This action might not be possible to undo. Are you sure you want to continue?
com/abstract=2210359
0
A General Closed Form Option Pricing Formula
Ciprian NECULA
1,2
Gabriel DRIMUS
1
Walter FARKAS
1,3
1
Department of Banking and Finance, University of Zurich, Plattenstrasse 14, CH8032, Zurich, Switzerland
Email: ciprian.necula@bf.uzh.ch, gabriel.drimus@bf.uzh.ch, walter.farkas@bf.uzh.ch.
2
Department of Money and Banking, DOFIN, Bucharest University of Economic Studies, Bucharest, Romania
3
Department of Mathematics, ETH Zurich, Raemistrasse 101, CH8092 Zurich, Switzerland
First version: January 7, 2013
This version: January 25, 2013
Abstract
A new method to retrieve the riskneutral probability measure from observed option prices is
developed and a closedform pricing formula for European options is obtained by employing
a modified GramCharlier series expansion, known as the GaussHermite expansion. This
new option pricing formula is an alternative to the inverse Fourier transform methodology
and can be employed in general models with probability distribution function or
characteristic function known in closed form. We calibrate the model to both simulated and
market option prices and find that the resulting implied volatility curve provides a good
approximation for a wide range of strikes.
Keywords: European options, GaussHermite series expansion, calibration.
JEL Classification: C63, G13
Acknowledgements: This work was supported by a SCIEXNMS Fellowship, Project Code
11.159.
Electronic copy available at: http://ssrn.com/abstract=2210359
1
1. Introduction
The riskneutral valuation framework is one of the pillars of modern finance theory.
Under this framework, the riskneutral probability measure is an essential ingredient for asset
valuation, since the value of a financial derivative is given by the expected value under the
riskneutral measure of the future payoffs generated by the derivative, discounted at the
riskless interest rate. This approach was pioneered in the context of European option pricing
by Black and Scholes (1973) and Merton (1973) who postulated a Gaussian riskneutral
measure and obtained the celebrated BlackScholesMerton (BSM) formula. The stylized fact
of volatility smiles has encouraged the development of various extensions that account for the
fact that the riskneutral density is negatively skewed and leptokurtic, such as stochastic
volatility models (Heston, 1993), jumpdiffusion models (Merton, 1976; Kou, 2002), models
based on the Generalized Hyperbolic process or other purejumps Levy processes (Bibby and
Sorensen, 1997).
It is nonetheless appealing to obtain the riskneutral density from option prices data by
using nonparametric methods (AitSahalia and Lo, 1998; AitSahalia and Duarte, 2003) or
using an expansion around a density which is easy to compute, such as the normal or the log
normal. Jarrow and Rudd (1982) pioneered the density expansion approach to option pricing
using an Edgeworth series expansion of the terminal underlying asset price riskneutral
density around the lognormal density. Corrado and Su (1996, 1997) adopted the Jarrow–
Rudd framework and derived an option pricing formula using a Gram–Charlier type A series
expansion of the underlying asset logreturn riskneutral density around the Gaussian density.
The pioneering papers of Jarrow and Rudd (1982) and Corrado and Su (1996) focused
on option pricing formulae based on a GramCharlier type A expansion, generated a seminal
branch of option pricing research. An overview on the GramCharlier density expansion
approach to option valuation is provided by Jurczenko, Maillet, and Negrea (2002). Brown
and Robinson (2002) corrected a typographic error in the initial Corrado and Su (1996)
formula and point out that callput parity is no longer verified when the riskneutral logreturn
density function is approximated by a Gram–Charlier Type A series expansion, due to the
lack of the martingale restriction (Longstaff, 1995). Jurczenko, Maillet, and Negrea (2004)
slightly modified the original formula to provide consistency with a martingale restriction.
They also employ CAC 40 index options and show that the differences between the various
modifications of the Corrado–Su model are minor, but could be economically significant in
2
specific cases. Corrado (2007) developed a martingale restriction that is hidden behind a
reduction in parameter space for the Gram–Charlier expansion coefficients. The resulting
restriction is invisible in the option price.
Although probability densities given by GramCharlier Type A series expansions are
as tractable as the Gaussian density, they have the drawback that they can yield negative
probability values. Jondeau and Rockinger (2001) developed numerical methods to impose
positivity constraints on the Gram–Charlier expansion. Rompolis and Tzavalis (2007) employ
a method to retrieve the risk neutral probability density function based on an exponential form
of a Gram–Charlier series expansion, known as type C Gram–Charlier expansion. This type of
expansion guarantees that the values of the risk neutral density will be always positive, but
there is no closed form formula for option values.
In general, from the Gram–Charlier type A density expansion, only the first two terms,
accounting for skewness and kurtosis, are kept in empirical studies related to option pricing. It
is quite probable, however, that including higher order terms in the expansion will produce a
decrease, rather, as one would expect, an increase in the performance of the option pricing
formula based on the GramCharlier type A expansion. This counterintuitive result is likely to
occur due to the lack of convergence of the GramCharlier type A expansion for heavy tailed
distributions that are of interest in finance, as it converges only if the probability density
function falls off faster than
( )
2
exp 4 x − at infinity (Cramér, 1957).
The aim of this paper is to develop a new method to retrieve the riskneutral
probability measure and to derive an option pricing formula by employing a modified Gram
Charlier series expansion. Instead of using the “probabilists” Hermite polynomials, as in the
classical GramCharlier type A series expansion, we replace them by the “physicists” Hermite
polynomials. The main advantages of the new expansion consist in its convergence for heavy
tailed distributions and in the possibility of obtaining option prices in closed form.
The rest of the paper is structured as follows. In section 2 we present the classical and
the modified GramCharlier series expansion and analyze an example based on a widely used
distribution in finance, namely the Normal Inverse Gaussian (NIG) distribution. In section 3
we derive a pricing formula for European call options in the context of the modified Gram
Charlier expansion. In section 4 we analyze various methods for obtaining the expansion
coefficients from observed option prices and present a simulation study based on the Heston
model. Although the contribution of the paper is methodological rather than empirical, we
3
also present, in section 4, a calibration exercise based on observed option data. The final
section concludes.
2. The GramCharlier and GaussHermite series expansions
A probability density function ( ) p x with mean µ and standard deviation σ can be
represented as a Gram–Charlier Type A series expansion in the following form:
( )
0
1
n n
n
x x
p x z c He
µ µ
σ σ σ
∞
=
− −
   
=
 
\ ¹ \ ¹
∑
(1)
where ( ) z x is the standard Gaussian density, ( )
n
He x denotes an nthorder “probabilists”
Hermite polynomial. These form an orthogonal basis on ( ) , −∞ +∞ with respect to the weight
function
2
2
( )
x
w x e
−
= . The “probabilists” Hermite polynomials (Abramowitz and Stegun,
1964) are defined recursively by ( ) ( ) ( )
1 1 n n n
He x xHe x nHe x
+ −
= − with ( )
0
1 He x = and
( )
1
He x x = . The expansion coefficients are given by ( )
1
!
n
x
c He p x dx
n
µ
σ
∞
−∞
−
 
=

\ ¹
∫
and are,
in fact, a linear combination of the moments of the random variable with probability
distribution function ( ) p x , a property which is quite convenient when it comes to estimate
them.
Figure 1. The classical GramCharlier approximation of the NIG distribution
4 3 2 1 0 1 2
2
1
0
1
2
3
4
5
6
7
8
standard deviations
p
d
f
GC N=3
GC N=4
GC N=5
Gaussian
NIG
4 3 2 1 0 1 2
4
2
0
2
4
6
8
10
standard deviations
p
d
f
GC N=3
GC N=4
GC N=5
Gaussian
NIG
A B
The graphs depict the probability distribution function of the target distribution (NIG), of the Gaussian distribution, and the
GramCharlier approximations truncated after N=3,4,5 terms. In panel a the target distribution has mean 0.0012, standard
deviation 0.10, skewness 1.6 and excess kurtosis 5, and in panel b the target distribution has mean 0.005, standard deviation
0.20, skewness 2 and excess kurtosis 10.
4
The GramCharlier series expansion has poor convergence properties for heavy tailed
distributions (Cramer, 1957). In order to illustrate the divergence of this expansion we present
an illustrative example based on the Normal Inverse Gaussian (NIG) distribution. Figure 1
presents the GramCharlier approximations with an increasing number of terms in the
expansion and the exact NIG distribution for two sets of parameters. Figure 1a depicts a NIG
distribution with mean 0.0012, standard deviation 0.1, skewness 1.6 and excess kurtosis 5,
and figure 1b the case with mean 0.005, standard deviation 0.2, skewness 2 and excess
kurtosis 10. It is obvious that the series quickly becomes inaccurate by including a larger
number of terms. The parameters of the NIG distributions were obtained from the first four
cumulants using the method described in Eriksson, Forsberg and Ghysels (2004).
In fact, the GramCharlier expansion is a result of the expansion of the function
( ) ( ) p x z x in the basis of the “probabilists” Hermite polynomial. However, one could
choose another set of orthogonal polynomials as the basis of the expansion. In particular, the
“physicists” Hermite polynomials are employed in astrophysics (van der Marel and Franx,
1993). This yields a modified GramCharlier expansion or the so called GaussHermite
expansion. A result in MyllerLebedeff (1907) implies that the GaussHermite expansion
converges even for heavy tailed distributions. More specifically, the convergence of the
expansion is assured if ( )
3
lim 0
x
x p x
→±∞
= .
Therefore, a probability density function ( ) p x with mean µ and standard deviation
σ can be represented as a GaussHermite (or modified Gram–Charlier) series expansion in
the following form:
( )
0
1
n n
n
x x
p x z a H
µ µ
σ σ σ
∞
=
− −
   
=
 
\ ¹ \ ¹
∑
(2)
where ( )
n
H x denotes an nthorder “physicists” Hermite polynomial. These form an
orthogonal basis on ( ) , −∞ +∞ with respect to the weight function
2
( )
x
w x e
−
= . The
“physicists” Hermite polynomials (Abramowitz and Stegun, 1964) are defined recursively by
( ) ( ) ( )
1 1
2 2
n n n
H x xH x nH x
+ −
= − with ( )
0
1 H x = and ( )
1
2 H x x = .
Using the orthogonality condition of the “physicists” Hermite polynomials it follows
that the expansion coefficients are given by ( )
1
2 !
n n n
x x
a z H p x dx
n
π µ µ
σ σ
∞
−
−∞
− −
   
=
 
\ ¹ \ ¹
∫
.
5
These coefficients are no longer a linear combination of the moments, but of some “weighted”
moments, i.e.
n
x x
E z
µ µ
σ σ
(
− −
   
(
 
\ ¹ \ ¹
(
¸ ¸
, of the random variable with probability distribution
function ( ) p x . Later in the paper we will present some methods for calibrating these
expansions coefficients.
The improved convergence properties for heavy tailed distributions of the Gauss
Hermite series expansion are illustrated in Figure 2, for the NIG distributions with the same
parameter sets as in the previous exercise.
Figure 2. The modified GramCharlier (GaussHermite) approximation of the NIG distribution
4 3 2 1 0 1 2
0
1
2
3
4
5
6
standard deviations
p
d
f
GaussHermite approx.
Gaussian
NIG
4 3 2 1 0 1 2
7
6
5
4
3
2
1
0
1
2
standard deviations
l
o
g
p
d
f
GaussHermite approx.
Gaussian
NIG
A
4 3 2 1 0 1 2
0
0.5
1
1.5
2
2.5
3
3.5
standard deviations
p
d
f
GaussHermite approx.
Gaussian
NIG
4 3 2 1 0 1 2
8
7
6
5
4
3
2
1
0
1
2
standard deviations
l
o
g
p
d
f
GaussHermite approx.
Gaussian
NIG
B
The graphs depict the probability distribution function (pdf) and the logarithm of the pdf of the target distribution (NIG), of
the Gaussian distribution, and the modified GramCharlier (GaussHermite) approximation truncated after N=25 terms. In
panel a the target distribution has mean 0.0012, standard deviation 0.10, skewness 1.6 and excess kurtosis 5, and in panel b
the target distribution has mean 0.005, standard deviation 0.20, skewness 2 and excess kurtosis 10.
There is one drawback of the GaussHermite approximation relative to the Gram
Charlier approximation related to condition that the total mass of the density should be 1. Due
6
to the properties of the “probabilists” Hermite polynomials, in the GramCharlier
approximation this condition is satisfied independent of the number of terms used in the
approximation. On the other hand for the GaussHermite series expansion the condition is
valid only in the limit since the requirement of unitary mass is equivalent to the identity
( )
2
0
2 !
1
!
k
k
k
a
k
∞
=
=
∑
. However, if the truncation is done after a large number of terms are
included, the mass of the distribution should be very close to 1. Alternatively one could
normalize the expansion coefficients such as the truncated sum adds to 1.
The following lemma points out that the characteristic function can also be easily
expanded using the same expansion coefficients.
Lemma 1. (Characteristic function representation) Consider a probability density
function ( ) p x with mean µ , standard deviation σ
and GaussHermite expansion coefficients
( )
n
n
a
∈N
. Then the associated characteristic function can be written as:
( ) ( )
2
2
0
exp
2
n
n n
n
i a i H
σ
ϕ φ µφ φ σφ
∞
=
 
= −

\ ¹
∑
(3)
where 1 i = − .
Proof: If one denotes by ( ) p x ɶ ɶ the standardized density, one has that ( )
1 x
p x p
µ
σ σ
−
 
=

\ ¹
ɶ .
Using a wellknown property of the “physicists” Hermite polynomials, namely
( )
2 2
2 2
n
x x
n
d
e H x x e
dx
− −
 
= −

\ ¹
⋅
, the GaussHermite expansion of ( ) p x ɶ ɶ can be written as
( ) ( ) ( ) ( )
0 0
n
n n n
n n
d
p x z x a H x a x z x
dx
∞ ∞
= =
 
= = −

\ ¹
∑ ∑
ɶ ɶ ɶ ɶ ɶ ɶ
ɶ
. Therefore, the Fourier transform of pɶ is
( )
2 2
0 0
1 1
exp exp
2 2
n
n n
n n n
n n
d
a i a i H
d
φ φ φ φ
φ
∞ ∞
= =
 
   
− − = −
  
\ ¹ \ ¹
\ ¹
∑ ∑
. The Fourier transform of p
follows immediately. ■
3. Option pricing
The GaussHermite series expansion is an attractive alternative for approximating the
riskneutral measure density and, as the following result points out, it allows for a closed form
formula for pricing European options.
7
Proposition 1. (Option pricing formula) Assume that the logreturn riskneutral
measure for time horizon τ is characterized by an annualized mean µ , an annualized
standard deviation σ and GaussHermite expansion coefficients ( )
n
n
a
∈N
.
Then the premium at time t of an European call option with strike price K and maturity
t τ + is given by:
( )
1 2
, , , , , ,
q r
t t
c S K r q S e Ke
τ τ
µ σ τ
− −
= Π − Π (4)
where
t
S is the spot price of the underlying, r is the riskfree interest rate, q denotes the
dividend yield. We have
2
1
0
exp ( )
2
n n
n
r q a I
σ
µ τ
∞
=
¦ ¹
 
Π = − − +
´ `

\ ¹ ¹ )
∑
and
2
0
n n
n
a J
∞
=
Π =
∑
where
n
I and
n
J satisfy the recursion equations ( ) ( )
1 1 2 1
2 2 2
n n n n
I z d H d I nI σ τ
+ −
= − − + + and
( ) ( )
1 2 2 1
2 2
n n n
J z d H d nJ
+ −
= − − + with ( )
0 1
I N d = , ( ) ( )
1 1 1
2 2 I z d N d σ τ = − + ,
( )
0 2
J N d = , ( )
1 2
2 J z d = − ,
( ) ( )
2
1
log
t
S K
d
µ σ τ
σ τ
+ +
= ,
2 1
d d σ τ = − , and ( ) N ⋅ is the
cumulative distribution function of the standard Gaussian distribution.
Proof: If one denotes by ( )
t t
p S
τ τ + +
the terminal underlying asset price riskneutral density
and by ( ) p x the standardized logreturn riskneutral density then we have that:
( ) ( ) ( ) , , , , , , max , 0
r
t t t t t
c S K r q e S K p S dS
τ
τ τ τ τ
µ σ τ
∞
−
+ + + +
−∞
= −
∫
( )
( ) max , 0
r x
t
e S e K p x dx
τ µτ σ τ
∞
− +
−∞
= −
∫
( )
( )
2
1 2
r x q r
t t
d
e S e K p x dx S e Ke
τ µτ σ τ τ τ
∞
− + − −
−
= − = Π − Π
∫
with ( )
2
( )
1
r q x
d
e e p x dx
µ τ σ τ
∞
− +
−
Π =
∫
and ( )
2
2
d
p x dx
∞
−
Π =
∫
.
Taking into account the GaussHermite expansion of the logreturn riskneutral density, one
has that
2
1
0
exp ( )
2
n n
n
r q a I
σ
µ τ
∞
=
¦ ¹
 
Π = − − +
´ `

\ ¹ ¹ )
∑
with ( ) ( )
2
2
2
x
n n
d
I e e z x H x dx
σ τ
σ τ
∞
−
−
= ⋅ =
∫
( )
( ) ( ) ( )
( )
2
1 2
2
x
n n
d d
e e z x H x dx H x z x dx
σ τ
σ τ σ τ
σ τ
σ τ σ τ σ τ
∞ ∞
−
+
− − −
+ + = + ⋅
∫ ∫
and
8
2
0
n n
n
a J
∞
=
Π =
∑
with ( ) ( )
2
n n
d
J H x z x dx
∞
−
=
∫
. Using the properties of GaussHermite
polynomials namely ( ) ( ) ( )
1 1
2 2
n n n
H x x H x n H x
+ −
= − , ( )
'
1
2 ( )
n n
H x n H x
−
= and integration
by parts, one can obtain the recursion equations for
n
I and
n
J as follows:
( )
( )
( ) ( ) ( )
( )
1 1
1 1
2
n n n n
d d
I H x z x dx x H x H x z x dx σ τ σ τ σ τ σ τ
∞ ∞
+ +
− −
(
′ = + = + + − +
¸ ¸
∫ ∫
( )
( )
( )
( )
( )
( )
1 1 1
2 2
n n n
d d d
H x z x dx H x z x dx H x z x dx σ τ σ τ σ τ σ τ
∞ ∞ ∞
− − −
′ ′ = − + + + − +
∫ ∫ ∫
( )
( )
1 1 1
2 2 2
n n n
H d z d I nI σ τ σ τ
−
= − + − + + where we have used '( ) ( ) z x xz x = − .
( ) ( ) ( ) ( ) ( )
2 2
1 1
2
n n n n
d d
J H x z x dx xH x H x z x dx
∞ ∞
+ +
− −
′ = = − (
¸ ¸ ∫ ∫
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
2 2
2 2 2 2 1
2 2 2
n n n n
d d
H d z d H x z x dx H d z d n H x z x dx
∞ ∞
−
− −
′ = − − + = − − +
∫ ∫
( ) ( )
2 2 1
2 2
n n
H d z d nJ
−
= − − + .
Finally, we have ( ) ( )
1
0 1
d
I z x dx N d
∞
−
= =
∫
and similarly ( ) ( )
2
0 2
d
J z x dx N d
∞
−
= =
∫
.■
Remark 1. If one assumes that the risk neutral measure density is normal then the
GaussHermite expansion coefficients are given by
0
1 a = and 0
n
a = , 1 n ≥ and, therefore,
equation (4) reduces to the BlackScholesMerton formula. ■
When using series expansions to approximate riskneutral densities one has to
determine the martingale restriction associated to that expansion. Lemma 1 allows us to easily
derive the martingale restriction in the case of the GaussHermite expansion.
Corollary 1. (Martingale restriction) Assume that the logreturn riskneutral measure
for time horizon τ is characterized by an annualized mean µ , an annualized standard
deviation σ , and GaussHermite expansion coefficients ( )
n
n
a
∈N
. Given that r is the riskfree
interest rate and q denotes the dividend yield then the martingale restriction associated to the
GaussHermite expansion is:
9
( )
2
0
exp ( ) 1
2
n
n n
n
r q a i H i
σ
µ τ σ τ
∞
=
¦ ¹
 
− − + − ⋅ =
´ `

\ ¹ ¹ )
∑
(5)
Proof: The martinagle condition requires
r q
t t
E e S e S
τ τ
τ
− −
+
( =
¸ ¸
. Denoting by ϕ the Fourier
transform of the logreturn riskneutral measure density, the result follows immediately from
Lemma 1 since ( ) { }
( ) r q r q
t t
E e S e S e i
τ τ τ
τ
ϕ
− − − −
+
( = −
¸ ¸
■
4. Calibrating the expansion coefficients to option data
Before presenting the result relating the expansion coefficients to observed option
data, it is interesting to point out that the GaussHermite expansion coefficients could be
computed from the associated characteristic function. More specifically, from equation (3)
and from the orthogonality condition of the “physicists” Hermite polynomials it follows that
the expansion coefficients of a density with mean µ , standard deviation σ and characteristic
function ϕ can be represented as:
( )
2
1
exp exp
2
2 !
n n
n n
a H i d
n i
φ φ µ
φ ϕ φ φ
σ σ
π
∞
−∞
 
   
= − −
  
\ ¹ \ ¹
\ ¹
∫
(6)
Therefore, given an option pricing model for which the characteristic function is
known, such as the Heston model, equation (6) together with equation (4) provide an
alternative method of obtaining option prices.
We now show how to obtain the GaussHermite expansion coefficients from option
market prices. The following proposition states the result.
Proposition 2. (Calibration to option data) Given option prices for time horizon τ
and given that the logreturn riskneutral measure density for time horizon τ is characterized
by an annualized mean µ and an annualized standard deviation σ then the GaussHermite
expansion coefficients can be computed as:
( )
( ) ( )
( )
( ) ( )
( )
( ) ( )
( )
( ) ( )
2 2 1
0
2 !
, , , ,
r q
t
r q
t
r q r q
n t n t n
S e
r
n t n t
S e
a z d S e H d S e
n
e F K c S K dK F K p S K dK
τ
τ
τ τ
τ
π
τ τ
−
−
− −
−
∞
¦
= − +
´
¹
¹ (
¦
( + +
`
(
¦
¸ ¸ )
∫ ∫
(7)
10
where
2
log
( )
t
S
K
d K
µτ
σ τ
 
+

\ ¹
= ,
t
S is the spot price of the underlying, r is the riskfree
interest rate, q is the dividend yield, ( ) , ,
t
c S K τ and ( ) , ,
t
p S K τ denotes the premium at time
t of an European call and put, respectively, with strike price K and maturity t τ + and the
function ( )
n
F ⋅ is given by
( )
( )
( )
( )
( )
( )
2 2
2 2 2 2 1 2 2 2 2 2
2
( )
1 4 4 ( 1) ( )
n n n n
z d
F K d d H d nd H d n n H d n
K
σ τ σ τ
σ τ
− −
= − − − + − + − − −
Proof: One has that ( )
1
2 !
n n n
x x
a z H p x dx
n
π µτ µτ
σ τ σ τ
∞
−
−∞
− −    
=
 
\ ¹ \ ¹
∫
where ( ) p x logreturn
riskneutral measure density for time horizon τ . By a change of variable we obtain that
( ) ( )
1
0
2 !
n n t t t t n
a G S p S dS
n
τ τ τ τ
π
∞
+ + + + −
=
∫
where ( )
t t
p S
τ τ + +
is the terminal underlying asset price
riskneutral density and ( )
( ) ( ) log log
t t
n n
S S S S
G S z H
µτ µτ
σ τ σ τ
 −   − 
=
 
\ ¹ \ ¹
. The proposition
follows in a straightforward way from a general result in Bakshi, Kapadia and Madan (2003)
based on the fact that any function with bounded expectations can be spanned by a continuum
of OTM European call and put options (Bakshi and Madan, 2000). More specifically, one has
to apply equation (3) in Bakshi, Kapadia and Madan (2003) to the function ( )
n
G S . ■
Remark 2. In order to implement equation (7) one needs the mean and the standard
deviation of the logreturn riskneutral measure density. These can be computed using the
method described in Bakshi, Kapadia and Madan (2003). ■
In what follows we conduct a series of calibration exercises in order to investigate the
performance of the option pricing model based on the GaussHermite approximation to
adequately reflect the observed volatility smile.
4.1. Calibration to simulated option data
In order to simulate the option prices we employ the Heston stochastic volatility model with
the following parameter values: 4.25 k = , 0.08 θ = , 0.8 σ = , 0.85 ρ = − and
0
0.03 v = .
These values are similar to those obtained in Gourier, Bardgett and Leippold (2012) by
calibrating the Heston model to all S&P 500 index options traded on 21 October 2010. We
11
assume that the current underlying price is 100 and we generate a set of 3 months European
call and put option prices with strike prices between 35 and 155 with a step of 2.5 units. The
riskfree interest rate is set to 0.05 r = and the dividend yield to 0.03 q = . Since the
characteristic function of the Heston model is known in closed form we can compute the
expansion coefficients either by using equation (6) or directly from the generated option
pricing using (7). In all the computation we truncated the GaussHermite expansion after
N=25 terms. Figure 3 depicts the approximated density and in the case the expansion
coefficients are computed using the characteristic function (panel a) or are inferred from
option prices (panel b).
Figure 3. The GaussHermite approximation of the Heston model distribution
4 3 2 1 0 1 2
0
1
2
3
4
5
6
standard deviations
p
d
f
GaussHermite approx.
Heston
Gaussian
4 3 2 1 0 1 2
7
6
5
4
3
2
1
0
1
2
standard deviations
l
o
g
p
d
f
GaussHermite approx.
Heston
Gaussian
a
4 3 2 1 0 1 2
0
1
2
3
4
5
6
standard deviations
p
d
f
GaussHermite approx.
Heston
Gaussian
4 3 2 1 0 1 2
7
6
5
4
3
2
1
0
1
2
standard deviations
l
o
g
p
d
f
GaussHermite approx.
Heston
Gaussian
b
The graphs depict the probability distribution function (pdf) and the logarithm of the pdf of the target distribution (Heston
model), of the Gaussian distribution, and the modified GramCharlier (GaussHermite) approximation truncated after N=25
terms. In panel a the expansion coefficients are computed using the characteristic function, and in panel b are inferred from
the simulated option prices.
Next we employed the option pricing formula (4) truncated after N=25 terms and
derived the implied volatility curve as depicted in Figure 4. In panel a the expansion
12
coefficients are computed using the characteristic function, and in panel b they are estimated
from option prices. We do not constrain the coefficients in order to observe the martingale
restriction. However, the truncated sum in equation (5) is quite close to 1 being equal to
1.002.
Figure 4. Implied volatility curves inferred from simulated option prices
0.5 0.4 0.3 0.2 0.1 0 0.1 0.2
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
i
m
p
l
i
e
d
v
o
l
a
t
i
l
i
t
y
log (K/F)
IV Heston
IV GaussHermite approx.
0.5 0.4 0.3 0.2 0.1 0 0.1 0.2
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
i
m
p
l
i
e
d
v
o
l
a
t
i
l
i
t
y
log (K/F)
IV Heston
IV GaussHermite approx.
a b
The graphs depict the implied volatility curves of the Heston model and of the option pricing model based on the Gauss
Hermite approximation truncated after N=25 terms. In panel a the expansion coefficients are computed using the
characteristic function, and in panel b are inferred from the simulated option prices
The GaussHermite implied volatility curve is a good approximation of the Heston
model one for a range of strike prices spanning five standard deviations of the logreturn risk
neutral density, four standard deviations to the left and one to the right of the mean return or,
equivalently, for log( ) K F in the interval (0.4, 0.1), where F is the forward price.
4.2. Calibration to market option data
In this section we employ market data about European options in order to infer the
implied volatility curve of option pricing model based on the GaussHermite approximation.
More specifically, we focus on SPX options quotes on 15 September 2011, the day before
triplewitching, and compute implied volatilities for maturities of 1 month and 3 months. The
implied volatility surface on this specific date was also analysed by Gatheral and Jacquier
(2012) with a SVI parameterization. Data on option prices, the term structure of interest rate
and the dividend yield on this specific date are obtained from OptionMetrics. In order to
estimate the expansion coefficients, we determined the implied volatilities associated to mid
prices of the put options and then the call and put prices that appear in the integrals in
equation (7) are computed by interpolation. We employed the option pricing formula (4)
13
truncated after N=15 terms. We do not constrain the coefficients in order to impose the
martingale restriction and the truncated sum in equation (5) is around 1.006. The resulting
implied volatility curves of the GaussHermite approximation are depicted in Figure 5.
Figure 5. Implied volatility curves inferred from market option prices
0.4 0.3 0.2 0.1 0 0.1 0.2
0.2
0.3
0.4
0.5
0.6
0.7
0.8
i
m
p
l
i
e
d
v
o
l
a
t
i
l
i
t
y
log (K/F)
IV GaussHermite approx.
IV bid
IV ask
1 0.8 0.6 0.4 0.2 0 0.2 0.4
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
i
m
p
l
i
e
d
v
o
l
a
t
i
l
i
t
y
log (K/F)
IV GaussHermite approx.
IV bid
IV ask
1 month 3 months
The graphs depict the implied volatility curves for 1 month and 3 months of the option pricing model based on the Gauss
Hermite approximation truncated after N=15 terms.
The GaussHermite approximation seems to perform quite well for an interval of
strike prices that spans more than five standard deviations of the logreturn riskneutral
density, four standard deviations to the left and one and a half to the right of the mean return
or for a log( ) K F in the interval (0.4, 0.15) for 1 month maturity and in the interval (0.8,
0.3) for 3 months maturity.
5. Concluding remarks
In this paper we developed a new method to retrieve the riskneutral measure density
from option prices using the GaussHermite series expansion around the Gaussian density and
pointed out its better convergence properties compared to the GramCharlier expansion. We
also derived several methods for obtaining the expansion coefficients. More specifically, one
can obtain the coefficients of the GaussHermite expansion from the probability distribution
function, from the characteristic function, or directly from market option prices.
Approximating the riskneutral density using the GaussHermite expansion is quite
appealing because it allows for a closed form option pricing model that embeds the classical
BlackScholesMerton formula. This option pricing formula based on the GaussHermite
expansion is an alternative to the inverse Fourier transform methodology and is quite general
14
since it can be employed for models with the probability distribution function known in
closed form, for models with the characteristic function known in closed form, and can also
be calibrated to market option prices.
We calibrated the new option pricing model to option prices simulated using Heston
stochastic volatility model and to market option prices. These calibration exercises have
revealed that the resulting implied volatility curves are quite accurate for a range of strike
prices that spans five standard deviations of the logreturn riskneutral density, four standard
deviations to the left and one to the right of the mean return.
References
Abramowitz, M., and I.A. Stegun, (Eds.) (1964). Handbook of Mathematical Functions,
National Bureau of Standards.
AitSahalia, Y. and A.W. Lo. (1998). Nonparametric estimation of stateprice densities
implicit in financial asset prices. Journal of Finance 53 (1998) 499–548
AitSahalia, Y., and J. Duarte. (2003). Nonparametric option pricing under shape restrictions.
Journal of Econometrics, 116, 9–47.
Bakshi, G. and D. Madan. (2000). Spanning and derivativesecurity valuation. Journal of
Financial Economics, 55 205–238.
Bakshi, G., N. Kapadia, and D. Madan. (2003). Stock Return Characteristics, Skew Laws, and
the Differential Pricing of Individual Equity Options, Review of Financial Studies,
16(1): 101143
Bibby, B. M. and M. Sorensen. (1997). A hyperbolic diffusion model for stock prices.
Finance & Stochastics 1, 2541
Black, F., and M. Scholes. (1973). The pricing of options and corporate liabilities. Journal of
Political Economy, 81, 637–654.
Brown, C. and D. Robinson. (2002). Skewness and kurtosis implied by option prices: A
correction. Journal Financial Research 25 279–81.
Corrado, C. J., and T. Su. (1996). S&P 500 index option tests of Jarrow and Rudd’s
approximate option valuation formula. Journal of Futures Markets, 16, 611–629.
15
Corrado, C. J., and T. Su. (1997). Implied volatility skews and stock index skewness and
kurtosis implied by S&P 500 index option prices. Journal of Derivatives, 4, 8–19.
Corrado, C. J. (2007) The hidden martingale restriction in GramCharlier option prices,
Journal of Futures Markets, vol. 27, no. 6, pp. 517534.
Cramer. H. (1957). Mathematical Methods of Statistics. Princeton University Press, Princeton.
Eriksson, A., L. Forsberg, and E. Ghysels. (2004). Approximating the Probability Distribution
of Functions of Random Variables: A New Approach, Discussion paper, CIRANO.
Gatheral, J., and A. Jacquier. (2012). ArbitrageFree SVI Volatility Surfaces. available at
SSRN: http://dx.doi.org/10.2139/ssrn.2033323
Gourier, E., C. Bardgett, and M. Leippold. (2012). Inferring volatility dynamics from the
S&P500 and VIX markets. Presentation, University of Zurich
Heston, S. L., (1993). A closedform solution for options with stochastic volatility with
applications to bond and currency options. Review of Financial Studies 6 327–343.
Jarrow, R., and A. Rudd. (1982). Approximate option valuation for arbitrary stochastic
processes. Journal of Financial Economics, 10, 347–369.
Jondeau, E., and M. Rockinger. (2001). Gram–Charlier densities. Journal of Economic
Dynamics and Control, 25, 1457–1483.
Jurczenko, E., B. Maillet, and B. Negrea. (2002). Revisited multimoment approximate option
pricing models: A general comparison (Discussion Paper of the LSEFMG, No. 430).
Jurczenko, E., B. Maillet, and B. Negrea. (2004). A note on skewness and kurtosis adjusted
option pricing models under the martingale restriction. Quantitative Finance, 4, 479–
488.
Kou, S. G. (2002). A JumpDiffusion Model for Option Pricing. Management Science, 48, 8,
10861101
Longstaff, F. A. (1995). Option pricing and the martingale restriction. Review of Financial
Studies, 8, 1091–1124.
Merton, R. (1973). Rational theory of option pricing. Bell Journal of Economics and
Management Science, 4, 141–183.
Merton, R. C. (1976). Option pricing when underlying stock returns are discontinuous.
Journal of Financial Economics 3, 125144.
16
MyllerLebedeff, W. (1907). Die Theorie der Integralgleichungen in Anwendung auf einige
Reihenentwicklungen. Mathematische Annalen. 64, 388416
Rompolis, L.S. and E. Tzavalis. (2007). Retrieving risk neutral densities based on risk neutral
moments through a Gram–Charlier series expansion. Mathematical and Computer
Modelling 46 225–234
van der Marel R.P. and Franx M., (1993). A new method for the identification of non
Gaussian line profiles in elliptical galaxies. Astrophysical Journal. 407, 525 – 539
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.