Sub-fractional Brownian motion as a model in

nance
Constantin Tudor: University of Bucharest
(Pitesti, May 2008)

CONTENTS

1. Fractional and sub-fractional Brownian motions.

General properties
2. Pathwise integral with respect to sfBm
3. Wiener integral. Sub-fractional Girsanov theorem
4. Anticipating stochastic calculus for sfBm
and Clark-Ocone representation formula
5. Sub-fractional Black-Scholes model
1. Fractional and sub-fractional Brownian motions.
General properties

The fractional Brownian motion (fBm for short) is the best known and most used
process with long-dependence property for models in telecommunication, turbu-
lence, nance, etc. This process was rst introduced by Kolmogorov (1940) and
later studied by Mandelbrot and his coworkers (1968). The fBm is a continuous
centered Gaussian process Btk , starting from zero, with covariance
t2R

1
CB k (s; t) := E Btk Bsk = jsj2k+1 + jtj2k+1 jt sj2k+1 ; s; t 2 R;
2
(1.1)
1 1
where k 2 ;
2 2
(H = k + 12 is called Hurst parameter). The case k = 0
corresponds to the Brownian motion.
The self-similarity and stationarity of the increments are two main properties for
which fBm enjoyed success as a modeling tool. The fBm is the only continuous
Gaussian process which is self-similar and has stationary increments.
An extension of Bm which preserves many properties of the fBm, but not the
stationarity of the increments, is so called sub-fractional Brownian motion (sfBm
for short), i.e.,a continuous Gaussian process Stk , starting from zero, with
t 0
covariance
1h i
CS k (s; t) := E Stk Ssk = s2k+1 + t2k+1 (s + t)2k+1 + jt sj2k+1 ; s; t 0:
2
(1.2)
Next we assume k 6= 0:
The sfBm has properties analogous to those of fBm (see Bojdecki, Gorostiza and
Talarczyk, 2004, Dzaparidze-Van Zanten, 2004 and Tudor, 2007):
k
(i1 ) Self-similarity: For each a > 0 the processes Sat has the same distribution
t 0
k+ 21
as a Stk:
t 0
(i2 ) Covariance: For all s; t 0;
CS k (s; t) > 0;

1
CS k (s; t) > CB k (s; t) if k 2 ;0 ;
2

2
 1
CS k (s; t) < CB k (s; t) if k 2 0; :
2
(i3 ) Non-stationarity of increments: For all s t;
2
E Stk Ssk = 22k (t2k+1 + s2k+1 ) + (t + s)2k+1 + (t s)2k+1 ;

2
E Stk = (2 22k )t2k+1 ;
2 1
(t s)2k+1 E Stk Ssk (2 22k ) (t
;0 ; s)2k+1 if k 2
2
2 1
(2 22k ) (t s)2k+1 E Stk Ssk (t s)2k+1 if k 2 0; :
2
(i4 ) Correlation of increments and long-range dependence: For 0 u < v s<
t; de ne h i
k
Ru;v;s;t = E Bvk Buk Btk Bsk ;
h i
k
Cu;v;s;t =E Svk Suk Stk Ssk :
Then
1h
k
Cu;v;s;t = (t + u)2k+1 + (t u)2k+1 + (s + v)2k+1 + (s v)2k+1
2
i
(t + v)2k+1 (t v)2k+1 (s + u)2k+1 (s u)2k+1 ;

k k 1
Ru;v;s;t < Cu;v;s;t < 0 if k 2
;0 ;
2
k k 1
0 < Cu;v;s;t < Ru;v;s;t if k 2 0; :
2
For u 0; r > 0 let B k (u; r) and S k (u; r) denote the correlation coe cients of
k
Bu+r Buk ; Bu+2r
k k
Bu+r and Su+r k
Suk ; Su+2r
k k
Su+r .
Then
j S k (u; r)j j B k (u; r)j :
and we have the long-range dependence
k 2k 1
Ru;v;s+ ;t+ k(2k + 1) (t s) (v u) as ! 1;

k
Cu;v;s+ ;t+ k(2k + 1)(1 2k) v 2 u2 2(k 1)
as ! 1:

3
Therefore the covariance of increments of sfBm over non-overlapping intervals
have the same sign but are smaller in absolute value than those of fBm and the
increments on the intervals [u; u + r] ; [u + r; u + 2r] are more weakly correlated
than those of fBm. Moreover the long-range dependence decays at a higher rate
for sfBm than for fBm (these properties justi es the name sfBm).
(i5 ) Short memory: For each a > 0;
X
k
cov Sa+1 Sak ; Sn+1
k
Snk < 1:
n a+1

The above mentioned properties make sfBm a possible candidate for models which
involve long-dependence, self-similarity and non-stationarity of the increments.
(i6 ) S k is not a Markov process.
(i7 ) H•older paths: For each " < k + 12 and each T > 0 there exists a random
variable K";T such that
1
Stk Ssk K";T jt sjk+ 2 "
; s; t 2 [0; T ] ; a:s:

(i8 ) Variation: For each T > 0;

n
X1 p
L2 2
S k(i+1)T S kiT ! 0 if p > ;
i=0 n n n!1 2k + 1
n
X1 p
L2 2
S k(i+1)T S kiT ! 2 T if p =
i=0 n n n!1 2k+1 2k + 1
n
X1 p
p 2
S k(i+1)T S kiT ! 1 if p < ;
i=0 n n n!1 2k + 1
where p = E (jN (0; 1)jp ) :
Such a result for fBm is obtained mainly by using self-similarity and stationarity of
the increments (in particular ergodic theorem). For sfBm the lack of stationarity
of the increments is replaced by linear regression.
(i9 ) If W is a Brownian motion independent of S k and k > 14 , then the process
S k + W is a semimartingale equivalent in law with W .
(i10 ) The fBm S k is not a quasimartingale. In particular S k is not a semimartin-
gale.

4
2. Pathwise integral with respect to sfBm
R
For functions of one variable the Riemann-Stieltjes integral 0T f (t)dg(t) was ex-
tended to functions with unbounded variation, essentially by using fractional inte-
grals or p-variation (Dudley and Norvaisa, Feyel and Pradelle, Kondurar, Mikosch
and Norvaisa, Young and Z•ahle).
For a function f : [0; T ] ! R a partition : 0 = t0 < ::: < tN = T and p 1 we
de ne the p-variation associated to by
X
vp (f; ) = jf (ti ) f (ti 1 )jp ;
i

Denote
vp0 (f ) = lim vp (f; n ); vp (f ) = sup vp (f; );
j n j!0

for all homogeneous partitions n = (iT n )i ; n ! 0:

We say that f has nite p-variation if vp0 (f ) < 1 and bounded p-variation if
vp (f ) < 1:

Remark 2.1. A function f : [0; T ] ! R has bounded p-variation if and only if

f = h g; where h : [0; T ] ! R is bounded nondecreasing nonnegative function
and g : [h(0); h(T )] ! R is p1 -H•older.

The family of functions with bounded p-variation is denoted by Wp and it becomes

a Banach space under the norm
1
kf k[p] = max vp (f ) p ; kf k1 :

We denote by H[0;T ]; the class of all -H•older functions f : [0; T ] ! R with

f (0) = 0 and de ne

jf (u) f (v)j
kf k[0;T ]; = sup :
u6=v;0 u<v T (v u)

Remark 2.2 (Young, Dudley-Norvaisa) If f 2 Wp ; g 2 Wq ; p1 + 1q > 1 and f; g

R
have no common discontinuities, then the Stieltjes integral 0T f (t)dg(t) exists as
limit of the corresponding Riemann-Stieltjes sums.

5
In particular
Rt
if f is -H•older, g is -H•older with + > 1; then the Stieltjes
integral 0 f (s)dg(s) exists and is -H•older. Moreover for every 0 < " < + 1;
Z T
f (s)dg(s) C( ; ) kf k[0;T ]; kgk[0;T ]; T 1+" ; (2.1)
0

(Feyel-Pradelle).
Concerning the variation of sfBm we have the following result.
Proposition 2.3.

2
vp0 (S k ) = 0; vp (S k ) < 1 if p > ; (2.2)
2k + 1

v 0 2 (S k ) = v 2 (S k ) = 2 ; (2.3)
2k+1 2k+1 2k+1

2
vp0 (S k ) = vp (S k ) = 1 if p < : (2.4)
2k + 1

Remark 2.4. From the above proposition it follows that a.s. S k 2 Wp if and
2
only if p > 2k+1 :
Moreover; for every process (ut )t2[0;T ] with paths a.s. in Wq with q < 1 22k ; the
R
Riemann-Stieltjes integral 0t ur dSrk is well de ned a.s. In particular ifR u has -
H•older paths for some > 1 22k ; then the Riemann-Stieltjes integral 0t ur dSrk is
well de ned and has {H•older paths, for every < k + 21 :

Since every Riemann-Stieltjes integral obeys the change of variable formula, we

have the following result.
Theorem 2.5. If F (t; x) 2 C 1 and the mapping t ! @F @x
(t; Stk ) 2 Wq with
2
q < 1 2k ; then for all s; t 2 [0; T ] ;

Z t Z t
@F @F
F (t; Stk ) F (s; Ssk )
= k k
(r; Sr )dSr + (r; Srk )dr: (2.5)
s @x s @t

Remark 2.6. (a) The pathwise integral may not exist : for example for k 2
R
1
2
; 0 the integral 0T Stk dStk does not exists. Indeed if we assume that the
integral exists, then using (2.4) we obtain
X 2 X
1 = v20 (S k ) = lim S kiT S k(i 1)T = n!1
lim S kiT S kiT S k(i 1)T
n!1 n n n n n
i i

6
 X Z T Z T
lim
n!1
S k(i 1)T S kiT S k(i 1)T = Stk dStk Stk dStk = 0:
n n n 0 0
i

(b) If the pathwise integral exists, then it may have not zero expectation: for
example if k 2 0; 12 ; by using (2.5) we obtain
Z !
T 1 2
E Stk dStk = E STk = (1 22k 1 )T 2k+1 :
0 2

3. Wiener integral and the sub-fractional Girsanov theorem

Next we restrict to a nite time interval [0; T ] : Denote by E the family of elemen-
P
tary deterministic functions f : [0; T ] ! R;i.e., f = jn 01 fj 1[tj ;tj+1 ) :
For f 2 E we de ne the Wiener integral by
n
X1
I(f ) = fj Stkj+1 Stkj ;
j 0

and the bilinear and symmetric form

hf; giS k = E (I(f )I(g)) :

The closure of the step functions with respect to the above inner product (denoted
by sf
k;T ) is called the domain of the Wiener integral.

Let f : [0; T ] ! R be a measurable application and 2 R; ; 2 R. We de ne

the Erdely-Kober-type fractional integral

s Z T t (1 ) 1
f (t)
IT ; ; f (s) = 1 dt; s 2 [0; T ] ; > 0;
( ) s (t s ) (3.1)

!n
d (n )
IT ; ; f (s) = s 1 ds
s IT +n; ; nf (s); s 2 [0; T ] ; > n;
s
(3.2)

( + ) Z + 1
s s t f (t)
I0+; ; f (s) = dt; s 2 [0; T ] ; > 0;
( ) 0 (s t )1 (3.3)

7
 !n
( + ) d ( +n+ ) +n
I0+; ; f (s) = s 1 ds
s I0+; ; f (s); s 2 [0; T ] ; > n:
s
(3.4)

We introduce the following kernels

p
n(t; s) = k ITk ;2; 1 2 k
uk 1[0;t) (s) ; (3.5)
2

Z
sk k t k
(t; s) = tk 1 2
t s 2
(k 1) u2 s2 uk 1 du 1(0;t) (s) :
(1 k) s
(3.6)
sf
Remark 3.1. The function (t; :) 2 k;T and satis es (uniquely) the equality
p
ITk ;2; 1 2 k
uk (t; :) (s) = 1(0;t) : (3.7)
2k
Theorem 3.2 (Dzhaparidze-Van Zanten, 2004, Tudor, 2007). The process
Z t
Wtk = (t; s)dSsk ;
0

is the unique Brownian motion such that

Z t
Stk = ck n(t; s)dWsk ; t 2 [0; T ] ; (3.8)
0

1
(2k + 2) sin k+ 2
c2k = : (3.9)

Moreover S k and W k generate the same ltration.

1 sf
Theorem 3.3 (Tudor, 2007). (i) If 2
< k < 0; then the space k;T ; h:; :i sf
k;T
;
where
sf
k;T =

( ! )
2 k 2k
f : [0; T ] ! R : 9'f 2 L ([0; T ]) ; IT ;2; k+1
p 'f (t) = tk f (t) ;
2
(3.10)

8
 Z T
hf; gi sf = c2k 'f (t)'g (t)dt; (3.11)
k;T 0

is the domain of the Wiener integral and

Z T Z T
f (t)dStk = ck 'f (t)dWtk : (3.12)
0 0

sf
(ii) If 0 < k < 21 ; then the space k;T ; h:; :i sf ; where
k;T

n
sf
k;T = f 2 D (0; T )0 : 9f 2 S 0 ; f odd; supp(f ) [ T; T ]

Z
2
f j[0;T ] = f; fc (x) jxj 2k
dx < 1 ; (3.13)
R

c2k Z c c 2k
hf; gi sf = f (x)g (x) jxj dx; (3.14)
k;T 2 R
is the domain of the Wiener integral:
If we de ne

j jsf k
k;T = f : [0; T ] ! R : IT ;2; 1 2 k
uk jf j 2 L2 ([0; T ]) ;
(3.15)

then we have the strict inclusion j jsf

k;T
sf
k;T and
Z Z p !
T T
f (t)dStk = ck ITk ;2; 1 k u f (t)dWtk ; f 2 L2 ([0; T ]) :
k
0 0 2 2k
(3.16)

Moreover, if k 2 0; 21 ; f 2 j jsf
k;T ;
* p ! p !+
hf; gi sf = c2k ITk ;2; 1 k k
u f ; ITk ;2; 1 k k
u g
k;T 2 2k 2 2k L2 ([0;T ])

Z T Z T
= f (u)g(v)'k (u; v)dudv: (3.17)
0 0

9
 h i
'k (u; v) = k(2k + 1) ju vj2k 1
(u + v)2k 1
:
Theorem 3.4 (Prediction). For every 0 t T;
h i Z t
k
S^Tk jt := E STk j FtS = Stk + k k
t;T (u)dSu
0

Z t
= ck n(T; u)dWuk ; (3.18)
0

Z k
k 2 sin k T (z 2 t2 ) k
t;T (u) = u(t2 u2 ) k
z dz: (3.19)
t z2 u2
k ^k
In particular (since n(T; u) > 0 on (0; T )) we have the equality FtS = FtS :

Denote
2k
dk = p ;
ck (1 k)
The process
Z t
Mtk = dk s k dWsk ; (3.20)
0

is called the sub-fractional fundamental martingale.

The following result is straightforward.
k k
Remark 3.5. For every s < t; Mtk Msk is independent of FsS and FsS =
k k
FsM = FsW :
RT
For f : [0; T ] ! R with 0 f 2 (s)s 2k
ds < 1 de ne the probability Qf by
Z t
dQf 1 Z t 2 D kE
jF Sk = exp f (s)dMsk f (s) M
dP t 0 2 0 s

Z !
t d2k Z t 2
= exp f (s)dMsk f (s)s 2k
ds ; (3.21)
0 2 0
and denote
1 k
( k f ) (s) = I0+;2; k f (s): (3.22)
(1 k)

10
Theorem 3.6 (Girsanov). For f as above, the process
Z t
Stk ( k f ) (s) ds; t 2 [0; T ] ;
0

is a Qf sfBm.
In particular if f a 2 R it follows that the process Stk at is Qa sfBm.
t2[0;T ]

4. Anticipating stochastic calculus for sfBm

and Clark-Ocone representation formula
Multiple integrals w.r.t. fBm were introduced by Dasgupta-Kallianpur and Duncan-
Hu and Pasik-Duncan for the fBm with. The techniques used in these papers
involve Wick product and reproducing kernel Hilbert space theory.
We study multiple fractional and subfractional integrals by using this transfer
principle from multiple Brownian integrals via a Gamma type operator. Then the
chaos form of the sub-fractional anticipating integral is considered.
We assume that k 2 0; 12 :
For a function f : [0; T ]n ! R we consider the n dimensional form IT ;n; ; f of
the Erdely-Kober-type fractional integrals (3.1),
!n
Q
n
IT ;n; ; f (s1 ; ::; sn ) = sj
( ) j=1

Z Z (1 ) 1
T T Q
n tj
:: 1 f (t1 ; ::; tn )dt1 ::dtn ; sj 2 [0; T ] ; > 0:
s1 sn j=1 tj sj (4.1)
!n
;n Q
n ( + )
I0+; ; f (s) = sj
( ) j=1

Z Z + 1
s1 sn Q
n tj f ((t1 ; ::; tn )
:: 1 dt1 ::dtn ; sj 2 [0; T ] ; > 0: (4.2)
0 0 j=1 sj tj
1
We shall denote by In2 (f ) the multiple Wiener-It^o integral with respect to W k
and we introduce the space
( ! )
n s;f n Q
n
n
j jk;T = f : [0; T ] ! R : ITk;n;2; 1 k ukj 2
jf j 2 L ([0; T ] ) :
2 j=1

11
De nition 4.2. If f 2 j n jsf k;T ; f symmetric, then we de ne the multiple sub-
fractionar integral of f with respect to S k by
p !n !!
k
1
k;n Q
n
k
In (f ) = ck k In2 IT ;2; 1 k uj f ; (4.3)
2 2 j=1

We have the equalities

p !2n ! ! 2
2 Q
n
kf k2j n jsf := Ink (f ) 2 = ck ITk;n;2; 1 k ukj f
k;T L ( ;F ;P ) 2k 2 j=1 L2 ([0;T ]n )

Z
Q
n
= 2n
f (u1 ; :::; un )f (v1 ; :::; vn ) 'k (uj ; vj )duj dvj : (4.4)
[0;T ] j=1

We can now de ne the space

( 1
)
X
n sf
L2k 2
= F 2 L ( ; F; P ) : F = Ink (fn ); fn 2 j jk ; fn symmetric :
n=0

Remark 4.3. Since j n jsf 2 2

k;T is not complete, jLk j is a strict subspace of L ( ; F; P ) :
Like in the fBm case (Bender, Bender-Elliot) the multiplication by an indicator
function can increase the norm in sf k;T :
Following the ideas of Hu-Oksendal we introduce the following
De nition 4.4. For F 2 jL2k j and t 2 [0; T ] ; we de ne the sub-fractional quasi-
k
conditional expectation of F with respect to FtS by
h k
i 1
X
E~ F j FtS = Ink (1(0;t)n fn ); (4.5)
n=0

provided the series converges in L2 ( ; F; P ) ;i.e.,

1
X 2
n! 1(0;t)n fn n jsf
< 1: (4.6)
j k;T
n=1

Remark 4.5. Te see the di erence between conditional expectation and quasi-
conditional expectation note that for t 2 (0; T ) ;
h k
i
E~ STk j FtS = Stk ;

12
 h i Z t
k
E STk j FtS = Stk + k
t;T (u)dSu :
0
The sub-fractional quasi-conditional expectation need not exist.
P1 k
De nition 4.6. The random variable F = n=1 In (fn ) 2 jL2k j is sub-fractional
Malliavin di erentiable if
1
X
Dtk F = nInk 1 (fn (:; t)) ; (4.7)
n=1

converges in jL2k j for a.e. t 2 [0; T ] ; i.e.,

1
X
nn! kfn (:; t)k2j n 1 jsf < 1: (4.8)
k;T
n=1

The sub-fractional Clark-Ocone derivative at time t of F 2 jL2k j is de ned by

h
k
i
rkt F = E~ Dtk F j FtS ; (4.9)

provided F is sub-fractional Malliavin di erentiable and the quasi-conditional

expectation exists, i.e.,
1
X 2
nn! 1(0;t)n 1 fn (:; t) n 1 jsf
< 1: (4.10)
j k;T
n=1

Remark 4.7. The sub-fractional

sf
Clark-Ocone derivative need not exist.
Consider the set S ch of all measurable processes (ut )t2[0;T ] such that:
k;T
(a) ut 2 jL2k j for a.e. t 2 [0; T ] and
1
X
ut = Ink (fn (:; t)); (4.11)
n=0

sf
with fn (:; t) 2 j n jsf
k;T ; fn 2 j
n+1
jk;T :
(b) The following series is convergent
1
X
(n + 1)! kjsym (fn )jk2j n+1 jsf < 1:
k;T
n=1

13
 sf
De nition 4.8. For a process u 2 S ch de ne the chaos sub-fractional Skoro-
k;T
k
hod integral o of u with respect to S by
1
X
ch k
k;T (u) := In+1 (sym (fn )) :
n=0

From (b) it follows that the previous converges in jL2k j :

Remark 4.9. Note that the chaos sub-fractional Skorohod integral has zero ex-
pectation (in contrast with the pathwise integral).
P1
De ne Dk1;2 as the family of all F = k
n=1 In (fn ) 2 jL2k j such that
1
X
nn! kjfn jk2j n jsf < 1: (4.12)
k;T
n=1

Theorem 4.10 (Clark-Ocone representation formula). If F 2 Dk1;2 then the

Clark-Ocone derivative exists and satis es
1
X
rkt F = nInk 1 1(0;t)n 1 fn (:; t) ; (4.13)
n=1

Z
E rks F rkt F 'k (s; t)dsdt < 1: (4.14)
[0;T ]2

Moreover, rk F is sub-fractional Skorohod integrable and the following Clark-

Ocone representation formula holds
ch
F = E(F ) + k;T rk F : (4.15)

Remark 4.11. A representation of a random variable F 2 jL2k j in the form

ch
F = E(F ) + k;T (u) ; (4.16)

need not be unique.

For example: By using the product formula it is easily see that qwe have the
relations 2
STk = k;T
ch
STk + (2 22k )T 2k+1 ;

14
 2
STk = ch
k;T 2S:k + (2 22k )T 2k+1 :
Proposition 4.12. A representation of the form (4.16) with u adapted is unique.
Remark 4.13. In (4.15) the integrand is adapted.

Proposition 4.14. For 0 < < and F 2 L2 (R; N (0; 2

)) de ne
Z (x y) 2
1
GF; (x) = q F (y)e 2( 2 2)
: (4.17)
2 ( 2 2) R

If f 2 j jsf
k;T ; (a; b) [0; T ] and 1(a;b) f sf < kf k sf ; then for every F 2
k;T k;T

L2 R; N (0; kf k2 sf ) we have the equality

k;T

h i
E~ F I1k (f ) j Stk : a t b = GFkf k ;k1(a;b) f k
I1k 1(a;b) f :
sf sf
k;T k;T (4.18)

Moreover, if F 2 C 1 (R); F 0 2 L2 R; N (0; kf k2 sf ) ; then the following relation

k;T
holds
0 1
h i 0
F I1k (f ) = E I1k (f ) + ch
k;T
@G F
kf k ;k1(a;b) f k
I1k 1(0;t) f fA :
sf sf
k;T k;T (4.19)

5. Sub-fractional Black-Scholes model

Next we consider the sub-fractional Black-Scholes model.
In this model the bank account has the dynamics

dBt = rBt dt; 0 t T; B0 = 1; (5.1)

so that Bt = exp(rt) and the price of the risky asset has sub-fractional log normal
dynamics

dSt = St dt + St dStk ; 0 t T; S0 = s0 > 0; (5.2)

where is the mean rate of return and > 0 is the volatility.

15
If we interpret the stochastic integral in (5.2) as Riemann-Stieltjes, then by the
change of variable formula (2.5) the solution of (5.2) is

St = s0 exp t + Stk : (5.3)

n o k
A couple = (ut )t2[0;T ] ; (vt )t2[0;T ] of FtS adapted processes is called portfolio.
The wealth or the value of the portfolio is the process

Vt = ut exp (rt) + vt St ; 0 t T: (5.4)

We say that the portfolio is self- nancing if

Z t Z t
Vt = V0 + r us exp (rs) ds + vs dSs ; 0 t T: (5.5)
0 0

An arbitrage is an self- nancing portfolio such that V0 = 0; VT 0 and

P (VT > 0) > 0:
As soon as we have the dynamics of the risky asset given by Riemann-Stieltjes
integral, always this leads to the existence of arbitrage opportunities.
As example we recall the Shiryaev construction or an arbitrage for = r; = 1 :
= (u; v) ; h i
ui = 1 exp(2Stk ); vt = 2 exp(Stk ) 1 :
From the change of variables formula (2.5) it follows that is self- nancing and
moreover is arbitrage, since
h i2
VT = exp(STk ) 1 exp(rT ) > 0:

An alternative which works in the case of pathwise cost is to restrict the class of
admissible portfolios, but to remain stil big enough to cover hedges for relevant
options and also to consider mixed cost models.
In this respect we consider the mixed model with the stock price given by
1 2
Stk;a = s0 exp Stk + aWt + t (1 22k 1 ) 2 t2k+1 at ; (5.6)
2

where ; s0 > 0; a 2 R; W is a Bm and the mixed process Stk + aWt is assumed

t
to be Gaussian (this heapens, for example, if S k and W are independent).

16
Remark 5.1. The mixed process Stk + aWt is a semimartingale equivalent
t
with (aWt )t if k 2 14 ; 12 and is not a semimartingale if 0 < k 14 :
The following class of restringed class of self- nancing portfolios is considered: a
self- nancing portfolio = (u; v) is nds-admissible (no-doubling strategy) if there
exists a 0 such that Vt a for all 0 t T P-a.s.
A nds-admissible portfolio = (u; v) is regular if there exists a di erentiable
4
function ' : [0; T ] R+ ! R such that
Z t
vt = ' t; Stk;a ; max Ssk;a ; min Ssk;a ; Ssk;a ds :
0 s t 0 s t 0

As a consequence of a more general result due to Bender-Sottined-Valkeila (2006)

the following result holds:
Theorem 5.2. The mixed model is arbitrage free in the class of regular portfolios.
Moreover, European, Assian, lookback options can be hedged with regular portfo-
lios, with the same functionals and hedging prices as in tht classical Black-Scholes
model.

A di erent alternative to the pathwise approach of the stock price is to consider

the chaos form of the stochastic
Rt
integral in (5.2).
ch
We shall use the notation 0 f (s) k;T BsH for the sub-fractional Skorohod integral
ch
k;T 1(0;t) f :
Therefore the price of the risky asset has the dynamics
ch k
dSt = St dt + St k;T St ; 0 t T; S0 = s0 > 0: (5.7)
In the present situation the portofolio is self- nancing if v: S 2 L1 ([0; T ]) ; 1(0;t) v: S 2
sf
S ch for a.a. t; and
k;T
Z t Z t
ch k
Vt = V0 + (rus exp (rs) + vs Ss ) ds + v s Ss k;T Ss ; 0 t T:
0 0
(5.8)
A self- nancing portofolio is admissible if Vt is bounded below for all 0 t T:
In order to nd the explicit form of the St in (5.7) we consider the following
sub-fractional a ne equation
Z t Z t
ch H
Xt = + [a0 (s) + a(s)Xs ] ds + [b0 (s) + b(s)Xs ] k;T Bs ; t 2 [0; T ] ;
0 0
(5.9)

17
 2 L2 ( ; F; P ) and a0 ; b0 ; a; b : [0; T ] ! R are measurable and bounded func-
tions.

De nition 5.3. A process (Xt )t2[0;T ] is a strong solution of (5.9) if

(i) X: 2 L1 ([0; T ]) and 1[0;t] (:)b(:)X: is sub-fractional Skorohod integrable for a.a.
t 2 [0; T ] :
(ii) For a.a. t 2 [0; T ] ; the equation (5.9) is satis ed P a.s.:
We introduce the following notation: Given t1 ; :::; tp and a symmetric function of
p j variables (j < p), we denote by fp j (t^i1 ; ::; t^ij ) the function fp j evaluated
in t's other that ti1 ; ::; tij :
Proposition 5.4. Assume that
1
X 1
X
= IpH;T ( p ); p! k p k2j n jsf < 1: (5.10)
k;T
p=0 p=1

De ne
Z t
U (t; s) = exp a(r)dr
(Zs Z )
t t 1 2
(t; s) = exp a(u)du + b(u)dBuH b1[s;t] sf :
s s 2 k;T

Then (5.9) has a unique strong solution X with the chaos decomposition
1
X
Xt = Ipk (fpt ); (5.11)
p=0

Z t
f0t = U (t; t0 ) 0+ U (t; s)a0 (s)ds; (5.12)
t0

f1t (t1 ) = U (t; t0 ) 1 (t1 ) + 1[t0 ;t] (t1 )U (t; t1 )b0 (t1 )+
Z t1
1[t0 ;t] (t1 )U (t; t0 )b(t1 ) 0 + 1[t0 ;t] (t1 )b(t1 ) U (t; s)a0 (s)ds; (5.13)
t0

and for p 2;
p
1X t
fpt (t1 ; :::; tp ) = U (t; t0 ) p (t1 ; :::; tp ) + 1[t0 ;t] (tj )U (t; tj )b(tj )fpj 1 (t^j );
p j=1
(5.14)

18
or in an explicit form

fpt (t1 ; :::; tp ) = U (t; t0 ) p (t1 ; :::; tp )+

p
X n o
1 ^ ^
U (t; t0 ) sym 1[t0 ;t]j (t1 ; :::; tj )b(t1 ):::b(tj ) p j (t1 ; ::; tj +
j=1 j!
n o
sym 1t0 <t1 <:::<tp <t U (t; t1 )b0 (t1 )b(t2 ):::b(tp ) +

Z t1 ^:::^tp
1
1[t ;t] (t1 ; :::; tp )b(t1 ):::b(tp )
p U (t; s)a0 (s)ds: (5.15)
p! 0 t0

In particular if = 0 2 R the solution is

Z t Z t
Xt = (t; t0 ) 0+ (t; s)a0 (s)ds + (t; s)b0 (s)dBsH (5.16)
t0 t0

Corollary 5.5. The stock price St in (5.7) is given by

n o
St = s0 exp t 1 22k 1 2 2k+1
t + Stk : (5.17)

Remark 5.6. The standard Black-Scholes model is markovian and the log-returns
Rt;t+s = log SSt+s
s
are stationary independent Gaussian random variables.
The fractional Black-Scholes model is nonmarkovian and the log-returns are sta-
tionary non-independent Gaussian random variables.
The sub-fractional Black-Scholes model di ers from fractional Black-Scholes model
by the non stationarity of the log-returns
h i
Rt;t+s = s 1 22k 1 2
(t + s)2k+1 t2k+1 + k
St+s Stk :

De nition 5.7. A probability measure Q on FTS which is equivalent with P

(Q P ) is called a quasi-martingale measure (or average risk neutral measure) if:
(i) There exists a Gaussian process (Zt )0 t T with respect to Q such that
1
St Bt = exp(Zt ); 0 t T: (5.18)

(ii) For every 0 t T;

EQ (St Bt 1 ) = s0 : (5.19)

19
Remark 5.8. It is clear that if Q is a quasi-martingale measure then Q is uniquely
k
determined on FTS = FTS :
De ne the probability measure Q by
( )
dQ r r 2 d2k
j Sk = exp Mtk t 1 2k
: (5.20)
dP Ft 2(1 2k)

Remark 5.9. (a) By Girsanov's theorem (Theorem 3.6) the process

r
Ztk := Stk + t
t

is a sfBm under Q.
(b) The relation (5.17) becomes in terms of Z k
n o
St = exp Ztk (1 22k 1 )t2k+1 ; (5.21)

and it is clear that St has under Q the dynamics

ch k
dSt = St k;T Zt ; 0 t T; S0 = s0 : (5.22)

The main result is the following

Theorem 5.10. (i) The sub-fractional Black-Scholes market is arbitrage free
and for every bounded contigent claim F 2 Dk1;2 (Q) there exist v0 2 R and an
admissible portofolio such that

V0 = v0 ; VT = F P a.s.

(ii) The following relation holds:

h k
i
EQ ST BT 1 j FtS = St Bt 1 exp(K(T; t)); 80 t T; (5.23)

where Z t
K(T; t) = exp dk (r ) ITk ;2; k+1
1(0;t) t;T (s) s k ds
0 2

Z t Z t
2 2 2 k 2
ck n (T; s)ds + t;T (s)dSs + (1 22k 1 )t2k+1 :
0 0
(5.24)

In particular Q is the unique quasi-martingale measure and Q is not a martingale

measure.

20
Remark 5.11. The price CT (F ) of the contigent claim F is given by

CT (F ) = EQ BT 1 F : (5.25)

The corresponding replicating portfolio = (u; v) also can be described.

The price of the contigent clain f (ST ) is given by the formula
exp( rT )
CT (f (ST )) = p
2

Z n o
1 y2
f s0 exp T k+ 2 y + rT 2
1 22k 1
T 2k+1 exp( )dy
R 2
(5.26)

In particular the price of of an European call is

CT ((ST K)+ ) = s0 (y1 ) K exp( rT ) (y2 ); (5.27)

log sK0 + rT + 2
1 22k 1
T 2k+1
y1 = p 1 ;
2 22k T k+ 2

log sK0 + rT 2
1 22k 1
T 2k+1
y1 = p 1 :
2 22k T k+ 2
Comments. (a) In the mixed sub-fractional model the class of regular portfolios
is a abitrage-free class that is su ciently large to cover hedges for most known
relevant options.
A recent result by Bender, Sottinen and Valkeila (2006) extends no-arbitrage
property and robust hedges to a class of non-semimartingale models larger than
the mixed processes and a larger class of portfolios.
It should be noted that the quadratic variation is the main property which is
necessary for pricing in non-semimartingale models.
(b) The use of chaos form of the Skorohod integral in the sub-fractional Black-
Scholes model does not have a nice economic interpretation (Bjork-Hult, 2005
for fBm case) and also this is problematic from the mathematical point of view

21
(Nualart-Taqqu, 2006). It happens that di erent Gaussian processes with the
same variation as S k give the same price for European call options.
Therefore in above mentioned both cases it is not the distribution of the process
which determines uniquely the prices, but the variation of the process.
(c) An alternative to rescue the sub-fractional Black-Scholes model is to use so
called market observers according to an idea by ksendal (Bender, 2003)

References
[1] T. ANDROSHCHUK AND Yu. MISHURA, Mixed Brownian-fractional
Brownian model: absence of arbitrage and related topics. Intern. J. Prob.
Stoch. Proc. 78 (2006), 281-300.

[2] E. ALOS, O. MAZET AND D. NUALART, Stochastic calculus with respect

to Gaussian processes. Annals of Probability 29(2) (2001), 766-801.tic di er-
ential equations and Wick products. Probab. Theory Relat. Fields 99 (1994),
501-526.

[3] C. BENDER, Integration with respect to a fractional Brownian motion and

related market models. Konstanzer Dissertationen 597. Hartung-Gorre Verlag
Konstanz, 2003.

[4] C. BENDER AND R. ELLIOT, A note on the Clark-Ocone theorem for frac-
tional Brownian motions with Hurst parameter bigger than a half. Preprint
2002.

[5] C. BENDER, T. SOTTINEN AND E. VALKEILA, Arbitrage with fractional

Brownian motion? Theory of Stochastic Processes 12(28), no. 3-4 (2006),
1-12.

[6] C. BENDER, T. SOTTINEN AND E. VALKEILA, No-arbitrage pricing be-

yond semimartingales. Preprint Nr. 110. Weierstra -Institur f•
ur Angewandte
Analysis und Stochastik, 2006.

[7] F. BIAGINI AND B. KSENDAL, Forward integrals and an It^

o formula for
fractional Brownian motion. Preprint, 2007.
•
[8] T. BJORK AND H. HULT, A note on Wick products and the fractional
Black-Scholes model. Finance ans Stochastics 9 (2005), 197-208.

22
 [9] T. Bojdecki, L. Gorostiza and A. Talarczyk, Sub-fractional Brownian motion
and its relation to occupation times. Stat. & Probab. Lett. 69 (2004), 405-419.

[10] A. DASGUPTA AND G. KALLIANPUR, Multiple fractional integrals. Prob.

Theory Rel. Fields 115 (1999), 505-525.

[11] |||, Chaos decomposition of multiple fractional integrals and applications.

Prob.Theory Rel. Fields 115 (1999), 527-548.

[12] L. DECREUSEFOND AND A.S. USTUNEL, Stochastic Analysis with respect

to fractional Brownian motion. Potential Analysis 10(2) (1999), 177-214.

[13] R.M. DUDLEY, R. NORVAISA, Di erentiability of Six Operators on Non-

smooth Functions and p-variation. Springer, Berlin, 1999.

[14] R.M. DUDLEY, R. NORVAISA, An introduction to p-variation and Young

integrals. MathPhySto. Lecture Notes 1 (1998).

[15] T. DUNCAN, Y. HU AND B. PASIK-DUNCAN, Stochastic calculus for frac-

tional Brownian motion I.Theory. SIAM J. Control Optim. 38(2) (2000),
582-612.

[16] K. DZHAPARIDZE AND H. VAN ZANTEN, A series expansion of fractional

Brownian motion. Probab. Theory Relat. Fields 103 (2004), 39-55.

[17] R.J. ELLIOT AND J. VAN DER NOEK, A general fractional white noise
theory and applications to nance. Mathematical Finance 11, 2 (2003), 301-
330.

[18] D. Feyel and A. De La Pradelle, On fractional Brownian processes. Potential

Analysis 10 (1999), 273-288.

[19] G. GRIPENBERG AND I. NORROS, On the prediction of fractional Brown-

ian motion. J.Appl. Prob. 33 (1996), 400-410.

[20] Y. HU AND B. OKSENDAL, Fractional white noise calculus and applica-

tions to nance. In: In nite Dimensional Analysis, QuantumProbability and
related Topics, Vol. 6 (2003), 1-32.

[21] S.T. HUANG AND S. CAMBANIS, Stochastic and multiple Wiener integrals
for Gaussian processes. Ann. Probability 6 (1978), 585-614.

23
[22] K. ITO, Multiple Wiener integral. J. Math. Soc. Japan 3 (1951), 157-169.

[23] M. JOLIS, On the Wiener integral with respect to the fractional Brownian
motion on an interval. J. Math. Anal. Appl. 330:22 (2007), 1115-1127.

[24] V. KONDURAR. Sur l'integrale de Stieltjes. Mat. Sbornik 2 (1937), 361-366.

[25] T. MIKOSCH, R. NORVAISA, Stochastic integral equations without proba-

bility. Bernoulli 6(3) (2000), 401-434.

[26] Yu. MISHURA AND E. VALKEILA, On arbitrage and replication in the

fractional Black-Scholes pricing model. Proc. of the Steklov Inst. of Math.
237 (2002), 215-224.

[27] I. NORROS, E. VALKEILA AND J. VIRTAMO, An elementary approach

to a Girsanov formula and other analytical results on fractional Brownian
motions. Bernoulli 5(4) (1999), 571-587.

[28] D. NUALART, The Malliavin Calculus and Related Topics. Springer-Verlag,

1995.

[29] D. NUALART, Fractional Brownian motion: stochastic calculus and applica-

tons. Proceedings of the International Congress of Mathematicians, Madrid,
Spain, 2006.

[30] D. NUALART AND M. TAQQU, Some issues concerning Wick integrals and
the Balck and Scholes formula. Preprint 2006.

[31] V. PIPIRAS AND M.S. TAQQU, Integration questions related to fractional

Brownian motion. Probability Theory and Related Fields 118 (2) (2000),
251-291.

[32] ||, Are classes of deterministic integrands for fractional Brownian motion
on an interval complete? Bernoulli 7 (6) (2001), 873-897.

[33] S.G. SAMKO, A.A. KILBAS AND O.I. MARICHEV, Fractional Integrals
and Derivatives. Gordon and Breach Science, 1993.

[34] A.N. SHYRYAEV, On arbitrage and replication for fractal models. Research
Reports 1998-20. MatPhySto, 1998.

24
[35] T. SOTTINENAND E. VALKEILA, Fractional Brownian motion as a model
in nance. Preprint, 2001.

[36] T. SOTTINENAND E. VALKEILA, On arbitrage and replication in the frac-

tional Black-Scholes pricing model. Statistics&Decisions 21 (2003), 137-151.

[37] C. Stricker, Quelques remarques sur les semimartingales gaissienes et le

probleme de l'innovation. Lecture Notes in Control and Information Sciences
61 (1984), 260-276.

[38] C. TUDOR, Some properties of the sub-fractional Brownian motion. Stochas-

tics 79,5 (2007), 431-448.

[39] C. TUDOR, Inner product spaces of integrands associated to sub-fractional

Brownian motion. Stat. Probab. Letters. doi:10.1016/j.spl.2008.01.087.

[40] C. TUDOR, On the Wiener integral with respect to a fractional Brownian

motion on an interval. Preprint 2007.

[41] L.C. Young, An inequality of the H•

older type connected with Stieltjes integra-
tion. Acta Math. 67 (1936), 251-282.
•
[42] M. ZAHLE, Integration with respect to fractal functions and stochastic cal-
culus I. Probab. Theory Related Fields 111 (1998), 333-374.
•
[43] M. ZAHLE, Long range dependence, no arbitrage and the Black-Scholes for-
mula. Stoch. Dyn. 2 (2002), 265-280.

25