a

r
X
i
v
:
1
3
0
3
.
6
0
9
0
v
1


[
q
-
f
i
n
.
P
R
]


2
5

M
a
r

2
0
1
3
Volatility Swap Under The SABR Model
Simon Bossoney
March 26, 2013
1
Contents
1 Introduction 3
1.1 The SABR Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Volatility Swaps 3
2.1 Principle of The swaps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Getting Rid of The Square Root . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 A dierential Equation for . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Solving the dierential equation 5
4 Recovering The Square Root 6
4.1 Value at t = t
0
+ T. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
5 Verication By Functional Calculus 9
6 Conclusion 10
2
Abstract
The SABR model is shortly presented and the volatility swap explained. The fair value for a
volatility swap is then computed using the usual theory in nancial mathematics. An analytical
solution using conuent hypergeometric functions is found. The solution is then veried using Rama
Conts functional calculus.
1 Introduction
Mathematical models for the computation of nancial derivatives have gained some importance in the
nancial industry. A breakthrough was realized with the Black & Scholes model. Though widely applied,
this model soon began to show its limits and shortcomings. Among other deciencies, the hypothesis of
constant volatility for the underlying appeared as a much to strong assumption. It is usually now dealt
with by quoting a so-called implied volatility, which, when inserted in the Black & Scholes equation for
the option pricing, gives the traded prices for the latter.
1.1 The SABR Model
As another Ansatz, models with time-dependent volatilities or even stochastic volatilities where intro-
duced. A famous example is the Heston-model, where the variance of the underlying is itself stochastic.
Another widely used model is the SABR-model, where this time the volatility is stochastic. Under the
forward measure, the forward prices F
t
for an underlying is governed by the stochastic process
dF
t
=
t
F
t
dW
t
,
d
t
=
t
dZ
t
, (1)
where is some positive constant and where W
t
and Z
t
are Wiener processes with a correlation coecient
1 1 [1]. In general, the rst stochastic dierential equation reads dF
t
=
t
F

t
dW
t
, with 0 1,
but we are going to work with = 1 throughout our computations. This will have no inuence in the
following, since in our denition of the volatility swap, only the stochastic process governing the volatility
will be considered. European call options are then priced under the forward measure as
S(F, , K, t) = p(t, T)E
_
(F
T
K)
+
|F
t
= F,
t
=
_
, (2)
where p(t, T) is the price at t of a zero-coupon bond paying 1 at maturity T (see [2]).
2 Volatility Swaps
As it became obvious that volatilities and variances for various underlyings behave in a stochastic manner
on the market, several derivative instruments dealing with those quantities where introduced.
2.1 Principle of The swaps
A variance swap consists of a strike K and a realized average variance 1/T
_
t0+T
to

2
t
dt over some time-
period t
0
, t
0
+ T. The pay-o function is then the dierence of these two quantities. There are known
closed form solution for variance swaps, but it seems, that these instruments are rather volatile and risky.
A volatility swap is a bet on a realized averaged volatility 1/T
_
_
t0+T
to

2
t
dt over some time-period t
0
, t
0
+
T. By taking the square root of the realized variance, this instrument has a somewhat less volatile behavior
than a variance swap. But mathematically, it is exactly this square root that makes computations more
dicult.
Here we propose to compute exact solutions to the fair value of a volatility swap
p(t, T)
_
E
_
1
T

_
t0+T
to

2
t
dt|
t
=
_
K
_
,
under the SABR model with t [t
0
, t
0
+ T]. In the following, we shall write

t
:= E
_
1
T

_
t0+T
to

2
t
dt|
t
=
_
,
so that the fair price of a volatility swap becomes p(t, T)(
t
K). The task is now to compute
t
.
3
2.2 Getting Rid of The Square Root
It is a well-known fact, that for y R
+

y =
1

_

0
1 exp(x
2
y)
x
2
dx (3)
we therefore obtain

t
= E
_
1
T

_
t0+T
to

2
t
dt|
t
=
_
=
1
T

E
_
_

0
1 exp(x
2
_
t0+T
to

2
t
dt)
x
2

t
=
_
=
1
T

_

0
1 E
_
exp(x
2
_
t0+T
to

2
t
dt)

t
=
_
x
2
. (4)
We will hence have to compute the quantity
f(t, , x) := E
_
exp(x
2
_
t0+T
to

2
t
dt)

t
=
_
(5)
and integrate along x as in the previous equality.
Note, that x is not stochastic but a variable of integration. Also, we did not fully justify the interchange
of the order of integration, which shall be justied a posteriori. Also, if t
0
t t
0
+ T, then f will also
depend on the realized volatility
t
:=
_
t
t0

2
t
dt "up to t". We suppose at rst, that t
0
t t
0
+ T and
write
f(t, ,
t
, x) := E
_
exp(x
2
_
t0+T
to

2
t
dt)

t
=
_
= E
_
exp(x
2

t
) exp(x
2
_
t0+T
t

2
t
dt)

t
=
_
= exp(x
2

t
)E
_
exp(x
2
_
t0+T
t

2
t
dt)

t
=
_
exp(x
2

t
)(t, , x). (6)
2.3 A dierential Equation for
The Fenman-Kac formula gives the correspondence between expectation computations for a stochastic
process and dierential equations. For a stochastic process governed by the equation
dX
t
= (X
t
, t)dt + (X
t
, t)dW
t
, (7)
the function
u(x, t) := E
_
_
T
t
exp(
_
s
t
V (X

)d)f(X
s
, s)ds + exp(
_
T
t
V (X
s
)ds)(X
T
)
_
(8)
is the solution to the dierential equation

t
u(x, t) + (x, t)
x
u(x, t) +

2
(x, t)
2

2
x
u(x, t) V (x, t)u(x, t) + f(x, t) = 0, (9)
subject to the terminal condition
u(x, T) = (x). (10)
4
In our situation, we have (
t
, t) = 0 = f(
t
, t), (x) = 1, V (
t
) = x
2

2
t
and the stochastic diusion
term is
t
, so that (t, , x) is, for a given x, solution to the dierential equation

t
(t, , x) +

2

2
2

2

(t, , x) x
2

2
(t, , x) = 0 (11)

t
(t, , x) =
2

(t, , x)
2x
2

2
(t, , x), (12)
subject to the terminal condition
(T, , x) = 1. (13)
Note, that here, x is only a parameter, over which we will have to integrate later on. Here, x is
indexing a family of dierential equations. We shall hence drop it from the variables in . Writing
y :=

2x/ and (t, y) := (t, (y)), we are to solve

t
(t, y) = y
2

2
y
(t, y) y
2
(t, y), (14)
subject to the terminal condition
(T, y) = 1. (15)
Clearly, we will have (t, , x) = (t,

2x

).
3 Solving the dierential equation
If it wasnt for the missing y
y
term, the dierential operator appearing on the right-hand side of equation
14 would be the modied Bessel dierential operator, whose solutions are well-known. In order to retrieve
this equation, we dene g(t, y) := y
1/2
(t, y). This function satises the equation
1
4
g(t, y)
2

t
g(t, y) = y
2

2
y
g(t, y) + y
y
g(t, y) y
2
g(t, y), (16)
subject to the terminal condition
g(T, y) =
1

y
. (17)
The solutions to the equation
k
2
f(y) = y
2

2
y
f(y) + y
y
f(y) y
2
f(y), (18)
are the modied Bessel functions of the rst and second kind I
k
(y) and K
k
(y). The functions I
k
(y) read
I
k
(y) =

m0
1
m!(k + m+ 1)
_
y
2
_
2m+k
. (19)
The terminal condition y
1/2
can be expressed in terms of these functions as
1

y
=
1

n0
(1)
n
(2n 1/2)(n 1/2)
n!
I
2n1/2
(y). (20)
The convergence of this series is pointwise. We may now write the solution to the equation 16 as
g(t, y) =

n0
(1)
n
(2n 1/2)(n 1/2)

2n!
I
2n1/2
(y) exp
_
E
n
(T + t
0
t)
_
. (21)
where E
n
:=

2
2
_
(2n 1/2)
2
1/4
_
=
2
n(2n 1). Consequently,
(t, y) =

n0
(1)
n
(2n 1/2)(n 1/2)
n!
_
y
2
_
1/2
I
2n1/2
(y) exp
_
E
n
(T + t
0
t)
_
. (22)
5
4 Recovering The Square Root
We shall now compute
t
for a volatility swap under the SABR model, by integrating over x. At rst,
we will suppose t
0
t t
0
+ T. The case t < t
0
will be treated afterwards. We thus have

t
=
1
T

_

0
1 exp(x
2

t
)(t, , x)
x
2
dx
=
1
T

_

0
1 exp(x
2

t
) + exp(x
2

t
)
_
1 (t, , x)
_
x
2
dx
=
1
T
_

t
+
1

_

0
exp(x
2

t
)
_
1 (t, , x)
_
x
2
dx
_
=
1
T
_

t
+
1

_

0
exp(x
2

t
)
_
1 (t,

2x

)
_
x
2
dx
_
=
1
T
_

t
+

_

0
exp(

2
t
2
2
y
2
)
_
1 (t, y)
_
y
2
dy
_
To lighten up the notations, we shall write J :=
_

0
exp(

t
2
2
y
2
)
_
1(t,y)
_
y
2
dy. Moreover, we write
a
n
:=
(1)
n
(2n1/2)(n1/2)
n!
and F
n
(y) :=
_
y
2
_
1/2
I
2n1/2
(y). We will split the integral as
J =
_

0
exp(

2
t
2
2
y
2
)
_
1 (t, y)
_
y
2
dy (z :=
2
2

t
)
=
_

0
exp(
y
2
z
)
_
1

n0
a
n
F
n
(y) exp(E
n
(T + t
0
t))
_
y
2
dy
= J
0
+ J

,
where
J
0
=
_

0
exp(
y
2
z
)
_
1 a
0
F
0
(y)
_
y
2
dy and
J

=
_

0
exp(
y
2
z
)
_
n1
a
n
F
n
(y) exp(E
n
(T + t
0
t))
_
y
2
dy.
Observe, that F
n
(y) :=

m0
1
m!(m+2n+1/2)
_
y
2
_
2m+2n
. To compute J
0
, we insert a
0
= 1/2(1/2) =

, and F
0
=

m0
1
m!(m+1/2)
_
y
2
_
2m
. Therefore,
J
0
=
_

0
exp(
y
2
z
)
_
1

F
0
(y)
_
y
2
dy
=
_

0
exp(
y
2
z
)

m1
(1/2)
m!(m+1/2)
_
y
2
_
2m
y
2
dy
=
_

0
exp(
y
2
z
)

m1
(1/2)
m!(m + 1/2)
_
y
2
_
2m1
4
_
y
2
_
2
dy
=

m1
(1/2)
4m!(m+ 1/2)
_

0
exp(
y
2
z
)
_
y
2
_
2m2
dy,
where the interchange of the integral and the sum is justied by the fact, that all terms which are to
integrate have the same sign, so that the convergence of the integral of the sum is equivalent to the
convergence of the sum of the integrals.
6
This integral is fairly easily computed:
_

0
exp(
y
2
z
)
_
y
2
_
2m2
dy =
_

0
exp(s)
_
sz
4
_
m1
_
z
4s
_
1/2
ds
=
_

0
exp(s)
_
z
4
_
m1/2
s
m3/2
ds
=
_
z
4
_
m1/2
(m1/2).
We arrive at
J
0
=

m1
(1/2)
4m!(m+ 1/2)
_
z
4
_
m1/2
(m1/2).
=

m1

4m!(m1/2)
_
z
4
_
m1/2
=

1
16
_
z
0
_
4

_
3/2
_
exp(/4) 1
_
d
=

2
_
er
_

z
2
_
+
2

z
_
1 exp(z/4)
_
_
,
where er() is the imaginary error function. Hence, we get

t
=

t
T
_
1

z
2
er
_

z
2
_
+ 1 exp(z/4)
_
+
_
z

_
We now procede to compute the other terms in J

. There, we need to compute for n 1

_

0
exp(
y
2
z
)F
n
(y)
y
2
dy =
_

0
exp(
y
2
z
)

m0
1
m!(m+ 2n + 1/2)
_
y
2
_
2m+2n dy
y
2
=

m0
1
4m!(m+ 2n + 1/2)
_

0
exp(
y
2
z
)
_
y
2
_
2m+2n2
dy
=

m0
1
4m!(m+ 2n + 1/2)
_

0
exp(
y
2
z
)
_
y
2
4
_
m+n3/2 ydy
2
=

m0
1
4m!(m+ 2n + 1/2)
_

0
exp(s)
_
sz
4
_
m+n3/2
_
z
4
_
ds
=

m0
1
4m!(m+ 2n + 1/2)
_
z
4
_
m+n1/2
_

0
exp(s)s
m+n3/2
ds
=

m0
(n + m1/2)
4m!(m+ 2n + 1/2)
_
z
4
_
m+n1/2
=

m0
(n 1/2)(n 1/2)
(m)
4m!(2n + 1/2)(2n + 1/2)
(m)
_
z
4
_
m+n1/2
=
(n 1/2)
4(2n + 1/2)
_
z
4
_
n1/2
1
F
1
(n 1/2; 2n + 1/2; z/4),
where
1
F
1
(a; b; z) is the conuent hypergeometric function. Inserting the exponential factors and the
a
n
s, we arrive at
J

n1
a
n
exp
_
E
n
(T + t
0
t)
_
_

0
exp(
y
2
z
)F
n
(y)
y
2
dy
=

n1
(1)
n+1
(n 1/2)
2
4n!(2n 1/2)
e
En(T+t0t)
_
z
4
_
n1/2
1
F
1
(n 1/2; 2n + 1/2; z/4)
7
Note, that J
0
may also be expressed as a conuent hypergeometric function, as
J
0
=

m1
(1/2)(m1/2)
4m!(m+ 1/2)
_
z
4
_
m1/2
.
=
(1/2)
4
_
z
4
_
1/2

m1
(m1/2)
m!(m + 1/2)
_
z
4
_
m
.
=
(1/2)
2
4(1/2)
_
z
4
_
1/2
_
1
F
1
(1/2; +1/2;
z
4
) 1
_
.
We nally arrive at

t
=

t
T

n0
b
n
e
En(T+t0t)

n
1
F
1
(n 1/2; 2n + 1/2; ), (23)
where
b
n
=
(1)
n+1
(n 1/2)
2
2

n!(2n 1/2)
,
E
n
=
2
n(2n 1),

t
=
_
t0+T
t0

2
t
dt is the realized volatility up to t,
=

2
2
2

t
=
z
4
,
=
t
is the volatility at time t and
1
F
1
(; ; ) are conuent hypergeometric functions.
The fair value of a volatility swap with t
0
t t
0
+ T is hence
p(t, T)
_

t
T

n0
b
n
e
En(T+t0t)

n
1
F
1
(n 1/2; 2n + 1/2; ) K
_
. (24)
4.1 Value at t = t
0
+T.
We replace t with T + t
0
. The formula 23 simplies to

t0+T
=

t
T
_

n0
b
n

n
1
F
1
(n 1/2; 2n + 1/2; )
_
. (25)
Inserting the series expansion of the conuent hypergeometric function, we arrive at

t0+T
=

t
T
_

n,m0
(1)
n+1
(n 1/2)
2
2

n!(2n 1/2)
(n 1/2)
(m)
(2n + 1/2)
(m)
m!

n+m
_
=

t
T
_

n,m0
(1)
n+1
(2n 1/2)(n + m1/2)
2

n!(2n + m+ 1/2)
(n 1/2)
m!

n+m
_
.
By regrouping the terms with same exponents, we get

t0+T
=

t
T
_

s0

s
(s 1/2)
2

0ns
(1)
n+1
(2n 1/2)(n 1/2)
n!(s n)!(s + n + 1/2)
_
.
The rst term, i.e. the constant multiplying
0
, is 1. The second, with s = 1 and n = 0, 1 is equal to
(1/2)(1/2)/(3/2) + (3/2)(1/2)/(5/2) = (1/2)/(3/2) + (1/2)/(3/2) = 0.
For s = 2 and n = 0, 1, 2, one obtains (1/2)(1/2)/2(5/2)+(3/2)(1/2)/(7/2)(7/2)(3/2)/2(9/2) =
8
(1/2)/2(5/2) +3(1/2)/5(5/2) (3/2)/2(7/2) = (5(1/2) +6(1/2) 2(3/2))/10(5/2) = 0,
and one may show by a recurrence relation, that all these sums equal zero. we thus arrive at

t0+T
=

t
T
,
which is the expected result.
This is actually the boundary value for the functional calculus we are going to study in the next section
5 Verication By Functional Calculus
In the computation of the solution, we somewhat ignored the mathematical diculties which could arise
from the exchange of orders of integration, especially when taking expectations and integrating over a
parameter was involved. In this paragraph, we investigate the solution (23) in the light of Rama Conts
papers [3] & [4] on functional stochastic processes. As given by (23),
t
is certainly well-dened. It can
even be shown, that it is an analytical function. The question is, if it is truly part of the solution of the
problem at hand. Observe, that

t
=
1
T
E
_
(
_
t0+T
t0

2
s
ds)
1/2

t
=
_
is in fact an integral depending on the path chosen for
t
. As explained in Rama Conts paper on
functional calculus, we must have
D
t

t
+
1
2

2
(
v

)
2

t
= 0, (26)
where D
t
=
t
+
h

and
h

is the horizontal derivative, while

v

is the vertical derivative. In our case,

= 0,
h

_
t0+T
t0

2
s
ds =
2
t
(t t
0
)(T +t
0
t),
v

_
t0+T
t0

2
s
ds = 0,
v

= 2/
t
and the usual chain
rules apply. We also should have
t0+T
=

t0+T
, which was already veried.
Let us lighten up the notations and put := T + t
0
t and f
n
() :=
1
F
1
(n 1/2; 2n + 1/2; ).
t
rewrites as:

t
=

t
T

n0
b
n
e
En

n
f
n
()
Computing derivatives, we get
D
t

t
=
t

t
+
h

t
=

t
_

n0
b
n
E
n
e
En

n
f
n
()
_
+

n0
b
n
e
En

n
f
n
()
_
2
2

t
_

n0
b
n
e
En

n1
_
f

n
() + nf
n
()
__
= 2
2

t
_

n0

n
f
n
()b
n
e
En
_

2

E
n
2
2
n
_

n0

n+2
f

n
()b
n
e
En
_
on the other hand,
(
v

)
2

t
=

t
_

n0
b
n
e
En

n1
_
f

n
() + nf
n
()
___
=
1

t
_

n0
b
n
e
En

n1
_
f

n
() + nf
n
()
__
+

2

4
(
t
)
3/2
_

n0
b
n
e
En

n2
_
2nf

n
() + n(n 1)f
n
() +
2
f

n
()
__
.
Now, use the fact, that the hypergeometric functions are solution to the dierential equation
z
1
F

1
(a; b; z) = (z b)
1
F

1
(a; b; z) + a
1
F
1
(a; b; z).
9
In our case,

2
f

n
() = ( 2n 1/2)f

n
() + (n 1/2)f
n
().
This yields

2
2
(
v

)
2

t
= 2
2

t
_

n0
b
n
e
En

n
_

2
f

n
() + ( + n)(n 1/2)f
n
()
__
.
By replacing E
n
with their values
2
n(2n1), we see that the functional equation (26) is satised, which
justies the equation (24) as the solution to the fair value for volatility swaps.
6 Conclusion
The fair value of a volatility swap under the SABR model and with t [t
0
, t
0
+ T] has the solution (24)
. The relation

0ns
(1)
n+1
(2n 1/2)(n 1/2)
n!(s n)!(s + n + 1/2)
= 0
was not yet proven but can be accepted by verication. As a next computation, one may look for the
value of a volatility swap for t < t
0
. The value for a volatility option is also a question that bears some
interest and should be addressed in the future.
References
[1] Patrick S. Hagan, Deep Kumar, Andrew S. Lebniewski and Diana E. Woodward L
A
T
E
X: Managing
Smile Risk.
[2] Tomas Bjoerk, L
A
T
E
X: Arbitrage Theory in Continuous Time. Oxford University Press, 3rd Edition,
2009.
[3] Rama Cont, L
A
T
E
X: A Functional Extension of The Ito Formula. Comptes Rendus Mathmatiques
de lAcadmie des Sciences, Volume 348, Issues 1-2 January 2010, pages 57-61.
[4] Rama Cont and David-Antoine Fourni L
A
T
E
X: Functional Ito Calculus and Stochastic Integral Repre-
sentation of Martingales. Comptes Rendus Mathmatiques de lAcadmie des Sciences, hal.archives-
ouvertes.fr February 2010.
10