CURRENCY DERIVATIVES

1 Pure Currency Contracts

Consider a situation where we have two currencies: the domestic currency (say US dollars), and
the foreign currency (say pounds). The spot exchange rate at time t is denoted by X(t), and by
denition it is quoted as
units of the domestic currency
unit of the foreign currency
,
i.e. in our example it is quoted as dollars per pound. We assume that the domestic short rate r
d
,
as well as the foreign short rate r
f
, are deterministic constants, and we denote the corresponding
riskless asset prices by B
d
and B
f
respectively. Furthermore we assume that the exchange rate is
modelled by geometric Brownian motion. We can summarize this as follows.
Assumption 1
We take as given the following dynamics (under the objective probability measure P):
dX = X
X
dt +X
X
d

W (1)
dB
d
= r
d
B
d
dt (2)
dB
f
= r
f
B
f
dt (3)
where
X
,
X
are deterministic constants, and

W is a scalar Wiener process.
Our problem is that of pricing a currency derivative, i.e. a T-claim Z of the form
Z = (X(T)),
European call which gives the owner the option to buy one unit of the foreign currency at the price
K (in the domestic currency).
First we formalize the institutional assumptions.
Assumption 2
All markets are frictionless and liquid. All holdings of the foreign currency are invested in the
foreign riskless asset, i.e. they will evolve according to the dynamics
dB
f
= r
f
B
f
dt.
Remark
Interpreted literally this means that, for example, pounds are invested in a British bank. In reality
this does not have to be the case.
Applying the standard theory of derivatives to the present situation we have the usual risk neutral
valuation formula
(t; Z) = e
r
d
(Tt)
E
Q
t,x
[(X(T))],
and our only problem is to gure out what the martingale measure Q looks like. To do this we use
the result that Q is characterized by the property that every domestic asset has the short rate r
d
1
as its local rate of return under Q. In order to use this characterization we have to translate the
possibility of investing in the foreign riskless asset into domestic terms. Since B
f
(t) units of the
foreign currency are worth B
f
(t) X(t) in the domestic currency we immediately have the following
result.
Lemma 1
The possibility of buying the foreign currency, and investing it at the foreign short rate of interest,
is equivalent to the possibility of investing in a domestic asset with price process

B
f
, where

B
f
(t) = B
f
(t) X(t).
The dynamics of

B
f
are given by
d

B
f
=

B
f
(
X
+r
f
)dt +

B
f

X
d

W.
Summing up we see that our currency model is equivalent to a model of a domestic market consisting
of the assets B
d
and

B
f
. It now follows directly from the general results that the martingale measure
Q has the property that the Q-dynamics of

B
f
are given by
d

B
f
= r
d

B
f
dt +

B
f

X
dW (4)
where W is a Q-Wiener process. Since by denition we have
X(t) =

B
f
B
f
(t)
, (5)
we can use Itos formula, (3) and (4) to obtain the Q-dynamics of X as
dX = X(r
d
r
f
)dt +X
X
dW. (6)
The basic pricing result follows immediately.
Proposition 2 (Pricing formulas)
The arbitrage free price (t; ) for the T-claim Z = (X(T)) is given by (t; ) = F(t, X(t)),
where
F(t, x) = e
r
d
(Tt)
E
Q
t,x
[(X(T))], (7)
and where the Q-dynamics of X are given by
dX = X(r
d
r
f
)dt +X
X
dW. (8)
Alternatively F(t, x) can be obtained as the solution to the boundary value problem

F
t
+x(r
d
r
f
)
F
x
+
1
2
x
2

2
X

2
F
x
2
r
d
F = 0
F(T, x) = (x).
(9)
Comparing (8) to the formulas in the last lecture, we see that a foreign currency is to be treated
exactly as a stock with a continuous dividend. We may thus draw upon the results in last lecture,
2
which allows us to use pricing formulas for stock prices (without dividends) to price currency
derivatives.
Proposition 3 (Option pricing formula)
Let F
0
(t, x) be the pricing function for the claim Z = (X(T)), in a world where we interpret X
as the price of ordinary stock without dividends. Let F(t, x) be the pricing function of the same
claim when X is interpreted as an exchange rate. Then the following relation holds
F(t, x) = F
0
(t, xe
r
f
(Tt)
).
In particular, the price of the European call, Z = max[X(T) K, 0], on the foreign currency, is
given by the modied Black-Scholes formula
F(t, x) = xe
r
f
(Tt)N[d
1
]
e
r
d
(Tt)KN[d
2
]
, (10)
where
d
1
(t, x) =
1

T t
{ln(
x
K
) + (r
d
r
f
+
1
2

2
X
)(T t)},
d
2
(t, x) = d
1
(t, x)
X

T t.
2 Domestic and Foreign Equity Markets
We will model a market which, apart from the objects of the previous section, also includes a
domestic equity with (domestic) price S
d
, and a foreign equity with (foreign) price S
f
.
We model the equity dynamics as geometric Brownian motion, and since we now have three risky
assets we use a three-dimensional Wiener process in order to obtain a complete market.
Assumption 1
The dynamic model of the entire economy, under the objective measure P, is as follows.
dX = X
X
dt +X
X
d

W, (11)
dS
d
= S
d

d
dt +S
d

d
d

W, (12)
dS
f
= S
f

f
dt +S
f

f
d

W, (13)
dB
d
= r
d
B
d
dt (14)
dB
f
= r
f
B
f
dt (15)
where

W =

W
1

W
2

W
3

is a three-dimensional Wiener process (as usual with independent components). Furthermore, the
(3 3)-dimensional matrix , given by
=

X1

X2

X3

d1

d2

d3

f1

f2

f3

,
3
assumed to be invertible.
Remark 1
The reason for the assumption about is that this is the necessary and sucient condition for
completeness. It is also possible, and in many situations convenient, to model the market using
three scalar correlated Wiener processes (one for each asset).
Typical T-contracts which we may wish to price (in terms of the domestic currency) are given by
the following list.
A foreign equity call, struck in foreign currency, i.e. an option to buy one unit of the
foreign equity at the strike price of K units of the foreign currency. The value of this claim at the
date of expiration is, expressed in the foreign currency, given by
Z
f
= max[S
f
(T) K, 0]. (16)
Expressed in terms of the domestic currency the value of the claim at T is
Z
d
= X(T) max[S
f
(T) K, 0]. (17)
A foreign equity call, struck in domestic currency, i.e. a European option to buy one unit
of the foreign equity at time T, by paying K units of the domestic currency. Expressed in domestic
terms this claim is given by
Z
d
= max[X(T) S
f
(T) K, 0]. (18)
An Exchange option which gives us the right to exchange one unit of the domestic equity
for one unit of the foreign equity. The corresponding claim, expressed in terms of the domestic
currency, is
Z
d
= max[X(T) S
f
(T) S
d
(T), 0]. (19)
More generally we will study pricing problems for T-claims of the form
Z = (X(T), S
d
(T), S
f
(T)), (20)
where Z is measured in the domestic currency. We know that the pricing function F(t, x, s
d
, s
f
) is
given by the risk neutral valuation formula
F(t, x, s
d
, s
f
) = e
r
d
(Tt)
E
Q
t,x,s
d
,s
f
[(X(T))],
so we only have to nd the correct risk adjusted measure Q. We transform all foreign traded assets
into domestic terms. The foreign bank account has already been taken care of, and it is obvious
that one unit of the foreign stock, worth S
f
(t) in the foreign currency, is worth X(t) S
f
(t) in
domestic terms. We thus have the following equivalent domestic model, where the asset dynamics
follow from the Ito formula.
Proposition 4
The original market (11)-(15) is equivalent to a market consisting of the price processes S
d
,

S
f
,

B
f
, B
d
,
where

B
f
(t) = X(t)B
f
(t),
4

S
f
(t) = X(t)S
f
(t).
The P-dynamics of this equivalent model are given by
dS
d
= S
d

d
dt +S
d

d
d

W, (21)
d

S
f
=

S
f
(
f
+
X
+
f

X
)dt +

S
f
(
f
+
X
)d

W, (22)
d

B
f
=

B
f
(
X
+r
f
)dt +

B
f

X
d

W, (23)
dB
d
= r
d
B
d
dt. (24)
Here we have used to denote transpose, so

X
=
3

i=1

fi

Xi
.
Note that, because of Assumption 1, the volatility matrix above is invertible, so the market is
complete.
Since S
d
,

S
f
,

B
f
can be interpreted as prices of domestically traded assets, we can easily obtain the
relevant Q-dynamics.
Proposition 5
The Q-dynamics are as follows.
dS
d
= S
d
r
d
dt +S
d

d
dW, (25)
d

S
f
=

S
f
r
d
dt +

S
f
(
f
+
X
)dW, (26)
d

B
f
=

B
f
r
d
dt +

B
f

X
dW, (27)
dX = X(r
d
r
f
)dt +X
X
dW, (28)
dS
f
= S
f
(r
f

f

X
)dt +S
f

f
dW. (29)
We can now immediately obtain the risk neutral valuation formula, and this can in fact be done in
two ways. We can either use (X, S
d
, S
f
) as state variables, or use the quivalent (there is one-to-one
mapping) set (X, S
d
,

S
f
). Which set to use is a matter of convenience, depending on the particular
claim under study, but in both cases the arbitrage free price is given by the discounted expected
value of the claim under the Q-dynamics.
Proposition 6 (Pricing formulas)
For a claim of the form
Z = (X(T), S
d
(T),

S
f
(T))
the corresponding pricing function F(t, x, s
d
, s
f
) is given by
F(t, x, s
d
, s
f
) = e
r
d
(Tt)
E
Q
t,x,s
d
, s
f
[(X(T)), S
d
(T),

S
f
(T))], (30)
where the Q-dynamics are given by Proposition 5.
5
The pricing PDE is
F
t
+x(r
d
r
f
)
F
x
+s
d
r
d
F
s
d
+ s
f
r
d
F
s
f
+
1
2
{x
2
||
X
||
2

2
F
x
2
+s
2
d
||
d
||
2

2
F
s
2
d
+ s
2
f
(||
f
||
2
+ ||
X
||
2
+ 2
f

X
)

2
F
s
2
f
}
+s
d
x
d

2
F
s
d
x
+ s
f
x(
f

X
+ ||
X
||
2
)

2
F
s
f
x
+s
d
s
f
(
d

f
+
d

X
)

2
F
s
d
s
f
r
d
F = 0
F(T, x, s
d
, s
f
) = (x, s
d
, s
f
)
Proposition 7
For a claim of the form
Z = (X(T), S
d
(T), S
f
(T))
the corresponding pricing function F(t, x, s
d
, s
f
) is given by
F(t, x, s
d
, s
f
) = e
r
d
(Tt)
E
Q
t,x,s
d
,s
f
[(X(T)), S
d
(T), S
f
(T))], (31)
where the Q-dynamics are given by Proposition 5.
The pricing PDE is
F
t
+x(r
d
r
f
)
F
x
+s
d
r
d
F
s
d
+s
f
(r
f

f

X
)
F
s
f
+
1
2
{x
2
||
X
||
2

2
F
x
2
+s
2
d
||
d
||
2

2
F
s
2
d
+s
2
f
||
f
||
2

2
F
s
2
f
}
+s
d
x
d

2
F
s
d
x
+s
f
x
f

2
F
s
f
x
+s
d
s
f

2
F
s
d
s
f
r
d
F = 0
F(T, x, s
d
, s
f
) = (x, s
d
, s
f
)
Remark 2.3
In many applications the claim under study is of the restricted form
Z = (X(T), S
f
(T)).
In this case all partial derivatives w.r.t. s
d
vanish from the PDEs above. A similar reduction will
of course also take place for a claim of the form
Z = (X(T), S
d
(T)).
Remark 2.4
In practical applications it may be more convenient to model the market, and easier to read the
6
formulas above, if we model the market using correlated Wiener processes. We can formulate our
basic model (under Q) as
dX = X
X
dt +X
X
d

V
X
,
dS
d
= S
d

d
dt +S
d

d
d

V
d
,
dS
f
= S
f

f
dt +S
f

f
d

V
f
,
dB
d
= r
d
B
d
dt,
dB
f
= r
f
B
f
dt.
where the three processes

V
X
,

V
d
,

V
f
are one-dimensional corelated Wiener processes. We assume
that
X
,
d
,
f
are positive. The instantaneous correlation between

V
X
and

V
f
is denoted by
X
f
and correspondingly for the other pairs. We then have the following set of translation rules between
the two formalisms
||
i
|| =
i
, i = X, d, f,

j
=
i

ij
, i, j = X, d, f,
||
i
+
j
|| =

2
i
+
2
j
+ 2
i

ij
, i, j = X, d, f.
3 Domestic and Foreign Market Prices of Risk
This section constitutes a small digression in the sense that we will not derive any new pricing
formulas. Instead we will take a closer look at the various market prices of risk. As will be shown
below, we have to distinguish between the domestic and the foreign market price of risk, and we
will clarify the connection between these two objects. As a by-product we will obtain a somewhat
deeper understanding of the concept of risk neutrality.
Let us therefore again consider the international model X, S
d
, S
f
, B
d
, B
f
, with dynamics under the
objective measure P given by (11)-(15). As before we transform the international model into the
domestically traded assets S
d
,

S
f
, B
d
,

B
f
with P-dynnamics given by (21)-(24).
In the previous section we infer the existence of a martingale measure Q, under which all domes-
tically traded assets command the domestic short rate r
d
as the local rate of return. Our rst
observation is that, from a logical point of view, we could just as well have chosen to transform
(11)-(15) into equivalent assets traded on the foreign market. Thus we should really denote our
old martingale measure Q by Q
d
in order to emphasize its dependence on the domestic point of
view. If we instead take a foreign investors point of view we will end up with a foreign martingale
measure, which we will denote by Q
f
, and an obvious project is to investigate the relationship
between these martingale measures. A natural guess is perhaps that Q
d
= Q
f
, but as we shall see
this is generically not the case. Since there is of risk, we will carry out the project above in terms
of market prices ot risk.
We start by taking the domestic point of view, and applying the previous result to the domestic
price processes (21)-(24), we infer the exstence of the domestic market price of risk process

d
(t) =

d1
(t)

d2
(t)

d3
(t)

7
with the property that if is the price process of any domestically traded asset in the model, with
price dynamics under P of the form
d(t) = (t)

(t)dt + (t)

(t)d

W(t),
then, for all t and P-a.s., we have

(t) r
d
=

(t)
d
(t)
Applying this to (21)-(23) we get the following set of equations:

d
r
d
=
d

d
, (32)

f
+
X
+
f

X
r
d
= (
f
+
X
)
d
, (33)

X
+r
f
r
d
=
X

d
. (34)
In passing we note that, since the coecient matrix
=

f
+
X

is invertible by Assumption 2.1,

d
is uniquely determind (and in fact constant). This uniqueness
is of course equivalent to the completeness of the model.
We now go on to take the perspective of a foreign investor, and the rst thing to notice is that the
model (11)-(15) of the international market does not treat the foreign and the domestic points of
view symmetrically. This is due to the fact that the exchange rate X by denition is quoted as
units of the domestic currency
unit of the foreign currency
.
From the foreign point of view the exchange rate X should thus be replaced by the exchange rate
Y (t) =
1
X(t)
which is then quoted as
unit of the foreign currency
units of the domestic currency
,
and the dynamics for X, S
d
, S
f
, B
d
, B
f
should be replaced by the dynamics for Y, S
d
, S
f
, B
d
, B
f
.
In order to do this we only have to compute the dynamics of Y , given those of X, and an easy
application of Itos formula gives us
dY = Y
Y
dt +Y
Y
d

W, (35)
where

Y
=
X
+ ||
X
||
2
, (36)

Y
=
X
. (37)
8
Following the arguments the domestic analysis, we now transform the processes Y, S
d
, S
f
, B
d
, B
f
into a set of asset prices on the foreign market, namely S
f
,

S
d
,

B
d
, where

S
d
= Y S
d
,

B
d
= Y B
d
.
If we want to obtain the P-dynamics of S
f
,

S
d
,

B
d
we noe only have to use (21)-(24), substituting
Y for X and d for f. Since we are not intrersted in these dynamics per se, we will, however, not
carry out these computations. The object that we are primarily looking for is the foreign market
price of risk
f
, and we can easily obtain that by writing down the foreign version of the system
and substituting d for f and Y for X directly in (32)-(34). We get

f
r
f
=
f

f
,

d
+
Y
+
d

Y
r
f
= (
d
+
Y
)
f
,

Y
+r
d
r
f
=
Y

f
.
and inserting (36)-(37), we nally obtain

f
r
f
=
f

f
, (38)

d

X
+ ||
X
||
2

X
r
f
= (
d

X
)
f
, (39)

X
+ ||
X
||
2
+r
d
r
f
=
X

f
. (40)
After some simple algebraic manipulations, the two systems (32)-(34) and (38)-(40) can be written
as

X
+r
f
r
d
=
X

d
,

d
r
d
=
d

d
,

f
+
f

X
r
f
=
f

d
,

X
||
X
||
2
+r
f
r
d
=
X

f
,

d

d

X
r
d
=
d

f
,

f
r
f
=
f

f
.
These equitions can be written as
=
d
,
=
f
,
where
=

X
+r
f
r
d

d
r
d

f
+sigma
f

X
r
f

, =

X
||
X
||
2
+r
f
r
d

d

d

X
r
d

f
r
f

So, since is invertible,

d
=
1
,
9

f
=
1
.
Thus we have

d

f
=
1
( ),
and since
=

X
,
we obtain

d

f
=
1
( ) =
1

X
=

X
.
We have thus proved the following central result.
Proposition 10
The foreign market price of risk is uniquely determined by the domestic market price of risk, and
by the exchange rate volatility vector
X
, through the formula

f
=
d

X
. (41)
Remark 3.1
For the benet of the probabilist we note that this result implies that the transition from Q
d
to Q
f
is eected via a Girsanov transformation, for which the likelihood process L has the dynamics
dL = L
X
dW,
L(0) = 1.
Propostion 10 has immediately consequences for the existence of risk neutral markets. If we focud
on the domestic market can we say that the market is (on the aggregate) risk neutral if the following
valuation formula holds, where
d
is the price process for any domestically traded asset.

d
(t) = e
r
d
(Tt)
E
P
[
d
(T)|F
t
]. (42)
In other words, the domestic market is risk neutral if and only if P = Q
d
. In many scientic
papers an assumption is made that the domestic market is in fact risk neutral, and this is of course
a behavioral assumption, typically made in order to facilitate computations. In an international
setting it then seems natural to asume that both the domestic market and the foreign market are
risk neutral, i.e. that, in addition to (42), the following formula also holds, where
f
is the foreign
price of any asset traded on the foreign market

f
(t) = e
r
f
(Tt)
E
P
[
f
(T)|F
t
]. (43)
This seems innocent enough, but taken together these assumptions imply that
P = Q
d
= Q
f
. (44)
Proposition 10 now tells us that (44) can never hold, unless
X
= 0,i.e. if and only if the
exchange rate is deterministic.
10
At rst glance this seems highly counter-intuitive, since the assumption about risk neutrality often
is interpreted as an assumption about the (aggregate) attitude towards risk as such. However,
from (42), which is an equation for objects measured in the domestic currency, it should be clear
that risk neutrality is a property which holds only relative to a specied numeraire. To put it
as a slogan, you may vary well be risk neutral w.r.t. pounds sterling, and still be risk averse w.r.t.
US dollars.
There is nothing very deep going on here: it is basically just the Jensen inequality. To see this
more clearly let us consider the following simplied situation. We assume that r
d
= r
f
= 0, and
we assume that the domestic market is risk neutral. This means in particular that the exchange
rate itself has the following risk neutral valuation formula
X(0) = E[X(T)]. (45)
Looking at the exchange rate from the foreign perspective we see that if the foreign market also is
risk neutral, then it must hold that
Y (0) = E[Y (T)], (46)
with Y = 1/X. The Jensen inequality together with (45) gives us, however,
Y (0) =
1
X(0)
=
1
E[X(T)]
E[
1
X(T)
] = E[Y (T)]. (47)
Thus (45) and (46) can never hold simultaneously with a stochastic exchange rate.
11