This action might not be possible to undo. Are you sure you want to continue?
com/abstract=1782063
Finance and Economics
Discussion Series
Division of Research & Statistics and Monetary Affairs
Federal Reserve Board, Washington, DC
Pricing Counterparty Risk at the Trade
Level and CVA Allocations
Michael Pykhtin and Dan Rosen
201010
Z.11
Electronic copy available at: http://ssrn.com/abstract=1782063
Finance and Economics Discussion Series
Divisions of Research & Statistics and Monetary Aﬀairs
Federal Reserve Board, Washington, D.C.
Pricing Counterparty Risk at the Trade Level and CVA
Allocations
Michael Pykhtin and Dan Rosen
201010
NOTE: Staﬀ working papers in the Finance and Economics Discussion Series (FEDS) are preliminary
materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth
are those of the authors and do not indicate concurrence by other members of the research staﬀ or the
Board of Governors. References in publications to the Finance and Economics Discussion Series (other than
acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.
Electronic copy available at: http://ssrn.com/abstract=1782063
1
Pricing Counterparty Risk at the Trade Level and CVA Allocations
1
Michael Pykhtin
2
and Dan Rosen
3
November 2009
Abstract
We address the problem of allocating the counterpartylevel credit valuation
adjustment (CVA) to the individual trades composing the portfolio. We show that
this problem can be reduced to calculating contributions of the trades to the
counterpartylevel expected exposure (EE) conditional on the counterparty’s
default. We propose a methodology for calculating conditional EE contributions
for both collateralized and noncollateralized counterparties. Calculation of EE
contributions can be easily incorporated into exposure simulation processes that
already exist in a financial institution. We also derive closedform expressions for
EE contributions under the assumption that trade values are normally distributed.
Analytical results are obtained for the case when the trade values and the
counterparty’s credit quality are independent as well as when there is a
dependence between them (wrongway risk).
1
The opinions expressed here are those of the authors and do not necessarily reflect the views or policies of their
employers.
2
Federal Reserve Board, Washington, DC, USA. michael.v.pykhtin@frb.gov.
3
R
2
Financial Technologies and The Fields Institute for Research in Mathematical Sciences, Toronto, Canada.
dan.rosen@R2financial.com and drosen@fields.utoronto.ca
2
1. Introduction
For years, the standard practice in the industry was to mark derivatives portfolios to
market without taking the counterparty credit quality into account. In this case, all cash flows are
discounted using the LIBOR curve, and the resulting values are often referred to as riskfree
values.
4
However, the true value of the portfolio must incorporate the possibility of losses due to
counterparty default. The credit valuation adjustment (CVA) is, by definition, the difference
between the riskfree portfolio value and the true portfolio value that takes into account the
counterparty’s default. In other words, CVA is the market value of counterparty credit risk.
5
There are two approaches to measuring CVA: unilateral and bilateral (see Picoult, 2005
or Gregory, 2009). Under the unilateral approach, it is assumed that the counterparty that does
the CVA analysis (we call this counterparty a bank throughout the paper) is defaultfree. CVA
measured this way is the current market value of future losses due to the counterparty’s potential
default. The problem with unilateral CVA is that both the bank and the counterparty require a
premium for the credit risk they are bearing and can never agree on the fair value of the trades in
the portfolio. Bilateral CVA takes into account the possibility of both the counterparty and the
bank defaulting. It is thus symmetric between the bank and the counterparty, and results in an
objective fair value calculation.
Under both, the unilateral and bilateral approaches, CVA is measured at the counterparty
level. However, it is sometimes desirable to determine contributions of individual trades to the
counterpartylevel CVA. The problem of calculating CVA contributions bears many similarities
to the calculation of risk contributions and capital allocation (see Aziz and Rosen 2004, Mausser
and Rosen 2007). There are several possible measures of CVA contributions. We refer to the
CVA of each transaction on a standalone basis as the transaction’s standalone CVA. Clearly,
when the given portfolio does not allow for netting between trades, the portfoliolevel CVA is
given by the sum of the individual trades’ standalone CVA. However, this is not the case when
netting and margin agreements are in place. We refer to the incremental CVA contribution of a
trade as the difference between the portfolio CVA with and without the trade.
6
This measure is
commonly seen as appropriate for pricing counterparty risk for new trades with the counterparty
(see Chapter 6 in Arvanitis and Gregory, 2001 for details). One problem with incremental CVA
contributions is that they are nonadditive – the sum of the individual trade’s CVA contributions
does not add up to the portfolio’s CVA. Hence neither standalone nor incremental contributions
can be used effective contributions of existing trades in the portfolio to the counterpartylevel
CVA, in the presence of netting and/or margin agreements. For this purpose we require additive
CVA contributions. In this case, we draw the analogy with the capital allocation literature and
refer to these as (continuous) marginal risk contributions.
4
More precisely, LIBOR rates roughly correspond to AA risk rating and incorporate the typical credit risk of large
banks.
5
See Canabarro and Duffie (2003) or Pykhtin and Zhu (2007) for an introduction to counterparty credit risk and
CVA.
6
Sometimes these are referred to as discrete marginal contributions.
3
The marginal CVA contributions with a given counterparty give the bank a clear picture
how much each trade contributes to the counterpartylevel CVA. However, the use of CVA
contributions is not limited to an analysis at a single counterparty level. Once the CVA
contributions have been calculated for each counterparty, the bank can calculate the price of
counterparty credit risk in any collection of trades without any reference to the counterparties.
For example, by selecting all trades booked by a certain business unit or product type (e.g., all
CDSs or all USD interest rate swaps), the bank can determine the contribution of that business
unit or product to the bank’s total CVA.
We show how to define and calculate marginal CVA contributions in the presence of
netting and margin agreements, and under a wide range of assumptions, including the
dependence of exposure on the counterparty’s credit quality. The theory of marginal risk
contributions, sometimes refer to as Euler Allocations (see Tasche 2008), is now well developed
and largely relies on the risk function being homogeneous (of degree one). We show that this
principle can be applied readily for CVA when the counterparty portfolio allows for netting (but
does not include collateral and margins). We further extend this allocation principle for the more
general case of collateralized/margined counterparties For the sake of simplicity, we assume the
unilateral framework throughout the paper. However, an extension of all the results to the
bilateral framework is straightforward.
The paper is organized as follows. In Section 2, we define counterparty credit exposure
for both collateralized and noncollateralized cases. We show how counterpartylevel CVA can
be calculated from the profile of the discounted risk neutral expected exposure (EE) conditional
on the counterparty’s default. In Section 3, we introduce CVA contributions of individual trades
and relate them to the profiles of conditional EE contributions. In Section 4, we adapt the
continuous marginal contribution (CMC) method often used for allocating economic capital to
calculating EE contributions for the case when the counterpartylevel exposure is a homogeneous
function of the trades’ weights in the portfolio. This is the case when there are no exposure
limiting agreements, such as margin agreements, with the counterparty. When such agreements
are present, the CMC method fails because the counterpartylevel exposure is not homogeneous
anymore. In Section 5, we propose an EE allocation scheme that is based on the CMC method,
but can be used for collateralized counterparties. In Section 6, we show how to incorporate EE
and CVA contribution calculations into exposure simulation process. In Section 7, we derive
closed form expressions for EE contributions under the assumption that all trade values are
normally distributed. We start with the case of independence between exposure and the
counterparty’s credit quality, and extend the results to incorporate dependence between them
(wrongway risk). We also provide an intuitive explanation to our closedform results. In Section
8, we show several numerical examples that illustrate the behavior of exposure (and hence CVA)
contributions for both, the collateralized and noncollateralized cases.
4
2. Counterparty credit risk and CVA
In this section, we review the basic concepts and notation for counterparty credit risk,
credit exposures and CVA.
Counterparty credit risk (CCR) is the risk that the counterparty defaults before the final
settlement of a transaction's cash flows. An economic loss occurs if the counterparty portfolio
has a positive economic value for the bank at the time of default. Unlike a loan, where only the
lending bank faces the risk of loss, CCR creates a bilateral risk: the market value can be positive
or negative to either counterparty and can vary over time with the underlying market factors. We
define the counterparty exposure ( ) E t of the bank to a counterparty at time t as the economic
loss, incurred on all outstanding transactions with the counterparty if the counterparty defaults at
t , accounting for netting and collateral but unadjusted by possible recoveries.
2.1 Counterparty exposures
Consider a portfolio of N derivative contracts of a bank with a given counterparty. The
maturity of the longest contract in the portfolio is T . The counterparty defaults at a random time
τ with a known riskneutral distribution ( ) Pr[ ] P t t τ ≡ ≤ .
7
We further assume that the
distribution of the trade values at all future dates is risk neutral.
8
Denote the value of the ith instrument in the portfolio at time t from the bank’s
perspective by ( )
i
V t . At each time t, the counterpartylevel exposure ( ) E t is determined by the
values of all trades with the counterparty at time t,
1
{ ( )}
N
i i
V t
=
. The value of the counterparty
portfolio at t is given by
1
( ) ( )
N
i
i
V t V t
=
=
∑
(1)
When netting is not allowed, the (gross) counterpartylevel exposure ( ) E t is
( ) { }
1
max 0, ( )
N
i
i
E t V t
=
=
∑
(2)
For a counterparty portfolio with a single netting agreement, the (netted) exposure is
{ } ( ) max ( ), 0 E t V t = (3)
7
The term structure of risk neutral probabilities of default can be obtained from credit default swaps spreads quoted
for the counterparty on the market for different of different maturities. See, for example, Schönbucher (2003).
8
See, for example, Brigo and Masetti (2005).
5
When the netting agreement is further supported by a margin agreement, the counterparty
must provide the bank with collateral whenever the portfolio value exceeds a threshold. As the
portfolio value drops below the threshold, the bank returns collateral to the counterparty.
Collateral transfer occurs only when the collateral amount that needs to be transferred exceeds a
minimum transfer amount. The counterpartylevel (margined) exposure is given by
{ } ( ) max ( ) ( ), 0 E t V t C t = − (4)
where ( ) C t is the collateral available to the bank at time t.
Counterparty portfolios with a combination of multiple netting agreements and trades
outside of these agreements can be modeled in a straightforward way by a combination of
Equations (2)(4).
2.2 Models of Collateral
We start modeling collateral with a simplifying assumption: we incorporate the minimum
transfer amount into the threshold H and treat the margin agreement as having no minimum
transfer amount. This approximation is rather crude, but it is very popular amongst banks
because it greatly simplifies modeling.
We consider two models of collateral. In the instantaneous collateral model, we assume
that collateral is delivered immediately and that the trades can be liquidated immediately as well.
Under these simplifying assumptions, the collateral available to the bank is
{ } ( ) max ( ) , 0 C t V t H = − (5)
The instantaneous collateral model is attractive because of its simplicity, but is rarely used in
practice because its assumptions materially affect the exposure distribution.
9
However, we use
this model to show the simple, intuitive interpretation of our results for collateralized netting.
A more realistic collateral model must account for the time lag between the last margin
call made before default and the settling of the trades with the defaulting counterparty. This time
lag, which we denote by t δ , is known as the margin period of risk. While the margin period of
risk is not known with certainty, we follow the standard practice and assume that it is a
deterministic quantity that is defined at the margin agreement level.
10
We assume that the
collateral available to the bank at time t is determined by the portfolio value at time t t δ −
according to
{ } ( ) max ( ) , 0 C t V t t H δ = − − (6)
9
When the threshold is not too small, the instantaneous collateral model works reasonably well for expected
exposure. See Pykhtin (2009).
6
We refer to this more realistic model as the lagged collateral model. While more difficult to
implement, it is often used by banks to obtain results, which have more practical value.
2.3 Credit losses and CVA
In the event that the counterparty defaults at time τ , the bank recovers a fraction R of the
exposure ( ) E τ . The bank’s discounted loss due to the counterparty’s default is
{ }
(1 ) ( ) ( ) 1
T
L R E D
τ
τ τ
≤
= − (7)
where
{ }
1
A
is the indicator function that takes value 1 when logical variable A is true and value 0
otherwise, ( ) D t is the stochastic discount factor process at time t, defined according to
( )
t
B B t D
0
= , with
t
B the value of the money market account at time t.
The unilateral counterpartylevel CVA is obtained by applying the expectation to
Equation (7). This results in
0
ˆ CVA (1 ) ( ) ( )
T
R dP t e t
∗
= −
∫
(8)
where ˆ ( ) e t
∗
is the riskneutral discounted expected exposure (EE) at time t, conditional on the
counterparty’s default at time t:
[ ]
ˆ
ˆ ( ) E ( ) ( ) E ( ) ( )
t
e t D t E t D t E t t τ
∗
= ≡ = (
¸ ¸
(9)
Throughout this paper we use “star” to designate discounting and “hat” to designate conditioning
on default at time t. Note that we have not made so far any assumptions on whether the exposure
depends on the counterparty’s credit quality.
3. CVA Contributions from EE Contributions
We would like to develop a general approach to calculating additive contributions of
individual trades to the counterpartylevel CVA. We denote the contribution of trade i by CVA
i
.
We say that CVA contributions are additive when they sum up to the counterpartylevel CVA:
1
CVA CVA
N
i
i =
=
∑
(10)
10
The margin period of risk depends on the contractual margin call frequency and the liquidity of the portfolio. For
example, 2 t δ = weeks is usually assumed for portfolios of liquid contracts and daily margin call frequency.
7
Note that the recovery rate R and the default probabilities ( ) P t are defined at the counterparty
level in Equation (8). Thus, the problem of calculating CVA contributions reduces to that of
calculating contributions of individual trades to the portfolio conditional discounted EE, ˆ ( )
i
e t
∗
, at
each future date. To obtain additive CVA contributions, then the conditional discounted EE
contributions must sum up to the portfolio conditional discounted EE:
1
ˆ ˆ ( ) ( )
N
i
i
e t e t
=
∗ ∗
=
∑
(11)
and the CVA contribution of trade i can be calculated from its EE contribution according to
0
ˆ CVA (1 ) ( ) ( )
T
i i
R dP t e t
∗
= −
∫
(12)
Thus, from now on we focus on defining and calculating EE contributions.
Note first that, without netting agreements, the allocation of the counterpartylevel EE
across the trades is trivial because the counterpartylevel exposure is the sum of the standalone
exposures (Equation (2)) and expectation is a linear operator. Furthermore, when there is more
than one netting set with the counterparty (e.g., multiple netting agreements, nonnettable
trades), we can focus on first calculating the CVA contribution of a transaction to its netting set.
The allocation of the counterpartylevel EE across the netting sets is then trivial again because
the counterpartylevel exposure is defined as the sum of the nettingsetlevel exposures. Thus,
our goal is to allocate the nettingsetlevel exposure to the trades belonging to that netting set. To
keep the notation simple, we assume from now on that all trades with the counterparty are
covered by a single netting set.
4. Additive EE Contributions for Noncollateralized Netting Sets
In this section, we develop the basic methodology to compute EE contributions and
allocate portfoliolevel EE for noncollateralized netting sets.
4.1 Continuous Marginal Contributions and Euler Allocation
We derive EE contributions by adapting the continuous marginal contributions (CMC)
method from the economic capital (EC) literature. EC is calculated at the portfolio level and then
it is allocated to individual obligors and transactions. Under the CMC method, the risk
contribution of a given transaction to the portfolio EC is determined by the infinitesimal
increment of the EC corresponding to the infinitesimal increase of the transaction’s weight in the
portfolio (see Chapter 4 in Arvanitis and Gregory (2001) or Tasche (2008) for details). This
follows from the fact that the risk function is homogeneous (of degree one) and the application of
Euler’s theorem.
8
A real function ( ) f x of a vector
1
( , ... , )
N
x x = x is said to be homogeneous of degree β
if for all 0 c > , ( ) ( ) f c c f
β
= x x . If the function ( ) f ⋅ is piecewise differentiable, then Euler’s
theorem states that:
1
( )
( )
N
i
i
i
f
f x
x
β
=
∂
⋅ = ⋅
∂
∑
x
x (13)
The risk measures most commonly used, such as standard deviation, valueatrisk (VaR) and
expected shortfall, are homogeneous functions of degree one ( 1 β = ) in the portfolio positions.
Thus, Euler’s theorem is applied to allocate EC and compute risk contributions across portfolios.
If x denotes the vector of positions in a portfolio, and EC( ) x the corresponding
economic capital, then Euler’s theorem implies additive capital contributions
1
EC( ) EC ( )
N
i
i =
=
∑
x x (14)
where the terms
i
EC( )
EC ( )
i i
x
x
∂
= ⋅
∂
x
x (15)
are referred to as the marginal capital contributions of the portfolio.
4.2 Continuous Marginal EE Contributions for netted exposures without collateral
Consider now the calculation of EE contributions. Assume that we can adjust the size of
any trade in the portfolio by any amount. Define the weight
i
α for trade i as a scale factor that
represents the relative size of the trade in the portfolio, ( , ) ( )
i i i i
V t V t α α = . These weights can
assume any real value, with 1
i
α = corresponding to the actual size of the trade and 0
i
α = being
the complete removal of the trade. We describe adjusted portfolios via the vector of weights
1
( , , )
N
α α = α K . For adjusted portfolios, we use the notations ( , ) E t α , ˆ ( , ) e t
∗
α , andCVA( ) α for
the exposure and EE at time t and CVA. Furthermore, for convenience, denote by (1, ,1) = K 1
the vector representing the original portfolio.
When there is no margin agreement between the bank and the counterparty, the
counterpartylevel exposure is a homogeneous function of degree one in the trade weights:
( , ) ( , ) E c t cE t = α α (16)
9
The intuition behind Equation (16) is simple: if the bank uniformly doubles the size of its
portfolio with the counterparty by entering into exactly the same trade with the counterparty for
each existing trade, the bank’s exposure doubles.
We define the continuous marginal EE contribution of trade i at time t as the infinitesimal
increment of the conditional discounted EE of the actual portfolio at time t resulting from an
infinitesimal increase of trade i’s presence in the portfolio, scaled to the full trade amount:
0
ˆ ˆ ˆ ( , ) ( ) ( , )
ˆ ( ) lim
i
i
i
e t e t e t
e t
δ
δ
δ α
→
∗ ∗ ∗
∗
=
+ ⋅ − ∂
= =
∂
α 1
1 u α
(17)
where
i
u describes a portfolio whose only component is one unit of trade i. Since the portfolio
exposure is homogeneous in the trades’ weights, the EE contributions defined by Equation (17)
automatically sum up to the counterpartylevel conditional discounted EE by Euler’s theorem
(Equation (13)).
We can derive an expression for the marginal EE contributions as follows. First,
substitute Equation (9) into Equation (17) and bring the derivative inside the expectation. This
results in
( , )
ˆ
ˆ ( ) E ( )
i t
i
E t
e t D t
α
∗
=
(
∂
=
(
∂
(
¸ ¸ α 1
α
(18)
where exposure of the adjusted portfolio (with weight vector
1
( , , )
N
α α = α K ) is given by
1
( , ) max ( ), 0
N
i i
i
E t V t α
=
¦ ¹
=
´ `
¹ )
∑
α (19)
Calculating the first derivative of the exposure with respect to the weight
i
α and setting all
weights to one, we have:
{ }
( , ) 0 ( ) 0
( , ) ( , )
max ( , ), 0 ( )
{ } { }
1 1
i
i i i
V t V t
E t V t
V t V t
α α α
> >
= = =
∂ ∂ ∂
= = =
∂ ∂ ∂
α
α 1 α 1 α 1
α α
α (20)
Substituting Equation (20) into Equation (18), we obtain the EE contribution of trade i:
( ) 0
ˆ
ˆ ( ) E ( ) ( )
{ }
1
i t i
V t
e t D t V t
∗
>
(
=
(
¸ ¸
(21)
The EE contribution of trade i is the expectation of a function which considers the discounted
values of the trade on all scenarios where the total counterparty exposure is positive, or zero
otherwise. As expected, the EE contributions sum up to the counterpartylevel discounted EE:
10
{ }
1
( ) 0
ˆ ˆ
ˆ ˆ ( ) E ( ) ( ) E ( ) max ( ), 0 ( )
{ }
1
N
i t t
i
V t
e t D t V t D t V t e t
=
∗ ∗
>
(
= = = (
¸ ¸
(
¸ ¸
∑
5. Additive EE Contributions for Collateralized Netting Sets
Consider now a counterparty that has a single netting agreement supported by a margin
agreement, which covers all the trades with the counterparty. As discussed in Section 2, the
counterpartylevel stochastic exposure is given by Equation (4), where the collateral available to
the bank is given either by the instantaneous collateral model (Equation (5)) or by the lagged
collateral model (Equation (6)). In what follows, we specify additive EE contributions for both
models, starting with the simpler instantaneous collateral model.
5.1 Instantaneous Collateral Model
Substituting Equation (5) into Equation (4), we obtain
{ } { }
0 ( ) ( )
( ) ( )1 1
V t H V t H
E t V t H
< < >
= + (22)
As can be seen from Equation (22), the expected exposure in not a homogeneous function of the
trades’ weights and, hence, the CMC approach cannot be applied directly. From the
mathematical point of view, the conditions of Euler’s theorem are not satisfied, and the CMCs,
as given earlier, do not sum to the counterpartylevel discounted EE anymore. To understand
conceptually how the CMC method fails, notice that when the portfolio value is above the
threshold, the counterpartylevel exposure equals the threshold. An infinitesimal increase of any
trade’s weight in the portfolio is not sufficient to bring the portfolio value below the threshold.
Thus, the counterpartylevel exposure is not affected by the infinitesimal weight changes, and the
exposure contribution is zero for all scenarios with the portfolio value above the threshold.
However, we still would like to “allocate” the nonzero collateralized counterpartylevel
exposure (equal to the threshold) to the individual trades, so that these allocations cannot be all
equal to zero.
We can derive additive contributions for this nonhomogeneous case, which are
consistent with the continuous marginal contributions as follows. First, notice that, while the
exposure function in Equation (22) is not homogeneous in the vector of weights
1
( , , )
N
α α = α K ,
the function
( ) ( )
( ) { } ( ) { }
0 , ,
, , 1 1
H
H H
V t H V t H
E t V t H
α α
α
< < >
′ = ⋅ + ⋅
α α
α α (23)
is a homogeneous function in the extended vector of weights
1
' ( ,..., , )
N H
α α α = α . That is, we
consider scaling the each of the trades as well as the threshold H. Thus, we can think of the
contribution of the threshold itself to the counterpartylevel exposure.
The first derivatives of the exposure with respect to the trade weights is given by
11
{ }
0 ( )
( , )
( ) 1
i
i
V t H
E t
V t
α
< ≤
= ′
′ ∂
= ⋅
∂
α 1
α
(24)
Similarly, the derivative with respect to the threshold weight is
{ }
( )
( , )
1
H
V t H
E t
H
α
>
= ′
′ ∂
= ⋅
∂
α 1
α
(25)
Note that these sum up to the counterpartylevel exposure given by Equation (22), as expected.
By applying discounting and taking conditional expectation of the righthand side of Equations
(24) and (25), we obtain the EE contributions of the trades
{ }
,
0 ( )
ˆ
ˆ ( ) E ( ) ( ) 1
i H t i
V t H
e t D t V t
∗
< ≤
(
= ⋅ ⋅
(
¸ ¸
(26)
and of the threshold
{ }
( )
ˆ
ˆ ( ) E ( ) 1
H t
V t H
e t H D t
∗
>
(
= ⋅ ⋅
(
¸ ¸
(27)
which satisfy
,
1
ˆ ˆ ˆ ( ) ( ) ( )
N
i H H
i
e t e t e t
=
∗ ∗ ∗
= +
∑
(28)
The contribution of the threshold can be interpreted as the change of the conditional discounted
EE associated with an infinitesimal shift of the threshold upwards scaled up by the actual size of
the threshold. Note that, when the threshold goes to infinity, the last term vanishes and we
recover the uncollateralized contributions.
As the final step, we “allocate back” the contribution adjustment of the collateral
threshold given by Equation (27) to the individual trades, so that Equation (28) can be written in
terms of EE contributions only of the trades (as in Equation (11)):
1
ˆ ˆ ( ) ( )
N
i
i
e t e t
=
∗ ∗
=
∑
There are several possibilities for allocating the amount ˆ ( )
H
e t
∗
in meaningful proportions
to each trade. Given that that the adjustment of the trade contributions occurs when the portfolio
value exceeds the threshold, a meaningful weighting scheme is given by the ratio of the
individual instrument’s expected discounted value when the threshold is crossed to the total
counterparty discounted value when this occurs:
12
{ } { }
{ }
{ }
1
( )
( ) ( )
( )
ˆ
E ( ) ( ) 1
ˆ ˆ
E ( ) 1 E ( ) 1
ˆ
E ( ) ( ) 1
N
t i
t t
i
t
V t H
V t H V t H
V t H
D t V t
D t D t
D t V t
=
>
> >
>
(
⋅ ⋅
(
¸ ¸
( (
⋅ = ⋅ ⋅
( (
¸ ¸ ¸ ¸
(
⋅ ⋅
(
¸ ¸
∑
(29)
Thus, the individual trade contributions to EE are given by
{ }
{ }
{ }
{ }
( )
0 ( ) ( )
( )
ˆ
E ( ) 1
ˆ ˆ
ˆ ( ) E ( ) ( ) 1 E ( ) ( ) 1
ˆ
E ( ) 1
t
i t i t i
t
V t H
V t H V t H
V t H
H D t
e t D t V t D t V t
D t V
>
∗
< < >
>
(
⋅ ⋅
(
¸ ¸
( (
= ⋅ + ⋅ ⋅
( (
¸ ¸ ¸ ¸
(
⋅ ⋅
(
¸ ¸
(30)
Both terms of Equation (30) have a straightforward interpretation: the first term is the
contribution of all scenarios where the bank holds no collateral at time t, while the second term is
the contribution of all scenarios where the bank holds nonzero collateral at time t.
We refer to the allocation scheme above as type A allocation. An alternative allocation
scheme (type B) is obtained by bringing the weighting scheme of the threshold contribution now
inside the expectation operator, so that instead of Equation (29) we now have:
{ } { } { }
1
1 1
( ) ( ) ( )
( )
( )
ˆ ˆ ˆ
E ( ) 1 E ( ) 1 E ( ) 1
( ) ( )
N
N N
i
i i
t t t
i i
V t H V t H V t H
V t
V t
D t D t D t
V t V t
=
= =
> > >
(
(
(
( ⋅ = ⋅ ⋅ = ⋅ ⋅
(
(
¸ ¸
(
¸ ¸
¸ ¸
∑
∑ ∑
(31)
This leads to the continuous marginal contributions given by
{ } { }
*
0 ( ) ( )
( )
ˆ ˆ
ˆ ( ) E ( ) ( ) 1 E ( ) 1
( )
i
i t i t
V t H V t H
V t
e t D t V t H D t
V t
< < >
(
(
= ⋅ + ⋅ ⋅ ⋅
(
(
¸ ¸
¸ ¸
(32)
Both terms on the right hand of Equation (32) have the same interpretation as before.
5.2 Lagged Collateral Model
We now apply the formalism developed for the continuous collateral model to the lagged
collateral model. In this case, the counterparty credit exposure is obtained by substituting
Equation (6) into Equation (4):
{ } { }
0 ( ) ( ) 0 ( ) ( )
( ) ( ) 1 [ ( )] 1
V t H V t H V t V t
E t V t H V t
δ δ
δ
< ≤ + < + <
= ⋅ + + ⋅ (33)
where ( ) ( ) ( ) V t V t V t t δ δ = − − is the change of the portfolio value from the lookback time point
t t δ − to the time point of interest t . Note that as the margin period of risk, t δ , vanishes, this
expression reduces to Equation (22), which gives the exposure with instantaneous collateral.
13
Indeed, a zero margin period of risk implies a zero change of portfolio value from t t δ − to t ,
since both time points are the same now.
As defined for the instantaneous model, we rewrite the exposure given by Equation (33)
as a homogeneous function in the extended vector of weights
1
' ( ,..., , )
N H
α α α = α :
{ } { }
0 ( , ) ( , ) 0 ( , ) ( , )
( , ) ( , ) 1 [ ( , )] 1
H H
H
V t H V t H V t V t
E t V t H V t
α δ α δ
α δ
< ≤ + < + <
′ = ⋅ + + ⋅
α α α α
α α α (34)
where
1
( , ) ( )
N
i i
i
V t V t δ α δ
=
=
∑
α (35)
and ( ) ( ) ( )
i i i
V t V t V t t δ δ = − − . The first derivatives of exposure with respect to the trades’
weights and with respect to the threshold weight are given, respectively, by
{ } { }
0 ( ) ( ) 0 ( ) ( )
( , )
( ) 1 ( ) 1
i i
i
V t H V t H V t V t
E t
V t V t
δ δ
δ
α
< ≤ + < + <
= ′
′ ∂
= ⋅ + ⋅
∂
α 1
α
(36)
{ }
0 ( ) ( )
( , )
1
i
H V t V t
E t
H
δ
α
< + <
= ′
′ ∂
= ⋅
∂
α 1
α
(37)
The sum these first derivatives across all the trades and the threshold, gives the counterparty
level exposure, Equation (33). By applying discounting and taking the conditional expectation
of the righthand side of Equations (36) and (37), we obtain the EE contributions of the trades
{ } { }
,
0 ( ) ( ) 0 ( ) ( )
ˆ ˆ
ˆ ( ) E ( ) ( ) 1 E ( ) ( ) 1
i H t i t i
V t H V t H V t V t
e t D t V t D t V t
δ δ
δ
∗
< ≤ + < + <
( (
= ⋅ ⋅ + ⋅ ⋅
( (
¸ ¸ ¸ ¸
(38)
and of the threshold
{ }
0 ( ) ( )
ˆ
ˆ ( ) E ( ) 1
H t
H V t V t
e t H D t
δ
∗
< + <
(
= ⋅ ⋅
(
¸ ¸
(39)
Now we need to allocate back the threshold contribution, Equation (39), to the individual
trades. Following the type A allocation scheme in the previous section, ˆ ( )
H
e t
∗
is allocated to
individual trades in proportion to the expectation of the discounted trade values when
0 ( ) ( ) H V t V t δ < + < . This results in the trade allocations given by
14
{ } { }
{ }
{ }
{ }
0 ( ) ( ) 0 ( ) ( )
0 ( ) ( )
0 ( ) ( )
0 ( ) ( )
ˆ ˆ
ˆ ( ) E ( ) ( ) 1 E ( ) ( ) 1
ˆ
E ( ) ( ) 1
ˆ
E ( ) 1
ˆ
E ( ) ( ) 1
i t i t i
t i
t
t
V t H V t H V t V t
H V t V t
H V t V t
H V t V t
e t D t V t D t V t
D t V t
H D t
D t V t
δ δ
δ
δ
δ
δ
∗
< ≤ + < + <
< + <
< + <
< + <
( (
= ⋅ ⋅ + ⋅ ⋅
( (
¸ ¸ ¸ ¸
(
⋅ ⋅
(
( ¸ ¸
+ ⋅ ⋅ ⋅
(
( ¸ ¸
⋅ ⋅
(
¸ ¸
(40)
In the type B allocation, we bring the weighting scheme inside the expectation operator. This
gives
{ } { }
{ }
0 ( ) ( ) 0 ( ) ( )
0 ( ) ( )
ˆ ˆ
ˆ ( ) E ( ) ( ) 1 E ( ) ( ) 1
( )
ˆ
E ( ) 1
( )
i t i t i
i
t
V t H V t H V t V t
H V t V t
e t D t V t D t V t
V t
H D t
V t
δ δ
δ
δ
∗
< ≤ + < + <
< + <
( (
= ⋅ ⋅ + ⋅ ⋅
( (
¸ ¸ ¸ ¸
(
+ ⋅ ⋅ ⋅
(
¸ ¸
(41)
It is straightforward to verify that the lagged EE contributions degenerate to the
instantaneous EE contributions when 0 t δ = . Substituting ( ) 0 V t δ = into Equations (40) and
(41), we obtain Equations (30) and (32), respectively.
6. Calculating CVA Contributions by Simulation
Banks commonly use Monte Carlo simulation in practice to obtain the distribution of
counterpartylevel exposures. Based on these simulations a bank can also compute the
counterpartylevel CVA. In this section, we show how the calculation of EE contributions can be
easily incorporated to the Monte Carlo simulation of the counterpartylevel exposure that banks
already perform.
6.1 Exposure Independent of Counterparty’s Credit Quality
Consider first the case where the exposures are independent of the counterparty’s credit
quality. In general, banks implicitly assume that each counterparty’s exposure is independent of
that counterparty’s credit quality when exposures are simulated separately. Let us now make this
assumption explicitly. Then, conditioning on t τ = in the expectations in Equations (19), (29),
(30) and (32) become unconditional, and these conditional expectations can be replaced by the
unconditional ones.
The simulation algorithm for calculating counterpartylevel CVA can be extended to
calculate CVA contributions. For the ease of exposition, we assume that all the trades with the
counterparty are nettable and that collateral (if there is any) can be described by the
instantaneous model.
First, the counterpartylevel CVA can be calculated in a Monte Carlo simulation as
follows:
15
1. Generate a market scenarios j (interest rates, FX rates, etc.) for each of the future
time points
k
t
2. For each simulation time point
k
t and scenario j :
a. For each trade i, calculate trade value
( )
( )
j
i k
V t
b. Calculate portfolio value
( ) ( )
1
( ) ( )
N
j j
k i k
i
V t V t
=
=
∑
c. If there is margin agreement, calculate collateral
( ) ( )
( ) max{ ( ) , 0}
j j
k k
C t V t H = − available at time
k
t .
d. Calculate counterpartylevel exposure
( ) ( ) ( )
( ) max{ ( ) ( ), 0}
j j j
k k k
E t V t C t = − (if
there is no margin agreement,
( )
( ) 0
j
k
C t ≡ ).
3. After running large enough number M of market scenarios, compute the discounted
EE by averaging over all the market scenarios at each time point:
( ) ( )
1
1
( ) ( ) ( )
M
j j
k k k
j M
e t D t E t
=
∗
=
∑
.
4. Finally, compute CVA as
1
CVA (1 ) ( )[ ( ) ( )]
k k k
k
R e t P t P t
−
∗
= − −
∑
,
where, as before, R denotes the (constant) recovery rate and ( ) P t is the unconditional
cumulative probability of default up to time t .
The calculation of EE and CVA contributions can be incorporated to this algorithm as
follows. Consider, for example, the EE contributions given by Equation (32). The following
calculations are added to Steps 24:
Step 2: For each trade i, calculate the trade’s exposure contribution for scenario j
( )
( )
j
i k
E t
∗
, which is equal to
( )
( )
j
i k
V t if
( )
0 ( )
j
k
V t H < ≤ ,
( ) ( )
( ) ( )
j j
i k k
HV t V t if
( )
( )
j
k
V t H > , and zero otherwise.
Step 3: For each trade i, compute the discounted EE contribution by averaging over
all the market scenarios at each time point:
( ) ( )
1
1
( ) ( ) ( )
M
j j
i k k i k
j M
e t D t E t
=
∗
=
∑
.
Step 4: CVA contributions are computed as
16
1
CVA (1 ) ( )[ ( ) ( )]
i i k k k
k
R e t P t P t
−
∗
= − −
∑
.
6.2 Exposure Dependent on Counterparty’s Credit Quality
The algorithm above assumes independence between the exposure and the counterparty’s
credit quality. More generally, there may be dependence between them which can come from
two sources:
Right/wrongway risk. The risk is called rightway (wrongway) if exposure tends to
decrease (increase) when counterparty quality worsens. Strictly speaking, right/wrong
way risk is always present, but it is usually ignored to simplify exposure modeling.
However, there are cases when right/wrong way risk is too significant to be ignored (e.g.,
credit derivatives, commodity trades with a producer of that commodity, etc.).
Exposurelimiting agreements that depend on the counterparty credit quality. One
example such agreements is a margin agreement with the threshold dependent on the
counterparty’s credit rating. Another example is an early termination agreement, under
which the bank can terminate the trades with the counterparty when the counterparty’s
rating falls below a prespecified level.
Both types of dependence of exposure on the counterparty’s credit quality can be
incorporated in the EE and CVA calculation if the trade values and credit quality of the bank’s
counterparties are simulated jointly. If a bank’s counterparty risk simulation environment is
capable of such joint simulation, the calculation of EE and CVA contributions is also
straightforward.
Let us introduce a stochastic default intensity process ( ) t λ without specifying its
underlying dynamics.
11
This intensity can be used as a measure of counterparty credit quality:
higher values of the intensity correspond to lower credit quality. The counterpartylevel exposure
( ) E t may depend either on the intensity value ( ) t λ at time t , or on the entire path of the
intensity process ( ) λ ⋅ from zero to t. We can use the intensity process to convert the expectation
conditional on default at time t in Equation (21) to an unconditional expectation so that the
conditional EE contribution becomes
0
1
ˆ ( ) E ( ) exp ( ) ( ) ( )
( )
t
i i
e t t s ds D t E t
P t
λ λ
∗
(  
= −
( 
′
(
\ ¹ ¸ ¸
∫
(42)
where ( ) P t ′ is the first derivative of the cumulative PD ( ) P t . A short derivation of Equation
(42) is given in Appendix 1.
11
For an overview of stochastic default intensity and the associated Cox process, see Chapter 5 in Schönbucher
(2003).
17
As described for unconditional EE contributions, the calculations for conditional EE
contributions can be performed during a Monte Carlo simulation of exposures. In this case, given
the dependence of exposures on the counterparty credit quality, the intensity process ( ) λ ⋅ needs
to be simulated jointly with the market risk factors that determine trade values. This joint
simulation is done pathbypath: simulated values of the intensity and of the market factors at
time
k
t are obtained from the corresponding simulated values at the earlier time points
(
1 2
, ,...
k k
t t
− −
).
12
Assuming that we have already simulated the market factors and the intensity for
times
j
t for all j k < , the algorithm for computing CVA contributions for time
k
t can be
expressed as follows:
1. Jointly simulate market risk factors and intensity ( )
k
t λ at time
k
t
2. For each trade i, calculate its trade value ( )
i k
V t
3. Calculate the portfolio value
1
( ) ( )
N
i
i
V t V t
=
=
∑
4. For each trade i, update the EE contribution counter
a. If there is no margin agreement, then
− if ( ) 0 V t > , add
( ) 1 1
1
( )
exp ( )( ) ( ) ( )
( )
k
k
j j j k i k
j
k
t
t t t D t V t
P t
λ
λ
− −
=
− −
′
∑
b. If there is a margin agreement with an intensitydependent threshold [ ( )] h λ ⋅ , then
− if 0 ( ) [ ( )]
k
V t h t λ < ≤ , add
( ) 1 1
1
( )
exp ( )( ) ( ) ( )
( )
k
k
j j j k i k
j
k
t
t t t D t V t
P t
λ
λ
− −
=
− −
′
∑
− if ( ) [ ( )]
k
V t h t λ > , add
( ) 1 1
1
( ) ( )
exp ( )( ) ( ) [ ( )]
( ) ( )
k
k i k
j j j k k
j
k k
t V t
t t t D t h t
P t V t
λ
λ λ
− −
=
− −
′
∑
After running large enough number of market scenarios, EE contributions are obtained by
dividing the EE contribution counter by the number of scenarios.
7. Analytical CVA Contributions under a Normal Approximation
It is also useful in practice to estimate EE and CVA contributions quickly outside of the
simulation system. To facilitate such calculations, we derive analytical EE contributions, for the
case when trade values are normally distributed. For simplicity, and to avoid dealing with
stochastic discounting factors, we assume that, at time t, the distribution of trade values is given
12
For a discussion on pathbypath vs. direct jump to simulation date, see Pykhtin and Zhu (2007).
18
under the forward (to time t) probability measure. Under this measure, the discounted
conditional EE in Equation (9) can be written as
ˆ ˆ ( ) (0, ) E ( ) (0, ) ( ) e t B t E t t B t e t τ
∗
= = ( ≡
¸ ¸
(43)
and the discounted unconditional EE is
[ ] ( ) (0, ) E ( ) (0, ) ( ) e t B t E t B t e t
∗
= ≡ (44)
where (0, ) B t is the timezero price of the riskfree zerocoupon bond maturing at time t. This
change of measure allows us to work with undiscounted EEs and EE contributions.
Assume that the value ( )
i
V t of trade i at each future time t is normally distributed with
expectation
i
µ and standard deviation
i
σ under the forward to t probability measure:
( ) ( ) ( )
i i i i
V t t t X µ σ = + (45)
where
i
X is a standard normal variable. Correlations between these standard normal variables
(and, therefore, between the discounted trade values) are denoted by
ij
r .
Since the sum of normal variables is also normal, the discounted portfolio value ( ) V t is
normally distributed:
( ) ( ) ( ) V t t t X µ σ = + (46)
where X is another standard normal variable, and the mean and standard deviation of the
portfolio value are given by
1
( ) ( )
N
i
i
t t µ µ
=
=
∑
,
2
1 1
( ) ( ) ( )
N N
ij i j
i j
t r t t σ σ σ
= =
=
∑∑
(47)
Denote by ( )
i
t ρ the correlation between the value ( )
i
V t of trade i and the portfolio value
( ) V t . We can calculate this correlation as follows:
1
1
cov[ ( ), ( )]
( )
cov[ ( ), ( )]
( )
( ) ( ) ( ) ( ) ( )
N
N
i j
j j
i
i ij
j
i i
V t V t
t
V t V t
t r
t t t t t
σ
ρ
σ σ σ σ σ
=
=
= = =
∑
∑
(48)
Using this correlation, we can represent
i
X as
2
( ) 1 ( )
i i i i
X t X t Z ρ ρ = + − (49)
19
where
i
Z is a standard normal random variable independent of X (and different for each trade).
7.1 Exposure Independent of Counterparty’s Credit Quality
We first calculate counterpartylevel EE and EE contributions assuming independence
between exposures and counterparty credit quality. We obtain the results for the general case of a
netting agreement with a margin agreement. The simpler case, with no margin agreement, is
obtained as the limiting case when the threshold goes to infinity. For the clarity of exposition, we
assume the instantaneous collateral model.
In the presence of a margin agreement, the counterpartylevel stochastic exposure is
given by Equation (22). Substituting Equation (46) into Equation (22) and taking the expectation,
we obtain
[ ]
{ } { }
[ ]
( )
( )
( ) ( )
( ) ( )
0 ( ) ( ) ( ) ( )
( ) E ( ) ( ) E
( ) ( ) ( ) ( )
1 1
H t
t
t H t
t t
t t X H t t X H
e t t t X H
t t x x dx H x dx
µ
σ
µ µ
σ σ
µ σ µ σ
µ σ
µ σ φ φ
−
∞
−
−
< + < + >
( (
= + +
( (
¸ ¸ ¸ ¸
= + +
∫ ∫
where ( ) φ ⋅ is the probability density of the standard normal distribution. Evaluating the integrals
yields an analytical formula for the EE:
( ) ( )
( ) ( )
( ) ( )
( ) ( ) ( )
( )
( ) ( ) ( )
t t H
e t t
t t
t t H t H
t H
t t t
µ µ
µ
σ σ
µ µ µ
σ φ φ
σ σ σ
(     −
= Φ − Φ
(  
\ ¹ \ ¹ ¸ ¸
(       − −
+ − + Φ
(   
\ ¹ \ ¹ \ ¹ ¸ ¸
(50)
where ( ) Φ ⋅ is the standard normal cumulative distribution function.
Let us consider now the EE contributions given by Equation (32) (type B allocations).
Removing the discounting and substituting Equations (45) and (46) into Equation (32), we obtain
( )
{ } { }
0 ( ) ( ) ( ) ( )
( ) ( )
( ) E ( ) ( ) E
( ) ( )
1 1
i i i
i i i i
t t X H t t X H
t t X
e t t t X H
t t X
µ σ µ σ
µ σ
µ σ
µ σ
< + < + >
( +
(
= + +
(
(
+ ¸ ¸
¸ ¸
(51)
Appendix 2 shows that, after some analytical manipulation, this EE contribution can be written
as
20
( )
( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( )
( )
( ) ( )
i i i i
i i i
H t
t
t t H t t H
e t t t t
t t t t
t t t x
H x dx
t t x
µ
σ
µ µ µ µ
µ σ ρ φ φ
σ σ σ σ
µ σ ρ
φ
µ σ
∞
−
( (         − −
= Φ − Φ + −
( (    
\ ¹ \ ¹ \ ¹ \ ¹ ¸ ¸ ¸ ¸
+
+
+
∫
(52)
where the remaining integral can be easily evaluated numerically. One can verify that EE
contributions given by Equation (52) sum up to the counterpartylevel EE, Equation (50).
Similarly, we obtain the EE contributions corresponding to type A allocations as
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( )
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) (
( ) ( )
( )
i i i i
i i i
t t H t t H
e t t t t
t t t t
t H t H
t t t
t t t H
H
t t H
t t
t
µ µ µ µ
µ σ ρ φ φ
σ σ σ σ
µ µ
µ σ ρ φ
σ σ µ
σ µ µ
µ σ φ
σ
( (         − −
= Φ −Φ + −
( (    
\ ¹ \ ¹ \ ¹ \ ¹ ¸ ¸ ¸ ¸
    − −
Φ +
 
  −
\ ¹ \ ¹
+ ⋅ Φ ⋅

  −
\ ¹
Φ +

\ ¹
)
( )
t H
t σ
  −

\ ¹
(53)
In contrast to Equation (52), the EE contributions given by Equation (53) are given in
closed form, and do not require numerical integration.
For the case without a margin agreement, we take the limit H → ∞ of Equations (50)
(53). This leads to the counterpartylevel EE
( ) ( )
( ) ( ) ( )
( ) ( )
t t
e t t t
t t
µ µ
µ σ φ
σ σ
   
= Φ +
 
\ ¹ \ ¹
(54)
and the EE contributions
( ) ( )
( ) ( ) ( ) ( )
( ) ( )
i i i i
t t
e t t t t
t t
µ µ
µ σ ρ φ
σ σ
   
= Φ +
 
\ ¹ \ ¹
(55)
7.2 Right/WrongWay Risk
We now lift the independence assumption to accommodate right/wrongway risk. Note
that the EE contributions obtained in the previous section are contributions of the trades in the
portfolio to the counterpartylevel unconditional discounted EE. We need to modify the
approach to obtain the contributions to the counterpartylevel EE conditional on the counterparty
defaulting at the time when the exposure is measured. An obvious approach is to define an
intensity process and compute the conditional EE contributions as the expectation over all
21
possible paths of the intensity process (See Appendix 1), but this requires a Monte Carlo
simulation. In this section, we develop an alternative, simpler approach that results in closed
form expressions for the conditional EE contributions.
For this purpose, we define a Normal copula
13
to model the codependence between the
counterparty’s credit quality and the exposures.
14
Thus, we first map the counterparty’s default
time τ to a standard normal random variable Y:
1
[ ( )] Y P τ
−
= Φ (56)
where
1
( )
−
Φ ⋅ is the inverse of the standard normal cumulative distribution function. The
counterpartylevel conditional EE is given by
ˆ( ) E ( ) e t E t t τ = = (
¸ ¸
(57)
while the conditional EE contribution of trade i is given by
ˆ ( ) E ( )
i i
e t E t t τ = = (
¸ ¸
(58)
where ( ) E t is the counterpartylevel exposure and ( )
i
E t is the stochastic exposure contribution
of trade i at time t. Since the counterparty’s cumulative probability of default ( ) P ⋅ is a
monotonic function, each possible default time is mapped to a unique value of Y. Thus, we can
replace the conditioning on τ in Equations (57) and (58) with the conditioning on Y and write
the counterpartylevel conditional EE and the EE contributions as
1
ˆ( ) E ( ) [ ( )] e t E t Y P t
−
( = =Φ
¸ ¸
(59)
1
ˆ ( ) E ( ) [ ( )]
i i
e t E t Y P t
−
( = =Φ
¸ ¸
(60)
We model the right/wrongway risk by allowing trade values to depend on Y. More
specifically, we assume that the standard normal risk factor
i
X , which drives the value of trade i,
depends on Y according to
2
ˆ
1
i i i i
X bY b X = + − (61)
13
The Normal copula framework was first proposed by Li (2000) to model correlated default times for pricing
portfolio credit derivatives.
14
See for example Garcia Cespedes et al. (2009) for the application of such a copula approach to compute
counterparty credit capital.
22
where
ˆ
i
X is a standard normal variable independent of Y. The parameter
i
b drives the
right/wrongway risk. When 0
i
b = , the value of trade i is independent of the counterparty credit
quality. Wrongway risk occurs when 1 0
i
b − ≤ < (the value of trade i tends to increase when Y
declines). Similarly, there is rightway risk when0 1
i
b < ≤ (the value of trade i tends to decrease
when Y declines). As the magnitude of
i
b , increases, so does the codependence between the trade
value and the counterparty credit quality.
If the portfolio contains trades with nonzero
i
b , the standard normal risk factor X, which
drives the portfolio value, also depends on Y:
2
ˆ
( ) 1 ( ) X t Y t X β β = + − (62)
with
ˆ
X a standard normal variable independent of Y. The portfolio factor loading ( ) t β can be
computed from the individual trade factor loadings
i
b as follows:
1 1
cov[ ( ), ] cov[ ( ), ] ( )
( ) cov[ , ]
( ) ( ) ( )
N N
i i
i
i i
V t Y V t Y t
t X Y b
t t t
σ
β
σ σ σ
= =
= = = =
∑ ∑
(63)
Now we have all the ingredients to derive the counterpartylevel EE and EE contributions
in the presence of right/wrongway risk. One approach may be to calculate conditional
expectations in the same manner as we have calculated the unconditional ones in the previous
Section. However, in Appendix 2 we show how this can be done in a faster and more elegant
way. In particular, the conditional exposure model can be formulated the in exactly the same
mathematical terms as the unconditional model. The only difference is that instead of the
unconditional expectations, standard deviations and correlations that specify the behavior of the
trade values, we now use the conditional ones. Therefore, we can use all the results of Subsection
7.1 (Equations (50)(55)) after putting “hats” on the parameters:
15
[ ]
1
ˆ ( ) E[ ( )  ] ( ) ( ) ( )
i i i i i
t V t t t t b P t µ τ µ σ
−
≡ = = + Φ (64)
2
ˆ ( ) StDev[ ( )  ] ( ) 1
i i i i
t V t t t b σ τ σ ≡ = = − (65)
[ ]
1
ˆ ( ) E[ ( )  ] ( ) ( ) ( ) ( ) t V t t t t t P t µ τ µ σ β
−
≡ = = + Φ (66)
15
This conclusion is consistent with the results in Redon (2006). Using a different model of right/wrongway risk,
they show that the conditional counterpartylevel uncollateralized EE is described by the same expression as the
unconditional EE, after replacing the unconditional expectations and standard deviations of the trade values with the
conditional ones.
23
2
ˆ ( ) StDev[ ( )  ] ( ) 1 ( ) t V t t t t σ τ σ β ≡ = = − (67)
2 2
( ) ( )
ˆ ( )
(1 )[1 ( )]
i i
i
i
t b t
t
b t
ρ β
ρ
β
−
=
− −
(68)
7.3 Remarks on the Analytical Formulae
In this section we briefly comment on the properties and interpretation of the analytical
contributions derived in this section.
Netting & no margin
Equation (55) can be understood from the incremental viewpoint of the CMC method.
According to Equation (17), the EE contribution of trade i is determined by the infinitesimal
change of the counterpartylevel EE resulting from an infinitesimal increase of the weight of
trade i in the portfolio. The effect of an increase of the weight of a trade on the portfolio value
distribution can be viewed as the sum of two effects:
o a uniform shift of the distribution
o a change of width of the distribution
Let us consider these two effects separately.
If the weight of trade i is increased by δ , the expectation of portfolio value changes by
( )
i
t δ µ ⋅ . Let us first ignore the change of the standard deviation and consider how a uniform
shift of the entire distribution by ( )
i
t δ µ ⋅ affects the counterpartylevel EE. Scenarios with
positive portfolio value contribute the same amount ( )
i
t δ µ ⋅ to the exposure change, while
scenarios with negative portfolio value contribute nothing. Therefore, the increment of the EE
will be given by the product of the magnitude of the shift ( )
i
t δ µ ⋅ and the probability of the
portfolio value being positive. It is straightforward to verify that Pr[ ( ) 0] [ ( ) / ( )] V t t t µ σ > = Φ .
Thus, the first term in the righthand side of Equation (55) describes the increment of the
counterpartylevel EE resulting from the infinitesimal uniform shift of the portfolio value
distribution associated with an increase of the weight of trade i.
The second term of Equation (55) describes the change of the width of the portfolio value
distribution. The change of the standard deviation of the portfolio value resulting from increasing
the weight of trade i by δ can be calculated as
( )
( )
1
2
2
1
2 2 2
2
StDev[ ( ) ( )] StDev[ ( )] var[ ( )] 2 cov[ ( ), ( )] var[ ( )] ( )
( ) 2 ( ) ( ) ( ) var[ ( )] ( ) ( ) ( ) ( )
i i i
i i i i i
V t V t V t V t V t V t V t t
t t t t V t t t t O
δ δ δ σ
σ δ ρ σ σ δ σ δ ρ σ δ
+ − = + + −
= + + − = +
24
where
2
( ) O δ denotes the terms of the second order and higher that can be ignored. Thus, the
portfolio value distribution is widening if the correlation ( )
i
t ρ is positive, and narrowing if the
correlation is negative. A widening distribution (with no accompanying shift) always increases
the counterpartylevel EE, while narrowing always decreases it. Indeed, for a given realization of
the portfolio value ( ) V t , the change of exposure associated with the change of the standard
deviation of the portfolio value from ( ) t σ to ( , ) ( ) ( ) ( )
i i
t t t t σ δ σ δ ρ σ = + is given by
16
{ } ( ) 0
( ) ( )
( , ) ( ) ( ) ( ) 1
( )
i i
V t
V t t
E t E t t t
t
µ
δ δ ρ σ
σ
>
−
− = (69)
The second term of Equation (55) can be obtained by taking the expectation of the righthand
side of Equation (69).
It appears that Equation (55) has simple linear dependence on ( )
i
t µ and the product
( ) ( )
i i
t t ρ σ . However, this is only part of the true dependence. Since trade i is part of the
portfolio, ( ) t µ depends on ( )
i
t µ and ( ) t σ depends on ( )
i
t σ and the correlation of trade i with
the rest of the portfolio. Moreover, correlation ( )
i
t ρ is the correlation between the values of
trade i and the portfolio that includes trade i itself. Because of this, ( )
i
t ρ depends on the ratio
( ) / ( )
i
t t σ σ (see Equation (48)). Thus, unless trade i represents a negligible fraction of the
portfolio, the true dependence of EE contribution on trade parameters is nonlinear.
Netting & margin
In this case, only the first two terms of Equation (52) allow interpretation from the
incremental viewpoint of the CMC method: the first term can be explained as the effect of the
uniform shift and the second term as the effect of the widening or narrowing of the portfolio
value distribution. The third term results from the allocation of exposure when the portfolio
value is above the threshold. An attempt to use the CMC method would give zero EE
contribution from ( ) V t H > scenarios.
Equation (52) can be rewritten as
( )
( )
( )
( )
( ) ( ) ( )
( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( )
( ) ( )
( ) ( ) ( ) ( )
i i
H t
t
i i
H t
t
t t H d
e t t H
t t t t
t t H d
t t H
t t t t
µ
σ
µ
σ
µ µ φ ξ ξ
µ
σ σ µ σ ξ
µ µ φ ξ ξ ξ
σ ρ φ φ
σ σ µ σ ξ
∞
−
∞
−
(
    ( −
= Φ −Φ +
  (
+
\ ¹ \ ¹
(
¸ ¸
(
    ( −
+ − +
  (
+
\ ¹ \ ¹
(
¸ ¸
∫
∫
(70)
16
This can be immediately seen from Equation (46).
25
As in the noncollateralized case, this EE contribution appears to be linear in ( )
i
t µ and the
product ( ) ( )
i i
t t ρ σ , but the true dependence on these quantities is more complex due to the extra
dependence of ( ) t µ and ( ) t σ on these quantities.
Right/wrongway risk
If the value of trade i is correlated with the counterparty’s credit quality, its value
distribution at time t conditional on the counterparty’s default at time t differs from its
unconditional value distribution. If the correlation is positive (rightway risk), the distribution
shifts down; if the correlation is negative (wrongway risk), the distribution shifts up. In both
cases, the distribution becomes narrower. Under the normal approximation, the shift of the
distribution is described by Equation (64), and the narrowing is described by Equation (65).
An interesting property of Equation (64) is its dependence on the counterparty’s PD. To
understand this, let us consider the bank entering into the same trade with an investmentgrade
counterparty A and with a speculativegrade counterparty B. We are interested in the trade value
distribution conditional on the counterparty’s default at the time of observation. For the case of
wrong (right) way risk, the deterioration of the counterparty’s credit quality to the point of
default pushes trade values higher (lower). Since counterparty A is “further away” from default
than counterparty B, the deterioration of credit quality to the point of default is larger for
counterparty A. Therefore, trade values conditional on default of A are shifted more than trade
values conditional on default of B. Note that this is not specific to the normal approximation, but
is a general property not related to any model.
8. Examples
In this section, we present some simple examples that illustrate the behavior of exposure
(and hence CVA) contributions. For ease of exposition, we assume that trade values are Normal,
as well as market and credit independence. However, as discussed earlier in Section 7.2, the
conclusions apply equally to the case of wrongway risk by simply using conditional
expectations, volatilities and correlations, instead of unconditional ones. We first present an
example when there is no collateral agreement in place, and then show the impact of adding a
collateral agreement to the portfolio.
8.1 Contributions for a Noncollateralized Portfolio
As a first step to understand this behavior, consider Equations (54) and (55), which give
the counterpartylevel EE and the EE trade contributions, in the case when there is no margin
agreement in place:
( ) ( )
( ) ( ) ( )
( ) ( )
t t
e t t t
t t
µ µ
µ σ φ
σ σ
   
= Φ +
 
\ ¹ \ ¹
,
( ) ( )
( ) ( ) ( ) ( )
( ) ( )
i i i i
t t
e t t t t
t t
µ µ
µ σ ρ φ
σ σ
   
= Φ +
 
\ ¹ \ ¹
The EE contribution of instrument i is a function of:
26
the mean value contribution, µ
i
the volatility contribution, ρ
i
σ
i
the overall value of the ratio µ/σ (for the entire portfolio)
where we have dropped the time variable notation for brevity.
Both the counterpartylevel EE and the trade contributions can be seen as the sum of two
components: a mean value component (first term in the equations), and a volatility component
(second term). These components weigh the mean value (or mean value contributions) and the
volatility (volatility contribution), respectively, by the Normal distributions and density
evaluated at the ratio µ/σ (for the entire counterparty portfolio). Thus, the overall level of the
counterparty portfolio’s mean value and volatility determine how the individual instrument’s
mean and volatility contribution are weighted to yield the EE contributions. Figure 1 plots these
weights as a function of µ/σ. A low ratio weighs the volatility contribution much higher; while a
high ratio weighs mean values much more. For example, if µ/σ = 2, the volatility component
weight is 2.4 times the mean value weight. In contrast, µ/σ = 2 results in mean values being
weighted 18 times the volatilities.
0.0
0.2
0.4
0.6
0.8
1.0
4 3 2 1 0 1 2 3 4
Mu/Sigma
Norm Dist
Norm
Density
Figure 1. Volatility and mean exposure weights for EE contributions.
To illustrate the impact of various parameters on EE contributions, consider now the
simple counterparty portfolio, which comprises of 5 transactions over a single step. Table 1 gives
the individual trade’s mean value, variance and volatility (in dollar values and % contributions).
The portfolio has a mean value and variance of 10. We assume that trade values are
independent.
17
In this case, the portfolio’s ratio µ/σ = 3.16.
P1 P2 P3 P4 P5 Total
17
This assumption is only made for simplicity, and bears no impact on the analysis. Alternatively we can simply use
the volatility contributions directly for any correlated model.
27
µ 0 1 2 3 4 10
% 0% 10% 20% 30% 40% 100%
σ
2
4 3 2 1 0 10
% 40% 30% 20% 10% 0% 100%
σ 2 1.7 1.4 1.0 0.0 3.2
Table 1. Portfolio – mean values, variances and volatilities.
The portfolio is constructed so that for each trade, its mean value and volatility are
inversely related; thus the first instrument, P1, has the lowest mean (0) and largest volatility (2),
while position 5 has the highest mean value (4), and lowest volatility (0). This may not only be
reasonably realistic, but it will also help highlight some of the points below.
Using Equations (54) and (55), we compute the EE and contributions for the portfolio.
The EE for the portfolio is 10.001, with most of this arising from the mean value component
(9.992). The trade contributions to EE are fairly close to the contributions to the mean values in
Table 1 (0.03%, 10.02%, 20.00%, 29.98%, 39.97%).
Now, we vary the overall mean value of the portfolio, µ, while leaving intact the
volatility, σ, as well as the percent contributions of each instrument to the mean exposure and
volatility in table 1. This allows us to express the trade contributions in terms of how deep in or
outofthemoney the counterparty portfolio is (relative to its volatility). Figure 2 plots the EE, as
well as its mean and volatility components, as functions of the portfolio’s µ/σ. For large negative
portfolio mean values, the EE (red line) is zero. In this case, the mean value component of EE is
actually negative, and the volatility component compensates for this to generate positive EEs. As
µ/σ increases beyond zero, the volatility component decreases and, once µ/σ > 2, the EE is
completely dominated by the mean value.
2
0
2
4
6
8
10
12
14
16
2 0 2 4
Mu/Sigma
E
x
p
e
c
t
e
d
E
x
p
o
s
u
r
e
EE
mu
component
sigma
component
28
Figure 2. EE as a function of the portfolio’s ratio µ/σ.
Figure 3 shows the EE contributions for each of the 5 trades as a function of µ/σ. There is
a clear shift in dominance between the mean and volatility components as the portfolio’s mean
value increases. At one side of the spectrum, when the mean portfolio values are negative, trades
4 and 5, which have the largest (negative) mean values and lowest volatilities, produce very large
negative EE contributions. The opposite occurs for trades 1 and 2 (with low negative means and
large volatilities). As the portfolio’s µ/σ increases, trades 4 and 5 end up dominating the
contributions, with the EE contribution converging to the mean value contributions themselves.
For this particular symmetric portfolio, every trade contributes 20% of EE at µ/σ = 0.506.
80%
60%
40%
20%
0%
20%
40%
60%
80%
100%
1 0.5 0 0.5 1 1.5 2 2.5 3
Mu/Sigma
E
E
C
o
n
t
r
i
b
u
t
i
o
n
s
P1
P2
P3
P4
P5
Figure 3. EE Contributions as a function of the portfolio’s ratio µ/σ.
8.2 Contributions for a Collateralized Portfolio
We consider now the case when there is a margin agreement, and demonstrate the impact of
the collateral on the trade contributions. In very general terms:
As the threshold becomes very large, trade contributions converge to those of
uncollateralized exposures;
With lower thresholds, the contributions of more volatile exposures are diminished (as
the threshold caps the exposures), and contributions of higher mean exposures (inthe
money positions) increase.
Consider the same portfolio in table 1, but assume now that there is a collateral agreement in
place where margins are placed instantaneously. First, we characterize the impact of the
threshold on the counterparty level EE. Figure 4 shows the reductions in EE as a result of the
margin agreement (as % of uncollateralized EE) as a function of µ/σ , and for various levels of
the (standardized) collateral threshold.
29
Exposure reductions with collateral
as a funciton of H/Sigma
0.0%
20.0%
40.0%
60.0%
80.0%
100.0%
5 4 3 2 1 0 1 2 3 4 5
Mu/Sigma
%
E
x
p
o
s
u
r
e
R
e
d
u
c
i
t
o
n
s
0
0.2
0.5
1
2
4
6
Figure 4. EE reductions from a collateral agreement.
As the threshold is increased, the EE reductions decrease, as expected. Also, the collateral
thresholds become more effective at reducing EE as the portfolio is deeper inthemoney (i.e.
when µ/σ increases). For example, a normalized threshold of 2 does not reduce EE until the
portfolio’s mean value is positive. At a value of µ/σ = 5, it reduces EE by about 60%.
Figure 5 plots the EE contributions for the case µ/σ = 1, as a function of the standardized
threshold, H/σ. At high threshold values, H/σ > 4, trade contributions are essentially the
uncollateralized contributions. Conversely at low H/σ values, EE contributions are basically the
mean value contributions. The presence of the collateral affects each instrument’s contributions
differently. In particular, a tighter threshold increases the percent contributions of trades P4 and
P5 (which have the highest mean values) while reducing the contributions of P1 and P2 (the
lowest mean values). These eventually converge in the limit to the mean value contributions.
10%
0%
10%
20%
30%
40%
0 1 2 3 4 5
H/Sigma
E
x
p
o
s
u
r
e
C
o
n
t
r
i
b
u
t
i
o
n
s
P1
P2
P3
P4
P5
Figure 5. EE contributions as a function of the collateral threshold.
Note finally that a trivial case arises when the ratio µ/σ =0. In this case, the EE contributions
are independent of the threshold level H/σ and equal the uncollateralized contributions.
30
9. Conclusions
Counterparty credit risk is usually measured and priced at the counterparty level. The
price of the counterparty risk for the entire portfolio of trades with a counterparty is known as
credit valuation adjustment (CVA). In this article we have proposed a methodology for allocation
of the counterpartylevel CVA to individual trades. These allocations are additive, so that one
can aggregate the CVA allocations for any collection of trades with different counterparties.
Thus, the contribution of all trades belonging to a certain class to the banklevel CVA can be
calculated. Such a class can be defined as “all trades booked by a certain business unit”, “all
EUR interest rate swaptions”, etc.
In this paper, we show that the calculation of CVA allocations can be reduced to the
calculation of contributions of individual trades to the counterpartylevel expected exposure (EE)
conditional on the counterparty’s default. To obtain conditional EE contributions, we adapt the
continuous marginal contribution method which is often used for allocating economic capital.
The method is directly applicable for CVA contributions only when the counterpartylevel
exposure is a homogeneous function of the trades’ weights in the portfolio. This is the case when
there are no collateral or margin agreements. We extend the methodology to deal with non
homogeneous exposures of the type encountered when the portfolios have margin agreements.
We further show how the calculations of conditional EE contributions can be
incorporated into an existing exposure simulation process. In addition, the ability to make quick
calculations of CVA allocations outside of the exposure simulation system may be also desirable.
To facilitate such calculations, we derive closed form expressions for unconditional EE
contributions under the assumption that trade values are normally distributed. By using
unconditional EEs in the CVA calculations, one implicitly assumes that exposures are
independent of the counterparty credit quality. To overcome this limitation, we extend the results
for conditional EE contributions in the normal approximation, which incorporate dependence
between the trade values and the counterparty’s credit quality.
References
Aziz A. and D. Rosen, 2004, Capital Allocation and RAPM, The Professional Risk Manager’s
Handbook (C. Alexander and E. Sheedy editors), PRMIA Publications, Wilmington, DE
(www.prmia.org), pages 1341.
A. Arvanitis and J. Gregory, 2001, “Credit: The Complete Guide to Pricing, Hedging and Risk
Management”, Risk Books
D. Brigo and M. Masetti, 2005, Risk Neutral Pricing of Counterparty Credit Risk in
“Counterparty Credit Risk Modelling” (M. Pykhtin, ed.), Risk Books
31
E. Canabarro and D. Duffie, 2003, Measuring and Marking Counterparty Risk
in “Asset/Liability Management for Financial Institutions” (L. Tilman, ed.), Institutional Investor
Books
Garcia Cespedes J. C., de Juan Herrero J. A., Rosen D., and Saunders D., 2009, Effective
modelling of CCR Capital and Alpha for Derivatives Portfolios, Working Paper, Fields Institute
and University of Waterloo
M. Gibson, 2005, Measuring Counterparty Credit Exposure to a Margined Counterparty in
“Counterparty Credit Risk Modelling” (M. Pykhtin, ed.), Risk Books
D. Li, 2000, On Default Correlation: a Copula Approach, Journal of Fixed Income 9, pages 43
54.
H. Mausser and D. Rosen, 2007, Economic credit capital allocation and risk contributions, in
Handbook of Financial Engineering, J. Birge and V. Linetsky, eds., vol. 15 of Handbooks in
Operations Research and Management Science, NorthHolland, 2007, pp. 681726.
E. Picoult, 2005, Calculating and Hedging Exposure, Credit Value Adjustment and Economic
Capital for Counterparty Credit Risk in “Counterparty Credit Risk Modelling” (M. Pykhtin,
ed.), Risk Books
M. Pykhtin and S. Zhu, 2006, Measuring Counterparty Credit Risk for Trading Products
under Basel II in “Basel Handbook”, 2
nd
Edition (M. Ong, ed.), Risk Books
M. Pykhtin and S. Zhu, 2007, A Guide to Modeling Counterparty Credit Risk, GARP Risk
Review, July/August, pages 1622.
M. Pykhtin, 2009, Modeling Credit Exposures for Collateralized Counterparties, Journal of
Credit Risk, 5(4).
C. Redon, 2006, Wrong Way Risk Modelling, Risk, April, pages 9095.
P. Schönbucher (2003), “Credit Derivatives Pricing Models”, Wiley Finance.
32
Appendix 1. Derivation of Equation (42)
Suppose we have a random variable W and we want to calculate its expectation
conditional on the counterparty defaulting at time t ( t τ = ). We can express this expectation as
{ }
{ }
E 1
E
E 1
t t dt
t t dt
W
W t
τ
τ
τ
< ≤ +
< ≤ +
(
(
¸ ¸
= ( =
¸ ¸
(
(
¸ ¸
(71)
Then, for the numerator of Equation (55) we can write
( )
0 0
0
0
{ }
E 1 E exp[ ( ) ] exp[ ( ) ]
E exp[ ( ) ] 1 exp[ ( ) ]
E exp[ ( ) ] ( )
t t dt
t
t
t t dt
W W s ds s ds
W s ds t dt
W s ds t dt
τ
λ λ
λ λ
λ λ
+
< ≤ +
(  
(
= − − −
( 
¸ ¸
(
\ ¹ ¸ ¸
(
= − − −
(
¸ ¸
(
= −
(
¸ ¸
∫ ∫
∫
∫
For the denominator, we can write
{ }
[ ] E 1 Pr ( )
t t dt
t t dt P t dt
τ
τ
< ≤ +
(
′ = < ≤ + =
(
¸ ¸
where ( ) P t ′ is the first derivative of the cumulative probability of default ( ) P t . Substituting both
expressions in Equation (71), we obtain
0
1
E E ( )exp[ ( ) ]
( )
t
W t W t s ds
P t
τ λ λ
(
= ( = −
(
¸ ¸
′
¸ ¸
∫
Appendix 2. Analytical Results under Normal Approximation.
A2.1 Exposure Independent of Counterparty’s Credit Quality
We derive now Equation (52) from Equation (51), which we restate here for convenience:
( )
{ } { }
0 ( ) ( ) ( ) ( )
( ) ( )
( ) E ( ) ( ) E
( ) ( )
1 1
i i i
i i i i
t t X H t t X H
t t X
e t t t X H
t t X
µ σ µ σ
µ σ
µ σ
µ σ
< + < + >
( +
(
= + +
(
(
+ ¸ ¸
¸ ¸
33
Substituting Equation (49) and inserting the expectation conditional on X inside each of the
unconditional expectations, we obtain
{ }
( )
{ }
[ ]
{ }
2
2
0 ( ) ( )
( ) ( )
0 ( ) ( )
( ) E E ( ) ( ) ( ) 1 ( )
( ) ( ) ( ) 1 ( )
E E
( ) ( )
( ) ( ) (
E ( ) ( ) ( ) E
1
1
1
i i i i i i
i i i i i
i i i
i i i
t t X H
t t X H
t t X H
e t t t t X t Z X
t t t X t Z
H X
t t X
t t t
t t t X H
µ σ
µ σ
µ σ
µ σ ρ ρ
µ σ ρ ρ
µ σ
µ σ ρ
µ σ ρ
< + <
+ >
< + <
(
(
= + + −
(
¸ ¸
¸ ¸
(  
(
+ + −
( 
¸ ¸
+
( 
+

(
\ ¹ ¸ ¸
+
(
= + +
(
¸ ¸
{ }
[ ]
( )
( )
( ) ( )
( ) ( )
( ) ( )
)
( ) ( )
( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
( ) ( )
1
H t
t
i i i
i i i
t H t
t t
t t X H
X
t t X
t t t x
t t t x x dx H x dx
t t x
µ
σ
µ µ
σ σ
µ σ
µ σ
µ σ ρ
µ σ ρ φ φ
µ σ
−
∞
−
−
+ >
(
(
+
¸ ¸
+
= + +
+
∫ ∫
While evaluating of the first integral is straightforward, there is no closed form solution for the
second one. This results in Equation (42):
( )
( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( )
( )
( ) ( )
i i i i
i i i
H t
t
t t H t t H
e t t t t
t t t t
t t t x
H x dx
t t x
µ
σ
µ µ µ µ
µ σ ρ φ φ
σ σ σ σ
µ σ ρ
φ
µ σ
∞
−
( (         − −
= Φ − Φ + −
( (    
\ ¹ \ ¹ \ ¹ \ ¹ ¸ ¸ ¸ ¸
+
+
+
∫
A2.2 EE Contributions under Right/Wrongway Risk
We present the derivation of the analytical EE contributions when exposures are
correlated with the counterparty credit quality. Specifically, we show that we can use Equations
(51)(55), with the only difference that instead of using the unconditional expectations, standard
deviations and correlations that specify the behavior of the trade values, we now use the
conditional ones.
From conditional expectation to conditional random variables
The conditional expectation of a random variable can always be formulated as the
unconditional expectation of a conditional random variable. For example, the counterpartylevel
conditional EE in Equation (59) can be represented as
ˆ
ˆ( ) E ( ) e t E t
(
=
¸ ¸
(72)
while the conditional EE contribution in Equation (60) can be written as
34
ˆ
ˆ ( ) E ( )
i i
e t E t
(
=
¸ ¸
(73)
where
ˆ
( ) E t is the conditional counterpartylevel exposure and
ˆ
( )
i
E t is the conditional stochastic
exposure contribution of trade i. These conditional quantities are determined from the
conditional trade values in exactly the same manner as the unconditional exposure and exposure
contributions are determined from the unconditional trade values.
Conditional trade values are calculated as follows. By substituting Equation (61) into
Equation (45) and setting
1
[ ( )] Y P t
−
=Φ , we obtain the value
ˆ
( )
i
V t of trade i conditional on the
counterparty defaulting at time t:
ˆ ˆ
ˆ ˆ ( ) ( ) ( )
i i i i
V t t t X µ σ = + (74)
where ˆ ( )
i
t µ is the conditional expectation of ( )
i
V t given by
[ ]
1
ˆ ( ) E[ ( )  ] ( ) ( ) ( )
i i i i i
t V t t t t b P t µ τ µ σ
−
≡ = = + Φ (75)
and ˆ ( )
i
t σ is the conditional standard deviation of ( )
i
V t given by
2
ˆ ( ) StDev[ ( )  ] ( ) 1
i i i i
t V t t t b σ τ σ ≡ = = − (76)
The conditional portfolio value
ˆ
( ) V t is obtained by substituting Equation (62) into Equation (46)
and setting
1
[ ( )] Y P t
−
=Φ :
ˆ ˆ
ˆ ˆ ( ) ( ) ( ) V t t t X µ σ = + (77)
where ˆ ( ) t µ is the conditional expectation of ( ) V t given by
[ ]
1
ˆ ( ) E[ ( )  ] ( ) ( ) ( ) ( ) t V t t t t t P t µ τ µ σ β
−
≡ = = + Φ (78)
and ˆ ( ) t σ is the conditional standard deviation of ( ) V t given by
2
ˆ ( ) StDev[ ( )  ] ( ) 1 ( ) t V t t t t σ τ σ β ≡ = = − (79)
Finally, we need the correlation between the conditional trade value and the conditional
portfolio value. Let us denote the correlation between
ˆ
( )
i
V t and
ˆ
( ) V t by ˆ ( )
i
t ρ . To obtain this
correlation, let us use Equations (61) and (62) to calculate the covariance between
i
X and X:
35
2 2
ˆ cov( , ) ( ) 1 1 ( ) ( )
i i i i
X X b t b t t β β ρ = + − −
where we have taken into account that both
ˆ
i
X and
ˆ
X are independent of Y. Realizing that
cov( , )
i
X X is nothing but the unconditional correlation ( )
i
t ρ given by Equation (48) , we obtain
2 2
( ) ( )
ˆ ( )
(1 )[1 ( )]
i i
i
i
t b t
t
b t
ρ β
ρ
β
−
=
− −
(80)
Note that ˆ ( )
i
t ρ can also be interpreted as the correlation between the value ( )
i
V t of trade
i at time t and the portfolio value ( ) V t at time t, conditional on the counterparty’s default at time
t. When no right/wrongway risk is present in the portfolio, all 0
i
b = and, as a consequence of
Equation (63), ( ) 0 t β = . Then, Equation (68) reduces to ˆ ( ) ( )
i i
t t ρ ρ = , and the conditional
correlation becomes the unconditional one, as one would expect. Using this correlation, we can
express
ˆ
i
X as
2
ˆ ˆ ˆ
ˆ ˆ ( ) 1 ( )
i i i i
X t X t Z ρ ρ = + − (81)
where
ˆ
i
Z is a standard normal random variable independent of
ˆ
X . Equation (81) is the
conditional version of Equation (49).
Now we have formulated the conditional exposure model in exactly the same
mathematical terms as the unconditional model of Section 7 and can use all the results of
Subsection 7.1 after putting “hats” on the parameters!
Finance and Economics Discussion Series
Divisions of Research & Statistics and Monetary Affairs
Federal Reserve Board, Washington, D.C.
Starting with working papers in 1996, electronic copies of FEDS papers
may be downloaded at the
Internet site: http://www.federalreserve.gov/pubs/feds
Single hard copies of FEDS papers may be obtained by request from:
Diane Linder
Finance and Economics Discussion Series
Mailstop 80
Federal Reserve Board
Washington, D.C. 20551
FAX: (202) 8724927
diane.c.linder@frb.gov
Comments about the FEDS working paper series may be addressed to:
Michael Palumbo
Finance and Economics Discussion Series
Mailstop 61
Federal Reserve Board
Washington, D.C. 20551
michael.g.palumbo@frb.gov
Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Aﬀairs Federal Reserve Board, Washington, D.C.
Pricing Counterparty Risk at the Trade Level and CVA Allocations
Michael Pykhtin and Dan Rosen
201010
NOTE: Staﬀ working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staﬀ or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.
Electronic copy available at: http://ssrn.com/abstract=1782063
Electronic copy available at: http://ssrn.com/abstract=1782063
Pricing Counterparty Risk at the Trade Level and CVA Allocations1
Michael Pykhtin2 and Dan Rosen3
November 2009
Abstract We address the problem of allocating the counterpartylevel credit valuation adjustment (CVA) to the individual trades composing the portfolio. We show that this problem can be reduced to calculating contributions of the trades to the counterpartylevel expected exposure (EE) conditional on the counterparty’s default. We propose a methodology for calculating conditional EE contributions for both collateralized and noncollateralized counterparties. Calculation of EE contributions can be easily incorporated into exposure simulation processes that already exist in a financial institution. We also derive closedform expressions for EE contributions under the assumption that trade values are normally distributed. Analytical results are obtained for the case when the trade values and the counterparty’s credit quality are independent as well as when there is a dependence between them (wrongway risk).
1
The opinions expressed here are those of the authors and do not necessarily reflect the views or policies of their employers.
2 3
Federal Reserve Board, Washington, DC, USA. michael.v.pykhtin@frb.gov.
R2 Financial Technologies and The Fields Institute for Research in Mathematical Sciences, Toronto, Canada. dan.rosen@R2financial.com and drosen@fields.utoronto.ca
1
.
2005 or Gregory. Mausser and Rosen 2007). the unilateral and bilateral approaches. and results in an objective fair value calculation.4 However. by definition. when the given portfolio does not allow for netting between trades. In this case. it is sometimes desirable to determine contributions of individual trades to the counterpartylevel CVA. In this case. in the presence of netting and/or margin agreements.1. Under both. The credit valuation adjustment (CVA) is. LIBOR rates roughly correspond to AA risk rating and incorporate the typical credit risk of large banks.6 This measure is commonly seen as appropriate for pricing counterparty risk for new trades with the counterparty (see Chapter 6 in Arvanitis and Gregory. Under the unilateral approach. Introduction For years. the true value of the portfolio must incorporate the possibility of losses due to counterparty default. CVA is the market value of counterparty credit risk. Hence neither standalone nor incremental contributions can be used effective contributions of existing trades in the portfolio to the counterpartylevel CVA. There are several possible measures of CVA contributions. For this purpose we require additive CVA contributions. we draw the analogy with the capital allocation literature and refer to these as (continuous) marginal risk contributions. all cash flows are discounted using the LIBOR curve. We refer to the CVA of each transaction on a standalone basis as the transaction’s standalone CVA. Sometimes these are referred to as discrete marginal contributions. this is not the case when netting and margin agreements are in place. However. it is assumed that the counterparty that does the CVA analysis (we call this counterparty a bank throughout the paper) is defaultfree. the standard practice in the industry was to mark derivatives portfolios to market without taking the counterparty credit quality into account.5 There are two approaches to measuring CVA: unilateral and bilateral (see Picoult. 2009). However. the difference between the riskfree portfolio value and the true portfolio value that takes into account the counterparty’s default. 5 4 See Canabarro and Duffie (2003) or Pykhtin and Zhu (2007) for an introduction to counterparty credit risk and CVA. and the resulting values are often referred to as riskfree values. More precisely. We refer to the incremental CVA contribution of a trade as the difference between the portfolio CVA with and without the trade. the portfoliolevel CVA is given by the sum of the individual trades’ standalone CVA. The problem of calculating CVA contributions bears many similarities to the calculation of risk contributions and capital allocation (see Aziz and Rosen 2004. 6 2 . One problem with incremental CVA contributions is that they are nonadditive – the sum of the individual trade’s CVA contributions does not add up to the portfolio’s CVA. The problem with unilateral CVA is that both the bank and the counterparty require a premium for the credit risk they are bearing and can never agree on the fair value of the trades in the portfolio. 2001 for details). Clearly. Bilateral CVA takes into account the possibility of both the counterparty and the bank defaulting. CVA is measured at the counterparty level. In other words. It is thus symmetric between the bank and the counterparty. CVA measured this way is the current market value of future losses due to the counterparty’s potential default.
we propose an EE allocation scheme that is based on the CMC method. In Section 2. we introduce CVA contributions of individual trades and relate them to the profiles of conditional EE contributions.The marginal CVA contributions with a given counterparty give the bank a clear picture how much each trade contributes to the counterpartylevel CVA. we show how to incorporate EE and CVA contribution calculations into exposure simulation process. the collateralized and noncollateralized cases. but can be used for collateralized counterparties. We start with the case of independence between exposure and the counterparty’s credit quality.g. such as margin agreements. In Section 8. The theory of marginal risk contributions. Once the CVA contributions have been calculated for each counterparty. the bank can calculate the price of counterparty credit risk in any collection of trades without any reference to the counterparties. In Section 5. we define counterparty credit exposure for both collateralized and noncollateralized cases. an extension of all the results to the bilateral framework is straightforward. In Section 6. the CMC method fails because the counterpartylevel exposure is not homogeneous anymore. is now well developed and largely relies on the risk function being homogeneous (of degree one). all CDSs or all USD interest rate swaps). we assume the unilateral framework throughout the paper. However. we derive closed form expressions for EE contributions under the assumption that all trade values are normally distributed. the bank can determine the contribution of that business unit or product to the bank’s total CVA. We also provide an intuitive explanation to our closedform results. sometimes refer to as Euler Allocations (see Tasche 2008).. When such agreements are present. we adapt the continuous marginal contribution (CMC) method often used for allocating economic capital to calculating EE contributions for the case when the counterpartylevel exposure is a homogeneous function of the trades’ weights in the portfolio. However. We show that this principle can be applied readily for CVA when the counterparty portfolio allows for netting (but does not include collateral and margins). by selecting all trades booked by a certain business unit or product type (e. the use of CVA contributions is not limited to an analysis at a single counterparty level. We show how counterpartylevel CVA can be calculated from the profile of the discounted risk neutral expected exposure (EE) conditional on the counterparty’s default. In Section 4. we show several numerical examples that illustrate the behavior of exposure (and hence CVA) contributions for both. and under a wide range of assumptions. The paper is organized as follows. with the counterparty. We show how to define and calculate marginal CVA contributions in the presence of netting and margin agreements. and extend the results to incorporate dependence between them (wrongway risk). This is the case when there are no exposurelimiting agreements. For example. including the dependence of exposure on the counterparty’s credit quality. We further extend this allocation principle for the more general case of collateralized/margined counterparties For the sake of simplicity. In Section 3. In Section 7. 3 .
CCR creates a bilateral risk: the market value can be positive or negative to either counterparty and can vary over time with the underlying market factors.2. The value of the counterparty = portfolio at t is given by V (t ) = ∑Vi (t ) i =1 N (1) When netting is not allowed.Vi (t )} i =1 N (2) For a counterparty portfolio with a single netting agreement. the (netted) exposure is E (t ) = max {V (t ). for example.8 Denote the value of the ith instrument in the portfolio at time t from the bank’s perspective by Vi (t ) . the (gross) counterpartylevel exposure E (t ) is E ( t ) = ∑ max {0. The counterparty defaults at a random time τ with a known riskneutral distribution P(t ) ≡ Pr[τ ≤ t ] . Counterparty credit risk (CCR) is the risk that the counterparty defaults before the final settlement of a transaction's cash flows. for example. where only the lending bank faces the risk of loss.1 Counterparty exposures Consider a portfolio of N derivative contracts of a bank with a given counterparty. The maturity of the longest contract in the portfolio is T . Unlike a loan. 8 4 . At each time t. Counterparty credit risk and CVA In this section. We define the counterparty exposure E (t ) of the bank to a counterparty at time t as the economic loss. {Vi (t )}iN 1 . 0} (3) 7 The term structure of risk neutral probabilities of default can be obtained from credit default swaps spreads quoted for the counterparty on the market for different of different maturities. See. An economic loss occurs if the counterparty portfolio has a positive economic value for the bank at the time of default. 2. incurred on all outstanding transactions with the counterparty if the counterparty defaults at t . Schönbucher (2003). accounting for netting and collateral but unadjusted by possible recoveries. we review the basic concepts and notation for counterparty credit risk. credit exposures and CVA. the counterpartylevel exposure E (t ) is determined by the values of all trades with the counterparty at time t.7 We further assume that the distribution of the trade values at all future dates is risk neutral. See. Brigo and Masetti (2005).
the instantaneous collateral model works reasonably well for expected exposure. Collateral transfer occurs only when the collateral amount that needs to be transferred exceeds a minimum transfer amount. As the portfolio value drops below the threshold. See Pykhtin (2009). we use this model to show the simple. 2.2 Models of Collateral (4) We start modeling collateral with a simplifying assumption: we incorporate the minimum transfer amount into the threshold H and treat the margin agreement as having no minimum transfer amount.0} (6) When the threshold is not too small. the counterparty must provide the bank with collateral whenever the portfolio value exceeds a threshold.0} (5) The instantaneous collateral model is attractive because of its simplicity. the bank returns collateral to the counterparty. In the instantaneous collateral model. the collateral available to the bank is C (t ) = max {V (t ) − H . While the margin period of risk is not known with certainty. intuitive interpretation of our results for collateralized netting. A more realistic collateral model must account for the time lag between the last margin call made before default and the settling of the trades with the defaulting counterparty. we follow the standard practice and assume that it is a deterministic quantity that is defined at the margin agreement level. Counterparty portfolios with a combination of multiple netting agreements and trades outside of these agreements can be modeled in a straightforward way by a combination of Equations (2)(4).9 However. but it is very popular amongst banks because it greatly simplifies modeling. The counterpartylevel (margined) exposure is given by E (t ) = max {V (t )− C (t ).10 We assume that the collateral available to the bank at time t is determined by the portfolio value at time t − δ t according to C (t ) = max {V (t − δ t ) − H . but is rarely used in practice because its assumptions materially affect the exposure distribution. This time lag. which we denote by δ t . we assume that collateral is delivered immediately and that the trades can be liquidated immediately as well. Under these simplifying assumptions.When the netting agreement is further supported by a margin agreement. 0} where C (t ) is the collateral available to the bank at time t. This approximation is rather crude. We consider two models of collateral. is known as the margin period of risk. 9 5 .
We say that CVA contributions are additive when they sum up to the counterpartylevel CVA: CVA = ∑ CVA i i =1 N (10) 10 The margin period of risk depends on the contractual margin call frequency and the liquidity of the portfolio. Note that we have not made so far any assumptions on whether the exposure depends on the counterparty’s credit quality.3 Credit losses and CVA In the event that the counterparty defaults at time τ . CVA Contributions from EE Contributions We would like to develop a general approach to calculating additive contributions of individual trades to the counterpartylevel CVA. While more difficult to implement. which have more practical value. 2. The bank’s discounted loss due to the counterparty’s default is L = 1 τ ≤T (1 − R) E (τ ) D(τ ) { } (7) where 1{ A} is the indicator function that takes value 1 when logical variable A is true and value 0 otherwise. D(t ) is the stochastic discount factor process at time t. For example. conditional on the counterparty’s default at time t: ˆ ˆ e∗(t ) = E t [ D(t ) E (t )] ≡ E D(t ) E (t ) τ = t (9) Throughout this paper we use “star” to designate discounting and “hat” to designate conditioning on default at time t. This results in ˆ CVA = (1 − R) ∫ dP(t ) e∗(t ) 0 T (8) ˆ where e∗(t ) is the riskneutral discounted expected exposure (EE) at time t.We refer to this more realistic model as the lagged collateral model. it is often used by banks to obtain results. defined according to D (t ) = B0 Bt . We denote the contribution of trade i by CVA i . 3. 6 . with Bt the value of the money market account at time t. the bank recovers a fraction R of the exposure E (τ ) . δ t = 2 weeks is usually assumed for portfolios of liquid contracts and daily margin call frequency. The unilateral counterpartylevel CVA is obtained by applying the expectation to Equation (7).
we can focus on first calculating the CVA contribution of a transaction to its netting set. our goal is to allocate the nettingsetlevel exposure to the trades belonging to that netting set. at each future date. multiple netting agreements. from now on we focus on defining and calculating EE contributions. The allocation of the counterpartylevel EE across the netting sets is then trivial again because the counterpartylevel exposure is defined as the sum of the nettingsetlevel exposures. This follows from the fact that the risk function is homogeneous (of degree one) and the application of Euler’s theorem. 4. Thus. Note first that. the allocation of the counterpartylevel EE across the trades is trivial because the counterpartylevel exposure is the sum of the standalone exposures (Equation (2)) and expectation is a linear operator.. when there is more than one netting set with the counterparty (e. the risk contribution of a given transaction to the portfolio EC is determined by the infinitesimal increment of the EC corresponding to the infinitesimal increase of the transaction’s weight in the portfolio (see Chapter 4 in Arvanitis and Gregory (2001) or Tasche (2008) for details). 4. we develop the basic methodology to compute EE contributions and allocate portfoliolevel EE for noncollateralized netting sets. nonnettable trades). then the conditional discounted EE contributions must sum up to the portfolio conditional discounted EE: ˆ ˆ e∗(t ) = ∑ ei∗(t ) i =1 N (11) and the CVA contribution of trade i can be calculated from its EE contribution according to ˆ CVAi = (1 − R ) ∫ dP(t ) ei∗(t ) 0 T (12) Thus. EC is calculated at the portfolio level and then it is allocated to individual obligors and transactions. without netting agreements. Under the CMC method. To keep the notation simple. the problem of calculating CVA contributions reduces to that of ˆ calculating contributions of individual trades to the portfolio conditional discounted EE. Thus. ei∗(t ) . To obtain additive CVA contributions. we assume from now on that all trades with the counterparty are covered by a single netting set.Note that the recovery rate R and the default probabilities P (t ) are defined at the counterparty level in Equation (8). Furthermore.1 Continuous Marginal Contributions and Euler Allocation We derive EE contributions by adapting the continuous marginal contributions (CMC) method from the economic capital (EC) literature.g. Additive EE Contributions for Noncollateralized Netting Sets In this section. 7 .
t ) . When there is no margin agreement between the bank and the counterparty. t ) . we use the notations E (α. f (c x ) = c β f ( x ) . the counterpartylevel exposure is a homogeneous function of degree one in the trade weights: E (cα. and CVA(α ) for the exposure and EE at time t and CVA. If x denotes the vector of positions in a portfolio. xN ) is said to be homogeneous of degree β if for all c > 0 . t ) (16) 8 .2 Continuous Marginal EE Contributions for netted exposures without collateral Consider now the calculation of EE contributions. then Euler’s theorem implies additive capital contributions EC(x) = ∑ ECi (x) i =1 N (14) where the terms ECi (x) = ∂EC(x) ⋅ xi ∂xi (15) are referred to as the marginal capital contributions of the portfolio.. We describe adjusted portfolios via the vector of weights ˆ α = (α1 .A real function f (x) of a vector x = ( x1 .. Thus. Euler’s theorem is applied to allocate EC and compute risk contributions across portfolios. Assume that we can adjust the size of any trade in the portfolio by any amount.K. For adjusted portfolios.K. e∗ (α . . . If the function f (⋅) is piecewise differentiable.1) the vector representing the original portfolio. and EC(x) the corresponding economic capital. t ) = cE (α. for convenience. 4. are homogeneous functions of degree one ( β = 1 ) in the portfolio positions. These weights can assume any real value. such as standard deviation. with α i = 1 corresponding to the actual size of the trade and α i = 0 being the complete removal of the trade. Furthermore. Define the weight α i for trade i as a scale factor that represents the relative size of the trade in the portfolio. denote by 1 = (1. Vi (α i . then Euler’s theorem states that: β ⋅ f ( x) = ∑ i =1 N ∂f (x) ⋅ xi ∂xi (13) The risk measures most commonly used. valueatrisk (VaR) and expected shortfall. t ) = α iVi (t ) .α N ) .
0} {V (t ) >0} ∂α i α =1 ∂α i ∂α i α =1 α =1 Substituting Equation (20) into Equation (18). t ) ˆ ˆ ei∗(t ) = Et D(t ) ∂α i α =1 (18) where exposure of the adjusted portfolio (with weight vector α = (α1 . We can derive an expression for the marginal EE contributions as follows. scaled to the full trade amount: ˆ ei (t ) = lim δ →0 ∗ ˆ ˆ e∗(t . we have: ∂ E (α.α N ) ) is given by N E (α. the bank’s exposure doubles. or zero otherwise. 0 i =1 (19) Calculating the first derivative of the exposure with respect to the weight α i and setting all weights to one.K. We define the continuous marginal EE contribution of trade i at time t as the infinitesimal increment of the conditional discounted EE of the actual portfolio at time t resulting from an infinitesimal increase of trade i’s presence in the portfolio. the EE contributions defined by Equation (17) automatically sum up to the counterpartylevel conditional discounted EE by Euler’s theorem (Equation (13)). This results in ∂E (α. As expected. t ) ∂ ∂ V (α. First. t ). α ) = ∂α i α =1 (17) where ui describes a portfolio whose only component is one unit of trade i. Since the portfolio exposure is homogeneous in the trades’ weights. t ) = max ∑ α iVi (t ). the EE contributions sum up to the counterpartylevel discounted EE: 9 .The intuition behind Equation (16) is simple: if the bank uniformly doubles the size of its portfolio with the counterparty by entering into exactly the same trade with the counterparty for each existing trade. substitute Equation (9) into Equation (17) and bring the derivative inside the expectation. 1 + δ ⋅ ui ) − e∗(t ) δ ˆ ∂ e∗(t . t ) = = = Vi (t ) 1 1{V (α. we obtain the EE contribution of trade i: ˆ ˆ ei∗(t ) = Et D(t )Vi (t ) 1 {V (t ) >0} (20) (21) The EE contribution of trade i is the expectation of a function which considers the discounted values of the trade on all scenarios where the total counterparty exposure is positive.t ) > 0} max {V (α.
we specify additive EE contributions for both models.1 Instantaneous Collateral Model Substituting Equation (5) into Equation (4). t ) ⋅ 1 0<V { (α . In what follows. the counterpartylevel exposure is not affected by the infinitesimal weight changes. The first derivatives of the exposure with respect to the trade weights is given by 10 . the function E ( α′. From the mathematical point of view. To understand conceptually how the CMC method fails. notice that. That is. which are consistent with the continuous marginal contributions as follows. and the exposure contribution is zero for all scenarios with the portfolio value above the threshold. As discussed in Section 2. while the exposure function in Equation (22) is not homogeneous in the vector of weights α = (α1 . α H ) . the counterpartylevel exposure equals the threshold. and the CMCs. hence. starting with the simpler instantaneous collateral model. the conditions of Euler’s theorem are not satisfied.K.. which covers all the trades with the counterparty.. do not sum to the counterpartylevel discounted EE anymore. we still would like to “allocate” the nonzero collateralized counterpartylevel exposure (equal to the threshold) to the individual trades. so that these allocations cannot be all equal to zero.. the counterpartylevel stochastic exposure is given by Equation (4). Thus.α N ) .t ) <α H H } + α H H ⋅1 V { (α . However.t ) >α H H } (23) is a homogeneous function in the extended vector of weights α ' = (α1 . we can think of the contribution of the threshold itself to the counterpartylevel exposure. the expected exposure in not a homogeneous function of the trades’ weights and. Thus. where the collateral available to the bank is given either by the instantaneous collateral model (Equation (5)) or by the lagged collateral model (Equation (6)). α N . the CMC approach cannot be applied directly. as given earlier. t ) = V ( α. notice that when the portfolio value is above the threshold. Additive EE Contributions for Collateralized Netting Sets Consider now a counterparty that has a single netting agreement supported by a margin agreement.. First. we obtain E (t ) = V (t )1 0<V (t ) < H + H 1 V (t ) > H { } { } (22) As can be seen from Equation (22). 5. An infinitesimal increase of any trade’s weight in the portfolio is not sufficient to bring the portfolio value below the threshold. We can derive additive contributions for this nonhomogeneous case.0} = e∗(t ) ˆ t ˆ {V (t ) >0} 5. we consider scaling the each of the trades as well as the threshold H.ˆ ˆ ∑ e∗(t ) = E i i =1 N t D (t )V (t ) 1 = E D (t ) max {V (t ).
By applying discounting and taking conditional expectation of the righthand side of Equations (24) and (25). ei∗H (t ) = E t D(t ) ⋅ Vi (t ) ⋅ 1{0<V (t ) ≤ H } and of the threshold ˆ ˆ∗ eH (t ) = H ⋅ E t D(t ) ⋅ 1{V (t ) > H } which satisfy ˆ ˆ. when the threshold goes to infinity. Given that that the adjustment of the trade contributions occurs when the portfolio value exceeds the threshold. a meaningful weighting scheme is given by the ratio of the individual instrument’s expected discounted value when the threshold is crossed to the total counterparty discounted value when this occurs: 11 . As the final step. so that Equation (28) can be written in terms of EE contributions only of the trades (as in Equation (11)): ˆ ˆ e∗(t ) = ∑ ei∗(t ) i =1 N ˆ∗ There are several possibilities for allocating the amount eH (t ) in meaningful proportions to each trade. t ) = H ⋅ 1 V (t ) > H { } ∂α H α ′=1 (24) (25) Note that these sum up to the counterpartylevel exposure given by Equation (22). we “allocate back” the contribution adjustment of the collateral threshold given by Equation (27) to the individual trades. t ) = Vi (t ) ⋅ 1 0<V (t ) ≤ H { } ∂α i α ′=1 Similarly. the derivative with respect to the threshold weight is ∂E (α′. ˆ∗ e∗(t ) = ∑ ei∗H (t ) + eH (t ) i =1 N (26) (27) (28) The contribution of the threshold can be interpreted as the change of the conditional discounted EE associated with an infinitesimal shift of the threshold upwards scaled up by the actual size of the threshold.∂E (α′. the last term vanishes and we recover the uncollateralized contributions. as expected. Note that. we obtain the EE contributions of the trades ˆ ˆ.
δ t . which gives the exposure with instantaneous collateral. the counterparty credit exposure is obtained by substituting Equation (6) into Equation (4): E (t ) = V (t ) ⋅ 1 0<V (t ) ≤ H +δ V (t ) + [ H + δ V (t )] ⋅ 1 0< H +δ V (t ) <V (t ) { } { } (33) where δ V (t ) = V (t ) − V (t − δ t ) is the change of the portfolio value from the lookback time point t − δ t to the time point of interest t . In this case. while the second term is the contribution of all scenarios where the bank holds nonzero collateral at time t. vanishes. An alternative allocation scheme (type B) is obtained by bringing the weighting scheme of the threshold contribution now inside the expectation operator. this expression reduces to Equation (22).2 Lagged Collateral Model (32) We now apply the formalism developed for the continuous collateral model to the lagged collateral model. Note that as the margin period of risk.ˆ E t D(t ) ⋅ Vi (t ) ⋅ 1{V (t ) > H } N = ⋅ ˆ D(t ) ⋅ 1 ˆ D(t ) ⋅ 1 Et ∑ Et V (t ) > H } V (t ) > H } { { E D(t ) ⋅ V (t ) ⋅ 1 ˆ i =1 t {V (t ) > H } Thus. the individual trade contributions to EE are given by (29) ˆ H ⋅ Et D(t ) ⋅ 1{V (t ) > H } ˆ ⋅ E D(t )V (t ) ⋅ 1 D(t )V (t ) ⋅ 1 + ˆ ˆ ei (t ) = E t i t i {0<V (t ) < H } {V (t ) > H } (30) ˆ E t D(t ) ⋅ V ⋅ 1{V (t ) > H } ∗ Both terms of Equation (30) have a straightforward interpretation: the first term is the contribution of all scenarios where the bank holds no collateral at time t. 5. 12 . so that instead of Equation (29) we now have: N ∑ i =1Vi (t ) = N E D(t ) ⋅ Vi (t ) ⋅1 ˆ ˆ ∑ ˆt (31) E t D (t ) ⋅ 1{V (t ) > H } = ∑ E t D(t ) ⋅ 1{V (t ) > H } ⋅ {V (t ) > H } V (t ) i =1 V (t ) i =1 N This leads to the continuous marginal contributions given by V (t ) ˆ ˆ ˆ ei* (t ) = Et D(t )Vi (t ) ⋅ 1{0<V (t ) < H } + H ⋅ E t D(t ) ⋅ i ⋅ 1{V (t ) > H } V (t ) Both terms on the right hand of Equation (32) have the same interpretation as before. We refer to the allocation scheme above as type A allocation.
t ) = Vi (t ) ⋅ 1 0<V (t ) ≤ H +δ V (t ) + δ Vi (t ) ⋅ 1 0< H +δ V (t ) <V (t ) { } { } ∂α i α ′=1 ∂E (α′. This results in the trade allocations given by 13 . ei∗H (t ) = E t D(t ) ⋅ Vi (t ) ⋅ 1 0<V (t ) ≤ H +δ V (t ) + Et D(t ) ⋅ δ Vi (t ) ⋅ 1 0< H +δ V (t ) <V (t ) { } { } (38) and of the threshold ˆ ˆ∗ eH (t ) = H ⋅ E t D(t ) ⋅ 1 0< H +δ V (t ) <V (t ) { } (39) Now we need to allocate back the threshold contribution. t )] ⋅ 1 0<α { H H +δ V (α . The first derivatives of exposure with respect to the trades’ weights and with respect to the threshold weight are given. As defined for the instantaneous model. eH (t ) is allocated to individual trades in proportion to the expectation of the discounted trade values when 0 < H + δ V (t ) < V (t ) .t )} + [α H H + δ V (α.. a zero margin period of risk implies a zero change of portfolio value from t − δ t to t . t ) = H ⋅ 1 0< H +δ V (t ) <V (t ) { } ∂α i α ′=1 (36) (37) The sum these first derivatives across all the trades and the threshold. we obtain the EE contributions of the trades ˆ ˆ ˆ. Following the type A allocation scheme in the previous section. By applying discounting and taking the conditional expectation of the righthand side of Equations (36) and (37). to the individual ˆ∗ trades. t ) = V (α. by ∂E (α′. Equation (39). t ) = ∑ α iδ Vi (t ) i =1 N (35) and δ Vi (t ) = Vi (t ) − Vi (t − δ t ) . we rewrite the exposure given by Equation (33) as a homogeneous function in the extended vector of weights α ' = (α1 . t ) ⋅ 1 0<V ( α .α N ..t ) ≤α { H H +δ V (α .Indeed.t )} (34) where δ V (α. α H ) : E (α′.. since both time points are the same now. respectively.. gives the counterpartylevel exposure.t ) <V (α . Equation (33).
In general. Then. 6. 6.1 Exposure Independent of Counterparty’s Credit Quality Consider first the case where the exposures are independent of the counterparty’s credit quality. banks implicitly assume that each counterparty’s exposure is independent of that counterparty’s credit quality when exposures are simulated separately. This gives ˆ ˆ ˆ ei∗(t ) = E t D(t ) ⋅ Vi (t ) ⋅ 1 0<V (t ) ≤ H +δ V (t ) + Et D(t ) ⋅ δ Vi (t ) ⋅ 1 0< H +δ V (t ) <V (t ) { } { } V (t ) ˆ + H ⋅ Et D(t ) ⋅ i ⋅ 1 0< H +δ V (t ) <V (t ) { } V (t ) (41) It is straightforward to verify that the lagged EE contributions degenerate to the instantaneous EE contributions when δ t = 0 . Calculating CVA Contributions by Simulation Banks commonly use Monte Carlo simulation in practice to obtain the distribution of counterpartylevel exposures. For the ease of exposition. conditioning on τ = t in the expectations in Equations (19). (30) and (32) become unconditional. we assume that all the trades with the counterparty are nettable and that collateral (if there is any) can be described by the instantaneous model. Substituting δ V (t ) = 0 into Equations (40) and (41). the counterpartylevel CVA can be calculated in a Monte Carlo simulation as follows: 14 . Based on these simulations a bank can also compute the counterpartylevel CVA. we show how the calculation of EE contributions can be easily incorporated to the Monte Carlo simulation of the counterpartylevel exposure that banks already perform. (29). we bring the weighting scheme inside the expectation operator. Let us now make this assumption explicitly. The simulation algorithm for calculating counterpartylevel CVA can be extended to calculate CVA contributions. In this section. respectively. First. and these conditional expectations can be replaced by the unconditional ones.ˆ ˆ ˆ ei∗(t ) = E t D(t ) ⋅ Vi (t ) ⋅ 1 0<V (t ) ≤ H +δ V (t ) + Et D(t ) ⋅ δ Vi (t ) ⋅ 1 0< H +δ V (t ) <V (t ) { } { } ˆ Et D(t ) ⋅ Vi (t ) ⋅ 1 0< H +δ V (t ) <V (t ) { } D(t ) ⋅ 1 ⋅ ˆ + H ⋅ Et 0 < H +δ V (t ) <V (t )} { E D(t ) ⋅ V (t ) ⋅ 1 ˆ t {0< H +δ V (t )<V (t )} (40) In the type B allocation. we obtain Equations (30) and (32).
After running large enough number M of market scenarios. Generate a market scenarios j (interest rates. where. Step 4: CVA contributions are computed as 15 .) for each of the future time points tk 2. calculate the trade’s exposure contribution for scenario j Ei∗( j ) (tk ) . 4. 0 } (if there is no margin agreement. calculate collateral d. If there is margin agreement. H Vi ( j ) (tk ) V ( j ) (tk ) if V ( j ) (tk ) > H . compute CVA as CVA = (1 − R ) ∑ k e∗(t k )[ P (tk ) − P (tk −1 )] . as before. FX rates. the EE contributions given by Equation (32). ( j) ( j) C (tk ) = max{ V (tk ) − H . R denotes the (constant) recovery rate and P(t ) is the unconditional cumulative probability of default up to time t . Step 3: For each trade i. Calculate portfolio value V ( j ) (tk ) = ∑ i =1Vi ( j ) (tk ) N c. which is equal to Vi ( j ) (tk ) if 0 < V ( j ) (tk ) ≤ H . 0 } available at time tk . etc. and zero otherwise. For each simulation time point tk and scenario j : a. The calculation of EE and CVA contributions can be incorporated to this algorithm as follows. calculate trade value Vi ( j ) (tk ) b. For each trade i. C ( j ) (tk ) ≡ 0 ). The following calculations are added to Steps 24: Step 2: For each trade i. Consider. for example. Calculate counterpartylevel exposure E ( j ) (tk ) = max{ V ( j ) (tk ) − C ( j ) (tk ). compute the discounted EE by averaging over all the market scenarios at each time point: 1 e∗(tk ) = M ∑ M j =1 D ( j ) (tk ) E ( j ) (tk ) . compute the discounted EE contribution by averaging over all the market scenarios at each time point: 1 ei∗(t k ) = M ∑ M j =1 D ( j ) (t k ) Ei( j ) (t k ) .1. Finally. 3.
g. or on the entire path of the intensity process λ (⋅) from zero to t. there may be dependence between them which can come from two sources: Right/wrongway risk. The risk is called rightway (wrongway) if exposure tends to decrease (increase) when counterparty quality worsens. However. Both types of dependence of exposure on the counterparty’s credit quality can be incorporated in the EE and CVA calculation if the trade values and credit quality of the bank’s counterparties are simulated jointly. right/wrongway risk is always present. but it is usually ignored to simplify exposure modeling. under which the bank can terminate the trades with the counterparty when the counterparty’s rating falls below a prespecified level.). If a bank’s counterparty risk simulation environment is capable of such joint simulation. credit derivatives. The counterpartylevel exposure E (t ) may depend either on the intensity value λ (t ) at time t .CVAi = (1 − R)∑ k ei∗(tk )[ P(tk ) − P (tk −1 )] . etc. 16 . More generally. One example such agreements is a margin agreement with the threshold dependent on the counterparty’s credit rating.. 6. 11 For an overview of stochastic default intensity and the associated Cox process. Another example is an early termination agreement. Let us introduce a stochastic default intensity process λ (t ) without specifying its underlying dynamics. Exposurelimiting agreements that depend on the counterparty credit quality. there are cases when right/wrong way risk is too significant to be ignored (e. commodity trades with a producer of that commodity.11 This intensity can be used as a measure of counterparty credit quality: higher values of the intensity correspond to lower credit quality. Strictly speaking. the calculation of EE and CVA contributions is also straightforward. see Chapter 5 in Schönbucher (2003).2 Exposure Dependent on Counterparty’s Credit Quality The algorithm above assumes independence between the exposure and the counterparty’s credit quality. We can use the intensity process to convert the expectation conditional on default at time t in Equation (21) to an unconditional expectation so that the conditional EE contribution becomes ˆ ei∗(t ) = t 1 E λ (t ) exp − ∫ λ ( s )ds D(t ) Ei (t ) P′(t ) 0 (42) where P′(t ) is the first derivative of the cumulative PD P(t ) . A short derivation of Equation (42) is given in Appendix 1.
To facilitate such calculations. For each trade i.. Analytical CVA Contributions under a Normal Approximation It is also useful in practice to estimate EE and CVA contributions quickly outside of the simulation system.. In this case. 17 . for the case when trade values are normally distributed. given the dependence of exposures on the counterparty credit quality. tk − 2 . add λ (tk ) P′(tk ) exp −∑ j =1 λ (t j −1 )(t j − t j −1 ) D(tk )h[λ (tk )] ( k ) Vi (tk ) V (tk ) After running large enough number of market scenarios. calculate its trade value Vi (tk ) 3. the intensity process λ (⋅) needs to be simulated jointly with the market risk factors that determine trade values. If there is no margin agreement. at time t.As described for unconditional EE contributions. add λ (tk ) P′(tk ) exp −∑ j =1 λ (t j −1 )(t j − t j −1 ) D(tk )Vi (tk ) ( k ) − if V (t ) > h[λ (tk )] . add λ (tk ) P′(tk ) exp −∑ j =1 λ (t j −1 )(t j − t j −1 ) D(tk )Vi (tk ) ( k ) b. For each trade i. then − if V (t ) > 0 . EE contributions are obtained by dividing the EE contribution counter by the number of scenarios. see Pykhtin and Zhu (2007). the calculations for conditional EE contributions can be performed during a Monte Carlo simulation of exposures. ). then − if 0 < V (t ) ≤ h[λ (tk )] .. Calculate the portfolio value V (t ) = ∑ i =1Vi (t ) N 4. This joint simulation is done pathbypath: simulated values of the intensity and of the market factors at time tk are obtained from the corresponding simulated values at the earlier time points ( tk −1 . the algorithm for computing CVA contributions for time tk can be expressed as follows: 1. For simplicity. If there is a margin agreement with an intensitydependent threshold h[λ (⋅)] . the distribution of trade values is given 12 For a discussion on pathbypath vs.12 Assuming that we have already simulated the market factors and the intensity for times t j for all j < k . we assume that. 7. and to avoid dealing with stochastic discounting factors. update the EE contribution counter a. Jointly simulate market risk factors and intensity λ (tk ) at time tk 2. we derive analytical EE contributions. direct jump to simulation date.
Assume that the value Vi (t ) of trade i at each future time t is normally distributed with expectation µi and standard deviation σ i under the forward to t probability measure: Vi (t ) = µi (t ) + σ i (t ) X i (45) where X i is a standard normal variable. t ) e(t ) (44) where B(0. the discounted portfolio value V (t ) is normally distributed: V (t ) = µ (t ) + σ (t ) X (46) where X is another standard normal variable. t ) E E (t ) τ = t ≡ B(0. between the discounted trade values) are denoted by rij . t ) is the timezero price of the riskfree zerocoupon bond maturing at time t. and the mean and standard deviation of the portfolio value are given by µ (t ) = ∑ µi (t ) i =1 N .V j (t )] σ i (t ) σ (t ) = ∑ rij j =1 N σ j (t ) σ (t ) (48) Using this correlation. the discounted conditional EE in Equation (9) can be written as ˆ ˆ e∗(t ) = B(0. We can calculate this correlation as follows: cov[Vi (t ). Under this measure. we can represent X i as X i = ρi (t ) X + 1 − ρi2 (t ) Z i (49) 18 . t ) e(t ) (43) and the discounted unconditional EE is e∗(t ) = B(0. t ) E [ E (t ) ] ≡ B(0. σ 2(t ) = ∑∑ rij σ i (t ) σ j (t ) i =1 j =1 N N (47) Denote by ρi (t ) the correlation between the value Vi (t ) of trade i and the portfolio value V (t ) .V (t )] = ρi (t ) = σ i (t ) σ (t ) ∑ N j =1 cov[Vi (t ).under the forward (to time t) probability measure. Since the sum of normal variables is also normal. Correlations between these standard normal variables (and. This change of measure allows us to work with undiscounted EEs and EE contributions. therefore.
Let us consider now the EE contributions given by Equation (32) (type B allocations). we obtain µ (t ) + σ i (t ) X i 1 µ (t )+σ (t ) X > H ei (t ) = E ( µi (t ) + σ i (t ) X i ) 1 0< µ (t )+σ (t ) X < H + H E i { } { } µ (t ) + σ (t ) X (51) Appendix 2 shows that. the counterpartylevel stochastic exposure is given by Equation (22). with no margin agreement. is obtained as the limiting case when the threshold goes to infinity. after some analytical manipulation. The simpler case.where Z i is a standard normal random variable independent of X (and different for each trade). Evaluating the integrals yields an analytical formula for the EE: µ (t ) µ (t ) − H e(t ) = µ (t ) Φ − Φ σ (t ) σ (t ) µ (t ) µ (t ) − H +σ (t ) φ −φ σ (t ) σ (t ) µ (t ) − H + H Φ σ (t ) (50) where Φ(⋅) is the standard normal cumulative distribution function. Substituting Equation (46) into Equation (22) and taking the expectation. this EE contribution can be written as 19 . For the clarity of exposition. Removing the discounting and substituting Equations (45) and (46) into Equation (32).1 Exposure Independent of Counterparty’s Credit Quality We first calculate counterpartylevel EE and EE contributions assuming independence between exposures and counterparty credit quality. 7. We obtain the results for the general case of a netting agreement with a margin agreement. In the presence of a margin agreement. we assume the instantaneous collateral model. we obtain e(t ) = E [ µ (t ) + σ (t ) X ] 1 0< µ (t ) +σ (t ) X < H + H E 1 µ (t ) +σ (t ) X > H { { } } H − µ (t ) σ (t ) = − ∫ µ ∞ [ µ (t ) + σ (t ) x ]φ ( x)dx + H (t ) σ (t ) H − (t ) σ (t ) ∫µ φ ( x)dx where φ (⋅) is the probability density of the standard normal distribution.
This leads to the counterpartylevel EE µ (t ) µ (t ) e(t ) = µ (t ) Φ + σ (t ) φ σ (t ) σ (t ) (54) and the EE contributions µ (t ) µ (t ) ei (t ) = µi (t ) Φ + σ i (t ) ρi (t ) φ σ (t ) σ (t ) 7.2 Right/WrongWay Risk (55) We now lift the independence assumption to accommodate right/wrongway risk. For the case without a margin agreement. Note that the EE contributions obtained in the previous section are contributions of the trades in the portfolio to the counterpartylevel unconditional discounted EE. Similarly. and do not require numerical integration. we take the limit H → ∞ of Equations (50)(53). We need to modify the approach to obtain the contributions to the counterpartylevel EE conditional on the counterparty defaulting at the time when the exposure is measured. µ (t ) µ (t ) µ (t ) − H µ (t ) − H ei (t ) = µi (t ) Φ − Φ + σ i (t ) ρi (t ) φ −φ σ (t ) σ (t ) σ (t ) σ (t ) µi (t ) + σ i (t ) ρi (t ) x +H ∫ φ ( x)dx µ (t ) + σ (t ) x H − µ (t ) σ (t ) ∞ (52) where the remaining integral can be easily evaluated numerically. One can verify that EE contributions given by Equation (52) sum up to the counterpartylevel EE. Equation (50). An obvious approach is to define an intensity process and compute the conditional EE contributions as the expectation over all 20 . we obtain the EE contributions corresponding to type A allocations as µ (t ) µ (t ) − H ei (t ) = µi (t ) Φ −Φ σ (t ) σ (t ) µ (t ) − H +H ⋅ Φ σ (t ) ⋅ µ (t ) µ (t ) − H + σ i (t ) ρi (t ) φ −φ σ (t ) σ (t ) µi (t ) Φ µ (t ) − H µ (t ) − H + σ i (t ) ρi (t ) φ σ (t ) σ (t ) µ (t ) − H µ (t ) − H µ (t ) Φ + σ (t ) φ σ (t ) σ (t ) (53) In contrast to Equation (52). the EE contributions given by Equation (53) are given in closed form.
depends on Y according to ˆ X i = bY + 1 − bi2 X i i (61) The Normal copula framework was first proposed by Li (2000) to model correlated default times for pricing portfolio credit derivatives. 14 13 See for example Garcia Cespedes et al. 21 . each possible default time is mapped to a unique value of Y. which drives the value of trade i. More specifically. we first map the counterparty’s default time τ to a standard normal random variable Y: Y = Φ −1[ P(τ )] (56) where Φ −1(⋅) is the inverse of the standard normal cumulative distribution function. but this requires a Monte Carlo simulation. Since the counterparty’s cumulative probability of default P(⋅) is a monotonic function.14 Thus. Thus. (2009) for the application of such a copula approach to compute counterparty credit capital. For this purpose. we define a Normal copula13 to model the codependence between the counterparty’s credit quality and the exposures. we assume that the standard normal risk factor X i . The counterpartylevel conditional EE is given by ˆ( e t ) = E E (t ) τ = t (57) while the conditional EE contribution of trade i is given by ˆ ei (t ) = E Ei (t ) τ = t (58) where E (t ) is the counterpartylevel exposure and Ei (t ) is the stochastic exposure contribution of trade i at time t. we develop an alternative. simpler approach that results in closed form expressions for the conditional EE contributions. we can replace the conditioning on τ in Equations (57) and (58) with the conditioning on Y and write the counterpartylevel conditional EE and the EE contributions as ˆ( e t ) = E E (t ) Y = Φ −1[ P(t )] ˆ ei (t ) = E Ei (t ) Y = Φ −1[ P(t )] (59) (60) We model the right/wrongway risk by allowing trade values to depend on Y. In this section.possible paths of the intensity process (See Appendix 1).
we now use the conditional ones. they show that the conditional counterpartylevel uncollateralized EE is described by the same expression as the unconditional EE. 22 .1 (Equations (50)(55)) after putting “hats” on the parameters: 15 ˆ µi (t ) ≡ E[ Vi (t )  τ = t ] = µi (t ) + σ i (t )bi Φ −1[ P(t )] ˆ σ i (t ) ≡ StDev[ Vi (t )  τ = t ] = σ i (t ) 1 − bi2 ˆ µ (t ) ≡ E[ V (t )  τ = t ] = µ (t ) + σ (t ) β (t ) Φ −1[ P(t )] (64) (65) (66) 15 This conclusion is consistent with the results in Redon (2006). Wrongway risk occurs when −1 ≤ bi < 0 (the value of trade i tends to increase when Y declines). One approach may be to calculate conditional expectations in the same manner as we have calculated the unconditional ones in the previous Section. As the magnitude of bi . after replacing the unconditional expectations and standard deviations of the trade values with the conditional ones. The only difference is that instead of the unconditional expectations. in Appendix 2 we show how this can be done in a faster and more elegant way. increases. When bi = 0 . Y ] = cov[V (t ). Therefore. also depends on Y: ˆ X = β (t )Y + 1 − β 2(t ) X (62) ˆ with X a standard normal variable independent of Y. The portfolio factor loading β (t ) can be computed from the individual trade factor loadings bi as follows: β (t ) = cov[ X . the conditional exposure model can be formulated the in exactly the same mathematical terms as the unconditional model. If the portfolio contains trades with nonzero bi . Y ] N cov[Vi (t ). the standard normal risk factor X. Similarly. so does the codependence between the trade value and the counterparty credit quality. which drives the portfolio value. the value of trade i is independent of the counterparty credit quality. The parameter bi drives the right/wrongway risk. Using a different model of right/wrongway risk. we can use all the results of Subsection 7. standard deviations and correlations that specify the behavior of the trade values.ˆ where X i is a standard normal variable independent of Y. Y ] N σ i (t ) =∑ = ∑ bi σ (t ) σ (t ) σ (t ) i =1 i =1 (63) Now we have all the ingredients to derive the counterpartylevel EE and EE contributions in the presence of right/wrongway risk. there is rightway risk when 0 < bi ≤ 1 (the value of trade i tends to decrease when Y declines). In particular. However.
The change of the standard deviation of the portfolio value resulting from increasing the weight of trade i by δ can be calculated as StDev[V (t ) + δ Vi (t )] − StDev[V (t )] = ( var[V (t )] + 2δ cov[V (t ). Scenarios with positive portfolio value contribute the same amount δ ⋅ µi (t ) to the exposure change. Thus. According to Equation (17). Netting & no margin Equation (55) can be understood from the incremental viewpoint of the CMC method. Let us first ignore the change of the standard deviation and consider how a uniform shift of the entire distribution by δ ⋅ µi (t ) affects the counterpartylevel EE. the first term in the righthand side of Equation (55) describes the increment of the counterpartylevel EE resulting from the infinitesimal uniform shift of the portfolio value distribution associated with an increase of the weight of trade i. The second term of Equation (55) describes the change of the width of the portfolio value distribution. the EE contribution of trade i is determined by the infinitesimal change of the counterpartylevel EE resulting from an infinitesimal increase of the weight of trade i in the portfolio. Vi (t )] + δ 2 var[Vi (t )]) 2 − σ (t ) 1 1 = (σ 2 (t ) + 2δ ρi (t ) σ i (t ) σ (t ) + δ 2 var[Vi (t )]) 2 − σ (t ) = δ ρi (t ) σ i (t ) + O(δ 2 ) 23 . while scenarios with negative portfolio value contribute nothing. the expectation of portfolio value changes by δ ⋅ µi (t ) . It is straightforward to verify that Pr[V (t ) > 0] = Φ[ µ (t ) /σ (t )] . The effect of an increase of the weight of a trade on the portfolio value distribution can be viewed as the sum of two effects: o a uniform shift of the distribution o a change of width of the distribution Let us consider these two effects separately. If the weight of trade i is increased by δ .ˆ σ (t ) ≡ StDev[ V (t )  τ = t ] = σ (t ) 1 − β 2(t ) (67) ˆ ρi (t ) = ρi (t ) − bi β (t ) (1 − bi2 )[1 − β 2(t )] (68) 7. Therefore. the increment of the EE will be given by the product of the magnitude of the shift δ ⋅ µi (t ) and the probability of the portfolio value being positive.3 Remarks on the Analytical Formulae In this section we briefly comment on the properties and interpretation of the analytical contributions derived in this section.
Moreover. Because of this. Indeed. correlation ρi (t ) is the correlation between the values of trade i and the portfolio that includes trade i itself. ρi (t ) depends on the ratio σ i (t ) / σ (t ) (see Equation (48)). t ) − E (t ) = δ ρi (t ) σ i (t ) V (t ) − µ (t ) 1{V (t )>0} σ (t ) (69) The second term of Equation (55) can be obtained by taking the expectation of the righthand side of Equation (69). t ) = σ (t ) + δ ρi (t ) σ i (t ) is given by16 E (δ . 24 . the true dependence of EE contribution on trade parameters is nonlinear. Thus. Since trade i is part of the portfolio. A widening distribution (with no accompanying shift) always increases the counterpartylevel EE. It appears that Equation (55) has simple linear dependence on µi (t ) and the product ρi (t ) σ i (t ) . while narrowing always decreases it. for a given realization of the portfolio value V (t ) . and narrowing if the correlation is negative. The third term results from the allocation of exposure when the portfolio value is above the threshold.where O(δ 2 ) denotes the terms of the second order and higher that can be ignored. Netting & margin In this case. unless trade i represents a negligible fraction of the portfolio. Equation (52) can be rewritten as ∞ µ (t ) − H φ (ξ )dξ µ (t ) ei (t ) = µi (t ) Φ −Φ +H ∫ σ (t ) H − µ ( t ) µ (t ) + σ (t )ξ σ (t ) σ (t ) ∞ µ (t ) − H µ (t ) φ (ξ )ξ dξ +σ i (t ) ρi (t ) φ −φ +H ∫ H − µ ( t ) µ (t ) + σ (t )ξ σ (t ) σ (t ) σ (t ) (70) 16 This can be immediately seen from Equation (46). Thus. the change of exposure associated with the change of the standard deviation of the portfolio value from σ (t ) to σ (δ . only the first two terms of Equation (52) allow interpretation from the incremental viewpoint of the CMC method: the first term can be explained as the effect of the uniform shift and the second term as the effect of the widening or narrowing of the portfolio value distribution. the portfolio value distribution is widening if the correlation ρi (t ) is positive. this is only part of the true dependence. However. An attempt to use the CMC method would give zero EE contribution from V (t ) > H scenarios. µ (t ) depends on µi (t ) and σ (t ) depends on σ i (t ) and the correlation of trade i with the rest of the portfolio.
the conclusions apply equally to the case of wrongway risk by simply using conditional expectations. An interesting property of Equation (64) is its dependence on the counterparty’s PD. 8. let us consider the bank entering into the same trade with an investmentgrade counterparty A and with a speculativegrade counterparty B. For the case of wrong (right) way risk. µ (t ) µ (t ) ei (t ) = µi (t ) Φ + σ i (t ) ρi (t ) φ σ (t ) σ (t ) The EE contribution of instrument i is a function of: 25 . instead of unconditional ones. To understand this. For ease of exposition. the shift of the distribution is described by Equation (64).1 Contributions for a Noncollateralized Portfolio As a first step to understand this behavior. If the correlation is positive (rightway risk). consider Equations (54) and (55). Note that this is not specific to the normal approximation. this EE contribution appears to be linear in µi (t ) and the product ρi (t ) σ i (t ) . the distribution shifts up. as discussed earlier in Section 7. We are interested in the trade value distribution conditional on the counterparty’s default at the time of observation. in the case when there is no margin agreement in place: µ (t ) µ (t ) e(t ) = µ (t ) Φ + σ (t ) φ σ (t ) σ (t ) . as well as market and credit independence. its value distribution at time t conditional on the counterparty’s default at time t differs from its unconditional value distribution. However. 8. but is a general property not related to any model. Therefore. the distribution becomes narrower. and the narrowing is described by Equation (65).As in the noncollateralized case. Right/wrongway risk If the value of trade i is correlated with the counterparty’s credit quality. Under the normal approximation. the deterioration of credit quality to the point of default is larger for counterparty A.2. which give the counterpartylevel EE and the EE trade contributions. we present some simple examples that illustrate the behavior of exposure (and hence CVA) contributions. the distribution shifts down. In both cases. the deterioration of the counterparty’s credit quality to the point of default pushes trade values higher (lower). volatilities and correlations. we assume that trade values are Normal. and then show the impact of adding a collateral agreement to the portfolio. trade values conditional on default of A are shifted more than trade values conditional on default of B. We first present an example when there is no collateral agreement in place. Examples In this section. but the true dependence on these quantities is more complex due to the extra dependence of µ (t ) and σ (t ) on these quantities. if the correlation is negative (wrongway risk). Since counterparty A is “further away” from default than counterparty B.
while a high ratio weighs mean values much more. 26 . and a volatility component (second term).16. µi the volatility contribution.0 Norm Dist 0. Figure 1 plots these weights as a function of µ/σ.2 0. respectively. A low ratio weighs the volatility contribution much higher. Volatility and mean exposure weights for EE contributions. which comprises of 5 transactions over a single step. Table 1 gives the individual trade’s mean value. the volatility component weight is 2. and bears no impact on the analysis. P1 P2 P3 P4 P5 Total 17 This assumption is only made for simplicity. if µ/σ = 2. Thus. For example. the overall level of the counterparty portfolio’s mean value and volatility determine how the individual instrument’s mean and volatility contribution are weighted to yield the EE contributions. consider now the simple counterparty portfolio. In contrast. 1. ρiσi the overall value of the ratio µ/σ (for the entire portfolio) where we have dropped the time variable notation for brevity. We assume that trade values are independent. To illustrate the impact of various parameters on EE contributions. the portfolio’s ratio µ/σ = 3. The portfolio has a mean value and variance of 10. by the Normal distributions and density evaluated at the ratio µ/σ (for the entire counterparty portfolio). Alternatively we can simply use the volatility contributions directly for any correlated model.6 0. Both the counterpartylevel EE and the trade contributions can be seen as the sum of two components: a mean value component (first term in the equations).the mean value contribution. variance and volatility (in dollar values and % contributions).4 times the mean value weight.4 0.8 Norm Density 0. µ/σ = 2 results in mean values being weighted 18 times the volatilities.0 4 3 2 1 0 Mu/Sigma 1 2 3 4 Figure 1.17 In this case. These components weigh the mean value (or mean value contributions) and the volatility (volatility contribution).
In this case.2 σ2 % σ Table 1. As µ/σ increases beyond zero. while leaving intact the volatility.001. Now. as functions of the portfolio’s µ/σ. The EE for the portfolio is 10.03%. variances and volatilities.0 10 100% 10 100% 3. the volatility component decreases and. 10. and the volatility component compensates for this to generate positive EEs. For large negative portfolio mean values. 20. µ.7 2 20% 2 20% 1. we compute the EE and contributions for the portfolio.0 4 40% 0 0% 0. and lowest volatility (0). while position 5 has the highest mean value (4). Using Equations (54) and (55). as well as its mean and volatility components. 39. once µ/σ > 2.00%. The portfolio is constructed so that for each trade. the mean value component of EE is actually negative.µ % 0 0% 4 40% 2 1 10% 3 30% 1. Figure 2 plots the EE. but it will also help highlight some of the points below. as well as the percent contributions of each instrument to the mean exposure and volatility in table 1. has the lowest mean (0) and largest volatility (2). 16 14 12 Expected Exposure 10 8 6 4 2 0 2 2 0 Mu/Sigma 2 4 EE mu component sigma component 27 .992).4 3 30% 1 10% 1. P1. This allows us to express the trade contributions in terms of how deep in. thus the first instrument. The trade contributions to EE are fairly close to the contributions to the mean values in Table 1 (0.or outofthemoney the counterparty portfolio is (relative to its volatility). we vary the overall mean value of the portfolio. This may not only be reasonably realistic. the EE is completely dominated by the mean value.02%.97%). its mean value and volatility are inversely related. Portfolio – mean values. 29. the EE (red line) is zero.98%. with most of this arising from the mean value component (9. σ.
produce very large negative EE contributions. trade contributions converge to those of uncollateralized exposures. Consider the same portfolio in table 1. In very general terms: As the threshold becomes very large. For this particular symmetric portfolio. With lower thresholds. There is a clear shift in dominance between the mean and volatility components as the portfolio’s mean value increases. when the mean portfolio values are negative. Figure 4 shows the reductions in EE as a result of the margin agreement (as % of uncollateralized EE) as a function of µ/σ .2 Contributions for a Collateralized Portfolio We consider now the case when there is a margin agreement.Figure 2. EE as a function of the portfolio’s ratio µ/σ. the contributions of more volatile exposures are diminished (as the threshold caps the exposures). and contributions of higher mean exposures (inthemoney positions) increase. we characterize the impact of the threshold on the counterparty level EE. with the EE contribution converging to the mean value contributions themselves. but assume now that there is a collateral agreement in place where margins are placed instantaneously. 100% 80% 60% EE Contributions 40% P1 20% 0% 1 0. trades 4 and 5 end up dominating the contributions. EE Contributions as a function of the portfolio’s ratio µ/σ. and demonstrate the impact of the collateral on the trade contributions. trades 4 and 5. 28 . Figure 3 shows the EE contributions for each of the 5 trades as a function of µ/σ. As the portfolio’s µ/σ increases.5 2 2. 8.506. and for various levels of the (standardized) collateral threshold. At one side of the spectrum.5 20% 40% 60% 80% Mu/Sigm a 0 0.5 1 1. First.5 3 P2 P3 P4 P5 Figure 3. The opposite occurs for trades 1 and 2 (with low negative means and large volatilities). every trade contributes 20% of EE at µ/σ = 0. which have the largest (negative) mean values and lowest volatilities.
a normalized threshold of 2 does not reduce EE until the portfolio’s mean value is positive.0% 5 4 3 2 1 0 Mu/Sigma 1 2 3 4 5 Figure 4. The presence of the collateral affects each instrument’s contributions differently. the collateral thresholds become more effective at reducing EE as the portfolio is deeper inthemoney (i. EE contributions as a function of the collateral threshold. 40% Exposure Contributions 30% 20% P1 P2 P3 10% P4 P5 0% 0 10% H/Sigm a 1 2 3 4 5 Figure 5. it reduces EE by about 60%. In this case. EE contributions are basically the mean value contributions. the EE reductions decrease.e. Figure 5 plots the EE contributions for the case µ/σ = 1. H/σ.0% 80. H/σ > 4. a tighter threshold increases the percent contributions of trades P4 and P5 (which have the highest mean values) while reducing the contributions of P1 and P2 (the lowest mean values). as a function of the standardized threshold. Note finally that a trivial case arises when the ratio µ/σ =0. At high threshold values. when µ/σ increases). trade contributions are essentially the uncollateralized contributions. Conversely at low H/σ values.0% 0 % Exposure Reducitons 0. In particular. As the threshold is increased. Also.0% 60.0% 20.0% 0.5 1 2 4 6 40. the EE contributions are independent of the threshold level H/σ and equal the uncollateralized contributions. At a value of µ/σ = 5. as expected.Exposure reductions with collateral as a funciton of H/Sigma 100.2 0. 29 . For example. EE reductions from a collateral agreement. These eventually converge in the limit to the mean value contributions.
We further show how the calculations of conditional EE contributions can be incorporated into an existing exposure simulation process. “all EUR interest rate swaptions”.9. one implicitly assumes that exposures are independent of the counterparty credit quality. In addition. ed. The Professional Risk Manager’s Handbook (C. we adapt the continuous marginal contribution method which is often used for allocating economic capital. PRMIA Publications.prmia. 2001. Risk Books 30 . A. Risk Books D. and D. These allocations are additive. Capital Allocation and RAPM. pages 1341. 2004. To overcome this limitation. “Credit: The Complete Guide to Pricing. This is the case when there are no collateral or margin agreements.). so that one can aggregate the CVA allocations for any collection of trades with different counterparties. Wilmington. By using unconditional EEs in the CVA calculations.org). DE (www. In this article we have proposed a methodology for allocation of the counterpartylevel CVA to individual trades. Such a class can be defined as “all trades booked by a certain business unit”. Masetti. we show that the calculation of CVA allocations can be reduced to the calculation of contributions of individual trades to the counterpartylevel expected exposure (EE) conditional on the counterparty’s default. Rosen. Thus. Hedging and Risk Management”. We extend the methodology to deal with nonhomogeneous exposures of the type encountered when the portfolios have margin agreements. Arvanitis and J. we extend the results for conditional EE contributions in the normal approximation. Risk Neutral Pricing of Counterparty Credit Risk in “Counterparty Credit Risk Modelling” (M. Pykhtin. Gregory. The method is directly applicable for CVA contributions only when the counterpartylevel exposure is a homogeneous function of the trades’ weights in the portfolio. To obtain conditional EE contributions. In this paper. 2005. The price of the counterparty risk for the entire portfolio of trades with a counterparty is known as credit valuation adjustment (CVA). the contribution of all trades belonging to a certain class to the banklevel CVA can be calculated. we derive closed form expressions for unconditional EE contributions under the assumption that trade values are normally distributed. Brigo and M. etc. the ability to make quick calculations of CVA allocations outside of the exposure simulation system may be also desirable. References Aziz A. Alexander and E. Conclusions Counterparty credit risk is usually measured and priced at the counterparty level. To facilitate such calculations. which incorporate dependence between the trade values and the counterparty’s credit quality. Sheedy editors).
and Saunders D. 31 . eds.. GARP Risk Review. ed.. 2003. Rosen D. Risk. 15 of Handbooks in Operations Research and Management Science. Schönbucher (2003). Redon. Calculating and Hedging Exposure. Risk Books D. ed. Pykhtin and S. Wiley Finance. Mausser and D. 5(4). “Credit Derivatives Pricing Models”.E. Credit Value Adjustment and Economic Capital for Counterparty Credit Risk in “Counterparty Credit Risk Modelling” (M. 681726. Gibson. 2009. Journal of Credit Risk. Fields Institute and University of Waterloo M. Canabarro and D. Tilman. J.. 2005. pages 1622. E. Measuring and Marking Counterparty Risk in “Asset/Liability Management for Financial Institutions” (L. July/August. 2007. 2005. April. NorthHolland. Li. Duffie. Picoult. 2006. C. Modeling Credit Exposures for Collateralized Counterparties. Pykhtin and S. Measuring Counterparty Credit Risk for Trading Products under Basel II in “Basel Handbook”. 2009. Journal of Fixed Income 9.). Wrong Way Risk Modelling. 2000. pages 9095. Ong.).). Measuring Counterparty Credit Exposure to a Margined Counterparty in “Counterparty Credit Risk Modelling” (M. in Handbook of Financial Engineering. Risk Books M. Institutional Investor Books Garcia Cespedes J. M.. 2nd Edition (M. Linetsky. vol. Pykhtin. Working Paper. Zhu. A Guide to Modeling Counterparty Credit Risk. 2007. ed. Effective modelling of CCR Capital and Alpha for Derivatives Portfolios. ed. pages 4354. Rosen. Economic credit capital allocation and risk contributions. de Juan Herrero J. Risk Books M. Pykhtin. A. 2007.. On Default Correlation: a Copula Approach.). Birge and V. H. P. Pykhtin. pp. C. 2006. Zhu.
1 Exposure Independent of Counterparty’s Credit Quality We derive now Equation (52) from Equation (51). A2. we can write E 1 t <τ ≤t + dt = Pr [ t < τ ≤ t + dt ] = P′(t ) dt { } where P′(t ) is the first derivative of the cumulative probability of default P(t ) . which we restate here for convenience: µ (t ) + σ i (t ) X i 1 µ ( t ) +σ ( t ) X > H ei (t ) = E ( µi (t ) + σ i (t ) X i ) 1 0< µ (t ) +σ (t ) X < H + H E i { } } µ (t ) + σ (t ) X { 32 . we obtain t 1 E W τ = t = P′(t ) E W λ (t )exp[− ∫ λ ( s )ds ] 0 Appendix 2. Substituting both expressions in Equation (71). We can express this expectation as E W 1 t <τ ≤ t + dt { } E W τ = t = E 1 t <τ ≤t + dt { } (71) Then.Appendix 1. Analytical Results under Normal Approximation. Derivation of Equation (42) Suppose we have a random variable W and we want to calculate its expectation conditional on the counterparty defaulting at time t ( τ = t ). for the numerator of Equation (55) we can write t t + dt W 1{t <τ ≤t + dt} = E W exp[− λ ( s )ds ] − exp[− λ ( s )ds ] E ∫ ∫ 0 0 t = E W exp[− ∫ λ ( s )ds ] (1 − exp[−λ (t )dt ]) 0 t = E W exp[− ∫ λ ( s )ds ] λ (t )dt 0 For the denominator.
we show that we can use Equations (51)(55). Specifically. with the only difference that instead of using the unconditional expectations. From conditional expectation to conditional random variables The conditional expectation of a random variable can always be formulated as the unconditional expectation of a conditional random variable.2 EE Contributions under Right/Wrongway Risk We present the derivation of the analytical EE contributions when exposures are correlated with the counterparty credit quality. we now use the conditional ones. This results in Equation (42): µ (t ) µ (t ) µ (t ) − H µ (t ) − H ei (t ) = µi (t ) Φ − Φ + σ i (t ) ρi (t ) φ −φ σ (t ) σ (t ) σ (t ) σ (t ) ∞ +H H− ∫µ µi (t ) + σ i (t ) ρi (t ) x φ ( x)dx µ (t ) + σ (t ) x (t ) σ (t ) A2. there is no closed form solution for the second one. the counterpartylevel conditional EE in Equation (59) can be represented as ˆ ˆ e(t ) = E E (t ) while the conditional EE contribution in Equation (60) can be written as (72) 33 . standard deviations and correlations that specify the behavior of the trade values. For example.Substituting Equation (49) and inserting the expectation conditional on X inside each of the unconditional expectations. we obtain ei (t ) = E 1 0< µ (t )+σ (t ) X < H E µi (t ) + σ i (t ) ρi (t ) X + 1 − ρi2 (t ) Z i X { } ( ) µ (t ) + σ (t ) ρ (t ) X + 1 − ρ 2 (t ) Z i i i i i X + H E 1 µ (t ) +σ (t ) X > H E { } µ (t ) + σ (t ) X µ (t ) + σ i (t ) ρi (t ) X 1 µ (t )+σ (t ) X > H = E [ µi (t ) + σ i (t ) ρi (t ) X ] 1 0< µ (t )+σ (t ) X < H + H E i { } { } µ (t ) + σ (t ) X H − µ (t ) σ (t ) = − ∫ µ ∞ [ µi (t ) + σ i (t ) ρi (t ) x ]φ ( x)dx + H (t ) H− ∫µ µi (t ) + σ i (t ) ρi (t ) x φ ( x)dx µ (t ) + σ (t ) x (t ) σ (t ) σ (t ) While evaluating of the first integral is straightforward.
These conditional quantities are determined from the conditional trade values in exactly the same manner as the unconditional exposure and exposure contributions are determined from the unconditional trade values. By substituting Equation (61) into ˆ Equation (45) and setting Y = Φ −1[ P(t )] . Conditional trade values are calculated as follows. Let us denote the correlation between Vi (t ) and V (t ) by ρi (t ) . let us use Equations (61) and (62) to calculate the covariance between X i and X: 34 . To obtain this correlation. we need the correlation between the conditional trade value and the conditional ˆ ˆ ˆ portfolio value.ˆ ˆ ei (t ) = E Ei (t ) (73) ˆ ˆ where E (t ) is the conditional counterpartylevel exposure and Ei (t ) is the conditional stochastic exposure contribution of trade i. we obtain the value Vi (t ) of trade i conditional on the counterparty defaulting at time t: ˆ ˆ ˆ ˆ Vi (t ) = µi (t ) + σ i (t ) X i ˆ where µi (t ) is the conditional expectation of Vi (t ) given by ˆ µi (t ) ≡ E[ Vi (t )  τ = t ] = µi (t ) + σ i (t )bi Φ −1[ P(t )] ˆ and σ i (t ) is the conditional standard deviation of Vi (t ) given by ˆ σ i (t ) ≡ StDev[ Vi (t )  τ = t ] = σ i (t ) 1 − bi2 (76) (75) (74) ˆ The conditional portfolio value V (t ) is obtained by substituting Equation (62) into Equation (46) and setting Y = Φ −1[ P(t )] : ˆ ˆ ˆ ˆ V (t ) = µ (t ) + σ (t ) X ˆ where µ (t ) is the conditional expectation of V (t ) given by (77) ˆ µ (t ) ≡ E[ V (t )  τ = t ] = µ (t ) + σ (t ) β (t ) Φ −1[ P(t )] ˆ and σ (t ) is the conditional standard deviation of V (t ) given by ˆ σ (t ) ≡ StDev[ V (t )  τ = t ] = σ (t ) 1 − β 2(t ) (78) (79) Finally.
as a consequence of ˆ Equation (63). X ) = bi β (t ) + 1 − bi2 1 − β 2(t ) ρi (t ) ˆ ˆ where we have taken into account that both X i and X are independent of Y. conditional on the counterparty’s default at time t. X ) is nothing but the unconditional correlation ρi (t ) given by Equation (48) . Then. β (t ) = 0 . as one would expect. we obtain ˆ ρi (t ) = ρi (t ) − bi β (t ) (1 − bi2 )[1 − β 2(t )] (80) ˆ Note that ρi (t ) can also be interpreted as the correlation between the value Vi (t ) of trade i at time t and the portfolio value V (t ) at time t. and the conditional correlation becomes the unconditional one. Using this correlation. Equation (68) reduces to ρi (t ) = ρi (t ) . we can ˆ express X i as ˆ ˆ ˆ ˆ ˆ X i = ρi (t ) X + 1 − ρi2 (t ) Z i (81) ˆ ˆ where Z i is a standard normal random variable independent of X . all bi = 0 and.ˆ cov( X i . Realizing that cov( X i .1 after putting “hats” on the parameters! 35 . When no right/wrongway risk is present in the portfolio. Now we have formulated the conditional exposure model in exactly the same mathematical terms as the unconditional model of Section 7 and can use all the results of Subsection 7. Equation (81) is the conditional version of Equation (49).
20551 FAX: (202) 8724927 diane. 20551 michael.Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board.gov/pubs/feds Single hard copies of FEDS papers may be obtained by request from: Diane Linder Finance and Economics Discussion Series Mailstop 80 Federal Reserve Board Washington.C.gov . D. D.gov Comments about the FEDS working paper series may be addressed to: Michael Palumbo Finance and Economics Discussion Series Mailstop 61 Federal Reserve Board Washington.palumbo@frb. Washington.C. Starting with working papers in 1996. electronic copies of FEDS papers may be downloaded at the Internet site: http://www.linder@frb.c.federalreserve.g. D.C.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.