Finance Stoch (2007) 11: 571589

DOI 10.1007/s00780-007-0049-1
On the short-time behavior of the implied volatility for
jump-diffusion models with stochastic volatility
Elisa Als Jorge A. Len Josep Vives
Received: 30 June 2006 / Accepted: 23 May 2007 / Published online: 8 August 2007
Springer-Verlag 2007
Abstract In this paper we use Malliavin calculus techniques to obtain an expression
for the short-time behavior of the at-the-money implied volatility skew for a gener-
alization of the Bates model, where the volatility does not need to be a diffusion or
a Markov process, as the examples in Sect. 7 show. This expression depends on the
derivative of the volatility in the sense of Malliavin calculus.
Keywords Black-Scholes formula Derivative operator Its formula for the
Skorohod integral Jump-diffusion stochastic volatility model
JEL Classication G12 G13
Mathematics Subject Classication (2000) 91B28 91B70 60H07
E. Als research is supported by grants MEC FEDER MTM 2006 06427 and SEJ2006-13537.
J.A. Lens research is partially supported by the CONACyT grant 45684-F.
J. Vives research is supported by grant MEC FEDER MTM 2006 06427.
E. Als ()
Dpt. dEconomia i Empresa, Universitat Pompeu Fabra, c/Ramon Trias Fargas, 2527, 08005
Barcelona, Spain
e-mail: elisa.alos@upf.edu
J.A. Len
Control Automtico, CINVESTAV-IPN, Apartado Postal 14-740, 07000 Mxico, D.F., Mexico
J. Vives
Dpt. de Matemtiques, Universitat Autnoma de Barcelona, 08193 Bellaterra (Barcelona), Spain
J. Vives
Dpt. Probabilitat, Lgica i Estadstica, Universitat de Barcelona, Gran Via, 585, 08007 Barcelona,
Spain
572 E. Als et al.
1 Introduction
In the last years several authors have studied different extensions of the classical
Black-Scholes model in order to explain the current market behavior. Among these
extensions, one of the most popular allows the volatility to be a stochastic process
(see, for example, [5, 14, 15, 23, 24], among others).
It is well known that classical stochastic volatility diffusion models, where the
volatility also follows a diffusion process, capture some important features of the im-
plied volatilityfor example, its variation with respect to the strike price, described
graphically as a smile or skew (see [21]). But the observed implied volatility exhibits
dependence not only on the strike price, but also on the time to maturity (term struc-
ture). Unfortunately, the term structure is not easily explained by classical stochastic
volatility models. For instance, a popular rule of thumb for the short-time behavior
with respect to time to maturity, based on empirical observations, states that the skew
slope is approximately O((T t )

1
2
), while the rate for these stochastic volatility
models is O(1); see [18, 19], or [17]. Note that in these models, for reasonable co-
efcients in their dynamics, volatility behaves almost as a constant, on a very short
time-scale. Consequently, returns are roughly normally distributed and the skew be-
comes quite at. This problem has motivated the introduction of jumps in the asset
price dynamic models. Although the rate of the skew slope for models with jumps is
still O(1), as is shown by Medvedev and Scaillet [19], they allow exible modeling
and generate skews and smiles similar to those observed in market data (see [68],
or [9]). Recently, Fouque et al. [13] have introduced continuous diffusion models
again to describe the empirical short-time skew. Their idea is to include suitable co-
efcients that depend on the time till the next maturity date and that guarantee the
variability to be large enough near the maturity time.
The difculties in tting classical stochastic volatility models or models with
jumps to observed marked prices have motivated, as an alternative approach, to model
directly the implied volatility surfaces. Some recent research in modeling and exis-
tence issues for stochastic implied volatility models can be found in [16, 22], and the
references therein.
The main goal of this paper is to provide a method based on the techniques of
Malliavin calculus to estimate the rate of the short-dated behavior of the implied
volatility (see Theorem7 below) for general jump-diffusion stochastic volatility mod-
els, where the volatility does not need to be a diffusion or a Markov process. It is
well known that the Malliavin calculus is a powerful tool to deal with anticipating
processes. Since the future volatility is not adapted, this theory becomes a natural
tool to analyze this problem. Hence, now it is possible to deal with a volatility in a
class that includes fractional processes with parameter in (0, 1), Markov processes
and processes with time-varying coefcients, among others.
The paper is organized as follows. In Sect. 2 we introduce the framework and the
notation that we utilize in this paper. In Sect. 3 we state our basic tool, namely, an
anticipating It formula for the Skorohod integral. As a consequence, in Sect. 4, we
obtain an extended Hull and White formula for a general class of jump-diffusion mod-
els with stochastic volatility. An expression for the derivative of the implied volatility
is given in Sect. 5. Section 6 is devoted to the main result of this article. This means
On the short-time behavior of the implied volatility 573
that we gure out the short-time limit behavior. Finally, in Sect. 7, we give some ex-
amples in order to show that we can not only extend some known results, but also
consider new volatility models so that we are able to capture the short-time behavior
of skew slopes of order (T t )

, for >1/2.
2 Statement of the model and notation
In this paper we consider the following model for the log-price of a stock under a
risk-neutral probability measure Q:
X
t
= x +(r k)t
1
2
_
t
0

2
s
ds
+
_
t
0

s
_
dW
s
+
_
1
2
dB
s
_
+Z
t
, t [0, T ]. (2.1)
Here x is the current log-price, r is the instantaneous interest rate, W and B are
independent standard Brownian motions, (1, 1) and Z is a compound Pois-
son process with intensity , Lvy measure , independent of W and B, and with
k =
1

_
R
(e
y
1)(dy) <. The volatility process is a square-integrable stochas-
tic process with right-continuous trajectories and adapted to the ltration generated
by W. In some parts of the paper we shall assume, in addition, that its trajectories
are bounded below by a positive constant and that the process satises some suitable
conditions in the Malliavin calculus sense (see hypotheses (H1)(H5) below).
Notice that this model is a generalization of the classical Bates model introduced
in [8], in the sense that we do not assume the volatility to be a diffusion process.
It is well known that any stopping time with respect to a Brownian ltration is
predictable. So, an extension of our results below allows the volatility to have non-
predictable jump times as advocated by Bakshi et al. [4] and Dufe et al. [11], among
others. In this case we have to use Malliavin calculus for Lvy processes. The details
of this extension are in preparation and will appear elsewhere.
In the following we denote by F
W
, F
B
and F
Z
the ltrations generated by W, B
and Z, respectively. Moreover we dene F :=F
W
F
B
F
Z
.
It is well known that if we price a European call with strike price K by the formula
V
t
=e
r(T t )
E
__
e
X
T
K
_
+
|F
t
_
, (2.2)
where E is the expectation with respect to Q, there is no arbitrage opportunity. Thus,
V
t
is a possible price for this derivative. Notice that any allowable choice of Qleads to
an equivalent martingale measure and to a different no arbitrage price. The approach
that we follow here is the same as in [12], where it is assumed that the market selects
a unique equivalent martingale measure under which derivative contracts are priced.
In the sequel we use the following notation:
v
t
:=(
Y
t
T t
)
1
2
, with Y
t
:=
_
T
t

2
s
ds, will denote the future average volatility.
For any >0, p(x, ) will denote the centered Gaussian kernel with variance
2
.
If =1 we write p(x).
574 E. Als et al.
BS(t, x, ) will denote the price of a European call option under the classical
Black-Scholes model with constant volatility , current log stock price x, time
to maturity T t, strike price K and interest rate r. Remember that in this case:
BS(t, x, ) =e
x
N(d
+
) Ke
r(T t )
N(d

),
where N denotes the cumulative probability function of the standard normal law
and
d

:=
x x

T t

T t ,
with x

t
:=lnK r(T t ).
L
BS
() will denote the Black-Scholes differential operator, in the log variable,
with volatility :
L
BS
() =
t
+
1
2

2
xx
+
_
r
1
2

2
_

x
r.
It is well known that L
BS
() BS(, , ) =0.
G(t, x, ) :=(
2
xx

x
) BS(t, x, ).
3 An anticipating It formula
First of all, we describe the basic notation that is used in this article. For this, we
assume that the reader is familiar with the elementary results of Malliavin calculus,
as given for instance in [20].
Let us consider a standard Brownian motion W = {W
t
, t [0, T ]} dened in a
complete probability space (, F, P). The set D
1,2
W
will denote the domain of the
derivative operator D
W
. It is well known that D
1,2
W
is a dense subset of L
2
() and
that D
W
is a closed and unbounded operator from L
2
() to L
2
([0, T ] ). We also
consider the iterated derivatives D
W,n
, for n > 1, whose domains will be denoted
by D
n,2
W
.
The adjoint of the derivative operator D
W
, denoted by
W
, is an extension of the
It integral in the sense that the set L
2
a
([0, T ] ) of square integrable and adapted
processes (with respect to the ltration generated by W) is included in Dom
W
and
the operator
W
restricted to L
2
a
([0, T ] ) coincides with the It integral. We shall
use the notation
W
(u) =
_
T
0
u
t
dW
t
. We recall that L
n,2
:= L
2
([0, T ]; D
n,2
W
) is in-
cluded in the domain of
W
for all n 1.
Now we can establish the following It formula, which follows from [1, 2], and is
the main tool of this paper. In the sequel we use the notation D =D
W
to simplify the
exposition.
Proposition 3.1 Assume the model (2.1) and L
1,2
. Let F : [0, T ] R
2
R be
a function in C
1,2
([0, T ] R
2
) such that there exists a positive constant C such that,
On the short-time behavior of the implied volatility 575
for all t [0, T ], F and its partial derivatives evaluated in (t, X
t
, Y
t
) are bounded
by C. Then it follows that
F(t, X
t
, Y
t
) = F(0, X
0
, Y
0
) +
_
t
0

s
F(s, X
s
, Y
s
) ds
+
_
t
0

x
F(s, X
s
, Y
s
)
_
r k

2
s
2
_
ds
+
_
t
0

x
F(s, X
s
, Y
s
)
s
_
dW
s
+
_
1
2
dB
s
_

_
t
0

y
F(s, X
s
, Y
s
)
2
s
ds +
_
t
0

2
xy
F(s, X
s
, Y
s
)
s
ds
+
1
2
_
t
0

2
xx
F(s, X
s
, Y
s
)
2
s
ds
+
_
t
0
_
R
_
F(s, X
s
+y, Y
s
) F(s, X
s
, Y
s
)
_

J
X
(ds, dy)
+
_
t
0
_
R
_
F(s, X
s
+y, Y
s
) F(s, X
s
, Y
s
)
_
ds (dy),
where
s
:= (
_
T
s
D
s

2
r
dr)
s
, J
X
is the Poisson random measure such that
Z
t
=
_
[0,t ]R
yJ
X
(ds, dy) and

J
X
(ds, dy) :=J
X
(ds, dy) ds (dy).
Proof Denote by T
i
, i =1, . . . , N
T
, the jump instants of X. On [T
i
, T
i+1
), X evolves
according to
dX
c
t
=
_
r k

2
t
2
_
dt +
t
_
dW
t
+
_
1
2
dB
t
_
.
Then, applying Theorem 1 in [1], and using the fact that Z is independent of W and
B, we have that
F(T
i+1
, X
T
i+1
, Y
T
i+1

) F(T
i
, X
T
i
, Y
T
i
)
=
_
T
i+1

T
i

s
F(s, X
s
, Y
s
) ds +
_
T
i+1

T
i

x
F(s, X
s
, Y
s
) dX
c
s

_
T
i+1

T
i

y
F(s, X
s
, Y
s
)
2
s
ds +
_
T
i+1

T
i

2
xy
F(s, X
s
, Y
s
)
s
ds
+
1
2
_
T
i+1

T
i

2
xx
F(s, X
s
, Y
s
)
2
s
ds.
Note that if a jump of size X
t
occurs then the resulting change in F(t, X
t
, Y
t
)
is given by F(t, X
t
+ X
t
, Y
t
) F(t, X
t
, Y
t
). Therefore, the total change in
576 E. Als et al.
F(t, X
t
, Y
t
) can be written as the sum of these two contributions. Thus, we deduce
the desired result.
4 An extension of the Hull and White formula
In this section, using the It formula and the arguments developed in [1], we prove
an extension of the Hull and White formula that gives the price of a European call
option as a sum of the price when the model has no jumps and no correlation plus
three terms: one describing the impact of the correlation on option prices and two of
them, which can be presented jointly, describing the impact of jumps on these prices.
Hence, this formula will be a useful tool to compare the effect of correlation and
jumps (see Sect. 5).
We need the following result, inspired by Lemma 5.2 in [12].
Lemma 4.1 Let 0 t s < T and G
t
:= F
t
F
W
T
F
Z
T
. Then for every n 0,
there exists C =C(n, ) such that

E
_

n
x
G(s, X
s
, v
s
)|G
t
_

C
__
T
t

2
s
ds
_

1
2
(n+1)
.
Proof A simple calculation gives
G(s, X
s
, v
s
) =Ke
r(T s)
p
_
X
s
, v
s

T s
_
,
where =lnK (r v
2
s
/2)(T s). This allows us to write
E
_

n
x
G(s, X
s
, v
s
)|G
t
_
=(1)
n
Ke
r(T s)

E
_
p
_
X
s
, v
s

T s
_
|G
t
_
. (4.1)
Since the conditional expectation of X
s
given G
t
is a normal random variable with
mean
=X
t
+
_
s
t
_
r
2

/2
_
d +Z
s
Z
t
k(s t ) +
_
s
t

dW

and variance (1
2
)
_
s
t

2

d, and using the semigroup property of the Gaussian

density function, it follows that
E
_
p
_
X
s
, v
s

T s
_
|G
t
_
=p
_
,
_
_
1
2
_
_
T
t

d +
2
_
T
s

d
_
.
Putting this result in (4.1), we have
E
_

n
x
G(s, X
s
, v
s
)|G
t
_
=(1)
n
Ke
r(T s)

p
_
,
_
_
1
2
_
_
T
t

d +
2
_
T
s

d
_
.
On the short-time behavior of the implied volatility 577
A simple calculation and the fact that, for all positive constants c, d, the function
x
c
e
dx
2
is bounded, give us that

p
_
,
_
_
1
2
_
_
T
t

2
s
ds +
2
_
T
s

2
s
ds
_

C
_
_
1
2
_
_
T
t

2
s
ds +
2
_
T
s

2
s
ds
_

1
2
(n+1)
C
__
T
t

2
s
ds
_

1
2
(n+1)
,
and thus the proof is complete.
Now we are able to prove the main result of this section, the extended Hull and
White formula.
Theorem 4.2 Assume the model (2.1) holds with L
1,2
. Then it follows that
V
t
= E
_
BS(t, X
t
, v
t
)|F
t
_
+

2
E
__
T
t
e
r(st )

x
G(s, X
s
, v
s
)
s
ds

F
t
_
+E
__
T
t
_
R
e
r(st )
_
BS(s, X
s
+y, v
s
) BS(s, X
s
, v
s
)
_
(dy) ds

F
t
_
kE
__
T
t
e
r(st )

x
BS(s, X
s
, v
s
) ds

F
t
_
.
Proof This proof is similar to the one of the main theorem in [1], so we only sketch it.
Notice that BS(T, X
T
, v
T
) =V
T
. Then, from (2.2), we have
e
rt
V
t
=E
_
e
rT
BS(T, X
T
, v
T
)|F
t
_
.
Now our idea is to apply Proposition 3.1 to the process e
rt
BS(t, X
t
, v
t
). As the
derivatives of BS(t, x, ) are not bounded, we use an approximating argument,
changing v
t
to
v

t
:=
_
1
T t
(Y
t
+),
and BS(t, x, ) to BS
n
(t, x, ) := BS(t, x, )
n
(x), where
n
(x) := (
1
n
x), for
some C
2
b
such that (x) =1 for all |x| <1 and (x) =0 for all |x| >2. Applying
Proposition 3.1 between t and T , proceeding as in Theorem 3 in [1] and observing
that
L
BS
(
s
) BS
n
_
s, X
s
, v

s
_
=
_
L
BS
(
s
) BS
_
s, X
s
, v

s
__

n
(X
s
) +A
n
(s),
where
A
n
(s) =
1
2

2
s
_
2
x
BS
_
s, X
s
, v

s
_

n
(X
s
) +BS
_
s, X
s
, v

s
__

n
(X
s
)

n
(X
s
)
__
+r BS
_
s, X
s
, v

s
_

n
(X
s
),
578 E. Als et al.
we obtain
E
_
e
rT
BS
n
_
T, X
T
, v

T
_
|F
t
_
=E
_
e
rt
BS
n
_
t, X
t
, v

t
_
+
_
T
t
e
rs
A
n
(s) ds k
_
T
t
e
rs

x
BS
n
_
s, X
s
, v

s
_
ds
+

2
_
T
t
e
rs
_
(
x
G)
_
s, X
s
, v

s
_

n
(X
s
) +G
_
s, X
s
, v

s
_

n
(X
s
)
_

s
ds
+
_
T
t
_
R
e
rs
_
BS
n
_
s, X
s
+y, v

s
_
BS
n
_
s, X
s
, v

s
__
(dy) ds

F
t
_
.
Now, letting rst n and then 0, using Lemma 4.1 and dominated convergence
arguments, the result follows.
5 An expression for the derivative of the implied volatility
Let I
t
(X
t
) denote the implied volatility process, which satises
V
t
=BS(t, X
t
, I
t
(X
t
)), by denition. In this section we prove a formula for its at-
the-money derivative that we use in Sect. 6 to study the short-time behavior of the
implied volatility.
Proposition 5.1 Assume the model (2.1) holds with L
1,2
and for every xed
t [0, T ), E(
_
T
t

2
s
ds|F
t
)
1
< a.s. Then it follows that
I
t
X
t
(x

t
) =
E(
_
T
t
(
x
F(s, X
s
, v
s
)
1
2
F(s, X
s
, v
s
)) ds|F
t
)

BS(t, x

t
, I
t
(x

t
))

X
t
=x

t
, a.s.,
where
F(s, X
s
, v
s
) :=

2
e
r(st )

x
G(s, X
s
, v
s
)
s
+
_
R
e
r(st )
_
BS(s, X
s
+y, v
s
) BS(s, X
s
, v
s
)
_
(dy)
ke
r(st )

x
BS(s, X
s
, v
s
).
Proof Taking partial derivatives of the expression V
t
= BS(t, X
t
, I
t
(X
t
)) with re-
spect to X
t
, we obtain
V
t
X
t
=
x
BS
_
t, X
t
, I
t
(X
t
)
_
+

BS
_
t, X
t
, I
t
(X
t
)
_
I
t
X
t
(X
t
). (5.1)
On the other hand, from Theorem 4.2 we deduce that
V
t
=E
_
BS(t, X
t
, v
t
)|F
t
_
+E
__
T
t
F(s, X
s
, v
s
) ds

F
t
_
,
On the short-time behavior of the implied volatility 579
which implies that
V
t
X
t
=E
_

x
BS(t, X
t
, v
t
)|F
t
_
+E
__
T
t

x
F(s, X
s
, v
s
) ds

F
t
_
. (5.2)
Using now the fact that E(
_
T
t

2
s
ds|F
t
)
1
< we can check that the conditional
expectation E(
_
T
t

x
F(s, X
s
, v
s
) ds|F
t
) is well-dened and nite a.s. Thus, (5.1)
and (5.2) imply
I
t
X
t
(x

t
)
=
E(
x
BS(t,x

t
,v
t
)|F
t
)
x
BS(t,x

t
,I
t
(x

t
))+E(
_
T
t

x
F(s,X
s
,v
s
) ds|F
t
)

BS(t,x

t
,I
t
(x

t
))

X
t
=x

t
.
(5.3)
Notice that
E
_

x
BS(t, x

t
, v
t
)|F
t
_
=
x
E
_
BS(t, x, v
t
)|F
t
_

x=x

t
=
x
BS
_
t, x, I
0
t
(x)
_

x=x

t
, (5.4)
where I
0
t
(X
t
) is the implied volatility in the case = =0.
Also, by the classical Hull and White formula, we have

x
_
BS
_
t, x, I
0
t
(x)
__

x=x

t
=
x
BS
_
t, x

, I
0
t
(x

)
_
+

BS
_
t, x

, I
0
t
(x

)
_
I
0
t
x
(x

t
).
(5.5)
From [21] we know that
I
0
t
x
(x

t
) =0. Then, (5.3), (5.4), and (5.5) imply that
I
t
X
t
(x

t
)
=

x
BS(t,x

t
,I
0
t
(x

t
))
x
BS(t,x

t
,I
t
(x

t
))+E(
_
T
t

x
F(s,X
s
,v
s
) ds|F
t
)

BS(t,x

t
,I
t
(x

t
))

X
t
=x

t
.
(5.6)
On the other hand, straightforward calculations lead us to

x
BS(t, x

t
, ) =e
x

t
N
_
1
2

T t
_
and
BS(t, x

t
, ) =e
x

t
_
N
_
1
2

T t
_
N
_

1
2

T t
__
.
580 E. Als et al.
Then

x
BS(t, x

t
, ) =
1
2
_
e
x

t
+BS(t, x

t
, )
_
and

x
BS
_
t, x

t
, I
0
t
(x

t
)
_

x
BS
_
t, x

t
, I
t
(x

t
)
_
=
1
2
_
BS
_
t, x

t
, I
0
t
(x

t
)
_
BS
_
t, x

t
, I
t
(x

t
)
__
=
1
2
_
E
_
BS(t, x

t
, v
t
) V
t
(x

t
)|F
t
__
=
1
2
E
__
T
t
F(s, X
s
, v
s
) ds

F
t
_

X
t
=x

t
.
This, together with (5.6), implies the result.
6 Short-time limit behavior
Here our purpose is to study the limit of
I
t
X
t
(x

t
) when T t. To this end, we need
the following lemma:
Lemma 6.1 Assume the model (2.1) is satised. Then I
t
(x

t
)

T t tends to 0 a.s.,
as T t.
Proof Using the dominated convergence theorem it is easy to see that
P
t
:=E
_
e
r(T t )
_
K e
X
T
_
+
|F
t
_

X
t
=x

T t
_
K e
x

t
_
+
=0, a.s.
Now, by the classical call-put parity relation, we obtain
V
t
=E
_
e
r(T t )
_
e
X
T
K
_
+
|F
t
_

X
t
=x

_
e
x

t
K
_
+
=0.
Hence, taking into account that, in the at-the-money case, V
t
=BS(t, x

t
, I
t
(x

t
)), we
deduce that
BS
_
t, x

t
, I
t
(x

t
)
_
=2Ke
r(T t )
_
N
_
I (x

t
)

T t
2
_

1
2
_
0,
and this allows us to complete the proof.
Henceforth, we consider the following hypotheses:
(H1) L
2,4
.
(H2) There exists a constant a >0 such that >a.
(H3) There exists a constant >
1
2
such that, for all 0 <t <s <r <T,
E
_
(D
s

r
)
2
|F
t
_
C(r s)
2
, (6.1)
E
_
(D

D
s

r
)
2
|F
t
_
C(r s)
2
(r )
2
. (6.2)
Proposition 6.2 Assume that the model (2.1) and hypotheses (H1)(H3) hold. Then
On the short-time behavior of the implied volatility 581

BS
_
t, x

t
, I
t
(x

t
)
_
I
t
X
t
(x

t
)
=

2
E
_
L(t, x

t
, v
t
)
_
T
t

s
ds

F
t
_
kE
_
G(t, x

t
, v
t
)(T t )|F
t
_
+O(T t )
(1+2)1
,
as T t and where L(t, x

t
, v
t
) =(
2
xx

1
2

x
)G(t, x

t
, v
t
).
Proof Proposition 5.1 gives us that

BS
_
t, x

t
, I
t
(x

t
)
_
I
t
X
t
(x

t
)
=

2
E
__
T
t
e
r(st )
_

x

1
2
_

x
G(s, X
s
, v
s
)
s
ds

F
t
_

X
t
=x

t
+ E
__
T
t
_
R
e
r(st )
_

x

1
2
_

_
BS(s, X
s
+y, v
s
) BS(s, X
s
, v
s
)
_
(dy) ds

F
t
_

X
t
=x

t
kE
__
T
t
e
r(st )
_

x

1
2
_

x
BS(s, X
s
, v
s
) ds

F
t
_

X
t
=x

t
=T
1
+T
2
+T
3
. (6.3)
Now the proof will be decomposed into several steps.
Step 1. Here we claim that
T
1
=

2
E
_
L(t, x

t
, v
t
)
_
T
t

s
ds

F
t
_
+O(T t )
1+2
, (6.4)
where L(s, X
s
, v
s
) =(
2
xx

1
2

x
)G(s, X
s
, v
s
). In fact, applying Its formula to

2
e
r(st )
L(s, X
s
, v
s
)
__
T
s

r
dr
_
as in the proof of Theorem 4.2 and taking conditional expectations with respect to F
t
,
we obtain that

2
E
__
T
t
e
r(st )
L(s, X
s
, v
s
)
s
ds

F
t
_
=

2
E
_
L(t, X
t
, v
t
)
__
T
t

s
ds
_

F
t
_
+

2
4
E
__
T
t
e
r(st )
_

3
xxx

2
xx
_
L(s, X
s
, v
s
)
__
T
s

r
dr
_

s
ds

F
t
_
582 E. Als et al.
+

2
2
E
__
T
t
e
r(st )

x
L(s, X
s
, v
s
)
__
T
s
D
s

r
dr
_

s
ds

F
t
_
+

2
E
__
T
t
_
R
e
r(st )
_
L(s, X
s
+y, v
s
) L(s, X
s
, v
s
)
_

__
T
s

r
dr
_
(dy) ds

F
t
_
k

2
E
__
T
t
e
r(st )

x
L(s, X
s
, v
s
)
__
T
s

r
dr
_
ds

F
t
_
=

2
E
_
L(t, X
t
, v
t
)
__
T
t

s
ds
_

F
t
_
+S
1
+S
2
+S
3
+S
4
.
Using Lemma 4.1 we can write
S
1
=

2
4
E
__
T
t
e
r(st )
E
__

3
xxx

2
xx
_
L(s, X
s
, v
s
)|G
t
_
__
T
s

r
dr
_

s
ds

F
t
_
C
6

k=4
E
___
T
t

2
s
ds
_

k
2
_
T
t

__
T
s

r
dr
_

ds

F
t
_
.
Hence, using hypotheses (H2) and (H3) we can write
S
1
C
2

k=0
E
___
T
t

d
_

k
2
__
T
t
_
T
t
(D
r

)
2
dr d
_

F
t
_
C(T t )
1
__
T
t
_

t
( r)
2
dr d
_
C(T t )
1+2
.
Using similar arguments it follows that S
2
+ S
3
+ S
4
= O(T t )
1+2
, which
proves (6.4).
Step 2. As |BS(t, x, )| + |
x
BS(t, x, )| 2e
x
+ K it follows that T
2
=
O(T t ).
Step 3. Let us prove that
T
3
=kE
_
G(t, x

t
, v
t
)(T t )|F
t
_
+O(T t ). (6.5)
In fact,
E
__
T
t
e
r(st )
_

x

1
2
_

x
BS(s, X
s
, v
s
) ds

F
t
_

X
t
=x

t
=E
__
T
t
e
r(st )
G(s, X
s
, v
s
) ds

F
t
_

X
t
=x

t
+
1
2
E
__
T
t
e
r(st )

x
BS(s, X
s
, v
s
) ds

F
t
_

X
t
=x

t
.
On the short-time behavior of the implied volatility 583
As |
x
BS(t, x, )| e
x
it follows easily that the second term on the right-hand side
of this equality is O(T t ). On the other hand, Its formula allows us to write
E
_
e
r(st )
G(s, X
s
, v
s
)|F
t
_
=E
_
G(t, X
t
, v
t
)|F
t
_
+

2
E
__
s
t
e
r(ut )
_

3
xxx

2
xx
_
G(u, X
u
, v
u
)
u
du

F
t
_
+ E
__
s
t
_
R
e
r(ut )
_
G(u, X
u
+y, v
u
) G(u, X
u
, v
u
)
_
(dy) du

F
t
_
kE
__
s
t
e
r(ut )

x
G(u, X
u
, v
u
) du

F
t
_
.
Now, using again the same arguments as in Step 1, (6.5) follows. Therefore, the proof
is complete.
Now we can state the main result of this paper. We consider the following hy-
potheses:
(H4) has a.s. right-continuous trajectories.
(H5) For every xed t >0, sup
s,r,[t,T ]
E((
s

r

2

)
2
|F
t
) 0, as T t.
Theorem6.3 Consider the model (2.1) and suppose that hypotheses (H1)(H5) hold.
1. Assume that in (H3) is nonnegative and that there exists an F
t
-measurable ran-
dom variable D
+
t

t
such that, for every t >0,
sup
s,r[t,T ]

E
__
D
s

r
D
+
t

t
_

F
t
_

0, a.s., (6.6)
as T t. Then
lim
T t
I
t
X
t
(x

t
) =
1

t
_
k +
D
+
t

t
2
_
. (6.7)
2. Assume that in (H3) is negative and that there exists an F
t
-measurable random
variable L
,+
t

t
such that, for every t >0,
1
(T t )
2+
_
T
t
_
T
s
E(D
s

r
|F
t
) dr ds L
,+
t

t
0, a.s., (6.8)
as T t. Then
lim
T t
(T t )

I
t
X
t
(x

t
) =

t
L
,+
t

t
. (6.9)
Proof Using Proposition 6.2 and the facts that

BS
_
t, x

t
, I
t
(x

t
)
_
=
Ke
r(T t )
e
I
t
(x

t
)
2
(T t )
8

T t

2
,
L(t, x

t
, v
t
) =Ke
r(T t )
1

2
e

v
2
t
(T t )
8
v
3
t
(T t )

3
2
584 E. Als et al.
and
G(t, x

t
, v
t
) =
Ke
r(T t )
e
v
t
2
(T t )
8
v
t

2(T t )
,
we can write
I
t
X
t
(x

t
) =

2
e
I
t
(x

t
)
2
(T t )
8
(T t )
2
E
_
e

v
2
t
(T t )
8
v
3
t
_
T
t

s
ds

F
t
_
ke
I
t
(x

t
)
2
(T t )
8
E
_
e
v
t
2
(T t )
8
v
1
t

F
t
_
+O(T t )
(
1
2
+2)
1
2
=: S
1
+S
2
+O(T t )
(
1
2
+2)1
.
By Lemma 6.1 we know that I
t
(x

t
)
2
(T t ) 0, as T t. Then
lim
T t
S
1
=

2
lim
T t
_
(T t )
2
E
_
e

v
2
t
(T t )
8
v
3
t
_
T
t

s
ds

F
t
__
.
Using again Lemma 6.1, observe that (H4) and the dominated convergence theorem
imply that
lim
T t
S
2
=
k

t
. (6.10)
Now the proof will be decomposed into two steps.
Step 1. Here we analyze the case 0. In this case we only need to show that
lim
T t
_
S
1
+

2
t
D
+
t

t
_
=0. (6.11)
Indeed, we can write
lim
T t
_
S
1
+

2
t
D
+
t

t
_
= lim
T t
E
_
A
T
B
T
+

2
t
D
+
t

t

F
t
_
,
where
A
T
:=

v
t
exp
_

(v
2
t
)(T t )
8
_
and
B
T
:=
1
v
2
t
(T t )
2
_
T
t
_
T
s

s
D
s

r
dr ds.
Notice that
lim
T t
E
_
A
T
B
T
+

2
t
D
+
t

t

F
t
_
= lim
T t
E
__
A
T

t
_
B
T

F
t
_
+

t
lim
T t
E
__
B
T
+
D
+
t

t
2
_

F
t
_
= lim
T t
U
1
+

t
lim
T t
U
2
.
On the short-time behavior of the implied volatility 585
Applying the CauchySchwarz inequality yields that
U
1

_
E
__
A
T

t
_
2

F
t
__1
2
_
E
_
B
2
T

F
t
__ 1
2
.
Using the dominated convergence theorem it is easy to see that E((A
T

t
)
2
|F
t
)
tends to zero, as T t, and a simple calculation gives us that E(B
2
T
|F
t
) is bounded;
whence, we deduce that lim
T t
U
1
=0. On the other hand,
|U
2
| =

1
(T t )
2
E
__
T
t
_
T
s
_

r
v
2
t
D
s

r
D
+
t

t
_
dr ds

F
t
_

C
(T t )
2

E
__
T
t
_
T
s
_

r
v
2
t
1
_
D
s

r
dr ds

F
t
_

+
C
(T t )
2

E
__
T
t
_
T
s
_
D
s

r
D
+
t

t
_
dr ds

F
t
_

=: |U
2,1
| +|U
2,2
|.
Using now the CauchySchwarz inequality and the fact that hypothesis (H3) holds
with 0, we obtain that
|U
2,1
|
C
(T t )
2
_
E
__
T
t
_
T
s
_

r
v
2
t
1
_
2
dr ds

F
t
__1
2

_
E
__
T
t
_
T
s
(D
s

r
)
2
dr ds

F
t
__1
2

C
(T t )
_
E
__
T
t
_
T
s
_

r
v
2
t
1
_
2
dr ds

F
t
__1
2
.
Now (H2) and (H4) allow us to write
|U
2,1
|
C
(T t )
__
T
t
_
T
s
E
__

r
v
2
t
_
2

F
t
_
dr ds
_1
2
=
C
(T t )
__
T
t
_
T
s
E
__

r

_
1
T t
_
T
t

d
__
2

F
t
_
dr ds
_1
2

C
(T t )
3
2
__
T
t
_
T
s
_
T
t
E
__

r

2

_
2

F
t
_
d dr ds
_1
2
,
which tends to zero by hypothesis (H5). Similarly,
|U
2,2
|
C
(T t )
2

_
T
t
_
T
s
E
__
D
s

r
D
+
t

t
_

F
t
_
dr ds

,
586 E. Als et al.
which tends to zero by (6.6). Now we have proved (6.11). Then, (6.10), (6.11), and
the fact that 0 give (6.7).
Step 2. Finally, we show that (6.9) is true. Let us prove that
lim
T t
_
S
1
(T t )

t
L
,+
t

t
_
=0. (6.12)
Note that
lim
T t
_
S
1
(T t )

t
L
,+
t

t
_
= lim
T t
E
_
A
T

B
T
+

t
L
,+
t

t

F
t
_
,
where A
T
is dened as in Step 1 and

B
T
:=
1
v
2
t
(T t )
2+
_
T
t
_
T
s

s
D
s

r
dr ds.
But
lim
T t
E
_
A
T

B
T
+

t
L
,+
t

t

F
t
_
= lim
T t
E
__
A
T

t
_

B
T

F
t
_
+

t
lim
T t
E
__

B
T
+L
,+
t

t
_

F
t
_
.
Then, using similar arguments as in the proof of Step 1, we can easily see that this
expression is equal to zero. Now we have proved (6.12). Finally, using (6.12), (6.10),
and the fact that
1
2
< <0 the result follows.
Remark Notice that (6.7) and (6.9) can be written in terms of
I
t
Z
, where Z =logK
is the log-strike, by simply changing the sign of the limits.
7 Examples
7.1 Diffusion stochastic volatilities
Assume that the volatility can be written as =f (Y), where f C
1
b
(R) and Y is
the solution of a stochastic differential equation
dY
r
=a(r, Y
r
) dr +b(r, Y
r
) dW
r
, (7.1)
for some real functions a, b C
1
b
(R). Then, classical arguments (see, for example,
Theorem 2.2.1 in [20]) give us that Y L
1,2
and that
D
s
Y
r
=
_
r
s
a
x
(u, Y
u
)D
s
Y
u
du +b(s, Y
s
) +
_
r
s
b
x
(u, Y
u
)D
s
Y
u
dW
u
. (7.2)
On the short-time behavior of the implied volatility 587
Taking now into account that D
s

r
= f

(Y
r
)D
s
Y
r
, it can be easily deduced
from (7.2) that (H3) holds with =0 and that
sup
s,r[t,T ]

E
__
D
s

r
f

(Y
t
)b(t, Y
t
)
_

F
t
_

0,
as T t . Then Theorem 6.3 gives us that
lim
T t
I
t
X
t
(x

t
) =
1

t
_
k +

2
f

(Y
t
)b(t, Y
t
)
_
,
which agrees with the results in [19].
In particular, if Y is an OrnsteinUhlenbeck process of the form
Y
r
=m+(Y
t
m)e
(rt )
+c
_
r
t

2 exp
_
(r s)
_
dW
s
, (7.3)
D
s
Y
r
=c

2 exp((r s)) for all t s <r and then it follows that

lim
T t
I
t
X
t
(x

t
) =
1

t
_
k +c

2
f

(Y
t
)
_
.
7.2 Fractional stochastic volatility models
Assume that the volatility can be written as =f (Y), where f C
1
b
(R) and Y is
a process of the form
Y
r
=m+(Y
t
m)e
(rt )
+c

2
_
r
t
e
(rs)
dW
H
s
, (7.4)
where W
H
s
:=
_
s
0
(s u)
H
1
2
dW
u
.
7.2.1 Case H >
1
2
As in [10], assume the volatility model (7.4), for some H >1/2. Notice that (see, for
example, [3])
_
r
t
e
(rs)
dW
H
s
can be written as
_
H
1
2
__
r
0
__
r
s
1
[t,r]
(u)e
(ru)
(u s)
H
3
2
du
_
dW
s
,
from which it follows easily that sup
s,r[t,T ]
|E(D
s

r
|F
t
)| 0, as T t . Then
Theorem 6.3 gives us that lim
T t
I
t
X
t
(x

t
) =
k

t
. That is, the at-the-money short-
dated skew slope of the implied volatility is not affected by the correlation in this
case.
588 E. Als et al.
7.2.2 Case H <
1
2
Assume again the model (7.4), taking 0 <H <1/2. It can be proved (see, for exam-
ple, [3]) that
_
r
t
e
(rs)
dW
H
s
can be expressed as
_
1
2
H
__
r
0
__
r
s
_
1
[t,r]
(u)e
(ru)
1
[t,r]
(s)e
(rs)
_
(u s)
H
3
2
du
_
dW
s
+
_
r
t
e
(rs)
(r s)
H
1
2
dW
s
.
Then it follows that (H3) holds for =H
1
2
and we can easily check that
E
_
1
(T t )
2+H
1
2
_
T
t
_
T
s
D
W
s

r
dr ds c

2f

(Y
t
)

F
t
_
0, as T t.
Then Theorem 6.3 gives us that
lim
T t
(T t )
1
2
H
I
t
X
t
(x

t
) =c

t
f

(Y
t
).
That is, the introduction of fractional components with Hurst parameter H <1/2 in
the denition of the volatility process allows us to reproduce a skew slope of order
O(T t )

, for every >1/2.

7.3 Time-varying coefcients
Fouque et al. [13] have introduced a new approach to capture the maturity-dependent
behavior of the implied volatility, by allowing the volatility coefcients to depend
on the time till the next maturity date. Namely, they assume that the volatility
can be written as = f (Y), where f is a regular enough function and Y is a
diffusion process of the form (7.3), with

(s) a suitable cutoff of the function
(T
n(s)
s)

1
2
, with xed maturity dates {T
k
} (the third Friday of each month) and
n(t ) =inf{n : T
n
>s}.
Following this idea, we can consider Y to be a diffusion process of the form (7.3),
with

(s) = (T
n(s)
s)

1
2
+
, for some > 0. It is now easy to see that Y L
1,2
and that
1
(T t )
2+(
1
2
)
_
T
t
_
T
s
E(D
s

r
|F
t
) dr ds +c
_
1
1/2 +
__
1
1/2 +
_
f

(Y
t
)
2
tends to zero, as T t tends to zero. Hence, we deduce that, in this case, the short-date
skew slope of the implied volatility is of the order O(T t )

1
2
+
.
8 Conclusions
We have seen that Malliavin calculus may provide a natural approach to deal with
the short-date behavior of the implied volatility for jump-diffusion models with sto-
chastic volatility. This theory does not require the volatility to be a diffusion or a
On the short-time behavior of the implied volatility 589
Markov process. Moreover, with these techniques the short-time behavior of the im-
plied volatility can be analyzed for known and new volatility modelsin particular,
models that reproduce short-date skews of order O(T t )

, for >
1
2
.
References
1. Als, E.: A generalization of the Hull and White formula with applications to option pricing approxi-
mation. Finance Stoch. 10, 353365 (2006)
2. Als, E., Nualart, D.: An extension of Its formula for anticipating processes. J. Theor. Probab. 11,
493514 (1998)
3. Als, E., Mazet, O., Nualart, D.: Stochastic calculus with respect to Gaussian processes. Ann. Probab.
29, 766801 (2001)
4. Bakshi, G., Cao, C., Chen, Z.: Empirical performance of alternative option pricing models. J. Finance
52, 20032049 (1997)
5. Ball, C., Roma, A.: Stochastic volatility option pricing. J. Finance Quant. Anal. 29, 589607 (1994)
6. Barndorff-Nielsen, O.E., Shephard, N.: Modelling by Lvy processes for nancial econometrics.
In: Barndorff-Nielsen, O.E., Mikosch, T., Resnick, S.I. (eds.) Lvy Processes: Theory and Appli-
cations, pp. 283318. Birkhuser, Basel (2001)
7. Barndorff-Nielsen, O.E., Shephard, N.: Econometric analysis of realized volatility and its use in esti-
mating stochastic volatility models. J. Roy. Stat. Soc. Ser. B Stat. Methodol. 64, 253280 (2002)
8. Bates, D.S.: Jumps and stochastic volatility: exchange rate processes implicit in Deutsche Mark op-
tions. Rev. Finance Stud. 9, 69107 (1996)
9. Carr, P., Wu, L.: The nite moment log stable process and option pricing. J. Finance 58, 753778
(2003)
10. Comte, F., Renault, E.: Long memory in continuous-time stochastic volatility models. Math. Finance
8, 291323 (1998)
11. Dufe, D., Pan, J., Singleton, K.: Transform analysis and asset pricing for afne jump-diffusions.
Econometrica 68, 13431376 (2000)
12. Fouque, J.-P., Papanicolaou, G., Sircar, R., Solna, K.: Singular perturbations in option pricing. SIAM
J. Appl. Math. 63, 16481665 (2003)
13. Fouque, J.-P., Papanicolaou, G., Sircar, R., Solna, K.: Maturity cycles in implied volatility. Finance
Stoch. 8, 451477 (2004)
14. Heston, S.L.: A closed-form solution for options with stochastic volatility with applications to bond
and currency options. Rev. Finance Stud. 6, 327343 (1993)
15. Hull, J., White, A.: The pricing of options on assets with stochastic volatilities. J. Finance 42, 281300
(1987)
16. Jacod, J., Protter, P.: Risk neutral compatibility with option prices. Preprint (2006). http://legacy.orie.
cornell.edu/~protter/nance.html
17. Lee, R.W.: Implied volatility: statics, dynamics, and probabilistic interpretation. In: Baeza-Yates, R.,
Glaz, J., Gzyl, H., et al. (eds.) Recent Advances in Applied Probability, pp. 241268. Springer, Berlin
(2004)
18. Lewis, A.L.: Option Valuation Under Stochastic Volatility with Mathematica Code. Finance Press,
Newport Beach (2000)
19. Medvedev, A., Scaillet, O.: Approximation and calibration of short-term implied volatilities under
jump-diffusion stochastic volatility. Rev. Finance Stud. 20(2), 427459 (2007)
20. Nualart, D.: The Malliavin Calculus and Related Topics. Springer, Berlin (1995)
21. Renault, E., Touzi, N.: Option hedging and implied volatilities in a stochastic volatility model. Math.
Finance 6, 279302 (1996)
22. Schweizer, M., Wissel, J.: Term structures of implied volatilities: absence of arbitrage and existence
results. Math. Finance (2006, to appear)
23. Scott, L.O.: Option pricing when the variance changes randomly: theory, estimation and an applica-
tion. J. Finance Quant. Anal. 22, 419438 (1987)
24. Stein, E.M., Stein, J.C.: Stock price distributions with stochastic volatility: an analytic approach. Rev.
Finance Stud. 4, 727752 (1991)