stochastic

processes
andtheir
applications
ELSEVIER Stochastic Processes and their Applications 55 (1995) 227--251

General framework for pricing derivative securities

M a r e k Musiela*
Department o[ Statistics, School q/' Mathematics, The University ~?fNSW, P.O. Box 1, Kensington,
NS W 2033, Australia

Received 20 November 1991; revised 27 June 1994

Abstract

This article describes a general methodology that can be used for financial risk management.
The approach is based on the model of Heath et al. (1992) of term structure movements but
deals with the case of incomplete market. Both, domestic and foreign economies are investig-
ated, Prices of various options are calculated using the forward measure introduced recently by
El Karoui and Rochet (1989).

Keyu'ords: Term structure models; HJM framework; Arbitrage free pricing; Martingale
measures

O. Introduction

The general aim of the paper is to develop an approach to management of financial

risks in an international economy. The key issue is then to control interest rate
risks. This is why we start with a model of the term structure of interest rates and
volatilities (Section 1.1). We identify conditions under which there are no arbitrage
opportunities between rates of different maturities. Using the Girsanov transforma-
tion we construct a model of arbitrage-free economy appropriate to price and hedge
claims contingent on interest rate-dependent assets. In Section 1.2 we extend our
model in order to cover other risky assets. This is done using the general methodology
developed in Section 1.1. Finally, in Sections 2.1 and 2.2 we analyse a model of
arbitrage-free international economy. The idea is to consider foreign assets as new
assets in the domestic economy.
Section 1.4 describes the concept of the forward measure recently introduced by El
Karoui and Rochet. Let us point out that this idea leads to significant simplifications
in the derivation of pricing formulas. Sections 1.5 and 2.3 make use of this new
approach to derive pricing formulas for various options.

*Tel: 612 3852959. Fax: 612 3851071.

0304-4149/95/$09.50 ~C) 1995 Elsevier Science B.V. All rights reserved

SSDI 0 3 0 4 - 4 1 4 9 ( 9 4 ) 0 0 0 4 5 - X
228 M. Musieht/Stochastic Processes am/their Applications 55 (1995) 227 251

I. Domestic economy

We present here a general mathematical framework for financial risk management.

Our approach is based on the model of Heath et al. (1992) (HJM) of term structure
movements and hence it allows for arbitrary term structure of interest rates and
volatilities. It can also be seen as a generalisation to the case of incomplete market of
the model presented in Musiela et al. (1993).

1.1. Domestic term structure

We begin with a financial market where investors can trade continuously over the
time interval [0, r]. The traded assets consist of a continuum of default-free discount
bonds with maturities time from 0 to r. We do not assume that our market model is
complete. This allows us subsequently to add additional assets as long as their price
processes are adapted to the information available to investors at any time over the
trading interval.
To be able to analyse mathematically financial risks we introduce a probability
space (f2, F, Po) equipped with a filtration (Ft), 0 < t _< r satisfying the usual condi-
tions. The sub-a-algebra Ft represents all information available to the market partici-
pants at time t. We assume that F = F~.
The most basic fixed income security is the pure discount bond, a bond that pays
a certain dollar at maturity with no other payment in the interim. The time t price of
a T maturity bond is denoted by P(t, T), indicating that the price of the bond is
a function of the trading date t and the maturity date T. The pure discount bond is
a theoretical instrument. Financial instruments that closely resemble these hypotheti-
cal bonds are US Treasury bills. We assume that for each 0 < t < T the price P(t, T)
depends only on information contained in Ft, that is, the process ~P(t, T); 0 _< t < T I
is adapted to the filtration (F,). We also require that P(T, T) = 1 for all 0 _< T_< r,
P(t, T) > 0 for all 0 < t < T < r, and that ~ log P(t, T)/~Texists for all 0 < t _< T _< r.
The last condition implies that the instantaneous forward rate at time T as seen from
date t is well defined. This rate, denotedf(t, T), corresponds to the rate at which one
could enter a contract at time t on a riskless loan over the forward period
[7, T + dT]. Obviously,

P(t,T)=exp(-frf(t,u)du).

Therefore, the process [f(t, T); 0 < t < T~ is also adapted to the filtration (Ft). The
spot rate at time t, denoted r(t), corresponds to the rate at which one could enter
a contract at time t on a riskless loan over the period It, t + dt]. Clearly we have
r(t) = f ( t , t) for all 0 < t < r. Given an initial investment of one dollar at time 0, the
amount generated at time t by continuously reinvesting in the instantaneous spot rate
is equal to

exp(r,,du)
 M. Musiela/Stochastic Processes and their Applicatimls 55 (1995) 227 251 229

The process IB(t), 0 _%<t _< r I is referred to as the accumulation factor. Note that
B(t)/B(T) represents the amount one needs to invest at time t in order to accumulate
one dollar at time T. Hence it seems we can say that P(t, T) and B(t)/B(T) are equal.
In fact they would be if we could know at time t the future behaviour of the spot rate
r over the interval [t, T]. This is exactly where the role of information available to
investors becomes important. We would not know at time t how much is B(t)/B(T),
that is B(t)/B(T) is not F~-measurable. On the other hand, the random variable Pit, T)
is F,-measurable. Therefore, we can interpret P(t, T) as the time t value of a bond that
pays one dollar at T. Finally, note that under the assumption of known deterministic
rates we have P(t, T) = B(t)/B(T). In the general situation, as it will become clear later
on, P(t, T) can be understood as a "'projection" of B(t)/B(T) on the information F,
available at t. Hence, it seems clear that the amount of information and the way
information is structured will be essential in any analysis of the interest rate risk. In
the HJM model the filtration (Ft) is generated by two independent Brownian motions.
Musiela et al. (1993) consider the case of filtration spanned by n independent
Brownian motions. It should be noted, however, that these assumptions are rather
technical. They imply that the corresponding market models are complete and hence
also that claims contingent on interest rate-dependent assets can be generated by
dynamic portfolio strategies based on the family of pure discount bonds with maturi-
ties covering the interval [0, r].
In this paper we assume that for fixed, but arbitrary 0 _< T _< r , f ( t , T) is given by

dr(t, T) = ~(t, T)dt + a*(t, T)dWo(t), tl.l)

where I Wo(t): 0 < t < r] is an n-dimensional F~-Brownian motion while the processes
l~(t, T); 0 _< t, T < r I and ~tcr(t, T); 0 _< t, T _< r) are bounded on [0, r]2 x Q adapted
with values in ~ and ~", respectively. Note that, because (Ft) is not necessarily the
filtration generated by Wo, there may exist contingent claims which cannot be hedged
with a self-financing portfolio constructed from discount bonds. Therefore, our
market model is not complete in general. For the vectors (matrices) x, y the symbols
Ix q, x* and x*y stand for the Euclidean norm, the transpose and the Euclidean scalar
product, respectively.
The differential (1.1) can be written in the following integral form

f(t, T)=.['(O, T)+ fi ~(u, T)du + fl a*(u, T)dWo(u)

for all 0 < t, r < ~, where [./"(0, T); 0 _< T < r~j is an arbitrary bounded initial forward
rate curve. The structure of volatilities is also arbitrary bounded because we only
assume that a ( . , T) is bounded and adapted. The spot rate process satisfies

r(t)=f(O,t)+ fi ~t(u,t)du + flcY*(u,t)dWo(u)

for all t > 0. It is interesting to note that the process r is not a semimartingale in
general.
230 M. Musiela/Stochastic Processesand theh"Applications 55 (1995) 227-251

T h r o u g h o u t the paper we assume the nonexistence of arbitrage opportunities in

financial markets. Based on the results of Harrison and Kreps (1979), in the context of
our model the above assumption will be satisfied if there exists a martingale measure
for discounted price processes of bonds of all maturities. Mathematically speaking this
means that there exists a probability measure P defined on (f2, F) which is equivalent
to Po and such that the processes {Z(t, T); 0 < t < T I, where

Z(t, T ) = P(t, T)/B(t)

are martingales under P for all 0 < T < ~.

In order to construct the measure P we shall henceforth impose the following
assumption.

Assumption I.I. There exists a bounded predictable process H = {H(t); 0 < t < r I
with values in ~" such that for all 0 < t, T < z

~(t, T) = ~*(t, T)
(f/ e(t, s)ds - H(t) .
) (I.2)

For all A e F~, we now define a set function P(A) by the formula

P(A)=fAexp(ffH*(t)dWo(t)-½f/ [H(t)12dOdPo. (1.3)

Because H is assumed bounded it is not difficult to show that P is in fact a probability

measure on (f2, F) which is equivalent to the measure Po. Also it follows from the
Girsanov theorem that under the measure P the process { W(t); 0 < t < r}, where

W(t) = Wo(t) - f l H(s)ds (1.4)

is an n-dimensional F, Brownian motion. This observation helps to describe the

arbitrage free behaviour of forward rates and bond prices.

Proposition I.I. We have

(i) df(t, T) = cr*(t, 7") ~(t, u)dudt + ~*(t, T)dW(t),

(ii) dP(t,T)=P(t,T)Ir(t)dt-ff~*(t,u)dudW(t) 1,

(iii) dZ(t, T) = - Z(t, T) ~*(t, u)dudW(t).

Under the measure P the processes [Z(t, 7"); 0 <_ t < T~ are martingales for all
0 <_ T <_ r. The spot rate process is square integrable and satisfies

(iv) r(t) = f ( 0 , t) ÷ ¢z*(u, t) a(u, s)dsdu + o'*(u, t)dW(u).

 M. Musiela/Stochastic Processesand their Applications 55 (1995) 227 251 231

Proof. Note that

d.f(t, u) = ~(t, u)dt + a*(t, u)dWo(t)

= a*(t, u) (;u )
a(t, s)ds - H(t) dt + a*(t, u)dWo(t)

= ~*(t, u) ~(t, s)ds dt + cr*(t, u)d W(t).

Therefore,

f T df(t, u)du

= a*(t, u) ~r(t, s)dsdudt + a*(t, u)dudW(t)

2 ~r(t, u)du + a*(t, u)dudW(t),

and, because

df(t, u)du - f ( t , t)dt = d f(t, u)du = - dlog P(t, T),

it follows that
( 1
dP(t, T) = P(t, T) dlog P(t, T) + 5 d ( l o g P ( . , T) )(t)
)
= P(t, T) t f(t,t)dt - 5 1 f,r a(t, u)du2dt - f, r ~*(t,u)dudW(t)

+ ~ f rtr(t,u)du2dt)=P(t,T)(r(t) at- ftra*(t,u)dudW(t)).

This finishes proofs of (i) and (ii). A straightforward application of the Ito formula plus
boundness of a(t, T) leads to (iii). Finally, (iv) follows from the integral representation
of (i). U3

1.2. Returns on risky assets

One of the building blocks of the arbitrage-flee pricing theory is the assumption
that traded assets can be classified as nonrisky and risky. Geometric Brownian
motion is frequently assumed as a simple model of the risky asset price dynamics. This
is justified by the belief that the corresponding return process satisfies the so-called
random walk hypothesis. However, there is growing practical and statistical evidence
that this assumption should be seen only as a first approximation.
For example, hedging over the life of an instrument is often done using forward
contracts on the underlying asset and requires parameters implied by the market. This
232 M. Musiela/ Stochastic Process'¢~ and their Applications 55 (1995) 227 251

may lead to some inconsistencies with the model assumptions particularly if one is
looking at more exotic products like average rate or compound options. Implied
volatilities will be in general different for different maturities of the forward contracts
used for hedging while geometric Brownian motion assumes constant volatility. Here
dealing room practice appears to be in contradiction with the theory. Statistical
evidence which contradicts the random walk hypothesis is provided by empirical
studies of returns on risky assets. For example, daily returns on US stocks indicate
abnormally small returns (from the model point of view) on Mondays. Corresponding
analysis of some European markets seems to confirm the existence of an analogous
phenomenon. Studies of currency exchange rates (Fama and Roll, 1971; McFarland
et al., 1982; So, 1987) have shown conclusively that they are abnormal. This abnormal-
ity has been seen for all major currency exchange rates, where the observed distribu-
tions have been invariably described as very peaked and thick tailed.
To be able to understand risks related to such anomalies we propose the following
nonhomogeneous model of the asset price dynamics:

dR(t) = R(t)(t~(t)dt + q*(t)dWo(t)), (1.5)

where R(t) denotes the asset price at time t, {Wo(t); 0 < t < r I is, as before, the
n-dimensional Ft-Brownian motion on ((2, F, Po) while the processes [ l+(t); 0 < t < r I
and {q(t); 0 < t < rl are adapted with values in N and N", respectively, R(0) is
a constant. The process, q is bounded on [0, r] x (~ while ll satisfies

Po(f~ ,/~(t)ldt < o c ) = 1.

Note that such a model allows to price options on the asset under stochastic interest
rates and with an arbitrary structure of market volatilities. It has also a potential to
explain the "day of the week" effect as well as the abnormality of currency exchange rates.
Consider now the economy E which consists of the money market account B, the
family [ P ( . , T); 0 < T < r) of pure discount bonds of all maturities and of the family
JR(', i); i t I} of risky assets which satisfy the hypotheses (1.5), of course with different
processes /.l(', i) and ~r(-, i) but with the same Brownian motion Wo. The set
I represents here an index set which could be finite or infinite. Assume that for each
i e l and each 0 < t _< r, the variable R(t, i) denotes the price at time t of the stock
i which pays no dividends. We already know that under the measure P there are no
arbitrage opportunities between B and the family {P(-, T); 0 _< T < r}. The following
assumption, valid throughout the paper, will help to eliminate any possible arbitrage
between B and the family {R(., i); i~l~ as well.

Assumption 1.2. For all 0 < t ~ r and all i t I

ll(t, i) = r(t) -- q*(t, i) H(t). (1.6)

It follows from (1.4) and (1.5) that
dR(t, i) = R(t, i)((/z(t, i) + q*(t, i) H(t))dt + r/*(t, i)dw(t)).
Hence we can formulate the following proposition.
 M. Musk, la/Stocha,stic Processes aml their Applicutions 55 1995~ 227 251 233

Proposition 1.2. Under the measure P we have

(i) dR(t, i) = R(t, i)(r(t)dt + q*(t, i ) ) d W ( t ) ) ,

(ii) dZ(t, i) = Z(t, i)~/*(t, i ) d W ( t ) ,

where Z(t, i) = R(t, i)/B(t), 0 <_ t < r, i ~ l .

Remark 1.1. Note that under the measure P the appreciation rate on the stock i is
equal to the spot rate for every i E I. In this sense the measure P is "risk-neutral".

Now let f o r 0 _ < t _ < r and T e [ 0 , r ] , i e l

M(t, T) = -
flf; ~*{s, u ) d u d W ( s ) ,

M(t, i) = f l q*(s, i)dW(s). (1.7)

Since W is a Brownian motion under P and the integrands are bounded adapted
processes, the process M mapping t into {M(t, T), T~[0, r], M ( t , i ) , i ~ l I, that is
M : t ~ M(t, • ) is a vector valued continuous martingale. Of course, we also have

( M ( ' , r), M ( . , i))(t) = - ;2f; a*(s, u)duq(s, i)ds, (1.8)

-, i), M(. ,jI)(t) = f ] q*(s, i)rl(s,j)ds,

which defines the "matrix" ( M ) of the joint quadratic variations of "co-ordinates'of

the martingale M. In particular, in terms of ( M ) one can calculate the quadratic
variation process for any linear combination of a finite number of "co-ordinates"
of M.
Also note that using the above compact notation we can write parts (ii) and (iii) of
Propositions 1.1 and 1.2, respectively, in the following form:

dZ(t, .) = Z ( t , - ) d M ( t , .). (1.9)

This shows that the process Z : t ~ Z(t, • ) is a vector valued martingale under P and in
practical terms it means that under Assumptions 1.1 and 1.2 there are no arbitrage
opportunities between the money market account B, the family [P(., T); 0 < T _< z~
of pure discount bonds and the family ~R(', i); i~ I I of stocks. From the pricing and
hedging point of view, however, the difficulty is that the measure P is not unique.
Let then ~ be the set of probability measures Q on (O, F) which are equivalent to
P (and hence to Po) and such that Z is a martingale under Q
234 M. Musiela/Stochastic Processesanti the#"Applications55 (1995) 227-251

Proposition 1.3. The process M defined in (1.7) is a continuous martingale, with the
quadratic variation process ( M ) given in (1.8), under any Q ~ ~. Moreover, (f there exist
T1 ..... TR and i k + 1 . . . . . i n such that.for every 0 <_ t <_ r the matrix

2;(0= a(t,u)du ..... a(t,u)du, tl(t, ik+l) ..... ~l(t, in)

is nonsingular and Z - 1 ( 0 is locally bounded then the process W defined in (1.4) is an

n-dimensional FrBrownian motion under any Q e ~ .

Proof. The quadratic variation under Q is the same as under P because P and Q are
equivalent. Also, Wis a semimartingale under Q a n d for any x belonging to [0, T] or I
Z(t, x) = Z(O, x)exp(M(t, x) - ½( M ( ' , x) )(t))
is a martingale. This implies that

M(t, x) = f l Z - l ( s , x)dZ(s, x), O~t~z

is a continuous local martingale with

( M ( ' , T))(t) = a(s,u)du ds

and

( M ( ' , i))(t) = Jo [~/(s, i)12ds

and hence, E(supo<t<~M(t, x)) 2 < ~ . Finally, let

U(t) f l Z* - l(s)dS(s),

where

N(t) = (M(t, 7"1) ..... M(t, Tk), M (t, ik+ 1)..... M(t, in))*

= fo Z* - 1 (s) d W(s).

It follows from the above that U(t) is a continuous local martingale with the tensor
quadratic variation ( U ) ( t ) = tl and hence (cf. Dellacherie and Meyer, 1980, p. 381)it
is a Brownian motion in R n []

Note that if the quadratic variation process ( M ) is deterministic then the martin-
gale M is Gaussian seen, for example, as a two parameter random field
{ M ( t , x ) ; O ~ t_<z, x E [ O , z ] u l } .
Consequently in this case the distribution of M, defined on the appropriate a-algebra
on C[O, z] × R t°'']u~ does not depend on the choice of Q ~ . This allows to formulate
the following proposition.
 M. Musieht/Stochastic Processes aml their Applications 55 (199~) .~ 7 .51 235

Proposition 1.4. If the quadratic variation process (M> given in (1.8) is deterministic
then the distribution of the process Z defined in (1.9) does not depend on the choice of the
arbitrage.free measure Q ~ ~.

Remark 1.2. Last two propositions are important from the point of view of pricing
and hedging contingent claims. Proposition 1.3 will find first applications already in
the next section where we will analyse some simple attainable claims. Proposition 1.4
in some sense solves the problem of incompleteness (under the assumption that (M>
is deterministic). It implies that the joint distribution of the collection of processes

E = IB('), P ( ' , T), R(-, i); TEl0, r], i ~ I ) (1.10}

does not depend on the choice of Q ~ . Consequently, calculations of expected

values, under Q, of functions depending on these processes can be carried out under
the measure P. In other words, the market model- is not complete but different
measures Q e ~ produce its identical copies.

1.3. Contingent claims

The T-maturity contingent claim is an Fr-measurable random variable C(T) which

represents a contract equivalent to a stochastic cash flow C(T) at time T. The claim
C(T) is called integrable if for every arbitrage free measure Q e ~,

Ee(IC(T)I/B(T)) < ~ .
We have the following examples.

Example 1.1. Constant contingent claims are integrable. Clearly it is sufficient to

show that for every O < T < r and every Q ~ Eo(I/B(T))< 3c. But
1/B(T) = Z(T, T) and the process Z(-, T) is a martingale under any Q for any T.
Consequently,

Eo(I/B(T) ) = EoZ(T, T) = EoZ(O, T) = P(O, T) < ~ .

Example 1.2. The time T price of the risky asset i is a T-maturity integrable claim,

Indeed EQ(R(T, i)/B(T)) = EcCZ(T, i) : R(O, i) < c~.

The question now is how to price (and hedge) contingent claims. When the market
model is complete it is well known that the unique time t price of the T-maturity
integrable claim C(T) is equal to

E(C(T) B(t)/B(T)I F,),

where the expectation is calculated with respect to the unique arbitrage-free measure.
When the market model is not complete as in our case the same statement remains
valid but only for attainable integrable claims (cf. Harrison and Pliska, 1981) and the
expectation is then calculated with respect to an arbitrary Q ~ . It turns out that
C(T) is attainable in our model if and only if C(T)/B(T) can be represented as
236 M. Musiela/Stochastic Processes aml theh" Applicatimls 55 (1995) 227-251

a stochastic integral with respect to the martingale Z or, due to Proposition 1.3, with
respect to the martingales M or W. Note that for the claim C(T) = l of a T-maturity
pure discount bond we can write

C(T)/B(T) = Z(T, T) = Z(t, T) + dZ(s, T),

t

which shows that this claim is attainable. Clearly, we also have

R(K i)/B(r) = Z(T, i) = Z(t, i) + dZ(s, i)

and hence, the claim C(T) = R(T, i) is attainable as well. The prices of claims are listed
in the proposition below where the expectation is calculated with respect to the
arbitrage-free measure P c ~.

Proposition1.5. The unique time t prices associated with the contingent claims: 15 at
time T and R(T) (t <_ T) are given by
E(B(t)/B(T)IFt) = P(t, T), E ( B ( t ) R ( T ) / B ( T ) I F , ) = g(t).

Remark 1.3. It seems that so far tw0 general approaches were developed to analyse
the term structure of interest rates. The first one (cf. Hull and White, 1990; Jamshidian,
1990: Artzner and Delbean, 1989) is based on a probabilistic model of the spot rate
from which forward rates and bond prices are calculated using the arbitrage argu-
ments. The second approach (cf. Heath et al., 1987; Musiela et al. 1993) assumes
a model for the instantaneous forward rates of all maturities from which one needs to
eliminate first any arbitrage possible. Then the spot rate as well as the forward rates
and bond prices can be analysed. Continuing discussions and comparisons exhibit
relative advantages and disadvantages of both methodologies and it remains an open
question which approach is more natural, better adapted to give answers to many
theoretical and practical problems or simply easier for the market to accept. Let us
point out here that Proposition 1.5 and namely the formula

establishes a one-to-one correspondence between the spot rate and the forward rates
under a relatively general set of assumptions concerning exclusively boundness of
volatilities. (These assumptions could be weakened even more and replaced by some
integrability conditions, we decided not to do it here for expositional clarity.) This
implies that both methods are equivalent in the above sense. It also shows that the spot
rate is the only parameter which "drives" the entire term structure of interest rates.

1.4. Forward contracts and./brward measures

A forward contract with maturity T o n a risky asset obligates its owner to purchase
the asset at time T for a fixed price, called the forward price. By convention, the
forward price is set to make the forward contract's value at initiation equal to zero.
 M. Musiela/Stochastic Processes and their Applications 55 f 1995) 227 251 237

The forward contract we consider here is written on the asset R which represents
either a stock from the family JR(-, i); iE I] or a zero coupon from the family {P(., T);
Tc [0, q].
Hence, under the measure P,
dR(t) = R(t)(r(tJdt + q*(t)dW(t)),
T
where q(t) equals q(t, i) or - ~, a(t, u) du for some i e I, T~ [0, ~]. As before we define
the martingale Z(t, R) = R(t)/B(t).
Now assume that the contract is initiated at time t < T and that F(t) denotes the
forward price. It is clear that the contract corresponds to an attainable integrable
claim with a cash flow of C(T) = R(T) - F(t) and hence from the above definition it
follows (cf. Jarrow, 1988) that
E((R(T) - F(t))B(t)/B(T)IFt) = O.
This and Proposition 1.5 give
F(t) = R(t)/P(t, T). (1.11)

Example 1.3. The forward price at time t in a T-maturity forward contract on

a discount bond with maturity ~ is equal to P(t, ~)/P(t, T).

From (1.11) it follows that F is a semimartingale.

Proposition 1.6. The semimartingale decomposition ofF under P is given by

dF(t) = F(t)( ftT ~(t, u)du + q(t))* ( frr ~(t, u)dudt + dW(t)).

Proof. The Ito formula and Propositions 1.1 and 1.2 allow to write
F(t) = R(t)(P(t, T))-1

= (R(t)/B(t))(P(t, T)/B(t)) ' = Z(t, R)(Z(t, T)) ~, (1.12)

d(Z(t, T)) -~ = - (Z(t, T)) 2dZ(/, T) + (Z(t, T ) ) - 3 d < Z ( . , T))(t)
~T ~T 2dr
=(Z(t,T)) 1 ~*(t,u)dudW(t)+(Z(t,T)) I ~(t,u)du

and hence,
dF(t) = (dZ(t, R))(Z(t, T))-~ + Z(t, R)d(Z(t, T))-
+ d<Z(., R),(Z(., T)) 1)(t) = F(t)~l*(t)dW(t)

+F(t)(ftTa*(t,u)dudW(t)+lfTa(t,u)du 2dt)

+ F(t)q*(t) cr(t, u)dudt

= F(t) (y/ a(t, u)du + rl(t) )*(// er(t, u)dudt + dW(t) . ) F5

238 M. Musiela/Stochastic Processesand their Applications 55 (1995) 227-251

Remark 1.4. Note that if

q(t) = - a(t, u)du,

then F ( t ) = F ( 0 ) = R(0). In particular the forward price at time t in a T-maturity

forward contract on a T-maturity pure discount bond is equal to 15. This simple
observation seems to support an intuitively clear more general belief that by looking
into forward contracts we in some sense eliminate, or rather control through the
volatility, risks related to fluctuations of interest rates. From this point there is only
a small step to another Girsanov transformation developed by E1 Karoui and Rochet
(1989) to price options on coupon bonds and then successfully used by E1 Karoui et al.
(1991) to value other claims that are contingent on interest rate dependent assets. The
idea is given below.
Let for all A e FT

Pr(A) = fA (P(O, T ) B ( T ) ) - ' dP. (1.13)

It was shown in Proposition 1.5 that E ( B ( T ) ) - I = P(0, T). Hence (1.13) defines
a probability measure on (f2, FT). We call it here the forward measure. Note that the
local density E((P(0, T)B(T))- 1 IF t) can be represented as follows:

E((P(O, T)B(T))- 11 F,) = (P(O, T)B(t))- ~E(B(t)/B(T)I Ft)

= (P(O, T)B(t))-1P(t, T) = (Z(0, T ) ) - ' Z ( t , T).

This together with Proposition 1.1 (iii) implies that

(P(O, T ) B ( T ) ) - ' = exp(M(T, T) - ½(M(., T))(T)),

where the martingale M(-, T) is defined in (1.7). Consequently, the Girsanov theorem
asserts that the process { W(t, T); 0 _< t _< T}, where

IV(t, T) = W(t) + a(u, s)dsdu (1.14)

is an Ft-Brownian motion under Pr- This together with Proposition 1.6 lead to the
following proposition.

Proposition 1.7. The process F is a martingale under Pr.

Remark 1.5. Concept of the forward measure is relatively new and therefore probably
not very well known to a wider audience. Let us try to explain it in a more intuitive
way. An "ordinary" arbitrage-free measure makes discounted (hence interest rate
influence) prices look like martingales. Speaking in general terms we place ourselves at
time 0 and looking into the future we construct a distribution under which present
values of securities, in terms of the dollar value at time O, behave like random walks.
The forward measure Pr produces the same effect but in terms of the dollar value at
time T. From the point of view of pricing T-maturity cash flows the measure
 M. Musiela/Stochastic Processes and their Applications 55 (1995) 227-251 239

P r integrates (in the mathematical and ordinary sense) risks related to interest rate
movements inside of the corresponding volatility parameter. Finally, note that
P0 = P, of course, and that PT = P for all 0 < T < r if the spot rate is deterministic,
because in this case
P(O, T ) = E(B(T) -1) = B(T) 1

and hence, the density between PT and P equals I.

1.5, Options involving finite number o f assets

Throughout this section we assume that the vectors a(t, T), q(t,i), t~[0, r],
T~ [0, r], i ~ I which describe the structure of volatilities in our economy are determin-
istic. Let us remind (cf. Remark 1.2) that under this assumption the distribution of the
collection of process E does not depend on the choice of the arbitrage-free measure
Q~. Now select a finite number of pure discount bonds with maturities say
T1 ..... Tte [0, z] and a finite number of stocks say it + 1..... i,, ~ I from (1. I 0) and form
the model

E(m) = {B(.), P(" T~),j = 1 .... , I, R ( ' , ij),j = I + 1 . . . . . m I,

which consists only of a finite number of risky assets. We will be interested here in the
problem of pricing and hedging contingent claims concerning exclusively assets of the
economy E(m). Note that the distribution of E(m) also does not depend on the choice
of Q ~ ~ , i.e. for each P, Q ~ ~,
P E ( m ) - l = Q E ( m ) - 1.

This implies that on the canonical probability space of E(m) i.e.

(C([0, r] "+ 1), B(C([-O, r] "+ ~)), PE(m)-1)

with the filtration Ft = F ( ~"~ generated by the co-ordinate process the set of arbitr-
age-free measures PE(m)-~ is a singleton. Consequently (cf. Harrison and Pliska,
1983) the market model is complete and hence any integrable T-maturity claim C(T)
defined on the above probability space is attainable. Moreover, the time t price of
C(T) is equal to

V(t) = E(C(T)B(t)/B(T)[ F,). (1.15)

Let us now consider several examples of claims C(T), which correspond to standard
or more exotic options, and show how to use previously developed ideas to price
them.

1.5.1. Options on coupon bonds

Consider a European call option with maturity T and exercise price K on the cash
flows C1 . . . . . Ct occurring at times T1 . . . . . Tt. One could think here of an option on
a coupon bond where the nonzero Cj's are the coupon payments except for the last
one which represents the coupon plus the face value. One could also think of
240 M. Musiela/Stochastic Processes and their Applications 55 (1995) 227 251

a portfolio of bonds or in general of any cash flows, positive or negative. It is clear that
the time Tvalue of this cash flows, expressed in terms of the Tfmaturity pure discount
bonds, is equal to
l
~" CjP(T, Tj)
j-1

and hence the European call option corresponds to the claim

C(T) = (Y.J=, CjP(T, T i ) - K) +.

Now let for j = l ..... l

r(t,j) = P(t, Tj)/P(t, T)

denote the time t price of a T-maturity forward contract on the Tj-maturity pure
discount bond (cf. Example 1.3). From the Propositions 1.6 and 1.7 it follows that

df(t,j) = - F(t,j) ;) a*(t, u)dudW(t, r),

where the process W(', T) defined in (1.14) is a Brownian motion under the forward
measure P r and consequently,

P(T, Ts)= F(T,j)

= F(t,j)exp - a*(s, u)dudW(s, T)

12frffJa(s,u)duZds).
But we also have

Er(C(T)[F,)E(B(t)/B(T)IF,) = E(C(T)B(t)/B(T)IF,),
where E T stands for the expectation under the measure PT. Using Proposition 1.5,
(l.15) and the above it follows that

V(t) = P(t, T)ET((EJ=, C~F(T,j) - K)+ I F,)

= P(t, T ) E r ( ( Z ~ : , C j F ( t , j ) e x p ( - f f ffJa*(s,u)dudW(s, T)

Now define for j = 1 ... /

Xi= ~*(s, u)dudW(s, T)

 M. Musiela/Stochastic Processes and their Applications 55 (1995) 227 251 241

and note that the P r - c o n d i t i o n a l distribution of the vector X = (X1 . . . . . Xt)* given
F, is N(0, F ' F ) where F = [71 . . . . . 7t] is a k × l matrix such that

F*F = EXX* =
((St; a*(s, u)du a(s, u ) d u d s
))
.

This implies that we can write the following representation of the price

V(t) = P(t, T) f~k ( Elj = , C j F ( t , j ) e x p ( - 7*x - ~l'/jI z) - K)+ ~p(x)dx,

where q~(x) = (2rr)-~/2 exp( - ½lxl 2) or equivalently that

t ~C~P(t, Tj)~p(x + 7~) - KP(t, T)q~(x))+dx.

V(t) = f~k (X j=

Note that if in addition a(s, u) dose not depend on s i.e. a(s, u) = a(u) then we can write

F*F = (T- t) ((f: a*(u)du f a(u)du )) = (T- t)B*B,

where B = [bl . . . . . bt] is a k × I matrix. In particular one could take bj = ~rrj a(u)du.
Using this notation we can finally write

V(t) = ( ~=tCjP(t, Tj)<p(x + ( T f)l/2bj) KP(t, T)~o(x)) + dx. (1.16)

It is perhaps interesting to note that the function

~(t, p~ . . . . . Pt) = f ~ (Y~= t pj~o(x + ( T - t)L'z hj) - K~o(x)) +dx

is the unique solution to the C a u c h y problem

+ ½Z[~, Z~=t b*bjp~p~p~?.pj = 0 on [0, T] x ~t,

~9(T, Pl . . . . . Pl) = (Y~- 1 P2 - K)+ P2 e ~

and that

V(t) = P(t, T)8(t, C~F(t, 1) . . . . . CIF(t, 1)).

Interpretation of the parameters is also obvious. One can show that B*B represents
the matrix of volatilities of returns on forward prices of pure discount bonds with
maturities Tj, / = 1 . . . . . I.
Let us now consider some particular cases. Assume first C~ = 1,
C2 = C3 . . . . . Ct = 0. Then formula (1.16) reduces to

V(t) = J'~ (P(t, Tj)~p(x + ( T - - t)l:zhl) -- KP(t, Tj~p(x)) + dx

242 M. Musiela / Stochastic Processes and their Applications 55 (1995) 227- 251

and it is easy to see that the integral is positive if and only if

b'~x/I b, I <- (log(P(t, T1)/KP(t, T)) - ½ ( T - t)lb212)/(T - t)l/2lb, [.

This allows to write that

V(t) = P(t, T 1)N(h) - KP(t, T)N(h - ( T - t)i/2[b ~I), (1.17)

where

h = (log(P(t, T1)/KP(t, T)) + ½(T - t)Ibl [2)/(T - t)I/2[bl I

and one can recognize the Biack-Scholes formula applied to the forward price as
obtained in (Heath et al. 1992).
Assume next k -- 1 and C s >_ 0, j -- 1 . . . . . I. Then it is not difficult to see that the
equation

Z~: 1CsP(t, Ts)q~(x + (T - t)l/Zbj) = KP(t, T)cp(x)

has a unique solution in x which we call s. In terms of s we can write

V(t) = YJ= ~CsP(t, Ts)N(s + ( T - t)I/2bs) - KP(t, T)N(s)

and hence (1.16) reduces to the formula derived by E1 Karoui and Rochet (1989) and
Jamshidian (1990).
Finally assume k = 2, C~ > 0 and Cz < 0. Then (1.16) can be expressed in terms of
the bivariate normal distribution and applied to price options on spreads between
forward rates.

1.5.2. Options on baskets qf stocks

Consider now a European call option with maturity T and exercise price K on the
basket {Cs, j = l + 1 . . . . . m} of stocks R ( ' , is). Clearly this option corresponds to the
claim

C(T) = (Zs"_-~+ ~CsR(T, is) - K) +.

Repeating arguments used in Section 1.5.1 one can verify that

dF(t,j) = F(t,j) a(t, u)du + q(t, is) dW(t, T),

where f o r j = l + 1. . . . . m

F(t,j) = R(t, is)/P(t, T).

Let f o r j = l + 1. . . . . m

£(t,j) = a(t, u) du + q(t, is),

then we have

R(t, is) = F(T,j) = F ( t , j ) e x p 2*(s,j)dW(s, T) - -~ ])~(s,j)12 ds .

 M. Musieht / Stochastic Processes and their Applications 55 (1995) 227- 251 243

The P r - c o n d i t i o n a l distribution of the vector Y = (Xt+l . . . . . X")* given F,, where

X~ = 2 * ( s , j ) d W ( s , T), j = I + 1..... m

is N(0, A ' A ) with a k x I matrix A = [6~ . . . . . 61] such that

A*A = E Y Y * = 2(s, i)2(s,j)ds.

Consequently, we can write the following representation of the price

Vit) = E I C ( T ) B ( t ) / B ( r ) l F,)

f~ ( 2 "j = / + l CiR(t, ij) q~(x 6i) - KP(t, T)~o(x)) + dx.

k

Again if 2(t,j) does not depend on t i.e. )dt, j) = 2(j) then we can write

A*A = E Y Y * = ( T - t)(2*(i)2(j)) = ( T - t)A*A,

where A = [bl+ ~. . . . . bin] is a k × l matrix. In particular one could take bj = 2(j).

Finally, we can represent the option price as follows

V(t) = ;~k (x"j=t+lCjR(t, ij)q~(x - (T - t)l/2bj) - KP(t, T)qqx)) + dx.

Interpretation of the parameters remains intuitive. The B * B matrix represents volatil-

ities of returns on forward prices of stocks. Moreover, the transformation x ~ - x
leads to

V(t) = ( ~=1+ 1CjR(t, i~)tp(x + ( T t)l/2bj) KP(t, T)to(x)) + dx. (1.18)

fRk X" -- --
and allows to c o m p a r e formulae (1.16) and (1.18). As a consequence all particular
cases discussed in Section 1.5.1 can be restated here with R(t, ij) substituting P(t, Tj).
For example if C1+ ~ = 1 and C1+ 2 . . . . . C,, = 0 we obtain the formula derived in
Jarrow (1988).

1.5.3. Multiple options

Consider a basket of E u r o p e a n call options with maturity T and exercise prices
Kj on stocks R ( . , ij),j = I + 1 . . . . . m. A multiple option is the option which gives the
right to exercise in exactly one of the stocks of the buyers choice. Clearly, the holder
will exercise in the stock which gives the maximal profit and consequently the multiple
option corresponds to the claim

C ( T ) = max(R(T, it+ 1) - Kl+ 1 . . . . . R ( T , i,,) -- K " , O)

= m a x l R ( T , it) -- K~)+; I + 1 <_j <_ m I.

244 M. Musiela/Stoehastic Processes and their Appfications 55 (1995) 227 251

Using methods and notation of Section 1.5.2 it is not difficult to see that the price V(t)
is given by

V(t) = f~k max(R(t, ij)cp(x + ( T - t)l/Zbj) -- KiP(t, T)~p(x))+dx.

For l + 1 = m we get the Black Scholes formula of course.

Remark 1.6. One could think that Eqs. (1.16)-(1.19) which in principle involve
integration over the space [~k will be very difficult to implement. In fact this is not true.
The common denominator in examples of options analysed in Sections 1.5.1-1.5.3. is
that in practice k will always be less than or equal to 2. This statement is based on
a statistical analysis of the matrix of volatilities of forward prices. It is remarkable that
in almost all cases analysed by us we found consistently only two significant eigen-
values.

2. Foreign economy

In this part of the paper we show how to expand our model into a larger economy
in which the foreign money market account, foreign bonds and other foreign securities
can be viewed as "domestic" assets. Our approach is based on the Amin and Jarrow
(1989) model of an arbitrage-free international economy. We use the same notation
and write the subscript f on the quantities defined in the first part of the paper to
indicate that they represent the corresponding quantities in the foreign economy.

2.1. Foreign term structure

We assume that the foreign instantaneous forward rate fi(t, T) is given by

dfi(t, T) = zcf(t, T)dt + a*(t, T)dWo(t), (2.1)

where Wo is the same as before while the processes {~f(t, T); 0 < t < T < r} and
l~rr(t, T); 0 < t < T < r I are bounded on [0, r ] 2 x ~2 adapted with values in I~ and I~",
respectively.
Denominated in foreign currency are the time t price Pf(t, T) of a T-maturity
foreign pure discount bond and the foreign accumulation factor Bf(t). Of course, we
have

P f ( t , T ) = e x p ( - ~r.ff(t,u)du)

and

of,,, exp( rf,u, du)

 M. Musieht/Stochastic Processes and their Applications 55 ~1995J 227 251 245

where rf(t) =ff(t, t) stands for the foreign instantaneous riskless rate of interest at
time t.
The exchange rate S(t) of the foreign currency, denominated in the domestic
currency per unit of the foreign currency, establishes the link between the two
economies. We assume that for 0 < t _<

dS(t) : S ( t ) ( f l ( t ) d t + v * ( t ) d W o ( t ) ) , (2.2)

where Wo is still the same while the processes {fl(t); 0 < t _< r~j and Iv(t); 0 < t <_ rl
are adapted with values in N and N", respectively, S(0) is a constant, the process v is
bounded on [0, r] x f2 and the process fl satisfies the condition

Po(f]lfl(t)[dt< , m ) = 1.

Note that for each 0 _< T < ~, Pr(t, T)S(t) represents the dollar value of a foreign pure
discount bond. In this sense the family of processes

[Pf(', T ) S ( ' ) ; 0 ~ T < ~) (2.3)

can be viewed as a new family of assets in the domestic economy. Also the dollar value
of the foreign money market account B d t ) S ( t ) can be interpreted as a new asset.
However, by bringing new assets into our economy we create new possibilities for
arbitrage. To eliminate them we proceed as before. That is, first we identify a condi-
tion under which the process

Zf(t) = Br(t)S(t)/B(t) (2.4)

is a P-martingale (the measure P is the one defined in (1.3)). A straightforward

application of the Ito formula to Zf(t) leads to the following differential representation

d Z d t ) = Zf(t)((rf(t) - r(t) + fl(t) + ,9*(t)H(t))dt + ~q*(t)dW(t)),

where the process W defined in Eq. 1.4 is Brownian motion under P. This invites to
formulate the following assumption.

Assumption 2.1. For all 0 _< T _< T

fl(t) = r(t) - rdt) - ~9*(t)H(t). (2.5)

The case of the family (2.3) is more involved but it can be analysed in exactly the same
way. We have, as in Proposition 1.1,

dPr(t, T) = Pdt, T ) ( d l o g P d t , T) + ½d(logPr( ", T ) ) ( t ) )

and

dlog Pr(t, T) = - d (iT) .l~(t, u)du = rdt)dt - dfi(t, u)du.

246 M. Musiela/StochasticProcessesandtheh"Applications55 (1995)227-251
But using (2.1) it follows that

ftr (dff(t, u)du) = ( ~tr ~tf(t,u)du)dt + ( f r tr'(t, u)du)dW(t)

and hence that

d(logPf(', T))(t)= i r trf(t, u)du 2dt.

Consequently, we have

dPf(t, T) = ((
Pf(t, 7") rr(t) - ~f(t, u)du

+~ ftr~rf(t,u)du 2)dt-(ftrG*(t,u)du)dWo(t))
and therefore, together with (2.2) also

d£Pf(., 7"),S('))(t) = - Pf(t, 7")S(t) a~(t,u)du~9(t)dt.

All this leads to

d(Pf(t, T)S(t)) = Pf(t, T)S(t) rf(t) - o~f(t, u)du

1 r ~rf(t,u)du 2 +fl(t)- ftrtr*(t,u)du g(t))dt

+2If, /

Now it is sufficient to analyse the process

Zf(t, T)= Pf(t, T)S(t)/B(t). (2.6)

Clearly, we have

dZf(t, T) = (dPf(t, T)S(t))/B(t) - (Pf(t, T)S(t)r(t)/B(t)) dt

= Zf(t, T) rf(t) - r(t) - el(t, u)du

+5l ftra,(t,u)du + fl(t)-

1 2
ftra*(t,u)dm9(t)

+ (~9*(t) - f f a*(t,u)du)H(t))dt

+(8*(t)-- ftra*(t,u)du)(dWo(t)-H(t)dt))
 M. Musiela/Stochastit"Processesand theirApplications55 ( 1995) 227-25I 247

=Zf(t,T)((- ff ~f(t,u)du+~ ff tTf(t,u)du 2

- ff ~,(t,u)du~(t)- ftr~*(t,u)duH(t)]dt

+ (~9*(t) - ftra~'(t,u)du)dW(t))
and hence we need the following assumption.

Assumption 2.2. For all 0 < t, T < z

:q(t, T) = o*(t, T) ~rf(t, s)ds - H(t) - ~(t) . (2.7)

Note that assumptions (2.5) and (2.7) are consistent with assumptions (1.6) and (1.2),
respectively. It seems intuitively clear that one would need to adjust the model only for
the exchange rate risk.
Let us summarize the above observations.

Proposition 2.1. Under the measure P the processes Zf(') and Zf(', T), 0 < T <_z,
defined in (2.4) and (2.6) are martingales. Moreover, we have
(i) d Z d t ) -- Zf(t)~9*(t)d W(t),

(ii) dZdt, T ) = Z f ( t , T ) ( 9 * ( t ) - f f t r * ( t , u ) d u ) d W ( t ) .

We finish this section with the definitions of martingales

Mf(t) = f ' 9*(s)dW(s) ~2.8)

Jo
and

Mf(t, T)= f i (~q*(s)- f f tr*(s,u)du)dW(s), (2.9)

where0_<t<zand0< T_<z.

2.2. Foreign risk)' assets

Assume now that we are interested in the family [ Rr(', i); i e If} of foreign stocks or
other foreign assets prices of which are given by

dRf{t, i) = Rf(t, i)(pAt, i)dt + qf(t, i)dWo(t)).

As usual assume that the processes {~q(t, i); 0 _< t < z] and {qf(t, i); 0 < t < ~, Jell,
are adapted with values in ~ and ~", respectively. For every ie If qf(., i) is bounded
248 M. Musiela/ Stochastic Processes and their Applications 55 (1995) 227- 251

on [0, ~] x f2 while tlf(', i) satisfies

l'o(ffll,~{t,i)ldt<~c) = 1.
O f course for each i6 It, Rdt, i)S(t) is the dollar value of a foreign asset and again all
we need to do is to find conditions under which the processes

Zr(t, i) = Rr(t, i)S(t)/B(t) (2.10)

are martingales under P. This leads to the following assumption.

Assumption 2.3. For all 0 _< t _< z and all i s It,

pr(t, i) = rf(t) -- rl*(t, i)H(t) - q~'{t, i)o(t). (2.1 1)

It is a simple exercise to show the following proposition.

Proposition 2.2. Under the measure P.]br all i e If the process Zr( ", i) is a martin#ale and
dZf(t, i) = Zf(t, i)dMf(t, i), (2.12)

where

Mr(t, i) = f~ (r/*(s, i) + ~9*(s))dW(s). (2.13)

Remark 2.1. It m a y be interesting to note at this stage that from Propositions 2.1 (i)
and 2.2 we can deduce the following intuitively obvious statement; portfolios consist-
ing of continuously rebalanced equivalent short and long positions in two foreign
assets do not carry any currency exchange risk. This is because for all 0 < t < ~,

foZrt(s,i)dZf(s,i)- ~ Z ( ' ( s , T)dZf(s, T)

=
fo( q~(s, i) +
)
a~'(s, u)du dW(s)

does not depend on O. As it is seen from the formula above we think here of
discounted prices.

Let us r e m e m b e r that we treat foreign assets as new assets in the domestic economy.
Therefore, we can also define new "larger" vectors M and Z. For example, the new
vector M is the family of martingales formed from all martingales defined in 1.7, 2.8,
2.9 and 2.13. The new vector Z consists of martingales defined in 1.9, 2.4, 2.6 and 2.10.
This allows us to restate Propositions 1.3 and 1.4 with the matrix Z(t) build from
vectors of"volatilities" of domestic and foreign assets. In particular if the process ( M )
is deterministic then the joint distribution of the family of processes

E = {B(-), P ( . , T). R ( . , i), B r ( ' ) S ( ' ) , Pr{', T)S('), R(',.]),

S('); T s [-0, r], i 6 l , . j 6 l f } (2.14)
 M. Musiela/Stochastic Processe~am/their Al~plic,tirm~55 /19957 227 251 249

representing our larger economy does not depend on the choice of the arbitrage free
measure Q~@.
All results concerning forward contracts and forward measures remain valid as well.
We simply need to remember to change names of vectors of volatilities.

Example 2.1. The time t price of a T-maturity forward contract on a foreign discount
bond with maturity T1 is equal to

F(t) = Pdt, T1)S(t)/P(t, T). (2.15)

In particular if T = T~ then

F(t) = Pf(t, T)S(t)/P(t, T)

is the value of a forward contract on foreign currency. In general when T :~ T~ we can
write, using Propositions 1.6 and 2.1 the following decomposition of F under P

dF(t) = F(t) or(t, u)du + ~9(t)

-
f/ ~f(t, u)du )'(f/ ~tt, u)dudt + dWtt) ) 12.16)

which again can be interpreted intuitively (take T = T~, and ¢r = err).

2.3. Options on.[breign assets

As in Section 1.5 we assume here throughout that the vectors a(t, T), ~l(t, i), af(t, T),
rh(t,j), O(t), T~ [0, ~], i6 l,,je If which define the term structure of volatilities in our
economy are deterministic. This implies that finite families containing B(') and other
processes from (2.14) can be viewed as complete models of economies spanned by
finite number of assets. We consider here several examples of such models in the
context of option pricing.

2.3.1. Options on.[breign discount bonds

To price a European call option with maturity T and exercise price K on a foreign
discount bond with maturity T1 we consider a finite dimensional "'projection" of
economy (2.14t which contains B(.), P(., T), B d ' ) S ( ' ) , P(', T~ )S(.). The T-maturity
forward contract on the asset P(., T1)S('t is priced at time t by Eq. (2.15). The
semimartingale decomposition of its value F(t) is given in (2.16). It allows us to
identify the new volatility parameter

;/' a(t, utdu + O{t) -

f) 6f(t, u)du

and the new 7~ vector as any vector such that

] ; ' 1 ] 2 ~--- o(s,u)du + O ( s ) - ar(s,u)du ds.

250 M. Musiela/Stochastic Processesattd their Applications 55 (1995) 227-251

If in addition the integrand in the above formula is constant in s and equal to 0.2 then
17112 = ( T - t ) a 2 and we can derive from (1.16) or (1.17) the following pricing formula.

V(t) = Pf(t, T1)S(t)N(h) - KP(t, T)N(h - ( T - t)1/2o'), (2.17)

where

h = (log(Pt(t, T1 )S(t)/KP(t, T)) + ½ ( T - t ) a 2 ) / ( T - t)1/26.

Interpretation of the p a r a m e t e r 0 .2 is the same as before i.e. 0.2 represents here
volatility of the T-maturity forward contract on the Tl-maturity foreign discount
bond.
Finally let us analyse some particular cases. If for all t, T e [0, z], 0.(t, T) = 0.f(t, T),
~9(t) = 0 and S(0) = 1 then Eq. (2.17) reduces to (1.17) and hence can be used to price
options on "domestic" discount bonds. If T = T1 then (2.17) reduces to the formula for
pricing options on foreign currency under stochastic interest rates as derived by Amin
and J a r r o w (1989).

2.3.2. Options on spread~ between the domestic and foreign discount functions
One way of thinking of a European call option with maturity T on the spread at
time T~ between the domestic and foreign discount functions is to think of an option
which corresponds to the contingent claim

C(T) = (Pf(T, T, )S(T) - CP(T, TI) - K) +,

where C and K are constants. To price such a claim we can follow arguments
presented in Section 1.5.1. The time t value V(t) of the option is given by

V(t) = [ " (Pr(t, Tt)S(t)cp(x + ( T - t)l/Zbl)

dak
- CP(t, Tl)cp(x + ( T - t)l/Zb2) - KP(t, T)~p(x)) + dx,
where ¢p(x)= ( 2 r t ) - k / Z e x p ( - ½lxl2), k _< 2, and the matrix (b* bj) represents the
matrix of returns on prices of the corresponding forward contracts i.e.
P(', TOS(" )/P(', T) and P(., TO~P(', T).

2.3.3. Options on .&reign stocks

It is a simple exercise to show that the price of a E u r o p e a n call option with maturity
T and exercise price K on the basket {Cj; j = 1 . . . . . m I of foreign stocks Rf(., ij) can
be obtained from Eq. (1.18) by substituting R(-, ij) with Rf( ", it)S(').

References
P. Artzner and F. Delbean, Term structure of interest rates: the Martingale approach, Adv. Appl. Math. 10
[1989) 95 129.
K.I. Amin and R.A. Jarrow, Pricing foreign currency options under stochastic interest rates, Cornell
University Working Paper (Ithaca, NY, 1989).
C. Dellacherie and P.A. Meyer, Probabilite et potential, Hermann [1980).
 M. Musieht/Stochastic Processes aml their Applications 55 (1995) 227 251 251

N. El Karoui, C. Lepage R. Myneni, N. Roseau and R. Viswanathan, Valuation and hedging with
markovian interest rates, Stanford University working paper (Stanford, CA, 1991).
N. E1 Karoui and J.C. Rochet, A pricing formula for options on coupon bonds, CREMAP working paper
(1989).
E. Fama and R. Roll, Parameter estimation for symmetric stable distributions, J. Amer. Statist. Assoc. 66
(1971) 331-338.
J.M. Harrison and D.M. Kreps, Martingales and arbitrage in multiperiod securities markets, J. Econom.
Theory 20 (1979) 381 408.
J.M. Harrison and S.R. Pliska, Martingales and stochastic integrals in the theory of continuous trading,
Stochastic Process Appl. 11 (3)tl981~ 215 260.
J. M. Harrison and S.R. Pliska, A Stochastic calculus model of continuous trading: complete markets,
Stochastic Process Appl. 15 (1983) 313 316.
D. Heath, R. Jarrow and A. Morton, Bond pricing and the term structure of interest rates: a new
methodology, Econometrica 60~1~ 11992) 77- 105.
J. Hull and A. White, Pricing interest rate derivative securities, The Review of Financial Studies 3 11990)
573 592.
F. Jamshidian, Bond and option evaluation in the gaussian interest rate model, Unpublished manuscript
(1990).
R.A. Jarrow, Option valuation of risky assets in a stochastic interest rate economy, Cornell University
working paper. (1988).
J. McFarland, P. Richardson and S. Sam, The distribution of foreign exchange price changes; trading day
effects and risk measurement, Journal of Finance 37 (19821 693 714.
M. Musiela, S. Turnbull and L. Wakeman, Interest rate risk management, The Review of Futures Markets
12 111 (1993) 221-261.
J. So, The distribution of foreign exchange price changes: trading day effects and risk measurement
a comment, Journal of Finance 42 (1987) 181 188.