Académique Documents
Professionnel Documents
Culture Documents
A) < spElAr for all \>0.
Hint: f |X|PdP > f |X|PdP, where A = {w:|X| >A}.
b) Suppose there exists & > 0 such that
M = Blexp(k|X|)] < 00.
Prove that Pi|X| >) < Me~™ for all A>0.Exercises 17
Let X,Y: —+ R be two independent random variables and assume
for simplicity that X and Y are bounded. Prove that
E[XY] = E[X]E|¥] .
(xine: Assume |X| < M, |Y| < N. Approximate X and Y by sim-
ple functions p(w) = > aiXr,(w), ¥w) = Do bj; XG, (w), respectively,
a =1
3
where Fy = X7¥(a;,ai41)), Gj = ¥7"((by,Bj41)), —M = ao aids PUR G)).. ) :
rey
Let (2,F,P) be a probability space and let Ai, Ap,... be sets in F
such that
SY P(Ak) < 00
ta
Prove the Borel-Cantelli lemma:
rr) U ay=o,
mal kom
ie. the probability that w belongs to infinitely many Aj.s is zero.
a) Suppose Gi,G2,...,Gn are disjoint subsets of 2 such that
Prove that the family G consisting of @ and all unions of some (or
all) of Gi,...,G, constitutes a o-algebra on 22.
b) Prove that any finite o-algebra F on 9? is of the type described in
a).
c) Let F be a finite o-algebra on @ and let X:2 + R be F-
measurable. Prove that X assumes only finitely many possible
values. More precisely, there exists a disjoint family of subsets
F,..., Fm © F and real numbers ¢),...,¢m such that
X(w) = Satn(u) :
Let B, be Brownian motion on R, By = 0. Put E = E°18.
2.9.
2.10.
2.11.
2, Some Mathematical Preliminaries
a) Use (2.2.3) to prove that
Ele] = exp(-3ut) for allueR.
b) Use the power series expansion of the exponential function on both
sides, compare the terms with the same power of u and deduce that
EB?) = 30?
and more generally that,
(2k)! a,
E[BM|=ae-qth: REN.
c) If you feel uneasy about the lack of rigour in the method in b), you
can proceed as follows: Prove that (2.2.2) implies that
EIs(B)] = pee [ S(e)e Fade
Z
for all functions f such that the integral on the right converges
Then apply this to f(z) = 2% and use integration by parts and
induction on k.
4) Prove (2.2.14), for example by using b) and induction on n.
To illustrate that the (finite-dimensional) distributions alone do not
give all the information regarding the continuity properties of a pro-
cess, consider the following example:
Let (2, F, P) = (0,00), B, 2) where B denotes the Borel a-algebra on
(0,00) and p is a probability measure on [0,00) with no mass on single
points. Define '
1 ift=w
HXalw) = { 0 otherwise
and
¥;,(w) =0 for all (t,w) € [0, 00) x (0,00) .
Prove that {X,} and {¥,} have the same distributions and that X; is
a version of Y;. And yet we have that t + ¥;(w) is continuous for all
w, while t + X;(w) is discontinuous for all w.
A stochastic process X; is called stationary if {X;} has the same dis-
tribution as {X;4,} for any h > 0. Prove that Brownian motion B,
has stationary increments, i.e. that the process {Bi4n — Bi}n>0 has
the same distribution for all t.
Prove (2.2.15).+12.
-13.
14,
15.
-16.
17.
Exercises 19
Let B be Brownian motion and fix to > 0. Prove that
Bu= Bust—-Byi t20
is a Brownian motion
Let B, be 2-dimensional Brownian motion and put
D,={cER?*;|2| 0-
Compute
P°|Be € Dy).
Let B, be n-dimensional Brownian motion and let K C R" have zero
n-dimensional Lebesgue measure. Prove that the expected total length
of time that B, spends in K is zero. (This implies that the Green
measure associated with B, is absolutely continuous with respect to
Lebesgue measure. See Chapter 9).
Let By be n-dimensional Brownian motion starting at 0 and let
UeR"*" be a (constant) orthogonal matrix, ie. UUT =I. Prove that
By: = UB,
is also a Brownian motion.
(Brownian scaling). Let B, be a 1-dimensional Brownian motion
and let c > 0 be a constant. Prove that
is also a Brownian motion.
If X,(-): @ — R is a continuous stochastic process, then for p > 0 the
p’th variation process of X;, (X,X)\”) is defined by
(XX) (w) = slim | ST [Xa )-Xe,(w))? (limit in probability)
test
where 0 = t) < tz <...< ty =t and At, = ty41 — te. In particular,
if p = 1 this process is called the total variation process and if p = 2
this is called the quadratic variation process. (See Exercise 4.7.) For
Brownian motion B, ¢ R we now show that the quadratic variation
process is simply
(B, B)(w) = (B, BY (w) =t as.
Proceed as follows:20 2. Some Mathematical Preliminaries
a) Define
ABy = Buys ~ Bu
and put.
¥(tw) = S>(ABx(w))?
teSt
Show that.
E\(Y0(ABx)? — #)7] = 2S (Ate)?
teSt teSt
and deduce that ¥(t,-) + t in L?(P) as Aty oo.
b) Use a) to prove that a.a. paths of Brownian motion do not have
a bounded variation on [0,t], ie. the total variation of Brownian
motion is infinite, a.s.
2.18. a) Let 2 = {1,2,3,4,5} and let U/ be the collection
U = {(1,2,3}, (3,4,5}}
of subsets of (2. Find the smallest o-algebra containing U (i.e. the
o-algebra Hy generated by U).
b) Define X : 2+ R by
X(1)=X(2)=0, X(3)=10, X(4) = X(5)=1
Is X measurable with respect to Hu?
c) Define Y : 2 4 R by
YQ) =0, Y(2)=¥(3)=¥(4)=¥(5) =1
Find the o-algebra Hy generated by ¥
2.19. Let (2,F,u) be a probability space and let p € {1,00]. A sequence
{fn}%21 of functions fy, € L?() is called a Cauchy sequence if
fn -fmllp 70 as nym — oo.
‘The sequence is called convergent if there exists f € L?(s) such that
fn f in L(y).
Prove that every convergent sequence is a Cauchy sequence
A fundamental theorem in measure theory states that the converse is
also true: Every Cauchy sequeence in L?(s) is convergent. A normed
linear space with this property is called complete. Thus the L?(s1)
spaces are complete.
2.20. Let B, be 1-dimensional Brownian motion, o € R. be constant and
OSs M,cM
(i.e. {M,} és increasing). An n-dimensional stochastic process {M}120 on
(Q,F,P) is called a martingale with respect to a filtration {M,}r>0 (and
with respect to P) if
(i) M, is My-measurable for all t,
(ii) El|Mil] < 00 for allt
and
(iii) E[M,|M,] = M, for all s > t.
Here the expectation in (ii) and the conditional expectation in (iii) is
taken with respect to P = P®. (See Appendix B for a survey of conditional
expectation),
Example 3.2.3. Brownian motion B; in R” is a martingale w.r.t. the o-
algebras F, generated by {B,;s < t}, because
E\|Bil? < EllBi?] =|Bol? +nt and if s > ¢ then
E[B.|Fi] = E[Bs ~ Be + BilFi)
= E(B, - BilFi] + E[BlF| = 0+ Be = Be
Here we have used that E[(B, — Br)|¥:] = E[B, — Bi] = 0 since B, — By is
independent of F; (see (2.2.11) and Theorem B.2.d)) and we have used that
E|B.\Fi] = B, since B, is F;-measurable (see Theorem B.2.c)).
For continuous martingales we have the following important inequality
due to Doob: (See e.g. Stroock and Varadhan (1979), Theorem 1.2.3 or Revuz
and Yor (1991), Theorem II.1.7)
Theorem 3.2.4 (Doob’s martingale inequality). If M, is a martingale
such that t + My(w) is continuous a.s., then for all p > 1,T > 0 and all
A>0
7
P{ sup |Mi| >A} < <>: El|Mrl?|
(sup pIM 2 AL S 5p Ello
We now use this inequality to prove that the It6 integral
:
J fewyan,
4
can be chosen to depend continuously on t :323. It6 Integrals
Theorem 3.2.5. Let f € V(0,7). Then there exists a t-continuous version
of
f ovoydate) O Up=ahy.
‘Then
(i) Up € £(Z,k)
(ii) X —U,L£(Z,k), since
BUX — Ug) 24] = BIXZ,| — on B22
= BIX(X + Wa) — nF > B1z;24)
4
~ foe Dw \(X+ WI = qaulke?+m?| =0.)
The result can be interpreted as follows:
For large k we put X ~ Zy, while for small k the relation between a?
and m? becomes more important. If m? >> a?, the observations are to a large
extent neglected (for small k) and X;, is put equal to its mean value, 0. See
also Exercise 6.11.6.2 The 1-Dimensional Linear Filtering Problem 89
This example gives the motivation for our approach:
We replace the process Z, by an orthogonal increment process N; (Step 2)
in order to obtain a representation for X; analogous to (6.2.5). Such a rep-
resentation is obtained in Step 3, after we have identified the best linear
estimate with the best measurable estimate (Step 1) and established the con-
nection between N; and Brownian motion.
Step 1. Z-Linear and Z-Measurable Estimates
Lemma 6.2.2. Let X,Z,; 8 < t be random variables in L?(P) and assume
that
(X, 2s1)Zeay-++1Zs,) € RY
has a normal distribution for all 51, 52,...,8n