day window, and
c ¼ p2 d: ðA:5Þ
zero otherwise. Let p1 ¼ lDt denote the probability that
the underlying firm defaults during the two-day window, Thus, given l, n, and r, we can solve for the probabilities a,
and p2 ¼ nDt denote the probability that the counterparty b, c, and d that define the joint default distribution for
defaults during the two-day window. Thus, with this each two-day window.
notation, E½I1  ¼ p1 and E½I2  ¼ p2 . Also, Var½I1  ¼ p1 p21 To simulate default outcomes for a five-year CDS
and Var½I2  ¼ p2 p22 . contract, we simulate a path for the default intensity
Now let a denote the probability that neither the processes l and n using the dynamics given in Eqs. (9) and
underlying firm nor the counterparty defaults during the (10). In doing this, we use two-day discretization
 N. Arora et al. / Journal of Financial Economics 103 (2012) 280–293 293

intervals. For each two-day window along the path, we Duffie, D., Huang, M., 1996. Swap rates and credit quality. Journal of
then apply the above algorithm to simulate the joint Finance 51, 921–949.
Duffie, D., Singleton, K.J., 1997. An econometric model of the term
default outcome (neither defaults, both default, etc.). We structure of interest rate swap yields. Journal of Finance 52,
then use the simulated joint default probabilities to define 1287–1323.
the cash flows along the path and evaluate the default Duffie, D., Singleton, K.J., 1999. Modeling term structures of defaultable
bonds. The Review of Financial Studies 12, 687–720.
costs. We repeat this process using 100,000 simulated Duffie, D., Zhu, H., 2009. Does a Central Clearing Counterparty Reduce
paths. Counterparty Risk? Unpublished Working Paper. Stanford
Finally, we note that there is a minor restriction on r University.
Gregory, J., 2010. Counterparty Credit Risk: The New Challenge for
that is needed to insure that b and c take positive values: Global Financial Markets. John Wiley & Sons, West Sussex, United
minðp1 ð1p2 Þ, p2 ð1p1 ÞÞ Kingdom.
ro pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ðA:6Þ Hull, J., White, A., 2001. Valuing credit default swaps II: modeling default
p1 p2 ð1p1 Þð1p2 Þ correlations. Journal of Derivatives 8 (Spring), 12–21.
ISDA Margin Survey, 2009. International Swaps and Derivatives Associa-
Whenever r exceeds this bound for a two-day window, tion. /http://www.isda.org/c_and_a/pdf/ISDA-Margin-Survey-2009.
pdfS.
we set r equal to this bound in solving for the joint
Jarrow, R., Yu, F., 2001. Counterparty risk and the pricing of defaultable
default probabilities for that two-day window. This securities. Journal of Finance 56, 1765–1799.
restriction, however, only affects a small fraction of the Johnson, N.L., Kotz, S., 1972. Distributions in Statistics: Continuous
100,000 simulated paths. Multivariate Distributions. John Wiley & Sons Inc., New York.
Kraft, H., Steffensen, M., 2007. Bankruptcy counterparty risk and con-
tagion. Review of Finance 11, 209–252.
References Lando, D., 1998. On Cox processes and credit risky securities. Review of
Derivatives Research 2, 99–120.
Longstaff, F.A., 2004. The flight-to-liquidity premium in U.S. Treasury
Blanchet-Scalliet, C., Patras, F., 2008. Counterparty Risk Valuation for bond prices. Journal of Business 77, 511–526.
CDS. Unpublished Working Paper. University of Lyon and University Longstaff, F.A., 2010. The subprime credit crisis and contagion in
of Nice. financial markets. Journal of Financial Economics 97, 436–450.
Bliss, R., Kaufman, G., 2006. Derivatives and systemic risk: netting, Longstaff, F.A., Mithal, S., Neis, E., 2005. Corporate yield spreads: default
collateral, and closeout. Journal of Financial Stability 2, 55–70. risk or liquidity? New evidence from the credit-default swap market.
Brigo, D., Pallavicini, A., 2006. Counterparty Risk and Contingent CDS Journal of Finance 60, 2213–2253.
Valuation under Correlation between Interest Rates and Default. Longstaff, F.A., Rajan, A., 2008. An empirical analysis of the pricing of
Unpublished Working Paper. Imperial College and Banca Leonardo. collateralized debt obligations. Journal of Finance 63, 509–563.
Buhlman, R., Lane, J., 2009. Counterparty risk: hard lessons learned. Segoviano, M., Singh, M., 2008. Counterparty Risk in the Over-the-
Practical Compliance & Risk Management (March–April), 35–42. Counter Derivatives Market. Working Paper 08/258. International
Cooper, I., Mello, A., 1991. The default risk on swaps. Journal of Finance Monetary Fund.
46, 597–620. Sorensen, E., Bollier, T., 1994. Pricing swap default risk. Financial
Duffie, D., Gârleanu, N., Pedersen, L.H., 2002. Securities lending, shorting, Analysts Journal 50 (May–June), 23–33.
and pricing. Journal of Financial Economics 66, 307–339. White, H., 1980. A heteroskedasticity-consistent covariance matrix
Duffie, D., Gârleanu, N., Pedersen, L.H., 2005. Over-the-counter markets. estimator and a direct test for heteroskedasticity. Econometrica 48,
Econometrica 73, 1815–1847. 817–838.
Duffie, D., Gârleanu, N., Pedersen, L.H., 2008. Valuation in over-the- Yu, F., 2007. Correlated defaults in intensity-based models. Mathema-
counter markets. Review of Financial Studies 20, 1865–1900. tical Finance 17, 155–173.