5

Bessel Processes, Asian Options,

and Perpetuities
Mathematical Finance, Vol. 3, No.4
(October 1993), 349-375
(with Helyette Geman)

Abstract. Using Bessel processes, one can solve several open problems involving
the integral of an exponential of Brownian motion. This point will be illustrated
with three examples. The first one is a formula for the Laplace transform of an
Asian option which is "out of the money." The second example concerns volatility
misspecification in portfolio insurance strategies, when the stochastic volatility is
represented by the Hull and White model. The third one is the valuation of perpe-
tuities or annuities under stochastic interest rates within the Cox-Ingersoll-Ross
framework. Moreover, without using time changes or Bessel processes, but only
simple probabilistic methods, we obtain further results about Asian options: the
computation of the moments of all orders of an arithmetic average of geometric
Brownian motion; the property that, in contrast with most of what has been writ-
ten so far, the Asian option may be more expensive than the standard option
(e.g., options on currencies or oil spreads); and a simple, closed-form expression
of the Asian option price when the option is "in the money," thereby illuminating
the impact on the Asian option price of the revealed underlying asset price as
time goes by. This formula has an interesting resemblance with the Black-Scholes
formula, even though the comparison cannot be carried too far.

1. Introduction

Bessel processes possess two major features: first, besides the Ornstein-
Uhlenbeck processes, they are essentially the only diffusion processes in addi-
tion to Brownian motion (with drift), for which a relatively simple expression
for the transition probability is known; second, they appear naturally in a
number of interesting problems in finance and insurance. For instance, hyper-
geometric functions (which are related to Bessel processes) are used for the
pricing of options on zero coupon bonds in the Cox-Ingersoll-Ross general
equilibrium model of interest rates. Another key point is that the standard
hypothesis in most financial papers assumes stock price dynamics driven by
the exponential of a Brownian motion with drift, which is in turn (the square
of) a time-changed Bessel process. Moreover, the class of laws of squares of
Bessel processes is stable under the convolution operation, a property which
is not shared by the commonly used lognormal models.
We shall illustrate these points with three examples of theoretical and
practical importance: Asian options, portfolio insurance, and perpetuities or
annuities.

M. Yor, Exponential Functionals of Brownian Motion and Related Processes

© Springer-Verlag Berlin Heidelberg 2001
.

64 5. Bessel Processes, Asian Options, and Perpetuities

The paper is organized as follows: Section 2 recalls the definition of Bessel

processes, their main properties and their relation to exponentials of Brownian
motion. A first application is given in Section 3, namely the study of the distri-
bution and the moments of the asset average price, together with a compari-
son between the Asian and standard option prices and an expression for (the
Laplace transform of) the Asian option price. Section 4 explains why Bessel
processes can help solve the problem of volatility misspecification in the classi-
cal strategies of portfolio insurance which have been extensively implemented
over the last decade. Section 5 addresses the pricing of perpetuities and
annuities in a stochastic interest rates environment, within the framework
of the Cox-Ingersoll-Ross model. Some concluding comments are given in
Section 6.

2. Some Properties of Bessel Processes

In this section, we collect some general results about Bessel processes. Detailed
proofs of these results may be found in the references (Shiga and Watanabe
1973, Revuz and Yor 1991, Yor 1992b). For simplicity, we first look at squares
of Bessel processes, or "Bessel squared" processes; these we denote by BESQ
(a Bessel process BES is the square root of a BESQ).

Definition. For any 6 2': 0, the 6-dimensional Bessel squared process BESQ"
is a continuous diffusion process Pt taking its values in jR+ and satisfying the
stochastic differential equation

Po = a 2: 0,
where W t is a standard one-dimensional Brownian motion.
The real v = 6/2 - 1 is called the index of the process BESQ".
As a consequence of the stochastic differential equation satisfied by
(Pt, t 2: 0), the process

is a local martingale as soon as f belongs to C 2 ((0, +00)), where

and LV is the (martingale) infinitesimal generator associated with BESQ".

We denote by Q~ the distribution of the process BESQ" starting at a 2': 0;
this distribution is defined on the canonical space of continuous functions
C(jR+, jR+) on which we consider the coordinate process (R t , t 2': 0) which
.

5.2. Some Properties of Bessel Processes 65

is defined by Rt(f) = f(t) for every f E C(~+,~+) and the a-field

g'= a{ R t , t ?: O}.
The following important property of the Bessel squared processes was
obtained by Shiga and Watanabe (1973).

Proposition 2.1. For every 8,8', x, x' ?: 0, Q~ ® Q~', = Q~t~/ where P ® Q

denotes the distribution of the process (Xt + yt, t ?: 0), for X t and yt, two
independent processes with respective distributions P and Q.

This property permits, whether 8 is an integer or not, the reduction of

a number of problems involving BESQ6 to the case 8 = 1, where BESQl is
precisely the square of one-dimensional Brownian motion.
Bessel (squared) processes are, by definition, Markov processes. Their tran-
sition functions have the following form.
Proposition 2.2. For 8 > 0, the semigroup of BESQ6 has a density in y
equal to

qt6 (x, y) = 1 (Y)V/2 exp ((x

2t;:; + y)) Iv (v'xY)
- ----u- -t- (t > O,x,y > 0),
(2.1)

where Iv is the Bessel function with index v == 8/2 - 1.

The density of the semigroup of the Bessel process BES 6 can be obtained
from (2.1) by a straightforward change of variable and is found equal, for
8> 0, to

pt (x, y) = t1 (Y)V
;:; yexp [x2
-
+ y2] Iv (XY)
2t t (t > O,x,y > 0). (2.2)

A key result for our applications in the following sections is due to Williams
(1974),<1)
Proposition 2.3. The exponential of Brownian motion with drift is a time-
changed Bessel process; more specifically,

exp(W(t) + vt) = R(v) (lot exp 2(W(s) + VS)dS) (t ?: 0),

where (R(v)(u),u?: 0) is a Bessel process with index v.

Random time changes are of constant use in the study of one-dimensional

diffusions and are likely to be more widely used in finance. The reader
interested in the issue of "stochastic clocks" can find a number of impor-
tant examples in Ito and McKean (1965). Moreover, one can observe that

(1)An earlier reference is the paper of Lamperti (1972) quoted in paper [8] in this
book; see also Postscript #1.
.

66 5. Bessel Processes, Asian Options, and Perpetuities

this issue already arises in adjustments made by practitioners to account for

different temporal rates of trading. The "business time scale" and the "trans-
action clock" which are often introduced in the study of intraday prices involve
indeed stochastic time scale changes.
We can give two extensions of Proposition 2.3.
1. A similar result holds more generally for (exp(aW(t) + vt), t 2 0). Thanks
to the scaling property of Brownian motion, we can write

(with TV another Brownian motion, and corresponding R from Proposition 2.3)

so that

exp(aW(t) + vt) = R(v/a 2

) (a 2 1t exp2 (aW(u) + VU)dU). (2.3)

2. The exponential of Brownian motion with drift is a time-changed Bessel

squared process.
This property is straightforward since, from (2.3), we deduce

exp(aW(t) + vt) =
a W(t)
exp 2 ( "2 vt)
+2

= [R(2v/a 2 ) (: 1texp2GW(S)+~S)ds)r
More generally, it is possible to write the exponential of a Brownian motion
with drift as any given power of a Bessel process, up to a time change. How-
ever, the additivity property of squares of Bessel processes makes it most
efficient to use the power 2.

Proposition 2.4. Denote by p~v) the law on C(JR+, JR+) of the Bessel process
R(v) starting at a, by (Rt, t 2 0) the canonical process on C(JR+, JR+), and by
:9t = a(Rs, s :::; t) the canonical filtration. Then, the following relationship
holds (from Girsanov's theorem):

p~v) Ig; = ( ~t ) v exp ( _ ~21t ~;). p~O) Ig;.

We shall denote by E(v) the expectation relative to p}v!

As a consequence of Proposition 2.4, we obtain

.

5.3. Asian Options Revisited 67

= V2,,\ + v 2 and Tt = J~ dsj(Rs)2.

°
Lemma 2.1. For any v ~ 0,"\ ~ 0, define JL
Then for every t ~ and a E lK we have

Proposition 2.5. Using the same notations as in the definition at the begin-
ning of this section, the following relationship holds for any () > 0, a ~ 0,
JL > () - 1:

Ea
(J.t) _1_
(
(R )
t
_ _1_
)2f) - r(()) 1
1
h
f)-l
(1
_
h)
J.t-f) ()f)
~
2t exp
( -a2t h) dh.

where r denotes the gamma function.

The proof of Proposition 2.5 may be found for instance in Yor (1992a,
Chapter 6).

Proposition 2.6. (Yor 1980). For every v E lK, a> 0, x> 0, t > 0,

where, on the right-hand side, (f jg)(y) stands for f(y)jg(y).

3. Asian Options Revisited

3.1. The General Problem

Even though a number of results have been established on the subject of
Asian options (defined below), so far no one has produced an exact formula for
their values. Nevertheless, Asian options are popular in the financial commu-
nity, because they might be superior to standard options for several reasons.
First, as we will see later, they are often cheaper than the equivalent classical
European option (or are thought to be so even when it is not the case). Second,
the fact that the option is based on an average price is an attractive feature
for thinly traded assets and commodities, e.g., gold or crude oil, where price
manipulations near the option expiration date are possible; this is the case for
index options in newly opened markets during the famous "triple-witching"
hour. Today, average options represent an enormous percentage of options on
oil; some are written directly on oil prices, others on spreads between two
types of oil. In the same way, Asian options on average exchange rates are
extremely popular because they are less expensive for corporations which
need to hedge a series of risky positions on the foreign exchange incurring at
a steady rate over a period of time. Some options on domestic interest rates
.

68 5. Bessel Processes, Asian Options, and Perpetuities

(and options on interest rate swaps) exhibit this Asian feature when the base
rate is an arithmetic average of spot rates.
In this paper, we shall address only the "fixed-strike" case, because the
"floating-strike" case, where the payoff depends on the relative values of the
arithmetic average and the underlying price at maturity, is less widely used in
practice. The exact pricing of fixed-strike Asian options is clearly a difficult
exercise because the distribution of an arithmetic average of stock prices is
unknown (and in particular not lognormal) when prices themselves are lognor-
mally distributed. The studies on this problem fall roughly into three groups.
The first approach is numerical: Kemna and Vorst (1990) use Monte Carlo
simulations with the corresponding geometric average option as a control vari-
able; Carverhill and Clewlow (1990) use the fast Fourier transform to calculate
the density of the sum of random variables as the convolution of individual
densities and then numerically integrate the payoff function against the den-
sity. Clearly, these methods, besides being computationally expensive, give no
information on the hedging portfolio.
A second approach, used by Ruttiens (1990) and Vorst (1992), consists of
modifying the geometric average option price. Although this method reduces
the amount of calculation, it may lead to significantly underpriced call options
and has the same shortcoming as the numerical one in terms of dynamic
hedging. Bouaziz, Bryis, and Crouhy (1991) provide an upper bound of the
error involved in their approximation of the law of the average.
A third approach, proposed by Levy (1992) and adopted by a number of
practitioners, assumes that the distribution of an arithmetic average is well
approximated-at least in some markets-by the lognormal distribution, and,
consequently, the problem is reduced to determining the necessary parameters.
This task is less complicated since the first two moments of the arithmetic
average are relatively simple to derive. In what follows, we shall give a closed-
form expression for all moments of the arithmetic average.

3.2. The Mathematical Setting

We take as given a complete probability space (0,:7, P) with a filtration

(S't)o<t<oo, which is right-continuous and such that Sib contains all the P-null
sets of ST. Following the seminal papers by Harrison and Kreps (1979) and
Harrison and Pliska (1981), we assume the existence of a "risk-neutral" prob-
ability measure Q (equivalent to P) under which the underlying stock price
dynamics are driven by

dS(t) = rS(t) dt + CfS(t) dWt .

Here, r is the instantaneous risk-free rate and Cf is the volatility. We suppose

that (S't) is the natural filtration of the Brownian motion Wt ; for simplicity,
we assume in this section that both rand Cf are constant over [to, T], the time
interval over which the average value of the stock is calculated, where T is
.

5.3. Asian Options Revisited 69

the maturity of the option. In what follows, we will use interchangeably the
notation Wt or W (t).
Consider the process (A(x), x ~ 0), where

A(x) = -1-
x - to
i to
X
S(u) duo

The payoff of the Asian option at maturity T is then

max[(A(T) - k), 0] = (A(T) - k)+

where k is the fixed-strike price of the option. By arbitrage arguments and
because we assumed the interest rate to be constant over the lifetime of the
option, the value at time t of the Asian call option is

We start with three simplifying steps.

Step 1.
-r(T-t)
Ct,T(k) = e-r(T-t) EQ[(A(T) - k)+ISf] = eT _ to Ct,T(k')

where

Ct,T(k') ~f EQ[(A(T) - k')+ ISf], A(T) = iT S(u) du, k' = k(T - to).
to

Step 2. We now show how to reduce the computation of this conditional

expectation to that of the deterministic function

where b E JR, a, s, 'Y E R

Consider the case to :::; t < T. (The case to > t of the "forward-start" Asian
option is solved in Appendix A). We have

using

A(T) = A(t) + S(t) io

r T- t
exp(Y(s)) ds,
.

70 5. Bessel Processes, Asian Options, and Perpetuities

with y(s) ~f aW(s) + (r - a 2/2)s, where W(s) = W(s + t) - W(t), s ~ 0, is

a new Brownian motion independent of Hence, Sr

where kif = (k' - A(t))/S(t). Consequently, Ct,T(k') = S(t)(<y,b(T - t, kif),

with b = r - a 2 /2.
From now on, we will drop the index Q in the expectation symbol.
Step 3. We show finally that the scaling property of Brownian motion
allows to replace a by 2 in the expression

Using the change of variable u = 4v / a 2 and the scaling property of Brownian

motion, we obtain

/"<y,b( ) = ~/"2,4b/<y2 (sa 2 'Ya 2 )

., s, 'Y a2" 4 ' 4 .

rl
Finally, defining

c(v) (x, q) = E [ (lax exp{2(W(u) + vu)} du - q

we can write

To summarize steps 1-3, we just expressed the desired Asian option price as

C (k) = e-r(T-t) (4S(t)) C(v) (h ) (3.1)

t,T T - ta a2 ,q

where

We must remark at this point that q may be negative if the observed

values of the underlying asset S over the time interval [to, t] are big enough
(this means we already know at time t that the Asian option will be "in the
money" at maturity T and that the put option is worthless).
.

5.3. Asian Options Revisited 71

3.3. First Results

We now compare the Asian option price to the standard option price. As
shown above, if very high values of S(t) have been observed on the interval
[to, tJ, the Asian option is already very valuable a long time before maturity.
So, for our comparison to be of interest, we assume that to = t = 0; that is, no
price of the underlying asset relevant to the average has been revealed when
we compare the Asian option and the standard option prices. The quantity k
represents the exercise price and is positive.

Proposition 3.1. The Asian call option price is smaller than the standard
option price when the underlying asset is a domestic stock; i.e., v> O.

Proof. This result can be proved in the general case, without the discretiza-
tion procedure used by Kemna and Vorst (1990), who were themselves looking
at a stock Asian option. Moreover, as observed by Turnbull and Wakeman
(1991), Kemna and Vorst did not emphasize the fact that the property does
not hold when the time to maturity of the call option is shorter than the
length of the averaging period.
From the equality

T1 Jro T1 Jor
T T
exp2(Ws + vs) ds - k = [exp2(Ws + vs) - k]ds

we deduce, thanks to Jensen's inequality,

The convexity of the exponential function implies that

{exp2(Ws + vs), s ?: O} is a submartingale as long as v = 2r/a 2 -1 is positive:
this is usually the case for the values of r and a observed in practice when the
underlying asset is a domestic stock, say r = 0, 1; a = 0,2; v = 4. Consequently,
{(exp2(Ws + vs) - k)+,s?: O} is also a submartingale and
E[(exp2(Ws + vs) - k)+] ::; E[(exp2(WT + vT) - k)+].

o
We can give a more precise result and show that the crucial element in
determining whether this property holds is the sign of v + 1 = 2r / a 2 , i.e.,
the sign of the drift r of the underlying asset. For currency Asian options, for
instance, the dynamics of the underlying asset are

dS
S = (r - y) dt + a dWt , where y is the foreign spot rate.
.

72 5. Bessel Processes, Asian Options, and Perpetuities

Consequently, the "risk-neutral" drift r - y may be negative. The same

observation holds for Asian options on oil spreads.
A more general version of Proposition 3.1 is the following.

Proposition 3.2. (i) If v + 1 2: 0, the Asian option price is smaller than

the standard option price for any exercise price k.
(ii) If v + 1 < 0, the Asian option price is greater than the standard option
price for all values of the exercise price k in a neighborhood of 0.

Proof. Using Ito's lemma, we can write

exp[2(W(s) + vs)]
= 1 + 210 8
exp[2(W(u) + vu)] dWu +210 8
(v+1) exp[2(W(u)+vu)] duo

Consequently, {exp[2(W(s) + vs)], s 2: o} is a submartingale if v + 1 2:

and is a supermartingale if v + 1 :s: 0. In particular, it is a martingale for
°
v = -1.

(i) When v + 1 2: 0, the submartingale property remains satisfied for

(exp 2(W(s) + vs) - k)+; hence, the result of Proposition 3.1 holds for
any value of k.
(ii) When v + 1 < 0,{exp2(W(s) + vs),s 2: O} is a strict supermartingale
and if we choose k = 0, i.e., if we look at a zero-exercise price option,2 we
obtain the inequality

E [~ loT exp 2(W(s) + vs) ds1> E(exp[2(W(T) + vT)]). (3.2)

We could have avoided the supermartingale argument because we can

make a simple explicit computation of both sides of (3.2). The left-hand side
is equal to (1 - exp( -b))/b, where b = -2(v + l)T, and, keeping in mind the
equality E[exp(2W(s))] = exp(2s), the right-hand side is equal to exp( -b).
Now, the elementary inequality (1 - e-b)/b > e- b concludes this second
proof of (3.2). Furthermore, for any k in the interval [0, {(I - exp( -b))/b} -
exp( -b)], the following inequality holds:

E [ (~ [ exp2(W(,) + v.,) d., - k) +1;> E[(exp2(W(J') + vT) - WI·

(3.3)

2 Zepos were first introduced by Garman and Kohlhagen (1983).

.

5.3. Asian Options Revisited 73

To prove (3.3), we first use Jensen's inequality to obtain

E [ (~ [ exp 2(W (,) + vs) d, - k f1

:> H(~ [exP2(W(S) + Vs) d')]- kf
From the simple computation relative to the case k = 0, the right-hand
side is equal to ((1- exp( -b)) /b - k)+, which, for k in the prescribed interval,
is greater than (or equal to)

exp( -b) = E[exp 2(W(T) + vT)] 2: E[(exp 2(W(T) + vT) - k)+],

the last inequality being obvious.

This result does not hold for all k, at least for v E [- 2, -1] and probably
for every v < -1. 0

If we set

A~V) = lot exp[2(Ws + vs)] ds,

the distribution of the pair (A~v), Wt + vt) was obtained by Yor (1992b).
First, it is useful to observe that, due to Girsanov's theorem relating the laws
of (Ws + vs; s :::; t) and (Ws; s :::; t), the quantity P(A}v) E du I W t + vt = x)
depends on t, x, u, but not on v. Consequently, we can take v = 0, in which
case

P(A~O) E du, W t E dx) = ~u exp [-~(1

2u
+ e2X )] ¢ex/u(t) du dx, (3.4)

where

and

(We use the French notation ch, sh, and th for, respectively, the hyperbolic
cosine, sine, and tangent functions.)
.

74 5. Bessel Processes, Asian Options, and Perpetuities

Given the complexity of the exact joint distribution of the pair {average,
underlying asset}, we shall look at the moments of the average A~v), which
give some partial information about its law. (However, the valuation of an
Asian option having a payoff max(A~) - ST, 0) would require the use of the
joint distribution of (A~);ST).)
The first two moments of A~v) are easily computed by integration. Higher
order moments have been computed by Dufresne (1989, 1990) using Ito's
lemma, time reversal, and a recurrence argument. A different method consists
of introducing the Laplace transform in time; see Geman and Yor (1992).
Using either procedure, one obtains the formula(1)

where

dj!3) = 2n II [(;3 + j)2 - (;3 + i)2rl.

iij
O~i~n

For,\ = 2, this expression gives the different moments of A~v). In particular,

for n = 1, we obtain the first-order moment of A~v):

E(A~v)) = (1 ) [exp(2(// + l)t) - 1], (3.5)

2//+1

an expression which could have been obtained by a direct calculation, inter-

changing the order of integration and expectation, thanks to Fubini's theorem.
The first- and second-order moments would give the parameters of the
distribution of A~v) if it were lognormal, which is the approximation made by
some practitioners, although there is clearly no mathematical justification for
this assumption. A possibly related fact is that practitioners are very inter-
ested in knowing the extent to which a distribution can be characterized by
its moments. A relevant mathematical result is the following (see for instance
Feller 1964, vol. 2, p. 224).

Carleman's Criterion. If a random variable XsatisjieSL,nEN(m2n)-1/2n =00,

where m2n = E(X 2n ), then, its distribution is determined by its moments.

The criterion does not apply to lognormal distributions nor to the average
of geometric Brownian motion.

(1) See Postscript #3.

.

5.3. Asian Options Revisited 75

Proposition 3.3. For any v E IR and t > 0, the quantity

L E[(A~v))2ntl/2n
00

Mt(v) ~f
n=l

is finite.

Proof. (a) We first reduce the proofto the case v = O. If v ~ 0, A~v) ~ A~O)
so Mt(v)-S Mt(O). On the other hand, if v -S 0,

A (v) > e 2vt A (0) and M(v) < e 2vt M(O).

t - t t - t

(b) To show that Mt(O) < 00, we use Bougerol's identity (1983). For any fixed
(law) (0)
t > 0, shBt = 'YAt' where we denote At by At and 'Y represents
a Brownian motion independent of At. From the scaling property of
Brownian motion, we deduce

Hence, in order to show that Mt(O) < 00, it remains to prove that the
series with general term en = an/b n , where an = [Ebtn)p/2n and bn =
[E(sh(Bt) )4n]1/2n, is convergent. We now observe that

hence, using Stirling's formula, we obtain

an ~ (4n/e).
n-+oo

On the other hand, we claim that

(3.6)

Assuming (3.6) for now, we deduce that, for n large enough, there exists
E E JO, 1[ such that

4n])1/2n
bn ~ (1- E) ( E [ (~exP(Bt)) ~ 1 ~ E exp(4nt).
.

76 5. Bessel Processes, Asian Options, and Perpetuities

Consequently, On = an/bn :::; Cnexp( -4nt) and the series (On)n~O

converges. The equivalence (3.6) is derived from the following chain of
equivalences:

(E[('h(B,)),n1(n n':= (E [('h( B,)) '" 1(IB'IO>>)]) 'i'n

n'O= (B [G exp(IB,Dr 1(IB&1)]) 'I"~
G
.'Ow (E [ exp(IB,D) 4n]) 'i'n
o
3.4. Asian Option Valuation
Let us return to (3.1), obtained at the end of Section 3.2. What remains
to be computed is the quantity C(v)(h, q) d~ E[(At) - q)+], where At) =
Ioh exp[2(Ws + vs)] ds.
(a) When q :::; 0, the calculation is straightforward:

C(v)(h, q) = E(At)) - q.

Using (3.5) we obtain

C(v)(h, q) = {2(V ~ 1) [exp(2(v + l)h) - I]} - q.

This quantity, plugged into (3.1) with the corresponding values of v, h, and q

1:
gives a closed-form expression of the Asian option price, namely,

Ct,T(k) = S(t) (1 ~;-~(:~t)) - e-(T-t) (k - T ~ to S(u) dU)' (3.7)

This expression deserves some comment.

(1) It has an interesting resemblance to the Black and Scholes (1973) formula
(it is easy to show that the coefficient of S(t) is smaller than 1), but, as
shown in the following remarks, the comparison should not be carried too
far.
(2) The volatility a does not appear explicitly in the call price, but it is
carried implicitly in S(t) and in the integral It:
S(u) duo This appears
clearly in a direct and simple proof of (3.7), using the property that the
discounted underlying asset price is a Q-martingale; moreover, the result
easily extends to stochastic interest rates (see Geman 1989), since it indeed
involves the forward price of the underlying asset.
.

5.3. Asian Options Revisited 77

(3) It is obviously satisfactory to observe that Ct,T(k) is simultaneously and

separately increasing in the integral It:
S(u) du and in Set).
(4) The "delta" and the "gamma" of the Asian call can be immediately
derived from (3.7). It is worth observing that gamma is zero while delta is
not constant, which is an uncommon situation. This property reflects the
fact that the hedging strategy consists in selling every day (or continu-
ously) the same fraction of the underlying asset.

For q > 0, we do not have such a simple reduction of C(v)(h,q), but

we are able to provide an expression of its Laplace transform with respect
to the variable h, which mirrors the fact that A(v) taken at an independent
exponential time has a remarkable form (see Yor 1992c).
First, we recall from (3.5) that

c(V)(h) d~f C(v)(h, 0) = (1 ) {exp[2(v + l)h]- 1}.

2v+l

Then, using Proposition 2.3 we can represent the process

Xx = exp(Wx + vx), x;::: 0 as

X x = R(v)(A(v))
x ,

where (R(v) (u), u ;::: 0) is a Bessel process with index v, starting at R(v) (0) = l.
Moreover, if T~v) = inf{ s; A~v) > u}, then

(v) _ (U ds
Tu - io (R(v)(s))2

Now, we write

Ar) = A (~~) +
Tq
lh
(v)
Tq
exp 2(Ws + vs) ds,

on the set (Ar) ;::: q). Since A(~~) = q and denoting R~v) = [X(TJv))], in
Tq

agreement with (2.3), then the strong Markov property and the independence
of the increments of Brownian motion imply that, for all h ;::: 0 and q ;::: 0,

Consequently,

(3.8)
.

78 5. Bessel Processes, Asian Options, and Perpetuities

From (3.8) we can derive the Laplace transform in h of C(v)(h, q). For
>. > [),

where

Using Proposition 2.6, i.e.,

E[exp( _>'T(v)) I R(v) = s] = I/-,(sjq)

q q Iv(sjq) '

where JL = V2>. + v 2 , we can write

D (q >.) = r=p(V)(l s)s2 I /-,(sjq) ds
v, Jo q , Iv(sjq)'

Here p~v) (1, s) denotes the density at time q of the Bessel semigroup, with index
v and starting position 1 at time 0; this density is given by Proposition 2.2
(where we used pb instead of p(v), with [) = 2(1 + v)):

p(v)(l, s)
q
= sV+l exp
q
(-~(1
2q
+ s2)) Iv (~).
q

where JL = V2>' + v 2 and where the integrals in (3.10) are two representations
of confluent hypergeometric functions (see Lebedev 1972, p. 278, Exercise 12,
and p. 266, (9.11.1) and (9.11.2)).
Finally, the Laplace transform of cry) (h, q) with respect to the variable h
can be written as

1 =
o
dh
e
->"hC(v) (h )-
,q -
fl/2q
Jo
dx e- X x(/-,-v)/2-2(1 - 2qx)(/-,+v)/2+1
>.(>.-2-2v)f((JL-v)j2-1)
(3 10)
,.
where f denotes the gamma function.
The inversion of this Laplace transform for a fixed h would provide the
quantity C(v)(h, q) and, hence, the Asian option price thanks to (3.1). This
inversion is not easy, since the parameter >. also appears in the gamma func-
tion. There is now off-the-shelf software for inverting Laplace transforms,
which may be useful for a numerical solution of this problem.
.

5.4. Quadratic-Variation-Based Strategies 79

4. Quadratic-Variation-Based Strategies
In this section, we consider the issue of portfolio insurance based on option
replication when the model of the underlying asset does not have a constant
volatility over the investment period, which is the case in practice.
The most straightforward method of insuring a portfolio of risky assets is
to purchase a put option on the portfolio, with a strike price equal to some
desired minimum value for the portfolio. Alternatively, a dynamic portfolio
of stock and cash can be created to replicate the stock-plus-put portfolio.
However, the construction of such synthetic portfolio insurance will depend
crucially on a correct estimation of the volatility of the portfolio being insured.
Rendleman and O'Brien (1990), among others, have shown that misspeci-
fication of the volatility can cause the outcome of a synthetic portfolio
insurance strategy to deviate significantly from its target. Assuming that the
Black-Scholes option pricing model (particularly the hypothesis of a con-
stant volatility over the lifetime of the option) is appropriate for determin-
ing the value of the put option, a very simple argument-which is the core
of the Black-Scholes formula-gives the number of shares of the stock one
must hold and the amount one must invest in a riskless asset (e.g., T-bills).
These numbers depend on time and on the market value of the risky asset,
and, consequently, portfolio adjustments have to be made on a continuous
basis; however, the relative amounts in stocks and T-bills also depend on the
volatility (through the coefficients d1 and d2 in the Black-Scholes formula).
Therefore, errors in volatility estimation will result in an incorrect mix during
the life of the insurance program. This issue has usually been addressed by
estimating the deviation from the target at the time horizon H (using, for
instance, simulations of volatility, as in Hill et al. 1988).
Bick (1991) offers a new approach, which is to build "quadratic-variation-
based" dynamic strategies: instead of facing an unwanted outcome of the
portfolio insurance strategy at the time H chosen as the horizon, he proposes
to stop the strategy at the first time Ta such that

laTa 0'2(8) d8 = a, (4.1)

where Va is the volatility parameter chosen at inception of the portfolio

strategy at time t = 0, and O'(t), the true volatility of the risky asset, is
not necessarily constant or deterministic but only Sf-measurable. It is easy
to see that, assuming 0' is continuous and Sf-adapted, the stopping time

Ta = inf { U 2: 0; loU 0'2(8) d8 = a}

is the first instant at which the option replication is correct and, consequently,
the portfolio value is equal to the desired target.
Pursuing this approach, we can calculate the distribution of Ta in the Hull
and White (1987) framework. Denoting by S the risky asset and by y = 0'2
.

80 5. Bessel Processes, Asian Options, and Perpetuities

the instantaneous variance, we suppose that their dynamics are driven by the
following stochastic differential equations:
dS dy
5 = 111 dt + a dW1(t); Y = 112 dt + ~ dW2(t), where y(O) = Yo,

and d(Wl' W 2)t = Pt dt; i.e., WI and W 2 are two S'[-Brownian motions with
a deterministic bracket.
It will be convenient to use the notation 11~ = 112/2 and ry = ~/2. Hence,
we have
a(t) = VYo exp{(ry) W 2(t) + (11~ - ry2)t}.
Making the change of variables t = U/ry2 and using the scaling property of
Brownian motion, we can write

a(t) =a (;) = aoexp [W2(U) + (11~ - ry2); l

where W2 is another Brownian motion. Recognizing the exponential of a
Brownian motion with drift, we use again its Bessel process representation:

(4.2)

where (R~;ht), t :::: 0) is a Bessel process starting at ao, with index v =

11~/ry2 - 1.
We now wish to identify the stopping time Ta in (4.1). To simplify com-
putations, we first assume ry = 1. Making in (4.1) the random time change
S = Tb, we obtain Ta = f oa db/a 2 (Tb). Now, replacing u by Tb in (4.2) gives

a(Tb) = R~;hb). Hence,

Similar computations for a general ry lead to

Ta = ~ {2a'/l db
ry2 io (R~)(b))2·
The probability density fa (x) of Ta does not have a simple expression (see
Yor 1980, Sections 4-6), but its Laplace transform is relatively simple, being
precisely
.

5.5. Valuation of a Perpetuity in a Stochastic Interest Rates Environment 81

where Lemma 2.1 was applied twice; here, JL = (2)../ry2 + v 2)1/2 and /'i, =
(JL - v)/2.
Using Proposition 2.5 we can transform the latter expectation into

Therefore, we have replaced uncertainty about the portfolio outcome by

uncertainty about the variable time horizon t of the strategy; however, we
know the distribution of this stopping time. For practical purposes, this
approach to volatility misspecification in classical portfolio insurance strate-
gies removes the following unpleasant feature of these strategies, which is that
the level of investment of the portfolio in the risky asset just prior to the fixed
horizon H is either 0% or 100%, depending on whether the value of the port-
folio at that time is below or above the desired floor. An investor wishing to
resume this strategy would see his portfolio reallocated to a drastically diffe-
rent asset mix, whereas his risk aversion has not changed from one day to
the next. In the "stopping time investment strategies" that we propose, the
investor, after having obtained the desired terminal wealth, can choose any
type of new strategy. Longstaff (1990) had already raised the issue of European
options with "extendible maturities"; in our approach, however, the date at
which the target portfolio is exactly obtained-i.e., the date t at which fo
was the correct volatility to use in the Black and Scholes formula-may be
earlier than the horizon H at which the target would have been reached in
the situation of constant volatility. In this case, this date t would be the right
time to resume the portfolio insurance.
In the same spirit, many fund managers have been particularly interested
since the market crash of 1987 in hedging specifically against sudden market
drops. Some financial institutions today offer index options (assuming the
manager holds a well-diversified portfolio) which have a fixed exercise price
and a stochastic maturity date as defined above. If there is a market collapse,
the high volatility in (4.1) implies that the put option is rapidly exercised
(i.e., Ta is small), protecting the fund value; otherwise, the option keeps its
time value. These options are less expensive than American options (which
also have a random maturity date but of another nature) and still allow one to
avoid the drawbacks of synthetic portfolio insurance strategies during market
crashes.

5. Valuation of a Perpetuity III a Stochastic Interest

Rates Environment

In a recent paper, Dufresne (1990) was able to establish the distribution of

a perpetuity making a continuous payment of one unit over time, when this
.

82 5. Bessel Processes, Asian Options, and Perpetuities

perpetuity is expressed as

Z= 1 00
dtexp-(aWt+vt)

where v and a are positive.

In fact, Dufresne showed that Z is distributed as the reciprocal of a gamma
variable. Taking into account, in the actuarial approach, the stochastic nature
of interest rates is certainly an interesting step. However, this approach has
two limitations: first, the result does not hold (see Section 3) when the upper
bound of the integral in Z is a fixed number T (which would better express
the finite nature of human life): second, and more troublesome, is that the
discount factor used by Dufresne, exp -(aWt + vt), is radically different from
the one which would be derived from any of the models presented in the abun-
dant financial literature on interest rates. This may be related to Dufresne's
discount factor being obtained as the limit of discrete discounting over a num-
ber of time intervals going to infinity. We are going to take these two points
into account in our approach, but at the cost of losing the knowledge of the
exact distribution of the random variable Z.
We take the classical setting in which the short-term rate is the state vari-
able of the interest rates movements and is driven by the following dynamics:
(5.1)
Here, Wt denotes a standard Brownian motion, b, (3, a, and'Y are constants
and 1/2 :::: 'Y :::: 1 or 'Y = 0 (see, for instance, Geman and Portait 1989).
We will first make an important observation: if'Y E ]0,1/2[, (4.1) may not
have a unique solution; at least in the case (3 = 6 = 0, it is well known that
this equation does not have a unique solution. This famous result is due to
Girsanov (see, for instance, Rogers and Williams 1987, vol. 2, pp. 175-176).
When 'Y = 1/2, it can be shown that if 26 2': a 2 , (5.1) has a unique solution
which does not hit O.
The situation 'Y = 1/2, (3 < 0 corresponds to the well-known Cox-Ingersoll-
Ross (1985) model of interest rates. The C.LR. term structure model is derived
in a general equilibrium setting; not only does it avoid arbitrage opportunities,
but it also provides the market price of interest rate risk as part of the equi-
librium. In the C.LR. case, the coefficient (3 in (4.1) under the risk-neutral
probability incorporates the market price of interest rate risk. Within this
setting, we are going to price perpetuities and annuities.
Denoting (3 = -2(j, we deduce from (5.1) that for any>. E IR~,

r(>.t) = ro
rAt
+ (b>.)t + a io v'r(S) dWs -
r'
2(j>. io r(>.s) ds.

Using the scaling property of Brownian motion, we can write

r(>.t) = ro + (b>.)t + a It Jr(>'s)~ dWs -

t
2(j>.l r(>'s) ds (5.2)
.

5.5. Valuation of a Perpetuity in a Stochastic Interest Rates Environment 83

where W"S represents another Brownian motion. Denoting 8 = S>., () = e>., and
r(t) = r(>.t) and choosing>. = 4/1J2, we observe that when 8 is an integer,
r(t) is the square of the norm of an Ornstein-Uhenbeck process (x(t), t ~ 0)
with dimension 8 and parameter e, i.e.,

x(t) = Xo + Zt -1J it x(s) ds, (5.3)

where (Zt,
t ~ 0) is a 8-dimensional Brownian motion and Xo E ]R6. A proof
of this result is obtained by using Ito's lemma for p(t) = Ix(t)j2:

where dW"s = (x(s) . dz(s))/lx(s)1 (the numerator being an inner product in

]R6).
We have thus recovered (5.1) since 1Jv0. = 2. When 8 = 1, (5.3) involves
the one-dimensional Brownian motion and r(t) is the square of an Ornstein-
Uhlenbeck process. This (one-dimensional) Ornstein-Uhlenbeck process has
become familiar in finance since it was introduced by Vasicek (1977) when
defining the short-term interest rate dynamics by the particular form of (5.1):

dr = a(b - r)dt + IJdWt , with a, b, and IJ constant.

We have thus exhibited a relationship between two very popular models

of the term structure dynamics. The Gaussian feature of the Vasicek model
has the drawback of possibly negative interest rates (but would lead to a
mathematically simple valuation of annuities and perpetuities). Consequently,
we are going to study the pricing of perpetuities in the Cox-Ingersoll-Ross
framework.
Prior to that, we want to show how the valuation of zero coupon bonds in
the C.I.R. model can be obtained within our methodology. The price at time
o of a zero coupon bond maturing at time t is

B(O,t) = EQ [exp (-it r(s) dS) l

Making the change of variable s = >.u = 4u/ 1J2 in the integral, we obtain

B(O,t)=EQ [ exp ( -Jtt/(J2

o
4)]
r(u)1J2du.
.

84 5. Bessel Processes, Asian Options, and Perpetuities

where r( u) is the square of the norm of x( u), as-dimensional Ornstein-

Uhlenbeck process with parameter e
= 40 I 0'2. Girsanov's relationship
between the law Po of the norm of x( u) and the law P of the S-dimensional
Bessel process R(u), assuming they both start at Fa at time 0, is

Moreover, formula (2.k) in Pitman and Yor (1982) establishes that for any a
and bin jR+,

E[exp ( - ~ R; - b; lot R; dS) ]

a ) -6/2 (-rob alb + th(bt) )
= ( ch(bt) + b"sh(bt) exp -2-1 + (alb)th(bt) ,

where again the 5-dimensional Bessel process R(t) originates at Fa. Com-
bining these two results, we obtain for the bond price at time 0 that

B(O, t) == UVW,
with

U ( 20 (ro + 16M))
= exp 0'2 0'4 '

_{ (16t (1- + - )1/2) + -

V - ch -
02 0 (1- + - )-1/2
02
sh
(16t
-
(1- + - )1/2)
02 }2/)/a
2

,
0'3 2 0'2 0' 2 0'2 0'3 2 0'2

A perpetuity making a continuous payment of one unit over time has the
form

=A 1 00
duexP(-A [r(As)ds), foranyA>O.
.

5.5. Valuation of a Perpetuity in a Stochastic Interest Rates Environment 85

To gain "simplicity," we first look at the case (3 = 0, 'Y = 1/2. Defining b2 =

8/a 2 = 2>', we see that Z takes the form

(5.4)

where (R(s), s 2: 0) denotes a Bessel process with dimension 8>' = 2(v + 1),
starting from Fa.
We are interested in computing the first-order moment of Z since its dis-
tribution is not simple. The details of this computation, which use roughly
the same tools as the ones used in the previous sections, are completed in
Appendix B and the result is given in (B.5):

E(Z) = J2
a
r _1_(1_
1

Jo v'h
h)(V-l)/2 exp (-brov'h)
2
dh,

where b = 2 J2 / a, v = 48/ a 2 - 1, and 8 and a are the parameters of the short-

term dynamics in the Cox-Ingersoll-Ross model defined in (5.1) by'Y = 1/2
(and (3 = 0 for now).
Denoting by ZT an annuity which is continuously paid between dates 0
and T, all the previous results may be slightly modified to show that the
financial value of this annuity at time 0 is

r bU)
J2
E[ZTJ = -;; Jo
th(bT)
(1 - u 2 )(V-l)/2 exp
(
--T"- du

with the same notation as above. (We remark that, taking the derivative of
this expression with respect to T, we recover the bond price B(O, T) calculated
above, for 0 = 0.)
As a consequence, we obtain

where (3 is the (Euler) beta function.

Now, from the expression of (3 in terms of r and the duplication formula
for r, Le.,

2
r(n+1)=y'1fr
n
2+ 1 r + 1)
(n ) (n- 2 -'
.

86 5. Bessel Processes, Asian Options, and Perpetuities

we deduce

E Z _ J2 {~ (_ bro)n f((v + 1)/2)y'1T }

[ ]- (J" f::o 2 2nf(n/2 + l)r((n + v)/2 + 1) .

In the general case of the Cox-Ingersoll-Ross model, (3 = -2(), with () > 0

and the same method gives

E(Z) = Aexp (()~o)

1o
+00 (()bA2
dtexp -2- t (ch wt+flshwt)
---6>../2 {-row(()+th(wt))}
exp -2- 1- + flth(wt)

where w = V2A + ()2 A2 , A = 4/ (J"2, fl = ()Ajw, and the parameters (3, b, and (J"
define the short-term rate dynamics in (5.1) with 'Y = 1/2. Introducing the
change of variables u = th(wt), we can write

_
E(Z) -A exp
(()Aro)
2
r w(l du- u2) (1+U)(0>../4W)!i
Jo
1
1- u

1 () u -0>../2 { -row ( fl + u ) }
.,. ( ~ + - V1- u 2 ) exp -2- 1 + flu .

After simple computations, this can be rearranged as

E(Z) = :;-exp). (().) 11 2Tb 0 (1 + u)P(l - u)'1(l + flu)--6>../2 ...

-r 0 w ( fl + u ) }
... exp { - 2 - 1 + flu du

where

The same type of computational developments as in the case (3 = () = 0 could

be made from here.

6. Conclusion

We have seen how Bessel processes can be used to bring two kinds of
answers to the problem of Asian options pricing, one through the calculation
of all moments of the average of the underlying asset price, the other through
.

5. Appendix A. The "Forward-Start" Asian Option 87

the expression (of the Laplace transform) of the Asian option price. In the
portfolio insurance framework, Bessel processes provide alternative answers
to the issue of deviating from the target at the time horizon in portfolio
insurance strategies when the volatility of the risky asset is stochastic. In the
same manner, Bessel processes appear to be the appropriate tool to price per-
petuities and annuities in a stochastic interest rates environment when the
dynamics of the short-term rate are defined by a particular case of the Cox-
Ingersoll-Ross model. These processes also appear very promising for other
models of the term structure dynamics.

Appendix A.
The "Forward-Start" Asian Option

A.1. Reduction of the Integral

We now show that for t < to < T, it is possible to reduce the valuation
problem to the case t = 0:

E[(A(T) - k)+IS{] = E{[E(A(T) - k)+IS{o] I S{} = E[¢(S(to)) I S{J,

where

since S(u) = exp[y(u)] = S(to) exp[y(u) - y(t o)] and (y(t), t 2: 0) has inde-
pendent increments. We observe that

¢(,) ~ ,E [ ( [ duexp(y(u) - y(to)) - ~ ) +1= SCto.T(k/s).

def -

Now, conditioning with respect to S{, we obtain

[ - (k)
S(tO) IS{] = X(S(t)),
E[¢(S(to)) I S{] = E S(tO) Cto,T def
.

88 5. Bessel Processes, Asian Options, and Perpetuities

where

X(s)=sE [~~;? Go,T(~~;?))]

=sE [exp{y(~ - t)}Cto,T (~exp-y(to - t))].
The gain is that the last expression is no longer a conditional expectation but
simply an expectation.

A.2. The Use of This Instrument

When, in 1989, Dow Chemical made a tender offer on Marion Laboratory, the
shareholders of the latter company received in exchange for an old share a
"contingent value right" attached to a new share. This contingent value right
(C.V.R.), which was traded on the American Stock Exchange between 1989
and 1991, was exactly an Asian put option, since the payoff in September
1991 was $45.77 - $A where $A was the average value of the stock during the
last 60 days. The same contingent value right was used in the tender offer of
the French company Rh6ne-Poulenc on the American firm Rorer. The Rh6ne-
Poulenc-Rorer contingent value rights are currently traded on the American
Stock Exchange; their expiration date is July 1993.

Appendix B.
Valuation of a Perpetuity
in the C.LR Framework

We start from (5.4):

Z = AZI = A 1 00
exp ( - b; lt [R(sWdS) dt.

It will be very helpful to use the relation

(B.1)

where (Rrn(t), t ~ 0) denotes a Bessel process with index m, and 1jp+ 1jq = 1
(see Revuz and Yor 1991, Chap. 11). [Note: In this Appendix, we use the
notation Rv instead of R(v), as in the rest of the paper, because here we deal
constantly with powers of R v , and R~(t) denotes the quantity Rv(t) raised to
the power o:.J
.

5. Appendix B. Valuation of a Perpetuity in the C.I.R Framework 89

If we take p = -1, we obtain q = 1/2. Denoting

Au = inf {t: 1t dsR~(s) > U}.

we deduce from (B.1), where we replace t by Au, that

U::::: 0,
This gives, with our choice of q,

U::::: 0. (B.2)

Now we can write ZI as

ZI = 1 00
dAu exp ( - b:U) = 1 R~~~u)
00
exp ( _ b:U ). (B.2')

Using (5.1) we obtain

(B.3)

An Expression of the First Moment of Z. To compute the first moment of Z,

we use Proposition 2.5:

E~m) ((R~)20 ) = rto) 11 dh h O- 1(1 - h)m-O (;t) 0 exp ( - ~~ ), (B.4)

where we take m = v /2,0 = 1/2, Z = (ro/2)2. We then deduce from (B.3) and
(B.4) that

E(Z) = 2r~O) 11 dh hO-l(l - h)m-O 1 (:~o 00

exp (~~) exp (_~2U)

11 1 J2u
(B.5)
2
= -oX- dh
--(1 - h)(v-l)/2
00
-du- exp (-Zh)
- - exp (-b
-- u) .
2"fir 0 Jh 0 2u 2
We could also have avoided the time change operation with the help of
the following formula, which is an important example of a Laplace trans-
form for quadratic functionals of Brownian motion (see Revuz and Yor 1991,
Corollary 1.8, p. 414):

E~~) [exp - b; 1t dSX s ] = exp ( - r~b th(bt)) [ch(bt)r(v+l). (B.6)

(Again, we use the notation ch and th for hyperbolic cosine and tangent.)
.

90 5. Bessel Processes, Asian Options, and Perpetuities

This gives

E[Z1J = 1
0
00 dt (rob)
(ch(bt))v+1 exp -Tth(bt)

b1
1 00 ds (rob)
= 0 (ch(s))v+1 exp -Tth(s) .

Making the change of variables h = th 2 s, we obtain

E[ZJ = ~
2b
11 Jh
0
dh (1 - h)(V-1)/2 exp( -bVzh). (B.7)

This is another expression for (B.5) as shown below.

We start using the classical integral representation for Bessel functions
KIJ:

( -b)a-1 2Ka_1(bc) = 100 -dy exp --1 ( b2 y + -c2 ) ,

c 0 ya 2 y

(see, for instance, Lebedev 1972, p. 119), and the important particular cases
a = 1/2 and a = 3/2:

(B.8)

Returning to (B.5) we obtain

E[Z1J = - 1
2y'1f
11 Jh
0
dh - h)(v-1)/2
-(1 100
0
exp (-Zh)
-
2u
exp (-b
--
2
2u) du
--;
ffu

therefore,

E[Z1J = - 1 11 - 1 ( b )
dh (1 - h)(V-1)/2_ -- -1/2 2K1/ 2(bVzh)
2y'1f Jh J2 Jdi
11 Jh
0

=-1 dh
-(1- h) (v-1)/2 -y'1f exp( -by/zh),
J
2y'1f 0 b

using (B.7). Remembering that E(Z) = AE(Zd, the last expression is pre-
cisely the righthand side of (B.7).
.

5. Appendix B. Valuation of a Perpetuity in the C.I.R Framework 91

References

Bick, A. (1991). Quadratic Variation Based Dynamic Strategies. European

Finance Assoc. Meeting, Rotterdam
Black, F. and Scholes, M. (1973). The Pricing of Options and Corporate
Liabilities. J. Political Econom., 81, 637-654
Bouaziz, L., Bryis, E. and Crouhy, M. (1994). The Pricing of Forward-Starting Asian
Options. J. Banking and Finance, 18, 623-639
Bougerol, P. (1983). Exemples de Theoremes Locaux sur les Groupes Resolubles.
Ann. Inst. H. Poincare, 19, 369-391
Carverhill, A.P. and Clewlow, L.J. (1990). Valuing Average Rate (Asian)
Options. Risk, 3, 25-29
Cox, J., Ingersoll, J. and Ross, S. (1985). A Theory of the Term Structure of Interest
Rates. Econometrica, 53, 363-384
Dufresne, D. (1989). Weak Convergence of Random Growth Processes with
Applications to Insurance. Insurance: Math. Econom., 8, 187-201
Dufresne, D. (1990). The Distribution of a Perpetuity, with Applications to Risk
Theory and Pension Funding. Scand. Actuarial J., 39-79
Feller, W. (1964). An Introduction to Probability Theory and Its Applications. 2,
New York: Wiley
Garman, M.B. and Kohlhagen, S.W. (1983). Foreign Currency Option Values.
J. Int. Money Finance, 2, 231-237
Geman, H. (1989). The Importance of the Forward-Neutral Probability Measure in
a Stochastic Approach to Interest Rates E.S.S.E.C. Working paper
Geman, H. (1992). Asset Allocation, Portfolio Insurance and Synthetic Securities.
Appl. Stach. Models Data Anal., 8,179-188
Geman, H. and Portait, R. (1989). A Framework for Interest Rate Risk Analysis and
Portfolio Management. American Stock Exchange Colloq., New York
Geman, H. and Yor, M. (1992). Quelques relations entre processus de Bessel, options
asiatiques, et fonctions confluentes hypergeometriques. C. R. Acad. Sci. Paris Ser.
I, 471-474. Paper [3] in this book
Harrison, J.M. and Kreps, D. (1979). Martingale and Arbitrage in Multiperiods
Securities Markets. J. Econom. Theory, 20, 381-408
Harrison, J.M. and Pliska, S.R. (1981). Martingales and Stochastic Integrals in the
Theory of Continuous Trading. Stach. Proc. Appl., 11, 215-260
Hill, J., Jain, A. and Wood, R.A. (1988). Portfolio Insurance: Volatility Risk and
Futures Mispricing. J. Portfolio Management, Winter
Hull, J. and White, A. (1987). The Pricing of Options on Assets with Stochastic
Volatilities. J. Finance, 42, 281-300
Ito, K. and McKean, H.P. (1965). Diffusion Processes and Their Sample Paths.
Berlin: Springer-Verlag
Kemna, A.G.Z. and Vorst, A.C.F. (1990). A Pricing Method for Options Based on
Average Asset Values. J. Banking Finance, 14, 113-129
Lebedev, N. (1972). Special Functions and Their Applications. New York: Dover
Levy, E. (1992). Pricing European Average Rate Currency Options. J. Int. Money
Finance, 11, 474-491
Levy, E. and Turnbull, S. (1992). Average Intelligence. Risk, 5, 53-59
Longstaff, F. (1990). Pricing Options with Extendible Maturities: Analysis and
Applications. J. Finance, 45, 935-956
.

92 5. Bessel Processes, Asian Options, and Perpetuities

Pitman, J.W. and Yor, M. (1982). A Decomposition of Bessel Bridges.

Z. Wahrscheinlichkeitstheorie, 59, 425-457
Rendleman, RJ. and O'Brien, T.J. (1990). The Effects of Volatility Misestimation
on Option Replication Portfolio Insurance. Financial Anal. J., May-June, 60-61
Revuz, D. and Yor, M. (1991). Continuous Martingales and Brownian Motion.
Berlin: Springer-Verlag
Rogers, L.C.G. and Williams, D. (1987). Diffusions, Markov Processes and Martin-
gales. New York: Wiley
Ruttiens, A. (1990). Average-Rate Options, Classical Replica. Risk, 3, 33-36
Shiga, T. and Watanabe, S. (1973). Bessel Diffusions as a One-Parameter Family of
Diffusion Processes. Z. Wahrscheinlichkeitstheorie, 27, 37-46
Turnbull, S. and Wakeman, L.M. (1991). A Quick Algorithm for Pricing
European Average Options. Financial Quant. Anal., 26, 377-389
Vasicek, O. (1977). An Equilibrium Characterization of the Term Structure.
J. Financial Econom., 5,177-188
Vorst, T. (1992). Prices and hedge ratios of average exchange rate options. Intern.
Review of Financial Analysis, 1 (3), 179-193
Williams, D. (1974). Path Decomposition and Continuity of Local Time for One-
Dimensional Diffusions. Proc. London Math. Soc., 28 (3), 738-768
Yor, M. (1980). Loi de l'indice du lacet brownien, et distribution de Hartman-
Watson. Z. Wahrscheinlichkeitstheorie, 53, 71-95
Yor, M. (1992a). Some Aspects of Brownian Motion. Part I: Some Special Function-
also Lecture Notes in Math., E.T.H. (Zurich). Basel: Birkhiiuser
Yor, M. (1992b). On Some Exponential Functionals of Brownian Motion. Adv. Appl.
Probab., 24, 509-531. Paper [2] in this book
Yor, M. (1992c). Sur les lois des fonctionnelles exponentielles du mouvement
brownien, considerees en certains instants aleatoires. C.R. Acad. Sci. Paris Sir. I,
314, 951-956. Paper [4] in this book

Postscript #5
A further discussion of this paper and perspectives in Mathematical finance
are found in H. Geman's contribution [0) at the beginning of this monograph.