BLACK-SCHOLES GOES HYPERGEOMETRIC

CLAUDIO ALBANESE, GIUSEPPE CAMPOLIETI, PETER CARR, AND ALEXANDER LIPTON
ABSTRACT. We introduce a general pricing formula that extends Black-Scholes’ and contains as particular
cases most analytically solveable models in the literature, including the quadratic and the constant-elasticity-
of-variance (CEV) models for European and barrier options. In addition, large families of new solutions can
be found, containing as many as seven free parameters.
1. INTRODUCTION
It has been known since the seventies that Black-Scholes pricing formulas are a special case of more
general families of pricing formulas with more than just the volatility as an adjustable parameter. The
list of the classical extensions includes affine, quadratic and the constant-elasticity-of-variance models.
These models admit up to three adjustable parameters and have found a variety of applications to solving
pricing problems for equity, foreign exchange, interest rate and credit derivatives. In a series of working
papers, the authors have recently developed new mathematical techniques that allow to go much further
and to build several families of pricing formulas with up to seven adjustable parameters in the stationary,
driftless case, and additional flexibility in the general time dependent case. The formulas extend to
barrier options and have a similar structure as the Black-Scholes formulas, the most notable difference
being that error functions (or cumulative normal distributions) are replaced by (confluent) hypergeometric
functions, i.e. the special transcendental functions of applied mathematics and mathematical physics.
Let F denote a generic financial observable which we know is driftless. Examples could be the
forward price of a stock or foreign currency under the forward measure, a LIBOR forward rate or a swap
rate with the appropriate choice of numeraire asset. Black or Black-Scholes formulas are obtained by
postulating that the time evolution of F obeys a stochastic differential equation of the form
(1.1) dF
t
= σ(F
t
)dW
t
where σ(F) = σ
1
F is linear. In this case, pricing formulas of calls and puts for both plain vanilla and
barrier options can be written in exact analytical form in terms of the error function or cumulative normal
distribution. Interestingly, quadratic volatility models with
(1.2) σ(F) = σ
0

1
F +σ
2
F
2
.
also allow for pricing formulas that reduce to the evaluation of an error function. The reason why error
functions suffice also to express pricing formulas in the more general case of quadratic volatility models
is that these models can all be reduced to a Wiener process by means of a simple measure change and
variable transformation of the form
(1.3) F
t
= F(x
t
),
where the underlying x
t
follows:
(1.4) dx
t
= dW
t
.
The formula (1.3) applies under a pricing measure where assets are valued in terms of a suitably defined
numeraire g
t
= g(x, t). For a review of change of numeraire methods in pricing theory we refer to [1] and
[2]. Both functions F(x) and g(x, t) can be derived explicitly for any choice of parameters σ
0
, σ
1
, σ
2
.
It also turns out that quadratic volatility models are the only stationary, driftless models for which
the combination of a non-linear transformation of the form (1.3) together with a change of numeraire
Date: August 6th, 2001.
C. A. was supported in part by the National Science and Engineering Council of Canada. We thank Dilip Madan, Stephan Lawi,
Vadim Linetsky and Andrei Zabidonov for discussions. Remaining errors are our own.
1
2 CLAUDIO ALBANESE, GIUSEPPE CAMPOLIETI, PETER CARR, AND ALEXANDER LIPTON
reduces the problem to a Wiener process x
t
as the one in (1.4). In [3], Carr, Lipton and Madan address
the question of whether it is possible to relax the condition of stationarity and find more general processes
with drift and volatility both dependent on calendar time as well as on F which still reduce to the Wiener
process and they find that the general solution admits as many as eleven time dependent functions.
A related line of reasoning leading to extensions of the Black-Scholes formula starts from the obser-
vation that the CEV models with state dependent volatility specified as follows:
(1.5) σ(F) = σ
0
(F −
¯
F)
1+θ
with constants θ and
¯
F, reduces to the Bessel process
(1.6) dx
t
= λ
0
dt +ν
0

x
t
dW
t
.
by means of a non-linear transformation combined with a measure change. In [4], Lipton derives general
reducibility conditions to the more general processes
(1.7) dx
t
= (λ
0

1
x
t
)dt +ν
0
x
β
t
dW.
which are solvable for β = 0, 1,
1
2
. The case β = 1 is the lognormal (or affine) model leading to the
Black-Scholes formula. The two cases β = 0,
1
2
correspond to well-known solvable short rate models,
namely the Vasicek and the Cox-Ingerssol-Ross (CIR) models. In [5], [6], Albanese and Campolieti find
a general solution to the reducibility conditions by Carr, Lipton and Madan for stationary, driftless pro-
cesses. In this article, we summarize our findings by presenting the general solution formula, originally
derived using reducibility conditions and illustrate its use in a few particular cases. Since the original
derivation is somewhat lengthy, in the Appendix we give a streamlined verification of its validity and
refer to our other papers for a constructive derivation.
2. GENERAL PRICING FORMULA
In this section, we derive a general pricing formula for the models which are solvable by the reduction
method.
Assume that the state variable x has a drift λ(x) for which one can find the pricing kernel for the
process
(2.1) dx
t
= λ(x
t
)dt +ν(x
t
)dW
t
.
The pricing kernel is the function u(x, t; x
0
, t
0
) which solves the forward Fokker-Plank equation in the
first pair of arguments and the backward Black-Scholes equation in the second pair. The latter equation
can be written as follows:
(2.2) u
t0
(x, t; x
0
, t
0
) +
ν(x
0
)
2
2
u
x0x0
(x, t; x
0
, t
0
) +λ(x
0
)u
x0
(x, t; x
0
, t
0
) = 0
for t ≥ t
0
, with final time condition at terminal time t
0
= t given by u(x, t; x
0
, t) = δ(x −x
0
) (a Dirac
delta function). Recall that the pricing kernel can be interpreted as the price of a limiting butterfly spread
option and is related to the second derivative of a call option written on the state variable x with respect
to the ”strike” x
0
. The processes in (1.7) are examples of analytically solvable models for which one can
compute the pricing kernel.
Solvable pricing models can be constructed starting from the price function of a generic European
style option written on x, i.e. a solution v(x, t) of the Black-Scholes equation in (2.2) with an arbitrary
final time condition at t = 0. The Laplace transform of such a function
(2.3) ˆ v(x, ρ) =


0
e
ρ(t0−t)
v(x, t
0
−t) dt.
is usually referred to as “time-independent Green’s function” and satisfies a second order ordinary differ-
ential equation with Dirac delta function source term δ(x − x
0
). Let us consider the homogeneous part
of this equation as given by
(2.4) −ρˆ v(x, ρ) +
ν(x)
2
2
ˆ v
xx
(x, ρ) +λ(x)ˆ v
x
(x, ρ) = 0.
BLACK-SCHOLES GOES HYPERGEOMETRIC 3
We find that functions ˆ v(x, ρ) solving this equation can be taken as the elementary building blocks for
the construction of solvable pricing models for the F space processes. Because of this reason, we name
ˆ v(x, ρ) the “generating function”.
Armed with a solution ˆ v(x, ρ), we define a volatility function σ(F) and an invertible monotonic
transformation F = F(x) and its inverse x = X(F) such that
(2.5) σ(F) =
σ
0
ν(X(F)) exp

−2

X(F) λ(s)ds
ν(s)
2

ˆ v(X(F), ρ)
2
with arbitrary constant σ
0
and where
(2.6)
dx
ν(x)
= ±
dF
σ(F)
.
The two signs correpond to either monotonic increasing or monotonic decreasing transformations. The
freedom in choosing the sign gives rise to 2 families of solutions which are different in the general case.
As we verify in the Appendix, the process
(2.7) g
t
=
e
ρt
ˆ v(x
t
, ρ)
can be regarded as a forward price process and under the measure with g as a numeraire the state variable
x
t
drifts at rate λ(x). Hence, the pricing kernel U(F, t; F
0
, 0) for the overlying forward price F at time
t can be evaluated in closed form as the expected reward from a limit butterfly spread contract with delta
function payoff
(2.8) U(F, t; F
0
, 0) = E
0
¸
g
t
g
0
δ(F(x
t
) −F)

conditional on the price having value F
0
at initial time t = 0. Here, the expectation is computed assuming
that g
t
is the numeraire and that the state variable x drifts at rate λ(x). The final formula for the pricing
kernel in F space is related to the kernel in the underlying x space as follows:
(2.9) U(F, t; F
0
, 0) =
ν(X(F))
σ(F)
ˆ v(X(F), ρ)
ˆ v(X(F
0
), ρ)
e
−ρt
u(X(F), t; x(F
0
), 0).
A European call option written on the forward price F
0
at current time t = 0, struck at K and maturing
at time t = T can be priced in this model by computing the following integral:
(2.10) C(K, T; F
0
) = e
−ρT


X(K)
dx
ˆ v(x, ρ)
ˆ v(X(F
0
), ρ)
(F(x) −K)u(x, T; X(F
0
), 0).
Barrier and lookback options can be handled by modifying the underlying kernel in x-space to account for
the appropriate boundary conditions. This is accomplished by means of either integral representations
or eigenfunction expansion methods, i.e. Green’s function methods that are standard in the theory of
Sturm-Liouville equations. See the articles of Davydov and Linetsky [7] for a discussion in an option
pricing context.
3. FOUR FAMILIES OF SOLVABLE MODELS
The case β = 1 is the usual lognormal (or affine) model. Of interest here is the other two families with
β = 0,
1
2
in equation (1.7) giving rise to solvable models in case λ
1
< 0 which admit as a singular limit
a family with different properties in the non mean-reverting limit λ
1
= 0. This provides four examples
of our methodology to generate exactly solvable models.
If β = 0 and λ
1
= 0 we recover the Wiener process with constant drift, which is readily transformed
into a driftless Wiener process and thus supports only quadratic volatility functions in F space, including
the lognormal Black-Scholes model as a special subcase. If β = 0 and λ
1
< 0 then the kernel in x space
can be shown to be given by
(3.1) u(x, t; x
0
, 0) =
2


1
sinh(λ
1
t)

πν
0
(e
2λ1t
−1)
sinh

¯ x
0
¯ x
ν
2
0
λ
1
sinh(λ
1
t)

exp
¸
¯ x
2
+ ¯ x
2
0
e
2λ1t
λ
1
ν
2
0
(e
2λ1t
−1)
+
λ
1
2
t

4 CLAUDIO ALBANESE, GIUSEPPE CAMPOLIETI, PETER CARR, AND ALEXANDER LIPTON
where ¯ x ≡ λ
0
+ λ
1
x, ¯ x
0
≡ λ
0
+ λ
1
x
0
. The generating function solves a special case of the confluent
hypergeometric equation (Hermite’s equation) with general solution
(3.2) ˆ v(x, ρ) = q
1
M


ρ

1
,
1
2
,

0

1
x)
2

1
ν
2
0
)
2

+q
2

0

1
x)M


ρ

1
+
1
2
,
1
2
,

0

1
x)
2

1
ν
2
0
)
2

with arbitrary constants q
1
,q
2
. Here M(a, b, z) is Kummer’s function[8], i.e. the confluent hypergeomet-
ric function that is regular at z = 0. In this expression, one can count 6 free dimensionless parameters.
For each choice of these parameters, there are two different families corresponding to the two different
choices of the sign in equation (2.6).
If β =
1
2
and λ
1
= 0, then the pricing kernel for the state variable is expressed in terms of modified
Bessel functions as follows:
(3.3) u(x, t; x
0
, 0) =

x
x
0
1
2
(

0
ν
2
0
−1)
e
−2(x+x0)/ν
2
0
t
ν
2
0
t/2
I2λ
0
ν
2
0
−1

4

xx
0
ν
2
0
t

.
The generating function is
(3.4) ˆ v(x, ρ) = x
1
2
(1−

0
ν
2
0
)
¸
q
1
I2λ
0
ν
2
0
−1

8ρx
ν
2
0

+q
2
K2λ
0
ν
2
0
−1

8ρx
ν
2
0

,
with arbitrary constants q
1
,q
2
. Here I
ν
(z) is the modified Bessel function of order ν and K
ν
(z) is the
associated McDonalds function. In this case we obtain a dual family with 6 adjustable parameters.
The case β =
1
2
and λ
1
< 0 is more general than the Vasicek case. Geometrically, the CIR process
describes the stochastic dynamics of the radial distance of a point whose Cartesian coordinates follow
an evolution given by the Vasicek model. This analogy applies only to integer dimensions, but analytic
continuation in the dimension parameter allows one to gain one additional degree of freedom. The pricing
kernel for the state variable x corresponds to that of the short rate CIR model, and can still be expressed
in terms of modified Bessel functions as follows:
(3.5) u(x, t; x
0
, 0) = c
t

xe
−λ1t
x
0
1
2
(

0
ν
2
0
−1)
exp

−c
t
(x
0
e
λ1t
+x)

I2λ
0
ν
2
0
−1

2c
t

xx
0
e
λ1t

,
where c
t
≡ 2λ
1
/(ν
2
0
(e
λ1t
−1)). For a derivation see [9]. The general solution of equation (2.4) reduces
to Whittaker’s equation and generating functions have the general form
(3.6) ˆ v(x, ρ) = x
−λ0/ν
2
0
e
−λ1x/ν
2
0
¸
q
1
W
k,m



1
ν
2
0
x

+q
2
M
k,m



1
ν
2
0
x

for arbitrary constants q
1
,q
2
. Here W
k,m
(·) and M
k,m
(·) are Whittaker functions which can also be
expressed in terms of confluent hypergeometric functions or in terms of Kummer functions.[8] This
construction gives rise to a dual family with 7 free parameters where
(3.7) k =
λ
0
ν
2
0
+
ρ
λ
1
, m =
λ
0
ν
2
0

1
2
.
The 7 parameter family which reduces to the CIR model has a local volatility function defined on
either an interval or on a half line and behaves asymptotically as the CEV volatility on one hand and as
a quadratic model on the other. This hibrid shape allows for a great deal of flexibility in reproducing
observed volatility skews. In Figure 1 we show
There is a way of gaining a visual understanding of the geometric meaning of the 7 parameters which
perhaps over-simplifies the picture but is intriguing. The allowed shapes, when confined to a finite inter-
vals, can be regarded as hybrids between the quadratic and the CEV model. The support of the volatility
function can be either a finite or an infinite interval. On one hand of the interval, the volatility behaves
asymptotically as that of a CEV model, blowing up or going to zero according to a power law. On the
other side of the interval the volatility goes to zero linearly like in a quadratic model. If on the CEV
endpoint the volatility blows up, there is no absorption, while if the volatility decays with a power be-
tween 1/2 and 1 there is absorption at finite rate. The other case of decay with power law between 0 and
1/2 is not acceptable because boundary conditions need to be reflective in this case, which implies the
existence of arbitrage opportunities. Hunch-back shapes with a local minimum and a local maximum are
BLACK-SCHOLES GOES HYPERGEOMETRIC 5
FIGURE 1. Examples of local volatility functions for a quadratic model.
FIGURE 2. Examples of local volatility unctions for the CEV model.
possible. The 7 parameters single out the interval endpoints, the blow up or decay rate at one end and
the location of the local minimum and the local maximum. This representation is a over-simplification
as the minimum and maximum disappear in certain parameter ranges while only inflection points persist.
The inflection points also disappear in other parameter ranges. Hence, our 7 parameter model supports a
varied zoology of skews, smiles, frowns, skews with a moderate upside twist. It also supports both cases
with and without absorption at F=0.
Additional extensions are possible. For instance, one can apply a deterministic time change and still
retain solvability. It is also possible to apply stochastic time changes and arrive to solvable extensions of
the variance-gamma model, but we refer to forthcoming articles for a discussion of this and other related
topics.
4. RECOVERING EXACT SOLUTIONS IN THE LITERATURE
In this section we show that the known exact solutions in the literature, namely quadratic and CEV
models, can all be rediscovered as particular cases of our general formula for the Bessel family where
we make use of the above solutions to the underlying x space process with β =
1
2
, λ
1
= 0 and λ ≡ λ
0
.
6 CLAUDIO ALBANESE, GIUSEPPE CAMPOLIETI, PETER CARR, AND ALEXANDER LIPTON
FIGURE 3. Examples of local volatility for the CIR family of solveable models.
Without loss of generality, we can fix ν
0
= 2. Let’s specialize further to the case where
(4.1) F(x) =
¯
F −a

2
−1
(

2ρx)

2
−1
(

2ρx)
which leads to a process for the forward price F with volatility
(4.2) σ(F) =
a

X(F)


2
−1
(

2ρX(F))

2
,
where x = X(F) is the inverse of the function in equation (4.1). In this family, a and ρ are positive,
¯
F is
arbitrary and λ > 2. The function F(x) maps the half line x ∈ [0, ∞) into F ∈ (−∞,
¯
F], where F(x) is
a strictly monotonically increasing function with dF(x)/dx = σ(F(x))/ν(x). This solution region can
be inverted so that F ∈ [
¯
F, ∞). This is accomplished by either replacing a by −a in equation (4.1) or
by applying a linear change of variables that maps F into 2
¯
F − F. In this special case, we make use of
the generating function in equation (3.4), with the choice q
2
= 0, and formula (2.9) reduces to
(4.3) U(F, t; F
0
, 0) =
e
−ρt−(X(F)+X(F0))/2t
at
X(F)


2
−1
(

2ρX(F))

3

2
−1
(

2ρX(F
0
))

2
−1

X(F)X(F
0
)
t

.
We note that this density integrates exactly to unity in F space (i.e. no absorption).
4.1. Quadratic volatility models. Pricing kernels for quadratic volatility models are readily obtained
as a subset of the above general family with the special choice of parameter λ = 3. After making the
substitution F →2
¯
F −F and setting a = (
¯
F −
¯
¯
F)/π the transformation function F(x) becomes
(4.4) F(x) =
¯
F +
(
¯
F −
¯
¯
F)
π
K1
2

0

x/2)
I1
2

0

x/2)
=
¯
F +
(
¯
F −
¯
¯
F)
exp(σ
0

x) −1
where σ
0
> 0. Here, we assume that
¯
F >
¯
¯
F. The inverse transformation X(F) is given by
(4.5) X(F) = (1/σ
2
0
) log
2
[1 + (
¯
F −
¯
¯
F)/(F −
¯
F)],
and the volatility function σ(F) is obtained by insertion into equation (4.2) while using the Bessel func-
tion of order
1
2
,
(4.6) σ(F) =
σ
0
(
¯
F −
¯
¯
F)
(F −
¯
F)(F −
¯
¯
F).
BLACK-SCHOLES GOES HYPERGEOMETRIC 7
Inserting the expression (4.5) into equation (4.3), one obtains the pricing kernel
(4.7) U(F, t; F
0
, 0) =
2e
−σ
2
0
t/8
σ(F)

2πt

(F
0

¯
F)(F
0

¯
¯
F)
(F −
¯
F)(F −
¯
¯
F)
e
−(φ(F)
2
+φ(F0)
2
)/2σ
2
0
t
sinh

φ(F
0
)φ(F)
σ
2
0
t

where φ(F) ≡ log((F −
¯
¯
F)/(F −
¯
F)). In the special case of a volatility function with a double root,
i.e.
(4.8) σ(F) = σ
0
(F −
¯
F)
2
the pricing kernel is computed by taking the limit as
¯
¯
F →
¯
F, and one finds
U(F, t; F
0
, 0) =
1
σ
0

2πt
(F
0

¯
F)
(F −
¯
F)
3
¸
e

(F−
¯
F)
−1
−(F0−
¯
F)
−1

2
/2σ
2
0
t
−e

(F−
¯
F)
−1
+(F0−
¯
F)
−1

2
/2σ
2
0
t

. (4.9)
4.2. Lognormal models. The pricing kernel for the log-normal Black-Scholes model with σ(F) = σ
0
F
is a particular case of the above formula for the quadratic model. The derivative with respect to F of the
quadratic volatility function in (4.6), evaluated at F =
¯
F, is σ
0
. Taking the limit
¯
¯
F →−∞(or
¯
¯
F <<
¯
F),
while holding the other parameters fixed, one obtains σ(F) = σ
0
(F −
¯
F). The pricing kernel in (4.7)
gives the kernel for the log-normal model in the limit
¯
¯
F →−∞, i.e.
(4.10) U(F, t; F
0
, 0) =
1
(F −
¯
F)σ
0

2πt
exp
¸

log((F
0

¯
F)/(F −
¯
F)) −
σ
2
0
2
t

2


2
0
t

.
4.3. CEV model. The constant-elasticity-of-variance (CEV) model is recovered in the limiting case as
ρ →0. Assume λ > 2 and let θ > 0 be defined so that λ = θ
−1
+ 2. The transformation F = F(x)
(4.11) F(x) =
¯
F + (σ
2
0
x)
−(2θ)
−1
has inverse x = X(F) given by
(4.12) X(F) = σ
−2
0
(F −
¯
F)
−2θ
,
for any constant
¯
F. The volatility function for this model is
(4.13) σ(F) =
σ
0
|θ|
(F −
¯
F)
1+θ
.
In the limit ρ → 0, the Laplace transform ˆ v(X(F), 0) = 1, which implies that the numeraire change is
trivial in this case. The pricing kernel can be evaluated by substitution into the general formula (2.9), and
after collecting terms, it turns out to be
U(F, t; F
0
, 0) =
|θ|
σ
2
0
t
(F
0

¯
F)
1
2
(F −
¯
F)
3
2
+2θ
e

(F−
¯
F)
−2θ
+(F0−
¯
F)
−2θ

/2σ
2
0
t
I 1

(F −
¯
F)(F
0

¯
F)

−θ
σ
2
0
t

. (4.14)
This formula was derived in the case θ > 0, for which the limiting value F =
¯
F is not attained and the
density is easily shown to integrate to unity (i.e. no absorption occurs and the density also vanishes at
the endpoint F =
¯
F). We note that the same formula solves the forward pricing equation for θ < 0,
leading to the same Bessel equation of order ±(2θ)
−1
. In the range θ < 0, however, the properties of
the above pricing kernel are generally more subtle. In particular, one can show that the density integrates
to unity for all values θ < −1/2, hence no absorption occurs for θ ∈ (−∞, −1/2). The boundary
conditions for the density can be shown to be vanishing at F =
¯
F (i.e. paths do not attain the lower
endpoint) for all θ < −1. In contrast, for θ ∈ (−1, −1/2) the density becomes singular at the lower
endpoint F =
¯
F (hence this corresponds to the case that the density has an integrable singularity for
which paths can also attain the lower endpoint, but are not absorbed). For the special case of θ = −1/2
the formula gives rise to absorption. [Note that only for the range θ ∈ (−1/2, 0) the above pricing kernel
is not useful since it gives rise to a density that has a non-integrable singularity at F =
¯
F. In this case,
8 CLAUDIO ALBANESE, GIUSEPPE CAMPOLIETI, PETER CARR, AND ALEXANDER LIPTON
however, another solution that is integrable is obtained by only replacing the order (2θ)
−1
by −(2θ)
−1
in the Bessel function. The latter solution for the density does not integrate to unity and hence gives rise
to absorption which can be of use to price options in a credit setting.] The special case of θ = −1 gives
a nonzero constant value at the lower endpoint, and recovers the Wiener process with reflection and no
absorption on the interval [
¯
F, ∞) with
U(F, t; F
0
, 0) =
1
σ
0

2πt

e
−(F−F0)
2
/2σ
2
0
t
+e
−(F+F0−2
¯
F)
2
/2σ
2
0
t

. (4.15)
5. BARRIER OPTIONS
The original motivation of two of us, C.A. and G.C., as we engaged in this project, was to stream-
line the derivation of pricing formulas for barrier options for our class of financial engineering master
students. The general expression for the pricing kernel in this article gives in fact a simple and straight-
forward derivation of pricing formulas for barrier options, by allowing one to reduce to the case of a
standard Brownian motion in x space.
Consider as an example a down-and-out option with barrier at F = H within the Black-Scholes model
with σ(F) = σ
0
F. This reduces to the driftless Wiener process with volatility ν(x) =

2, by means of
the transformation where
(5.1) x = X(F) = (

2/σ
0
) log F
with inverse F = F(x) = e
σ0x/

2
. Specializing equation (2.9) gives
(5.2) U(F, t; F
0
, 0) =

2
σ
0
F
exp
¸
1
2
log(F
0
/F) −
σ
2
0
8
t

u(X(F), t; X(F
0
), 0) .
The region x ∈ (−∞, ∞) maps into F ∈ (0, ∞). A barrier located at F = H corresponds to H =
F(x
H
) = e
σ0x
H
/

2
, so x
H
= X(H) = (

2/σ
0
) log H. The upper region F ∈ [H, ∞) maps into
x ∈ [x
H
, ∞). The x-space kernel with absorbing boundary condition at x = x
H
is obtained by the
method of images, as
u(x, t; x
0
, 0) =
1

4πt

e
−(x−x0)
2
/4t
−e
−(x+x0−2x
H
)
2
/4t

. (5.3)
Inserting this kernel into the general pricing formula in (5.2), immediately gives the pricing kernel in F
space:
U
H
(F, t; F
0
, 0) = U(F, t; F
0
, 0)
¸
1 −exp
¸

log(F/H) log(F
0
/H)
σ
2
0
t/2

(5.4)
where U(F, t; F
0
, 0) is the barrier-free pricing kernel
(5.5) U(F, t; F
0
, 0) =
1
σ
0
F

2πt
exp
¸

log(F
0
/F) −
σ
2
0
2
t

2
/2σ
2
0
t

.
A down-and-out call maturing at time T and struck at K > H, has price at time t = 0 given by the
integral
C
DO
(F
0
, K, T) =


H
dF U
H
(F, T; F
0
, 0)(F −K)
+
. (5.6)
where F
0
is the current forward of maturity T. This integral can be evauated in terms of cumulative
normal distribution functions as follows:
C
DO
(F
0
, K, T) = F
0
N(d
1
(F
0
/K)) −KN(d
2
(F
0
/K)) −HN(d
1
(H
2
/F
0
K))
+(KF
0
/H)N(d
2
(H
2
/F
0
K)) (5.7)
where
d
1
(x) =
log x +
1
2
σ
2
0
T
σ
0

T
(5.8)
and d
2
(x) = d
1
(x) −σ
0

T.
BLACK-SCHOLES GOES HYPERGEOMETRIC 9
In the more general case of the other solvable models such as the dual family of 7 parameter models
reducible to the CIR process, one can also obtain analytic closed form solutions for a variety of exotic
payoffs, including simple barrier options. Based on our general results, the derivation of pricing formulas
is straightforward and will be presented elsewhere.
REFERENCES
[1] Karoui Geman and Rochet. Changes in numeraire, changes in probability measure and option pricing. Journal of Applied
Probability, 1995.
[2] M. Schroder. Changes in numeriare for pricing futures, forwards and options. Review of Financial Studies, 1999.
[3] P. Carr, A. Lipton, and D. Madan. The reduction method for valuing derivative securities. working paper., April 2000.
[4] A. Lipton. Interactions between mathematics and finance: Past, present and future. Risk Math Week 2000, New York., November
2000.
[5] C. Albanese and G. Campolieti. Extensions of the black-scholes formula. working paper, www.mathpoint.ca, 2001.
[6] C. Albanese and G. Campolieti. New families of integrable diffusions. working paper, www.mathpoint.ca, 2001.
[7] D. Davydov and V. Linetsky. The valuation and hedging of barrier and lookback options under the cev process. working paper,
forthcoming in Management Science, 1999.
[8] M. Abramowitz and I.A. Stegun. Handbook of Mathematical Functions. 1972.
[9] V. Giorno, A.G. Nobile, L.M. Ricciardi, and L. Sacerdote. Some remarks on the rayleigh process. Journal of Applied Proba-
bility, 1988.
6. APPENDIX
In this appendix we derive the main formula in section 2. Consider the situation of a generic pricing
measure whereby the process for x
t
obeys the equation
(6.1) dx
t
= µ(x
t
)dt +ν(x
t
)dW
t
for some drift µ(x). Then the process g
t
defined in (2.7) satisfies the equation
(6.2) dg =

ρ −µ
ˆ u
x
ˆ u

2
¸

ˆ u
x
ˆ u

2

1
2
ˆ u
xx
ˆ u
¸
gdt +σ
g
gdW
where
(6.3) σ
g
= −
ˆ u
x
ν
ˆ u
is defined as the log-normal volatility of g. Note that in this appendix we write the function σ
g
(with the
superscript g) to mean the volatility function of the underlying g
t
. This volatility is of course a different
function from the volatility function σ(F) of the forward price. Note also that here the subscript variable
x stands for differentiation and subscript xx stands for double differentiation of the function with respect
to x, respectively. Substituting equation (2.4)
(6.4)
ν
2
2
ˆ u
xx
= ρˆ u −λˆ u
x
into this equation, we find that
(6.5)
dg
g
=

µ −λ
ν
σ
g
+ (σ
g
)
2

dt +σ
g
dW.
To demonstrate that g defines a forward price process, consider this equation in the original forward
measure where the forward price F follows a martingale process. In this case, using Ito’s Lemma on the
mapping x
t
= X(F
t
) and equation 1.1 we arrive at an SDE of the form in equation (6.1) with drift given
by
(6.6) µ(x) =
σ(F)
2
2
d
dF
dX(F)
dF
=
σ(F)
2
2
d
dF
ν(x)
σ(F)
where the monotonic mapping dX(F)/dF = ν(x)/σ(F) has been used. Using the chain rule for differ-
entiation, and expressing all functions in terms of x, we then have
(6.7) µ(x) =
σν
2
d
dx

ν
σ

=
ν
2

ν
x

ν
σ
σ
x

,
10 CLAUDIO ALBANESE, GIUSEPPE CAMPOLIETI, PETER CARR, AND ALEXANDER LIPTON
where σ ≡ σ(F(x)) is the volatility function for the forward price F. Hence, by substitution the drift of
g in the forward measure given by equation 6.5 is
(6.8)
µ −λ
ν
σ
g
+ (σ
g
)
2
=
¸
λ +
ν
2
2
σ
x
σ

1
2
νν
x

ˆ u
x
ˆ u

2

ˆ u
x
ˆ u

2
.
By using the volatility function of the forward price defined in terms of the nonlinear transformation in
the x variable, namely
(6.9) σ =
σ
0
ν(x)e
−2

x λ(y)dy
ν(y)
2
ˆ u(x, ρ)
2
,
we find
(6.10)
σ
x
σ
=
ν
x
ν


ν
2

2ˆ u
x
ˆ u
.
Substituting into (6.8), we find that the drift of g under the forward measure vanishes. Hence, g describes
a forward price process, as stated.
Next, consider equation (6.5) again, but this time under the measure having the forward g as numeraire.
Under this pricing measure the price of risk is σ
g
and
(6.11) dg = (σ
g
)
2
gdt +σ
g
gdW.
Comparison with equation (6.5) shows that under this measure the drift of the underlying process x
t
is
λ, as stated. This implies that the representation (2.9) for the pricing kernel is correct and completes the
proof.
We refer the reader interested in gaining further insight into this derivation to our article [6]. There
we provide a proof of this result, which is more elaborate and fully constructive, and is not based on the
above stochastic analysis argument.
CLAUDIO ALBANESE, DEPARTMENT OF MATHEMATICS, 100 ST. GEORGE STREET, UNIVERSITY OF TORONTO, M5S
3G3, TORONTO, CANADA
E-mail address: albanese@math.utoronto.ca
GIUSEPPE CAMPOLIETI, MATH POINT LTD., 720 SPADINA AV., TORONTO, M5S 2T9 ONTARIO, CANADA
E-mail address: jcampoli@mathpoint.ca
PETER CARR, INDEPENDENT CONSULTANT.
E-mail address: pcarr@nyc.rr.com
ALEXANDER LIPTON, FX PRODUCT DEVELOPMENT, DEUTSCHE BANK, 31 WEST 52ND STREET, 13TH FLOOR, NEW
YORK NY 10019
E-mail address: alex.lipton@db.com

2 .2 CLAUDIO ALBANESE. ρ) = 0. we derive a general pricing formula for the models which are solvable by the reduction method.e.2) ut0 (x.7) dxt = (λ0 + λ1 xt )dt + ν0 xβ dW.2) with an arbitrary final time condition at t = 0. t 1 which are solvable for β = 0. Let us consider the homogeneous part of this equation as given by (2. Solvable pricing models can be constructed starting from the price function of a generic European style option written on x. Lipton and Madan address the question of whether it is possible to relax the condition of stationarity and find more general processes with drift and volatility both dependent on calendar time as well as on F which still reduce to the Wiener process and they find that the general solution admits as many as eleven time dependent functions. in the Appendix we give a streamlined verification of its validity and refer to our other papers for a constructive derivation. AND ALEXANDER LIPTON reduces the problem to a Wiener process xt as the one in (1.4). t0 ) = 0 2 for t ≥ t0 . t.1) dxt = λ(xt )dt + ν(xt )dWt . The case β = 1 is the lognormal (or affine) model leading to the Black-Scholes formula. ρ) + v ν(x)2 vxx (x. t0 − t) dt. The two cases β = 0. PETER CARR. Lipton derives general reducibility conditions to the more general processes (1.5) ¯ σ(F ) = σ0 (F − F )1+θ √ dxt = λ0 dt + ν0 xt dWt . ρ) = ˆ 0 eρ(t0 −t) v(x. 2. Recall that the pricing kernel can be interpreted as the price of a limiting butterfly spread option and is related to the second derivative of a call option written on the state variable x with respect to the ”strike” x0 . t. 1. t0 ) which solves the forward Fokker-Plank equation in the first pair of arguments and the backward Black-Scholes equation in the second pair. [6]. G ENERAL P RICING F ORMULA In this section. we summarize our findings by presenting the general solution formula. reduces to the Bessel process (1. originally derived using reducibility conditions and illustrate its use in a few particular cases. x0 . In [4]. t0 ) + ∞ (2. ρ) + λ(x)ˆx (x. t. t) = δ(x − x0 ) (a Dirac delta function). Since the original derivation is somewhat lengthy. In [3]. In [5]. x0 . A related line of reasoning leading to extensions of the Black-Scholes formula starts from the observation that the CEV models with state dependent volatility specified as follows: (1. 2 namely the Vasicek and the Cox-Ingerssol-Ross (CIR) models. Lipton and Madan for stationary. GIUSEPPE CAMPOLIETI. with final time condition at terminal time t0 = t given by u(x. ¯ with constants θ and F . In this article. a solution v(x. x0 .7) are examples of analytically solvable models for which one can compute the pricing kernel. The Laplace transform of such a function (2. t. Albanese and Campolieti find a general solution to the reducibility conditions by Carr. Assume that the state variable x has a drift λ(x) for which one can find the pricing kernel for the process (2. The pricing kernel is the function u(x. is usually referred to as “time-independent Green’s function” and satisfies a second order ordinary differential equation with Dirac delta function source term δ(x − x0 ).6) by means of a non-linear transformation combined with a measure change.4) −ρˆ(x. t) of the Black-Scholes equation in (2. t. driftless processes. The latter equation can be written as follows: ν(x0 )2 ux0 x0 (x. ˆ v 2 . x0 . x0 . i. 1 correspond to well-known solvable short rate models.3) v (x. t0 ) + λ(x0 )ux0 (x. The processes in (1. Carr.

0) = ν(X(F )) v (X(F ).6) dx dF =± . struck at K and maturing at time t = T can be priced in this model by computing the following integral: ∞ (2. F0 . x0 . ρ) the “generating function”. This is accomplished by means of either integral representations or eigenfunction expansion methods.5) σ(F ) = with arbitrary constant σ0 and where (2. i. Of interest here is the other two families with β = 0. including the lognormal Black-Scholes model as a special subcase. we define a volatility function σ(F ) and an invertible monotonic ˆ transformation F = F (x) and its inverse x = X(F ) such that σ0 ν(X(F )) exp −2 (2. ρ)2 ˆ The two signs correpond to either monotonic increasing or monotonic decreasing transformations. 0) for the overlying forward price F at time t can be evaluated in closed form as the expected reward from a limit butterfly spread contract with delta function payoff (2. we name v (x. X(F0 ).e.9) U (F. t. The freedom in choosing the sign gives rise to 2 families of solutions which are different in the general case. ρ) ˆ Barrier and lookback options can be handled by modifying the underlying kernel in x-space to account for the appropriate boundary conditions. ρ) −ρt ˆ e u(X(F ). t. x(F0 ). t. which is readily transformed into a driftless Wiener process and thus supports only quadratic volatility functions in F space. ν(x) σ(F ) X(F ) λ(s)ds ν(s)2 v (X(F ).7) giving rise to solvable models in case λ1 < 0 which admit as a singular limit 2 a family with different properties in the non mean-reverting limit λ1 = 0. Green’s function methods that are standard in the theory of Sturm-Liouville equations. F OUR FAMILIES OF SOLVABLE MODELS The case β = 1 is the usual lognormal (or affine) model. As we verify in the Appendix. 1 in equation (1. See the articles of Davydov and Linetsky [7] for a discussion in an option pricing context.BLACK-SCHOLES GOES HYPERGEOMETRIC 3 We find that functions v (x. 0). 0) = E0 gt δ(F (xt ) − F ) g0 conditional on the price having value F0 at initial time t = 0. If β = 0 and λ1 = 0 we recover the Wiener process with constant drift. 0) = √ sinh πν0 (e2λ1 t − 1) x0 x ¯ ¯ 2 ν0 λ1 sinh(λ1 t) exp x2 + x2 e2λ1 t ¯ ¯0 λ1 + t 2 λ1 ν0 (e2λ1 t − 1) 2 . ρ) ˆ (F (x) − K)u(x.1) 2 2λ1 sinh(λ1 t) u(x. Hence. The final formula for the pricing kernel in F space is related to the kernel in the underlying x space as follows: (2. ρ) ˆ can be regarded as a forward price process and under the measure with g as a numeraire the state variable xt drifts at rate λ(x). T . t.8) U (F. T . t. the expectation is computed assuming that gt is the numeraire and that the state variable x drifts at rate λ(x). F0 . ρ). the pricing kernel U (F. F0 . σ(F ) v (X(F0 ).10) C(K. Because of this reason. v (X(F0 ). This provides four examples of our methodology to generate exactly solvable models. F0 ) = e−ρT X(K) dx v (x. ˆ Armed with a solution v (x. If β = 0 and λ1 < 0 then the kernel in x space can be shown to be given by (3. ρ) solving this equation can be taken as the elementary building blocks for ˆ the construction of solvable pricing models for the F space processes. ρ) ˆ A European call option written on the forward price F0 at current time t = 0. the process (2.7) gt = eρt v (xt . 3. 0). Here.

t.6). Here Iν (z) is the modified Bessel function of order ν and Kν (z) is the associated McDonalds function. 0) = I 2λ0 −1 . x0 ≡ λ0 + λ1 x0 . 2 2 2 x0 ν0 t/2 ν0 t ν0 The generating function is 1 2λ0 2 ν0 (3. Here Wk.4) reduces to Whittaker’s equation and generating functions have the general form (3. The other case of decay with power law between 0 and 1/2 is not acceptable because boundary conditions need to be reflective in this case. 2 2λ1 2 2 (λ1 ν0 )2 with arbitrary constants q1 . there is no absorption. 0) = ct xe−λ1 t x0 1 2λ0 2 ( ν2 0 −1) exp −ct (x0 eλ1 t + x) I 2λ0 −1 2ct 2 ν0 xx0 eλ1 t .q2 . The general solution of equation (2. In this case we obtain a dual family with 6 adjustable parameters. z) is Kummer’s function[8]. For each choice of these parameters.6) v (x.4) v (x. t. ρ) = q1 M ˆ − ρ 1 (λ0 + λ1 x)2 .m ν0 − 2λ1 2 x ν0 for arbitrary constants q1 . x0 . The pricing kernel for the state variable x corresponds to that of the short rate CIR model.3) u(x. AND ALEXANDER LIPTON where x ≡ λ0 + λ1 x. which implies the existence of arbitrage opportunities. In Figure 1 we show There is a way of gaining a visual understanding of the geometric meaning of the 7 parameters which perhaps over-simplifies the picture but is intriguing. ν0 λ1 ν0 2 The 7 parameter family which reduces to the CIR model has a local volatility function defined on either an interval or on a half line and behaves asymptotically as the CEV volatility on one hand and as a quadratic model on the other.7) k= 2 + . PETER CARR. there are two different families corresponding to the two different choices of the sign in equation (2. 2 2λ1 2 (λ1 ν0 )2 + q2 (λ0 + λ1 x)M − ρ 1 1 (λ0 + λ1 x)2 + . . The generating function solves a special case of the confluent ¯ ¯ hypergeometric equation (Hermite’s equation) with general solution (3. The allowed shapes.q2 . with arbitrary constants q1 .5) u(x. then the pricing kernel for the state variable is expressed in terms of modified 2 Bessel functions as follows: 1 2λ0 √ ( 2 −1) −2(x+x0 )/ν 2 t 0 4 xx0 e x 2 ν0 (3. m= 2 − .m (·) are Whittaker functions which can also be expressed in terms of confluent hypergeometric functions or in terms of Kummer functions. ρ) = x−λ0 /ν0 e−λ1 x/ν0 q1 Wk. when confined to a finite intervals. . can be regarded as hybrids between the quadratic and the CEV model. i. Hunch-back shapes with a local minimum and a local maximum are . ρ) = x 2 ˆ (1− ) q1 I 2λ0 −1 2 ν0 8ρx 2 ν0 + q2 K 2λ0 −1 2 ν0 8ρx 2 ν0 . The support of the volatility function can be either a finite or an infinite interval. On one hand of the interval. On the other side of the interval the volatility goes to zero linearly like in a quadratic model. The case β = 1 and λ1 < 0 is more general than the Vasicek case.q2 .[8] This construction gives rise to a dual family with 7 free parameters where λ0 ρ λ0 1 (3. For a derivation see [9]. the confluent hypergeometric function that is regular at z = 0. the CIR process 2 describes the stochastic dynamics of the radial distance of a point whose Cartesian coordinates follow an evolution given by the Vasicek model. GIUSEPPE CAMPOLIETI. In this expression. blowing up or going to zero according to a power law. while if the volatility decays with a power between 1/2 and 1 there is absorption at finite rate. This analogy applies only to integer dimensions.e.m (·) and Mk. Geometrically. If on the CEV endpoint the volatility blows up. the volatility behaves asymptotically as that of a CEV model. Here M (a. and can still be expressed in terms of modified Bessel functions as follows: (3. one can count 6 free dimensionless parameters. x0 . This hibrid shape allows for a great deal of flexibility in reproducing observed volatility skews. 2 where ct ≡ 2λ1 /(ν0 (eλ1 t − 1)).2) v (x. If β = 1 and λ1 = 0.4 CLAUDIO ALBANESE. b. but analytic continuation in the dimension parameter allows one to gain one additional degree of freedom.m ˆ 2 2 − 2λ1 2 x + q2 Mk.

the blow up or decay rate at one end and the location of the local minimum and the local maximum. one can apply a deterministic time change and still retain solvability. 2 .BLACK-SCHOLES GOES HYPERGEOMETRIC 5 F IGURE 1. It also supports both cases with and without absorption at F=0. The 7 parameters single out the interval endpoints. F IGURE 2. smiles. possible. 4. Additional extensions are possible. The inflection points also disappear in other parameter ranges. skews with a moderate upside twist. Examples of local volatility unctions for the CEV model. Hence. For instance. This representation is a over-simplification as the minimum and maximum disappear in certain parameter ranges while only inflection points persist. but we refer to forthcoming articles for a discussion of this and other related topics. λ1 = 0 and λ ≡ λ0 . our 7 parameter model supports a varied zoology of skews. It is also possible to apply stochastic time changes and arrive to solvable extensions of the variance-gamma model. frowns. namely quadratic and CEV models. Examples of local volatility functions for a quadratic model. R ECOVERING EXACT SOLUTIONS IN THE LITERATURE In this section we show that the known exact solutions in the literature. can all be rediscovered as particular cases of our general formula for the Bessel family where we make use of the above solutions to the underlying x space process with β = 1 .

We note that this density integrates exactly to unity in F space (i. Examples of local volatility for the CIR family of solveable models. ∞) into F ∈ (−∞. we make use of the generating function in equation (3. where F (x) is a strictly monotonically increasing function with dF (x)/dx = σ(F (x))/ν(x). Let’s specialize further to the case where √ K λ ( 2ρx) ¯ − a 2 −1 √ (4.1).3) U (F. After making the ¯ ¯ ¯ substitution F → 2F − F and setting a = (F − F )/π the transformation function F (x) becomes √ ¯ ¯ ¯ ¯ (F − F ) K 1 (σ0 x/2) (F − F ) 2 ¯ ¯ √ √ (4. we assume that F > F . 0) = 2 at I λ −1 ( 2ρX(F0 )) 2 3 X(F )X(F0 ) t . F0 . 0 and the volatility function σ(F ) is obtained by insertion into equation (4. The function F (x) maps the half line x ∈ [0.2) σ(F ) = 2.1) F (x) = F I λ −1 ( 2ρx) 2 which leads to a process for the forward price F with volatility a (4. ∞). F ].1) or ¯ by applying a linear change of variables that maps F into 2F − F . F is ¯ arbitrary and λ > 2. In this family. This solution region can ¯ be inverted so that F ∈ [F . AND ALEXANDER LIPTON F IGURE 3. Pricing kernels for quadratic volatility models are readily obtained as a subset of the above general family with the special choice of parameter λ = 3.4) F (x) = F + =F + π I 1 (σ0 x/2) exp(σ0 x) − 1 2 ¯ ¯ where σ0 > 0. with the choice q2 = 0. 4. In this special case. The inverse transformation X(F ) is given by ¯ ¯ ¯ (4. t. a and ρ are positive. Quadratic volatility models. This is accomplished by either replacing a by −a in equation (4.1. Here.2) while using the Bessel func1 tion of order 2 . ¯ (F − F ) . GIUSEPPE CAMPOLIETI. and formula (2. σ0 ¯ ¯ (4. X(F ) I λ −1 ( 2ρX(F )) 2 ¯ where x = X(F ) is the inverse of the function in equation (4. no absorption). Without loss of generality.e.6) σ(F ) = ¯ (F − F )(F − F ). we can fix ν0 = 2.6 CLAUDIO ALBANESE.9) reduces to e−ρt−(X(F )+X(F0 ))/2t X(F ) I λ −1 ( 2ρX(F )) 2 I λ −1 (4. PETER CARR.4).5) X(F ) = (1/σ 2 ) log2 [1 + (F − F )/(F − F )].

Assume λ > 2 and let θ > 0 be defined so that λ = θ−1 + 2. The volatility function for this model is σ0 ¯ (F − F )1+θ . The boundary ¯ conditions for the density can be shown to be vanishing at F = F (i.14) I 2θ . it turns out to be ¯ 1 2 |θ| (F0 − F ) 2 − (F −F )−2θ +(F0 −F )−2θ /2σ0 t ¯ ¯ U (F. The pricing kernel for the log-normal Black-Scholes model with σ(F ) = σ0 F is a particular case of the above formula for the quadratic model. The pricing kernel in (4.6). 0) = sinh 2 ¯ e ¯ )(F − F ) σ0 t σ(F ) 2πt (F − F ¯ ¯ where φ(F ) ≡ log((F − F )/(F − F )). −1/2). no absorption occurs and the density also vanishes at ¯ the endpoint F = F ). F0 . t. [Note that only for the range θ ∈ (−1/2. −1/2) the density becomes singular at the lower ¯ endpoint F = F (hence this corresponds to the case that the density has an integrable singularity for which paths can also attain the lower endpoint. F0 . however. leading to the same Bessel equation of order ±(2θ)−1 .7) U (F. i. Taking the limit F → −∞ (or F << F ). one can show that the density integrates to unity for all values θ < −1/2. F0 . F0 .3.9) ¯ 1 (F0 − F ) − √ ¯ )3 e σ0 2πt (F − F −e− ¯ ¯ (F −F )−1 −(F0 −F )−1 2 2 2 /2σ0 t ¯ ¯ (F −F )−1 +(F0 −F )−1 2 /2σ0 t . for which the limiting value F = F is not attained and the density is easily shown to integrate to unity (i.13) σ(F ) = |θ| In the limit ρ → 0. (4. The derivative with respect to F of the ¯ ¯ ¯ ¯ quadratic volatility function in (4. In contrast.8) U (F. The constant-elasticity-of-variance (CEV) model is recovered in the limiting case as ρ → 0. is σ0 . For the special case of θ = −1/2 the formula gives rise to absorption. The transformation F = F (x) (4. which implies that the numeraire change is ˆ trivial in this case. .11) has inverse x = X(F ) given by (4. ¯ σ(F ) = σ0 (F − F )2 ¯ ¯ the pricing kernel is computed by taking the limit as F → F . 0) = (4. −1 2 ¯ F (x) = F + (σ0 x)−(2θ) ¯ for any constant F . 2 σ0 t ¯ This formula was derived in the case θ > 0. 4.e.9). for θ ∈ (−1.12) −2 ¯ X(F ) = σ0 (F − F )−2θ .3). (4. the properties of the above pricing kernel are generally more subtle. ¯ while holding the other parameters fixed.2.10) U (F. In particular. We note that the same formula solves the forward pricing equation for θ < 0. In the special case of a volatility function with a double root. t. 0) = 1. and one finds (4. paths do not attain the lower endpoint) for all θ < −1. In this case. t. evaluated at F = F . one obtains the pricing kernel 2 ¯ ¯ 2 φ(F0 )φ(F ) 2e−σ0 t/8 (F0 − F )(F0 − F ) −(φ(F )2 +φ(F0 )2 )/2σ0 t √ (4.5) into equation (4. CEV model. the Laplace transform v (X(F ).e.BLACK-SCHOLES GOES HYPERGEOMETRIC 7 Inserting the expression (4. 0) the above pricing kernel ¯ is not useful since it gives rise to a density that has a non-integrable singularity at F = F . In the range θ < 0. Lognormal models.e. 0) = 2 e 3 +2θ ¯ σ0 t (F − F ) 2 ¯ ¯ −θ (F − F )(F0 − F ) 1 (4. one obtains σ(F ) = σ0 (F − F ).e.7) ¯ gives the kernel for the log-normal model in the limit F → −∞. and after collecting terms. i. 4. 0) = 1 √ exp − ¯ )σ0 2πt (F − F σ2 ¯ ¯ log((F0 − F )/(F − F )) − 0 t 2 2 2 2σ0 t . The pricing kernel can be evaluated by substitution into the general formula (2. t. hence no absorption occurs for θ ∈ (−∞. but are not absorbed).

GIUSEPPE CAMPOLIETI. t. as 2 2 1 u(x. 0) 1 − exp − log(F/H) log(F0 /H) 2 σ0 t/2 2 σ0 t 2 2 where U (F. was to streamline the derivation of pricing formulas for barrier options for our class of financial engineering master students. ∞). ∞) maps into x ∈ [xH . d1 (x) = 2 log x + 1 σ0 T √2 σ0 T C DO (F0 . immediately gives the pricing kernel in F space: (5. where F0 is the current forward of maturity T .4) U H (F. T ) = F0 N (d1 (F0 /K)) −KN (d2 (F0 /K)) − HN (d1 (H 2 /F0 K)) +(KF0 /H)N (d2 (H 2 /F0 K)) . x0 . t. t. and G. F0 . F0 . U (F.] The special case of θ = −1 gives a nonzero constant value at the lower endpoint.5) 2 /2σ0 t . by means of the transformation where √ (5. ∞) maps into F ∈ (0. σ0 2πt 5. σ0 F 2πt A down-and-out call maturing at time T and struck at K > H. and recovers the Wiener process with reflection and no ¯ absorption on the interval [F . t. T ) = H dF U H (F. X(F0 ).A. 0) = √ (5. by allowing one to reduce to the case of a standard Brownian motion in x space. AND ALEXANDER LIPTON however. t. The general expression for the pricing kernel in this article gives in fact a simple and straightforward derivation of pricing formulas for barrier options. F0 . F0 .2). ∞) with (4. The x-space kernel with absorbing boundary condition at x = xH is obtained by the method of images.6) C DO (F0 . T .8 CLAUDIO ALBANESE. Specializing equation (2.9) gives √ 2 1 σ2 exp log(F0 /F ) − 0 t u(X(F ). 0) . another solution that is integrable is obtained by only replacing the order (2θ)−1 by −(2θ)−1 in the Bessel function. t. t. BARRIER O PTIONS The original motivation of two of us. A barrier located at F = H corresponds to H = √ √ F (xH ) = eσ0 xH / 2 . C. has price at time t = 0 given by the integral U (F. 0) = σ0 F 2 8 √ The region x ∈ (−∞.15) U (F.3) e−(x−x0 ) /4t − e−(x+x0 −2xH ) /4t . ∞).8) √ and d2 (x) = d1 (x) − σ0 T . t. K. 0) is the barrier-free pricing kernel (5. 4πt Inserting this kernel into the general pricing formula in (5. PETER CARR. This reduces to the driftless Wiener process with volatility ν(x) = 2.C. F0 . The upper region F ∈ [H. F0 .. F0 . 0) = U (F.1) x = X(F ) = ( 2/σ0 ) log F with inverse F = F (x) = eσ0 x/ (5. This integral can be evauated in terms of cumulative normal distribution functions as follows: (5. 0) = 1 √ exp − log(F0 /F ) − ∞ (5. 0)(F − K)+ . 0) = 2 2 2 1 ¯ 2 √ e−(F −F0 ) /2σ0 t + e−(F +F0 −2F ) /2σ0 t . so xH = X(H) = ( 2/σ0 ) log H. K.2) 2 .7) where (5. The latter solution for the density does not integrate to unity and hence gives rise to absorption which can be of use to price options in a credit setting. as we engaged in this project. Consider as an example a down-and-out option with barrier at F = H within the Black-Scholes model √ with σ(F ) = σ0 F .

New families of integrable diffusions. forthcoming in Management Science. Linetsky. [5] C. ν ν2 uxx = ρˆ − λˆx ˆ u u 2 dg = ρ−µ ux ˆ + ν2 u ˆ ux ˆ u ˆ − 1 uxx ˆ 2 u ˆ gdt + σ g gdW To demonstrate that g defines a forward price process. forwards and options. Nobile.ca. Albanese and G. Davydov and V.7) µ(x) = σν d ν 2 dx σ = ν ν νx − σ x . www. L. Carr.4) into this equation. Giorno. consider this equation in the original forward measure where the forward price F follows a martingale process. we then have (6. April 2000. 1999. 1972.5) dg = g µ−λ g σ + (σ g )2 dt + σ g dW. we find that (6.4) (6. [2] M. [7] D. R EFERENCES [1] Karoui Geman and Rochet. November 2000. In this case. A PPENDIX In this appendix we derive the main formula in section 2. [6] C.6) µ(x) = σ(F )2 d dX(F ) σ(F )2 d ν(x) = 2 dF dF 2 dF σ(F ) where the monotonic mapping dX(F )/dF = ν(x)/σ(F ) has been used. New York.3) σg = − (6.1 we arrive at an SDE of the form in equation (6. Substituting equation (2. respectively. Abramowitz and I. 1988. working paper.G. Campolieti. [4] A.. and D. changes in probability measure and option pricing. A. This volatility is of course a different function from the volatility function σ(F ) of the forward price. Using the chain rule for differentiation. Ricciardi. 1999. using Ito’s Lemma on the mapping xt = X(Ft ) and equation 1..1) dxt = µ(xt )dt + ν(xt )dWt 2 for some drift µ(x). including simple barrier options. Note that in this appendix we write the function σ g (with the superscript g) to mean the volatility function of the underlying gt . working paper. Risk Math Week 2000.2) where ux ν ˆ u ˆ is defined as the log-normal volatility of g. the derivation of pricing formulas is straightforward and will be presented elsewhere. Then the process gt defined in (2. www.ca. 1995. Stegun. Journal of Applied Probability. Changes in numeriare for pricing futures. Lipton. working paper. Schroder. working paper. Madan. 2001. and expressing all functions in terms of x. The valuation and hedging of barrier and lookback options under the cev process. The reduction method for valuing derivative securities. [8] M. present and future.M. [9] V. Note also that here the subscript variable x stands for differentiation and subscript xx stands for double differentiation of the function with respect to x.mathpoint. Interactions between mathematics and finance: Past. Sacerdote. [3] P.1) with drift given by (6.7) satisfies the equation (6. Changes in numeraire. Review of Financial Studies. Some remarks on the rayleigh process. Albanese and G. Handbook of Mathematical Functions. one can also obtain analytic closed form solutions for a variety of exotic payoffs. 2 σ .mathpoint.A. Campolieti. Consider the situation of a generic pricing measure whereby the process for xt obeys the equation (6. 2001. Based on our general results. Lipton. 6.BLACK-SCHOLES GOES HYPERGEOMETRIC 9 In the more general case of the other solvable models such as the dual family of 7 parameter models reducible to the CIR process. Extensions of the black-scholes formula. and L. A. Journal of Applied Probability.

10) (6. C ANADA E-mail address: albanese@math. consider equation (6. by substitution the drift of g in the forward measure given by equation 6. Under this pricing measure the price of risk is σ g and (6. as stated. 31 W EST 52 ND S TREET.10 CLAUDIO ALBANESE. M5S G IUSEPPE C AMPOLIETI .utoronto. D EPARTMENT OF M ATHEMATICS . We refer the reader interested in gaining further insight into this derivation to our article [6].5 is (6. namely (6. PETER CARR. we find that the drift of g under the forward measure vanishes. This implies that the representation (2. C ANADA E-mail address: jcampoli@mathpoint.com . Hence. M ATH P OINT LTD . g describes a forward price process. Hence. = σ ν ν u ˆ Substituting into (6. 13 TH F LOOR .rr. but this time under the measure having the forward g as numeraire.ca OF 2 σ0 ν(x)e σ= u(x. Comparison with equation (6. I NDEPENDENT CONSULTANT.com A LEXANDER L IPTON . AND ALEXANDER LIPTON where σ ≡ σ(F (x)) is the volatility function for the forward price F . 100 S T. ρ)2 ˆ −2 x λ(y)dy ν(y)2 . Next.8). FX P RODUCT D EVELOPMENT. There we provide a proof of this result. T ORONTO .9) for the pricing kernel is correct and completes the proof. U NIVERSITY 3G3.ca P ETER C ARR .8) µ−λ g ν 2 σx 1 ux ˆ ux ˆ σ + (σ g )2 = λ + − ννx + ν2 .9) we find σx νx 2λ 2ˆx u − 2 − . E-mail address: pcarr@nyc. D EUTSCHE BANK . 720 S PADINA AV. T ORONTO .. T ORONTO .. C LAUDIO A LBANESE .11) dg = (σ g )2 gdt + σ g gdW. G EORGE S TREET. N EW YORK NY 10019 E-mail address: alex.5) shows that under this measure the drift of the underlying process xt is λ. GIUSEPPE CAMPOLIETI. ν 2 σ 2 u ˆ u ˆ By using the volatility function of the forward price defined in terms of the nonlinear transformation in the x variable.lipton@db. M5S 2T9 O NTARIO .5) again. and is not based on the above stochastic analysis argument. which is more elaborate and fully constructive. as stated.

Sign up to vote on this title
UsefulNot useful