Merton Jump - Diffusion.model
Merton Jump - Diffusion.model
Merton Jump - Diffusion.model
March 5, 2012
In Mertons jump-diffusion model for option pricing, the price process of an underlying asset
St is assumed to follow the stochastic differential equation (sde)
N (t)
X
dSt = St dt + St dWt + St d (Vj 1) ,
j=0
where Vj are i.i.d. log-normal random variables. This SDE can be solved to obtain
St = S0 exp
NY
(t)
1 2
t + Wt
Vj .
2
j=0
t + Wt +
N (t)
X
Yj .
j=0
2
N (t)
iu 2 t
Yj
E eiuXt = E e
E eiuWt E exp iu
j=0
N (t)
X
2
2
2
iu 2 t u t
2
= e
Yj ,
E exp iu
e
j=0
where
N (t)
X
E exp iu
Yj = E eiuY0 +iuY1 +iuY2 ++iuYN (t)
j=0
N (t)iuY0
= E E e
N (t)
X
et (t)k iuY0 k
E e
.
=
k!
k=0
(1)
2 u2
E eiuY0 = eiu E eiuZ = eiu 2 .
Substitute this into Eq. (1) to obtain
2 u2
N (t)
teiu 2
X
E exp iu
Yj = et
k!
j=0
k=0
= e
t eiu
2 u2
2
!
1
2 u2
2
2 2
iu 2 t u 2 t+t eiu 2 1
E eiuXt = e
Cumulants
E esXt = e
2 s2
2
2 2
s 2 t+ s 2 t+t es+ 2 1
2 2
sXt
2
s2 2
s+ 2s
t + t e
1 ,
ln E e
=s
t+
2
2
(2)
the power series expansion of which gives the cumulants. Here are the first few cumulants
n :
2
1 = t + t
,
2
2 = t 2 + 2 + 2 ,
3 = t 3 + 3 2 ,
4 = t 4 + 62 2 + 3 4 ,
5 = t 5 + 103 2 + 15 4 ,
6 = t 6 + 154 2 + 452 4 + 15 6 ,
E eiuXt = e
2 2
2
iu 2 t u 2 tt
teiu
k!
k=0
k!
k=0
X
et (t)k
2 u2
2
2
2
iu 2 t+k u2 ( 2 t+k 2 )
Taking the inverse Fourier transform, we obtain the probability density function (pdf) of Xt
fXt (x) =
X
et (t)k
k=0
k!
1
2( 2 t
k 2 )
12
2
x
2 t+k
2 t+k 2
Calibration
There are five parameters in Mertons jump-diffusion model: , , , , . One way to determine their values to fit the market data is to solve the following system of equations:
2
t + t
= m1
(3)
2
t 2 + 2 + 2 = m2
(4)
3
t + 3 2 = m3
(5)
4
2 2
4
t + 6 + 3 = m4
(6)
5
3 2
4
t + 10 + 15 = m5
(7)
where the left hand side are the first five cumulants of Mertons jump-diffusion model in
terms of the five parameters, which we have discussed before, and the right hand side are the
first five sample cumulants obtained from market data. This system can be solved explicitly.
First we set = 2 , = 2 to simplify the notations. If we multiply Eq. (5) by 4 and
Eq. (6) by and then add the resulting equations together, we have
m4 + 4m3 = m5 .
(8)
(m2 t) + 2 m1
t = m3 .
2
(9)
(10)
Now Eqs. (8) and (9) can be thought of as a linear system of and , which gives the
solution
3m5 (t m2 ) + 4m3 m4
,
3m4 (t m2 ) + 4m23
m3 m5 m24
=
.
3m4 (t m2 ) + 4m23
(11)
(12)
(13)
Now , , and are given in Eqs. (11), (12) and (14) respectively, all in terms of .
From Eq. (3), we have
1 h m1
i
=
;
t
2
(15)
m2 /t
.
2 +
= 0.
+
t
t
2
Finally, substitute Eqs. (11), (12) and (14) into the above equation and further simplify it
to obtain a quartic equation of
a4 4 + a3 3 + a2 2 + a1 + a0 = 0,
where
a2
a3
3
2
3
2
3
2
3
3
2 2
2
2
6
2 5
4
16m5 m3 + 48m4 m3 168m2 m4 m5 m3 + 8m2 9m2 m5 5m4 m3 + 225m2 m4 m5 m3 9m2 m4 5m4 + 18m2 m5 m3 + 27m2 m5 m4 + m2 m5 ,
3
2
3
2
2
4
3
2
3
2
2
t,
168m4 m5 m3 + 8 5m4 18m2 m5 m3 450m2 m4 m5 m3 + 18m2 m4 5m4 + 27m2 m5 m3 27m2 m5 3m4 + 4m2 m5
3
2
2
2 3
2
2
4
2
t ,
9 8m5 m3 + 25m4 m5 m3 5m4 + 54m2 m5 m4 m3 + 9m2 m5 m4 + 2m2 m5
3
2
3
27m5 m4 6m3 m5 m4 + 4m2 m5 t ,
a4
27m5 t .
a0
a1
3 4
A quartic equation can be solved analytically. Once we obtain the value of , Eqs. (11),
(12), (15) and (14) can be used to find , , and , respectively. In many cases the above
quartic equation has positive real roots. When there are more then one, all the positive real
roots of will gives parameters that fit the market data very well. When no positive real
root exist, an alternative choice for calibration is the method of least squares.
t+t
E eXt = e
e+
2
2
!
1
Comparing the above two identities gives the under risk-neutral probability
2
+ 2
=r e
1 .
5
A General Formula
Before going to option pricing, we first derivative a more general formula to help to simplify
the computations. In many cases, the expectation E [(S0 D K)+ ] is needed when deriving
the price of an option with strike price K and the initial price of underlying asset is S0 . Now
we derive a general formula for this expectation with log D N (A, B 2 ) A + BZ. Note
that
where
B = KP(S0 D > K)
!
ln SK0 A
ln SK0 + A
= KP Z >
= KN
.
B
B
Here we use the identity P(Z > x) = N (x). On the other hand, we have
A = E S0 eA+BZ 1(
Z>
Z
A
= S0 e
ln K A
S0
B
2
A+ B2
= S0 e
Bz
ln K A
S0
B
z2
1
e 2 dz = S0 eA
2
ln K A
S0
B
B
(zB)2 B 2
1
2
e
dz
2
!
S0
+
A
ln
B2
1 w2
K
e 2 dw = S0 eA+ 2 N
+B ,
B
2
ln K A
S0
B
!
S0
S0
2
ln
+
A
ln
+
A
B
K
K
E (S0 D K)+ = S0 eA+ 2 N
+ B KN
.
B
B
(16)
2
with A = (r /2)T , B = T .
Perhaps the most convenient way to derive the option prices is to use the law of iterated
expectations. Notice that
PN (T )
2
2 T +WT + j=1 Yj
= erT E E S0 e
K
N (T ) = k
"
2
+ #
Pk
T
k
X
e
(T
)
2 T +WT + j=1 Yj
= erT
E S0 e
K
,
(17)
k!
k=0
6
where
k
p
2
X
2
2
2
2
T + WT +
Yj N
T + k,
T + k
.
2
2
j=1
p
Substituting Eq. (16) into Eq. (17) with A = ( 2 /2) T + k, B = 2 T + k 2 yields
S0
2
2
2
+
+
T
+
k
(
+
)
T
k
ln
X
K
2
e
(T ) T +k + 2
p
S0 e
C = erT
N
2
2
k!
T + k
k=0
2
ln SK0 + 2 T + k
.
p
KN
2 T + k 2