T
About this book:
The book could be described as Stochastic Integration without tears or fear or even as
Stochastic Integration made easy.
F
This book can be used as a 2 semester graduate level course on Stochastic Calculus.
The background required is a course on measure theoretic probability.
A
The book begins with conditional expectation and martingales and basic results on
martingales are included with proofs (in discrete time as well as in continuous time). Then
a chapter on Brownian motion and Ito integration w.r.t. Brownian motion follows, which
R
includes stochastic di↵erential equations. These three chapters over 80 pages give a soft
landing to a reader to the more complex results that follow. The first three chapters form
the Introductory material.
D
Taking a cue from the Ito integral, a Stochastic Integrator is defined and its properties
as well as the properties of the integral are defined at one go. In most treatments, we
start by defining the integral for a square integrable martingales and where integrands
themselves are in suitable Hilbert space. Then over several stages the integral is extended,
and at each step one has to reaffirm its properties. Here this is avoided. Just using the
definition, various results including quadratic variation, Ito formula, Emery topology are
defined and studied.
We then define quadratic variation of a square integrable martingale and using this show
that these are stochastic integrators. Using an inequality by Burkholder, we show that all
martingales and local martingales are stochastic integrators and thus semimartingales are
2
stochastic integrators. We then show that stochastic integrators are semimartingales and
obtain various results such as a description of the class of integrands for the stochastic
integral. We complete the chapter by giving a proof of the Bichteler-Dellacherie-Meyer-
Mokobodzky Theorem.
These two chapters (of about 100 pages) form the basic material. Most results used in
these two chapter are known, but we have put them in a order to make the presentation
easy flowing. We have avoided invoking results from functional analysis but rather included
the required steps. Thus instead of saying that the integral is a continuous linear functional
on a dense subset of a Banach space and hence can be extended to the Banach space, we
explicitly construct the extension!
Next we introduce Pathwise formulae for the stochastic integral. These as yet have not
found a place in any textbook on stochastic integration. We briefly specialise to continuous
T
semimartingales and obtain growth estimates and study solution of a stochastic di↵erential
equation (SDE) using the technique of random time change. We also prove a pathwise
formulae for the solution of an SDE driven by continuous semimartingales.
F
Then we move onto a study of predictable increasing process and introduce predictable
stopping times etc and prove the Doob Meyer decomposition theorem
We conclude by proving the Metivier Pellaumail inequality and using it introduce a
A
notion of dominating process of a semimartingale. We then obtain existence, uniqueness of
solutions to an SDE driven by a general semimartingale and also obtain a pathwise formula
by showing that modified Eular-Peano approximations converge almost surely.
R
D
Contents
T
1.3 Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 Martingales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.6 Stop Times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.7
1.8
F
Doob’s Maximal Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . .
Martingale Convergence Theorem . . . . . . . . . . . . . . . . . . . . . . . .
20
22
A
1.9 Square Integrable Martingales . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.10 Burkholder-Davis-Gundy inequality . . . . . . . . . . . . . . . . . . . . . . . 34
3
4 CONTENTS
4 Stochastic Integration 91
4.1 The Predictable -Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2 Stochastic Integrators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.3 Properties of the Stochastic Integral . . . . . . . . . . . . . . . . . . . . . . 101
4.4 Locally Bounded Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.5 Approximation by Riemann Sums . . . . . . . . . . . . . . . . . . . . . . . 115
4.6 Quadratic Variation of Stochastic Integrators . . . . . . . . . . . . . . . . . 119
4.7 Quadratic Variation of the Stochastic Integral . . . . . . . . . . . . . . . . . 126
4.8 Ito’s Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.9 The Emery Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.10 Extension Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
T
5 Semimartingales 159
5.1 Notations and terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.2 The Quadratic Variation Map . . . . . . . . . . . . . . . . . . . . . . . . . . 162
5.3
5.4
5.5
F
Quadratic Variation of a Square Integrable Martingale . . . . . . . . . . . . 165
Square Integrable Martingales are Stochastic Integrators . . . . . . . . . . . 173
Semimartingales are Stochastic Integrators . . . . . . . . . . . . . . . . . . . 177
A
5.6 Stochastic Integrators are Semimartingales . . . . . . . . . . . . . . . . . . . 180
5.7 The Class L(X) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
5.8 The Bichteler-Dellacherie-Meyer-Mokobodzky Theorem . . . . . . . . . . . 198
R
T
9.4 Sigma-Martingales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
9.5 Auxiliary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
T
F
A
R
D
Chapter 1
In this chapter, we will discuss martingales indexed by integers (mostly positive integers)
T
and obtain basic inequalities on martingales and other results which are the basis of most of
the developments in later chapters on stochastic integration. We will begin with a discus-
sion on conditional expectation and then on filtration two notions central to martingales.
1.1 Notations
F
A
For an integer d 1, Rd denotes the d-dimensional Euclidean space and B(Rd ) will denote
the Borel -field on Rd . Further, C(Rd ) and Cb (Rd ) will denote the classes of continuous
functions and bounded continuous functions on Rd respectively. When d = 1, we will write
R
R in place of R1 .
(⌦, F, P) will denote a generic probability space and B(⌦, F) will denote the class of
real-valued bounded F-measurable functions.
D
For a collection A ✓ F, (A) will denote the smallest -field which contains A and
for a collection G ✓ B(⌦, F), (G) will likewise denote the smallest -field with respect to
which each function in G is measurable.
It is well known and easy to prove that
7
8 Chapter 1. Discrete Parameter Martingales
T
Let X and Y be random variables. Suppose we are going to observe Y and are required to
make a guess for the value of X. Of course, we would like to be as close to X as possible.
Suppose the penalty function is square of the error. Thus we wish to minimize
where a is the guess or the estimate. For this to be meaningful, we should assume E[X 2 ] <
A
1. The value of a that minimizes (1.2.1) is the mean µ = E[X]. On the other hand, if we
are allowed to use observations Y while making the guess, then our estimate could be a
function of Y . Thus we should choose the function g such that
R
takes the minimum possible value. It can be shown that there exists a function g (Borel
measurable function from R to R) such that
D
for all (Borel measurable) functions f . Further, if g1 , g2 are two functions satisfying (1.2.2),
then g1 (Y ) = g2 (Y ) almost surely P. Indeed, A = L2 (⌦, F, P) - the space of all square
integrable random variables on (⌦, F, P) with inner product hX, Y i = E[XY ] giving rise
p
to the norm kZk = E[Z 2 ] is a Hilbert space and
T
E[(X g(Y ))f (Y )] = 0 8f bounded measurable. (1.2.3)
Indeed, (1.2.3) characterizes g(Y ). Also, (1.2.3) is meaningful even when X is not square
F
integrable but only E[|X| ] < 1. With a little work we can show that given X with
E[|X| ] < 1, there exists a unique g(Y ) such that (1.2.3) is valid. To see this, suffices to
consider X 0, and then taking Xn = min(X, n), gn such that E[Xn | Y ] = gn (Y ), it
A
follows that gn (Y ) gn+1 (Y ) a.s. and hence defining g̃(x) = lim sup gn (x), g(x) = g̃(x) if
g̃(x) < 1 and g(x) = 0 otherwise, one can show that
E[(X g(Y ))f (Y )] = 0 8f bounded measurable.
R
It is easy to see that in (1.2.3) it suffices to take f to be {0, 1}-valued, i.e. indicator function
of a Borel set. We are thus led to the following definition: For random variables X, Y with
E[|X| ] < 1,
D
E[X | Y ] = g(Y )
where g satisfies
E[(X g(Y1 , . . . Ym ))1B (Y1 , . . . Ym )] = 0, 8B 2 B(Rm ).
10 Chapter 1. Discrete Parameter Martingales
Also if we were to observe an infinite sequence, we have to proceed similarly, with g being
a Borel function on R1 . Of course, the random variables could be taking values in Rd .
In each case we will have to write down properties and proofs thereof and keep doing the
same as the class of observable random variables changes.
Instead, here is a unified way. Let (⌦, F, P) be a probability space and Y be a random
variable on ⌦. The smallest -field (Y ) with respect to which Y is measurable (also called
the -field generated by Y ) is given by
(Y ) = {A 2 F : A = {Y 2 B}, B 2 B(R)}.
A random variable Z can be written as Z = g(Y ) for a measurable function g if and only
if Z is measurable with respect to (Y ). Likewise, for random variables Y1 , . . . Ym , the
-field (Y1 , . . . Ym ) generated by Y1 , . . . Ym is the smallest -field with respect to which
Y1 , . . . Ym are measurable and is given by
T
(Y1 , . . . Ym ) = {A 2 F : A = {(Y1 , . . . Ym ) 2 B}, B 2 B(Rm )}.
Similar statement is true even for an infinite sequence of random variables. In view of
F
these observations, one can define conditional expectation given a -field as follows. It
should be remembered that one mostly uses it when the -field in question is generated by
a collection of observable random variables.
A
Definition 1.1. Let X be a random variable on (⌦, F, P) with E[|X| ] < 1 and let G be
a sub- -field of F. Then the conditional expectation of X given G is defined to be the G
measurable random variable Z such that
R
Existence of Z can be proven on the same lines as given above, first for the case when
D
X is square integrable and then for general X. Also, Z is uniquely determined up to null
sets if Z and Z 0 are G measurable and both satisfy (1.2.5), then P(Z = Z 0 ) = 1. Some
properties of conditional expectation are given in the following two propositions.
Throughout the book, we adopt a convention that all statements involving random vari-
ables are to be interpreted in almost sure sense- i.e. X = Y means X = Y a.s., X Y
means X Y a.s. and so on.
Proposition 1.3. Let X be a random variable with E[X 2 ] < 1 and H be a sub -field.
Then for all H measurable square integrable random variables U
T
E[(X E[X | H])2 ] E[(X U )2 ].
1.3 Independence
F
Two events A, B in a probability space (⌦, F, P), i.e. A, B 2 F are said to be independent
if
A
P(A \ B) = P(A)P(B).
For j = 1, 2, . . . m, let Xj be an Rd -valued random variable on a probability space (⌦, F, P).
Then X1 , X2 , . . . Xm are said to be independent if for all Aj 2 B(Rd ), 1 j m
R
m
Y
P(Xj 2 Aj , 1 j m) = P(Xj 2 Aj ).
j=1
(i) X, Y are independent if and only if for all bounded Borel measurable functions f, g on R,
one has
E[f (X)g(Y )] = E[f (X)]E[g(Y )]. (1.3.1)
12 Chapter 1. Discrete Parameter Martingales
(ii) X, Y are independent if and only if for all bounded Borel measurable functions f, on R,
one has
E[f (X) | (Y )] = E[f (X)]. (1.3.2)
Exercise 1.5. Let U be an Rd -valued random variable and let G be a -field. Show that U
is independent of G if and only if for all 2 Rd
1.4 Filtration
T
Suppose Xn denotes a signal being transmitted at time n over a noisy channel (such as
voice over telephone lines) and let Nn denote the noise at time n and Yn denote the noise
F
corrupted signal that is observed. Under the assumption of additive noise, we get
Yn = Xn + Nn , n 0.
A
Now the interest typically is in estimating the signal Xn at time n. Since the noise as well
the true signal are not observed, we must require that the estimate X̂n of the signal at
time n be a function of only observations upto time n, i.e. X̂n must only be a function of
R
{Yk : 0 k n}, or X̂n is measurable with respect to the -field Gn = {Yk : 0 k n}.
A sequence of random variables X = {Xn } will also be referred to as a process. Usually, the
index n is interpreted as time. This is the framework for filtering theory. See Kallianpur
D
for each n, ⇡n is a (measurable) function of {Sk : 0 k < n}. In particular, the strategy
cannot be a function of the unobserved processes A, B, C.
Thus it is useful to define for n 0, Gn to be the -field generated by all the random
variables observable before time n namely, S0 , S1 , S2 , . . . , Sn 1 and then require any action
to be taken at time n (say investment decision) should be measurable with respect to Gn .
These observations lead to the following definitions
Definition 1.7. A stochastic process X i.e. a sequence X={Xn } of random variables is said
to be adapted to a filtration (F⇧ ) if for all n 0, Xn is Fn measurable.
We will assume that the underlying probability is complete i.e. N 2 F, P(N ) = 0 and
T
N1 ✓ N implies N1 2 F and that F0 contains all sets N 2 F with P(N ) = 0. We will
refer to a stochastic process as a process. Let N be the class of all null sets (sets with
P(N ) = 0), and for a process Z, possibly vector-valued, let
and
F
FnZ = (Zk : 0 k n)
FenZ = (N [ FnZ ).
(1.4.1)
(1.4.2)
A
While it is not required in the definition, in most situations, the filtration (F⇧ ) under
consideration would be chosen to be (for a suitable process Z)
(F⇧Z ) = {FnZ : n 0}
R
or
(Fe⇧Z ) = {FenZ : n 0}.
D
1.5 Martingales
In this section, we will fix a probability space (⌦, F, P) and a filtration (F⇧ ). We will only
be considering R-valued processes in this section.
When there are more than one filtration in consideration, we will call it a (F⇧ )-martingale
T
or martingale w.r.t. (F⇧ ). Alternatively, we will say that {(Mn , Fn ) : n 1} is a martingale.
It is easy to see that for a martingale M , for any m < n, one has
F
EP [Mn | Fm ] = Mm
and similar statement is also true for submartingales. Indeed, one can define martingales
and submartingales indexed by an arbitrary partially ordered set. We do not discuss these
A
in this book.
If M is a martingale and is a convex function on R, then Jensen’s inequality implies
that the process X = {Xn } defined by Xn = (Mn ) is a submartingale if Xn is integrable
R
submartingale.
When we are having only one filtration under consideration, we will drop reference to
it and simply say M is a martingale. It is easy to see also that sum of two martingales with
respect to the same underlying filtration is also a martingale. We note here an important
property of martingales that would be used later.
Proof. Note that for any X m converging to X in L1 (P), for any -field G, using (i), (ii),
(iii) and (iv) in Proposition 1.2, one has
Mnm = E[Mn+1
m
| Fn ] ! E[Mn+1 | Fn ] in L1 (P).
The following decomposition result, called the Doob decomposition, is simple to prove
T
but its analogue in continuous time was the beginning of the theory of stochastic integra-
tion.
An =
Xn
F
Theorem 1.11. Let X be a submartingale. Let A = {An } be defined by A0 = 0 and for
n 1,
E[(Xk Xk 1 ) | Fk 1 ].
A
k=1
Mn = X n An
P(An = Bn 8n 1) = 1.
Mn Mn 1 = Xn Xn 1 E[Xn Xn 1 | Fn 1 ]
= E[(Bn Bn 1 ) | Fn 1 ]
= Bn Bn 1.
Now B0 = 0 implies
n
X
Bn = E[(Xk Xk 1) | Fk 1]
k=1
completing the proof.
The uniqueness in the result above depends strongly on the assumption that Bn is Fn 1
measurable. The process A is called the compensator of the submartingale X.
Let M = {Mn } be a martingale. The sequence D defined by Dn = Mn Mn 1 , for
T
n 1 and D0 = M0 is called a martingale di↵erence sequence and satisfies
E[Dn | Fn 1] = 0 8n 1. (1.5.1)
F
Definition 1.12. A sequence of random variables U = {Un } is said to be a predictable if for
all n 1, Un is Fn 1 measurable and U0 is F0 measurable.
Since a gambler can use the outcomes of the games that have been played earlier, Un
the amount she bets on nth game can be random instead of a fixed number but has to be
predictable with respect to the underlying filtration since the gambler has to decide how
D
much to bet before the nth game is played. If the game is fair, i.e. E[Dn | Fn 1 ] = 0, then
the partial sums Mn = D0 + . . . + Dn is a martingale and the total reward Rn at time n
P
is then given by Rn = nk=0 Uk Dk . One can see that it is also a martingale, if say Un is
bounded. This leads to the following definition.
The following result gives conditions under which the transformed sequence is a mar-
tingale.
E[|Mn Un | ] = E[E[|Mn Un | | Fn 1 ]]
E[|E[Mn Un | Fn 1 ]| ]
(1.5.4)
T
= E[|Un E[Mn | Fn 1 ]| ]
= E[|Un Mn 1| ]
where we have used properties of conditional expectation and the fact that Un is Fn 1
F
measurable and that M is a martingale. Thus, (1.5.3) implies E[|Un Mn 1 | ] < 1. This is
needed to justify splitting the expression in the next step.
A
E[Un (Mn Mn 1 ) | Fn 1 ] = E[Un Mn | Fn 1] E[Un Mn 1 | Fn 1 ]
= Un E[Mn | Fn 1] U n Mn 1
(1.5.5)
= U n Mn 1 U n Mn 1
R
= 0.
This implies Cn = Un (Mn Mn 1) is a martingale di↵erence sequence and thus Z defined
by (1.5.2) is a martingale.
D
Definition 1.17. A stop time ⌧ is a function from ⌦ into {0, 1, 2, . . . , } [ {1} such that for
all n 0,
{⌧ = n} 2 Fn , 8n < 1.
T
Definition 1.18. Let ⌧ be a stop time and X be an adapted process. The stopped random
variable X⌧ is defined by
X1
X⌧ (!) = Xn (!)1{⌧ =n} .
n=0
F
Note that by definition, X⌧ = X⌧ 1{⌧ <1} . The following results connecting martingales
and submartingales and stop times (and their counterparts in continuous time) play a very
A
important role in the theory of stochastic processes.
⇠=⌧_ and ⌘ = ⌧ ^ .
Show that ⇠ and ⌘ are also stop times. Here and in the rest of this book, a _ b = max(a, b)
D
Exercise 1.20. Let ⌧ be a random variable taking values in {0, 1, 2, . . .} and for n 0 let
Fn = (⌧ ^ n). If ⌘ is a stop time for this filtration then show that ⌘ = (⌘ _ ⌧ ) ^ r for some
r 2 {0, 1, 2 . . . , 1}
Theorem 1.21. Let M = {Mn } be a submartingale and ⌧ be a stop time. Then the process
N = {Nn } defined by
Nn = Mn^⌧ ,
Proof. Without loss of generality, we assume that M0 = 0. Let Un = 1{⌧ <n} and Vn =
1{⌧ n} . Since
{Un = 1} = [nk=01 {⌧ = k}
Theorem 1.22. Let M = {Mn } be a submartingale and , ⌧ be stop times such that ⌧.
T
Let R = {Rn }, S = {Sn } be defined as follows: R0 = S0 = 0 and for n 1
S n = Mn Mn^⌧ ,
Rn = Mn^⌧
F Mn^ .
Then R, S are submartingales. Further, if M is a martingale then so are S, R.
k=1
Xn
Rn = Vk (Mk Mk 1 ).
k=1
D
Corollary 1.23. Let M = {Mn } be a submartingale and , ⌧ be stop times such that
⌧ . Then for all n 1
E[Mn^ ] E[Mn^⌧ ].
It is easy to see that for a martingale M , E[Mn ] = E[M0 ] for all n 1. Of course, this
property does not characterize martingales. However, we do have the following result.
20 Chapter 1. Discrete Parameter Martingales
Theorem 1.24. Let M = {Mn } be an adapted process such that E[|Mn | ] < 1 for all
n 0. Then M is a martingale if and only if for all bounded stop times ⌧ ,
E[M⌧ ^n ] = E[M0 ].
T
stop time. Using E[M⌧ ] = E[M0 ], it follows that
F
Likewise, using the (1.6.1) for ⌧ = n, we get E[Mn ] = E[M0 ], or equivalently
Now (1.6.2) follows from (1.6.3) and (1.6.4) completing the proof.
(1.6.4)
A
1.7 Doob’s Maximal Inequality
R
We will now derive an inequality for martingales known as Doob’s maximal inequality. It
plays a major role in stochastic calculus as we will see later.
D
Further, for 1 < p < 1, there exists a universal constant Cp depending only on p such that
Proof. Under the assumptions, the process N defined by Nk = |Mk | is a positive sub-
martingale. Let
⌧ = inf{k : Nk > }.
1.7. Doob’s Maximal Inequality 21
Here, and in what follows, we take infimum of the empty set as 1. Then ⌧ is a stop time
and further, {max0kn Nk } ✓ {⌧ > n}. By Theorem 1.22, the process S defined by
Sk = Nk Nk^⌧ is a submartingale. Clearly S0 = 0 and hence E[Sn ] 0. Note that ⌧ n
implies Sn = 0. Thus, Sn = Sn 1{max0kn Nk > } . Hence
This yields
T
N⌧ ^n 1{max0kn Nk > } 1{max0kn Nk > } , (1.7.4)
F
combining (1.7.3) and (1.7.4) we conclude that (1.7.1) is valid.
The conclusion (1.7.2) follows from (1.7.1). To see this, fix 0 < ↵ < 1 and let us write
A
f = (max0kn |Mk |) ^ ↵ and g = |Mn |. Then the inequality (1.7.1) can be rewritten as
1
P(f > ) E[g 1{f > } ]. (1.7.5)
R
Now consider the product space (⌦, F, P) ⌦ ((0, ↵], B((0, ↵]), µ) where µ is the Lebesgue
measure on (0, ↵]. Consider the function h : ⌦ ⇥ (0, ↵] 7! (0, 1) defined by
D
h(!, t) = ptp 1
1{t<f (!)} , (!, t) 2 ⌦ ⇥ (0, ↵].
On the other hand, using Fubini’s theorem in the first and fourth step below and using the
22 Chapter 1. Discrete Parameter Martingales
T
p
= E[gf (p 1) ].
(p 1)
The first step in (1.7.8) below follows from the relations (1.7.6)-(1.7.7) and the next one
E[f p ]
(p
p
p
1)
F
from Holder’s inequality, where q = (p p 1) so that p1 + 1q = 1:
E[gf (p 1)
]
A
1 1
= (E[g p ]) p (E[f (p 1)q
]) q (1.7.8)
(p 1)
p 1 1
= (E[g p ]) p (E[f p ]) q .
(p 1)
R
Since (1.7.9) holds for all ↵, taking limit as ↵ " 1 (via integers) and using monotone
convergence theorem, we get that (1.7.2) is true.
a sequence of real numbers and ↵ < be real numbers. Let sk , tk be defined (inductively)
as follows: s0 = 0, t0 = 0, and for k = 1, 2, . . . m
Recall our convention- infimum of an empty set is taken to be 1. It is easy to see that if
tk = j < 1, then
0 s1 < t1 < . . . < sk < tk = j < 1 (1.8.1)
and for i j
asi ↵, ati . (1.8.2)
We define
Um ({aj : 1 j m}, ↵, ) = max{k : tk m}.
T
sequence {aj : 1 j m}. The equations (1.8.1) and (1.8.2) together also imply
tk > (2k 1) and also writing cj = max(↵, aj ) that
m
X
j=1
(ctj ^m csj ^m )
F
( ↵)Um ({aj }, ↵, ).
↵ to
A
the sum, one term could be non-negative and rest of the terms are zero.
m!1
Proof. If lim inf n!1 an < ↵ < < lim supn!1 an then
It follows that if (1.8.5) holds, then limn!1 an exists in R̄ = R [ { 1, 1}. The next
result gives an estimate on expected value of
Um ({Xj : 1 j m}, ↵, )
for a submartingale X.
24 Chapter 1. Discrete Parameter Martingales
k = {n ⌧k 1 : Xn ↵}, ⌧k = {n k : Xn }.
Then for each k, k and ⌧k are stop times. Writing Yk = (Xk ↵)+ , we see that Y is also
sub- martingale and as noted in (1.8.3)
m
X
(Y⌧j ^m Y j ^m ) ( ↵)Um ({Xj : 1 j m}, ↵, ). (1.8.7)
j=1
T
have
Xm m
X1
Ym Y 1 ^m = (Y⌧j ^m Y j ^m ) + (Y j+1 ^m Y⌧j ^m ). (1.8.8)
j=1 j=1
Since
have
j+1 ^m
E[Y j+1 ^m
F
⌧j ^ m are stop times and Y is a submartingale, using Corollary 1.23 we
Y⌧j ^m ] 0. (1.8.9)
A
Putting together (1.8.7), (1.8.8) and (1.8.9) we get
Since Yk 0 we get
E[Ym ]
E[Um ({Xj : 1 j m}, ↵, )] . (1.8.11)
↵
↵)+ |Xm | + |↵|.
D
We recall the notion of uniform integrability of a class of random variables and related
results.
A collection {Z↵ : ↵ 2 } of random variables is said to be uniformly integrable if
Exercise 1.28. If {Xn : n 1} is uniformly integrable and |Yn | |Xn | for each n then
show that {Yn : n 1} is also uniformly integrable.
1.8. Martingale Convergence Theorem 25
Exercise 1.31. Show that {Z↵ : ↵ 2 } satisfies (i), (ii) in Exercise 1.30 if and only if
{Z↵ : ↵ 2 } is uniformly integrable.
T
Exercise 1.32. Suppose {X↵ : ↵ 2 } and {Y↵ : ↵ 2 } are uniformly integrable and for
each ↵ 2 , let Z↵ = X↵ + Y↵ . Show that {Z↵ : ↵ 2 } is also uniformly integrable.
Exercise 1.34. Prove Lemma 1.33. For (i), use Exercise 1.31.
We are now in a position to prove the basic martingale convergence theorem.
D
T
⇠ ⇤ (!) = lim inf Xn (!) = lim sup Xn (!).
n!1 n!1
Defining ⇠ ⇤ (!) = 0 for ! 2 N ⇤, by Fatou’s lemma we get
F
E[|⇠ ⇤ | ] < 1
so that P(⇠ ⇤ 2 R) = 1. Then defining ⇠(!) = ⇠ ⇤ (!)1{|⇠⇤ (!)|<1} we get
Xn ! ⇠ a.e.
A
If {Xn } is uniformly integrable, the a.e. convergence also implies L1 (P) convergence.
If {Xn } is a martingale or a positive submartingale, then by Doob’s maximal inequality,
R
and hence by monotone convergence theorem Z = (supk 1 |Xk |)p is integrable. Now the
convergence in Lp (P) follows from the dominated convergence theorem.
D
Theorem 1.36. Let {Fm } be an increasing family of sub -fields of F and let F1 =
([1
m=1 Fm ). Let Z 2 L (P) and for 1 n < 1, let
1
Zn = E[Z | Fn ].
Then Zn ! Z ⇤ = E[Z | F1 ] in L1 (P) and a.s..
Proof. From the definition it is clear that Z is a martingale and it is uniformly integrable
by Lemma 1.33. Thus Zn converges in L1 (P) and a.s. to say Z ⇤ . To complete the proof,
we will show that
Z ⇤ = E[Z | F1 ]. (1.8.16)
1.8. Martingale Convergence Theorem 27
Since [1n=1 Fn is a field that generates the -field F1 , the monotone class theorem implies
that (1.8.18) holds for all A 2 F1 and hence (1.8.16) holds.
The previous result has an analogue when the -fields are decreasing. Usually one
introduces a reverse martingale (martingale indexed by negative integers) to prove this
T
result. We avoid it by incorporating the same in the proof.
Theorem 1.37. Let {Gm } be a decreasing family of sub -fields of F, i.e. Gm ◆ Gm+1 for
all m 1. Let G1 = \1
m=1 Gm . Let Z 2 L (P) and for 1 n 1, let
1
E[|Z0 | + |↵| ]
E[Um ({Zm j : 1 j m}, ↵, )] .
↵
and hence proceeding as in Theorem 1.35, we can conclude that there exists N ⇤ ✓ ⌦ with
P(N ⇤ ) = 0 and for ! 62 N ⇤ one has
D
By Jensen’s inequality |Zn | E[|Z0 | | Gn ]. It follows that {Zn } is uniformly integrable (by
Lemma 1.33) and thus Zn converge a.e. and in L1 (P) to a real-valued random variable Z ⇤ .
Since Zn for n m is Gm measurable, it follows that Z ⇤ is Gm -measurable for every m and
hence Z ⇤ is G1 -measurable.
28 Chapter 1. Discrete Parameter Martingales
Thus E[Z0 | G1 ] = Z ⇤ .
Exercise 1.38. Let ⌦ = [0, 1], F be the Borel -field on [0, 1] and let P denote the Lebesgue
measure on (⌦, F). Let Q be another probability measure on (⌦, F) absolutely continuous
with respect to P, i.e. satisfying : P(A) = 0 implies Q(A) = 0. For n 1 let
2 n
X
n
X (!) = 2n Q((2 n
(j 1), 2 n
j])1(2 (!).
T
n (j 1),2 n j]
j=1
Let Fn be the field generated by the intervals {(2 n (j 1), 2 n j] : 1 j 2n }. Show that
(Xn , Fn ) is a uniformly integrable martingale and thus converges to a random variable ⇠ in
L1 (P). Show that ⇠ satisfies Z
A
F
⇠ dP = Q(A) 8A 2 F. (1.8.19)
A
Hint: To show uniform integrability of {Xn : n 1} use Exercise 1.31 along with the fact that
absolute continuity of Q w.r.t. P implies that for all ✏ > 0, 9 > 0 such that P(A) < implies
Q(A) < ✏.
R
Exercise 1.39. Let F be a countably generated -field on a set ⌦, i.e. there exists a
sequence of sets {Bn : n 1} such that F = ({Bn : n 1}). Let P, Q be probability
measures on (⌦, F) such that Q absolutely continuous with respect to P. For n 1, let
D
Fn = (C1 , C2 . . . , Ckn ).
The random variable X in (1.8.19) and (1.8.20) is called the Radon-Nikodym derivative
of Q w.r.t. P.
T
A
are called square integrable martingales and they play a special role in the theory of stochas-
(1.9.1)
A
tic integration as we will see later. Let us note that for p = 2, the constant Cp appearing
in (1.7.2) equals 4. Thus for a square integrable martingale M , we have
0kn
for n 1,
Xn
An = E[(Xk Xk 1 ) | Fk 1 ].
k=1
Thus hM, M i is the unique predictable increasing process with hM, M i0 = 0 such that
Mn2 hM, M in is a martingale. Let us also define another increasing process [M, M ]
associated with a martingale M : [M, M ]0 = 0 and
n
X
2
[M, M ]n = (Mk Mk 1) . (1.9.5)
k=1
The process [M, M ] is called the quadratic variation of M and the process hM, M i is called
the predictable quadratic variation of M . It can be easily checked that
n
X
Mn2 [M, M ]n = 2 Mk 1 (Mk Mk 1)
k=1
and hence using Theorem 1.14 it follows that Mn2 [M, M ]n is also a martingale. If M0 = 0,
then it follows that
T
E[Mn2 ] = E[hM, M in ] = E[[M, M ]n ]. (1.9.6)
We have already seen that if U is a bounded predictable sequence then the transform
Z defined by (1.5.2)
Zn =
Xn
Uk (Mk Mk 1 )
k=1
F
A
is itself a martingale. The next result includes an estimate on the L2 (P) norm of Zn .
As a consequence
n
X N
X
2
E[ max | Uk (Mk Mk 1 )| ] 4E[ Uk2 ([M, M ]k [M, M ]k 1 )]. (1.9.9)
1nN
k=1 k=1
1.9. Square Integrable Martingales 31
Proof. Since U is bounded and predictable Zn is square integrable for each n. That Z is
square integrable martingale follows from Theorem 1.14. Since
2
(Zk Zk 1) = Uk2 (Mk Mk 1)
2
(1.9.10)
the relation (1.9.8) follows from the definition of the quadratic variation. Further, taking
conditional expectation given Fk 1 in (1.9.10) and using that Uk is Fk 1 measurable, one
concludes
2
E[(Zk Zk 1) | Fk 1] = Uk2 E[(Mk Mk 1)
2
| Fk 1]
and this proves (1.9.7). Now (1.9.9) follows from (1.9.8) and the Doob’s maximal inequality,
T
Theorem 1.25.
E[ max |
1nN
n
X
k=1
Uk (Mk
F
Mk 1 )|
2
] 4E[ [M, M ]N ]. (1.9.12)
A
Further,
[Z, Z]n [M, M ]n , n 1. (1.9.13)
The inequality (1.9.12) plays a very important role in the theory of stochastic integra-
R
tion as later chapters will reveal. We will obtain another estimate due to Burkholder [6]
on martingale transform which is valid even when the martingale is not square integrable.
Proof. Let
⌧ = inf{n : 0 n N, |Mn | } ^ (N + 1)
4
so that ⌧ is a stop time. Since {⌧ N } = {max0nN |Mn | 4 }, using Doob’s maximal
inequality (Theorem 1.25) for the positive submartingale {|Mk | : 0 k N }, we have
4
P(⌧ N ) E(|MN |). (1.9.15)
32 Chapter 1. Discrete Parameter Martingales
We will now prove that for any predictable sequence of random variables {Vn } bounded
by 1, we have
N
X
|E[ Vk (⇠k ⇠k 1 )]| E[|MN | ]. (1.9.17)
T
k=1
PN
Let us define M̃k = Mk^⌧ for k 0. Since M̃ is a martingale, E[ k=1 Vk (M̃k M̃k 1 )] = 0.
Writing Yn = ⇠n M̃n , it follows that
E[
N
X
k=1
Vk (⇠k ⇠k
F
1 )] = E[
N
X
k=1
Vk (Yk Yk 1 )]. (1.9.18)
A
Since Yn = Mn 1{n<⌧ } Mn^⌧ = Mn^⌧ 1{⌧ n} = M⌧ 1{⌧ n} , it follows that Yk Yk 1 =
M⌧ 1{⌧ (k 1)} M⌧ 1{⌧ k} and thus Yk Yk 1 = M⌧ 1{⌧ =k} . Using (1.9.18), we get
N
X N
X
|E[ Vk (⇠k ⇠k 1 )]| =|E[ Vk (Yk Yk 1 )]|
R
k=1 k=1
N
X
E[ |Yk Yk 1| ]
D
k=1
N
(1.9.19)
X
E[ |M⌧ |1{⌧ =k} ]
k=1
E[|M⌧ |1{⌧ N } ]
E[|MN | ]
where for the last step we have used that |Mn | is a submartingale. We will decompose
{⇠n : 0 n N } into a martingale {Rn : 0 n N } and a predictable process
{Bn : 0 n N } as follows: B0 = 0, R0 = ⇠0 and for 0 n N
Bn = Bn 1 + (E[⇠n | Fn 1] ⇠n 1)
1.9. Square Integrable Martingales 33
Rn = ⇠ n Bn .
Bn Bn 1 = E[⇠n | Fn 1] ⇠n 1
and hence
Rn Rn 1 = (⇠n ⇠n 1) E[⇠n ⇠n 1 | Fn 1 ].
As a consequence
2 2
E[(Rn Rn 1) ] E[(⇠n ⇠n 1) ]. (1.9.20)
For x 2 R, let sgn(x) = 1 for x 0 and sgn(x) = 1 for x < 0, so that |x| =
T
sgn(x)x. Now taking Vk = sgn(Bk Bk 1 ) and noting that Vk is Gk 1 measurable, we have
E[Vk (Bk Bk 1 )] = E[Vk (⇠k ⇠k 1 )] (since ⇠ = R + B and R is a martingale) and hence
E[ max |
1nN
n
X
k=1
Uk (Bk Bk F
1 )| ] E[
N
X
k=1
XN
|(Bk Bk 1 )| ]
A
= E[ Vk (Bk Bk 1 )]
k=1 (1.9.21)
XN
= E[ Vk (⇠k ⇠k 1 )]
R
k=1
E[ |MN | ]
n
X 2
P( max | Uk (Bk Bk 1 )| ) E[ |MN | ]. (1.9.22)
1nN 2
k=1
Pn
Since R is a martingale and U is predictable, Xn = k=1 Uk (Rk Rk 1) is a martingale
and hence Xn2 is a submartingale and hence
n
X n
X 2
2
P( max | Uk (Rk Rk 1 )| ) =P( max | Uk (Rk Rk 1 )| )
1nN 2 1nN 4
k=1 k=1 (1.9.23)
4 2
2
E[XN ].
34 Chapter 1. Discrete Parameter Martingales
T
over k, one gets
X N N
X
(⇠k ⇠k 1 )2 =⇠N
2
⇠02 2 ⇠k 1 (⇠k ⇠k 1)
k=1 k=1
where Wk = 4 ⇠k
4
F
|MN | + 2
4
N
X
k=1
Wk (⇠k ⇠k 1)
(1.9.26)
A
1. Note that |Wk | 1 and Wk is Gk 1 measurable. Using (1.9.19) and
(1.9.26), we get
N
X
2 3
E[ (⇠k ⇠k 1) ] E[|MN | ]. (1.9.27)
4
R
k=1
Now (1.9.25) and (1.9.27) together yield
X n
3
P( max | Uk (Rk Rk 1 )| ) E[|MN | ]. (1.9.28)
1nN 2
D
k=1
Since ⇠k = Rk + Bk , (1.9.22) and (1.9.28) give
n
X 5
P( max | Uk (⇠k ⇠k 1 )| ) E[|MN | ]. (1.9.29)
1nN
k=1
Finally, (1.9.16) and (1.9.29) together imply the required estimate (1.9.14).
E[[M, M ]n ] = E[Mn2 ]
E[( max |Mk |)2 ] (1.10.1)
1kn
4E[[M, M ]n ].
Burkholder and Gundy proved that indeed for 1 < p < 1, there exist constants c1p , c2p such
that
p p
c1p E[([M, M ]n ) 2 ] E[( max |Mk |)p ] c2p E[([M, M ]n ) 2 ].
1kn
Note that fr p = 2, this reduces to (1.10.1). Davis went on to prove the above inequality
for p = 1. This case plays an important role in the result on integral representation of
T
martingales that we will later consider. Hence we include a proof for the case p = 1
essentially this is the proof given by Davis [14]
1
F
Theorem 1.44. Let M be a martingale with M0 = 0. Then there exist universal constants
c1 , c2 such that for all N 1
Wn = max |Mk Mk 1|
1kn
and Vn1 = Un2 , Vn2 = Un1 . The reason for unusual notation is that we will prove
and this will prove both the inequalities in (1.10.2). Note that by definition, for all n 1,
Sn = E[Rn | Fn 1]
n
X
Xn = (Rk Sk )
k=1
Y n = Mn Xn
and X0 = Y0 = 0. Let us note that X is a martingale by definition and hence so is Y . Also
that
Xn
Yn = (Tk + Sk ).
k=1
T
If |Rn | > 0 then |(Mn Mn 1 )| > 2Wn 1 and so Wn = |(Mn Mn 1 )| and in turn,
|Rn | = Wn < 2(Wn Wn 1 ). Thus (noting Wn Wn 1 for all n)
Xn
|Rk | 2Wn (1.10.5)
X n
k=1
and using that E[ |Sn | ] E[ |Rn | ], it also follows that F
E[ |Sk | ] 2E(Wn ). (1.10.6)
A
k=1
Thus (using (1.10.4)) we have
Xn n
X
E[ |Rk | + |Sk | ] 4E(Wn ) 8E[Vnt ], t = 1, 2. (1.10.7)
R
k=1 k=1
Since Xk Xk 1 = Rk Sk , (1.10.7) gives us for all n 1,
Xn
E[ |Xk Xk 1 | ] 8E[Vnt ], t = 1, 2.
D
(1.10.8)
k=1
Let us define for n 1,
A1n = max |Xk |
1kn
Xn
1
A2n = ( (Xk Xk 1)
2
)2
k=1
Pn
and Bn1 = A2n , Bn2 = A1n , G1n = Fn2 , G2n = Fn1 . Since for 1 j n, |Xj | k=1 |Rk | +
Pn
k=1 |Sk |, the estimates (1.10.7) - (1.10.8) immediately gives us
and
Gtn Bnt + Vnt . (1.10.11)
T
Now using Fubini’s theorem (as used in proof of Theorem 1.25) it follows that
Z 1
t
E[Fn ] = P(Fnt > x)dx. (1.10.13)
0
F
We now will estimate P(FNt > x). For this we fix t 2 {1, 2} and for x 2 (0, 1), define a
stop time
A
x = inf{n 1 : Vnt > x or Gtn > x or |Sn+1 | > x} ^ N.
Since Sn+1 is Fn measurable, it follows that x is a stop time. If x < N then either
P
VNt > x or GtN > x or Nk=1 |Sn | > x. Thus
R
PN
P( x < N ) P(VNt > x) + P(GtN > x) + P( k=1 |Sk | > x)
and hence
Z Z Z
D
1 1 1
P( x < N )dx P(VNt > x)dx + P(GtN > x)dx
0 0 0
Z 1
P
+ P( Nk=1 |Sn | > x)dx
0
PN
= E[VNt ] + t
E[GN ] + E[ k=1 |Sn | ]
|Gt x Gt x 1| |T x + S x|
2W x 1 +x
4V tx 1 +x
5x
Gt x 6x.
Gt x min{BN
t
+ VNt , 6x}. (1.10.15)
T
Let Zn = Yn^ x . Then Z is a martingale with Z0 = 0 and
N
X
2 2
E[ZN ]= E[(Zk Zk 1) ] = E[(G1 x )2 ].
k=1
1
= 2 E[(G1 x )2 ].
x
On the other hand
PN
D
T
14E[VNt ] + E[36Qt ] + E[36Qt ].
Using the estimate (1.10.9), we have
1
E[Qt ] = (E[BNt
] + E[VNt ])
6
9
E[VNt ]
6
F (1.10.18)
A
and putting together (1.10.17)-(1.10.18) we conclude
E[FNt ] 122E[VNt ]
t
E[UN ] 130E[VNt ].
D
Chapter 2
In this chapter, we will give definitions, set up notations that will be used in the rest of the
T
book and give some basic results. While some proofs are included, several results are stated
without proof. The proofs of these results can be found in standard books on stochastic
processes.
L(m, d) will denote the space of m ⇥ d matrices. For x 2 Rd and A 2 L(m, d), |x| and kAk
will denote the Euclidean norms of x, A respectively.
(⌦, F, P) will denote a generic probability space and B(⌦, F) will denote the class of
D
For an integer d 1, let Cd = C([0, 1), Rd ) with the ucc topology, i.e. uniform
convergence on compact subsets of [0, 1). With this topology, Cd is itself a complete
40
2.1. Notations and Basic Facts 41
T
Exercise 2.1. Let ↵ 2 Dd .
(i) Show that for any ✏ > 0, and T < 1, the set
is a finite set.
F
{t 2 [0, T ] : |( ↵)(t)| > ✏}
A
(ii) Show that the set {t 2 [0, 1) : |( ↵)(t)| > 0} is a countable set.
-field is same as the Borel -field for the Skorokhod topology (see Ethier and Kurtz[18] ).
An E-valued random variable X defined on a probability space (⌦, F, P) is a measurable
function from (⌦, F) to (E, B(E)). For such an X and a function f 2 Cb (E), E[f (X)] will
denote the integral Z
E[f (X)] = f (X(!))dP(!).
⌦
If there are more than one probability measures under consideration, we will denote it as
EP [f (X)].
An E-valued stochastic process X is a collection {Xt : 0 t < 1} of E-valued random
variables. While one can consider families of random variables indexed by sets other than
42 Chapter 2. Continuous Time Processes
[0, 1), say [0, 1] or even [0, 1] ⇥ [0, 1], unless stated otherwise we will take the index set to
be [0, 1). Sometimes for notational clarity we will also use X(t) to denote Xt .
From now on unless otherwise stated, a process will mean a continuous time stochastic
process X = (Xt ) with t 2 [0, 1) or t 2 [0, T ]. For more details and discussions as well
as proofs of statements given without proof in this chapter, see Breiman [5], Ethier and
Kurtz[18], Ikeda and Watanabe[23] Jacod [25], Karatzas and Shreve [42], Metivier [48],
Meyer [49], Protter [50], Revuz-Yor [51], Stroock and Varadhan [57], Williams [56].
Definition 2.2. Two processes X, Y , defined on the same probability space (⌦, F, P) are
said to be equal (written as X = Y ) if
T
8t 0, 8! 2 ⌦0 , Xt (!) = Yt (!).
v
Definition 2.3. A process Y is said to be a version of another process X (written as X !
F
Y ) if both are defined on the same probability space and if
v
to see that X ! Y but P(Xt = Yt for all t 0) = 0.
Definition 2.4. Two E-valued processes X, Y are said to have same distribution (written
d
as X = Y ), where X is defined on (⌦1 , F1 , P1 ) and Y is defined on (⌦2 , F2 , P2 ), if for all
D
Thus, two processes have same distribution if their finite dimensional distributions are
v d
the same. It is easy to see that if X ! Y then X = Y .
Definition 2.7. An E-valued process X is said to be an r.c.l.l. process (or a process with
right continuous paths with left limits) if for all ! 2 ⌦, the path t 7! Xt (!) is right continuous
and admits left limits for all t > 0.
Definition 2.8. An E-valued process X is said to be an l.c.r.l. process (or a process with
left continuous paths with right limits) if for all ! 2 ⌦, the path t 7! Xt (!) is left continuous
on (0, 1) and admits right limits for all t 0.
For an r.c.l.l. process X, X will denote the r.c.l.l. process defined by Xt = X(t ),
i.e. the left limit at t, with the convention X(0 ) = 0 and let
T
X=X X
so that ( X)t = 0 at each continuity point and equals the jump otherwise. Note that by
the above convention
F
( X)0 = X0 .
v
Let X, Y be r.c.l.l. processes such that X ! Y . Then it is easy to see that X = Y .
A
The same is true if both X, Y are l.c.r.l. processes.
is measurable and induces a measure P X 1 on (Cd , B(Cd )). This is so because the Borel
-field B(Cd ) is also the smallest -field with respect to which the coordinate process is
measurable. Likewise, when X is an r.c.l.l. process, the mapping ! 7! X· (!) from ⌦ into Dd
is measurable and induces a measure P X 1 on (Dd , B(Dd )). In both cases, the probability
measure P X 1 is called the distribution of the process X
(i) P(X0 = 0) = 1.
44 Chapter 2. Continuous Time Processes
(ii) For 0 s < t < 1, the distribution of Xt Xs is Normal (Gaussian) with mean 0 and
co-variance matrix (t s)I, i.e. for u 2 Rd
(iii) For m 1, 0 = t0 < t1 < . . . < tm , the random variables Y1 , . . . , Ym are independent,
where Yj = Xtj Xtj 1 .
Remark 2.11. The process X in the definition above is sometimes called a standard Brownian
T
motion and Y given by
Yt = µt + Xt
where µ 2 Rd and is a positive constant, is also called a Brownian motion for any µ and .
F
When X is a d-dimensional Brownian motion, its distribution, i.e. the induced measure
µ = P X 1 on (Cd , B(Cd )) is known as the Wiener measure. The Wiener measure was
A
constructed by Wiener before Kolmogorov’s axiomatic formulation of Probability theory.
The existence of Brownian motion or the Wiener measure can be proved in many di↵erent
ways. One method is to use Kolmogorov’s consistency theorem to construct a process
X satisfying (i), (ii) and (iii) in the definition, which determine the finite dimensional
R
distributions of X and then to invoke the following criterion for existence of a continuous
v
process X̃ such that X ! X̃.
D
Theorem 2.12. Let X be an Rd -valued process. Suppose that for each T < 1, 9m >
0, K < 1 , > 0 such that
Exercise 2.13. Let X be a Brownian motion and let Y be defined as follows: Y0 = 0 and
for 0 < t < 1, Yt = tXs where s = 1t . Show that Y is also a Brownian motion.
Definition 2.14. A Poisson Process (with rate parameter ) is an r.c.l.l. non-negative integer
valued process N such that
2.2. Filtration 45
(i) N0 = 0.
n
(ii) P(Nt Ns = n) = exp{ t} ( n!t) .
(iii) For m 1, 0 = t0 < t1 < . . . < tm , the random variables Y1 , . . . , Ym are independent,
where Yj = Ntj Ntj 1 .
Exercise 2.15. Let N 1 , N 2 be Poisson processes with rate parameters 1 and 2 respectively.
Suppose N 1 and N 2 are independent. Show that N defined by Nt = Nt1 + Nt2 is also a Poisson
process with rate parameter = 1 + 2 .
Brownian motion and Poisson process are the two most important examples of contin-
uous time stochastic processes and arise in modelling of phenomenon occurring in nature.
T
2.2 Filtration
As in the discrete time case, it is useful to define for t 0, Gt to be the - field generated
F
by all the random variables observable up to time t and then require any action to be taken
at time t (an estimate of some quantity or investment decision) should be measurable with
respect to Gt . These observations lead to the following definitions.
A
Definition 2.16. A filtration on a probability space (⌦, F, P) is an increasing family of sub
-fields (F⇧ ) = {Ft : t 0} of F indexed by t 2 [0, 1).
R
We will always assume that the underlying probability space is complete i.e. N 2 F,
D
P(N ) = 0 and N1 ✓ N implies N 2 F and (ii) that F0 contains all sets N 2 F with
P(N ) = 0.
Note that if X is (F⇧ )-adapted and Y = X (see Definition 2.2) then Y is also (F⇧ )-
adapted in view of the assumption that F0 contains all null sets.
Given a filtration (F⇧ ), we will denote by (F⇧+ ) the filtration {Ft+ : t 0} where
Ft+ = \u>t Fu .
Let N be the class of all null sets (sets with P(N ) = 0), and for a process Z, possibly
vector-valued, let
FtZ = (N [ (Zu : 0 u t)). (2.2.1)
46 Chapter 2. Continuous Time Processes
Then (F⇧Z ) is the smallest filtration such that F0 contains all null sets with respect to
which Z is adapted.
While it is not required in the definition, in most situations, the filtration (F⇧ ) under
consideration would be chosen to be (for a suitable process Z)
(F⇧Z ) = {FtZ : t 0}
Sometimes, a filtration is treated as a mere technicality, specially in continuous time
setting as a necessary detail just to define stochastic integral. We would like to stress that
it is not so. See discussion in Section 1.4.
T
Definition 2.18. M is said to be (F⇧ ) martingale if M is (F⇧ ) adapted, Mt is integrable
for all t and for 0 s < t one has
F
EP [Mt | Fs ] = Ms .
When there is only one filtration under consideration, we will drop reference to it and
R
have r.c.l.l. paths. The next result shows that under minimal conditions on the underlying
filtration one can assume that every martingale has r.c.l.l. paths.
Proof. For k, n 1, let snk = k2 n and Xkn = Msnk and Gkn = Fsnk . Then for fixed n,
{Xkn : k 0} is a martingale w.r.t. the filtration {Gkn : k 0}. Thus the Doob’s
upcrossings inequality Theorem 1.27 yields, for rationals ↵ < and an integer T , writing
T n = T 2n ,
E[|MT | ] + |↵|
E[UTn ({Xkn : 0 k Tn }, ↵, )] .
↵
Thus, using the observation that
UTn ({Xkn : 0 k Tn }, ↵, ) UTn+1 ({Xkn+1 : 0 k Tn+1 }, ↵, )
we get by the monotone convergence theorem,
E[|MT | ] + |↵|
E[sup UTn ({Xkn : 0 k Tn }, ↵, )] .
n 1 ↵
Thus we conclude that there exists a set N with P(N ) = 0 such that for ! 62 N ,
T
sup UTn ({Xkn : 0 k Tn }, ↵, ) < 1 (2.3.4)
n 1
for all rationals ↵ < and integers T . Thus if {tk : k 1} are diadic rationals increasing
or decreasing to t, then for ! 62 N , Mtk (!) converges. Thus defining ✓k (t) = ([t2k ] + 1)2 k ,
F
where [r] is the largest integer less than or equal to r (the integer part of r), and for ! 62 N ,
M̃t (!) = lim M✓k (t) (!), 0 t < 1
k!1
(2.3.5)
A
it follows that t 7! M̃t (!) is a r.c.l.l. function and agrees with Mt (!) for diadic rationals t.
Let us define M̃t (!) = 0 for ! 2 N . Since F0 contains all P null sets, it follows that M̃ is
an r.c.l.l. adapted process and by definition
R
construction. Further,
Mtn = E[Mt+1 | Ftn ] ! E[Mt+1 | Ft ] = Mt a.s.
by Theorem 1.37. Hence (2.3.3) follows.
The conditions (2.3.1) and (2.3.2) together are known as usual Hypothesis in the lit-
erature. We will assume (2.3.1) but will not assume (2.3.2). We will mostly consider
martingales with r.c.l.l. paths, and refer to it as an r.c.l.l. martingale.
When we are having only one filtration under consideration, we will drop reference to
it and simply say M is a martingale. We note here an important property of martingales
that would be used later.
48 Chapter 2. Continuous Time Processes
Mtn ! Mt in L1 (P), 8t 0.
Proof. Note that for any X n converging to X in L1 (P), for any -field G, using (i), (ii),
(iii) and (iv) in Proposition 1.2, one has
E[|E[X n | G] E[X | G]| ] = E[|E[(X n X) | G]| ]
E[E[|X n X| | G]]
n
= E[|X X| ]
! 0.
T
For s < t, applying this to X n = Mtn , one gets
Proof. Here {Mn : n 1} is a uniformly integrable martingale and Theorem 1.35 yields
that Mn ! ⇠ in L1 (P). Similarly, for any sequence tn " 1, Mtn converges to say ⌘ in
D
The Doob’s maximal inequality for martingales in continuous time follows from the
discrete version easily.
Example 2.25. Let (Wt ) denote a one dimensional Brownian motion on a probability space
(⌦, F, P). Then by the definition of Brownian motion, for 0 s t, Wt Ws has mean zero
and is independent of {Wu , 0 u s}. Hence (see (1.3.2)) we have
E[Wt Ws | FsW ] = 0
and hence W is an (F⇧W )-martingale. Likewise,
E[(Wt Ws )2 | FsW ] = E[(Wt Ws ) 2 ] = t s
and using this and the fact that (Wt ) is a martingale, it is easy to see that
E[Wt2 Ws2 | FsW ] = t s
T
and so defining Mt = Wt2 t, it follows that M is also an (F⇧W )-martingale.
Example 2.26. Let (Nt ) denote a Poisson process with rate parameter = 1 on a probability
F
space (⌦, F, P). Then by the definition of Poisson process, for 0 s t, Nt Ns has mean
(t s) and is independent of {Nu , 0 u s}. Writing Mt = Nt t, we can see using (1.3.2)
that
A
E[Mt Ms | FsN ] = 0
and hence M is an (F⇧N )-martingale. Likewise,
E[(Mt Ms )2 | FsM ] = E[(Mt Ms ) 2 ] = t s
R
and using this and the fact that (Mt ) is a martingale, it is easy to see that
E[Mt2 Ms2 | FsN ] = t s
D
The notion of stop time, as mentioned in Section 1.6, was first introduced in the context
of Markov chains. Martingales and stop times together are very important tools in the
theory of stochastic process in general and stochastic calculus in particular.
Definition 2.27. A stop time with respect to a filtration (F⇧ ) is a mapping ⌧ from ⌦ into
[0, 1] such that for all t < 1,
{⌧ t} 2 Ft .
50 Chapter 2. Continuous Time Processes
If the filtration under consideration is fixed, we will only refer to it as a stop time. Of
course, for a stop time, {⌧ < t} = [n {⌧ t n1 } 2 Ft . For stop times ⌧ and , it is easy
to see that ⌧ ^ and ⌧ _ are stop times. In particular, ⌧ ^ t and ⌧ _ t are stop times.
Example 2.28. Let X be a Rd -valued process with continuous paths adapted to a filtration
(F⇧ ) and let C be a closed set in R. Then
⌧C = inf{t 0 : Xt 2 C} (2.3.6)
is a stop time. To see this, define open sets Uk = {x : d(x, C) < k1 }. Writing Ar,k = {Xr 2
Uk } and Qt = {r : r rational, 0 r t},
{⌧C t} = \1
k=1 [[r2Qt Ar,k ]. (2.3.7)
⌧C is called the hitting time of C.
T
If X is an r.c.l.l. process, then the hitting time ⌧C may not be a random variable and
hence may not be a stop time. Let us define the contact time C by
C = inf{t 0 : Xt 2 C or Xt 2 C}.
Exercise 2.29. Construct an example to show that this alternate description may be incorrect
R
when 0 2 C.
⌧m = 2 m
([2m ⌧ ] + 1) (2.3.10)
is also a stop time since {⌧ m t} = {⌧ m 2 m ([2m t])} = {⌧ < 2 m ([2m t])} and it follows
that {⌧ m t} 2 (F⇧ ). Clearly, ⌧ m # ⌧ .
One can see that if k is an increasing sequence of stop times ( k k+1 for all k 1),
then = limk"1 k is also a stop time. However, if k # , may not be a stop time as
seen in the next exercise.
Exercise 2.30. Let ⌦ = { 1, 1} and P be such that 0 < P(1) < 1. Let Xt (1) = Xt ( 1) = 0
for 0 t 1. For t > 1, let Xt (1) = sin(t 1) and Xt ( 1) = sin(t 1). Let
k
k = inf{t 0 : Xt 2 },
2.3. Martingales and Stop Times 51
Show that FtX = { , ⌦} for 0 t 1 and FtX = { , {1}, { 1}, ⌦} for t > 1, k are stop
times w.r.t. (F·X ) and k # but is not a stop time w.r.t. (F·X ). Note that { < s} 2 FsX
for all s and yet is not a stop time.
Exercise 2.31. Let ⌧ be a [0, 1)-valued random variable and Ft = (⌧ ^ t). Show that if
⌘ is a stopping time for this filtration then ⌘ = (⌘ _ ⌧ ) ^ r for some r 2 [0, 1].
Definition 2.32. For a stochastic process X and a stop time ⌧ , X⌧ is defined via
Remark 2.33. A random variable ⌧ taking countably many values {sj }, 0 sj < 1 8j is
a stop time if and only if {⌧ = sj } 2 Fsj for all j. For such a stop time ⌧ and a stochastic
T
process X,
X1
X⌧ = 1{⌧ =sj } Xsj
j=1
F
and thus X⌧ is a random variable (i.e. a measurable function).
In general, X⌧ may not be a random variable, i.e. may not be a measurable function.
A
However, if X has right continuous paths X⌧ is a random variable. To see this, given ⌧
that is [0, 1)-valued, X⌧ m is measurable where ⌧ m is defined via (2.3.10) and X⌧ m ! X⌧
and hence X⌧ is a random variable. Finally, for a general stop time ⌧ , ⇠n = X⌧ ^n is a
R
Lemma 2.34. Let Z be an r.c.l.l. adapted process and ⌧ be a stop time with ⌧ < 1. Let
Ut = Z⌧ ^t and Vt = Z⌧ 1[⌧,1) (t). Then U, V are adapted processes (with respect to the same
filtration (F⇧ ) that Z is adapted to).
Proof. When ⌧ takes finitely many values it is easy to see that U, V are adapted and for
the general case, the proof is by approximating ⌧ by ⌧ m defined via (2.3.10).
Definition 2.35. For a stop time ⌧ with respect to a filtration (F⇧ ), the stopped -field is
defined by
F⌧ = {A 2 ([t Ft ) : A \ {⌧ t} 2 Ft 8t}.
52 Chapter 2. Continuous Time Processes
We have seen that for a right continuous process Z, Z⌧ is a random variable, i.e.
measurable. The next lemma shows that indeed Z⌧ is F⌧ measurable.
Lemma 2.36. Let X be a right continuous (F⇧ )-adapted process and ⌧ be a stop time. Then
X⌧ is F⌧ measurable.
Proof. For a t < 1, we need to show that {X⌧ a} \ {⌧ t} 2 Ft for all a 2 R. Note
that
{X⌧ a} \ {⌧ t} = {Xt^⌧ a} \ {⌧ t}.
As seen in Lemma 2.34, {Xt^⌧ a} 2 Ft and of course, {⌧ t} 2 Ft .
T
Also, we have
F
Proof. Let Xt = 1[ (!),1) (t). We have seen that X is r.c.l.l. adapted and hence X⌧ is F⌧
measurable. Note that X⌧ = 1{ ⌧ } . This completes the proof.
A
Here is another observation.
Lemma 2.39. Let ⌧ be a stop time and ⇠ be F⌧ measurable random variable. Then Z =
⇠ 1[⌧,1) is an adapted r.c.l.l. process.
R
Proof. Note that for any t, {Zt a} = {⇠ a} \ {⌧ t} if a < 0 and {Zt a} = ({⇠
a} \ {⌧ t}) [ {⌧ > t} if a 0. Since {⇠ a} 2 F⌧ , we have {⇠ a} \ {⌧ t} 2 Ft
c
and also {⌧ > t} = {⌧ t} 2 Ft . This shows Z is adapted. Of course it is r.c.l.l. by
D
definition.
Lemma 2.40. Let X be any real-valued r.c.l.l. process adapted to a filtration (F⇧ ). Let
be any stop time. Then for any a > 0, ⌧ defined as follows is a stop time : if = 1 then
⌧ = 1 and if < 1 then
⌧ = inf{t > : |Xt X | a or |Xt X | a}. (2.3.11)
If ⌧ < 1 then ⌧ > and either |X⌧ X | a or |X⌧ X | a. In other words, when
the infimum is finite, it is actually minimum.
2.3. Martingales and Stop Times 53
T
For the other part, if ! 2 ⌦2 , then there exist {rn 2 Qt : n 1} such that |Yrn (!)|
a n1 . Since Qt ✓ [0, t], it follows that we can extract a subsequence rnk that converges
to s 2 [0, t]. By taking a further subsequence if necessary, we can assume that either
F
rnk s 8k or rnk < s 8k and thus |Yrnk (!)| ! |Ys (!)| a in the first case and |Yrnk (!)| !
|Ys (!)| a in the second case. Also, Y0 (!) = 0 and Y0 (!) = 0 and hence s 2 (0, t]. This
shows ⌧ (!) s t and thus ! 2 ⌦1 . This proves (2.3.12).
A
In each of the cases, we see that either |Y⌧ | a or |Y⌧ | a. Since Yt = 0 for t , it
follows that ⌧ > . This completes the proof.
Remark 2.41. Note that ⌧ is the contact time for the set C = [a, 1) for the process Y .
R
Exercise 2.43. Construct an example to convince yourself that in the definition (2.3.11) of
⌧ , t > cannot be replaced by t . However, when X is continuous, in (2.3.13), t > can
be replaced by t .
Theorem 2.44. Let X be an r.c.l.l. adapted process with X0 = 0. For a > 0, let { i , : i
0} be defined inductively as follows: 0 = 0 and having defined j : j i, let i+1 = 1 if
i = 1 and otherwise
Then each i is a stop time. Further, (i) if i < 1, then i < i+1 and (ii) limi"1 i = 1.
Proof. That each i is a stop time and observation (i) follow from Lemma 2.40. Remains
to prove (ii). If for some ! 2 ⌦,
lim i (!) = t0 < 1
i"1
then for such an !, the left limit of the mapping s ! Xs (!) at t0 does not exist, a
contradiction. This proves (ii).
Exercise 2.45. Let X be the Poisson process with rate parameter . Let 0 = 0 and for
i 0, i+1 be defined by (2.3.14) with a = 1. For n 1 let
⌧n = n n 1.
Show that
T
(i) Nt (!) = k if and only if k (!) t< k+1 (!).
(ii) {⌧n : n 1} are independent random variables with P(⌧n > t) = exp{ t} for all n 1
and t 0.
F
Recall that (F⇧+ ) denotes the fight continuous filtration corresponding to the filtration
(F⇧ ) We now show that the hitting time of the interval ( 1, a) by an r.c.l.l. adapted
A
process is a stop time for the filtration (F⇧+ ).
Proof. Note that for any s > 0, right continuity of t 7! Yt (!) implies that
{! : ⌧ (!) < s} = {! : Yr (!) < a for some r rational, r < s}
and hence {⌧ < s} 2 Fs .
Now for any t, for all k
1
{⌧ t} = \1
n=k {⌧ < t + } 2 Ft+ 1
n k
Remark 2.47. Similarly, it can be shown that the hitting time of an open set by an r.c.l.l.
adapted process is a stop time for the filtration (F⇧+ ).
2.3. Martingales and Stop Times 55
{ < t} 2 Ft 8t 0.
{ t} = \1
n=m { < tn } 2 Ftm .
+
Thus { t} 2 \1
m=1 Ftm = Ft .
Corollary 2.49. Let {⌧m : m 1} be a sequence of (F⇧ )-stop times. Let = inf{⌧m :
m 1} and ⇣ = sup{⌧m : m 1}. Then ⇣ is a (F⇧ )-stop time whereas is (F⇧+ )-stop
T
time.
and
F
{⇣ t} = \m {⌧m t}
is (F⇧ ) adapted.
D
F ✓ F⌧ .
{ = ⌧ } 2 F \ F⌧ .
56 Chapter 2. Continuous Time Processes
Martingales and stop times are intricately related as the next result shows.
Theorem 2.53. Let M be a r.c.l.l. martingale and ⌧ be a bounded stop time, ⌧ T . Then
E[MT | F⌧ ] = M⌧ . (2.3.16)
Proof. We have observed that M⌧ is F⌧ measurable. First let us consider the case when ⌧
is a stop time taking finitely many values, s1 < s2 < . . . < sm T . Let Aj = {⌧ = sj }.
T
Since ⌧ is a stop time, Aj 2 Fsj . Clearly, {A1 , . . . , Am } forms a partition of ⌦. Let B 2 F⌧ .
Then, by definition of F⌧ it follows that Cj = B \ Aj 2 Fsj . Since M is a martingale,
E[MT | Fsj ] = Msj and hence
E[MT 1B ] = E[M⌧ 1B ].
A
This proves (2.3.16) when ⌧ takes finitely many values. For the general case, given ⌧ T ,
let
([2n ⌧ ] + 1)
⌧n = ^ T.
R
2n
Then for each n, ⌧n is a stop time that takes only finitely many values and further the
sequence {⌧n } decreases to ⌧ . By the part proven above we have
D
It should be noted that we have not assumed that the underlying filtration is right
continuous. This result leads to the following
As in the discrete case, here too we have this characterization of martingales via stop
times.
Theorem 2.55. Let X be an r.c.l.l. (F⇧ ) adapted process with E[|Xt | ] < 1 for all t < 1.
T
Then X is an (F⇧ )-martingale if and only if for all (F⇧ )-stop times ⌧ taking finitely many
values in [0, 1), one has
E[X⌧ ] = E[X0 ]. (2.3.20)
F
Further if X is a martingale then for all bounded stop times
E[X ] = E[X0 ].
one has
(2.3.21)
A
Proof. Suppose X is a martingale. Then (2.3.20) and (2.3.21) follow from (2.3.16).
On the other hand suppose (2.3.20) is true for stop times taking finitely many values.
Fix s < t and A 2 Fs . To show E[Xt | Fs ] = Xs , suffices to prove that E[Xt 1A ] = E[Xs 1A ].
Now take
R
⌧ = s 1 A + t1 A c
to get
D
Theorem 2.57. Suppose M is an r.c.l.l. (F⇧ )-martingale and and ⌧ are (F⇧ )-stop times
with ⌧ . Suppose X is an r.c.l.l. (F⇧ ) adapted process. Let
Nt = X ^t (M⌧ ^t M ^t ).
Then N is a (F⇧ )-martingale if either (i) X is bounded or, if (ii) E[X 2 ] < 1 and M is
square integrable.
Proof. Clearly, N is adapted. First consider the case when X is bounded. We will show
that for any bounded stop time ⇣ (bounded by T ),
E[N⇣ ] = 0
and then invoke Theorem 2.53. Note that
T
N⇣ =X ^⇣ (M⌧ ^⇣ M ^⇣ )
=X ˜ (M⌧˜ M˜ )
where ˜ = ^ ⇣ ⌧˜ = ⌧ ^ ⇣ are also bounded stop times. Now
F
E[N⇣ ] = E[E[N⇣ | F ˜ ]]
= E[E[X ˜ (M⌧˜ M ˜ ) | F ˜ ]]
A
= E[X ˜ (E[M⌧˜ | F ˜ ] M ˜ )]
=0
as E[M⌧˜ | F ˜ ] = M ˜ by part (ii) of Corollary 2.54. This proves the result when X is
R
bounded. For (ii), approximating X by X n , where Xtn = max{min{Xt , n}, n} and using
(i) we conclude
E[X n^⇣ (M⌧ ^⇣ M ^⇣ )] = 0.
D
Corollary 2.58. Suppose M is a r.c.l.l. (F⇧ )-martingale and and ⌧ are stop times with
⌧ . Suppose U is a F measurable random variable. Let
Nt = U (M⌧ ^t M ^t ).
Then N is a (F⇧ )-martingale if either (i) U is bounded or, if (ii) E[U 2 ] < 1 and M is
square integrable.
Then X is adapted by Lemma 2.39. Now the result follows from Theorem 2.57.
Theorem 2.59. Suppose M, N are r.c.l.l. (F⇧ )-martingales, E(Mt2 ) < 1, E(Nt2 ) < 1 and
T
and ⌧ are (F⇧ )-stop times with ⌧ . Suppose B, X are r.c.l.l. (F⇧ ) adapted processes
with E(|Bt |) < 1 for all t and X bounded. Suppose that Z is also a martingale where
Zt = Mt Nt Bt . Let
Yt = X ^t [(M⌧ ^t
=0
by Theorem 2.57 since M, N, Z are martingales. This completes the proof.
Definition 2.60. A subset A ✓ B(⌦, F) is said to be closed under uniform and monotone
convergence if f 2 B(⌦, F), f n , g n , hn 2 A for n 1 are such that
then f, g, h 2 A.
Theorem 2.61. Let A ✓ B(⌦, F) be closed under uniform and monotone convergence.
Suppose G ✓ A is an algebra such that
T
(i) (G) = F.
K0 (f ) = {g 2 K : ↵f + g 2 K, 8↵, 2 R}.
Note that K0 (f ) is closed under uniform and monotone convergence. First fix f 2 G. Since
G is an algebra, it follows that G ✓ K0 (f ) and hence K = K0 (f ). Now fix f 2 K. The
D
statement proven above implies that G ✓ K0 (f ) and since K0 (f ) is closed under uniform
and monotone convergence, it follows that K = K0 (f ). Thus K is a vector space.
To show that K is an algebra, for f 2 B(⌦, F) let
K1 (f ) = {g 2 K : f g 2 K}.
Since we have shown that K is a vector space containing constants, it follows that
K1 (f ) = K1 (f + c)
is closed under monotone and uniform convergence. Now proceeding as in the proof of K
being vector space, we can show that K is closed under multiplication and thus K is an
algebra.
Let C = {A 2 F : 1A 2 K}. In view of the assumption (ii), ⌦ 2 C. It is clearly a -field
and also B(⌦, C) ✓ K. Since (G) = F, one has (A) = F where
A = { {f ↵} : f 2 G, ↵ 2 R}
and hence to complete the proof, suffices to show A ✓ C as that would imply F = C and
in turn
B(⌦, F) ✓ K ✓ A ✓ B(⌦, F)
implying K = A = B(⌦, F). Now to show that A ✓ C, fix f 2 G and ↵ 2 R and let
A = {f ↵} 2 A. Let |f | M . Let (x) = 1 for x ↵ and (x) = 0 for x ↵ + 1 and
T
(x) = 1 x+↵ for ↵ < x < ↵+1. Using Weierstrass’s approximation theorem, we can get
polynomials pn that converge to uniformly on [ M, M ] and hence boundedly pointwise
on [ M, M ]. Now pn (f ) 2 G as G is an algebra and pn (f ) converges boundedly pointwise
F
to (f ). Thus (f ) 2 K. Since K is an algebra, it follows that ( (f ))m 2 K for all m 1.
Clearly ( (f ))m converges monotonically to 1A . Thus g m = 1 ( (f ))m 2 K increases
monotonically to 1 1A and hence 1 1A 2 K i.e. A 2 C completing the proof.
A
Here is a useful variant of the Monotone class Theorem.
bp
fn 2 A 8n 1, fn ! f implies f 2 A.
If a set A is bp-closed then it is also closed under monotone and uniform limits and thus
we can deduce the following useful variant of the Monotone class Theorem from Theorem
2.61
(i) (G) = F.
Theorem 2.65. Let F be a -field on ⌦ and let Q be a probability measure on (⌦, F).
Suppose G ✓ B(⌦, F) be an algebra such that (G) = F. Further, 9f n 2 G such that
bp
f n ! 1 (where 1 denotes the constant function 1). Then G is dense in L2 (⌦, F, Q).
Proof. Let K denote the closure of G in L2 (⌦, F, Q) and let A be the set of bounded
functions in K. Then G ✓ A and hence by Theorem 2.64 it follows that A = B(⌦, F).
Hence it follows that K = L2 (⌦, F, Q) as every function in L2 (Q) can be approximated by
bounded functions.
T
2.5 The UCP Metric
Let R0 (⌦, (F⇧ ), P) denote the class of all r.c.l.l. (F⇧ ) adapted processes. For processes
X, Y 2 R0 (⌦, (F⇧ ), P), let
ducp (X, Y ) =
1
X
m=1
2 m
F
E[min(1, sup |Xt
0tm
Yt |)]. (2.5.1)
A
Noting that ducp (X, Y ) = 0 if and only if X = Y (i.e. P(Xt = Yt 8t) = 1), it follows that
ducp is a metric on R0 (⌦, (F⇧ ), P). Now ducp (X n , X) ! 0 is equivalent to
0tT
ucp
also called uniform convergence in probability, written as X n ! X.
D
Remark 2.66. We have defined ducp (X, Y ) when X, Y are real-valued r.c.l.l. processes. We
can similarly define ducp (X, Y ) when X, Y are Rd -valued r.c.l.l. or l.c.r.l. processes. We will
use the same notation ducp in each of these cases.
In the rest of this section, d is a fixed integer and we will be talking about Rd -valued
processes, and |·| will be the Euclidean norm on Rd .
ucp
When ducp (X n , X) ! 0, sometimes we will write it as X n ! X (and thus the two
mean the same thing). Let X, Y 2 R0 (⌦, (F⇧ ), P). Then for > 0 and integers N 1,
observe that
N
ducp (X, Y ) 2 + + P( sup |Xt Yt | > ) (2.5.2)
0tN
2.5. The UCP Metric 63
and
2N
P( sup |Xt Yt | > ) ducp (X, Y ). (2.5.3)
0tN
For an r.c.l.l. process Y , given T < 1, " > 0 one can choose K < 1 such that
P[ sup|Yt | > K ] < " (2.5.5)
tT
ucp
and hence using (2.5.4) it follows that if Z n ! Z, then given T < 1, " > 0 there exists
T
K < 1 such that
sup P[ sup|Ztn | > K ] < ". (2.5.6)
n 1 tT
The following result uses standard techniques from measure theory and functional anal-
F
ysis but a proof is included as it plays an important part in subsequent chapters.
Theorem 2.68. The space R0 (⌦, (F⇧ ), P) is complete under the metric ducp .
A
Proof. Let {X n : n 1} be a Cauchy sequence in ducp metric. By taking a subsequence if
necessary, we can assume without loss of generality that ducp (X n+1 , X n ) < 2 n and hence
1 X
X 1
R
2 m
E[min(1, sup |Xtn+1 Xtn |)] < 1
n=1 m=1 0tm
or equivalently
1
X 1
X
m
E[min(1, sup |Xtn+1 Xtn |)] < 1
D
2
m=1 n=1 0tm
and thus for all m 1 one has (using, for a random variable Z with Z 0, E[Z] < 1
implies Z < 1 a.s.)
1
X
[min(1, sup |Xtn+1 Xtn |)] < 1 a.s.
n=1 0tm
P P
Note that for a [0, 1)-valued sequence {an }, n min(an , 1) < 1 if and only if n an <
1. Hence for all m 1 we have
X1
sup |Xtn+1 Xtn | < 1 a.s.
n=1 0tm
64 Chapter 2. Continuous Time Processes
P
Again, noting that for a real-valued sequence {bn }, n |bn+1 bn | < 1 implies that {bn }
is cauchy and hence converges, we conclude that outside a fixed null set (say N ), Xtn
converges uniformly on [0, m] for every m. So we define the limit to be Xt , which is an
r.c.l.l. process. On the exceptional null set N , Xt is defined to be zero. Note that
n+k
X1
ducp (X n+k , X n ) ducp (X j+1 , X j )
j=n
n+k
X1
j
2
j=n
n+1
2 .
As a consequence, (using definition of ducp ) we get that for all integers m,
E[min(1, sup |Xtn Xtn+k |)] 2m 2 n+1
.
T
0tm
Taking limit as k ! 1 and invoking Dominated convergence theorem we conclude
E[min(1, sup |Xtn Xt |)] 2m 2 n+1
.
0tm
It follows that for any T < 1, 0 <
P( sup |Xtn
0tT
F
< 1 we have
1
Xt |) > ) 2(T +1) 2 n+1
. (2.5.7)
A
and hence, invoking Remark 2.67, it follows that X n converges in ucp metric to X. Thus
every Cauchy sequence converges and so the space is complete under the metric ducp .
k
Then there exists a subsequence {nk } such that Y k = Z n satisfies
Proof. Since ducp (Z n , Z) ! 0, for k 1, we can choose nk with nk+1 > nk such that
k
ducp (Z n , Z) 2 k . Then as seen in the proof of Theorem 2.68, this implies
1
X
[sup|Ytk Zt |] < 1, 8T < 1 a.s.
k=1 tT
2.5. The UCP Metric 65
Hence (i) above holds for this choice of {nk }. Further, defining
1
X
Ht = sup [ |Ysk Zs | + |Zs |]
0st
k=1
Remark 2.70. If we have two, or finitely many sequences {Z i,n } converging to Z i in ducp ,
i = 1, 2, . . . , p then we can get one common subsequence {nk } and a process H such that (i),
(ii) above hold for i = 1, 2, . . . p. All we need to do is to choose {nk } such that
k
ducp (Z i,n , Z i ) 2 k
, i = 1, 2 . . . p.
Exercise 2.71. An alternative way of obtaining the conclusion in Remark 2.70 is to apply
Theorem 2.69 to an appropriately defined Rdp -valued processes.
T
The following Lemma will play an important role in the theory of stochastic integration.
Lemma 2.72. Let Z n , Z be adapted processes and let ⌧ m be a sequence of stop times such
that ⌧ m " 1. Suppose that for each m
n
Zt^⌧ m
ucp F
! Zt^⌧ m as n " 1. (2.5.9)
A
Then
ucp
Zn ! Z as n " 1.
Proof. Fix T < 1, " > 0 and ⌘ > 0. We need to show that 9n0 such that for n n0
R
Now,
{ sup|Ztn n
Zt | > ⌘} ✓ {{sup|Zt^⌧ m Zt^⌧ m | > ⌘} [ {⌧ m < T }}
tT tT
Corollary 2.73. Let Z n be r.c.l.l. adapted processes and let ⌧ m be a sequence of stop
times such that ⌧ m " 1. For n 1, m 1 let
Ytn,m = Zt^⌧
n
m.
Suppose that for each m, {Y n,m : n 1} is Cauchy in ducp metric then Z n is Cauchy in
in ducp metric.
T
{ti }:a=t0 <t1 <...<tm =b, m 1 j=1
F
If Var[0,t] (G) < 1 for all t, G will be called a function with finite variation. It is
well known that a function has finite variation paths if and only if it can be expressed as
di↵erence of two increasing functions.
A
If Var[0,t] (G) < 1 for all t, the function |G|t = Var[0,t] (G) is then an increasing [0, 1)-
valued function. Let us fix such a function G.
For any T fixed, there exists a unique countably additive measure and a countably
additive signed measure µ on the Borel -field of [0, T ] such that
R
Rt
It follows that if h is a bounded measurable function on [0, 1), then hdG is defined
0
bp Rt
and further ifhn ! h, then the dominated convergence theorem yields that Htn = 0 hn dG
Rt
converges to H(t) = 0 hdG uniformly on compact subsets.
Exercise 2.74. Let G be an r.c.l.l. function on [0, 1) such that |G|t < 1 for all t < 1.
Show that for all T < 1
X
|( G)t | < 1. (2.6.7)
tT
Note that as seen in Exercise 2.1, {t : |( G)t | > 0} is a countable set and thus the sum
appearing above is a sum of countably many terms.
Hint: Observe that the left hand side in (2.6.7) is less than or equal to |G|T .
Let us denote by V+ = V+ (⌦, (F⇧ ), P) the class of (F⇧ ) adapted r.c.l.l. increasing
processes A with A0 0 and by V = V(⌦, (F⇧ ), P) the class of r.c.l.l. adapted processes B
T
such that A defined by
At (!) = Var[0,t] (B· (!)) (2.6.8)
F
belongs to V+ . A process A 2 V will be called process with finite variation paths. It is
easy to see that B 2 V if and only if B can be written as di↵erence of two processes in
V+ : indeed, if A is defined by (2.6.8), we have B = D C where D = 12 (A + B) and
A
C = 12 (A B) and C, D 2 V+ . Let V0 and V+ 0 denote the class of processes A in V and
V respectively such that A0 = 0. For B 2 V, we will denote the process A defined by
+
(2.6.8) as A = Var(B).
A process A 2 V will be said to be purely discontinuous if
R
X
At = A0 + ( A)s .
0<st
Lemma 2.76. Let B 2 V and let X be a bounded l.c.r.l. adapted process. Then
Z t
Ct (!) = Xs (!)dBs (!) (2.6.9)
0
is well defined and is an r.c.l.l. adapted process. Further, C 2 V.
Proof. For every ! 2 ⌦, t 7! Xt (!) is a bounded measurable function and hence C is well
defined. For n 1 and i 0 let tni = ni . Let
Xtn = Xtni for tni t < tni+1 .
68 Chapter 2. Continuous Time Processes
bp
Then X n ! X. Clearly
Z t X
Ctn (!) = Xsn (!)dBs (!) = Xtni (Btni+1 ^t Btni ^t )
0 i : tn
i t
bp
is r.c.l.l. adapted and further, C n ! C. Thus C is also an r.c.l.l. adapted process. Let
A = Var(B). For any s < t,
Z t
|Ct (!) Cs (!)| |Xu (!)|dAs (!)
s
K(At (!) As (!)).
where K is a bound for the process X. Since A is an increasing process, it follows that
Var[0,T ] (C)(!) KAT (!) < 1 for all T < 1 and for all !.
T
Exercise 2.77. Show that the conclusion in Lemma 2.76 is true even when X is a bounded
r.c.l.l. adapted process. In this case, Dtn defined by
Z t X
n
Dt (!) = Xsn (!)dBs (!) = Xtni+1 ^t (Btni+1 ^t Btni ^t )
converges to
Rt
0 X dB.
0
Fi : tn
i t
A
R
D
Chapter 3
T
Let (Wt ) denote a one dimensional Brownian motion on a probability space (⌦, F, P). We
have seen that W is a martingale w.r.t. its natural filtration (F⇧W ) and with Mt = Wt2 t,
M is also (F⇧W ) martingale. These properties are easy consequence of the independent
increment property of Brownian motion. F
Wiener and Ito realized the need to give a meaning to limit of what appeared to be
A
Riemann-Stieltjes sums for the integral
Z t
fs dWs (3.1.1)
0
in di↵erent contexts- while in case of Wiener, the integrand was a deterministic function,
R
Ito needed to consider a random process (fs ) that was a non-anticipating function of W -
i.e. f is adapted to (F⇧W ).
It is well known that paths s 7! Ws (!) are nowhere di↵erentiable for almost all !
D
n n
Vt ! 1 a.s. and (b) Qt ! t a.s.
Proof. We will first prove (b). Let us fix t < 1 and let
69
70 Chapter 3. The Ito Integral
Then from properties of Brownian motion it follows that {Xin , i 0} are independent
random variables with normal distribution and E(Xi ) = 0, E(Xi ) = (tni+1 ^ t tni ^ t).
n n 2
So, {Zin , i 0} are independent random variables with E(Zin ) = 0 and E(Zin )2 = 2((tni+1 ^
t tni ^ t))2 . Now
1
X
n 2
E(Qt t) =E( Zin )2
i=0
1
X
= E(Zin )2
i=0
X1
(3.1.2)
=2 (tni+1 ^ t tni ^ t)2
i=0
1
X
n+1
2 (tni+1 ^ t tni ^ t)
T
i=0
n+1
=2 t.
Note that each of the sum appearing above is actually a finite sum. Thus
so that
P1 n
n=1 (Qt t)2
E
X1
(Qnt t)2 t < 1
n=1
< 1 a.s. and hence Qnt ! t a.s.
F
A
For (a), let ↵( , !, t) = sup{|Wu (!) Wv (!)| : |u v| , u, v 2 [0, t]}. Then uniform
continuity of u 7! Wu (!) implies that for each !, t < 1,
lim ↵( , !, t) = 0. (3.1.3)
R
#0
Now note that for any !,
1
X
Qnt (!) ( max |Wtni+1 ^t (!) Wtni ^t (!)|)( |Wtni+1 ^t Wtni ^t |)
D
0i<1 (3.1.4)
i=0
n
=↵(2 , !, t)Vtn (!).
So if lim inf n Vtn (!) < 1 for some !, then lim inf n Qnt (!) = 0 in view of (3.1.3) and (3.1.4).
For t > 0, since Qnt ! t a.s., we must have Vtn ! 1 a.s.
m (T ) =( sup (tm
n+1 tm
n )) ! 0 as m " 1 (3.1.6)
{n : tm
n T }
3.2. Levy’s Characterization of the Brownian motion 71
let
1
X
Qm
t = (Wtm
n+1 ^t
Wtm
n ^t
)2 . (3.1.7)
n=0
Show that for each t, Qm
t converges in probability to t.
Remark 3.3. It is well known that the paths of Brownian motion are nowhere di↵erentiable.
For this and other path properties of Brownian motion, see Breiman [5], McKean [45], Karatzas
and Shreve[42].
Remark 3.4. Since the paths of Brownian motion do not have finite variation on any interval,
R
we cannot invoke Riemann-Stieltjes integration theory for interpreting X dW , where W is
Brownian motion. The following calculation shows that the Riemann-Stieltjes sums do not
R
converge in any weaker sense (say in probability) either. Let us consider W dW . Let tni =
T
i2 n , i 0, n 1. The question is whether the sums
1
X
Wsni ^t (Wtni+1 ^t Wtni ^t ) (3.1.8)
i=0
i=0
Now Ant and Btn cannot converge to the same limit as their di↵erence satisfies
Thus even in this simple case, the Riemann-Stieltjes sums do not converge to a unique limit.
In this case, it is possible to show that Ant and Btn actually do converge but to two di↵erent
limits. It is possible to choose {sni } so that the Riemann sums in (3.1.8) do not converge.
It follows that Mt = Wt2 t is also a martingale w.r.t (Ft ). Levy had proved that if W is
any continuous process such that both W, M are (F⇧ )-martingales then W is a Brownian
motion and (3.2.1) holds. Most proofs available in texts today deduce this as an application
of Ito’s formula. We will give an elementary proof of this result which uses interplay of
martingales and stop times. The proof is motivated by the proof given in Ito’s lecture notes
[24], but the same has been simplified using partition via stop times instead of deterministic
partitions.
We will use the following inequalities on the exponential function which can be easily
proven using Taylor’s theorem with remainder. For a, b 2 R
1
|eb (1 + b)| e|b| |b|2
2
ia
|e 1| |a|
1 2 1
T
|eia (1 + ia a )| |a|3 .
2 6
Using these inequalities, we conclude that for a, b such that |a| , |b| , < loge (2),
we have
|eia+b (1 + ia
1 2
2
a + b)| F
|eia [eb (1 + b)] + b(eia 1) + (eia (1 + ia
1 2
a ))|
A
2 (3.2.2)
1 |b| 2 1 3
( e |b| + |b||a| + |a| )
2 6
2
(2|b| + |a| ).
R
Theorem 3.5. Let X be a continuous process adapted to a filtration (F⇧ ) and let Mt =
Xt2 t for t 0. Suppose that (i) X0 = 0, (ii) X is a (F⇧ )-martingale and (iii) M is a
(F⇧ ) martingale. Then X is a Brownian motion and further, for all s
D
{(Xt Xs ) : t s} is independent of Fs .
To see this, Corollary 2.58 and the hypothesis that M is a martingale imply that
Yt = X⌧2^t X 2^t (⌧ ^ t ^ t)
T
m
X1
Z ⌧m 1= (Z⌧k+1 Z⌧k ). (3.2.6)
k=0
Note that
E[(Z⌧k+1 Z⌧k )] = E[Z⌧k (ei
= E[Z⌧k E(ei
F
(X⌧k+1 X⌧k )+ 12
(X⌧k+1 X⌧k )+ 12
2 (⌧
2 (⌧
k+1
k+1
⌧k )
⌧k )
1)]
1 | F⌧k )]
A
Since E[X⌧k+1 X⌧k | F⌧k ] = 0 as X is a martingale and
E[(X⌧k+1 X⌧k ) 2 (⌧k+1 ⌧k ) | F⌧k ] = 0 (3.2.7)
R
1 2 1 2
2 (X⌧k+1 X⌧k ) 2 + 2 (⌧k+1 ⌧k )} | F⌧k ]
Using (3.2.2), (3.2.5) and the choice of , we can conclude that the expression on the right
hand side inside the conditional expectation is bounded by
2
((X⌧k+1 X⌧k )2 + (⌧k+1 ⌧k ))
Putting together these observations and (3.2.7), we conclude that
2
|E[(Z⌧k+1 Z⌧k )]| 2 E[(⌧k+1 ⌧k )].
As a consequence
2
|E[Z⌧m 1]| 2 E[⌧m ].
74 Chapter 3. The Ito Integral
1 2T
Now Z⌧m is bounded by e 2 and converges to Z , we conclude
2
|E[Z 1]| 2 T.
Since this holds for all small > 0 it follows that
E[Z ] = 1
and this completes the proof of step 2.
step 3: For s < t, 2 R,
1 2 (t
E[e(i (Xt Xs ))
| Fs ] = e 2
s)
. (3.2.8)
We have seen that Zt is a martingale and (3.2.8) follows from it since Xs is Fs measurable.
As a consequence
1 2
E[e(i (Xt Xs )) ] = e 2 (t s) . (3.2.9)
T
and so the distribution of Xt Xs is Gaussian with mean 0 and variance (t s).
step 4:
For s < t, , ✓ 2 R, a Fs measurable random variable Y we have
E[e(i (Xt Xs )+i✓Y )
F
] = E[e(i (Xt Xs ))
]E[e(i✓Y ) ].
The relation (3.2.10) is an immediate consequence of (3.2.8). We have already seen that
(3.2.10)
A
Xt Xs has Gaussian distribution with mean 0 and variance t s and (3.2.10) implies
that Xt Xs is independent of Fs , in particular, Xt Xs is independent of {Xu : u s}.
This completes the proof.
R
Indeed, using Theorem 3.5, we will show that (3.2.13) characterizes multidimensional
Brownian motion.
3.3. The Ito Integral 75
Theorem 3.6. Let W be an Rd -valued continuous process such that W0 = 0. Suppose (F⇧ )
is a filtration such that W is (F⇧ ) adapted. Suppose W satisfies
where X ✓ and M ✓ are defined via (3.2.11) and (3.2.12). Then W is a d-dimensional
Brownian motion and further, for 0 s t
(Wt Ws ) is independent of Fs .
This implies that W is a Brownian motion. Independence of (Wt Ws ) and Fs also follows
T
as in Theorem 3.5.
Theorems 3.5 and 3.6 are called Levy’s characterization of Brownian motion (one di-
where 0 = s0 < s1 < s2 < . . . < sm+1 < 1, aj is bounded Fsj 1 measurable random
variable for 1 j (m + 1), and a0 is bounded F0 measurable. Elements of S will be
D
R
called simple processes. For an f given by (3.3.1), we define X = f dW by
m
X
Xt (!) = aj+1 (!)(Wsj+1 ^t (!) Wsj ^t (!)). (3.3.2)
j=0
R
a0 does not appear on the right side because W0 = 0. It can be easily seen that f dW
defined via (3.3.1) and (3.3.2) for f 2 S does not depend upon the representation (3.3.1).
In other words, if g is given by
n
X
gt (!) = b0 (!)1{0} (s) + bj+1 (!)1(rj ,rj+1 ] (t) (3.3.3)
j=0
76 Chapter 3. The Ito Integral
where 0 = r0 < r1 < . . . < rn+1 and bj is Frj 1 measurable bounded random variable,
R R
1 j (n + 1), and b0 is bounded F0 measurable and f = g, then f dW = gdW i.e.
m
X
aj+1 (!)(Wsj+1 ^t (!) Wsj ^t (!))
j=0
n (3.3.4)
X
= bj+1 (!)(Wrj+1 ^t (!) Wrj ^t (!)).
j=0
Rt
By definition, X is a continuous adapted process. We will denote Xt as 0 f dW . We
will obtain an estimate on the growth of the integral defined above for simple f 2 S and
then extend the integral to an appropriate class of integrands - those that can be obtained
as limits of simple processes. This approach is di↵erent from the one adopted by Ito and
we have adopted this approach with an aim to generalize the same to martingales.
R
T
We first note some properties of f dW for f 2 S and obtain an estimate.
Proof. Let f, g have representations (3.3.1) and (3.3.3) respectively. Easy to see that we
(3.3.5)
A
can get 0 = t0 < t1 < . . . < tk such that
and then represent both f, g over common time partition. Then the result (3.3.5) follows
easily.
Rt Rt Rt
Lemma 3.8. Let f, g 2 S, and let Yt = 0 f dW , Zt = 0 gdW and At = 0 fs gs ds,
D
Proof. By linearity property (3.3.5) and the fact that sum of martingales is a martingale,
suffices to prove the lemma in the following two cases:
Case 1: 0 s < r and
Here in both cases, a, b are assumed to be bounded. In both the cases, Yt = a(Wt^r Wt^s ).
That Y is a martingale follows from Theorem 2.57. Thus in both cases, Y is an (F⇧ )-
martingale and similarly, so is Z. Remains to show that M is a martingale. In Case 1,
writing Nt = Wt2 t
Mt =ab((Wt^r Wt^s )2 (t ^ r t ^ s))
2 2
=ab((Wt^r Wt^s ) (t ^ r t ^ s) 2Wt^s (Wt^r Wt^s ))
=ab(Nt^r Nt^s ) 2abWt^s (Wt^r Wt^s ).
Recalling that N, W are martingales, it follows from Theorem 2.57 that M is a martingale
as
abWt^s (Wt^r Wt^s ) = abWs (Wt^r Wt^s ).
In case 2, recalling 0 s r u v, note that
T
Mt =a(Wt^r Wt^s )b(Wt^v Wt^u )
=a(Wr Ws )b(Wt^v Wt^u )
as Mt = 0 if t u. This completes the proof.
Rt
Theorem 3.9. Let f 2 S, Mt = 0 f dW and Nt = Mt2
(F⇧ )-martingales. Further, For any T < 1,
Z t
F
Z T
Rt 2
0 fs ds. Then M and N are
A
2
E[sup| f dW | ] 4E[ fs2 ds]. (3.3.6)
tT 0 0
Proof. The fact that M and N are martingales follow from Lemma 3.8. As a consequence
R
We will use the growth inequality (3.3.7) to extend the integral to a larger class of
functions that can be approximated in the norm defined by the right hand side in (3.3.7).
Each f 2 S can be viewed as a real valued function on ⌦ e = [0, 1) ⇥ ⌦. It is easy to
e
see that S is an algebra. Let P be the -field generated by S, i.e. the smallest -field on ⌦
such that every element of S is measurable w.r.t. P. Then by Theorem 2.65, the class of
functions that can be approximated by S in the norm
Z T
1
(E[ fs2 ds]) 2
0
78 Chapter 3. The Ito Integral
T
E[( sup |Yt hdW |)2 ] 4E[ (fs hs )2 ds] 8T < 1. (3.3.9)
0tT 0 0
Further, Y and Z are (F⇧ )-martingales where
Z t
Zt = Yt2
F 0
fs2 ds.
for every r > 0 and hence for integers m 1, we can get f m 2 S such that
kf f m k2,µm 2 m 1
. (3.3.10)
Using k·k2,µr k·k2,µs for r s it follows that for k 1
D
kf m+k f m k2,µm 2 m
. (3.3.11)
Denoting the L2 (⌦, F, P) norm by k·k2,P , the growth inequality (3.3.6) can be rewritten as,
for g 2 S, m 1, Z t
k sup | gdW | k2,P 2kgk2,µm (3.3.12)
0tm 0
Rt Rt
Recall that f k 2 S and hence 0 f k dW is already defined. Let Ytk = 0 f k dW . Now using
(3.3.11) and (3.3.12), we conclude that for k 1
k[ sup |Ytm+k Ytm | ]k2,P 2 m+1
. (3.3.13)
0tm
3.3. The Ito Integral 79
and hence
1
X 1
X
k[ sup |Ytm+1 Ytm | ]k2,P k[ sup |Ytm+1 Ytm | ]k2,P
m=n 0tn m=n 0tn
X1
m+1
2
m=n
<1.
Hence,
1
X
[ sup |Ytm+1 Ytm |] < 1 a.s. P. (3.3.15)
T
m=n 0tn
So let
1
X
Nn = {! : [ sup |Ytm+1 (!) Ytm (!)|] = 1}
m=n 0tn
and let N = [1 F
n=1 Nn . Then N is a P-null set. For ! 62 N , let us define
and for ! 2 N , let Yt (!) = 0. It follows from (3.3.15) that for all T < 1, ! 62 N
Thus Y is a process with continuous paths. Now using (3.3.12) for f m h 2 S we get
Z t Z T
m 2
E[( sup |Yt hdW |) ] 4E[ (f m h)2s ds]. (3.3.17)
0tT 0 0
D
As a consequence,
Z t
E[ |(fs )2 (fsm )2 |ds] ! 0
0
and hence Z Z
t t
E[| (fsm )2 ds fs2 ds| ] ! 0.
0 0
Rt n 2
By Theorem 3.9 , we have Y n and Z n are martingales where Ztn = (Ytn )2 0 (fs ) ds.
n 1 n 1
As observed above Yt converges in L (P) to Yt and Zt converges in L (P) to Zt for each
t and hence Y and Z are martingales.
Remark 3.11. From the proof of the Lemma it also follows that Y is uniquely determined by
the property (3.3.9), for if Ỹ is another process that satisfies (3.3.9), then using it for h = f m
as in the proof above, we conclude that Y m converges almost surely to Ỹ and hence Y = Ỹ .
T
RT
Definition 3.12. For a predictable process f such that E[ 0 fs2 ds] < 1 8T < 1, we define
Rt
the Ito integral 0 f dW to be the process Y that satisfies (3.3.9).
The next result gives the basic properties of the Ito integral
essentially been proved above.
F R
f dW , most of them have
A
Theorem 3.13. Let f, g be predictable processes satisfying (3.3.8).Then
Z t Z t Z t
(af + bg)dW = a f dW + b gdW. (3.3.18)
0 0 0
R
Rt Rt
Let Mt = 0 f dW and Nt = Mt 0 fs2 ds. Then M and N are (F⇧ )-martingales. Further,
for any T < 1,
Z t Z T
2
E[sup| f dW | ] 4E[ fs2 ds]. (3.3.19)
D
tT 0 0
Proof. The linearity (3.3.18) follows by linearity for the integral for simple functions ob-
served in Lemma 3.7 and then for general predictable processes via approximation. That
M , N are martingales has been observed in Lemma 3.10. The Growth inequality (3.3.19)
follows from (3.3.9).
R
Remark 3.14. For a bounded predictable process f , let IW (f ) = f dW . Then the Growth
inequality (3.3.19) and linearity of IW implies that for fn , f bounded predictable processes
bp ucp
fn ! f implies IW (fn ) ! IW (f ).
3.4. Multidimensional Ito Integral 81
T
Rt Rt
Exercise 3.16. Let ft = t. Show that 0 f dW = tWt 0 Ws ds.
Exercise 3.17. Let A 2 V be a bounded r.c.l.l. adapted process with finite variation paths.
Show that Z t
A dW = At Wt
Z t
0
W dA.
F (3.3.20)
0
A
Hint: Let tni = i2 n, i 0, n 1. Observe that
1
X 1
X
Atni ^t (Wtni+1 ^t Wtni ^t ) = At Wt Wtni+1 ^t (Atni+1 ^t Atni ^t ).
R
i=0 i=0
Rt
The left hand side converges to 0 A dW while the second term on right hand side converges
Rt
to 0 W dA as seen in Exercise 2.77.
D
Remark 3.18. The Ito integral can be extended to a larger class of predictable integrands f
satisfying Z T
fs2 ds < 1 a.s. 8T < 1.
0
We will outline this later when we discuss integration w.r.t. semimartingales.
are independent. Suppose further that (F⇧ ) is a filtration such that W is (F⇧ ) adapted and
for each s, {Wt Ws : t s} is independent of Fs (and hence (3.2.14) holds). Recall that
defining
Xd
✓
Xt = ✓j Wtj (3.4.1)
j=1
The argument given in the proof of next lemma is interesting and will be used later.
Throughout this section, the filtration (F⇧ ) will remain fixed.
T
Proof. Let Xt = p12 (Wtj + Wtk ). Then, as seen above X is a Brownian motion and hence
Xt2 t is a martingale. Note that
Xt2
1
2
1
F
t = [(Wtj )2 + (Wtk )2 + 2Wtj Wtk ] t
1
= [(Wtj )2 t] + [(Wtk )2 t] + Wtj Wtk .
A
2 2
Since the left hand side above as well as the first two terms of right hand side above are
martingales, it follows that so is the third term.
R
Xtj = f jk dW k .
k=1 0
Proof. Let Xt = p12 (Wtj + Wtk ). Then, as seen above X is a Brownian motion. For simple
functions f it is easy to check that
Z t Z t Z t
1
f dX = p ( f dW j + f dW k )
0 2 0 0
and hence for all f satisfying (3.3.8) via approximation. Thus,
Z t
1
hdX = p (Ytj + Ytk )
0 2
Rt R t 2
and so ( 0 hdX)2 0 hs ds is a martingale. Now as in the proof of Lemma 3.19
Z t Z t Z t
2 2 1 j 2 k 2 j k
( hdX) hs ds = [(Yt ) + (Yt ) + 2Yt Yt ] h2s ds
0 0 2 0
Z t Z t
1 1
= [(Ytj )2 h2s ds] + [(Ytk )2 h2s ds] + Ytj Ytk .
2 0 2 0
T
and once again the left hand side as well as the first two terms on the right hand side are
martingales and hence so is the last term completing the proof.
Rt
Lemma 3.21. Let f, g be predictable processes satisfying (3.3.8). Let Ytk = 0 f dW k ,
Rt
F
Ztk = 0 gdW k . Then for j 6= k, Ytk Ztj is a martingale.
Proof. Let Xt = Ytk Ztj . We will first prove that X is a martingale when f, g are simple, the
A
general case follows by approximation. The argument is similar to the proof of Theorem
3.8. By linearity, suffices to prove the required result in the following cases.
Case 1: 0 s < r and
R
In Case 1,
Z t Z t
Xt = Ytk Ztj = ab(Wr^t
k k
Ws^t j
)(Wr^t j
Ws^t )=( hdW k )( hdW j )
0 0
p
where h = (ab)1(s,r] (t) and hence by Lemma 3.20, X is a martingale.
In Case 2,
Xt = Ytk Ztj = ab(Wr^t
k k
Ws^t j
)(Wv^t j
Wu^t ).
k
with ⇠ = ab(Wr^t k ) is F measurable and hence by Theorem 2.57, X is a martingale.
Ws^t u
This proves the required result for simple predictable processes f, g. The general case
follows by approximating f, g by simple predictable processes {f n }, {g n } such that for all
T <1 Z T
[|fsn fs |2 + |gsn gs |2 ]ds ! 0.
0
T
Rt Rt
Ytk Ztj = ( 0 f dW k )( 0 gdW j ) in L1 (P) and thus Ytk Ztj is a martingale.
We are now ready to prove the main growth inequality. Recall that qP L(d, m) denotes
Pm 2
2
j=1 xj denotes
0
we have Z Z
t T
2
E[ sup | f dW | ] 4E[ kfs k2 ds]. (3.4.4)
0tT 0 0
Pd R
Proof. Let Xtj = k=1 f jk dW k . Then
d Z
X t d
X Z t Z t
(Xtj )2 jk 2
(f ) ds = [( f jk dW k )2 (f jk )2 ds]
k=1 0 k=1 0 0
d X
X d Z t Z t
jk k
+ 1{k6=l} f dW f jl dW l
l=1 k=1 0 0
3.5. Stochastic Di↵erential Equations 85
and each term on the right hand side above is a martingale and thus summing over j we
conclude Z t Z t
| f dW |2 kfs k2 ds
0 0
is a martingale. Thus Z Z
t t
E[| f dW |2 ] = E[ kfs k2 ds] (3.4.5)
0 0
Rt
and | 0 f dW |2 is a submartingale. The required estimate (3.4.4) now follows from Doob’s
maximal inequality (2.24) and (3.4.5).
T
tions (SDE) of the type
dXt = (t, Xt )dWt + b(t, Xt )dt. (3.5.1)
The equation (3.5.1) is to be interpreted as an integral equation :
Xt = X0 +
Z t
(s, Xs )dWs +
0
Z t
F
b(s, Xs )ds.
We will need the following lemma, known as Gronwall’s lemma for proving uniqueness of
solution to (3.5.2) under the Lipschitz conditions.
Lemma 3.23. Let (t) be a bounded measurable function on [0, T ] satisfying, for some
0 a < 1, 0 < b < 1,
Z t
(t) a + b (s) ds, 0 t T. (3.5.6)
0
Then
(t) aebt . (3.5.7)
Proof. Let
Z t
bt
g(t) = e (s) ds.
T
0
where almost everywhere refers to the Lebesgue measure on R. Using (3.5.6), it follows
that
A
g 0 (t) ae bt
a.e.
Hence (using g(0) = 0 and that g is absolutely continuous) g(t) ab (1 e bt ) from which
R
we get
Z t
a
(s) ds (ebt 1).
0 b
The conclusion (t) aebt follows immediately from (3.5.6).
D
Note that in view of the growth condition (3.5.5) the Ito integral above is defined. Using
the growth estimate (3.4.4) we see that
Z T
2 2
E[ sup |⇠t | ] =3[E[|Y0 | ] + 4E[ k (s, Ys )k2 ds]
0tT 0
Z T
+ E[( |b(s, Ys )|ds)2 ]]
0
Z T
2
3E[|Y0 | ] + 3KT2 (4 + T) (1 + E[|Ys |2 ])ds
0
and hence ⇠ 2 Km . Let us define a mapping ⇤ from Km into itself as follows: ⇤(Y ) = ⇠
where ⇠ is defined by (3.5.8). Thus solving the SDE (3.5.2) amounts to finding a fixed
point Z of the functional ⇤ with Z0 = X0 , where X0 is pre-specified. We are going to
prove that given X0 , there exists a unique solution (or a unique fixed point of ⇤) with the
T
given initial condition. The following lemma is an important step in that direction.
Lemma 3.24. Let Y, Z 2 Km and let ⇠ = ⇤(Y ) and ⌘ = ⇤(Z). Then for 0 t T one
has Z t
2 2 2
E[ sup |⇠s ⌘s | ] 3E[|Y0 Z0 | ] + 3CT (4 + T )
0st
t
+ E[( b(s, Zs )|ds)2 ]]
|b(s, Ys )
0
Z t
2 2
3E[|Y0 Z0 | ] + 3CT (4 + T ) E[|Ys Zs |2 ]ds.
0
Theorem 3.26. Suppose , b satisfy conditions (3.5.3) and (3.5.4) and X0 is a F0 mea-
surable Rm -valued random variable with E[|X0 |2 ] < 1. Then there exists a process X such
RT
that E[ 0 |Xs |2 ds] < 1 8T < 1 and
Z t Z t
Xt = X0 + (s, Xs )dWs + b(s, Xs )ds. (3.5.9)
0 0
RT
Further if X̃ is another process such that X̃0 = X0 , E[ 0 |X̃s |2 ds] < 1 for all T < 1 and
Z t Z t
X̃t = X̃0 + (s, X̃s )dWs + b(s, X̃s )ds
0 0
then X = X̃, i.e. P(Xt = Yt 8t) = 1.
Proof. Let us first prove uniqueness. Let X and X̃ be as in the statement of the theorem.
Then, using Lemma (3.24) it follows that
T
u(t) = E[ sup|Xs X̃s |2 ]
st
t
E[|Xs X̃s |2 ]ds.
A
u(t) 3CT2 (4 + T) u(s)ds, 0 t T.
0
By (Gronwall’s) Lemma (3.23), it follows that u(t) = 0, 0 t T for every T < 1. Hence
X = X̃.
R
We will now construct a solution. Let Xt1 = X0 for all t 0. Note that X 1 2 Km .
Now define X n inductively by
X n+1 = ⇤(X n ).
D
Note that X0n = X01 = X0 for all n 1 and hence from Lemma (3.24) it follows that for
n 2, for 0 t T ,
Z t
E[sup|Xsn+1 Xsn |2 ] 3CT2 (4 + T ) E[|Xsn Xsn 1 2
| ]ds
st 0
3.5. Stochastic Di↵erential Equations 89
As seen in (3.5.10),
u1 (t) 2KT2 (4 + T )(1 + E[|X0 |2 ])t
and hence using (3.5.11), which is true for n 2, we can deduce by induction on n that
for a constant C̃T = 3(CT2 + KT2 )(4 + T )(1 + E[|X0 |2 ])
(C̃T )n tn
un (t) , 0 t T.
n!
P p
Thus n un (T ) < 1 for every T < 1 which is same as
T
1
X
ksup|Xsn+1 Xsn |k2 < 1 (3.5.12)
n=1 sT
kZk2 denoting the L2 (P) norm here. The relation (3.5.12) implies
k[
1
X
sup|Xsn+1
n=1 sT
F Xsn | ]k2 < 1 (3.5.13)
A
as well as
n+k
X
supk[sup|Xsn+k Xsn | ]k2 supk[ sup|Xsj+1 Xsj | ]k2
R
! 0 as n tends to 1.
P1
Let N = [1 T =1 {! :
n+1 (!)
n=1 supsT |Xs Xsn (!)| = 1}. Then by (3.5.13) P(N ) = 0
and for ! 62 N , Xsn (!) converges uniformly on [0, T ] for every T < 1. So let us define X
as follows: 8
<lim n c
n!1 Xt (!) if ! 2 N
Xt (!) =
:0 if ! 2 N.
(3.5.14) we get
n+k
X
k[sup|Xs Xsn | ]k2 lim inf k[ sup|Xsj+1 Xsj | ]k2
sT k!1
j=n sT
1
X (3.5.15)
[ ksup|Xsj+1 Xsj |k2 ]
j=n sT
! 0 as n tends to 1.
In particular, X 2 Km . Since ⇤(X n ) = X n+1 by definition, (3.5.15) also implies that
while (3.5.15) and Corollary 3.25 (remembering that X0n = X0 for all n) imply that
T
n!1 sT
From (3.5.16) and (3.5.17) it follows that X = ⇤(X) or that X is a solution to the SDE
(3.5.9).
F
A
R
D
Chapter 4
Stochastic Integration
In this chapter we will pose the question as to what processes X are good integrators: i.e.
Z t
T
JX (f )(t) = f dX
0
can be defined for a suitable class of integrands f and has some natural continuity prop-
erties. We will call such a process a Stochastic integrator. In this chapter, we will prove
F
Rt
basic properties of the stochastic integral 0 f dX for a stochastic integrator X.
In the rest of the book, (⌦, F, P) will denote a complete probability space and (F⇧ ) will
denote a filtration such that F0 contains all null sets in F. All notions such as adapted,
A
stop time, martingale will refer to this filtration unless otherwise stated explicitly.
For some of the auxiliary results, we need to consider the corresponding right continuous
filtration (F⇧+ ) = {Ft+ : t 0} where
R
Ft+ = \s>t Fs .
We begin with a discussion on the predictable -field.
D
91
92 Chapter 4. Stochastic Integration
appeared naturally in the definition of the stochastic integral w.r.t. Brownian motion and
play a very significant role in the theory of stochastic integration with respect to general
semimartingales as we will see. A process f will be called a predictable process if it is P
measurable. Of course, P depends upon the underlying filtration and would refer to the
filtration that we have fixed. If there are more than one filtration under consideration, we
will state it explicitly. For example P(G· ) to denote the predictable -field corresponding
to a filtration (G· ) and S(G· ) to denote simple predictable process for the filtration (G· ).
The following proposition lists various facts about the -field P.
(i) Let f be P measurable. Then f is (F⇧ ) adapted. Moreover, for every t < 1, ft is
([s<t Fs )-measurable.
T
(ii) Let Y be a left continuous adapted process. Then Y is P measurable.
(iii) Let T be the class of all bounded adapted continuous processes. Then P = (T) and
F e P).
the smallest bp- closed class that contains T is B(⌦,
(iv) For any stop time ⌧ , U = 1[0,⌧ ] (i.e. Ut = 1[0,⌧ ] (t)) is P measurable.
A
(v) For an r.c.l.l. adapted process Z and a stop time ⌧ , the process X defined by
is predictable.
(vi) For a predictable process g and a stop time ⌧ , g⌧ is a random variable and h defined
D
by
ht = g⌧ 1(⌧,1) (t) (4.1.3)
is itself predictable.
Proof. It suffices to prove the assertions assuming that the processes f , Y , g are bounded
(by making a tan 1 transformation, if necessary). Now, for (i) let
e P) : ft is ([s<t Fs )- measurable}.
K1 = {f 2 B(⌦,
It is easily seen that K1 is bp-closed and contains S and thus by Theorem 2.64 equals
e P) proving (i).
B(⌦,
4.1. The Predictable -Field 93
For (ii), given a left continuous bounded adapted process Y , let Y n be defined by
n2 n
X
Ytn = Y0 1{0} (t) + Y k 1( kn , k+1
n ]
(t). (4.1.4)
2n 2 2
k=0
bp
Then Y n 2 S and Y n ! Y and this proves (ii).
For (iii), Let K2 be the smallest bp-closed class containing T. From part (ii) above, it
follows that T ✓ B(⌦,e P) and hence K2 ✓ B(⌦, e P). For t 2 [0, 1), n 1 let
T
f (s) = a0 1{0} (s) + aj+1 1(sj ,sj+1 ] (s)
j=0
where 0 = s0 < s1 < s2 . . . < sm+1 < 1, aj+1 is bounded Fsj measurable random variable,
n n
Ys = a0 (s) +
Xm F
0 j m, and a0 is bounded F0 measurable random variable let
s s
aj+1 n ( sj+1 jsj ).
j=0
A
bp
Then it follows that Y n 2 T and Y n ! Y . Thus S ✓ K2 and hence B(⌦, e P) ✓ K2
completing the proof of (iii).
For part (iv) note that U is adapted left continuous process and hence P measurable
R
by part (ii).
For (v), suffices to prove that X is adapted since X is left continuous by construction.
We have shown earlier (Lemma 2.36) that Z⌧ is F⌧ measurable and hence Da,t = {Z⌧
D
If X is P(F⇧Y ) measurable, then part (i) of the result proven above says that for every
t, Xt is measurable w.r.t. (Yu : 0 u < t). Thus having observed Yu , u < t, the value Xt
can be known (predicted with certainty) even before observing Yt . This justifies the name
predictable -field for P.
Theorem 4.3. Let f be a (F⇧+ )-predictable process (i.e. P(F⇧+ ) measurable). Then g
defined by
gt (!) = ft (!)1{(0,1)⇥⌦} (t, !) (4.1.5)
T
is a (F⇧ )-predictable process.
Proof. Let f a (F⇧+ )-adapted bounded continuous process. A crucial observation is that
F
then for t > 0, ft is Ft measurable and thus defining for n 1
hnt = (nt ^ 1)ft
bp
it follows thhat hn are (F⇧ ) adapted continuous processes and hn ! g where g is defined
A
by (4.1.5) and hence in this case g is (F⇧ ) predictable.
Now let H be the class of bounded P(F⇧+ ) measurable f for which the conclusion is
true. Then easy to see that H is an algebra that is bp-closed and contains all (F⇧+ )-adapted
R
continuous processes. The conclusion follows by the monotone class theorem, Theorem
2.64.
D
Corollary 4.4. Let f be a process (F⇧+ )-predictable such that f0 is F0 measurable. Then
f is (F⇧ )-predictable.
The following example will show that the result may not be true if f0 = 0 is dropped
in the Corollary 4.4 above.
4.2. Stochastic Integrators 95
Exercise 4.6. Let ⌦ = C([0, 1) and let Xt denote the coordinate process and Ft = (Xs :
0 s t). Let A be the set of all ! 2 ⌦ that take positive as well as negative values in
(0, ✏) for every ✏ > 0. Show that A 2 F0+ but A does not belong to F0 . Use this to show the
relevance of the hypothesis on f0 in the Corollary given above.
T
m
X
f (s) = a0 1{0} (s) + aj+1 1(sj ,sj+1 ] (s) (4.2.1)
j=0
where 0 = s0 < s1 < s2 < . . . < sm+1 < 1, aj is bounded Fsj 1 measurable random
by
F
variable, 1 j (m + 1), and a0 is bounded F0 measurable.
For a simple predictable f 2 S given by (4.2.1), let JX (f ) be the r.c.l.l. process defined
A
m
X
JX (f )(t) = a0 X0 + aj+1 (Xsj+1 ^t Xsj ^t ). (4.2.2)
j=0
One needs to verify that JX is unambiguously defined on S. i.e. if a given f has two
R
representations of type (4.2.1), then the corresponding expression in (4.2.2) agree. This as
well as linearity of JX (f ) for f 2 S can be verified using elementary algebra. By definition,
for f 2 S, JX (f ) is an r.c.l.l. adapted process. In analogy with the Ito integral with respect
D
to Brownian motion discussed in the earlier chapter, we wish to explore if we can extend JX
to the smallest bp-closed class of integrands that contain S. Each f 2 S can be viewed as a
real valued function on ⌦ e = [0, 1) ⇥ ⌦. Since P is the -field generated by S, by Theorem
2.64, the smallest class of functions that contains S and is closed under bp-convergence is
e P).
B(⌦,
When the space, filtration and the probability measure are clear from the context, we
will write the class of adapted r.c.l.l. processes R0 (⌦, (F⇧ ), P) simply as R0 .
Definition 4.8. An r.c.l.l. adapted process X is said to be a Stochastic integrator if the map-
e P) 7! R0 (⌦, (F⇧ ), P) satisfying
ping JX from S to R0 (⌦, (F⇧ ), P) has an extension JX : B(⌦,
96 Chapter 4. Stochastic Integration
It should be noted that for a given r.c.l.l. process X, JX may not be continuous on S.
See the next Exercise. So this definition , in particular, requires that JX is continuous on
S and has a continuous extension to B(⌦,e P). Though not easy to prove, continuity of JX
on S does imply that it has a continuous extension and hence it is a stochastic integrator.
We will see this later in Section 5.8.
Exercise 4.9. Let G be a real valued function from [0, 1) with G(0) = 0 that does not
have bounded variation. So for some T < 1, Var[0,T ] (G) = 1. We will show that JG is not
continuous on S. From the definition of Var[0,T ] (G), it follows that we can get a sequence of
partitions of [0, T ]
T
0 = tm m m
0 < t1 < . . . < tnm = T (4.2.4)
such that
Let sgn : R !
↵m = (
nX
m 1
j=0
0 and sgn(x) =
(4.2.5)
(ii) JG (f m ) converges to 1.
0 from
Theorem 4.10. Let X be an r.c.l.l. process. Suppose there exist mappings JX , JX
e P)into R0 (⌦, (F⇧ ), P), such that for f 2 S (given by (4.2.1)),
B(⌦,
m
X
0
JX (f )(t) = JX (f )(t) = a0 X0 + aj+1 (Xsj+1 ^t Xsj ^t ). (4.2.6)
j=1
0 satisfy (4.2.3). Then
Further suppose that both JX , JX
0
P[JX (f )(t) = JX e P).
(f )(t) 8t] = 1 8f 2 B(⌦,
4.2. Stochastic Integrators 97
is bp-closed and by our assumption (4.2.6), contains S. Since P = (S), by Theorem 2.64
e P).
it follows that K1 = B(⌦,
The following result which is almost obvious in this treatment of stochastic integration
is a deep result in the traditional approach to stochastic integration and is known as
Stricker’s theorem.
Theorem 4.11. Let X be a stochastic integrator for the filtration (F⇧ ). Let (G⇧ ) be a
filtration such that Ft ✓ Gt for all t. Suppose X is a stochastic integrator for the filtration
(G⇧ ) as well. Denoting the mapping defined by (4.2.2) for the filtration (G⇧ ) and its extension
T
by HX , we have
e P(F⇧ )).
JX (f ) = HX (f ) 8f 2 B(⌦, (4.2.7)
F
as (4.2.3). Thus (4.2.7) follows from uniqueness of extension, Theorem 4.10.
e P),
Theorem 4.13. Let X be a stochastic integrator. Then f, g 2 B(⌦,
implies
P(! 2 ⌦ : JX (f )t (!) = JX (g)t (!) 8t 0) = 1. (4.2.10)
In other words, the mapping JX maps equivalence classes of process under the relation
f = g (see Definition (2.2)) to equivalence class of processes.
98 Chapter 4. Stochastic Integration
Then the assumption that F0 contains all P-null sets implies that ⌦0 2 F0 and thus ⇠ = 1⌦0
is F0 measurable and ⇠f = ⇠g in the sense that these are identical processes:
Now we have
⇠JX (f ) = JX (⇠f )
= JX (⇠g)
= ⇠JX (g).
Since P(⇠ = 1) = 1, (4.2.10) follows.
T
Corollary 4.14. Let fn , f be bounded predictable be such that there exists ⌦0 ✓ ⌦ with
P(⌦0 ) = 1 and such that
Then
F
bp
1⌦0 fn ! 1⌦0 f.
ucp
A
JX (fn ) ! JX (f ).
Then it is well known and easy to see that (for a sequence of random variables) convergence
D
We have thus seen that the stochastic integrator does not quite depend upon the under-
lying probability measure. We have also seen that it does not depend upon the underlying
filtration either - see Theorem 4.11.
R
For a stochastic integrator X, we will be defining the stochastic integral Y = f dX,
for an appropriate class of integrands, given as follows:
4.2. Stochastic Integrators 99
Definition 4.16. For a stochastic integrator X, let L(X) denote the class of predictable
processes f such that
ucp
e P), hn ! 0 pointwise, |hn | |f | ) JX (hn )
hn 2 B(⌦, ! 0. (4.2.11)
From the definition, it follows that if f 2 L(X) and g is predictable such that |g| |f |,
then g 2 L(X). Here is an interesting consequence of this definition.
T
Given any subsequences {mk }, {nk } of integers, increasing to 1, let
1 k k
hk = (g m g n ).
2
ucp
Then {hk } satisfies (4.2.11) and hence JX (hk )
k
(JX (g m )
F ! 0 and as a consequence
k
JX (g n ))
ucp
! 0. (4.2.13)
Since (4.2.13) holds for all subsequences {mk }, {nk } of integers, increasing to 1, it follows
A
that JX (g n ) is Cauchy. Conversely, suppose f satisfies (4.2.12). Given {hn } as in (4.2.11),
let a sequence g k be defined as g 2k = hk and g 2k 1 = 0 for k 1. Then g k converges
to g = 0 and hence JX (g k ) is Cauchy. Since for odd integers n, JX (g n ) = 0 and thus
R
JX (g 2k ) = JX (hk ) converges to 0.
R
This result enables us to define f dX for f 2 L(X).
D
Note that in the result given above, we did not require g to be bounded. Even if g were
T
bounded, the convergence was not required to be bounded pointwise.
R
The process f dX is called the stochastic integral of f with respect to X and we will
also write Z Z
Rt
( F
f dX)t =
0
t
f dX.
We interpret 0 f dX as the definite integral of f with respect to X over the interval [0, t].
A
We sometime need the integral of f w.r.t. X over (0, t] and so we introduce
Z t Z t Z t
f dX = f 1(0,1) dX = f dX f0 X0 .
0+ 0 0
R Rt
R
the mapping t 7! Xt (!) satisfies Var[0,T ] (X· (!)) < 1 for all ! 2 ⌦, for all T < 1.
Theorem 4.21. Let X 2 V be a process with finite variation paths, i.e. X be an r.c.l.l.
adapted process such that
Var[0,T ] (X· (!)) < 1 for all T < 1. (4.2.15)
Then X is a stochastic integrator. Further, for f 2 B(⌦, e P) the stochastic integral JX (f ) =
R
f dX is the Lebesgue-Stieltjes integral for every ! 2 ⌦:
Z t
JX (f )(t)(!) = f (s, !)dXs (!) (4.2.16)
0
where the integral above is the Lebesgue-Stieltjes integral.
4.3. Properties of the Stochastic Integral 101
Proof. For f 2 S, the right hand side of (4.2.16) agrees with the specification in (4.2.1)-
(4.2.2). Further the dominated convergence theorem (for Lebesgue-Stieltjes integral) im-
plies that the right hand side of (4.2.16) satisfies (4.2.3) and thus X is a stochastic integrator
and (4.2.16) is true.
Remark 4.22. If X 2 V and At = |X|t is the total variation of X on [0, t] and f is predictable
such that Z t
|fs |dAs < 1 8t < 1 a.s. (4.2.17)
0
then f 2 L(X) and the stochastic integral is the same as the Lebesgue-Stieltjes integral. This
follows from the dominated convergence theorem and Theorem 4.21. However, L(X) may
include processes f that may not satisfy (4.2.17). We will return to this later.
T
Remark 4.23. Suppose X is an r.c.l.l. adapted such that
ucp
f n ! 0 uniformly ) JX (f n ) ! 0. (4.2.18)
F
Of course every stochastic integrator satisfies this property. Let S1 denote the class of f 2 S
(simple functions) that are bounded by 1. For T < 1 the family of random variables
Z T
A
{ f dX : f 2 S1 } (4.2.19)
0
is bounded in probability or tight in the sense that
Z T
R
8" > 0 9 M < 1 such that sup P(| f dX| M ) ". (4.2.20)
f 2S1 0
To see this, suppose (4.2.18) is true but (4.2.20) is not true. Then we can get an " > 0 and
for each m 1, f m 2 S with |f m | 1 such that
D
Z T
P(| f m dX| m) ". (4.2.21)
0
1 m
RT
Writing g m = mf , it follows that g m ! 0 uniformly but in view of (4.2.21), 0 g m dX does
not converge to zero in probability- which contradicts (4.2.18). Indeed, the apparently waeker
property (4.2.18) characterises stochastic integrators as we will see later. See Theorem 5.78.
Proof. We will begin by showing that (4.3.1) is true for f, g bounded predictable processes.
For a bounded predictable process f , let
Z Z Z
T
e
K(f ) = {g 2 B(⌦, P) : (↵f + g)dX = ↵ f dX + gdX, 8↵, 2 R}.
If f 2 S, it easy to see that S ✓ K(f ) and Theorem 4.20 implies that K(f ) is bp-closed.
e P).
Hence invoking Theorem 2.64, it follows that K(f ) = B(⌦,
Now we take f 2 B(⌦, F
e P) and the part proven above yields S ✓ K(f ). Once again,
e P). Thus (4.3.1) is true when
using that K(f ) is bp-closed we conclude that K(f ) = B(⌦,
A
f, g are bounded predictable process.
Now let us fix f, g 2 L(X). We will show (|↵f | + | g|) 2 L(X), let un be bounded
predictable processes converging to u pointwise and
R
Let v n = un 1{|↵f || g|} and wn = un 1{|↵f |>| g|} . Then v n and wn converge pointwise to
v = u1{|↵f || g|} and w = u1{|↵f |>| g|} respectively and further
D
|v n | 2| g|
|wn | 2|↵f |.
Note that since v n , wn are bounded and un = v n + wn , from the part proven above, we
have Z Z Z
v n dX + wn dX = un dX.
R R
Since f, g 2 L(X), it follows that v n dX and wn dX are Cauchy in ducp metric and hence
R
so is their sum un dX. Thus (|↵f |+| g|) 2 L(X) and as a consequence, (↵f + g) 2 L(X)
as well.
4.3. Properties of the Stochastic Integral 103
Now let
f n = f 1{|f |n} , g n = g 1{|g|n} .
R R R R
Then by definition, f n dX converges to f dX and g n dX converges to gdX in ducp
metric. Also (↵f n + g n ) are bounded predictable processes, converge pointwise to (↵f + g)
and is dominated by (|↵f | + | g|) 2 L(X). Hence by Theorem 4.20 we have
Z Z
n n ucp
(↵f + g )dX ! (↵f + g)dX.
On the other hand, the validity of (4.3.1) for bounded predictable processes yields
Z Z Z
(↵f n + g n )dX = ↵f n dX + g n dX
Z Z
= ↵ f n dX + g n dX
Z Z
T
ucp
! ↵ f dX + gdX.
is their sum, which equals g d(↵X + Y ) = g n dZ. Thus f 2 L(Z). The equation
(4.3.2) follows by using (4.3.2) for the bounded process f n = f 1{|f |n} and passing to the
limit.
D
Thus the class of stochastic integrators is a linear space. We will see later that it is
R
indeed an Algebra. Let us note that when X is a continuous process, then so is f dX.
Proof. Let Z
e
K = {f 2 B(⌦, P) : f dX is a continuous process}.
R
By using the definition of f dX it is easy to see that S ✓ K. Also that K is bp-closed
ucp
since Z n continuous, Z n ! Z implies Z is also continuous. Hence invoking Theorem 2.64
104 Chapter 4. Stochastic Integration
we conclude K = B(⌦, e P). The general case follows by noting that limit in ducp metric of
R
continuous process is a continuous process and using that for f 2 L(X), f dX is the limit
R
in ducp -metric of f 1{|f |n} dX.
T
Proof. Let g n = hn 1{|hn |n} and f n = hn 1{|hn |>n}
Note that in view of (4.3.4) and the assumption f 2 L(X) it follows that hn 2 L(X).
R R F
Pointwise convergence of hn to h also implies |h| |f | which in turn yields h 2 L(X).
Thus hn dX, hdX are defined. Clearly, g n ! h pointwise and also f n ! 0 pointwise.
Further, hn = g n + f n , |g n | |f | and |f n | |f |.
R ucp R
A
From Theorem 4.20 it follows that g n dX ! hdX and from the definition of L(X),
R ucp
it follows that f n dX ! 0. Now linearity of the stochastic integral, Theorem 4.24, shows
that (4.3.5) is true.
R
Remark 4.27. The condition (4.3.3) that hn ! h pointwise can be replaced by requiring
that convergence holds pointwise outside a null set, namely that there exists ⌦0 ✓ ⌦ with
P(⌦0 ) = 1 such that
D
It should be noted that the hypothesis in the dominated convergence theorem given
above are exactly the same as in the case of Lebesgue integrals.
Recall, for an r.c.l.l. process X, X denotes the l.c.r.l. process defined by Xt = X(t ),
i.e. the left limit at t with the convention X(0 ) = 0 and X = X X . Note that
( X)t = 0 at each continuity point and equals the jump otherwise. Note that by the above
convention
( X)0 = X0 .
4.3. Properties of the Stochastic Integral 105
The next result connects the jumps of the stochastic integral with the jumps of the inte-
grator.
Theorem 4.28. Let X be a stochastic integrator and let f 2 L(X). Then we have
Z
( f dX) = f · ( X). (4.3.7)
Proof. For f 2 S, (4.3.7) can be verified from the definition. Now the class K of f such that
(4.3.7) is true can be seen to be bp-closed and hence is the class of all bounded predictable
processes. The case of general f 2 L(X) can be completed as in the proof of Theorem
4.25.
Rt
The equation (4.3.7) is to be interpreted as follows: if Yt = 0 f dX then ( Y )t =
ft ( X)t .
T
We have already seen that the class of stochastic integrators (with respect to a filtration
on a given probability space) is a linear space.
R
Theorem 4.29. Let X be a stochastic integrator and let f 2 L(X). Let Y = f dX. Then
Z t
0
gdY =
Z tF
Y is also a stochastic integrator and for a bounded predictable process g,
gf dX, 8t.
0
(4.3.8)
A
Further, for a predictable process g, g 2 L(Y ) if and only if gf 2 L(X) and then (4.3.8)
holds.
m
X
f (s) = a0 1{0} (s) + aj+1 1(sj ,sj+1 ] (s) (4.3.9)
j=0
D
n
X
g(s) = b0 1{0} (s) + bj+1 1(tj ,tj+1 ] (s) (4.3.10)
j=0
where a0 and b0 are bounded F0 measurable random variables, 0 = s0 < s1 < s2 < . . . , <
sm+1 < 1, 0 = t0 < t1 < t2 < . . . , < tn+1 < 1; aj+1 is bounded Fsj measurable random
variable, 0 j m and bj+1 is bounded Ftj measurable random variable, 0 j n . Let
us put
A = {sj : 0 j (m + 1)} [ {tj : 0 j (n + 1)}.
Let us enumerate the set A as
A = {ri : 0 i k}
106 Chapter 4. Stochastic Integration
where 0 = r0 < r1 < . . . < rk+1 . Note k may be smaller than m + n as there could be
repetitions. We can then represent f, g as
k
X
f (s) = c0 1{0} (s) + cj+1 1(rj ,rj+1 ] (s) (4.3.11)
j=0
k
X
g(s) = d0 1{0} (s) + dj+1 1(rj ,rj+1 ] (s) (4.3.12)
j=0
and hence
Z
T
t k
X
(gf )dX = d0 c0 X0 + dj+1 cj+1 (Xrj+1 ^t Xrj ^t ). (4.3.13)
0 j=0
Pk
Also, Y0 = c0 X0 and Yt = c0 X0 + j=0 cj+1 (Xrj+1 ^t Xrj ^t ) and hence
(Yrj+1 ^t
where Z t
Yt = f dX. (4.3.16)
0
R
The right hand side in (4.3.15) is gdY and thus in view of (4.3.10) we also have
D
Z t n
X
(gf )dX = b0 Y0 + bj+1 (Ytj+1 ^t Ytj ^t ) (4.3.17)
0 j=0
or in other words,
Z t n
X Z tj+1 ^t Z tj ^t
(gf )dX = b0 Y0 + bj+1 ( f dX f dX). (4.3.18)
0 j=0 0 0
Thus the result is proven when both f, g 2 S. Now fix g 2 S. Note that this fixes n as well.
Let
e P) : (4.3.18) holds}.
K = {f 2 B(⌦,
4.3. Properties of the Stochastic Integral 107
We have seen that S ✓ K. Easy to see using Theorem 4.26 (dominated convergence
e P). Thus, for all
theorem) that K is bp- closed and since it contains S, it equals B(⌦,
bounded predictable f , (4.3.18) is true.
If f 2 L(X), then approximating f by f n = f 1{|f |n} , using (4.3.18) for f n and taking
limits, we conclude (invoking dominated convergence theorem, which is justified as g is
bounded) that (4.3.18) is true for g 2 S and f 2 L(X).
Now fix f 2 L(X) and let Y be given by (4.3.16). Note that right hand side in (4.3.18)
is JY (g)(t), as defined by (4.2.1)-(4.2.2), so that we have
Z t
(gf )dX = JY (g)(t), 8g 2 S. (4.3.19)
0
Rt
Let us define J(g) = 0 gf dX for bounded predictable g, then J is an extension of
bp ucp
JY as noted above. The Theorem 4.26 again yields that if g n ! g then J(g n ) ! J(g).
T
Thus, J is the extension of JY as required in the definition of stochastic integrator. Thus
Y is a stochastic integrator and (4.3.8) holds (for bounded predictable g).
Now suppose g is predictable such that f g 2 L(X). We will prove that g 2 L(Y ) and
that (4.3.8) holds for such a g. F
To prove g 2 L(Y ), let hk be bounded predictable, converging pointwise to h such that
R
hk are dominated by g. We need to show that hk dY is Cauchy in ducp metric.
A
Let uk = hk f . Then uk are dominated by f g 2 L(X) and converge pointwise to hf .
R R
Thus, Z k = uk dX converges to Z = hf dX and hence is Cauchy in ducp metric. On the
R
other hand, since hk is bounded, invoking (4.3.8) for hk , we conclude hk dY = Z k and
R
hence is Cauchy in ducp metric. This shows g 2 L(Y ). Further, taking h = g, hk = g 1{|g|k}
R R R
above we conclude that the limit of hk dY is gdY which in turn equals hf dX as noted
above. This also shows that (4.3.8) is true when f g 2 L(X) and g 2 L(Y ).
D
To complete the proof, we need to show that if g 2 L(Y ) then f g 2 L(X). For this,
suppose un are bounded predictable, |un | |f g| and un converges to 0 pointwise. Need to
R ucp
show un dX ! 0. Let
8
< 1 if fs (!) 6= 0
˜
fs (!) = fs (!)
:0 if f (!) = 0.
s
and v n = un f˜. Now v n are predictable, are dominated by |g| and v n converges pointwise
to 0. Thus, Z
ucp
v n dY ! 0.
108 Chapter 4. Stochastic Integration
Since |un | |f g|, it follows that v n f = un and thus v n f is bounded and hence in L(X).
Thus by the part proven above, invoking (4.3.8) for v n we have
Z Z Z
v dY = v f dX = un dX.
n n
R ucp
This shows un dX ! 0. Hence f g 2 L(X). This completes the proof.
We had seen in Theorem 4.3 that the class of predictable processes is essentially the
same for the filtrations F⇧ and F⇧+ - the only di↵erence being at t = 0.
We now observe :
Theorem 4.30. For an r.c.l.l. F⇧ adapted process X, it is a stochastic integrator w.r.t. the
filtration (F⇧ ) if and only if it is a stochastic integrator w.r.t. the filtration (F⇧+ ).
T
R
Proof. Let X be a stochastic integrator w.r.t. the filtration (F⇧ ), so that hdX is defined
for bounded (F⇧ ) predictable processes h. Given a bounded (F⇧+ )-predictable processes f ,
let g be defined by
F
gt (!) = ft (!)1{(0,1)⇥⌦} (t, !).
A
Then by Theorem 4.3, g is a bounded (F⇧ ) predictable processes. So we define
Z
JX (f ) = gdX + f0 X0 .
R
R
f we can define by JX (f ) = f dX and JX will have the required continuity properties.
However, we need to check that JX (f ) so defined is (F⇧ ) adapted or in other words, belongs
to R0 (⌦, (F⇧ ), P). For f 2 S, it is clear that JX (f ) 2 R0 (⌦, (F⇧ ), P) since X is (F⇧ ) adapted.
Since the space R0 (⌦, (F⇧ ), P) is a closed subspace of R0 (⌦, (F⇧+ ), P) in the ducp metric,
it follows that JX (f ) 2 R0 (⌦, (F⇧ ), P) for f 2 B(⌦, e P(F⇧ )) and thus X is a stochastic
integrator for the filtration (F⇧ ).
Exercise 4.31. Let X be a (F⇧ )- stochastic integrator. For each t let {Gt : t 0} be a
+
filtration such that for all t, Ft ✓ Gt ✓ Ft . Show that X is a (G⇧ )- stochastic integrator.
4.4. Locally Bounded Processes 109
Lemma 4.32. Let X be a stochastic integrator and f 2 L(X). Let ⌧ be a stop time. Let
g = f 1[0,⌧ ] . Let
Z t
Yt = f dX (4.4.1)
0
R
and V = gdX. Then Vt = Yt^⌧ , i.e.
Z t
T
Yt^⌧ = f 1[0,⌧ ] dX. (4.4.2)
0
Proof. When f 2 S is a simple predictable process and ⌧ is a stop time taking only finitely
many values, then g 2 S and (4.4.1) - (4.4.2) can be checked as in that case, the integrals
R R
F
f dX and gdX are both defined directly by (4.2.2). Thus fix f 2 S. Approximating a
bounded stop time ⌧ from above by stop time taking finitely many values (as seen in the
A
proof of Theorem 2.52), it follows that (4.4.2) is true for any bounded stop time, then any
stop time ⌧ can be approximated by ⌧˜n = ⌧ ^ n and one can check that (4.4.2) continues
to be true. Thus we have proven the result for simple integrands.
Now fix a stop time ⌧ and let
R
e P) : (4.4.1)
K = {f 2 B(⌦, (4.4.2) is true for all t 0.}.
Then it is easy to see that K is closed under bp-convergence and as noted above it contains
D
S. Hence by Theorem 2.64, it follows that K = B(⌦, e P). Finally, for a general f 2 L(X),
the result follows by approximating f by f n = f 1{|f |n} and using dominated convergence
theorem. This completes the proof.
Exercise 4.33. In the proof given above, we first proved the required result for f 2 S and
any stop time ⌧ . Complete the proof by first fixing a simple stop time ⌧ and prove it for all
f 2 L(X) and subsequently prove it for all stop times ⌧ .
R t^⌧
Remark 4.34. We can denote Yt^⌧ as 0 f dX so that (4.4.2) can be recast as
Z t^⌧ Z t
f dX = f 1[0,⌧ ] dX. (4.4.3)
0 0
110 Chapter 4. Stochastic Integration
Definition 4.36. A process f is said to be locally bounded if there exist stop times ⌧ n ,
0 ⌧ 1 ⌧ 2 . . ., ⌧ n " 1 such that for every n,
f 1[0,⌧ n ] is bounded.
Thus if a process f is locally bounded, then we can get stop times ⌧ n increasing to
T
infinity (can even choose each ⌧ n to be bounded) and constants Cn such that
localizing sequence {⌧ n : n
F
Note that given finitely many locally bounded processes one can choose a common
1}. A continuous adapted process X such that X0 is
bounded is easily seen to be locally bounded. We can take the localizing sequence to be
A
⌧n = inf{t 0 : |X(t)| n or t n}.
For an r.c.l.l. adapted process X, recall that X is the process defined by X (t) =
R
X(t ), where X(t ) is the left limit of X(s) at s = t for t > 0 and X (0) = X(0 ) = 0.
Let ⌧n be the stop times defined by
Then it can be easily seen that X 1[0,⌧n ] is bounded by n and that ⌧n " 1 and hence
X is locally bounded. Easy to see that sum of locally bounded processes is itself locally
bounded. Further, if X is a r.c.l.l. process with bounded jumps : | X| K, then X is
locally bounded, since X is locally bounded and ( X) is bounded and X = X + ( X).
Exercise 4.37. Let X be an r.c.l.l. adapted process. Show that X is locally bounded if and
only if X is locally bounded.
Theorem 4.39. Let X be an integrator and f be a predictable process such that there exist
stop times ⌧m increasing to 1 with
Then f 2 L(X).
T
Z n = g n dX. Now for each m,
|g n 1[0,⌧m ] | |f 1[0,⌧m ] |
(4.4.10)
A
satisfies
Ytn,m = Zt^⌧
n
m
. (4.4.11)
R
In view of (4.4.8), (4.4.9) the assumption (4.4.7) implies that for each m, {Y n,m : n 1}
is Cauchy in ducp metric. Thus invoking Corollary 2.73, we conclude that Z n is Cauchy in
ducp metric and hence f 2 L(X).
D
As noted earlier, f 2 L(X), h predictable, |h| C|f | for some constant C > 0 implies
h 2 L(X). Thus the previous result gives us
In particular, we have
Corollary 4.41. Let g be a locally bounded predictable process. Then g belongs to L(X)
for every stochastic integrator X. And if Y is an r.c.l.l. adapted process, then Y 2 L(X)
for every stochastic integrator X.
112 Chapter 4. Stochastic Integration
Exercise 4.42. Let X be a stochastic integrator. Let s0 = 0 < s1 < s2 < . . . < sn < . . .
with sn " 1. Let ⇠j , j = 1, 2 . . . , be such that ⇠j is Fsj 1 measurable.
P
(i) For n 1 let hn = nj=1 ⇠j 1(sj 1 ,sj ] . Show that hn 2 L(X) and
Z t Xn
n
h dX = ⇠j (Xsj ^t Xsj 1 ^t ).
0 j=1
P1
(ii) Let h = j=1 ⇠j 1(sj 1 ,sj ]
. Show that h 2 L(X) and
Z t X1
hdX = ⇠j (Xsj ^t Xsj 1 ^t ).
0 j=1
For an r.c.l.l. process X and a stop time , let X [ ] denote the process X stopped at
defined as follows
[ ]
T
Xt = Xt^ . (4.4.12)
Next result shows that if X is a stochastic integrator and is a stop time, then Y = X [ ]
Z t Z t
F
Lemma 4.43. Let X be a stochastic integrator and be a stop time. Then X [
Rt
stochastic integrator and for f 2 L(X), writing Zt = 0 f dX, one has
] is also a
A
[ ]
Zt^ = f dX = (f 1[0, ] )dX. (4.4.13)
0 0
Proof. First one checks that (4.4.13) is true for f 2 S. Then for any bounded predictable
f , defining the process Z
R
J0 (f ) = (f 1[0, ] )dX
bp ucp
one can check that fn ! f implies J0 (fn ) ! J0 (f ) and hence it follows that X [ ]
D
is a stochastic integrator. Using (4.4.4), it follows that (4.4.13) is true for all bounded
predictable processes f . Finally, for a general f 2 L(X), the result follows by approximat-
ing f by f n = f 1{|f |n} and using dominated convergence theorem. This completes the
proof.
R
Exercise 4.44. Deduce the previous result using Theorem 4.29 by identifying X [ ] as gdX
for a suitable g 2 L(X).
The next result shows that if we localize the concept of integrator, we do not get
anything new- i.e. if a process is locally a stochastic integrator, then it is already a stochastic
integrator.
4.4. Locally Bounded Processes 113
Theorem 4.45. Suppose X is an adapted r.c.l.l. process such that there exist stop times
n
⌧ n such that ⌧ n ⌧ n+1 for all n and ⌧ n " 1 and the stopped processes X n = X [⌧ ] are
stochastic integrators. Then X is itself a stochastic integrator.
R
Proof. Fix f 2 B(⌦, e P) and for m 1 let U m = f dX m . Without loss of generality we
assume that ⌧ 0 = 0. Then using (4.4.4), it follows that Utm = Ut^⌧
k
m for m k. We define
J0 (f ) by J0 (f )0 = f0 X0 and for m 1,
J0 (f )t = Utm , ⌧m 1
< t ⌧ m. (4.4.14)
It follows that Z t
J0 (f )t^⌧ m = f dX m . (4.4.15)
0
Rt
Of course, for simple predictable f , J0 (f ) = 0 f dX and thus J0 is an extension of JX .
T
bp
Now let f n be bounded predictable such that f n ! f . Using (4.4.15), for n, m 1 we
have Z t
J0 (f n )t^⌧ m = f n dX m .
Writing Zn = J0(f n ) F
satisfy(2.5.9) and hence by Lemma 2.72 it follows that Z
n bp n ucp
0
and Z = J0 (f ), it follows using (4.4.15) that Z n , n
n ucp
1 and Z
! Z. We have thus proved
A
that f ! f implies J0 (f ) ! J0 (f ) and since J0 (f ) agrees with JX (f ) for simple
predictable f , it follows that X is a stochastic integrator.
We have seen a version of the dominated convergence theorem for stochastic integration.
R
Z Z
ucp
(Y n ) dX ! Y dX.
To prove that bn ! 0 suffices to prove the following: For any subsequence {nk : k 1},
there exists a further subsequence {mj : j 1} of {nk : k 1} (i.e. 9 subsequence
{kj : j 1} such that mj = nkj ) such that
bmj ! 0. (4.4.16)
114 Chapter 4. Stochastic Integration
So now, given a subsequence {nk : k 1}, using ducp (Y nk , Y ) ! 0, let us choose mj = nkj
with kj+1 > kj and ducp (Y mj , Y ) 2 j . Then as seen in the proof of Theorem 2.68, this
would imply
X1
m
[sup|Yt j Yt |] < 1, 8T < 1.
j=1 tT
Thus defining
1
X mj
Ht = | Yt Yt | (4.4.17)
j=1
it follows that (outside a fixed null set ) the convergence in (4.4.17) is uniform on t 2 [0, T ]
for all T < 1 and as a result H is an r.c.l.l. adapted process. Thus the processes (Y mj )
are dominated by (H + Y ) which is a locally bounded process, as H + Y is an r.c.l.l.
adapted process. Thus the dominated convergence Theorem 4.26 yields
T
Z Z
bmj = ducp ( (Y mj ) dX, Y dX) ! 0.
ucp
Proposition 4.48. Let Y n ! Y , where Y n , Y are Rd -valued r.c.l.l. processes and g n , g :
[0, 1) ⇥ Rd 7! R be continuous functions such that g n converges to g uniformly on compact
ucp
subsets of [0, 1) ⇥ Rd . Let Ztn = g n (t, Ytn ) and Zt = g(t, Yt ). Then Z n ! Z.
D
Proof. Like in the previous proof, let bn = ducp (Z n , Z) and given any subsequence {nk :
k 1}, using ducp (Y nk , Y ) ! 0, choose mj = nkj with kj+1 > kj and ducp (Y mj , Y ) 2 j .
It follows that
X1
m
[ sup|Yt j Yt | ] < 1, 8T < 1.
j=1 tT
m
and so [ suptT |Yt j Yt |] converges to zero for all T < 1 a.s. and now uniform convergence
of g n to g on compact subsets would yield convergence of Z mj to Z uniformly on [0, T ] for
all T and thus bmj = ducp (Z mj , Z) converges to 0. Thus every subsequence of {bn } has a
further subsequence converging to zero and hence limn!1 bn = 0.
4.5. Approximation by Riemann Sums 115
Essentially the same proof as given above for Theorem 4.46 gives the following result,
only di↵erence being that if X 2 V i.e. if Var[0,t] (X) < 1 for all t < 1, and Y is an r.c.l.l.
R R
adapted process, the integral Y dX is defined in addition to Y dX, both are defined
as Lebesgue-Stieltjes integrals while the later agrees with the stochastic integral (as seen
in Theorem 4.21).
ucp
Proposition 4.49. Suppose Y n , Y 2 R0 (⌦, (F⇧ ), P), Y n ! Y and X 2 V (a process
with finite variation paths). Then
Z Z
n ucp
(Y )dX ! Y dX.
T
The next result shows that for an r.c.l.l. process Y and a stochastic integrator X, the
R
stochastic integral Y dX can be approximated by Riemann like sums, with one di↵erence-
the integrand must be evaluated at the lower end point of the interval as opposed to any
point in the interval in the Riemann-Stieltjes integral.
0 = tm m m
0 < t1 < . . . < tn < . . . ;
F
Theorem 4.50. Let Y be an r.c.l.l. adapted process and X be a stochastic integrator. Let
tm
n " 1 as n " 1 (4.5.1)
A
be a sequence of partitions of [0, 1) such that for all T < 1,
m (T ) =( sup (tm
n+1 tm
n )) ! 0 as m " 1. (4.5.2)
{n : tm
n T }
R
Let
1
X
Ztm = Ytm
n ^t
(Xtm
n+1 ^t
Xtm
n ^t
) (4.5.3)
n=0
R
D
and Z = Y dX. Note that for each t, m, the sum in (4.5.3) is a finite sum since
tm
n ^ t = t from some n onwards. Then
ucp
Zm !Z (4.5.4)
or in other words Z
1
X t
ucp
Ytm
n ^t
(X tm
n+1 ^t
Xtm
n ^t
) ! Y dX.
n=0 0
For this, let Vt = supst |Y |. Then V is an r.c.l.l. adapted process and hence V is locally
bounded.
Let us fix m and let k (x) = max(min(x, k), k), so that | k (x)| k for all x 2 R. Let
k
X
Utk = k
(Ytm
n ^t
)1(tm m
n ,tn+1 ]
(t).
n=0
T
0 n=0
R m dX.
converges to Y This proves (4.5.5).
m
Now, Y converges pointwise to Y and are dominated by the locally bounded process
V . Hence again by Theorem 4.26,
Z
m ucp
Z
Y dX ! Y dX
F
A
which is same as (4.5.4).
We will next show that the preceding result is true when the sequence of deterministic
partitions is replaced by a sequence of random partitions via stop times. For this, we need
R
and Z is locally bounded, it follows that h is locally bounded and thus h 2 L(X).
4.5. Approximation by Riemann Sums 117
If Z is a bounded r.c.l.l. adapted process and ⌧ takes finitely many values, then easy to
see that h belongs to S and that (4.5.7) is true. Now if ⌧ is a bounded stop time, then for
m 1, ⌧ m defined by
⌧ m = 2 m ([2m ⌧ ] + 1) (4.5.8)
are stop times, each taking finitely many values and ⌧ m # ⌧ . One can then verify (4.5.7) by
approximating ⌧ by ⌧ m defined via (4.5.8) and then using the fact that hm = Z⌧ m 1(⌧ m ,1)
converges boundedly pointwise to h, validity of (4.5.7) for ⌧ m implies the same for ⌧ . For
a general ⌧ , we approximate it by ⌧n = ⌧ ^ n. Thus, it follows that (4.5.6)-(4.5.7) are true
for bounded r.c.l.l. adapted processes Z. For a general Z, let
Noting that hn = Z⌧n 1(⌧,1) converges to h and |hn | |h|, the validity of (4.5.7) for Z n
T
implies the same for Z.
Corollary 4.52. Let Z and ⌧ be as in the previous lemma and be another stop time
with ⌧ . Let
Then Z t
g = Z⌧ 1(⌧, ] . F (4.5.9)
A
gdX = Z⌧ ^t (Xt^ Xt^⌧ ) = Z⌧ (Xt^ Xt^⌧ ). (4.5.10)
0
The first equality follows from the observation that g = h1[0, ] where h is as in Lemma
Rt R·
4.51 and hence 0 gdX = ( 0 hdX)t^ (using Lemma 4.32). The second equality can be
R
directly verified.
Exercise 4.53. Express 1(⌧, ] as 1(⌧,1) 1( ,1) and thereby deduce the above Corollary from
Lemma 4.51.
D
f = ⇠ 1(⌧, ] .
Proof. This follows from the Corollary 4.52 by taking Z = ⇠ 1[⌧,1) and noting that as shown
in Lemma 2.39, Z is adapted.
118 Chapter 4. Stochastic Integration
Definition 4.55. For > 0, a -partition for an r.c.l.l. adapted process Z is a sequence of
stop times {⌧n ; : n 0} such that 0 = ⌧0 < ⌧1 < . . . < ⌧n < . . . ; ⌧n " 1 and
|Zt Z ⌧n | for ⌧n t < ⌧n+1 , n 0. (4.5.11)
Remark 4.56. Given r.c.l.l. adapted processes Z i , 1 i k and > 0, we can get a
sequence of partitions {⌧n : n 0} such that {⌧n : n 0} is a partition for each of
Z 1 , Z 2 , . . . Z k . Indeed, let {⌧n : n 0} be defined inductively via ⌧0 = 0 and
⌧n+1 = inf{t > ⌧n : max( max |Zti Z⌧in |, max |Zti Z⌧in |) }. (4.5.12)
1ik 1ik
Invoking Theorem 2.44, we can see that {⌧n : n 0} are stop times and that limn"1 ⌧n = 1.
T
n=0
ucp
Then it follows that |Ztm Zt | m and hence Z m ! Z . For k 1, let
k
X
Z t
Ztm,k =
Xk
n=0 F
Zt^⌧nm 1(⌧nm ,⌧n+1
m ] (t).
Rt ucp R t
4.26, 0 Z m,k dX ! 0 Z m dX. Thus
Z t X1
m
Z dX = Zt^⌧nm (X⌧n+1
m ^t X⌧nm ^t ). (4.5.13)
D
0 n=0
Since Zm converges pointwise to Z and |Z n | V with V locally bounded, invoking
Theorem 4.26 it follows that Z Z
m ucp
Z dX ! Z dX. (4.5.14)
We have thus proved another version of Theorem 4.50.
Theorem 4.57. Let X be a stochastic integrator. Let Z be an r.c.l.l. adapted process. Let
m # 0 and for m 1 let {⌧nm : n 1} be a m -partition for Z. Then
X1 Z t
ucp
Zt^⌧nm (X⌧n+1 ^t X⌧nm ^t ) !
m Z dX. (4.5.15)
n=0 0
4.6. Quadratic Variation of Stochastic Integrators 119
P
Remark 4.58. When m ( m )2 < 1, say m = 2 m , then the convergence in (4.5.15) is
stronger: it is uniform convergence on [0, T ] almost surely for each T < 1. We will prove this
in Chapter (6).
Theorem 4.59. Let X be a stochastic integrator. Then there exists an adapted increasing
process A, written as [X, X], such that
Z t
2 2
Xt = X0 + 2 X dX + [X, X]t , 8t. (4.6.1)
0
T
Further, Let m # 0 and for m 1 let {⌧nm : n 1} be a m -partition for X. Then one
has
1
X ucp
(X⌧n+1
m ^t X⌧nm ^t )2 ! [X, X]t . (4.6.2)
Proof. For a, b 2 R,
n=0
b2 a2 = 2a(b
F a) + (b a)2 .
A
Using this with b = X⌧n+1
m ^t and a = X⌧ m ^t and summing with respect to n, we get
n
where
1
X
Vtm = X⌧nm ^t (X⌧n+1
m ^t X⌧nm ^t )
n=0
D
and
1
X
Qm
t = (X⌧n+1
m ^t X⌧nm ^t )2 .
n=0
Note that after some n that may depend upon ! 2 ⌦, ⌧nm > t and hence X⌧n+1
m ^t = Xt . In
view of this, the two sums above have only finitely many non-zero terms.
ucp R
By Theorem 4.57, V m ! X dX. Hence, writing
Z t
2 2
At = X t X 0 2 X dX, (4.6.3)
0
we conclude
ucp
Qm
t ! At .
120 Chapter 4. Stochastic Integration
Thus
Qm m
s Qt +
2
m. (4.6.4)
ucp
Since Qm ! A, it follows that A is an increasing process.
Remark 4.60. From the identity (4.6.1), it follows that the process [X, X] does not depend
T
upon the choice of partitions {⌧nm : n 1}.
Definition 4.61. For a stochastic integrator X, the process [X, X] obtained in the previous
theorem is called the Quadratic Variation of X.
F
From the definition of quadratic variation, it follows that for a stochastic integrator X
and a stop time
A
[X [ ] , X [ ] ]t = [X, X]t^ 8t. (4.6.5)
For stochastic integrators X, Y , let us define cross-quadratic variation between X, Y
via the polarization identity
R
1
[X, Y ]t = ([X + Y, X + Y ]t [X Y, X Y ]t ) (4.6.6)
4
By definition [X, Y ] 2 V since it is defined as di↵erence of two increasing processes. Also,
D
Remark 4.63. Like (4.6.5), one also has for any stop time and stochastic integrators X, Y :
[X [ ] , Y ]t = [X, Y [ ] ]t = [X [ ] , Y [ ] ]t = [X, Y ]t^ 8t. (4.6.9)
This follows easily from (4.6.8).
Exercise 4.64. For stochastic integrators X, Y and stops and ⌧ show that
^⌧ ]
[X [ ] , Y [⌧ ] ] = [X, Y ][ .
T
(iv) If X (or Y ) is a continuous process, then [X, Y ] is also a continuous process.
Proof. For (i), using (4.6.1) and (4.3.7), we get for every t > 0
Using b2 a2
Xt2
2a(b
Xt2 = 2Xt (Xt
a) = (b a)2 , we get
F
Xt ) + [X, X]t [X, X]t .
A
(Xt Xt )2 = [X, X]t [X, X]t
which is same as (i). (ii) follows from (i) as [X, X]t is an increasing process. (iii) follows
from (i) via the polarization identity (4.6.6). And lastly, (iv) is an easy consequence of
R
(iii).
The next result shows that for a continuous process A 2 V, [X, A] = 0 for all stochastic
integrators X.
Theorem 4.67. Let X be a stochastic integrator and A 2 V, i.e. an r.c.l.l. process with
finite variation paths. Then
X
[X, A]t = ( X)s ( A)s . (4.6.10)
0<st
In particular, if X, A have no common jumps, then [X, A] = 0. This is clearly the case if
one of the two processes is continuous.
122 Chapter 4. Stochastic Integration
and Z
1
X t
ucp
X m ^t
⌧n+1 (Am ^t
⌧n+1 A ⌧nm ^t ) ! X dA.
n=0 0+
T
j(X, Y )t = ( X)s ( Y )s .
0<st
The sum above is absolutely convergent in view of part (ii) Corollary 4.65. Clearly,
j(X, X) is an increasing process. Also we have seen that
F
j(X, X)t [X, X]t . (4.6.11)
A
We can directly verify that j(X, Y ) satisfies the polarization identity
1
[j(X, Y )t = (j(X + Y, X + Y )t j(X Y, X Y )t ). (4.6.12)
4
The identity (4.6.7) characterizes [X, Y ]t and it shows that (X, Y ) 7! [X, Y ]t is a
R
bilinear form. The relation (4.6.8) also yields the parallelogram identity for [X, Y ]:
Proof. Let m # 0 and for m 1 let {⌧nm : n 1} be a m -partition for X. and take
m
an = (X⌧n+1m ^t m
X⌧nm ^t ), bn = (Y⌧n+1
m ^t Y⌧nm ^t ). Use the identity (a + b)2 + (a b)2 =
2(a2 + b2 ) with a = am m
n and b = bn ; sum over n and take limit over m. We will get the
required identity by (4.6.2).
p
Here is an analogue of the inequality |a2 b2 | 2(a b)2 (a2 + b2 ).
and taking limit over m, using (4.6.2), we get the required result (4.6.14).
T
Indeed, one has to do this carefully (in view of the null sets lurking around). We can prove
a little bit more.
p
p F
Theorem 4.70. Let X, Y be stochastic integrators. Then for any s t
Var(s,t] ([X, Y ]) ([X, X]t [X, X]s ).([Y, Y ]t [Y, Y ]s ),
Proof. Let
⌦a,b,s,r = {! 2 ⌦ : [aX + bY, aX + bY ]r (!) [aX + bY, aX + bY ]s (!)}
D
and
⌦0 = [{⌦a,b,s,r : s, r, a, b 2 Q, r s}.
Then it follows that P(⌦0 ) = 1 (since for any process Z, [Z, Z] is an increasing process).
For ! 2 ⌦0 , for 0 s r, s, r, a, b 2 Q
(a2 ([X, X]r [X, X]s ) + b2 ([Y, Y ]r [Y, Y ]s ) + 2ab([X, Y ]r [X, Y ]s ))(!) 0.
Since the quadratic form above remains positive, we conclude
|([X, Y ]r (!) [X, Y ]s (!))|
p (4.6.19)
([X, X]r (!) [X, X]s (!))([Y, Y ]r (!) [Y, Y ]s (!)).
124 Chapter 4. Stochastic Integration
Since all the process occurring in (4.6.19) are r.c.l.l., it follows that (4.6.19) is true for
all s r, s, r 2 [0, 1). Taking s = 0, this proves (4.6.21). Now given s < t and
s = t0 < t1 < . . . < tm = t, we have
mX1
|[X, Y ]tj+1 [X, Y ]tj |
j=0
X1 q
m
(4.6.20)
([X, X]tj+1 [X, X]tj )([Y, Y ]tj+1 [Y, Y ]tj )
j=0
p
([X, X]t [X, X]s )([Y, Y ]t [Y, Y ]s )
where the last step follows from Cauchy-Schwarz inequality using the fact that [X, X], [Y, Y ]
are increasing processes. Now taking supremum over partitions of [s, t] in (4.6.20) we get
(4.6.15). For (4.6.16), recalling definition of Var[a,b] (G) we have
T
Var[s,t] ([X, Y ]) = Var(s,t] ([X, Y ]) + |( [X, Y ])s |
p
([X, X]t [X, X]s )([Y, Y ]t [Y, Y ]s ) + |( X)s ( Y )s |
p
([X, X]t [X, X]s + ( X)2s )([Y, Y ]t [Y, Y ]s + ( Y )2s )
p
([X, X]t F
[X, X]s )([Y, Y ]t
Now (4.6.17) follows from (4.6.16) taking s = 0. As for (4.6.18), note that
Var[0,t] (j(X, Y )) =
X
|( X)s ( Y )s |
[Y, Y ]s ).
A
0<st
sX X
( X)2s ( Y )2s
0<st 0<st
R
p
[X, X]t [Y, Y ]t .
D
The next inequality is a version of the Kunita-Watanabe inequality, that was proven in
the context of square integrable martingales.
Proof. Let us write At = |[X, Y ]|t = Var[0,t] ([X, Y ]). Note A0 = 0 by definition of quadratic
variation. We first observe that (4.6.23) holds for simple predictable processes f, g 2 S.
As seen in the proof of Theorem 4.29, we can assume that f, g are given by (4.3.11) and
(4.3.12). Using (4.6.15), it follows that for 0 s < t
T
1 1
|At As | ([X, X]t [X, X]s ) 2 ([Y, Y ]t [Y, Y ]s ) 2
and hence
Z T
=
|fs gs |d|[X, Y ]|s
k
X
|cj+1 dj+1 |(Arj+1 Ar j )
F
A
j=0
k
X 1 1
|cj+1 dj+1 |([X, X]rj+1 [X, X]rj ) 2 ([Y, Y ]rj+1 [Y, Y ]rj ) 2
j=0
R
k
X k
X
1 1
( c2j+1 ([X, X]rj+1 [X, X]rj )) 2 ( d2j+1 ([Y, Y ]rj+1 [Y, Y ]rj )) 2
j=0 j=0
Z T Z T
1 1
|fs |2 d[X, X]s ) 2 · ( |gs |2 d[Y, Y ]s ) 2 .
D
=(
0 0
This proves (4.6.23) for f, g 2 S. Now using functional version of monotone class theorem,
Theorem 2.64 one can deduce that (4.6.23) continues to hold for all bounded predictable
processes f, g. Finally, for general f, g, the inequality follows by approximating f, g by
f n = f 1{|f |n} and g n = g 1{|g|n} respectively and using monotone convergence theorem
(recall, that integrals appearing in this result are Lebesgue-Stieltjes integrals with respect
to increasing processes.)
that X is also a stochastic integrator on (⌦, F, Q) and the class L(X) under the two measures
R
is the same and for f 2 L(X), the stochastic integral f dX on the two spaces is identical.
It follows (directly from definition or from (4.6.1)) that the quadratic variation of X is the
same when X is considered on (⌦, F, P) or (⌦, F, Q).
T
bounded random variable. Then
h = b1(u,1) f
is predictable, h 2 L(X) and
Z t
0
b1(u,1) f dX = b
Z t
F Z t
1(u,1) f dX = b ( f dX
0 0
Z
0
Proof. When f 2 S, validity of (4.7.2) can be verified directly as then h is also simple
u^t
f dX). (4.7.2)
A
predictable. Then, the class of f such that (4.7.2) is true can be seen to be closed under
bp-convergence and hence by Theorem 2.64, (4.7.2) is valid for all bounded predictable
processes. Finally, since b is bounded, say by c, |h| c|f | and hence h 2 L(X). Now
R
(4.7.2) can be shown to be true for all f 2 L(X) by approximating f by f n = f 1{|f |n}
and using Dominated Convergence Theorem - Theorem 4.26.
R
bounded random variable. Let g = b1(u,1) and Y = gdX. Then
Z t Z t Z t
Yt Zt = Zs dYs + Ys dZs + gs d[X, Z]s . (4.7.3)
0 0 0
Proof. Noting that Yt = b(Xt Xu^t ) and writing Vt = Xt Xu^t , we have,
Yt Zt =b (Zt Xt Zt Xu^t )
=b (Xt Zt Xu^t Zu^t Xu^t Zt + Xu^t Zu^t )
Z t Z u^t Z t Z u^t
=b ( X dZ X dZ + Z dX Z dX)
0 0 0 0
+ b ([X, Z]t [X, Z]u^t Xu^t (Zt Zu^t )).
4.7. Quadratic Variation of the Stochastic Integral 127
Noting that Z Z Z
t u^t t
X dZ X dZ Xu^t (Zt Zu^t ) = V dZ
0 0 0
R u^t
since Vs = 0 for 0 s u, V
dZ = 0 and we get
0
Z t Z t Z u^t
Yt Zt = b ( U dZ + Z dX Z dX + [X, Z]t [X, Z]u^t ).
0 0 0
R u^t
Invoking Lemma 4.74 and again using 0 V dZ = 0 we conclude that (4.7.3) holds.
T
f such that Z
Wt = f dX (4.7.5)
Z t Z t Z t
Wt Z t = W0 Z 0 +
0
W dZ +
F 0
Z dW +
Easy to see that K is a linear space and that it is closed under bounded pointwise conver-
gence of sequences. It trivially contains constants and we have seen in Lemma 4.75 that K
0
fs d[X, Z]s . (4.7.6)
A
contains g = b1(u,1) , where 0 u < 1 and b is Fu measurable bounded random variable
and since S is the linear span of such processes, it follows that S ✓ K.
Now Theorem 2.64 implies that (4.7.5)-(4.7.6) hold for all bounded predictable pro-
R
cesses. Comparing with (4.6.7), we conclude that for any stochastic integrator Z
Z Z
[ f dX, Z] = f d[X, Z]. (4.7.7)
R
D
We would like to show that (4.7.4) is true for all f 2 L(X) and h 2 L(Y ).
Theorem 4.77. Let X, Y be stochastic integrators and let f 2 L(X), g 2 L(Y ) and let
R R
U = f dX, V = gdY . Then
Z t
[U, V ]t = f gd[X, Y ]. (4.7.10)
0
R
Proof. Let us approximate f, g by f n = f 1{|f |n} and g n = g 1{|g|n} and let U n = f n dX,
R ucp ucp
V n = g n dY . Since f 2 L(X) and g 2 L(Y ), by definition U n ! U and V n ! V . Now
using Theorem 2.69 (also see the Remark following the result) we can get a subsequence
{nk } and a r.c.l.l. adapted increasing process H such that
k k
|Utn | + |Vtn | Ht 8k 1. (4.7.11)
Using (4.7.4) for f n , g n (since they are bounded) we get invoking Theorem 4.29
T
Z t Z t
n n n n n n
U t V t = U 0 V0 + Us dVs + Vsn dUsn + [U n , V n ]t
0 0
Z t Z t Z t (4.7.12)
n n n n n n n n
= U 0 V0 + Us gs dYs + Vs fs dXs + fs gs d[X, Y ]s .
n 2
0
Taking Y = X and g = f in (4.7.12) we get
n 2
Z t
n n
0
Z t
F
(fsn )2 d[X, X]s .
0
A
(Ut ) = (U0 ) + 2 Us fs dXs + (4.7.13)
0 0
In (4.7.13), we would like to take limit as n ! 1. Since (fsn )2 = fs2 1{|f |n} increases to
fs2 , using monotone convergence theorem, we get
R
Z t Z t
n 2
(fs ) d[X, X]s ! fs2 d[X, X]s . (4.7.14)
0 0
For the stochastic integral term, taking limit along the subsequence {nk } (chosen so that
D
(4.7.11) holds) and using H f 2 L(X) (see Corollary 4.40) and dominated convergence
theorem (Theorem 4.26), we get
Z t Z t
nk nk ucp
Us fs dXs ! Us fs dXs . (4.7.15)
0 0
ucp
Thus putting together (4.7.13), (4.7.14) and (4.7.15) along with U n ! U we conclude
Z t Z t
2 2
(Ut ) = (U0 ) + 2 Us fs dXs + (fs )2 d[X, X]s . (4.7.16)
0 0
This implies Z t
[U, U ]t = (fs )2 d[X, X]s . (4.7.17)
0
4.7. Quadratic Variation of the Stochastic Integral 129
and then using dominated convergence theorem (for signed measures) we conclude
Z t Z t
n n
fs (!)gs (!)d[X, Y ]s (!) ! fs (!)gs (!)d[X, Y ]s (!) 8t < 1 a.s. (4.7.21)
T
0 0
In view of (4.7.21), taking limit in (4.7.12) along the subsequence {nk } and using argument
similar to the one leading to (4.7.16), we conclude
Z t Z t Z t
U t Vt = U 0 V0 +
[U, V ]t = fs gs d[X, Y ]s .
0
D
The earlier proof contains a proof of the following Theorem. Of course, this can also
be deduced by taking f = h and X = Y .
R
Theorem 4.78. Let X be stochastic integrators and let f 2 L(X) and let U = f dX.
Then Z t
[U, U ]t = f 2 d[X, X] (4.7.22)
0
Remark 4.79. In Particular, it follows that for a stochastic integrator X, if f 2 L(X) then
Z t
fs2 d[X, X]s < 1 a.s. 8t < 1.
0
130 Chapter 4. Stochastic Integration
T
0+
Now this is true as long as G is a continuous function with finite variation. Let |G(s)| K
for 0 s t. Let
a( ) = sup{|G(t1 )
f 0 (x2 )| :
FK x1 , x2 K, |x1
G(t2 )| : 0 t1 , t2 t, |t1
x2 | "},
t2 | },
A
so that h(a( nt )) ! 0 as n ! 1 in view of uniform continuity of f on [ K, K] and G on
[0, t].
it
Let us write tni = n. Now using the mean value theorem, we get
R
that
n
X1
f (G(t)) f (G(0)) = [f (G(tni+1 )) f (G(tni ))] (4.8.4)
i=0
since ✓in = G(tni ) + uni (G(tni+1 ) G(tni )), with uni , 0 uni 1. Hence,
n
X1
| [f 0 (✓in ) f 0 (G(tni ))](G(tni+1 ) G(tni ))|
i=0
n
X1
1
h(a(tn )) |G(tni+1 ) G(tni )| (4.8.7)
i=0
1
h(a(tn ))Var(0,t] (G)
! 0.
T
quadratic variation is finite. This means we should keep track of first two terms in Taylor
expansion and take their limits (and prove that remainder goes to zero). Note that (4.8.3)
is essentially using Taylor expansion up to one term with remainder, but we had assumed
that f is only once continuously di↵erentiable.
F
The following lemma is a crucial step in the proof of the Ito formula. First part is
proven earlier, stated here for comparison and ease of reference.
A
Lemma 4.80. Let X, Y be stochastic integrators and let Z be an r.c.l.l. adapted process.
Let m # 0 and for m 1 let {⌧nm : n 1} be a m -partition for X, Y and Z. Then
R
1
X Z t
ucp
Z⌧nm ^t (X⌧n+1
m ^t X⌧nm ^t ) ! Z dX (4.8.8)
n=0 0
D
and
1
X Z t
ucp
Z⌧nm ^t (X m ^t
⌧n+1 X⌧nm ^t )(Y m ^t
⌧n+1 Y⌧nm ^t ) ! Z d[X, Y ]. (4.8.9)
n=0 0
Rt Rt
Remark 4.81. Observe that if Z is continuous 0 Z dX = 0+ Z dX
Proof. The first part, (4.8.8) has been proved in Theorem 4.57. The second part for the
special case Z = 1 is proven in Theorem 4.62. For (4.8.9), note that
where
1
X
Ant = Z⌧nm ^t (X⌧n+1
m ^t X⌧nm ^t )(Y⌧n+1
m ^t Y⌧nm ^t ),
n=0
X1
Btn = Z⌧nm ^t (X⌧n+1
m ^t Y⌧ m ^t
n+1
X⌧nm ^t Y⌧nm ^t ),
n=0
X1
Ctn = Z⌧nm ^t X⌧nm ^t (Y⌧n+1
m ^t Y⌧nm ^t ),
n=0
X1
Dtn = Z⌧nm ^t Y⌧nm ^t (X⌧n+1
m ^t X⌧nm ^t ).
n=0
Now using (4.8.8), we have
Z t
ucp
Btn ! Z d(XY )
T
0
Z t Z t Z
n ucp
Ct ! Z X dY = Z dS where S = X dY
0 0
Z t Z t Z
n ucp
Dt ! Z Y dX =
0 0 F
Z dR where R = Y dX.
Here we have used Theorem 4.29. Using bilinearity of integration, it follows that
n ucp
Z t
A
At ! Z dV
0
Rt Rt
where Vt = Xt Yt X0 Y0 0 X dY 0 Y dX. As seen in Theorem 4.62, Vt = [X, Y ]t .
This completes the proof.
R
We will first prove a single variable change of variable formula for a continuous stochastic
integrator and then go on to the multivariate version.
Now let us fix a twice continuously di↵erentiable function f . The standard version of
Taylor’s theorem gives
Z b
0
f (b) = f (a) + f (a)(b a) + f 00 (s)(b s)ds. (4.8.10)
a
4.8. Ito’s Formula 133
However, we need the expansion up to two terms with an estimate on the remainder. Let
us write
1
f (b) = f (a) + f 0 (a)(b a) + f 00 (a)(b a)2 + Rf (a, b) (4.8.11)
2
where Z b
Rf (a, b) = [f 00 (s) f 00 (a)](b s)ds.
a
Since f 00 is assumed to be continuous, for any K < 1, f 00 is uniformly continuous and
bounded on [ K, K] so that
lim ⇤f (K, ) = 0
!0
where
2
⇤f (K, ) = sup{|Rf (a, b)(b a) | : a, b 2 [ K, K], 0 < |b a| < }.
Here is the univariate version of the Ito formula for continuous stochastic integrators.
T
Theorem 4.83. (Ito’s formula - first version) Let f be a a twice continuously di↵eren-
tiable function on R and X be a continuous stochastic integrator. Then
Z t Z
1 t 00
f (Xt ) = f (X0 ) +
0
0
f (Xu )1{u>0} dXu +
f (Xtni ), (4.8.12)
Vin = f 0 (Xtni )(Xtni+1 Xtni ), (4.8.13)
1
Win = f 00 (Xtni )(Xtni+1 Xtni )2 , (4.8.14)
D
2
n
Ri = Rf (Xtni , Xtni+1 ). (4.8.15)
Then one has
Uin = Vin + Win + Rin (Xtni+1 Xtni )2 , (4.8.16)
n
X1
Uin = f (Xt ) f (X0 ). (4.8.17)
i=0
Further using (4.8.8) and (4.8.9) (see remark 4.81), we get
n
X1 Z t
n
Vi ! f 0 (Xu )1{u>0} dXu in probability, (4.8.18)
i=0 0
134 Chapter 4. Stochastic Integration
n
X1 Z t
1
Win ! f 00 (Xu )d[X, X]u in probability. (4.8.19)
2 0
i=0
T
|Rin (!)| |Rf (Xtni (!), Xtni+1 (!))|.
n (!)
and since n (!) ! 0, it follows that (4.8.20) is valid completing the proof.
Z t Z
m(m 1) t m
Xtm = X0m + mXum 1 1{u>0} dXu + Xu 2
1{u>0} d[X, X]u
0 2 0
and taking f (x) = exp(x) we get
D
Z t Z t
1
exp(Xt ) = exp(X0 ) + exp(Xu )dXu + exp(Xu )d[X, X]u .
0+ 2 0
We now turn to the multidimensional version of the Ito formula. Its proof given below
is in the same spirit as the one given above in the one dimensional case and is based on
the Taylor series expansion of a function. This idea is classical and the proof given here is
a simplification of the proof presented in Metivier [48]. A similar proof was also given in
Kallianpur and Karandikar [31].
We will first prove the required version of Taylor’s theorem. Here, | · | denotes the
Euclidean norm on Rd , U will denote a fixed open convex subset of Rd and C 1,2 ([0, 1) ⇥ U )
4.8. Ito’s Formula 135
denotes the class of functions f : [0, 1) ⇥ U 7! R that are once continuously di↵erentiable
in t and twice continuously di↵erentiable in x 2 U . Also, for f 2 C 1,2 ([0, 1) ⇥ U), f0
denotes the partial derivative of f in the t variable, fj denotes the partial derivative of f
w.r.t. j th coordinate of x = (x1 , . . . , xd ) and fjk denote the partial derivative of fj w.r.t.
k th coordinate of x = (x1 , . . . , xd ).
T
j,k=1
Then there exist continuous functions gjk : [0, 1) ⇥ U ⇥ U ! R such that for t 2 [0, 1),
y = (y 1 , . . . , y d ), x = (x1 , . . . , xd ) 2 U ,
h(t, y, x) =
d
X
j,k=1
> 0,
(4.8.22)
A
2
(T, K, ) = sup{|h(t, y, x)||y x| : 0 t T, x 2 K, y 2 K, 0 < |x y| }
and
R
2
⇤(T, K) = sup{|h(t, y, x)||y x| : 0 t T, x 2 K, y 2 K, x 6= y}.
⇤(T, K) < 1
D
and
lim (T, K, ) = 0.
#0
d
X
g 00 (s, y, x) = (fjk (t, x + s(y x)) fjk (t, x))(y j xj )(y k xk ).
j,k=1
Noting that g(0, y, x) = g 0 (0, y, x) = 0, by Taylor’s theorem (see remainder form given in
equation (4.8.11)) we have
h(t, y, x) =g(1, y, x)
Z 1 (4.8.23)
= (1 s)g 00 (s, y, x)ds.
0
Thus (4.8.22) is satisfied where {gjk } are defined by
Z 1
T
gjk (t, y, x) = (1 s)(fjk (t, x + s(y x)) fjk (t, x))ds. (4.8.24)
0
The desired estimates on h follow from (4.8.22) and (4.8.24).
F
Theorem 4.86. (Ito’s Formula for Continuous Stochastic Integrators) Let f 2
C 1,2 ([0, 1) ⇥ U ) where U ✓ Rd is a convex open set and let Xt = (Xt1 , . . . , Xtd ) be an
U -valued continuous process where each X j is a stochastic integrator. Then
A
Z t d Z t
X
f (t, Xt ) =f (0, X0 ) + f0 (s, Xs )ds + fj (s, Xs )dXsj
0 j=1 0+
Z (4.8.25)
d d
1X X t
R
Proof. Suffices to prove the equality (4.8.25) for a fixed t a.s. since both sides are continuous
D
Now
n
X1
f (t, Xt ) f (0, X0 ) = (f (tni+1 , Xtni+1 ) f (tni , Xtni ))
i=0
n
X1
= (f (tni+1 , Xtni+1 ) f (tni , Xtni+1 ))
i=0
n
X1
+ (f (tni , Xtni+1 ) f (tni , Xtni ))
i=0
n
X1
= (Uin + Vin + Win + Rin )
i=0
in view of (4.8.26), where
Uin =f (tni+1 , Xtni+1 ) f (tni , Xtni+1 )
T
Z tn (4.8.27)
i+1
= f0 (u, Xtni+1 )du
tn
i
and hence, by the dominated convergence theorem for Lebesgue integrals, we have
n
X1
n
Ui !
i=0
Z t
f0 (s, Xs )ds.
F
0
(4.8.28)
A
Using (4.8.8) and (4.8.9), we get (see remark 4.81)
n
X1 X d Z t
n
Vi ! fj (s, Xs )dXsj in probability (4.8.29)
0+
R
i=0 j=1
and
n
X1 d Z
1X t
Win ! fjk (s, Xs )d[X j , X k ]s in probability. (4.8.30)
2 0
i=0 j,k=1
D
In view of these observations, the result, namely (4.8.25) would follow once we show that
n
X1
Rin ! 0 in probability. (4.8.31)
i=0
Now let K t,! = {Xs (!) : 0 s t} and n (!) = supi |(Xtni+1 (!) Xtni (!))|. Then K t,! is
compact, n (!) ! 0 and hence by Lemma 4.85 for every ! 2 ⌦
(K t,! , n (!)) ! 0. (4.8.32)
Since
|Rin | (K t,! , n (!))(Xtn
i+1
(!) Xtni (!))2
138 Chapter 4. Stochastic Integration
we have
n
X1 n
X1
|Rin | (K t,! , n (!)) (Xtni+1 (!) Xtni (!))2 .
i=0 i=0
The first factor above converges to 0 pointwise as seen in (4.8.32) and the second factor
converges to [X, X]t in probability, we conclude that (4.8.31) is true completing the proof
as noted earlier.
Theorem 4.87. (Ito’s Formula for r.c.l.l. Stochastic Integrators) Let U be a convex
open subset of Rd . Let f 2 C 1,2 ([0, 1) ⇥ U ). Let X 1 , . . . , X d be stochastic integrators
Xt := (Xt1 , . . . , Xtd ) is U -valued. Further, suppose X is also U -valued. Then
Z t d Z t
X
f (t, Xt ) =f (0, X0 ) + f0 (s, Xs )ds + fj (s, Xs )dXsj
0 j=1 0+
T
d d Z
1X X t
+ fjk (s, Xs )d[X j , X k ]s
2 0
j=1 k=1
(4.8.33)
X d
X
+
0<st
{f (s, Xs )
d X
X d
1
F
f (s, Xs )
j=1
Proof. Let us begin by examining the last term. Firstly, what appears to be a sum of
uncountably many terms is really a sum of countable number of terms- since the summand
R
0<st
where
K t,! = {Xt (!) : 0 t T } [ {Xt (!) : 0 t T } (4.8.35)
Here K t,! is compact (see Exercise 2.1) and thus by Lemma 4.85, ⇤(K t,! ) < 1. As a
consequence, we have
X X
|h(s, Xs (!), Xs (!))| ⇤(K t,! ) ( Xs (!))2
0<st 0<st
4.8. Ito’s Formula 139
and invoking Corollary 4.65 we conclude that series in (4.8.34) converges absolutely, a.s.
The rest of the argument is on the lines of the proof in the case of continuous stochastic
integrators except that this time the remainder term after expansion up to two terms does
not go to zero yielding an additional term. Also, the proof given below requires use of
partitions via stop times.
For each n 1, define a sequence {⌧in : i 1} of stop times inductively as follows:
⌧0n = 0 and for i 0,
n
⌧i+1 = inf{t > ⌧in : max{|Xt X⌧in |, |Xt X⌧in |, |t ⌧in |} 2 n
} (4.8.36)
Let us note that each ⌧in is a stop time (see Theorem 2.44),
T
8n 1, ⌧m " 1 as m ! 1
and
n
(⌧i+1 ⌧in ) 2 n
.
n : m
Thus, {⌧m 0}, n F
1 satisfies the conditions of Lemma 4.80.
Fix t > 0. On the lines of the proof in the continuous case let
A
Ani = f (⌧i+1
n
^ t, X⌧i+1
n ^t ) f (⌧in ^ t, X⌧i+1
n ^t )
d
X
Vin = fj (⌧in ^ t, X⌧in ^t )(X⌧j n X⌧j n ^t )
i+1 ^t i
j=1
R
d
1 X
Win = fjk (⌧in ^ t, X⌧in ^t )(X⌧j n ^t X⌧j n ^t )(X⌧ki+1
n ^t X⌧kin ^t )
2 i+1 i
j,k=1
D
Then
n
f (⌧i+1 ^ t, X⌧i+1
n ^t ) f (⌧in ^ t, X⌧in ^t ) = Ani + Vin + Win + Rin
and hence
1
X
f (t, Xt ) f (0, X0 ) = (Ani + Vin + Win + Rin ). (4.8.37)
i=0
Now,
1
X 1 Z
X n ^t
⌧i+1
Ani = f0 (s, X⌧i+1
n ^t )ds
Since for every !, Xs (!) = Xs (!) for all but countably many s, we can conclude
X1 Z t
n
Ai ! f0 (s, Xs )ds in probability. (4.8.38)
i=0 0
and
1
X d d Z
1 XX t
Win ! fjk (s, Xs )d[X j , X k ] (4.8.40)
T
2 0
i=0 j=1 k=1
Rn =
1
X
i=0
h(⌧in ^ t, X⌧i+1
F
n ^t , X⌧ n ^t ) !
i
X
0<st
h(s, Xs , Xs ). (4.8.41)
A
Let us partition indices i into three sets: h(⌧in ^ t, X⌧i+1
n ^t , X⌧ n ^t ) is zero, is small and is
i
large as follows:
H n (!) ={i n
0 : ⌧i+1 (!) ^ t = ⌧in (!) ^ t}
R
i
X
B n (!) = h(⌧in (!) ^ t, X⌧i+1
n ^t (!), X⌧ n ^t (!)),
i
i2E n (!)
X
C n (!) = h(⌧in (!) ^ t, X⌧i+1
n ^t (!), X⌧ n ^t (!))
i
i2F n (!)
we observe that
Rn = B n + C n .
| Xs (!)| > 2·2 n for s 2 (0, t], then s must equal ⌧jn (!) ^ t for some j with i = j 1 2 F n .
i.e.
if s 2 (0, t] and | Xs (!)| > 2 · 2 n then s = ⌧i+1
n
^ t for i 2 F n (!). (4.8.42)
Hence for i 2 E n (!),
|X⌧i+1
n ^t (!) X⌧in ^t (!)| |(X )⌧i+1
n ^t (!) X⌧in ^t (!)| + | X⌧i+1
n ^t (!)|
n
3 · 2
and hence
X
|B n (!)| |h(⌧in (!) ^ t, X⌧i+1
n ^t (!), X⌧ n ^t (!))|
i
i2E n (!)
X
(K t,! , 3 · 2 n
) |X⌧i+1
n ^t (!) X⌧in ^t (!)|2
i
P
where K t,! defined by (4.8.35) is compact. Since i |X⌧i+1 ^t X⌧in ^t |2 converges to
T
n
P j j t,! , 3 · 2 n ) ! 0 for all !, invoking Lemma 4.85 it
j [X , X ]t in probability and (K
follows that
B n ! 0 in probability.
Cn !
F
Thus to complete the proof of (4.8.41), it would suffice to show that
X
h(s, Xs , Xs ) in probability. (4.8.43)
A
0<st
Let G(!) = {s 2 (0, t] : | Xs (!)| > 0}. Since X is an r.c.l.l. process, G(!) is a countable
set for every !. Fix ! and for s 2 G(!), define
X
R
Then
X X
C n (!) = h(⌧in ^ t(!), X⌧i+1
n ^t (!), X⌧ n ^t (!)) = ans (!).
D
i
i2F n (!) s2G(!)
If | Xs (!)| > 2 · 2 n,
then ans (!) = h(⌧in (!) ^ t, Xs (!), X⌧in ^t (!)) n (!) ^ t (as
with s = ⌧i+1
seen in (4.8.42)) and hence
ans (!) ! h(s, Xs (!), Xs (!)) for all !. (4.8.44)
For i 2 F n (!),
|X⌧i+1
n ^t (!) X⌧in ^t (!)| |(X )⌧i+1
n ^t (!) X⌧in ^t (!)| + | X⌧i+1
n ^t (!)|
n
2 + | X⌧i+1
n ^t (!)|
2| X⌧i+1
n ^t (!)|
142 Chapter 4. Stochastic Integration
Using Weierstrass’s M -test (series version of the dominated convergence theorem) along
T
with |ans (!)| Cs (!) (recall (4.8.44)), we get
X X
C n (!) = ans (!) ! h(s, Xs (!), Xs (!)) a.s. (4.8.46)
s2G(!) s2G(!)
Proof. In the Ito formula (4.8.33) that expresses f (t, Xt ), it is clear that the terms involving
R
integral are stochastic integrators. We had seen that the last term is
X
Dt = h(s, Xs Xs )
0<st
D
Hence
Var[0,T ] (D)(!) ⇤(K T,! )[X, X]T .
Thus D is a process with finite variation and hence is a stochastic integrator, completing
the proof.
4.8. Ito’s Formula 143
We have seen earlier in Lemma 4.65 that [X, Y ]t = ( X)t ( Y )t . Thus carefully
examining the right hand side of the Ito formula (4.8.33), we see that we are adding and
subtracting a term
Xd X d
1
fjk (s, Xs )( Xsj )( Xsk ).
2
j=1 k=1
Let us introduce - for now in an ad-hoc manner [X, Y ](c) for semimartingales X, Y :
(c)
X
[X, Y ]t = [X, Y ]t ( X)s ( Y )s . (4.8.47)
0<st
Later we will show that this is the cross-quadratic variation of continuous martingale parts
of X and Y (See Theorem 8.76). For now we observe the following.
T
Then [X, Y ](c) is a continuous process and further
Z Z X
(c)
hd[X, Y ] = hd[X, Y ] hs ( X)s ( Y )s . (4.8.48)
F 0<st
Proof. Continuity of [X, Y ](c) follows from its definition (4.89) and part (iii) of Lemma
4.65. The identity (4.8.48) for simple functions h 2 S follows by direct verification and
A
hence follows for all bounded predictable processes by monotone class theorem.
R
Using Lemma 4.89, we can recast the Ito formula in an alternate form
d d Z
1X X t
+ fjk (s, Xs )d[X j , X k ](c) (4.8.49)
2 0+
j=1 k=1
X d
X
+ {f (s, Xs ) f (s, Xs ) fj (s, Xs ) Xsj }.
0<st j=1
144 Chapter 4. Stochastic Integration
Exercise 4.91. Let S be a (0, 1)-valued continuous stochastic integrator. Show that Rt =
(St ) 1 is also a stochastic integrator and
Z t Z t
2
Rt = R0 (Ss ) dSs + (Ss ) 3 d[S, S]s .
0+ 0+
Exercise 4.92. Let S be a (0, 1)-valued r.c.l.l. stochastic integrator with St > 0 for t > 0.
Show that Rt = (St ) 1 is also a stochastic integrator and
Z t Z t X
2
Rt = R0 (Ss ) dSs + (Ss ) 3 d[S, S](c)
s + u(Ss , Ss )
0+ 0+ 0<st
1 1
where u(y, x) = y x + (y x) x12 .
1
Exercise 4.93. Let X be a continuous stochastic integrator and let St = exp(Xt 2 [X, X]t ).
T
(i) Show that S satisfies
Z t
St = exp(X0 ) + Su dXu . (4.8.50)
0+
Z t
A
Zt = exp(X0 ) + Zu dXu . (4.8.51)
0+
Show that Z = S.
Hint: Let Yt = Zt exp( Xt + 12 [X, X]t ) and applying Ito formula, conclude that Yt =
R
Y0 = 1.
Xt = X0 + Su 1 dSu .
0+
Exercise 4.94. Let X, Y be a (0, 1)-valued continuous stochastic integrators and let U, V
be solutions to Z t
Ut = exp(X0 ) + Uu dXu (4.8.54)
0+
and Z t
Vt = exp(Y0 ) + Vu dYu . (4.8.55)
0+
Let Wt = Ut Vt . Show that W is the unique solution to
Z t Z t Z t
Wt = exp(X0 Y0 ) + Wu dXu + Wu dYu + Wu d[X, Y ]u . (4.8.56)
0+ 0+ 0+
T
be the class of simple predictable processes f such that |f | 1. For stochastic integrators
X, Y , let Z Z
dem (X, Y ) = sup{ducp ( f dX, f dY ) : f 2 S1 } (4.9.1)
F
Easy to see that dem is a metric on the class of stochastic integrators. This metric was
defined by Emery [16] and the induced topology is called the Emery topology. If X n
em
A
converges to X in dem metric, we will write it as X n ! X. Taking f = 1 in (4.9.1), it
follows that
ducp (X, Y ) dem (X, Y ) (4.9.2)
R
and likewise that {X n }is Cauchy in dem is and only if for all T > 0, > 0
Z t Z t
n
lim [ sup P( sup | f dX f dX k | > )] = 0. (4.9.6)
n,k!1 f :2S1 0tT 0 0
Proof. Let K1 be the class of bounded predictable processes for which (4.9.7) is true. Then
T
using the dominated convergence theorem and the definition of ducp , it follows that K1 is
closed under bp-convergence and contains S1 .
Let K be the class of bounded predictable processes g such that g̃ = max(min(g, 1), 1) 2
K1 . Then it is easy to see that K is closed under bp-convergence and contains S. Thus K
F
is the class of all bounded predictable processes g and hence K1 contains all predictable
processes bounded by 1.
A
Corollary 4.97. If Xn , X are semimartingales such that dem (X n , X) ! 0 then for all
T > 0, > 0 Z t Z t
lim [ sup P( sup | f dX n f dX| > )] = 0 (4.9.8)
R
Linearity of the stochastic integral and the definition of the metric dem yields the
D
inequality
dem (U + V, X + Y ) dem (U, X) + dem (V, Y )
em em em
which in turn implies that if X n ! X and Y n ! Y then (X n + Y n ) ! (X + Y ).
We will now prove an important property of the metric dem .
Theorem 4.98. The space of stochastic integrators is complete under the metric dem .
Proof. Let {X n } be a Cauchy sequence in dem metric. Then by (4.9.2) it is also cauchy
in ducp metric and so by Theorem 2.68 there exists an r.c.l.l. adapted process X such that
ucp em
X n ! X. We will show that X is a stochastic integrator and X n ! X.
4.9. The Emery Topology 147
T
uniformly bounded. Dividing by the uniform upper bound if necessary, we can assume that
|f m | 1. We wish to show that ducp (Y (f m ), Y (f )) ! 0 as m ! 1.
Given " > 0, first choose and fix n⇤ such that an⇤ < 3" . Then
⇤
⇤
F ⇤
+ ducp (Y n (f ), Y (f ))
⇤
an⇤ + ducp (Y n (f m ), Y n (f )) + an⇤
⇤ ⇤
ducp (Y (f m ), Y (f )) ducp (Y (f m ), Y n (f m )) + ducp (Y n (f m ), Y n (f ))
(4.9.11)
A
" ⇤ ⇤ "
+ ducp (Y n (f m ), Y n (f )) + .
3 3
m bp m
R m n ⇤ ucp R ⇤
Since f ! f and {f } is bounded by 1, f dX ! f dX n and hence we can
R
choose m⇤ (depends upon n⇤ which has been chosen and fixed earlier) such that for m m⇤
one has
⇤ ⇤ "
ducp (Y n (f m ), Y n (f ))
3
D
We will now strengthen the conclusion in Theorem 4.46 by showing that the convergence
is actually in the Emery topology.
ucp
Theorem 4.99. Suppose Y n , Y 2 R0 (⌦, (F⇧ ), P), Y n ! Y and X is a stochastic inte-
grator. Then Z Z
em
(Y n ) dX ! Y dX.
R R
Proof. As noted in Thoerem 4.29, (Y n ) dX and Y dX are stochastic integrators.
Further, we need to show that we need to show that for T < 1, > 0
Z t Z t
n
lim [ sup P( sup | (Y ) f dX Y f dX| > )] = 0. (4.9.12)
n!1 f :2S 0tT 0 0
1
We will prove this by contradiction. Suppose (4.9.12) is not true. Then there exists an "
and a subsequence {nk } such that
T
Z t Z t
nk
sup P( sup | (Y ) f dX Y f dX| > ) " 8k 1. (4.9.13)
f :2S1 0tT 0 0
2 S1 such that
For each k get fk
Z t
k
P( sup | (Y n ) f k dX
0tT 0
F Z
0
t
Y f k dX| > ) ". (4.9.14)
A
k ucp
Now let g k = (Y n Y ) f k . Since Y n ! Y and f k are uniformly bounded by 1, it
ucp
follows that g k ! 0 and hence by Theorem 4.46,
Z t Z t
nk k
P( sup | (Y ) f dX Y f k dX| > ) ! 0.
R
0tT 0 0
We will first prove a lemma and then go on to the result mentioned above.
So we need to show that (see (4.9.7)) for T < 1, > 0 and " > 0
Z t Z t
sup P( sup | (U n ) f dX n (U n ) f dX| > ) < ". (4.9.15)
f :2S1 0tT 0 0
Recall, S1 is the class of predictable processes that are bounded by 1.
First, get < 1 such that
"
P(HT > )
2
and let a stop time be defined by
= inf{t > 0 : Ht or Ht } ^ (T + 1).
Then we have
"
P( < T ) = P(HT > )
2
and for f 2 S1 writing hn = (U n ) f 1 1[0, ] we see that
T
Z t Z t
n n
P( sup | (U ) f dX (U n ) f dX| > ) (4.9.16)
0tT 0 0
Z t Z t
n n
P( sup | (U ) f dX (U n ) f dX| > ) + P( < T )
0t ^T
P( sup |
0t ^T
Z
0
0
t
n
h dX n
ZF 0
t
0
hn dX| >
"
)+ .
2
A
em
Xn !
Finally, since X, for the choice of 0 = and "0 = 2" invoking Corollary 4.97
get n0 such that for n n0 one has
Z t Z t
n "
R
n n
P( sup | (U ) f dX (U n ) f dX| > ) < ".
0tT 0 0
Note that the choice of and n0 did not depend upon f 2 K1 and hence (4.9.15) holds
completing the proof.
Here is our main result that connects convergence in the Emery topology and stochastic
integration.
ucp
Theorem 4.101. Let Y n , Y 2 R0 (⌦, (F⇧ ), P) be such that Y n ! Y and let X n , X be
em R R
stochastic integrators such that X n ! X. Let Z n = (Y n ) dX n and Z = Y dX.
em
Then Z n ! Z.
150 Chapter 4. Stochastic Integration
So now, given a subsequence {nk : k 1}, using ducp (Y nk , Y ) ! 0, let us choose mj = nkj
T
with kj+1 > kj and ducp (Y mj , Y ) 2 j for each j 1. Then as seen earlier, this would
imply
1
X m
Thus defining
1
X
F
[sup|Yt j Yt |] < 1, 8T < 1.
j=1 tT
A
mj
Ht = sup [ |Ys Ys | + |Ys |]
0st j=1
we have that H is an r.c.l.l. process and |Y mj | H. Then by Lemma 4.100, it follows that
Z Z
R
mj mj em
(Y ) dX (Y mj ) dX ! 0.
We will now show that X 7! [X, X] and (X, Y ) 7! [X, Y ] are continuous mappings in
the Emery topology.
4.9. The Emery Topology 151
em em
Theorem 4.103. Let X n , X, Y n , Y be stochastic integrators such that X n ! X, Y n !
Y . Then
dvar ([X n , Y n ], [X, Y ]) ! 0 (4.9.18)
and as a consequence
em
[X n , Y n ] ! [X, Y ]. (4.9.19)
ucp
it follows that [U n , U n ] ! 0 and so
T
dvar ([X n X, X n X]) ! 0. (4.9.20)
and hence [U n , U n ]
ucp
Var[0,T ] ([U n , Z])
F [U n , U n ]T [Z, Z]T
Noting that
[X n , X n ] = [X n X, X n X] + 2[X n , X] [X, X]
D
Now the required relation (4.9.18) follows from (4.9.22) by polarization identity (4.6.6).
T
dvar (j(X n , X n ), j(X, X)) ! 0.
The required result, namely (4.9.23) follows from this by polarization identity (4.6.12).
F
Next we will prove that (Xt ) 7! (f (t, Xt )) is a continuous mapping in the Emery
topology for smooth functions f . Here is a Lemma that would be needed in the proof.
X
At = Zs ( X)s ( Y )s .
0<st
We have seen in (4.9.23) that dvar (B n , B) ! 0. The conclusion now follows from Propo-
sition 4.102.
Theorem 4.106. Let X (n) = (X 1,n , X 2,n , . . . , X d,n ) be a sequence of Rd -valued r.c.l.l.
processes such that X j,n are stochastic integrators for 1 j d, n 1. Let
em
X j,n ! X j , 1 j d.
4.9. The Emery Topology 153
T
Ytn = fj (s, Xs )dXsj,n
j=1 0+
d Z
X t
Yt = fj (s, Xs )dXsj
Btn =
2
j=1
1X X
d
j=1 k=1
0+
d Z t
0+
F (n)
fjk (s, Xs )d[X j,n , X k,n ]s
A
d d Z
1X X t
Bt = fjk (s, Xs )d[X j , X k ]s
2 0
j=1 k=1
X (n)
R
n
V = h(s, Xs(n) , Xs )
0<st
X
V = h(s, Xs , Xs ).
0<st
D
If X n , X were continuous processes so that V n , V are identically equal to zero and the
result follows from (4.9.24). For the r.c.l.l. case, recall that h can be expressed as
d
X
h(t, y, x) = gjk (t, y, x)(y j xj )(y k xk )
j,k=1
T
Now combining this with (4.9.24) we finally get Z n ! Z.
Essentially the same proof (invoking Proposition 4.48) would yield the slightly stronger
result:
F
Theorem 4.107. Let X (n) = (X 1,n , X 2,n , . . . , X d,n ) be a sequence of Rd -valued r.c.l.l.
processes such that X j,n are stochastic integrators for 1 j d, n 1 such that
A
em
X j,n ! X j , 1 j d.
Let X = (X 1 , X 2 , . . . , X d ). Let f n , f 2 C 1,2 ([0, 1)⇥Rd ) be functions such that f n , f0n , fjn , fjk
n
f0n , f0 are derivatives in the t variable of f n = f n (t, x), f = f (t, x) respectively; fjn , fj
are derivatives in xj and fjk n ,f n n
jk are derivatives in xj , xk of f = f (t, x), f = f (t, x)
respectively with x = (x1 , . . . xd ). Let
D
(n)
Ztn = f n (t, Xt ), Zt = f (t, Xt ).
em
Then Z n ! Z.
it can be shown that JX admits a required extension. The next exercise gives steps to
construct such an extension. The steps are like the usual proof of Caratheodory extension
theorem from measure theory. This exercise can be skipped on the first reading.
T
n=1 n=1
Let
e P) : 9f m 2 S s.t.
A = {⇠ 2 B(⌦, ⇤
(⇠ f m ) ! 0}
X
bp
(i) Let f n , f 2 S be such that f n ! f . Show that
bp
f n ! f implies JX (f n )
F ucp
! JX (f ). (4.10.5)
A
(ii) Let f, f 1 , f 2 2 S and c 2 R. Show that
X X
1 2 1 2
|f | |f | ) X (f ) X (f ). (4.10.8)
X (f 1 + f 2 ) X (f 1 ) + X (f 2 ). (4.10.9)
D
e P)
(iii) Show that for ⇠1 , ⇠2 2 B(⌦,
⇤ ⇤ ⇤
X
(⇠1 + ⇠2 ) X
(⇠1 ) + X
(⇠2 ).
X (f n ) ! 0. (4.10.10)
bp
(vi) Let hn , h 2 S for n 1 be such that and hn ! h Show that
(vii) Let f n 2 S be such that f n f n+1 K for all n, where K is a constant. Show that
8✏ > 0 9n⇤ such that for m, n n⇤ , we have
X (f m f n ) < ✏. (4.10.12)
[Hint: Prove by contradiction. If not true, get ✏ > and subsequences {nk }, {mk } both
increasing to 1 such that
k k
X (f n fm ) ✏ 8k 1.
k k bp
Observe that g k defined by g k = f n f m satisfy g k ! 0.]
e P)
(viii) For ⇠ 1 , ⇠ 2 2 B(⌦,
⇤ ⇤
|⇠ 1 | |⇠ 2 | ) X
(⇠ 1 ) X
(⇠ 2 ). (4.10.13)
T
j=1
1
X 1
X
⇤ ⇤
X
( ⇠j ) X
(⇠ j ). (4.10.14)
j=1 j=1
|g|
F 1
X
|f n |. (4.10.15)
A
n=1
Show that
1
X
X (|g|) X (|f k |). (4.10.16)
k=1
R
Pn bp
[Hint: Let g n = min(|g|, k=1 |f
k |) and note that g n ! |g|.]
⇤
X
(f ) = X (f ). (4.10.17)
e P) for n
(xv) Let ⇠, ⇠ n 2 B(⌦, 1 be such that ⇠ n ! ⇠ uniformly. Show that
⇤
lim X
(⇠ n ⇠) = 0. (4.10.21)
n!1
T
(c) For k 1 show that 9nk such that nk > nk 1 (here n0 = 1) and
⇤ k k 1
X
(⇠ n ⇠n )<2 k
.
e P) be such that f n bp
(xxi) Let f n , f 2 B(⌦, ! f . Show that JX (f n ) is Cauchy in ducp -metric.
T
F
A
R
D
Chapter 5
Semimartingales
The reader would have noticed that in the development of stochastic integration in the
T
previous chapter, we have not talked about either martingales or semi martingales.
A semimartingale is any process which can be written as a sum of a local martingale
and a process with finite variation paths.
F
The main theme of this chapter is to show that the class of stochastic integrators is the
same as the class of semimartingales, thereby showing that stochastic integral is defined
for all semimartingales and the Ito formula holds for them. This is the Dellacherie-Meyer-
A
Mokobodzky-Bichteler Theorem.
Traditionally, the starting point for integration with respect to square integrable mar-
tingales is the Doob-Meyer decomposition theorem. We follow a di↵erent path, proving
that for a square integrable martingale M , the quadratic variation [M, M ] can be defined
R
directly and then Xt = Mt2 [M, M ]t is itself a martingale. This along with Doob’s maxi-
mal inequality would show that square integrable martingales (and locally square integrable
martingales) are integrators. We would then go on to show that a local martingale (and
D
159
160 Chapter 5. Semimartingales
M is a martingale and
E[Mt2 ] < 1 8t < 1.
Definition 5.3. An r.c.l.l. adapted process L is said to be a local martingale if there exist stop
n
times ⌧ n " 1 such that for each n, the process M n defined by M n = L[⌧ ] , i.e. Mtn = Lt^⌧ n
is a martingale.
Exercise 5.4. Let W be the Brownian motion and let U = exp(W12 ). Let
T
8
<W if t < 1
t
Lt =
:W + U (W W1 ) if t 1.
1 t
Let ⌧ n = 1 if U
that
F
n and ⌧ n = n if U n. Let M n = L[⌧
n]
be the stopped process. Show
A
(i) For each n, ⌧ n is a stop time for the filtration (F⇧W ).
(ii) ⌧ n " 1.
R
For a local martingale L, one can choose the localizing sequence {⌧ n } such that ⌧ n n
n
and then M n = L[⌧ ] is uniformly integrable.
Here is a simple condition under which a local martingale is a martingale.
Then L is a martingale.
5.1. Notations and terminology 161
n
Proof. Let ⌧ n " 1 such that for each n, the process M n defined by M n = L[⌧ ] is a
martingale. Then Mtn converges to Lt pointwise and (5.1.1) implies that sup0st |Ls |
serves as a dominating function. Thus, for each t, Mtn converges to Lt in L1 (P). Now the
required result follows using Theorem 2.21.
Definition 5.6. An r.c.l.l. adapted process N is said to be a locally square integrable mar-
tingale if there exist stop times ⌧ n " 1 such that for each n, the process M n defined by
n
M n = N [⌧ ] , i.e. Mtn = Nt^⌧ n is a square integrable martingale.
Exercise 5.7. Show that the process L constructed in Exercise 5.4 is a locally square inte-
grable martingale.
T
Definition 5.8. An r.c.l.l. adapted process X is said to be a semimartingale if X can be
written as X = M + A where M is an r.c.l.l. local martingale and A is an r.c.l.l. process whose
paths have finite variation on [0, T ] for all T < 1, i.e. Var[0,T ] (A) < 1 for all T < 1.
F
Let us denote by M the class of all r.c.l.l. martingales M with M0 = 0, M2 the class
of all r.c.l.l. square integrable martingales with M0 = 0. We will also denote by Mloc the
A
class of r.c.l.l. local martingales with M0 = 0 and M2loc the class of r.c.l.l. locally square
integrable martingales with M0 = 0. Thus, M 2 M2loc (Mloc ) if there exist stop times ⌧ n
n
increasing to infinity such that the stopped processes M [⌧ ] belong to M2 (respectively
R
belong to M).
Let Mc be the class of all continuous martingales M with M0 = 0, Mc,loc be the class of
all continuous local martingales M with M0 = 0 and M2c be the class of square integrable
D
Exercise 5.10. Suppose fn 2 D([0, 1), R) are such that fn converges to a function f
uniformly on compact subsets of [0, 1), i.e. sup0tT |fn (t) f (t)| converges to zero for all
T < 1. Show that
T
(iii) fn (s ) converges to f (s ) for all s 2 [0, 1).
We will now define quadratic variation (↵) of a function ↵ 2 D([0, 1), R).
F
For each n 1; let {tni (↵) : i 1} be defined inductively as follows : tn0 (↵) = 0 and
having defined tni (↵), let
A
tni+1 (↵) = inf{t > tni (↵) : |↵(t) ↵(tni (↵))| 2 n
or |↵(t ) ↵(tni (↵))| 2 n
}.
If limi tni (↵) = t⇤ < 1, then the function ↵ cannot have a left limit at t⇤ . Hence for each
↵ 2 D([0, 1), R) , tni (↵) " 1 as i " 1 for each n. Let
R
1
X
n (↵)(t) = (↵(tni+1 (↵) ^ t) ↵(tni (↵) ^ t))2 . (5.2.1)
i=0
Since tni (↵) increases to infinity, for each ↵ and n fixed, the infinite sum appearing above
D
is essentially a finite sum and hence n (↵) is itself an r.c.l.l. function. The space D =
D([0, 1), R) is equipped with the topology of uniform convergence on compact subsets
(abbreviated as ucc). Let D e denote the set of ↵ 2 D such that n (↵) converges in the ucc
topology and 8
<lim if ↵ 2 De
n n (↵)
(↵) = (5.2.2)
:0 e
if ↵ 62 D.
Here are some basic properties of the quadratic variation map .
e
Lemma 5.11. For ↵ 2 D
5.2. The Quadratic Variation Map 163
Proof. For (i), note that for s t, if tnj s < tnj+1 , then |(↵(s) ↵(tnj ))| 2 n, and
j 1
X
n (↵)(s) = (↵(tni+1 (↵)) ↵(tni (↵)))2 + (↵(s) ↵(tnj ))2
i=0
jX 1
n (↵)(t) = (↵(tni+1 (↵)) ↵(tni (↵)))2
i=0
T
1
X
+ (↵(tni+1 (↵) ^ t) ↵(tni (↵) ^ t))2
i=j
and hence
n (↵)(s) F
n (↵)(t) +2 2n
Compare with (4.6.4). Thus (5.2.3) is valid for all n 1 and s t and hence it follows that
. (5.2.3)
A
the limiting function (↵) is an increasing function. Uniform convergence of the r.c.l.l.
function n (↵) to (↵) implies that (↵) is an r.cl.l. function.
For (ii), it is easy to see that the set of points of discontinuity of n (↵) are contained
in the set of points of discontinuity of ↵ for each n. Uniform convergence of n (↵)(t) to
R
(↵)(t) for t 2 [0, T ] for every T < 1 implies that the same is true for (↵) i.e. for t > 0,
(↵)(t) 6= 0 implies that ↵(t) 6= 0.
On the other hand, let t > 0 be a discontinuity point for ↵. Let us note that by the
D
Thus for n such that 2.2 n < (↵)(t), t must be equal to tnk (↵) for some k 1 since(5.2.4)
implies ↵(v) 2.2 n for any v 2 [j (tnj (↵), tnj+1 (↵)). Let sn = tnk 1 (↵) where k is such
that t = tnk (↵). Note that sn < t. Let s⇤ = lim inf n sn . We will prove that
If s⇤ < t, using (5.2.4) it follows that |↵(u) ↵(v)| = 0 for u, v 2 (s⇤ , t) and hence
the function ↵(u) is constant on the interval (s⇤ , t) and implying that sn ! s⇤ . Also,
↵(s⇤ ) = ↵(t ) and (↵)(s⇤ ) = (↵)(t ). So if ↵ is continuous at s⇤ , once again uniform
convergence of n (↵) to (↵) on [0, t] shows that (5.2.5) is valid in this case too.
Remains to consider the case s⇤ < t and ↵(s⇤ ) = > 0. In this case, for n such that
2.2 n < , sn = s⇤ and uniform convergence of n (↵) to (↵) on [0, t] shows that (5.2.5)
is true in this case as well.
We have (for large n)
T
(iii) follows from (i) and (ii) since for an increasing function that is non-negative at
zero, the sum of jumps up to t is at most equal to its value at t:
X
( ↵(s))2 (↵)(t).
0<st
The last part, (iv) follows from (ii), (iii).
F
A
Remark 5.12. is the quadratic variation map. It may depend upon the choice of the
partitions. If instead of 2 n , we had used any other sequence {"n }, it would yield another
mapping ˜ which will have similar properties. Our proof in the next section will show that if
P
R
We note two more properties of the quadratic variation map . Recall that the total
D
If Var(0,T ] (↵) < 1, ↵ is said to have finite variation on [0, T ] and then on [0, T ] it can be
written as di↵erence of two increasing functions.
(iii) Suppose ↵ is continuous, and Var(0,T ] (↵) < 1. Then (↵)(t) = 0, 8t 2 [0, T ].
Proof. (i) is immediate. For (ii), it can be checked from the definition that
k
n (↵)(t ^ sk ) = n (↵ )(t) , 8t. (5.2.8)
Since ↵k 2 D,e it follows that n (↵)(t) converges uniformly on [0, sk ] for every k and hence
using (5.2.8) we conclude that ↵ 2 D e and that (5.2.7) holds.
For (iii), note that ↵ being a continuous function,
|↵(tni+1 (↵) ^ t) ↵(tni (↵) ^ t)| 2 n
T
1
X
n (↵)(t) = (↵(tni+1 (↵) ^ t) ↵(tni (↵) ^ t))2
i=0
1
X
2
2
n
n
⇥
i=0 F
|↵(tni+1 (↵) ^ t)
⇥ Var[0,T ] (↵).
↵(tni (↵) ^ t)|
A
This shows that (↵)(t) = 0 for t 2 [0, T ].
The next lemma connects the quadratic variation map and r.c.l.l. martingales.
Lemma 5.14. Let (Nt , Ft ) be an r.c.l.l. martingale such that E(Nt2 ) < 1 for all t > 0.
D
At (!) = [ (N.(!))](t).
Then (At ) is an (Ft ) adapted r.c.l.l. increasing process such that Xt := Nt2 At is also a
martingale.
166 Chapter 5. Semimartingales
T
i=0
1
X 1
X
2 2
= (N n
i+1 ^t
N n
i ^t
) (N n
i+1 ^t
N n
i ^t
)2
i=0 i=0
Let us define
=2
1
X
i=0
N n
i ^t
(N n
i+1 ^t
F N n
i ^t
)
A
Xtn,i = N n
i ^t
(N n
i+1 ^t
N n
i ^t
).
Then
1
X
Ytn = 2 Xtn,i . (5.3.2)
R
i=0
Noting that for s < ⌧ , |Ns | C and for s ⌧ , Ns = N , it follows that
|N n
i+1 ^t
N n
i ^t
| > 0 implies that |N n
i ^t
| C.
D
Since n increases to 1 as i tends to infinity, E(N 2sn ^t ) and E(N 2n ^t ) both tend to E[Nt2 ]
i r+1
P
as r, s tend to 1 and hence ri=1 Xtn,i converges in L2 (P). In view of (5.3.2), one has
r
X
2 Xtn,i ! Ytn in L2 (P) as r ! 1
i=1
Ntn = N n
i
if n
i t< n
i+1 .
T
Observe that by the choice of { n :i 1}, one has
i
|Nt Ntn | 2 n
for all t. (5.3.7)
H(!) = { n
i (!) : i F
1} [ { n+1
i (!)
It may be noted that for ! such that t 7! Nt (!) is continuous, each jn (!) is necessarily
: i 1} (5.3.8)
A
equal to in+1 (!) for some i, but this need not be the case when t 7! Nt (!) has jumps. Let
⇣0 (!) = 0 and for j 0, let
{⇣i t} = [ik=0 ({ n
i k t} \ Bk )
Using (5.3.9) and using the fact that Ntn = Nt^ jn for jn t < n
j+1 , one can write Y n
and Y n+1 as
X1
n n
Yt = 2Nt^⇣ j
(Nt^⇣j+1 Nt^⇣j ),
j=0
X1
Ytn+1 = n+1
2Nt^⇣j
(Nt^⇣j+1 Nt^⇣j ).
j=0
Hence
1
X
Ytn+1 Ytn = 2 Ztn,j (5.3.10)
j=0
where
Ztn,j = (Nt^⇣
n+1
j
n
Nt^⇣j
)(Nt^⇣j+1 Nt^⇣j ).
Also, using (5.3.7) one has
T
|Ntn+1 Ntn | |Ntn+1 Nt | + |Nt Ntn | 2 (n+1)
+2 n
2.2 n
(5.3.11)
and hence (using that (Ns ) is a martingale), one has
4 4
for i 6= j
2 2 F
E[(Ztn,j )2 ] 2n E[(Nt^⇣j+1 Nt^⇣j )2 ] = 2n E[(Nt^⇣j+1 )2 (Nt^⇣j )2 ].
j=0 (5.3.13)
X1
16
E[(Nt^⇣j+1 )2 (Nt^⇣j )2 ]
22n
j=0
16
E[(Nt )2 ].
22n
Thus, recalling that Ytn+1 , Ytn are martingales, it follows that Ytn+1 Ytn is also a martingale
and thus invoking Doob’s maximal inequality, one has (using (5.3.13))
E[supsT |Ysn+1 Ysn |2 ] 4E(YTn+1 YTn )2
64 (5.3.14)
2n E[NT2 ].
2
5.3. Quadratic Variation of a Square Integrable Martingale 169
T
completes the proof.
Exercise 5.15. Use completeness of the underlying -field to show that the set {! : N· (!) 2
e appearing above is measurable.
D}
F
We are now in a position to prove an analogue of the Doob-Meyer decomposition
theorem for the square of an r.c.l.l. locally square integrable martingale. We will use the
A
notation [N, N ] for the process A = (N.(!)) of the previous result and call it quadratic
variation of N . We will later show that square integrable martingales and locally square
integrable martingales are stochastic integrators. Then it would follow that the quadratic
R
variation defined for a stochastic integrator X via (4.6.2) agrees with the definition given
below for a square integrable martingale and a locally square integrable martingale.
Theorem 5.16. Let (Mt , Ft ) be an r.c.l.l. locally square integrable martingale i.e. there
D
exist stop times ⇣n increasing to 1 such that for each n, Mtn = Mt^⇣n is a martingale with
E[(Mt^⇣n )2 ] < 1 for all t, n. Let
[M, M ]t (!) = [ (M.(!))](t). (5.3.16)
Then
(i) ([M, M ]t ) is an (Ft ) adapted r.c.l.l. increasing process such that Xt = Mt2 [M, M ]t
is also a local martingale.
(ii)
P( [M, M ]t = ( Mt )2 , 8t > 0) = 1.
170 Chapter 5. Semimartingales
P( Bt = ( Mt )2 , 8t > 0) = 1
(iv) If M is a martingale and E(Mt2 ) < 1 for all t, then E([M, M ]t ) < 1 for all t and
Xt = Mt2 [M, M ]t is a martingale.
(v) If E([M, M ]t ) < 1 for all t and M0 = 0, then E(Mt2 ) < 1 for all t, (Mt ) is a
martingale and Xt = Mt2 [M, M ]t is a martingale.
T
Then ⌧k increases to 1 and Mtk = Mt^⌧k is a martingale satisfying conditions of Lemma
5.14 with C = k and ⌧ = ⌧k . Hence Xtk = (Mtk )2 [M k , M k ]t is a martingale, where
[M k , M k ]t = (M·k (!))t . Also,
F
e = 1, 8k
P({! : M·k (!) 2 D})
and
P({! : [M k , M k ]t (!) = [M, M ]t^⌧k (!) (!) 8t}) = 1.
R
It follows that Xt^⌧k = Xtk a.s. and since X k is a martingale for all k, it follows that
Xt is a local martingale. This completes the proof of part (i).
D
Ut = [M, M ]t Bt
Hence Wt = Ut2 is a local martingale. Now if k are stop times increasing to 1 such that
Wt^ k is a martingale for k 1, then we have
2
E[Wt^ k ] = E[Ut^ k
] = E[U02 ] = 0.
and hence Ut^ 2 = 0 for each k. This yields Ut = 0 a.s. for every t. This completes the
k
proof of (iii).
For (iv), we have proven in (i) that Xt = Mt2 [M, M ]t is a local martingale. Let k
be stop times increasing to 1 such that Xtk = Xt^ k are martingales. Hence, E[Xtk ] = 0,
or
2
E([M, M ]t^ k ) = E(Mt^ k
) E(M02 ). (5.3.19)
Hence
2
E([M, M ]t^ k ) E(Mt^ k
) E(Mt2 ). (5.3.20)
T
Now Fatou’s lemma (or Monotone convergence theorem) gives
Corollary 5.17. For an r.c.l.l. martingale M with M0 = 0 and E[MT ]2 < 1, one has
The inequality (5.3.22) now follows from Doob’s maximal inequality, Theorem 2.24.
172 Chapter 5. Semimartingales
Corollary 5.18. For an r.c.l.l. locally square integrable martingale M , for any stop time
, one has
E[ sup |Ms |2 ] 4E[ [M, M ] ] (5.3.24)
0s
Proof. If ⌧n " 1, then using (5.3.22) for the square integrable martingale Mtn = Mt^⌧n we
get
E[ sup |Ms |2 ] 4E[ [M, M ] ^⌧n ] 4E[ [M, M ] ].
0s ^⌧n
T
Proof. Invoking (iii) in Lemma 5.13, we conclude that [M, M ]t = 0 for all t and thus the
conclusion follows from Corollary 5.18.
F
Remark 5.20. The pathwise formula for quadratic variation of a continuous local martingale
M was proven in Karandikar [33], but the proof required the theory of stochastic integration.
A proof involving only Doob’s inequality as presented above for the case of continuous local
A
martingales was the main theme of Karandikar [34]. The formula for r.c.l.l. case was given in
Karandikar [37] but the proof required again the theory of stochastic integration. The treatment
given above is adapted from Karandikar - Rao [41].
R
T
Proof. The proof is almost exactly the same as proof of Lemma 3.8 and it uses Mt Nt
[M, N ]t is a martingale along with Theorem 2.57, Corollary 2.58 and Theorem 2.59.
Rt
Rt 2
0 fs d[M, M ]s are martingales and
Z t
2
F
Corollary 5.23. Let M 2 M2 and f 2 S. Then Yt = 0 f dM and Zt = (Yt )2
Z T
fs2 d[M, M ]s ].
A
E[ sup | f dM | ] 4E[ (5.4.1)
0tT 0 0
Proof. Lemma 5.22 gives Y, Z are martingales. The estimate (5.4.1) now follows from
Doob’s inequality.
R
Z t Z T
2
E[ sup | f dM | ] 4E[ fs2 d[M, M ]s ], 8T < 1. (5.4.2)
0tT 0 0
Proof. Fix T < 1. Suffices to prove the result for the case when Mt = Mt^T . The rest
follows by localization. See Theorem 4.45. Recall that ⌦ e = [0, 1) ⇥ ⌦ and P is the
e Let µ be the measure on (⌦,
predictable -field on ⌦. e P) defined for A 2 P
Z Z T
µ(A) = [ 1A (!, s)d[M, M ]s (!)]dP(!). (5.4.3)
0
Note that
e = E[ [M, M ]T ] = E[|MT |2 ] < 1
µ(⌦)
174 Chapter 5. Semimartingales
e P) ✓ L2 (⌦,
Clearly, B(⌦, e P, µ). Since (S) = P, it follows from Theorem 2.65 that S is
e P, µ). Thus, given f 2 B(⌦,
dense in L2 (⌦, e P), we can get f n 2 S such that
kf f n k2,µ 2 n 1
. (5.4.5)
Rt
Letting Ytn = 0 f n dM for t T and Ytn = YTn for t > T one has for m, n k
It then follows that (as in the proof of Lemma 3.10) Ytn converges uniformly in t to Yt
(a.s.), where Y is an r.c.l.l. adapted process with Yt = Yt^T . For any g 2 S, using the
T
estimate (5.4.1) for f n g, we get
Z t
E[ sup |Ytn gdM |2 ] 4kf n gk22,µ
0tT 0
E[ sup |Yt
F
and taking limit as n tends to infinity in the inequality above we get
Z t
gdM |2 ] 4kf gk22,µ . (5.4.6)
A
0tT 0
Rt
Let us denote Y as JM (f ). The equation (5.4.6) implies that for f 2 S, JM (f ) = 0 f dM .
Also, (5.4.6) implies that the process Y does not depend upon the choice of the particular
e P), a sequence hm 2 S approximating
sequence {f n } in (5.4.5). Further, taking h 2 B(⌦,
R
bp
The estimate (5.4.7) implies that if fn ! f , then JM (f n ) converges to JM (f ) in ucp
topology and thus M is a stochastic integrator.
The estimate (5.4.2) follows from (5.4.7) by taking h = 0.
Rt RT
Remains to show that Yt = 0 f dM and Zt = Yt2 0 fs2 d[M, M ]s are martingales. We
RT n 2
have seen in Corollary 5.23 that Y n and Ztn = (Ytn )2 0 (f )s d[M, M ]s are martingales.
Here Yt converges to Yt in L (P) (and hence in L (P)) and so Y is a martingale. Further,
n 2 1
and thus Z t
E| ((fsn )2 fs2 )d[M, M ]s | ! 0.
0
R
Let us now introduce a class of integrands f such that f dM is a locally square
integrable martingale.
Definition 5.25. For M 2 M2loc let L2m (M ) denote the class of predictable processes f such
T
that there exist stop times k " 1 with
Z k
E[ fs2 d[M, M ]s ] < 1 for k 1. (5.4.8)
0
R t 2 Rt 2
Ut = Yt2 0 f s d[M, M ] s is a local martingales, [Y, Y ] t = 0 fs d[M, M ]s . Further, for
any stop time , Z Zt
E[ sup | f dM |2 ] 4E[ fs2 d[M, M ]s ]. (5.4.10)
D
0t 0 0
Proof. Let {⇣k } be stop times such that M k = M [⇣k ] 2 M2 . Then M k is a stochastic
integrator by Theorem 5.24 and thus so is M by Theorem 4.45.
Now given f 2 L2m (M ), let k " 1 be stop times such that (5.4.8) is true and let
⌧k = k ^ ⇣k ^ k.
Let gn be bounded predictable processes, with |gn | |f | such that gn converges to 0
R
pointwise. Let Z n = gn dM . To prove f 2 L(M ), we need to show that ducp (Z n , 0)
converges to 0. In view of Lemma 2.72, suffices to show that for each k 1,
ucp
Y n,k = (Z n )[⌧k ] !0 as n ! 1. (5.4.11)
176 Chapter 5. Semimartingales
R
Note that Y n,k = gn 1[0,⌧k ] dM k . Also, Y n,k is a square integrable martingale since gn is
bounded and M k is a square integrable martingale. Moreover,
Z T
n,k n,k
E([Y , Y ]T ) = E( (gn )2 1[0,⌧k ] d[M k , M k ])
0
Z ⇣k
E( (gn )2 d[M, M ].
0
Sine gn converge pointwise to 0 and |gn | |f | and f satisfies (5.4.8), it follows that for
each k fixed, E([Y n,k , Y n,k ]T ) ! 0 as n ! 1. Invoking (5.3.22), we thus conclude
T
similar to the proof of part (iv) in Theorem 5.16. The estimate (5.4.10) follows invoking
(5.3.24) as Y 2 M2loc .
Remark 5.27. Now that we have shown that locally square integrable r.c.l.l. martingales M
F
are stochastic integrators, the quadratic variation of M defined via the mapping is consistent
with the definition of [M, M ] given in Theorem 4.59. As a consequence various identities and
inequalities that were proven for the quadratic variation of a stochastic integrator in Section
A
4.6 also apply to quadratic variation of locally square integrable martingales. Thus from now
on we will drop the superfix in [M, M ] .
R
D itself is continuous. Thus D is locally bounded and hence f 2 L2m (M ). It is easy to see
that if f 2 L2m (M ) then f satisfies (5.4.12).
E[ [M n M, M n M ]T ] ! 0 8T < 1. (5.4.13)
T
Earlier we have shown that processes whose paths have finite variation on [0, T ] for every
T are stochastic integrators. It would then follow that all semimartingales are stochastic
integrators. Here is the continuous analogue of the Burkholder’s inequality, Theorem 1.43.
F
Lemma 5.30. Let Z be a r.c.l.l. martingale with Z0 = 0 and f 2 S1 , namely a simple
predictable process bounded by 1. Then for all > 0, T < 1 we have
Z t
20
A
P( sup | f dZ| > ) E[|ZT | ]. (5.5.1)
0tT 0
Rt R sk 1
Note that for sk 1 t sk , defining Vtk = 0 f dZ 0 f dZ, we have Vtk = ak (Zt
Zsk 1 ) and hence
sup |Vtk | 2 sup |Zt | (5.5.5)
sk 1 tsk 0tT
R sk P
Also, note that 0 f dZ = kj=1 Uj (Mj Mj 1 ) and hence
Z t Z sk
sup | f dZ| max | f dZ| + max sup |Vtk |
0tT 0 1km 0 1km sk 1 tsk
k (5.5.6)
X
max | Uj (Mj Mj 1 )| + 2 sup |Zt |.
1km 0tT
j=1
Thus using (5.5.4) and (5.5.6) along with Theorem 2.24 we get
Z t k
X 3
P( sup | f dZ| > ) P( max | Uj (Mj Mj 1 )| > )
T
0tT 0 1km 4
j=1
1
+ P(2 sup |Zt | > )
0tT 4 (5.5.7)
36
3
20
F 8
E[|ZT | ] + E[|ZT | ]
= E[|ZT | ].
A
Lemma 5.31. Let M be a square integrable r.c.l.l. martingale with M0 = 0 and let g be a
bounded predictable process, |g| C. Then
R
Z t
20
P( sup | gdM | > ) C E[|MT | ]. (5.5.8)
0tT 0
D
Proof. When g is simple predictable process, the inequality (5.5.8) follows from Lemma
5.30. Since M is a stochastic integrator, the class of predictable processes g bounded by C
for which (5.5.8) is true is bp- closed and hence it includes all such processes.
Proof. In view of Theorem 4.30, in order to show that M is a stochastic integrator for the
filtration (F⇧ ), suffices to show that it is a stochastic integrator w.r.t. (F⇧+ ). Recall that M
being r.c.l.l., remains a uniformly integrable martingale w.r.t. (F⇧+ ).
5.5. Semimartingales are Stochastic Integrators 179
T
Since M k is a bounded (F⇧+ )-martingale, it is a square integrable (F⇧+ )-martingale and
hence a (F⇧+ )-stochastic integrator. We will first prove that M k is Cauchy in dem metric.
Note that for any f 2 S1 , using (5.5.8) we have for k, j n,
Z t Z t
P( sup | f dM k
0tT 0 0
F 20
f dM j | > ) E[|MTk MTj | ] = an
40
R
gdM in ducp , it follows that (5.5.8) continues to be true for uniformly integrable martin-
gales M .
D
The proof given above essentially also contains the proof of the following observations.
Proof. We have shown that uniformly integrable r.c.l.l. martingales are stochastic integra-
tors and hence by localization, all r.c.l.l. local martingales are integrators. Thus M is an in-
tegrator. Of course, r.c.l.l. processes A with finite variation paths, i.e. Var[0,T ] (A· (!)) < 1
for all T < 1 for all ! 2 ⌦, are integrators and thus X = M + A is a stochastic integra-
tor.
T
The aim of this section is to prove the converse to Theorem 5.34. These two results taken
constitute one version of The Bichteler-Dellacherie-Meyer-Mokobodzky Theorem.
Let Z be a stochastic integrator. Let
Bt = Z0 +
X
0<st
F
( Z)s 1{|( Z)s |>1} . (5.6.1)
A
Since paths of Z are r.c.l.l., for every !, there only finitely many jumps of Z of size greater
than 1 in [0, t] and thus B is a well defined r.c.l.l. adapted process whose paths are of finite
variation and hence B itself is a stochastic integrator. Thus Y = Z B is a stochastic
R
integrator and now jumps of Y are of magnitude at most 1. Now defining stop times ⌧n
for n 1 via
⌧ n = inf{t > 0 : |Yt | n or |Yt | n} (5.6.2)
D
n
and Y n = Y [⌧ ] (i.e. Ytn = Yt^⌧ n ), it follows that for each n, Y n is a stochastic integrator
by Lemma 4.43. Further Y n is bounded by n + 1, since its jumps are bounded by 1.
We will show that bounded stochastic integrators X can be decomposed as X = M + A
where M is a r.c.l.l. square integrable martingale and A is an r.c.l.l. process with finite
variation paths. We will also show that this decomposition is unique under a certain condi-
tion on A. This would help in piecing together {M n }, {An } obtained in the decomposition
Y n = M n + An of Y n to get a decomposition of Y into an r.c.l.l. locally square integrable
martingale and an r.c.l.l. process with finite variation paths.
The proof of these steps is split into several lemmas.
5.6. Stochastic Integrators are Semimartingales 181
Proof. The relation (5.3.23) and the hypothesis (5.6.3) implies that
T
Hence the sequence of processes {M n } is Cauchy in ucp metric and thus in view of Theorem
2.68 converges to an r.c.l.l. adapted process M and (5.6.6) is satisfied. Further, (5.6.6) also
implies that Msn converges to Ms in L2 (P) for each s and hence using Theorem 2.21 it
F
follows that M is a martingale, indeed a square integrable martingale. As a consequence
(MTn MT ) ! 0 in L2 (P) and using (5.3.23) it follows that (5.6.4) is true. The convergence
of M n to M in Emery topology follows from Theorem 5.29.
A
If M n in the Lemma above are continuous, it follows that M is also continuous. This
gives us the following.
Proof. The proof is very similar to the proof of existence of the projection operator on a
Hilbert space onto a closed subspace of the Hilbert space. Let
↵ = inf{E[[X M, X M ]T ] : M 2 M2 }.
Since E[[X, X]T ] < 1, it follows that ↵ < 1. For k 1, let M̃ k 2 M2 be such that
1
E[[X M̃ k , X M̃ k ]T ] ↵ + .
k
T
Define Mtk = M̃t^T
k , t 0, k 1. Then
1
E[[X M k, X M k ]T ] = E[[X M̃ k , X M̃ k ]T ] ↵ + .
k
we get
[Y k Y n, Y k
F
Applying the parallelogram identity (4.6.13) to Y k = 12 (X
Y n ]T = 2[Y k , Y k ]T + 2[Y n , Y n ]T
M k ), Y n = 21 (X
[Y k + Y n , Y k + Y n ]T
M n)
A
(5.6.11)
1
Note that Y k + Y n = X 2 (M
n + M k ) and since 12 (M n + M k ) 2 M2 , we have
E[[Y k + Y n , Y k + Y n ]T ] ↵
R
and hence
1 1 1 1
E[[Y k Y n, Y k Y n ]T ] 2( (↵ + )) + 2( (↵ + )) ↵ (5.6.12)
4 k 4 n
D
We now show that ↵ is attained for this M . Let us define Y = 12 (X M ). Then we have
Y n Y = 12 (M M n ) and hence (5.6.13) yields
E[[Y n Y, Y n Y ]T ] ! 0. (5.6.14)
5.6. Stochastic Integrators are Semimartingales 183
T
since M + uN 2 M2 . We thus have
u2 E[[N, N ]T ] 2uE[[N, X M ]T ] 0 for all u 2 R. (5.6.16)
F
This implies E[[N, X M ]T ] = 0. Now the result follows by setting A = X M . A is a
stochastic integrator because X is so by hypothesis and M has been proven to be so.
Recall that M2c denotes the class of continuous square integrable martingales. A small
A
modification of the proof above yields the following.
The next lemma shows that (5.6.10) implies an apparently stronger conclusion- that
[N, A] is a martingale for N 2 M2 .
184 Chapter 5. Semimartingales
Proof. Fix N 2 M2 . For any stop time , N [ ] is also a square integrable martingale and
[ ]
(4.6.9) gives [N [ ] , A]T = [N, A]T = [N, A]T ^ and thus we conclude E[[N, A]T ^ ] = 0.
Theorem 2.53 now implies that [N, A] is a martingale.
T
Remark 5.40. If U is a continuous square integrable martingale and is a stop time, U [ ]
is also a continuous square integrable martingale. Hence, arguments as in the proof of the
previous result yields that if a stochastic integrator Y satisfies (5.6.17), then
F
[Y, U ] is a martingale 8U 2 M2c .
The next result would tell us that essentially, the integrator A obtained above has finite
A
variation paths (under some additional conditions).
Then A is a process with finite variation paths: Var[0,T ] (A) < 1 a.s.
We will show that for all " > 0, 9K < 1 such that
where the supremum above is taken over all partitions of [0, T ]. Taking a sequence ⇡ n of
successively finer partitions such that (⇡ n ) ! 0 (for example, ⇡ n = {kT 2 n : 0 k
n
2n }), it follows that V ⇡ " Var[0,T ] (A) and thus (5.6.18) would imply
P(Var[0,T ] (A) K) < "
and hence that P(Var[0,T ] (A) < 1) = 1. This would complete the proof.
Fix " > 0. Since A is a stochastic integrator for the filtration (F⇧ ), it is also a stochastic
integrator for the filtration (F⇧+ ) in view of Theorem 4.3. Thus we can get (see Remark
4.23) J1 < 1 such that
Z T
"
sup P(| f dA| J1 ) . (5.6.19)
f 2S1 (F + ) 0 6
⇧
T
6
Let J = max{J1 , J2 , E[B]} and n be such that 24 n < ". Let K = (n + 1)J. We will show
that (5.6.18) holds for this choice of K. Note that the choice has been made independent
of a partition.
Now fix a partition ⇡ = {0 = t0 < t1 < . . . < tm = T }. Recall that for x 2 R,F
sgn(x) = 1 for x 0 and sgn(x) = 1 for x < 0, so that |x| = sgn(x)x. For 1 j m, let
A
us consider the (F⇧+ )-martingale
Z̃tj = E[sgn(Atj Atj 1 ) | Ft+ ].
Since the filtration (F⇧+ ) is right continuous, the martingale Z̃tj admits an r.c.l.l. version
R
0 1 ,tj ]
(4.6.7)
|(Atj Atj 1 )| = Ztjj Ctjj
Z tj Z tj
j
= j
Zs dCs + Csj dZsj + [Z j , C j ]tj
0 0
Z tj Z tj
j
= Zs dAs + Csj dZsj + [Z j , A]tj [Z j , A]tj 1
tj 1 tj 1
and for tj 1 t tj
Z t Z t
Ztj Ctj = Zsj dAs + Csj dZsj + [Z j , A]t [Z j , A]tj 1 .
tj 1 tj 1
186 Chapter 5. Semimartingales
Let us define
m
X
Zt = Ztj 1(tj 1 ,tj ]
(t),
j=1
m
X
Ct = Ctj 1(tj 1 ,tj ]
(t),
j=1
m
X j j
Mt = (Zt^t j
Zt^t j 1
).
j=1
T
t t
Yt = Zs dAs + Cs dMs + [M, A]t
0 0
we get, for tk 1 t < tk , 1 k m
k 1
and thus Yt
Yt =
j=1
|(Atj F Atj 1 )| + Ztk Ctk
A
m
X
YT = |(Atj Atj 1 )| = V ⇡ .
j=1
Rt
Let Ut = 0 Cs dMs + [M, A]t , it follows that U is a (F⇧+ )-local martingale since by
R
Rt
assumption on A, [M, A] is itself a (F⇧+ )-martingale and thus 0 Cs dMs is a (F⇧+ )local
martingale by Theorem 5.26. Further
Z t
D
Yt = Zs dAs + Ut .
0
Now defining
⌧ = inf{t 0 : Ut < 3J} ^ T
it follows that ⌧ is a stop time (see Lemma 2.46) and
{⌧ < T } ✓ {U⌧ 3J}.
Since Y⌧ 2B, we note that
({⌧ < T } \ {B < J}) ✓ ({U⌧ 3J} \ {Y⌧ 2J}
Z ⌧
✓{ Zs dAs J}.
0
5.6. Stochastic Integrators are Semimartingales 187
T
Since V ⇡ = YT = 0 Zs dAs + UT and recalling that K = (n + 1)J
Z T
⇡
P(V K) P( Zs dAs J) + P(UT nJ)
0
"
6
" "
+ +
6 3 nJ
F
+ P(⌧ < T ) + P(U⌧ nJ)
1
E[(U⌧ )+ ]
A
" 11J
+
2 nJ
" "
< + ="
2 2
24
R
since by our choice n < ". This proves (5.6.18) and completes the proof as noted earlier.
We now put together the results obtained earlier in this chapter to get the following
D
key step in the main theorem of the section. We have to avoid assuming right continuity
of the filtration in the main theorem. However, it required in the previous result and we
avoid the same by an interesting argument. Here is a lemma that is useful here and in
later chapter.
Lemma 5.42. Let (⌦, F, P) be a complete probability space and let H ✓ G be sub- fields
of F such that H contains all the P null sets in F. Let Z be an integrable G measurable
random variable such that for all G measurable bounded random variables U one has
E[ZU ] = E[ZE[U | H]]. (5.6.24)
Then Z is H measurable.
188 Chapter 5. Semimartingales
using (5.6.24) it follows that for all G measurable bounded random variables U
Taking U = sgn(Z E[Z | H]) (here sgn(x) = 1 for x 0 and sgn(x) = 1 for x < 0), we
conclude from (5.6.25) that
E[ |(Z E[Z | H])| ] = 0.
T
(i) Xt = Xt^T for all t,
such that
[N, A] is a martingale for all N 2 M2 (5.6.27)
R
and
E[[X, X]T ] = E[[M, M ]T ] + E[[A, A]T ] (5.6.28)
D
and further, the decomposition (5.6.26) is unique under the requirement (5.6.27).
Proof. As seen in Theorem 4.30, X being a stochastic integrator for the filtration (F⇧ )
implies that X is also a stochastic integrator for the filtration (F⇧+ ). Also, (4.6.2) implies
that [X, X] does not depend upon the underlying filtration, so the assumptions of the
Theorem continue to be true when we take the underlying filtration to be (F⇧+ ). Now
Theorem 5.37 yields a decomposition X = M̃ + Ã with M̃ 2 M2 (F⇧+ ) and E[[N, Ã]T ] = 0
for all N 2 M 2 (F⇧+ ). Let Mt = M̃t^T , At = Ãt^T . Then M 2 M2 (F⇧+ ) and E[[N, A]T ] = 0
for all N 2 M 2 (F⇧+ ) since [N, A]t = [N, Ã]t^T (by (4.6.9)). As a consequence,
Thus E[[A, A]T ] < 1 and then by Lemma 5.39, we have [N, A] is a (F⇧+ )-martingale
for all N 2 M2 (F⇧+ ). Since the underlying filtration (F⇧+ ) is right continuous, Lemma 5.41
implies that A 2 V, namely paths of A have finite variation. By construction, M, A are
(F⇧+ ) adapted. Since for s < t, Fs+ ✓ Ft , it follows that At = At is Ft measurable. We
will show that
At is Ft measurable 8t > 0. (5.6.29)
Since Xt is Ft measurable, it would follow that Mt = Xt At is also Ft measurable. This
will also imply M 2 M2 = M2 (F⇧ ) completing the proof.
Fix t > 0. Let U be a bounded Ft+ measurable random variable and let V = U E[U |
Ft ]. Let
Ns = V 1[t,1) (s)
i.e. Ns = 0 for s < t and Ns = V for s t. It is easy to see that N 2 M2 (F⇧+ ) and
T
that [N, A]s = V ( A)t 1[t,1) (s) (using (4.6.10)). Thus [N, A] is a (F⇧+ )-martingale, and in
particular E[[N, A]t ] = 0. Hence we conclude that for all bounded Ft+ measurable U
E[( A)t U ] = E[( A)t E[U | Ft ]]. (5.6.30)
F
Invoking Lemma 5.42 we conclude that ( A)t is Ft measurable and hence that At is Ft
measurable. As noted earlier, this implies M 2 M2 . Only remains to prove uniqueness of
A
decomposition satisfying 5.6.27. Let X = Z + B be another decomposition with Z 2 M2 ,
B 2 V and [N, B] being a martingale for all N 2 M2 .
Now X = M + A = Z + B, B A = M Z 2 M2 and [N, B A] is a martingale
R
M Z 2 M2 and M0 = Z0 = 0)
E[sup|Ms Zs |] = 0.
sT
Proof. Note that Xt = Xt^T . Let Rt = Mt^T and Bt = At^T . Then X = R + B is also a
decomposition that satisfies (5.6.26) since R is also a square integrable martingale, B is a
190 Chapter 5. Semimartingales
process with finite variation paths and if R is any square integrable martingale, then
[N, B]t = [N, A]t^T
and hence [N, B] is also a martingale. Now uniqueness part of Theorem 5.43 implies
(5.6.31).
T
We now introduce two important definitions.
Definition 5.46. An adapted process B is said to be locally integrable if there exist stop
P(! : sup
0t⌧n (!)
F
times ⌧n increasing to 1 and random variables Dn such that E[Dn ] < 1 and
|Bt (!)| Dn (!)) = 1 8n 1.
A
The condition above is meaningful even if sup0t⌧n (!) |Bt (!)| is not measurable. It
is to be interpreted as - there exists a set ⌦0 2 F with P(⌦0 ) = 1 such that the above
inequality holds for ! 2 ⌦0 .
R
Definition 5.47. An adapted process B is said to be locally square integrable if there exist
stop times ⌧n increasing to 1 and random variables Dn such that E[Dn2 ] < 1 and
D
Theorem 5.48. Let X be locally square integrable stochastic integrator. Then X admits a
decomposition X = M + A where M 2 M2loc (M is a locally square integrable martingale)
and A 2 V (A is a process with finite variation paths) satisfying
[A, N ] 2 M2loc 8N 2 M2loc . (5.6.32)
Further, such a decomposition is unique.
Proof. Since X is a locally square integrable process, it follows that X is locally square
integrable. Since [X, X] = ( X)2 , it follows that [X, X] is locally integrable and thus so
is Dt = supst |Xs |2 + [X, X]t . Let n " 1 be stop times such that E[D n ] < 1 and let
n
⌧ n = n ^ n . Let X n = X [⌧ ] . Then X n satisfies conditions of Theorem 5.43 (with T = n)
and thus we can get decomposition X n = M n + An such that M n 2 M2 . An 2 V and
[U, An ] is a martingale for all U 2 M2 . (5.6.33)
T
Using Corollary 5.45, we can see that
P(Mtn = Mtk 8t ⌧ n ^ ⌧ k ) = 1, 8n, k.
n n
Thus we can define r.c.l.l. processes M, A such that M [⌧ ] = M n and A[⌧ ] = An . This
decomposition satisfies the asserted properties.F
Uniqueness follows as in Theorem 5.43 and the observation that if Y 2 M2loc , Y0 = 0
A
and [Y, Y ]t = 0 for all t then Y = 0 (i.e. P(Yt = 0 8t) = 1.)
Remark 5.49. The process A with finite variation r.c.l.l. paths appearing in the above theorem
was called a Natural process by Meyer and it appeared in the Doob Meyer decomposition of
R
supermartingales. Later it was shown that such a process is indeed a predictable process. A is
also known as the compensator of X. We will come back to this in Chapter 8 later
Corollary 5.50. Let X be a locally square integrable stochastic integrator and A be its
D
compensator and M = X A 2 M2loc . Then for any stop time such that E[ [X, X] ] < 1,
E[ [A, A] ] E[ [X, X] ] (5.6.34)
and
E[ [M, M ] ] E[ [X, X] ] (5.6.35)
Proof. If ⌧n " 1 are as in the proof of Theorem 5.48, then by Theorem 5.43 we have
E[[X, X] ^⌧n ] = E[[M, M ] ^⌧n ] + E[[A, A] ^⌧n ].
and the required inequalities folow by taking limit as n ! 1 and using monotone conver-
gence theorem.
192 Chapter 5. Semimartingales
Arguments similar to the ones leading to Theorem 5.48 yield the following (we use
Theorem 5.38 and Remark 5.40). We also use the fact that every continuous process is
locally bounded and hence locally square integrable.
Theorem 5.51. Let X be locally square integrable stochastic integrator. Then X admits
a decomposition X = N + Y with N 2 Mc,loc ( N is a continuous locally square integrable
martingale with N0 = 0) and Y is a locally square integrable stochastic integrator satisfying
Proof. The only new part is to show that (5.6.36) yields (5.6.37). For this note that on
T
one hand [Y, U ] is continuous as [Y, U ] = ( Y )( U ) and U is continuous and on the
other hand by definition [Y, U ] has finite variation paths. Thus [Y, U ] is a continuous local
martingale with finite variation paths. Hence by Theorem 5.19, [Y, U ] = 0.
F
As an immediate consequence of the Theorem 5.48, here is a version of the Bichteler-
Dellacherie-Meyer-Mokobodzky Theorem. See Theorem 5.78 for the final version.
A
Theorem 5.52. Let X be an r.c.l.l. adapted process. Then X is a stochastic integrator if
and only if X is a semimartingale.
R
Proof. We have already proved (in Theorem 5.34) that if X is a semimartingale then X is
a stochastic integrator.
For the other part, let X be a stochastic integrator. Let us define
D
X
Bt = X0 + ( X)s 1{|( X)s |>1} . (5.6.38)
0<st
Then as noted at the beginning of the section, B is an adapted r.c.l.l. process with finite
variation paths and is thus a stochastic integrator. Hence Z = X B is also a stochastic
integrator. By definition,
( Z) = ( X)1{| X|1}
and hence jumps of Z are bounded by 1. Hence Z is locally square integrable. Hence by
Theorem 5.48, Z admits a decomposition Z = M + A where M 2 M2loc and A is a process
with finite variation paths. Thus X = M + (B + A) and thus X is a semimartingale.
5.6. Stochastic Integrators are Semimartingales 193
The result proven above contains a proof of the following fact, which we record here
for later reference.
T
Using the technique of separating large jumps from a semimartingale to get a locally
bounded semimartingale used in proof of Theorem 5.78, we can get the following extension
of Theorem 5.51.
F
Theorem 5.55. Let X be a stochastic integrator. Then X admits a decomposition X =
N +S with N 2 Mc,loc ) ( N is a continuous locally square integrable martingale with N0 = 0)
A
and S is a stochastic integrator satisfying
Proof. Let B be defined by (5.6.38) and let Z = X B. Then Z is locally bounded and
D
Definition 5.56. Let X be a semimartingale. The continuous local martingale N such that
X = N + S and [S, U ] = 0 for all U 2 Mc,loc is said to be the continuous local martingale
part of X and is denoted by X (c) .
194 Chapter 5. Semimartingales
For a semimartingale X with X = X (c) + Z, we can see that [X, X] = [X (c) , X (c) ] +
[Z, Z]. We will later show in Theorem 8.76 that indeed [X (c) , X (c) ] = [X, X](c) - the
continuous part of the quadratic variation of X.
The next result shows that if X is a continuous process that is a semimartingale, then
it can be uniquely decomposed as a sum of a continuous local martingale and a continuous
process with finite variation paths.
Theorem 5.57. Let X be a continuous process and further let X be a semimartingale.
Then X can be uniquely decomposed as
X =M +A
where M and A are continuous processes, M0 = 0, M a local martingale and A a process
with finite variation paths.
T
Proof. Without loss of generality, we assume X0 = 0. Now X being continuous, it follows
that X is locally square integrable and thus X admits a decomposition X = M + A with
M 2 M2loc and A 2 V with A satisfying (5.6.32). On one hand, continuity of X implies
( M )t = ( A)t for all t > 0 and since A 2 V, we have
[A, M ]t =
X
F
( A)s ( M )s .
0<st
A
Thus,
X
[A, M ]t = ( A)2s = [A, A]t .
0<st
Since A satisfies (5.6.32), it follows that [A, M ] = [A, A] is a local martingale. If n is a
R
localizing sequence, it follows that E[[A, A]t^ n ] = 0 for all n. Since [A, A]s 0 for all s, it
follows that [A, A]t^ n = 0 a.s. for all t, n. This implies
X
[A, A]t = ( A)2s = 0 a.s. 8t
D
0<st
and hence A is a continuous process and hence so is M = X A. Uniqueness follows from
Theorem 5.19.
T
Proof. Let
X
Bt = X0 + ( X)s 1{|( X)s |>1}
0<st
F
and let Y = X B. Then B is an adapted r.c.l.l. process with finite variation paths and Y
is a stochastic integrator with jumps bounded by 1 and hence is locally square integrable.
Then invoking Theorem 5.51, we get a decomposition
A
Y =M +A
where M 2 Mc,loc and A is a locally square integrable stochastic integrator such that
R
[A, S]t = 0 for all S 2 M2c . Since B 2 V, it follows that [B, S] = 0 for all S 2 M2c . Defining
N = A + B = X M , we get that [N, S] = 0 for all S 2 M2c and since X, M are local
martingales, it follows that so is N .
D
Remark 5.60. Here, if X 2 M2loc then M 2 M2c,loc and N 2 M2d,loc . We will later show in
Theorem 8.73 that in this case [M, M ] is the continuous part of [X, X] and [N, N ] is the sum
of squares of jumps of X.
T
R
Y = (1 + |f |)dX. Then as noted in Theorem 4.29, Y is a stochastic integrator. Hence by
Corollary 5.53, Y admits a decomposition Y = N + B and with N 2 M2loc and B 2 V. Let
Z
M = (1 + |f |) 1 dN,
Z F
A = (1 + |f |) 1 dB.
A
The two integrals are defined as (1 + |f |) 1 is bounded. Further, this also yields, M 2 M2loc
and A 2 V. Clearly
Z Z
M + A = (1 + |f |) 1 dY = (1 + |f |) 1 (1 + |f |)dX = X.
R
T
{M,A :X=M +A, M 2M2loc , A2V}
(iv) A 2 V.
Rt
(v) Let Bt = |A|t . Show that 0 f (s)dBs = 1.
P1 P 2m 1
1
(vi) Show that [A, A]t = m=1 k=1 1 n,m ,1) (t).
24m [a
R1
(vii) Show that 0 f 2 (s)d[A, A]s < 1.
Rt
T
Thus 0 f dA is defined as a stochastic integral but not defined as a Riemann-Stieltjes integral.
Definition 5.67. Let X be an r.c.l.l. adapted process. Then X is said to be a weak stochastic
D
integrator if
Towards this goal, we introduce some notation. Let S+ be the class of stochastic
processes f of the form
m
X
fs (!) = aj+1 (!)1(sj ,sj+1 ] (s) (5.8.1)
j=0
where 0 = s0 < s1 < s2 < . . . < sm+1 < 1, aj+1 is bounded Fs+j measurable random
variable, 0 j m, m 1 and let C be the class of stochastic processes f of the form
Xn
fs (!) = bj+1 (!)1( j (!), j+1 (!)] (s) (5.8.2)
j=0
where 0 = 0 1 2 . . . , n+1 < 1 are (F·+ ) bounded stop times and bj+1 is
bounded F +j measurable random variable, 0 j n, n 1.
For f 2 C, let us define
n
X
T
IX (f )t (!) = bj+1 (!)(X j+1 ^t (!) X j ^t (!)). (5.8.3)
j=0
n
X
gs = bj+1 1( j ^⌧, j+1 ^⌧ ]
(s)
j=0
D
Since IX (f ) and IX (g) are defined pathwise, we can verify that the relation (5.8.4) is
true.
Lemma 5.69. Let X be an r.c.l.l. adapted process. Then X is a weak stochastic integrator
if and only if X satisfies the following condition for each t < 1:
8" > 0 9K" < 1 s.t. [ sup P(|JX (f )t | > K" )] ". (5.8.5)
f 2S, |f |1
200 Chapter 5. Semimartingales
Proof. If X satisfies (5.8.5), then given f n 2 S, an = supt,! |ftn (!)| ! 0, we need to show
JX (f n )t ! 0 in probability. So given " > 0, ⌘ > 0, get K" as in (5.8.5). Let n0 be such
that for n n0 , we have an K" < ⌘. Then gn = a1n f n 2 S and |gn | 1. Hence
T
This contradicts the assumption that X is a weak stochastic integrator.
Lemma 5.70. Let X be an r.c.l.l. adapted process. Then X satisfies (5.8.5) if and only if
for each t < 1
where 0 = s0 < s1 < s2 < . . . < sm < 1, aj+1 is bounded Fs+j measurable random variable,
D
0 j (m 1). Let 0 < k < k1 be such that sj + k < sj+1 , 0 j m and let
m
X1
gsk (!) = aj+1 (!)1(sj + k ,sj+1 ]
(s).
j=0
Since Fs+j ✓ Fsj + k , it follows that g k 2 S. Noting that g k converges to f and using the
explicit formula for IX , JX , we see that (the number of terms remain fixed!)
JX (g k )t ! IX (f )t .
Let us note that if f 2 C is given by (5.8.2) with j being simple stop times, namely taking
finitely many values, then f 2 S+ . To see this, let us order all the values that the stop
times j j = 0, 1, . . . , n take and let the ordered list be s0 < s1 < s2 < s3 < . . . < sk .
Recall that by construction, f is l.c.r.l. adapted and thus
bj = lim f (u)
u#sj
T
Now returning to the proof of (5.8.7), let f 2 C be given by
n
X1
fs = bj+1 1( j , j+1 ]
(s)
j=0
F
with 0 = 0 1 2 . . . , n < 1 are (F·+ ) bounded stop times and bj is bounded
F +j measurable random variable, 0 j (n 1). Let
A
[2m
j] + 1
⌧jm = , m 1
2m
be simple stop times decreasing to j , 0 j n, and let
R
n
X1
gsm = bj+1 1(⌧jm ,⌧j+1
m ] (s).
j=0
Then g m converges to f and from the explicit expression for IX (g m ) and IX (f ), it follows
D
The preceding two lemmas lead to the following interesting result. The convergence
for each t in the definition of weak stochastic integrator leads to the apparently stronger
result- namely convergence in ducp .
202 Chapter 5. Semimartingales
Proof. Fix f n 2 C such that f n ! 0 uniformly. Proceeding as in the proof of Lemma 5.69,
one can show using (5.8.7) that for each t,
IX (f n )t ! 0 in probability. (5.8.10)
Fix ⌘ > 0, T < 1. For n 1, let
⌧ n = inf{t 0 : |IX (f n )t | > 2⌘}.
⌧ n is a stop time with respect to the filtration (F·+ ). Let g n = f n 1[0,⌧ n ] . Note that g n 2 C.
T
In view of Lemma 5.68 we have
IX (g n )T = IX (f n )T ^⌧ n .
Also, from definition of ⌧ n , we have
0tT
F
{ sup |IX (f n )t | > 2⌘} ✓ {|IX (g n )T | > ⌘}.
ucp ucp
hn ! 0 ) IX (hn ) ! 0. (5.8.12)
Proof. Fix T < 1 and let n1 = 1. For each k 2, get nk such that nk > nk 1 and
n nk ) P( sup |hnt | > k1 ) k1 .
0tT
satisfies
|ftn | 1
k for nk n < nk+1
and
{ sup |hnt | > k1 } = { n T }.
0tT
So
1
P( n T) k for nk n < nk+1 . (5.8.13)
Thus
f n converges to 0 uniformly (5.8.14)
and
P(IX (f n )t = IX (hn )t 8t 2 [0, T ]) P( n > T ). (5.8.15)
T
ucp
In view of Theorem 5.71, (5.8.14) implies IX (f n ) ! 0 and then (5.8.13) and (5.8.15)
yield the desired conclusion, namely
ucp
IX (hn ) ! 0.
F
In view of this result, for a weak stochastic integrator, we can extend IX continuously
A
to the closure C̄ of C in the ducp metric.
ucp
For g 2 C̄, let IX (g) be defined as limit of IX (g n ) where g n ! g. It is easy to see
R
that IX is well defined. If X is a stochastic integrator, IX agrees with gdX.
R
C̄ = {Z : Z 2 R0 }.
ucp
Proof. Let f 2 C̄ and let f n 2 C, f n ! f . Then Vtn = limu#t fun are r.c.l.l. adapted
processes and V n can be seen to be Cauchy in ducp and hence converge to V . Further,
using Theorem 2.69, it follows that a subsequence of V n converges to V uniformly on [0, T ]
for every T < 1, almost surely. Thus V 2 R0 . Now it follows that f = V .
n=0
Then
1
X
IX (Z m ) = Zt^⌧nm (X⌧n+1
m ^t X⌧nm ^t ) (5.8.17)
n=0
and
ucp
IX (Z m )
! IX (Z ). (5.8.18)
R
Thus, if X is also a stochastic integrator, IX (f ) = f dX.
T
Proof. Let us note that given m # 0, {⌧nm : n 1} m 1, satisfying (5.8.16) exist, (say
given in (4.5.12)).
For k 1, let
k
X
ucp
IX (Z m,k ) ! IX (Z m ).
This proves (5.8.17). Now Z m converges to Z uniformly and hence Z 2 C̄ and
ucp
IX (Z m ) ! IX (Z ).
D
Remark 5.75. Note that while dealing with weak stochastic integrators, we considered the
filtration F·+ for defining the class C, but the definition of weak stochastic integrator did not
require the underlying filtration to be right continuous.
Remark 5.76. A careful look at results in section 4.6 show that Theorem 4.59, Theorem
4.62 continue to be true if the underlying processes are weak stochastic integrators instead of
stochastic integrators. Instead of invoking Theorem 4.57, we can invoke Theorem 5.74. Thus
weak stochastic integrators X, Y admit quadratic variations [X, X], [Y, Y ] and cross-quadratic
variation [X, Y ]. Various results on quadratic variation obtained in section 4.6 continue to be
true for weak stochastic integrators.
Moreover, Theorem 5.37, Theorem 5.38 and Lemma 5.39 are true for weak stochastic
integrators as well since the proof only relies on quadratic variation. Likewise, Lemma 5.41 is
true for weak stochastic integrators since apart from quadratic variation, it relies on (4.2.20),
a property that holds for weak stochastic integrators as noted earlier in Remark 4.23.
As a consequence, Theorem 5.43, Theorem 5.48 and Theorem 5.51 are true for weak
stochastic integrators.
T
This discussion leads to the following result, whose proof is same as that of Theorem
5.52.
F
Theorem 5.77. Let X be a weak stochastic integrator. Then X is a semimartingale.
in probability 8t < 1.
ucp
(ii) If f n 2 S, f n ! 0 uniformly, then JX (f n ) ! 0.
D
ucp ucp
(iii) If f n 2 S, f n ! 0 , then JX (f n ) ! 0.
bp ucp
(iv) If f n 2 S, f n ! 0 , then JX (f n ) ! 0.
(v) X is a stochastic integrator, i.e. the mapping JX from S to R0 (⌦, (F⇧ ), P) has an
e P) 7! R0 (⌦, (F⇧ ), P) satisfying
extension JX : B(⌦,
bp ucp
f n ! f implies JX (f n ) ! JX (f ).
Proof. Equivalence of (v), (vi), (vii) has been proven in Theorem 5.52 and Corollary 5.53.
Clearly,
(v) ) (iv) ) (iii) ) (ii) ) (i).
And we have observed in Theorem 5.77 that (iii) implies (vi). This completes the proof.
T
The main result of the section is about enlargement of the underlying filtration (F⇧ ) by
adding a set A 2 F to each Ft . The surprising result is that a semimartingale for the
original filtration remains a semimartingale for the enlarged filtration. In the traditional
F
approach, this was a deep result as it required decomposition of the semimartingale into
local martinagle w.r.t. the enlarged filtration and a finite variation process.
Let A 2 F be fixed and let us define a filtration (G⇧ ) by
A
Gt = {(B \ A) [ (C \ Ac ) : B, C 2 Ft }. (5.9.1)
It is easy to see that Gt is a -field for all t 0 and (G⇧ ) is a filtration. Using the
R
⇠ = ⌘ 1A + ⇣ 1Ac . (5.9.2)
D
Let S(G⇧ ) denote the class of simple predictable process for the filtration (G⇧ ). Using (5.9.2)
it is easy to verify that for f 2 S(G⇧ ) 9 g, h 2 S(F⇧ ) such that
e
f (t, !) = g(t, !)1A (!) + h(t, !)1Ac (!) 8(t, !) 2 ⌦. (5.9.3)
With this we can now describe the connection between predictable processes for the two
filtrations.
Theorem 5.79. Let f be an (G⇧ )-predictable process. We can choose (F⇧ ) predictable
processes g, h such that (5.9.3) holds. Further, if f is bounded by a constant c, then g, h
can also be chosen to be bounded by the same constant c.
5.9. Enlargement of Filtration 207
Proof. Let K be the set of (G⇧ )-predictable process f for which the conclusion is true.
We have seen that K contains S(G⇧ ). We next show that K is closed under pointwise
convergence. This will show in particular that K is closed under bp-covergence and hence
that the conclusion is true for all bounded predictable processes. If f n 2 K with
f n = g n 1 A + hn 1 A c
T
f n = f 1{|f n |n} converges pointwise to f .
If f is bounded by c, we can replace g by g̃ = g 1{|g|c} and h by h̃ = h1{|h|c} .
Theorem 5.80. Let X be a semimartingale for the filtration (F⇧ ). Let A 2 F and (G⇧ ) be
defined by (5.9.1). F
Then X is also a semimartingale for the filtration (G⇧ ).
A
Proof. Let us denote the mapping defined by (4.2.1) and (4.2.2) for the filtration (G⇧ ) by
HX and let JX be the mapping for the filtration (F⇧ ). Since these mappings are defined
pathwise, it is easy to verify that if f 2 S(G⇧ ) and g, h 2 S(F⇧ ) are as in (5.9.3), then
R
We will prove that X is a weak-stochastic integrator for the filtration (G⇧ ). For this, let
e P(G⇧ )) decrease to 0 uniformly. Let an # 0 be such that |f n | an . Then for each
f n 2 B(⌦,
D
f n = g n 1A + g n 1Ac .
Remark 5.81. As noted in Theorem 4.11, when f is (F⇧ )-predictable bounded process,
R
HX (f ) = JX (f ) and thus the stochastic integral f dX is unambiguously defined.
Theorem 5.82. Let X be a stochastic integrator on (⌦, F, P) with a filtration (F⇧ ). Let
Q be a probability measure on (⌦, F) that is absolutely continuous w.r.t. P. Then X is a
stochastic integrator on (⌦, F, Q) and the stochastic integral under P is a version of the
integral under Q.
Proof. Let H be the completion of F under Q and let Ht be the -field obtained by adding
all Q-null sets in every Ht . It can be checked that a process f is (H⇧ )-predictable if and
only if f 1⌦0 is (G⇧ )-predictable. Let
e P(H⇧ )).
HX (f ) = JX (f 1⌦0 ) f 2 B(⌦,
It is easy to see that HX is the required extension of the integral of simple predictable
T
processes.
Remark 5.83. Suppose we start with a filtration (F̃⇧ ) that may not satisfy the condition that
each F̃t contains all null sets. Suppose X is a (F̃⇧ )-adapted process that satisfies (4.10.2) for
F
this filtration. Let Ft be the smallest -field containing F̃t and all the null sets. It is easy
to see that X continues to satisfy (4.10.2) w.r.t. the filtration (F⇧ ) and is thus a stochastic
integrator.
A
Exercise 5.84. Let X be a semimartingale for a filtration (F⇧ ) on (⌦, F, P). Let {Am :
m 1} be a partition of ⌦ with An 2 F for all n 1. For t 0 let
R
(i) For every (G⇧ )-predictable process f , there exists (F⇧ )-predictable processes f m such that
D
1
X
f= 1Am f m .
m=1
ucp
(ii) Suppose for each m 1, {Y m,n : n 1} are r.c.l.l. processes such that Y m,n ! Ym
as n ! 1. Let
1
X
Zn = 1Am Y m,n
m=1
and
1
X
Z= 1Am Y m .
m=1
5.9. Enlargement of Filtration 209
ucp
Then prove that Z n ! Z as n ! 1.
T
F
A
R
D
Chapter 6
6.1 Preliminaries
T
R
For a simple predictable process f , the stochastic integral f dX has been defined explicitly,
Rt
path by path. In other words, the path t 7! ( 0 f dX)(!) is a function of the paths
F
{fs (!) : 0 s t} and {Xs (!) : 0 s t} of the integrand f and integrator X. For
a general (bounded) predictable f the integral has been defined as limit in probability of
suitable approximations and it is not clear if we can obtain a pathwise version. In statistical
A
inference for stochastic process the estimate, in stochastic filtering theory the filter and in
stochastic control theory the control in most situations involves stochastic integral, where
the integrand and integrator are functionals of the observation path and to be meaningful,
R
If we have one probability measure Q such that each P↵ is absolutely continuous w.r.t.
Q and the underlying process X is a semimartingale under Q then the answer to the
question above is yes- simply take the integral defined under Q and that will agree with
the integral under P↵ for each ↵ by Remark 4.15.
When the family {P↵ } is countable family, such a Q can always be constructed. How-
ever, such a Q may not exist in general. Once concrete situation where such a situation
arises and has been considered in the literature is when we have a Markov Process- when
one considers the family of measures {Px : x 2 E}, where Px represents the distribution of
Markov Process conditioned on X0 = x. See [18]. In this context, Cinlar, Jacod, Protter
210
6.2. Pathwise Formula for the Stochastic Integral 211
and Sharp [9] showed the following : For processes S such that S is semimartingale under
Px for every x and for f in a suitable class of predictable processes, there exists a process
R
Z such that Z is a version of f dS under Px for every x.
In Bichteler [3], Karandikar [32], [33], [37] it was shown that for an r.c.l.l. adapted
process Z and a semimartingale X, suitably constructed Riemann sums converge almost
R
surely to the stochastic integral Z dX . This result was recast in [40] to obtain a universal
mapping : D([0, 1), R) ⇥ D([0, 1), R) 7! D([0, 1), R) such that if X is a semimartingale
and Z is an r.c.l.l. adapted process, then (Z, X) is a version of the stochastic integral
R
Z dX.
As in the previous chapter, we fix a filtration (F⇧ ) on a complete probability space
(⌦, F, P) and we assume that F0 contains all P-null sets in F.
First we will prove a simple result which enables us to go from L2 estimates to almost
T
sure convergence.
Lemma 6.1. Let V m be a sequence of r.c.l.l. process and ⌧k an increasing sequence of stop
times, increasing to 1 such that
1
X F
ksup |Vtm |k2 < 1.
m=1 t⌧k
(6.1.1)
A
Then we have
sup|Vtm | ! 0 8T < 1, a.s.
tT
R
Fix ↵ 2 D([0, 1), R). For each n 1; let {tni (↵) : i 1} be defined inductively as
follows : tn0 (↵) = 0 and having defined tni (↵), let
Note that for each ↵ 2 D([0, 1), R) and for n 1, tni (↵) " 1 as i " 1 (if limi tni (↵) = t⇤ <
1, then the function ↵ cannot have a left limit at t⇤ ). For ↵, 2 D([0, 1), R) let
1
X
n (↵, )(t) = ↵(tni (↵) ^ t)( (tni+1 (↵) ^ t) (tni (↵) ^ t)). (6.2.2)
i=0
Since tni (↵) increases to infinity, for each ↵ and t fixed, the infinite sum appearing above
is essentially a finite sum and hence n (↵, ) is itself an r.c.l.l. function. We now define a
mapping : D([0, 1), R) ⇥ D([0, 1), R) 7! D([0, 1), R) as follows: Let D⇤ ✓ D([0, 1), R) ⇥
D([0, 1), R) be defined by
T
D⇤ = {(↵, ) : n (↵, ) converges in ucc topology}
(↵, ) =
<lim
:0
n F
n (↵, ) if (↵, ) 2 D⇤
otherwise.
(6.2.3)
A
Note that the mapping has been defined without any reference to a probability measure
or a process. Here is the main result on Pathwise integration formula.
Then Z
Z= U dX. (6.2.5)
and is thus a stop time (for all n, i). Let us note that n = tni (U· (!)). Let Z·n (!) =
i (!)
n (U· (!), X· (!)). Then we can see that
1
X
Ztn = Ut^ jn (Xt^ n
j+1
Xt^ jn )
j=0
6.2. Pathwise Formula for the Stochastic Integral 213
R
and thus Z n = U n dX where
1
X
Utn = Ut^ jn 1( n n
j , j+1 ]
(t).
j=0
Since by definition of { n {U n }, we have
i },
|Utn Ut | 2 n
(6.2.6)
R
and hence Un! U in ucp. Then by Theorem 4.46, ! Z = U dX in the ucp Zn
metric.
The crux of the argument is to show that the convergence is indeed almost sure-
Z t Z t
n
sup| U dX U dX| ! 0 8T < 1 a.s. (6.2.7)
tT 0 0
Once this is shown, it would follow that (U· (!), X· (!)) 2 D⇤ a.s. and then by definition of
and Z we conclude that Zn = (U n , X) converges to (U, X) in ucc topology almost
T
surely. Since Z n ! Z in ucp, we have Z = (U, X) completing the proof.
Remains to prove (6.2.7). For this, first using Corollary 5.53, let us decompose X as
X = M + A, M 2 M2loc and A 2 V. Now using the fact that the dA integral is just the
Lebesgue - Stieltjes integral and the estimate (6.2.6) we get
Z t
| U dA
0
n
Z t
U dA|
Z t
0
|Usn Us |d|A|t
F 0
A
n
2 |A|t .
and hence Z Z
t t
sup| U n dA U dA| 2 n
|A|T
R
tT 0 0 (6.2.8)
! 0.
Thus (6.2.7) would follow in view of linearity of the integral once we show
Z t Z t
D
n
[sup| U dM U dM | ] ! 0 8T < 1 a.s. (6.2.9)
tT 0 0
k]
Let ⌧k be stop times increasing to 1 such that ⌧k k and M [⌧ is a square integrable
martingale so that
k k
E[ [M, M ]⌧k ] = E[ [M [⌧ ] , M [⌧ ] ]k ] < 1. (6.2.10)
Thus using the estimate (5.4.10) on the growth of the stochastic integral with respect to a
local martingale (Theorem 5.26), we get
Z t Z t Z ⌧k
E[sup | U n dM U dM |2 ] 4E[ |Usn Us |2 d[M, M ]s ]
t⌧k 0 0 0 (6.2.11)
2n
4(2 )E[ [M, M ]⌧k ].
214 Chapter 6. Pathwise Formula for the Stochastic Integral
Rt Rt p
Thus, writing ⇠tn = 0 U n dM 0 U dM and ↵k = E[ [M, M ]⌧k ], we have
Since ↵k < 1 as seen in (6.2.10), Lemma 6.1 implies that (6.2.9) is true completing the
proof.
R
Remark 6.3. This result implies that the integral U dX for an r.c.l.l. adapted process U
does not depend upon the underlying filtration or the probability measure or on the decom-
position of the semimartingale X into a (local) martingale and a process with finite variation
R·
paths. A path ! 7! 0 U dX(!) of the integral depends only on the paths ! 7! U· (!) of
the integrand and ! 7! X· (!) of the integrator. The same however cannot be said in general
R
about f dX if f is given to be a predictable process.
T
Remark 6.4. In Karandikar [37, 40] the same result was obtained with tni (↵) defined via
F
instead of (6.2.1). The result is of course true, but requires the underlying -fields were required
to be right continuous and to prove that the resulting jn are stop times required deeper results
which we have avoided in this book.
A
6.3 Pathwise Formula for Quadratic Variation
R
In Section 5.2 we had obtained a pathwise formula for the quadratic variation process of
a locally square integrable martingale - namely (5.3.16). We now observe that the same
formula gives quadratic variation of local martingales as well - indeed of semimartingales.
D
We will need to use notations from Section 5.2 as well as Section 6.2
Theorem 6.5. Let be the mapping defined in Section 5.2 by (5.2.2). For a semimartin-
gale X, let
[X, X]t (!) = [ (X.(!))](t). (6.3.1)
Proof. For ↵ 2 D([0, 1), R) and for n 1, let {tni (↵) : i 1} be defined inductively
by ba8. Recall the definition (5.2.1) of n (↵) and (6.2.2) of n . Using the identity
6.3. Pathwise Formula for Quadratic Variation 215
(b a)2 = b2 a2 2a(b a) with b = ↵(tni+1 (↵) ^ t) and a = ↵(tni (↵) ^ t) and summing
over i 2 {0, 1, 2 . . .}, we get the identity
Then using (6.3.2) along with the definition (6.2.3) of , it follows that
b✓D
D e
and
(↵) = (↵(t))2 (↵(0))2 b
2 (↵) ↵ 2 D. (6.3.3)
T
Z
X dX = (X, X) (6.3.4)
0
t
X dX. (6.3.5)
A
This along with (4.6.1) implies [X, X]t = [X, X]t completing the proof.
R
D
Chapter 7
Continuous Semimartingales
In this chapter we will consider continuous semimartingales and show that stochastic dif-
T
ferential equations driven by these can be analyzed essentially using the same techniques
as in the case of SDE driven by Brownian motion. This can be done using random time
change. The use of random time change in study of solutions to stochastic di↵erential
equations was introduced in [32], [33].
F
We introduce random time change and we then obtain a growth estimate on f dX
R
can assume that the driving semimartingale satisfies (7.2.2) and then use techniques used
for Brownian motion case. We thus show that stochastic di↵erential equation driven by
continuous semimartingales admit a solution when the coefficients are Lipschitz functions.
D
We also show that in this case, one can get a pathwise formula for the solution, like the
formula for the integral obtained in the previous chapter.
216
7.1. Random Time Change 217
which was derived in section 4.8. We had seen that when G is absolutely continuous, this
formula follows by the chain rule for derivatives. Let a : [0, 1) 7! [0, 1) be a continuous
strictly increasing one-one onto function. Let us write G̃(s) = G(a(s)). It can be seen that
(7.1.1) can equivalently be written as
Z t
f (G̃(t)) = f (0) + f 0 (G̃(s))dG̃(s). (7.1.2)
0
Exercise 7.1. Show that (7.1.1) holds if and only if (7.1.2) is true.
So to prove (7.1.1), suffices to prove (7.1.2) for a suitable choice of a(t). Let
a(s) = inf{t 0 : (t + G(t)) s}.
For this choice of a it can be seen that G̃ is a continuous increasing function and that for
0 u v < 1, G̃(v) G̃(u) v u so that G̃ is absolutely continuous and thus (7.1.2)
follows from chain rule.
T
When working with continuous semimartingales, the same idea yields interesting results-
of course, the time change t 7! a(t) has to be replaced by t 7! t , where t is a stop time.
F
Definition 7.2. A (F⇧ )-random time change = ( t ) is a family of (F⇧ )- stop times { t :
0 t < 1} such that for all ! 2 ⌦, t 7! t (!) is a continuos strictly increasing function from
[0, 1) onto [0, 1).
A
Example 7.3. Let A be a (F⇧ )-adapted continuous increasing process with A0 = 0. Then
s = inf{t 0 : (t + At ) s}
can be seen to be a (F⇧ )-random time change.
R
Example 7.4. Let B be a (F⇧ )-adapted continuous increasing process with B0 = 0 such that
B is strictly increasing and limt!1 Bt = 1 a.s.. Then
D
s = inf{t 0 : Bt s}
can be seen to be a (F⇧ )-random time change.
Recall the Definition 2.35 of the stopped -field. Given a (F⇧ )-random time change
= ( t ), we define a new filtration (G⇧ ) = (Gt ) as follows:
Gt = F t , 0 t < 1. (7.1.3)
Clearly, for s t, we have s t and hence Gs ✓ Gt and so {Gs } is a filtration. Further,
G0 = F0 . We will denote the filtration (G⇧ ) defined by (7.1.3) as ( F⇧ ). Given a process f ,
we define the process g = [f ] via
gs = f s 0 s < 1. (7.1.4)
218 Chapter 7. Continuous Semimartingales
The map f ! g is linear and preserves limits: if f n converges to f pointwise or in ucp then
so does the sequence g n . We also define = { t : 0 t < 1} via
t = inf{s 0 : s t} (7.1.5)
= [ ]⌧ 1 = ⌧. (7.1.6)
Note the appearance of [ ] 1 in the definition above. It is not difficult to see that
= ⌧.
T
Recall the definition (4.4.12) of X [⌧ ] , the process X stopped at ⌧ . For Y = [X], note that
[⌧ ]
( [X [⌧ ] ])s = X s
=X s ^⌧ =X s^ =X s^ = Ys^ = Ys[ ]
We will now prove few relations about random time change and its interplay with
(7.1.7)
A
notions discussed in the earlier chapters such as stop times, predictable processes, local
martingales, semimartingales and stochastic integrals.
(iii) Let U be a (F⇧ ) adapted r.c.l.l. process. Then V = [U ] is a (G⇧ ) adapted r.c.l.l.
process.
(iv) Let f be a (F⇧ ) bounded predictable process. Then g = [f ] is a (G⇧ ) bounded pre-
dictable process. If f is a (F⇧ ) locally bounded predictable process then g = [f ] is a
(G⇧ ) locally bounded predictable process.
(v) Let A be a (F⇧ ) adapted r.c.l.l. process with finite variation paths. Then B = [A] is
a (G⇧ ) adapted r.c.l.l. process with finite variation paths.
7.1. Random Time Change 219
Proof. Note that by Corollary 2.50, is (G⇧ ) adapted. For any a, t 2 [0, 1), note that
{ a t} = {a t }.
T
time. Since s 7! s is continuous strictly increasing function from [0, 1) onto itself, same
is true of s 7! s and hence = ( s ) is a random time change. This proves (i).
Now, for s 2 [0, 1)
bp-closed and by the part proved already, it contains bounded continuous (F⇧ ) adapted
processes and thus also contains bounded (F⇧ )-predictable processes. Now if f is (F⇧ )-
predictable and locally bounded, let ⌧ n be sequence of stop times, ⌧ n " 1 such that
D
n
fn = f [⌧ ] is bounded predictable. Then as shown above, [fn ] is also bounded predictable.
Let n = [⌧ n ]. As seen in (7.1.7),
n] n
g[ = [f [⌧ ] ] = [fn ]
n
and thus g [ ] is predictable. Now ⌧ n " 1 implies n " 1 and thus g is locally bounded
(G⇧ )-predictable process. This proves (iv).
For (v), we have already noted that B is (G⇧ ) adapted. And clearly,
For (vi), in order to prove that N is a (G⇧ ) local martingale, we will obtain a sequence
n of (G⇧ ) stop times increasing to 1 such that for all (G⇧ )- stop times ⇣,
E[N n ^⇣ ] = E[N0 ].
This will prove N [ n ] is a martingale and hence that N is a local martingale. Since M is a
local martingale, let ⌧˜n be stop times such M [⌧˜n ] is a martingale for each n where ⌧˜n " 1.
Let
⌧n = ⌧˜n ^ n ^ n .
Then ⌧n ⌧˜n and ⌧n " 1. Then M [⌧n ] is a martingale for each n and hence for all stop
times ⌘ one has
E[M⌧n ^⌘ ] = E[M0 ]. (7.1.8)
T
n
Let ⌘ = [⇣] = ⇣. Then by part (ii) above, ⌘ is a (F⇧ )-stop time. Note that
N n ^⇣
F
= M⌧n ^⌘ .
Further, M0 = N0 and thus (7.1.8) and (7.1.10) together imply (7.1.9) proving (vi).
(7.1.10)
A
Part (vii) follows from (v) and (vi) by decomposing the semimartingale X into a local
martingale M and a process with finite variation paths A : X = M + A. Then [X] =
Y = [M ] + [A] = N + B.
R
We can verify the validity of (viii) when f is a simple predictable process and then
easy to see that the class of processes for which (viii) is true is bp-closed and thus contains
all bounded predictable processes. We can then get the general case (of f being locally
D
bounded) by localization.
For (ix), note that
Z t
2 2
[X, X] = Xt X0 2 X dX.
0
Changing time in this equation, and using (viii), we get
Z t
2 2
[[X, X]]t =X t X0 2 X dX
0
Z t
2 2
=Yt Y0 2 Y dY
0
=[Y, Y ].
7.2. Growth estimate 221
For the last part, note that for T < 1, T0 < 1, > 0 one has (using Ys = X s )
Now given T < 1, > 0 and ✏ > 0, first get T0 such that P( T T0 ) < 2✏ and then for this
ucp
T0 , using X n ! X get n0 such that for n n0 one has P(sup0tT0 |Xsn Xs | ) < 2✏ .
Now, for n n0 we have
P( sup |Ysn Ys | ) < ✏.
0tT
T
Exercise 7.8. Let W be a Brownian motion and ⇠ be a (0, 1)-valued random variable
independent of W such that E[⇠] = 1. Let Ft = (⇠, Ws : 0 s t). Let t = t⇠. Show
that
(i)
(ii)
t is a stop time for each t.
R R R
To see this, we note f dX = f dM + f dA and for the dM integral we use Theorem
R R
5.26 and for the dA integral we use | f dA| |f |d|A|.
For process A, B 2 V+ (increasing adapted processes), we define A << B if Ct = Bt At
is an increasing process. The following observation will be used repeatedly in the rest of
this chapter: if A << B, then for all f
Z t Z t
| fs dAs | |fs |dBs .
0+ 0+
We introduce a notion of a amenable semimartingale and obtain a growth estimate on
integrals w.r.t. a amenable semimartingale which is similar to the one for Brownian motion.
T
[N, N ]t [N, N ]s (t s), |B|t |B|s (t s). (7.2.2)
Theorem 7.10. Suppose Y is a continuous amenable semimartingale. Then for any locally
bounded predictable f , and a stop time , one has
E[ sup |
0s ^T
Z s
0+
F
f dX|2 ] 2(4 + T )E[
Z ^T
|fs |2 ds].
0+
(7.2.3)
A
Proof. The condition (7.2.2) implies that t [N, N ]t and t |B|t are increasing processes.
This observation along with (7.2.1) yields
Z s Z ^T Z ^T
2 2
E[ sup | f dX| ] 8E[ |fs | ds] + 2(E[ |fs |ds])2 .
R
0s ^T 0+ 0+ 0+
0+ 0+
Remark 7.13. In the definition of amenable semimartingale, instead of (7.2.2) we could have
required that for some constant K < 1
[N, N ]t [N, N ]s K(t s), |B|t |B|s K(t s). (7.2.6)
The only di↵erence is that a constant K 2 would appear in the estimate (7.2.3)
Z s Z ^T
E[ sup | f dX|2 ] 2(4 + T )K 2 E[ |fs |2 ds]. (7.2.7)
0s ^T 0+ 0+
T
(7.2.2). Indeed, given finitely many semimartingales, we can choose one random time
change that does it as we see in the next result.
Xm
Vt = t + ([M j , M j ]t + |Aj |t ).
j=1
t = inf{s 0 : Vs t}
it follows that = ( t ) is a random time change. As noted earlier, Y j = [X j ] is a
semimartingale with canonical decomposition Y j = N j + B j where N j = [M j ], B j =
[Aj ]. Further, observing that for 1 j m, 0 s t < 1,
[N j , N j ]t [N j , N j ]s = [M j , M j ] t [M j , M j ] s V t V s =t s
and
|B j |t |B j |s = |Aj | t |Aj | s V t V s =t s,
it follows that this random time change satisfies the required condition.
224 Chapter 7. Continuous Semimartingales
T
Let Y 1 , Y 2 , . . . Y m be continuous semimartingales w.r.t. the filtration (F⇧ ). Here we
will consider an SDE
dUt = b(t, ·, U )dYt , t 0, U0 = ⇠0 (7.3.1)
F
where the functional b is given as follows. Recall that Cd = C([0, 1), Rd ). Let
and there is an increasing r.c.l.l. adapted process K such that for all ⇣1 , ⇣2 2 Cd ,
(ii) For any continuous (F⇧ ) adapted process V , Z defined by Zt = a(t, ·, V ) ( i.e. Zt (!) =
a(t, !, V (!)) ) is an r.c.l.l. (F⇧ ) adapted process.
7.3. Stochastic Di↵erential Equations 225
Proof. Since for fixed t, ⇣, the mapping ! 7! a(t, !, ⇣) is Ft measurable and in view of
(7.3.4), ⇣ 7! a(t, !, ⇣) is continuous for fixed t, !, it follows that (7.3.6) is true since Cd is
separable.
For part (ii), let us define a process V t by Vst = Vs^t . In view of assumption (7.3.3), Z
is an r.c.l.l. process.The fact that ! 7! V t (!) is Ft measurable along with (7.3.6) implies
that Zt = a(t, ·, V t ) is Ft measurable.
For part (iii), when ⌧ is a simple stop time, (7.3.7) follows from (7.3.6). For a general
bounded stop time ⌧ , the conclusion (7.3.7) follows by approximating ⌧ from above by
simple stop times and using right continuity of a(t, !, ⇣). For a general stop time ⌧ , (7.3.7)
follows by approximating ⌧ by ⌧ ^ n.
T
Let b(t, !, ⇣) = a(t , !, ⇣). Let 0 denote the process that is identically equal to zero.
Since (t, !) 7! b(t, !, 0) is an r.c.l.l. adapted process, using hypothesis (7.3.4), it follows
that for ⇣ 2 Cd
where
0st
F
sup ka(t, !, ⇣)k Kt0 (!)(1 + sup |⇣(s)|)
0st
(7.3.8)
A
Kt0 (!) = Kt (!) + sup ka(t, !, 0)k. (7.3.9)
0st
Here too, as in the Brownian motion case, a continuous (Rd -valued) adapted process U
is said to be a solution to the equation (7.3.1) if
Z t
U t = ⇠0 + b(s, ·, U )dYs (7.3.10)
D
0+
i.e. for 1 j d,
m Z
X t
Utj = ⇠0j + bjk (s, ·, U )dYsk
k=1 0+
T
Proof.
Z s
E[ sup | f dX|2 ]
0s ^T 0+
d
X
j=1
E[ sup |
F
0s ^T k=1
m Z
X
0+
s
fjk dX k |2 ]
A
d X
X m Z s
m E[ sup | fjk dX k |2 ]
j=1 k=1 0s ^T 0+
d X
X m Z ^T
R
0+
We will prove existence and uniqueness of solution of (7.3.1). When the driving semi-
martingale satisfies (7.2.2), the proof is almost the same as the proof when the driving
semimartingale is a Brownian Motion. We will prove this result without making any in-
tegrability assumptions on the initial condition ⇠0 and the uniqueness assertion is without
any moment condition. For this the following simple observation is important.
Remark 7.17. Suppose M is a square integrable martingale w.r.t. a filtration (F⇧ ) and let
⌦0 2 F0 . Then Nt = 1⌦0 Mt is also a square integrable martingale and further [N, N ]t =
7.3. Stochastic Di↵erential Equations 227
T
We will first prove uniqueness of solution in the special case when the driving semi-
martingale is a amenable semimartingale.
F
gale for each j. Let the functional a satisfy conditions (7.3.2)- (7.3.4) and b be defined by
(7.3.5). Let ⇠0 be any F0 measurable random variable. Then if U, Ũ are (F⇧ ) adapted
continuous process satisfying
A
Z t
U t = ⇠0 + b(s, ·, U )dYs , (7.3.14)
0+
Z t
Ũt = ⇠0 + b(s, ·, Ũ )dYs . (7.3.15)
R
0+
then
P(Ut = Ũt 8t 0) = 1. (7.3.16)
D
Proof. For i 1, let ⌧i = inf{t 0 : Kt0 (!) i or Kt0 (!) i} ^ i where Kt0 (!) is
the r.c.l.l. adapted process given by (7.3.9). Thus each ⌧i is a stop time, ⌧i " 1 and for
0 t < ⌧i (!), we have
0 Kt (!) Kt0 (!) i.
Recalling that b(t, ·, ⇣) = a(t , ·, ⇣), we conclude that for ⇣, ⇣1 , ⇣2 2 Cd
sup kb(s, !, ⇣2 ) b(s, !, ⇣1 )k i sup |⇣2 (s) ⇣1 (s)|. (7.3.17)
0s(t^⌧i (!)) 0s(t^⌧i (!))
and
sup kb(s, !, ⇣)k i(1 + sup |⇣(s)|). (7.3.18)
0s(t^⌧i (!)) 0s(t^⌧i (!))
228 Chapter 7. Continuous Semimartingales
Let us fix k, i for now. For j 1, let j = inf{t 0 : |Vt | j}. Since V is a continuous
adapted process with V0 = ⇠0 , it follows that j is a stop time, limj!1 j = 1 and
Thus using the estimate (7.3.12) along with (7.3.18), we get for, i, j, k 1
E[1{|⇠0 |k} sup |Vt |2 ]
T
0t(u^⌧i ^ j)
Z (u^⌧i ^ j)
2
2[E[1{|⇠0 |k} (|⇠0 | + 2m(4 + i) kb(s, ·, Vs )k2 ds)]]
0+
Z (u^⌧i ^
Writing
E[1{|⇠0 |k} (2k 2 + 8m(4 + i)i2
F 0+
j)
(1 + sup
0t(s^⌧i ^ j)
|Vs |2 )ds)].
A
j (u) = E[1{|⇠0 |k} sup |Vt |2 ],
0t(u^⌧i ^ j)
2 3 2
j (u) [2E[1{|⇠0 |k} |⇠0 | ] + 8m(4 + i)i + 8m(4 + i)i j (s)ds]
0+
and further, (7.3.21) yields that j is a bounded function. Thus, using (Gronwall’s) Lemma
3.23, we conclude
D
Now Z t
Ut Ũt = (b(s, ·, U ) b(s, ·, Ũ ))dYs
0+
7.3. Stochastic Di↵erential Equations 229
and hence using the Lipschitz condition (7.3.17) and the growth estimate (7.3.13), we
conclude that for 0 t T ,
Z (t^⌧i )
E[1{|⇠0 |k} sup |Us Ũs |2 ] 2m(4 + i)i2 E[ 1{|⇠0 |k} sup |Us Ũs |2 ]du
0s(t^⌧i ) 0+ 0su
T
P({|⇠0 | k} \ { sup |Us Ũs | ✏}) = 0
0s(t^⌧i )
for all ✏ > 0, i 1 and k 1. Since ⌧i " 1, this proves (7.3.16) completing the proof.
F
We have thus seen that if the Y is a amenable semimartingale then uniqueness of
solution holds for the SDE (7.3.14) and the proof is on the lines of the proof in the case
of Brownian motion. One big di↵erence is that uniqueness is proven without a priori
A
requiring the solution to satisfy a moment condition. This is important as while the
stochastic integral is invariant under time change, moment conditions are not. And this
is what enables us to prove uniqueness when the driving semimartingale may not be a
R
amenable semimartingale.
Using random time change we extend the result on uniqueness of solutions to the SDE
(7.3.14) to the case when the driving semimartingale may not be a amenable semimartin-
gale.
D
T
K s (!) sup |✓! (⇣1 )(v) ✓! (⇣2 )(v)| (7.3.27)
0v s (!)
Bs = A s
= b( s , ·, [U ])
= d(s, ·, U )
T
Since c, d satisfy (7.3.2)-(7.3.5), Theorem 7.18 implies
P(Ut = Ũt 8t 0) = 1
(n)
(7.3.33)
or ka(s , ·, V (n 1)
) a(⌧j , ·, V (n 1)
)k 2 n
}.
232 Chapter 7. Continuous Semimartingales
(n)
Since the process s 7! a(s, ·, V (n 1) ) is an adapted r.c.l.l. process, it follows that each ⌧j
(n) (n) (n) (n)
is a stop time and limj"1 ⌧j = 1. Let V0 = ⇠0 and for j 0, ⌧j < t ⌧j+1 let
(n) (n) (n)
Vt =V (n) + a(⌧j , ·, V (n 1)
)(Xt X⌧ (n) ).
⌧j j
Equivalently,
1
X
(n) (n)
Vt = ⇠0 + a(⌧j , ·, V (n 1)
)(Xt^⌧ (n) Xt^⌧ (n) ). (7.3.34)
j+1 j
j=0
Thus we have defined V (n) and we will show that these process converge almost surely and
the limit process V is the required solution. Let F (n) , Z (n) , R(n) be defined by
(n)
Ft =b(t, ·, V (n 1) ) (7.3.35)
Z t
(n)
Zt =⇠0 + Fs(n) dXs (7.3.36)
T
0+
1
X
(n) (n)
Rt = a(⌧j , ·, V (n 1)
)1 (n) (n)
(t) (7.3.37)
(⌧ ,⌧ ]
j j+1
j=0
Vt
(n)
=⇠0 +
Z t
0+
(n)
F
Rs(n) dXs . (7.3.38)
A
Let us note that by the definition of {⌧j : j 0}, we have
(n) (n) n
|Ft Rt | 2 . (7.3.39)
We will prove convergence of V (n) employing the technique used in the proof of unique-
R
T
(n) (n) n
|Gt St | 2 . (7.3.46)
Recall that we had shown in (7.3.31) that c satisfies Lipschitz condition with coefficient
H = [K]. For j 1, let
⇢j = inf{t
Then ⇢j are (G⇧ )-stop times and
0 : |Ht |
j
j or |Ht | F j or |Ut
" 1. Further, for all ⇣1 , ⇣2 2 Cd we have
(1)
⇠0 | j} ^ j.
A
sup kc(a, !, ⇣2 ) c(a, !, ⇣1 )k j sup |⇣2 (a) ⇣1 (a)|. (7.3.47)
0a(s^⇢j ) 0a(s^⇢j )
Now for k 1 and j 1 (fixed), we observe that for all n 2 (using (7.3.13) along with
(7.3.42) and (7.3.43))
R
Combining (7.3.48) and (7.3.49), we observe that for n 2 (using for positive numbers
x, y, z, (x + y + z)2 3(x2 + y 2 + z 2 ))
E[1{|⇠0 |k} sup |Us(n) Us(n 1) 2
| ] 6m(4 + j)(4 n
+4 (n 1)
)j
0s(t^⇢j )
Z (t^⇢j ) (7.3.50)
+ 6m(4 + j)j 2 E[1{|⇠0 |k} sup |Us(n 1)
Us(n 2) 2
| dr]
0 0s(r^⇢j )
Let
f (n) (t) = E[1{|⇠0 |k} sup |Us(n) Us(n 1) 2
| ] (7.3.51)
0s(t^⇢j )
T
(0) (1)
Since Ut = ⇠0 , by the definition of ⇢j , ft j 2 Cm,j . Now (7.3.51) implies (via
induction on n) that
(Cm,j t)n
f (n) (t) Cm,j . (7.3.52)
F n!
and as a consequence (writing |·|2 for the L2 (P ) norm, and recalling ⇢j j)
1
X
A
k1{|⇠0 |k} sup |Us(n) Us(n 1)
|k2 < 1.
n=1 0s⇢j
and as a consequence
1
R
X
k1{|⇠0 |k} sup |Us(n) Us(n 1)
|k2 < 1. (7.3.53)
n=1 0s⇢j
As in the proof of Theorem 3.26, it now follows that for k 1, j 1, P(Nk,j ) = 0 where
D
1
X
Nk,j = {! : 1{|⇠0 (!)|k} sup ( |Us(n) (!) Us(n 1)
(!)|) < 1}.
0s⇢j (!) n=1
which is same as
lim E[1{|⇠0 |k} sup |Us(n) Us |2 ] = 0. (7.3.54)
n!1 0s⇢j
Rt
Now defining Gt = d(t, ·, U ) and Wt = ⇠0 + 0 Gs dYs , it follows using the Lipschitz condition
(7.3.47) along with (7.3.54) that
T
lim E[1{|⇠0 |k} sup |Ws(n) Ws |2 ] = 0. (7.3.55)
n!1 0s⇢j
0
It now follows that V = [U ] satisfies (7.3.32).
We have shown the existence and uniqueness of solution to the SDE (7.3.32). Indeed,
D
we have explicitly constructed processes V (n) that converge to V . We record this in the
next theorem
Then V (n) em
! V and Vt
(n)
converges to Vt in ucc topology almost surely.
236 Chapter 7. Continuous Semimartingales
Proof. We had constructed in the proof of Theorem 7.20 a (F⇧ ) random time change
and filtration (G⇧ ) = ( F⇧ ) such that U (n) = [V (n) ] converges to U in ucp metric:
ducp (U (n) , U ) ! 0. Now = 1 is also a (G ) random time change, V (n) = [U (n) ] and
⇧
V = [U ]. It follows from Theorem 7.6 that ducp (V (n) , V ) ! 0. Let Ft = b(t, ·, V ). Now
ucp
the Lipschitz condition on a, b- (7.3.4) implies F (n) ! F , where F (n) is defined by (7.3.35)
ucp Rt
and then (7.3.39) implies R(n) ! F . Now Vt = ⇠0 + 0 Fs dXs and (7.3.38) together imply
em
that V (n) ! V . As for almost sure convergence in ucc topology, we had observed that it
holds for U (n) , U and then we can see that same holds for V (n) , V
T
dVt = f (t , H, V )dXt (7.4.1)
where f : [0, 1) ⇥ Dr ⇥ Cd 7! L(d, m), H is an Rr -valued r.c.l.l. adapted process and X
is a continuous semimartingale. Here Dr = D([0, 1), Rd ), Cd = C([0, 1), Rd ). For t < 1,
F
⇣ 2 Cd and ↵ 2 Dr , let ↵t (s) = ↵(t ^ s) and ⇣ t (s) = ⇣(t ^ s). We assume that f satisfies
f (t, ↵, ⇣) = f (t, ↵t , ⇣ t ), 8↵ 2 Dr , ⇣ 2 Cd , 0 t < 1, (7.4.2)
A
t 7! f (t, ↵, ⇣) is an r.c.l.l. function 8↵ 2 Dr , ⇣ 2 Cd (7.4.3)
and that there exists constants for T < 1, CT < 1 such that 8↵ 2 Dr , ⇣1 , ⇣2 2 Cd , 0
tT
R
SDE (7.4.1).
Theorem 7.22. Suppose f satisfies (7.4.2), (7.4.3) and (7.4.4). Then there exists a map-
ping
: Rd ⇥ Dr ⇥ Cd 7! C([0, 1), L(d, m))
such that for an adapted r.c.l.l. process H and a continuous semimartingale X,
V = (⇠0 , H, X)
yields the unique solution to the SDE
Z t
Vt = ⇠ 0 + f (s , H, V )dX. (7.4.5)
0
7.4. Pathwise Formula for solution of SDE 237
inductively for n 0. Let (0) (u, ↵, ⇣)(s) = u for all s 0. Having defined (0) , (1) , . . . ,
let
(n) (n) (n 1) (n 1) (n) n
tj+1 = inf{t tj :k (u, ↵, ⇣)(s) (u, ↵, ⇣)(tj )k 2
(n)
T
(n 1) (n 1) n
or k (u, ↵, ⇣)(s ) (u, ↵, ⇣)(tj )k 2 }
(n 1) (u, ↵, ⇣) (n)
(since is an r.c.l.l. function, tj " 1 as j " 1) and
1
X
This defines
(n)
(u, ↵, ⇣)(s) = u +
(n) (n)
(⇠0 (!), H(!), X(!)) = Vt (!).
(n)
As shown in Theorem 7.21, Vt (!) converges to Vt (!) in ucc topology almost surely
and hence it follows that
This pathwise formula was obtained in [35] and [39]. It was recast in [40] in the form
given in this section.
238 Chapter 7. Continuous Semimartingales
In this section, we will consider matrix valued semimartingles. The notations introduced
here will be used only in this section and in a corresponding section later. Recall that
L(m, k) is the set of all m ⇥ k matrices. Let L0 (d) denote the set of non- singular d ⇥ d
matrices.
Let X = (X pq ) be an L(m, k)-valued process. X is said to be a semimartingale if each
X pq is a semimartingale. Likewise, X will be said to be a local martingale if if each X pq is
a local martingale and we will say that X 2 V if each X pq 2 V.
If f = (f ij ) is an L(d, m)-valued predictable process such that f ij 2 L(X jq (for all
R
i, j, q), then Y = f dX is defined as an L(d, k)-valued semimartingale as follows: Y =
(Y iq ) where
m Z
X
T
Y iq = f ij dX jq .
j=1
Likewise, if g = (g ij ) is an L(k, d)-valued predictable process such that g ij 2 L(X pi (for all
R
Z pj =
k Z
X
F
i, j, p), then Z = (dX)g is defined as follows: Z = (Z pj ) where
g ij dX pi .
A
i=1
For L(d, d)-valued semimartingales X, Y let [X, Y ] = ([X, Y ]ij ) be the L(d, d)-valued
process defined by
R
d
X
[X, Y ]ij
t = [X ik , Y kj ].
k=1
D
The relation (7.5.1) is the matrix analogue of the integration by parts formula (4.6.7).
Recall our terminology : We say that a L(d, d)-valued process h is L0 (d)-valued if
P(ht 2 L(d) 8t 0) = 1.
R R R R
f dX, Z = (dX)g, U = (dX)h and V = h 1 dY . Show that
Z t
[W, Y ]t = f d[X, Y ]. (7.5.2)
0
Z t
[Y, Z]t = (d[Y, X])g. (7.5.3)
0
Z t Z t
gdW = gf dX. (7.5.4)
0 0
Z t Z t
(dZ)f = (dX)gf. (7.5.5)
0 0
Z t Z t
f dZ = (dW )g. (7.5.6)
0 0
[U, V ]t = [X, Y ]t . (7.5.7)
T
R
In view of (7.5.6), with f, g, X, W, Z as in the previous exercise, we will denote f dZ =
R R
(dW )g = f (dX)g.
We can consider an analogue of the SDE (7.3.1)
F
dUt = b(t, ·, U )dYt , t 0, U0 = ⇠0
Essentially the same arguments as given earlier in the section would give analogues of
existence and uniqueness results for the equation (7.5.8).
Exercise 7.25. Formulate and prove analogues of Theorem 7.19, Theorem 7.20 and Theorem
D
Exercise 7.26. Let X be an L(d, d)-valued continuous semimartingale with X(0) = 0 and
let I denote the d ⇥ d identity matrix. Show that the equations
Z t
Yt = I + Ys dXs (7.5.9)
0
and Z t
Zt = I + (dXs )Zs (7.5.10)
0
admit unique solutions.
240 Chapter 7. Continuous Semimartingales
The solutions Y, Z are denoted respectively by e(X) and e0 (X) and are the left and
right exponentials of X.
(iv) W Z = I
The relation (iv) above implies that for any L(d, d)-valued continuous semimartingale X with
X0 = 0, e(X) is L0 (d)- valued and
T
[e(X)] 1
= e0 ( X + [X, X]). (7.5.11)
Hint: For (i), Start with right hand side and use integration by parts formula (7.5.1) and
simplify. For (ii), note that X 1 + X̃ 2 + [X 1 , X̃ 2 ] = X 1 + X 2 and use (i). For (iii) note that
R
if we let X 1 = X̃ 3 then Y 2 (dX 1 )(Y 2 ) 1 = X 3 .
D
e(log(Y ) = Y, e0 (log0 (Y )) = Y.
Likewise, for any L(d, d)-valued continuous semimartingale X with X0 = 0, we have
log(e(X) = X, log0 (e0 (X)) = X.
7.5. Matrix-valued Semimartingales 241
Exercise 7.30. Let Let X be an L(d, d)-valued continuous semimartingales with X0 = 0 and
Y be an L0 (d)-valued continuous semimartingale with Y0 = I . Then show that
T
additive decomposition of X.
F
A
R
D
Chapter 8
We have discussed predictable -field and seen the crucial role played by predictable inte-
T
grands in the theory of stochastic integration. In our treatment of the integration, we have
so far suppressed another role played by predictable processes. In the decomposition of
semimartingales (Theorem 5.48) the process A with finite variation paths turns out to be
In this chapter, we will make this identification and prove the Doob-Meyer decomposi-
A
tion theorem obtaining the predictable quadratic variation hM, M i of a square integrable
martingale. We will also introduce the notion of a predictable stop time.
An important step towards the proof of Doob-Meyer decomposition theorem is: An
R
r.c.l.l. adapted process A with finite variation paths, A0 = 0, E[sup0tT |At | ] < 1 is
predictable if and only if it is natural, i.e. for all bounded r.c.l.l. martingales N , [N, A] is
also a martingale.
D
This result is usually stated assuming that the underlying filtration is right continuous.
We will prove its validity without assuming this. However, some of the auxiliary results do
require right continuity of -fields, which we state explicitly.
Recall that for a stop time ⌧ with respect to a filtration (F· ), the stopped field F⌧ is
defined by
F⌧ = {A 2 ([t Ft ) : A \ {⌧ t} 2 Ft 8t < 1.}
242
8.1. The -field F⌧ 243
We had seen that for every r.c.l.l. adapted process X and a stop time ⌧ , X⌧ is F⌧ measur-
able. We now define the pre-stopped -field F⌧ as follows.
Definition 8.1. Let ⌧ be a stop time with respect to a filtration (F· ). Then
Exercise 8.2. Let ⌧ be the constant stop time ⌧ = t, t > 0. Show that
Ft = ([s<t Fs ).
We note some basic properties of the pre-stopped -field in the next result. Recall the
definition (2.32) of f⌧ , (for a process f ) whereby f⌧ = f⌧ 1{⌧ <1} .
T
Theorem 8.4. Let ⌧ , be stop times with respect to a filtration (F· ). Then
(i) ⌧ is F⌧ measurable.
(ii) F⌧
(iii) If
✓ F⌧ .
F
< ⌧ on ⌧ > 0 ( i.e. ⌧ (!) > 0 implies (!) < ⌧ (!)), then F ✓ F⌧ .
A
(iv) If A 2 F then (A \ { < ⌧ }) 2 F⌧ and in particular, { < ⌧ } 2 F⌧ .
(vi) Let W be a F⌧ measurable random variable. Then there exists a predictable process
f such that f⌧ 1{⌧ <1} = W 1{⌧ <1} .
(vii) Let U be a F⌧ measurable random variable with E[ |U | ] < 1 and E[U | F⌧ ] = 0. Let
D
and A \ { r} 2 Fr . Thus A 2 F⌧ .
244 Chapter 8. Predictable Increasing Processes
where 0 = s0 < s1 < s2 < . . . < sm+1 < 1, m 1, a0 is bounded F0 measurable and for
1 j (m + 1), , aj is a bounded Fsj 1 measurable random variable. For such an f and
↵ < 0,
T
m
[
{f⌧ ↵} = ({a0 ↵} \ {⌧ = 0}) [ ( {aj+1 ↵} \ {sj < ⌧ sj+1 }). (8.1.1)
j=0
Now aj+1 is Fsj measurable implies {aj+1 ↵} \ {sj < ⌧ } 2 F⌧ and since ⌧ is F⌧
B 2 H = F0 [ {A \ {t < ⌧ } : A 2 Ft , t 0}.
D
Thus if G denotes the class of simple functions over H, if follows that the result (vi) is true
if W 2 G. Note that H is closed under intersection and hence G is an algebra. Denoting
by A the class of W such that (vi) is true, it follows that G ✓ A. It is easy to check that A
is bp-closed. Since F⌧ = (H), the result (vi) follows from the Monotone class theorem,
Theorem 2.64.
For (vii), invoking Theorem 2.39 it follows that M is an r.c.l.l. (F⇧ ) adapted stochastic
process. To show that M is a martingale, suffices to show (see Theorem 2.55) that for all
bounded stop times ,
E[M ] = 0. (8.1.2)
8.2. Predictable Stop Times 245
Theorem 8.5. Let be an (F⇧+ )-stop time and ⌧ be a (F⇧ )-stop time. Then
A 2 F + ) A \ { < ⌧ } 2 F⌧ .
F + ✓ F⌧ .
T
A \ { < ⌧ } = { A \ { r} \ {s < ⌧ } : r < s rationals in [0, t]}
e:
[ , ⌧ ] = {(t, !) 2 ⌦ (!) t ⌧ (!)}
and likewise, [ , ⌧ ), ( , ⌧ ) are also defined. The graph [⌧ ] of a stop time is defined by
D
e
[⌧ ] = {(⌧ (!), !) 2 ⌦}.
With this notation, [⌧, ⌧ ] = [⌧ ]. Note that for any , ⌧ , the processes f, g defined by
ft = 1[ ,⌧ ) (t), gt = 1( ,⌧ ] (t) and ht = 1[0,⌧ ] (t) are adapted processes. While f is r.c.l.l., g
and h are l.c.r.l. and thus predictable. As a consequence we get that for stop times , ⌧
with 0 ⌧ , we have
[0, ⌧ ] 2 P, ( , ⌧ ] 2 P. (8.2.1)
On the other hand if ⌧ is a [0, 1)-valued random variable such that ft = 1[0,⌧ ) (t) is adapted,
then ⌧ is a stop time, since in that case {ft = 0} = {⌧ t} 2 Ft .
246 Chapter 8. Predictable Increasing Processes
(iii) n " ⌧ as n ! 1.
T
(iv) [⌧ ] 2 P.
(vii) E[( M )⌧ | F⌧ ] = 0.
F
The stop time ⌧ in the exercise above has some special properties. Such stop times are
A
called predictable.
[⌧ ] 2 P. (8.2.4)
We have noted that for every stop time ⌧ , (⌧, 1) 2 P. Thus ⌧ is predictable if and
only if
D
[⌧, 1) 2 P. (8.2.5)
It follows that maximum as well as minimum of finitely many predictable stop times is
predictable. Indeed, supremum of countably many predictable stop times {⌧k : k 1} is
predictable since
[ sup ⌧k , 1) = \1
k=1 [⌧k , 1).
1k<1
Exercise 8.8. Let be any stop time and a 2 [0, 1) be a constant. Let ⌧ = + a. Show
that ⌧ is predictable.
8.2. Predictable Stop Times 247
e P) and let
Theorem 8.9. Let µ be a finite measure on (⌦,
e : Xt (!) = 0} : X is a bounded continuous adapted process.}
C = {{(t, !) 2 ⌦
Then for all ✏ > 0 and for all 2 P there exist ⇤0 , ⇤1 such that ⇤0 2 C, ⇤c1 2 C,
T
⇤0 ✓ ✓ ⇤1
and
F
µ(⇤1 \ (⇤0 )c ) < ✏.
Proof. Easy to see that C is closed under finite unions and finite intersections: if X 1 , X 2
A
are bounded continuous processes, so are Y = X 1 X 2 and Z = |X 1 | + |X 2 |. Indeed, C is
closed under countable intersections as X j , j 1 bounded continuous (w.l.g. bounded by
P
1) yields that Z = 1 j j
j=1 2 |X | is a bounded continuous adapted process and
R
{Z = 0} = \j {X j = 0}.
Let G be the class of sets in P for which the desired conclusion holds. Now it can be
checked using properties of C noted in the previous paragraph that G is a -field.
D
{X ↵} = {Y = 0} where Y = max(X, ↵) ↵.
Thus {X ↵} 2 C. Since
1 1 c
{X = 0} = \n {|X| < } = \n { |X| }
n n
it follows that C ✓ G. Invoking Proposition 4.1, part (iii) we now conclude that G = P.
The following result gives some insight as to why stop times satisfying (8.2.4) are called
predictable.
248 Chapter 8. Predictable Increasing Processes
Theorem 8.10. Let ⌧ be (F⇧ )-stop time. Then ⌧ is predictable if and only if there exist
(F⇧+ )-stop times ⌧ n such that ⌧ n ⌧ n+1 ⌧ , ⌧ n " ⌧ and ⌧ n < ⌧ on ⌧ > 0.
Proof. Let us take the easy part first. If {⌧ n : n 1} as in the statement exist then noting
that
[⌧, 1) \ {(0, 1) ⇥ ⌦} = \1 n
n=1 (⌧ , 1)
it follows that [⌧, 1) \ {(0, 1) ⇥ ⌦} 2 P(F⇧+ ). Thus using Corollary 4.5 we conclude
[⌧, 1) 2 P(F⇧ ).
Thus ⌧ is predictable.
T
e P)
For the other part, suppose (8.2.4) holds. Consider the finite measure µ on (⌦,
defined by
µ( ) = P({! : (⌧ (!), !) 2 })
Z
F
or equivalently for a positive bounded predictable f
f dµ = E[f⌧ ].
A
For m 1, get bounded continuous adapted processes X m such that
{X m = 0} ✓ [⌧ ] (8.2.6)
R
and
µ([⌧ ] \ ({X m = 0}c )) 2 m
. (8.2.7)
D
Let ⇣ m = inf{t 0 : |Xtm | = 0}. Then (8.2.6) and (8.2.7) together imply that ⇣ m is either
equal to ⌧ or 1 and
P(⌧ 6= ⇣ m ) 2 m
. (8.2.8)
Let = inf{t : |Xtm | 2 n }. Easy to see that m,n ⇣ m and if 0 < ⇣ m < 1 then
m,n
m m ⇤ 1 1
m,n < ⇣ and m,n " ⇣ as n ! 1. For u, v 2 [0, 1], let d (u, v) = | tan (u) tan (v)|.
Then d⇤ is a metric for the usual topology on [0, 1]. Since m,n " ⇣ m as n ! 1, we can
choose n = nm large enough so that denoting m,nm = m we have m ⇣ m and further
m < ⇣ m on 0 < ⇣ m < 1 and
P(d⇤ ( m
, ⇣ m) 2 m
)2 m
(8.2.9)
8.2. Predictable Stop Times 249
T
Remark 8.11. The stopping times {⌧ n : n 1} in the Theorem above are said to announce
the predictable stop time ⌧ . Indeed, this characterization was the definition of predictability of
stop times in most treatments.
F
Here are two observations linking the two filtrations (F⇧ ), (F⇧+ ) and stop times and
martingales.
A
Lemma 8.12. Let ⌧ be a (F⇧ )-stop time. Then
[⌧ ] 2 P(F⇧+ ) =) [⌧ ] 2 P(F⇧ ).
Proof. Let f = 1[⌧ ] . Then f0 = 1{⌧ =0} is F0 measurable as ⌧ is (F⇧ )-stop time. Now the
R
Lemma 8.13. Let M be an r.c.l.l. (F⇧ )-martingale. Then M is also a (F⇧+ )-martingale.
D
We reiterate here that when the underlying filtration is required to be right continuous,
we will state it explicitly. Otherwise martingales, stop times, predictable etc. refer to the
filtration (F⇧ ). Here is a consequence of predictability of stop times.
250 Chapter 8. Predictable Increasing Processes
Theorem 8.14. Let ⌧ be a bounded predictable stop time. Then for all martingales M
with M0 = 0 we have
E[( M )⌧ | F⌧ ] = 0 (8.2.10)
Proof. Let T be a bound for ⌧ and let ⌧ n be a sequence of F⌧+ -stop times announcing ⌧
as in Theorem 8.10 above. If ⌧ > 0, then M⌧ n converges to M⌧ almost surely whereas if
⌧ = 0 then ⌧ n = 0 for all n and M⌧ n = 0 and M⌧ = 0 by definition of M0 = 0. Thus
we conclude that M⌧ n converges to M⌧ almost surely. On the other hand ⌧ n ⌧ and the
martingale property of M gives E[M⌧ | F⌧+n ] = M⌧ n - see Corollary 2.54. Now Theorem
1.36 along with (8.2.21) yields
E[M⌧ | F⌧+ ] = M⌧ . (8.2.11)
T
We now observe that (8.2.10) charaterizes predictable stop times when the filtration is
right continuous.
are equivalent.
E[( M )⌧ ] = 0. (8.2.13)
D
Proof. If [⌧ ] 2 P(F⇧ ) then of course [⌧ ] 2 P(F⇧+ ) and hence (ii) holds as seen in Theorem
8.14. Thus (i) implies (ii).
That (ii) implies (iii) is obvious.
Let us now assume that (8.2.13) holds for all bounded (F⇧+ )-martingales M . We will
show that there exists a sequence of stop times announcing ⌧ . In view of Theorem 8.10,
this will prove (i) completing the proof.
Let N be the r.c.l.l. version of the martingale E[ ⌧ | Ft+ ]. Let Mt = Nt N0 and
Zt = Nt t. Noting that N⌧ = E[ ⌧ | F⌧+ ] = ⌧ by Theorem 2.53, we have Z⌧ = 0. Let
n
n = inf{t 0 : Zt < 2 }.
8.2. Predictable Stop Times 251
We will show that n are (F⇧+ )-stop times and announce ⌧ . Clearly, n n+1 ⌧ for all
n. As (F⇧+ ) is right continuous, n is a (F⇧+ )-stop time by Lemma 2.46.
By definition of n we have Z n 2 n i.e. N n n 2
n . Further, if
n > 0, then
by left continuity of paths of Z , we also have
n
Z n 2 . (8.2.14)
We have seen that E[NT ] = E[⌧ ] and N being a martingale, we have E[N n ] = E[N0 ] =
n , we conclude
E[NT ] and hence E[N n ] = E[⌧ ]. Since N n n 2
n
E[⌧ n] 2 . (8.2.15)
Since n n+1 ⌧ for all n we conclude that
lim n = ⌧. (8.2.16)
n!1
Remains to prove that on {⌧ > 0}, < ⌧ . For this, we will first prove that
T
n
F
E[1A Zt 1{t⌧ } ] = E[1A (⌧ t)1{t⌧ } ] 0.
Since this holds for all A 2 Ft and Zt 1{t⌧ } is Ft measurable, it follows that for each t
A
Zt 1{t⌧ } 0 a.s.
Now right continuity of t 7! Zt shows the validity of (8.2.17). It now follows that Z⌧
0 a.s. and hence
R
N⌧ ⌧ a.s. (8.2.18)
On the other hand, M is a bounded martingale with M0 = 0 and hence in view of the
assumption (8.2.13) on ⌧ , it follows that E[( M )⌧ ] = 0. Noting that
D
( M )⌧ = ( N )⌧ 1{⌧ >0}
we have E[( N )⌧ 1{⌧ >0} ] = 0. Now using N⌧ = ⌧ we conclude
E[N⌧ 1{⌧ >0} ] = E[N⌧ 1{⌧ >0} ] = E[⌧ 1{⌧ >0} ]. (8.2.19)
In view of (8.2.18)-(8.2.19) we conclude
N⌧ 1{⌧ >0} = ⌧ 1{⌧ >0} a.s. (8.2.20)
If ⌧ > 0 and n > 0, then as seen in (8.2.14), Z n 2 n and so N n > n a.s. In
view of (8.2.20), this implies n < ⌧ a.s. on ⌧ > 0. Thus { n } announces ⌧ and as a
consequence ⌧ is predictable.
252 Chapter 8. Predictable Increasing Processes
Theorem 8.16. Let ⌧ be a (F⇧+ )-predictable stop time and let ⌧ n be as in Theorem 8.10
announcing ⌧ . Then
1
[
( F⌧+n ) = F⌧+ . (8.2.21)
n=1
Proof. As seen in Theorem 8.4 F⌧+n ✓ F⌧+ . To see the other inclusion, let B 2 F⌧+ .
If B 2 F0+ , then B 2 F +n for each n 1. If B = A \ {t < ⌧ }, A 2 Ft+ . Then
Bn = A \ {t < ⌧ n } 2 F⌧+n ✓ F⌧+n . Thus
1
[
Bn 2 ( F⌧+m )
m=1
and of course easy to see that B = [1
n=1 Bn completing the proof.
Exercise 8.17. Let ⌧ be a (F⇧ )-predictable stop time and let ⌧ n be as in Theorem 8.10
T
announcing ⌧ . Suppose F0 = F0+ . Show that
1
[
( F⌧+n ) = F⌧ . (8.2.22)
Theorem 8.18. Let ⌧ be a predictable stop time and let ⇠ be a F⌧ measurable random
variable. Then f = ⇠ 1[⌧,1) and g = ⇠ 1[⌧ ] are predictable processes.
Proof. Note that f = g + h where h = ⇠ 1(⌧,1] and that h is predictable being l.c.r.l.
D
We will next show that the jump times of an r.c.l.l. adapted process X are stop times
and if the process is predictable, then the stop times can also be chosen to be predictable.
Lemma 8.19. Let X be an r.c.l.l. (F⇧ ) adapted process with X0 = 0. For ↵ > 0 let
Then ⌧ is a stop time with ⌧ (!) > 0 for all !. Further, ⌧ < 1 implies | X|t ↵.
{t 2 [0, T ] : | X(!)|t ↵}
is a finite set since X has r.c.l.l. paths. Thus ⌧ (!) < 1 implies | X(!)|⌧ (!) ↵. Moreover,
T
It now follows that (writing Qt = {r 2 [0, t] : r is rational} [ {t}) for ! 2 ⌦, ⌧ (!) t if
and only if 8n 1, 9sn , rn 2 Qt , 0 < sn < rn < sn + n1 ,
1
|Xrn (!)
F
Xsn (!)| ↵ n.
Since |Xsnk Xrnk | ↵ n1k and Xsnk , Xrnk are converging, the only possibility is that
Xsnk (!) ! Xu , Xrnk (!) ! Xu and |Xu (!) Xu (!)| ↵. Hence ⌧ (!) t. For the
other part, if ⌧ (!) = s t, using Q is dense in [0, t] and t 2 Qt , we can get sn , rn 2 Qt ,
t
D
{⌧ t} = \1
n=1 ([{s,r2Qt ,0<s<rs+ 1 } {|Xs Xr | ↵ 1
n }).
n
The next result shows that the jumps of an r.c.l.l. process can be covered by a countable
sequence of stop times.
254 Chapter 8. Predictable Increasing Processes
Lemma 8.20. Let A be an r.c.l.l. (F⇧ ) adapted process with A0 = 0. For n 1 and ! 2 ⌦,
let 0n = 0 and for i 1 let i+1
n (!) = 1 if n (!) = 1 and
i
n n 1
i+1 (!) = inf{t > i (!) : |( A)|t (!) n }. (8.2.25)
Then
(i) For all n 1, i 1, n is a stop time and n > 0.
i i
n 1 n n
(ii) 8!, i (!) < 1 implies |( A)| n
i (!)
(!) n and i (!) < i+1 (!).
e = [0, 1) ⇥ ⌦)
(iv) (recall ⌦
e : |( A)t (!)|
{(t, !) 2 ⌦ 1 n e
n} = {( i (!), !) : i 1} \ ⌦. (8.2.26)
T
i
Proof. Fix n 1. The Lemma 8.19 implies that 1n is a stop time. We will prove that
n
i are stop times by induction. Assuming this to be the case for i = j, consider the
e : |( A)t (!)|
{(t, !) 2 ⌦ 1 n e
n} ✓ {( i (!), !) : i 1} \ ⌦.
Part (ii) proven above implies that the equality holds proving (8.2.26). For (v), if A is
predictable, then A = A A is also predictable (since A is l.c.r.l. adapted). Also
D
The previous result shows that the jumps of an r.c.l.l. adapted process can be covered
by countably many stop times. We now show that one can choose finite or countably many
stop times that cover jumps and have mutually disjoint graphs. Note that stop times are
allowed to take 1 as its value and the graph of a stop time is a subset of [0, 1) ⇥ ⌦ and
thus several (or all!) may take value 1 for an ! without violating the requirement that
the graphs are mutually disjoint.
8.2. Predictable Stop Times 255
Theorem 8.21. Let X be an r.c.l.l. (F⇧ ) adapted process with X0 = 0. Then there exists
a sequence of stop times {⌧m : m 1}, such that
[1
{( X) 6= 0} = [⌧m ] (8.2.27)
m=1
and further that for m 6= n, [⌧m ] \ [⌧n ] = ;. As a consequence
X1
( X) = ( X)⌧m 1[⌧m ] . (8.2.28)
m=1
Further, if the process X is also predictable, then the stop times {⌧m : m 1} can be
chosen to be predictable and then ( X)⌧m are F⌧m measurable.
T
Then the sequence of stop times {⇠k : k 1} is just an enumeration of { n :i 1, n 1}
i
and thus in view of (8.2.26) we have
e : |( X)t (!)| > 0} = {(⇠k (!), !) : k
{(t, !) 2 ⌦ e
1} \ ⌦. (8.2.29)
8
<⇠ if ! 2 \m
F
However, the graphs of {⇠k : k 1} may not be disjoint. Define ⌧1 = ⇠1 and define stop
times ⌧k : k 2 inductively as follows. Having defined stop times ⌧k , 1 k m let
A
m+1 (!) j=1 {⇠m+1 (!) 6= ⌧j (!), ⌧j (!) < 1}
⌧m+1 (!) = (8.2.30)
:1 otherwise .
Fix t. Note that
R
and A = \m
j=1 {⇠m+1 6= ⌧j } 2 F⇠m+1 by Theorem 2.52. Thus,
{⌧m+1 t} = {⇠m+1 t} \ A 2 Ft .
D
and hence each ⌧m+1 is also a stop time. Thus {⌧k : k 1} is a sequence of stop times.
In view of (8.2.29) and the definition (8.2.30) of ⌧m+1 , we can check that the sequence
{⌧m : m 1} satisfies the required conditions, (8.2.27) and (8.2.28).
When the process X is predictable, we have seen that the stop times { in : n 1, i 1}
are predictable and thus here {⇠k : k 1} are predictable. Since
[⌧m+1 ] = [⇠m+1 ] \ ([m
j=1 [⌧j ])
c
it follows inductively that {⌧m } are also predictable. Predictability of X implies that
X is also predictable and then part (v) Theorem 8.4 now shows that ( X)⌧m are F⌧m
measurable
256 Chapter 8. Predictable Increasing Processes
Remark 8.22. It is possible that in the construction given above P(⌧m = 1) = 1 for some
m.
Here is an observation.
Corollary 8.23. Let A be a (F⇧ )-predictable r.c.l.l. process with finite variation paths.
Let |A| = Var(A) denote the total variation of A. Then |A| is (F⇧ )-predictable.
Lemma 8.24. Let H be an r.c.l.l. adapted process. Then H is predictable if and only if
( H) admits a representation
T
1
X
( H) = ( H)⌧m 1[⌧m ] (8.2.31)
m=1
where {⌧m : m
able. F
1} is a sequence of predictable stop times and ( H)⌧m is F⌧m
Proof. One part is proven in Theorem 8.21.For the other part, if H admits such a repre-
measur-
A
sentation then by Theorem 8.18, ( H) is predictable. Since H = H + ( H) and H is
predictable being left continuous, it follows that H is predictable.
The following result would show that an r.c.l.l. predictable process is locally bounded-
R
in a sense proving that since we can predict the jumps, we can stop just before a big jump.
Lemma 8.25. Let A be an r.c.l.l. predictable process with A0 = 0. Then for every n, the
D
stop time
⌧n = inf{t > 0 : |At | n or |At | n}
is predictable. As a consequence, A is locally bounded as a (F⇧+ )-adapted process.
Proof. We have shown in Lemma 2.40 that ⌧n is a stop time, ⌧n > 0 and if ⌧n < 1 then
|A⌧n | n or |A⌧n | n. Further, it follows that limn ⌧n = 1. Let
n
e : |At |
= {(t, !) 2 ⌦ n or |At | n}.
Since A is predictable and so is A (being l.c.r.l.), it follows that n 2 P and hence
[⌧n ] = [0, ⌧n ] \ n 2 P.
8.2. Predictable Stop Times 257
This shows ⌧n is predictable. For the second part, since ⌧n > 0, there exist (F⇧+ )-stop times
⌧n,m that increase to ⌧n strictly from below. Thus we can get mn such that ⌧n⇤ = ⌧n,mn
satisfies
P(⌧n⇤ < ⌧n ) = 1, P(⌧n⇤ ⌧n 2 n
)2 n
. (8.2.32)
Let n = max{⌧1⇤ , ⌧2⇤ , . . . , ⌧n⇤ }. Then { n n 1} are (F⇧+ )-stop times and
n n
P( n < ⌧n ) = 1, P( n n+1 ) = 1, P( n ⌧n 2 )2 . (8.2.33)
Since ⌧n " 1, it follows that n " 1. And < ⌧n implies that |At | n for t n. Thus
A[ n ] is bounded (by n) and so A is locally bounded as a (F⇧+ )-adapted process.
T
Theorem 8.26. Let M be an r.c.l.l. martingale. If M is also predictable, then M has
continuous paths almost surely.
Proof. Let ⌧ be a bounded predictable stop time. Since M is also a (F⇧+ )-martingale as
seen in Lemma 8.13, by Theorem 8.15, we have F
E[( M )⌧ | F⌧+ ] = 0.
A
On the other hand, as seen in Theorem 8.4, part (v), M⌧ is F⌧+ measurable and thus so
is ( M )⌧ and so ( M )⌧ = 0 a.s. and hence we get M⌧ = M⌧ (a.s.). Now if is any
predictable stop time, ^ k is also predictable and hence we conclude
R
( M) ^k =0 a.s. 8k 1
( M) = 0 a.s.
By Theorem 8.21, the jumps of M can be covered by countably many predictable stop
times. This shows
P(( M )t = 0 8t) = 1
Corollary 8.27. Let M be an r.c.l.l. (F⇧ ) local martingale. If M is also (F⇧ )-predictable,
then M has continuous paths almost surely.
258 Chapter 8. Predictable Increasing Processes
Theorem 8.28. Let A be an r.c.l.l. (F⇧ )-predictable process with finite variation paths with
A0 = 0. If A is also a (F⇧ )-martingale, then
P(At = 0 8t) = 1. (8.2.34)
Proof. Theorem 8.26 implies that A is continuous. Now Theorem 4.67 implies
P([A, A]t = 0 8t) = 1.
Now part (v) of Theorem 5.16 and its Corollary 5.17 together imply that (8.2.34) is true.
Corollary 8.29. Let A be an r.c.l.l. (F⇧ )-predictable process with finite variation paths
with A0 = 0. If A is also a (F⇧ ) local martingale, then
T
P(At = 0 8t) = 1.
Definition 8.30. Let A 2 V0 i.e. A isan adapted process with finite variation paths and
A0 = 0. Suppose |A| is locally integrable where |A|t = Var[0,t] (A). A is said to be Natural if
for all bounded r.c.l.l. martingales M
[M, A] is a local martingale. (8.3.1)
Remark 8.31. Let A 2 V0 be such that [A, A] is locally integrable. Since [A, A] = ( A)2 ,
it follows that |( A)| is locally integrable and as a consequence A is locally integrable and thus
A 2 W.
8.3. Natural FV Processes are Predictable 259
Theorem 8.32. Let A 2 W be natural. Let n " 1 be such that |A|[ n ] is integrable. Then
[M, A[ n ] ] is a martingale for all bounded martingales M and n 1.
[ ]
Proof. Let An = A[ n ] . As seen in (4.6.9), [M, A]t n = [M, An ]t and since A is natural,
[M, An ]t is also a local martingale for all bounded martingales M . Invoking Theorem 4.67
we have
X
[M, An ]t = ( M ) s ( An ) s . (8.3.2)
0<st
T
E[sup |[M, An ]s |] 2CE[|A|t^ n ] < 1.
st
Now Lemma 5.5 implies that the local martingale [M, An ] is indeed a martingale.
all n. Fix n.
Predictability of An implies that ( An ) is predictable since (An ) is predictable being
l.c.r.l. adapted. Let {⌧m } be predictable stop times covering jumps of An as constructed
D
in Theorem 8.20. Now predictability of ( An ) and part (v) of Theorem 8.4 implies that
( An )⌧m is F⌧m measurable for all m 1. Since {⌧m } cover the jumps of An , it follows
from (8.3.2) that
X1
n
[M, A ]t = ( An )⌧m ( M )⌧m 1[⌧m ,1) (t). (8.3.4)
m=0
2C|An |t
< 1.
The dominated convergence theorem implies that the series in (8.3.4) converges in L1 (P)
and as a consequence, E[[M, An ]t ] = 0 for all t < 1.
Now given a bounded martingale N and bounded stop time say bounded by T , apply
this to N = M to get
which in turn proves that [N, An ]t is a martingale (see Theorem 2.53). Thus [N, A] is a
T
local martingale for every bounded martingale N and so A is natural.
Our next observation will play an important role in the converse that we will take up
later.
F
Lemma 8.34. Let A 2 W be natural. Then for all stop times ⌧ , ( A)⌧ is F⌧ measurable.
A
Proof. To see this, first let ⌧ be bounded, say by k and let U be any bounded F⌧ measurable
random variable and let
Wt = (U E[U | F⌧ )1[⌧,1) (t).
R
As seen in Theorem 8.4, part (vi), W is a martingale with r.c.l.l. paths. Since W is bounded
and A is natural
[W, A]t = ( A)⌧ (U E[U | F⌧ ])1[⌧,1) (t)
D
is a local martingale.
Let n " 1 be stop times such that |A| n is integrable for all n. As seen in Theorem
8.32, [W, A[ n ] is a martingale and since ⌧ k we have E[[W, A[ n ] ]k ] = 0. Thus
E[( A[ n]
)⌧ U ] = E[( A[ n]
)⌧ E[U | F⌧ ]]
for all bounded F⌧ measurable random variables U . Thus by Lemma 5.42, it follows that
( A[ n ] )⌧ is F⌧ measurable. Since n " 1, we conclude that A⌧ is F⌧ measurable.
For a general stop time ⌧ , noting that
it follows that
( A)⌧ is F⌧ measurable. (8.3.5)
Next we will show that if A is natural for (F⇧ ) then it is so for (F⇧+ ) filtration as well.
Theorem 8.35. Let A be an (F⇧ ) adapted r.c.l.l. process with finite variation paths such
that A0 = 0 and |A| is locally integrable where |A|t = Var[0,t] (A). Suppose that A is natural
for the filtration (F⇧ ). Then it is also natural for the filtration (F⇧+ ).
Proof. Let n " 1 be such that |A|[ n ] is integrable and let An = A[ n ] . Let us fix a
(F⇧+ )-martingale N bounded by C with r.c.l.l. paths. To show [N, A] is a local martingale,
we will prove that for all n
T
[N, An ] is a (F⇧+ )-martingale. (8.3.6)
We divide the proof in steps.
Step 1:. We will first prove that for r 0 fixed,
U has a single jump at t = r and thus U is also a (F⇧+ )-martingale and (8.3.8) is true.
Invoking Lemma 8.34 we observe that
( An )r is Fr ✓ Fr+ measurable. (8.3.9)
D
As a consequence
E[( [U, An ])r | Fr+ ] = E[( N )r ( An )r | Fr+ ]
= ( An )r E[( N )r | Fr+ ] (8.3.10)
=0
and thus [U, An ] is a (F⇧+ )-martingale. This completes Step 1.
Step 2: Let D = {t 2 [0, 1) : P(( N )t 6= 0) > 0}. Then D is countable.
To see this, for t 0 let h(t) = E[ [N, N ]t ]. Then h is an [0, 1)-valued increasing
function since N is a bounded martingale. Since (h(t) h(t ) = E[ [N, N ]t [N, N ]t ] =
262 Chapter 8. Predictable Increasing Processes
T
{t : P[( M m )t 6= 0] > 0} = {tk : k (n + 1)}. (8.3.12)
X
E[ (( N )tk )2 1[tk ,1) (t)] E[[N, N ]t ] < 1.
k=1
and
em
Z m ! Z. (8.3.14)
Let V n be defined by
1
X
Vtn = ( N )tk ( An )tk 1[tk ,1) (t). (8.3.16)
k=1
The series defining V n converges almost surely and in L1 (P) in view of (8.3.15). Then
1
X
E[ |Vtn m n
[Z , A ]t | ] 1[tk ,1) (t)E[ |( N )tk ( An )tk | ]
k=m+1 (8.3.17)
! 0 as n ! 1
in view of (8.3.15). Since [Z m , An ] is a (F⇧+ )-martingale, (8.3.17) implies that V n is a
(F⇧+ )-martingale. On the other hand (8.3.14) implies that
[Z m , An ] ! [Z, An ] as m ! 1
in the Emery topology (see Theorem 4.103). Thus V n = [Z, An ] is a (F⇧+ )-martingale.
T
This completes Step 3.
em
Since M m = N Z m , we get M m ! M where M = N Z is also a (F⇧+ )-martingale.
ucp
Also, M m ! M . It then follows from (8.3.12) that
F
P[( M )t 6= 0] = 0 8t 0.
This observation (8.3.18) and the assumption that F0 contains all P null sets together
(8.3.18)
A
imply that Mt is Ft+ measurable. Since Ft+ ✓ Ft ✓ Ft+ , we conclude that M is a
(F⇧ )-martingale. In view of the assumption on M , we conclude that [M, A] is a (F⇧ )-local
martingale. The process [M, A] is r.c.l.l. and thus is also a (F⇧+ )-local martingale.
R
Since N = M +Z, [N, A] = [M, A]+[Z, A]. We have already shown in Step 3 that [Z, A]
is a (F⇧+ )-local martingale and thus it follows that [N, A] is a (F⇧+ )-local martingale.
Theorem 8.36. Let B 2 W be a natural (for the filtration (F⇧ ) ) i.e. for all bounded
(F⇧ )-martingales M
[M, B] is a local martingale. (8.3.19)
Then B is (F⇧ )-predictable.
Proof. Let ⌧m " 1 be stop times such that B m = B [⌧m ] satisfies |B m |t is integrable for all
m, t. Suffices to prove that for all m, B m is predictable. So fix m 1 and let us write
A = B m . As seen in Theorem 8.32, for all bounded martingales M ,
[M, A] is a martingale. (8.3.20)
264 Chapter 8. Predictable Increasing Processes
Let
1
X
gn = f n,i 1( n n
i 1, i ]
.
i=1
T
Then g n is predictable since f n,i is predictable and 1( n n
i 1, i ]
is l.c.r.l. adapted process and
hence predictable. Further, note that by definition
g nin = ( A) in .
In view of the definition of { n
i}
n
g 1{| A| 1
n
}
F
as seen in (8.2.26), it follows that
= ( A)1{| A| 1
n
} . (8.3.23)
A
In particular, for all m n
g n 1{| A| n 1
m
} = g 1{| A| 1
}
n
A| }
m
R
Now given a bounded (F⇧+ )-martingale M with M0 = 0, let = Since hm Nm hm dM .
is bounded predictable and M is bounded (F⇧+ )-martingale, N m is also a (F⇧+ )-martingale
and thus [N m , A] is a (F⇧+ )-martingale. On the other hand
Z t X
m
[N , A]t = hm d[M, A] = hm
s ( M )s ( A)s
0 0st
X
= ( M )s 1{|( A)s | 1
}
m
0st
X1
= ( M) m
j ^t
.
j=0
8.4. Decomposition of Semimartingales revisited 265
T
n
i ]
and since At is (F⇧+ )-predictable, being an l.c.r.l. adapted process, we conclude that A
is (F⇧+ )-predictable. Since B 2 W, it follows that B0 = A0 = 0 and then Corollary 4.4
implies that A is (F⇧ )-predictable.
D
T
with finite variation paths which is also a martingale and hence by Theorem 8.28, M = N
and B = A.
Remark 8.38. We note that in Theorem 8.37 we have not assumed that the underlying
filtration is right continuous. F
We need two more results before we can deduce the Doob-Meyer decomposition result.
A
First, we need to extend Theorem 8.37 to all locally integrable semimartingales. Then
we need to show that when X is a submartingale, then the FV process appearing in the
decomposition is an increasing process. We begin with the second result.
R
Theorem 8.39. Let U be an r.c.l.l. (F⇧ )-predictable process with finite variation paths with
U0 = 0 and
E[Var[0,T ] (U )] < 1 for all T < 1. (8.4.2)
D
Proof. Let V = |U | denote the total variation process of U (Vt = Var[0,t] (U )). As seen in
Corollary 8.23, V is also predictable and of course, V is an increasing process. Fix T < 1
and define measures µ and on (⌦, e P) as follows. For a bounded predictable process f
Z Z T
f dµ = E[ fs dUs ] (8.4.4)
0
Z Z T
f d = E[ fs dVs ]. (8.4.5)
0
8.4. Decomposition of Semimartingales revisited 267
T
f 0 and thus µ is a positive measure. If f is a non-negative bounded predictable process,
then Z Z
T T
| fs dUs | fs dVs
Z
f dµ
FZ
0
fd .
A
Thus µ is absolutely continuous w.r.t. and thus denoting the Radon-Nikodym derivative
by ⇠, it follows that ⇠ is a [0, 1)-valued predictable process such that
Z Z
R
f dµ = f ⇠ d . (8.4.6)
We will show that C is a martingale. For this, using (8.4.6) and (8.4.7), we have for
any bounded predictable f
Z T Z T
E[ fs dBs ] =E[ fs ⇠s dVs ]
0 Z 0
= f⇠d (8.4.9)
Z
= f dµ.
Using (8.4.4) and (8.4.9) and recalling that C = B U we conclude that for bounded
predictable processes f we have
Z T
E[ fs dCs ] = 0.
0
Taking f = a1(s1 ,s2 ] with a being Fs1 measurable, s1 < s2 T , we conclude
T
E[a((Cs2 Cs1 )] = 0.
This implies C is a martingale (since this is true for all T < 1).
By Theorem 8.28, it follows that C = 0 and as a consequence, U = B. Since by
F
construction, B is an increasing process, this completes the proof.
Z t
Ut = ⇠s dVs . (8.4.10)
0
Here are some auxiliary results.
D
Lemma 8.42. Let ⌧ be a stop time and let ⇣ be a F⌧ measurable [0, 1)-valued integrable
random variable. Let
Xt = ⇣ 1[⌧,1) (t).
Then X admits a decomposition
X = M + A, M 2 M, A 2 V+ ; A is predictable. (8.4.11)
The decomposition (8.4.11) is unique. Further, for all T < 1,
E[AT ] E[⇣] (8.4.12)
E[ |MT | ] 2E[⇣]. (8.4.13)
Proof. Let ⇣ m = ⇣ 1{⇣m} and X m = ⇣ m 1[⌧,1) (t). Then, X m is bounded FV processes and
hence a stochastic integrator. Thus by Theorem 8.37, it admits a decomposition
T
X m = M m + Am , M m 2 M 2 , A m m
0 = 0 and A is predictable.
P(Am
t Ant 8t) = 1.
Thus defining
At (!) = lim sup Ant (!)
D
n!1
it follows that A is an increasing process. Since An is predictable, so is A and of course,
A0 = 0. Let M = X A. We will now show that M is a martingale.
Note that X0m X0n = 0 implies that M0m M0n = 0 and hence E[Mtm Mtn ] = 0. As
a consequence for n m
E[Am
t Ant ] = E[Xtm Xtn ] = E[⇣ 1[⌧,1) (t)1{n<⇣m} ].
Since Amt Ant is non negative (being an increasing process that is zero at t = 0), it follows
that for each t,
E[ |Am
t Ant | ] E[⇣ 1{n<⇣m} ].
270 Chapter 8. Predictable Increasing Processes
T
E[ |MT | ] E[AT ] + E[XT ] 2E[⇣].
F
Corollary 8.43. Let ⌧n be a sequence of stop times and let ⇣n be a sequence of [0, 1)-
valued random variables, with ⇣n being F⌧n measurable and
X1
A
E[⇣n ] < 1. (8.4.14)
n=1
Let
1
X
Zt = ⇣n 1[⌧n ,1) . (8.4.15)
R
n=1
Then there exists a unique predictable increasing process B with B0 = 0 and a martingale
N such that
D
Z = N + B.
Proof. Let Xtn = ⇣n 1[⌧n ,1) and let An be the predictable increasing process with An0 = 0
given by Lemma 8.42 such that Mtn = Xtn Ant is a martingale. Then as seen in Lemma
8.42 we have
E[AnT ] E[⇣n ]
P
and hence the assumption (8.4.14) implies that Bt = 1 n
n=1 At defines a integrable pre-
Pk
dictable process. Further n=1 Mtn converges in L1 (P) to Nt = Zt Bt . By Theorem 2.21
N is a martingale. This proves existence part. The uniqueness again follows from Theorem
8.28.
8.4. Decomposition of Semimartingales revisited 271
T
Let ↵n be the stop times defined by
↵n = inf{t 0 : |Xt | n or |Xt | n} (8.4.18)
and let ⌧n = ↵n ^ n ^ n. Let X n = X [⌧n ] , ⇠ n = ( X)⌧n , U n = (⇠ n )+ 1[⌧n ,1) , V n =
(⇠ n ) 1[⌧n ,1) and Z n = X n U n + V n . Then
8
<X
t
F
if t < ⌧n
A
Ztn =
:X if t ⌧n .
⌧n
|( X)| a
T
decomposition with M 2 Mloc ,A 2 V, A0 = 0 and A predictable. Show that
|( A)| a.
stop time ⌧ ,
F
Hint: If ↵n are defined by (8.4.18), then X [↵n ] are bounded. Observe that for any predictable
[M, N ]t At is a local martingale (in other words the process A appearing in the decom-
position (8.4.16)) is denoted by hM, N i.
Definition 8.47. For local martingales M, N such that [M, N ] is locally integrable, the
D
predictable cross quadratic variation hM, N i is the unique predictable process with FV paths
which is zero at t = 0 and such that
[M, N ]t hM, N it
It is easy to see that (M, N ) 7! hM, N i is bilinear. And when M, N are locally square
integrable, [M, N ] is locally integrable and hence the predictable quadratic variation hM, N i
is defined.
8.5. Doob-Meyer Decomposition 273
Lemma 8.48. Let A be an adapted increasing integrable process with A0 = 0 and let U be
a predictable process, U 2 V such that M = A U is a martingale. Then U 2 V+ , i.e. U
T
is an increasing process.
Proof. Let {⌧m : m 2 F } be the sequence of stop times given by Theorem 8.21 (F is a
subset of natural numbers) so that (8.2.27) and (8.2.28) are true. Let
Ct =
X
m2F
F
( A)⌧m 1[⌧m ,1)
A
and
D t = At Ct .
continuous. Since
0 Ct At 8t
it follows that C is also integrable and thus by Corollary 8.43 we can get a predictable
D
U = D + B.
Corollary 8.49. Let M be a locally square integrable martingale. Then the predictable
quadratic variation hM, M i (defined in Definition 8.47 ) is an increasing process.
274 Chapter 8. Predictable Increasing Processes
Proof. The result follows from Lemma 8.48 since [M, M ] is an increasing process.
Proof. Since Nt2 [N, N ]t is a local martingale (see Theorem 5.16), for a predictable process
A, N 2 A is a local martingale if and only if [N, N ] A is a local martingale. Now the
T
first part follows from Corollary 8.49.
For the remaining part, let ⌧n be a localizing sequence for the local martingale M =
N 2 hN, N i. Then for n 1
Ytn = N⌧2n ^t
is a martingale. Hence for any bounded stop time
F hN, N i⌧n ^t
we have,
A
E[N⌧2n ^ ] = E[hN, N i⌧n ^ ] (8.5.3)
Now Doob’s maximal inequality, Theorem 2.24, we get
R
0s
Remark 8.51. Using (8.5.3), it follows that if N is a locally square integrable martingale
then for all stop times ,
E[ [N, N ] ] = E[N 2 ] = E[hN, N i ]. (8.5.5)
All the quantities can be 1, but if one is finite, so are the others and they are equal.
The process hN, N i is now known as the predictable quadratic variation of the (locally)
square integrable martingale N .
8.5. Doob-Meyer Decomposition 275
Lemma 8.52. Let M be a locally square integrable martingale and let f be a predictable
process such that Z t
Bt = |fs |2 dhM, M is < 1 a.s. (8.5.6)
0
Then B is a predictable increasing process.
Proof. Let us note that B is an r.c.l.l. increasing adapted process and for any stop time ⌧ ,
( B)⌧ = |f⌧ |2 ( hM, M i)⌧ .
Now the result follows from Lemma 8.24 and part (v) of Theorem 8.4.
Lemma 8.53. Let M be a locally square integrable martingale and let f be a locally bounded
R
predictable process and let N = f dM . Then
Z t
hN, N it = |fs |2 dhM, M is < 1. (8.5.7)
T
0
Proof. That Bt defined by (8.5.6) is predictable has been noted above. One can show that
Nt2 Bt is a local martingale staring with f simple and then by approximation. Thus
hN, N it = Bt .
F
Remark 8.54. For f, M, N as above, we have seen that
Z t
A
[N, N ]t = |fs |2 d[M, M ]s < 1 (8.5.8)
0
and hence it follows that for a stop time ,
Z Z
2
E[ |fs | dhM, M is ] = E[ |fs |2 d[M, M ]s ]. (8.5.9)
R
0 0
The estimate (5.4.10) on the growth of the stochastic integral can be recast as :
Theorem 8.55. Let M be a locally square integrable martingale with M0 = 0. For a locally
D
Rt Rt 2
bounded predictable process f , the processes Yt = 0 f dM and Zt = Yt2 0 fs dhM, M is
R 2
are local martingales and for any stop time such that E[ 0 fs dhM, M is ] < 1,
Z t Z
2
E[ sup | f dM | ] 4E[ fs2 dhM, M is ], 8T < 1. (8.5.10)
0t 0 0
is a local martingale. We can see that the predictable cross quadratic variation also satisfies
the polarization identity (for locally square integrable martingales M, N )
1
hM, N it = (hM + N, M + N it hM N, M N it ) (8.5.12)
4
We had seen that (M, N ) 7! hM, N i is bilinear in M and N . This yields an analogue of
Theorem 4.69 which we note here.
Theorem 8.56. Let M, N be locally square integrable martingales. Then for any s t
p
Var(s,t] (hM, N i) (hM, M it hM, M is ).(hN, N it hN, N is ) (8.5.13)
and
p
Var[0,t] (hM, N i) hM, M it hN, N it , (8.5.14)
Proof. Let
T
⌦a,b,s,r = {! 2 ⌦ : haM + bN, aM + bN ir (!) haM + bN, aM + bN is (!)}
and
⌦0 = [{⌦a,b,s,r : s, r, a, b 2 Q, r s}.
F
Then it follows that P(⌦0 ) = 1 (since for any locally square integrable martingale Z, hZ, Zi
is an increasing process) and that for ! 2 ⌦0 , for 0 s r, s, r, a, b 2 Q
A
(a2 (hM, M ir hM, M is ) + b2 (hN, N ir hN, N is )
+ 2ab(hM, N ir hM, N is ))(!) 0,
and since the quadratic form above remains positive, we conclude
R
s r, s, r 2 [0, 1). Now given s < t and s = t0 < t1 < . . . < tm = t, we have
m
X1
|hM, N itj+1 hM, N itj |
j=0
X1 q
m
(8.5.16)
(hM, M itj+1 hM, M itj )(hN, N itj+1 hN, N itj )
j=0
p
(hM, M it hM, M is )(hN, N it hN, N is )
where the last step follows from Cauchy-Schwarz inequality since hM, M i, hN, N i are in-
creasing processes. Now taking supremum over partitions of [0, t] in (8.5.16) we get (8.5.13).
8.5. Doob-Meyer Decomposition 277
Now (8.5.14) follows from (8.5.13) taking s = 0 since hM, M i0 = 0, hN, N i0 = 0 and
hM, N i0 = 0.
And here is an anlaogue of 4.69. However, the proof is di↵erent from that given earlier.
Lemma 8.57. Let U, V be locally square integrable martingales. Then for any t we have
p
|hU, U it hV, V it | 2hU V, U V it (hU, U it + hV, V it ). (8.5.17)
Using Lemma 8.24, one can show that if A is a predictable r.c.l.l. FV process and f is
a [0, 1)-valued predictable process, then B defined by
T
Z t
Bt = fs dAs
0
is predictable. It can be checked (first for simple integrands and then by limiting arguments)
X = f dM and Y = gdN ,
R
Z t
F
that for f, g predictable and locally bounded, M, N locally square integrable martingales,
R
A
U t = Xt Yt f gdhM, N i
0
is a local martingale. Thus
Z Z Z
R
These observations lead us to the following analogue of Theorem 5.26 which we record here
for use later.
D
Theorem 8.58. Let M 2 M2loc ( i.e. M be a locally square integrable martingale) with
M0 = 0. Then for a locally bounded predictable process f and for any stop time such
R
that E[ 0 fs2 dhM, M is ] < 1,
Z t Z
E[ sup | f dM |2 ] 4E[ fs2 dhM, M is ], 8T < 1. (8.5.20)
0t 0 0
R
Proof. We had observed that ZY = f dM is a local martingale above and that
Z t
hY, Y it = fs2 dhM, M is .
0
Now the estimate (8.5.20) follows from (8.5.2).
278 Chapter 8. Predictable Increasing Processes
Definition 8.59. Two locally square integrable martingales M, N are said to be strongly
orthogonal if hM, N it = 0 for all t < 1.
Proof. Let ⌧n be a sequence of stop times, ⌧n n, such that Nt^⌧n is a square integrable
T
2
martingale. Let an = E[Nt^⌧ ]. Let
n
P R ⌧n
↵ = inf{ 1 1
n=1 2n (1+an ) E[(N⌧n 0 gdM ) ] : g 2 Lm (M )}.
2 2
F
It can be seen that ↵ < 1, indeed, ↵ 1 since g = 0 2 L2m (M ). Now it can be shown
(proceeding as in the proof of existence of orthogonal projections onto a closed subspace
in a Hilbert space and in the proof of Theorem 5.37) that the infimum is attained, say for
A
f 2 L2m (M ) and then for every n
Z ⌧n Z ⌧n
E[(N⌧n f dM )( gdM )] = 0 8g 2 L2m (M ).
0 0
R
E[U⌧n N ^⌧n ] =0
Lemma 8.61. Let N 2 V \ M2loc and M 2 M2loc be such that for all stop times ⌧ ,
( N )⌧ ( M )⌧ = 0 a.s.
Then hM, N i = 0.
Using Theorem 8.21, we can get stop times {⌧n : n 2 F } (F is a finite or countable set)
that cover the jumps of N and then it follows that
X
[M, N ]t = ( M )⌧m ( N )⌧m 1{⌧m t} .
n2F
Now in view of the assumption, ( M )⌧m ( N )⌧m = 0 (a.s.) for each m and as a con-
T
sequence, [M, N ]t = 0 (a.s.) for all t. Thus Zt = Mt Nt is a local martingale and as a
consequence hM, N i = 0.
Theorem 8.62. Let ⌧ be a predictable stop time and let ⇠ be a F⌧ measurable square
R
Proof. From part (vii) in Theorem 8.4 it follows that M is a martingale. Since ⌧ is
predictable, by Theorem 8.18 it follows that A is predictable and clearly A is an increasing
process. Noting that
Mt2 At = (⇠ 2 E[⇠ 2 | F⌧ ])1[⌧,1) (t)
it follows, again invoking part (vii) in Theorem 8.4, that Nt = Mt2 At is a martingale.
It is clear from the definition of A that A0 = 0 on the set ⌧ > 0. On the other hand,
⇠ is F⌧ measurable and hence ⇠ 1{⌧ =0} is F0 measurable. Thus E[⇠ | F⌧ ] = 0 implies
280 Chapter 8. Predictable Increasing Processes
follows from the uniqueness part of the Doob-Meyer decomposition (Theorem 8.50).
We are going to prove a structural decomposition result for square integrable martin-
gales. This would play an important role in the proof of Metivier Pellaumail inequality in
the next chapter. Here are some preparatory results.
Lemma 8.63. Let X be a square integrable martingale with Xt = Xt^T for all t, for some
T < 1 and let ⌧ be a predictable stop time such that
T
Let N be the martingale defined by
and let Y = X F
N . Then P(( Y )⌧ 6= 0) = 0, Y is a square integrable martingale,
hY, N i = 0
A
and
hN, N it = E[( X)2⌧ | F⌧ ]1[⌧,1) (t). (8.6.4)
R
|( X)⌧ | 2 sup|Xt |
tT
D
and hence in view of the Doob’s maximal inequality, Theorem 2.24, we have that ⇠ = ( X)⌧
is square integrable. Theorem 8.62 implies that N is a square integrable martingale and
that (8.6.4) is true. It follows that Y is a square integrable martingale and by construction,
P(( Y )⌧ 6= 0) = 0. Hence
Corollary 8.64. Let X be a locally square integrable martingale and let ⌧ be a predictable
stop time such that
P(( X)⌧ 6= 0) > 0.
Let N, Y be defined by
Nt = ( X)⌧ 1[⌧,1) (t)
and Y = X N . Then P(( Y )⌧ 6= 0) = 0, N, Y are square integrable martingales with
hY, N i = 0.
Further, hN, N i has a single jump at ⌧ and if m " 1 are bounded stop times such that
X [ m ] 2 M2 then
hN, N i⌧ 1{⌧ m } = E[( X [ m ] )2⌧ | F⌧ ]. (8.6.5)
T
Theorem 8.65. Let M 2 M2loc be a locally square integrable martingale with M0 = 0. Let
{⌧k : k 1} be a sequence of stop times with disjoint graphs. Let
E[ [M, M ] i ]
T
< 1.
Thus S n 2 M2loc . Since S n 2 V by construction, it follows that S n 2 M2d,loc . Similarly, for
for m n we have
F
E[ [S n S m , S n S m ] i ] E[
Xn
[V j , Vj ] i ].
j=m+1
(8.6.12)
A
Now (8.6.11) and (8.6.11) imply that S n is Cauchy in dem and thus converges to say W
such that S 2 M2loc (see Lemma 5.35) and further (8.6.7) holds. Since (S n )[ i ] 2 M2d , the
relation (8.6.7) implies that W [ i ] 2 M2d and hence W 2 M2d,loc .
R
j=m+1
! 0 as n ! 1.
Thus [S n , S n ]t ! [S, S]t and hS n , S n it ! hS, Sit for all t 0.
Lemma 8.66. Let M 2 M2 be such that Mt = Mt^T for t 0 for some T and let ⌧ be a
predictable stop time. Then
and
E[( M )2⌧ ] = E[( M 2 )⌧ ]. (8.6.15)
8.6. Square Integrable Martingales 283
Proof. Let ⌧n " ⌧ be a sequence of (F + ·)-stop time announcing ⌧ so that ⌧n < ⌧ on ⌧ > 0.
Note that since Mt2 hM, M it is a martingale, for m n we have
Taking limit as n ! 1 and using that M⌧n ! M⌧ (we need to use M0 = 0), it follows that
Now taking limit as m ! 1 and using Theorem 8.16 along with Theorem 1.36, we conclude
T
Since hM, M i is predictable, hM, M i⌧ is F⌧
Both the required relations (8.6.14) and (8.6.15) follow from this.
F
Corollary 8.67. Let M 2 M2loc and let ⌧ be a predictable stop time. Then for all bounded
stop times such that M [ ] 2 M2
A
( hM, M i⌧ )1{⌧ } = E[( M [ ] )2⌧ | F⌧ ]. (8.6.16)
Theorem 8.68. Let Y 2 M2loc be a locally square integrable martingale with Y0 = 0. Then
there exist sequence {⌧k : k 1} of predictable stop times such that the local martingale
D
Y =Z +U (8.6.17)
where
P1 k , U k i.
(v) hU, U i = k=1 hU
and
1
X
hY, Y i = hZ, Zi + hU k , U k i. (8.6.19)
k=1
T
Further, hU k , U k i is a process with a single jump at ⌧k and if m " 1 are bounded stop
times such that Y [ m ] 2 M2
hU k , U k i⌧k 1{⌧k m}
= E[( Y [ m]
)2⌧k | F⌧k ]. (8.6.20)
F
Proof. Let A = hY, Y i. By definition A is an increasing predictable process with A0 = 0.
Using Theorem 8.21, we can get predictable stop times {⌧m : m 1} with disjoint graphs
A
that cover jumps of A, i.e.
X1
( A) = ( A)⌧m 1[⌧m ] . (8.6.21)
m=1
R
For k 1, let
Utk = ( Y )⌧k 1[⌧k ,1) (t). (8.6.22)
As observed in Lemma 8.63 and Corollary 8.64, U k is a locally square integrable martingale
D
and (8.6.20) holds. Since graphs of {⌧m : m 1} are disjoint, it follows that
[U k , U j ] = 0, for j, k 1 (8.6.23)
and as a consequence
hU j , U k i = 0 for j, k 1. (8.6.24)
Pn
where U 2 M2d,loc . In view of (8.6.23) and (8.6.24) we have [S n , S n ] = k=1 [U
k, U k] and
P
hS n , S n i = nk=1 hU k , U k i. Part (iv) in Theorem 8.65 now implies
1
X
[U, U ] = [U k , U k ] (8.6.25)
k=1
1
X
hU, U i = hU k , U k i. (8.6.26)
k=1
T
[Y, Y ] = [Z, Z] + [U, U ].
This along with (8.6.25) implies (8.6.18). [Z, U ] = 0 also gives hZ, U i = 0 and thus
Remains to prove that hZ, Zi is continuous. By the choice of the stop times {⌧k : k
A
1}, it follows that the hY, Y i does not have jumps other than at {⌧k : k 1}. From
Corollary 8.67 we get
( hY, Y i⌧k )1{⌧k = E[( Y [ m]
)2⌧k | F⌧k ]
R
m}
Since hU j , U j i has a single jump at ⌧j and since the graphs of {⌧k : k 1} are disjoint, it
follows that
hU, U i⌧k = hU k , U k i⌧k .
Thus we conclude that for all k 1
hY, Y i⌧k = hU, U i⌧k .
This implies hZ, Zi is continuous.
Remark 8.69. Note that in the decomposition (8.6.17), Z and U are such that hU, U i is
purely discontinuous and hZ, Zi is continuous.
286 Chapter 8. Predictable Increasing Processes
We will explore further structure of elements in M2d,loc and conclude with identifying the
continuous part [X, X](c) of the quadratic variation of a semimartingale X as the quadratic
variation of its continuous local martingale martingale part, to be defined below.
Lemma 8.70. Let Z 2 M2loc be such that hZ, Zi is continuous. Then for any predictable
stopping time ,
( Z) = 0. (8.6.27)
Proof. Let ⌧k " 1 be stopping times such that Z k = Z [⌧k ] 2 M2 and let Ytk = (Ztk )2
hZ, Zit^⌧k . Then Y k is a martingale and hence using Theorem 8.14 it follows that
E[( Y k ) ^n ] = 0.
T
E[(Z k^n )2 (Z(k ^n) )2 ] = 0.
Thus
E[(Z k^n
F
Z(k ^n) )2 ] = 0.
Since this holds for all n and for all k and ⌧k " 1, this completes the proof.
Lemma 8.71. Let Z 2 M2loc be such that hZ, Zi is continuous. Let ⌧ be a stop time and let
D = ( Z)⌧ 1[⌧,1) . Let A be the compensator of D (A is the unique predictable process in
D
Proof. Let ⌧ be any predictable stop time. By Lemma 8.70, we conclude that ( Z)⌧ = 0.
Since D = R + A this gives
( A)⌧ = ( R)⌧ .
E[( R)⌧ | F⌧ =0
8.6. Square Integrable Martingales 287
for all bounded predictable stop times. Since A is predictable, this shows that A is contin-
uous.
Theorem 8.72. Let Z 2 M2loc be such that hZ, Zi is continuous. Then Z admits a decom-
position
Z =S+R (8.6.29)
T
Proof. By Theorem 8.21, we can get stop times { j : j 1} that cover the jumps of Z.
Let
Vtj = ( Z) j 1[ (t), (8.6.31)
n
X Xn
[W n , W n ] = [Rj , Rj ]t = ( Z)2 j 1[ j ,1) (t) (8.6.32)
j=1 j=1
and
D
[Z W n , W n ] = 0. (8.6.33)
[Z R, R] = 0.
Defining M = Z R, we conclude
In turn, (8.6.35) and (8.6.36) together yield the validity of (8.6.30). Since ( [Z, Z])s =
( Z)2s and as the stop times { j : j 1} cover jumps of Z, (8.6.30) implies that [M, M ]
is continuous and hence S is a continuous local martingale. This completes the proof.
Theorem 8.73. Let Y 2 M2loc be a locally square integrable martingale with Y0 = 0. There
exist predictable stop times {⌧k : k 1} and a sequence of stop times { j : j 1} such
T
k j
that U and V defined by
Utk = ( Y )⌧k 1[⌧k ,1) (t)
Vtj = ( N ) j 1[
satisfy
Pm k em
(iv) k=1 R ! R 2 M2d,loc .
Proof. The proof is just putting together various parts proven in Theorem 8.68 and Theo-
rem 8.72. First get {⌧k : k 1} as in Theorem 8.68 that cover jumps of hY, Y i and define
U k , U as above. Writing Z = Y U , we conclude hZ, Zi is continuous. Now we get j to
cover jumps of Z and use results proven in Theorem 8.72.
T
[N, N ] = ( N )2s .
0<st
Theorem 8.76. Let X be a semimartingale. Then there exists a continuous local martin-
(c)
gale X (c) with X0 = 0 and such that S = X X (c) is a semimartingale that satisfies
Proof. First, we invoke Corollary 5.53 and decompose X = Y + A where Y 2 M2loc and
A 2 V. Then we use Theorem 8.73, we get a decomposition Y = M + N with M 2 M2c,loc
and N 2 M2d,loc . Let X (c) = M . Since N 2 M2d,loc ,
[U, N ] = 0 8U 2 M2c,loc .
290 Chapter 8. Predictable Increasing Processes
T
F
A
R
D
Chapter 9
In this Chapter, we would give the continuous time version of the Burkholder-Davis-Gundy
T
inequality - p = 1 case. This is due to Davis. This plays an important role in answering
various questions on the stochastic integral w.r.t. a martingale M - including condition on
R
f 2 L(M ) under which f dM is a local martingale. This naturally leads us to the notion
of a sigma- martingale which we discuss.
F
We will begin with a result on martingales obtained from process with a single jump.
A
9.1 Preliminaries
Lemma 9.1. Let ⌧ be a stop time and let ⇣ be a F⌧ measurable [0, 1)-valued integrable
R
( A) = E[( X) | F ]
291
292 Chapter 9. The Davis Inequality
E[ |( A) | ] = E[ |E[( X) | F ]| ]
E[E[ |( X) | | F ]]
(9.1.2)
= E[ |( X) | ]
= E[ |⇣|1{⌧ <1} 1{ =⌧ } ].
T
Since
|( A) | lim inf ( A)⌧k
k!1
and F
|⇣|1{⌧ <1} 1{⌧k =⌧ } " |⇣|1{⌧ <1} 1{ =⌧ } a.s.
A
we can take limit as k ! 1 and use Fatou’s lemma on left hand side and monotone
convergence theorem on right hand side to conclude
{( A) 6= 0} = [m 1[ m]
D
(existence of such stop times was proven in Theorem 8.21). Recall that the graphs being
disjoint means
P( n = m, n < 1) = 0 8n, m, n 6= m. (9.1.5)
Thus
v
u 1
p uX
[A, A]T = t ( A)2 m
m=1
1
X
|( A) m |.
m=1
9.2. Burkholder-Davis-Gundy inequality - Continuous Time 293
Noting that by Lemma 4.68 [M, M ]T 2([X, X]T + [A, A]T ) and by definition of X,
[X, X]T = ⇣ 2 1{⌧ T } , we conclude that
p p p p
E[ [M, M ]T ] 2E[ [X, X]T + [A, A]T ]
1
X
p p
2E[⇣] + 2E[ |( A) m |]
m=1
1 (9.1.6)
p p X
2E[⇣] + 2E[ 1{⌧ <1} ⇣ 1{ m =⌧ }
]
m=1
p p
2E[⇣] + 2E[⇣]
where we have used (9.1.4) and (9.1.5). This completes the proof of (9.1.1).
T
We will prove the p = 1 case of the Burkholder-Davis-Gundy inequality: For 1 p < 1,
there exist universal constants c1p , c2p such that for all martingales M and T < 1,
p
F
c1p E[([M, M ]T ) 2 ] E[ sup |Mt |p ]
0tT
c2p E[([M, M ]T ) 2 ].
p
A
We have given a proof for p = 1 in the discrete case and here we will approximate the
continuous time martingale by its restriction to a discrete skeleton and then pass to the
limit.
One inequality follows easily from the discrete case. For the other we first note it for
R
the case of square integrable martingale and then later we will prove the same without this
restriction.
D
Theorem 9.2. Let c1 , c2 be the universal constants appearing in Theorem 1.44. Let M be
a martingale with M0 = 0. Then
1
c1 E[([M, M ]T ) 2 ] E[ sup |Mt | ]. (9.2.1)
0tT
and
Z n = max n |Mtk,n |.
1k2
Also let
p
Q= [M, M ]T .
Applying the discrete version of the inequality proven in Theorem 1.44, we have
c1 E[Qn ] E[Z n ] c2 E[Qn ]. (9.2.3)
We have seen in Theorem 4.59 that Qn converges to Q in probability and hence a sub-
sequence Qnk converges almost surely. Then applying Fatou’s Lemma we conclude from
(9.2.3) that
c1 E[Q] c1 lim inf E[Qnk ] lim inf E[Z nk ].
k!1 k!1
Since Zn
increases to Z, E[Z nk ] converges to E[Z] and thus (9.2.1) follows. Also we get
T
from (9.2.3)
E[Z] c2 lim inf E[Qnk ].
k!1
If E[MT2 ] < 1, it follows that
F
E[(Qn )2 ] = E[MT2 ] < 1.
Noting that E[MT2 ] = E[[M, M ]T ] = E[Q2 ], it follows that Qn converges to Q in L2 (P) and
A
hence in L1 (P) and thus
lim inf E[Qnk ] = E[Q]
k!1
and thus (9.2.2) follows.
R
conclusion follows by passing to the limit and invoking monotone convergence theorem.
We first consider a special case and show that (9.2.2) holds in this case.
Lemma 9.4. Let M be a martingale with M0 = 0 and let be a stop time bounded by T
such that
|Mt | K 8t < . (9.2.4)
and
Mt = Mt^ 8t. (9.2.5)
p
Then E[ [M, M ]T ] < 1 and (9.2.2) holds for M .
9.2. Burkholder-Davis-Gundy inequality - Continuous Time 295
and hence sup0tT |Mt | is integrable. Hence by first part of Theorem 9.2, it follows that
p
E[ [M, M ]T ] < 1. Let
Ajt = ⇠ + 1{|⇠| j} 1[ ,1) (t)
T
Vtj = Btj Dtj
q
F
E[ [U j , U j ]T ] 3E[⇠ + 1{|⇠| j} ]
E[ [V j , V j ]T ] 3E[⇠ 1{|⇠| j} ]
(9.2.6)
A
(9.2.7)
Let
D
Then M j = M U j + V j . Now
Now |Mt (Ajt Btj )| K for t < (as in that case Ajt = Btj = 0). Also
M (Aj Bj ) = M +⇠ ⇠ + 1{|⇠| j} + ⇠ 1{|⇠| j}
=M + ⇠ 1{|⇠|<j} .
Further,
Mt (Ajt Btj ) = Mt^ (Ajt^ j
Bt^ ).
296 Chapter 9. The Davis Inequality
Hence
|Mt (Ajt Btj )| K + j 8t.
C j and Dj being predictable r.c.l.l. processes are locally bounded and hence it follows that
M j is locally bounded, hence locally square integrable. Thus by Remark 9.3, we have
1
E[ sup |Mtj | ] c2 E[([M j , M j ]T ) 2 ]. (9.2.10)
0tT
In view of (9.2.8) and (9.2.9), UTj ! 0 and VTj ! 0 in L1 (P) and hence MTj ! MT in
L1 (P). Thus, by Doob’s maximal inequality,
By going through a subsequence, and using Fatou’s Lemma we conclude from (9.2.10) that
T
1
E[ sup |Mt | ] c2 lim inf E[([M j , M j ]T ) 2 ]. (9.2.11)
0tT j!1
Thus
E[([M Mj, M
[M Mj, M
1
M j ]T ) 2 ]
p
F
M j ]t 2([U j , U j ]t + [V j , V j ]t )
1
2E[([U j , U j ]T ) 2 ] +
p 1
2E[([V j , V j ]T ) 2 ]
A
p (9.2.12)
3 2E[ |⇠|1{|⇠| j} ]
where we have used (9.2.6), (9.2.7) and hence
R
1
lim E[([M Mj, M M j ]T ) 2 ] = 0. (9.2.13)
j!1
Theorem 9.5. There exist universal constants c1 , c2 such that for all local martingales M
with M0 = 0 and for all t > 0 one has
1 1
c1 E[([M, M ]T ) 2 ] E[ sup |Mt | ] c2 E[([M, M ]T ) 2 ]. (9.2.15)
0tT
Proof. Let {⌧n : n 1} be stop times increasing to 1 such that Mt^⌧n is a martingale.
For n 1 let
⇣n = inf{t 0 : |Mt | n or |Mt | n}
T
Theorem 9.2. Thus we have
1 1
c1 E[([N n , N n ]T ) 2 ] E[ sup |Ntn ]| ] c2 E[([N n , N n ]T ) 2 ] < 1. (9.2.16)
0tT
1 1
F
Now as n ! 1, ([N n , N n ]T ) 2 increases to ([M, M ]T ) 2 and sup0tT |Ntn | increases to
sup0tT |Mt ]| and thus (9.2.15) follows from (9.2.16) using Monotone Convergence Theo-
rem.
A
During the proof given above we have shown the following:
Corollary 9.6. Let M be a local martingale. Then there exist stop times n increasing
R
to 1 such that
p
E[ [M, M ] n ] < 1 8n 1 (9.2.17)
and as a consequence
D
Corollary 9.7. Let M be a local martingale. For any stop time , one has
1 1
c1 E[([M, M ] ) 2 ] E[ sup |Mt | ] c2 E[([M, M ] ) 2 ]. (9.2.19)
0t
then it follows that E[sup0t |Mt ]| < 1 for all T < 1 and hence Nt = Mt^ is a
martingale.
298 Chapter 9. The Davis Inequality
Proof. We will show that for all k 1, there exists a bounded martingale Z k such that
1
E[ sup |Ztk Xt | ] . (9.2.23)
0tk k
T
The required result follows from this.
Thus we fix an integer k < 1. Let n be the stop times constructed in the proof
of Theorem 9.5 with M = X and N n denote the martingale X stopped at n . Since
E[ sup |Yt
0tk
F
sup0tT |Ntn | increases to sup0tT |Xt |, we can get integer n such that Y = N n satisfies
Xt | ]
1
3k
. (9.2.24)
A
As noted in the proof of Theorem 9.5, N n and hence Y satisfies the conditions of Lemma
9.4. Thus for M = Y , we can get locally bounded martingales M j such that (9.2.13) holds,
i.e.
R
1
lim E[([Y M j , Y M j ]k ) 2 ] = 0. (9.2.25)
j!1
Now using Burkholder-Davis-Gundy inequality Theorem 9.5, we can get j such that W =
M j satisfies
D
1
E[ sup |W Yt | ] . (9.2.26)
0tk 3k
Finally, W being locally bounded, we can get stop times ⌧n increasing to 1 such that U n
given by Utn = Wt^⌧n is a bounded martingale and
E[ sup |Utn W | ] ! 0 as n ! 1
0tk
and hence can get n such that Z = U n satisfies
1
E[ sup |Zt W|] . (9.2.27)
0tk 3k
Now (9.2.25)-(9.2.27) together imply (9.2.23) with Z k = Z.
9.3. On Stochastic Integral w.r.t. a Martingale 299
T
9.3 On Stochastic Integral w.r.t. a Martingale
R
F
For a local martingale M , the stochastic integral Y = f dM for f 2 L(M ) is defined (since
M is also a stochastic integrator), we have only observed that when M is locally square
integrable martingale and f 2 L2m (M ), Y is also a locally square integrable martingale.
A
We now explore as to when is Y a martingale or a local martingale. We begin with an
observation.
1
Theorem 9.12. Let M be a martingale such that E[([M, M ]T ) 2 ] < 1 8T < 1 and f be a
R
R 1
bounded predictable process. Then N = f dM is also a martingale and E[([N, N ]T ) 2 ] < 1
8T < 1.
R
D
Hence
p
E[ [N n N, N n N ]T ] ! 0 as n, ! 1
E[ sup |Nsn Ns | ] ! 0 as n ! 1.
0tT
T
process. Then Y = f dM is also a local martingale.
R
Y = X dM is also a local martingale.
F
Corollary 9.14. Let M be a local martingale and X be an r.c.l.l. adapted process. Then
A
Earlier we have defined L2m (M ) for a locally square integrable martingale M . We now
define L1m (M ) for a local martingale M .
R
Definition 9.15. For a local martingale M , L1m (M ) is the class of predictable processes f
such that there exist stop times ⇣n increasing to 1 with
Z ⇣n
1
E[( fs2 d[M, M ]s ) 2 ] < 1. (9.3.1)
D
Theorem 9.16. Let M be a local martingale. Then L1m (M ) ✓ L(M ) and for f 2 L1m (M ),
R
N = f dM is a local martingale.
Proof. Let f 2 L1m (M ) be such that (9.3.1) holds. We first show that f 2 L(M ). Let g k
be bounded predictable processes converging pointwise to g such that |g k | |f |.
R
For k 1, let Y k = g k dM . Then we have seen that Y k is a local martingale. From
properties of stochastic integrators, we have
Z t
k k
[Y , Y ]t = (g k )2 d[M, M ]
0
9.3. On Stochastic Integral w.r.t. a Martingale 301
and hence Z ⇣n 1
E[ sup |Ysk | ] 2
c E[( (fs )2 d[M, M ]s ) 2 ] < 1 (9.3.2)
0t⇣n 0
T
lim ( sup E[ sup |Ytk Ytj | ] ) = 0. (9.3.3)
m!1 j,k m 0t⇣n
Thus {Y k}are Cauchy in ducp and hence f 2 L(M ). Let Y be the limit of Y k . By
R
k!1 0t⇣n
F
dominated convergence theorem, we get Y = gdM We also get from (9.3.3) that
lim E[ sup |Ytk Yt | ] = 0.
A
This and Utk,n = Yt^⌧ k
n
being a martingale for each k, n implies that Utn = Yt^⌧n is a
martingale and hence Y is a local martingale.
To show that N is a local martingale, let us take g k = f 1{|f |k} . The process Y for
R
this choice of {g k } equals N which has been shown to be a local martingale completing the
proof.
and thus f 2 L1m (M ). The second part follows from Theorem 9.16.
Proof. Since Z is bounded, f satisfies (9.3.4). Thus by Corollary 9.17, Z is a local martin-
gale. Since it is bounded, it follows that Z is a martingale.
T
Corollary 9.19. Let M be a continuous local martingale. Then
Corollary 9.18. F
Proof. Since M is continuous, (9.3.4) is trivially satisfied and hence the result follows from
9.4 Sigma-Martingales
R
We have seen that if M is a local martingale and f 2 L1m (M ) then X = f dM is a local
martingale. On the other hand if f 2 L(M ) but does not belong to L1m (M ) then X is
defined and is a semimartingale but it is not a local martingale. Nonetheless, it shares
some properties of a local martingale and is called a sigma-martingale.
1
R
Proof. If such a M , exist, then = 2 L(M ) and X = dM .
R
For the converse part, suppose N is a local martingale and f 2 L(N ) with X = f dN .
R R
Then taking g = (1 + |f |) 1 , we observe that Y = gdX = f gdN . Since N is a local
R
martingale and f g is bounded by 1, invoking Corollary 9.13 we conclude that Y = f gdN
is itself a local martingale. As seen in Corollary 9.6, there exist stop times n increasing
T
to 1 such that
p
an = E[ [Y, Y ] n ] < 1.
hs =
1
1 + |Y0 |
1{0} (s) +
X
n=1
2 n
F
1
1
1
1 + an ( n 1 , n ]
(s).
A
R
Then h in (0, 1) valued and thus M = hdY is a local martingale. It can be checked that
p
E[ [M, M ]T ] < 1 and |M0 | 1. As a result M is a martingale. Let = hg. Then is
R
(0, 1) valued and dX = M .
R
From the definition, it is not obvious that sum of sigma-martingales is also a sigma-
martingale, but this is so as the next result shows.
D
are martingales. Then, writing ⇠ = min( 1 , 2 ) and ⌘si = ⇠si , it follows that
s
Z t Z t
Nti = ⌘si dMsi = ⇠s dXsi
0 0
304 Chapter 9. The Davis Inequality
The following result gives conditions under which a sigma-martingale is a local martin-
gale.
T
Then X is a local martingale.
R
Proof. Let N be a local martingale and f 2 L(N ) be such that X = f dN . Note that
Z t
Let
[X, X]t =
Xtk =
Z t
0 F
(fs )2 d[N, N ]s . (9.4.2)
A
fs 1{|fs |k} dNs . (9.4.3)
0
Noting that fs 1{|fs |k} is bounded, it follows that X k is a local martingale. Since
sZ
q ⌧n
R
<1
we conclude that for k, n fixed, Ztk,n = Xt^⌧
k
n
is a martingale. Let Ztn = Xt^⌧n . Further,
sZ
q ⌧n
E[ [X X k , X X k ]t^⌧n ] E[ fs2 1{|fs |>k} d[N, N ]s ]. (9.4.4)
0
The assumption (9.4.1) and the estimate (9.4.4) implies that for n fixed,
q
lim E[ [X X k , XX k]
t^⌧n ] = 0. (9.4.5)
k!1
The Burkholder-Davis-Guindy inequality- Corollary 9.7 now gives
lim E[sup |Ztk,n Ztn |] = 0. (9.4.6)
k!1 st
9.5. Auxiliary Results 305
Since Z k,n is a martingale for each k, n, (9.4.6) implies Z k is a martingale and thus X is a
local martingale.
Proof. Let Yt = Xt X0 . Observe that [Y, Y ]t = [X, X]t for all t. Now the previous result
gives that Y is a local martingale and hence X is a local martingale.
Proof. Since X is bounded, say by K, it follows that jumps of X are bounded by 2K. Thus
jumps of the increasing process [X, X] are bounded by 4K 2 and thus X satisfies (9.4.1) for
T
⌧n = inf{t 0 : [X, X]t n}.
F
Here is a variant of the example given by Emery [17] of a sigma-martingale that is not
a local martingale. Let ⌧ be a random variable with exponential distribution (assumed
A
to be (0, 1)-valued without loss of generality) and ⇠ with P(⇠ = 1) = P(⇠ = 1) = 0.5,
independent of ⌧ . Let
Mt = ⇠ 1[⌧,1) (t)
R
1
and Ft = (Ms : s t). Easy to see that M is a martingale. Let ft =
Rt t 1(0,1) (t) and
Xt = 0 f dM . Then X is a sigma-martingale and
1
[X, X]t = 1 (t).
⌧ 2 [⌧,1)
D
For any stop time , it can be checked that is a constant on < ⌧ and thus if is not
p 1
identically equal to 0, (⌧ ^ a) for some a > 0. Thus, [X, X]
p ⌧ 1{⌧ <a} . It follows
that for any stop time , not identically zero, E[ [X, X] ] = 1 and so X is not a local
martingale.
Lemma 9.28. Let N be martingale such that E[sup0tT |Nt | ] < 1 for all T < 1 and let
A 2 V be a predictable process with A0 = 0 such that |A|T KT for some constant KT < 1
Rt
for all T < 1, where |A|T is the variation of At on [0, T ]. Then Yt = At Nt 0 Ns dAs
is a martingale.
T
Proof. Invoking Theorem 9.9, obtain bounded martingales N k such that
The integral appearing above is a local martingale by Corollary 9.14 and [A, N k ] is a
martingale by Theorem 8.33. Thus Y k is a local martingale. Lemma 5.5 along with the
observation (9.5.1) implies that Y k is a martingale for each k and then (9.5.2) along with
Theorem 2.21 shows that Y is a martingale.
Theorem 9.29. Let M be local martingale such that M0 = 0 and let A 2 V be a predictable
Rt
process with A0 = 0. Then Yt = At Mt 0 Ms dAs is a local martingale.
Proof. Since A 2 V is predictable, so is its total variation process B = |A| (see Corollary
n
8.23). Thus B is locally bounded and we can get stop times ⌧ n such that B ⌧ is bounded.
9.5. Auxiliary Results 307
n]
Invoking Corollary 9.8, get stop times n " 1 such that M n = M [ are martingales
satisfying, for each n 1
E[ sup |Mtn ]| < 1 8T < 1.
0tT
T
Lemma 9.30. Let M, N be local martingales such that M0 N0 is integrable. Then the process
Z defined by Zt = Mt Nt is locally integrable if and only [M, N ] is locally integrable.
Mt N t = M0 N 0 +
Z t
Ms dNs +
0
F
Z t
Ns dMs + [M, N ]s .
0
Theorem 9.16 yields that the stochastic integral terms in the right hand side above are
A
local martingales and thus the result follows.
Remark 9.31. If M, N are local martingales such that M N is locally integrable, then so is
[M, N ] and thus hM, N i exists and is the unique predictable process in V+
R
0 such that
Mt N t hM, N it
is a local martingale.
D
Chapter 10
In this chapter we will consider the question as to when do all martingales adapted to
T
a filtration (F· ) admit a representation as a stochastic integral with respect to a given
local martingale M . This result was proved by Ito when the underlying filtration is the
filtration generated by a multidimensional Wiener process. Ito had proven the integral
martingales by Clark.
F
representation property for square integrable martingales and this was extended to all
Jacod and Yor investigated this aspect and proved that the integrable representation
A
property holds if and only if there does not exist any other probability measure Q equivalent
to the underlying probability measure with the property that X is a Q - local martingales.
Such a measure Q is called an Equivalent Martingale Measure (EMM). In other words,
R
martingale representation property holds if and only if EMM is unique. Jacod-Yor proved
this result in one dimension and in a special case for multidimensional local martingale,
which was subsequently extended.
D
This result is important from the point of view of mathematical finance. We will give
brief introduction to same and prove the second fundamental theorem of asset pricing.
10.1 Preliminaries
Throughout this chapter, we will be working with one fixed filtration (F⇧ ). All notions
- martingale, local martingale, stop time, adapted process, predictable process are with
reference to this fixed filtration.
Let X 1 , X 2 , . . . , X d be semimartingales. We introduce the class C(X 1 , X 2 , . . . , X d ) of
308
10.1. Preliminaries 309
Let us note that if Y 2 C(X 1 , X 2 , . . . , X d ) then for any stop time ⌧ , Ỹ defined by Ỹt = Yt^⌧
also belongs to C(X 1 , X 2 , . . . , X d ).
A semimartingale Y is said to have an integral representation w.r.t. semimartingales
X 1 , X 2 , . . . , X d if Y 2 C(X 1 , X 2 , . . . , X d ). Here is an elementary observation on the class
C(X 1 , X 2 , . . . , X d ).
Lemma 10.1. Let Y be a semimartingale such that for a sequence of stop times ⌧n " 1,
Y n defined by Ytn = Yt^⌧n admits an integral representation w.r.t. X 1 , X 2 , . . . , X d for each
n 1. Then Y also admits an integral representation w.r.t. X 1 , X 2 , . . . , X d .
T
Proof. Let f n,j 2 L(X j ), 1 j d, n 1 be such that for all n,
Xd Z t
Ytn = Y0n + f n,j dX j .
Define f j by
j
f =
1
X
F
j=1
1(⌧n
0
f n,j .
A
1 ,⌧n ]
n=1
Yt = Y0 + f j dX j .
j=1 0
Exercise 10.2. Let Y be an r.c.l.l. process such that it is a local martingale under probability
measure Q1 as well as Q2 . Show that Y is a local martingale under Q = 12 (Q1 + Q2 ).
T
10.2 One Dimensional Case
F
In this section, we will fix a local martingale M and explore as to when C(M ) contains all
martingales. The next lemma gives an important property of C(M ).
A
Lemma 10.3. Let M be a local martingale and let N n 2 C(M ) be martingales such that
E[ |Ntn Nt | ] ! 0 for all t. Then N 2 C(M ).
Proof. The assumptions imply that N is a martingale (see Theorem 2.21). In view of
R
[N n N, N n N ]T ! 0 in probability as n ! 1. (10.2.1)
[N n N m, N n N m ]T ! 0 in probability as n, m ! 1. (10.2.2)
T
1 X T
1 1
[ ( |gsk+1 gsk |)2 d[M, M ]s ] 2 ( (gsk+1 gsk )2 d[M, M ]s ) 2
0 m=1 m=1 0
and also
[
Z T X
0
1
( |gsk+1
k=1
F
gsk |)2 d[M, M ]s ] < 1 a.s. (10.2.6)
A
Z T
lim sup |gsm gsn |2 d[M, M ]s = 0 a.s. (10.2.7)
k!1 0 m,n k
Using (10.2.7), one can conclude that for all t < 1, for all ✏ > 0
Z t
1
lim P(( (gsk gs )2 d[M, M ]s ) 2 ✏) = 0. (10.2.9)
k!1 0
312 Chapter 10. Integral Representation of Martingales
and Z Z
t t
(gsk )2 d[M, M ]s ! (gs )2 d[M, M ]s in probability.
0 0
On the other hand convergence of N n to N in Emery topology and Theorem 4.103 implies
for all t < 1
Z t
(gsk )2 d[M, M ]s = [N k , N k ]t ! [N, N ]t in probability.
0
Thus Z t
[N, N ]t = (gs )2 d[M, M ]s . (10.2.10)
0
Now N being a martingale, as seen in Corollary 9.6, there exist stop times n increasing
to 1 such that
p
E[ [N, N ] n ] < 1 8n 1 (10.2.11)
Rt
T
and hence (10.2.10) implies that g 2 L1m (M ). Let Yt = N0 + 0 gdM . By definition, Y is
a local martingale with Y0 = N0 . Now
Z T
n n
[N Y, N Y ]T = (gsn gs )2 d[M, M ]s
Y ]T ! 0 in probability (10.2.12)
A
On the other hand N n converges to N in Emery topology and thus invoking Theorem 4.103
we conclude that
[N n Y, N n
R
Y ]T ! [N Y, N Y ]T in probability. (10.2.13)
Thus [N Y, N Y ]T = 0 and recalling that N, Y are local martingales such that N0 = Y0 ,
it follows (once again invoking Burkholder-Davis-Gundy inequality for p = 1) that N = Y .
Rt
D
Proof. Let N n 2 C(M ) \ M be such that NTn converges in L1 (⌦, FT , P) to ⇠. Without loss
of generality, we assume that Ntn = Nt^T
n for all n 1 and for all t < 1. Now for each n,
N n is a uniformly integrable martingale. It follows that for all t
E[|Ntn Ntm | ] ! 0 as n, m ! 1. (10.2.15)
10.2. One Dimensional Case 313
Thus, by Theorem 5.33, it follows that {N n } is Cauchy in the dem metric for the Emery
topology. Since this metric is complete, the sequence N n converges in the Emery topology
to say N . Then N n also converges in ducp to N and in turn, in view of (10.2.15), we
conclude NT = ⇠ and for each t
E[|Ntn Nt | ] ! 0 as n ! 1. (10.2.16)
Thus Lemma 10.3 implies N 2 C(M ) and thus NT = ⇠ 2 KT (M ).
Exercise 10.5. Show that ⇠ 2 KT if and only if ⇠ 2 L1 (⌦, FT , P) and there exist ⌘ 2
RT
L1 (⌦, F0 , P) and g 2 L1m (M ) such that ⇠ = ⌘ + 0 gdM .
We now come to the main result of this section, due to Jacod-Yor [28]. This character-
izes martingales M with property that all martingales N admit an integral representation
w.r.t. M .
T
Definition 10.6. A process Y is said to admit an integral representation w.r.t. a semimartin-
gale X if Z t
F
9f 2 L(X) such that Yt = Y0 +
0
fs dXs a.s. 8t. (10.2.17)
Lemma 10.7. Let M be an r.c.l.l. local martingale. Then all martingales N admit a
D
Proof. Suppose all martingales N admit a representation w.r.t. M . Given ⇠ 2 L1 (⌦, FT , P),
consider the martingale Nt = E[⇠ | Ft ]. Note that N may not be r.c.l.l. to begin with. In
view of our assumption, get f 2 L(M ) such that
Z t
Nt = N0 + f dM a.s. 8t.
0
Rt
This implies that Vt = N0 + 0 f dM is an r.c.l.l. martingale and is a version of N . Thus
by definition of KT (M ), it follows that NT = ⇠ 2 KT (M ).
314 Chapter 10. Integral Representation of Martingales
Conversely, if (10.2.18) holds, then given a martingale N , fix n and let ⇠ = Nn Then
RT
⇠ 2 Kn (M ) and so we get f n 2 L(X) such that ⇠ = 0 f n dM and further that Z n =
R n
f dM is a martingale. It follows that Ztn = Nt a.s. for t T . Let
X
fs = fsn 1(n 1,n] .
n
R
Then one can check that f 2 L(M ), Z = f dM is a martingale and Zt = Nt a.s. for all
t. Hence N admits a representation w.r.t. M .
Corollary 10.8. Let M be an r.c.l.l. local martingale. Then all bounded martingales N
admit a representation w.r.t. M if and only if
T
Theorem 10.9. Let M be an r.c.l.l. local martingale on (⌦, F, P). Suppose that F0 is
trivial and F = ([t Ft ). Then the following are equivalent.
F
(i) Every bounded martingale N admits an integral representation w.r.t. M .
e P (M ) = {P}.
(iv) E
R
(v) EP (M ) = {P}.
Proof. We have seen that (i) is same as L1 (⌦, FT , P) ✓ KT (M ) 80 < T < 1 and (ii) is
D
Till now we have proved (i) () (ii) and (iii) ( (iv) () (v).
To complete the proof, we will show (i) ) (v) and (iii) ) (ii).
First we come to the proof of (iii) ) (ii). Suppose P is an extreme point of E(M )
but (ii) is not true. We will show that this leads to a contradiction. Since (ii) is not
true, it follows that KT (M ) is a closed proper subspace of L1 (⌦, F, P). Since KT (M ) is
not equal to L1 (⌦, FT , P), by the Hahn-Banach Theorem, there exists ⇠ 2 L1 (⌦, FT , P),
P(⇠ 6= 0) > 0 such that Z
✓⇠ dP = 0 8✓ 2 KT (M ).
T
P(c|⇠| < 0.5) = 1.
Now, let Q be the measure with density ⌘ = (1 + c⇠). Then Q is a probability measure.
Thus (10.2.20) yields Z Z
F
✓dQ = ✓dP 8✓ 2 KT (M ).
EQ [M⌧n_T ] = EP [⌘M⌧n_T ]
= EP [EP [⌘M⌧n_T | FT ]]
= EP [⌘EP [M⌧n_T | FT ]]
D
(10.2.23)
= EP [⌘MTn ]
= EQ [MTn ]
= M0 .
where we have used the facts that ⌘ is FT measurable, M n is a P-martingale and (10.2.22).
Now noting that M⌧n = M⌧n^T + M⌧n_T MTn , we conclude
EQ [M⌧n ] = EQ [M⌧n^T ] + EQ [M⌧n_T ] EQ [MTn ] = M0 .
Thus Mtn = Mt^ n is a Q-martingale for every n so that M is a Q local martingale and
thus Q 2 E(M ). Similarly, if Q̃ is the measure with density ⌘ = (1 c⇠), we can prove
316 Chapter 10. Integral Representation of Martingales
that Q̃ 2 E(M ). Here P = 12 (Q + Q̃) and P 6= Q (since P(⇠ 6= 0) > 0). This contradicts the
assumption that P is an extreme point of E(M ). Thus (iii) ) (ii).
To complete the proof we will prove (i) implies (v). Suppose (i) is true and let Q 2
EP (M ). Fix T < 1 and let ⌘ be any FT measurable bounded random variable. Since
L1 (⌦, FT , P) ✓ KT (M ) and F0 is trivial, we can get g 2 L(M ) with
Z T
⌘ =c+ gdM
0
Rt Rt
such that 0 gdM is a martingale. Let Zt = 0 gs 1[0,T ] (s)dMs . It follows that Zt =
EP [(⌘ c) | Ft ] and since ⌘ is bounded, it follows that Z is bounded. As noted earlier,
since P and Q are equivalent, the stochastic integrals under P and Q are identical. Under
R
Q, M being a local martingale and Z = f dM being bounded, we conclude invoking
Corollary 9.18 that Z is also a martingale under Q. Thus, EQ [ZT ] = 0 = EP [ZT ] and
T
thus using ⌘ = c + ZT we get EQ [⌘] = c = EP [⌘]. Since this holds for all FT measurable
bounded random variables ⌘, we conclude Q and P agree on FT . In view of the assumption
F = ([t Ft ), we get Q = P proving (v). This completes the proof.
F
We can now deduce the integral representation property for Brownian Motion, due to
Ito [24] and Clark [10].
A
Theorem 10.10. Let W be one-dimensional Brownian motion and let Ft = FtW and
F = ([t FtW ). Then every martingale M admits a integral representation
R
Z t
Mt = M0 + f dW, 8t 0 (10.2.24)
0
Proof. We will prove that EP (W ) = {P }. The conclusion then would follow from Theorem
10.9. If Q 2 EP (W ), then by definition, W is a Q-local martingale and [W, W ]Q t = t
P
since Q is equivalent to P and [W, W ]t = t, see Remark 4.73. Now part (v) in Theorem
5.16 implies that Wt and Wt2 t are Q-martingales and then Levy’s characterisation of
Brownian motion, Theorem 3.5 implies that W is a Brownian motion under Q. Thus for
t1 , t2 , . . . , tm 2 [0, 1) and B 2 B(Rm ),
we cannot conclude that the sequence of integrands g n,j is converging as was the case in 1-
dimensional case. See counter example in [28]. This prompts us to introduce a condition
under which the class of martingales that admit integral representation is closed under L1
convergence.
First note that for semimartingales X 1 , X 2 , . . . , X d and 1, . . . , d 2 R, defining Y =
T
Pd jXj,
j=1
d
X
i j
[Y, Y ]t = [X i , X j ]t
and hence
d
X
i j
i,j=1
([X i , X j ]t
F [X i , X j ]s ) 0 a.s. (10.3.1)
A
i,j=1
In other words, for s < t fixed, the matrix (([X i , X j ]t [X i , X j ]s )) is non-negative definite.
quasi-elliptic if there exists a sequence of stop times ⌧n " 1 and constants ↵n > 0 such that
for all 1 , . . . , d 2 R, one has
d
X d
X
D
i j i j i j
([X , X ]t [X , X ]s ) ↵n2 ( i )2 ([X i , X i ]t [X i , X i ]s ) 8s < t ⌧n a.s.
i,j=1 i=1
(10.3.2)
T
where n and ↵n are as in (10.3.2).
Proof. Clearly, the assumption (10.3.2) implies that (10.3.4) is true for simple predictable
processes h1 , h2 , . . . , hd 2 S. Yow fixing h2 , . . . , hd 2 S, the class of h1 for which (10.3.4) is
F
true is seen to be bp-closed and hence by Monotone class theorem, (Theorem 2.64) contains
all bounded predictable processes. Similarly by (finite) induction on j we can show that
A
(10.3.4) is true for bounded predictable h1 , . . . , hj and hj+1 , . . . , hd 2 S, for 1 j d.
This proves validity of (10.3.4) when h1 , h2 , . . . , hd are bounded.
Now note that by the Kunita-Watanabe inequality (Theorem 4.72) and Remark 4.79,
P
the right hand side in (10.3.4) is finite a.s. for hj 2 L(X j ), 1 j d. Let ⌘ = dj=1 |hj |2
R
Thus validity of (10.3.4) for {hn,j : 1 j d} for all n 1 implies validity of (10.3.4) for
{hj 2 L(X j ) : 1 j d}.
10.3. Quasi-elliptical multidimensional Semimartingales 319
Then
d Z
X t
(gsn,i gsm,i )2 d[X i , X i ]s ! 0 in probability as n, m ! 1.
i,j=1 0
T
X = (X 1 , X 2 , . . . , X d ) is quasi-elliptic local martingale.
Here is the analogue of Lemma 10.3 in multidimensional case for a quasi-elliptic semi-
martingale.
F
Lemma 10.17. Let M 1 , M 2 , . . . , M d be a quasi-elliptic local martingale (each component is
a local martingale). Let N n 2 C(M 1 , M 2 , . . . , M d ) be martingales such that E[ |Ntn Nt | ] !
0 for all t. Then N 2 C(M 1 , M 2 , . . . , M d ).
A
Proof. The proof follows the steps in the proof Lemma 10.3. First we get g n,j 2 L(M j ) for
1 j d, n 1 such that
d Z t
R
X
Ntn = N0n + g n,j dM j .
j=1 0
We then conclude that N n is converging in Emery topology and as a consequence, for all
D
T < 1,
[N n N m , N n N m ]T ! 0 in probability as n, m ! 1. (10.3.5)
Here note that
d Z
X T
n m n m
[N N ,N N ]T = (g n,j g m,j )(g n,k g m,k )d[M j , M k ]. (10.3.6)
j,k=1 0
Getting n " 1 such that (10.2.11) holds and invoking the assumption that M 1 , . . . , M d
is quasi- elliptic, we conclude
Z n ^⌧n
T
1
E[( (gsn,j )2 d[M j , M j ]s ) 2 ] < 1 1 j d, n 1.
0
P Rt
Thus gj
2 Now defining Yt = N0 + dj=1 0 g j dM j , we can show that (10.2.12)
L(M j ).
and (10.2.13) hold and thus N = Y completing the proof that N 2 C(M 1 , M 2 , . . . , M d ).
F
Now the same proof as that of Theorem 10.4 gives us the following.
A
Theorem 10.18. For local martingales M 1 , M 2 , . . . , M d and T < 1 let
KT (M 1 , M 2 , . . . , M d ) = {NT : N 2 C(M 1 , M 2 , . . . , M d ) \ M}. (10.3.9)
R
We are now ready to prove the multidimensional version of the Theorem 10.9.
D
e P (M 1 , M 2 , . . . , M d ) = {P}.
(iv) E
(v) EP (M 1 , M 2 , . . . , M d ) = {P}.
Proof. The proof closely follows that of Theorem 10.9. Once again we can observe that
(i) is same as L1 (⌦, FT , P) ✓ KT (M 1 , M 2 , . . . , M d ) 80 < T < 1 and (ii) is same as
L1 (⌦, FT , P) = KT (M 1 , M 2 , . . . , M d ) 80 < T < 1.
Proofs of (i) () (ii) and (iii) ( (iv) () (v) are exactly the same.
In the proof of (i) ) (v) is also on similar lines, invoking Theorem 10.18 in place of
Theorem 10.4 to conclude that the class of FT measurable random variables that admit
representation is a closed subspace of L1 (⌦, F, P).
For the proof of the last part, namely (i) implies (v), assume (i) is true and let Q 2 MP .
Let Q 2 EP (M 1 , M 2 , . . . , M d ).
T
Fix T < 1 and let ⌘ be any FT measurable bounded random variable. Since L1 (⌦, FT , P) ✓
KT (M 1 , M 2 , . . . , M d ) and F0 is trivial, we can get g j 2 L(M j ) for 1 j d with
d Z
X
P Rt
⌘ =c+
F
j=1 0
T
g j dM j
P Rt
such that Vt = c + dj=1 0 g j dM j is a P-martingale. Let Zt = dj=1 0 gsj 1[0,T ] (s)dMsj .
A
Then Zt = E[(⌘ c) | Ft ] and thus Z is a bounded P-martingale.
Since M 1 , M 2 , . . . , M d are Q-local martingales and g j 2 L(M j ), it follows that Z is
a Q-sigma-martingale. But Z is a bounded process and now invoking Corollary 9.26 we
R
We can now deduce the integral representation property for d-dimensional Brownian
Motion, due to Ito [24] and Clark [10].
Proof. The proof is on the same lines as in the case of 1-dimensional version, Theorem 10.10.
First we note that [W j , W k ] = 0 for j 6= k implies that W is a quasi-elliptic semimartingale
so that we can use Theorem 10.19. We will show that if Q 2 EP (W 1 , W 2 , . . . , W d ) then
Q = P. Once again as in Theorem 10.10, we deduce that for each j W j is a square integrable
martingale with [W j , W j ]Q j j Q
t = t and for j 6= k, [W , W ]t = 0. Thus Levy’s characterisa-
tion Theorem 3.6 implies that W is a d-dimensional Brownian motion on (⌦, F, Q). The
assumption that F is generated by {Wt : t 0} yields P = Q completing the proof
T
d
X d
X
2
i j ij (t, x) ↵ i. (10.3.12)
i,j=1 i=1
Wt =
F
Then as noted earlier, X = (X 1 , X 2 , . . . , X d ) is a quasi-elliptic semimartingale. Moreover, the
condition (10.3.12) implies that (t, x) is invertible and then
Z t
1
(s, Xs )dXs . (10.3.13)
A
0
Thus, Wt is FtX
measurable and as a consequence, FtX = Ft and as a consequence, every
martingale M admits a representation
d Z t
R
X
M t = M0 + g j dX j 8t (10.3.14)
j=1 0
Since (F·X ) = (F· ), g above is also (F·X )-predictable. As a consequence, we also get that
EP (X 1 , X 2 , . . . , X d ) = {P}.
satisfies for j 6= k
[Y j , Y k ]t = 0 8t
and thus is quasi-elliptic. As a result, Theorem 10.19 would hold for (Y 1 , . . . , Y d ) if
X 1 , . . . , X d were local martingales.
We will first show that such a transformation is always possible. In order to achieve
this, we need some auxiliary results.
Lemma 10.22. Let N[d] , O[d] , D[d] be the class of d ⇥ d symmetric non-negative definite
matrices, d ⇥ d Orthogonal matrices and d ⇥ d diagonal matrices. Then There exists a
Borel measurable mapping ✓ : N[d] 7! O[d] ⇥ D[d] such that
✓(C) = (B, D) satisfies C = B T DB.
T
istence of orthogonal B and diagonal D such that C = B T DB. Since for all C 2 N[d] , the
set
{(B, D) 2 O[d] ⇥ D[d] : C = B T DB}
F
is compact, measurable selection theorem (See [19] or or Corollary 5.2.6 of [54]) yields the
existence of Borel measurable ✓.
A
Lemma 10.23. Let D be a -field on a non-empty set and for 1 i, j d, ij be -finite
signed measures on ( , D) such that for all E 2 D, the matrix(( ij (E))) is a symmetric
P
non-negative definite matrix. Let ⇤(E) = di=1 ii (E). Then for 1 i, j d there exists
R
d
a version cij of the Radon-Nikodym derivative d⇤ij such that for all 2 , the matrix
((cij ( ))) is non- negative definite.
d ij
Proof. or 1 i j d let f ij be a version of the Radon-Nikodym derivative and let
D
d✓
f ji = f ij . For rationals r1 , r2 , . . . , rd , let
X
Ar1 ,r2 ,...,rd = { : ri rj f ij ( ) < 0}.
ij
satisfies
[N j , N k ]t = 0 8t. (10.4.2)
Further,
d Z
X t
Mtk = bjk dN j . (10.4.3)
j=1 0
T
Proof. First let us assume that [M k , M k ]t C for all t and 1 k d. Recall that the
predictable -field P is the smallest -field on ⌦ e = [0, 1) ⇥ ⌦ with respect to which all
continuous adapted processes are measurable.
Pd
ij (E) =
Z Z 1
⌦ 0
F
Let signed measures ij on P be defined as follows: for E 2 P, 1 i, j d,
d ij
= cij (10.4.4)
d⇤
and that C = ((cij )) is a non-negative definite matrix (see Lemma 10.23). By construction
|cij | 1. Using Lemma 10.22, we can obtain obtain predictable processes bij , dj such that
D
d
X
bij kj
s bs = ik (10.4.5)
j=1
d
X
bji jk
s bs = ik (10.4.6)
j=1
d
X
bij jl kl
s c s bs =
i
ik ds (10.4.7)
j,l=1
10.4. Continuous Multidimensional Semimartingales 325
Since ((cij i
s )) is non-negative definite, it follows that ds 0. For 1 j d, let N j be
defined by (10.4.1). Using (10.4.6), it follows that
d Z
X t Z tX
d X
d
jk j
b dN = bjk bji
s dM
i
T
0 ⌦ 0 j,l=1
Z d
X
= h bij bkl d jl
¯
⌦ (10.4.9)
=
Z
¯
⌦
h
j,l=1
d
X
j,l=1
F
bij bkl cjl d⇤
A
=0
where the last step follows from (10.4.7). Given a bounded stop time , taking h =
1(0, ] , it follows that h is predictable and thus using (10.4.9) we conclude Taking hs (!) =
R
E[ [N i , N k ] ] = 0.
D
[n],k
Then n " 1 and for each n, M [n],1 , M [n],2 , . . . , M [n],d defined by Mt k
= Mt^ n
satisfy
Let ((b[n],ij )) be the predictable processes obtained for M [n],1 , M [n],2 , . . . , M [n],d . Then
defining
1
X
bij
s = b[n],ij 1( n 1 , n ] (s)
n=1
Remark 10.25. Let M, N be as in Theorem 10.24. Then it follows that M 1 , . . . , M d are local
martingales if and only if N 1 , . . . , N d are local martingales since bik are bounded predictable
processes. Further, 10.4.1, 10.4.3 imply that
E(M 1 , M 2 , . . . , M d ) = E(N 1 , N 2 , . . . , N d ),
EP (M 1 , M 2 , . . . , M d ) = EP (N 1 , N 2 , . . . , N d ),
e P (M 1 , M 2 , . . . , M d ) = E
E e P (N 1 , N 2 , . . . , N d ).
T
In view of Remark 10.25, we have the following result as a direct consequence of The-
orem 10.19.
F
Theorem 10.26. Let M 1 , M 2 , . . . , M d be continuous local martingales on (⌦, F, P). Sup-
pose that F0 is trivial and F = ([t Ft ). Let N 1 , N 2 , . . . , N d be as in Theorem 10.24 so
that (10.4.1),(10.4.2) and (10.4.3) hold. Then the following are equivalent.
A
(i) For every bounded martingale u, 9f j 2 L(N j ), 1 j d such that
Xd Z t
Ut = U0 + fsj dNsj 8t. (10.4.10)
R
j=1 0
e P (M 1 , M 2 , . . . , M d ) = {P}.
(iv) E
(v) EP (M 1 , M 2 , . . . , M d ) = {P}.
In order to discuss the general case of integral representation theorem, we need to extend
the notion of vector stochastic integral. See Jacod [26], Cherny and Shiryaev[7].
(i) | n| 1,
T
(ii) hj n is bounded for all j, n, 1 j d, n 1,
(iii) n ! 0 pointwise
the processes Z n =
Pd
Here is an observation.
j=1
R
hj n dX j
F
converge to 0 in ducp metric.
A
Lemma 10.28. Let X 1 , . . . , X d be semimartingales and h = (h1 , . . . , hd ) be an Rd -valued
predictable process. Let
Xd Z
1
R
h
Y = hj dX j (10.5.1)
|h|
j=1
Pd j )2 1
where |h| = j=1 (h and hj |h| is taken to be 0 by convention when hj = 0. Then
D
Noting that
Z d Z
X
f n dY h = hj n
dX j
j=1
we conclude that h 2 Lv (X 1 , . . . , X d ).
T
w.r.t. X = (X 1 , . . . , X d ). Let
Z t
v 1 d 1 d
C (X , . . . , X ) = {Y : 9g 2 Lv (X , . . . , X ), with Yt = Y0 + hg, dXi 8t < 1}
Note that if Zt =
Rt
0 hh,dXi, then
Z t
F 0
A
[Z, Z]t = |hs |2 d[Y h , Y h ]s
0
d Z
X t
1
= |hs |2 (hj hk )d[X j , X k ]s
|hs |2 s s (10.5.3)
j,k=1 0
R
d Z
X t
= (hjs hks )d[X j , X k ]s .
j,k=1 0
Rt
D
T
h 2 Lv (X 1 , X 2 , . . . , X d ) if and only if f 2 Lv (Y 1 , Y 2 , . . . , Y d ) (10.5.7)
and then Z Z
hh, dXi = hf, dY i. (10.5.8)
As a consequence F
Cv (X 1 , X 2 , . . . , X d ) = Cv (Y 1 , Y 2 , . . . , Y d ). (10.5.9)
A
hj gj
Proof. Since |h| = |g| ,
d Z
X
h hj
Z = dX j
|h|
R
j=1
Xd Z
gj
= dY j
|g|
j=1
Z
D
= dY g
Pd R gj
where Y g = j=1
j
|g| dY .
We observe that an analogue of Theorem 4.39 holds for vector integral as well.
Ztm,n = Zt^⌧
n
m
.
Now the required conclusion, namely (10.5.12) follows from (10.5.11) and Corollary (2.73).
T
R
Exercise 10.34. Show that the mapping (h, X) 7! hh, dXi is linear in h and X.
F
In analogy with Lemma 10.1, here we have the following result, with very similar proof.
Lemma 10.35. Let Y be a semimartingale such that for a sequence of stop times ⌧n " 1,
A
Y n defined by Ytn = Yt^⌧n satisfies
Y n 2 Cv (X 1 , X 2 , . . . , X d ).
Then
R
Y 2 Cv (X 1 , X 2 , . . . , X d ).
Z t
Ytn n
= Y0 + hf n , dXi.
0
Define f j by
1
X
fj = 1(⌧n 1 ,⌧n ]
f n,j .
n=1
With the introduction of vector integral, we can now prove the multidimensional ana-
logue of Lemma 10.3.
Proof. The proof follows the steps in the proof Lemma 10.3. We note that Theorem 5.33
implies that X n is converging to X in Emery topology and as a consequence, for each T ,
[X n X, X n X]T ! 0 in probability as n ! 1, (10.5.13)
[X n , X n ]T ! [X, X]T in probability as n ! 1 (10.5.14)
and
[X n X m, X n X m ]T ! 0 in probability as n, m ! 1.
T
(10.5.15)
By taking a subsequence and relabeling if necessary, we assume that
P([X n X, X n X]k 2 k
)2 k
, 8n, m k. (10.5.16)
1 p
X
[X n X, X n X]t < 1 a.s.
F
Using this estimate and invoking Borel-Cantelli lemma it follows that
A
n=1
Let
d q
X 1 p
X p
Bt = [M j , M j ]t + [X n X, X n X]t + [X, X]t
R
j=1 n=1
p
Using (4.6.22), it follows that [X n , X n ]t Bt and also
p
[X n X m , X n X m ]t Bt (10.5.17)
D
We are going to carry out a orthogonalization as in Theorem 10.24. However, this time
[M j , M k ]
are not continuous and thus we cannot assume them to be locally integrable.
Thus we introduce an equivalent measure Q as follows: let
bn = EP [exp { Bn }],
1
X 1
⇠= exp { Bn }
2n a n
n=1
and Q be the probability measure on (⌦, F) defined by
dQ
= ⇠.
dP
332 Chapter 10. Integral Representation of Martingales
Since EQ [(Bt )k ] < 1 for all t and for all k, and [X n X m , X n X m ]t converges to 0 in P
and hence in Q probability as n, m ! 1, it follows using (10.5.17) that
EQ [ [X n X m, X n X m ]T ] ! 0 as n, m ! 1. (10.5.18)
Likewise
EQ [ [X n X, X n X]T ] ! 0 as n ! 1 (10.5.19)
and
EQ [ [X n , X n ]T ] ! EQ [ [X, X]T ] as n ! 1. (10.5.20)
T
ference. Here we do not have continuity of [M i , M j ] but do have integrability of [M j , M j ]T
under probability measure Q for each j.
Let signed measures on P be defined as follows: for E 2 P, 1 i, j d,
Pd
ij
ij
(E) =
Z Z 1
⌦ 0
F
1E (s, !)d[M i , M j ]s (!)dQ(!).
A
Let ⇤ = j=1 jj . From the properties of quadratic variation [M i , M j ], it follows that
for all E 2 P, the matrix (( ij (E))) is non-negative definite. Further, ij is absolutely
continuous w.r.t. ⇤ 8i, j. It follows that we can get predictable processes cij such that
d ij
R
= cij (10.5.22)
d⇤
and that C = ((cij )) is a non-negative definite matrix (see Lemma 10.23). By construction
|cij | 1. Using Lemma 10.22, we can obtain obtain predictable processes bij , dj such that
D
d
X
bij kj
s bs = ik (10.5.23)
j=1
d
X
bji jk
s bs = ik (10.5.24)
j=1
d
X
bij jl kl
s c s bs =
i
ik ds (10.5.25)
j,l=1
10.5. General Multidimensional case 333
Since ((cij
s )) is non-negative definite, it follows that ds
i 0. For 1 j d, let N j be
defined by
Xd Z t
j
Nt = bji dM i (10.5.26)
i=1 0
T
Z 1 Z Z 1 d
X
i k
EQ [ hs d[N , N ]s ] = hs bij kl j l
s bs d[M , M ]s dQ(!)
0 ⌦ 0 j,l=1
Z
=
Z
¯
⌦
h
d
X
j,l=1
d
X
F
bij bkl d jl
(10.5.28)
A
= h bij bkl cjl d⇤
¯
⌦ j,l=1
=0
R
where the last step follows from (10.5.25). As a consequence, for bounded predictable hi ,
for T < 1
Xd Z T d Z T
X
i k i k
EQ [ hs hs d[N , N ]s ] = EQ [ (hks )2 d[N k , N k ]s ] (10.5.29)
D
i,k=1 0 k=1 0
Let us observe that (10.5.29) holds for any predictable processes {hi : 1 i d} provided
the right hand side if finite: we can first note that it holds for h̃i = hi 1{|h|c} where
P
|h| = di=1 |hi | and then let c " 1.
Let us define
Xd
n,k
g = f n,j bkj (10.5.30)
j=1
n
Pd
and let = k=1 (|g n,k | + |f n,k |).Then note that
Z Z
n n
hg , dN i = dW n
334 Chapter 10. Integral Representation of Martingales
where
d Z Z
X
n g n,k
W = n
dN k .
k=1
Note that
d Z
d X
X
n 1
W = n
f n,j bkj dN k
k=1 j=1
d X
X d Z
d X
1
= n
f n,j bkj bkl
s dM
l
(10.5.31)
k=1 j=1 l=1
d Z
X 1
= n
f n,j dM j
j=1
and hence
T
Z Z
hf n , dM i = n
dW n .
Thus we have Z Z
Thus
d Z
n
hg , dN i =
F hf n , dM i (10.5.32)
A
X T
n m n m
[X X ,X X ]T = (g n,j g m,j )(g n,k g m,k )d[N j , N k ]. (10.5.33)
j,k=1 0
d Z
X T
n m n m
EQ [ [X X ,X X ] T ] = EQ [ (gsn,k gsm,k )2 d[N k , N k ]s ]
k=1 0
(10.5.34)
XZ
d
D
n,k m,k 2
= (g g ) 1⌦⇥[0,T ] d kk
k=1
Since left hand side in (10.5.34) converges to 0 (see (10.5.18)), invoking completeness of
¯ P, kk ), we can get predictable processes g k such that
L2 (⌦,
Z
(g n,k g k )2 1⌦⇥[0,T ] d kk ! 0.
As a consequence
X d Z T
(g n,j g j )(g n,k g k )d[N j , N k ] ! 0 in Q - probability as n ! 1, (10.5.35)
j,k=1 0
10.5. General Multidimensional case 335
and as a consequence
d Z
X t
[X, X]t = (g j )(g k )d[N j , N k ] (10.5.38)
j,k=1 0
T
Let us define bounded predictable processes j and predictable process hn , h and a P-
martingale X as follows:
d
X
hs = 1 + |gsi | (10.5.39)
j
s
F
=
i=1
gsj
hs
(10.5.40)
A
d Z
X t
j j
Zt = s dNs (10.5.41)
j=1 0
R
Then
X d Z
d X t
j ji i
Zt = s bs dM (10.5.42)
i=1 j=1 0
D
Since j , bji are predictable and bounded by 1, it follows that Z is a P-martingale. Let us
note that
Z t d X
X d
h2s d[Z, Z]s = h2s js ks d[N j , N k ]s
0 j=1 k=1
(10.5.43)
d X
X d
= gsk gsj d[N j , N k ]s .
j=1 k=1
We now forget Q and focus only on P. Since X is a martingale, we can get stop times
n " 1 such that E[ [X, X] n ] < 1 and thus using (10.5.44), we conclude that h 2 Lm (Z).
1
Rt Rt
Defining Yt = X0 + 0 hdZ, we note that Yt = X0 + 0 < g, dN >. Thus Y is a local
martingale and further
Z t X
d Z T
n n
[X Y, X Y ]t = (g n,j g j )(g n,k g k )d[N j , N k ]. (10.5.45)
0 j,k=1 0
and thus using (10.5.35) and the observation that Q and P are equivalent, we conclude
[X n Y, X n Y ]t ! o in P probability as n ! 1. (10.5.46)
Since for all n 1
p p p
[X Y, X Y ]t [X n X, X n X]t + [X n Y, X n Y ]t (10.5.47)
using (10.5.34) and (10.5.13), we conclude
T
[X Y, X Y ]t = 0 8t. (10.5.48)
Since X, Y are local martingales and X0 = Y0 , (10.5.48) implies Xt = Yt for all t. To
complete the proof, we will show that
Xt = X0 +
0
hf, dM i
F 0
A
d
X
i
f = g k bki
k=1
Pd
(|f j | |g j |).
R
Xd Xd Z t
1
= f i bji dN j
i=1 j=1 0
d X
X d Z
d X t
1
= g k bki bji dN j
i=1 k=1 j=1 0
X d Z
d X t
1
= gk jk dN
j
k=1 j=1 0
d Z
X t
1
= g j dN j
j=1 0
10.5. General Multidimensional case 337
Thus Z Z Z
hf, dM i = dW = hg, dN i.
Thus Z t
Xt = X0 + hf, dM i.
0
T
Theorem 10.37. Let M 1 , . . . , M d be local martingales and T < 1. Then KvT (M 1 , . . . , M d )
is a closed linear sub-space of L1 (⌦, FT , P).
F
Now using Theorem 10.18 instead of Theorem 10.37, we can obtain this result on the
integral representation property for general multidimensional local martingales - rest of the
argument is essentially same as in the proof of Theorem 10.19, but we will give it here.
A
Theorem 10.38. Let M 1 , M 2 , . . . , M d be local martingales on (⌦, F, P). Suppose that F0
is trivial and F = ([t Ft ). Then the following are equivalent.
Z t
St = S0 + hf, dM i a.s. 8t. (10.5.50)
0
e P (M 1 , M 2 , . . . , M d ) = {P}.
(iv) E
(v) EP (M 1 , M 2 , . . . , M d ) = {P}.
Proof. It can be seen that (i) is same as L1 (⌦, FT , P) ✓ KvT (M 1 , . . . , M d ) 80 < T < 1
and (ii) is same as L1 (⌦, FT , P) = KvT (M 1 , . . . , M d ) 80 < T < 1. As seen in Theorem
10.18 KvT (M 1 , . . . , M d ) is a closed subspace of L1 (⌦, FT , P). Since L1 (⌦, FT , P) is dense
in L1 (⌦, FT , P), it follows that (i) and (ii) are equivalent.
338 Chapter 10. Integral Representation of Martingales
On the other hand, suppose (iv) holds and suppose Q1 , Q2 2 E(M 1 , M 2 , . . . , M d ) and
P = ↵Q1 + (1 ↵)Q2 . It follows that Q1 , Q2 are absolutely continuous w.r.t. P and hence
Q1 , Q2 2 E e P (M 1 , M 2 , . . . , M d ). In view of (iv), Q1 = Q2 = P and thus P is an extreme
point of E(M 1 , M 2 , . . . , M d ) and so (iii) is true. Thus (iv) ) (iii).
Since {P} ✓ EP (M 1 , M 2 , . . . , M d ) ✓ E e P (M 1 , M 2 , . . . , M d ), it follows that (iv) implies
(v).
If (v) is true and Q 2 E e P (M 1 , M 2 , . . . , M d ), then Q1 = 1 (Q+P) 2 EP (M 1 , M 2 , . . . , M d ).
2
Then (v) implies Q1 = P and hence Q = P. Thus (v) ) (iv) holds.
To see that (iii) ) (ii) let P be an extreme point of E(M 1 , M 2 , . . . , M d ) but (ii) is
not true. Then KvT (M 1 , . . . , M d ) is a closed proper subspace of L1 (⌦, F, P) and by the
Hahn-Banach Theorem, there exists ⇠ 2 L1 (⌦, FT , P), P(⇠ 6= 0) > 0 such that
Z
✓⇠ dP = 0 8✓ 2 KvT (M 1 , . . . , M d ).
T
Then for c 2 R, we have
Z Z
✓(1 + c⇠)dP = ✓dP 8✓ 2 KvT (M 1 , . . . , M d ). (10.5.51)
EQ [M⌧j,n j,n j
^T ] = EP [M⌧ ^T ] = M0 (10.5.53)
On the other hand,
EQ [M⌧j,n j,n
_T ] = EP [⌘M⌧ _T ]
= EP [EP [⌘M⌧j,n
_T | FT ]]
= EP [⌘EP [M⌧j,n
_T | FT ]]
(10.5.54)
= EP [⌘MTj,n ]
= EQ [MTj,n ]
= M0j .
10.6. Integral Representation w.r.t. Sigma-Martingales 339
where we have used the facts that ⌘ is FT measurable, M j,n is a P-martingale and (10.5.53).
Now noting that M⌧j,n = M⌧j,n j,n
^T + M⌧ _T MTj,n , we conclude
EQ [M⌧j,n ] = EQ [M⌧j,n j,n
^T ] + EQ [M⌧ _T ] EQ [MTj,n ] = M0j .
Thus Mtj,n = Mt^ j
n
is a Q-martingale for every n so that M j is a Q local martingale and
thus Q 2 E(M 1 , M 2 , . . . , M d ). Similarly, if Q̃ is the measure with density ⌘ = (1 c⇠), we
can prove that Q̃ 2 E(M 1 , M 2 , . . . , M d ). Here P = 12 (Q + Q̃) and P 6= Q (since P(⇠ 6= 0) >
0). This contradicts the assumption that P is an extreme point of E(M 1 , M 2 , . . . , M d ).
Thus (iii) ) (ii).
To complete the proof, we need to show that (i) implies (v). Suppose (i) is true and let
Q 2 EP (M 1 , M 2 , . . . , M d ). Fix T < 1 and let ⌘ be any FT measurable bounded random
variable. Since L1 (⌦, FT , P) ✓ KT (M 1 , M 2 , . . . , M d ) and F0 is trivial, we can get for
1 j d g j 2 L(M j ) with
Z T
T
⌘ =c+ hg, dM i
0
Rt Rt
such that 0 hg, dM i is a martingale. Let hs = gs 1[0,T ] (s) and Zt = 0 hh, dM i. It follows
F
that Zt = EP [(⌘ c) | Ft ] and since ⌘ is bounded, it follows that Z is bounded. As noted
earlier, since P and Q are equivalent, the stochastic integrals under P and Q are identical.
R
Under Q, M 1 , M 2 , . . . , M d being local martingales and Z = hh, dM i is a local martingale.
A
Since it is also bounded, we conclude invoking Corollary 9.18 that Z is also a martingale
under Q. Thus, EQ [ZT ] = 0 = EP [ZT ] and thus using ⌘ = c + ZT we get EQ [⌘] = c = EP [⌘].
Since this holds for all FT measurable bounded random variables ⌘, we conclude Q and P
R
R
(i) N j = dX j is a martingale for each j.
(iii) EP (N 1 , N 2 , . . . , N d ) ✓ EP (X 1 , X 2 , . . . , X d ).
Proof. We have seen in Lemma 9.22 that we can choose (0, 1) valued predictable processes
j 2 L(X j ) such that M j =
R j
dX j are martingales. Let = min( 1 , . . . , d ). Let
j = j is bounded by 1. Then N j =
R R
j . Note that dX j = j dM j and then using
T
(g 1 , g 2 , . . . , g d ) 2 Lv (X 1 , X 2 , . . . , X d ) and then
Z Z
hg, dN i = h g, dXi.
(iv)
Part (ii) above along with Lemma 10.36 yields the following.
D
The proof of the next result is on the lines of corresponding results given in previous
sections. The proof closely follows the proof of Theorem 10.38.
e (X 1 , X 2 , . . . , X d ) = {P}.
(iv) E
T
P
(v) EP (X 1 , X 2 , . . . , X d ) = {P}.
Proof. Once again it can be seen that (i) is same as L1 (⌦, FT , P) ✓ KvT (X 1 , . . . , X d )
F
80 < T < 1 and (ii) is same as L1 (⌦, FT , P) = KvT (X 1 , . . . , X d ) 80 < T < 1. As seen in
Theorem 10.18 KvT (X 1 , . . . , X d ) is a closed subspace of L1 (⌦, FT , P). Since L1 (⌦, FT , P)
A
is dense in L1 (⌦, FT , P), it follows that (i) and (ii) are equivalent. The proofs of (iv) )
(iii), (iv) implies (v) and (v) ) (iv) are exactly the same as that given in Theorem 10.38.
To see that (iii) ) (ii) let P be an extreme point of E(X 1 , X 2 , . . . , X d ). Let and N j
be as in Lemma 10.39. Part (iv) in Lemma 10.39 now implies that P be an extreme point of
R
To complete the proof, we will show that (i) implies (v). This is exactly as in Theorem
10.38. Suppose (i) is true and let Q 2 EP (X 1 , X 2 , . . . , X d ). Fix T < 1 and let ⌘ be any
FT measurable bounded random variable. Since L1 (⌦, FT , P) ✓ KT (X 1 , X 2 , . . . , X d ) and
F0 is trivial, we can get for 1 j d g j 2 L(X j ) with
Z T
⌘ =c+ hg, dXi
0
Rt Rt
such that 0 hg, dXi is a martingale. Let hs = gs 1[0,T ] (s) and Zt = 0 hh, dXi. It follows
that Zt = EP [(⌘ c) | Ft ] and since ⌘ is bounded, it follows that Z is bounded. As noted
earlier, since P and Q are equivalent, the stochastic integrals under P and Q are identical.
342 Chapter 10. Integral Representation of Martingales
R
Under Q, X 1 , X 2 , . . . , X d are sigma-martingales and thus Z = hh, dM i is also a sigma-
martingale. Since it is also bounded, we conclude invoking Corollary 9.26 that Z is also
a martingale under Q. Thus, EQ [ZT ] = 0 = EP [ZT ] and thus using ⌘ = c + ZT we get
EQ [⌘] = c = EP [⌘]. Since this holds for all FT measurable bounded random variables ⌘,
we conclude Q and P agree on FT . In view of the assumption F = ([t Ft ), we get Q = P
proving (v). This completes the proof.
This result has strong connections to mathematical finance and in particular to the
theory of asset pricing. We will give a brief background in the next section.
T
lier [1] studied the question of option pricing in his Ph. D. thesis. Here he had modelled
the stock price movement as a Brownian Motion. This was before Einstein used Brownian
Motion in the context of physics and movement of particles. Samuelson, Merton worked
F
extensively on this question [53], [47]. The paper by Black - Scholes brought the connec-
tion to the forefront. The papers by Harrison and Pliska around 1980 built the formal
connection between mathematical finance and stochastic calculus [21, 22]. The fundamen-
A
tal papers by Kreps [43], Yan [58], Stricker [55] laid the foundation for the so-called First
Fundamental Theorem of Asset Pricing. The final version of this result is due to Delbaen
and Schachermayer [11], [12]. Also see [52], [42], [29], [31], [2].
R
We will give a brief account of the framework. We consider a market with d stocks,
modelled as stochastic processes X 1 , X 2 , . . . , X d , assumed to be processes with r.c.l.l. paths.
The market is assumed to be ideal where there are no transaction costs and rate of interest
D
where k are stop times and ajk are F k measurable bounded random variables. For such
a trading strategy, the value function (representing gain or loss from the strategy) is given
by
d m
X X1 j j
Vf (f ) = ak (S k+1 ^t S j k ^t ). (10.7.2)
j=1 k=0
Rt
When S 1 , . . . S j are semimartingales, then we see that Vt (f ) = 0 hf, dSi.
An simple trading strategy f is said to be an arbitrage opportunity if for some T (i)
P(VT (f ) 0) = 1 and (ii) P(VT (f ) > 0) > 0. One of the economic principles is that such
a strategy cannot exist in a market in equilibrium, for if it existed, all investors will follow
the strategy as it gives an investor a shot at making money without taking any risk, thus
disturbing the equilibrium. This is referred to as the No Arbitrage principle or simply NA.
If each S j is a martingale, then Vt (f ) is a martingale for every simple strategy f and
T
thus E[VT (f )] = 0 and thus NA holds. It can be seen that NA is true even when each
S j is martingale under a probability measure Q that is equivalent to P. The converse to
this statement is not true. However, it was recognized that if one rules out approximate
F
arbitrage (in an appropriate sense) then indeed the converse is true. We will not trace
the history of this line of thought (see references given above for the same) but give three
results on this theme. The following result is Theorem 7.2 in [11].
A
Theorem 10.43. Suppose the processes S 1 , S 2 . . . S d are locally bounded and that for any
sequence of simple strategies f n 2 Sd such that 0 < T < 1, the condition
1
R
P(VT (f n ) ) = 1, 8n 1 (10.7.3)
n
implies that for all ✏ > 0
P(|VT (f n )| ✏) ! 0 as n ! 1. (10.7.4)
D
The condition (10.7.3) ) (10.7.4) is essentially ruling out approximate arbitrage and
has been called NFLVR - No Free Lunch with Vanishing Risk by Delbaen - Schachermayer.
Thus we now assume that S 1 , S 2 . . . S d are semimartingales. Lv (X 1 , . . . , X d ) is taken
as the class of trading strategies and for f 2 Lv (X 1 , . . . , X d ), the value process for the
Rt
trading strategy f is defined to be Vt (f ) = 0 hf, dXi. A trading strategy f is said to be
admissible if for some constant K, one has
Z t
P( hf, dXi K 8t) = 1 (10.7.5)
0
344 Chapter 10. Integral Representation of Martingales
and Z t
hf, dXi converges in probability (to say V (f )) as t ! 1. (10.7.6)
0
The following theorem for one dimensional case was proven in [11] (Corollary 1.2). For the
multidimensional case see Theorem 8.2.1 in [13]. This also follows from the Theorem 10.45
T
martingale on (⌦, F, Q).
Here is the final version of the (first) Fundamental Theorem of Asset Pricing - Theorem
14.1.1 in [12]. Also see [29], who independently proved the result.
Then there exists a probability measure Q equivalent to P such that each S j is a sigma-
martingale on (⌦, F, Q).
D
across the world for close to a century. It was in the context of pricing of options that
Bachelier had introduced in 1900 Brownian motion as a model for stock prices.
Suppose that ⇠ (FT measurable random variable) is a contingent claim, x 2 R and
f 2 Lv (S 1 , S 2 . . . S d ) is a trading strategy with
Z T
x+ < f, dS >= ⇠ a.s. (10.7.11)
0
Even if ⇠ is not o↵ered for trade, an investor can always replicate it with an initial invest-
ment x following the strategy f . If (10.7.11) holds, (x, f ) is called replicating strategy. In
such a case, the price p of the contingent claim (assuming that the market is in equilibrium)
must be equal to x. For if p > x, an investor could sell one the contingent claim at p, keep
aside p x, invest x and follow the strategy f . At time T the portfolio is worth exactly
what the investor has to pay for the contingent claim. Thus the investor has made a profit
T
of (p x) without any risk, in other words, it is an arbitrage opportunity. The possibility
p < x can be ruled out likewise, this time the investor buys a contingent claim at p and
follows strategy ( s, f ).
F
Thus if (10.7.11) holds, in other words, a replicating strategy exists for a contingent
claim ⇠, the price of the contingent claim equals the initial investment needed for the
strategy.
A
The market consisting of (discounted) stocks S 1 , S 2 , . . . , S d is said to be complete if for
all bounded FT measurable random variables ⇠, there exists x 2 R and f 2 Lv (S 1 , S 2 , . . . , S d )
such that for some K < 1,
Z t
R
Proof. Let Q 2 EP (S 1 , S 2 , . . . , S d ). Suppose the market is complete. Fix T > 0 and let
⇠ 2 L1 (⌦, FT , Q). Consider the contingent claim ⇠. Using completeness of market, obtain
x, f satisfying (10.7.12) and (10.7.13). Under Q, S 1 , S 2 , . . . , S d being sigma-martingales
Rt
Nt = 0 < f, dS > for t T and Nt = Nt^T is also a sigma-martingale. Being bounded (in
346 Chapter 10. Integral Representation of Martingales
T
F
A
R
D
Chapter 11
In chapter 7, we saw that using random time change, any continuous semimartingale can
T
be transformed into a amenable semimartingale, and then one can have a growth estimate
on the stochastic integral similar to the one satisfied by integrals w.r.t. Brownian motion.
When it comes to r.c.l.l. semimartingales, this is impossible in view of the jumps. Here
we are faced with a difficulty as the stochastic integral is essentially created via an L2
F
estimate while the integral w.r.t. a process with finite variation is essentially defined as an
L1 object - as in Lebesgue-Stieltjes integral. The problem is compounded by the fact that
A
not every semimartingale need be locally integrable.
Metivier-Pellaumail obtained an inequality that makes all semimartingales amenable to
the L2 treatment. Indeed, P. A. Meyer in a private correspondence had drawn our attention
to the Metivier-Pellaumail inequality when he had seen the random change technique in
R
[33] - both have an e↵ect of making every semimartingale amenable to the L2 theory. As
in earlier chapters, we fix a filtration (F⇧ ) on a complete probability space (⌦, F, P) and
we assume that F0 contains all P-null sets in F.
D
347
348 Chapter 11. Dominating Process of a Semimartingale
Let us examine this in a special case, where H is the -field generated by a countable
partition {Hn : n 1} of ⌦. Let En = Hn \ A and Fn = Hn \ Ac . Let pn = P(En ) and
qn = P(Fn )
an = E[⇠ 1En ], bn = E[⇠ 1Fn ].
Since E[⇠ | H] = 0, it follows that an + bn = 0. It follows that for any 2 L, E[ 1Fn ] = bn
since E[ | H] = 0 and 1A = ⇠ 1A . Since there is no other restriction on , it is clear that
T
in this case, the minimum is attained when is a constant on each Fn , equal to 0 if pn = 0
or qn = 0 and equal to qbnn when qn > 0. Thus let N 0 = {n : pn > 0, qn > 0} and
X bn
= ⇠ 1A + 1 Fn
]
F
n2N
E[ 2
].
A
We would like to get a description of as well as E[ 2 ] in terms of ⇠, H and A. For this,
let G = (H, A) and ⌘ = E[⇠ | G]. Then
X an X bn
⌘= 1 En + 1 Fn
R
0
pn q
0 n
n2N n2N
and
= ⇠ 1A + ⌘ 1Ac .
D
Thus
X b2
2 n
E[ ] = E[⇠ 2 1A ] + .
q
0 n
n2N
Now it can be checked that (using a2n = b2n )
X 1 a2 b2
E[⌘ 2 | H] = ( n + n ) 1Hn
pn + q n pn qn
n2N0
1 pn + q n
= b2n ( ) 1Hn
pn + q n pn q n
b2
= n 1Hn
pn q n
11.1. An optimization result 349
and hence
2
E[ ] = E[⇠ 2 1A ] + E[1A E[⌘ 2 | H]].
We will now show that the result is true in general. The calculations done above give
us a clue as to the answer.
Theorem 11.1. Let H be a sub- -field of F and let ⇠ be a random variable with E[⇠ 2 ] < 1
such that E[⇠ | H] = 0. Let A 2 F and
T
L = { : E[ | H] = 0, 1A = ⇠ 1A }.
Then for 2L
Further,
2
E[ ] = E[⇠ 2 1A ] + E[1A E[⌘ 2 | H]]. (11.1.1)
R
as E[ | H] = 0 and E[ | H] = 0. This proves the first part. For the second part, let
↵ = E[1A | H]. Since E[⌘ | H] = 0, we have
⇣↵ + ✓(1 ↵) = 0. (11.1.3)
Thus
E[⌘ 2 | H] =E[⇣ 2 1A + ✓2 1Ac | H]
=⇣ 2 ↵ + ✓2 (1 ↵)
and hence
E[1A E[⌘ 2 | H]] = E[E[⌘ 2 | H]E[1A | H]]
= E[⇣ 2 ↵ + ✓2 (1 ↵)E[1A | H]]
2 2 2 (11.1.4)
= E[⇣ ↵ + ✓ (1 ↵)↵]
= E[✓2 (1 ↵)2 + ✓2 (1 ↵)↵]
T
= E[✓2 (1 ↵)]
where we have used (11.1.3). On the other hand
E[⌘ 2 1Ac ] =E[✓2 1Ac ]
(11.1.7)
=E[⇠ 2 1A ] + E[1A E[⌘ 2 | H]].
where the last step follows from (11.1.6).
D
Let us observe that ⌘ = E[⇠ | G] and hence by Jensen’s inequality, one has.
E[1A E[⌘ 2 | H]] E[1A E[⇠ 2 | H]].
Thus the previous result leads to
Theorem 11.2. Let ⇠ be a random variable such that E[⇠ | H] = 0 and E[⇠ 2 ] < 1. For
A 2 F, there exists a random variable such that
(i) 1A = ⇠ 1A .
(ii) E[ | H] = 0.
Proof. Suffices to prove it for bounded as the general case follows by using (11.2.1) for
^ m and taking limit over m. So we assume T . Now we can assume that Mt = Mt^T .
Let {⌧k : k 1} be predictable stop times as in Theorem 8.68. Let ⇠k = ( M )⌧k ,
P
U = ⇠k 1[⌧k ,1) , Z = k2F U k , N = M Z. Since ⌧k is predictable, U k is a martingale.
k
We have seen in Theorem 8.68 that hN, N i is a continuous increasing process. Moreover
X
⇠k2 1[⌧k ,1) (t)
T
[Z, Z]t = (11.2.2)
k2F
and
X
hZ, Zit = E[⇠k2 | F⌧k ]1[⌧k ,1) (t) (11.2.3)
k
k2F
For k 2 F , let Gk = (F⌧k , { > ⌧k }), ⌘k = E[⇠k | Gk ] and
= ⇠k 1{ >⌧k }
F + ⌘k 1 { ⌧k } .
A
Since E[⇠k | F⌧k ] = 0 and F⌧k ✓ Gk , it follows that E[ k | F⌧k ] = 0 and hence
Vtk = k 1[⌧k ,1) (t)
[V k , V j ] = 0, hV k , V j i = 0. (11.2.4)
Moreover, for all k 2 F , Theorem 11.2 implies
D
2 2
E[( k) ] E[1{ >⌧k } (⇠k + E[⇠k2 | F⌧k ])] (11.2.5)
This of course also gives
2
E[( k) ] 2E[(⇠k )2 ].
P
Let Y = k2F V k . If F is finite, clearly, Y is a martingale. In case F is infinite, the
series converges and Y is a martingale as in the proof of Theorem 8.68. Noting that
k 1{ >⌧k } = ⇠k 1{ >⌧k } , we have
E[sup|Mt |2 ] = E[sup|Xt |2 ]
T
t< t<
E[sup|Xt |2 ] (11.2.9)
t
4E[[X, X] ]
F
Finally, using (11.2.8) along with [N, Y ] = 0, we get
E[[X, X] ] = E[[N, N ] ] + E[[Y, Y ] ]
A
(11.2.10)
E[hN, N i ] + E[[Z, Z] ] + E[hZ, Zi ]
Since hN, N i is continuous, we have hN, N i = hN, N i . As seen in Theorem 8.68
hN, Zi = 0 and hence we get
R
R
and B = f dA. Then Y = N + B, N is a locally square integrable martingale and B 2 V.
Further,
Z t
[N, N ]t = fs2 d[M, M ]s , (11.3.1)
0
Z t
hN, N it = fs2 dhM, M is (11.3.2)
0
T
0 0
and hence Z Z
t ⌧
E[sup| f dA|2 ] E[ |A|⌧ |fs2 |d|A|s ] (11.3.4)
t<⌧ 0 0
F
We can combine the two estimates (11.3.3) and (11.3.4) as follows: let
The significant point about the estimate (11.3.3) is that given any locally bounded pre-
dictable f and any semimartingale X, we can get a sequence of stop times ⌧n increasing
to 1 such that the expression on right hand in (11.3.6) is finite. This may not be the case
D
which indeed can be obtained without any need for the Metivier-Pellaumail inequality.
The process V introduced above (modulo a constant) was called a control process of the
semimartingale X.
While the control process was used to successfully deal with stochastic di↵erential
equations driven by semimartingales, the notion is not natural as even if the semimartingale
is small in Emery topology, the control process may not be small. Further, if V is control
process for a semimartingale X, for a constant c, cV may not be a control process for cX.
354 Chapter 11. Dominating Process of a Semimartingale
T
is a dominating process.
Remark 11.6. One of the reason that we did not define U given by (11.3.8) for some
decomposition X = M + A as the dominating process is that now we can have a common
F
dominating process for finitely many semimartingales.
0t<
p p
2 2([M, M ]t + hM, M it )1/2 + 2|A|t Ut
and as a result
8([M, M ]t + hM, M it ) + 2|A|2t Ut2 (11.3.10)
On the other hand, for any stop time , we have
E[ sup |Xt |2 ] 2E[ sup |Mt |2 ] + 2E[ sup |At |2 ] (11.3.11)
0t< 0t< 0t<
At the same time |At | |A|t (where |A| is the total variation of A). As a result
E[sup|At |2 ] E[ |A|2 ]. (11.3.13)
t<
Combining (11.3.11)-(11.3.13), we get
E[ sup |Xt |2 ] E[8([M, M ] + hM, M i ) + 2|A|2 ]. (11.3.14)
0t<
Now the required estimate (11.3.9) follows from (11.3.10) and (11.3.14).
Ideally, we would have liked to have a notion of dominating process such that if U 1 , U 2
be dominating processes of semimartingale X 1 , X 2 respectively, then V = U 1 + U 2 is a
dominating process for Y = X 1 + X 2 . While this is not quite true, we will show that Y
admits a dominating process W such that Wt Vt . To prove this, we need the following
result.
T
Lemma 11.8. For M, N be locally square integrable martingales, let q(M, N ) = [M, N ] +
hM, N i. Then, for all t
1 1 1
(q(M + N, M + N )t ) 2 (q(M, M )t ) 2 + (q(N, N )t ) 2 . (11.3.15)
(q(
k
X
i
M ,
k
X
i
M )t )
F
Also, if M i are locally square integrable martingales, i = 1, 2, . . . , k then
1
2
k
X 1
(q(M i , M i )t ) 2 . (11.3.16)
A
i=1 i=1 i=1
Proof. Since M, N 7! [M, N ] and M, N 7! hM, N i are bilinear maps, it follows that same
is true of M, N 7! q(M, N ). Further, q(M, M )t 0. Then proceeding as in the proof of
R
Let N = M 1 + M 2 , B = A1 + A2 and
p p
Vt = 2 2([N, N ]t + hN, N it )1/2 + 2|B|t .
Then V is a dominating process for Y and to complete the proof of the first part, we need
to show (11.3.18). Clearly,
T
|B|t |Ai |t + |Ai |t 8t. (11.3.19)
We now move to exploring the connection of dominating process with stochastic inte-
gral. Here are a sequence of auxiliary results that we need later.
D
Remark 11.11. Essentially the same argument as in the Lemma 11.8 (see also (4.6.21)) gives
us for locally square integrable martingales N j , j = 1, 2, . . . k
1 1
hN i , N j it (hN i , N i it ) 2 (hN j , N j it ) 2 (11.3.21)
1 1
[N i , N j ]t ([N i , N i ]t ) 2 ([N j , N j ]t ) 2 (11.3.22)
and as a consequence
k
X k
X k X
X k
1 1
(h N i, N j it ) 2 = ( hN i , N j it ) 2
i=1 j=1 i=1 j=1
k X
X k
1 1
( (hN i , N i it ) 2 (hN j , N j it ) 2 (11.3.23)
i=1 j=1
k
X 1
= (hN i , N i it ) 2
T
i=1
and similarly,
k
X k
X k
X
1 1
i j
([ N , N ]t ) ( 2 ([N i , N i ]t ) 2 (11.3.24)
i=1 j=1
F
For a locally bounded predictable process f and an increasing process V , let ✓t (f, V )
be defined by
i=1
A
Z t Z t
2 2 12
✓t (f, V ) = ( |fs | dVs ) + |fs |dVs . (11.3.25)
0 0
Note that
R
and also
Z t Z t Z t
2
|fs | dVs2 ✓t2 (f, V ) 2( |fs | 2
dVs2 + Vt |fs |2 dVs ) (11.3.28)
0 0 0
The following result gives interplay of this notion of dominating process with that of
stochastic integral.
T
2 2 1/2
= ( |f | dC ) + 2 |f |d|A|
0 0
and U
p
2|A| 2 V+ implies
|f |2 dC 2
p Z t
0
Z t
F
|f |2 dU 2
0
(11.3.31)
A
2 |f |d|A| |f |d|U | (11.3.32)
0 0
Combining (11.3.30)-(11.3.32), we get
Z t Z t
R
2 2 1/2
Vt ( |f | dU ) + |f |d|U | = ✓t (f, U ). (11.3.33)
0 0
This proves (11.3.29).
D
Putting together Theorems 11.7 and 11.12 we now obtain an estimate on the growth
of a stochastic integral.
Proof. As noted earlier, (11.3.34) follows from Theorems 11.7 and 11.12 and then (11.3.35)
follows from (11.3.26).
Remark 11.14. Before proceeding, we would like to stress that given a locally bounded
predictable f and an increasing process V 2 V+ , one can always get stop times ⌧m increasing
to 1 such that
E[✓⌧2m (f, V )] < 1
and thus the estimate (11.3.34) is meaningful for any locally bounded predictable f and any
semimartingale.
We can now add to the Bichteler-Dellacherie-Meyer-Mokobodzky Theorem. Each of
the six equivalent conditions in Theorem 5.78 is equivalent to existence of a dominating
process. We will list here only two out of the six.
T
Theorem 11.15. Let X be an r.c.l.l. (F· ) adapted process. Let JX be defined by (4.2.1)-
(4.2.2). Then the following are equivalent.
(i) X is a weak stochastic integrator, i.e. if f n 2 S, f n ! 0 uniformly, then JX (f n )t ! 0
in probability 8t < 1. F
(ii) X is a semimartingale, i.e. X admits a decomposition X = M + A where M is a local
A
martingale and A is a process with finite variation paths.
(iii) There exists an increasing adapted process V such that for all stop times ⌧ and for
all f 2 S, one has
Z Z
R
Proof. We have already shown that (i) and (ii) are equivalent. Using Theorem 11.13, it
D
follows that (ii) implies (iii) follows. To see that (iii) implies (i), note that given any
adapted increasing process V , s < 1 and ✏ > 0, we can get a stop time ⌧ such that V⌧ is
bounded and P(s < ⌧ ) (1 ✏). See Remark 11.14. Now the result follows from Theorem
11.13 and the estimate (11.3.27).
T
1
X
n
E[ 2 [ |Vnk | ^ 1]] 1.
k=1
Vtk
Thus (noting 0)
1
X
k=1
2 n F
[Vnk ^ 1] < 1 a.s.
A
and hence for every t < 1
1
X
Ut = [ Vtk ] < 1 a.s.
k=1
R
Now let ⌧m be stop times increasing to 1 such that U⌧m m. In view of (11.3.9), we
have for every k
E[ sup |Xt Yt |2 ] E[(V⌧km )2 ]. (11.4.1)
0t<⌧m
D
Now (V⌧km )2 converges to zero in probability and is dominated by m2 and thus the right
hand since in (11.4.1) converges to zero. Thus for every m
E[ sup |Xt Yt | 2 ] = 0
0t<⌧m
showing that X = Y .
Proof. Given " > 0, get U, V 2 V+ such that U is a dominating process for X Y , V is a
dominating process for Y Z and
T
dsm (X, Y ) + dsm (Y, Z) + 2".
Since " is arbitrary, this proves 11.4.2.
F
Using the previous two results, we conclude that dsm is a metric on the space of
semimartingales. We will show that this metric also induces the Emery topology. The first
step is to show that the space of semimartingales is complete in this metric.
A
Theorem 11.19. Let X n be a sequence of semimartingales that is Cauchy in dsm metric.
Then there exists a semimartingale X such that dsm (X n , X) ! 0.
R
ducp (V n , 0) dsm (X n , X n 1
)+2 n
2.2 n
. (11.4.4)
T
by cj ) for each j. Let
X1
Dtm = (hM n , M n it )1/2 .
n=m+1
Dm
Then j
D j
P
N k N m = kn=m+1 M n and hence using (11.3.23) we get
k
X k
X
k m k m 1/2 n
(hN N ,N N it ) = (h M , M n it )1/2
D
n=m+1 n=m+1
k
X (11.4.10)
(hM n , M n it )1/2
n=m+1
Dtm
and thus in view of (11.4.9), we have
lim E[ sup hN k N m, N k N m i j ] = 0. (11.4.11)
m!1 k>m
In turn, using the Doob’s maximal inequality, we get
lim E[ sup sup |Ntk Ntm |2 ] = 0. (11.4.12)
m!1 k>m t j
11.4. Alternate metric for Emery topology 363
T
B B m and thus U m defined by
p p
U m = 2 2([N N m , N N m ]t + hN N m , N N m it )1/2 + 2|B B m |t
Utm
F1
X
n=m+1
Vtn .
A
P1
In view of (11.4.7), it follows that n=m+1 Vtn converges to zero almost surely (as m ! 1).
Thus, Utm converges to 0 in probability, and so dsm (X, X m ) converges to 0 completing the
proof.
R
1
X
dsm (X n , X) < 1. (11.4.15)
n=1
Then dem (X n , X) ! 0.
ducp (V n , 0) dsm (X n , X) + 2 n
.
T
Z t Z t
n 1
P(sup | f dX f dX| ] ⌘) " (11.4.19)
t< j 0 0 2
Recall that choice of n0 is independent of f and thus (note: S1 is the set of bounded
predictable processes bounded by 1) for n n0 we have
Z t
sup P(sup| f dX n
Z t
f dX| ] ⌘)
F
A
f 2S1 tT 0 0
Z t Z t
n
sup P(sup | f dX f dX| ] ⌘) + P( j T)
f 2S1 t< j 0 0
R
"
As noted in the proof of Lemma 4.100, this shows dem (X n , X) ! 0.
Now we are in a position to prove that dsm and dem give rise to the same topology.
D
Proof. Let X n , X be such that dsm (X n , X) ! 0. We will first show that dem (X n , X) ! 0.
k
Take any subsequence {nk } and let Y k = X n . Since dsm (Y k , X) ! 0, we can choose a
m
subsequence {k m } such that Z m = Y k satisfies
1
X
dsm (Z m , X) ! 0.
k=1
11.4. Alternate metric for Emery topology 365
Now Lemma 11.20 yields that dem (Z m , X) ! 0.. Thus the sequence {X n } satisfies the
property that given any sub-sequence, there exists a further subsequence that converges to
X in dem metric. Hence dem (X n , X) ! 0.
From the definition of the metrics dem and dsm , it follows that the space of semi-
martingales is a linear topological space under each. Further, we have shown that the
space is complete under each of the metric. The identity mapping being continuous is then
a homeomorphism in view of the Open mapping theorem.
The following result gives a technique to prove almost sure convergence of stochastic
integrals.
T
and
1
X
(Vtn )2 < 1 8t < 1. (11.4.21)
n=1
X
[ sup |ftn ft |2 ] < 1 a.s.
n=1 0tT
F
Suppose f n , f locally bounded predictable f , such that for all T < 1
1
(11.4.22)
A
Then for all T < 1
1
X Z t Z t
n n
[ sup | f dX f dX|2 ] < 1 a.s. (11.4.23)
n=1 0tT 0 0
R
and as a consequence
Z t Z t
n n
lim [ sup | f dX f dX|2 ] = 0 a.s. (11.4.24)
n!1 0tT 0 0
D
For j 1, let
⌧j = inf{t 0 : Vt j or Vt j}.
Now noting that sup0s<⌧j |fs | j,
sup |fsn | sup |fs | + sup |fsn fs |
0s<⌧j 0s<⌧j 0s<⌧j
2j,
366 Chapter 11. Dominating Process of a Semimartingale
the fact that V is a dominating process for X n X as well as for X and that V⌧j j we
have (using (11.3.35))
Z t Z t
n n
E[ sup | f dX f n dX|2 ] 2j 2 E[(V⌧nj )2 ]
0t<⌧j 0 0
Z t Z t
E[ sup | f n dX f dX|2 ] 2j 2 E[ sup |fsn fs |2 ]
0t<⌧j 0 0 0s<⌧j
and hence
1
X Z t Z t
E[ sup | f n dX n f dX|2 ]
n=1 0t<⌧j 0 0
Z t Z t Z t Z t
n n n 2 n
2E[ sup | f dX f dX| ] + 2E[ sup | f dX f dX|2 ]
0t<⌧j 0 0 0t<⌧j 0 0
1
X (11.4.25)
T
4j 2 [E[(V⌧nj )2 ] + E[ sup |fsn fs |2 ]]
n=1 0s<⌧j
4j 2 E[2V⌧2j ]
8j 4 .
This shows that for all j
1
X Z t
F Z t
A
n n
sup | f dX f dX|2 < 1 a.s.
n=1 0t<⌧j 0 0
In this Chapter, we will consider stochastic di↵erential equations as in section 7.3 where
T
the driving semimartingale need not be continuous.
We will consider the SDE (7.3.1), where b would be as in the section 7.3 but Y would
be an r.c.l.l. semimartingale. We will continue to use the conventions used in that section
F
on matrix-vector-valued processes and stochastic integrals.
Here we will use the Metivier-Pellaumail inequality and the notion of dominating pro-
cess introduced earlier and we will see that invoking these, the proofs of existence and
A
uniqueness is essentially same as in the case of SDE’s driven by Brownian Motion. In
section 7.3 we had used random time change to achieve the same.
We begin with an analogue of the Gronwall’s inequality, a key step in study of di↵er-
ential equations.
R
We will obtain an analogue of Gronwall’s inequality that would be useful in dealing with
the stochastic di↵erential equations driven by semimartingales in the next section. The
first one is from Metivier [MeP] and the second one is essentially based on the same idea.
Theorem 12.1. Let A, B 2 V+ (increasing processes with A0 0, B0 0) and a stop
time ⌧ be such that B⌧ M . Suppose that for all stop times ⌧
Z
E[A ] a + E[ A dB]. (12.1.1)
[0, )
P[↵] j.
For ↵ > 0 let C(↵) = j=0 ↵ Then we have
E[A⌧ ] 2aC(2 M ) (12.1.2)
367
368 Chapter 12. SDE driven by r.c.l.l. Semimartingales
1
Proof. Let us define Vt = M Bt . Then V⌧ 1 and we have
Z
E[A ] a + M E[ A dV ] (12.1.3)
[0, )
For integers i 1, let ⌧i = inf{t 0 : At i or At i} ^ ⌧ . Note that A⌧i i, ⌧i " ⌧
and since A is increasing process, it follows that
E[A⌧i ] ! E[A⌧ ] as i ! 1. (12.1.4)
1
Fix i and = 2 M and let k be defined inductively by 0 = 0 and for k 0
k+1 = inf{t > k : (Vt V k) or (Vt V k) } ^ ⌧i (12.1.5)
If k+1 < ⌧i , then (V k+1
V k) and hence
N = ⌧i , for N = [2 M ] + 1. (12.1.6)
Moreover, for all k, (V k+1 V k) .
T
For k 0, let Zk = A k and k = E[Zk ]. Then
Z Z
E[Zk+1 ] a + M E[ As dVs ] + M E[( As dVs ]
[0, k) [ k , k+1 )
a + M E[Zk ] + M E[Zk+1 ]
1
a + M E[Zk ] + E[Zk+1 ]
2
F
A
Thus we have for k 1 (note that a priori we know that A k+1
i and hence k+1 is
finite)
k+1 2a + 2 M k.
R
Likewise, we can conclude that 1 = E[Z1 ] 2a. Then by induction it follows that
k
X
k+1 2a(1 + (2 M )j ).
j=1
D
1
Proof. Let us define Vt = M Bt . Then V⌧ 1 and we have
E[(A )2 ] a + M 2 E[(✓ (A , V ))2 ] (12.1.9)
For integers i 1, let ⌧i = inf{t 0 : At i or At i} ^ ⌧ . Note that A⌧i i, ⌧i " ⌧
and since A is increasing process, it follows that
E[A2⌧i ] ! E[A2⌧ ] as i ! 1. (12.1.10)
1
Fix = 10 M 2
and let k be defined inductively by 0 = 0 and for k 0
T
[0,t) [0,t)
and using the inequality (12.1.9), we have (writing Ut = Vt2 for convenience)
E[(A k+1
)2 ] a + M 2 E[(✓ k+1
(A , V ))2 ]
Z Z
For k 0, let Zk = (A k
a + 2bE[(
and )2 k
[0, k+1 )
F
A2t
= E[Zk ] = E[(A
dUt +
k
[0, k+1 )
)2 ].
A2t dVt )].
Since k ⌧i , k
(12.1.12)
is finite for
A
each k. Using (12.1.12) for k 1
Z Z
2
k+1 a + 2 M E[( A2t dUt + A2t dVt )]
[0, k) [0, )
Z Zk
R
a + 2 M 2 E[Zk (U k
U0 ) + Zk (V k
V0 )]
D
+ 2 M 2 E[Zk+1 (U k+1
U k ) + Zk+1 (V k+1
V k )]
Using Ut 1, Vt 1 for all t < ⌧ and (V k+1
V k) and
(U k+1
U k ) = (V k+1
V k )(V k+1
+ V k) 2
the inequality (12.1.13) yields
k+1 a + 2 M 2 (2 k ) + 2 M 2 (3 k+1 )
6 (12.1.14)
a + 4 M 2 k + k+1
10
Thus, for k 1
10
k+1 a + 10 M 2 k 3a + 10 M 2 .
4
370 Chapter 12. SDE driven by r.c.l.l. Semimartingales
Same argument as above also yields 1 3a. Thus by induction it follows that
k
X
k+1 3a[ (10 M 2 )j ]. (12.1.15)
j=0
T
Let Y 1 , Y 2 , . . . Y m be r.c.l.l. semimartingales w.r.t. the filtration (F⇧ ). Here we will consider
an SDE
dUt = b(t, ·, U )dYt , t 0, U0 = ⇠0 (12.2.1)
F
where the functional b is given as follows. Recall that Dd = D([0, 1), Rd ). Let B(Dd ) be
the smallest -field on Dd under which the coordinate mappings are measurable. Let
A
a : [0, 1) ⇥ ⌦ ⇥ Dd ! L(d, m) (12.2.2)
and suppose there there is an increasing r.c.l.l. adapted process K such that for all
↵, ↵1 , ↵2 2 Dd ,
sup ka(s, !, ↵)k Kt (!) sup (1 + |↵(s)|) (12.2.5)
0st 0st
(i) For an r.c.l.l. (F⇧ ) adapted process V , Z defined by Zt = b(t, ·, V ) ( i.e. Zt (!) =
a(t, !, V (!)) ) is an r.c.l.l. (F⇧ ) adapted process.
Proof. For part (i), let us define a process V t by Vst = Vs^t . Note that in view of (12.2.6),
Zt = a(t, ·, V t ). The fact that ! 7! V t (!) is Ft measurable along with (12.2.3) implies
that Zt is also Ft measurable. For part (ii), when ⌧ is a simple stop time, (12.2.8) follows
from (12.2.3). For a general bounded stop time ⌧ , the conclusion (12.2.8) follows by
approximating ⌧ from above by simple stop times and using right continuity of a(t, !, ⇣).
For a general stop time ⌧ , (12.2.8) follows by approximating ⌧ by ⌧ ^ n.
T
Recall that we had introduced matrix-vector-valued processes and stochastic integral
R
F
f dX where f, X are matrix-vector- valued while dealing with SDEs driven by continuous
semimartingales. We will continue to use the same convention and notions. As in the case
of continuous semimartingales, here too, an r.c.l.l. (Rd -valued) adapted process U is said
A
to be a solution to the equation (12.2.1) if
Z t
Ut = U0 + b(s, ·, U )dYs (12.2.9)
0
R
i.e. for 1 j d,
m Z
X t
Utj = U0j + bjk (s, ·, U )dYsk
D
k=1 0+
Proof.
Z s
E[ sup | f dX|2 ]
0s ^T 0+
d
X m Z
X s
= E[ sup | fjk dX k |2 ]
j=1 0s ^T k=1 0+
d X
X m Z s
m E[ sup | fjk dX k |2 ]
j=1 k=1 0s ^T 0+
d X
X m
m E[✓⌧2 (fjk , V )].]
j=1 k=1
T
We are now in a position to prove uniqueness of solution to the SDE (12.2.1). The
F
proof is essentially the same as for SDE driven by Brownian motion or by a continuous
semimartingale. Here we use the growth estimate (7.3.11) in place of (3.4.4) or (7.2.3)
for a continuous semimartingale satisfying (7.2.2). The technique of time change used for
A
continuous semimartingale is replaced here by the notion of dominating process and the
estimate in Theorem 12.2 replacing Gronwall’s lemma - Lemma 3.23.
Lemma 12.5. Let Y 1 , Y 2 , . . . Y m be r.c.l.l. semimartingales w.r.t. the filtration (F⇧ ). Let
R
a satisfy (12.2.2) - (12.2.6) and let b be defined by (12.2.7). Suppose H and G be r.c.l.l.
adapted processes and let X and Z satisfy
Z t
D
Let V be a (common) dominating process for Y j , 1 j m and let be a stop time such
that V M and K . Then
Using the Lipschitz condition (12.2.6), the fact that K , it follows that for s <
Thus writing As = sup0ts |Xt Zt |, we get for any stop time ⌧ , using (12.2.10),
Now V M and Theorem 12.2 together imply the required estimate (12.2.13).
T
This result immediately leads to:
F
Theorem 12.6. Let Y 1 , Y 2 , . . . Y m be r.c.l.l. semimartingales w.r.t. the filtration (F⇧ ). Let
a satisfy (12.2.2) = (12.2.6) and let b be defined by (12.2.7). Let H be an adapted r.c.l.l.
process. Suppose X and Z satisfies
A
Z t
X t = Ht + b(s, ·, X)dYs (12.2.14)
0+
Z t
R
Then X = Z.
D
n = inf{t 0 : Ut n or Ut n}.
E[ sup |Xt Zt |2 ] = 0
0s< n
Theorem 12.7. Let Y 1 , Y 2 , . . . Y m be r.c.l.l. semimartingales w.r.t. the filtration (F⇧ ). Let
a satisfy (12.2.2) = (12.2.6) and let b be defined by (12.2.7). Let H be an adapted r.c.l.l.
process. Suppose ⌧ is a stop time and X and Z satisfies
Z t^⌧
Xt^⌧ = Ht^⌧ + b(s, ·, X)dYs (12.2.16)
0+
Z t^⌧
Zt^⌧ = Ht^⌧ + b(s, ·, Z)dYs (12.2.17)
0+
Then
P(Xt^⌧ = Zt^⌧ 8t) = 1. (12.2.18)
T
n = inf{t 0 : Ut n or Ut n} ^ ⌧.
E[ sup |Xt Zt |2 ] = 0
for n
this.
0s< n
F
1. This proves P(Xt = Zt 8t < ⌧ ) = 1. The required result (12.2.18) follows from
A
Having proven uniqueness of solution to the SDE (12.2.14), we now move onto proving
existence of solution to the equation. We will show this by showing that Picard’s successive
R
approximation method converges to a solution of the equation. The proof will be very
similar to the proof in the Brownian motion case.
Theorem 12.8. Let Y 1 , Y 2 , . . . Y m be r.c.l.l. semimartingales w.r.t. the filtration (F⇧ ). Let
D
where K is as in conditions (12.2.6) and (12.2.5) and let j be the stop times defined by
j = inf{t 0 : Ut j or Ut j}.
For s < j,
[n 1]
k(b(s, ·, X [n] ) b(s, ·, Xs[n 1]
)k jAs
T
and thus using (12.2.20) along with the estimate (12.2.10), we get for any stop time ⌧ j,
for n 1 (using (11.3.28) for the last step)
[n] [n+1] [n]
E[(A⌧ )2 ] = E[sup|Xt Xt | 2 ]
t<⌧
Z t
= E[sup| (b(s, ·, X [n] ) b(s, ·, Xs[n
t<⌧ 0
dm j E[✓⌧2 (A[n 1] , V )
2 2
F 1]
)dYs |2 ]
A
(12.2.21)
Z Z
2 2 [n 1] 2 2 [n 1] 2
2dm j E[ (As ) dVs + Vt (As ) dVs ]
[0,⌧ ) [0,⌧ )
Z
[n 1] 2
R
2 3
4dm j E[ (As ) dDs ]
[0,⌧ )
P1 n (A[n] )2 ,
where Ds = Vs2 + Vs . Hence writing Bt = n=0 4 t we thus get for any stop time
⌧ j
D
Z
[0]
E[B⌧ ] E[A⌧ ] + 16dm2 j 3 E[ Bs dU ] (12.2.22)
[0,⌧ )
[0]
Also, recalling that |Xt | = |Ht | j for all t < j, we have
[0] [1] [0]
E[A⌧ ] = E[sup|Xt Xt | 2 ]
t<⌧
Z t
= E[sup| b(s, ·, X [0] )dY |2
t<⌧ 0
2 2
dm E[✓⌧ (j, V )]
dm2 j 4 .
376 Chapter 12. SDE driven by r.c.l.l. Semimartingales
and then using the version of Gronwall inequality given in Theorem 12.1, we conclude
E[B j ] < 1.
T
n=0 n=0
The relation (12.2.23) implies
1
X
k[ sup |Xs[n+1] Xs[n] | ] k2 < 1 (12.2.24)
as well as
! 0 as n tends to 1.
P1 [n+1] [n]
Let N = [1
j=1 {! : n=1 sups< j
|Xs (!) Xs (!)| = 1}. Then by (12.2.24) P(N ) = 0
D
[n]
and for ! 62 N , Xs (!) converges uniformly on [0, j (!)) for every j < 1. So let us define
X as follows: 8
<lim [n] c
n!1 Xt (!) if ! 2 N
Xt (!) =
:0 if ! 2 N.
Since P(N ) = 0, it follows that
Recalling the Lipschitz condition (12.2.6) and the fact that K j j, we have
! 0 as n ! 1.
This along with (12.2.19) and (12.2.26) yields that
Z t
X t = Ht + (b(s, ·, X)dYs ,
0
T
in other words X is a solution to the equation (12.2.14).
For t < 1, ↵ 2 Dd and 2 Dr , let ↵t (s) = ↵(t ^ s) and t (s) = (t ^ s) and we assume
that f satisfies
t
f (t, , ↵) = f (t, , ↵t ), 8 2 Dr , ↵ 2 Dd , 0 t < 1. (12.3.5)
T
As in section 6.2, we will now obtain a mapping
: Dd ⇥ Dr ⇥ Dm 7! D([0, 1), Rd )
such that for adapted r.c.l.l. process H, G (Rm , Rr -valued respectively) and an r.c.l.l. semi-
martingale Y ,
X=
F
(H, G, Y )
A
yields the unique solution to the SDE
Z t
X t = Ht + g(s, G, X)dY. (12.3.9)
0
R
inductively for n 1. Let (0) (⌘, , ↵)(s) = ⌘ for all s 0 and having defined (0) , (1) , . . . ,
D
let
(n) (n) (n 1) (n 1) (n) n
tj+1 = inf{t tj :k (⌘, , ↵)(s) (⌘, , ↵)(tj )k 2
(n 1) (n 1) (n) n
or k (⌘, , ↵)(s ) (⌘, , ↵)(tj )k 2 }
12.3. Pathwise Formula for solution to an SDE 379
(n 1) (⌘, (n)
(since , ↵) is an r.c.l.l. function, tj " 1 as j " 1) and
1
X
(n) (n 1) (n) (n)
(⌘, , ↵)(s) = ⌘ + (⌘, , ↵)(s)(↵(t ^ tj+1 ) ↵(t ^ tj )).
j=0
T
X(!) = (H(!), G(!), Y (!)). (12.3.11)
Note the an ! path of X has been defined directly in terms of the ! paths of G, H, Y via
the functional . We will prove
(0)
F
Theorem 12.9. X defined by (12.3.11) is the (unique) solution to the SDE (12.3.9).
Proof. Let Zt = H0 . The processes Z (n) are defined by induction on n. Assuming that
A
Z (0) , . . . , Z (n 1) have been defined, we now define Z (n) : Fix n.
(n) (n) (n)
Let ⌧0 = 0 and let {⌧j : j 1} be defined inductively as follows: if ⌧j = 1 then
(n) (n)
⌧j+1 = 1 and if ⌧j < 1 then
R
(n)
(12.3.12)
or kf (s , G, Z (n 1)
) f (⌧j , G, Z (n 1)
)k 2 n
}.
D
(n)
Since the process s 7! f (s, G, Z (n 1) ) is an adapted r.c.l.l. process, it follows that each ⌧j
(n) (n) (n) (n)
is a stop time and limj"1 ⌧j = 1. Let Z0 = H0 and for j 0, ⌧j < t ⌧j+1 let
(n) (n) (n)
Zt =Z (n) + f (⌧j , G, Z (n 1)
)(Yt Y⌧ (n) ).
⌧j j
Equivalently,
1
X
(n) (n)
Zt = Ht + f (⌧j , G, Z (n 1)
)(Yt^⌧ (n) Yt^⌧ (n) ) (12.3.13)
j+1 j
j=0
Thus to complete the proof, suffices to show that Z (n) converges to a solution Z of the
SDE (12.3.9). Uniqueness would then imply that Z = X.
For n 1, let us define W n and S n by
1
X
(n) (n)
St = f (⌧j , G, Z (n 1)
)1[⌧ (n) ,⌧ (n) ) (t) (12.3.14)
j j+1
j=0
Z t
(n)
Wt = Ht + f (s , G, Z (n 1)
)dYs . (12.3.15)
0
T
0
(n)
Noting that by definition of {⌧j : j 1},
kSt f (t, G, Z (n
F 1)
)k 2
where Kt (!) = C(t, G(!)) and C is as in the Lipschitz and growth conditions (12.3.6) -
(12.3.7), Ds = Vs2 + Vs and let j be the stop times defined by
D
j = inf{t 0 : Ut j or Ut j}.
Note that j " 1 as j " 1. Using (12.3.15), (12.3.16), (12.3.17) and the fact that V j j
along with the fact that V is a common dominating process for Y j , 1 j m we get
(n) (n)
E[ sup |Wt Zt |2 ] dm2 j 2 2 2n
(12.3.18)
0s< j
For n 0 let
[n]
At = sup |Zs[n+1] Zs[n] |.
0st
12.4. Euler-Peano approximations 381
For any stop time ⌧ j, for n 1 (using (11.3.28) for the last step)
[n] [n+1] [n]
E[(A⌧ )2 ] = E[sup|Zt Zt |2 ]
t<⌧
(n+1) (n+1) 2 (n) (n)
3E[ sup |Wt Zt | ] + 3E[ sup |Wt Zt |2 ]
0s< j 0s< j
[n+1] [n]
+ 3E[sup|Wt Wt | 2 ]
t<⌧
Z t
2 2 2n
6dm j 2 + 3E[sup| (g(s, G, Z [n] ) g(s, G, Zs[n 1]
)dYs |2 ] (12.3.19)
t<⌧ 0
6dm2 j 2 2 2n
+ 3dm2 j 2 E[✓⌧2 (A[n 1] , V )
Z Z
[n 1] 2 [n 1] 2
6dm2 j 2 2 2n + 6dm2 j 2 E[ (As ) dVs2 + Vt (As ) dVs ]
[0,⌧ ) [0,⌧ )
Z
[n 1] 2
6dm2 j 2 2 2n + 12dm2 j 3 E[ (As ) dDs ]
T
[0,⌧ )
P1 n (A[n] )2 ,
Hence writing Bt = n=0 2 we thus get for any stop time ⌧ j
t
1
X Z
E[B⌧
[0]
] E[A⌧ ] + 2n 6dm2 j 2 2 2n + 24dm2 j 3 E[
n=0
2 2 4 2 2 2 3
E[(B⌧ ) dm j + 6dm j + 12dm j E[ (Bs )2 dDs ]. (12.3.21)
[0,⌧ )
Now proceeding exactly as in the proof of Theorem 12.8, we can conclude that Z (n) con-
verges to a solution Z of the equation (12.3.9). Once again from the definition, it follows
D
computational point of view as compared to the formula (12.3.10). These results were
obtained in [39]. The formulation given here is taken from [40]
We will essentially consider the framework as in section 12.2. Let Y 1 , Y 2 , . . . Y m be
r.c.l.l. semimartingales w.r.t. the filtration (F⇧ ), H be an r.c.l.l. adapted process. Consider
the SDE Z 1
U t = Ht + b(t, ·, U )dYt , (12.4.1)
0+
where the functional b is given as follows. Let
a : [0, 1) ⇥ ⌦ ⇥ Dd ! L(d, m) (12.4.2)
be such that for all t 2 [0, 1)
(!, ↵) 7! a(t, !, ↵) is Ft ⌦ B(Dd ) measurable, (12.4.3)
for all (!, ↵) 2 ⌦ ⇥ Dd ,
T
t 7! a(t, !, ↵) is an r.c.l.l. mapping (12.4.4)
and suppose there there is an increasing r.c.l.l. adapted process K such that for all
↵, ↵1 , ↵2 2 Dd ,
0st
sup ka(s, !, ↵2 )
F
sup ka(s, !, ↵)k Kt (!) sup (1 + |↵(s)|)
0st
↵1 (s)|. (12.4.6)
A
0st 0s<t
Note the slight di↵erence between (12.2.6) and (12.4.6)- here the sup on the right hand
side is over 0 s < t.
Let b : [0, 1) ⇥ ⌦ ⇥ Dd ! L(d, m) be given by
R
⌧i+1 = inf{t > ⌧i : |Ait | " or |Ait | " or |Bti | " or |Bti | "}
and 8
<W i for t < ⌧i+1
t
Wti+1 =
: W i + Ai for t ⌧i+1 .
t ⌧i+1
Thus, W i+1 is a process that has jumps at ⌧1 , ..., ⌧i+1 and is constant on the intervals
[0, ⌧1 ), · · · , [⌧j , ⌧j+1 ), · · · [⌧i , ⌧i+1 ), [⌧i+1 , 1). Also W i and W i+1 agree on [0, ⌧i+1 ) by defini-
tion. For k 1 define Z k by Z0k = H0 and
8
<W k + H H⌧i + a(⌧i , ·, W k )(Yt Y⌧i ) for ⌧i t < ⌧i+1
⌧i t
Ztk =
:Z k = W k for t ⌧ .
⌧k ⌧k k
As a consequence, we have
k 1
X
Ztk = Ht^⌧k + a(⌧i , ·, W k )(Yt^⌧i+1
T
Y⌧i ) (12.4.9)
i=0
Noting that for i k, Wtk = Wti
for t ⌧i and thus a(⌧i , ·, W k ) = a(⌧i , ·, W i ) and Ztk = Zti
for t ⌧i . Thus Ztk Wtk = Ait
for ⌧i t < ⌧i+1 . As a consequence, we have (using the
definition of the sequence {⌧i }),
|Ztk Wtk | ".
Since W i and W i+1 agree on [0, ⌧i+1 ), we also have
F (12.4.10)
A
a(⌧i+1 ·, W i+1 ) = a(⌧i+1 ·, W i ). (12.4.11)
We will show later that ⌧i " 1. However, a priori it is not clear that this is so. The next
R
i=0
Then S is an r.c.l.l. adapted process.
Proof. Let = limi!1 ⌧i . Fix T < (!). For t 2 [0, T ], St (!) is a finite sum of r.c.l.l.
functions and hence is r.c.l.l. while if (!) < 1, then for t (!), St (!) = 0. Here we
have used the fact that if ⌧i < 1 then ⌧i < ⌧i+1 and as a consequence, ⌧i < for all i.
Thus only remains to show that when (!) < 1, the left limit of S at (!) exists. Fix
! such that a = (!) < 1. In this case, U⌧i (!) (!) ! Ua (!) and if tn " a with tn < a,
then Stn (!) is a subsequence of U⌧i (!) (!) and hence left limit of S· (!) at a exists and equals
Ua (!).
384 Chapter 12. SDE driven by r.c.l.l. Semimartingales
Lemma 12.10 ensures that I(!, ↵) is an r.c.l.l. function. We now define mapping J that
maps r.c.l.l. adapted processes into r.c.l.l. adapted processes by
or equivalently,
1
X
J (U ) = U⌧i 1[⌧i ,⌧i+1 ) . (12.4.13)
i=0
J (Z k ) = W k . (12.4.14)
T
Let us define ã : [0, 1) ⇥ ⌦ ⇥ Dd ! L(d, m) as follows:
F
Easy to check that ã satisfies (12.2.2)-(12.2.6) and hence defining
i=0
and so
1
X
k
b̃(t, ·, Z ) = a(⌧i , ·, W k )1(⌧i ,⌧i+1 ] (t).
i=0
Lemma 12.11.
lim ⌧i = 1 a.s.
i!1
Proof. The definition of the sequence {⌧j } gives that if ⌧i+1 < 1 then at least one of
|a(⌧i+1 , ·, W i ) a(⌧i , ·, W i )|, |a(⌧i+1 , ·, W i ) a(⌧i , ·, W i )|, |Z⌧i+1
i+1
Z⌧i+1
i
| and |Z⌧i+1
i+1
Z⌧i+1
i
|,
i
exceeds ". Since W , J (Z) and W i+1 agree on [0, ⌧i+1 ), using (12.4.6), it follows that
a(⌧i+1 , ·, W i ) = a(⌧i+1 , ·, W i+1 ) = a(⌧i+1 , ·, J (Z))
and likewise
a(⌧i+1 , ·, W i ) = a(⌧i+1 , ·, W i+1 ) = a(⌧i+1 , ·, J (Z)).
Thus if ⌧i+1 < 1, at least one of |Z⌧i+1 Z⌧i |, |Z⌧i+1 Z⌧i |,
|a(⌧i+1 , ·, J (Z)) a(⌧i , ·, J (Z))|, |a(⌧i+1 , ·, J (Z)) a(⌧i , ·, J (Z))| exceeds ". So if limi ⌧i =
< 1, either Zt or a(t, ·, J (Z)) would fail to have left limit at and this would contradict
T
the fact that J (Z) has r.c.l.l. paths or the assumptions on a. Thus = 1.
1
X
kb(s, ·, Z) b̃(s, ·, Z)k = ka(s , ·, Z) a(⌧i , ·, W i+1 )k1(⌧i ,⌧i+1 ] (s)
i=0
1
X
D
(Ks + 1)"
386 Chapter 12. SDE driven by r.c.l.l. Semimartingales
Lemma 12.12. Let X be the solution to (12.4.1) and Z ⌘ Z " be as defined in preceding
paragraphs (satisfying (12.4.19)) for fixed ". Let V be a common dominating process for
Y j , 1 j m. Let Ut = Vt + Kt (where K appears in condition (12.4.6) on a) and for
j 1 let j be defined by
j = inf{t 0 : Ut j or Ut j}.
T
Z t
2
E[|A⌧ | ] E[ sup | (b(s, ·, X) b̃(s, ·, Z))dY |2 ]
0t< 0+
Z t
2E[ sup |
0t<
+ E[ sup |
0t<
0+
Z
F
(b(s, ·, X)
t
(b(s, ·, Z)
b(s, ·, Z))dY |2 ]
b̃(s, ·, Z))dY |2 ]
(12.4.21)
A
0+
If X n denotes the approximation Z " for " = 2 n , then the estimate (12.4.22) along
with Lemma 6.1 imply that X n converges to the solution X of the SDE (12.4.1). This in
turn helps us obtain a pathwise formula for solution to the SDE (12.4.1).
We will essentially consider the framework from section 12.3 and obtain a pathwise
formula involving a single limit rather than an iterative limit. Let f, g : [0, 1) ⇥ Dr ⇥ Dd 7!
L(d, m) be such that
For t < 1, ↵ 2 Dd and 2 Dr , let ↵t (s) = ↵(t ^ s) and t (s) = (t ^ s) and we assume
that f satisfies
t
f (t, , ↵) = f (t, , ↵t ), 8 2 Dr , ↵ 2 Dd , 0 t < 1. (12.4.26)
T
0s<t
For n 1, we define
F
t ! C(t, ) is r.c.l.l. (12.4.29)
A
˜ (n) : Dd ⇥ Dr ⇥ Dm 7! D([0, 1), Rd )
themselves functions of n, (⌘, , ↵), which are fixed for now and we will suppress writing
these as a function) if tj = 1 then tj+1 = 1 and if tj < 1 then
D
and 8
<⇠ i for t < ti+1
t
⇠ti+1 =
:⇠ i + ↵ i for t ti+1 .
t ti+1
Thus, ⇠ i+1 is a function that has jumps at t1 , ..., ti+1 and is constant on the intervals
[0, t1 ), . . . , [tj , tj+1 ), . . . [ti , ti+1 ), [ti+1 , 1). Also ⇠ i and ⇠ i+1 agree on [0, ti+1 ) by definition.
388 Chapter 12. SDE driven by r.c.l.l. Semimartingales
We finally define
1
X
˜ (n) (⌘, , ↵)(t) = ⌘t + f (t ^ ti , , ⇠ i )(↵t^ti+1 ↵t^ti ) (12.4.31)
i=0
˜ : Dd ⇥ Dr ⇥ Dm 7! D([0, 1), Rd )
has been defined without any reference to a probability measure or any semimartingale.
T
As in Theorem 12.9, this also yields a pathwise formula. This one is preferable from
computation point of view as here, in order to construct nth approximation, we do not
need the (n 1)th approximation.
F
Theorem 12.13. Let f, g satisfy conditions (12.4.23)-(12.4.29). Let Y be a semimartingale
w.r.t. a filtration (F⇧ ) and let H, G be r.c.l.l. (F⇧ ) adapted processes taking values in Rd ,
Rr respectively. Let ˜ be as defined in (12.4.32) and let
A
X̃ = ˜ (H, G, Y ).
Z t
X̃t = Ht + g(t, G, X̃)dY. (12.4.33)
0+
˜ (n) (H, G, Y ) = X n
where X n is the 2 n approximation constructed in this section earlier. It now follows that
˜ (H, G, Y ) = X̃
and that for L(d, d)-valued semimartingales X, Y let [X, Y ] = ([X, Y ]ij ) be the L(d, d)-
valued process defined by
T
d
X
[X, Y ]ij
t = [X ik , Y kj ].
k=1
Exercise 12.14. Formulate and prove analogues of Theorem 12.6, Theorem 12.8 and The-
orem 12.13 for the equation (12.5.1).
Exercise 12.15. Let X be an L(d, d)-valued semimartingale with X(0) = 0 and let I denote
D
The solutions Y, Z are denoted respectively by e(X) and e0 (X) and are the left and
right exponentials of X.
390 Chapter 12. SDE driven by r.c.l.l. Semimartingales
Exercise 12.16. Let X be an L(d, d)-valued semimartingale with X(0) = 0 and let Y = e(X)
and Z = e0 (X). Show that
For a matrix A 2 L(d, d) we will denote (only in this section) the Hilbert-Schmidt norm
of A by kAk. The following facts are standard. The norm is defined as
d
X
2
kAk = (aij )2 .
i,j=1
If kAk < 1 then B = (I + A) belongs to L0 (d). Further, for kAk ↵ < 1, one has
1 1
kAk2 .
T
k(I + A) I + Ak (12.5.4)
1 ↵
Exercise 12.17. For an L(d, d)-valued semimartingale X, show that
X
k( X)s k2 T race([X, X]t ).
0<st F
Exercise 12.18. Let X be an L(d, d)-valued semimartingale with X(0) = 0 such that (I +
A
X) is L0 (d)-valued. Then
P 1}
(i) Show that Wt = 0<st [{(I + X) I + ( X) + ( X)2 ] is well defined.
1} ( X)2 .
R
X + U + [X, U ] = 0 (12.5.5)
D
and
e(U )e0 (X) = I. (12.5.7)
(v) Let Y = e(X) and Z = e0 (X). Show that Y , Y , Z and Z are L0 (d)-valued.
Hint: For (i), separate jumps bigger than half- these are finitely many. For the rest of the
jumps, use estimate (12.5.4). For (iv) use Integration by parts formula, (7.5.1).
12.5. Matrix-valued Semimartingales 391
For a L(d, d)-valued semimartingale Y such that Y0 = I and such that Y and Y are
L0 (d)-valued, let
Z t
log(Y )t = (Y ) 1 dY
0+
and Z t
0 1
log (Y ) = (dY )(Y ) .
0+
The next exercise is to show that e and log are inverses of each other. We will say that a
matrix valued process is a local martingale (or a process with finite variation) if each of its
components is so.
Exercise 12.19. Let X be an L(d, d)-valued semimartingale with X(0) = 0 such that (I +
X) is L0 (d)-valued and let Y be a L(d, d)-valued semimartingale such that Y and Y are
L0 (d)-valued. Then show that
T
(i) e(log(Y )) = Y, e0 (log0 (Y )) = Y.
Exercise 12.20. Let X i be L(d, d)-valued semimartingale with X i (0) = 0 such that (I +
R
X i ) is L0 (d)-valued, for i = 1, 2. Let Y = e(X 2 ) and U 1 = Y (dX 1 )(Y ) 1 . Then show
D
that
e(X 1 + X 2 + [X 1 , X 2 ]) = e(U 1 )e(X 2 ) (12.5.8)
The formula (12.5.8) has an important consequence. Given a L(d, d)-valued semi-
martingale Y such that Y and Y are L0 (d)-valued, let X = e(Y ). If we can write
X = M + A + [M, A] such that M 2 Mloc and A 2 V with (I + M ), (I + M ) are
L0 (d)-valued then it would follow that
Y = NB
where N = e(M ) 2 Mloc and B = e(A) 2 V yielding a mutiplicative decomposition of Y .
The next exercise is about this.
392 Chapter 12. SDE driven by r.c.l.l. Semimartingales
Exercise 12.21. Let Y be a L(d, d)-valued semimartingale such that Y0 = I with Y and Y
being L0 (d)-valued. Let X = e(Y ). Let
X
Dt = ( X)s 1{k( X)s k 1 }
3
0<st
Z t = Xt Dt .
T
1
(a) P(k( A)t k 3 8t) = 1.
2
(b) P(k( M )t k 3 8t) = 1.
(c) (I + M ) is L0 (d)-valued.
Girsanov Theorem
In this chapter, we will obtain Girsanov Theorem and its generalizations by Meyer. Let M
T
be a martingale on (⌦, F, P) and let Q be another probability measure on (⌦, F), absolutely
continuous w.r.t. P. Then as noted in Remark 4.15, M is a semimartingale on (⌦, F, Q).
We will obtain a decomposition of M into N and B, where N is a Q-martingale. This
F
result for Brownian motion was due to Girsanov and we are presenting the generalizations
due to Meyer.
A
13.1 Girsanov Theorem
Suppose Z is a (0, 1) valued r.c.l.l. uniformly integrable martingale with EP [Zt ] = 1. Let
⇠ be the limit of Zt (in L1 ) as t ! 1 and suppose that P(⇠ > 0) = 1. Let Q be the
R
393
394 Chapter 13. Girsanov Theorem
Thus Ms is Q integrable if and only if Ms Zs is P-integrable. Further, for any stop time ,
EQ [M ] = EP [M Z ]. (13.1.2)
Thus (i) follows from Theorem 2.55. For (ii), if M is a Q-local martingale, then get stop
times ⌧n " 1 such that for each n, Mt^⌧n is a Q-martingale. Then we have
Thus Mt^⌧n Zt^⌧n is a P-martingale and thus M Z is a P- local martingale. The converse
T
follows similarly.
The following result is due to Meyer building upon the idea by Girsanov in the context
of a Wiener process.
N t Z t = Mt Z t Ut Zt
Z t Z t
= (Mt Zt [M, Z]t ) + [M, Z]t ( Us dZs + Zs dUs + [U, Z]t )
0 0
D
where we have used integration by parts formula, (4.6.7) along with U0 = 0. Now
Rt
(Mt Zt [M, Z]t ) is P-local martingale (see Theorem 9.27). Further, 0 Us dZs is a P-
local martingale (see Corollary 9.14). It thus follows that
Z t
Nt Zt = Lt + [M, Z]t Zs dUs [U, Z]t (13.1.5)
0
Rt
where Lt = (Mt Zt [M, Z]t ) 0 Us dZs and thus L is a P- local martingale. Since U 2 V
P
is a process with finite variation paths, [U, Z]t = 0<st ( U )s ( Z)s and as a consequence
Z t Z t
Zs dUs + [U, Z]t = Zs dUs . (13.1.6)
0 0
13.1. Girsanov Theorem 395
T
0
is a Q-local martingale.
Proof. Once again we need to show that LZ is a P-local martingale. We have observed
Nt = hM, Zit Zt
Z t
F
that (see Remark 9.31), Mt Zt hM, Zit is a local martingale. So in order to show that LZ
is a P- local martingale, it suffices to show that N defined by
1
( V ) = (Z ) ( hM, Zi) .
Thus predictability of hM, Zi and the fact that for a predictable stop time , Z is F
measurable, it follows using Lemma 8.24 that V is predictable. Then we have (note V0 = 0)
D
Z t Z t
Z t Vt = Zs dVs + Vs dZs + [V, Z]t . (13.1.10)
0 0
Rt
Since 0 Zs dVs = hM, Zit , we conclude
Z t
Nt = Vs dZs + [V, Z]t .
0
Rt
Now 0 Vs dZs is a P-local martingale by Corollary 9.14 and [V, Z]t is a P-local martingale
by Theorem 9.29. Thus N is a P-local martingale. As noted earlier, this completes the
proof that L is Q-local martingale.
396 Chapter 13. Girsanov Theorem
T
F
A
R
D
Bibliography
[1] Bachelier, L.; Theorie de la Speculation, Ann. Sci. Ecole Norm. Sup., vol. 17,(1900),
pp. 21-86. English translation in: The Random Character of stock market prices (P.
Cootner, editor), (1964), MIT Press.
T
[2] Bhatt, A. G. and Karandikar, R. L.:On the Fundamental Theorem of Asset Pricing,
Communications on Stochastic Analysis 9, (2015), 251 - 265.
[3] Bichteler, K.; Stochastic integration and Lp -theory of semi martingales, Annals of
Probability, 9, (1981), 49-89.
F
[4] Billingsley, P.: Probability and Measure, (1995), John Wiley, New York.
A
[5] Breiman, L.: Probability, (1968), Addison-Wesley, Reading, Mass.
[6] Burkholder, D. L.;A Sharp Inequality for Martingale Transforms, Annals of Probabil-
R
[7] Cherny, A. S. and Shiryaev, A. N.: Vector stochastic integrals and the fundamental
D
[9] Cinlar, E., Jacod, J., Protter, P., and Sharpe, M. J.: Semimartingales and Markov
processes Z. Wahrscheinlichkeitstheorie verw. Gebiete 54, (1980), 161 - 219.
397
398 Bibliography
[11] Delbaen, F. and Schachermayer, W.: A general version of the fundamental theorem of
asset pricing, Mathematische Annalen 300, (1994), 463 - 520.
[12] Delbaen, F. and Schachermayer, W.: The Fundamental Theorem of Asset Pricing for
Unbounded Stochastic Processes, Mathematische Annalen, 312, (1998), 215-250.
[13] Delbaen, F. and Schachermayer, W.: The Mathematics of Arbitrage, (2006), Springer-
Verlag.
[14] Davis, B.: On the integrability of the martingale square function, Israel J. Math. 8,
(1970), 187-190.
T
[16] Emery, M.: Une topologie sur l’espace des semi-martingales, Seminaire de Probabilites
XIII, Lecture Notes in Mathematics, 721, (1979), 260 - 280, Springer-Verlag.
F
[17] Emery, M. : Compensation de processus a variation finie non localement integrables.
Seminaire de Probabilites XIV, Lecture Notes in Mathematics, 784, (1980), 152 - 160,
Springer-Verlag.
A
[18] Ethier, S. N. and Kurtz, T. G. : Markov processes: Characterization and Convergence.
Wiley, New York. (1986)
R
[20] Harrison, J.M, Kreps, D.M.Martingales and Arbitrage in Multi-period Securities Mar-
kets, (1979), Journal of Economic Theory 20, 381-408.
[21] Harrison, M. and Pliska, S.: Martingales and Stochastic Integrals in the theory of
continuous trading, Stochastic Processes and Applications, 11 (1981), 215 - 260.
[22] Harrison, M. and Pliska, S.: A stochastic calculus model of continuous trading: Com-
plete markets, Stochastic Processes and their Applications 15, (1983), 313-316.
[23] Ikeda, N. and Watanabe, S.: Stochastic di↵erential equations and di↵usion processes,
(1981), North Holland.
Bibliography 399
[24] Ito, K.: Lectures on stochastic processes, Tata Institute of Fundamental Research,
Bombay, (1961).
[25] Jacod, J.: Calcul Stochastique et Problemes de Martingales. Lecture Notes in Mathe-
matics, 714 , (1979), Springer-Verlag.
[27] Jacod, J.: Grossissement initial, hypothese (H’) et theoreme de Girsanov In: Grossisse-
ments de filtrations: exemples et applications, Jeulin and Yor (eds.), Lecture Notes in
Mathematics 1118, (1985), 15-35, Springer-Verlag.
T
[28] Jacod, J. and Yor, M.: Etude des solutions extremales et representation intgrale des
solutions pour certains problemes de martingales, Z. Wahrscheinlichkeitstheorie ver.
Geb., 125, (1977), 38 - 83.
F
[29] Kabanov, Y.M.: On the FTAP of Kreps-Delbaen-Schachermayer, (English). Y.M. Ka-
banov (ed.) et al., Statistics and control of stochastic processes. The Liptser Festschrift.
Papers from the Steklov seminar held in Moscow, Russia, 1995-1996 , 191-203, World
A
Scientific, Singapore.
[33] Karandikar, R.L.: Pathwise solution of stochastic di↵erential equatios. Sankhya A, 43,
(1981), 121-132.
T
[40] Karandikar, R.L.: On pathwise stochastic integration, Stochastic processes and their
applications, 57, (1995), 11-18.
F
Indian Academy of Sciences, 124, (2014), 457-469.
[42] Karatzas, I. and Shreve, S. E.: Brownian Motion and Stochastic Calculus, (1988),
A
Springer-Verlag.
[43] Kreps, D.M.: Arbitrage and Equilibrium in Economics with in- finitely many Com-
modities, (1981), Journal of Mathematical Economics, 8, 15-35.
R
[44] Kunita, H. and Watanabe, S.: On square integrable martingales, Nagoya Math. J., 30,
(1967), 209-245.
D
[45] McKean, H. P.: Stochastic Integrals, (1969), Academic Press, New York.
[47] Merton, R.C.: The theory of rational option pricing, Bell J. Econ. Manag. Sci. 4,
(1973), 141-183.
[49] Meyer, P. A.: Un cours sur les integrales stochastiques, Seminaire de Probabilites X,
Lecture Notes in Mathematics, 511, (1976), 245 - 400, Springer-Verlag.
Bibliography 401
[50] Protter, P.: Stochastic Integration and Di↵erential Equations, (1980), Springer-Verlag.
[51] Revuz, D. and Yor, M.; Continuous Martingales and Brownian Motion, Grundlehren
der Mathematischen Wissenschaften, 293, (1991), Springer-Verlag.
[52] Ross, S.: The arbitrage theory of capital asset pricing, J. Econ. Theor. 13, (1976),
341-360.
[53] Samuelson, P.A.: Proof that properly anticipated prices fluctuate randomly, (1965),
Industrial Management Review 6, 41-50.
[55] Stricker, Ch.: Arbitrage et Lois de Martingale, Annales de l’Institut Henri Poincare -
Probabilites et Statistiques 26, (1990), 451-460.
T
[56] Williams, D.: Probability with Martingales, (1991), Cambridge University Press.
[59] Yor, M.: Sous-espaces denses dans L1 ou H 1 et representation des martingales Sem-
R
T
F
A
R
D
Index
B(Rd ), 7 P, 91
bp
B(E), 40 !, 61
B(⌦, F), 7 R0 , 95
C(E), 40 Rd , 7
T
C(Rd ) , 7 S, 91, 95
Cb (Rd ), 7 S1 , 101
Dd , 41 (A), 7
dem , 145
↵, 41
F ( , ⌧ ], [ , ⌧ ], 245
[⌧ ], 245
ucp
!, 62
A
ducp , 62
X, 43 V,V0 , V+ , V+ 0 , 67
em
!, 145 Var[a,b] , 66
v
X ! Y , 42
R
(F⇧ ), 45
(F⇧+ ), 45 W, 258
kf kp , 8 X⌧ , 51
X , 43
F⌧ , 51
D
403
404 INDEX
T
Doob’s maximal inequality, 20, 48 r.c.l.l., 43
Doob’s upcrossings inequality, 24
quadratic variation
Emery topology, 145 of stochastic integrator, 119
filtration, 13, 45
independent, 11
F of Brownian motion, 69
of square integrable martingales, 165
of stochastic integrator, 120
A
integration by parts formula, 121
Ito’s formula, 133, 136, 138 semimartingale, 161
stochastic integrator, 95
l.c.r.l. process, 43
R
continuous time, 46
convergence theorem, 25 contact time, 50
discrete time, 14 discrete, 18
locally square integrable, 161 hitting time, 50
square integrable, 160 predictable, 246
submartingale, 14, 46
No Arbitrage,NA, 343
trading strategies, 343
pathwise formula, 210
admissible, 343
for stochastic integral, 211
Poisson process, 44 ucc topology, 40