This action might not be possible to undo. Are you sure you want to continue?
DERIVATIVES PRICING
riskmagazine.net 97
derivatives pricing theory (see, for
example, Hull, 2006) relies on the
assumption that one can borrow and lend at a unique riskfree
rate. e realities of being a derivatives desk are, however, rather
diﬀerent these days, as historically stable relationships between
bank funding rates, government rates, Libor rates, etc, have bro
ken down.
e practicalities of funding, that is, how dealers borrow and
lend money, are of central importance to derivatives pricing,
because replicating naturally involves borrowing and lending
money and other assets. In this article, we establish derivatives
valuation formulas in the presence of such complications start
ing from ﬁrst principles, and study the impact of market fea
tures such as stochastic funding and collateral posting rules on
values of fundamental derivatives contracts, including forwards
and options.
Simplifying considerably, we can describe a derivatives desk’s
activities as selling derivatives securities to clients while hedging
them with other dealers. Should the desk default, a client would
join the queue of the bank’s creditors. e situation is a bit diﬀer
ent for trading among dealers where, to reduce credit risk, agree
ments have been put in place to collateralise mutual exposures.
Such agreements are based on the socalled credit support
annex (CSA) to the International Swaps and Derivatives Associa
tion master agreement, so we often refer to collateralised trades as
CSA trades. As collateral is used to oﬀset liabilities in case of a
default, it could be thought of as an essentially riskfree invest
ment, so the rate on collateral is usually set to be a proxy of a risk
free rate such as the fed funds rate for dollar transactions, Eonia
for euro, etc. Often, purchased assets are posted as collateral
against the funds used to buy them, such as in the ‘repo’ market
for shares used in delta hedging.
Secured borrowing will normally attract a better rate than
unsecured borrowing. In a bank, funding functions are often
centralised within a treasury desk. e unsecured rates that the
treasury desk provides to the trading desks are generally linked to
the unsecured funding rate at which the bank itself can borrow/
lend, a rate typically based on the bank credit rating, that is, its
perceived probability of default.
e money that a derivatives desk uses in its operations comes
from a multitude of sources, from the collateral posted by coun
terparties to funds secured by various types of assets. We show in
this article how to aggregate these rates to come up with the value
of a derivatives security given the rules for collateral posting and
repo rates available for the underlying. Note that some desks may
be required to borrow at rates diﬀerent from those that they can
lend at – a complication we avoid in this article as our formalism
does not extend readily to the nonlinear partial diﬀerential equa
tions that such a setup would require.
Having derived an appropriate extension to the standard no
arbitrage result, we then look carefully at the diﬀerences in value
of CSA (that is, collateralised) and nonCSA (not collateralised)
versions of the same derivatives security. is is important as
dealers often calibrate their models to marketobserved prices of
derivatives, which typically reﬂect CSAbased valuations, yet they
also trade a large volume of nonCSA overthecounter deriva
tives. We demonstrate that a number of often signiﬁcant adjust
ments are required to reﬂect the diﬀerence between CSA and
nonCSA trades.
e ﬁrst adjustment is to use diﬀerent discounting rates for
CSA and nonCSA versions of the same derivative. e second
adjustment is a convexity, or quanto, adjustment and aﬀects for
ward curves – such as equity forwards or Libor forward rates – as
they turn out to depend on collateralisation used. is is a conse
quence of the stochastic funding spread and, in particular, of the
correlation between the bank funding spread and the underlying
assets. e third adjustment that may be required is to volatility
information used for options – in particular, the volatility smile
changes depending on collateral. We show some numerical results
for these eﬀects.
Preliminaries
We start with the riskfree curve for lending, a curve that corre
sponds to the safest available collateral (cash). We denote the cor
responding short rate at time t by r
C
(t); ‘C’ here stands for ‘CSA’,
as we assume this is the agreed overnight rate paid on collateral
among dealers under CSA. It is convenient to parameterise term
curves in terms of discount factors; we denote corresponding risk
free discount factors by P
C
(t, T), 0 ≤ t ≤ T < ∞. Standard Heath
JarrowMorton theory applies, and we specify the following
dynamics for the yield curve:
dP
C
t,T / P
C
t,T r
C
t dt
C
t,T
T
dW
C
t
(1)
Funding beyond discounting: collateral
agreements and derivatives pricing
Standard theory assumes traders can lend
and borrow at a riskfree rate, ignoring the
intricacies of the repo and collateralisation
markets. Here, Vladimir Piterbarg
shows that these force adjustments to
discounting, forward prices and implied
volatilities, depending on the particulars of
collateral posting
Standard
98 Risk February 2010
CUTTING EDGE. DERIVATIVES PRICING
where W
C
(t) is a ddimensional Brownian motion under the risk
neutral measure P and σ
C
is a vectorvalued (dimension d) sto
chastic process.
In what follows, we shall consider derivatives contracts on a
particular asset, whose price process we denote by S(t), t ≥ 0. We
denote by r
R
(t) the short rate on funding secured by this asset
(here ‘R’ stands for ‘repo’). e diﬀerence r
C
(t) – r
R
(t) is some
times called the stock lending fee. Finally, let us deﬁne the short
rate for unsecured funding by r
F
(t), t ≥ 0. As a rule, we would
expect that r
C
(t) ≤ r
R
(t) ≤ r
F
(t).
e existence of nonzero spreads between short rates based
on diﬀerent collateral can be recast in the language of credit
risk, by introducing joint defaults between the bank and vari
ous assets used as collateral for funding. In particular, the
funding spread s
F
(t) r
F
(t) – r
C
(t) could be thought of as the
(stochastic) intensity of default of the bank. We do not pursue
this formalism here (see, for example, Gregory, 2009, or Bur
gard & Kjaer, 2009), postulating the dynamics of funding
curves directly instead. Likewise, we ignore the possibility of a
counterparty default, an extension that could be developed
rather easily.
BlackScholes with collateral
Let us look at how the standard BlackScholes pricing formula
changes in the presence of a CSA. Let S(t) be an asset that fol
lows, in the real world, the following dynamics:
dS t / S t
S
t dt
S
t dW t
Let V(t, S) be a derivatives security on the asset; by Itô’s lemma it
follows that:
dV t LV t dt t dS t
where L is the standard pricing operator:
L
t
S
t
2
S
2
2
2
S
2
and Δ is the option’s delta:
t
V t
S
Let C(t) be the collateral (cash in the collateral account) held at
time t against the derivative. For ﬂexibility, we allow this amount
to be diﬀerent
1
from V(t).
To replicate the derivative, at time t we hold Δ(t) units of stock
and γ(t) cash. en the value of the replication portfolio, which
we denote by Π(t), is equal to:
V t t t S t t
(2)
e cash amount γ(t) is split among a number of accounts:
N Amount C(t) is in collateral.
N Amount V(t) – C(t) needs to be borrowed/lent unsecured from
the treasury desk.
N Amount Δ(t)S(t) is borrowed to ﬁnance the purchase of Δ(t)
stocks. It is secured by stock purchased.
N Stock is paying dividends at rate r
D
.
e growth of all cash accounts (collateral, unsecured, stock
secured, dividends) is given by:
dy t ( ) r
C
t ( )C t ( ) + r
F
t ( ) V t ( ) ÷ C t ( ) ( )
¸
÷r
R
t ( ) A t ( )S t ( ) + r
D
t ( ) A t ( )S t ( )1
]
dt
On the other hand, from (2), by the selfﬁnancing condition:
d t dV t t dS t
which is, by Itô’s lemma:
dV t ( ) ÷ A t ( )dS t ( )
LV t ( ) ( )dt
o
ot
+
o
S
t ( )
2
2
S
2
o
2
oS
2
¸
¸
_
,
V t ( )dt
us we have:
o
ot
+
o
S
t ( )
2
2
S
2
o
2
oS
2
¸
¸
_
,
V
r
C
t ( )C t ( ) + r
F
t ( ) V t ( ) ÷ C t ( ) ( ) + r
D
t ( ) ÷ r
R
t ( ) ( )
oV
oS
S
which, after some rearrangement, yields:
oV
ot
+ r
R
t ( ) ÷ r
D
t ( ) ( )
oV
oS
S +
o
S
t ( )
2
2
S
2
o
2
V
oS
2
r
F
t ( )V t ( ) ÷ r
F
t ( ) ÷ r
C
t ( ) ( )C t ( )
e solution, obtained by essentially following the steps that lead
to the FeynmanKac formula (see, for example, Karatzas &
Shreve, 1997, theorem 4.4.2), is given by:
V t ( ) E
t
e
÷ r
F
u ( )du
t
T
[
V T ( )
¸
¸
+ e
÷ r
F
v ( )dv
t
u
[
r
F
u ( ) ÷ r
C
u ( ) ( )C u ( )du
t
T
(
]
(
_
,
(3)
in the measure in which the stock grows at rate r
R
(t) – r
D
(t),
that is:
dS t / S t r
R
t r
D
t dt
S
t dW
S
t
(4)
Note that if our probability space is rich enough, we can take it to
be the same riskneutral measure P as used in (1). We note that
this derivation validates the view of Barden (2009) (who also cites
Hull, 2006) that the repo rate r
R
(t) is the right ‘riskfree’ rate to
use when valuing assets on S(t).
By rearranging terms in (3), we obtain another useful formula
for the value of the derivative:
V t ( ) E
t
e
÷ r
C
u ( )du
t
T
[
V T ( )
¸
¸
_
,
÷ E
t
e
÷ r
C
v ( )dv
t
u
[
t
T
(
]
r
F
u ( ) ÷ r
C
u ( ) ( ) V u ( ) ÷ C u ( ) ( )du
¸
¸
_
,
(5)
We note that:
E
t
dV t ( ) ( ) = r
F
t ( )V t ( ) − r
F
t ( ) − r
C
t ( ) ( )C t ( ) ( )
dt
= r
F
t ( )V t ( ) − s
F
t ( )C t ( ) ( )dt
(6)
So, the rate of growth in the derivatives security is the funding
spread r
F
(t) applied to its value minus the credit spread s
F
(t)
applied to the collateral. In particular, if the collateral is equal to
1
In what follows we use (3), (5) with either C = 0 or C = V. However, these formulas, in their full
generality, could be used to obtain, for example, the value of a derivative covered by oneway (asymmetric)
CSA agreement, or a more general case where the collateral amount tracks the value only approximately
riskmagazine.net 99
the value V then:
E
t
dV t ( ) ( ) r
C
t ( )V t ( )dt, V t ( ) E
t
e
÷ r
C
u ( )du
t
T
[
V T ( )
¸
¸
_
,
(7)
and the derivative grows at the riskfree rate. e ﬁnal value is the
only payment that appears in the discounted expression as the
other payments net out given the assumption of full collateralisa
tion. is is consistent with the drift in (1) as P
C
(t, T) corresponds
to deposits secured by cash collateral. On the other hand, if the
collateral is zero, then:
E
t
dV t ( ) ( ) = r
F
t ( )V t ( )dt
(8)
and the rate of growth is equal to the bank’s unsecured funding
rate or, using credit risk language, adjusted for the possibility of
the bank default. We show later that the case C = V could be
handled by using a measure that corresponds to the riskfree bond
P
C
(t, T) = E
t
(e
–∫
T
t
r
C
(u)du
) as a numéraire and, likewise, the case C = 0
could be handled by using a measure that corresponds to the risky
bond P
F
(t, T) = E
t
(e
–∫
T
t
r
F
(u)du
) as a numéraire.
Before we proceed with valuing derivatives securities in our set
up, let us comment on the portfolio eﬀects of the collateral. When
two dealers are trading with each other, the collateral is applied to
the overall value of the portfolio of derivatives between them, with
positive exposures on some trades oﬀsetting negative exposures on
other trades (socalled netting). Hence, potentially, valuation of
individual trades should take into account the collateral position
on the whole portfolio. Fortunately, in the simple case of the col
lateral requirement being a linear function of the exact value of the
portfolio (the case that includes both the nocollateral case C = 0
and the full collateral case C = V), the value of the portfolio is just
the sum of values of individual trades (with collateral attributed to
trades by the same linear function). is easily follows from the
linearity of the pricing formula (3) in V and C.
Zerostrike call option
Probably the simplest derivatives contract on an asset is a promise
to deliver this asset at a given future time T. e contract could be
seen as a zerostrike call option with expiry T. In the standard the
ory, of course, the value of this derivative is equal to the value of the
asset itself (in the absence of dividends). Let us see what the situa
tion is in our case. e payout of the derivative is given by V(T) =
S(T) and the value, at time t, assuming no CSA, is given by:
V
zsc
t ( ) E
t
e
÷ r
F
u ( )du
t
T
[
S T ( )
¸
¸
_
,
On the other hand, if r
D
(t) = 0, then:
S t ( ) E
t
e
÷ r
R
u ( )du
t
T
[
S T ( )
¸
¸
_
,
as follows from (4) and, clearly, S(t) ≠ V
zsc
(t). e diﬀerence in values
between the derivative and the asset are now easily understood, as
the zerostrike call option carries the credit risk of the bank, while
the asset S(⋅) does not. Or, in our language of funding, the asset S(⋅)
can be used to secure funding – which is reﬂected in the discount
rate applied – while V
zsc
cannot be used for such a purpose.
Forward contract
We now consider a forward contract on S(⋅), where at time t the
bank agrees to deliver the asset at time T, against a cash payment
at time T.
N Without CSA. A noCSA forward contract could be seen as a
derivative with the payout S(T) – F
noCSA
(t, T) at time T, where
F
noCSA
(t, T) is the forward price at t for delivery at T. As the for
ward contract is costfree, we have by (3) that:
0 E
t
e
÷ r
F
u ( )du
t
T
[
S T ( ) ÷ F
noCSA
t,T ( ) ( )
¸
¸
_
,
so we get:
F
noCSA
t,T ( )
E
t
e
÷ r
F
u ( )du
t
T
[
S T ( )
¸
¸
_
,
E
t
e
÷ r
F
u ( )du
t
T
[
¸
¸
_
,
(9)
Going back to (9), let us deﬁne:
P
F
t,T ( ) @ E
t
e
÷ r
F
u ( )du
t
T
[
¸
¸
_
,
Note that this is essentially a creditrisky bond issued by the bank.
en we can rewrite (9) as:
F
noCSA
t,T ( ) =
%
E
t
T
S T ( ) ( )
where the measure P
~
T
is deﬁned by the numeraire P
F
(t, T) as:
e
÷ r
F
u ( )du
0
t
[
P
F
t,T ( ) E
t
e
÷ r
F
u ( )du
0
T
[
¸
¸
_
,
is a Pmartingale. Finally we see that F
noCSA
(t, T) is a P
~
T

martingale.
We note that the value of an asset under no CSA at time t with
payout V(T) is given, by (8), to be:
V t ( ) E
t
e
÷ r
F
u ( )du
t
T
[
V T ( )
¸
¸
_
,
P
F
t,T ( )
%
E
t
T
V T ( ) ( )
so it could be calculated by simply taking the expected value of
the payout in the risky Tforward measure.
N With CSA. Now let us consider a forward contract covered by
CSA, where we assume that the collateral posted C is always equal
to the value of the contract V. Let the CSA forward price F
CSA
(t,
T) be ﬁxed at t, then the value, from (5), is given by:
0 V t ( ) E
t
e
÷ r
C
u ( )du
t
T
[
V T ( )
¸
¸
_
,
E
t
e
÷ r
C
u ( )du
t
T
[
S T ( ) ÷ F
CSA
t,T ( ) ( )
¸
¸
_
,
so that:
F
CSA
t,T ( )
E
t
e
÷ r
C
u ( )du
t
T
[
S T ( )
¸
¸
_
,
E
t
e
÷ r
C
u ( )du
t
T
[
¸
¸
_
,
(10)
Comparing this with (9), we see that in general:
F
CSA
t,T ( ) ≠ F
noCSA
t,T ( )
By the arguments similar to the noCSA case, we obtain:
F
CSA
t,T ( ) = E
t
T
S T ( ) ( )
where the measure P
T
is the standard Tforward measure, that is,
a measure deﬁned by P
C
(t, T) = E
t
(e
–∫
T
t
r
C
(u)du
) as a numeraire.
We note that the value of an asset under CSA at time t with
payout V(T) is given, by (7), to be:
100 Risk February 2010
CUTTING EDGE. DERIVATIVES PRICING
V t ( ) E
t
e
÷ r
C
u ( )du
t
T
[
V T ( )
¸
¸
_
,
P
C
t,T ( ) E
t
T
V T ( ) ( )
so it could be calculated by simply taking the expected value of
the payout in the (riskfree) Tforward measure.
N Calculating CSA convexity adjustment. Let us now calcu
late the diﬀerence between CSA and nonCSA forward prices.
We have:
F
noCSA
t,T ( )
%
E
t
T
S T ( ) ( )
E
t
e
÷ r
F
u ( )du
t
T
[
S T ( )
¸
¸
_
,
P
F
t,T ( )
E
t
e
÷ r
C
u ( )du
t
T
[
e
÷ r
F
u ( )÷r
C
u ( ) ( )du
t
T
[
S T ( )
¸
¸
_
,
P
F
t,T ( )
P
C
t,T ( )
P
F
t,T ( )
E
t
T
e
÷ s
F
u ( )du
t
T
[
S T ( )
¸
¸
_
,
E
t
T
M T,T ( )
M t,T ( )
S T ( )
¸
¸
_
,
(11)
where:
M t,T ( ) @
P
F
t,T ( )
P
C
t,T ( )
e
− s
F
u ( )du
0
t
∫
(12)
is a P
T
martingale, as:
M t,T ( ) E
t
T
e
÷ s
F
u ( )du
0
T
[
¸
¸
_
,
We note that, trivially:
E
t
T
M T,T ( )
M t,T ( )
= 1
so:
F
noCSA
t,T ( ) ÷ F
CSA
t,T ( )
E
t
T
M T,T ( )
M t,T ( )
÷ E
t
T
M T,T ( )
M t,T ( )
¸
¸
_
,
¸
¸
_
,
S T ( ) ÷ F
CSA
t,T ( ) ( )
¸
¸
_
,
1
M t,T ( )
Cov
t
T
M T,T ( ), F
CSA
T,T ( ) ( )
(13)
To obtain the actual value of the adjustment we would need to
postulate joint dynamics of s
F
(u) and S(u), u ≥ t. We present a
simple model below where we carry out the calculations.
N Relationship with futures contracts. At ﬁrst sight, a forward
contract with CSA looks rather like a futures contract on the
asset. Recall that with futures contracts, the (daily) diﬀerence in
the futures price gets credited/debited to the margin account. In
the same way, as forward prices move, a CSA forward contract
also speciﬁes that money exchanges hands. ere is, however, an
important diﬀerence. Consider the value of a forward contract at
t′ > t, a contract that was entered at time t (so V(t) = 0). en:
V ´ t ( ) E
´ t
e
÷ r
C
u ( )du
´ t
T
[
S T ( ) ÷ F
CSA
t,T ( ) ( )
¸
¸
_
,
E
´ t
e
÷ r
C
u ( )du
´ t
T
[
S T ( )
¸
¸
_
,
÷ E
´ t
e
÷ r
C
u ( )du
´ t
T
[
¸
¸
_
,
F
CSA
t,T ( )
By (10):
V ´ t ( ) ÷ V t ( ) E
´ t
e
÷ r
C
u ( )du
´ t
T
[
¸
¸
_
,
F
CSA
´ t ,T ( ) ÷ F
CSA
t,T ( ) ( )
so the diﬀerence in contract values on t′ and t that exchanges
hands at t′ is equal to the discounted (to T) diﬀerence in forward
prices. For a futures contract, the diﬀerence will not be dis
counted. erefore, the type of convexity eﬀects we see in futures
contracts are diﬀerent from what we see in CSA versus noCSA
forward contracts, a conclusion diﬀerent from that reached in
Johannes & Sundaresan (2007).
Europeanstyle options
Consider now a Europeanstyle call option on S(T) with strike K.
Depending on the presence or absence of CSA, we get two
prices:
V
noCSA
t ( ) E
t
e
÷ r
F
u ( )du
t
T
[
S T ( ) ÷ K ( )
+ ¸
¸
_
,
V
CSA
t ( ) E
t
e
÷ r
C
u ( )du
t
T
[
S T ( ) ÷ K ( )
+ ¸
¸
_
,
(where for the CSA case we assumed that the collateral posted, C,
is always equal to the option value, V
CSA
). By the same measure
change arguments as in the previous section:
V
noCSA
t ( ) = P
F
t,T ( )
%
E
t
T
S T ( ) − K ( )
+
( )
V
CSA
t ( ) = P
C
t,T ( ) E
t
T
S T ( ) − K ( )
+
( )
e diﬀerence between measures P
~
T
t
and P
T
t
not only manifests
itself in the mean of S(T) – as already established in the previous
section – but also shows up in other characteristics of the distri
bution of S(⋅), such as its variance and higher moments. We
explore these eﬀects in the next section.
N Distribution impact of convexity adjustment. Let us see
how a change of measure aﬀects the distribution of S(⋅). In the
spirit of (11), we have:
V
noCSA
t ( ) = P
F
t,T ( ) E
t
T
M T,T ( )
M t,T ( )
S T ( ) − K ( )
+
where M(t, T) is deﬁned in (12). en, by conditioning on S(T),
we obtain:
V
noCSA
t P
F
t,T E
t
T
t,T, S T S T K
(14)
where the deterministic function α(t, T, x) is given by:
o t,T, x ( ) E
t
T
M T,T ( )
M t,T ( )
S T ( ) x
¸
¸
_
,
Inspired by Antonov & Arneguy (2009), we approximate the
function α(t, T, x) by a linear (in x) function:
t,T, x
0
t,T
1
t,T x
and obtain α
0
and α
1
by minimising the squared diﬀerence
(while using the fact that E
T
t
(M(T, T)/M(t, T)) = 1 and E
T
t
(S(T))
= F
CSA
(t, T)):
1
t,T
E
t
T M T ,T
M t ,T
S T
F
CSA
t,T
Var
t
T
S T
0
t,T 1
1
F
CSA
t,T
riskmagazine.net 101
We recognise the term:
E
t
T
M T,T ( )
M t,T ( )
S T ( )
− F
CSA
t,T ( )
as the convexity adjustment of the forward between the noCSA
and CSA versions (see (13)), and rewrite:
1
t,T
F
noCSA
t,T F
CSA
t,T
Var
t
T
S T
Diﬀerentiating (14) with respect to K twice, we obtain the fol
lowing relationship between the probability density functions
(PDFs) of S(T) under the two measures:
%
P
t
T
S T dK
0
t,T
1
t,T K P
t
T
S T dK
(15)
so the PDF of S(T) under the noCSA measure is obtained from
the density of S(T) under the CSA measure by multiplying it with
a linear function. It is not hard to see that the main impact of
such a transformation is on the slope of the volatility smile of S(⋅).
We demonstrate this impact numerically below.
Example: stochastic funding model
Let us consider a simple model that we can use to estimate the
impact of collateral rules on forwards and options. We start with
an asset that follows a lognormal process:
dS t / S t O dt
S
dW
S
t
and funding spread that follows dynamics inspired by a simple
onefactor Gaussian model of interest rates
2
:
ds
F
t
F
s
F
t dt
F
dW
F
t
with 〈dW
S
(t), dW
F
(t)〉 = ρdt. Here ρ is the correlation between the
asset and the funding spread. We also assume for simplicity that
r
C
(t), r
R
(t) are deterministic, while r
D
(t) = 0. en:
F
CSA
t,T ( ) = E
t
S T ( ) ( )
and:
dF
CSA
t,T / F
CSA
t,T
S
dW
S
t
with W
S
(t) being a Brownian motion in the riskneutral measure
P. On the other hand:
dP
F
t,T / P
F
t,T O dt
F
b T t dW
F
t
where:
b T t
1 e
F
T t
F
As M(t, T) is a martingale under P (since r
C
(t) is deterministic,
the measures P and P
T
coincide), we have from (12) that:
dM t,T / M t,T
F
b T t dW
F
t
Also both M(t, T) and F
CSA
(t, T) are martingales under P. We
then have:
d M t,T F
CSA
t,T / M t,T F
CSA
t,T
S
F
b T t dt O dW t
Recall that:
F
noCSA
0,T ( ) ÷ F
CSA
0,T ( )
E
M T,T ( )
M 0,T ( )
F
CSA
T,T ( ) ÷ F
CSA
0,T ( ) ( )
¸
¸
_
,
so that:
F
noCSA
0,T ( ) F
CSA
0,T ( )exp ÷ o
S
o
F
b T ÷ t ( )pdt
0
T
[ ( )
F
CSA
0,T ( )exp ÷o
S
o
F
p
T ÷ b T ( )
×
F
¸
¸
_
,
(16)
and, in the case ℵ
F
= 0:
F
noCSA
0,T ( ) ÷ F
CSA
0,T ( )
F
CSA
0,T ( ) exp ÷o
S
o
F
pT
2
/ 2
( )
÷1
( )
We note that the adjustment grows as (roughly) T
2
. A similar for
–50
–40
–30
–20
–10
0
10
20
30
40
50
M
a
y
2
0
,
2
0
0
5
A
u
g
3
0
,
2
0
0
5
D
e
c
1
3
,
2
0
0
5
M
a
r
2
7
,
2
0
0
6
J
u
l
6
,
2
0
0
6
O
c
t
1
6
,
2
0
0
6
J
a
n
3
1
,
2
0
0
7
M
a
y
1
4
,
2
0
0
7
A
u
g
2
2
,
2
0
0
7
D
e
c
6
,
2
0
0
7
M
a
r
1
9
,
2
0
0
8
J
u
n
2
7
,
2
0
0
8
O
c
t
7
,
2
0
0
8
J
a
n
2
1
,
2
0
0
9
M
a
y
1
,
2
0
0
9
Credit/rates correlation
Credit/equity correlation
%
1 Historical credit spread/interest rates and credit
spread/equity correlation calculated with a rolling
oneyear window
A. Relative dierences between nonCSA and CSA
forward prices with σ
S
= 30%, σ
F
= 1.50%, ℵ
F
= 5.00%
Time/ρ –30% –20% –10% 0% 10%
1 0.07% 0.04% 0.02% 0.00% –0.02%
2 0.26% 0.17% 0.09% 0.00% –0.09%
3 0.58% 0.39% 0.19% 0.00% –0.19%
4 1.02% 0.68% 0.34% 0.00% –0.34%
5 1.57% 1.04% 0.52% 0.00% –0.52%
6 2.23% 1.48% 0.74% 0.00% –0.73%
7 3.00% 1.99% 0.99% 0.00% –0.98%
8 3.87% 2.56% 1.27% 0.00% –1.26%
9 4.85% 3.20% 1.59% 0.00% –1.56%
10 5.92% 3.91% 1.94% 0.00% –1.90%
2
While a diﬀusion process for the funding spread may be unrealistic, the impact of more complicated
dynamics on the convexity adjustment is likely to be muted
mula was obtained by Barden (2009) using a model in which
funding spread is functionally linked to the value of the asset.
Let us perform a couple of numerical experiments. We start
with an equityrelated example. Let us set σ
F
= 30%, a number
roughly in line with implied volatilities of options on the S&P
500 equity index (SPX). We estimate the basispoint volatility of
the funding spread to be σ
F
= 1.50% and mean reversion to be ℵ
F
= 5% by looking at historical data of credit spreads on US banks.
Figure 1 shows a rolling historical estimate of correlations between
credit spreads and the SPX (as well as credit spread and interest
rates in the form of a ﬁveyear swap rate). From this graph, we
estimate a reasonable range for the correlation ρ to be [–30%,
10%]. In table A, we report relative adjustments:
F
noCSA
0,T ( ) − F
CSA
0,T ( )
F
CSA
0,T ( )
for diﬀerent values of correlations and for diﬀerent T from one to
10 years. Clearly, the adjustments could be quite signiﬁcant.
Next we look at the diﬀerence in implied volatilities for CSA
and nonCSA options. We look at options expiring in 10 years
across diﬀerent strikes, with F
CSA
(0, T) = 100. We assume that the
market prices of CSA options are given by the 30% implied vola
tility (for all strikes), so that the ‘CSA distribution’ of the asset is
lognormal with 30% volatility. en we express the distribution
of the underlying asset for nonCSA options as given by (15) in
terms of implied volatilities (using put options and the original
value of the forward, 100, to ensure fair comparison). Figure 2
demonstrates the impact – nonCSA options have lower volatility
(lower put option values), and the volatility smile has a higher
(negative) skew.
Finally, let us look at CSA convexity adjustments to forward
Libor rates. Table B presents absolute diﬀerences (that is, F
noCSA
(0,
T) – F
CSA
(0, T)) in nonCSA versus CSA forward Libor rates ﬁx
ing in one to 30 years over a reasonable range of possible correla
tions. We use the same parameters for the funding spread as above
together with recent marketimplied caplet volatilities and for
ward Libor rates. Again, the diﬀerences are not negligible, espe
cially for longerexpiry Libor rates.
Conclusions
In this article, we have developed valuation formulas for derivative
contracts that incorporate the modern realities of funding and col
lateral agreements that deviate signiﬁcantly from the textbook
assumptions. We have shown that the pricing of noncollateralised
derivatives needs to be adjusted, as compared with the collateral
ised version, with the adjustment essentially driven by the correla
tion between market factors for a derivative and the funding spread.
Apart from rather obvious diﬀerences in discounting rates used for
CSA and nonCSA versions of the same derivative, we have exposed
the required changes to forward curves and, even, the volatility
information used for options. In a simple model with stochastic
funding spreads we demonstrated the typical sizes of these adjust
ments and found them signiﬁcant. N
Vladimir Piterbarg is head of quantitative research at Barclays Capital. He
would like to thank members of the quantitative and trading teams at
Barclays Capital for thoughtful discussions, and referees for comments
that greatly improved the quality of the article. Email: vladimir.piterbarg@
barcap.com
102 Risk February 2010
CUTTING EDGE. DERIVATIVES PRICING
Antonov A and M Arneguy, 2009
Analytical formulas for pricing CMS
products in the Libor market model
with the stochastic volatility
SSRN eLibrary
Barden P, 2009
Equity forward prices in the presence
of funding spreads
ICBI Conference, Rome, April
Burgard C and M Kjaer, 2009
Modelling and successful
management of creditcounterparty
risk of derivative portfolios
ICBI Conference, Rome, April
Gregory J, 2009
Being twofaced over counterparty
credit risk
Risk February, pages 86–90
Hull J, 2006
Options, futures and other derivatives
Prentice Hall
Johannes M and S Sundaresan,
2007
Pricing collateralized swaps
Journal of Finance 62, pages 383–410
Karatzas I and S Shreve, 1997
Brownian motion and stochastic
calculus
Springer
References
25
26
27
28
29
30
31
32
33
34
35
40 60 80 100 120 140 160
Strike
Original (CSA) implied volatilities
Adjusted (nonCSA) implied volatilities, corr = –30%
Adjusted (nonCSA) implied volatilities, corr = –10%
Adjusted (nonCSA) implied volatilities, corr = 10%
%
Note: T = 10 years, F
CSA
(0, T) = 100,
F
= 1.50%,
F
= 5.00%
2 Difference in CSA v. nonCSA implied distribution for
European options using (15), expressed in implied vol
across strikes, for different levels of correlation ρ
B. Absolute dierences between nonCSA and CSA
forward Libor rates, using marketimplied caplet
volatilities and σ
F
= 1.50%, ℵ
F
= 5.00%
Time/r –20% 0% 20% 40%
1 0.00% 0.00% 0.00% 0.00%
2 0.01% 0.00% –0.01% –0.01%
3 0.01% 0.00% –0.01% –0.02%
4 0.02% 0.00% –0.02% –0.04%
5 0.03% 0.00% –0.03% –0.05%
7 0.05% 0.00% –0.05% –0.10%
10 0.09% 0.00% –0.09% –0.18%
15 0.18% 0.00% –0.18% –0.37%
20 0.30% 0.00% –0.30% –0.60%
25 0.42% 0.00% –0.42% –0.84%
30 0.54% 0.00% –0.54% –1.07%
Stock is paying dividends at rate rD. We denote by rR (t) the short rate on funding secured by this asset (here ‘R’ stands for ‘repo’). is given by: V t Et e T t t dS t S2 2 rF u du where L is the standard pricing operator: L t S V T T t 2 S2 e u t (3) rF v dv rF u rC u C u du and Δ is the option’s delta: t V t S in the measure in which the stock grows at rate rR (t) – rD (t). by Itô’s lemma: dV t t dS t t 2 S LV t dt 2 t 2 S2 2 S2 V t dt us we have: t S 2 t 2 S2 S2 V C t rD t rR t V S S rC t C t rF t V t which. yields: V t rR t rD t rF t V S S S Let us look at how the standard BlackScholes pricing formula changes in the presence of a CSA. postulating the dynamics of funding curves directly instead. the funding spread sF (t) rF (t) – rC (t) could be thought of as the (stochastic) intensity of default of the bank. unsecured. for example. whose price process we denote by S(t). obtained by essentially following the steps that lead to the FeynmanKac formula (see. By rearranging terms in (3). It is secured by stock purchased.2). these formulas. 1997. BlackScholes with collateral d t rC t C t rR t d t rF t V t t S t rD t C t t S t dt On the other hand.CUTTING EDGE. the rate of growth in the derivatives security is the funding spread rF(t) applied to its value minus the credit spread sF(t) applied to the collateral. by Itô’s lemma it follows that: dV t LV t dt t 2 e solution. theorem 4. by introducing joint defaults between the bank and various assets used as collateral for funding. In particular. in the real world. we allow this amount to be diﬀerent1 from V(t). e growth of all cash accounts (collateral. could be used to obtain. which we denote by Π(t). Likewise. we shall consider derivatives contracts on a particular asset. en the value of the replication portfolio. Amount Δ(t)S(t) is borrowed to ﬁnance the purchase of Δ(t) stocks. the value of a derivative covered by oneway (asymmetric) CSA agreement. For ﬂexibility. we can take it to be the same riskneutral measure P as used in (1).4. for example. DERIVATIVES PRICING where WC (t) is a ddimensional Brownian motion under the riskneutral measure P and σC is a vectorvalued (dimension d) stochastic process. Gregory. Karatzas & Shreve. for example. dividends) is given by: In what follows we use (3). at time t we hold Δ(t) units of stock and γ(t) cash. stocksecured. let us deﬁne the short rate for unsecured funding by rF(t). e existence of nonzero spreads between short rates based on diﬀerent collateral can be recast in the language of credit risk. we obtain another useful formula for the value of the derivative: V t Et e Et T t rC u du e cash amount γ(t) is split among a number of accounts: Amount C(t) is in collateral. In particular. We do not pursue this formalism here (see. (5) with either C = 0 or C = V. We note that this derivation validates the view of Barden (2009) (who also cites Hull. we would expect that rC (t) ≤ rR (t) ≤ rF(t). S) be a derivatives security on the asset. 2009. As a rule. is equal to: V t t t S t t (2) Note that if our probability space is rich enough. or Burgard & Kjaer. by the selfﬁnancing condition: dV t t dS t which is. that is: dS t / S t rR t rD t dt S t dW S t (4) Let C(t) be the collateral (cash in the collateral account) held at time t against the derivative. In what follows. from (2). e diﬀerence rC (t) – rR (t) is sometimes called the stock lending fee. To replicate the derivative. an extension that could be developed rather easily. after some rearrangement. or a more general case where the collateral amount tracks the value only approximately 1 V T T t e u t rC v dv (5) rF u rC u V u C u du We note that: Et ( dV ( t ) ) = rF ( t ) V ( t ) − ( rF ( t ) − rC ( t ) ) C ( t ) dt = ( rF ( t ) V ( t ) − s F ( t ) C ( t ) ) dt ( ) (6) So. Amount V(t) – C(t) needs to be borrowed/lent unsecured from the treasury desk. However. we ignore the possibility of a counterparty default. if the collateral is equal to 98 Risk February 2010 . Let S(t) be an asset that follows. t ≥ 0. 2006) that the repo rate rR (t) is the right ‘riskfree’ rate to use when valuing assets on S(t). Finally. in their full generality. the following dynamics: dS t / S t S rF t V t rC t C t 2 t 2 S2 2 S2 V t dt S t dW t Let V(t. 2009). t ≥ 0.
we obtain: FCSA ( t.T EtT V T Zerostrike call option Probably the simplest derivatives contract on an asset is a promise to deliver this asset at a given future time T. at time t. We note that the value of an asset under CSA at time t with payout V(T) is given. then the value. Let the CSA forward price FCSA(t.T Et e T 0 rF u du ~ is a Pmartingale.T S T as follows from (4) and. T C t T F t Et e rF u du (9) Going back to (9). A noCSA forward contract could be seen as a where the measure PT is the standard Tforward measure.T On the other hand.net 99 . When two dealers are trading with each other. adjusted for the possibility of the bank default. Let us see what the situation is in our case. Or. T) as: e r 0 F t ~ u du PF t. where we assume that the collateral posted C is always equal to the value of the contract V. from (5). we see that in general: FCSA ( t.T @ Et e T t rF u du Note that this is essentially a creditrisky bond issued by the bank. to be: V t Et e T t rF u du V T % PF t. e diﬀerence in values between the derivative and the asset are now easily understood. likewise.T ) ≠ FnoCSA ( t.the value V then: Et dV t rC t V t dt. the collateral is applied to the overall value of the portfolio of derivatives between them. Without CSA. T) corresponds to deposits secured by cash collateral. T) = Et(e–∫ r (u)du) as a numéraire and. the asset S(⋅) can be used to secure funding – which is reﬂected in the discount rate applied – while Vzsc cannot be used for such a purpose. We note that the value of an asset under no CSA at time t with payout V(T) is given. is is consistent with the drift in (1) as PC (t. let us comment on the portfolio eﬀects of the collateral. let us deﬁne: PF t. e ﬁnal value is the only payment that appears in the discounted expression as the other payments net out given the assumption of full collateralisation. then: S t Et e T t so that: Et e T t rR u du S T rC u du T t FCSA t. Before we proceed with valuing derivatives securities in our setup. using credit risk language. then: Et ( dV ( t ) ) = rF ( t ) V ( t ) dt rF u du S T FnoCSA t. in our language of funding.T ) = EtT ( S ( T ) ) where the measure PT is deﬁned by the numeraire PF(t. against a cash payment at time T. T) be ﬁxed at t. en we can rewrite (9) as: % FnoCSA ( t. the value of the portfolio is just the sum of values of individual trades (with collateral attributed to trades by the same linear function). Finally we see that FnoCSA(t. where FnoCSA(t. of course. In the standard theory. T) is the forward price at t for delivery at T. a measure deﬁned by PC (t.T ) = EtT ( S ( T ) ) We now consider a forward contract on S(⋅). We show later that the case C = V could be handled by using a measure that corresponds to the riskfree bond PC (t. V t Et e T t rC u du V T (7) derivative with the payout S(T) – FnoCSA(t. potentially. in the simple case of the collateral requirement being a linear function of the exact value of the portfolio (the case that includes both the nocollateral case C = 0 and the full collateral case C = V). we have by (3) that: 0 Et e T t and the derivative grows at the riskfree rate. as the zerostrike call option carries the credit risk of the bank. With CSA. is given by: 0 V t Et e T t Et e rC u du T t rC u du V T rF u du S T S T FCSA t. that is. to be: T C t riskmagazine.T ) By the arguments similar to the noCSA case. if the collateral is zero. As the forward contract is costfree. e payout of the derivative is given by V(T) = S(T) and the value. the case C = 0 could be handled by using a measure that corresponds to the risky bond PF(t. clearly. is easily follows from the linearity of the pricing formula (3) in V and C. valuation of individual trades should take into account the collateral position on the whole portfolio. if rD (t) = 0. T) is a PTmartingale. Hence. where at time t the bank agrees to deliver the asset at time T. assuming no CSA. by (8). Fortunately. Forward contract Et e rC u du (10) Comparing this with (9). e contract could be seen as a zerostrike call option with expiry T.T S T and the rate of growth is equal to the bank’s unsecured funding rate or. S(t) ≠ Vzsc(t). T) = Et(e–∫ r (u)du) as a numeraire. by (7). T) at time T. with positive exposures on some trades oﬀsetting negative exposures on other trades (socalled netting).T so we get: Et e T t rF u du T t (8) FnoCSA t. Now let us consider a forward contract covered by CSA. while the asset S(⋅) does not. T) = Et(e–∫ r (u)du) as a numéraire. On the other hand. is given by: Vzsc t Et e T t so it could be calculated by simply taking the expected value of the payout in the risky Tforward measure. the value of this derivative is equal to the value of the asset itself (in the absence of dividends).
is always equal to the option value.T EtT EtT en.T ) EtT VCSA ( t ) = PC ( t. u ≥ t.T EtT V T V t V t Et e T t rC u du FCSA t . x) by a linear (in x) function: t. trivially: M ( T . In the same way.T .T M t. we obtain: VnoCSA t PF t. T)) = 1 and ET(S(T)) t t = FCSA(t.T EtT FCSA t. the type of convexity eﬀects we see in futures contracts are diﬀerent from what we see in CSA versus noCSA forward contracts.T so it could be calculated by simply taking the expected value of the payout in the (riskfree) Tforward measure.T rC u du e T t rF u rC u du S T PF t.T rC u du S T where: M ( t. FCSA T .T FCSA t.T . Calculating CSA convexity adjustment.T EtT e T t (11) S T Consider now a Europeanstyle call option on S(T) with strike K.T ) EtT ~ is a PTmartingale. Let us see how a change of measure aﬀects the distribution of S(⋅). T)): 1 0 T t rC u du rC u du FCSA t.T M t. we approximate the function α(t. For a futures contract. en: V t Et Et e e T t (14) where the deterministic function α(t. such as its variance and higher moments.T ) EtT ( S (T ) − K ) M ( t. a forward contract with CSA looks rather like a futures contract on the asset. DERIVATIVES PRICING V t Et e T t rC u du V T PC t.T EtT T t Et e T t rF u du S T so the diﬀerence in contract values on t′ and t that exchanges hands at t′ is equal to the discounted (to T) diﬀerence in forward prices.T ) so: FnoCSA t. a CSA forward contract also speciﬁes that money exchanges hands. T. Relationship with futures contracts.T T 0 s F u du (( S (T ) − K ) ) (( S (T ) − K ) ) + + We note that. Depending on the presence or absence of CSA. VCSA). Let us now calculate the diﬀerence between CSA and nonCSA forward prices.T t.T ) + VnoCSA ( t ) = PF ( t. C. x) is given by: t. ere is. an important diﬀerence. as: M t. t. Recall that with futures contracts.T FCSA t.T S T FCSA t. a conclusion diﬀerent from that reached in Johannes & Sundaresan (2007).T M t.T where M(t. we get two prices: VnoCSA t VCSA t Et e Et e T t s F u du rF u du S T S T K K T t M T . a contract that was entered at time t (so V(t) = 0).T e diﬀerence between measures PT and PT not only manifests t t itself in the mean of S(T) – as already established in the previous section – but also shows up in other characteristics of the distribution of S(⋅).T S T x Inspired by Antonov & Arneguy (2009). the (daily) diﬀerence in the futures price gets credited/debited to the margin account.T % EtT S T Et e PC t. x 0 t.T 1 t. At ﬁrst sight.T M t .T EtT 1 M T . we have: M T . however. erefore. By the same measurechange arguments as in the previous section: % VnoCSA ( t ) = PF ( t.T t.CUTTING EDGE.T EtT M T .T ) @ PC ( t. as forward prices move.T M t.T (13) M ( T . x M T . We present a simple model below where we carry out the calculations. Europeanstyle options PF t. We explore these eﬀects in the next section.T . the diﬀerence will not be discounted. T)/M(t. Distribution impact of convexity adjustment. Consider the value of a forward contract at t′ > t.T ) EtT e PF ( t. We have: FnoCSA t. In the spirit of (11).T PF t.T VartT By (10): 1 FCSA 100 Risk February 2010 .T ) − s F ( u )du e ∫0 t (12) (where for the CSA case we assumed that the collateral posted. T) is deﬁned in (12).T ) EtT =1 M ( t. S T S T K To obtain the actual value of the adjustment we would need to postulate joint dynamics of sF(u) and S(u).T x rC u du S T S T FCSA t. T.T S T S T t.T ) 1 CovtT M T .T .T M t.T Et e T t and obtain α0 and α1 by minimising the squared diﬀerence (while using the fact that ET(M(T. by conditioning on S(T).
T exp S F T2 /2 1 We note that the adjustment grows as (roughly) T2 .T Let us consider a simple model that we can use to estimate the impact of collateral rules on forwards and options. 2006 Mar 19. We then have: d M t.87% 4. T) and FCSA(t. Relative di erences between nonCSA and CSA forward prices with σS = 30%.T FCSA T .1 Historical credit spread/interest rates and credit spread/equity correlation calculated with a rolling oneyear window 50 40 30 20 10 0 % –10 –20 –30 –40 –50 May 20.85% 5.T 1 t.T / M t.T FCSA 0.52% –0. 2008 Dec 6.T As M(t. we obtain the following relationship between the probability density functions (PDFs) of S(T) under the two measures: % PtT S T dK 0 T t dW F t Also both M(t.T / FCSA t.27% 1. We also assume for simplicity that rC (t).09% 0.net 101 . A similar for riskmagazine.98% –1.99% 2.19% 0. ℵF = 5. 2005 May 14.56% 3.T FnoCSA t.T ) = Et ( S ( T ) ) and: dFCSA t. σF = 1.19% –0.00% 10% –0.T FCSA 0.T FCSA 0.00% 0.T FCSA t. 2008 Jan 31.23% 3.T M 0.00% Time/ρ 1 2 3 4 5 6 7 8 9 10 –30% 0.00% 0.56% –1.73% –0.00% 0. we have from (12) that: dM t.02% 1.26% 0.T ) M ( t.04% 1.17% 0. while rD (t) = 0. 2006 May 1.T VartT S T FCSA t.T E M T . T) are martingales under P. Here ρ is the correlation between the asset and the funding spread.T K PtT S T dK (15) so the PDF of S(T) under the noCSA measure is obtained from the density of S(T) under the CSA measure by multiplying it with a linear function.T exp FCSA 0. 2007 Aug 22.59% 1. the impact of more complicated dynamics on the convexity adjustment is likely to be muted 2 and. and rewrite: 1 t.91% –10% 0.T ) S ( T ) − FCSA ( t.94% 0% 0.T exp T 0 S S Fb t T t dt and funding spread that follows dynamics inspired by a simple onefactor Gaussian model of interest rates2: ds F t F F T b T F (16) s F t dt F dW F t with 〈dWS (t).68% 1. T) is a martingale under P (since rC (t) is deterministic.02% –0.90% FCSA ( t. On the other hand: dPF t. 2007 Oct 7. rR (t) are deterministic.T FCSA 0.T O dW t T t dt Recall that: FnoCSA 0.50%. 2009 A. 2005 Credit/rates correlation Credit/equity correlation Mar 27.T / PF t.T ) with WS (t) being a Brownian motion in the riskneutral measure P. We start with an asset that follows a lognormal process: dS t / S t O dt S dW S so that: FnoCSA 0.99% 1. 2007 Jul 6.58% 1.00% 3.00% 0.00% 0.T FCSA t.00% 0.20% 3. 2005 Aug 30.00% 0. We demonstrate this impact numerically below.T S dW S t We recognise the term: EtT M ( T . the measures P and PT coincide).74% 0.04% 0.09% –0. 2006 Dec 13.T FCSA 0.26% –1. It is not hard to see that the main impact of such a transformation is on the slope of the volatility smile of S(⋅).00% 0.39% 0. Example: stochastic funding model / M t.92% –20% 0.52% 0.T Fb Diﬀerentiating (14) with respect to K twice. in the case ℵF = 0: FnoCSA 0.T O dt 1 e F Fb T t dW F t where: b T t T t F as the convexity adjustment of the forward between the noCSA and CSA versions (see (13)). 2007 Jun 27.48% 1.00% 0.57% 2.T S Fb t. 2009 Oct 16.34% 0.34% –0. en: While a diﬀusion process for the funding spread may be unrealistic. dWF(t)〉 = ρdt.07% 0.02% 0. 2008 Jan 21.
30% 0. to ensure fair comparison). with the adjustment essentially driven by the correlation between market factors for a derivative and the funding spread. We have shown that the pricing of noncollateralised derivatives needs to be adjusted.com for diﬀerent values of correlations and for diﬀerent T from one to 10 years. 2009 Modelling and successful management of creditcounterparty risk of derivative portfolios ICBI Conference.00% 0. en we express the distribution of the underlying asset for nonCSA options as given by (15) in terms of implied volatilities (using put options and the original value of the forward. FCSA(0. 10%]. and referees for comments that greatly improved the quality of the article.T ) − FCSA ( 0.54% 0% 0. We assume that the market prices of CSA options are given by the 30% implied volatility (for all strikes).42% 0. especially for longerexpiry Libor rates.piterbarg@ barcap.00% 0. expressed in implied vol across strikes.05% 0.03% –0. DERIVATIVES PRICING 2 Difference in CSA v. we report relative adjustments: FnoCSA ( 0.02% –0.00% –0.04% –0.00% Time/r 1 2 3 4 5 7 10 15 20 25 30 –20% 0. 2009 Being twofaced over counterparty credit risk Risk February.30% –0. In table A. Rome. 2006 Options.50%.50%.01% –0.00% 0. Apart from rather obvious diﬀerences in discounting rates used for CSA and nonCSA versions of the same derivative. Clearly.00% –0.T ) FCSA ( 0. Email: vladimir. 2007 Pricing collateralized swaps Journal of Finance 62. corr = –10% Adjusted (nonCSA) implied volatilities. corr = 10% B.42% –0.00% 0. pages 86–90 Hull J.00% 0.01% –0. Conclusions mula was obtained by Barden (2009) using a model in which funding spread is functionally linked to the value of the asset. let us look at CSA convexity adjustments to forward Libor rates.05% –0.00% 0.54% 40% 0.18% –0. We look at options expiring in 10 years across diﬀerent strikes. He would like to thank members of the quantitative and trading teams at Barclays Capital for thoughtful discussions. Rome.00% 20% 0. April Gregory J. ℵF = 5.01% –0. T)) in nonCSA versus CSA forward Libor rates ﬁxing in one to 30 years over a reasonable range of possible correlations.05% –0.02% –0.07% Note: T = 10 years.00% together with recent marketimplied caplet volatilities and forward Libor rates. F = 1.01% 0.50% and mean reversion to be ℵF = 5% by looking at historical data of credit spreads on US banks. In a simple model with stochastic funding spreads we demonstrated the typical sizes of these adjustments and found them signiﬁcant. Absolute di erences between nonCSA and CSA forward Libor rates. From this graph.00% 0.01% 0. we have exposed the required changes to forward curves and. 2009 Equity forward prices in the presence of funding spreads ICBI Conference.00% 0. for different levels of correlation ρ 35 34 33 32 31 30 29 28 27 26 25 40 60 80 100 Strike 120 140 160 % Original (CSA) implied volatilities Adjusted (nonCSA) implied volatilities.37% –0. a number roughly in line with implied volatilities of options on the S&P 500 equity index (SPX). nonCSA implied distribution for European options using (15).CUTTING EDGE. We estimate the basispoint volatility of the funding spread to be σF = 1.18% 0. Figure 1 shows a rolling historical estimate of correlations between credit spreads and the SPX (as well as credit spread and interest rates in the form of a ﬁveyear swap rate). 2009 Analytical formulas for pricing CMS products in the Libor market model with the stochastic volatility SSRN eLibrary Barden P. futures and other derivatives Prentice Hall Johannes M and S Sundaresan. Vladimir Piterbarg is head of quantitative research at Barclays Capital. Let us set σF = 30%. Table B presents absolute diﬀerences (that is. pages 383–410 Karatzas I and S Shreve.60% –0. Again. corr = –30% Adjusted (nonCSA) implied volatilities. FnoCSA(0.02% 0.03% 0. F = 5.00% 0. the volatility information used for options. so that the ‘CSA distribution’ of the asset is lognormal with 30% volatility. 1997 Brownian motion and stochastic calculus Springer 102 Risk February 2010 . T) = 100. T) = 100.18% –0. Let us perform a couple of numerical experiments. T) – FCSA(0.T ) In this article. the adjustments could be quite signiﬁcant. we have developed valuation formulas for derivative contracts that incorporate the modern realities of funding and collateral agreements that deviate signiﬁcantly from the textbook assumptions. Finally. the diﬀerences are not negligible. Figure 2 demonstrates the impact – nonCSA options have lower volatility (lower put option values). we estimate a reasonable range for the correlation ρ to be [–30%.84% –1. We use the same parameters for the funding spread as above References Antonov A and M Arneguy.00% 0.00% 0. and the volatility smile has a higher (negative) skew. using marketimplied caplet volatilities and σF = 1. as compared with the collateralised version. We start with an equityrelated example.09% –0. Next we look at the diﬀerence in implied volatilities for CSA and nonCSA options. April Burgard C and M Kjaer. even. 100. with FCSA(0.10% –0.09% 0.
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 listening from where you left off, or restart the preview.