Documente Academic
Documente Profesional
Documente Cultură
Calculus
Alan Bain
1. Introduction
The following notes aim to provide a very informal introduction to Stochastic Calculus,
and especially to the It
o integral and some of its applications. They owe a great deal to Dan
Crisans Stochastic Calculus and Applications lectures of 1998; and also much to various
books especially those of L. C. G. Rogers and D. Williams, and Dellacherie and Meyers
multi volume series Probabilities et Potentiel. They have also benefited from insights
gained by attending lectures given by T. Kurtz.
The present notes grew out of a set of typed notes which I produced when revising
for the Cambridge, Part III course; combining the printed notes and my own handwritten
notes into a consistent text. Ive subsequently expanded them inserting some extra proofs
from a great variety of sources. The notes principally concentrate on the parts of the course
which I found hard; thus there is often little or no comment on more standard matters; as
a secondary goal they aim to present the results in a form which can be readily extended
Due to their evolution, they have taken a very informal style; in some ways I hope this
may make them easier to read.
The addition of coverage of discontinuous processes was motivated by my interest in
the subject, and much insight gained from reading the excellent book of J. Jacod and
A. N. Shiryaev.
The goal of the notes in their current form is to present a fairly clear approach to
the It
o integral with respect to continuous semimartingales but without any attempt at
maximal detail. The various alternative approaches to this subject which can be found
in books tend to divide into those presenting the integral directed entirely at Brownian
Motion, and those who wish to prove results in complete generality for a semimartingale.
Here at all points clarity has hopefully been the main goal here, rather than completeness;
although secretly the approach aims to be readily extended to the discontinuous theory.
I make no apology for proofs which spell out every minute detail, since on a first look at
the subject the purpose of some of the steps in a proof often seems elusive. Id especially
like to convince the reader that the It
o integral isnt that much harder in concept than
the Lebesgue Integral with which we are all familiar. The motivating principle is to try
and explain every detail, no matter how trivial it may seem once the subject has been
understood!
Passages enclosed in boxes are intended to be viewed as digressions from the main
text; usually describing an alternative approach, or giving an informal description of what
is going on feel free to skip these sections if you find them unhelpful.
In revising these notes I have resisted the temptation to alter the original structure
of the development of the It
o integral (although I have corrected unintentional mistakes),
since I suspect the more concise proofs which I would favour today would not be helpful
on a first approach to the subject.
These notes contain errors with probability one. I always welcome people telling me
about the errors because then I can fix them! I can be readily contacted by email as
alanb@chiark.greenend.org.uk. Also suggestions for improvements or other additions
are welcome.
Alan Bain
[i]
2. Contents
1. Introduction . . . . . . . . . . . . . . . . . .
2. Contents . . . . . . . . . . . . . . . . . . . .
3. Stochastic Processes . . . . . . . . . . . . . .
3.1. Probability Space . . . . . . . . . . . . . .
3.2. Stochastic Process . . . . . . . . . . . . .
4. Martingales . . . . . . . . . . . . . . . . . .
4.1. Stopping Times
. . . . . . . . . . . . . .
5. Basics . . . . . . . . . . . . . . . . . . . . .
5.1. Local Martingales
. . . . . . . . . . . . .
5.2. Local Martingales which are not Martingales . .
6. Total Variation and the Stieltjes Integral . . . .
6.1. Why we need a Stochastic Integral . . . . . .
6.2. Previsibility . . . . . . . . . . . . . . . .
6.3. Lebesgue-Stieltjes Integral . . . . . . . . . .
7. The Integral . . . . . . . . . . . . . . . . . .
7.1. Elementary Processes . . . . . . . . . . . .
7.2. Strictly Simple and Simple Processes . . . . .
8. The Stochastic Integral . . . . . . . . . . . .
8.1. Integral for H L and M M2 . . . . . . .
8.2. Quadratic Variation
. . . . . . . . . . . .
8.3. Covariation . . . . . . . . . . . . . . . .
8.4. Extension of the Integral to L2 (M ) . . . . . .
8.5. Localisation . . . . . . . . . . . . . . . .
8.6. Some Important Results . . . . . . . . . . .
9. Semimartingales . . . . . . . . . . . . . . . .
10. Relations to Sums . . . . . . . . . . . . . .
10.1. The UCP topology
. . . . . . . . . . . .
10.2. Approximation via Riemann Sums . . . . . .
11. It
os Formula . . . . . . . . . . . . . . . . .
11.1. Applications of It
os Formula . . . . . . . .
11.2. Exponential Martingales . . . . . . . . . .
12. L
evy Characterisation of Brownian Motion . .
13. Time Change of Brownian Motion . . . . . .
13.1. Gaussian Martingales
. . . . . . . . . . .
14. Girsanovs Theorem . . . . . . . . . . . . .
14.1. Change of measure
. . . . . . . . . . . .
15. Brownian Martingale Representation Theorem
16. Stochastic Differential Equations . . . . . . .
17. Relations to Second Order PDEs . . . . . . .
17.1. Infinitesimal Generator . . . . . . . . . . .
17.2. The Dirichlet Problem . . . . . . . . . . .
[ii]
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
i
ii
1
1
1
4
4
8
8
9
11
11
12
13
15
15
15
17
17
19
22
23
26
27
29
31
31
32
35
40
41
46
48
49
51
51
53
56
61
61
62
Contents
18.
19.
20.
21.
22.
iii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
64
66
69
69
70
70
70
76
78
86
87
89
89
90
92
92
93
95
3. Stochastic Processes
The following notes are a summary of important definitions and results from the theory of
stochastic processes, proofs may be found in the usual books for example [Durrett, 1996].
F =
Ft :=
t0
Ft .
t0
If (Xt )t0 is a stochastic process, then the natural filtration of (Xt )t0 is given by
FtX := (Xs : s t).
The process (Xt )t0 is said to be (Ft )t0 adapted, if Xt is Ft measurable for each t 0.
The process (Xt )t0 is obviously adapted with respect to the natural filtration.
[1]
Stochastic Processes
{Xs > 1}
0ss
:=
n
2X
1
k=0
n
2
X
k=1
Note that X1n is a left continuous process, so if X is left continuous, working pointwise
(that is, fix ), the sequence X1n converges to X.
But the individual summands in the definition of X1n are by the adpatedness of X
clearly B[0,s] Fs measurable, hence X1n is also. But the convergence implies X is also;
hence X is progressively measurable.
Consideration of the sequence X2n yields the same result for right continuous, adapted
processes.
Stochastic Processes
Define
t (0, ) Ft =
Fs
0s<t
t [0, ) Ft+ =
Fs ,
ts<
Definition 3.3.
A process (Xt )t0 is said to be bounded if there exists a universal constant K such that
for all and t 0, then |Xt ()| < K.
Definition 3.4.
Let X = (Xt )t0 be a stochastic process defined on (, F, P), and let X 0 = (Xt0 )t0 be a
stochastic process defined on (, F, P). Then X and X 0 have the same finite dimensional
distributions if for all n, 0 t1 < t2 < < tn < , and A1 , A2 , . . . , An E,
P(Xt1 A1 , Xt2 A2 , . . . , Xtn An ) = P0 (Xt01 A1 , Xt02 A2 , . . . , Xt0n An ).
Definition 3.5.
Let X and X 0 be defined on (, F, P). Then X and X 0 are modifications of each other if
and only if
P ({ : Xt () = Xt0 ()}) = 1
t 0.
Definition 3.6.
Let X and X 0 be defined on (, F, P). Then X and X 0 are indistinguishable if and only if
P ({ : Xt () = Xt0 ()t 0}) = 1.
There is a chain of implications
indistinguishable modifications same f.d.d.
The following definition provides us with a special name for a process which is indistinguishable from the zero process. It will turn out to be important because many definitions
can only be made up to evanescence.
Definition 3.7.
A process X is evanescent if P(Xt = 0 t) = 1.
4. Martingales
Definition 4.1.
Let X = {Xt , Ft , t 0} be an integrable process then X is a
(i) Martingale if and only if E(Xt |Fs ) = Xs a.s. for 0 s t <
(ii) Supermartingale if and only if E(Xt |Fs ) Xs a.s. for 0 s t <
(iii) Submartingale if and only if E(Xt |Fs ) Xs a.s. for 0 s t <
Theorem (Kolmogorov) 4.2.
V
Let X = {Xt , Ft , t 0} be an integrable process. Then define Ft+ := >0 Ft+ and also
the partial augmentation of F by Ft = (Ft+ , N ). Then if t 7 E(Xt ) is continuous there
= {X
t , Ft , t 0} with sample paths which are
exists an Ft adapted stochastic process X
are modifications of each
right continuous, with left limits (CADLAG) such that X and X
other.
Definition 4.3.
A martingale X = {Xt , Ft , t 0} is said to be an L2 -martingale or a square integrable
martingale if E(Xt2 ) < for every t 0.
Definition 4.4.
A process X = {Xt , Ft , t 0} is said to be Lp bounded if and only if supt0 E(|Xt |p ) < .
The space of L2 bounded martingales is denoted by M2 , and the subspace of continuous
L2 bounded martingales is denoted Mc2 .
Definition 4.5.
A process X = {Xt , Ft , t 0} is said to be uniformly integrable if and only if
sup E |Xt |1|Xt |N 0 as N .
t0
for t s.
Proof
Via Cauchy Schwartz inequality E|Y (Mt Ms )| < , and so
E(Y (Mt Ms )) = E (E(Y (Mt Ms )|Fs )) = E (Y E(Mt Ms |Fs )) = 0.
[4]
Martingales
[
1
T t
Ft .
{T < t} =
n
n=1
To prove the converse, note that if for each t [0, ) we have that {T < t} Ft ,
then for each such t
1
T <t+
Ft+1/n ,
n
as a consequence of which
^
\
1
{T t} =
T <t+
Ft+1/n = Ft+ .
n
n=1
n=1
Martingales
The condition which is most often forgotten is that in (iii) that the stopping time T
be bounded. To see why it is necessary consider Bt a Brownian Motion starting from zero.
Let T = inf{t 0 : Xt = 1}, clearly a stopping time. Equally Bt is a martingale with
respect to the filtration generated by B itself, but it is also clear that EBT = 1 6= EB0 = 0.
Obviously in this case T < is false.
Theorem (Doobs Martingale Inequalities).
Let M = {Mt , Ft , t 0} be a uniformly integrable martingale, and let M := supt0 |Mt |.
Then
(i) Maximal Inequality. For > 0,
P(M ) E [|M |1M < ] .
(ii) Lp maximal inequality. For 1 < p < ,
kM kp
p
kM kp .
p1
Note that the norm used in stating the Doob Lp inequality is defined by
1/p
kM kp = [E(|M |p )]
Let Mc2 denote the set of L2 -bounded CADLAG martingales which are continuous. A norm
2
may be defined on the space M2 by kM k2 = kM k22 = E(M
).
2
From the conditional Jensens inequality, since f (x) = x is convex,
2
2
E M
|Ft (E(M |Ft ))
2
E M
|Ft (EMt )2 .
Hence taking expectations
2
EMt2 EM
,
2
and since by martingale convergence in L2 , we get E(Mt2 ) E(M
), it is clear that
2
E(M
) = sup E(Mt2 ).
t0
Martingales
Theorem 4.8.
The space (M2 , kk) (up to equivalence classes defined by modifications) is a Hilbert space,
with Mc2 a closed subspace.
Proof
We prove this by showing a one to one correspondence between M2 (the space of square
integrable martingales) and L2 (F ). The bijection is obtained via
f :M2 L2 (F )
f :(Mt )t0 7 M lim Mt
t
g :L (F ) M2
g :M 7 Mt E(M |Ft )
Notice that
2
sup EMt2 = kM k22 = E(M
) < ,
t
that is M (n) M uniformly in L2 . Hence there exists a subsequence n(k) such that
M n(k) M uniformly; as a uniform limit of continuous functions is continuous, M Mc2 .
Thus Mc2 is a closed subspace of M.
5. Basics
5.1. Local Martingales
A martingale has already been defined, but a weaker definition will prove useful for stochastic calculus. Note that Ill often drop references to the filtration Ft , but this nevertheless
forms an essential part of the (local) martingale.
Just before we dive in and define a Local Martingale, maybe we should pause and
consider the reason for considering them. The important property of local martingales
will only be seen later in the notes; and as we frequently see in this subject it is one of
stability that is, they are a class of objects which are closed under an operation, in this case
under the stochastic integral an integral of a previsible process with a local martingale
integrator is a local martingale.
Definition 5.1.
M = {Mt , Ft , 0 t } is a local martingale if and only if there exists a sequence of
stopping times Tn tending to infinity such that M Tn are martingales for all n. The space
of local martingales is denotes Mloc , and the subspace of continuous local martingales is
denotes Mcloc .
Recall that a martingale (Xt )t0 is said to be bounded if there exists a universal
constant K such that for all and t 0, then |Xt ()| < K.
Theorem 5.2.
Every bounded local martingale is a martingale.
Proof
Let Tn be a sequence of stopping times as in the definition of a local martingale. This
sequence tends to infinity, so pointwise XtTn () Xt (). Using the conditional form of the
dominated convergence theorem (using the constant bound as the dominating function),
for t s 0
lim E(XtTn |Fs ) = E(Xt |Fs ).
n
Basics
(ii) (i)It is required to prove that E(M0 ) = E(MT ) for any bounded stopping time T .
Then by local martingale property for any n,
E(M0 ) = E(MT Tn ),
uniform integrability then implies that
lim E(MT Tn ) = E(MT ).
for some constant C, which on tranformation into polar co-ordinates yields a bound of the
form
Z
1
C0
with C 0 another constant. This is finite if and only if (d 2)p + (d 1) > 1 (standard
integrals of the form 1/rk ). This in turn requires that p < d/(d 2). So clealry Ex |Xt | will
be finite for all d 3.
Now although Ex |Xt | < and Xt is a local martingale, we
shall show that it is not
a martingale. Note that (Bt x) has the same distribution as t(B1 x) under Px (the
Basics
probability measure induced by the BM starting from x). So as t , |Bt | in
probability and Xt 0 in probability. As Xt 0, we see that Ex (Xt ) = Ex |Xt | < .
Now note that for any R < , we can construct a bound
Z
1
Ex Xt
|y|(d2) dy + R(d2) ,
d/2
(2t)
|y|R
which converges, and hence
lim sup Ex Xt R(d2) .
t
10
n()
X
At t At t : 0 = t0 t1 tn = t .
Vt := sup
k
k1
k=1
Vt0
:= lim
X
Ak2n t A(k1)2n t .
1
These can be shown to be equivalent (for CADLAG processes), since trivially (use the
dyadic partition), Vt0 Vt . It is also possible to show that Vt0 Vt for the total variation
of a CADLAG process.
Definition 6.1.
A process A is said to have finite variation if the associated variation process V is finite
(i.e. if for every t and every , |Vt ()| < .
Xs ()dMs (),
0
for M Mc2 ; but for an interesting martingale (i.e. one which isnt zero a.s.), the total
variation is not finite, even on a bounded interval like [0, T ]. Thus the Lebesgue-Stieltjes
integral definition isnt valid in this case. To generalise we shall see that the quadratic
variation is actually the right variation to use (higher variations turn out to be zero and
lower ones infinite, which is easy to prove by considering the variation expressed as the
limit of a sum and factoring it by a maximum multiplies by the quadratic variation, the
first term of which tends to zero by continuity). But to start, we shall consider integrating
a previsible process Ht with an integrator which is an increasing finite variation process.
First we shall prove that a continuous local martingale of finite variation is zero.
[11]
N
X
E(Mt2 ) =E
Mtk Mtk1
2
k=1
E Vt sup Mtk Mtk1
The integrand is bounded by n2 (from definition of the stopping time Sn ), hence the
expectation converges to zero as the modulus of the partition tends to zero by the bounded
convergence theorem. Hence M 0.
6.2. Previsibility
The term previsible has crept into the discussion earlier. Now is the time for a proper
definition.
Definition 6.3.
The previsible (or predictable) -field P is the -field on R+ generated by the processes
(Xt )t0 , adapted to Ft , with left continuous paths on (0, ).
Remark
The same -field is generated by left continuous, right limits processes (i.e. c`
agl`
ad processes) which are adapted to Ft , or indeed continuous processes (Xt )t0 which are adapted
to Ft . It is gnerated by sets of the form A (s, t] where A Fs . It should be noted that
c`
adl`
ag processes generate the optional field which is usually different.
Theorem 6.4.
The previsible fieldis also generated by the collection of random sets A {0} where
A F0 and A (s, t] where A Fs .
Proof
Let the field generated by the above collection of sets be denotes P 0 . We shall show
P = P 0 . Let X be a left continuous process, define for n N
X
X n = X0 10 (t) +
Xk/2n 1(k/2n ,(k+1)/2n ] (t)
k
12
Hs ()dAs (),
0
HdX,
0
13
n=1
P
Consider P(A) = P(
n=1 An )
n=1 P(An ). Suppose for some n, P(An ) > , then note
that
An : Xs E(Xt |Fs ) > 1/n
/An : Xs E(Xt |Fs ) 0
Hence
Xs E(Xt |Fs )
1
1A ,
n n
,
n
but by the constant mean property the left hand side is zero; hence a contradiction, thus
all the P(An ) are zero, so
Xs = E(Xt |Fs ) a.s.
E(Xs ) E(Xt ) >
14
7. The Integral
We would like eventually to extend the definition of the integral to integrands which are
previsible processes and integrators which are semimartingales (to be defined later in these
notes). In fact in these notes well only get as far as continuous semimartingales; but it is
possible to go the whole way and define the integral of a previsible process with respect to
a general semimartingale; but some extra problems are thrown up on the way, in particular
as regards the construction of the quadratic variation process of a discontinuous process.
Various special classes of process will be needed in the sequel and these are all defined
here for convenience. Naturally with terms like elementary and simple occurring many
books have different names for the same concepts so beware!
Sn = S +
1
,
n
Tn = T +
1
.
n
So it is sufficient to show that Z1[U,V ) is adapted when U and V are stopping times which
satisfy U V , and Z is a bounded FU measurable function. Let B be a borel set of R,
then the event
{Z1[U,V ) (t) B} = [{Z B} {U t}] {V > t}.
By the definition of U as a stopping time and hence the definition of FU , the event enclosed
by square brackets is in Ft , and since V is a stopping time {V > t} = /{V t} is also
in Ft ; hence Z1[U,V ) is adapted.
n1
X
k=0
[15]
The Integral
This can be extended to H is a simple processes (H L), if there exists a sequence
of stopping times 0 T0 Tk , and Zk uniformly bounded FTk measurable
random variables such that
H = H0 ()10 (t) +
Zk 1(Tk ,Tk+1 ] .
k=0
Similarly a simple process is also a previsible process. The fundamental result will
follow from the fact that the -algebra generated by the simple processes is exactly the
previsible -algebra. We shall see the application of this after the next section.
16
Hs dMs = (H M )t :=
0
Zk MTk+1 t MTk t
k=0
Proposition 8.1.
If H is a simple process, M a L2 bounded martingale, and T a stopping time. Then
(i) (H M )T = (H1(0,T ] ) M = H (M T ).
(ii) (H M ) M2 . P
2
(iii) E[(H M )2 ] = k=0 E[Zk2 (MT2k+1 MT2k )] kHk2 E(M
).
Proof
Part (i)
As H L we can write
H=
Zk 1(Tk ,Tk+1 ] ,
k=0
for Tk stopping times, and Zk an FTk measurable bounded random variable. By our definition for M M2 , we have
(H M )t =
Zk MTk+1 t MTk t ,
k=0
Zk MTk+1 T t MTk T t .
k=0
Similar computations can be performed for (H M T ), noting that MtT = MT t and for
(H1(0,T ] M ) yielding the same result in both cases. Hence
(H M )T = (H1(0,T ] M ) = (H M T ).
[17]
18
19
E (H M )2 kH k2 E
MT2k+1 MT2k
k=0
2
kH k E
MT2K+1
MT20
2
kH k2 EM
,
since we require T0 = 0, and M M2 , so the final bound is obtained via the L2 martingale
convergence theorem.
Now extend this to the case of an infinite sum; let n m, we have that
(H M )Tm (H M )Tn = (H1(Tn ,Tm ] M ),
applying the result just proven for finite sums to the right hand side yields
m1
X
2
2
2
(H M )Tm (H M )Tn
2 =
E Zk MTk+1 MTk
2
k=n
2
kH k22 E M
MT2n .
But by the L2 martingale convergence theorem the right hand side of this bound tends to
zero as n ; hence (H M )Tn converges in M2 and the limit must be the pointwise
limit (H M ). Let n = 0 and m and the result of part (iii) is obtained.
20
Proof
For each n define stopping times
S0n
n
Sk+1
:= 0,
o
n
n
n
n
for k 0
:= inf t > Sk : Mt MSk > 2
Define
Tkn := Skn t
Then
Mt2 =
X
2
2
MtS
n MtS n
k
k1
k1
X
MT2 n MT2 n
k
()
k1
k1
=2
X
2
n
n
n
MTk1
MTkn MTk1
+
MTkn MTk1
k1
k1
We can then think of the first term in the decomposition () as (H n M ). Now define
2
X
n
,
Ant :=
MTkn MTk1
k1
()
From the construction of the stopping times Skn we have the following properties
kH n H n+1 k = sup |Htn Htn+1 | 2(n+1)
t
kH H
n+m
kH M k = sup |Htn Mt | 2n
n
The following corollary will be needed to prove the integration by parts formula, and
can be skipped on a first reading; however it is clearer to place it here, since this avoids
having to redefine the notation.
Corollary 8.3.
Let M be a bounded continuous martingale, starting from zero. Then
Z t
2
Mt = 2
Ms dMs + hM it .
0
Proof
In the construction of the quadratic variation process the quadratic variation was constructed as the uniform limit in L2 of processes Ant such that
Ant = Mt2 2(H n M )t ,
where each H n was a bounded previsible process, such that
sup |Htn M | 2n ,
t
21
Theorem 8.4.
The quadratic variation process hM it of a continuous local martingale M is the unique
increasing process A, starting from zero such that M 2 A is a local martingale.
Proof
We shall use a localisation technique to extend the definition of quadratic variation from
L2 bounded martingales to general local martingales.
The mysterious seeming technique of localisation isnt really that complex to understand. The idea is that it enables us to extend a definition which applies for X widgets to
one valid for local X widgets. It achieves this by using a sequence of stopping times which
reduce the local X widgets to X widgets ; the original definition can then be applied to
the stopped version of the X widget. We only need to check that we can sew up the pieces
without any holes i.e. that our definition is independent of the choice of stopping times!
Let Tn = inf{t : |Mt | > n}, define a sequence of stopping times. Now define
hMit := hM Tn i for 0 t Tn
To check the consistency of this definition note that
hM Tn iTn1 = hM Tn1 i
and since the sequence of stopping times Tn , we see that hMi is defined for all t.
Uniqueness follows from the result that any finite variation continuous local martingale
starting from zero is identically zero.
The quadratic variation turns out to be the right sort of variation to consider for
a martingale; since we have already shown that all but the zero martingale have infinite
total variation; and it can be shown that the higher order variations of a martingale are
zero a.s.. Note that the definition given is for a continuous local martingale; we shall
see later how to extend this to a continuous semimartingale.
8.3. Covariation
From the definition of the quadratic variation of a local martingale we can define the covariation of two local martingales N and M which are locally L2 bounded via the polarisation
identity
hM + N i hM N i
.
hM, N i :=
4
We need to generalise this slightly, since the above definition required the quadratic
variation terms to be finite. We can prove the following theorem in a straightforward
manner using the definition of quadratic variation above, and this will motivate the general
definition of the covariation process.
22
23
Theorem 8.5.
For M and N two local martingales which are locally L2 bounded then there exists a unique
finite variation process A starting from zero such that M N A is a local martingale. This
process A is the covariation of M and N .
This theorem is turned round to give the usual definition of the covariation process of
two continuous local martingales as:
Definition 8.6.
For two continuous local martingales N and M , there exists a unique finite variation
process A, such that M N A is a local martingale. The covariance process of N and M
is defined as this process A.
It can readily be verified that the covariation process can be regarded as a symmetric bilinear form on the space of local martingales, i.e. for L,M and N continuous local
martingales
hM + N, Li =hM, Li + hN, Li,
hM, N i =hN, M i,
hM, N i =hM, N i, R.
M )2
2
Zi1
hMiTi hMiTi1 ,
=E (H 2 hMi) .
=E
Z
1/2
2
= E (H hMi)
= E
0
Hs2 dhMis
1/2
.
The space L2 (M ) is then defined as the subspace of the previsible processes, where this
seminorm is finite, i.e.
L2 (M ) := {previsible processes H such that kHkM < }.
24
k n
nk
X
H(i1)/n
i=2
i1 i
,
,
n n
25
26
Proposition 8.8.
For M M2 , for any H L2 (M ) and for any stopping time T then
(H M )T = (H1(0,T ] M ) = (H M T ).
Proof
Consider the following linear maps in turn
f1 :L2 (F ) L2 (F )
f1 :Y
7 E(Y |FT )
This map is a contraction on L2 (F ) since by conditional Jensens inequality
2
E(Y |FT )2 E(Y
|FT ),
Z
0
Z
=
0
Hs2 dhMis
27
8.5. Localisation
Weve already met the idea of localisation in extending the definition of quadratic variation
from L2 bounded continuous martingales to continuous local martingales. In this context
a previsible process {Ht }t0 , is locally previsible if there exists a sequence of stopping
times Tn such that for all n H1(0,Tn ] is a previsible process. Fairly obviously every
previsible process has this property. However if in addition we want the process H to be
locally bounded we need the condition that there exists a sequence Tn of stopping times,
tending to infinity such that H1(0,Tn ] is uniformly bounded for each n.
For the integrator (a martingale of integrable variation say), the localisation is to a
local martingale, that is one which has a sequence of stopping times Tn such that
for all n, X Tn is a genuine martingale.
If we can prove a result like
(H X)T = (H1(0,T ] X T )
for H and X in their original (i.e. non-localised classes) then it is possible to extend the
definition of (H X) to the local classes.
Note firstly that for H and X local, and Tn a reducing sequence1 of stopping times
for both H and X then we see that (H1(0,T ] X T ) is defined in the existing fashion. Also
note that if T = Tn1 we can check consistency
(H1(0,Tn ] X Tn )Tn1 = (H X)Tn1 = (H1(0,Tn1 ] X Tn1 ).
Thus it is consistent to define (H X)t on t [0, ) via
(H X)Tn = (H1(0,Tn ] X Tn ),
n.
We must check that this is well defined, viz if we choose another regularising sequence Sn ,
we get the same definition of (H X). To see this note:
(H1(0,Tn ] X Tn )Sn = (H1(0,Tn Sn ] X Tn Sn ) = (H1(0,Sn ] X Sn )Tn ,
hence the definition of (H X)t is the same if constructed from the regularising sequence
Sn as if constructed via Tn .
The reducing sequence is the sequence of stopping times tending to infinity which makes the local
version of the object into the non-local version. We can find one such sequence, because if say {Tn }
reduces H and {Sn } reduces X then Tn Sn reduces both H and X.
HK)2 hMi min kHk2 kKk2M , kHk2M kKk2 .
28
9. Semimartingales
I mentioned at the start of these notes that the most general form of the stochastic integral
would have a previsible process as the integrand and a semimartingale as an integrator.
Now its time to extend the definition of the It
o integral to the case of semimartingale
integrators.
Definition 9.1.
A process X is a semimartingale if X is an adapted CADLAG process which has a decomposition
X = X0 + M + A,
where M is a local martingale, null at zero and A is a process null at zero, with paths of
finite variation.
Note that the decomposition is not necessarily unique as there exist martingales which
have finite variation. To remove many of these difficulties we shall impose a continuity
condition, since under this most of our problems will vanish.
Definition 9.2.
A continuous semimartingale is a process (Xt )t0 which has a Doob-Meyer decomposition
X = X0 + M + A,
where X0 is F0 -measurable, M0 = A0 = 0, Mt is a continuous local martingale and At is
a continuous adapted process of finite variation.
Theorem 9.3.
The Doob-Meyer decomposition in the definition of a continuous semimartingale is unique.
Proof
Let another such decomposition be
X = X0 + M 0 + A0 ,
where M 0 is a continuous local martingale and A a continuous adapted process of finite
variation. Then consider the process N , where
N = M 0 M = A0 A,
by the first equality, N is the difference of two continuous local martingales, and hence is
itself a continuous local martingale; and by the second inequality it has finite variation.
Hence by an earlier proposition (5.2) it must be zero. Hence M 0 = M and A0 = A.
We define the quadratic variation of the continuous semimartingale as that of the
continuous local martingale part i.e. for X = X0 + M + A,
hXi := hMi.
These definitions can be made to look natural by considering the quadratic variation defined in terms
of a sum of squared increments; but following this approach, these are result which are proved later
using the It
o integral, since this provided a better approach to the discontinuous theory.
[29]
Semimartingales
Similarly if Y + Y0 + N + B is another semimartingale, where B is finite variation and N
is a continuous local martingale, we define
hX, Y i := hM, N i.
We can extend the definition of the stochastic integral to continuous semimartingale
integrators by defining
(H X) := (H M ) + (H A),
where the first integral is a stochastic integral as defined earlier and the second is a
Lebesgue-Stieltjes integral (as the integrator is a process of finite variation).
30
0st
At first sight this may seem to be quite an esoteric definition; in fact it is a natural
extension of convergence in probability to processes. It would also appear to be quite
difficult to handle, however Doobs martingale inequalities provide the key to handling it.
Let
Ht = sup |Hs |,
0st
X
1
d(X, Y ) =
E (min(1, (X Y )n )) ,
n
2
n=1
for X and Y CADLAG processes. The metric space can also be shown to be complete. For
details see Protter.
[31]
Relations to Sums
Since we have just met a new kind of convergence, it is helpful to recall the other usual
types of convergence on a probability space. For convenience here are the usual definitions:
Pointwise
A sequence of random variables Xn converges to X pointwise if for all not in some null
set,
Xn () X().
Probability
A sequence of r.v.s Xn converges to X in probability, if for any > 0,
P (|Xn X| > ) 0, as n .
Lp convergence
A sequence of random variables Xn converges to X in Lp , if
E|Xn X|p 0, as n .
It is trivial to see that pointwise convergence implies convergence in probability. It is
also true that Lp convergence implies convergence in probability as the following theorem
shows
Theorem 10.2.
If Xn converges to X in Lp for p > 0, then Xn converges to X in probability.
Proof
Apply Chebyshevs inequality to f (x) = xp , which yields for any > 0,
P (|Xn | ) p E (|Xn |p ) 0, as n .
Theorem 10.3.
If Xn X in probability, then there exists a subsequence nk such that Xnk X a.s.
Theorem 10.4.
If Xn X a.s., then Xn X in probability.
32
Relations to Sums
33
Theorem 10.2.
Let X be a semimartingale, and H a locally bounded previsible CADLAG process starting
from zero. Then
Z t
X
Hs dXs = lim
Htk2n Xt(k+1)2n Xtk2n u.c.p.
n
k=0
Proof
Let Ks = Hs 1st , and define the following sequence of simple process approximations
Ksn
:=
k=0
Clearly this sequence Ksn converges pointwise to Ks . We can decompose the semimartingale
X as X = X0 + At + Mt where At is of finite variation and Mt is a continuous local
martingale, both starting from zero. The result that
Z t
Z t
n
Ks dAs
Ks dAs , u.c.p.
0
is standard from the Lebesgue-Stieltjes theory. Let Tk be a reducing sequence for the
continuous local martingale M such that M Tk is a bounded martingale. Also since K is
locally bounded we can find a sequence of stopping times Sk such that K Sk is a bounded
previsible process. It therefore suffices to prove for a sequence of stopping times Rk such
that Rk , then
Rk
k
(K n M )R
s (K M )s , u.c.p..
By Doobs L2 inequality, and the It
o isometry we have
2
2
E ((K n M ) (K M ))
4E [(K n M ) (K M )] ,
4kK n Kk2M ,
Z
4 (Ksn Ks )2 dhMis
Doob L2
It
o Isometry
E [(K n M ) (K M )]
So
0, as n .
[(K n M ) (K M )] 0 in L2 ,
Relations to Sums
34
[(K n M ) (K M )] 0 in probability.
Hence
[(K n M ) (K M )] 0 u.c.p.
as required, and putting the two parts together yields
Z t
Z t
n
Ks dXs , u.c.p.
Ks dXs
0
Xt(k+1)2n Xtk2n
2
in probability.
k=0
Proof
In the theorem (7.2) establishing the existence of the quadratic variation process, we noted
in () that
Ant = Mt2 2(H n M )t .
Now from application of the previous theorem
Z t
X
2
Xs dXs = lim
Xtk2n Xt(k+1)2n Xtk2n .
n
In addition,
Xt2
X02
k=0
X
2
2
Xt(k+1)2
.
n Xtk2n
k=0
X
2
At = X02 + lim
Xt(k+1)2n Xtk2n ,
n
k=0
where the limit is taken in probability. Hence the process A is increasing and positive on
the rational numbers, and hence on the whole of R by right continuity.
Remark
The theorem can be strengthened still further by a result of Doleans-Dade to the effect
that for X a continuous semimartingale
X
2
Xt(k+1)2n Xtk2n ,
hXit = lim
n
k=0
where the limit is in the strong sense in L1 . This result is harder to prove (essentially the
uniform integrability of the sums must be proven) and this is not done here.
11. It
os Formula
It
os Formula is the analog of integration by parts in the stochastic calculus. It is also
the first place where we see a major difference creep into the theory, and realise that our
formalism has found a new subtlety in the subject.
More importantly, it is the fundamental weapon used to evaluate It
o integrals; we
shall see some examples of this shortly.
The It
o isometry provides a clean-cut definition of the stochastic integral; however it
was originally defined via the following theorem of Kunita and Watanabe.
Theorem (Kunita-Watanabe Identity) 11.1.
Let M M2 and H and K are locally bounded previsible processes. Then (H M ) is
the unique element of L2 (F ) such that for every N M2 we have:
E [(H M ) N ] = E [(H hM, N i) ]
Moreover we have
h(H M ), N i = H hM, N i.
Proof
Consider an elementary function H, so H = Z1(S,T ] , where Z is an FS measurable bounded
random variable, and S and T are stopping times such that S T . It is clear that
E [(H M ) N ] =E [Z (MT MS ) N ]
=E [Z(MT NT MS NS )]
=E [M (H N ) ]
Now by linearity this can be extended to establish the result for all simple functions (in
L). We finally extend to general locally bounded previsible H, by considering a sequence
(provided it exists) of simple functions H n such that H n H in L2 (M ). Then there exists
a subsequence nk such that H nk converges to H is L2 (N ). Then
E (H nk M ) N E (H M ) N =E (H nk H) M N
r
r
2
E [(H nk H) M )]
r
2 )
E(N
r
2
E [(H nk M ) (H M )]
By construction H nk H in L2 (M ) which means that
kH nk HkM 0, as k .
By the It
o isometry
h
i
2
E ((H nk H) M ) = kH nk Hk2M 0, as k ,
[35]
2 )
E(N
It
os Formula
that is (H nk M ) (H M ) in L2 . Hence as N is an L2 bounded martingale, the right
hand side of the above expression tends to zero as k . Similarly as (H nk N )
(H N ) in L2 , we see also that
E ((H nk N ) M ) 0, as k .
Hence we can pass to the limit to obtain the result for H.
To prove the second part of the theorem, we shall first show that
h(H N ), (K M )i + h(K N ), (H M )i = 2HKhM, N i.
By polarisation
hM, N i =
also
hM + N i hM N i
,
4
(H + K)2 (H K)2
.
HK =
4
Hence
1
2
2
2(HK hM, N i) =
(H + K) (H K) {hM+Ni hM-Ni} .
8
Now we use the result that h(H M )i = (H 2 hMi) which has been proved previously in
theorem (7.9(iii)), to see that
1
2(HK hM, N i) = h(H + K) (M + N )i h(H + K) (M N )i
8
h(H K) (M + N )i + h(H K) (M N )i .
Considering the first two terms
h(H + K)(M + N )i h(H + K) (M N )i =
=h(H + K) M + (H + K) N i h(H + K) M (H + K) N i
=4h(H + K) M, (H + K) N i
by polarisation
=4 (hH M, H N i + hH M, K N i + hK M, H N i + hK M, K N i) .
Similarly for the second two terms
h(H K)(M + N )i h(H K) (M N )i =
=h(H K) M + (H K) N i h(H K) M (H K) N i
=4h(H K) M, (H K) N i by polarisation
=4 (hH M, H N i hH M, K N i hK M, H N i + hK M, K N i) .
36
It
os Formula
Adding these two together yields
2(HK hM, N i) = h(H N ), (K M )i + h(K N ), (H M )i
Putting K 1 yields
2H hM, N i = hH M, N i + hM, H N i.
So it suffices to prove that h(H M ), N i = hM, (H N )i, which is equivalent to showing
that
(H M )N (H N )M
is a local martingale (from the definition of covariation process). By localisation it suffices
to consider M and N bounded martingales, whence we must check that for all stopping
times T ,
E ((H M )T NT ) = E ((H N )T MT ) ,
but by the first part of the theorem
E ((H M ) N ) = E ((H N ) M ) ,
which is sufficient to establish the result, since
T
(H M )T NT =(H M )T N
T
(H N )T MT =(H N )T M
Corollary 11.2.
Let N, M be continuous local martingales and H and K locally bounded previsible processes, then
h(H N ), (K M )i = (HK hN, M i).
Proof
Note that the covariation is symmetric, hence
h(H N ), (K M )i =(H hX, (K M )i)
=(H h(K M ), X)i)
=(HK hM, N i).
We can prove a stochastic calculus analogue of the usual integration by parts formula.
However note that there is an extra term on the right hand side, the covariation of the
processes X and Y . This is the first major difference we have seen between the Stochastic
Integral and the usual Lebesgue Integral.
Before we can prove the general theorem, we need a lemma.
37
It
os Formula
Lemma (Parts for Finite Variation Process and a Martingale) 11.3.
Let M be a bounded continuous martingale starting from zero, and V a bounded variation
process starting from zero. Then
Z t
Z t
Vs dMs .
Ms dVs +
Mt V t =
0
Proof
For n fixed, we can write
X
X
Mt V t =
Mk2n t Vk2n t V(k1)2n t +
V(k1)2n t Mk2n t M(k1)2n t
k1
k1
Mk2n t Vk2n t V(k1)2n t +
Hsn dMs ,
k1
Vs dMs ,
0
and by the Dominated Convergence Theorem for Lebesgue-Stieltjes integrals, the second
term converges to
Z t
Ms dVs ,
0
as n .
Theorem (Integration by Parts) 11.4.
For X and Y continuous semimartingales, then the following holds
Z t
Z t
Xt Yt X0 Y0 =
Xs dYs +
Ys dXs + hX, Y it .
0
Proof
It is trivial to see that it suffices to prove the result for processes starting from zero.
Hence let Xt = Mt + At and Yt = Nt + Bt in Doob-Meyer decomposition, so Nt and Mt
are continuous local martingales and At and Bt are finite variation processes, all starting
from zero. By localisation we can consider the local martingales M and N to be bounded
martingales and the FV processes A and B to have bounded variation. Hence by the usual
(finite variation) theory
Z t
Z t
At Bt =
As dBs +
Bs dAs .
0
38
It
os Formula
39
It only remains to prove for bounded martingales N and M starting from zero that
Z t
Z t
Mt N t =
Ms dNs +
Ns dMs + hM, N it .
0
Reflect for a moment that this theorem is implying another useful closure property of
continuous semimartingales. It implies that the product of two continuous semimartingales
Xt Yt is a continuous semimartingale, since it can be written as a stochastic integrals with
respect to continuous semimartingales and so it itself a continuous semimartingale.
Theorem (It
os Formula) 11.5.
n
Let f : R Rn be a twice continuously differentiable function, and also let X =
(X 1 ,X 2 , . . . , X n ) be a continuous semimartingale in Rn . Then
f (Xt ) f (X0 ) =
n Z
X
i=1
n Z
f
1 X t f
i
(Xs )dXs +
(Xs )dhX i , X j is .
xi
2 i,j=1 0 xi xj
Proof
To prove It
os formula; first consider the n = 1 case to simplify the notation. Then let
A be the collection of C 2 (twice differentiable) functions f : R R for which it holds.
Clearly A is a vector space; in fact we shall show that it is also an algebra. To do this we
must check that if f and g are in A, then their product f g is also in A. Let Ft = f (Xt )
and Gt = g(Xt ) be the associated semimartingales. From the integration by parts formula
Z t
Z t
Ft G t F0 G 0 =
Fs dGs +
Gs dFs + hFs , Gs i.
0
It
os Formula
Also by the Kunita-Watanabe formula extended to continuous local martingales we have
Z t
f 0 (Xs )g 0 (Xs )dhX, Xis .
hF, Git =
0
40
It
os Formula
41
11.1. Applications of It
os Formula
Let Bt be a standard Brownian motion; the aim of this example is to establish that:
t
Bs dBs =
0
1 2 1
B t.
2 t
2
This example gives a nice simple demonstration that all our hard work has achieved something. The result isnt the same as that which would be given by the logical extension of
the usual integration rules.
To prove this we apply It
os formula to the function f (x) = x2 . We obtain
t
Z
f (Bt ) f (B0 ) =
0
1
f
(Bs )dBs +
x
2
2f
(Bs )dhB, Bis ,
x2
Bt2
=2
0
1
Bs dBs + 2ds,
2
Bs dBs =
0
Bt2
t
.
2
2
For those who have read the foregoing material carefully, there are grounds to complain that
there are simpler ways to establish this result, notably by consideration of the definition of
the quadratic variation process. However the point of this example was to show how It
os
formula can help in the actual evaluation of stochastic integrals; not to establish a totally
new result.
Zs dXs ,
0
so clearly if X is a continuous local martingale, i.e. X Mcloc then this implies, by the
stability property of stochastic integration, that Z Mcloc also.
It
os Formula
Proof
For existence, apply It
os formula to f (x) = exp(x) to obtain
d (exp(Yt )) = exp(Yt )dYt +
1
exp(Yt )dhY, Y it .
2
Hence
1
1
1
d exp(Xt hXit ) = exp(Xt hXit )d Xt hXit
2
2
2
1
1
1
1
+ exp Xt hXit d Xt hXit , Xt hXit
2
2
2
2
1
1
1
= exp(Xt hXit )dXt exp Xt hXit dhXit
2
2
2
1
1
+ exp Xt hXit dhXit
2
2
=Zt dXt
Hence Zt certainly solves the equation. Now to check uniqueness, define
1
Yt = exp Xt + hXit ,
2
we wish to show that for every solution of the Stochastic Differential Equation Zt Yt is a
constant. By a similar application of It
os formula
dYt = Yt dXt + Yt dhXit ,
whence by integration by parts (alternatively consider It
o applied to f (x, y) = xy),
d(Zt Yt ) =Zt dYt + Yt dZt + hZ, Y it ,
=Zt (Yt dXt + Yt dhXit ) + Yt Zt dXt + (Yt Zt )dhXit ,
=0.
So Zt Yt is a constant, hence the unique solution of the stochastic differential equation
dZt = Zt dXt , with Z0 = 1, is
1
Zt = exp Xt hXit .
2
42
It
os Formula
Example
Let Xt = Bt , for an arbitrary scalar . Clearly Xt is a continuous local martingale, so
the associated exponential martingale is
1 2
Mt = exp Bt t .
2
It is clear that the exponential semimartingale of a real valued martingale must be
non-negative, and thus by application of Fatous lemma we can show that it is a supermartingale, thus E(Mt ) 1 for all t.
Theorem 11.6.
Let M be a non-negative local martingale, such that EM0 < then M is a supermartingale.
Proof
Let Tn be a reducing sequence for Mn M0 , then for t > s 0,
E(MtTn |Fs ) = E(M0 |Fs ) + E(MtTn M0 |Fs )
= M0 + MsTn M0 = MsTn .
Now by application of the conditional form of Fatous lemma
E(Mt |Fs ) = E(lim inf MtTn |Fs ) lim inf E(MtTn |Fs ) = Ms .
n
Ms dBs ,
0
hence M is a local martingale. Fix T a constant time, which is of course a stopping time,
then B T is an L2 bounded martingale (E(BtT )2 = t T T ). We then show that M T is
43
It
os Formula
44
in L2 (B T ) as follows
Z
Ms2 dhBis
kM kB = E
!
Ms2 ds ,
=E
0
Z
E
exp(2Bs )ds ,
0
T
Z
=
E (exp(2Bs )) ds,
0
T
=
0
In the final equality we use that fact that Bs is distributed as N (0, s), and we use the characteristic function of the normal distribution. Thus by the integration isometry theorem
we have that (M T B)t is an L2 bounded martingale. Thus for every such T , Z T is an L2
bounded martingale, which implies that M is a martingale.
The Exponential Martingale Inequality
We have seen a specific proof that a certain (important) exponential martingale is a true
martingale, we now show a more general argument.
Theorem 11.7.
Let M be a continuous local martingale, starting from zero. Suppose for each t, there exists
a constant Kt such that hM it < a.s., then for every t, and every y > 0,
P
0st
st
exp(y + 1/22 Kt ).
Optimizing over now gives the desired result. For the second part, we establish the
It
os Formula
45
following bound
E
sup Zs
E exp
0st
sup Zs
0st
()
We have previously noted that Z is a local martingale; let Tn be a reducing sequence for
Z, hence Z Tn is a martingale, hence
E [ZtTn |Fs ] = ZsTn .
()
We note that Zt is dominated by exp( sup0st Zs ), and thus by our bound we can
apply the dominated convergence theorem to () as n to establish that Z is a true
martingale.
Corollary 11.8.
For all , > 0,
P sup Mt & hM i exp(2 /2).
t0
Proof
Set T = inf{t 0 : Mt }, the conditions of the previous theorem now apply to M T ,
with Kt = .
From this corollary, it is clear that if H is any bounded previsible process, then
Z t
Z
1 t
2
exp
|Hs | ds
Hs dBs
2 0
0
is
R a true martingale, since this is the exponential semimartingale associated with the process
HdB.
Corollary 11.9.
If the bounds Kt on hM i are uniform, that is if Kt C for all t, then the exponential
martingale is Uniformly Integrable. We shall use the useful result
Z
Z
Z
P(X log )d =
E 1eX d = E
1eX d = E(eX ).
0
Proof
Note that the bound () extends to a uniform bound
Z
exp (log )2 /2C d < .
E sup Zt 1 +
t0
12. L
evy Characterisation of Brownian Motion
A very useful result can be proved using the It
o calculus about the characterisation of
Brownian Motion.
Theorem 12.1.
Let {B i }t0 be continuous local martingales starting from zero for i = 1, . . . , n. Then
Bt = (Bt1 , . . . , Btn ) is a Brownian motion with respect to (, F, P) adapted to the filtration
Ft , if and only iff
hB i , B j it = ij t i, j {1, . . . , n}.
Proof
In these circumstances it follows that the statement Bt is a Brownian Motion is by definition
equivalent to stating that Bt Bs is independent of Fs and is distributed normally with
mean zero and covariance matrix (t s)I.
Clearly if Bt is a Brownian motion then the covariation result follows trivially from the
definitions. Now to establish the converse, we assume hB i , B j it = ij t for i, j {1, . . . , n},
and shall prove Bt is a Brownian Motion.
Observe that for fixed Rn we can define Mt by
Mt
1 2
:= f (Bt , t) = exp i(, x) + || t .
2
By application of It
os formula to f we obtain (in differential form using the Einstein
summation convention)
f
f
1 2f
j
(B
,
t)dB
+
(B
,
t)dt
+
(Bt , t)dhB j , B k it ,
t
t
t
xj
t
2 xj xk
1 2
1
=ij f (Bt , t)dBtj + || f (Bt , t)dt j k jk f (Bt , t)dt
2
2
j
=ij f (Bt , t)dBt .
d (f (Bt , t)) =
Hence
Mt
Z
=1+
and is a sum of stochastic integrals with respect to continuous local martingales and is
hence itself a continuous local martingale. But note that for each t,
1 2
|Mt | = e 2 || t <
Hence for any fixed time t0 , (M t0 )t satisfies
|(M t0 )t | |(M t0 ) | < ,
[46]
47
1
2
E (exp (i(, Bt Bs )) |Fs ) = exp (t s) ||
2
a.s.
However this is just the characteristic function of a normal random variable following
N (0, t s); so by the Levy character theorem Bt Bs is a N (0, t s) random variable.
Proof
We may assume that the map t 7 hMit is strictly increasing. Note that the map s 7 s is
the inverse to t 7 hMit . Hence the results (i),(ii) and (iii).
Define
Tn := inf{t : |M |t > n},
[Un := hMiTn .
Note that from these definitions
tUn = inf{s > 0 : hMis > t Un }
= inf{s > 0 : hMis > t hMiTn }
=Tn t
So
Tn
n
AU
s = AsUn = Mt .
E AU
|F
=E
A
|
F
=
E
M
|
F
s
tUn
s
s
t
t
n
=E MTtn |Fs = MTsn = AU
s .
[48]
h
E
MS q Mq
2
i
|Fq = 0,
49
f (s)dBs ,
Mt :=
0
satisfies
Z t
2
Mt N 0,
|f (s)| ds .
0
Proof
From the definition of M via a stochastic integral with respect to a continuous martingale,
it is clear that M is a continuous local martingale, and by the Kunita-Watanabe result,
the quadratic variation of M is given by
Z
hMit =
|f (s)|ds,
0
1 2
Zt = exp iMt + hMit ,
2
as in the Levy characterisation proof, we see that this is a continuous local martingale,
and by boundedness furthermore is a martingale, and hence
E(Z0 ) = E(Zt ),
whence
1
E (exp(iMt )) = E exp 2
2
f (s) ds
2
50
Ls = EP
dQ
Fs .
dP
Here Lt is the Radon-Nikodym derivative of Q with respect to P. The first result basically
shows that this is a martingale, and the second is a continuous time version of Bayes
theorem.
Proof
The first part is basically the statement that the Radon-Nikodym derivative is a martingale.
This follows because the measures P and Q are equivalent, but this will not be proved in
detail here. Let Y be an Ft measurable random variable, such that EQ (|Y |) < . We shall
prove that
1
EQ (Y |Fs ) =
EP [Y Lt |Fs ] a.s. (P and Q).
Ls
Then for any A Fs , using the definition of conditional expectation we have that
1
EQ 1A EP [Y Lt |Fs ] =EP (1A EP [Y Lt |Fs ])
Ls
=EP [1A Y Lt ] = EQ [1A Y ] .
Substituting Y = Xt gives the desired result.
[51]
Girsanovs Theorem
Theorem (Girsanov).
Let M be a continuous local martingale, and let Z be the associated exponential martingale
1
Zt = exp Mt hM it .
2
If Z is uniformly integrable, then a new measure Q, equivalent to P may be defined by
dQ
= Z .
dP
Then if X is a continuous P local martingale, X hX, M i is a Q local martingale.
Proof
Since Z exists a.s. it defines a uniformly integrable martingale (the exponential martingale), a version of which is given by Zt = E(Z |Ft ). Hence Q constructed thus is
a probability measure which is equivalent to P. Now consider X, a P local martingale.
Define a sequence of stopping times which tend to infinity via
Tn := inf{t 0 : |Xt | n, or |hX, M it | n}.
Now consider the process Y defined via
Y := X Tn hX Tn , M i.
By It
os formula for 0 t Tn , remembering that dZt = Zt dMt as Z is the exponential
martingale associated with M ,
d(Zt Yt ) =Zt dYt + Yt dZt + hZ, Y i
=Zt (dXt dhX, M i) + Yt Zt dMt + hZ, Y i
=Zt (dXt dhX, M i) + (Xt hX, M it )Zt dMt + Zt dhX, M i
=(Xt hX, M it )Zt dMt + Zt dXt
Where the result hZ, Y it = Zt hX, M it follows from the Kunita-Watanabe theorem. Hence
ZY is a P-local martingale. But since Z is uniformly integrable, and Y is bounded (by
construction of the stopping time Tn ), hence ZY is a genuine P-martingale. Hence for s < t
and A Fs , we have
EQ [(Yt Ys )1A ] = E [Z (Yt Ys )1A ] = E [(Zt Yt Zs Ys )1A ] = 0,
hence Y is a Q martingale. Thus X hX, M i is a Q local martingale, since Tn is a reducing
sequence such that (X hX, M i)Tn is a Q-martingale, and Tn as n .
Corollary 14.3.
t := Wt hW, M it is a Q Brownian motion.
Let Wt be a P Brownian motion, then W
Proof
t is continuous and
Use Levys characterisation of Brownian motion to see that since W
is a Q Brownian
hW , W it = hW, W it = t, since Wt is a P Brownian motion, then W
motion.
52
The proof of this result can seem hard if you are not familiar with functional analysis
style arguments. The outline of the proof is to describe all Y s which are GT measurable
which cannot be represented in the form (1) as belonging to the orthogonal complement of
a space. Then we show that for Z in this orthogonal complement that E(ZX) = 0 for all X
in a large space of GT measurable functions. Finally we show that this space is sufficiently
big that actually we have proved this for all GT measurable functions, which includes Z so
E(Z 2 ) = 0 and hence Z = 0 a.s. and we are done!
Proof
Without loss of generality prove the result in the case EY = 0 where Y is L2 integrable
and measurable with respect to GT for some constant T > 0.
Define the space
(
!
)
Z
T
L2T (B)
kHs k2 ds
H : H is Gt previsible and E
<
Hs dBs .
I(H) =
0
As a consequence of the It
o isometry theorem, this map is an isometry. Hence the image V
under I of the Hilbert space L2T (B) is complete and hence a closed subspace of L20 (GT ) =
{H L2 (GT ) : EH = 0}. The theorem will be proved if we can establish that the image is
the whole space.
We follow the usual approach in such proofs; consider the orthogonal complement of
V in L20 (GT ) and we aim to show that every element of this orthogonal complement is zero.
Suppose that Z is in the orthogonal complement of L20 (GT ), thus
E(ZX) = 0 for all X L20 (GT )
[53]
(2)
1 2
Zs exp || (t s) = E ( Zt exp (i (Bt Bs )| Gs )) .
2
Consider a partition 0 < t1 < t2 < < tm T , and by repeating the above argument,
conditioning on each Gtj we establish that
X
1X
2
(n)
K
m
X (r)
X
(r)
P (r) (x1 , . . . , xm ) =
ck exp i
aj,k xj
k=1
j=1
54
m
exp i
s dBs : L ([0, t], R )
0
is total in L1 . A set S is said to be total if E(af ) = 0 for all a S implies a = 0 a.s.. The
full proof of this result will reappear in a more abstract form in the stochastic filtering
section of these notes.
55
Lipshitz Conditions
Let k k denote the usual Euclidean norm on Rd . Recall that a function f is said to be
Lipshitz if there exists a constant K such that
kf (x) f (y)k Kkx yk,
we shall generalise the norm concept to a (d r) matrix by defining
2
kk =
d X
r
X
2
ij
.
i=1 j=1
The concept of Liphshitz continuity can be extended to that of local Lipshitz continuity,
by requiring that for each n there exists Kn , such that for all x and y such that kxk n
and kyk n then
kf (x) f (y)k Kn kx yk.
[56]
57
strong solutions of () relative to B with initial condition then X and X are indistinguishable, that is
h
i
t t : 0 t < = 1.
P Xt = X
The proof of this result is importanty inasmuch as it illustrates the first example of a
technique of bounding which recurs again and again throughout the theory of stochastic
differential equations. Therefore I make no apology for spelling the proof out in excessive
detail, as it is most important to understand exactly where each step comes from!
Proof
are strong solutions of (), relative to the same brownian motion
Suppose that X and X
B and initial condition on the same probability space (, F, P). Define a sequence of
stopping times
n = inf{t 0 : kXt k n}, and n = inf{t 0 : kYt k n}.
Now set Sn = min(n , n ). Clearly Sn is also a stopping time, and Sn a.s. (P) as
n . These stopping times are only needed because b and are being assumed merely
to be locally Lipshitz. If they are assumed to be Lipshitz, as will be needed in the existence
part of the proof, then this complexity may be ignored.
Hence
Z tSn
XtSn XtSn =
b(u, Xu ) b(u, Xu ) du
0
tSn
Z
+
u ) dWu .
(u, Xu ) (u, X
"Z
tSn
b(u, Xu ) b(u, Xu ) du
#2
"Z
+ 4E
0
tSn
u ) dWu
(u, Xu ) (u, X
#2
58
tSn
u ) dWu
(u, Xu ) (u, X
#2
"Z
tSn
=E
#
u )|2 du
|(u, Xu ) (u, X
The classical H
older inequality (in the form of the Cauchy Schwartz inequality) for
Lebsgue integrals which states that for p, q (1, ), with p1 + q 1 = 1 the following
inequality is satisfied.
Z
Z
|f (x)g(x)|d(x)
1/p Z
1/q
q
|f (x)| d(x)
|g(x)| d(x)
p
This result may be applied to the other term, taking p = q = 2 which yields
"Z
E
tSn
u ) du
b(u, Xu ) b(u, X
#2
"Z
tSn
"Z
u ) du
b(u, Xu ) b(u, X
tSn
Z
1ds
tSn
#2
#
2
u ) ds
b(u, Xu ) b(u, X
" Z
E t
tSn
2
b(u, Xu ) b(u, Xu )
Thus combining these two useful inequalities and using the nth local Lipshitz relations we
have that
"Z
#
tSn
2
2
tS k 4tE
u)
EkXtS X
b(u, Xu ) b(u, X
n
"Z
+ 4E
tSn
#
u )|2 du
|(u, Xu ) (u, X
4(T +
1)Kn2 E
Z t
uS
XuSn X
n
2
du
Now by Gronwalls lemma, which in this case has a zero term outside of the integral, we
tS k2 = 0, and hence that P(XtS = X
tS ) = 1 for all t < .
see that EkXtSn X
n
n
n
That is these two processes are modifications, and thus indistinguishable. Letting n
t }t0 .
we see that the same is true for {Xt }t0 and {X
Now we impose Lipshitz conditions on the functions b and to produce an existence
result. The following form omits some measure theoretic details which are very important;
for a clear treatment see Chung & Williams chapter 10.
59
. Define Xt
(0)
recursively, with Xt
(k+1)
Xt
= , and
Z t
Z t
k
(k)
= F (X )t = +
b(s, Xs )ds +
(s, Xs(k) )dBs
0
[Note: we have left out checking that the image of X under F is indeed adapted!] Now note
that using (a + b)2 2a2 + ab2 , we have using the same bounds as in the uniqueness result
that
Z
2
"
2 #
T
E
sup F (X)t F (Y )t
2E sup
((Xs ) (Ys )) dBs
0tT
tT 0
Z
2
T
+ 2E sup
(b(Xs ) b(Ys )) ds
tT 0
"
#
Z
2
2K (4 + T )
E
0
sup |Xt Yt |
tT
n!
tT
tT
dt.
Xt = Xt
for t N, N N,
which solves the SDE, and has already been shown to be unique.
60
k
ik (Xt )kj (Xt ) i j .
A=
b (Xt ) j +
x
2 i=1 j=1
x x
j=1
k=1
It is conventional to set
aij =
d
X
ik kj ,
k=1
1 XX
aij (Xt ) i j .
b (Xt ) j +
A=
x
2 i=1 j=1
x x
j=1
j
Z t X
d
d X
d X
d
2
X
f
f
1
f (Xt ) f (X0 ) =
bj (Xs ) j (Xs ) +
ik kj i j (Xs ) dt
x
2 i=1 j=1
x x
0
j=1
k=1
Z tX
d
ij (Xs )
0 j=1
Z
=
Af (Xs )ds +
0
f
(Xs )dBs
xj
Z tX
d
0 j=1
[61]
ij (Xs )
f
(Xs )dBs
xj
Definition 17.1.
We say that Xt satisfies the martingale problem for A, if Xt is Ft adapted and
Z
Mt = f (Xt ) f (X0 )
Af (Xs )ds,
0
Z t
:= (t, Xt ) (0, X0 )
0
+ A (s, Xs )ds.
s
then Mt is a local martingale, for Xt a solution of the SDE associated with the infinitesimal
generator A. The proof follows by an application of It
os formula to Mt , similar to that
of the above discussion.
A=
d
X
j=1
bj (Xt )
1 XX
2
+
a
(X
)
,
ij
t
xj
2 i=1 j=1
xi xj
which is associated as before to an SDE. This SDE will play an important role in what is
to follow.
A simple example of a Dirichlet Problem is the solution of the Laplace equation in
the disc, with Dirichlet boundary conditions on the boundary, i.e.
2 u = 0 on D,
u = f on D.
62
63
Theorem 17.2.
For each f Cb2 () there exists a unique u Cb2 () solving the Dirichlet problem for f .
Moreover there exists a continuous function m : (0, ) such that for all f Cb2 ()
this solution is given by
Z
u(x) =
m(x, y)f (y)(dy).
Now remember the SDE which is associated with the infinitesimal generator A:
dXt =b(Xt )dt + (Xt )dBt ,
X0 =x0
Often in what follows we shall want to consider the conditional expectation and probability
measures, conditional on x0 = x, these will be denoted Ex and Px respectively.
Theorem (Dirichlet Solution).
Define a stopping time via
T := inf{t 0 : Xt
/ }.
Then u(x) given by
"Z
u(x) := Ex
tT
Mt := u(XT t ) +
(Xs )ds.
0
We shall now show that this Mt is a martingale. For t T , it is clear that dMt = 0. For
t < T by It
os formula
dMt = du(Xt ) + (Xt )dt.
Also, by It
os formula,
d
d
d
X
1 X X 2u
u
i
(Xt )dXt +
(Xt )dhX i , X j it
du(Xt ) =
i
i xj
x
2
x
i=1 j=1
j=1
d
d
d
d
X
u
1 XXX
du
=
(X
)
[b(X
)dt
+
(X
)dB
]
+
(Xt )dt
t
t
t
t
ik
kj
j
i xj
x
2
x
j=1
i=1 j=1
k=1
=Au(Xt )dt +
d
X
j=1
(Xt )
u
(Xt )dBt .
xj
d
X
j=1
(Xt )
u
(Xt )dBt .
xj
64
d
X
(Xt )
j=1
d
X
j=1
(Xt )
u
(Xt )dBjt ,
j
x
u
(Xt )dBjt .
j
x
from which we conclude by the stability property of the stochastic integral that Mt is a
local martingale. However Mt is uniformly bounded on [0, t], and hence Mt is a martingale.
In particular, let (x) 1, and f 0, by the optional stopping theorem, since T t
is a bounded stoppping time, this gives
u(x) = Ex (M0 ) = Ex (MT t ) = Ex [u(XT t ) + (T t)].
Letting t , we have via monotone convergence that Ex (T ) < , since we know that
the solutions u is bounded from the PDE solution existance theorem; hence T < a.s..
We cannot simply apply the optional stopping theorem directly, since T is not necessarily
a bounded stopping time. But for arbitrary and f , we have that
|Mt | kuk + T kk = sup |u(x)| + T sup |(x)|,
x
Z
M = f (XT ) +
(Xs )ds.
0
#
(Xs )ds .
Theorem 17.4.
Let u Cb1,2 (R Rd ) be the solution of the Cauchy Problem for f . Then define
T := inf{t 0 : Xt
/ },
a stopping time. Then
u(t, x) = Ex [f (XT t )]
65
66
Proof
Fix s (0, ) and consider the time reversed process
Mt := u((s t) T, XtT ).
There are three cases now to consider; for 0 T t s, Mt = u((s t) T, XT ),
where XT , so from the boundary condition, Mt = f (XT ), and hence it is clear that
dMt = 0. For 0 s T t and for 0 t s T , the argument is similar; in the latter
case by Itos formula we obtain
d
X
u
u
(s t, Xt )dt +
(s t, Xt )dXtj
dMt =
j
t
x
j=1
d
1 X X 2u
+
(s t, Xt )dhX i , X j it ,
2 i=1 j=1 xi xj
d
d
X
X
u
u
(s t, Xt )
jk (Xt )dBkt .
=
+ Au (s t, Xt )dt +
j
t
x
j=1
k=1
We obtain a similar result in the 0 t T s, case but with s replaced by T . Thus for
u solving the Cauchy Problem for f , we have that
u
+ Au = 0,
t
we see that Mt is a local martingale. Boundedness implies that Mt is a martingale, and
hence by optional stopping
u(s, x) = Ex (M0 ) = Ex (Ms ) = Ex (f (XsT )),
17.4. Feynman-Ka
c Representation
Feynman observed the following representation for the representation of the solution of a
PDE via the expectation of a suitable function of a Brownian Motion intuitively and the
theory was later made rigorous by Kac.
In what context was Feynman interested in this problem? Consider the Schr
odinger Equation,
~2 2
d
X
bj (Xt )
j=1
1 XX
2
a
(X
)
+
.
ij
t
xj
2 i=1 j=1
xi xj
Now consider the more general form of the same Cauchy problem where we consider a
Cauchy Problem with generator L of the form:
d
X
1 XX
2
LA+v =
b (Xt ) j +
aij (Xt ) i j + v(Xt ).
x
2 i=1 j=1
x x
j=1
j
1 2
,
2
L=
1 2
+ v(Xt ),
2
Proof
Note that the SDE for Xt has the form
dXt = a(Xt )dBt + b(Xt )dt.
67
v(Xr )dr .
v(Xr )dr ,
Et = exp
0
whence by It
os formula dEt = Et v(Xt )dt. For 0 t s, we have
d
X
u
u
j
dMt =
(s t, Xt )Et dBt +
+ Au + vu (s t, Xt )Et dt
j
x
t
j=1
d
X
u
=
(s t, Xt )Et dBtj .
j
x
j=1
68
Stochastic Filtering
70
Y0 = 0,
(5)
(6)
for t [0, T ] with suitable constants C and c which may be functions of T . As a result of
this bound
"Z
#
T
2
E
kh(s, Xs )k ds < .
0
_
t0
Yt .
Stochastic Filtering
71
t [0, T ].
(7)
The following proofs are in many cases made complex by the fact that we do not
assume that the functions f , h and are bounded.
Given this expression, Zt is a positive continuous local martingale and hence is a supermartingale. To prove that it is a true martingale we must prove in addition that it has
constant mean.
Let Tn be a reducing sequence for the local martingale Zt , i.e. an increasing sequence
of stopping times tending to infinity as n such that for each n, Z Tn is a genuine
martingale. By Fatous lemma, and the local martingale property
EZt = E lim ZtTn lim inf EZtTn = EZ0Tn = 1,
n
so
EZt 1 t.
(9)
x
,
1 + x
Stochastic Filtering
72
we obtain that
t
Z
f (Zt ) = f (Z0 ) +
0
Hence
Zt
1
=
1 + Zt
1+
Z
0
1
f (Zs )dZs +
2
0
Z
0
f 00 (Zs )dhZs , Zs i,
Zs hT (s, Xs )
dWs
(1 + Zs )2
Z
0
(10)
Zs hT (s, Xs )
dWs ,
(1 + Zs )2
clearly this a local martingale, since it is a stochastic integral. The next step in the proof
is explained in detail as it is one which occurs frequently. We wish to show that the above
stochastic integral is in fact a genuine martingale. From the earlier theory (for L2 integrable
martingales) it suffices to show that integrand is in L2 (W ). We therefore compute
"Z
2 #
t
T
Zs hT (s, Xs )
Z
h
(s,
X
)
s
s
(1 + Zs )2
= E
(1 + Zs )2
ds ,
0
W
with a view to showing that it is finite, whence the integrand must be in L2 (W ) and hence
the integral a genuine martingale. We note that since Zs 0, that
1
Zs2
1
1, and
2,
2
(1 + Zs )
(1 + Zs )
so,
Zs hT (s, Xs )
2
Zs2
2
=
kh(s, Xs )k
(1 + Zs )2
4
(1 + Zs )
Zs2
1
2
=
kh(s, Xs )k
(1 + Zs )2
(1 + Zs )2
1
2
2 kh(s, Xs )k .
Using this inequality we obtain
"Z
"Z
#
2 #
t
T
T
Z
h
(s,
X
)
1
s
s
2
E
kh(s, Xs )k ds ,
(1 + Zs )2
ds 2 E
0
0
however as a consequence of our linear growth condition on h, the last expectation is finite.
Therefore
Z t
Zs hT (s, Xs )
dWs
(1 + Zs )2
0
is a genuine martingale, and hence taking the expectation of (10) we obtain
Z t 2
Zs kh(s, Xs )k2
Zt
1
E
=
E
ds .
1 + Zt
1+
(1 + Zs )3
0
Stochastic Filtering
Consider the integrand on the right hand side; clearly it tends to zero as 0. But also
Zs
Zs2 kh(s, Xs )k2
=
Zs kh(s, Xs )k2
3
(1 + Zs )
(1 + Zs )3
1 + Zs
Zs kh(s, Xs )k2
(1 + Zs )3
kZs 1 + kXs k2 ,
where we have used the fact that kh(s, Xs )k2 k(1 + kXs k2). So from the fact that
EZs 1 and using the next lemma we shall see that E Zs kXs k2 C. Hence we conclude
that kZs (1 + kXs k2 ) is an integrable dominating function for the integrand on interest.
Hence by the Dominated Convergence Theorem, as 0,
Z t 2
Zs kh(s, Xs )k2
E
ds 0.
(1 + Zs )3
0
In addition, since E(Zs ) 1, the bounded convergence theorem yields
Zs
E
E(Zs ), as 0.
1 + Zs
Hence we conclude that
E(Zt ) = 1
t,
Zs kXs k ds < C(T ) t [0, T ].
2
Proof
To establish this result It
os formula is used to derive two important results
d(kXt k2 ) =2XTt (f dt + dWs ) + tr( T )dt
d(Zt kXt k2 ) = Zt kXt k2 hT dVs + Zt 2XTt (f dt + dWs ) + tr( T ) dt
By It
os formula,
Zt kXt k2
1
2
d
=
d
Z
kX
k
+
dhZt kXt k2 i.
t
t
1 + Zt kXt k2
(1 + Zt kXt k2 )2
(1 + Zt kXt k2 )3
Substituting for d(Zt kXt k2 ) into this expression yields
Zt kXt k2
1
2 T
T
Z
kX
k
h
dW
+
Z
2X
dV
d
=
t
t
t
t
s
t
2
1 + Zt kXt k2
(1 + Zt kXt k2 )
"
#
Zt2 kXt k4 hT h + 4Zt2 2XTt T Xt
Zt 2XTt f + tr( T )
+
dt.
2
3
(1 + Zt kXt k2 )
(1 + Zt kXt k2 )
73
Stochastic Filtering
74
After integrating this expression from 0 to t, the terms which are stochastic integrals
are local martingales. In fact they can be shown to be genuine martingales. For example
consider the term
Z t
Zs 2XTs
dVs ,
2 2
0 (1 + Zs kXs k )
to show that this is a martingale we must therefore establish that
Z t"
E
0
#2
Zs 2XTs
"Z
ds = 4E
(1 + Zs kXs k2 )
Zs2 XTs T Xs
#
4 ds
(1 + Zs kXs k2 )
< .
In order to establish this inequality consider the term XTt T Xt which is a sum over terms
of the form
|Xti ij (t, Xt )kj (t, Xt )Xtk | kXt k2 kk2 ,
of which there are d3 terms (in Rd ). But by the linear increase condition on , we have for
some constant
kk2 (1 + kXk2 ),
and hence
|Xti ij (t, Xt )kj (t, Xt )Xtk | kXt k2 1 + kXt k2 ,
so the integral may be bounded by
t
Zs2 XTs T Xs
4 ds
(1 + Zs kXs k2 )
(1 + Zs kXs k2 )
= d3
"Z
0
Zs2 kXs k2
ds
Z
4 ds
(1 + Zs kXs k2 )
+
0
Zs2 kXs k4
4
(1 + Zs kXs k2 )
ds
Zs2 kXs k2
4 ds
(1 + Zs kXs k2 )
Zs kXs k2
1
ds
2
(1 + Zs kXt k ) (1 + Zs kXs k2 )3
Z
1 t
Zs
ds
Zs .
0
Zs
0
The last integral has a bounded expectation since E (Zs ) 1. Similarly for the second
term,
Z
0
Zs2 kXs k4
4 ds
(1 + Zs kXs k2 )
Z
0
Zs2 kXs k4
(1 + Zs kXs
2
k2 )
1
(1 + Zs kXs
2 ds
k2 )
1
t < .
2
A similar argument holds for the other stochastic integrals, so they must both be genuine
martingales, and hence if we take the expectations of the integrals they are zero. Integrating
Stochastic Filtering
75
the whole expression from 0 to t, then taking the expectation and finally differentiation
with respect to t yields
d
E
dt
Zt kXt k2
1 + Zt kXt k2
Zt (2XTt f + tr( T )
E
(1 + Zt kXt k2 )2
Zt kXt k2
k 1+E
.
1 + Zt kXt k2
An application of Gronwalls inequality (see the next section for a proof of this useful
result) establishes,
Zt kXt k2
E
C(T ); t [0, T ].
1 + Zt kXt k2
Applying Fatous lemma the desired result is obtained.
may
Given that Zt has been shown to be a martingale a new probability measure P
be defined by the Radon-Nikodym derivative:
dP
= ZT .
dP o
FT
This change of measure allows us to establish one of the most important results in stochastic
filtering theory.
Proposition 18.3.
the observations process Yt is a Brownian motion, which
Under the probability measure P,
is independent of Xt .
Proof
Define a process Mt taking values in R by
Z
Mt =
hT (s, Xs ) dWs ,
(11)
dP
1 t
T
2
Yt =
h (s, Xs ) dWs
kh(s, Xs )k ds = Zt .
= E(Mt ) = exp
dP
2 0
0
t
local martingale.
As a corollary of Girsanovs theorem Wt hW, M it is a continuous P
the process
Levys characterisation theorem then implies that under P
Z
Wt hW, M it = Wt +
h(s, Xs ) ds,
0
Stochastic Filtering
76
is an m dimensional Brownian Motion. But this process is just the observation process Yt
the observation process Yt is a Brownian
by definition. Thus we have proven that under P,
Motion. To establish the second part of the proposition we must prove the independence
of Y and X. The law of the pair (X, Y) on [0, T ] is absolutely continuous with respect to
that of the pair (X, V) on [0, T ], since the latter is equal to the former plus a drift term.
Their Radon-Nikodym derivative (which exists since they are absolutely continuous
with respect to each other) is T , which means that for f any bounded measurable function
EP [f (Xt , Yt ) ] = EP [EP [f (Xt , Yt )t |Ft ]]
= EP [f (Xt , Yt )EP [ |Ft ]]
= EP [f (Xt , Yt )t ] = EP [f (Xt , Vt )] .
Z
0
1
h (s, Xs )dYt
2
T
Z
0
kh(s, Xs )k ds .
2
Stochastic Filtering
Also for any stopping time T , EP ZT = EP ZT ZT = 1, the constant mean condition
martingale and thus by the martingale convergence theorem
implies that ZT is a UI P
Zt Z in L1 . So we may write
dP
= Z ,
dP F
whence
dP
= Zt , for all t 0.
F
dP
t
Now define, for every bounded measurable function , the unnormalised conditional
distribution of X via
h
i
t () :=EP (Xt )Zt |Yt
h
i
i
h
i
h
EP t ()EP Zt |Yt b = EP EP (Xt )Zt |Yt b .
which is equivalent to
EP (t ()t (1)b) = EP (t ()b) .
This implies that
EP ( t ()t (1)| Yt ) = EP ( t ()| Yt )
As the random variables t (), t () and t (1) from their definitions are all Yt measurable
this implies that
t ()t (1) = t ()
Since Zt > 0 a.s. which implies that t (1) > 0 a.s. this is sufficient to prove the required
result.
77
Stochastic Filtering
78
Remark
The above proof is simply a specific proof of the abstract form of Bayes theorem, the
general form of which is a standard result, see e.g. A.0.3 in [Musiela and Rutkowski, 2005].
S a = 0.
Lemma 18.7.
Let
Z t
Z
1 t
2
T
krs k ds
St := t = exp i
rs dYs +
2 0
0
rs L
([0, t], R ) ,
m
kTj Ytj
j=1
p
X
Ti (Ytj
Z
Ytj1 ) =
bTs dYs
j=1
where
p = kp ,
p1 = kp + kp1 , . . . ,
and
bs =
Hence
j
0
1 = kp + k1 .
for t (tj1 , tj )
for t (tp , T )
Z t
p
X
a exp i
a exp i
E
bs dYs
=E
kTj Ytj = 0
j=1
viz
and the same holds for linear combinations by the linearity of E
a
E
K
X
k=1
ck exp i
p
X
j=1
kTj,k Ytj = 0
Stochastic Filtering
where c1 , . . . , cK are complex coefficients. If F (x1 , . . . , xp ) is a continuous bounded complex
valued function defined on (Rm )p then since the set
p
K
X
X
T
f (x1 , . . . , xp ) =
ck exp i
kj,k xj
: ck C k, kj,k C j, k
j=1
k=1
is closed under complex conjugation, and separates points in (Rm )p , the Stone-Weierstrass
approximation theorem for complex valued functions implies that there exists a uniformly
bounded sequence of functions of the form
(n)
p
K
T
X (n)
X
(n)
xj
P (n) (x1 , . . . , xp ) =
ck exp i
kj,k
j=1
k=1
such that
lim P (n) (x1 , . . . , xp ) = F (x1 , . . . , xp )
hence
aF (Yt , . . . , Yt ) = 0
E
1
p
for every continuous bounded F and by the usual approximation argument this extends
to every F bounded Borel measurable with respect to (Yt1 , . . . , Ytp ). As 0 < t1 < . . . <
Ut2 dt
< ,
T 0,
Z
0
Us dYsj Yt
(Us |Ys ) dY j .
E
s
=
0
Proof
it is sufficient to prove that for any t St that
As St is a total set in L1 (, Yt , P),
Z t
Z t
j
j
t
t
(Us |Ys ) dY
E
Us dYs = E
E
.
s
0
79
Stochastic Filtering
80
t = 1 +
0
and hence
Z t
Z t
Z t
j
T
j
t
Us dYs
E
Us dYs = E
1+
is rs dYs
0
0
0
Z t
Z t
j
j
Us dYs + E
is rs Us ds .
=E
0
Here the last term is computed by recalling the definition of the covariation process, in
conjunction with the Kunita Watanabe identity. The first term on the right hand side above
is a martingale (from the boundedness condition on Us ) and hence its expectation vanishes.
Thus using the fact that the lebesgue integral and the expectation E(|Y)
commute
Z t
Z t
j
j
E t
Us dYs = E
is rs Us ds
0
0
Z t
j
E
=E
is rs Us ds Y
0
Z t
j
=E
iE s rs Us Y ds
0
Z t
j
=E
is rs E (Us |Y) ds
0
Z t
Z t
T
j
(Us |Y) dY
=E
is rs dYs
E
s
0
0
Z t
j
= E t
E (Us |Y) dYs .
0
s |Y) = E(U
s |Ys )
But from proposition 18.4, it follows that since Us is Ys measurable E(U
whence
Z t
Z t
j
j
E t
Us dYs = E t
E (Us |Ys ) dYs .
0
As this holds for all t in St , and the latter is a total set, by the earlier remarks, this
establishes the result.
Lemma 18.9.
Let {Ut }t0 be a real valued Ft adapted continuous process such that
Z
0
!
Us2 ds
< ,
T 0,
(14)
Stochastic Filtering
and let Rt be another continuous Ft adapted process such that hR, Y it = 0 for all t. Then
Z t
Us dRs |Yt = 0.
E
0
Proof
As before use the fact that St is a total set, so it suffices to show that for each t St
Z t
E t
Us dRs = 0.
0
E
Us dRs =E
Us dRs + E
is rs dYs
Us dRs
0
0
0
0
Z t
Z t
T
=E
Us dRs + E
is Us rs dhY, Ris
0
0
Z t
=E
Us dRs
0
=0.
The last term vanishes, because by the condition 14, the stochastic integral is a genuine
martingale. The other term vanishes because from the hypotheses hY, Rit = 0.
Remark
In the context of stochastic filtering a natural application of this lemma will be made by
setting Rt = Vt , the stochastic noise process driving the signal process.
We are now in a position to state and prove the Zakai equation. The Zakai equation is important because it is a parabolic stochastic partial differential equation which is
satisfied
by t ,and indeed its solution provides a practical means of computing t () =
E (Xt )Zt |Yt . Indeed the Zakai equation provides a method by which numerical solution of the non-linear filtering problem may be approached by using recursive algorithms
to solve the stochastic differential equation.
Theorem 18.10.
Let A be the infinitesimal generator of the signal process Xt , let the domain of this infinitesimal generator be denoted D(A). Then subject to the usual conditions on f , and
h, the un-normalised conditional distribution on X satisfies the Zakai equation which is
Z t
Z t
t () = 0 () +
s (As )ds +
s (hT )dYs , t 0, D(A).
(15)
0
81
Stochastic Filtering
Proof
We approximate Zt by
Zt =
Zt
,
1 + Zt
4
.
272
As (Xs )ds + Mt ,
d Z tX
d
X
=
(Xt )ij (t, Xt ) dWtj .
i
x
j=1 0 i=1
1 XX X
ik (s, x)kj (s, x)
(x),
As (x) =
fi (s, x) i (x) +
i xj
x
2
x
i=1 j=1
i=1
k=1
82
Stochastic Filtering
we shall do this using the previous expression which on integration gives us
Z th
i
Note that
Z0 (X0 ) Yt
=E
1
1
(X0 ) Y0 = 0 ()
.
1+
1+
Zt (Xt ) Yt
Z t
0 ()
2
T
+E
(Xs )(1 + Zs ) Zs h (s, Xs ) dYs Yt
=
1+
0
Z t
3 2
2
+E
Zs A(Xs ) (Xs )(1 + Zs ) Zs kh(s, Xs )k ds Yt
0
Z t
Zs dMs Yt
+E
0
Applying lemma 18.8 since Zt is bounded allows us to interchange the (stochastic) integrals
and the conditional expectations to give
() Z t
0
2 T
+
E (Xs )(1 + Zs ) Zs h (s, Xs ) Yt dYs
E Zt (Xt ) Yt =
1+
0
Z t h
i
Z A(Xs ) (Xs )(1 + Zs )3 Z 2 kh(s, Xs )k2 Yt ds
+
E
s
s
0
Z t h i
Z Yt dM
+
E
s
s
0
Now we take the 0 limit. Clearly Zt Zt pointwise, and thus by the monotone
convergence theorem since Zt Zt as 0,
(Z As (Xs )|Ys ) kAs kE
Zs |Ys ,
E
s
83
Stochastic Filtering
84
and the right hand side is an L1 bounded quantity. Hence by the dominated convergence
theorem we have
Z t
Z t
E Zs As (Xs )|Ys ds
s (As ) ds a.s.
0
2
1
lim (Xs ) Zs
1 + Zs
kh(s, Xs )k2 = 0.
0
kk
kh(s,
X
)k
s
(17)
Where the final inequality follows by (6). As a consequence of these two results, as the right
hand side of the above inequality is integrable, by the conditional form of the Dominated
Convergence theorem it follows that
2
1
2
(Xs ) Z
E
1 + Zs
kh(s, Xs )k Y 0
s
as 0. By proposition 18.4 the sigma field Y may be replaced by Yt which yields that
as 0
2
1
2
(Xs ) Z
E
1 + Zs
kh(s, Xs )k Ys 0.
s
Now note again that the dominating function (17) is also Lebesgue integrable and bounded
thus a second application of the Dominated convergence theorem yields
t
(Xs )
E
Zs
2
1 + Zs
1
kh(s, Xs )k Ys ds 0,
2
as 0.
(Xs )Zs
1 + Zs
1
Z t
h (s, Xs ) Ys dYs
s (hT (Xs )) dYs .
T
Stochastic Filtering
To this end define
Z
I (t) :=
0
1
T
I(t) :=
s (h (Xs ))dYs =
E Zs h (s, Xs )(Xs ) Yt dYs .
0
Both I(t) and I (t) are continuous bounded square integrable martingales. Consider the
difference
Z t
s
T
Z
(X
)h
(s,
X
)
I (t) I(t) =
E
2
s
s
s Y dYs
0
1 + Zs
Z t
s2 2 + Zs
Z
T
=
E
(X
)h
(s,
X
)
2
s
s Y dYs ,
0
1 + Zs
Brownian motion
Consider the L2 norm of this difference and use the fact that Y is a P
2
Z t
s2 2 + Zs
Z
T
(I (t) I(t))2 = E
E
E
2 (Xs )h (s, Xs ) Y dYs
0
1 + Zs
2
2
Zs 2 + Zs
T
E
2 (Xs )h (s, Xs ) Y
ds
0
1 + Zs
2
kXk2 Y . Thus
By the conditional form of Jensens inequality kE(X|Y)k
E
2
2
s 2 + Zs
Z
T
(X
)h
(s,
X
)
Y
E
2
s
s
1 + Zs
2
2
Zs 2 + Zs
E
2 (Xs )h (s, Xs )
Y
1 + Zs
= E
A dominating function can can readily be found for the quantity inside the conditional
expectation viz
2
2
2
Zs 2 + Zs
Z
1
s
T
2
1+
Zs kh(s, Xs )k2
2 (Xs )h (s, Xs )
kk
1 + Zs
1 + Zs
1 + Zs
4kk2 Zs2 kh(s, Xs )k2
4kkk2 Zs2 1 + kXs k2
85
Stochastic Filtering
Since
Zs Zs + Zs kXs k2
= E Zs + E Zs kXs k2
E
2
2
Z t
Zs 2 + Zs
E
E
(X
)
2 h (s, Xs )|Ys ds 0. a.s.
s
0
1 + Zs
Thus we have shown that
h
i
(I (t) I(t))2 0.
E
The process Yt
Rt
0
86
x(s)y(s)ds,
(19)
then
t
Z
x(t) z(t) +
z(s)y(s) exp
0
y(r)dr ds.
s
Proof
R
t
Multiplying both sides of the inequality (19) by y(t) exp 0 y(s)ds yields
Z t
Z t
Z t
y(s)ds
x(s)y(s)ds y(t) exp
y(s)ds
x(t)y(t) exp
0
Z t
z(t)y(t) exp
y(s)ds .
0
d
dt
Z t
Z t
x(s)y(s)ds exp
y(s)ds
z(t)y(t) exp
y(s)ds .
Z t
Z t
Z
x(s)y(s)ds exp
y(s)ds
z(s)y(s) exp
y(r)dr ds,
or equivalently
Z
Z
x(s)y(s)ds
Z
z(s)y(s) exp
y(r)dr ds.
Combining this with the original equation (19) gives the desired result.
.
[87]
Gronwalls Inequality
88
Corollary 19.1.
If x, y, and z satisfy the conditions for Gronwalls theorem then
Z
sup z(t) exp
sup x(t)
0tT
0tT
y(s)ds .
Proof
From the conclusion of Gronwalls theorem we see that for all 0 t T
!
Z
Z
T
z(s)y(s) exp
0tT
y(r)dr ds,
which yields
T
Z
sup x(t) sup z(t) +
0tT
0tT
z(s)y(s) exp
sup z(t) +
0tT
0tT
Z
sup z(t)
0tT
0tT
y(r)dr ds
s
Z
sup z(t)
exp
T
y(s) exp
Z
sup z(t) exp
0tT
y(r)dr ds,
sup z(t) +
!
y(s)ds .
!
y(s)ds
!
1
N
X
i=1
(i)
bt (Xt )
1 XX
2u
u
(F F T )ij
+
xi
2 i=1 j=1
xi xj
Kalman Filter
90
Z
0
(i)
E (Bs Xs )(i) + bt Yt dt
Z t
(i)
T
(i)
T
+
E Xs (Hs Xs ) |Ys Xs (Hs Xs ) dNs
0
s ds
Hs X
Whence writing the equation for the evolution of the conditional mean in vector form, we
obtain (since RT = R)
t = Bt X
t + bs dt + Rt H T (dYt Ht X
t dt)
dX
t
(20)
(i)
(j)
(k)
(k)
(i)
(j)
It is clear that
n
o
(i) (j)
(i) (j)
dR = d E(Xt Xt |Yt ) d Xt Xt
ij
(22)
Kalman Filter
91
i
(Rt HtT dNt )(i) , (Rt HtT dNt )(j) = Rtil H kl dN k , Rjm Hnm dN n
= Ril H kl Rjm Hnm kn dt
= Ril H kl H km Rjm dt
= (RH T HRT )ij dt
= (RH T HR)ij dt
(24)
where the last equality follows since RT = R. Substituting (24) into (23) and then using
this and (21) in (22) yields the following equation for the evolution of the ijth element of
the covariance matrix
1
ij
T ij
ij
T T ij
T
ij
(Ft Ft ) + (Bt Rt ) + (Rt Bt ) (RH HR)
dRt = dt
2
t(i) Rtjm + X
t(j) Rtjm )Hlm dN l (X
ti Rtjm + X
tj Rtjm )H lm dN l
+ (X
Thus we obtain the final differential equation for the evolution of the conditional covariance
matrix; notice that all of the stochastic terms have cancelled
dRt
= Ft FtT + Bt Rt + Rt BtT Rt HtT Ht Rt
dt
(25)
The two equations (20) and (25) therefore provide a complete description of the best
prediction of the system state Xt , conditional on the observations up to time t. This is the
Kalman-Bucy filter.
It is most frequently encountered in a discrete time version, since in this form it is
practical for use in some of the problems described in the introduction to the stochastic
filtering section of these notes.
21.1. Compensators
Let p(t) be an Ft adapted jump process with bounded jumps, such that p(0) = 0 and
Ep(t) < for all t. Let (t) be a bounded non-negative Ft adapted process with RCLL
sample paths. If
Z
t
p(t)
(s)ds
0
is and Ft martingale then () is the intensity of the jump process p() and the process
Rt
(s)ds is the compensator of the process p().
0
[92]
93
g(x) =
94
0 for x < 1 + ,
1 for x 1 + .
For arbitrary small these functions are always one unit apart in the sup metric. We
attempt to surmount this problem by considering the class T of increasing maps from
[0, T ] to [0, T ]. Now define a metric
dT (f, g) := inf : sup |s (s)| ,
0sT
sup |f (s) g((s))| for some T
0sT
This metric would appear to nicely have solved the problem described above. Unfortunately
it is not complete! This is a major problem because Prohorovs theorem (among others)
requires a complete separable metric space! However the problem can be fixed via a simple
trick; for a time transform T define a norm
(t)
(s)
.
|| := sup log
ts
0sT
Thus we replace the condition |(s) s| by || and we obtain a new metric on DT
which this time is complete and separable;
d0T (f, g)
:= inf : || ,
sup |f (s) g((s))| , for some T
0sT
This is defined on the space of CADLAG functions on the compact time interval [0, T ]
and can be exteneded via another standard trick to D ,
0
d (f, g) :=
0
22. References
Bensoussan, A. (1982). Stochastic control of partially observable systems. CUP.
Bollobas, B. (1990). Linear Analysis. Cambridge.
Dellacherie, C. and Meyer, P.-A. (1975). Probabilites et Potentiel A. Hermann, Paris.
Dellacherie, C. and Meyer, P.-A. (1980). Probabilites et Potentiel B. Hermann, Paris.
Durrett, R. (1996). Probability: Theory and Examples. Duxbury.
Ethier, S. and Kurtz, T. (1986). Markov Processes Characterization and Convergence.
Wiley.
Karatzas, I. and Shreve, S. E. (1987). Brownian Motion and Stochastic Calculus. Springer.
Mandelbaum, A., Massey, W. A., and Reiman, M. I. (1998). Strong approximations for
markovian service networks. Queueing Systems Theory and Applications, 30:149
201.
Musiela, M. and Rutkowski, M. (2005).
Springer, 2nd edition.
Pardoux, E. (1989). Filtrage non lineaire et equations aux derivees partielles stochastiques
associees, pages 67163. Number 1464 in Lecture notes in Mathematics. Springer
Verlag.
Protter, P. (1990). Stochastic Integration and Differential Equations. Springer.
Rogers, L. C. G. and Williams, D. (1994). Diffusions, Markov Processes and Martingales:
Volume One: Foundations. Wiley.
Rogers, L. C. G. and Williams, D. (2000). Diffusions, Markov Processes and Martingales:
Volume Two: It
o Calculus. Cambridge University Press, 2nd edition.
[95]