Counterparty Risk and the Impact of Collateralization

in CDS Contracts
arXiv:1104.2625v3 [q-fin.RM] 21 Aug 2011

Tomasz R. Bielecki∗ Igor Cialenco∗

Department of Applied Mathematics, Department of Applied Mathematics,
Illinois Institute of Technology, Illinois Institute of Technology,
Chicago, 60616 IL, USA Chicago, 60616 IL, USA
[email protected] [email protected]

Ismail Iyigunler
Department of Applied Mathematics,
Illinois Institute of Technology,
Chicago, 60616 IL, USA
[email protected]

First Circulated: April 11, 2011

This version: September 15, 2018

Abstract
We analyze the counterparty risk embedded in CDS contracts, in presence of a bilat-
eral margin agreement. First, we investigate the pricing of collateralized counterparty
risk and we derive the bilateral Credit Valuation Adjustment (CVA), unilateral Credit
Valuation Adjustment (UCVA) and Debt Valuation Adjustment (DVA). We propose
a model for the collateral by incorporating all related factors such as the thresholds,
haircuts and margin period of risk. We derive the dynamics of the bilateral CVA in a
general form with related jump martingales. We also introduce the Spread Value Ad-
justment (SVA) indicating the counterparty risk adjusted spread. Counterparty risky
and the counterparty risk-free spread dynamics are derived and the dynamics of the
SVA is found as a consequence. We finally employ a Markovian copula model for de-
fault intensities and illustrate our findings with numerical results.

∗
TRB and IC acknowledge support from the NSF grant DMS-0908099

1
Contents
1 Introduction 3

2 Pricing Counterparty Risk: CVA, UCVA and DVA 4

2.1 Dividend Processes and Marking-to-Market . . . . . . . . . . . . . . . . . . 5
2.1.1 Bilateral Margin Agreement and Collateral Modeling . . . . . . . . . 8
2.2 Bilateral Credit Valuation Adjustment . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Unilateral CVA and Debt Value Adjustment . . . . . . . . . . . . . 12
2.2.2 CVA via Credit Exposures . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Dynamics of CVA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.1 Dynamics of CVA when the immersion property holds . . . . . . . . 21
2.4 Fair Spread Value Adjustment . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.1 SVA Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3 Multivariate Markovian Default Model 25

3.1 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4 Conclusion 27

2
1 Introduction
Not very long after the collapse of prestigious institutions like Long-Term Capital Man-
agement, Enron and Global Crossing, the financial industry has again witnessed dramatic
downfalls of financial institutions such as Lehman Brothers, Bear Stearns and Wachovia.
These recent collapses have stressed out the importance of measuring, managing and miti-
gating counterparty risk appropriately.
Counterparty risk is defined as the risk that a party in an over-the-counter (OTC)
contract will default and will not be able to honor its contractual obligations. Since the
exchange-traded derivative contracts are subject to clearing by the exchange, counterparty
risk arises from OTC derivatives only. The main challenge in the counterparty risk as-
sessment and hedging is that the exposures of OTC derivatives are stochastic and involve
dependencies and systemic risk factors such as wrong way risks; the additional level of com-
plexity is introduced by risk mitigation techniques such as collateralization and netting.
Therefore, one needs to model potential future exposures and to price the counterparty risk
appropriately according to margin agreements that underlie the collateralization procedures.
Brigo and Capponi [BC09] focuses on a Gaussian copula model and study bilateral
counterparty risk using a CIR++ intensity model. Recently, Brigo et al. [BCPP11] extended
this framework to the collateralized contracts with an application to interest rate swaps
under bilateral margin agreements. Hull and White [HW01] propose a static copula model
and study unilateral counterparty risk on credit default swaps. Bielecki et al. [BCJZ11]
study unilateral counterparty risk with the absence of any margin agreements. Assefa et al.
[ABCJ11] consider the portfolio of credit default swaps under Markovian copula model and
consider only fully collateralized contracts. Jarrow and Yu [JY01] deal with the counterparty
risk by using a dependence structure based on the default intensities of the counterparties.
This approach, that also addresses the contagion risk issue, is considered in Leung and Kwok
[LK05]. All these works mentioned above employ the reduced form modeling technology.
However, structural models have also been used to model counterparty risk. Good examples
of this approach are papers by Lipton and Sepp [LS09] and Blanchet-Scalliet and Patras
[BSP11]. Moreover, Stein and Lee [SL11] study and illustrate credit valuation adjustment
computations in the fixed income markets.
Various issues regarding the simulation of credit valuation adjustments under margin
agreements are studied by Pykhtin in [Pyk09]. Furthermore, Cesari et al. [CAC10] and
Gregory [Gre09] provide thorough treatments of the methods and the applications used in
practice regarding the counterparty risk.
In this paper, we analyze the counterparty risk in a Credit Default Swap (CDS) contract
in presence of a bilateral margin agreement. There are three risky names associated with
the contract: the reference entity, protection seller (the counterparty) and the protection
buyer (the investor). Contrary to the common approach which starts with defining the
Potential Future Exposure (PFE) and derives the Credit Valuation Adjustment (CVA) as
the price of the counterparty risk, we find the CVA as the difference between the market

3
values of a counterparty risk-free and a counterparty risky CDS contracts and deduct the
relevant credit exposures accordingly. We consider the problem of bilateral counterparty
risk assessment; that is, we consider the situation where the two counterparties of the CDS
contract, i.e. the investor and the counterparty, are subject to default risk in a counterparty
risky CDS contract.
We focus on the collateralized contracts, where there is a bilateral margin agreement is in
force as a vital risk mitigation tool, and it requires the counterparty and the investor to post
collateral in case their exposure exceeds specific threshold values. We propose a model for
the collateral by incorporating all related factors, such as thresholds, margin period of risk
and minimum transfer amount. Then, we derive the dynamics of the bilateral CVA which
is essential for dynamic hedge of the counterparty risk. We also compute the decomposition
of the fair spread for the CDS, and we analyze so called Spread Value Adjustment (SVA).
Essentially, SVA represents the adjustment to be made to the fair spread to incorporate the
counterparty risk into the CDS contract.
Using the bilateral CVA formula, we derive relevant formulas for assessment of credit ex-
posures, such as PFE, Expected Positive Exposure (EPE) and Expected Negative Exposure
(ENE).
In our model, the dependence between defaults and the wrong way risk is represented
in a Markovian copula framework that accounts for simultaneous defaults among the three
names represented in a CDS contract. In this way, our model takes broader systemic risk
factors into account and quantifies the wrong way risk and the double defaults in a tangible
manner.
This paper is organized as follows. In Section 2, we first define the dividend processes
regarding the counterparty risky and the counterparty risk-free CDS contract in case of a
bilateral margin agreement. We also define the CVA, UCVA and the DVA terms as well
as the credit exposures such as PFE, EPE and ENE. We then prove the dynamics of the
CVA in Section 2.3. Moreover, we find the fair spread adjustment term and its dynamics
in Section 2.4. In Section 3, we simulate the collateralized exposures and the CVA using
our Markovian copula model of default dependence.

2 Pricing Counterparty Risk: CVA, UCVA and DVA

We consider a standard CDS contract, and we label by 1 the reference name, by 2 the
counterparty, and by 3 the investor. Each of the three names may default before the
maturity of the CDS contract, and we denote by τ1 , τ2 and τ3 their respective default
times. These times are modeled as non-negative random variables given on a underlying
probability space (Ω, G, Q). We let T and κ to denote the maturity and the spread of our
CDS contract, respectively. We assume the recovery at default covenant; that is, we assume
that recoveries are paid at times of default.
We introduce right-continuous processes Hti by setting Hti =I{τi ≤t} and we denote by Hi


the associated filtrations so that Hti = σ Hui : u ≤ t for i = 1, 2, 3.

4
 We assume that we are given a market filtration F, and we define the enlarged filtration


G = F∨H1 ∨H2 ∨H3 , that is Gt = σ Ft ∪ Ht1 ∪ Ht2 ∪ Ht3 for any t ∈ R+ . For each t ∈ R+
total information available at time t is captured by the σ-field Gt . In particular, processes
H i are G-adapted and the random times τi are G-stopping times for i = 1, 2, 3.
Next, we define the first default time as the minimum of τ1 , τ2 and τ3 : τ = τ1 ∧τ2 ∧τ3 ; the
corresponding indicator process is Ht = I{τ ≤t} . In addition, we define the first default time
of the two counterparties: τ̂ = τ2 ∧ τ3 , and the corresponding indicator process Ĥt = I{τ̂ ≤t} .
We also denote by Bt the savings account process, that is
Rt
rs ds
Bt = e 0 ,

where the F-progressively measurable process r models the short-term interest rate. We
also postulate that Q represents a martingale measure associated with the choice of the
savings account B as a discount factor (or numeraire).

2.1 Dividend Processes and Marking-to-Market

In this paper, all cash flows and the prices are considered from the perspective of the
investor.
We start by introducing the counterparty-risk-free dividend process D, which describes
all cash flows associated with a counterparty-risk-free CDS contract;1 that is, D does not
account for the counterparty risk.

Definition 2.1. The cumulative dividend process D of a counterparty risk-free CDS con-
tract maturing at time T is given as,
Z Z
1 1


Dt = δu dHu − κ 1 − Hu1 du, (1)
]0,t] ]0,t]

for every t ∈ [0, T ], where δ1 : [0, T ] → R is an F-predictable processes.

Process δ1 represents the loss given default (LGD); that is δ1 = 1 − Rt1 , where R1 is the
fraction of the nominal that is recovered in case of the default of the reference name. We
assume unit nominal, for simplicity.
The ex-dividend price processes of the counterparty risk-free CDS contract, say S, de-
scribes the current market value, or the Mark-to-Market (MtM) value of this contract.

Definition 2.2. The ex-dividend price process S of a counterparty risk-free CDS contract
maturing at time T is given by,
Z  !


S t = Bt E Q Bu−1 dDu  Gt , t ∈ [0, T ]. (2)
]t,T ] 
1
We shall sometimes refer to such contract as to the clean contract.

5
 b of a
Remark 2.3. Accordingly, we define the cumulative (dividend) price process, say S,
counterparty risk-free CDS contract as
Z
b
S t = S t + Bt Bu−1 dDu , t ∈ [0, T ].
]0,t]

Now, we are in position to define the dividend process D C of a counterparty risky CDS
contract, that is the CDS contract that accounts for the counterparty risk associated with
the two counterparties of the contract.

Definition 2.4. The dividend process D C of a T -maturity counterparty-risky CDS contract

is given as
Z Z Z
DtC = Cu dHu + δeu1 (1 − Hu− ) dHu1 + δeu2 (1 − Hu− ) dHu2
]0,t] ]0,t] ]0,t]
Z Z
+ δeu3 (1 − Hu− ) dHu3 + δeu4 (1 − Hu− ) d[H 2 , H 3 ]u (3)
]0,t] ]0,t]
Z Z
+ e 5 1
δu (1 − Hu− ) d[Ĥ, H ]u − κ (1 − Hu ) du, t ∈ [0, T ],
]0,t] ]0,t]

where δei : [0, T ] → R is an F-predictable processes for i = 1, . . . , 5 and C : [0, T ] → R is an

F-predictable process representing the collateral amount kept in the margin account.

Margin account is a contractual tool that supplements the CDS contract so to reduce
potential losses that may be incurred by one of the counterparties in case of the default of
the other counterparty, while the CDS contract is still alive. For the detailed description
of the mechanics of the collateral formation in the margin account we refer to Section 2.1.1
(see also [BC11]).
In case of any credit event, associated with the collateralized CDS contract, the first cash
flow that takes place is the “transfer” of the collateral amount; for example, in case when
the underlying entity defaults at time t = τ = τ 1 , (before any of the counterparties defaults)
the collateral in the margin account is acquired by one of the counterparties (depending on
the sign of Cτ ). Thus, consistently with the conventions of so called close-out cash flows
(cf. [BC11]) we define δei s as follows:

• We set δet1 = δt1 − Ct . This is because after the collateral transfer the counterparty
pays the remaining recovery amount δt1 − Ct .

• At time t = τ = τ 2 , when the counterparty defaults, then, after the collateral transfer
takes place, if the uncollateralized mark-to-market (MtM) of the CDS contract, that
is St + I{t=τ1 } δt1 − Ct 2 , is negative, the investor closes out the position by paying
the defaulting counterparty the uncollateralized MtM. If the uncollateralized MtM
2
The term I{t=τ 1 } δt1 represents the exposure in case when the counterparty and the underlying entity
default simultaneously.

6
 is positive, the investor closes out the position and receives a fraction R2 of the
uncollateralized MtM from the counterparty. Therefore, in this case, the close-out
payment is defined as,

+
−
δet2 = R2 St + I{t=τ 1 } δt1 − Ct − St + I{t=τ1 } δt1 − Ct .

• In case of investor default, that is at time t = τ = τ 3 , if the uncollateralized MtM is

positive, that is if St + I{t=τ1 } δt1 − Ct > 0, the counterparty closes out the position
by paying the uncollateralized MtM. If the uncollateralized MtM is negative, the
counterparty receives a fraction R3 of the uncollateralized MtM. Hence, the close-out
payment is defined as,

+
−
δet3 = St + I{t=τ 1 } δt1 − Ct − R3 St + I{t=τ1 } δt1 − Ct .

• If the investor and the counterparty default simultaneously at time t = τ = τ 2 = τ3 ,

and if the uncollateralized MtM negative, the counterparty receives a fraction R3
of the uncollateralized MtM; however, if the uncollateralized MtM is positive, the
investor receives a fraction R2 of the uncollateralized MtM. Therefore, we set,


δet4 = − St + I{t=τ1 } δt1 − Ct .

• If t = τ = τ̂ = τ1 , that is when the investor or the counterparty default simultaneously

with the reference entity, investor receives a fraction R2 of the remaining recovery

+
amount, δt1 − Ct , when the counterparty defaults. Likewise, if the investor defaults,

−
the counterparty receives a portion R3 of the remaining recovery amount, δt1 − Ct .
The close-out payment in joint defaults including the underlying entity has the form,


δet5 = − δt1 − Ct .

We are now ready to define the price processes associated with a counterparty risky
CDS contract.

Definition 2.5. The ex-dividend price process S C of a counterparty risky CDS contract
maturing at time T is given as,
Z  !


StC = Bt EQ Bu−1 dDuC  Gt , t ∈ [0, T ]. (4)
]t,T ] 

The cumulative price process SbC of a counterparty risky CDS contract is given by,
Z
SbtC = StC + Bt Bu−1 dDuC , t ∈ [0, T ].
]0,t]

7
2.1.1 Bilateral Margin Agreement and Collateral Modeling

Collateralization is one of the most important techniques of mitigation of counterparty risk,

and modeling the collateral process (also termed the margin call process) is of great practical
importance (cf. [Alg09]). In this section, we propose a model to describe formation of the
required collateral amount at every time t ∈ [0, T ], with regard to bilateral margin agree-
ments.3 The following contractual parameters are essential in bilateral margin agreements
and they are precisely defined in CSA documents.
Margin Period of Risk : The margin period of risk consists of several components. Firms
usually monitor their exposure on a periodic basis and receive or make appropriate margin
calls considering other collateral parameters. The frequency of this process is called the
margin call period and it is typically one day. This period includes a number of phases
such as computation, negotiation, verification and settlement of the margin call also with
possible disputes during the process. According to the ISDA Master Agreement, in case of
a potential default, the defaulting counterparty enters into a short forbearance period to
recover from a potential default event where the collateral is pledged by the other firm. This
time interval is called the cure period. If the default is uncured, liquidation process of the
collateral assets starts (cf. [ISD10b], page 26). This period mainly depends on the collateral
portfolio selection, precise assessment of asset correlation and concentration risks as well as
their liquidity, volatility and credit quality parameters. Therefore, the time interval from
the last margin call plus the cure period until all collateral assets are liquidated and the
resulting market risk is re-hedged is called the margin period of risk (cf. [Pyk09]); we shall
denote it as ∆.
Threshold: The threshold is the unsecured credit exposure that both counterparties
are willing to tolerate without holding any collateral. Bilateral margin agreements specify
thresholds for both counterparties and require them to post collateral whenever the current
credit exposure exceeds their thresholds (cf. [ISD10b], page 11). These threshold amounts
are defined in the related CSA documents and often set to react to the changes in the credit
rating of the counterparties (cf.[ISD10a], page 13). We will denote the counterparty and
the investor’s thresholds by Γcpty and Γinv , respectively. Since we are doing our analysis
from the point of view of the investor, we set the counterparty’s threshold Γcpty to be a
non-negative constant, and the investor’s threshold Γinv to be a non-positive constant.
Minimum Transfer Amount: Margin calls for amounts smaller than the M T A are not
allowed. The purpose of the M T A is to prevent calling small amounts; this is done so to
avoid the operational costs involved in small transactions (cf. [ISD10b], page 13). Minimum
transfer amount is usually the same for the investor and the counterparty.
Re-hypothecation Risk and Segregation: Collateral assets can be reused as a funding
source on another derivatives transactions. This is known as rehypothecation. An investor
3
A bilateral margin agreement is a contractual agreement governed by a Credit Support Annex (CSA),
which is a regulatory part of the ISDA Master Agreement (cf. [ISD05], page 34) describing the use of
collateral which is either directly transferred between counterparties or held by a third party such as a
clearing house (cf. [ISD05], page 68).

8
(counterparty) can rehypothecate the collateral received from the counterparty (investor)
by selling or lending out the assets to a third party, which dramatically increases the credit
risk associated with the collateral. Elimination of this rehypothecation risk is essentially
done by segregating the collateral to a third party, such as a clearing house. This procedure
carries certain funding risks, since the counterparties will not be getting funding benefit from
the collateral posted, so they need to raise funding in connection with their transactions
using their own funding rates.
The construction of the collateral process presented below builds upon the construction
given in [BC11]. Let us denote the margin call dates by 0 < t1 < · · · < tn < T . On
each margin call date, if the exposure is above the counterparty’s threshold, Γcpty , and the
difference between the current exposure and the collateral amount is greater than the MTA
the counterparty posts collateral and updates the margin account; otherwise, no collateral
exchange takes place since the transfer amount is less than the MTA. Likewise, the investor
delivers collateral on each margin call date, if the exposure is below investor’s threshold,
Γinv , and the difference between the current exposure and the collateral amount is greater
than MTA (cf. [ISD05], pages 52-56). Note that in this model a collateral transfer are
allowed only if it is greater than the M T A amount.
In accordance with the above discussion the collateral process is modeled as follows:
We set C0 = 0. Then, for i = 1, 2, . . . , n we postulate that

Cti + = I{Sti −Γcpty −Cti >M T A} (Sti − Γcpty − Cti )

+ I{St −Γinv −Cti <−M T A} (Sti − Γinv − Cti ) + Cti ,
i

on the set {ti < τ̂ }, and it is constant on interval (ti , ti+1 ]. Moreover, Ct = Cτ̂ on the set
{τ̂ < t < τ̂ + ∆}.
Observe that the collateral increments at each margin call date ti < τ̂ can now be
represented as,

∆Cti : = Cti + − Cti

= I{Sti −Γcpty −Cti >M T A} (Sti − Γcpty − Cti )
+ I{St } (Sti − Γinv − Cti )
i −Γinv −Cti <−M T A

One should also note that the collateral construction given in [Pyk09], which reads

Ct = I{St >Γcpty +M T A} (St − Γcpty ) + I{St <Γbuy −M T A} (St − Γinv ) ,

allows intermediate collateral updates that are smaller than MTA. In our case, we avoid this
intricacy by defining the collateral process as a left-continuous, piecewise constant process.
Remark 2.6. The collateral construction described above is cash based. The net cash value
of the collateral portfolio is determined using haircuts.
The haircut (or, valuation percentage) describes the amount that will be charged from a
particular collateral asset. Effective value of the collateral asset is determined by subtract-
ing the mark-to-market value of the asset multiplied by an appropriate haircut (cf. [ISD05],

9
page 67). Therefore, the haircuts applied to collateral assets should reflect the market risk
on those assets. The haircut is defined as a percentage, where 0% haircut implies complete
mark-to-market value of the asset to be used as collateral without any discounting. Gov-
ernment securities having high credit rating such as Treasury bonds and Treasury bills are
usually subjected to 1% to 10% haircut, while for more risky, volatile or illiquid securities,
such as a stock option, the haircut might be as high as 30%. The only asset that is not
subjected to any haircut as collateral is cash where usually both parties mutually agree the
use of an overnight index rate (cf. [ISD10b], page 27). The term valuation percentage is
also used in Credit Support Annex (CSA) documents. The valuation percentage defines the
amount that the market value of the asset is multiplied by to yield the effective collateral
value of the asset. Hence, we have VPt = 1 − ht , where VPt is the valuation percentage
and ht is the total haircut applied to the collateral assets at time t. We will not go into the
details of the formation of the haircut since it is either pre-determined in the CSA docu-
ments or related to market risk measures such as VaR of the collateral assets. (cf. [ISD05],
page 68). The main purpose of the haircut is to mitigate amortization or depreciation in the
collateral asset value at the time of a default and in the margin period of risk. Moreover, the
haircut should be updated as frequently as possible to reflect the changes in the volatility
or liquidity of the collateral assets (cf. [ISD05], page 63).
Therefore, the total value of the collateral portfolio at time t is equal to (1 + ht )Ct ,
where ht is the appropriate haircut applied to the collateral portfolio.

2.2 Bilateral Credit Valuation Adjustment

In this section, we shall compute the CVA on a CDS contract, subject to a bilateral margin
agreement.

Definition 2.7. The bilateral Credit Valuation Adjustment process on a CDS contract
maturing at time T is defined as

CVAt = St − StC , (5)

for every t ∈ [0, T ].

We now present an alternative representation for the bilateral CVA, which is convenient
for computational purposes.

Proposition 2.8. The bilateral CVA process on a CDS contract maturing at time T satisfies

+  
CVAt = Bt EQ I{t<τ =τ 2 ≤T } Bτ−1 (1 − R2 ) Sτ + I{τ =τ 1 } δτ1 − Cτ  Gt

−  
− Bt EQ I{t<τ =τ 3 ≤T } Bτ−1 (1 − R3 ) Sτ + I{τ =τ 1 } δτ1 − Cτ  Gt , (6)

for every t ∈ [0, T ].

10
Proof. We begin by observing that
Z
Bu−1 δeui (1 − Hu− ) dHui = Bτ−1 δeτi I{t<τ =τ i ≤T } ,
]t,T ]

for i = 1, 2, 3. Consequently,
Z
Bu−1 dDuC = Bτ−1 δeτ1 I{t<τ =τ 1 ≤T } + Bτ−1 δeτ2 I{t<τ =τ 2 ≤T }
]t,T ]

+ Bτ−1 δeτ3 I{t<τ =τ 3 ≤T } + Bτ−1 δeτ4 I{t<τ =τ 2 =τ3 ≤T }

+ Bτ−1 δeτ5 I{t<τ =τ ∗ =τ1 ≤T } + Bτ−1 Cτ I{t<τ ≤T }
Z
−κ Bu−1 I{τ >u} du. (7)
]t,T ]

Using the definitions of the close-out cash-flows δeτi , i = 1, . . . , 5, we get from (7)
Z


Bu−1 dDuC = Bτ−1 δτ1 − Cτ I{t<τ =τ 1 ≤T }
]t,T ]
Z
−κ Bu−1 I{τ >u} du + Bτ−1 Cτ I{t<τ ≤T } (8)
]t,T ]

+
−1
+ Bτ R2 Sτ +I{τ =τ 1 } δτ1 − Cτ
1

− 
− Sτ + I{τ =τ 1 } δτ − Cτ I{t<τ =τ 2 ≤T }

+
+ Bτ−1 Sτ + I{τ =τ 1 } δτ1 − Cτ

− 
−R3 Sτ + I{τ =τ 1 } δτ1 − Cτ I{t<τ =τ 3 ≤T }


− Bτ−1 Sτ + I{τ =τ 1 } δτ1 − Cτ I{t<τ =τ 2 =τ3 ≤T }


− Bτ−1 δτ1 − Cτ I{t<τ =τ ∗ =τ1 ≤T } .
Since
I{t<τ ≤T } = I{t<τ =τ 1 ≤T } + I{t<τ =τ 2 ≤T } + I{t<τ =τ 3 ≤T }
− I{t<τ =τ 2 =τ3 ≤T } − I{t<τ ∗ =τ 1 ≤T } ,
using the equality
Ri (Sτ − Cτ )+ − (Sτ − Cτ )− + Cτ = Sτ − (1 − Ri ) (Sτ − Cτ )+
and observing that I{τ =τ 1 } Sτ = 0, we can rearrange the terms in (8) as follows,
Z Z
−1 −1 1
C
Bu dDu = Bτ δτ I{t<τ =τ 1 ≤T } − κ Bu−1 I{τ >u} du (9)
]t,T ] ]t,T ]

+ Bτ−1 Sτ I{t<τ =τ 2 ≤T } + I{t<τ =τ 3 ≤T }

−I{t<τ =τ 2 =τ3 ≤T } I{τ 6=τ 1 }

+
− Bτ−1 (1 − R2 ) Sτ + I{τ =τ 1 } δτ1 − Cτ I{t<τ =τ 2 ≤T }

−
+ Bτ−1 (1 − R3 ) Sτ + I{τ =τ 1 } δτ1 − Cτ I{t<τ =τ 3 ≤T } .

11
Now, combining (9) with (1) we see that
Z  !



StC = Bt EQ I{t<τ =τ 1 ≤T } + I{τ >T } Bu−1 dDu  Gt
]t,T ] 
+ Bt E Q I{t<τ =τ 2 ≤T } + I{t<τ =τ 3 ≤T }
Z  ! !


 
 
−I{t<τ =τ 2 =τ3 ≤T } I{τ 6=τ 1 } EQ Bu−1 dDu  Gτ  Gt
]τ,T ]  

−1 1

+  
− Bt EQ I{t<τ =τ 2 ≤T } Bτ (1 − R2 ) Sτ + I{τ =τ 1 } δτ − Cτ  Gt

−  
+ Bt EQ I{t<τ =τ 3 ≤T } Bτ−1 (1 − R3 ) Sτ + I{τ =τ 1 } δτ1 − Cτ  Gt .

From here, observing that

I{τ ≤t} + I{τ >T } + I{t<τ =τ 1 ≤T } + I{t<τ =τ 2 ≤T } + I{t<τ =τ 3 ≤T } − I{t<τ =τ 2 =τ3 ≤T } I{τ 6=τ 1 } = 1,

we get
Z  !


StC = Bt EQ Bu−1 dDu  Gt
]t,T ] 


− Bt EQ I{t<τ =τ 2 ≤T } Bτ−1 (1 − R2 ) (Sτ − Cτ )+  Gt


+ Bt EQ I{t<τ =τ ≤T } Bτ−1 (1 − R3 ) (Sτ − Cτ )−  Gt ,
3

(10)

which is


StC = St − Bt EQ I{t<τ =τ 2 ≤T } Bτ−1 (1 − R2 ) (Sτ − Cτ )+  Gt


+ Bt EQ I{t<τ =τ ≤T } Bτ−1 (1 − R3 ) (Sτ − Cτ )−  Gt .
3

This proves the result.

Remark 2.9. The above results shows that the value of the bilateral CVA is the same as
the sum of the value of a long position in a zero-strike call option on the uncollateralized
amount and the value of a short position in a zero-strike put option on the uncollateralized
amount.

2.2.1 Unilateral CVA and Debt Value Adjustment

The bilateral nature of the counterparty risk is a consequence of possible default of the coun-
terparty and the possible default of the investor. The values of potential losses associated
with these two components are called unilateral CVA (UCVA) and debt value adjustment
(DVA), respectively, and defined below.

12
Definition 2.10. The Unilateral Credit Value Adjustment is defined as,

+  
UCVAt = Bt EQ I{t<τ =τ 2 ≤T } Bτ−1 (1 − R2 ) Sτ + I{τ =τ 1 } δτ1 − Cτ  Gt , t ∈ [0, T ] ,

and symmetrically the Debt Value Adjustment is defined as,


−  
DVAt = Bt EQ I{t<τ =τ 3 ≤T } Bτ−1 (1 − R3 ) Sτ + I{τ =τ 1 } δτ1 − Cτ  Gt , t ∈ [0, T ] .

Remark 2.11. DVA accounts for the risk of investor’s own default, and it represents the
value of any potential outstanding liabilities of the investors that will not be honored at the
time of the investor’s default:
In fact, at time of his/her default, the investor only pays to the counterparty the recov-

−
ery amount, that is R3 Sτ + I{τ =τ 1 } δτ1 − Cτ . Therefore, the investor gains the remaining

−
amount, which is equal to (1 − R3 ) Sτ + I{τ =τ 1 } δτ1 − Cτ , on his/her outstanding liabili-
ties by defaulting. Risk management of this component is of great importance for financial
institutions.
When considering the unilateral counterparty risk DVA is set to zero.
In view of Proposition 2.8 and of the above definition we have that

CVAt = UCVAt − DVAt , t ∈ [0, T ].

Note that the bilateral CVA amount may be negative for the investor due to “own
default risk”. This also indicates that the price S C of counterparty risky CDS contract may
be greater than the price S of counterparty risk-free contract.
Remark 2.12. (Upfront CDS Conversion)
After the “CDS Big Bang” (cf. [Mar09]) a process has been originated to replace
standard CDS contracts with so called upfront CDS contracts. An upfront CDS contract is
composed of an upfront payment, which is an amount to be exchanged upon the inception
of the contract, and of a fixed spread. The fixed spread, say κ
b, will be 100bps for investment
grade CDS contracts, and 500bps for high yield CDS contracts. The recovery rate is also
standardized to two possible values: 20% or 40%, depending on the credit worthiness of the
reference name. The corresponding cumulative dividend process of a counterparty-risk-free
CDS contract is described in the following definition.
b of a counterparty-risk-free upfront CDS
Definition 2.13. The cumulative dividend process D
contract, maturing at time T , is given as
Z Z


bt =
D δu1 dHu1 − UP − κb 1 − Hu1 du , t ∈ [0, T ] ,
]0,t] ]0,t]

where UP is the upfront payment, and κ

b is the fixed spread.

Recall that the spread κ0 of a standard CDS contract is set such that the protection leg
P L0 and fixed leg κ0 DV 010 are equal at initiation (making the price of the contract to be

13
zero). Similarly, in the case of an upfront CDS contract, with κ b being fixed, the upfront
payment UP is chosen such that the contract has zero value at initiation. It is easy to
convert the conventional spread κ0 into an upfront payment PU and vise versa. Indeed,
directly from the Definition 2.13, and definitions of P L0 and DV 010 , we have

P L0 − U P − κ
bDV 010 = P L0 − κ0 DV 010 = 0 ,

which implies the following representations

UP
U P = (κ0 − κ
b) DV 010 , κ0 = +κ
b.
DV 010
In view of the conversion formulae presented above the discussion of CVA, DVA and
UCVA done for standard CDS contracts can be adopted to the case of the upfront CDS
contracts in a straightforward manner.

2.2.2 CVA via Credit Exposures

Credit exposure is defined as the potential loss that may be suffered by either one of the
counterparties due to the other party’s default. Here, we discuss some measures commonly
used to quantify credit exposure, such as Potential Future Exposure (PFE), Expected Posi-
tive Exposure (EPE) and Expected Negative Exposure (ENE), and their relation to CVA.
Potential Future Exposure is the basic measure of credit exposure:

Definition 2.14. Potential Future Exposure of a CDS contract with a bilateral margin
agreement is defined as follows,

+
P F E = I{τ =τ 2 } (1 − R2 ) Sτ + I{τ =τ 1 } δτ1 − Cτ

−
− I{τ =τ 3 } (1 − R3 ) Sτ + I{τ =τ 1 } δτ1 − Cτ .

Note that there exists several forms in which the potential future exposure is defined by
financial institutions. The PFE definition given above, as a random variable, is in line with
the PFE definitions in (cf. [DPR]), as opposed to the rather classical definition of the PFE
as the quantile of the exposure distribution (cf. [CAC10]).
Remark 2.15. Observe that the CVA is related to PFE as follows,


CV At = Bt EQ I{t<τ ≤T } Bτ−1 P F E  Gt , t ∈ [0, T ] .

Expected Positive Exposure is defined as the expected amount the investor will lose if
the counterparty default happens at time t, and Expected Negative Exposure is defined as
the expected amount the investor will lose if his own default happens at time t. Note that
the losses are conditional on default at time t. EPE and ENE are necessary quantities to
price and hedge counterparty risk.

14
Definition 2.16. The Expected Positive Exposure of a CDS contract with a bilateral
margin agreement is defined as,

+  
EP Et = EQ (1 − R2 ) Sτ + I{τ =τ 1 } δτ1 − Cτ  τ = τ2 = t ,
and the Expected Negative Exposure is defined as,

−  
EN Et = EQ (1 − R3 ) Sτ + I{τ =τ 1 } δτ1 − Cτ  τ = τ 3 = t

for every t ∈ [0, T ].

2.3 Dynamics of CVA

In this section we derive the dynamics for CVA. This is important for deriving formulae for
dynamic hedging of counterparty risk, the issue that will be discussed in a different paper.
We begin with defining some auxiliary stopping times, that will come handy later on:

( (
{1} τ1 if τ1 6= τ2 , τ1 6= τ3 {2} τ2 if τ2 6= τ1 , τ2 6= τ3
τ := , τ := ,
∞ otherwise ∞ otherwise
( (
{3} τ3 if τ3 6= τ1 , τ3 6= τ2 {4} τ2 if τ2 = τ3 , τ2 6= τ1
τ := , τ := ,
∞ otherwise ∞ otherwise
( (
τ1 if τ1 = τ2 , τ1 6= τ3 τ1 if τ1 = τ3 , τ1 6= τ 2
τ {5} := , τ {6} := ,
∞ otherwise ∞ otherwise
(
τ1 if τ1 = τ2 = τ3
τ {7} := .
∞ otherwise
Accordingly, we define the default indicator processes:
{1} {2}
Ht : = I{τ1 ≤t,τ1 6=τ2 ,τ1 6=τ3 } = I{τ {1} ≤t,} , Ht : = I{τ2 ≤t,τ2 6=τ1 ,τ2 6=τ3 } = I{τ {2} ≤t,} ,
{3} {4}
Ht : = I{τ3 ≤t,τ3 6=τ1 ,τ3 6=τ2 } = I{τ {3} ≤t,} , Ht : = I{τ2 =τ3 ≤t,τ1 6=τ2 } = I{τ {4} ≤t,} ,
{5} {6}
Ht : = I{τ1 =τ2 ≤t,τ1 6=τ3 } = I{τ {5} ≤t,} , Ht : = I{τ1 =τ3 ≤t,τ1 6=τ2 } = I{τ {6} ≤t,} ,
{7}
Ht := I{τ1 =τ2 =τ3 ≤t} = I{τ {7} ≤t} .
{i}
Remark 2.17. Note that one can represent processes Ht , i = 1, . . . , 7, as follows
{7}    {6}   {7}
Ht = H 1 , H 2 , H 3 t , Ht = H 1 , H 3 t − Ht ,
{5}   {7} {4}   {7}
Ht = H 1 , H 2 t − Ht , Ht = H 2 , H 3 t − Ht ,
{3} {4} {6} {7}
Ht = Ht3 − Ht − Ht − Ht ,
{2} {4} {5} {7}
Ht = Ht2 − Ht − Ht − Ht ,
{1} {5} {6} {7}
Ht = Ht1 − Ht − Ht − Ht .

15
In particular, these processes are G-adapted processes.
Let G(t) = Q ( τ > t| Ft ) be the survival probability process of τ with respect to filtration
F. It is a F supermartingale and it admits unique Doob-Meyer decomposition G = µ − ν
where µ is the martingale part and ν is a predictable increasing process. We assume that G
is a continuous process and v is absolutely continuous with respect to the Lebesgue measure,
so that dνt = vt dt for some F-progressively measurable, non-negative process v. We denote
by l the F-progressively measurable process defined as lt = G(t)−1 vt . Finally, we assume
that all F martingales are continuous.
We assume that hazard process of each stopping time τ {i} admits an (F, G)-intensity
process q i for every i = 1, . . . , 7, so that the process M {i} , given by the formula,
Z t 
{i} {i}
Mt = Ht − 1 − Hu{i} qui du
0

is a G-martingale for every t ∈ [0, T ] and i = 1, . . . , 7.

We now have the following technical result,

Lemma 2.18. The processes

Z t∧τ
{i} {i}
Mti := Mt∧τ = Ht∧τ − lui du, t ≥ 0, i = 1, 2, . . . , 7,
0

and Z t∧τ
Mt := Ht∧τ − lu du, t ≥ 0,
0
where
7
X
lti = I{τ ≥t} qti and lt = lti t ≥ 0, i = 1, 2, . . . , 7,
i=1

are G-martingales

Proof. Fix i = 1, . . . , 7. Process M i follows a G-martingale, since it is G-martingale M {i}

P
stopped at the G stopping time τ. Moreover, we have that Mt = 7i=1 Mti , so that process
M is also a G-martingale.

We shall now proceed with deriving some useful representations for the processes S C
and S.

Lemma 2.19. The ex-dividend price process, S C , of a counterparty risky CDS contract,
given in (4), can be represented as follows,
Z  !
X7 
i 
StC = Bt EQ Bτ−1 I{t<τ =τ {i} ≤T } δ τ − κ Bu−1 I{τ >u} du Gt (11)
]t,T ] 
i=1

16
where
1 2
δt = δt1 , δ t = St − (1 − R2 ) (St − Ct )+
3
δt = St + (1 − R3 ) (St − Ct )− ,
4
δt = St − (1 − R2 ) (St − Ct )+ + (1 − R3 ) (St − Ct )−
5
+ 6
−
δt = δt1 − (1 − R2 ) δt1 − Ct , δ t = δt1 + (1 − R3 ) δt1 − Ct
7
+
−
δt = δt1 − (1 − R2 ) δt1 − Ct + (1 − R3 ) δt1 − Ct .
Proof. Let us rewrite (9) in the following form,

X X
StC = Bt EQ Bτ−1 δτ1 I{t<τ =τ {i} ≤T } + Bτ−1 Sτ I{t<τ =τ {i} ≤T }
i=1,5,6,7 i=2,3,4

+ X
− Bτ−1 (1 − R2 ) Sτ + I{τ =τ 1 } δτ1 − Cτ I{t<τ =τ {i} ≤T }
i=2,4,5,7

−
+ Bτ−1 (1 − R3 ) Sτ + I{τ =τ 1 } δτ1 − Cτ
 
X Z 

I{t<τ =τ {i} ≤T } − κ Bu I{τ >u} du Gt  ,
−1

i=3,4,6,7 ]t,T ] 
which, after rearranging terms, leads to


StC = Bt EQ Bτ−1 δτ1 I{t<τ =τ {1} ≤T } + Bτ−1 St − (1 − R2 ) (Sτ − Cτ )+ I{t<τ =τ {2} ≤T }


+ Bτ−1 Sτ + (1 − R3 ) (Sτ − Cτ )− I{t<τ =τ {3} ≤T }


+ Bτ−1 Sτ − (1 − R2 ) (Sτ − Cτ )+ + (1 − R3 ) (Sτ − Cτ )− I{t<τ =τ {4} ≤T }

+ 
+ Bτ−1 δτ1 − (1 − R2 ) δτ1 − Cτ I{t<τ =τ {5} ≤T }

− 
+ Bτ−1 δτ1 + (1 − R3 ) δτ1 − Cτ I{t<τ =τ {6} ≤T }

+
− 
+ Bτ−1 δτ1 − (1 − R2 ) δτ1 − Cτ + (1 − R3 ) δτ1 − Cτ I{t<τ =τ {7} ≤T }
Z  !


−κ Bu−1 I{τ >u} du Gt .
]t,T ] 
This proves the result.

In case when R2 = R3 = 1 process S is the same as process S C . Thus, we obtain from the
above
Corollary 2.20. The ex-dividend price process S of a counterparty risk-free CDS contract,
can be represented as follows 4 ,
Z  !
X 7 
−1 i −1 
S t = Bt E Q Bτ I{t<τ =τ {i} ≤T } δ̂τ − κ Bu I{τ >u} du Gt , (12)
]t,T ] 
i=1
4
We note that formula (13) provides a representation of St , which is convenient for our purposes. The

17
where δ̂t1 = δ̂t5 = δ̂t6 = δ̂t7 = δt1 , and δ̂t2 = δ̂t3 = δ̂t4 = St . Thus,
4
X
S t = Bt E Q Bτ−1 I{t<τ =τ 1 ≤T } δτ1 + Bτ−1 I{t<τ =τ {i} ≤T } Sτ (13)
i=2
Z  !


−κ Bu−1 I{τ >u} du Gt .
]t,T ] 

The following result is borrowed from [BJR08] (see Lemma 3.1 therein)

Lemma 2.21. The following equality holds (Q a.s.)

   Z  
i Bt T
i 
Bt EQ I{t<τ =τ {i} ≤T } Bτ−1 δ τ  Gt = I{t<τ } EQ Bu−1 lui δ u G (u) du Ft , (14)
G(t) t

for every t ∈ [0, T ].

The pre-default ex-dividend price processes, say Se and SeC , are defined as the (unique)
F-adapted processes (cf. [BJR08]) such that

StC = I{t<τ } SetC , St = I{t<τ } Set .

In view of the above we thus obtain the following result

Lemma 2.22. We have that, for every t ∈ [0, T ],

Z  !!
Bt T 7
X 
i 
SetC = EQ Bu−1 G (u) lui δ u − κ du Ft , (15)
G(t) t 
i=1

and  !!
Z T 7
X 
Bt 
Set = EQ Bu−1 G (u) lui δ̂ui − κ du Ft . (16)
G(t) t 
i=1

Proof. From Lemma 2.19 we have that

 ! Z  !
7
X  
i 
StC = Bt E Q Bτ−1 I{t<τ =τ {i} ≤T } δ t  Gt − κBt EQ Bu−1 I{τ >u} du Gt .
 ]t,T ] 
i=1

Now, in view of (14) we see that

 !  !
7
X  7 Z
X T 
i Bt i 
Bt E Q Bτ−1 I{t<τ =τ {i} ≤T } δ t  Gt = I{t<τ } EQ Bu−1 lui δu G (u) du Ft .
 G(t) t 
i=1 i=1

traditional representation of St , typically used in the context of counterparty risk free CDS contracts is
Z  !

−1 1 −1
St = Bt EQ Bτ1 I{t<τ1 ≤T } δτ1 − κ Bu I{τ1 >u} du Gt

]t,T ] 
.

18
 R −1
Let us now fix t ≥ 0, and define Ys := −κ ]t,s] Bu du for s ≥ t. Thus, we get
Z  !
 


−κBt EQ Bu−1 I{τ >u} du Gt = Bt EQ I{τ >T } YT  Gt
]t,T ] 


+ Bt EQ I{t<τ ≤T } Yτ  Gt .

It is known from [BJR08], that

Z  

Bt T 
Bt E Q I{t<τ ≤T } Yτ  Gt = −I{t<τ } EQ Yu dG(u) Ft
G(t) t

and

Bt
Bt EQ I{τ >T } YT  Gt = I{t<τ } EQ ( G(T )YT | Ft ) .
G(t)
Finally, since Y is of finite variation, (15) follows by applying the integration by parts
formula Z T Z T Z T
G(t)YT − Ys dG (s) = G (s) dYs = −κ G (s) Bu−1 du.
t t t
Equality (16) is obtained as a special case of (15), by setting R2 = R3 = 1.

We are ready now to derive dynamics of the pre-default price processes, that we shall
use in order to derive the dynamics of the CVA process.

Lemma 2.23. (i) The pre-default ex-dividend price of a counterparty risky CDS contract
follows the dynamics given as
!!
7
X  
i
dSeC = (rt + lt ) SeC −
t li δ − κ
t dt + G−1 (t) Bt dmC − SeC dµ
t t t t
i=1


+ G−2 (t) SetC d hµit − Bt d µ, mC t , t ∈ [0, T ],

where  !
!
Z T 7
X 
i 
mC
t = EQ Bu−1 G (u) lui δu − κ du Ft
0 
i=1

(ii) The pre-default ex-dividend price of a counterparty risk-free CDS contract follows the
dynamics given as
!!
X7  
dSet = (rt + lt ) Set − i i
l δ̂ − κ dt + G−1 (t) Bt dmt − Set dµ
t t
i=1
 
+ G−2 (t) Set d hµit − Bt d hµ, mit , t ∈ [0, T ],

where !  !
Z T 7
X 

mt = E Q Bu−1 G (u) lui δ̂ui − κ du Ft .
0 
i=1

19
Proof. The argument below follows the one in the proof of Proposition 1.2 in [BJR08].
In view of (15) we may write SetC as

SetC = Bt G−1 (t)Ut ,

where !
Z t 7
X i
Ut = mC
t − Bu−1 G (u) lui δ u − κ du.
0 i=1
Since G = µ − v, then applying Itô’s formula one obtains
7
!

X i
d G−1 (t)Ut = G−1 (t)dmC −1
t − Bt lti δt − κ dt
i=1


+ Ut G−3 (t)d hµit − G −2
(t) (dµt − dvt )


− G−2 (t)d µ, mC t .

Consequently,

7
!
X i
dSetC = Bt G−1 (t)dmC
t − lti δ t − κ dt
i=1


+ Bt Ut G−3 (t)d hµit − G−2 (t) (dµt − lt G(t)dt)


− Bt G−2 (t)d µ, mC + rt Bt G−1 (t)Ut dt
!!
7
X i
 
= (rt + lt ) SetC − lti δ t − κ dt + G−1 (t) Bt dmC e
t − St dµ
i=1


+G −2
(t) Set d hµit − Bt d µ, mC t ,

which verifies the result stated in (i).

Starting from (16), and using computations analogous to the ones done in (i), one can
derive the result stated in (ii).

The dynamics of the CVA process are easily derived with help of the above lemma,

Proposition 2.24. The bilateral CVA process satisfies,

7
!
X
dCVAt = rt CVAt dt − CVAt− dMt − (1 − Ht ) lti ξti dt
i=1
+ (1 − Ht ) Bt G−1 (t)dnt − G−1 (t)CVAt dµt + G−2 (t)CVAt d hµit



− (1 − Ht ) G−2 (t)Bt d hµ, mit − d µ, mC t ,

where !  !
Z T 7
X 

nt = EQ Bu−1 G (u) lui ξui du Ft , t ∈ [0, T ],
0 
i=1

20
with
ξt1 = 0, ξt2 = (1 − R2 ) (St − Ct )+ , ξt3 = − (1 − R3 ) (St − Ct )− ,
ξt4 = (1 − R2 ) (St − Ct )+ − (1 − R3 ) (St − Ct )− ,

+
−
ξt5 = (1 − R2 ) δt1 − Ct , ξt6 = − (1 − R3 ) δt1 − Ct ,

+
−
ξt7 = (1 − R2 ) δt1 − Ct − (1 − R3 ) δt1 − Ct .
Proof. Applying the integration by parts formula we get that
   
dCVAt = (1 − Ht ) dSet − dSetC − Set − SetC dHt .
This together with Lemma 2.23 implies
!


7
X  
C i
dCVAt = − St− − St− dMt + (1 − Ht ) rt St − StC − i i
lt δ̂t − δ t dt
i=1



+ (1 − Ht ) Bt G−1 (t) dmt − dmC −1 C
t − (1 − Ht ) G (t) St − St dµt


+ (1 − Ht ) G−2 (t) St − StC d hµit



− (1 − Ht ) G−2 (t)Bt d hµ, mit − d µ, mC t ,
which proves the result.

2.3.1 Dynamics of CVA when the immersion property holds

Here we adapt the results derived above to the case when the immersion property holds
between filtrations F and G, that is the case when every F-martingale is a G-martingale
under Q. In this case, the continuous martingale µ in the Doob-Meyer decomposition of G
vanishes, so that the survival process G is a non-increasing process represented as G = −v.
Frequently, the immersion property is referred to as Hypothesis (H). For an excellent
discussion of the immersion property we refer to [JLC09].
Assumption 2.25. Hypothesis (H) holds between the filtrations F and G under Q.
In view of the results (and the notation) from Proposition 2.24 we obtain
Corollary 2.26. Assume that Assumption 2.25 is satisfied. Then,
7
!
X
dCVAt = rt CVAt dt − CVAt− dMt − (1 − Ht ) lti ξti dt
i=1
+ (1 − Ht ) Bt G−1 (t)dnt , t ∈ [0, T ].
Remark 2.27. If we assume that the filtration F is generated by a Brownian motion, then,
in view of the Brownian martingale representation theorem, there exists an F-predictable
process ζ such that dnt = ζt dWt and
7
!
X
dCVAt = rt CVAt dt − CVAt− dMt − (1 − Ht ) lti ξti dt
i=1
−1
+ (1 − Ht ) Bt G (t)ζt dWt .

21
2.4 Fair Spread Value Adjustment
Let us fix t ∈ [0, T ], and let us denote by κt the market spread of the counterparty risk-free
CDS contract at time t; that is, κt is this level of spread that makes the pre-default values
of the two legs of a counterparty risk-free CDS contract equal to each other at time t,

Set (κt ) = 0. (17)

It is convenient to write the above equation in the form that is common in practice:

P Lt − κt RDV 01t = 0, (18)

where P L and RDV 01 are processes representing (pre-default) values of the protection leg
and the risky annuity, respectively, so that 5
    
Z 
Bt X 
P Lt = 1 E Q  −1 1 1
Bu G (u) δu lu du Ft  ,
i
(19)
G (t) ]t,T ]  i=1,5,6,7

and Z  !
Bt 

RDV 01t = 1 EQ Bu−1 G1 (u) du Ft , (20)
G (t) ]t,T ] 
where
G1 (t) = Q ( τ1 > t| Ft ) .
Therefore, we get,
R P   
EQ B −1 G1 (u) δ 1 l i du F
]t,T ] u u i=1,5,6,7 u  t
κt = R   . (21)

EQ ]t,T ∧τ1 ] Bu−1 G1 (u) du Ft

We denote by κCt the spread which makes the values of the two pre-first-default legs of
a counterparty risky CDS contract equal to each other at every t ∈ [0, T ] as


SetC κC
t = P LC C C
t − κt RDV 01t = 0. (22)

Similarly, we use the spread κC

0 initiated at time t = 0 in order to compute the fair price
of a counterparty risky CDS contract at any time t ∈ [0, T ]. Using Lemma 3.1, κC t admits
the following representation for every t ∈ [0, T ] ,

P LC
t
κC
t = ,
RDV 01C
t
5
We note that formula (19) provides a representation of P Lt , which is convenient for our purposes. The
traditional representation of P Lt , typically used in the context of counterparty risk free CDS contracts is
Z  !
Bt

−1 1 1 1
P Lt = 1 EQ Bu G (u) δu λu du Ft ,

G (t) ]t,T ] 

where λ1 is the F intensity of τ1 .

22
where
Z !  !
Bt T 7
X 
i 
P LC
t = EQ Bu−1 G (u) lui δ u du Ft (23)
G (t) t 
i=1
and Z  !
Bt 

RDV 01C
t = EQ Bu−1 G (u) du Ft . (24)
G (t) ]t,T ] 
We may now introduce the following definition,
Definition 2.28. The Spread Value Adjustment process of a counterparty risky CDS con-
tract maturing at time T is defined as,
SVAt = κt − κC
t

for every t ∈ [0, T ] .

Monitoring SVA is of great importance since it provides a more practical way to quantify
the counterparty risk. Moreover, the spread difference is a very useful indicator for the
trading decisions in practice (cf. [Gre09]).
Proposition 2.29. The SVA of a counterparty risky CDS contract maturing at time T
equals,
]t
CVA
SVAt = R  

Bt G−1 (t)EQ ]t,T ] Bu−1 G (u) du Ft
for every t ∈ [0, T ] , where the pre-first-default bilateral Credit Valuation Adjustment process
] is given as
CVA
] t = Set − SetC ,
CVA (25)
for every t ∈ [0, T ].
Proof. Let us rewrite P LC as
P LC C C C
t = P Lt − κt RDV 01t + κt RDV 01t

by a simple modification. Now, using (5) and (22), we conclude that

P LC eC C
t = St (κt ) + κt RDV 01t
] t + κt RDV 01C
= Set (κt ) − CVA t .

Since Set (κt ) = 0, then κC

t has the following form,

] t + κt RDV 01C
−CVA t
κC
t = C
,
RDV 01t
which is
]t
CVA
κC
t =− + κt .
RDV 01C
t

23
2.4.1 SVA Dynamics

Applying Itô formula one obtains the dynamics of the fair spread process and of the coun-
terparty risk adjusted spread process as

1
κt

dκt = Bt−1 G1 (t) κt − δt1 lt1 dt + d η2 t (26)
^01t
RDV ^01t
RDV
!
1
 1

− d η1 , η2 t + dηt1 − κt dηt2 , t ∈ [0, T ],
^01t
RDV ^01t
RDV
where Z  !

^01t := EQ 
RDV Bu−1 G1 (u) du Ft ,
]t,T ] 
Z  !


ηt1 := EQ Bu−1 G1 (u) δu1 lu1 du Ft ,
]0,T ] 
Z  ! Z

 ^01t +
ηt2 = EQ −1 1
Bu G (u) du Ft = RDV Bu−1 G1 (u) du,
]0,T ]  ]0,t]

and
 !
7
X
1 κC

dκC
t = C
Bt−1 G(t) κC
t − δeti lti dt + C
t
d ζ2 t (27)
^01t
RDV i=1 ^01t
RDV

1
 1

− C
d ζ 1, ζ 2
t
+
C
dζt1 − κC 2
t dζt ,
^01t
RDV ^01t
RDV
where Z  !

^01t = EQ 
RDV Bu−1 G1 (u) du Ft ,
]t,T ] 
with !  !
Z 7
X 
i 
ζt1 = EQ Bu−1 G (u) lui δ u du Ft ,
]0,T ] 
i=1
and Z  ! Z
 C
 ^01t +
ζt2 = EQ −1
Bu G (u) du Ft = RDV Bu−1 G (u) du.
]0,T ]  ]0,t]

Combining the above results, we find the dynamics of the SVA process:

d SVAt = dκt − dκC

t , t ∈ [0, T ].

Dynamics of the SVA is of great importance for observing the behavior of the difference
between the fair spread and the counterparty risk adjusted spread. Counterparty risk
dynamics can be assessed in a more intuitive manner by computing the SVA dynamics.

24
3 Multivariate Markovian Default Model
In this section, we propose an underlying stochastic model following the lines of [BCJZ11].
Towards this end we define a Markovian model of multivariate default times with factor


processes X = X 1 , X 2 , X 3 which will have the following key features,

• The pair (X, H) is Markov in its natural filtration,

• Each pair X i , H i is a Markov process,

• At every instant, either each counterparty defaults individually or simultaneously with

other counterparties.

Note that the second property grants quick valuation of the CDS and independent


calibration of each model marginal X i , H i , whereas the third property will allow us to
account for dependence between defaults. We present here some numerical results as an
application of above theory. The default intensities are assumed to be of the affine form


li t, Xti = ai + Xti ,

where ai is a constant and X i is a homogenous CIR process generated by,

q
dXt = ζi µi − Xt dt − σi Xti dWti ,
i i

for i = 1, 2, 3. Each collection of the parameters (ζi , µi , σi ) may take values corresponding
to a low, a medium or a high regime which are given as follows.

Credit Risk Level ζ µ σ X0

Low 0.9 0.001 0.01 0.001
Medium 0.8 0.02 0.1 0.02
High 0.5 0.05 0.2 0.05

Moreover, following the methodology in [BCJZ11], we specify the marginal default in-
tensity processes as follows

qt1 = lt1 + lt5 + lt6 + lt7 , qt2 = lt2 + lt4 + lt5 + lt7 , qt3 = lt3 + lt4 + lt6 + lt7

where the related survival probabilities are found as

 Rt i   Rt 
Q (τi > t) = EQ e− 0 qu du and Q (τ > t) = EQ e− 0 lu du .

For a detailed discussion including implementation and the calibration of the model, we
refer to [BCJZ11] and [ABCJ11].

25
3.1 Results
Our aim here is to assess by means of numerical experiments the impact of collateralization
on the counterparty risk exposure. We present numerical results for different collateraliza-
tion regimes distinguished by different threshold values. The numerical experiments below
have been done using the three factor (2F) parametrization given in [BCJZ11], the recovery
rates are fixed to 40%, the risk-free rate r is taken as 0 and the maturity is set to T = 5
years.
Table 3.1 shows the values of CVA0 and SVA0 for different threshold regimes. Threshold
values are chosen as a fraction of the notional (cf. [Pyk09]). Computations are done
assuming that (refer to Table 3) the underlying entity, the counterparty, and the investor
has high risk levels. Simulated fair spread without counterparty risk is found as 153bps.
Case A represents the uncollateralized regime where there is no collateral exchanged (this
is done by setting the thresholds infinity), whereas other Case F corresponds to the full
collateralization where the thresholds are set to 0. In each case, computations are done
by setting M T A to zero and assuming there is no margin period. One can observe that
decreasing threshold value dramatically decreases the initial CVA and therefore the SVA
values.

Γcpty Γinv CVA0 SVA0

Case A ∞ -∞ 1.01 × 10−4 0.2153
Case B 1.5 × 10−3 0.4 × 10−3 6.13 × 10−5 0.1305
Case C 1 × 10−3 0.2 × 10−3 4.36 × 10−5 0.0931
Case D 0.5 × 10−3 0.1 × 10−3 2.18 × 10−5 0.0464
Case E 0.25 × 10−3 0.05 × 10−3 1.14 × 10−5 0.0243
Case F 0 0 0 0

In Figure 1, we present the EP E and EN E curves for each case A to F, and we also
plot the mean collateral values. Computations are carried out by running 104 Monte Carlo
simulations. It is apparent that the behavior of the EP E and EN E values decreases as a
result of increased collateralization. Note that there are peaks in the collateral value in the
very beginning and through the maturity. This effect can be explained as follows: Observe
from Table 1 that the investor has lower threshold than the counterparty in each cases from
A to F. As a result, having a lower threshold value, investor will be posting collateral before
the counterparty. Therefore, until the counterparty’s exposure reaches the threshold, the
collateral value remains negative; meaning that there will be margin calls for the investor
before the counterparty.
Figure 2 plots the mean of sample CVA paths. Starting from CVA0 we compute the
mean sample paths in each case. The behavior of CVA as a credit hybrid option, as indicated
in Remark 2.9, can be clearly observed in the graphs. CVA values decrease over time as
a result of time decay since the expected loss decreases close to the expiration. The effect
of collateralization on the CVA values is apparent in the graphs. Observe that increased

26
initial threshold values are of great importance since one can significantly reduce the future
CVA values by appropriately setting the collateral thresholds. Moreover, one can also use
dynamic thresholds by linking the threshold values to the counterparties’ default intensities
or credit ratings. In this way, counterparties will have more control on the future values
of the CVA of the CDS contract and dynamically manage the CVA since the collateral
thresholds will be reacting to the changes in the default intensities or credit ratings. This
approach will be further investigated in a future research.
−3 −3
x 10 x 10

2.5 EPE 2.5 EPE

ENE ENE
Collateral Collateral
2 2

1.5 1.5

1 1

0.5 0.5

0 0

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
−3 −3
x 10 x 10

2.5 EPE 2.5 EPE

ENE ENE
Collateral Collateral
2 2

1.5 1.5

1 1

0.5 0.5

0 0

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
−3 −3
x 10 x 10

2.5 EPE 2.5 EPE

ENE ENE
Collateral Collateral
2 2

1.5 1.5

1 1

0.5 0.5

0 0

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Figure 1: EPE, ENE and the Collateral curves for Case A, B, C, D, E, and F

4 Conclusion
In this paper, we discussed the modeling of counterparty risk in the presence of bilateral
margin agreements. We defined an appropriate collateral process which takes various margin
agreement parameters into account. The dynamics of the counterparty risk adjustment,
CVA, has been found for the bilateral case. This achievement helps us to better understand
and monitor the behavior of the bilateral CVA as well as the unilateral CVA and the DVA.
We observed the impact of collateral agreements on counterparty risk adjustments as
well as the credit exposures such as the EPE and the ENE. The presence of simultaneous
defaults in our model represents the wrong way risk involved in the CDS contracts. We
formulate the fair spread value adjustment, which is named as SVA, that indicates the

27
 −5 −5
x 10 x 10
10 10

9 9

8 8

7 7

6 6

5 5

4 4

3 3

2 2

1 1

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
−5 −5
x 10 x 10
10 10

9 9

8 8

7 7

6 6

5 5

4 4

3 3

2 2

1 1

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
−5 −5
x 10 x 10
10 10

9 9

8 8

7 7

6 6

5 5

4 4

3 3

2 2

1 1

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Figure 2: Forward CVA curves for Case A, B, C, D, E, and F

additional spread value to incorporate the counterparty risk into the fair spread value.
Moreover, we derive the dynamics of the fair spread and the counterparty risky spread and
therefore the spread value adjustment, SVA. Finally, as in [BCJZ11] and [ABCJ11], we
present our numerical results using a Markovian model of counterparty credit risk.

References
[ABCJ11] S. Assefa, T.R. Bielecki, S. Crepey, and M. Jeanblanc. CVA computation for
counterparty risk assessment in credit portfolios. In Credit Risk Frontiers:
Subprime Crisis, Pricing and Hedging, CVA, MBS, Ratings, and Liquidity.
Bloomberg, 2011.

[Alg09] Algorithmics. Credit Value Adjustment and the changing environment for pric-
ing and managing counterparty risk. 2009.

[BC09] D. Brigo and A. Capponi. Bilateral counterparty risk valuation with stochastic
dynamical models and application to credit default swaps. Working Paper, 2009.

[BC11] T.R. Bielecki and S. Crepey. Dynamic hedging of counterparty exposure.

In T. Zariphopoulou, M. Rutkowski, and Y. Kabanov, editors, The Musiela
Festschrift. Spring, 2011.

28
[BCJZ11] T.R. Bielecki, S. Crepey, M. Jeanblanc, and B. Zargari. Valuation and Hedging
of CDS Counterparty Exposure in a Markov Copula Model. Forthcoming in
International Journal of Theoretical and Applied Finance, 2011.

[BCPP11] D. Brigo, A. Capponi, A. Pallavicini, and V. Papatheodorou. Collateral

margining in arbitrage-free counterparty valuation adjustment including re-
hypotecation and netting. Working Paper, 2011.

[BJR08] T.R. Bielecki, M. Jeanblanc, and M. Rutkowski. Pricing and trading credit de-
fault swaps in a hazard process model. Annals of Applied Probability, 18(6):2495–
2529, 2008.

[BSP11] Ch. Blanchet-Scalliet and F Patras. Structural counterparty risk valuation for
cds. In Brigo D. Bielecki, T.R. and F. Patras, editors, Credit Risk Frontiers:
Subprime crisis, Pricing and Hedging, CVA, MBS, Ratings and Liquidity. Wiley,
2011.

[CAC10] G. Cesari, J. Aquilina, and N. Charpillon. Modelling, Pricing, and Hedging

Counterparty Credit Exposure. Springer Finance, 2010.

[DPR] B. De Prisco and D. Rosen. Modelling stochastic counterparty credit exposures

for derivatives portfolios. Edited By Michael Pykhtin, RISK Books, London,
2005., Chapter 1 of Counterparty Credit Risk Modelling: Risk Management,
Pricing and Regulation.

[Gre09] John Gregory. Counterparty Credit Risk: The New Challenge for Global Finan-
cial Markets. Wiley, 2009.

[HW01] J. Hull and A. White. Valuing credit default swaps ii: Modeling default correla-
tion. The Journal of derivatives, 8(3):12–22, 2001.

[ISD05] ISDA. ISDA Collateral Guidelines. 2005.

[ISD10a] ISDA. ISDA Margin Survey. 2010.

[ISD10b] ISDA. ISDA Market Review of OTC Derivative Bilateral Collateralization Prac-
tices. 2010.

[JLC09] M. Jeanblanc and Y. Le Cam. Immersion property and credit risk modelling. In
Optimality and Risk - Modern Trends in Mathematical Finance, pages 99–132.
Springer, 2009.

[JY01] R.A. Jarrow and F. Yu. Counterparty risk and the pricing of defaultable secu-
rities. the Journal of Finance, LVI(5):1765–1799, Oct. 2001.

[LK05] S.Y. Leung and Y.K. Kwok. Cds valuation with counterparty risk. The Kyoto
Economic Review, 74(1):24–45, 2005.

29
[LS09] A. Lipton and A. Sepp. Credit value adjustment for credit default swaps via the
structural default model. The Journal of Credit Risk, 5(2), Summer 2009.

[Mar09] Markit. The CDS Big Bang: Understanding the Changes to the Global CDS
Contract and North American Conventions. 2009.

[Pyk09] M. Pykhtin. Modeling Credit Exposure for Collateralized Counterparties. Jour-

nal of Credit Risk, 5(4):3–27, 2009.

[SL11] H. Stein and K.P Lee. Counterparty valuation adjustments. In Brigo D. Bielecki,
T.R. and F. Patras, editors, Credit Risk Frontiers: Subprime crisis, Pricing and
Hedging, CVA, MBS, Ratings and Liquidity,. Wiley, 2011.

30