Pricing Financial Derivatives Subject to

Multilateral Credit Risk and Collateralization

Tim Xiao1

ABSTRACT

This article presents a new model for valuing financial contracts subject to credit risk and

collateralization. Examples include the valuation of a credit default swap (CDS) contract that is affected

by the trilateral credit risk of the buyer, seller and reference entity. We show that default dependency has

a significant impact on asset pricing. In fact, correlated default risk is one of the most pervasive threats in

financial markets. We also show that a fully collateralized CDS is not equivalent to a risk-free one. In

other words, full collateralization cannot eliminate counterparty risk completely in the CDS market.

Key Words: asset pricing; credit risk modeling; collateralization; comvariance; comrelation; correlation,

CDS.

1
Risk Models, Capital Markets, BMO, First Canadian Place, 51th Floor, 100 King West, Toronto, ON
M5X 1H3. Email: [email protected] Url: https://finpricing.com/

Electronic copy available at: https://ssrn.com/abstract=3410640

 1 Introduction

There are two primary types of models that attempt to describe default processes in the literature:

structural models and reduced-form (or intensity) models. Many practitioners in the credit trading arena

have tended to gravitate toward the reduced-from models given their mathematical tractability.

Central to the reduced-form models is the assumption that multiple defaults are independent

conditional on the state of the economy. In reality, however, the default of one party might affect the

default probabilities of other parties. Collin-Dufresne et al. (2003) and Zhang and Jorion (2007) find that

a major credit event at one firm is associated with significant increases in the credit spreads of other

firms. Giesecke (2004), Das et al. (2006), and Lando and Nielsen (2010) find that a defaulting firm can

weaken the firms in its network of business links. These findings have important implications for the

management of credit risk portfolios, where default relationships need to be explicitly modeled.

The main drawback of the conditionally independent assumption or the reduced-form models is

that the range of default correlations that can be achieved is typically too low when compared with

empirical default correlations (see Das et al. (2007)). The responses to correct this weakness can be

generally classified into two categories: endogenous default relationship approaches and exogenous

default relationship approaches.

The endogenous approaches include the contagion (or infectious) models and frailty models. The

frailty models (see Duffie et al. (2009), Koopman et al. (2011), etc) describe default clustering based on

some unobservable explanatory variables. In variations of contagion or infectious type models (see Davis

and Lo (2001), Jarrow and Yu (2001), etc.), the assumption of conditional independence is relaxed and

default intensities are made to depend on default events of other entities. Contagion and frailty models fill

an important gap but at the cost of analytic tractability. They can be especially difficult to implement for

large portfolios.

The exogenous approaches (see Li (2000), Laurent and Gregory (2005), Hull and White (2004),

Brigo et al. (2011), etc) attempt to link marginal default probability distributions to the joint default

Electronic copy available at: https://ssrn.com/abstract=3410640

probability distribution through some external functions. Due to their simplicity in use, practitioners lean

toward the exogenous ones.

Given a default model, one can value a risky derivative contract and compute credit value

adjustment (CVA) that is a relatively new area of financial derivative modeling and trading. CVA is the

expected loss arising from the default of a counterparty (see Brigo and Capponi (2008), Lipton and Sepp

(2009), Pykhtin and Zhu (2006), Gregory (2009), Bielecki et al (2013) and Crepey (2015), etc.)

Collateralization as one of the primary credit risk mitigation techniques becomes increasingly

important and widespread in derivatives transactions. According the ISDA (2013), 73.7% of all OTC

derivatives trades (cleared ad non-cleared) are subject to collateral agreements. For large firms, the figure

is 80.7%. On an asset class basis, 83.0% of all CDS transactions and 79.2% of all fixed income

transactions are collateralized. For large firms, the figures are 96.3% and 89.4%, respectively. Previous

studies on collateralization include Johannes and Sundaresan (2007), Fuijii and Takahahsi (2012),

Piterbarg (2010), Bielecki, et al (2013) and Hull and White (2014), etc.

This paper presents a new framework for valuing defaultable financial contracts with or without

collateral arrangements. The framework characterizes default dependencies exogenously, and models

collateral processes directly based on the fundamental principals of collateral agreements. For brevity we

focus on CDS contracts, but many of the points we make are equally applicable to other derivatives. CDS

has trilateral credit risk, where three parties – buyer, seller and reference entity – are defaultable.

In general, a CDS contract is used to transfer the credit risk of a reference entity from one party to

another. The risk circularity that transfers one type of risk (reference credit risk) into another

(counterparty credit risk) within the CDS market is a concern for financial stability. Some people claim

that the CDS market has increased financial contagion or even propose an outright ban on these

instruments.

The standard CDS pricing model in the market assumes that there is no counterparty risk.

Although this oversimplified model may be accepted in normal market conditions, its reliability in times

of distress has recently been questioned. In fact, counterparty risk has become one of the most dangerous

Electronic copy available at: https://ssrn.com/abstract=3410640

threats to the CDS market. For some time now it has been realized that, in order to value a CDS properly,

counterparty effects have to be taken into account (see ECB (2009)).

We bring the concept of comvariance into the area of credit risk modeling to capture the

statistical relationship among three or more random variables. Comvariance was first introduced to

economics by Deardorff (1982), who used this measurement to correlate three factors in international

trading. Furthermore, we define a new statistics, comrelation, as a scaled version of comvariance.

Accounting for default correlations and comrelations becomes important in determining CDS premia,

especially during the credit crisis. Our analysis shows that the effect of default dependencies on a CDS

premium from large to small accordingly is i) the default correlation between the protection seller and the

reference entity, ii) the default comrelation, iii) the default correlation between the protection buyer and

the reference entity, and iv) the default correlation between the protection buyer and the protection seller.

In particular, we find that the default comvariance/comrelation has substantial effects on the asset pricing

and risk management, which have never been documented.

There is a significant increase in the use of collateral for CDS after the recent financial crises.

Many people believe that, if a CDS is fully collateralized, there is no risk of failure to pay. Collateral

posting regimes are originally designed and utilized for bilateral risk products, e.g., interest rate swap

(IRS), but there are many reasons to be concerned about the success of collateral posting in offsetting the

risk of CDS contracts. First, the value of CDS contracts tends to move very suddenly with big jumps,

whereas the price movements of IRS contracts are far smoother and less volatile than CDS prices. Second,

CDS spreads can widen very rapidly. Third, CDS contracts have many more risk factors than IRS

contracts. In fact, our model shows that full collateralization cannot eliminate counterparty risk

completely for a CDS contract.

The rest of this paper is organized as follows: Pricing multilateral defaultable financial contract is

elaborated on in Section 2; numerical results are provided in Section 3; the conclusions are presented in

Section 4. All proofs and some detailed derivations are contained in the appendices.

Electronic copy available at: https://ssrn.com/abstract=3410640

 2 Pricing Financial Contracts Subject to Multilateral Credit Risk

We consider a filtered probability space (  , F , Ft t 0 , P ) satisfying the usual conditions,

where  denotes a sample space, F denotes a  -algebra, P denotes a probability measure, and

Ft t 0 denotes a filtration.

In the reduced-form approach, the stopping (or default) time  i of firm i is modeled as a Cox

arrival process (also known as a doubly stochastic Poisson process) whose first jump occurs at default and

is defined by,

 t
 i  inf t :  hi ( s, Z s )ds  H i
0
 (1)

where hi (t ) or hi (t , Z t ) denotes the stochastic hazard rate or arrival intensity dependent on an exogenous

common state Z t , and H i is a unit exponential random variable independent of Z t .

It is well-known that the survival probability from time t to s in this framework is defined by

pi (t , s) : Pi (  s |   t , Z t )  exp   hi (u )du 
s
(2a)
 t 

The default probability for the period (t, s) in this framework is given by

qi (t , s) : Pi (  s |   t , Z t )  1  pi (t , s)  1  exp   hi (u )du 
s
(2b)
 t 

There is ample evidence that corporate defaults are correlated. The default of a firm’s

counterparty might affect its own default probability. Thus, default correlation/dependence occurs due to

the counterparty relations.

The interest in the financial industry for the modeling and pricing of multilateral defaultable

instruments arises mainly in two respects: in the management of credit risk at a portfolio level and in the

valuation of credit derivatives. Central to the valuation and risk management of credit derivatives and

risky portfolios is the problem of default relationship.

Let us discuss a three-party case first. A CDS is a good example of a trilateral defaultable

instrument where the three parties are counterparties A, B and reference entity C. In a standard CDS

Electronic copy available at: https://ssrn.com/abstract=3410640

contract one party purchases credit protection from another party, to cover the loss of the face value of a

reference entity following a credit event. The protection buyer makes periodic payments to the seller until

the maturity date or until a credit event occurs. A credit event usually requires a final accrual payment by

the buyer and a loss protection payment by the protection seller. The protection payment is equal to the

difference between par and the price of the cheapest to deliver (CTD) asset of the reference entity on the

face value of the protection.

A CDS is normally used to transfer the credit risk of a reference entity between two

counterparties. The contract reduces the credit risk of the reference entity but gives rise to another form of

risk: counterparty risk. Since the dealers are highly concentrated within a small group, any of them may

be too big to fail. The interconnected nature, with dealers being tied to each other through chains of OTC

derivatives, results in increased contagion risk. Due to its concentration and interconnectedness, the CDS

market seems to pose a systemic risk to financial market stability. In fact, the CDS is blamed for playing a

pivotal role in the collapse of Lehman Brothers and the disintegration of AIG.

For years, a widespread practice in the market has been to mark CDS to market without taking the

counterparty risk into account. The realization that even the most prestigious investment banks could go

bankrupt has shattered the foundation of the practice. It is wiser to face frankly the real complexities of

pricing a CDS than to indulge in simplifications that have proved treacherous. For some time now it has

been realized that, in order to value a CDS properly, counterparty effects have to be taken into account.

Let A denote the protection buyer, B denote the protection seller and C denote the reference

entity. The binomial default rule considers only two possible states: default or survival. Therefore, the

default indicator Y j for firm j (j = A or B or C) follows a Bernoulli distribution, which takes value 1 with

default probability q j , and value 0 with survival probability p j . The marginal default distributions can be

determined by the reduced-form models. The joint distributions of a multivariate Bernoulli variable can

be easily obtained via the marginal distributions by introducing extra correlations. The joint probability

representations of a trivariate Bernoulli distribution (see Teugels (1990)) are given by

Electronic copy available at: https://ssrn.com/abstract=3410640

 p 000 : P(Y A  0, YB  0, YC  0)  p A p B pC  pC  AB  p B AC  p A BC   ABC (3a)

p100 : P(Y A  1, YB  0, YC  0)  q A p B pC  pC  AB  p B AC  q A BC   ABC (3b)

p 010 : P(Y A  0, YB  1, YC  0)  p A q B pC  pC  AB  q B AC  p A BC   ABC (3c)

p 001 : P(Y A  0, YB  0, YC  1)  p A p B qC  qC  AB  p B AC  p A BC   ABC (3d)

p110 : P(Y A  1, YB  1, YC  0)  q A q B pC  pC  AB  q B AC  q A BC   ABC (3e)

p101 : P(Y A  1, YB  0, YC  1)  q A p B qC  qC  AB  p B AC  q A BC   ABC (3f)

p011 : P(Y A  0, YB  1, YC  1)  p A q B qC  qC  AB  q B AC  p A BC   ABC (3g)

p111 : P(Y A  1, YB  1, YC  1)  q A q B qC  qC  AB  q B AC  q A BC   ABC (3h)

where

 ABC : E( YA  q A )( YB  q B )( YC  qC ) (3i)

Equation (3) tells us that the joint probability distribution of three defaultable parties depends not

only on the bivariate statistical relationships of all pair-wise combinations (e.g.,  ij ) but also on the

trivariate statistical relationship (e.g.,  ABC ).  ABC was first defined by Deardorff (1982) as comvariance,

who use it to correlate three random variables that are the value of commodity net imports/exports, factor

intensity, and factor abundance in international trading.

We introduce the concept of comvariance into credit risk modeling arena to exploit any statistical

relationship among multiple random variables. Furthermore, we define a new statistic, comrelation, as a

scaled version of comvariance (just like correlation is a scaled version of covariance) as follows:

Definition 1: For three random variables X A , X B , and X C , let  A ,  B , and C denote the means of

X A , X B , and X C . The comrelation of X A , X B , and X C is defined by

E( X A   A )( X B   B )( X C   C )
 ABC  (4)
3 3 3
3 E X A   A  E X B   B  E X C  C

According to the Holder inequality, we have

Electronic copy available at: https://ssrn.com/abstract=3410640

 E ( X A   A )( X B   B )( X C   C )   E ( X A   A )( X B   B )( X C   C )
(5)
 3 E X A   A  E X B   B  E X C  C
3 3 3

Obviously, the comrelation is in the range of [-1, 1]. Given the comrelation, Equation (3i) can be

rewritten as

 ABC : E (YA  q A )(YB  qB )( X C  qC )    ABC 3 E YA  q A  E YB  qB  E YC  qC

3 3 3

(6)
  ABC 3 p Aq A ( p A2  q A2 ) pB qB ( pB2  qB2 ) pC qC ( p C2  qC2 )

3
where E(Y j )  q j and E Y j  q j  p j q j ( p 2j  q 2j ) , j=A, B, or C.

If we have a series of n measurements of X A , X B , and X C written as x Ai , xBi and xCi where i =

1,2,…,n, the sample comrelation coefficient can be obtained as:

i 1 ( x Ai   A )(xBi   B )(xCi  C )
n
 ABC  (7)
i 1 x Ai   A  i 1 xBi   B  i 1 xCi  C
3 n 3 n 3 n 3

More generally, we define the comrelation in the context of n random variables as

Definition 2: For n random variables X 1 , X 2 ,…, X n , let  i denote the mean of X i where i=1,..,n. The

comrelation of X 1 , X 2 ,…, X n is defined as

E( X 1  1 )( X 2   2 )    ( X n   n )
 12...n  (8)
n n n
n E X 1  1  E X 2   2    E X n   n

Correlation is just a specific case of comrelation where n = 2. Again, the comrelation  12...n is in

the range of [-1, 1] according to the Holder inequality.

2.1 Risky valuation without collateralization

Recovery assumptions are important for pricing credit derivatives. If the reference entity under a

CDS contract defaults, the best assumption, as pointed out by J. P. Morgan (1999), is that the recovered

value equals the recovery rate times the face value plus accrued interest2. In other words, the recovery of

2
In the market, there is an average accrual premium assumption, i.e., the average accrued premium is half
the full premium due to be paid at the end of the premium.

Electronic copy available at: https://ssrn.com/abstract=3410640

par value assumption is a better fit upon the default of the reference entity, whereas the recovery of

market value assumption is a more suitable choice in the event of a counterparty default3.

Let valuation date be t. Suppose that a CDS has m scheduled payments represented as

X i  sN (Ti 1 , Ti ) with payment dates T1 ,…, Tm where i=1,,,,m,  (Ti 1 , Ti ) denotes the accrual factor for

period (Ti 1 , Ti ) , N denotes the notional/principal, and s denotes the CDS premium. Party A pays the

premium/fee to party B if reference entity C does not default. In return, party B agrees to pay the

protection amount to party A if reference entity C defaults before the maturity. We have the following

proposition.

Proposition 1: The value of the CDS is given by

V (t )   i 1 E
m
 i 1
j 0
 
O (T j , T j 1 ) X i Ft   i 1 E
m
 i 2
j 0

O(T j , T j 1 ) (Ti 1 , Ti ) R(Ti 1 , Ti ) Ft  (9a)

where t  T0 and

O(T j , T j 1 )  1V (T j 1 ) X j 1 0  B (T j , T j 1 )  1V (T j 1 ) X j 1 0  A (T j , T j 1 ) (9b)

 A (T j , T j 1 )  p A (T j , T j 1 ) p B (T j , T j 1 ) pC (T j , T j 1 )  q A (T j , T j 1 ) p B (T j , T j 1 ) pC (T j , T j 1 ) A (T j 1 )
 p A (T j , T j 1 )q B (T j , T j 1 ) pC (T j , T j 1 ) A (T j 1 )  q A (T j , T j 1 )q B (T j , T j 1 ) pC (T j , T j 1 ) AB (T j 1 )
 pC (T j , T j 1 ) AB (T j , T j 1 )1   A (T j 1 )   A (T j 1 )   AB (T j 1 )  (9c)

  AC (T j , T j 1 ) p B (T j , T j 1 )1   A (T j 1 )   q B (T j , T j 1 ) A (T j 1 )   AB (T j 1 )  
  BC (T j , T j 1 ) p (T , T )1   (T )   q (T , T ) (T
A j j 1 A j 1 A j j 1 A j 1 )   AB (T j 1 ) 
  ABC (T j , T j 1 ) 1   (T )   (T )   (T ) D(T , T
A j 1 AB j 1 A j 1 j j 1 )

 B (T j , T j 1 )  p A (T j , T j 1 ) p B (TJ , T j 1 ) pC (T j , T j 1 )  q A (T j , T j 1 ) p B (T j , T j 1 ) pC (T j , T j 1 ) B (T j 1 )
 p A (T j , T j 1 )q B (T j , T j 1 ) pC (T j , T j 1 ) B (T j 1 )  q A (T j , T j 1 )q B (T j , T j 1 ) pC (T j , T j 1 ) AB (T j 1 )
 pC (T j , T j 1 ) AB (T j , T j 1 )1   B (T j 1 )   B (T j 1 )   AB (T j 1 )  (9d)

  AC (T j , T j 1 ) p B (T j , T j 1 )1   B (T j 1 )   q B (T j , T j 1 ) B (T j 1 )   AB (T j 1 )  
  BC (T j , T j 1 ) p A (T j , T j 1 )1   B (T j )   q A (T j , T j 1 ) B (T j 1 )   AB (T j 1 )  
  (T j , T j 1 ) 1   B (T j 1 )   AB (T j 1 )   B (T j 1 ) D(T j , T j 1 )

3
Three different recovery models exist in the literature. The default payoff is either i) a fraction of par
(Madan and Unal (1998)), ii) a fraction of an equivalent default-free bond (Jarrow and Turnbull (1995)),
or iii) a fraction of market value (Duffie and Singleton (1999)).

Electronic copy available at: https://ssrn.com/abstract=3410640

 (T j , T j 1 )  p A (T j , T j 1 ) p B (T j , T j 1 )qC (T j , T j 1 )  q A (T j , T j 1 ) p B (T j , T j 1 )qC (T j , T j 1 ) B (T j 1 )
 p A (T j , T j 1 )q B (T j , T j 1 )qC (T j , T j 1 ) B (T j 1 )  q A (T j , T j 1 )q B (T j , T j 1 )qC (T j , T j 1 ) AB (T j 1 )
 qC (T j , T j 1 ) AB (T j , T j 1 )1   B (T j 1 )   B (T j 1 )   AB (T j 1 )  (9e)

  AC (T j , T j 1 ) p B (T j , T j 1 )1   B (T j 1 )   q B (T j , T j 1 ) B (T j 1 )   AB (T j 1 )  
  BC (T j , T j 1 ) p (T , T )1  
A j j 1 B (T j 1 )   q A (T j , T j 1 ) B (T j 1 )   AB (T j 1 ) 
  ABC (T j , T j 1 )1   (T )  
B j 1 AB (T j 1 )   B (T j 1 ) D(T j , T j 1 )

where R(T j , T j 1 )  N 1  C (T j 1 )   (T j , T j 1 ) ,  (T j , T j 1 )  sN (TS , T ) / 2 , and X i  sN (T j , T j 1 ) .

Proof: See the Appendix.

We may think of O(t , T ) as the risk-adjusted discount factor for the premium and (t , T ) as the

risk-adjusted discount factor for the default payment. Proposition 1 says that the pricing process of a

multiple-payment instrument has a backward nature since there is no way of knowing which risk-adjusted

discounting rate should be used without knowledge of the future value. Only on the maturity date, the

value of an instrument and the decision strategy are clear. Therefore, the evaluation must be done in a

backward fashion, working from the final payment date towards the present. This type of valuation

process is referred to as backward induction.

Proposition 1 provides a general form for pricing a CDS. Applying it to a particular situation in

which we assume that counterparties A and B are default-free, i.e., p j  1 , q j  0 ,  kl  0 , and  ABC  0 ,

where j=A or B and k, l=A, B, or C, we derive the following corollary.

Corollary 1: If counterparties A and B are default-free, the value of the CDS is given by

V (t )   i 1 E
m
 i 1
j 0
 
O(T j , T j 1 ) X i Ft   i 1 E
m
 i 2
j 0

O(T j , T j 1 ) (Ti 1 , Ti ) R(Ti 1 , Ti ) Ft  (10)
  i 1 E D(t , Ti ) pC (t , Ti ) X i Ft   i 1 E D(t , Ti ) pC (t , Ti 1 )qC (Ti 1 , Ti ) R(Ti 1 , Ti ) Ft 
m m

where O(Ti 1 , Ti )  D(Ti 1 , Ti ) pC (Ti 1 , Ti ) ; (Ti 1 , Ti )  D(Ti 1 , Ti )qC (Ti 1 , Ti ) .

The proof of this corollary becomes straightforward according to Proposition 1 by setting kl =0,

i 1 i 1
 AB  0 ,  ABC  0 , p j  1 , q j  0 , pC (t ,Ti )   g 0 p(Tg ,Tg 1) , and D(t , Ti )   g 0 D(Tg , Tg 1 ) .

If we further assume that the discount factor and the default probability of the reference entity are

uncorrelated and the recovery rate C is constant, we have

Electronic copy available at: https://ssrn.com/abstract=3410640

Corollary 2: Assume that i) counterparties A and B are default-free, ii) the discount factor and the default

probability of the reference entity are uncorrelated; iii) the recovery rate C is constant; the value of the

CDS is given by

V (t )   i 1 P(t , Ti ) pC (t , Ti 1 )qC (Ti 1, Ti )N 1  C    (Ti 1, Ti )   i 1 P(t , Ti ) pc (t , Ti ) sN  (Ti 1, Ti )
m m
(11)

where P(t,Ti )  ED(t ,Ti ) Ft  denotes the bond price, pc (t , Ti )  Epc (t , Ti ) Ft  , qc (t , Ti )  1  pc (t , Ti ) ,

p(t , Ti 1 )q (Ti 1 , Ti )  p(t , Ti 1 )  p(t , Ti ) .

This corollary is easily proved according to Corollary 1 by setting EXY Ft   EX Ft EY Ft 

when X and Y are uncorrelated. Corollary 2 is the formula for pricing CDS in the market.

Our methodology can be extended to the cases where the number of parties n  4 . A generating

function for the (probability) joint distribution (see details in Teugels (1990)) of n-variate Bernoulli can

be expressed as

p  1  pn 1  1  p  1 ( n )
p (n)   n      1  (12)
 qn 1   qn 1 1   q1 1 

 
where  denotes the Kronecker product; p ( n)  pk( n) and  ( n)   k( n ) are vectors containing 2n  

components: pk(n)  pk1 ,k2 ,...,kn , k  1  i 1 ki 2i 1 , ki  0,1;  k(n)   k1 ,k2 ,...,kn  E i 1 Yi  qi ki .
n n

2.2 Risky valuation with collateralization

Collateralization is the most important and widely used technique in practice to mitigate credit

risk. The posting of collateral is regulated by the Credit Support Annex (CSA) that specifies a variety of

terms including the threshold, the independent amount, and the minimum transfer amount (MTA), etc.

The threshold is the unsecured credit exposure that a party is willing to bear. The minimum transfer

amount is the smallest amount of collateral that can be transferred. The independent amount plays the

same role as the initial margin (or haircuts).

In a typical collateral procedure, a financial instrument is periodically marked-to-market and the

collateral is adjusted to reflect changes in value. The collateral is called as soon as the mark-to-market

10

Electronic copy available at: https://ssrn.com/abstract=3410640

(MTM) value rises above the given collateral threshold, or more precisely, above the threshold amount

plus the minimum transfer amount. Thus, the collateral amount posted at time t is given by

V (t )  H (t ) if V (t )  H (t )
C (t )   (13)
 0 otherwise

where H (t ) is the collateral threshold. In particular, H (t )  0 corresponds to full-collateralization4; H  0

represents partial/under-collateralization; and H  0 is associated with over-collateralization. Full

collateralization becomes increasingly popular at the transaction level. In this paper, we focus on full

collateralization only, i.e., C (t )  V (t ) .

The main role of collateral should be viewed as an improved recovery in the event of a

counterparty default. According to Bankruptcy law, if there has been no default, the collateral is returned

to the collateral giver by the collateral taker. If a default occurs, the collateral taker possesses the

collateral. In other words, collateral does not affect the survival payment; instead, it takes effect on the

default payment only.

According to the ISDA (2013), almost all CDSs are fully collateralized. Many people believe that

full collateralization can eliminate counterparty risk completely for CDS.

Collateral posting regimes are originally designed and utilized for bilateral risk products, e.g.,

IRS, but there are many reasons to be concerned about the success of collateral posting in offsetting the

risks of CDS contracts. First, the values of CDS contracts tend to move very suddenly with big jumps,

whereas the price movements of IRS contracts are far smoother and less volatile than CDS prices.

Second, CDS spreads can widen very rapidly. The amount of collateral that one party is required to

provide at short notice may, in some cases, be close to the notional amount of the CDS and may therefore

4
There are three types of collateralization: Full-collateralization is a process where the posting of
collateral is equal to the current MTM value. Partial/under-collateralization is a process where the posting
of collateral is less than the current MTM value. Over-collateralization is a process where the posting of
collateral is greater than the current MTM value.

11

Electronic copy available at: https://ssrn.com/abstract=3410640

exceed that party’s short-term liquidity capacity, thereby triggering a liquidity crisis. Third, CDS

contracts have many more risk factors than IRS contracts.

We assume that a CDS is fully collateralized, i.e., the posting of collateral is equal to the amount

of the current MTM value: C(t )  V (t ) . For a discrete one-period (t, u) economy, there are several

possible states at time u: i) A, B, and C survive with probability p000 . The instrument value is equal to the

market value V (u) ; ii) A and B survive, but C defaults with probability p001 . The instrument value is the

default payment R(u) ; iii) For the remaining cases, either or both counterparties A and B default. The

instrument value is the future value of the collateral V (t ) / D(t , u) (Here we consider the time value of

money only). The value of the collateralized instrument at time t is the discounted expectation of all the

payoffs and is given by

V (t )  ED(t , u ) p000 (t , u )V (u )  p001 (t , u ) R(u )

  p100 (t , u )  p010 (t , u )  p110 (t , u )  p101 (t , u )  p011 (t , u )  p111 (t , u )  V (t ) / D(t , u ) Ft  (14a)
 ED(t , u ) p000 (t , u )V ( s)  p001 (t , u ) R(u )   1  p000 (t , u )  p001 (t , u ) V (t ) Ft 

or

 
E  p A (t , u ) p B (t , u )   AB (t , u )  Ft V (t )
 ED(t , u ) pC (t , u )V (u )  qC (t , u ) R(u )   p A (t , u ) p B (t , u )   AB (t , u )  (14b)
 D(t , u )V (u )  R(u )  p B (t , u ) AC (t , u )  p A (t , u ) BC (t , u )   ABC (t , u )  Ft 

If we assume that  p A (t, u) p B (t, u)   AB (t, u) and D(t , u) pC (t , u)V (u)  qC (t , u) R(u) are

uncorrelated, we have

V (t )  V F (t )   ABC (t , u) / AB (t , u) (15a)

where

V F (t )  ED(t , u ) pC (t , u )V (u )  qC (t , u ) R(u ) Ft  (15b)

 AB (t , u)  E p A (t , u) p B (t , u)   AB (t , u) Ft  (15c)

 ABC (t, u)  ED(t, u)  p B (t, u) AC (t, u)  p A (t, u) BC (t , u)   ABC (t, u)V (u)  R(u) Ft  (15d)

12

Electronic copy available at: https://ssrn.com/abstract=3410640

 The first term V F (t ) in equation (15) is the counterparty-risk-free value of the CDS and the

second term is the exposure left over under full collateralization, which can be substantial.

Proposition 2: If a CDS is fully collateralized, the risky value of the CDS is NOT equal to the

counterparty-risk-free value, as shown in equation (15).

Proposition 2 or equation (15) provides a theoretical explanation for the failure of full

collateralization in the CDS market. It tells us that under full collateralization the risky value is in general

not equal to the counterparty-risk-free value except in one of the following situations: i) the market value

is equal to the default payment, i.e., V (u)  R(u) ; ii) firms A, B, and C have independent credit risks, i.e.,

 ij =0 and  ABC  0 ; or iii) the following relationship holds p B AC  p A BC   ABC .

3 Numerical Results

Our goal in this section is to study the quantitative relationship between CDS premia and the

credit quality of counterparties and reference entities, including the default correlations and comrelations.

In our study, we choose a new 5-year CDS with a quarterly payment frequency. Two

counterparties are denoted as A and B. Counterparty A buys a protection from counterparty B. All

calculations are from the perspective of party A. By definition, a breakeven CDS spread is a premium that

makes the market value of a given CDS at inception zero.

The current (spot) market data are shown in Table 1 provided by FinPricing (2013). Assume that

the reference entity C has an “A+200bps” credit quality throughout this subsection. The 5-year

counterparty-risk-free CDS premium is 0.027 (equals the 5-year ‘A’ rated CDS spread in Table 1 plus

200 basis points).

Since the payoffs of a CDS are mainly determined by credit events, we need to characterize the

evolution of the hazard rates. Here we choose the Cox-Ingersoll-Ross (CIR) model. The CIR process has

been widely used in the literature of credit risk and is given by

dht  a(b  ht )dt   ht dWt (16)

13

Electronic copy available at: https://ssrn.com/abstract=3410640

where a denotes the mean reversion speed, b denotes the long-term mean, and  denotes the volatility.

Table 1: Current/spot market data

This table displays the current (spot) market data used for all calculations in this paper, including the term

structure of continuously compounded interest rates, the term structure of A-rated breakeven CDS

spreads, and the curve of at-the-money caplet volatilities.

Term (days) 31 91 182 365 548 730 1095 1825 2555 3650 5475
Interest Rate 0.0028 0.0027 0.0029 0.0043 0.0071 0.0102 0.016 0.0249 0.0306 0.0355 0.0405
Credit Spread 0.0042 0.0042 0.0042 0.0045 0.0049 0.0052 0.0058 0.007 0.0079 0.0091 0.0106
Caplet Volatility 0.3267 0.331 0.3376 0.3509 0.3641 0.3773 0.308 0.2473 0.2141 0.1678 0.1634

Table 2: Risk-neutral parameters for CIR model

This table presents the risk-neutral parameters that are calibrated to the current market shown in Table 1.

‘A+100bps’ represents a ‘100 basis points’ parallel shift in the A-rated CDS spreads.

Credit Quality A A+100bps A+200bps A+300bps

Long-Term Mean a 0.035 0.056 0.077 0.099
Mean Reverting Speed b 0.14 0.18 0.25 0.36
Volatility  0.022 0.028 0.039 0.056

The calibrated parameters are shown in table 2. We assume that interest rates are deterministic

and select the regression-based Monte-Carlo simulation (see Longstaff and Schwartz (2001)) to perform

risky valuation.

We first assume that counterparties A, B, and reference entity C have independent default risks,

i.e.,  AB   AC   BC   AB   ABC  0 , and examine the following cases: i) B is risk-free and A is risky;

and ii) A is risk-free and B is risky. We simulate the hazard rates using the CIR model and then determine

the appropriate discount factors according to Proposition 1. Finally we calculate the prices via the

regression-based Monte-Carlo method. The results are shown in Table 3 and 4.

14

Electronic copy available at: https://ssrn.com/abstract=3410640

 Table 3: Impact of the credit quality of the protection buyer on CDS premia

This table shows how the CDS premium increases as the credit quality of party A decreases. The 1st data

column represents the counterparty-risk-free results. For the remaining columns, we assume that party B

is risk-free and party A is risky. ‘A+100bps’ represents a ‘100 basis points’ parallel shift in the A-rated

CDS spreads. The results in the row ‘Difference from Risk-Free’ = current CDS premium – counterparty-

risk-free CDS premium.

Party A - A A+100bps A+200bps A+300bps

Credit Quality
Party B - - - - -
CDS premium 0.027 0.02703 0.02708 0.02713 0.02717
Difference from Risk-Free 0 0.003% 0.008% 0.013% 0.017%

Table 4: Impact of the credit quality of the protection seller on CDS premia

This table shows the decrease in the CDS premium with the credit quality of party B. The 1st data column

represents the counterparty-risk-free results. For the remaining columns, we assume that party A is risk-

free and party B is risky. ‘A+100bps’ represents a ‘100 basis points’ parallel shift in the A-rated CDS

spreads. The results in the row ‘Difference from Risk-Free’ = current CDS premium – counterparty-risk-

free CDS premium.

Party A - - - - -
Credit Quality
Party B - A A+100bps A+200bps A+300bps
CDS premium 0.027 0.02695 0.02687 0.0268 0.02672
Difference from Risk-Free 0.00% -0.005% -0.013% -0.020% -0.028%

From table 3 and 4, we find that a credit spread of about 100 basis points maps into a CDS

premium of about 0.4 basis points for counterparty A and about -0.7 basis points for counterparty B. The

credit impact on the CDS premia is approximately linear. As would be expected, i) the dealer’s credit

quality has a larger impact on CDS premia than the investor’s credit quality; ii) the higher the investor’s

15

Electronic copy available at: https://ssrn.com/abstract=3410640

credit risk, the higher the premium that the dealer charges; iii) the higher the dealer’s credit risk, the lower

the premium that the dealer asks. Without considering default correlations and comrelations, we find that,

in general, the impact of counterparty risk on CDS premia is relatively small. This is in line with the

empirical findings of Arora, Gandhi, and Longstaff (2009).

Figure 1: Impact of default correlations and comrelation on CDS premia

Each curve in this figure illustrates how CDS premium changes as default correlations and comrelation

move from -1 to 1. For instance, the curve ‘cor_BC’ represents the sensitivity of the CDS premium to

changes in the correlation  BC when  AB   AC   ABC  0 .

Impact of Default Correlations and Comrelation on CDS Premia

0.034

0.032

0.03
CDS premia

cor_AB
0.028 cor_AC

0.026 cor_BC
comr_ABC
0.024

0.022

0.02
1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1

correlation or comrelation

Next, we study the sensitivity of CDS premia to changes in the joint credit quality of associated

parties. Sensitivity analysis is a very popular way in finance to find out how the value and risk of an

instrument/portfolio changes if risk factors change. One of the simplest and most common approaches

involves changing one factor at a time to see what effect this produces on the output. We are going to

examine the impacts of the default correlations  AB ,  AC ,  BC , and the comrelation  ABC separately.

16

Electronic copy available at: https://ssrn.com/abstract=3410640

Assume that party A has an ‘A+100bps’ credit quality and party B has an ‘A’ credit quality. The 5-year

risky CDS premium is calculated as 0.02703.

Assume  AB =0.5. The impact diagrams of the default correlations and comrelation are shown in

Figure 1. From this graph, we can draw the following conclusions: First, the CDS premium and the

default correlations/comrelation have a negative relation. Intuitively, a protection seller who is positively

correlated with the reference entity (a wrong way risk) should charge a lower premium for selling credit

protection. Next, the impacts of the default correlations and comrelation are approximately linear. Finally,

the sensitivity slopes of the CDS premium to the default correlations and comrelation are -0.06 to  AB ; -

0.09 to  AC ; -53 to  BC ; and -14 to  ABC . Slope measures the rate of change in the premium as a result

of a change in the default dependence. For instance, a slope of -53 implies that the CDS premium would

have to decrease by 53 basis points when a default correlation/comrelation changes from 0 to 1.

As the absolute value of the slope increases, so does the sensitivity. The results illustrate that  BC

has the largest effect on CDS premia. The second biggest one is  ABC . The impacts of  AB and  AC are

very small. In particular, the effect of the comrelation is substantial and has never been studies before. A

natural intuition to have on CDS is that the party buying default protection should worry about the default

correlations and comrelation.

4 Conclusion

This article presents a new valuation framework for pricing financial instruments subject to credit

risk. In particular, we focus on modeling default relationships.

To capture the default relationships among more than two defaultable entities, we introduce a

new statistic: comrelation, an analogue to correlation for multiple variables, to exploit any multivariate

statistical relationship. Our research shows that accounting for default correlations and comrelations

becomes important, especially under market stress. The existing valuation models in the credit derivatives

17

Electronic copy available at: https://ssrn.com/abstract=3410640

market, which take into account only pair-wise default correlations, may underestimate credit risk and

may be inappropriate.

We study the sensitivity of the price of a defaultable instrument to changes in the joint credit

quality of the parties. For instance, our analysis shows that the effect of default dependence on CDS

premia from large to small is the correlation between the protection seller and the reference entity, the

comrelation, the correlation between the protection buyer and the reference entity, and the correlation

between the protection buyer and the protection seller.

The model shows that a fully collateralized CDS is not equivalent to a risk-free one. Therefore,

we conclude that collateralization designed to mitigate counterparty risk works well for financial

instruments subject to bilateral credit risk, but fails for ones subject to multilateral credit risk.

Appendix

Proof of Proposition 1. Let t  T0 . On the first payment date T1 , let V (T1 ) denote the market

value of the CDS excluding the current cash flow X 1 . There are a total of eight ( 2 3  8 ) possible states

shown in Table A_1.

Table A_1. Payoffs of a trilaterally defaultable CDS

This table shows all possible payoffs at time T1 . In the case of V (T1 )  X 1  0 where V (T1 ) is the market

value excluding the current cash flow X 1 , there are a total of eight ( 2 3  8 ) possible states: i) A, B, and C

survive with probability p000 . The instrument value equals the market value: V (T1 )  X 1 . ii) A defaults,

but B and C survive with probability p100 . The instrument value is a fraction of the market value:

 B (T1 )V (T1 )  X 1  where  B represents the non-default recovery rate of party B .  B =0 represents the
5

5
There are two default settlement rules in the market. The one-way payment rule was specified by the

early ISDA master agreement. The non-defaulting party is not obligated to compensate the defaulting

18

Electronic copy available at: https://ssrn.com/abstract=3410640

one-way settlement rule, while  B =1 represents the two-way settlement rule. iii) A and C survive, but B

defaults with probability p010 . The instrument value is given by  B (T1 )V (T1 )  X 1  where  B represents

the default recovery rate of defaulting party B. iv) A and B survive, but C defaults with probability p001 .

The instrument value is the default payment: R (T0 , T1 ) . v) A and B default, but C survives with probability

p110 . The instrument value is given by  AB (T1 )V (T1 )  X 1  where  AB denotes the joint recovery rate

when both parties A and B default simultaneously. vi) A and C default, but B survives with probability

p101 . The instrument value is a fraction of the default payment:  B (T ) RT0 , T1  . vii) B and C default, but A

survives with probability p011 , The instrument value is given by  B (T ) RT0 , T1  . viii) A, B, and C default

with probability p111 . The instrument value is given by  AB (T ) RT0 , T1  . A similar logic applies to the case

of V (T1 )  X 1  0 .

Status Probability Payoff if V (T1 )  X 1  0 Payoff if V (T1 )  X 1  0

YA  0, YB  0, YC  0 p000 V (T1 )  X 1 V (T1 )  X 1

YA  1, YB  0, YC  0 p100  B (T1 )V (T1 )  X 1   A (T1 )V (T1 )  X 1 

YA  0, YB  1, YC  0 p 010  B (T1 )V (T1 )  X 1   A (T1 )V (T1 )  X 1 
YA  0, YB  0, YC  1 p 001 R (T0 , T1 ) R (T0 , T1 )

YA  1, YB  1, YC  0 p110  AB (T1 )V (T1 )  X 1   AB (T1 )V (T1 )  X 1 

YA  1, YB  0, YC  1 p101  B (T ) RT0 , T1   B (T ) RT0 , T1 

YA  0, YB  1, YC  1 p 011  B (T ) RT0 , T1   B (T ) RT0 , T1 

YA  1, YB  1, YC  1 p111  AB (T ) RT0 , T1   AB (T ) RT0 , T1 

The risky price is the discounted expectation of the payoffs and is given by

party if the remaining market value of the instrument is positive for the defaulting party. The two-way

payment rule is based on current ISDA documentation. The non-defaulting party will pay the full market

value of the instrument to the defaulting party if the contract has positive value to the defaulting party.

19

Electronic copy available at: https://ssrn.com/abstract=3410640

 
V (t )  E 1(V (T1 ) X 1 0  p 000 (T0 , T1 )  p100 (T0 , T1 ) B (T1 )  p 010 (T0 , T1 ) B (T1 )  p110 (T0 , T1 ) AB (T1 ) (V (T1 )  X 1 ) Ft
 1(V (T1 )  X 1 )0  p 000 (T0 , T1 )  p100 (T0 , T1 ) A (T1 )  p 010 (T0 , T1 ) A (T1 )  p110 (T0 , T1 ) AB (T1 ) (V (T1 )  X 1 ) Ft
  p 001 (T0 , T1 )  p101 (T0 , T1 ) B (T1 )  p 011 (T0 , T1 ) B (T1 )  p111 (T0 , T1 ) AB (T1 ) R(Ts , T1 ) Ft D(t , T )
 EO(T0 , T1 )(V (T1 )  X 1 )  (T0 , T1 ) R(Ts , T1 ) Ft 
(A1a)

where

O(T0 ,T1 )  1(V (T1 ) X1 )0 B (T0 ,T1 )  1(V (T1 ) X1 )0  A (T0 ,T1 ) (A1b)

 A (T0 , T1 )  p A (T0 , T1 ) p B (T0 , T1 ) pC (T0 , T1 )  q A (T0 , T1 ) p B (T0 , T1 ) pC (T0 , T1 ) A (T1 )

 p A (T0 , T1 )q B (T0 , T1 ) pC (T0 , T1 ) A (T1 )  q A (T0 , T1 )q B (T0 , T1 ) pC (T0 , T1 ) AB (T1 )
 pC (T0 , T1 ) AB (T0 , T1 )1   A (T1 )   A (T1 )   AB (T1 )  (A1c)
  AC (T0 , T1 ) p B (T0 , T1 )1   A (T1 )   q B (T0 , T1 ) A (T1 )   AB (T1 ) 
  BC (T0 , T1 ) p A (T0 , T1 )1   A (T1 )   q A (T0 , T1 ) A (T1 )   AB (T1 ) 
  ABC (T0 , T1 ) 1   A (T1 )   AB (T1 )   A (T1 ) D(T0 , T1 )

 B (T0 , T1 )  p A (T0 , T1 ) p B (T0 , T1 ) pC (T0 , T1 )  q A (T0 , T1 ) p B (T0 , T1 ) pC (T0 , T1 ) B (T1 )

 p A (T1 , T1 )q B (T0 , T1 ) pC (T0 , T1 ) B (T1 )  q A (T0 , T1 )q B (T0 , T1 ) pC (T0 , T1 ) AB (T1 )
 pC (T0 , T1 ) AB (T0 , T1 )1   B (T1 )   B (T1 )   AB (T1 )  (A1d)
  AC (T0 , T1 ) p B (T0 , T1 )1   B (T1 )   q B (T0 , T1 ) B (T1 )   AB (T1 ) 
  BC (T0 , T1 ) p A (T0 , T1 )1   B (T1 )   q A (T0 , T1 ) B (T1 )   AB (T1 ) 
  ABC (T0 , T1 ) 1   B (T1 )   AB (T1 )   B (T1 ) D(T0 , T1 )

(T0 , T )  p A (T0 , T ) p B (T0 , T )qC (T0 , T )  q A (T0 , T ) p B (T0 , T )qC (T0 , T ) B (T1 )
 p A (T0 , T )q B (T0 , T )qC (T0 , T ) B (T1 )  q A (T0 , T )q B (T0 , T )qC (T0 , T ) AB (T1 )
 qC (T0 , T ) AB (T0 , T )1   B (T1 )   B (T1 )   AB (T1 )  (A1e)
  AC (T0 , T ) p B (T0 , T )1   B (T1 )   q B (T0 , T ) B (T1 )   AB (T1 ) 
  BC (T0 , T ) p A (T0 , T )1   B (T1 )   q A (T0 , T ) B (T1 )   AB (T1 ) 
  ABC (T0 , T )1   B (T1 )   AB (T1 )   B (T1 ) D(T0 , T )

Similarly, we have


V (T1 )  E O (T1 , T2 ) X 2  V (T2 )   (T1 , T2 ) R (T1 , T2 ) FT 1  (A2)

Note that O(T0 , T1 ) is FT -measurable. By definition, an FT -measurable random variable is a

1 1

random variable whose value is known at time T1 . According to taking out what is known and tower

properties of conditional expectation, we have

20

Electronic copy available at: https://ssrn.com/abstract=3410640

 V (t )  EO(T0 , T1 ) X 1  V (T1 )   (T0 , T1 ) R(T0 , T1 ) Ft   E O(T0 , T1 ) X 1 Ft 

 E (T0 , T1 ) R(T0 , T1 ) Ft   E O(T0 , T1 ) E O(T1 , T2 ) X 2  V (T2 )   (T1 , T2 ) R(T1 , T2 ) FT 1 Ft  (A3)
 i 1 E
2
 i 1
j 0
 
O(T j , T j 1 ) X i  i 1 E
2
 i 2
j 0

O(T j , T j 1 ) (Ti 1 , Ti ) R(Ti 1 , Ti ) 
E  1
j 0

O(T j , T j 1 ) V (T2 ) Ft 
By recursively deriving from T2 forward over Tm , where V (Tm )  X m , we have

V (t )   i 1 E
m
 i 1
j 0
 
O (T j , T j 1 ) X i Ft   i 1 E
m
 i 2
j 0

O(T j , T j 1 ) (Ti 1 , Ti ) R(Ti 1 , Ti ) Ft  (A4)

References

Arora, Navneet, Priyank Gandhi, and Francis A. Longstaff (2009), “Counterparty credit risk and the

credit default swap market,” Working paper, UCLA.

Brigo, D., A. Pallavicini, and R. Torresetti (2011), “Credit Models and the Crisis: default cluster

dynamics and the Generalized Poisson Loss model,” Journal of Credit Risk, 6: 39-81.

Brigo, D., and Capponi, A., 2008, Bilateral counterparty risk valuation with stochastic dynamical models

and application to Credit Default Swaps, Working paper.

Collin-Dufresne, P., R. Goldstein, and J. Helwege (2003), “Is credit event risk priced? Modeling

contagion via the updating of beliefs,” Working paper, Haas School, University of California, Berkeley.

Das, S., D. Duffie, N. Kapadia, and L. Saita (2007), “Common failings: How corporate defaults are

correlated,” Journal of Finance, 62: 93-117.

Das, S., L. Freed, G. Geng, N. Kapadia (2006), “Correlated default risk,” Journal of Fixed Income, 16: 7-

32.

21

Electronic copy available at: https://ssrn.com/abstract=3410640

Davis, M., and V. Lo (2001), “Infectious defaults,” Quantitative Finance, 1: 382-387.

Bielecki, T., I. Cialenco and I. Iyigunler, (2013) “Collateralized CVA valuation with rating triggers and

credit migrations,” International Journal of Theoretical and Applied Finance, 16 (2).

Crepey, S. (2015) “Bilateral counterparty risk under funding constraints – part II: CVA,” Mathematical

Finance, 25(1), 23-50.

Deardorff, Alan V. (1982): “The general validity of the Heckscher-Ohlin Theorem,” American Economic

Review, 72 (4): 683-694.

Duffie, D., A. Eckner, G. Horel, and L. Saita (2009), “Frailty correlated default,” Journal of Finance, 64:

2089-2123.

Duffie, D., and K. Singleton (1999), “Modeling term structure of defaultable bonds,” Review of Financial

Studies, 12: 687-720.

ECB (2009), “Credit default swaps and counterparty risk,” European central bank.

FinPricing, 2013, Investment banking solution, https://finpricing.com/product.html

Fuijii, M. and A. Takahahsi (2012), “Collateralized CDS and default dependence – Implications for the

central clearing,” Journal of Credit Risk, 8(3): 97-113.

22

Electronic copy available at: https://ssrn.com/abstract=3410640

Giesecke, K. (2004), “Correlated default with incomplete information,” Journal of Banking and Finance,

28: 1521-1545.

Gregory, Jon, 2009, Being two-faced over counterparty credit risk, RISK, 22, 86-90.

Hull, J. and A. White (2004), “Valuation of a CDO and a nth to default CDS without Monte Carlo

simulation,” Journal of Derivatives, 12: 8-23.

Hull, J. and A. White (2014), “Collateral and Credit Issues in Derivatives Pricing,” Journal of Credit Risk,

10 (3), 3-28.

ISDA (2013), “ISDA margin survey 2013.”

Jarrow, R., and S. Turnbull (1995), “Pricing derivatives on financial securities subject to credit risk,”

Journal of Finance, 50: 53-85.

Jarrow, R., and F. Yu (2001), “Counterparty risk and the pricing of defaultable securities,” Journal of

Finance, 56: 1765-99.

Johannes, M. and S. Sundaresan (2007), “The impact of collateralization on swap rates,” Journal of
Finance, 62: 383-410.

J. P. Morgan (1999), “The J. P. Morgan guide to credit derivatives,” Risk Publications.

Koopman, S., A. Lucas, and B. Schwaab (2011), “Modeling frailty-correlated defaults using many

macroeconomic covariates,” Journal of Econometrics, 162: 312-325.

23

Electronic copy available at: https://ssrn.com/abstract=3410640

Lando, D. and M. Nielsen (2010), “Correlation in corporate defaults: contagion or conditional

independence?” Journal of Financial Intermediation, 19: 355-372.

Laurent, J. and J. Gregory (2005), “Basket default swaps, CDOs and factor copulas,” Journal of Risk, 7:

03-122.

Li, D. (2000), “On default correlation: A copula function approach,” Journal of Fixed Income, 9: 43-54.

Lipton, A., and Sepp, A., 2009, Credit value adjustment for credit default swaps via the structural default

model, Journal of Credit Risk, 5(2), 123-146.

Longstaff, F., and E. Schwartz (2001): “Valuing American options by simulation: a simple least-squares

approach,” The Review of Financial Studies, 14 (1): 113-147.

Madan, D., and H. Unal (1998), “Pricing the risks of default,” Review of Derivatives Research, 2: 121-

160.

Piterbarg, V. (2010), “Funding beyond discounting: collateral agreements and derivatives pricing,” Risk
Magazine, 2010 (2): 97-102.

Pykhtin, Michael, and Steven Zhu, 2007, A guide to modeling counterparty credit risk, GARP Risk

Review, July/August, 16-22.

Teugels, J. (1990), “Some representations of the multivariate Bernoulli and binomial distributions,”

Journal of Multivariate Analysis, 32: 256-268.

24

Electronic copy available at: https://ssrn.com/abstract=3410640

Zhang, G., and P. Jorion, (2007), “Good and bad credit contagion: Evidence from credit default swaps,”

Journal of Financial Economics 84: 860–883.

25

Electronic copy available at: https://ssrn.com/abstract=3410640