This action might not be possible to undo. Are you sure you want to continue?
1
CVA AND WRONG WAY RISK
John Hull and Alan White*
Joseph L. Rotman School of Management
University of Toronto
First Draft: June 14, 2011
This Draft:August 1, 2011
ABSTRACT
This paper proposes a simple model for incorporating wrongway and rightway risk into CVA
(credit value adjustment) calculations. These are the calculations made by a dealer to determine
the reduction in the value of its derivatives portfolio because of the possibility of a counterparty
default. The model relates the hazard rate of the counterparty to the value of the transactions
outstanding between the dealer and the counterparty. Numerical results for portfolios of 25
instruments dependent on five underlying market variables are presented. The paper finds that
wrongway and rightway risk have a significant effect on the Greek letters of CVA as well as on
CVA itself. It also finds that the nature of the effect depends on the collateral arrangements.
*We are grateful to Jon Gregory, Dan Rosen, Alexander Sokol, and Roger Stein for helpful
comments on an earlier version of this paper.
2
CVA AND WRONG WAY RISK
1. Introduction
It has for some time been standard practice for derivatives dealers to adjust the reported value of
their derivatives transactions with a counterparty to reflect the possibility of losses being
incurred because of a default by the counterparty. The adjustment is known as the credit value
adjustment or CVA.
A derivatives dealer has one CVA for each counterparty. These CVAs are themselves derivatives
and must be managed similarly to other derivatives. CVAs are particularly complex derivatives.
In fact, the CVA for a counterparty is more complex, and more difficult to value, than any of the
transactions between the dealer and the counterparty. This is because the CVA for a counterparty
is, as we will see, contingent on the net value of the portfolio of derivatives outstanding with that
counterparty.
Calculating CVAs is very computationally intensive. Gregory (2009) provides an excellent
discussion of the issues. Statistics published about Lehman Brothers give a sense of the scale of
the problem. Reuters (2008) reported that, at the time of its failure Lehman, had about 1.5
million derivatives transactions outstanding with 8,000 different counterparties. This means that
it had to calculate 8,000 different CVAs and the number of derivative transactions on which each
CVA was dependent averaged about 200.
Another measure, more controversial than CVA, is debit value adjustment or DVA.
1
The DVA
calculated by the dealer for a counterparty is an estimate of the costs to the counterparty of the
possibility that the dealer might default. The possibility that it might default is in theory a benefit
to the dealer and some accounting standards require the book value of the derivatives outstanding
with a counterparty to be calculated as their nodefault value minus the CVA plus the DVA.
2
The
1
DVA is sometimes also called debt value adjustment.
2
There is an interdependence here in that if the dealer defaults subsequent defaults by the counterparty are irrelevant
and vice versa. See Hull and White (2001) for a discussion of this point in the context of credit default swaps.
3
reason why DVA is controversial is that there is no way a dealer can monetize DVA without
actually defaulting.
It is interesting to note that when the credit spread of a derivatives dealer increases, DVA
increases. If the dealer uses DVA accounting, this in turn leads to an increase in the reported
value of the derivatives on the books of the dealer and a corresponding increase in its profits.
Some banks reported several billion dollars of profits from this source in 2008. Not surprisingly,
DVA gains and losses have now been excluded from the definition of common equity in
determining regulatory capital. In this paper, we will focus on CVA, but many of the points we
make are equally applicable to DVA.
Market variables that affect the nodefault value of a dealer’s outstanding transactions with a
counterparty also affect the dealer’s CVA for that counterparty. We will refer to these market
variables as the “underlying market variables.” In addition, CVA is affected by the
counterparty’s term structure of credit spreads (the “counterparty credit spreads”). CVA
therefore gives rise to two types of exposures. One arises from potential movements in the
underlying market variables; the other from potential movements in counterparty credit spreads.
In December 2010, the Basel Committee on Banking Supervision published a new regulatory
framework for banks known as Basel III.
3
It requires a dealer’s CVA risk arising from changes
in a counterparty’s credit spreads to be identified and included in the calculation of capital for
market risk. However, the dealer’s CVA risk arising from underlying market variables are not
included in this calculation. Some dealers have developed their own sophisticated systems for
managing both types of risk. These dealers feel that Basel III proposals are inadequate because,
if a dealer hedges against the underlying market variables, the hedging trades will lead to an
increase, not a decrease, in its required capital.
4
Since the crisis, governments have moved to require “standardized” overthe counter derivatives
to be cleared through central clearing parties. This paper focuses on how counterparty credit risk
3
See Basel Committee on Banking Supervision (2010)
4
See, for example, Pengelly (2011)
4
is handled for those transactions that continue to be cleared bilaterally.
5
This includes all
derivatives transactions classified as “nonstandard.” It also includes some categories of
derivatives, such as forward foreign exchange contracts, that are excluded from the central
clearing legislation.
Transactions between a dealer and a counterparty are typically governed by an International
Swaps and Derivatives Association (ISDA) Master Agreement. This specifies that all
transactions between the two parties are to be netted and considered as a single transaction in the
event that there is an early termination. The circumstances under which one side can send an
early termination notice to the other side and the procedures that are then used are specified in
the ISDA Master Agreement.
Collateralization has become an important feature of overthecounter derivatives markets. An
ISDA Master Agreement typically has a Credit Support Annex (CSA) which specifies the rules
governing the collateral that has to be posted by the two sides. In particular, it specifies a variety
of terms including the threshold, the independent amount, the minimum transfer amount, haircuts
that will apply to assets that are posted as collateral, etc. Suppose that the two sides are Party A
and Party B and Party B is required to post collateral. The threshold is the unsecured credit
exposure to Party B that Party A is willing to bear. If the value of the derivatives portfolio to
Party A is less than the threshold, no collateral is required from Party B. If the value of the
derivatives portfolio to Party A is greater than the threshold, the required collateral is equal to the
difference between the value and the threshold. The independent amount plays the same role as
the initial margin in a futures contract and can be regarded as a negative threshold. Failure to
post the required collateral by Party B would allow Party A to terminate its outstanding
transactions with Party B after what is known as a “cure period” has elapsed.
In the calculation of CVA, it is usually assumed that the counterparty’s probability of default is
independent of the dealer’s exposure to the counterparty. A situation where there is a positive
dependence between the two, so that the probability of default by the counterparty tends to be
high (low) when the dealer’s exposure to the counterparty is high (low), is referred to as “wrong
5
How derivatives dealers will assess their credit exposure arising from transactions that are cleared through a CCP
remains to be seen. The dealer is exposed to a) a default by any of the other clearing house members and b) a default
by the clearing house itself.
5
way risk.” A situation where there is negative dependence, so that the probability of default by
the counterparty tends to be high (low) when the dealer’s exposure to the counterparty is low
(high) is referred to “rightway risk.”
A subjective judgment of the amount of wrongway or rightway risk in transactions with a
counterparty requires a good knowledge of the counterparty’s business, in particular the nature
of the risks facing the business. It also requires knowledge of the transactions the counterparty
has entered into with other dealers.
6
The latter is difficult to know precisely but the extra
transparency provided by postcrisis legislation may help.
One situation in which wrongway risk tends to occur is when a company is selling credit
protection to the dealer. (AIG and monolines are obvious examples here.) This is because credit
spreads are correlated. When credit spreads are high the value of the protection to the dealer is
high and as a result the dealer has a large exposure to the company. At the same time the credit
spreads of the company are also likely to be high indicating a relatively high probability of
default for the company. Similarly, rightway risk tends to occur when a company is buying
credit protection from the dealer.
A situation in which a company is speculating by entering into many similar trades with one or
more dealers is likely to lead to wrongway risk for the dealers. This is because the company’s
financial position and therefore its probability of default is likely to be affected adversely if the
trades move against the company. If the company enters into transactions to partially hedge an
existing exposure there should in theory be rightway risk. This is because, when the transactions
move against the counterparty, it will be benefitting from the unhedged portion of its exposure so
that its probability of default will tend to be relatively low.
7
This paper explains the way CVA is calculated and the advanced approach for calculating CVA
risk capital under Basel III. The paper then proposes a model for incorporating wrongway risk
into CVA calculations. Other authors who have attempted to tackle the wrongway risk problem
6
A complication is that sometimes a company will trade with many different market participants in an attempt to
conceal its true exposure.
7
An exception could be when the counterparty is liable to run into liquidity problems. Although the assets being
hedged have increased in value, the counterparty might be unable to post collateral when required. An example here
is Ashanti Goldfields in September 1999. See for example Hull (2010, page 395) for a description of what happened.
6
include Cespedes et al (2010) and Sokol (2010). These authors propose an “exposure sampling”
approach where the process followed by the exposure is approximated with a onefactor Markov
process and a Gaussian (or other) copula is used to model the dependence between this process
and the time to default.
Our model is simpler than the exposure sampling approach and involves a relatively small
adjustment to the method used to calculate CVA when the usual assumption of no dependence
between exposure and probability of default is made. We specify a deterministic model relating
the hazard rate of the counterparty to the value of the dealer’s transactions with the counterparty.
Numerical results from the implementation of the model are presented.
2. Calculating CVA
Suppose that T is the longestmaturity derivative outstanding between a derivatives dealer and
one of its counterparties. Define v(t) as the value of a derivative that pays off the dealer’s net
exposure to the counterparty at time t and R as the recovery rate (assumed to be constant). If we
assume that the probability of a counterparty default at time t is independent of v(t) then, as
discussed by, for example, Hull and White (1995), Canabarro and Duffie (2003), and Picault
(2005),
}
=
÷ =
T
t
dt t v t q R
0
) ( ) ( ) 1 ( CVA (1)
where q(t) is the probability density function of the riskneutral time to default for the
counterparty.
If no collateral is posted the net exposure, E
NC
(t), at time t is given by
( ) ( ) ( )
max , 0
NC
E t w t = (2)
where w(t) is the value to the dealer of the derivatives portfolio at time t.
8
If transactions are
collateralized with zero threshold and an early termination is declared as soon as the counterparty
8
We assume that collateral is posted continuously and that there is no minimum transfer amount. Results can be
modified to relax these assumptions.
7
fails to post the required collateral, then in the event of default at time t the available collateral is
max(w(t), 0). If there is a threshold, K, the available collateral is max(w(t)−K, 0).
In practice, a period of time elapses between the time when the counterparty ceases to post
collateral and an early termination event being declared by the dealer. This is to allow any
disputes about the value of the portfolio between the dealer and the counterparty to be resolved
and is referred to the “cure period.” Suppose that the length of the cure period is c. The collateral
available if there is termination event at time t is
( ) ( ) ( )
max , 0 C t w t c K = ÷ ÷ (3)
An independent amount, I, can be treated as a negative threshold so that
( ) 0 , ) ( max ) ( I c t w t C + ÷ =
The net exposure, E
NET
(t), at time t is
( ) ( ) ( ) ( )
max , 0
NET NC
E t E t C t = ÷ (4)
The variable v(t) is the value of a derivative that pays off E
NET
(t) at time t. It can be calculated as
the expected value of E
NET
(t) in a riskneutral world discounted at the riskfree rate.
To approximate the integral in equation (1) we can choose times t
i
(0 ≤ i ≤ n) with t
0
= 0, t
n
= T
and t
0
< t
1
< t
2
< …< t
n
and set
¿
=
÷ =
n
i
i i
v q R
1
) 1 ( CVA (5)
where q
i
is the probability of default between times t
i−1
and t
i
and
( )
*
i i
v v t = with
( )
*
1
0.5
i i i
t t t
÷
= + . Note that q
i
is the unconditional riskneutral probability of default between
8
times t
i−1
and t
i
(as seen at time zero). It is not the probability of default conditional on no earlier
default.
9
The q
i
’s are usually calculated from credit spreads. Sometimes a complete term structure of
credit spreads for the counterparty can be observed in the market; sometimes it has to be
estimated using credit spread data for other companies. If s
i
is the credit spread for a maturity of
t
i
, an estimate of the average riskneutral hazard rate between times 0 and t
i
is
approximately ( ) 1
i
s R ÷ so that the probability of no default between times 0 and t
i
is
( ) exp 1
i i
s t R ( ÷ ÷
¸ ¸
. It follows that

.

\

÷
÷ ÷

.

\

÷
÷ =
÷ ÷
R
t s
R
t s
q
i i i i
i
1
exp
1
exp
1 1
(6)
Using a delta/gamma approximation, the impact on CVA of a small change As in all the s
i
’s is
( )
( )
( )
1 1
1
1
2
2 2 1 1
1
1
CVA exp exp
1 1
1
exp exp
2 1 1 1
n
i i i i
i i i
i
n
i i i i
i i i
i
s t s t
t t v s
R R
s t s t
t t v s
R R R
÷ ÷
÷
=
÷ ÷
÷
=
(
   
A = ÷ ÷ ÷ A
  (
÷ ÷
\ . \ .
¸ ¸
(    
+ ÷ ÷ ÷ A
  (
÷ ÷ ÷
\ . \ .
¸ ¸
¿
¿
(7)
This equation enables the dependence of CVA on counterparty credit spreads to be included in
the bank’s model for calculating market risk capital. Equations (5), (6), and (7) correspond to the
equations used in the Basel III advanced approach for determining capital for CVA risk.
10
DVA can be handled similarly. If R is the recovery rate of the dealer, v(t) is the value of a
derivative that pays off the counterparty’s exposure to the dealer at time t, and q(t) is the
probability density function of the dealer’s time to default, then the right hand side of equation
(1) gives the DVA. The counterparty’s net exposure to the dealer, after taking collateral posted
by the dealer into account, can be calculated in an analogous way to that indicated above and the
9
A confusion concerning the use of conditional and unconditional default probabilities in calculating CVA was
pointed out by Rebonato et al (2010).
10
Charging capital for the CVA risk associated with credit spreads is likely to encourage dealers to hedge spread
risk by buying credit default swaps. This might have the unintended consequence of increasing credit spreads.
9
approximations indicated in equations (5) and (6) can be used with q
i
defined as the dealer’s risk
neutral probability of default between times t
i−1
and t
i
, v
i
defined as the net exposure of the
counterparty to the dealer at time
*
i
t , and s
i
defined as the dealer’s credit spread for a maturity t
i
.
3. Monte Carlo Simulation
In practice, the v
i
’s are almost invariably calculated using Monte Carlo simulation. The market
variables affecting the nodefault value of a dealer’s derivatives with a counterparty are
simulated between times 0 and T in a riskneutral world.
11
One approach is to arrange the
simulation so that the value of the dealer’s portfolio with the counterparty is calculated at times
*
i
t c ÷ and
*
(1 )
i
t i n s s . (This means, On each simulation trial at each time ) 1 (
*
n i c t
i
s s ÷ , the
collateral is determined using equation (3). Then at time
*
i
t the value of the dealer’s portfolio
with the counterparty is determined and the net exposure of the dealer to the counterparty is
calculated using equation (4). The variable v
i
is estimated as the present value of the average of
the calculated net exposures at time ) 1 (
*
n i t
i
s s .
12
The variable q
i
is calculated using equation
(6).
In practice, some approximations are usually made so that computations are feasible. The model
used to value the portfolio during the Monte Carlo simulation may be simpler than either
a) the model used to simulate the underlying market variables, or
b) the model used by the dealer for marking to market its book.
Also, the riskneutral measure for the v
i
’s may be different from that used for the q
i
’s.
The Monte Carlo simulation is computationally quite time consuming. Some dealers calculate
CVA once a day, others less frequently. It is normal for the values of the market variables used
in the simulation and the portfolio values to be stored. If a new transaction is contemplated with
a counterparty, this stored data can then be used to quickly calculate the value of this transaction
at each valuation time for each simulation trial. This enables the portfolio value at each valuation
11
When interest rates are stochastic a convenient numeraire is the value of a riskfree zero coupon bond maturing at
the next valuation time. The Monte Carlo simulation can then be implemented by assuming that a) the returns on
market variables between valuation dates and b) the discount rates between valuation dates equal the yield on the
numeraire bond.
12
Only a small amount of additional computation time is necessary to calculate DVA at the same time as CVA.
10
time for each simulation trial to be updated relatively quickly to reflect the impact of the new
transaction.
13
Dealers also often estimate “peak exposures” which are high percentiles (e.g., 95%) of the
exposures at times ) 1 (
*
n i t
i
s s . The “maximum peak exposure” is the maximum of the
exposures calculated at these times. An interesting point (usually ignored in practice) is that
these estimates should in theory be made using the realworld measure to calculate exposures,
not the riskneutral measure.
4. CVA Exposures
Calculating the sensitivity of CVA to a small parallel shift in a counterparty’s credit spread is
straightforward using equation (7).
Calculating the first and second partial derivatives of CVA with respect to the underlying market
variables is liable to be more time consuming. Consider a market variable x with initial value x
0
.
It is necessary to calculate the effect on the paths sampled of changing x
0
to x
0
+c and x
0
−c for a
small c when all random number streams are kept the same. When the variable follows geometric
Brownian motion this is not too difficult. A small percentage change at time zero leads to the
same small percentage change at all future times on all simulation trials. (This is true both when
the volatility is deterministic and when the volatility is stochastic.) For other variables such as
those following mean reverting processes, the impact of a change at time zero on the change at
future times is liable to depend on the path followed by the market variable.
Suppose that
+
i
v and
÷
i
v are the values calculated for v
i
when the initial value of the market
variable is x
0
+c and x
0
–c, respectively. From equation (5)
) (
2
1 CVA
1
÷
=
+
÷
÷
=
c
c
¿ i
n
i
i i
v v q
R
x c
13
Note that the impact of a new transaction on CVA can be positive or negative.
11
¿
=
÷ +
÷ +
÷
=
c
c
n
i
i i i i
v v v q
R
x
1
2 2
2
) 2 (
1 CVA
c
These equations enable CVA risks relating to the underlying market variables to be assessed and
hedged. As already mentioned, under Basel III CVA exposures arising from the underlying
market variables are not included in the calculation of market risk capital.
5. A Model for WrongWay and RightWay Risk
The situation where q
is positively dependent on v is referred to as “wrongway” risk. In this
case, there is a tendency for a counterparty to default when the dealer’s exposure is relatively
high. The situation where
q is negatively dependent on v is referred to as “rightway” risk. In this
case, there is a tendency for a counterparty to default when the dealer’s exposure is relatively
low.
A simple way of dealing with wrongway risk is to use what is termed the “alpha” multiplier to
increase v(t) in the version of the model in which v(t) and q(t) are assumed to be independent.
The effect of this is to increase CVA by the alpha multiplier. The Basel II rules set alpha equal to
1.4 or allows banks to use their own models, with a floor for alpha of 1.2. This means that, at
minimum, the CVA has to be 20% higher than that given by the model where q(t) and v(t) are
independent. If a bank does not have its own model for wrong way risk it has to be 40% higher.
Estimates of alpha reported by banks range from 1.07 to 1.10.
The usual approach to modeling wrongway risk is to reflect it in the way the v’s are calculated.
The alpha multiplier approach just mentioned is one example of this approach. Another
sometimes used is to set v(t) equal to the present value of the exposure that is k standard
deviations above the average exposure for some k.
We will use a different approach. Instead of changing the calculation of v(t) we change the
calculation of q(t) so that q(t) depends on the path followed the value of the portfolio up until
time t.
12
Define h(t) as the hazard rate of the counterparty at time t. As before we define w(t) as the value
to the dealer of its portfolio with the counterparty at time t. A simple model of wrongway/right
way is to define h(t) as a deterministic function of w(t). Two models we have used are:
( ) ( ) ( ) exp h t a t bw t ( = +
¸ ¸
(8)
and
( ) ( ) ( ) ( )
ln 1 exp h t a t bw t
(
= + +
¸ ¸
(9)
Each of these models has the property that h(t) is a monotonic function of w(t) with h(t) > 0. The
parameter b measures the sensitivity of h(t) to w(t). The function a(t), as will be explained in
more detail later, is determined using an iterative search procedure so that the average survival
probability, calculated across all simulations, up to any time T matches that calculated from
credit spreads. When b > 0, h(t) is an increasing function of w(t), which corresponds to wrong
way risk; when b < 0, h(t) is a decreasing function of w(t), which corresponds to rightway risk.
In the model given by equation (8), h(t) increases exponentially as w(t) takes more extreme
values.
14
In the model in equation (9) h(t) increases linearly as w(t) takes on more extreme
values.
15
For moderate values of w(t) the two models are very similar. In the results we present,
we use the model in equation (9), which is illustrated in Figure 1.
There are two approaches to determining the parameter b. One is to base it on historical data.
Data is collected for past dates on a) credit spreads for the counterparty and b) what the value of
the current portfolio with the counterparty would have been. The credit spreads can be converted
into hazard rates and the parameter b can be estimated from whichever of equations (8) and (9)
are used. The disadvantage of this is that the counterparty’s credit spread may have been heavily
influenced by other factors in the past.
14
When b > 0 the relevant extreme values are high; when b < 0 they are low.
15
Either model can be modified to incorporate a cap on h(t).
13
The other approach involves some subjective judgement and it best illustrated with an example.
Suppose that the current value to the dealer of the portfolio, w(0), is $3 million and the
counterparty’s fiveyear credit spread is 300 basis points. Assuming a recovery rate of 40%, this
means that the average fiveyear hazard rate is 5% per year. Also, suppose that it is estimated
that, if w(0) increased to $20 million, the spread would rise to 600 basis points, corresponding to
an average fiveyear hazard rate of 10%. Assuming the term structure of hazard rates is flat,
16
when w(0) = 3 million, h(0) = 5% and when w(0) = 20 million, h(0) = 10%. Solving a pair of
simultaneous equations for a(0) and b, we find that at time zero, for the model in equation (8), b
is 0.0408 per million dollars while, for the model in equation (9), b is 0.0423 per million dollars.
The implementation of the model requires only a small change to the procedure for calculating
CVA described in Section 3. The values of ) (
*
i
t a for 1 ≤ i ≤ n must be determined so that the
average survival probability, across all simulations, up to time
*
i
t equals the survival probability
calculated from the term structure of credit spreads. This means that we require
1 1
1
exp( ) exp for 1
1
m k
k k
ij
j i
s t
h t k n
m R
= =
(  
÷ A = ÷ s s
 (
÷
\ . ¸ ¸
¿ ¿
where h
ij
and w
ij
are the values of ) (
*
i
t h and ) (
*
i
t w on the jth simulation trial and m is the number
of simulation trials.
For the model in equation (8)
( ) ( )
*
exp
ij i ij
h a t bw = +
For the model in equation (9)
( ) ( )
*
ln 1 exp
ij i ij
h a t bw
(
= + +
¸ ¸
16
Given the imprecision of any attempt to quantify wrongway risk this is a reasonable assumption.
14
First, k is set equal to 1 and an iterative search is used to determine ) (
*
1
t a from the w
1j
. This
determines the h
1j
. Second, k is set equal to 2 and an iterative search is carried out to determine
) (
*
2
t a from the w
2j
values and the h
1j
. This determines the h
2j
; and so on.
6. Numerical Results
We first illustrate the model in equation (9) by assuming that the dealer has a simple portfolio
consisting of a single oneyear forward foreign exchange transaction.
17
We assume that the
principal is $100 million, the domestic and foreign risk free rates are 5%, the initial exchange
rate is 1.0, the delivery exchange rate specified in the forward contract is also 1.0, and the
volatility of the exchange rate is 15%. We suppose that the counterparty’s credit spread (all
maturities) is 125 basis points. Table 1 considers the case where b is 0.03 per million dollars and
the dealer’s position is long. Four different situations are considered: no collateral, a threshold of
$10 million with a cure period of 15 days, a threshold of zero with a cure period of 15 days, and
an independent amount of $5 million with a cure period of 15 days. Table 2 provides similar
results to Table 1 for the case where b is 0.03 per million dollars and the dealer is short; Table 3
does so for the case where b = −0.03 per million dollars and the dealer is long; Table 4 does so
for the case where b = −0.03 per million dollars and the dealer is short.
The results illustrate that wrongway and rightway risk have a material effect on the deltas and
gammas of CVA as well as on CVA itself. This is true for both the deltas and gammas with
respect to the exchange rate and the deltas and gammas with respect to the credit spread. The
magnitude of the effect is difficult to predict. Indeed, in some cases, even the direction of effect
can be difficult to predict. As shown by the tables, sometimes the direction of the effect depends
on the collateral arrangements.
In general, the impact of wrongway and rightway risk on CVA depends in a complex way on
CVA itself and the collateral arrangements. To illustrate this, we randomly generated 250
portfolios. Each portfolio consists of 25 options on one of five different assets. The asset prices
17
The results from using the model in equation (8) are similar.
15
are assumed to follow geometric Brownian motion with pairwise correlations of 0.36. Each
option has the following properties.
(a) It is equally likely to be long or short
(b) It is equally likely to be a call or a put
(c) The underlying is equally likely to any one of the five assets
(d) All maturities between one and five years are equally likely
(e) All strike prices within 30% of the current asset price are equally likely
(f) The underlying principal is $25 million.
The assets do not provide any income. They have an initial price of $25 and a volatility of 25%.
The riskfree rate is 5%, the credit spread of the counterparty (all maturities) is 125 basis points,
and the recovery rate is 40%.
Figure 2 provides a scatter plot of the relationship between the dollar change in CVA caused
increasing b from 0 to 0.01 per million dollars and the CVA for b=0 for the 250 portfolios when
there is no collateralization. It can be seen that the change in CVA tends to increase as CVA
increases. The reason is that as CVA increases there is a tendency for both the mean and standard
deviation of the value of the portfolio at future times to increase. The gap between the exposures
on highw(t) paths and loww(t) paths increases. Wrongway risk causes the hazard rate to
increase dramatically on the highw(t) paths and decrease (modestly) on the loww(t) paths.
Figure 3 provides a similar plot to Figure 2 for the situation where the threshold is $10 million
and the cure period is 15 days. The average relationship between the change in CVA and CVA is
in this case is one where the change first increases and then decreases. To understand the reason
for this, we ignore the cure period and focus on the threshold. The impact of the threshold is to
restrict the net exposure to less than $10 million. For low w(t)’s this restriction has little effect.
However, as w(t) increases, the net exposure is increasingly impacted by the $10 million
restriction. The gap between the exposure on highw(t) paths and loww(t) paths is much less
than in the no collateral case. The highest theoretical CVA is achieved when the value of w(t) is
16
certain to be above $10 million at all times. In this case, the net exposure is the same on all w(t)
paths and wrongway risk has no effect. This explains the pattern observed in Figure 3. The cure
period does have an effect on the results, but its effect is less than the effect of the threshold.
Figure 4 provides a similar plot to Figures 2 and 3 for the situation where the threshold is zero
and the cure period is 15 days. In this case the net exposure is entirely as a result of the cure
period. When w(t) is positive, the net exposure increases as the standard deviation of the change
in w(t) during the cure period increases. High standard deviations of w(t) tend to be associated
with high w(t)’s. High w(t)’s are in turn associated with increases in the hazard rate when there is
wrongway risk. This leads to the pattern shown in Figure 4.
Figure 5 provides a plot for the situation where there is an independent amount equal to $5
million and a cure period of 15 days. As in the case of Figure 4 the impact of the cure period on
the exposure at time t for a particular w(t) depends on the standard deviation of the change in
w(t) during the cure period. Indeed, for positive exposures to be generated, the standard deviation
has to be sufficiently high that there is a reasonable chance of a $5 million increase in w(t) during
the cure period. The reasons for the pattern observed are similar to those for Figure 4.
The impact of rightway risk can be examined by changing b from 0 to −0.01 instead of from 0
to +0.01. The results are similar to those shown in Figures 2 to 5, except that the change in CVA
is negative instead of positive. We have carried out other experiments involving portfolios of
interest rate swaps and portfolios. The results are similar to those in Figures 2 to 5.
7. Conclusions
We have proposed a simple model for handling wrongway/rightway risk in the calculation of
CVA. The model can be implemented by making a small change to usual method for calculating
CVA. It is necessary for the dealer to make a single estimate describing the sensitivity of the
counterparty’s credit spread to the value of its portfolio with the counterparty.
Tests of the model show that wrongway and rightway risk have a significant effect on the
Greek letters of CVA as well as on CVA itself. Because CVA is such a complex derivative, it is
difficult to estimate these effects without a model. Indeed, even the sign of the effect can be
counterintuitive.
17
Further tests involving the random generation of portfolios indicate that when there is no
collateral or when collateral is posted with zero threshold or when collateral is posted with an
independent amount, the dollar impact of wrongway risk on CVA tends to increase as CVA
increases. The situation where the threshold is materially positive is more interesting. As CVA
increases, the impact of wrong way risk tends to first increase and then decrease.
18
REFERENCES
Basel Committee on Banking Supervision (2010), “Basel III: A Global Regulatory Framework
for More Resilient Banks and Banking Systems,” http://www.bis.org/publ/bcbs189_dec2010.pdf,
December 2010.
Canabarro, Eduardo and Darrell Duffie (2003), “Measuring and Marking Counterparty Risk,”
Chapter 9 in Asset/Liability Management for Financial Institutions, Edited by Leo Tilman, New
York: Institutional Investor books.
Cepedes, Juan Carlos Garcia, Juan Antonio de Juan Herrero, Dan Rosen, and David Saunders
(2010), “Effective Modeling of Wrong Way Risk, Counterparty Credit Risk Capital and Alpha
in Basel II,” Journal of Risk Model Validation, 4, 1, pp 7198
Gregory, Jon (2009), “Counterparty Credit Risk: The New Challenge for Financial Markets”
John Wiley and Sons.
Hull, John (2010), “Risk Management and Financial Institutions,” Second Edition, Pearson.
Hull, John and Alan White (1995), "The Impact of Default Risk on the Prices of Options and
Other Derivative Securities,'' Journal of Banking and Finance, Vol 19, pp 299322.
Hull, John and Alan White (2001), “Valuing Credit Default Swaps II: Modeling Default
Correlations,” Journal of Derivatives, 8, 3 (Spring 2001), pp1222.
Pengelley, Mark (2011), “CVA Melee” Risk, 24, 2, pp 3739.
Picault, Evan (2005), “Calculating and Hedging Exposure, CVA, and Economic Capital for
Counterparty Credit Risk,” in Counterparty Credit Risk Modelling edited by Michael Pykhtin,
London: Risk Books.
Rebonato, Riccardo, Mike Sherring, and Ronnie Barnes (2010), “Credit Risk, CVA, and the
Equivalent Bond,” Risk, 23, 9, pp 118121.
Reuters (2008), Update 1Lehman Brothers Holdings is Focus of Grand Jury Probe,” October
16.
Sokol, Alexander (2010), “A Practical Guide to Monte Carlo CVA,” Chapter 14 in Lessons
From the Crisis edited by Arthur Berd, London: Risk Books
19
Table 1: Impact of wrongway risk on CVA for a long forward contract to buy 100 million units
of a foreign currency in one year. The current exchange rate is 1.0, the domestic and foreign risk
free interest rates are both 5%, and the volatility of the exchange rate is 15%. The credit spread is
125 basis points for all maturities and the recovery rate is 40%. K is the threshold in $ millions,
c is the cure period in days, and b is the parameter in equation (9).
No
Collateral
K = 10 K = 0 K = –5
c = 15 c = 15 c = 15
CVA ($ millions) for b = 0 0.048 0.036 0.011 0.002
Impact of b = 0.03 per $mm on:
CVA 54.8% 41.7% 37.3% 53.5%
Delta wrt Exch Rate 32.0% 15.6% 12.8% 39.3%
Gamma wrt Exch Rate 2.6% –25.4% 17.7% –0.7%
Delta wrt Spread 53.8% 41.2% 36.8% 52.8%
Gamma wrt Spread 181.8% 124.3% 122.8% 184.3%
Table 2: Impact of wrongway risk on CVA for a short forward contract to sell 100 million units
of a foreign currency in one year. The current exchange rate is 1.0, the domestic and foreign risk
free interest rates are both 5%, and the volatility of the exchange rate is 15% . The credit spread
is 125 basis points for all maturities and the recovery rate is 40%. K is the threshold in $
millions, c is the cure period in days, and b is the parameter in equation (9).
No
Collateral
K = 10 K = 0 K = –5
c = 15 c = 15 c = 15
CVA ($ millions) for b = 0 0.048 0.039 0.011 0.001
Impact of b = 0.03 per $mm on:
CVA 40.5% 34.0% 27.6% 28.9%
Delta wrt Exch Rate 16.2% 7.7% –1.9% –341.9%
Gamma wrt Exch Rate –7.0% –21.4% 16.4% 26.5%
Delta wrt Spread 40.0% 33.7% 27.4% 28.8%
Gamma wrt Spread 114.8% 91.0% 77.0% 70.7%
20
Table 3: Impact of rightway risk on CVA for a long forward contract to buy 100 million units
of a foreign currency in one year. The current exchange rate is 1.0, the domestic and foreign risk
free interest rates are both 5%, and the volatility of the exchange rate is 15%. The credit spread is
125 basis points for all maturities and the recovery rate is 40%. K is the threshold in $ millions,
c is the cure period in days, and b is the parameter in equation (9).
No
Collateral
K = 10 K = 0 K = –5
c = 15 c = 15 c = 15
CVA ($ millions) for b = 0 0.048 0.036 0.011 0.002
Impact of b = –0.03 per $mm on:
CVA –37.5% –32.7% –29.1% –35.7%
Delta wrt Exch Rate –26.7% –18.8% –14.8% –28.9%
Gamma wrt Exch Rate –8.2% 11.7% –16.0% 6.2%
Delta wrt Spread –37.2% –32.5% –28.9% –35.6%
Gamma wrt Spread –79.2% –74.5% –72.1% –77.3%
Table 4: Impact of rightway risk on CVA for a short forward contract to buy 100 million units
of a foreign currency in one year. The current exchange rate is 1.0, the domestic and foreign risk
free interest rates are both 5%, and the volatility of the exchange rate is 15%. The credit spread is
125 basis points for all maturities and the recovery rate is 40%. K is the threshold in $ millions,
c is the cure period in days, and b is the parameter in equation (9).
No
Collateral
K = 10 K = 0 K = –5
c = 15 c = 15 c = 15
CVA ($ millions) for b = 0 0.048 0.039 0.011 0.001
Impact of b = –0.03 per $mm on:
CVA –33.9% –30.8% –25.9% –26.9%
Delta wrt Exch Rate –19.3% –13.6% –4.9% 209.1%
Gamma wrt Exch Rate 0.9% 14.4% –16.7% –37.5%
Delta wrt Spread –33.6% –30.6% –25.7% –26.7%
Gamma wrt Spread –78.8% –75.5% –71.3% –69.0%
21
Figure 1: The model in equation (9) when (i) a(t)= −4 and b=0.01 and (ii) a(t)= −4 and b=−0.01
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
‐400 ‐300 ‐200 ‐100 0 100 200 300 400
h(t)
w(t)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
‐400 ‐300 ‐200 ‐100 0 100 200 300 400
h(t)
w(t)
(i) (ii)
22
Figure 2: Impact of wrongway risk for 250 portfolios of options when there is no collateral.
The horizontal axis shows CVA when b=0. The vertical axis shows the change in CVA when
b is increased from 0 to 0.01 per million.
0
0.2
0.4
0.6
0.8
1
1.2
0 0.5 1 1.5 2 2.5 3
Change in CVA
($ millions)
CVA (no wrong‐way‐risk, $ million
23
Figure 3: Impact of wrongway risk for 250 portfolios of options when the threshold is $10
million and the cure period is 15 days. The horizontal axis shows CVA when b=0. The vertical
axis shows the change in CVA when b is increased from 0 to 0.01 per million
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
0 0.1 0.2 0.3 0.4 0.5 0.6
Change in CVA
($ millions)
CVA (no wrong‐way risk, $ million)
24
Figure 4: Impact of wrongway risk for 250 portfolios of options when the threshold is zero and
the cure period is 15 days. The horizontal axis shows CVA when b=0. The vertical axis shows
the change in CVA when b is increased from 0 to 0.01.
0
0.01
0.02
0.03
0.04
0.05
0.06
0 0.05 0.1 0.15
Change in CVA
($ millions)
CVA (no wrong‐way risk, $ millions)
25
Figure 5: Impact of wrongway risk for 250 portfolios of when there is an independent amount
of $5 million and a cure period of 15 days. The horizontal axis shows CVA when b=0. The
vertical axis shows the change in CVA when b is increased from 0 to 0.01 per million
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
Change in CVA
($ millions)
CVA (no wrong‐way risk, $ millions)
CVA AND WRONG WAY RISK
1. Introduction It has for some time been standard practice for derivatives dealers to adjust the reported value of their derivatives transactions with a counterparty to reflect the possibility of losses being incurred because of a default by the counterparty. The adjustment is known as the credit value adjustment or CVA. A derivatives dealer has one CVA for each counterparty. These CVAs are themselves derivatives and must be managed similarly to other derivatives. CVAs are particularly complex derivatives. In fact, the CVA for a counterparty is more complex, and more difficult to value, than any of the transactions between the dealer and the counterparty. This is because the CVA for a counterparty is, as we will see, contingent on the net value of the portfolio of derivatives outstanding with that counterparty. Calculating CVAs is very computationally intensive. Gregory (2009) provides an excellent discussion of the issues. Statistics published about Lehman Brothers give a sense of the scale of the problem. Reuters (2008) reported that, at the time of its failure Lehman, had about 1.5 million derivatives transactions outstanding with 8,000 different counterparties. This means that it had to calculate 8,000 different CVAs and the number of derivative transactions on which each CVA was dependent averaged about 200. Another measure, more controversial than CVA, is debit value adjustment or DVA.1 The DVA calculated by the dealer for a counterparty is an estimate of the costs to the counterparty of the possibility that the dealer might default. The possibility that it might default is in theory a benefit to the dealer and some accounting standards require the book value of the derivatives outstanding with a counterparty to be calculated as their nodefault value minus the CVA plus the DVA.2 The
DVA is sometimes also called debt value adjustment. There is an interdependence here in that if the dealer defaults subsequent defaults by the counterparty are irrelevant and vice versa. See Hull and White (2001) for a discussion of this point in the context of credit default swaps.
2 1
2
the other from potential movements in counterparty credit spreads. Pengelly (2011) 3 . if a dealer hedges against the underlying market variables. Market variables that affect the nodefault value of a dealer’s outstanding transactions with a counterparty also affect the dealer’s CVA for that counterparty. This paper focuses on how counterparty credit risk 3 4 See Basel Committee on Banking Supervision (2010) See. but many of the points we make are equally applicable to DVA. DVA increases.reason why DVA is controversial is that there is no way a dealer can monetize DVA without actually defaulting. One arises from potential movements in the underlying market variables. It is interesting to note that when the credit spread of a derivatives dealer increases. CVA therefore gives rise to two types of exposures. this in turn leads to an increase in the reported value of the derivatives on the books of the dealer and a corresponding increase in its profits. However. not a decrease.3 It requires a dealer’s CVA risk arising from changes in a counterparty’s credit spreads to be identified and included in the calculation of capital for market risk. for example. We will refer to these market variables as the “underlying market variables. CVA is affected by the counterparty’s term structure of credit spreads (the “counterparty credit spreads”).4 Since the crisis. the hedging trades will lead to an increase. Not surprisingly. Some banks reported several billion dollars of profits from this source in 2008. the Basel Committee on Banking Supervision published a new regulatory framework for banks known as Basel III. governments have moved to require “standardized” overthe counter derivatives to be cleared through central clearing parties. These dealers feel that Basel III proposals are inadequate because. in its required capital. If the dealer uses DVA accounting. the dealer’s CVA risk arising from underlying market variables are not included in this calculation. we will focus on CVA. In this paper.” In addition. DVA gains and losses have now been excluded from the definition of common equity in determining regulatory capital. In December 2010. Some dealers have developed their own sophisticated systems for managing both types of risk.
If the value of the derivatives portfolio to Party A is less than the threshold. so that the probability of default by the counterparty tends to be high (low) when the dealer’s exposure to the counterparty is high (low). no collateral is required from Party B. In the calculation of CVA. such as forward foreign exchange contracts.5 This includes all derivatives transactions classified as “nonstandard. the independent amount. Collateralization has become an important feature of overthecounter derivatives markets. In particular. The circumstances under which one side can send an early termination notice to the other side and the procedures that are then used are specified in the ISDA Master Agreement. haircuts that will apply to assets that are posted as collateral. The dealer is exposed to a) a default by any of the other clearing house members and b) a default by the clearing house itself. The threshold is the unsecured credit exposure to Party B that Party A is willing to bear. the minimum transfer amount. it is usually assumed that the counterparty’s probability of default is independent of the dealer’s exposure to the counterparty. The independent amount plays the same role as the initial margin in a futures contract and can be regarded as a negative threshold. 5 4 . A situation where there is a positive dependence between the two. Transactions between a dealer and a counterparty are typically governed by an International Swaps and Derivatives Association (ISDA) Master Agreement.” It also includes some categories of derivatives. Suppose that the two sides are Party A and Party B and Party B is required to post collateral. An ISDA Master Agreement typically has a Credit Support Annex (CSA) which specifies the rules governing the collateral that has to be posted by the two sides. This specifies that all transactions between the two parties are to be netted and considered as a single transaction in the event that there is an early termination. Failure to post the required collateral by Party B would allow Party A to terminate its outstanding transactions with Party B after what is known as a “cure period” has elapsed.is handled for those transactions that continue to be cleared bilaterally. etc. that are excluded from the central clearing legislation. is referred to as “wrong How derivatives dealers will assess their credit exposure arising from transactions that are cleared through a CCP remains to be seen. If the value of the derivatives portfolio to Party A is greater than the threshold. it specifies a variety of terms including the threshold. the required collateral is equal to the difference between the value and the threshold.
7 An exception could be when the counterparty is liable to run into liquidity problems. 6 5 . If the company enters into transactions to partially hedge an existing exposure there should in theory be rightway risk. so that the probability of default by the counterparty tends to be high (low) when the dealer’s exposure to the counterparty is low (high) is referred to “rightway risk. When credit spreads are high the value of the protection to the dealer is high and as a result the dealer has a large exposure to the company. the counterparty might be unable to post collateral when required. This is because the company’s financial position and therefore its probability of default is likely to be affected adversely if the trades move against the company.way risk. 6 The latter is difficult to know precisely but the extra transparency provided by postcrisis legislation may help. Similarly.” A situation where there is negative dependence. This is because. when the transactions move against the counterparty.7 This paper explains the way CVA is calculated and the advanced approach for calculating CVA risk capital under Basel III. Other authors who have attempted to tackle the wrongway risk problem A complication is that sometimes a company will trade with many different market participants in an attempt to conceal its true exposure. A situation in which a company is speculating by entering into many similar trades with one or more dealers is likely to lead to wrongway risk for the dealers. Although the assets being hedged have increased in value. An example here is Ashanti Goldfields in September 1999. in particular the nature of the risks facing the business. rightway risk tends to occur when a company is buying credit protection from the dealer. (AIG and monolines are obvious examples here. One situation in which wrongway risk tends to occur is when a company is selling credit protection to the dealer.” A subjective judgment of the amount of wrongway or rightway risk in transactions with a counterparty requires a good knowledge of the counterparty’s business. it will be benefitting from the unhedged portion of its exposure so that its probability of default will tend to be relatively low.) This is because credit spreads are correlated. It also requires knowledge of the transactions the counterparty has entered into with other dealers. See for example Hull (2010. The paper then proposes a model for incorporating wrongway risk into CVA calculations. At the same time the credit spreads of the company are also likely to be high indicating a relatively high probability of default for the company. page 395) for a description of what happened.
Results can be modified to relax these assumptions. CVA (1 R) T t 0 q(t )v(t )dt (1) where q(t) is the probability density function of the riskneutral time to default for the counterparty. at time t is given by ENC t max w t . as discussed by. Canabarro and Duffie (2003). These authors propose an “exposure sampling” approach where the process followed by the exposure is approximated with a onefactor Markov process and a Gaussian (or other) copula is used to model the dependence between this process and the time to default. Hull and White (1995). Define v(t) as the value of a derivative that pays off the dealer’s net exposure to the counterparty at time t and R as the recovery rate (assumed to be constant). ENC(t). We specify a deterministic model relating the hazard rate of the counterparty to the value of the dealer’s transactions with the counterparty. Calculating CVA Suppose that T is the longestmaturity derivative outstanding between a derivatives dealer and one of its counterparties. for example. Numerical results from the implementation of the model are presented. 6 . Our model is simpler than the exposure sampling approach and involves a relatively small adjustment to the method used to calculate CVA when the usual assumption of no dependence between exposure and probability of default is made.include Cespedes et al (2010) and Sokol (2010). If no collateral is posted the net exposure.8 If transactions are collateralized with zero threshold and an early termination is declared as soon as the counterparty 8 (2) We assume that collateral is posted continuously and that there is no minimum transfer amount. 0 where w(t) is the value to the dealer of the derivatives portfolio at time t. and Picault (2005). 2. If we assume that the probability of a counterparty default at time t is independent of v(t) then.
K.5 ti 1 ti . 0). tn = T and t0 < t1 < t2 < …< tn and set CVA (1 R) qi vi i 1 n (5) where qi is the probability of default between times ti−1 and ti and vi v ti* with ti* 0. 0 An independent amount.fails to post the required collateral. It can be calculated as the expected value of ENET (t) in a riskneutral world discounted at the riskfree rate. 0 (4) The variable v(t) is the value of a derivative that pays off ENET (t) at time t. ENET(t). This is to allow any disputes about the value of the portfolio between the dealer and the counterparty to be resolved and is referred to the “cure period. at time t is ENET t max ENC t C t . 0 (3) The net exposure. 0). can be treated as a negative threshold so that C (t ) max w(t c) I . I. a period of time elapses between the time when the counterparty ceases to post collateral and an early termination event being declared by the dealer. To approximate the integral in equation (1) we can choose times ti (0 ≤ i ≤ n) with t0 = 0. If there is a threshold. then in the event of default at time t the available collateral is max(w(t). the available collateral is max(w(t)−K.” Suppose that the length of the cure period is c. Note that qi is the unconditional riskneutral probability of default between 7 . The collateral available if there is termination event at time t is C t max w t c K . In practice.
This might have the unintended consequence of increasing credit spreads.9 The qi’s are usually calculated from credit spreads. can be calculated in an analogous way to that indicated above and the A confusion concerning the use of conditional and unconditional default probabilities in calculating CVA was pointed out by Rebonato et al (2010). 10 Charging capital for the CVA risk associated with credit spreads is likely to encourage dealers to hedge spread risk by buying credit default swaps. The counterparty’s net exposure to the dealer. If si is the credit spread for a maturity of ti. If R is the recovery rate of the dealer. and (7) correspond to the equations used in the Basel III advanced approach for determining capital for CVA risk. Equations (5). and q(t) is the probability density function of the dealer’s time to default. Sometimes a complete term structure of credit spreads for the counterparty can be observed in the market.10 DVA can be handled similarly. 9 8 .times ti−1 and ti (as seen at time zero). It is not the probability of default conditional on no earlier default. sometimes it has to be estimated using credit spread data for other companies. the impact on CVA of a small change s in all the si’s is n st s t CVA ti exp i i ti 1 exp i 1 i 1 vi s 1 R 1 R i 1 n 1 2 s 1t s t ti21 exp 1i iR1 ti2 exp 1 i iR vi s 2 1 R i 1 (6) (7) This equation enables the dependence of CVA on counterparty credit spreads to be included in the bank’s model for calculating market risk capital. (6). after taking collateral posted by the dealer into account. then the right hand side of equation (1) gives the DVA. v(t) is the value of a derivative that pays off the counterparty’s exposure to the dealer at time t. It follows that s t st qi exp i 1 i 1 exp i i 1 R 1 R Using a delta/gamma approximation. an estimate of the average riskneutral hazard rate between times 0 and ti is approximately si 1 R so that the probability of no default between times 0 and ti is exp si ti 1 R .
(This means. the riskneutral measure for the vi’s may be different from that used for the qi’s. some approximations are usually made so that computations are feasible. others less frequently. 9 . 3. On each simulation trial at each time t * c (1 i n) . The model used to value the portfolio during the Monte Carlo simulation may be simpler than either a) the model used to simulate the underlying market variables. Also. Then at time ti* the value of the dealer’s portfolio with the counterparty is determined and the net exposure of the dealer to the counterparty is calculated using equation (4).12 The variable qi is calculated using equation (6). the vi’s are almost invariably calculated using Monte Carlo simulation. This enables the portfolio value at each valuation 11 When interest rates are stochastic a convenient numeraire is the value of a riskfree zero coupon bond maturing at the next valuation time. Monte Carlo Simulation In practice.approximations indicated in equations (5) and (6) can be used with qi defined as the dealer’s riskneutral probability of default between times ti−1 and ti. It is normal for the values of the market variables used in the simulation and the portfolio values to be stored. this stored data can then be used to quickly calculate the value of this transaction at each valuation time for each simulation trial. or b) the model used by the dealer for marking to market its book. Some dealers calculate CVA once a day. the i collateral is determined using equation (3). vi defined as the net exposure of the counterparty to the dealer at time ti* . The market variables affecting the nodefault value of a dealer’s derivatives with a counterparty are simulated between times 0 and T in a riskneutral world. and si defined as the dealer’s credit spread for a maturity ti. The variable vi is estimated as the present value of the average of the calculated net exposures at time ti* (1 i n) . If a new transaction is contemplated with a counterparty.11 One approach is to arrange the simulation so that the value of the dealer’s portfolio with the counterparty is calculated at times ti* c and ti* (1 i n) . In practice. The Monte Carlo simulation is computationally quite time consuming. The Monte Carlo simulation can then be implemented by assuming that a) the returns on market variables between valuation dates and b) the discount rates between valuation dates equal the yield on the numeraire bond. 12 Only a small amount of additional computation time is necessary to calculate DVA at the same time as CVA.
Consider a market variable x with initial value x0. When the variable follows geometric Brownian motion this is not too difficult. It is necessary to calculate the effect on the paths sampled of changing x0 to x0+ and x0− for a small when all random number streams are kept the same.g. 4.) For other variables such as those following mean reverting processes. An interesting point (usually ignored in practice) is that these estimates should in theory be made using the realworld measure to calculate exposures. 95%) of the exposures at times ti* (1 i n) .. not the riskneutral measure.13 Dealers also often estimate “peak exposures” which are high percentiles (e. the impact of a change at time zero on the change at future times is liable to depend on the path followed by the market variable. Suppose that vi and vi are the values calculated for vi when the initial value of the market variable is x0+ and x0–respectively From equation (5) CVA 1 R n qi (vi vi ) x 2 i 1 13 Note that the impact of a new transaction on CVA can be positive or negative. 10 . A small percentage change at time zero leads to the same small percentage change at all future times on all simulation trials. The “maximum peak exposure” is the maximum of the exposures calculated at these times. CVA Exposures Calculating the sensitivity of CVA to a small parallel shift in a counterparty’s credit spread is straightforward using equation (7). Calculating the first and second partial derivatives of CVA with respect to the underlying market variables is liable to be more time consuming. (This is true both when the volatility is deterministic and when the volatility is stochastic.time for each simulation trial to be updated relatively quickly to reflect the impact of the new transaction.
10. with a floor for alpha of 1. under Basel III CVA exposures arising from the underlying market variables are not included in the calculation of market risk capital.2. As already mentioned. In this case. The effect of this is to increase CVA by the alpha multiplier. A Model for WrongWay and RightWay Risk The situation where q is positively dependent on v is referred to as “wrongway” risk. Instead of changing the calculation of v(t) we change the calculation of q(t) so that q(t) depends on the path followed the value of the portfolio up until time t.07 to 1. We will use a different approach. In this case. Estimates of alpha reported by banks range from 1. 11 . The alpha multiplier approach just mentioned is one example of this approach. the CVA has to be 20% higher than that given by the model where q(t) and v(t) are independent. This means that. A simple way of dealing with wrongway risk is to use what is termed the “alpha” multiplier to increase v(t) in the version of the model in which v(t) and q(t) are assumed to be independent. Another sometimes used is to set v(t) equal to the present value of the exposure that is k standard deviations above the average exposure for some k. The situation where q is negatively dependent on v is referred to as “rightway” risk. If a bank does not have its own model for wrong way risk it has to be 40% higher.4 or allows banks to use their own models. there is a tendency for a counterparty to default when the dealer’s exposure is relatively low. 2CVA 1 R n 2 qi (vi vi 2vi ) 2 i 1 x These equations enable CVA risks relating to the underlying market variables to be assessed and hedged. 5. there is a tendency for a counterparty to default when the dealer’s exposure is relatively high. at minimum. The usual approach to modeling wrongway risk is to reflect it in the way the v’s are calculated. The Basel II rules set alpha equal to 1.
Define h(t) as the hazard rate of the counterparty at time t. as will be explained in more detail later. h(t) is an increasing function of w(t). One is to base it on historical data. Two models we have used are: h t exp a t bw t and h t ln 1 exp a t bw t (9) (8) Each of these models has the property that h(t) is a monotonic function of w(t) with h(t) > 0. Data is collected for past dates on a) credit spreads for the counterparty and b) what the value of the current portfolio with the counterparty would have been. when b < 0 they are low. calculated across all simulations. As before we define w(t) as the value to the dealer of its portfolio with the counterparty at time t. The disadvantage of this is that the counterparty’s credit spread may have been heavily influenced by other factors in the past. 14 15 When b > 0 the relevant extreme values are high.15 For moderate values of w(t) the two models are very similar. There are two approaches to determining the parameter b. we use the model in equation (9). The credit spreads can be converted into hazard rates and the parameter b can be estimated from whichever of equations (8) and (9) are used. when b < 0. which corresponds to rightway risk. 12 . which is illustrated in Figure 1. h(t) increases exponentially as w(t) takes more extreme values. The parameter b measures the sensitivity of h(t) to w(t). The function a(t). which corresponds to wrongway risk.14 In the model in equation (9) h(t) increases linearly as w(t) takes on more extreme values. When b > 0. is determined using an iterative search procedure so that the average survival probability. up to any time T matches that calculated from credit spreads. Either model can be modified to incorporate a cap on h(t). In the model given by equation (8). A simple model of wrongway/rightway is to define h(t) as a deterministic function of w(t). In the results we present. h(t) is a decreasing function of w(t).
the spread would rise to 600 basis points. h(0) = 5% and when w(0) = 20 million. Assuming a recovery rate of 40%. h(0) = 10%. we find that at time zero.16 when w(0) = 3 million. w(0). The implementation of the model requires only a small change to the procedure for calculating CVA described in Section 3. Suppose that the current value to the dealer of the portfolio. across all simulations. if w(0) increased to $20 million. For the model in equation (8) hij exp a ti* bwij For the model in equation (9) hij ln 1 exp a ti* bwij 16 Given the imprecision of any attempt to quantify wrongway risk this is a reasonable assumption. is $3 million and the counterparty’s fiveyear credit spread is 300 basis points. Also.0423 per million dollars. b is 0. Assuming the term structure of hazard rates is flat. this means that the average fiveyear hazard rate is 5% per year. suppose that it is estimated that. The values of a(ti* ) for 1 ≤ i ≤ n must be determined so that the average survival probability.0408 per million dollars while. Solving a pair of simultaneous equations for a(0) and b. corresponding to an average fiveyear hazard rate of 10%.The other approach involves some subjective judgement and it best illustrated with an example. b is 0. 13 . for the model in equation (8). This means that we require 1 m k sk tk exp(hij t ) exp 1 R for 1 k n m j 1 i 1 where hij and wij are the values of h(ti* ) and w(ti* ) on the jth simulation trial and m is the number of simulation trials. for the model in equation (9). up to time ti* equals the survival probability calculated from the term structure of credit spreads.
and the volatility of the exchange rate is 15%.17 We assume that the principal is $100 million. Table 3 does so for the case where b = −0. and so on. we randomly generated 250 portfolios.0. the initial exchange rate is 1. As shown by the tables. k is set equal to 1 and an iterative search is used to determine a (t1* ) from the w1j.First. 14 . This is true for both the deltas and gammas with respect to the exchange rate and the deltas and gammas with respect to the credit spread. Table 1 considers the case where b is 0.03 per million dollars and the dealer is short. a threshold of zero with a cure period of 15 days. The asset prices 17 The results from using the model in equation (8) are similar. The magnitude of the effect is difficult to predict. the domestic and foreign risk free rates are 5%. Four different situations are considered: no collateral. Table 2 provides similar results to Table 1 for the case where b is 0. Numerical Results We first illustrate the model in equation (9) by assuming that the dealer has a simple portfolio consisting of a single oneyear forward foreign exchange transaction. the impact of wrongway and rightway risk on CVA depends in a complex way on CVA itself and the collateral arrangements. and an independent amount of $5 million with a cure period of 15 days. Second. 6. a threshold of $10 million with a cure period of 15 days.03 per million dollars and the dealer is long.03 per million dollars and the dealer’s position is long. in some cases. Indeed. the delivery exchange rate specified in the forward contract is also 1. We suppose that the counterparty’s credit spread (all maturities) is 125 basis points. Each portfolio consists of 25 options on one of five different assets.03 per million dollars and the dealer is short. The results illustrate that wrongway and rightway risk have a material effect on the deltas and gammas of CVA as well as on CVA itself. This determines the h2j. In general. Table 4 does so for the case where b = −0.0. k is set equal to 2 and an iterative search is carried out to determine * a(t2 ) from the w2j values and the h1j. This determines the h1j. sometimes the direction of the effect depends on the collateral arrangements. To illustrate this. even the direction of effect can be difficult to predict.
are assumed to follow geometric Brownian motion with pairwise correlations of 0. Figure 3 provides a similar plot to Figure 2 for the situation where the threshold is $10 million and the cure period is 15 days. The reason is that as CVA increases there is a tendency for both the mean and standard deviation of the value of the portfolio at future times to increase. To understand the reason for this. (a) It is equally likely to be long or short (b) It is equally likely to be a call or a put (c) The underlying is equally likely to any one of the five assets (d) All maturities between one and five years are equally likely (e) All strike prices within 30% of the current asset price are equally likely (f) The underlying principal is $25 million. The highest theoretical CVA is achieved when the value of w(t) is 15 . The gap between the exposures on highw(t) paths and loww(t) paths increases. the credit spread of the counterparty (all maturities) is 125 basis points. For low w(t)’s this restriction has little effect. It can be seen that the change in CVA tends to increase as CVA increases. Each option has the following properties. The riskfree rate is 5%.01 per million dollars and the CVA for b=0 for the 250 portfolios when there is no collateralization. They have an initial price of $25 and a volatility of 25%. as w(t) increases. Wrongway risk causes the hazard rate to increase dramatically on the highw(t) paths and decrease (modestly) on the loww(t) paths. the net exposure is increasingly impacted by the $10 million restriction. The assets do not provide any income. The gap between the exposure on highw(t) paths and loww(t) paths is much less than in the no collateral case. However. we ignore the cure period and focus on the threshold. The average relationship between the change in CVA and CVA is in this case is one where the change first increases and then decreases. The impact of the threshold is to restrict the net exposure to less than $10 million. Figure 2 provides a scatter plot of the relationship between the dollar change in CVA caused increasing b from 0 to 0.36. and the recovery rate is 40%.
It is necessary for the dealer to make a single estimate describing the sensitivity of the counterparty’s credit spread to the value of its portfolio with the counterparty.01. As in the case of Figure 4 the impact of the cure period on the exposure at time t for a particular w(t) depends on the standard deviation of the change in w(t) during the cure period. it is difficult to estimate these effects without a model. but its effect is less than the effect of the threshold. except that the change in CVA is negative instead of positive. for positive exposures to be generated.01 instead of from 0 to +0.certain to be above $10 million at all times. Indeed. High standard deviations of w(t) tend to be associated with high w(t)’s. Because CVA is such a complex derivative. The model can be implemented by making a small change to usual method for calculating CVA. 7. 16 . In this case. Indeed. Tests of the model show that wrongway and rightway risk have a significant effect on the Greek letters of CVA as well as on CVA itself. High w(t)’s are in turn associated with increases in the hazard rate when there is wrongway risk. the standard deviation has to be sufficiently high that there is a reasonable chance of a $5 million increase in w(t) during the cure period. In this case the net exposure is entirely as a result of the cure period. The reasons for the pattern observed are similar to those for Figure 4. This leads to the pattern shown in Figure 4. Figure 4 provides a similar plot to Figures 2 and 3 for the situation where the threshold is zero and the cure period is 15 days. The results are similar to those in Figures 2 to 5. Conclusions We have proposed a simple model for handling wrongway/rightway risk in the calculation of CVA. We have carried out other experiments involving portfolios of interest rate swaps and portfolios. The results are similar to those shown in Figures 2 to 5. Figure 5 provides a plot for the situation where there is an independent amount equal to $5 million and a cure period of 15 days. When w(t) is positive. This explains the pattern observed in Figure 3. even the sign of the effect can be counterintuitive. the net exposure increases as the standard deviation of the change in w(t) during the cure period increases. The cure period does have an effect on the results. The impact of rightway risk can be examined by changing b from 0 to −0. the net exposure is the same on all w(t) paths and wrongway risk has no effect.
the dollar impact of wrongway risk on CVA tends to increase as CVA increases. the impact of wrong way risk tends to first increase and then decrease. 17 . As CVA increases.Further tests involving the random generation of portfolios indicate that when there is no collateral or when collateral is posted with zero threshold or when collateral is posted with an independent amount. The situation where the threshold is materially positive is more interesting.
” Chapter 14 in Lessons From the Crisis edited by Arthur Berd. Hull.” October 16. 2. 23.REFERENCES Basel Committee on Banking Supervision (2010). December 2010.” Second Edition. "The Impact of Default Risk on the Prices of Options and Other Derivative Securities. and David Saunders (2010). “Valuing Credit Default Swaps II: Modeling Default Correlations.'' Journal of Banking and Finance. Juan Antonio de Juan Herrero. Mark (2011). 8. Pengelley.pdf. Cepedes. Pearson. Reuters (2008).org/publ/bcbs189_dec2010. Update 1Lehman Brothers Holdings is Focus of Grand Jury Probe. Juan Carlos Garcia.” Journal of Risk Model Validation. “Counterparty Credit Risk: The New Challenge for Financial Markets” John Wiley and Sons.” http://www. Evan (2005). pp1222.” Journal of Derivatives. “Credit Risk. London: Risk Books. and Ronnie Barnes (2010). 1. Jon (2009).” Chapter 9 in Asset/Liability Management for Financial Institutions. “Effective Modeling of Wrong Way Risk. Edited by Leo Tilman. Eduardo and Darrell Duffie (2003). and Economic Capital for Counterparty Credit Risk.” in Counterparty Credit Risk Modelling edited by Michael Pykhtin. 9. pp 7198 Gregory. Counterparty Credit Risk Capital and Alpha in Basel II. CVA. “Risk Management and Financial Institutions. John and Alan White (1995). Picault. pp 3739. pp 299322. and the Equivalent Bond. John and Alan White (2001). Sokol. Dan Rosen. Canabarro. Alexander (2010). “A Practical Guide to Monte Carlo CVA. Hull. 4. “Measuring and Marking Counterparty Risk. Rebonato. Riccardo. Hull. “CVA Melee” Risk. New York: Institutional Investor books. London: Risk Books 18 .bis. “Calculating and Hedging Exposure. 24.” Risk. Mike Sherring. 3 (Spring 2001). pp 118121. Vol 19. “Basel III: A Global Regulatory Framework for More Resilient Banks and Banking Systems. John (2010). CVA.
4% 77.8% 122.048 40.048 54.0. K is the threshold in $ millions.3% –0.7% 52.001 28.4% 33.0% 7.7% –7. The credit spread is 125 basis points for all maturities and the recovery rate is 40%.9% –1.5% 16.3% 12.039 34.5% 39. The current exchange rate is 1. No Collateral CVA ($ millions) for b = 0 Impact of b = 0.9% 16.7% 15. and the volatility of the exchange rate is 15%.8% 70.2% K = 10 c = 15 0. No Collateral CVA ($ millions) for b = 0 Impact of b = 0.9% 26.8% K = –5 c = 15 0. and b is the parameter in equation (9).4% 41. and the volatility of the exchange rate is 15% .5% 28.6% K=0 c = 15 0.8% 32. the domestic and foreign riskfree interest rates are both 5%. c is the cure period in days.0% 114.8% 17. the domestic and foreign riskfree interest rates are both 5%.4% 27. and b is the parameter in equation (9).011 37.6% 53.0% . The credit spread is 125 basis points for all maturities and the recovery rate is 40%.8% K = 10 c = 15 0.03 per $mm on: CVA Delta wrt Exch Rate Gamma wrt Exch Rate Delta wrt Spread Gamma wrt Spread 19 0.8% –21.3% Table 2: Impact of wrongway risk on CVA for a short forward contract to sell 100 million units of a foreign currency in one year. c is the cure period in days.7% 36.03 per $mm on: CVA Delta wrt Exch Rate Gamma wrt Exch Rate Delta wrt Spread Gamma wrt Spread 0.011 27.7% 91. K is the threshold in $ millions.6% K = –5 c = 15 0.2% 124.036 41.8% 184.0% –341.7% K=0 c = 15 0.Table 1: Impact of wrongway risk on CVA for a long forward contract to buy 100 million units of a foreign currency in one year. The current exchange rate is 1.002 53.0.0% 40.0% 2.8% 181.3% –25.
001 –33.3% Table 4: Impact of rightway risk on CVA for a short forward contract to buy 100 million units of a foreign currency in one year.5% –25.Table 3: Impact of rightway risk on CVA for a long forward contract to buy 100 million units of a foreign currency in one year.9% –16. K is the threshold in $ millions.8% –16.0.9% 209.011 K = –5 c = 15 0.3% 0. No Collateral CVA ($ millions) for b = 0 Impact of b = –0.5% –26.048 K = 10 c = 15 0.2% –35.7% –71.7% –69. c is the cure period in days.9% –30.8% 11. The credit spread is 125 basis points for all maturities and the recovery rate is 40%.6% –75.036 K=0 c = 15 0. The current exchange rate is 1. and the volatility of the exchange rate is 15%. and b is the parameter in equation (9).2% –37. and b is the parameter in equation (9).6% –78.4% –33.2% –79.9% 6.1% –35.1% –14.0% –28.8% –13.03 per $mm on: CVA Delta wrt Exch Rate Gamma wrt Exch Rate Delta wrt Spread Gamma wrt Spread 0. the domestic and foreign riskfree interest rates are both 5%. The credit spread is 125 basis points for all maturities and the recovery rate is 40%.0% 20 .8% –30.011 K = –5 c = 15 0.1% –37.7% –18.7% –8.9% –19.7% –32.5% –26.2% –32.039 K=0 c = 15 0.6% 14. the domestic and foreign riskfree interest rates are both 5%. K is the threshold in $ millions. and the volatility of the exchange rate is 15%.7% –28.03 per $mm on: CVA Delta wrt Exch Rate Gamma wrt Exch Rate Delta wrt Spread Gamma wrt Spread 0.048 K = 10 c = 15 0.5% –29. The current exchange rate is 1.6% –77.7% –25.9% –72. No Collateral CVA ($ millions) for b = 0 Impact of b = –0.5% –74.0.9% –4.002 –37. c is the cure period in days.3% –26.
7 0.4 0.4 0.01 and (ii) a(t)= −4 and b=−0.Figure 1: The model in equation (9) when (i) a(t)= −4 and b=0.7 0.8 0.3 0.8 0.1 h(t) w(t) 300 400 ‐400 ‐300 ‐200 ‐100 0 0 100 200 w(t) 300 400 (i) (ii) 21 .2 0.1 0 ‐400 ‐300 ‐200 ‐100 0 100 200 h(t) 0.5 0.6 0.2 0.6 0.3 0.5 0.01 0.
2 1 0.8 0. $ million Change in CVA ($ millions) 22 .5 2 2.5 1 1. The horizontal axis shows CVA when b=0.01 per million.2 0 0 0.Figure 2: Impact of wrongway risk for 250 portfolios of options when there is no collateral. 1.5 3 CVA (no wrong‐way‐risk. The vertical axis shows the change in CVA when b is increased from 0 to 0.4 0.6 0.
02 0.3 0.07 0.1 0.08 0.2 0.01 0 0 0.03 0. $ million) Change in CVA ($ millions) 23 .5 0.4 0.06 0.09 0.Figure 3: Impact of wrongway risk for 250 portfolios of options when the threshold is $10 million and the cure period is 15 days. The vertical axis shows the change in CVA when b is increased from 0 to 0.01 per million 0.1 0.6 CVA (no wrong‐way risk.05 0. The horizontal axis shows CVA when b=0.04 0.
04 0.Figure 4: Impact of wrongway risk for 250 portfolios of options when the threshold is zero and the cure period is 15 days.1 0. $ millions) Change in CVA ($ millions) 24 .15 CVA (no wrong‐way risk.06 0. 0.03 0. The horizontal axis shows CVA when b=0.05 0. The vertical axis shows the change in CVA when b is increased from 0 to 0.05 0.01 0 0 0.01.02 0.
The horizontal axis shows CVA when b=0.035 0.015 0.005 0 0 0.04 0.02 0.07 CVA (no wrong‐way risk.03 0.01 0.06 0.05 0.025 0. The vertical axis shows the change in CVA when b is increased from 0 to 0. $ millions) Change in CVA ($ millions) 25 .03 0.01 0.04 0.045 0.Figure 5: Impact of wrongway risk for 250 portfolios of when there is an independent amount of $5 million and a cure period of 15 days.01 per million 0.02 0.