Escolar Documentos
Profissional Documentos
Cultura Documentos
Stochastic Calculus
with Applications
Ovidiu Calin
Eastern Michigan University, USA
World Scientific
NEW JERSEY LONDON SINGAPORE BEIJING SHANGHAI HONG KONG TA I P E I CHENNAI
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center,
Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from
the publisher.
Printed in Singapore
Preface
Deterministic Calculus has been proved extremely useful in the last few hun-
dred years for describing the dynamics laws for macro-objects, such as plan-
ets, projectiles, bullets, etc. However, at the micro-scale, the picture looks
completely dierent, since at this level the classical laws of Newtonian me-
chanics cease to function normally. Micro-particles behave dierently, in
the sense that their state cannot be determined accurately as in the case of
macro-objects; their position or velocity can be described using probability
densities rather than exact deterministic variables. Consequently, the study
of nature at the micro-scale level has to be done with the help of a special
tool, called Stochastic Calculus. The fact that nature at a small scale has a
non-deterministic character makes Stochastic Calculus a useful and important
tool for the study of Quantum Mechanics.
In fact, all branches of science involving random functions can be ap-
proached by Stochastic Calculus. These include, but they are not limited to,
signal processing, noise ltering, stochastic control, optimal stopping, elec-
trical circuits, nancial markets, molecular chemistry, population evolution,
etc.
However, all these applications assume a strong mathematical background,
which takes a long time to develop. Stochastic Calculus is not an easy theory
to grasp and, in general, requires acquaintance with probability, analysis and
measure theory. This fact makes Stochastic Calculus almost always absent
from the undergraduate curriculum. However, many other subjects studied at
this level, such as biology, chemistry, economics, or electrical circuits, might be
more completely understood if a minimum knowledge of Stochastic Calculus
is assumed.
The attribute informal, present in the title of the book, refers to the fact
that the approach is at an introductory level and not at its maximum math-
ematical detail. Many proofs are just sketched, or done naively without
putting the reader through a theory with all the bells and whistles.
The goal of this work is to informally introduce elementary Stochastic
Calculus to senior undergraduate students in Mathematics, Economics and
Business majors. The authors goal was to capture as much as possible of the
v
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page vi
spirit of elementary Calculus, which the students have already been exposed
to in the beginning of their majors. This assumes a presentation that mimics
similar properties of deterministic Calculus as much as possible, which facili-
tates the understanding of more complicated concepts of Stochastic Calculus.
The reader of this text will get the idea that deterministic Calculus is just
a particular case of Stochastic Calculus and that Itos integral is not a too
much harder concept than the Riemannian integral, while solving stochastic
dierential equations follows relatively similar steps as solving ordinary dif-
ferential equations. Moreover, modeling real life phenomena with Stochastic
Calculus rather than with deterministic Calculus brings more light, detail and
signicance to the picture.
The book can be used as a text for a one semester course in stochastic
calculus and probabilities, or as an accompanying text for courses in other
areas such as nance, economics, chemistry, physics, or engineering.
Since deterministic Calculus books usually start with a brief presentation
of elementary functions, and then continue with limits, and other properties
of functions, we employed here a similar approach, starting with elementary
stochastic processes, dierent types of limits and pursuing with properties
of stochastic processes. The chapters regarding dierentiation and integration
follow the same pattern. For instance, there is a product rule, a chain-type rule
and an integration by parts in Stochastic Calculus, which are modications of
the well-known rules from elementary Calculus.
In order to make the book available to a wider audience, we sacriced rigor
and completeness for clarity and simplicity, emphasizing mainly on examples
and exercises. Most of the time we assumed maximal regularity conditions for
which the computations hold and the statements are valid. Many complicated
proofs can be skipped at the rst reading without aecting later understand-
ing. This will be found attractive by both Business and Economics students,
who might get lost otherwise in a very profound mathematical textbook where
the forests scenery is obscured by the sight of the trees. A ow chart indicat-
ing the possible order the reader can follow can be found at the end of this
preface.
An important feature of this textbook is the large number of solved prob-
lems and examples which will benet both the beginner as well as the advanced
student.
This book grew from a series of lectures and courses given by the author
at Eastern Michigan University (USA), Kuwait University (Kuwait) and Fu-
Jen University (Taiwan). The student body was very varied. I had math,
statistics, computer science, economics and business majors. At the initial
stage, several students read the rst draft of these notes and provided valuable
feedback, supplying a list of corrections, which is far from exhaustive. Finding
any typos or making comments regarding the present material are welcome.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page vii
Preface vii
ix
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page x
( ) Gamma function
B( , ) Beta function
Nt Poisson process
Sn Waiting time for Poisson process
Tn Interarrival time for Poisson process
1 2 The minimum between 1 and 2 (= min{1 , 2 })
1 2 The maximum between 1 and 2 (= max{1 , 2 })
n Sequence superior limit (= supn1 n )
n Sequence inferior limit (= inf n1 n )
Drift rate
Volatility, standard deviation
xk , Partial derivative with respect to xk
xk
Rn n-dimensional Euclidean space
x Euclidean norm (= x21 + + x2n )
f Laplacian of f
1A , A The characteristic function of A
f L2 2
The L -norm (=
b 2
a f (t) dt)
2
L [0, T ] Squared integrable functions on [0, T ]
C 2 (Rn ) Functions twice dierentiable with second derivative continuous
C02 (Rn ) Functions with compact support of class C 2
Rt Bessel process
t The mean square estimator of t
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page xi
Contents
Preface v
2 Basic Notions 11
2.1 Probability Space . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Sample Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Events and Probability . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Random Variables . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5 Integration in Probability Measure . . . . . . . . . . . . . . . . 14
2.6 Two Convergence Theorems . . . . . . . . . . . . . . . . . . . . 15
2.7 Distribution Functions . . . . . . . . . . . . . . . . . . . . . . . 16
2.8 Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.9 Expectation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.9.1 The best approximation of a random variable . . . . . . 20
2.9.2 Change of measure in an expectation . . . . . . . . . . . 21
2.10 Basic Distributions . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.11 Sums of Random Variables . . . . . . . . . . . . . . . . . . . . 25
2.12 Conditional Expectations . . . . . . . . . . . . . . . . . . . . . 27
2.13 Inequalities of Random Variables . . . . . . . . . . . . . . . . . 32
2.14 Limits of Sequences of Random Variables . . . . . . . . . . . . 37
xi
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page xii
Contents xiii
Bibliography 309
Index 313
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 1
Chapter 1
Even if deterministic Calculus is an excellent tool for modeling real life prob-
lems, however, when it comes to random exterior inuences, Stochastic Cal-
culus is the one which can allow for a more accurate modeling of the problem.
In real life applications, involving trajectories, measurements, noisy signals,
etc., the eects of many unpredictable factors can be averaged out, via the
Central Limit Theorem, as a normal random variable. This is related to the
Brownian motion, which was introduced to model the irregular movements of
pollen grains in a liquid.
In the following we shall discuss a few problems involving random pertur-
bations, which serve as motivation for the study of the Stochastic Calculus
introduced in next chapters. We shall come back to some of these problems
and solve them partially or completely in Chapter 11.
dP (t) = rP (t)dt.
This dierential equation has the solution P (t) = P0 ert , where P0 is the initial
population size. The evolution of the population is driven by its growth rate
1
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 2
40 100
K
80
30
60
20
40
10
20
0 2 4 6 8 10 0 2 4 6 8 10 12
a b
Figure 1.1: (a) Noisy population with exponential growth. (b) Noisy population
with logistic growth.
r. In real life this rate is not constant. It might be a function of time t, or even
more general, it might oscillate irregularly around some deterministic average
function a(t):
rt = a(t) + noise.
In this case, rt becomes a random variable indexed over time t. The associated
equation becomes a stochastic dierential equation
dP (t) = a(t) + noise P (t)dt. (1.1.1)
Solving an equation of type (1.1.1) is a problem of Stochastic Calculus, see
Fig. 1.1(a).
Logistic growth model The previous exponential growth model allows the
population to increase indenitely. However, due to competition, limited space
and resources, the population will increase slower and slower. This model
was introduced by P.F. Verhust in 1832 and rediscovered by R. Pearl in the
twentieth century. The main assumption of the model is that the amount
of competition is proportional with the number of encounters between the
population members, which is proportional with the square of the population
size
dP (t) = rP (t)dt kP (t)2 dt. (1.1.2)
The solution is given by the logistic function
P0 K
P (t) = ,
P0 + (K P0 )ert
where K = r/k is the saturation level of the population. One of the stochastic
variants of equation (1.1.2) is given by
dP (t) = rP (t)dt kP (t)2 dt + (noise)P (t),
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 3
where R is a measure of the size of the noise in the system. This equation is
used to model the growth of a population in a stochastic, crowded environment,
see Fig. 1.1(b).
dB(t) = rB(t)dt
with the nal condition B(T ) = B. In a noisy market the constant interest
rate r is replaced by rt = r(t) + noise, a fact that makes the bond pricing
more complicated. This treatment can be achieved by Stochastic Calculus.
How does the noisy force inuence the deviation angle (t)? Stochastic Cal-
culus can be used to answer this question.
Now consider a drop of ink (which is made out of a very large number of
tiny particles) left to diuse in a liquid. Each ink particle performs a noisy
trajectory in the liquid. Let p(x, t) represent the density of particles that
arrive about x at time t. After some diusion time, the darker regions of
the liquid represent the regions with higher density p(x, t), while the lighter
regions correspond to smaller density p(x, t). Knowing the density p(x, t)
provides control over the dynamics of the diusion process and can be used to
nd the probability that an ink particle reaches a certain region.
dC(t)
= a C0 C(t) + bE,
dt
where C0 is the natural level of cholesterol and E denotes the daily rate of
intaken cholesterol; the constants a and b model the production and absorption
of cholesterol in the organism. The solution of this linear dierential equation
is
b
C(t) = C0 eat + C0 + E (1 eat ),
a
which in the long run tends to the saturation level of cholesterol C0 + ab E.
Due to either observation errors or variations in the intake amount of food,
the aforementioned equation will get the following noisy form
dC(t)
= a C0 C(t) + bE + noise.
dt
This equation can be explicitly solved using Stochastic Calculus. Furthermore,
we can also nd the probability that the cholesterol level is over the allowed
organism limit.
dx(t) x(t)
=
dt |x(t)|
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 5
3
14
x0
12 2
10
1
6 3 2 1 1 2 3
4 xt 1
2 o 2
0 2 4 6 8 10 3
a b
Figure 1.2: (a) The trajectory of the electron x(t) tends towards the origin.
(b) White noise.
Like in the case of the pollen grain, whose motion is agitated by the neigh-
boring molecules, we assume that the electron is subject to bombardment by
some aether particles, which makes its movement unpredictable, with con-
stant tendency to go towards the origin, see Fig. 1.2 (a). Then its equation
becomes
dx(t) x(t)
= + noise
dt |x(t)|
any specic frequency,1 see Fig. 1.2. If Nt denotes the white noise at
time t, the trajectory (t) of a diused particle satises
(t) = Nt , t 0.
for any compact supported, smooth function f . From this point of view, the
white noise Nt is a generalized function or a distribution. We shall get back to
the notion of white noise in section 11.1.
1
The white light is an equal mixture of radiations of all visible frequencies.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 7
14 14 14
12 12 12
10 10 10
8 8 8
6 6 6
4 4 4
2 2 2
0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
a b c
n1
n1
= sup |Pk Pk+1 | = sup (x)2 + (f )2 .
xi xi
k=0 k=0
where we used that the limit of an increasing sequence is equal to its superior
limit.
Denition 1.8.1 The function f (x) has bounded variation on the interval
[a, b] if for any division a = x0 < x1 < < xn = b the sum
n1
|f (xk+1 ) f (xk )|
k=0
n1
V (f ) = sup |f (xk+1 ) f (xk )|. (1.8.3)
xi
k=0
n1
V (f ) = sup |f (xk+1 ) f (xk )|
xi
k=0
n1
|f (xk+1 ) f (xk )| b
= lim x = |f (x)| dx.
n xk+1 xk a
k=0
The next result states a relation between the length of the graph and the
total variation of a function f , which is not dierentiable.
V (f ) (b a) + V (f ),
n1 b
b
dBt
V (Bt ) = sup |Btk+1 Btk | =
dt = |Nt | dt = ,
tk a dt a
k=0
n1
(2)
V (f ) = sup |f (xk+1 ) f (xk )|2 . (1.8.4)
xi
k=0
where the sup is taken over all divisions a = x0 < x1 < < xn = b.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 9
The total variation for the Brownian motion does not provide much in-
formation. It turns out that the correct measure for the roughness of the
Brownian motion is the quadratic variation
n1
V (2) (Bt ) = sup |B(tk+1 ) B(tk )|2 . (1.8.5)
ti
k=0
Chapter 2
Basic Notions
11
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 12
Remark 2.2.1 Pick a natural number at random. Any subset of the sample
space corresponds to a sequence formed with 0 and 1. For instance, the subset
{1, 3, 5, 6} corresponds to the sequence 10101100000 . . . having 1 on the 1st,
3rd, 5th and 6th places and 0 in rest. It is known that the number of these
sequences is innite and can be put into a bijective correspondence with the
real number set R. This can also be written as |2N | = |R|, and stated by saying
that the set of all subsets of natural numbers N has the same cardinal as the
real numbers set R.
1. P () = 1;
Example 2.3.1 In the case of a coin ipping, the probability space has the
following elements: = {H, T }, F = {, {H}, {T }, {H, T }} and P is dened
by P () = 0, P ({H}) = 12 , P ({T }) = 12 , P ({H, T }) = 1.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 13
Basic Notions 13
Figure 2.1: If any set X 1 (a, b) is known, then the random variable X :
R is 2 -measurable.
Example 2.3.3 Let = [0, 1] and consider the -eld B([0, 1]) given by the
set of all open or closed intervals on [0, 1], or any unions, intersections, and
complementary sets. Dene P (A) = (A), where stands for the Lebesgue
measure (in particular, if A = (a, b), then P (A) = b a is the length of the
interval). It can be shown that (, B([0, 1]), P ) is a probability space.
Example 2.4.1 Let X() be the number of people who want to buy houses,
given the state of the market . Is X measurable? This would mean that
given two numbers, say a = 10, 000 and b = 50, 000, we know all the market
situations for which there are at least 10, 000 and at most 50, 000 people
willing to purchase houses. Many times, in theory, it makes sense to assume
that we have enough knowledge so that we can assume X is measurable.
Example 2.4.2 Consider the experiment of ipping three coins. In this case
is the set of all possible triplets, which can be made with H and T . Consider
the random variable X which gives the number of tails obtained. For instance
X(HHH) = 0, X(HHT ) = 1, etc. The sets
Basic Notions 15
Positivity: If X 0 then
X dP 0.
Then
(1) f is measurable;
(2) lim fn dP = f dP.
n
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 16
f () = lim fn (), ;
n
(ii) there is an integrable function g, (i.e. |g| dP < ) such that
It is worth observing that since X is a random variable, then the set {; X()
x} belongs to the information set F.
The distribution function is non-decreasing and satises the limits
If we have
d
F (x) = p(x),
dx X
then we say that p(x) is the probability density function of X.
It is important to note the following relation among distribution function,
probability and probability density function of the random variable X
x
FX (x) = P (X x) = dP () = p(u) du. (2.7.1)
{Xx}
Basic Notions 17
The rst one is a consequence of the fact that the distribution function
FX (x) is non-decreasing. The second follows from (2.7.1) by making x
p(u) du = dP = P () = 1.
For more details the reader is referred to a traditional probability book, such
as Wackerly et al. [13].
2.8 Independence
Roughly speaking, two random variables X and Y are independent if the
occurrence of one of them does not change the probability density of the other.
More precisely, if for any two open intervals A, B R, the events
E = {; X() A}, F = {; Y () B}
Dropping the factor dxdy yields the desired result. We note that the converse
also holds true.
The -algebra generated by a random variable X : R is the -algebra
generated by the unions, intersections and complements of events of the form
{; X() (a, b)}, with a < b real numbers. This will be denoted by AX .
Two -elds G and H included in F are called independent if
P (G H) = P (G)P (H), G G, H H.
The random variable X and the -eld G are called independent if the
algebras AX and G are independent.
2.9 Expectation
A random variable X : R is called integrable if
|X()| dP () = |x|p(x) dx < ,
R
where p(x) denotes the probability density function of X. The previous iden-
tity is based on changing the domain of integration from to R.
The expectation of an integrable random variable X is dened by
E[X] = X() dP () = x p(x) dx.
R
1
x+dx
We are using the useful approximation P (x < X < x + dx) = x
p(u) du = p(x)dx.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 19
Basic Notions 19
Proposition 2.9.1 The expectation operator E is linear, i.e. for any inte-
grable random variables X and Y
1. E[cX] = cE[X], c R;
2. E[X + Y ] = E[X] + E[Y ].
Proof: It follows from the fact that the integral is a linear operator.
Proof: This is a variant of Fubinis theorem, which in this case states that a
double integral is a product of two simple integrals. Let pX , pY , pX,Y denote
the probability densities of X, Y and (X, Y ), respectively. Since X and Y are
independent, by Proposition 2.8.1 we have
E[XY ] = xypX,Y (x, y) dxdy = xpX (x) dx ypY (y) dy = E[X]E[Y ].
From Exercise 2.9.4 (b), we have V ar(X) 0, so, there is a real number
> 0 such that V ar(X) = 2 . The number is called standard deviation.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 20
Exercise 2.9.5 Let and denote the mean and the standard deviation of
the random variable X. Show that
E[X 2 ] = 2 + 2 .
Exercise 2.9.6 Consider two random variables with the following table of
joint probabilities:
Y \X 1 0 1
1 1/16 3/16 1/16
0 3/16 0 3/16
1 1/16 3/16 1/16
Show the following:
(a) E[X] = E[Y ] = E[XY ] = 0;
(b) Cov(X, Y ) = 0;
(c) P (0, 0) = PX (0)PY (0);
(d) X and Y are not independent.
(b) Use part (a) to show that for any random variables X and Y we have
1 (X, Y ) 1.
(c) What can you say about the random variables X and Y if (X, Y ) = 1?
Basic Notions 21
It follows that the mean, , is the best approximation of the random variable
X in the least squares sense.
X N (, 2 ).
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 22
Y N ( + , 2 2 ).
The name comes from the fact that the nth moments of X, given by n =
E[X n ], are generated by the derivatives of mX (t)
dn mX (t)
= n .
dtn |t=0
It is worth noting the relation between the Laplace transform
and the mo-
ment generating function, in the case x 0, L(p(x))(t) = etx p(x) dx =
0
mX (t).
Exercise 2.10.3 Find the moment generating function for the exponential
distribution p(x) = ex , x 0, > 0.
Exercise 2.10.4 Show that if X and Y are two independent random vari-
ables, then mX+Y (t) = mX (t)mY (t).
Basic Notions 23
0.4
0.5
0.3 0.4
0.3
0.2
0.2
0.1
0.1
4 2 0 2 4 0 2 4 6 8
a b
3.5
3, 9 8, 3
3.0
0.20
3, 2 2.5
0.15
2.0
4, 3
0.10 1.5
1.0
0.05
0.5
c d
Figure 2.2: (a) Normal distribution. (b) Log-normal distribution. (c) Gamma
distributions. (d) Beta distributions.
2 2 /2
(a) E[Y n ] = en+n , where Y = eX .
(b) Show that the mean and variance of the log-normal random variable
Y = eX are
2 2 2
E[Y ] = e+ /2 , V ar[Y ] = e2+ (e 1).
x1 ex/
p(x) = , x 0,
()
x1 (1 x)1
p(x) = , 0 x 1,
B(, )
where B(, ) denotes the beta function.2 See see Fig. 2.2(d) for two partic-
ular density functions. In this case
E[X] = , V ar[X] = .
+ ( + )2 ( + + 1)
k
P (X = k) = e , k = 0, 1, 2, . . . ,
k!
with > 0 parameter, see Fig. 2.3(b). In this case E[X] = and V ar[X] = .
Pearson 5 distribution Let , > 0. A random variable X with the density
function
1 e/x
p(x) = , x0
() (x/)+1
is said to have a Pearson 5 distribution3 with positive parameters and . It
can be shown that
2
, if > 1 , if > 2
E[X] = 1 V ar(X) = ( 1)2 ( 2)
,
otherwise, , otherwise.
2 ()()
Two denition formulas for the beta functions are B(, ) = (+)
and B(, ) =
1
0
y (1 y)
1 1
dy.
3
The Pearson family of distributions was designed by Pearson between 1890 and 1895.
There are several Pearson distributions, this one being distinguished by the number 5.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 25
Basic Notions 25
0.10
0.35
0.08
15, 0 k 30
0.30
0.25
0.06
0.20
3
0.15 0.04
0.10
0.02
0.05
0 2 4 6 8 10 5 10 15 20 25 30
a b
The mode of this distribution is equal to .
+1
The Inverse Gaussian distribution Let , > 0. A random variable X
has an inverse Gaussian distribution with parameters and if its density
function is given by
2
(x)
p(x) = e 22 x , x > 0. (2.10.3)
2x3
We shall write X IG(, ). Its mean, variance and mode are given by
3 92 3
E[X] = , V ar(X) = , M ode(X) = 1+ ,
42 2
where the mode denotes the value xm for which p(x) is maximum, i.e., p(x0 ) =
maxx p(x). This distribution will be used to model the time instance when a
Brownian motion with drift exceeds a certain barrier for the rst time.
Exercise 2.11.2 If F (s) = L(f (t))(s), G(s) = L(g(t))(s) both exist for s >
a 0, then
H(s) = F (s)G(s) = L(h(t))(s),
for
t t
h(t) = (f g)(t) = f (t )g( ) d = f ( )g(t ) d.
0 0
Using the associativity of the convolution
(f g) k = f (g k) = f g k
we obtain that if f , g and k are the probability densities of the positive,
independent random variables X, Y and Z, respectively, then f g k is the
probability density of the sum X + Y + Z. The aforementioned result can be
easily extended to the sum of n random variables.
Example 2.11.3 Consider two independent, exponentially distributed ran-
dom variables X and Y . We shall investigate the distribution of the sum
X + Y . Consider f (t) = g(t) = et in Theorem 2.11.1 and obtain the
probability density of the sum
t
h(t) = e(t ) e d = 2 tet , t 0,
0
which is Gamma distributed, with parameters = 2 and = 1/.
Exercise 2.11.4 Consider the independent, exponentially distributed random
variables X 1 e1 t and Y 2 e2 t , with 1 = 2 . Show that the sum is
distributed as
1 2
X +Y (e2 t e1 t ), t 0.
1 2
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 27
Basic Notions 27
Figure 2.4: The orthogonal projection of the random variable X on the space
SG is the conditional expectation Y = E[X|G].
X, Y = E[XY ],
see Exercise 2.12.10. This denes the norm X2 = E[X 2 ], which induces the
distance d(X, Y ) = X Y . Denote by SG the set of all G-measurable random
variables on . We shall show that the element of SG that is the closest to
X in the aforementioned distance is the conditional expectation Y = E[X|G],
see Fig. 2.4. Let X denote the orthogonal projection of X on the space SG .
This satises
E[(X X )(Z X )] = 0, Z SG . (2.12.6)
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 28
X X 2 + Z X 2 = X Z2 , Z SG
X X 2 X Z2 , Z SG ,
which is equivalent to
E[(X Y )U ] = 0, U SG .
E[XU ] = E[Y U ], U SG .
then X = 0 a.s.
Basic Notions 29
Proof: We need to show that E[X] satises conditions 1 and 2. The rst one
is obviously satised since any constant is G-measurable. The latter condition
is checked on each set of G. We have
X dP = E[X] = E[X] dP = E[X]dP
X dP = E[X]dP.
Example 2.12.4 Show that E[E[X|G]] = E[X], i.e. all conditional expecta-
tions have the same mean, which is the mean of X.
E[X|F] = X.
Proof: The random variables X and E[X|F] are both F-measurable (from
the denition of the random variable). From the denition of the conditional
expectation we have
E[X|F] dP = X dP, A F.
A A
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 30
4. Positivity:
E[X|G] 0, if X 0;
5. Expectation of a constant is a constant:
E[c|G] = c.
E[X|G] = E[X],
if X is independent of G.
Exercise 2.12.7 Prove the property 3 (tower property) given in the previous
proposition.
Exercise 2.12.8 Toss a fair coin 4 times. Each toss yields either H (heads)
or T (tails) with equal probability.
(a) How many elements does the sample space have?
(b) Consider the events A = {Two of the 4 tosses are H}, B = {The rst
toss is H}, and C = {3 of the 4 tosses are H}. Compute P (A), P (B) and
P (C).
(c) Compute P (A B) and P (B C).
(d) Are the events A and B independent?
(e) Are the events B and C independent? Find P (B|C).
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 31
Basic Notions 31
G = {we know the outcomes of the tosses but not the order}.
X = number of H number of T
Figure 2.5: Jensens inequality (E[X]) < E[(X)] for a convex function .
(E[X]) E[(X)].
Basic Notions 33
(x) () + ()(x ),
which means the graph of (x) is above the tangent line at x, (x) . Replac-
ing x by the random variable X, and taking the expectation yields
E[X]2 E[X 2 ].
Since the right side is nite, it follows that E[X] < , so X is integrable.
mX (t) et .
Using the denition of the moment generating function mX (t) = E[etX ] and
that E[tX] = tE[X] = t, then (2.13.8) leads to the desired inequality.
The variance of a square integrable random variable X is dened by
(E[X|G]) E[(X)|G].
= p
dP () = P (A) = P (|X| ).
p p
A
Basic Notions 35
2 dP = 2 P (A) = 2 P (; |X() | ).
A
P (X ) = P (tX t) = P (etX et )
E[Y ] E[etX ]
= P (Y et ) t = .
e et
2. The case t < 0 is similar.
In the following we shall present an application of the Cherno bounds for
the normal distributed random variables.
Let X be a random variable normally distributed with mean and variance
2
. It is known that its moment generating function is given by
1 2 2
m(t) = E[etX ] = et+ 2 t
.
which implies
1
min[( )t + t2 2 ]
P (X ) e t>0 2 .
Basic Notions 37
Exercise 2.14.4 If Xn tends to X in mean square, show that E[Xn |H] tends
to E[X|H] in mean square.
Basic Notions 39
lim P (|Yn Y | ) = 0,
n
Remark 2.14.7 The conclusion still holds true even in the case when there
is a p > 0 such that E[|Xn |p ] 0 as n .
We make the remark that this type of limit is even weaker than the stochastic
convergence, i.e. it is implied by it.
An application of the limit in distribution is obtained if we consider (x) =
itx
e . In this case the expectation becomes the Fourier transform of the prob-
ability density
E[(X)] = eitx p(x) dx = p(t),
ms-lim (Xn + Yn ) = 0.
n
(x + y)2 2x2 + 2y 2 .
Basic Notions 41
Exercise 2.15.5 Let X be a random variable with the probability density func-
tion
5 1
p(x) = , x 0.
(1/5)(4/5) x5 + 1
(a) Show that E[X 2 ] < and E[X 4 ] = ;
1
(b) Construct the sequences of random variables Xn = Yn = n X. Show
that ms-lim Xn = 0, ms-lim Yn = 0, but ms-lim (Xn Yn ) = .
n n n
The evolution in time of a given state of the world given by the function
t Xt () is called a path or realization of Xt . The study of stochastic
processes using computer simulations is based on retrieving information about
the process Xt given a large number of its realizations.
Next we shall structure the information eld F with an order relation
parameterized by the time t. Consider that all the information accumulated
until time t is contained by the -eld Ft . This means that Ft contains the
information containing events that have already occurred until time t, and
which did not. Since the information is growing in time, we have
Fs Ft F
Example 2.16.3 Don Joe goes to a doctor to get an estimation of how long
he still has to live. The age at which he will pass away is a random variable,
denoted by X. Given his medical condition today, which is contained in Ft ,
the doctor can infer an average age, which is the average of all random in-
stances that agree with the information to date; this is given by the conditional
expectation Xt = E[X|Ft ]. The stochastic process Xt is adapted to the medical
knowledge Ft .
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 43
Basic Notions 43
Remark 2.16.5 The rst condition states that the unconditional forecast is
nite E[|Xt ]] = |Xt | dP < . Condition 2 says that the value Xt is known,
given the information set Ft . This can also be stated by saying that Xt is
Ft -measurable. The third relation asserts that the best forecast of unobserved
future values is the last observation on Xt .
Example 2.16.6 Let Xt denote Mr. Li Zhus salary after t years of work at
the same company. Since Xt is known at time t and it is bounded above, as all
salaries are, then the rst two conditions hold. Being honest, Mr. Zhu expects
today that his future salary will be the same as todays, i.e. Xs = E[Xt |Fs ],
for s < t. This means that Xt is a martingale.
4
The concept of martingale was introduced by Levy in 1934.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 44
Chapter 3
This chapter deals with the most common used stochastic processes and their
basic properties. The two main basic processes are the Brownian motion and
the Poisson process. The other processes described in this chapter are derived
from the previous two. For more advanced topics on the Brownian motion,
the reader may consult Freedman [19], Hida [22], Knight [27], Karatzas and
Shreve [26], or Morters and Peres [34].
45
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 46
where we used that Bs is Fs -predictable (from where E[Bs |Fs ] = Bs ) and that
the increment Bt Bs is independent of previous values of Bt contained in the
information set Ft = (Bs ; s t).
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 47
s
2. Corr(Ws, Wt ) = .
t
1
These type of processes are called Markov processes.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 49
Cov(Ws , Wt ) = Cov(Ws , Ws + Wt Ws )
= Cov(Ws , Ws ) + Cov(Ws , Wt Ws )
= V ar(Ws ) + E[Ws (Wt Ws )] E[Ws ]E[Wt Ws ]
= s + E[Ws ]E[Wt Ws ]
= s,
since E[Ws ] = 0.
We can also arrive at the same result starting from the formula
Using that conditional expectations have the same expectation, factoring out
the predictable part, and using that Wt is a martingale, we have
so Cov(Ws , Wt ) = s.
2. The correlation formula yields
Cov(Ws , Wt ) s s
Corr(Ws , Wt ) = = = .
(Wt )(Ws ) s t t
Remark 3.1.8 Removing the order relation between s and t, the previous
relations can also be stated as
The following exercises state the translation and the scaling invariance
properties of the Brownian motion.
Exercise 3.1.9 For any t0 0, show that the process Xt = Wt+t0 Wt0 is a
Brownian motion. It can also be stated that the Brownian motion is translation
invariant.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 50
Exercise 3.1.11 Let 0 < s < t < u. Show the following multiplicative prop-
erty
Corr(Ws , Wt )Corr(Wt , Wu ) = Corr(Ws , Wu ).
3
0.5
1.0
a b
Figure 3.1: (a) Three simulations of the Brownian motion Wt . (b) Two sim-
ulations of the exponential Brownian motion eWt .
Exercise 3.1.20 Show that the following processes are Brownian motions
(a) Xt = WT WT t , 0 t T ;
(b) Yt = Wt , t 0.
ax2 +bx b2
e dx = e 4a , a>0
a
1
with a = 2t and b = .
1
2
e(ln x) /(2t) , if x > 0,
d
p(x) = F (x) = x 2t
dx Xt
0, elsewhere.
E[eWt Ws ] = e
ts
2 , s < t.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 53
Figure 3.2: To be used in the proof of formula (3.3.1); the area of the blocks
can be counted in two equivalent ways, horizontally and vertically.
1 2
(c) Show that for any constant c R, the process Yt = ecWt 2 c t is a
martingale.
Ws 1 + + Ws n
= n(Ws1 W0 ) + (n 1)(Ws2 Ws1 ) + + (Wsn Wsn1 )
= X1 + X2 + + Xn . (3.3.1)
This formula becomes clear if one sums the area of the blocks in Fig. 3.2
horizontally and then vertically. Since the increments of a Brownian motion
are independent and normally distributed, we have
X1 N 0, n2 s
X2 N 0, (n 1)2 s
X3 N 0, (n 2)2 s
..
.
Xn N 0, s .
Recall now the following well known theorem on the addition formula for
Gaussian random variables.
Theorem 3.3.1 If Xj are independent random variables normally distributed
with mean j and variance j2 , then the sum X1 + + Xn is also normally
distributed with mean 1 + + n and variance 12 + + n2 .
Then
n(n + 1)(2n + 1)
X1 + +Xn N 0, (1+22 +32 + +n2 )s = N 0, s ,
6
t
with s = . Using (3.3.1) yields
n
Ws + + Ws n (n + 1)(2n + 1)
t 1 N 0, t3 .
n 6n2
Taking the limit with n , we get
t3
Zt N 0, .
3
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 55
The mean and the variance can also be computed in a direct way as follows.
By Fubinis theorem we have
t t
E[Zt ] = E[ Ws ds] = Ws ds dP
0 R 0
t t
= Ws dP ds = E[Ws ] ds = 0,
0 R 0
where
Exercise 3.3.4 (a) Prove that the moment generating function of Zt is given
by
2 3
m(u) = eu t /6 .
(b) Use the rst part to nd the mean and variance of Zt .
Exercise 3.3.5 Let s < t. Show that the covariance of the integrated Brown-
ian motion is given by
t s
Cov Zs , Zt = s2 , s < t.
2 6
Exercise 3.3.6 Show that
(a) Cov(Zt , Zt Zth ) = 12 t2 h + o(h), where o(h) denotes a quantity such
that limh0 o(h)/h = 0;
t2
(b) Cov(Zt , Wt ) = .
2
Exercise 3.3.7 Show that
E[eWs +Wu ] = e
emin{s,u} .
u+s
2
t
Exercise 3.3.8 Consider the process Xt = eWs ds.
0
(a) Find the mean of Xt ;
(b) Find the variance of Xt .
10
4
8
6
2
2 2 4 6 8 10
2
4
4
a b
Figure 3.3: (a) Brownian bridge pinned down at 0 and 1. (b) Brownian motion
with drift Xt = t + Wt , with positive drift > 0.
2t3 t3
V ar(Vt ) = E[Vt2 ] E[Vt ]2 = e 3 e3
t+3s
Cov(Vs , Vt ) = e 2 .
(T t)3
Exercise 3.4.1 Show that E[VT |Ft ] = Vt e(T t)Wt + 3 for t < T .
This can also be stated by saying that the Brownian bridge tied at
0 and 1 is
a Gaussian process with mean 0 and variance t(1 t), so Xt N 0, t(1 t) .
E[Yt ] = t + E[Wt ] = t
and variance
V ar[Yt ] = V ar[t + Wt ] = V ar[Wt ] = t.
Exercise 3.6.1 Find the distribution and the density functions of the process
Yt .
10
Wt
5
Rt
10 5 0 5 10
5
10
2
2
n1 e 2t , 0;
(2t)n/2(n/2)
pt () =
0, <0
with
n
n ( 2 1)! for n even;
=
2
( n2 1)( n2 2) 32 12 , for n odd.
Proof: Since the Brownian motions W1 (t), . . . , Wn (t) are independent, their
joint density function is
1 2 2
fW1 Wn (x) = fW1 (x) fWn (x) = e(x1 ++xn )/(2t) , t > 0.
(2t)n/2
d (Sn1 ) n1 2
pt () = FR () = e 2t
d (2t)n/2
2
2 n1 2t
= e , > 0, t > 0.
(2t)n/2 (n/2)
1 x2
pt (x) = xe 2t , x > 0, t > 0.
t
2 2 2
1 < P (Rt t) < .
2t 4t 2t
Rt
Exercise 3.7.4 Let Xt = , t > 0, where Rt is a 2-dimensional Bessel
t
process. Show that Xt 0 as t in mean square.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 61
k (t s)k (ts)
P (Nt Ns = k) = e .
k!
It can be shown that condition 4 in the previous denition can be replaced by
the following two conditions:
where o(h) denotes a quantity such that limh0 o(h)/h = 0. Then the prob-
ability that a jump of size 1 occurs in the innitesimal interval dt is equal to
dt, and the probability that at least 2 events occur in the same small interval
is zero. This implies that the random variable dNt may take only two values,
0 and 1, and hence satises
P (dNt = 1) = dt (3.8.6)
P (dNt = 0) = 1 dt. (3.8.7)
2 4 6 8 10 12 14
Condition 4 also states that Nt has stationary increments. The fact that
Nt Ns is stationary can be stated as
n
(t)k
P (Nt+s Ns n) = P (Nt N0 n) = P (Nt n) = et .
k!
k=0
s
2. Corr(Ns, Nt ) = .
t
Proof: 1. Using (3.8.8) we have
Cov(Ns , Nt ) s s
Corr(Ns , Nt ) = 1/2
= = .
(V ar[Ns ]V ar[Nt ]) (st)1/2 t
E[Nt t|Fs ] = Ns s,
Exercise 3.8.6 Compute E[Nt2 |Fs ] for s < t. Is the process Nt2 an Fs -
martingale?
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 64
Exercise 3.8.7 (a) Show that the moment generating function of the random
variable Nt is
mNt (x) = et(e 1) .
x
E[Nt ] = t
E[Nt2 ] = 2 t2 + t
E[Nt3 ] = 3 t3 + 32 t2 + t
E[Nt4 ] = 4 t4 + 63 t3 + 72 t2 + t.
(c) Show that the rst few central moments are given by
E[Nt t] = 0
E[(Nt t)2 ] = t
E[(Nt t)3 ] = t
E[(Nt t)4 ] = 32 t2 + t.
Exercise 3.8.8 Find the mean and variance of the process Xt = eNt .
Exercise 3.8.9 (a) Show that the moment generating function of the random
variable Mt is
mMt (x) = et(e x1) .
x
E[Mt Ms ] = 0,
E[(Mt Ms )2 ] = (t s),
E[(Mt Ms )3 ] = (t s),
E[(Mt Ms )4 ] = (t s) + 32 (t s)2 .
V ar[(Mt Ms )2 ] = (t s) + 22 (t s)2 .
Proof: We start by noticing that the events {T1 > t} and {Nt = 0} are the
same, since both describe the situation that no events occurred until time t.
Then
P (T1 > t) = P (Nt = 0) = P (Nt N0 = 0) = et ,
and hence the distribution function of T1 is
called the integrated Poisson process. The next result provides a relation
between the process Ut and the partial sum of the waiting times Sk .
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 67
Figure 3.6: The Poisson process Nt and the waiting times S1 , S2 , Sn . The
area of the shaded rectangle is n(Sn+1 t).
Nt
Ut = tNt Sk .
k=1
Since Sn < t < Sn+1 , the dierence of the last two terms represents the area
of the last rectangle, which has the length t Sn and the height n. Using
associativity, a computation yields
where we replaced n by Nt .
The conditional distribution of the waiting times is provided by the fol-
lowing useful result.
Exercise 3.11.9 (a) Let Tk be the kth interarrival time. Show that
E[eTk ] = , > 0.
+
(Hint: If we know that there are exactly n jumps in the interval [0, T ], it makes
sense to consider the arrival time of the jumps Ti independent and uniformly
distributed on [0, T ]).
(d) Find the expectation
E eUt .
3.12 Submartingales
A stochastic process Xt on the probability space (, F, P ) is called a sub-
martingale with respect to the ltration Ft if:
(a) |Xt | dP < (Xt integrable);
The integrability follows from the inequality |Xt ()| t + |Wt ()| and in-
tegrability of Wt . The adaptability of Xt is obvious, and the last property
follows from the computation:
Example 3.12.2 We shall show that the square of the Brownian motion, Wt2 ,
is a submartingale.
E[Xt ]
P (sup Xs x) , x > 0.
st x
E[Xt+ ]
P (sup Xt x) ,
st x
supst |Ws |
Exercise 3.12.7 Show that p-lim = 0.
t t
Exercise 3.12.8 Show that for any martingale Xt we have the inequality
E[Xt2 ]
P (sup Xt2 > x) , x > 0.
st x
E[Xt ]
P (sup Xs x)
st x
E[Xt ]
P (Xt x) .
x
Exercise 3.12.9 Let Nt denote the Poisson process and consider the infor-
mation set Ft = {Ns ; s t}.
(a) Show that Nt is a submartingale;
(b) Is Nt2 a submartingale?
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 72
Exercise 3.12.10 It can be shown that for any 0 < < we have the in-
equality
N 2 4
t
E 2
t
t
Nt
Using this inequality prove that mslim = .
t t
The following famous inequality involving expectations was also found by
Doob. The proof can be found for instance in Chung and Williams [12].
Chapter 4
Properties of Stochastic
Processes
Example 4.1.1 Let Ft be the information available until time t regarding the
evolution of a stock. Assume the price of the stock at time t = 0 is $50 per
share. The following decisions are stopping times:
73
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 74
(a) Sell the stock when it reaches for the rst time the price of $100 per
share;
(b) Buy the stock when it reaches for the rst time the price of $10 per
share;
(c) Sell the stock at the end of the year;
(d) Sell the stock either when it reaches for the rst time $80 or at the end
of the year.
(e) Keep the stock either until the initial investment doubles or until the
end of the year;
The following decision is not a stopping time:
(f ) Sell the stock when it reaches the maximum level it will ever be.
Exercise 4.1.3 Let () = inf{t > 0; |Wt ()| K}, with K > 0 constant.
Show that is a stopping time with respect to the ltration Ft = (Ws ; s t).
The random variable is called the rst exit time of the Brownian motion Wt
from the interval (K, K). In a similar way one can dene the rst exit time
of the process Xt from the interval (a, b):
() = inf{t > 0; Xt ()
/ (a, b)} = inf{t > 0; Xt () b or Xt () a)}.
Let X0 < a. The rst entry time of Xt in the interval [a, b] is dened as
Exercise 4.1.4 Let Xt be a continuous stochastic process. Prove that the rst
exit time of Xt from the interval (a, b) is a stopping time.
Proposition 4.1.5 Let 1 and 2 be two stopping times with respect to the
ltration Ft . Then
1. 1 2
2. 1 2
3. 1 + 2
are stopping times.
Proof: 1. We have
{; 1 2 t} = {; 1 t} {; 2 t} Ft ,
1 c, 2 t c.
since
{; 1 c} Fc Ft , {; 2 t c} Ftc Ft .
It follows that 1 + 2 is a stopping time.
A ltration Ft is called right-continuous if Ft = Ft+ 1 , for t > 0.
n
n=1
This means that the information available at time t is a good approximation
for any future innitesimal information Ft+
; or, equivalently, nothing more
can be learned by peeking innitesimally far into the future. If denote by
Ft+ = Ft+ 1 , then the right-continuity can be written conveniently as
n
n=1
Ft = Ft+ .
s
Exercise 4.1.6 (a) Let Ft = {Ws ; s t} and Gt = { 0 Wu du; s t},
where Wt is a Brownian motion. Is Ft right-continuous? What about Gt ?
(b) Let Nt = {Ns ; s t}, where Nt is a Poisson motion. Is Nt right-
continuous?
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 76
The next result states that in the case of a right-continuous ltration the
inequality { t} from the denition of the stopping time can be replaced by
a strict inequality.
Exercise 4.1.10 Let be a stopping time and c > 0 a constant. Prove that
+ c is a stopping time.
In particular, E[Mt ] = E[M0 ], for any t > 0. The next result states necessary
conditions under which this identity holds if t is replaced by any stopping time
. The reader can skip the proof at the rst reading.
Theorem 4.2.1 (Optional Stopping Theorem) Let (Mt )t0 be a right con-
tinuous Ft -martingale and be a stopping time with respect to Ft such that
is bounded, i.e. N < such that N .
Then E[M ] = E[M0 ]. If Mt is an Ft -submartingale, then E[M ] E[M0 ].
Proof: Consider the following convenient notation for the indicator function
of a set
1, () > t;
1{ >t} () =
0, () t.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 78
M = M t + (M Mt )1{ >t} ,
Since M t is a martingale, see Exercise 4.2.4 (b), then E[M t ] = E[M0 ]. The
previous relation becomes
since for t > N the integrand vanishes. Hence relation (4.2.1) yields E[M ] =
E[M0 ].
It is worth noting that the previous theorem is a special case of the more
general Optional Stopping Theorem of Doob:
E[M |F ] = M a.s.
2.5
Wt
2.0
a
1.5
Ta
50 100 150 200 250 300 350
M = M t + (M Mt )1{ >t} ,
Lemma 4.3.1 Let Ta be the rst time the Brownian motion Wt hits a. Then
the distribution function of Ta is given by
2 y 2 /2
P (Ta t) = e dy.
2 |a|/ t
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 80
P (A) = P (A B) + P (A B)
= P (A|B)P (B) + P (A|B)P (B). (4.3.2)
If Ta > t, the Brownian motion did not reach the barrier a yet, so we must
have Wt < a. Therefore
P (Ta t) = 2P (Wt a)
2 x2 /(2t) 2 y 2 /2
= e dx = e dy.
2t a 2 a/ t
Remark 4.3.2 The previous proof is based on a more general principle called
the reection principle: If is a stopping time for the Brownian motion Wt ,
then the Brownian motion reected at is also a Brownian motion.
Theorem 4.3.3 Let a R be xed. Then the Brownian motion hits a (in a
nite amount of time) with probability 1.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 81
The previous result stated that the Brownian motion hits the barrier a
almost surely. The next result shows that the expected time to hit the barrier
is innite.
dFTa (t) a a2 3
p(t) = = e 2t t 2 , t > 0.
dt 2
(t)
d
1
One may use Leibnizs formula f (u) du = f ((t)) (t) f ((t)) (t).
dt (t)
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 82
0.4
0.3
0.2
0.1
2
a3
1 2 3 4
a2 a2
Using the inequality e 2t > 1 , t > 0, we have the estimation
2t
a 1 a3 1
E[Ta ] > dt 3/2
dt = , (4.3.4)
2 0 t 2 2 0 t
1
1
since dt is divergent and
0 t 0 t3/2 dt is convergent.
Remark 4.3.5 The distribution has a peak at a2 /3. Then if we need to pick
a small time interval [t dt, t + dt] in which the probability that the Brownian
motion hits the barrier a is maximum, we need to choose t = a2 /3, see Fig. 4.2.
Remark 4.3.6 Formula (4.3.4) states that the expected waiting time for Wt
to reach the barrier a is innite. However, the expected waiting time for the
Brownian motion Wt to hit either a or a is nite, see Exercise 4.3.9.
Exercise 4.3.8 Try to apply the proof of Lemma 4.3.1 for the following stochas-
tic processes
(a) Xt = t + Wt , with , > 0 constants;
t
(b) Xt = Ws ds.
0
Where is the diculty?
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 83
Exercise 4.3.10 (a) Show that the distribution function of the process
Xt = max Ws
s[0,t]
is given by
a/ t
2 2 /2
P (Xt a) = ey dy,
2 0
and the probability density is
2 x2
pt (x) = e 2t , x 0.
2t
2
(b) Show that E[Xt ] = 2t/ and V ar(Xt ) = t 1 .
Exercise 4.3.11 (a) Show that the probability density of the absolute value
of a Brownian motion Xt = |Wt |, t 0, is given by
2 x2
pt (x) = e 2t , x 0.
2t
(b) Consider the processes Xt = |Wt | and Yt = |Bt |, with Wt and Bt indepen-
dent Brownian motions. Use Theorem 2.11.1 to obtain the probability density
of the sum process Zt = Xt + Yt .
The fact that a Brownian motion returns to the origin or hits a barrier
almost surely is a property characteristic to the rst dimension only. The
next result states that in larger dimensions this is no longer possible.
D
(x0 ) = {x R2 ; |x x0 | },
Theorem 4.3.13 The 2-dimensional Brownian motion W (t) = W1 (t), W2 (t)
hits the disk D
(x0 ) with probability one.
Theorem 4.3.14 Let n > 2. The n-dimensional Brownian motion W (t) hits
the ball D
(x0 ) with probability
|x | 2n
0
P = < 1.
The previous results can be stated by saying that that Brownian motion
is transient in Rn , for n > 2. If n = 2 the previous probability equals 1. We
shall come back with proofs to the aforementioned results in a later chapter
(see section 9.6).
0.4
0.2
t1 t2
50 100 150 200 250 300 350
0.2
a
0.4
Wt
0.6
P (A) = P A {X = x}
x
P (A {X = x})
= P ({X = x})
x
P ({X = x})
= P (A|X = x)P (X = x).
x
(b) In the case when X is continuous, the sum is replaced by an integral and
the probability P ({X = x}) by fX (x)dx, where fX is the density function of
X.
The zero set of a Brownian motion Wt is dened by {t 0; Wt = 0}. Since
Wt is continuous, the zero set is closed with no isolated points almost surely.
The next result deals with the probability that the zero set does not intersect
the interval (t1 , t2 ).
Theorem 4.4.2 (The Arc-sine Law) The probability that a Brownian mo-
tion Wt does not have any zeros in the interval (t1 , t2 ) is equal to
2 t1
0, t1 t t2 ) = arcsin
P (Wt = .
t2
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 86
Proof: The proof follows Ross [43]. Let A(a; t1 , t2 ) denote the event that the
Brownian motion Wt takes on the value a between t1 and t2 . In particular,
A(0; t1 , t2 ) denotes the event that Wt has (at least) a zero between t1 and
t2 . Substituting A = A(0; t1 , t2 ) and X = Wt1 into the formula provided by
Proposition 4.4.1
P (A) = P (A|X = x)fX (x) dx
yields
P A(0; t1 , t2 ) = P A(0; t1 , t2 )|Wt1 = x fWt (x) dx (4.4.5)
1
1 x2
= P A(0; t1 , t2 )|Wt1 = x e 2t1 dx.
2t1
Using the properties of Wt with respect to time translation and symmetry we
have
P A(0; t1 , t2 )|Wt1 = x = P A(0; 0, t2 t1 )|W0 = x
= P A(x; 0, t2 t1 )|W0 = 0
= P A(|x|; 0, t2 t1 )|W0 = 0
= P A(|x|; 0, t2 t1 )
= P T|x| t2 t1 ,
the last identity stating that Wt hits |x| before t2 t1 . Using Lemma 4.3.1
yields
2 y
2
2 t1
P A(0; t1 , t2 ) = 1 arcsin .
t2
Using P (Wt = 0, t1 t t2 ) = 1 P A(0; t1 , t2 ) we obtain the desired
result.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 87
1 x2 2 t1
2(t yt ) 2t
e 2 1 1 dydx = 1 arcsin .
t1 (t2 t1 ) 0 |x| t2
Exercise 4.4.5 Find the probability that a Brownian motion Wt does not take
the value a in the interval (t1 , t2 ).
We provide below without proof a few similar results dealing with arc-sine
probabilities, whose proofs can be found for instance in Kuo [30]. The rst
result deals with the amount of time spent by a Brownian motion on the
positive half-axis.
t
Theorem 4.4.7 (Arc-sine Law of L evy) Let L+ +
t = 0 sgn Ws ds be the
amount of time a Brownian motion Wt is positive during the time interval
[0, t]. Then
+ 2
P (Lt ) = arcsin .
t
The next result deals with the Arc-sine Law for the last exit time of a
Brownian motion from 0.
Theorem 4.4.8 (Arc-sine Law of exit from 0) Let t = sup{0 s
t; Ws = 0}. Then
2
P (t ) = arcsin , 0 t.
t
The Arc-sine Law for the time the Brownian motion attains its maximum
on the interval [0, t] is given by the next result.
Theorem 4.4.9 (Arc-sine Law of maximum) Let Mt = max Ws and de-
0st
ne
t = sup{0 s t; Ws = Mt }.
Then
2 s
P (t s) = arcsin , 0 s t, t > 0.
t
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 88
e2 1
P (Xt goes up to before down to ) = .
e2 e2
P (Wt hits ) = 1.
If = we obtain
1
P (Wt goes up to before down to ) = ,
2
e2 1
P (T T ) = .
e2 e2
P (Xt goes up to ) = 1.
P (sup(Wt + t) ) = 1, 0,
t0
P (sup(Wt + t) ) = e2 , < 0,
t0
or
P (sup(Wt t) ) = e2 , > 0,
t0
which is known as one of the Doobs inequalities. This can also be described
in terms of stopping times as follows. Dene the stopping time
= inf{t > 0; Wt t }.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 90
Using
P ( < ) = P sup(Wt t)
t0
P ( < ) = e2 , > 0,
P ( < ) = 1, 0.
e2 + e2
E[XT ] = .
e2 e2
(b) Find E[XT2 ];
(c) Compute V ar(XT ).
The next result deals with the time one has to wait (in expectation) for
the process Xt = t + Wt to reach either or .
Proposition 4.5.5 The expected value of T is
e2 + e2
E[T ] = .
(e2 e2 )
Proof: Using that Wt is a martingale, with E[Wt ] = E[W0 ] = 0, applying the
Optional Stopping Theorem, Theorem 4.2.1, yields
E[XT ] e2 + be2
E[T ] = = .
(e2 e2 )
Proposition 4.5.9 The expected waiting time for Nt to reach the barrier a
is E[ ] = a .
Proposition 4.6.1 The moments of the rst hitting time are all innite
E[ n ] = , n 1.
dn 2sx
E[ n ] = (1)n e
.
dsn
s=0
dn 2sx Mk xnk n1
n 2sx
e = (1) e ,
dsn 2rk /2 s(n+k)/2
k=0
d 2sx
2sx x
E[ ] = e
= lim e = +.
ds
s0 +
2 2sx
s=0
Another application involves the inverse Laplace transform to get the prob-
ability density. This way we can retrieve the result of Proposition 4.3.4.
|x| x2
p(t) = e 2t , t > 0. (4.6.9)
2t3
Proof: Let x > 0. The expectation
E[es ] = es p( ) d = L{p( )}(s)
0
E[etX ] E[esX ]
P ( ) t
= s
= esx 2s , s > 0.
e e
Then P ( ) emins>0 f (s) , where f (s) = s x 2s. Since f (s) = x2s ,
x 2
then f (s) reaches its minimum at the critical point s0 = 2 2 . The minimum
value is
x2
min f (s) = f (s0 ) = .
s>0 2
Substituting in the previous inequality leads to the required result.
The case of Brownian motion with drift Consider the Brownian motion
with drift Xt = t + Wt , with , > 0. Let
= inf{t > 0; Xt x}
denote the rst hitting time of the barrier x, with x > 0. We shall compute
the distribution of the random variable and its rst two moments.
Applying the Optional Stopping Theorem (Theorem 4.2.1) to the martin-
1 2
gale Mt = ecWt 2 c t yields
E[M ] = E[M0 ] = 1.
c 1 2
Substituting s = + c and completing to a square yields
2
2 2
2s + = c + .
2
Solving for c we get the solutions
2 2
c = + 2s + 2 , c= 2s + .
2
Assume c < 0. Then substituting the second solution into (4.6.10) yields
1
2 2
E[es ] = e 2 (+ 2s + )x .
1
s 2 (+ 2s2 +2 )x
This relation is contradictory since e < 1 while e > 1,
where we used that s, x, > 0. Hence it follows that c > 0. Substituting the
rst solution into (4.6.10) leads to
1
s ( 2s2 +2 )x
E[e ] = e 2 .
x (x )2
p( ) = e 2 2 , > 0. (4.6.12)
2 3/2
x x 2
E[ ] = , V ar( ) = .
3
d d 1
E[es ]
= e 2 ( 2s + )x
2 2
E[ ] =
ds ds
s=0
s=0
x 1
( 2s2 +2 )x
= e 2
2 + 2s s=0
x
= .
d2 s
d2 12 ( 2s2 +2 )x
E[ 2 ] = (1)2 E[e ]
= e
ds2 ds2
s=0
s=0
x( 2 + x 2 + 2s 2 ) 12 ( 2s2 +2 )x
= e
(2 + 2s 2 )3/2 s=0
x 2 x2
= + 2.
3
Hence
x 2
V ar( ) = E[ 2 ] E[ ]2 = .
3
It is worth noting that we can arrive at the formula E[ ] = x in the following
heuristic way. Taking the expectation in the equation + W = x yields
E[ ] = x, where we used that E[W ] = 0 for any nite stopping time
(see Exercise 4.5.8 (a)). Solving for E[ ] yields the aforementioned formula.
Even if the computations are more or less similar to the previous result,
we shall treat next the case of the negative barrier in its full length. This
is because of its particular importance in being useful in practical problems,
such as pricing perpetual American puts.
(a) We have
1
s (+ 2s2 +2 )x
E[e ]=e 2 , s > 0. (4.6.13)
(b) Then the density function of is given by
x (x+ )2
p( ) = e 2 2 , > 0. (4.6.14)
2 3/2
(c) The mean of is
x 2x
E[ ] = e 2 .
Proof: (a) Consider the stopping time = inf{t > 0; Xt = x}. By the
Optional Stopping Theorem (Theorem 4.2.1) applied to the martingale Mt =
c2
ecWt 2 t yields
c2 c c2
1 = M0 = E[M ] = E ecW 2 = E e (X ) 2
c c c2 c c2
= E e x 2 = e x E e( + 2 ) .
c
Therefore
c c2
E e( + 2 ) = e x .
c
(4.6.15)
2 2
If let s = c
+ 2 , then solving for c yields c =
c
2s + 2 , but only the
negative solution works out; this comes from the fact that both terms of the
equation (4.6.15) have to be less than 1. Hence (4.6.15) becomes
1
2 2
E[es ] = e 2 (+ 2s + )x , s > 0.
d 2 x 2 x 2x
E[ ] = E[es ]
= e 2 x 2 = e 2 .
ds s=0 2
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 97
Exercise 4.6.7 Assume the hypothesis of Proposition 4.6.6 are satised. Find
V ar( ).
Exercise 4.6.8 Find the modes of distributions (4.6.12) and (4.6.14). What
do you notice?
Exercise 4.6.10 Does 4t + 2Wt hit 9 faster (in expectation) than 5t + 3Wt
hits 14?
Exercise 4.6.11 Let be the rst time the Brownian motion with drift Xt =
t + Wt hits x, where , x > 0. Prove the inequality
x2 +2 2
P ( ) e 2
+x
, > 0.
The double barrier case In the following we shall consider the case of double
barrier. Consider the Brownian motion with drift Xt = t + Wt , > 0. Let
, > 0 and dene the stopping time
T = inf{t > 0; Xt or Xt }.
1 2 1
E[e(c+ 2 c )T
]=
ec p + ec (1 p )
If we substitute s = c + 12 c2 , then
1
E[esT ] = (4.6.16)
e(+ 2s+2 ) p + e(+ 2s+2 ) (1 p )
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 98
The probability density of the stopping time T is obtained by taking the inverse
Laplace transform of the right side expression
1
p(T ) = L1 2
2
( ),
e(+ 2s+ ) p + e(+ 2s+ ) (1 p )
an expression which is not feasible for having a closed form solution. However,
expression (4.6.16) would be useful for computing the price for double barrier
derivatives.
2.5
2b 2.0
Wt
2ba
1.5
b 1.0
0.5
a
O Wt
0.2 0.4
Tb 0.6 0.8
t 1.0
0.5
= I1 I2 ,
with
I1 = f (x, y) dx dy, I2 = f (x, y) dx dy.
{yxu,y0} {0yx}
This writes the integral over a strip as a dierence of integrals over the interior
of two angles, see Fig. 4.5(a). A great simplication of computation is done
by observing that the second integral vanishes
x
I2 = f (x, y) dy dx
0 0
x
2 (2yx)2
= (2y x)e 2t dy dx
2t3/2 0
x0
1 z2
= ze 2t dz dx = 0,
2t3/2 0 x
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 101
3.0 3.0
2.5 2.5
A
2.0 2.0
1.5 1.5
u yx
y xu
1.0
y x u 2 1.0
0.5
yx 0.5
C O B
1.0 0.5 0.5 1.0 1.0 0.5 0.5 1.0
u O
u u
0.5 0.5
a b
Figure 4.5: (a) The integration strip for the proof of Levys theorem. (b) The
domains DAOC , DAOB , and DABC for the proof of Pitmans theorem.
mt = min Ws .
0st
DAOC = {2y x u, y 0, x y}
DAOB = {0 x, x y}
DABC = {2y x u, y 0}.
We note rst that the second integral vanishes, as an integral of an odd func-
tion over a symmetric interval
u x
2 (2yx)2
I2 (u) = (2y x)e 2t dy dx
0 0 2t3/2
u x
1 z2
= ze 2t dz dx = 0.
2t3/2 0 x
u (x+u)/2
2 (2yx)2
I1 (u) = (2y x)e 2t dy dx
u 0 2t3/2
u u u u2
1 z2 1
e 2t dr dx
r
= ze 2t dz dx =
2t3/2 2 2t3/2 u x2
u
u
x
2 2
t
e 2t e 2t dx
x u
=
2t3/2 u
u
1 x2 2 u2
= e 2t dx ue 2t
2t u 2t
u 2
2 u2
e 2t dx ue 2t .
x
=
2t 0
d 2 u2
P (Zt u) = u2 e 2t ,
du 2 t3/2
lim Xt () = X().
t
Mean Square Limit We say that the process Xt converges to X in the mean
square if
lim E[(Xt X)2 ] = 0.
t
It is worthy to note that the previous statement holds true if the limit to
innity is replaced by a limit to any other number.
Next we shall provide a few applications.
Application 4.9.2 If > 1/2, then
Wt
ms-lim = 0.
t t
Wt E[Wt ]
Proof: Let Xt =
. Then E[Xt ] = = 0, and
t t
1 t 1
V ar[Xt ] = 2 V ar[Wt ] = 2 = 21 ,
t t t
1
for any t > 0. Since 21 0 as t , applying Proposition 4.9.1 yields
t
Wt
ms-lim = 0.
t t
Wt
Corollary 4.9.3 We have ms-lim = 0.
t t
t
Application 4.9.4 Let Zt = 0 Ws ds. If > 3/2, then
Zt
ms-lim = 0.
t t
Zt E[Zt ]
Proof: Let Xt =
. Then E[Xt ] = = 0, and
t t
1 t3 1
V ar[Xt ] = 2 V ar[Zt ] = 2 = 23 ,
t 3t 3t
1
for any t > 0. Since 23 0 as t , applying Proposition 4.9.1 leads
3t
to the desired result.
V ar[eWt ] e2t et 1 1 1
V ar[Xt ] = = = 0,
t2p e2ct t2p e2ct t2p e2t(c1) et(2c1)
as t , Proposition 4.9.1 leads to the desired result.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 106
max Ws
0st
Proof: Let Xt = . Since by Exercise 4.3.10
t
E[ max Ws ] = 0
0st
V ar( max Ws ) = 2 t,
0st
then
E[Xt ] = 0
2 t
V ar[Xt ] = 0, t .
t2
Apply Proposition 4.9.1 to get the desired result.
Remark 4.9.7 One of the strongest results regarding limits of Brownian mo-
tions is called the law of iterated logarithms and was rst proved by Lamperti:
Wt
lim sup = 1,
t 2t ln(ln t)
almost certainly.
Exercise 4.9.8 Find a stochastic process Xt such that the following both con-
ditions are satised:
(i) ms-lim Xt = 0
t
(ii) ms-lim Xt2 = 0.
t
Since E[|Xn |]2 E[Xn2 ], the condition (4.10.21) can be replaced by its stronger
version
M > 0 such that E[Xn2 ] M, n 1.
The following result deals with the continuous version of theMartingale
Convergence Theorem. Denote the innite knowledge by F = t Ft .
(k)
n1
X, Xt () = p- lim |Xti+1 () Xti ()|k ,
maxi |ti+1 ti |
i=0
It can be shown that the quadratic variation exists and is unique (up to in-
distinguishability) for continuous square integrable martingales Xt , i.e. mar-
tingales satisfying X0 = 0 a.s. and E[Xt2 ] < , for all t 0. Furthermore,
the quadratic variation, X, Xt , of a square integrable martingale Xt is an
increasing continuous process satisfying
(i) X, X0 = 0;
(ii) Xt2 X, Xt is a martingale.
Next we introduce a symmetric and bilinear operation.
Denition 4.11.2 The quadratic covariation of two continuous square inte-
grable martingales Xt and Yt is dened as
1
X, Y t = X + Y, X + Y t X Y, X Y t .
4
Exercise 4.11.3 Prove that:
(a) X, Y t = Y, Xt ;
(b) aX + bY, Zt = aX, Zt + bY, Zt .
Exercise 4.11.5 Prove that the total variation on the interval [0, t] of a Brow-
nian motion is innite a.s.
n1
ms- lim (Wti+1 Wti )2 = T. (4.11.22)
n
i=0
n1
Xn = (Wti+1 Wti )2 .
i=0
n1
n1
E[Xn ] = E[(Wti+1 Wti )2 ] = (ti+1 ti )
i=0 i=0
= tn t0 = T ;
n1
n1
2
V ar(Xn ) = V ar[(Wti+1 Wti ) ] = 2(ti+1 ti )2
i=0 i=0
T 2 2T 2
= n2 = ,
n n
where we used that the partition is equidistant. Since Xn satises the condi-
tions
E[Xn ] = T, n 1;
V ar[Xn ] 0, n ,
n1
ms- lim (Wti+1 Wti )2 = T. (4.11.23)
n
i=0
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 110
Exercise 4.11.7 Prove that the quadratic variation of the Brownian motion
Wt on [a, b] is equal to b a.
n1
ms- lim (Wti+1 Wti )2 = T. (4.11.24)
n
i=0
while the left side can be regarded as a stochastic integral with respect to dWt2
T
n1
2
(dWt ) = ms- lim (Wti+1 Wti )2 .
0 n
i=0
dWt2 = dt.
In fact, this expression also holds in the mean square sense, as it can be inferred
from the next exercise.
Roughly speaking, the process dWt2 , which is the square of innitesimal in-
crements of a Brownian motion, is deterministic. This relation plays a central
role in Stochastic Calculus and will be useful when dealing with Itos lemma.
The following exercise states that dt dWt = 0, which is another important
stochastic relation useful in Itos lemma.
Exercise 4.11.9 Consider the equidistant partition 0 = t0 < t1 < tn1 <
tn = T . Show that
n1
ms- lim (Wti+1 Wti )(ti+1 ti ) = 0. (4.11.25)
n
i=0
n1
V (Wt ) = sup |Wtk+1 Wtk |,
tk
k=0
for all partitions 0 = t0 < t1 < < tn1 < tn = T . Without losing generality,
we may assume the partition is equidistant, i.e. tk+1 tk = Tn . Equivalently,
V (Wt ) = lim Yn ,
n
where
n1
Yn = |Wtk+1 Wtk |.
k=0
Using Exercise 3.1.15 and the independent increments of the Brownian motion
provides the mean and variance of the random variable Yn
n1
2
n1
2nT
= E[Yn ] = E[|Wtk+1 Wtk |] = (tk+1 tk ) =
k=0 k=0
n1 2
n1
2 = V ar(Yn ) = V ar[|Wtk+1 Wtk |] = 1 (tk+1 tk )
k=0 k=0
2
= 1 T.
Since
{; Yn () < k} {; |Yn () | > k}
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 112
1
P (Yn < k) P (|Yn | > k) < , k 1.
k2
This states that the probability of the left tail is smaller than the probability
of both tails, the second being bounded by k12 . Using the probability of the
complement event, the foregoing relation implies
1
P (Yn k) 1 , k 1.
k2
Substituting for and yields
2nT 2 1
P Yn k 1 T 1 2, k 1.
k
Considering k = n, we get
1
P Yn C n 1 , n 1,
n
where
2T 2
C= 1 T > 0.
Then for any constant M > 0, there is an integer n such that C n M and
hence 1
P (Yn M ) P Yn C n 1 , n n0 .
n
Taking the limit over n yields
lim P (Yn M ) = 1.
n
P (V (Wt ) = ) = 1.
The rest of this chapter deals with similar properties regarding quadratic
variation of compensated Poisson process and can be skipped at a rst reading.
Proposition 4.12.1 Let a < b and consider the partition a = t0 < t1 < <
tn1 < tn = b. Then
n1
ms lim (Mtk+1 Mtk )2 = Nb Na , (4.12.26)
n 0
k=0
Proof: For the sake of simplicity we shall use the following notations
tk = tk+1 tk , Mk = Mtk+1 Mtk , Nk = Ntk+1 Ntk .
The relation we need to prove can also be written as
n1
ms-lim (Mk )2 Nk = 0.
n
k=0
Let
Yk = (Mk )2 Nk = (Mk )2 Mk tk .
It suces to show that
n1
E Yk = 0, (4.12.27)
k=0
n1
lim V ar Yk = 0. (4.12.28)
n
k=0
The rst identity follows from the properties of Poisson processes (see Exercise
3.8.9)
n1
n1
n1
E Yk = E[Yk ] = E[(Mk )2 ] E[Nk ]
k=0 k=0 k=0
n1
= (tk tk ) = 0.
k=0
For the proof of the identity (4.12.28) we need to rst nd the variance of Yk .
V ar[Yk ] = V ar[(Mk )2 (Mk + tk )] = V ar[(Mk )2 Mk ]
= V ar[(Mk )2 ] + V ar[Mk ] 2Cov[Mk2 , Mk ]
= tk + 22 t2k + tk
2 E[(Mk )3 ] E[(Mk )2 ]E[Mk ]
= 22 (tk )2 ,
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 114
where we used Exercise 3.8.9 and the fact that E[Mk ] = 0. Since Mt is a
process with independent increments, then Cov[Yk , Yj ] = 0 for i = j. Then
n1
n1
n1
V ar Yk = V ar[Yk ] + 2 Cov[Yk , Yj ] = V ar[Yk ]
k=0 k=0 k=j k=0
n1
n1
2 2 2
= 2 (tk ) 2 n tk = 22 (b a)n ,
k=0 k=0
n1
and hence V ar Yn 0 as n 0. According to the Proposition
k=0
2.14.1, we obtain the desired limit in mean square.
The previous result states that the quadratic variation of the martingale
Mt between a and b is equal to the jump of the Poisson process between a and
b.
The Relation dMt2 = dNt Recall relation (4.12.26)
n1
ms- lim (Mtk+1 Mtk )2 = Nb Na . (4.12.29)
n
k=0
The right side can be regarded as a Riemann-Stieltjes integral
b
Nb Na = dNt ,
a
while the left side can be regarded as a stochastic integral with respect to
(dMt )2
b
n1
(dMt )2 := ms- lim (Mtk+1 Mtk )2 .
a n
k=0
Substituting in (4.12.29) yields
b b
(dMt )2 = dNt ,
a a
1. E[Xn ] = 0;
2. lim V ar[Xn ] = 0.
n
n1
n1
E[Xn ] = E tk Mk = tk E[Mk ] = 0.
k=0 k=0
Since the Poisson process Nt has independent increments, the same property
holds for the compensated Poisson process Mt . Then tk Mk and tj Mj
are independent for k = j, and using the properties of variance we have
n1
n1
n1
2
V ar[Xn ] = V ar tk Mk = (tk ) V ar[Mk ] = (tk )3 ,
k=0 k=0 k=0
where we used
n1
n1
V ar[Xn ] = (tk )3 n 2 tk = (b a)n 2 0
k=0 k=0
dt dMt = 0. (4.12.32)
Since the Brownian motion Wt and the process Mt have independent incre-
ments and Wk is independent of Mk , we have
n1
n1
E[Yn ] = E[Wk Mk ] = E[Wk ]E[Mk ] = 0,
k=0 k=0
Chapter 5
Stochastic Integration
This chapter deals with one of the most useful stochastic integrals, called the
Ito integral. This type of integral was introduced in 1944 by the Japanese
mathematician Ito [24], [25], and was originally motivated by a construction
of diusion processes. We shall keep the presentation to a maximum sim-
plicity, integrating with respect to a Brownian motion or Poisson process only.
The reader interested in details regarding a larger class of integrators may con-
sult Protter [40] or Kuo [30]. For a more formal introduction into stochastic
integration see Revuz and Yor [41].
Here is a motivation
b for studying an integral of stochastic type. The Rie-
mann integral a F (x) dx represents the work done by the force F between
positions x = a and x = b. The element F (x) dx represents the work done by
the force for the innitesimal displacement dx. Similarly, F (t) dWt represents
the work done by F during an innitesimal b Brownian jump dWt . The cum-
mulative eect is described by the object a F (t) dWt , which will be studied
in this chapter. This represents the work eect of the force F done along
the trajectory of a particle modeled by a Brownian motion during the time
interval [a, b].
117
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 118
1
A 2 -distributed random variable with n degrees of freedom has mean n and variance
2n.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 119
The role of the previous condition will be made more clear when we discuss
the martingale property of the Ito integral, see Proposition 5.5.7. Divide the
interval [a, b] into n subintervals using the partition points
a = t0 < t1 < < tn1 < tn = b,
and consider the partial sums
n1
Sn = Fti (Wti+1 Wti ).
i=0
We emphasize that the intermediate points are the left endpoints of each
interval, and this is the way they should always be chosen. Since the process
Ft is nonanticipative, the random variables Fti and Wti+1 Wti are always
independent; this is an important feature in the denition of the Ito integral.
The Ito integral is the limit of the partial sums Sn
b
ms-lim Sn = Ft dWt ,
n a
provided the limit exists. It can be shown that the choice of partition does
not inuence the value of the Ito integral. This is the reason why, for practical
purposes, it suces to assume the intervals equidistant, i.e.
(b a)
ti+1 ti = , i = 0, 1, , n 1.
n
The previous convergence is taken in the mean square sense, i.e.
b 2
lim E Sn Ft dWt = 0.
n a
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 120
n1
n1
Sn = Fti (Wti+1 Wti ) = c(Wti+1 Wti )
i=0 i=0
= c(Wb Wa ),
Since
1
xy = [(x + y)2 x2 y 2 ],
2
letting x = Wti and y = Wti+1 Wti yields
1 2 1 1
Wti (Wti+1 Wti ) = W W 2 (Wti+1 Wti )2 .
2 ti+1 2 ti 2
Then after pair cancelations the sum becomes
1 2 1 2 1
n1 n1 n1
Sn = Wti+1 Wti (Wti+1 Wti )2
2 2 2
i=0 i=0 i=0
1 2 1
n1
= W (Wti+1 Wti )2 .
2 tn 2
i=0
Using tn = T , we get
1
n1
1
Sn = WT2 (Wti+1 Wti )2 .
2 2
i=0
Since the rst term on the right side is independent of n, using Proposition
4.11.6, we have
1
n1
1 2
ms- lim Sn = WT ms- lim (Wti+1 Wti )2 (5.4.2)
n 2 n 2
i=0
1 2 1
= W T. (5.4.3)
2 T 2
We have now obtained the following explicit formula of a stochastic inte-
gral:
T
1 1
Wt dWt = WT2 T.
0 2 2
In a similar way one can obtain
b
1 1
Wt dWt = (W 2 Wa2 ) (b a).
a 2 b 2
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 122
It is worth noting that the right side contains random variables depending on
the limits of integration a and b.
2. Homogeneity:
T T
cf (Wt , t) dWt = c f (Wt , t) dWt .
0 0
3. Partition property:
T u T
f (Wt , t) dWt = f (Wt , t) dWt + f (Wt , t) dWt , 0 < u < T.
0 0 u
n1
Xn = f (Wti , ti )(Wti+1 Wti )
i=0
n1
Yn = g(Wti , ti )(Wti+1 Wti ).
i=0
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 123
T T
Since ms-lim Xn = f (Wt , t) dWt and ms-lim Yn = g(Wt , t) dWt , using
n 0 n 0
Proposition 2.15.2 yields
T
f (Wt , t) + g(Wt , t) dWt
0
n1
= ms-lim f (Wti , ti ) + g(Wti , ti ) (Wti+1 Wti )
n
i=0
n1
n1
= ms-lim f (Wti , ti )(Wti+1 Wti ) + g(Wti , ti )(Wti+1 Wti )
n
i=0 i=0
= ms-lim (Xn + Yn ) = ms-lim Xn + ms-lim Yn
n n n
T T
= f (Wt , t) dWt + g(Wt , t) dWt .
0 0
The proofs of parts 2 and 3 are left as an exercise for the reader.
3. Covariance:
b b b
E f (Wt , t) dWt g(Wt , t) dWt = E f (Wt , t)g(Wt , t) dt .
a a a
We shall discuss the previous properties giving rough reasons of proof. The
detailed proofs are beyond the goal of this book.
b
1. The Ito integral I = a f (Wt , t) dWt is the mean square limit of the partial
sums Sn = n1 i=0 fti (Wti+1 Wti ), where we denoted fti = f (Wti , ti ). Since
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 124
because the increments have mean zero. Applying the Squeeze Theorem in
the double inequality
2
0 E[Sn I] E[(Sn I)2 ] 0, n
yields E[Sn ] E[I] 0. Since E[Sn ] = 0 it follows that E[I] = 0, i.e. the Ito
integral has zero mean.
2. Since the square of the sum of partial sums can be written as
n1
2
Sn2 = fti (Wti+1 Wti )
i=0
n1
= ft2i (Wti+1 Wti )2 + 2 fti (Wti+1 Wti )ftj (Wtj+1 Wtj ),
i=0 i=j
n1
= E[ft2i ](ti+1 ti ),
i=0
b b
which are the Riemann sums of the integral E[ft2 ] dt =E ft2 dt , where
a a
the last identity follows from Fubinis theorem. Hence E[Sn2 ] converges to the
aforementioned integral.
3. Consider the partial sums
n1
n1
Sn = fti (Wti+1 Wti ), Vn = gtj (Wtj+1 Wtj ).
i=0 j=0
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 125
Their product is
n1
n1
Sn Vn = fti (Wti+1 Wti ) gtj (Wtj+1 Wtj )
i=0 j=0
n1
n1
= fti gti (Wti+1 Wti )2 + fti gtj (Wti+1 Wti )(Wtj+1 Wtj ).
i=0 i=j
it follows that
n1
E[Sn Vn ] = E[fti gti ]E[(Wti+1 Wti )2 ]
i=0
n1
= E[fti gti ](ti+1 ti ),
i=0
b
which is the Riemann sum for the integral E[ft gt ] dt. a
b
From 1 and 2 it follows that the random variable a f (Wt , t) dWt has mean
zero and variance
b b
V ar f (Wt , t) dWt = E f (Wt , t)2 dt .
a a
Proof: It follows from the previous theorem and from the correlation formula
|Cov(X, Y )|
|Corr(X, Y )| = 1.
[V ar(X)V ar(Y )]1/2
Let Ft be the information set at time t. This implies that fti and Wti+1
Wti are known at time t, for any ti+1 t. It follows that the partial sum
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 126
n1
Sn = fti (Wti+1 Wti ) is Ft -measurable. The following result, whose proof
i=0
is omitted for technical reasons, states that this is also valid after taking the
limit in the mean square:
t
Proposition 5.5.4 The Ito integral 0 fs dWs is Ft -measurable.
The following two results state that if the upper limit of an Ito integral is
replaced by the parameter t we obtain a continuous martingale.
t
Proposition 5.5.6 Consider the process Xt = 0 f (Ws , s) dWs . Then Xt is
continuous, i.e. for almost any state of the world , the path t Xt ()
is continuous.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 127
Proof: A rigorous proof is beyond the purpose of this book. We shall provide
just a rough sketch. Assume the process f (Wt , t) satises E[f (Wt , t)2 ] < M ,
for some M > 0. Let t0 be xed and consider h > 0. Consider the increment
Yh = Xt0 +h Xt0 . Using the aforementioned properties of the Ito integral we
have
t0 +h
E[Yh ] = E[Xt0 +h Xt0 ] = E f (Wt , t) dWt = 0
t0
t0 +h 2 t0 +h
E[Yh2 ] = E f (Wt , t) dWt = E[f (Wt , t)2 ] dt
t0 t0
t0 +h
< M dt = M h.
t0
The process Yh has zero mean for any h > 0 and its variance tends to 0 as
h 0. Using a convergence theorem yields that Yh tends to 0 in mean square,
as h 0. This is equivalent to the continuity of Xt at t0 .
t
Proposition 5.5.7 Let Xt = 0 f (Ws , s) dWs , with E 0 f 2 (Ws , s) ds <
. Then Xt is a continuous Ft -martingale.
and then from the inequality E[|Xt |]2 E[Xt2 ] we obtain E[|Xt |] < , for all
t 0.
Measurability: Xt is Ft -measurable from Proposition 5.5.4.
Forecast: E[Xt |Fs ] = Xs for s < t by Proposition 5.5.5.
Continuity: See Proposition 5.5.6.
All properties of Ito integrals also hold for Wiener integrals. The Wiener
integral is a random variable with zero mean
b
E f (t) dWt = 0
a
and variance
b 2 b
E f (t) dWt = f (t)2 dt.
a a
However, in the case of Wiener integrals we can say something more about
their distribution.
b
Proposition 5.6.1 The Wiener integral I(f ) = a f (t) dWt is a normal ran-
dom variable with mean 0 and variance
b
V ar[I(f )] = f (t)2 dt := f 2L2 .
a
Proof: Since increments Wti+1 Wti are normally distributed with mean 0
and variance ti+1 ti , then
f (ti )(Wti+1 Wti ) N (0, f (ti )2 (ti+1 ti )).
Since these random variables are independent, by Theorem 3.3.1, their sum is
also normally distributed, with
n1 n1
Sn = f (ti )(Wti+1 Wti ) N 0, f (ti )2 (ti+1 ti ) .
i=0 i=0
T 1
Exercise 5.6.3 Show that the random variable X = dWt is normally
1 t
distributed with mean 0 and variance ln T .
T
Exercise 5.6.4 Let Y = 1 t dWt . Show that Y is normally distributed
with mean 0 and variance (T 2 1)/2.
t
Exercise 5.6.5 Find the distribution of the integral 0 ets dWs .
t t
Exercise 5.6.6 Show that Xt = 0 (2t u) dWu and Yt = 0 (3t 4u) dWu are
Gaussian processes with mean 0 and variance 73 t3 .
t
1
Exercise 5.6.7 Show that ms- lim u dWu = 0.
t0 t 0
t bu
Exercise 5.6.8 Find all constants a, b such that Xt = 0 a + t dWu is
normally distributed with variance t.
n1
Sn = Fti (Mti+1 Mti ),
i=0
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 130
where Fti is the left-hand limit at ti . We note that the intermediate points
are the left-handed limit to the endpoints of each interval. Since the process Ft
is nonanticipative, the random variables Fti and Mti+1 Mti are independent.
provided the limit exists. More precisely, this convergence means that
b 2
lim E Sn Ft dMt = 0.
n a
b
Exercise 5.7.1 Let c be a constant. Show that c dMt = c(Mb Ma ).
a
n1
Sn = Mti (Mti+1 Mti ).
i=0
1
Using xy = [(x + y)2 x2 y 2 ], by letting x = Mti and y = Mti+1 Mti ,
2
we get
1 1 1
Mti (Mti+1 Mti ) = (Mti+1 Mti + Mti )2 Mt2i (Mti+1 Mti )2 .
2 2 2
Let J be the set of jump instances between 0 and T . Using that Mti = Mti
for ti
/ J, and Mti = 1 + Mti for ti J yields
Mti+1 , if ti
/J
Mti+1 Mti + Mti =
Mti+1 1, if ti J.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 131
Splitting the sum, canceling in pairs, and applying the dierence of squares
formula we have
1 1 2 1
n1 n1 n1
Sn = (Mti+1 Mti + Mti )2 Mti (Mti+1 Mti )2
2 2 2
i=0 i=0 i=0
1 1 1 1 2
= (Mti+1 1)2 + Mt2i+1 Mt2i Mti
2 2 2 2
ti J ti J
/ ti J
/ ti J
1
n1
(Mti+1 Mti )2
2
i=0
1 1 1
n1
= (Mti+1 1)2 Mt2i + Mt2n (Mti+1 Mti )2
2 2 2
ti J i=0
1
= (Mti+1 1 Mti )(Mti+1 1 + Mti )
2
t J
!" #
i
=0
1 1
n1
+ Mt2n (Mti+1 Mti )2
2 2
i=0
1 2 1
n1
= Mtn (Mti+1 Mti )2 .
2 2
i=0
Remark 5.8.2 (a) Let be a xed state of the world and assume the sample
path t Nt () has a jump in the interval (a, b). Even if beyond the scope of
this book, it can be shown that the integral
b
Nt () dNt
a
t n1
E f (s) dNs = lim E f (si )(Nsi+1 Nsi )
0 n
i=0
n1
= lim f (si )E Nsi+1 Nsi
n
i=0
n1 t
= lim f (si )(si+1 si ) = f (s) ds.
n 0
i=0
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 133
= s + 2 (s)2 ,
E[(Nsi+1 Nsi )(Nsj+1 Nsj )] = E[(Nsi+1 Nsi )]E[(Nsj+1 Nsj )]
n1
2
n1
f (si )(Nsi+1 Nsi ) = f (si )2 (Nsi+1 Nsi )2
i=0 i=0
+2 f (si )f (sj )(Nsi+1 Nsi )(Nsj+1 Nsj )
i=j
yields
n1
2
E f (si )(Nsi+1 Nsi )
i=0
n1
= f (si )2 (s + 2 (s)2 ) + 2 f (si )f (sj )2 (s)2
i=0 i=j
n1 n1
2 2
= f (si ) s + f (si )2 (s)2 + 2 f (si )f (sj )(s)2
i=0 i=0 i=j
n1 n1
2
= f (si )2 s + 2 f (si ) s
i=0 i=0
t t 2
f (s)2 ds + 2 f (s) ds , as n .
0 0
(c) Using that Nt is stationary with independent increments and has the
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 134
k 1)t
moment generating function E[ekNt ] = e(e , we have
t
E e 0 f (s) dNs
n1 n1
$
f (si )(Nsi+1 Nsi )
= lim E e i=0 = lim E ef (si )(Nsi+1 Nsi )
n n
i=0
$
n1 $
n1
= lim E ef (si )(Nsi+1 Nsi ) = lim E ef (si )(Nsi+1 si )
n n
i=0 i=0
$
n1
f (si ) 1)(s
n1
i+1 si ) (ef (si ) 1)(si+1 si )
= lim e(e = lim e i=0
n n
i=0
t f (s) 1) ds
= e 0 (e .
t
Since f is continuous, the Poisson integral f (s) dNs can be computed
0
in terms of the waiting times Sk
t
Nt
f (s) dNs = f (Sk ).
0 k=1
This formula can be used to give a proof for the previous result. For instance,
taking the expectation and using conditions over Nt = n, yields
t
Nt
n
E f (s) dNs = E f (Sk ) = E f (Sk )|Nt = n P (Nt = n)
0 k=1 n0 k=1
n t
(t)n
= f (x) dx et
t 0 n!
n0
1 (t)n
t
= et f (x) dx
0 t (n 1)!
n0
t t
= et f (x) dx et = f (x) dx.
0 0
Exercise 5.8.6 Solve parts (b) and (c) of Proposition 5.8.5 using a similar
idea with the one presented above.
Exercise 5.8.7 Show that
t 2 t
E f (s) dMs = f (s)2 ds,
0 0
where Mt = Nt t is the compensated Poisson process.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 135
Proposition 5.8.10 Let Ft = (Ns ; 0 s t). Then for any constant c, the
process
Mt = ecNt +(1e )t ,
c
t0
is an Ft -martingale.
Proof: Let s < t. Since Nt Ns is independent of Fs and Nt is stationary,
we have
E[ec(Nt Ns ) |Fs ] = E[ec(Nt Ns ) ] = E[ecNts ]
c 1)(ts)
= e(e .
On the other side, taking out the deterministic part yields
E[ec(Nt Ns ) |Fs ] = ecNs E[ecNt |Fs ].
Equating the last two relations we arrive at
E[ecNt +(1e )t |Fs ] = ecNs +(1e
c c )s
,
which is equivalent to the martingale condition E[Mt |Fs ] = Ms .
We shall present an application of the previous result. Consider the waiting
time until the nth jump, Sn = inf{t > 0; Nt = n}, which is a stopping time,
and the ltration Ft = (Ns ; 0 s t). Since
Mt = ecNt +(1e
c )t
T
5.9 The Distribution Function of XT = 0 g(t) dNt
In this section we consider the function g(t) continuous. Let S1 < S2 < <
SNt denote the waiting times until time t. Since the increments dNt are equal
to 1 at Sk and 0 otherwise, the integral can be written as
T
XT = g(t) dNt = g(S1 ) + + g(SNt ).
0
T
The distribution function of the random variable XT = 0 g(t) dNt can be
obtained conditioning over the Nt
P (XT u) = P (XT u|NT = k) P (NT = k)
k0
= P (g(S1 ) + + g(SNt ) u|NT = k) P (NT = k)
k0
= P (g(S1 ) + + g(Sk ) u) P (NT = k). (5.9.5)
k0
(a) if 0 u < T , then Dk is a 21k -part of a k-dimensional sphere;
(b) if T u < T k, then Dk has a complicated shape;
(c) if T k u, then Dk is the entire k-dimensional cube, and then vol(Dk ) =
T k.
k/2 Rk
Since the volume of the k-dimensional ball of radius R is given by ,
( k2 + 1)
then the volume of Dk in case (a) becomes
k/2 uk/2
vol(Dk ) = .
2k ( k2 + 1)
T
Exercise 5.9.2 Compute the distribution function of Xt = 0 s dNs .
Exercise 5.9.3 The following stochastic dierential equation has been used
in [2] to model the depreciation value of a car with stochastic repair payments
where k > 0 is the depreciation rate, > 0 is the average repair payment, and
Nt is a Poisson process with rate .
(a) Show that the solution is given by
t
Vt = V0 ekt ekt eks dNs ;
0
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 138
Show that
kV0 +
E[ek ] = .
+ kK
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 139
Chapter 6
Stochastic Dierentiation
Most stochastic processes are not dierentiable. For instance, the Brownian
motion process Wt is a continuous process which is nowhere dierentiable.
Hence, derivatives like dW t
dt do not make sense in stochastic calculus. The only
quantities allowed to be used are the innitesimal changes of the process, in
our case, dWt .
The innitesimal change of a process The change in the process Xt be-
tween instances t and t + t is given by Xt = Xt+t Xt . When t is
innitesimally small, we obtain the innitesimal change of a process Xt
dXt = Xt+dt Xt .
Sometimes it is useful to use the equivalent formula Xt+dt = Xt + dXt .
139
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 140
and hence
d(f (t)Yt ) = f (t) dYt + Yt df (t).
This relation looks like the usual product rule.
The quotient rule If Xt and Yt are two stochastic processes, then
X Y dX X dY dX dY Xt
+ 3 (dYt )2 .
t t t t t t t
d = 2
Yt Yt Yt
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 141
The proof follows from Itos formula and will be addressed in section 6.2.3.
When the process Yt is replaced by the deterministic function f (t), and Xt
is a process satisfying an equation of type (6.1.1), then the previous formula
becomes
X f (t)dX X df (t)
t t t
d = 2
.
f (t) f (t)
Applying the product rule and the fundamental relation (dWt )2 = dt, yields
t
Example 6.1.5 Let At = 1t Zt = 1t 0 Wu du be the average of the Brownian
motion on the time interval [0, t]. Show that
1 1
dAt = Wt Zt dt.
t t
We have
1 1 1
dAt = d Zt + dZt + d dZt
t t t
1 1 1 2
= 2
Zt dt + Wt dt + 2 Wt dt
!"#
t t t
=0
1 1
= Wt Zt dt.
t t
t
Exercise 6.1.6 Let Gt = 1t 0 eWu du be the average of the geometric Brown-
ian motion on [0, t]. Find dGt .
df (x) = f (x)dx.
dXt = bt dt + t dWt ,
Remark 6.2.2 Itos formula can also be written under the following equiva-
lent integral form
t t
1 2
Ft = F0 + bs f (Xs ) + s f (Xs ) ds + s f (Xs ) dWs .
0 2 0
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 144
1
dFt = f (Wt )dt + f (Wt ) dWt . (6.2.5)
2
Particular cases In the following we shall present the most often used cases:
1. If f (x) = x , with constant, then f (x) = x1 , f (x) = ( 1)x2 .
Then (6.2.5) becomes the following useful formula
1
d(Wt ) = ( 1)Wt2 dt + Wt1 dWt .
2
1
d(ekWt ) = kekWt dWt + k2 ekWt dt.
2
1
d(sin Wt ) = cos Wt dWt sin Wt dt.
2
In the case when the function f = f (t, x) is also time dependent, the analog
of (6.2.2) is given by
1
df (t, x) = t f (t, x)dt + x f (t, x)dx + x2 f (t, x)(dx)2 + O(dx)3 + O(dt)2 .
2
(6.2.6)
Substituting x = Xt yields
1
df (t, Xt ) = t f (t, Xt )dt + x f (t, Xt )dXt + x2 f (t, Xt )(dXt )2 . (6.2.7)
2
If Xt is a process satisfying an equation of type (6.1.1), then we obtain an
extra-term in formula (6.2.4)
b(Wt , t)2 2
dFt = t f (t, Xt ) + a(Wt , t)x f (t, Xt ) + x f (t, Xt ) dt
2
+b(Wt , t)x f (t, Xt ) dWt . (6.2.8)
Exercise 6.2.5 Show that
d(tWt2 ) = (t + Wt2 )dt + 2tWt dWt .
Exercise 6.2.6 Find the following increments
f f
dFt = dWt1 + dWt2 . (6.2.12)
x y
Exercise 6.2.12 Let Wt1 , Wt2 be two independent Brownian motions. If the
function f is harmonic, show that Ft = f (Wt1 , Wt2 ) is a martingale. Is the
converse true?
Exercise 6.2.13 Use the previous formulas to nd dFt in the following cases
(a) Ft = (Wt1 )2 + (Wt2 )2
(b) Ft = ln[(Wt1 )2 + (Wt2 )2 ].
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 148
1 W1 W2
dRt = dt + t dWt1 + t dWt2 .
2Rt Rt Rt
Example 6.2.15 (The product rule) Let Xt and Yt be two Ito diusions.
Show that
d(Xt Yt ) = Yt dXt + Xt dYt + dXt dYt .
Example 6.2.16 (The quotient rule) Let Xt and Yt be two Ito diusions.
Show that
X Y dX X dY dX dY Xt
+ 3 (dYt )2 .
t t t t t t t
d =
Yt Yt2 Yt
Chapter 7
Stochastic Integration
Techniques
Computing a stochastic integral starting from the denition of the Ito integral
is not only dicult, but also rather inecient. Like in elementary Calculus,
several methods can be developed to compute stochastic integrals. We tried
to keep the analogy with elementary Calculus as much as possible. The inte-
gration by substitution is more complicated in the stochastic environment and
we have considered only a particular case of it, which we called the method of
heat equation.
after canceling the terms in pairs. Substituting into formula (7.1.1) yields the
equivalent form t
Xt = Xa + f (s, Ws )dWs .
a
149
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 150
t
This can also be written as dXt = d f (s, Ws )dWs , since Xa is a constant.
a
Using dXt = f (t, Wt )dWt , the previous formula can be written in the following
two equivalent ways:
(i) For any a < t, we have
t
d f (s, Ws )dWs = f (t, Wt )dWt . (7.1.2)
a
These formulas are equivalent ways of writing the stochastic dierential equa-
tion (7.1.1), and will be useful in future computations. A few applications
follow.
t t 2
Consider the stochastic processes Xt = 0 sWs dWs , Yt = Wt 1 , and
t 2
Zt = 12 0 Ws2 ds. Formula (7.1.2) yields
T
Application 7.2.1 Consider the Wiener integral IT = t dWt . From the
0
general theory, see Proposition 5.6.1, it is known that I is a random variable
normally distributed with mean 0 and variance
T
T3
V ar[IT ] = t2 dt = .
0 3
Recall the denition of integrated Brownian motion
t
Zt = Wu du.
0
Cov(IT + ZT , IT + ZT ) = V ar[T WT ]
V ar[IT ] + V ar[ZT ] + 2Cov(IT , ZT ) = T 2 V ar[WT ]
T 3 /3 + T 3 /3 + 2Cov(IT , ZT ) = T 3
Cov(IT , ZT ) = T 3 /6,
where we used that V ar[ZT ] = T 3 /3. The processes It and Zt are not inde-
pendent. Their correlation coecient is 0.5 as the following calculation shows
Cov(IT , ZT ) T 3 /6
Corr(IT , ZT ) = 1/2 = T 3 /3
V ar[IT ]V ar[ZT ]
= 1/2.
x2
Application 7.2.2 If we let g(x) = 2 in formula (7.2.4), we get
b
Wb2 Wa2 1
Wt dWt = (b a).
a 2 2
x3
Similarly, if we let g(x) = 3 in (7.2.4) yields
b
b
Wt3
b
Wt2 dWt =
Wt dt.
a 3 a a
In
T atsimilar way, we can obtain an exact formula for the stochastic integral
0 e sin Wt dWt as follows
T T
e sin Wt dWt =
t
et (cos Wt ) dWt
0 0
T T
1 T t
= et cos Wt
+ et cos Wt dt e cos Wt dt.
0 0 2 0
1
Taking = 2 yields the closed form formula
T t T
e 2 sin Wt dWt = 1 e 2 cos WT . (7.2.6)
0
This formula is of theoretical value. In practice, the term dXt dYt needs to be
computed using the rules dWt2 = dt, and dt dWt = 0.
t
(d) Prove that Corr(Wt, eWt ) = , and compute the limits as t 0
et 1
and t .
Exercise 7.2.7 (a) Let T > 0. Show the following relation using integration
by parts
T T
2Wt 2 1 Wt2
2 dW t = ln(1 + W T ) 2 2 dt.
0 1 + Wt 0 (1 + Wt )
(b) Show that for any real number x the following double inequality holds
1 1 x2
1.
8 (1 + x2 )2
(c) Use part (b) to show that
1 T
1 Wt2
T dt T.
8 0 (1 + Wt2 )2
(d) Use parts (a) and (c) to get
T
E[ln(1 + WT2 )] T.
8
(e) Use Jensens inequality to get
1
t + x2 = 0. (7.3.7)
2
This is called the heat equation without sources. The non-homogeneous equa-
tion
1
t + x2 = G(t, x) (7.3.8)
2
is called the heat equation with sources. The function G(t, x) represents the
density of heat sources, while the function (t, x) is the temperature at the
point x at time t in a one-dimensional wire. If the heat source is time inde-
pendent, then G = G(x), i.e. G is a function of x only.
Example 7.3.1 Find all solutions of the equation (7.3.7) of type (t, x) =
a(t) + b(x).
1
b (x) = a (t).
2
Since the left side is a function of x only, while the right side is a function of
variable t, the only case where the previous equation is satised is when both
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 158
sides are equal to the same constant C. This is called a separation constant.
Therefore a(t) and b(x) satisfy the equations
1
a (t) = C, b (x) = C.
2
Integrating yields a(t) = Ct + C0 and b(x) = Cx2 + C1 x + C2 . It follows
that
(t, x) = C(x2 t) + C1 x + C3 ,
with C0 , C1 , C2 , C3 arbitrary constants.
Example 7.3.2 Find all solutions of the equation (7.3.7) of the type (t, x) =
a(t)b(x).
(t, x) = a(t)b(x) = c1 x + c0 , c0 , c1 R
In particular, the functions x, x2 t, ext/2 , ext/2 , et/2 sin x and et/2 cos x,
or any linear combination of them, are solutions of the heat equation (7.3.7).
However, there are other solutions which are not of the previous type.
Sometimes it is useful to generate new solutions for the heat equation from
other solutions. Below we present a few ways to accomplish this:
(i) by linear combination: if 1 and 2 are solutions, then a1 1 + a1 2 is
a solution, where a1 , a2 are constants.
(ii) by translation: if (t, x) is a solution, then (t , x ) is a solution,
where (, ) is a translation vector.
(iii) by ane transforms: if (t, x) is a solution, then (t, 2 x) is a
solution, for any constant .
n+m
(iv) by dierentiation: if (t, x) is a solution, then n m (t, x) is a
x t
solution.
(v) by convolution: if (t, x) is a solution, then so are
b
(t, x )f () d
a
b
(t , x)g(t) dt.
a
For more detail on the subject the reader can consult Widder [46] and Cannon
[11].
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 160
Theorem 7.3.8 Let (t, x) be a solution of the heat equation (7.3.7) and
denote f (t, x) = x (t, x). Then
b
f (t, Wt ) dWt = (b, Wb ) (a, Wa ).
a
Consider the function (t, x) = 13 x3 tx, which is a solution of the heat equa-
tion (7.3.7), see Exercise 7.3.3. Then f (t, x) = x (t, x) = x2 t. Applying
Theorem 7.3.8 yields
T T
T
2
1
(Wt t) dWt = f (t, Wt ) dWt = (t, Wt )
= WT3 T WT .
0 0 0 3
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 161
2 t
Consider the function (t, x) = e 2 x , which is a solution of the homoge-
neous heat equation (7.3.7), see Example 7.3.2. Then f (t, x) = x (t, x) =
2 t
e 2
x
. Apply Theorem 7.3.8 to get
T T
T
2
2 t x
2 T
e dWt = f (t, Wt ) dWt = (t, Wt )
= e 2 WT 1.
0 0 0
2 t
From the Example 7.3.2 we know that (t, x) = e 2 sin(x) is a solution of the
heat equation. Applying Theorem 7.3.8 to the function f (t, x) = x (t, x) =
2 t
e 2 cos(x), yields
T T
T
2 t
e 2 cos(Wt ) dWt = f (t, Wt ) dWt = (t, Wt )
0 0 0
2
t
T 2
T
= e 2 sin(Wt )
= e 2 sin(WT ).
0
2 t
Choose (t, x) = e 2 cos(x) to be a solution of the heat equation. Apply
2 t
Theorem 7.3.8 for the function f (t, x) = x (t, x) = e 2 sin(x) to get
T
T
2 t
()e 2 sin(Wt ) dWt = (t, Wt )
0 0
T
2 T
2 T
= e 2 cos(Wt )
= e 2 cos(WT ) 1,
0
2
From Exercise 7.3.4 we have that (t, x) = t1/2 ex /(2t) is a solution of the
2
homogeneous heat equation. Since f (t, x) = x (t, x) = t3/2 xex /(2t) ,
applying Theorem 7.3.8 yields the desired result. The reader can easily ll in
the details.
Integration techniques will be used when solving stochastic dierential
equations in the next chapter.
T
1 T
14. sin Wt dWt = 1 cos WT cos Wt dt;
0 2 0
t
15. d f (s, Ws ) dWs = f (t, Wt ) dWt ;
a
b
16. Yt dWt = Fb Fa , if Yt dWt = dFt ;
a
b b
17. f (t) dWt = f (t)Wt |ba f (t)Wt dt;
a a
b
b 1 b
18. g (Wt ) dWt = g(Wt )
g (Wt ) dt.
a a 2 a
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 165
Chapter 8
Stochastic Dierential
Equations
where the last integral is taken in the Ito sense. Relation (8.1.2) is taken as the
denition for the stochastic dierential equation (8.1.1). However, since it is
convenient to use stochastic dierentials informally, we shall approach stochas-
tic dierential equations by analogy with the ordinary dierential equations,
165
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 166
and try to present the same methods of solving equations in the new stochastic
environment.
Most of the stochastic dierential equations considered here describe dif-
fusions, and are of the type
with a(t, x) and b(t, x) measurable functions. The functions a(t, x) and b(t, x)
are called the drift rate and the volatility of the process Xt , respectively. Given
these two functions as input, one may seek for the solution Xt of the stochastic
dierential equation as an output. The desired outputs Xt are the so-called
strong solutions. The precise denition of this concept is given in the following.
The beginner can skip this denition; all solutions in this book will be solutions
in the strong sense anyway.
A few comments regarding the previous denition. Part (i) states that
given the information induced by and the history of the Brownian motion
until time t, one can determine the value Xt . Part (ii) states that X0 takes
the value with probability 1. Part (iii) deals with a non-explosive condition
for the coecients. Part (iv) states that Xt veries the associated integral
equation.
We shall start with an example.
Example 8.1.2 (The Brownian bridge) Let a, b R. Show that the pro-
cess t
1
Xt = a(1 t) + bt + (1 t) dWs , 0 t < 1
0 1s
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 167
If the drift and the volatility depend only on variables t and Wt , the
stochastic dierential equation
dXt = a(t, Wt )dt + b(t, Wt )dWt , t0
denes a stochastic process that can be expressed in terms of Ito integrals
t t
Xt = X0 + a(s, Ws ) ds + b(s, Ws ) dWs .
0 0
There are several cases when both integrals can be computed explicitly. In
order to compute the second integral we shall often use the table of usual
stochastic integrals provided in section 7.4.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 169
dXt = dt + Wt dWt , X0 = 1.
Integrating yields
t t
Xt = s2 ds + es/2 cos Ws dWs
0 0
t3
= + et/2 sin Wt , (8.2.5)
3
where we used (7.3.10). Even if the process Xt is not Gaussian, we can still
compute its mean and variance. Since Ito integrals have zero expectation,
t t t t3
E[Xt ] = E s2 ds + es/2 cos Ws dWs = s2 ds = .
0 0 0 3
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 170
1
d(sin Wt ) = cos Wt dWt sin Wt dt
2
Integrating between 0 and t yields
t t
1
sin Wt = cos Ws dWs sin Ws ds,
0 2 0
where we used that sin W0 = sin 0 = 0. Taking the expectation in the previous
relation yields
t 1
t
E[sin Wt ] = E cos Ws dWs E[sin Ws ] ds.
0 2 0
From the properties of the Ito integral, the rst expectation on the right side
is zero. Denoting (t) = E[sin Wt ], we obtain the integral equation
t
1
(t) = (s) ds.
2 0
1
(t) = (t)
2
Taking the expectation and using that Ito integrals have zero expectation,
yields t
E[cos 2Wt ] = 1 2 E[cos 2Ws ] ds.
0
If we denote m(t) = E[cos 2Wt ], the previous relation becomes an integral
equation t
m(t) = 1 2 m(s) ds.
0
Dierentiate and get
m (t) = 2m(t),
with the solution m(t) = ke2t . Since k = m(0) = E[cos 2W0 ] = 1, we have
m(t) = e2t . Substituting into (8.2.6) yields
et et et
V ar[Xt ] = (1 e2t ) = = sinh t.
2 2
In conclusion, the solution Xt has the mean and the variance given by
t3
E[Xt ] = , V ar[Xt ] = sinh t.
3
Example 8.2.6 Solve the following stochastic dierential equation
and nd the distribution of the solution Xt and its mean and variance.
Dividing by et/2 , integrating between 0 and t, and using formula (7.3.9) yields
t t
s/2
Xt = e ds + es/2+Ws dWs
0 0
t/2 t/2 Wt
= 2(1 e )+e e 1
t/2
= 1+e Wt
(e 2).
F (y) = P (Xt y) = P 1 + et/2 (eWt 2) y
W 1
t
= P Wt ln 2 + et/2 (y 1) = P ln 2 + et/2 (y 1)
t t
1
= N ln 2 + et/2 (y 1) ,
t
u
1 2
where N (u) = es /2 ds is the distribution function of a standard
2
normal distributed random variable.
1
a(t, x) = t f (t, x) + x2 f (t, x) (8.3.8)
2
b(t, x) = x f (t, x). (8.3.9)
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 173
The coecient functions are a(t, x) = 2tx3 + 3t2 (1+ x) and b(t, x) = 3t2 x2 + 1.
The associated system is given by
1
2tx3 + 3t2 (1 + x) = t f (t, x) + x2 f (t, x)
2
3t2 x2 + 1 = x f (t, x).
Then t f = 2tx3 + T (t) and x2 f = 6t2 x, and substituting into the rst
equation we get
1
2tx3 + 3t2 (1 + x) = 2tx3 + T (t) + 6t2 x.
2
After cancelations we get T (t) = 3t2 , so T (t) = t3 + c. Then
f (t, x) = t2 x3 + x + t3 + c.
1
x a = t b + x2 b. (8.3.10)
2
Remark 8.3.4 The equation (8.3.10) has the meaning of a heat equation.
The function b(t, x) represents the temperature measured at x at the instance
t, while x a is the density of heat sources. The function a(t, x) can be regarded
as the potential from which the density of heat sources is derived by taking
the gradient in x.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 175
exact?
Exercise 8.3.7 Verify the closeness condition and then solve the following
exact stochastic dierential equations
(a) dXt = Wt + 32 Wt2 dt + (t + Wt3 )dWt , X0 = 0;
(b) dXt = 2tWt dt + (t2 + Wt )dWt , X0 = 0;
(c) dXt = et Wt + 12 cos Wt dt + (et + sin Wt )dWt , X0 = 0;
(d) dXt = eWt (1 + 2t )dt + teWt dWt , X0 = 2.
X f (t)dX X df (t)
t t t
d = .
f (t) f (t)2
For instance, if a stochastic dierential equation can be written as
the product rule brings the equation into the exact form
dXt = d f (t)Wt ,
Xt = X0 + f (t)Wt .
dXt = d(tWt2 ),
we note the exact expression formed by the last two terms Wt dt + tdWt =
d(tWt ). Then
dXt = d(t3 ) + d(tWt ),
which is equivalent to d(Xt ) = d(t3 + tWt ). Hence Xt = t3 + tWt + c, c R.
Integrating yields
t t
A(t) A(s)
e Xt = X0 + e
(s) ds + eA(s) b(s, Ws ) dWs
0 0
t t
A(s)
Xt = X0 eA(t)
+eA(t)
e (s) ds + eA(s) b(s, Ws ) dWs .
0 0
The rst integral within the previous parentheses is a Riemann integral, and
the latter one is an Ito stochastic integral. Sometimes, in practical applications
these integrals can be computed explicitly.
When b(t, Wt ) = b(t), the latter integral becomes a Wiener integral. In
this case the solution Xt is Gaussian with mean and variance given by
t
E[Xt ] = X0 eA(t) + eA(t) eA(s) (s) ds
0
t
V ar[Xt ] = e2A(t) e2A(s) b(s)2 ds.
0
Integrating yields
t t
et/2 Xt = X0 + es/2 ds + es/2 cos Ws dWs .
0 0
is given by
t
Xt = m + (X0 m)et + est dWs . (8.5.12)
0
Hence
t
t t t
Xt = X0 e +me
+ e es dWs
0
t
= m + (X0 m)et + est dWs .
0
lim E[Xt ] = m.
t
b Xt
dXt = dt + dWt , 0 t < 1, X0 = a
1t
Making t = 0 yields c = a b, so
b Xt t
1
=ab dWs .
1t 0 1s
E[Xt ] = a(1 t) + bt
V ar(Xt ) = V ar(Ut ) = t(1 t).
t 1.
Exercise 8.5.8 Find Cov(Xs , Xt ), 0 < s < t for the following cases:
(a) Xt is a mean reverting Ornstein-Uhlenbeck process;
(b) Xt is a Brownian bridge process.
d(e3t Xt ) = dNt .
Xt = eWt +c .
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 185
The nomination pseudo stands for the fact that Xt does not satisfy the
initial equation. We shall nd a correct solution by letting the parameter c be
a function of t. In other words, we are looking for a solution of the following
type:
Xt = eWt +c(t) , (8.7.16)
where the function c(t) is subject to be determined. Using Itos formula we
get
dXt = d(eWt +c(t) ) = eWt +c(t) (c (t) + 2 /2)dt + eWt +c(t) dWt
= Xt (c (t) + 2 /2)dt + Xt dWt .
Substituting the last term from the initial equation (8.7.15) yields
c (t) + 2 /2 = 0
2
with the solution c(t) = 2 t + k. Substituting into (8.7.16) yields
2
Xt = eWt 2
t+k
.
Example 8.7.1 Use the method of variation of parameters to solve the stochas-
tic dierential equation
dXt = Xt dt + Xt dWt ,
After dividing by Xt we bring the equation into the equivalent integral form
dXt
= dt + dWt .
Xt
ln Xt = t + Wt + c,
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 186
2
dXt = Xt + c (t) + dt + Xt dWt .
2
Subtracting the initial equation yields
2
c (t) + dt = 0,
2
2 2
which is satised for c (t) = 2 , with the solution c(t) = 2 t + k, k R.
Substituting into (8.7.17) yields the solution
2 2 2
Xt = et+Wt 2
t+k
= e( 2
)t+Wt +k
= X0 e( 2
)t+Wt
.
Exercise 8.7.2 Use the method of variation of parameters to solve the equa-
tion
dXt = Xt Wt dWt
by following the next two steps:
(a) Divide by Xt and integrate blindly to get the pseudo-solution
Wt2
2t +c
Xt = e 2 ,
with c constant.
(b) Consider c = c(t, Wt ) and nd a solution of type
Wt2
2t +c(t,Wt )
Xt = e 2 .
which has the solution Xt = Ceqt , with C constant. Now, look for a solution
of the type Xt = C(t)eqt , and determine the function C(t) such that the
initial equation is satised. Comparing
dXt = d C(t)eqt = qC(t)eqt dt + eqt dC(t)
= qXt dt + eqt dC(t)
with the initial stochastic dierential equation of Xt implies that C(t) satises
The rst condition says that the drift and volatility increase no faster
than a linear function in x. This condition ensures that the solution Xt does
not explode in nite time, i.e. does not tend to for nite t. The second
conditions states that the functions are Lipschitz in the second argument; this
condition guarantees the solution uniqueness.
The following example deals with an exploding solution. Consider the
nonlinear stochastic dierential equation
where a is a nonzero constant. It is clear that condition 1. does not hold, since
the drift increases cubically.
We shall look for a solution of the type Xt = f (Wt ). Itos formula yields
1
dXt = f (Wt )dWt + f (Wt )dt.
2
Equating the coecients of dt and dWt in the last two equations yields
1
Xt = .
a Wt
Let Ta be the rst time the Brownian motion Wt hits a. Then the process Xt
is dened only for 0 t < Ta . Ta is a random variable with P (Ta < ) = 1
and E[Ta ] = , see section 4.3.
Example 8.9.2 Show that that the following stochastic dierential equations
have a unique (strong) solution, without solving the equations explicitly:
(a) dXt = Xt dt + dWt (Langevin equation);
(b) dXt = (m Xt ) dt + dWt (Mean reverting Ornstein-Uhlenbeck pro-
cess);
(c) dXt = Xt dWt (Linear noise);
(d) dXt = dt + X2t dWt (Squared Bessel process);
(e) dXt = Xt dt + Xt dWt (Geometric Brownian motion);
(f ) dXt = (0 + 1 Xt ) dt + X2t dWt (CIR process),
with m, , 0 , 1 and positive constants.
Exercise 8.10.1 If dXt = (2Xt + e2t )dt + b(t, Wt , Xt )dWt , then show that
E[Xt ] = e2t (X0 + t).
Proof: The expression of E[Xt ] follows directly from formula (8.10.24) with
= 0. In order to compute the second moment we rst compute
where we used Itos formula. If we let Yt = Xt2 , the previous equation becomes
dYt = 2(t)Yt + b2 (t) dt + 2b(t) Yt dWt .
Applying formula (8.10.24) with (t) replaced by 2(t) and (t) by b2 (t),
yields t
E[Yt ] = e2A(t) Y0 + e2A(s) b2 (s) ds ,
0
which is equivalent to
t
E[Xt2 ] =e 2A(t) 2
X0 + e2A(s) b2 (s) ds .
0
Remark 8.10.3 We note that the previous equation is of linear type. This
shall be solved explicitly in a future section.
The mean and variance for a given stochastic process can be computed by
working out the associated stochastic equation. We shall provide next a few
examples.
Example 8.10.4 Find the mean and variance of ekWt , with k constant.
If we let f (t) = E[ekWt ], then dierentiating the previous relations yields the
dierential equation
1
f (t) = k2 f (t)
2
2 t/2
with the initial condition f (0) = E[ekW0 ] = 1. The solution is f (t) = ek ,
and hence
2
E[ekWt ] = ek t/2 .
The variance is
2 t/2 2t
V ar(ekWt ) = E[e2kWt ] (E[ekWt ])2 = e4k ek
2 2
= ek t (ek t 1).
We shall set up a stochastic dierential equation for Wt eWt . Using the product
formula and Itos formula yields
Let f (t) = E[Wt eWt ]. Using E[eWs ] = es/2 , the previous integral equation
becomes t
1
f (t) = ( f (s) + es/2 ) ds.
0 2
Dierentiating yields the following linear dierential equation
1
f (t) = f (t) + et/2
2
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 194
with the initial condition f (0) = 0. Multiplying by et/2 yields the following
exact equation
(et/2 f (t)) = 1.
The solution is f (t) = tet/2 . Hence we obtained that
(2k)! k
E[Wt2k ] = t , E[Wt2k+1 ] = 0.
2k k!
Since the expectation of the rst integral on the right side is zero, taking the
expectation yields the following recursive relation
n(n 1) t
E[Wt ] =
n
E[Wsn2 ] ds.
2 0
Using the initial values E[Wt ] = 0 and E[Wt2 ] = t, the method of mathematical
induction implies that E[Wt2k+1 ] = 0 and E[Wt2k ] = (2k)!2k k!
tk . The details are
left to the reader.
Let f (t) = E[sin Wt ]. Dierentiating yields the equation f (t) = 12 f (t) with
f (0) = E[sin W0 ] = 0. The unique solution is f (t) = 0. Hence
E[sin Wt ] = 0.
Exercise 8.10.11 Use the previous exercise and the denition of expectation
to show that
2 1/2
(a) ex cos x dx = 1/4 ;
e
2 2
(b) ex /2 cos x dx = .
e
Exercise 8.10.13 Using the result given by Example 8.10.7 show that
3 /2
(a) E[cos(tWt )] = et ;
(b) E[sin(tWt )] = 0;
(c) E[etWt ] = 0.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 196
For general drift rates we cannot nd the mean, but in the case of concave
drift rates we can nd an upper bound for the expectation E[Xt ]. The following
classical result will be used for this purpose.
Lemma 8.10.14 (Gronwalls inequality) Let f (t) be a non-negative con-
tinuous function satisfying the inequality
t
f (t) C + M f (s) ds
0
f (t) CeM t , 0 t T.
Proof: The proof follows Revuz and Yor [41]. Iterating the integral inequality
one gets
t
f (t) C + M f (s) ds
0
t s
C +M C+M f (u) du ds
0 0
t s
= C + M Ct + M 2 f (u) du ds
0 0
t
2
= C + M Ct + M t f (u) du.
0
M n+1 tn
lim = 0.
n n!
Taking the limit in (8.10.25) it is not hard to obtain
M n tn
f (t) C = CeM t .
n!
k=0
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 197
Proof: From the mean value theorem there is (0, x) such that
Exercise 8.10.16 State the previous result in the particular case when a(x) =
sin x, with 0 x .
Not in all cases can the mean and the variance be obtained directly from
the stochastic equation. In these cases one may try to produce closed form
solutions. Some of these techniques were developed in the previous sections of
the current chapter.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 199
Chapter 9
Applications of Brownian
Motion
n
f dxk (t)
n
f
(x0 )vk (0) = f (x0 ), v,
dt
Dv f (x0 ) = (x(t)) =
xk xk
k=1 t=0+
k=1
where f stands for the gradient of f and , denotes the scalar product.
The linear dierential operator Dv is called the directional derivative with
respect to the vector v. In the next section we shall extend this denition
to the case when the curve x(t) is replaced by an Ito diusion Xt ; in this
case the corresponding directional derivative will be a second order partial
dierential operator.
199
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 200
Ex [f (Xt )] f (x)
Af (x) = lim ,
t
0 t
for any smooth function (at least of class C 2 ) with compact support, i.e.
f : Rn R, f C02 (Rn ). Here Ex stands for the expectation operator given
the initial condition X0 = x, i.e.,
E [f (Xt )] = E[f (Xt )|X0 = x] =
x
f (y)pt (x, y) dy,
Rn
In the following we shall nd the generator associated with the Ito diusion
f 1 2f
dFt = (Xt ) dXti + (Xt ) dXti dXtj , (9.2.2)
xi 2 xi xj
i i,j
Using the stochastic relations dt2 = dt dWk (t) = 0 and dWk (t) dWr (t) = kr dt,
a computation provides
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 201
= ( T )ij dt.
Therefore
dXti dXtj = ( T )ij dt. (9.2.4)
Substituting (9.2.3) and (9.2.4) into (9.2.2) yields
% &
1 2f f
dFt = (Xt )( T )ij + bi (Xt ) (Xt ) dt
2 xi xj xi
i,j i
f
+ (Xt )ik (Xt ) dWk (t).
xi
i,k
t% f
&
1 2f T
Ft = F0 + ( )ij + bi (Xs ) ds
0 2 xi xj xi
i,j i
t f
+ ik (Xs ) dWk (s).
0 x i
k i
t
Using the commutativity between the operator Ex and the integral 0, apply-
ing lHospital rule (see Exercise 9.2.7), yields
1 2
A= ( T )ij + bk . (9.2.6)
2 xi xj xk
i,j k
t
E [f (Xt )] = f (x) + E
x x
Af (Xs ) ds , (9.2.7)
0
Exercise 9.2.2 Find the generator operator associated with the n-dimensional
Brownian motion.
Exercise 9.2.3 Find the Ito diusion corresponding to the generator Af (x) =
f (x) + f (x).
k
lim E g(Xs ) ds = E g(Xs ) ds . (9.3.9)
k 0 0
Proof: Let |g| < K. Using the properties of Ito integrals, we have
k 2 2
E g(Xs ) dWs g(Xt ) dWs =E g(Xs ) dWs
0 0 k
= E g2 (Xs ) ds K 2 E[ k] 0, k .
k
Exercise 9.3.2 Assume the hypothesis of the previous lemma. Let 1{s< } be
the characteristic function of the interval (, )
1, if u <
1{s< } (u) =
0, otherwise.
Show that
k k
(a) g(Xs ) dWs = 1{s< } g(Xs ) dWs ,
0 0
k k
(b) g(Xs ) ds = 1{s< } g(Xs ) ds.
0 0
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 204
Exercise 9.3.4 Write Dynkins formula for the case of a function f (t, Xt ).
Use Exercise 9.2.6.
More details in this direction can be found in Dynkin [16]. In the following
sections we shall present a few important results of stochastic calculus that
can be obtained as direct consequences of Dynkins formula.
provides
v
= Ex [Af (Xt )].
t
Since we are allowed to dierentiate inside of an integral, the operators A and
Ex commute
Ex [Af (Xt )] = AEx [f (Xt )].
Therefore
v
= Ex [Af (Xt )] = AEx [f (Xt )] = Av(t, x).
t
Hence, we arrive at the following result.
v
= Av, t>0
t
v(0, x) = f (x),
E[X ] = x0 . (9.5.13)
Exercise 9.5.1 Prove relation (9.5.13) using the Optional Stopping Theorem
for the martingale Xt .
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 206
apa + b(1 pa ) = x0 .
vector, see Fig. 9.1. Let R > 0 be such that R > |a|. Consider the exit time
of the process Xt from the ball B(0, R)
yields
2 2
R = |a| + E n ds ,
0
and hence
R2 |a|2
E[ ] = . (9.6.18)
n
In particular, if the Brownian motion starts from the center, i.e. a = 0, the
expectation of the exit time is
R2
E[ ] = .
n
We make a few remarks:
(i) Since R2 /2 > R2 /3, the previous relation implies that it takes longer for
a Brownian motion to exit a disk of radius R rather than a ball of the same
radius.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 208
(ii) The probability that a Brownian motion leaves the interval (R, R) is
twice the probability that a 2-dimensional Brownian motion exits the disk
B(0, R).
Exercise 9.6.2 Apply the Optional Stopping Theorem for the martingale Wt =
Wt2 t to show that E[ ] = R2 , where
is the rst exit time of the Brownian motion from (R, R).
where k > 0 such that b Ak . Consider the process Xt = b + W (t) and let
k = inf{t > 0; Xt
/ Ak }
(i) If n = 2 we obtain
pk ln R qk (ln k + ln R) = ln b.
ln( Rb )
pk = 1 .
ln k
Hence
P ( < ) = lim pk = 1,
k
where = inf{t > 0; |Xt | R} is the rst time Xt hits the ball B(0, R).
Hence in R2 a Brownian motion hits with probability 1 any ball. This is
stated equivalently by saying that the Brownian motion is recurrent in R2 .
(ii) If n > 2 the equation (9.6.20) becomes
pk qk 1
n2
+ n2 n2 = n2 .
R k R b
Taking the limit k yields
R n2
lim pk = < 1.
k b
Then in Rn , n > 2, a Brownian motion starting outside of a ball hits it with
a probability less than 1. This is usually stated by saying that the Brownian
motion is transient.
3. We shall recover the previous results using the n-dimensional Bessel process
Rt = dist 0, W (t) = W1 (t)2 + + Wn (t)2 .
Consider the process Yt = + Rt , with 0 < R, see section 3.7. It can be
shown that the generator of Yt is the Bessel operator of order n, see Example
10.2.4
1 d2 n1 d
A= 2
+ .
2 dx 2x dx
Consider the exit time
= {t > 0; Yt R}.
Applying Dynkins formula
E f (Y ) = f (Y0 ) + E (Af )(Ys ) ds
0
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 210
for f (x) = x2 yields R2 = 2 + E 0 n ds . This leads to
R2 2
E[ ] =
n
which recovers (9.6.18) with = |a|.
In the following assume n 3 and consider the annulus
pr r 2n + pR R2n = 2n ,
where
R n2 r n2
1 1
pr = R n2 , pR = n2 .
r 1 r
R 1
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 211
2n Rn2 1 r n2
pr = lim pr,R = lim = ,
R R r 2n Rn2 1
where pr is the probability that a Brownian motion starting outside the ball
of radius r will hit the ball, see Fig. 9.2.
dX(s)
= a s, X(s) , tsT
ds
X(t) = x,
and dene the cumulative cost between t and T along the solution
T
u(t, x) = c s, (s) ds, (9.7.21)
t
Exercise 9.7.1 Using the previous method solve the following nal boundary
problems:
(a)
t u + xx u = x
u(T, x) = 0.
(b)
t u + txx u = ln x, x>0
u(T, x) = 0.
Applying Itos formula on one side and the Fundamental Theorem of Calculus
on the other, we obtain
1
t u(t, x)dt + x u(t, Xt )dXt + x2 u(t, t, Xt )dXt2 = c(t, Xt )dt.
2
1
This is a well known method of solving linear partial dierential equations.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 213
(b)
1
t u + x u + x2 u = ex ,
2
u(T, x) = 0.
(c)
1
t u + xx u + 2 x2 x2 u = x,
2
u(T, x) = 0.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 215
Chapter 10
is an Ft -martingale.
The integrability of Xt follows from
t 2
2 2
E[|Xt |] E[Xt ] = E v(s) dWs
0
t t
= E v(s)2 ds = v(s)2 ds < .
0 0
215
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 216
is an Ft -martingale.
yields t
E[Xt2 ] = v 2 ( ) d,
0
which leads to the following estimation
t t
E[|Mt |] E[Xt2 ] + v 2 ( ) d 2 v 2 ( ) d < .
0 0
is an Ft -martingale for 0 t T .
1
dUt = u(t)dWt u2 (t)dt
2
(dUt )2 = u(t)dt.
It is worth noting that the conclusion still holds if the function u(s) is replaced
by a stochastic process u(t, ) satisfying Novikovs condition
1 T 2
E e 2 0 u (s,) ds < .
The previous process has a distinguished importance in the theory of martin-
gales and will be useful in the proof of Girsanov theorem.
Denition 10.1.5 Let u L2 [0, T ] be a deterministic function. Then the
stochastic process t
1 t 2
Mt = e 0 u(s) dWs 2 0 u (s) ds
is called the exponential process induced by u.
t3
= etWt 6 Zt
is an Ft -martingale.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 219
Then Yt = Mt Xt is an Ft -martingale.
and hence
E[Yt |Fs ] = Ys .
1
Exercise 10.1.7 Prove that (Wt + t)eWt 2 t is an Ft -martingale.
Exercise 10.1.9 Let Mt be the exponential process (10.1.2). Use the previous
exercise to show that for any t > 0
t
u(s)2 ds
(a) E[Mt ] = 1 (b) E[Mt2 ] = e 0 .
Exercise 10.1.10 Let Ft = {Wu ; u t}. Show that the following processes
are Ft -martingales:
(a) et/2 cos Wt ;
(b) et/2 sin Wt .
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 220
Recall that
nthe Laplacian of a twice dierentiable function f is dened by
2
f (x) = j=1 xj f .
Exercise 10.1.12 Let W1 (t) and W2 (t) be two independent Brownian mo-
tions. Show that Xt = eW1 (t) cos W2 (t) is a martingale.
Hence Xt is an Ft -martingale.
Exercise 10.1.14 Use Proposition 10.1.13 to show that the following pro-
cesses are martingales:
(a) Xt = Wt2 t;
t
(b) Xt = Wt3 3 0 Ws ds;
1
t
(c) Xt = n(n1) Wtn 12 0 Wsn2 ds;
t
(d) Xt = ecWt 12 c2 0 eWs ds, with c constant;
t
(e) Xt = sin(cWt ) + 12 c2 0 sin(cWs ) ds, with c constant.
where
1, if x > 0
sgn(x) =
1, if x 0.
We note that dXt = sgn(Bt ) dBt and hence (dXt )2 = dt. Since Xt is a
continuous martingale (because it is an Ito integral) and its quadratic variation
is given by
t t
X, Xt = (dXs )2 = ds = t,
0 0
then Levys theorem implies that Xt is a Brownian motion.
Hence, the squared Bessel process, Zt = Rt2 , satises the following stochas-
tic dierential equation
dZt = 2 Zt dt + n dt, (10.2.4)
Example 10.2.4 (The Bessel process) From Example 10.2.3 we recall that
1/2
(dZt )2 = 4Rt2 dt. Then, using Rt = Zt , Itos formula yields
1 1/2 1 3/2
dRt = Zt dZt Zt (dZt )2
2 8
1 1 1 1
= (2Rt dt + ndt) dt
2 Rt 2 Rt
n1
= dt + dt.
2Rt
Hence the Bessel process satises the following stochastic dierential equation
n1
dRt = dt + dt, (10.2.5)
2Rt
where t is a Brownian motion. It is worth noting that the innitesimal gen-
erator of Rt is the operator
1 n1
A = x2 + x , (10.2.6)
2 2x
which is the Bessel operator of order n.
f f 1 2f 2f
dXt = dW1 (t) + dW2 (t) + + dt.
x1 x2 2 x21 x22
2f 2f
+ = 0. (10.2.7)
x21 x22
Then f 2 f 2
(dXt )2 = + dt = |f |2 dt,
x1 x2
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 224
so we have t t
2
X, Xt = (dXt ) = |f |2 ds.
0 0
Then the condition X, Xt = t, for any t 0, implies |f | = 1, i.e.
f 2 f 2
+ = 1. (10.2.8)
x1 x2
Equation (10.2.8) is called the eiconal equation. We shall show that if a func-
tion f satises both equations (10.2.7) and (10.2.8), then it is a linear func-
tion. Or, equivalently, a harmonic solution of the eiconal equation is a linear
function.
From equation (10.2.8) there is a continuous function = (x1 , x2 ) such
that
f f
= cos , = sin . (10.2.9)
x1 x2
The closeness condition implies
(cos ) (sin )
= ,
x2 x1
which is equivalent to
cos + sin = 0. (10.2.10)
x1 x2
Dierentiating in (10.2.9) with respect to x1 and x2 yields
2f 2f
= sin , = cos .
x21 x1 x22 x2
Mt = BM,M t ,
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 226
Bs = MT (s)
Then there is a Brownian motion W .t such that Xt = W .c2 t . This can also be
1. 2 .s = cWs/c2 . There-
written as c Wc2 t = Wt . Substituting s = c t, yields W
fore, if Ws is a Brownian motion, then the process cWs/c2 is also a Brownian
motion. In particular, if c = 1, then Ws is a Brownian motion.
This relation can be easily veried for n = 1, when both sides are normally
distributed as N (0, T 3 /3), see section 3.3.
t
1
(1 t) dWs = (1 t)B t = B
t(1t) ,
0 1s 1t
Xt = a(1 t) + bt + B
t(1t) , o t 1.
This process has the characteristics of a Brownian motion, while satisfying the
boundary conditions X0 = a and X1 = b.
ATu = u, u 0.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 228
eBt = RAt .
Lemma 10.3.7 Let (t) be a continuous dierentiable, with (t) > 0, (0) =
0 and limt (t) = . Then given the one-dimensional Brownian motion
Bt , there is another one-dimensional Brownian motion Wt such that
t
W(t) = (s) dBs , t 0. (10.3.15)
0
t
Proof: The process Xt = 0 (s) dBs is a continuous martingale, with the
quadratic variation
t t t
2 2
X, Xt = (dXs ) =
( (s) dBs ) = (s) ds = (t).
0 0 0
Ws = cBs/c2 , s0
Remark 10.3.9 The right side of (10.3.15) is a Wiener integral. Hence, under
certain conditions, a Wiener integral becomes a time-scaled Brownian motion.
Xt = et/2 B
et .
Proof: First, we will prove formula (10.3.16) informally, and then we will
check that the identity holds in law. Formula (10.3.15) can be written in the
equivalent dierential form as
dW(t) = (t) dBt .
1
If F (u) = and = 0, then we obtain
1 + u2
tan1
1 1
2
dW u = dBt ,
0 1+u 0 sec t
which, after substituting v = tan1 , implies
tan v v
1
dWu = cos t dBt .
0 1 + u2 0
Making v
2 yields
/2
1
dWu = cos t dBt .
0 1 + u2 0
We have not used the upper script until now since there was no doubt which
probability measure was used. In this section we shall also use another prob-
ability measure given by
dQ = MT dP,
Q(A) > 0, A = ;
Q() = MT dP = EP [MT ] = EP [MT |F0 ] = M0 = 1,
The following result will play a central role in proving Girsanovs theorem:
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 233
t
Proof: Denote U (t) = 0 u(s) ds. Then
Taking the expectation and using the property of Ito integrals we have
t t
E[Wt Mt ] = u(s)E[Ms ] ds = u(s) ds = U (t). (10.4.24)
0 0
Xt = t + Wt , 0tT
1 2 T W
dQ = e 2 T
dP.
This result states that a Brownian motion with drift can be viewed as a regular
Brownian motion under a certain change of the probability measure.
EP [X] = EQ [XMT1 ].
on (, F, Q) and we have
2 T
E[f (t + Wt )] = EP [f (t + Wt )] = EQ [f (Xt )MT1 ] = EQ [f (Xt )e 2
+WT
]
2 T
+(XT T )
= EQ [f (Xt )e 2 ]
2
2T
= e EQ [f (Xt )eXt e(XT Xt ) ]
2 T
= e 2 EQ [f (Xt )eXt ] EQ [e(XT Xt ) ]
2 T 2
= e 2 EQ [f (Xt )eXt ]e 2
(T t)
2 t
= e 2 EQ [f (Xt )eXt ]
2
2 t
= e E[f (Wt )eWt ].
(ii) We apply Girsanovs theorem for the Q-Brownian motion Xt = t + Wt
and obtain
2 T
EQ [f (Xt )] = EP [f (t + Wt )MT ] = EP [f (t + Wt )e 2
WT
]
2
= EP [f (t + Wt )eWt e 2T (WT Wt )
]
2
2T
= e EP [f (t + Wt )eWt ] EP [e(WT Wt ) ]
2 t
= e 2 EP [f (t + Wt )eWt ].
Replacing Xt by Wt in the rst term yields the desired formula.
Exercise 10.4.8 Use the reduction of drift formulas and Example 8.10.7 to
show
(2k)! 2 t
(a) E[(t + Wt )2k eWt ] = k tk e 2 ;
2 k!
(b) E[(t + Wt )2k+1 eWt ] = 0.
Exercise 10.4.9 Use the reduction of drift formula to show
2
(a) E[sin(t + Wt )] = e t/2 sin t;
2
(b) E[cos(t + Wt )] = e t/2 cos t.
Exercise 10.4.10 Use the reduction of drift formulas to nd
(a) E[cos(t + Wt )eWt ];
(b) E[sin(t + Wt )eWt ].
Exercise 10.4.11 Let Wt be a Brownian motion on the space (, F, P ), and
dQ = MT dP . Show that
t
EQ [et+Wt ] = e 2
(a) by a direct computation;
(b) using Girsanovs theorem.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 239
Exercise 10.4.13 Use Jensens inequality to show that for any convex, mea-
surable function f we have
2 t
E[f (t + Wt )eWt ] f (0)e 2 .
t2
Exercise 10.4.15 Consider the stochastic process Xt = 2 + Wt , where Wt
is a Brownian motion on (, F, P ).
(a) Find the probability measure dQ such that Xt becomes a Q-Brownian
motion;
(b) Compute explicitly EP [Xt MT ];
(c) Use EP [Xt MT ] = EQ [Xt ] = 0 to nd a formula for
t
EP [Wt e 0
s dWs
].
Exercise 10.4.17 Use the drift reduction formula to express V ar[f (t+Wt )].
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 241
Chapter 11
Some Applications of
Stochastic Calculus
241
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 242
cess. The eect of the noise during the time interval dt is normally distributed
with mean zero and variance dt, and is given as an innitesimal jump of a
Brownian motion, Nt dt = dBt . Thus, it is convenient sometimes to represent
the white noise informally as a derivative of a Brownian motion,
dBt
Nt = .
dt
Since Bt is nowhere dierentiable, the aforementioned derivative does not make
sense classically. However, it makes sense in the following generalized sense:
Nt f (t) dt = Bt f (t) dt,
R R
for any compact supported, smooth function f . Hence, from this point of view,
the white noise Nt is a generalized function or a distribution. In the following
we shall state its relation with the Dirac distribution 0 , which is dened in
the generalized sense as
0 (t)f (t) dt = f (0),
R
(
) 1
Xt = B(t + ) B(t) , t 0,
which models the rate of change of a Brownian motion B(t). Since we have
(
) 1
E[Xt ] = E[B(t + )] E[B(t)] = 0
(
) 1 1
V ar(Xt ) = 2
(t + t) = ,
(
)
the limiting process Nt = lim Xt will have zero mean and innite variance.
0
Consider s < t and choose > 0 small enough, such that (s, s + )
(t, t + ) = . Using the properties of Brownian motions, the dierences
B(s + ) B(s) and B(t + ) B(t) are independent. Hence, the random
variables Ns and Nt are independent for s < t.
In the following we shall compute the covariance of the process Nt . For
reasons which will be clear later we shall extend the parameter t to take values
in the entire real line R. This can be done by dening the Brownian motion
B(t) as
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 243
1 |t s|
(
)
Cov(Xs(
) , Xt ) = max 1 ,0 . (11.1.1)
Consider the test function
1
1 , if | |
1
( ) = max 1 , 0 =
0, if | | > ,
1
which veries
( ) 0,
(0) = and
1
( ) d = 1 d = 1.
R
Therefore, we have
lim
( ) = 0 ( ),
0
where 0 is the Dirac distribution centered at 0. In fact, the above limit has
the following meaning
lim
( )f ( ) d = f (0),
0 R
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 244
then
(
)
Cov(Ns , Nt ) = lim Cov(Xs(
) , Xt ) = lim
(t s) = 0 (t s).
0
0
E[Nt ] = 0
Cov(Ns , Nt ) = 0 (t s).
This shows that the error between the expected coordinate and the coordinate
of a uniform move is at most linear in time and is controlled by the dierence
v0 m. Therefore, the expected coordinate is the classical coordinate, E[xt ] =
xunif (t), if and only if v0 = m.
The acceleration at is obtained as the derivative of velocity vt with respect
to time
dvt
at =
dt
t
at at dWt
= a(v0 m)e ae eas dWs +
0 dt
t
dW
= a0 eat aeat
t
eas dWs + , (11.2.3)
0 dt
where a0 is the initial acceleration. The rst term is a deterministic function,
the second term is a normally distributed random variable of zero mean, while
dWt dW
t
is the white noise term. Since E = 0, the long run limit of the
dt dt
expectation becomes E[at ] = a0 e at 0, as t .
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 246
Let M denote the mass of the cyclist and Ft = M at be the muscle force
developed at time t. The work done by the cyclist between instances 0 and t
is given by
xt xt t
W = Fs dxs = M as dxs = M as vs ds,
x0 x0 0
where the velocity and the acceleration are given by (11.2.2) and (11.2.3).
Computing the exact expression of W is tedious. However, using the properties
of Ito integrals one can compute E[W], see Exercise 11.2.1.
Since the square of velocity is given by
2 t
vt2 = m + (v0 m)e at
+ 2e at
m + (v0 m)eat
eas dWs
0
t 2
2 2at
+ e eas dWs ,
0
dvt = g dt + vt dWt , v0 = 0,
(a) What is the probability that at the time t = 3 sec the velocity of the
particle is less than 0.1 m/sec?
(b) Find the work done by the environment on the particle in order to
decelerate it from v0 = 1 m/sec to v = 0.5 m/sec.
Exercise 11.2.4 A snowball rolls downhill with the velocity given by the equa-
tion
dvt = 0.6 vt dt + 0.2 dWt , v0 = 0.
(a) Find the velocity vt ;
(b) What is the probability that the velocity is greater than 10 m/sec at
t = 20 sec?
i.e. the probability of the occurrence of one decay in a small time interval is
proportional with the time interval. The probability of the complementary
event is
P N (t) N (t + t) = 0 = 1 t. (11.3.4)
Divide the interval [0, t] into n equidistant subintervals
with t = tk+1 tk = t/n. The event of not having any decays during the
interval [0, t] can be expressed as
n1
{N (0) N (t) = 0} = {N (tk ) N (tk+1 ) = 0}.
k=0
For N (t) large enough, this probability represents the fraction of nuclei that
survived the decay during the time interval [0, t]. Since the percentage of nuclei
that are still alive after time t is represented by the quotient N (t)/N (0), we
have
N (t)
= et .
N (0)
This relation, written as N (t) = N (0)et , is the law of radioactive decay.
Now we shall develop a dierential equation for N (t). Relation (11.3.4)
states that the fraction of nuclei that resist the decay during the time interval
[t, t + t] is
N (t + t)
= 1 t.
N (t)
Cross multiplying and subtracting N (t) yields
N (t) N (t + t) = N (t)t. (11.3.5)
Assuming that the period of observation is innitely ne, t dt, the equa-
tion becomes
dN (t) = N (t) dt.
This describes the kinetics of the radioactive decay, stating that the change
in the number of nuclei dN (t) during the time interval dt is proportional with
the number of nuclei N (t). Solving the aforementioned equation, we obtain
again the law of radioactive decay
N (t) = N (0)ekt .
Noisy radioactive decay In real life relation (11.3.5) does not hold exactly,
and some errors of measurement or counting are involved. These will be added
as a noisy term
N (t) N (t + t) = N (t)t + noise.
For t small, this becomes a stochastic dierential equation
dN (t) = N (t) dt + dWt ,
with positive constant and Wt Brownian motion. The obtained equation is
called Langevins equation. We shall solve it as a linear stochastic dierential
equation. Multiplying by the integrating factor et yields
d(et N (t)) = et dWt .
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 249
Integrating yields t
et N (t) = N (0) + es dWs .
0
Hence the solution is
t
t t
N (t) = N (0)e + e es dWs . (11.3.6)
0
This is the Ornstein-Uhlenbeck process. Since the last term is a Wiener inte-
gral, by Proposition 8.2.1 we have that N (t) is Gaussian with the mean
t
t
E[N (t)] = N (0)e +E e(st) dWs = N (0)et
0
and variance
t 2
V ar[N (t)] = V ar e(st) dWs = (1 e2t ).
0 2
e2t = 1 + 2t + o(t2 ), t 0,
then (e2t 1)/(2) = t + o(t2 ), and hence the following approximation holds
for t small
N (t) = N (0)et + et Bt .
Exercise 11.3.1 Let N (t) be a noisy radioactive decay. Dene the half time
h as
1
h = inf{t > 0; N (t) N (0)}.
2
(a) Prove that E[eh ] = 2;
ln 2
(b) Use Jensens inequality to show that E[h] .
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 250
Exercise 11.3.3 A living organism has initially N0 cells. Assume the number
of cells which die during the time interval t is Poisson distributed
n (t)n
P N (t) N (t + t) = n = et .
n!
The organism dies when at least 30% of its cells are dead. Find an approxi-
mation of the death time of the organism.
+ k 2 (t) = Nt ,
(t) (11.4.8)
where k and are constants and the noise is given informally as Nt = dBt
dt .
The general solution of equation (11.4.8) can be expressed as the sum
where p (t) is a particular solution of (11.4.8) and 0 (t) is the solution of the
associated homogeneous equation (11.4.7).
Standard ODE methods provide
where u1 (t) and u2 (t) are two dierentiable functions, which will be determined
later. Assuming
dierentiating yields
p (t) = u1 (t) cos(kt) + u2 (t) sin(kt) ku1 (t) sin(kt) + ku2 (t) cos(kt)
= ku1 (t) sin(kt) + ku2 (t) cos(kt).
dBt
ku1 (t) sin(kt) + ku2 (t) cos(kt) = , (11.4.13)
dt
dBt
where we used the informal notation for the white noise Nt = . Equations
dt
(11.4.11) and (11.4.13) yield the ODEs system in u1 and u2
t
u1 (t) = sin(ks) dBs (11.4.16)
k 0
t
u2 (t) = cos(ks) dBs . (11.4.17)
k 0
These represent the eect of the white noise dBs along the solutions trajec-
tories sin(ks) and cos(ks). From the properties of Wiener integrals, it follows
that u1 (t) and u2 (t) have normal distributions with the mean, variances and
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 252
covariance given by
E[u1 (t)] = E[u2 (t)] = 0
2 t 2 2 t sin(2kt)
V ar[u1 (t)] = sin (ks) ds =
k2 0 k2 2 4k
t
2
2
2
t sin(2kt)
V ar[u2 (t)] = cos (ks) ds = +
k2 0 k2 2 4k
Denoting / 0 / 0 / 0
X1 0 1 0
Xt = , A= , K= ,
X2 k2 0
the aforementioned system becomes a linear matrix stochastic dierential
equation
dXt = AXt dt + K dBt .
Multiplying by the integrating factor eAt yields the exact equation
d eAt Xt = eAt K dBt .
(c) Find the general solution, Qt , and show that E[Qt ] satises the homo-
geneous equation.
Exercise 11.4.2 (a) Solve the following stochastic dierential equation
dXt = Yt dt + dWt1
dYt = Xt dt + dWt2 ,
where (Wt1 , Wt2 ) is a 2-dimensional Brownian motion and and are con-
stants.
(b) Use part (a) to nd a solution for the following stochastic pendulum
equation
2 + W
t + t = W 1.
t t
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 255
rt = a(t) + noise.
dt = t ( 2 dt dt),
dYt = a(t)Yt dt
2 t/2
Using E[eBt ] = e , we obtain
t t
a(s) ds 2 t/2
E[Pt ] = P0 e 0 E[eBt ] = P0 e 0 a(s) ds
.
It is worth noting that the function P (t) = E[Pt ] satises the deterministic
equation
dP (t) = a(t) P (t)dt.
The population Pt provided by formula (11.5.22) is log-normally distributed.
Pt
In fact, ln is normally distributed
P0
Pt
ln N (m(t), 2 t),
P0
with the mean given by
t
2t
m(t) = a(s)ds .
0 2
E[PT ] = 2P0 .
The constant r is the intrinsic growth rate, i.e. the relative rate at which the
population would increase if there were no restrictions on the population. The
positive constant k reects the damping eect on the population growth caused
by competition for resources between the members of the same population.
The solution of the equation (11.5.23) is given by the logistic function
P0 K
P (t) = , (11.5.24)
P0 + (K P0 )ert
where K = r/k is the saturation level, or carrying capacity of the environment.
This represents the equilibrium level to which the population, regardless of its
initial size, will tend in the long run
K = lim P (t).
t
One of the stochastic models for the population growth in a stochastic and
competitive environment is obtained keeping in equation (11.5.23) the rate k
constant, while considering a noisy intrinsic rate of growth
It can be shown that this equation has a unique strong solution, but the
discussion of this subject is beyond the level of this textbook.
Population growth in a stochastic catastrophic environment In the
previous model the population tends to decrease due to competition and lim-
ited space. In the present model the population decreases suddenly due to
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 259
with the solution given by the usual formula Pt = P0 ert , for t [0, S1 ). In
particular, when t = S1 , we have
dPS1 PS PS1
= 1 = ,
PS1 PS1
Because there are no jumps in the interval [S1 , S2 ) the following dierential
equation holds
dPt = rPt dt, S1 t < S2 ,
with the solution
Pt = PS1 er(tS1 ) .
Combining with (11.5.30) yields
The eect of passing over a jump is to multiply the solution by the factor
(1 ). Using this observation we obtain inductively
E[Pt ] = P0 ert E[(1 )Nt ] = P0 ert E[(1 )Nt |Nt = n]P (Nt = n)
n0
n tn t
= P0 ert (1 )n P (Nt = n) = P0 ert (1 )n e
n!
n0 n0
rt t (1)t (r)t
= P0 e e e = P0 e .
k tk
P (Pt x) = et .
k!
kyt
Exercise 11.5.4 An ant colony of 1, 000 ants grows at the intrinsic rate r =
0.30 per month. However, each rainfall kills 2% of the ant population, and it
rains 5 times per year.
(a) Write the stochastic dierential equation for the ant population size;
(b) What is the probability that there are at least 2, 000 ants in the colony
at the end of the rst year?
(c) What is the expected size of the colony after 2 years?
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 261
This implies that Pt is normally distributed with the mean and variance given
by
E[Pt ] = ert P0 (11.5.34)
r
2 2rt
V ar[Pt ] = (e 1). (11.5.35)
2r
Exercise 11.5.5 Show that the stochastic process (11.5.33) is the solution of
equation (11.5.32).
where is a positive constant that controls the size of the noise. This leads
to the following stochastic process
T T T
Bt = Be t rs ds
= Be t r ds
e t dWs
= Ber(T t) e(WT Wt ) .
According to the previous model, we note that in the case when the market
2
noise is small, r > 2 , the bond value appreciates, having the maximum value,
2
B, at t = T . If the noise is large, 2 > r, the bond depreciates and has to be
sold as soon as possible.
Exercise 11.6.1 Use Itos formula to show that the process Bt satises the
equation
2
dB(t) = r+ B(t)dt + B(t)dWt
2
B(T ) = B.
b
E[Ct ] = C0 eat + (C0 + E)(1 eat )
a
and variance
2
V ar[Ct ] = (1 e2at ).
2a
If the diet is kept for a long time, then the cholesterol level becomes the normal
random variable
b 2
Ct N C0 + E, , t .
a 2a
In order to evaluate the health of a particular person who is on a given diet
(i.e. when E is kept constant for a long time) we have to nd the probability
that the cholesterol level is over a given acceptable level M . Using the long
run normality of Ct , this probability can be evaluated as
a 1 (xC0 2 ab E)
2
Exercise 11.7.1 Assuming that the intaken food does not have a constant
cholesterol level, the term E is replaced by E + noise. Write and solve the
corresponding stochastic dierential equation.
Exercise 11.7.2 The normal cholesterol level in the blood is C0 = 200 (mil-
ligrams per deciliter), the production parameter is a = 0.1 and the absorption
parameter for a particular person is b = 0.15. What is the maximum daily
intake E of cholesterol such that the long run level of cholesterol is less than
220 with a probability of 95%?
Xt
photon's trajectory
solar plasma
Figure 11.1: The photon bounces back and forth in its eort to emerge to the
suns surface.
coordinates vector in R3 of the photon at time t, where the center of the sun
is assigned the zero coordinate, then
Xt = a + Wt , (11.8.38)
where a = X0 is the location where the photon was initially created, is
the dispersion function and Wt = (Wt1 , Wt2 , Wt3 ) is a 3-dimensional Brownian
process in R3 . The dispersion is given by Einsteins formula
2 = 2kT /(6),
where k is Boltzmanns constant, T denotes the absolute temperature, is
the diameter of the photon, and is the viscosity of the solar plasma. For the
sake of simplicity we shall assume in the following that is constant.
The time when the photon reaches the surface of the sun is a random
variable denoted by . We shall use a similar method as the one described in
section 9.6 to nd the expected time E[ ]. If RS denotes the radius of the sun,
the time necessary for a photon to emerge to the suns surface is the exit time
= inf{t > 0; |Xt | RS }. (11.8.39)
Since the innitesimal generator of the process (11.8.38) is the operator
2 2 2
= ( + x22 + x23 ),
2 2 x1
using the function f (x) = |x|2 = x21 + x22 + x23 in Dynkins formula
2
E f (X ) = f (x) + E f (Xs ) ds
0 2
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 265
yields
RS2 2
= |a| + E 3 2 ds ,
0
and hence
RS2 |a|2
E[ ] = . (11.8.40)
3 2
In particular, if the photon is emitted from the suns center, the expected
emerging time to the surface is
RS2
E[ ] = . (11.8.41)
3 2
Using the numerical values for the suns radius and photons diusion given
by RS = 6.955 105 km and 2 = 0.0025 km2 /sec (this corresponds to a 50
meters per second radial photon displacement), formula (11.8.41) yields the
approximate value E[ ] 2 million years. Some other sources compute slightly
dierent values, but the idea is that it takes a really long time for a photon to
leave the suns interior. This is huge compared with the only 8 minutes spent
by the photon on its way to earth.
It is worth noting that if a star has its radius 100 times the suns radius,
then the expected emerging time multiplies by a factor of 104 . This means
E[ ] 2 1010 years (20 billion years), which is longer than the entire age of
the universe ( 14 billion years)! Hence, it is possible that a photon created
in the center of the star has not found its way out to the surface yet. Since the
life span of a star is usually around 10 billion years, the photon will probably
not get out of the star during its life span.
noise
t St
input signal observed signal
where a(t) and b(t) are given deterministic functions and Wt is a Brownian
motion, independent of the initial value 0 . We assume that t is observed
continuously with the actual observations St = t + noise, see Fig. 11.2. If
dBt
the white noise is given by noise = (t) , then
dt
St dt = t dt + (t)dBt . (11.9.42)
and using that the information (-algebras) induced by the processes Qt and
St are the same,
FtQ = FtS ,
it follows that equation (11.9.42) can be replaced by the more mathematically
useful formula
dQt = t dt + (t)dBt , (11.9.43)
with Bt independent of Wt . From now on, Qt will be considered as the ob-
servation process instead of St . It is easier to work with the cummulative
observation process Qt rather than the actual observations St .
The ltering problem can now be stated as:
Given the observations Qs , 0 s t, satisfying equation (11.9.43), nd
the best estimate t of the state t .
One of the best estimators t , which is mathematically tractable, is the
one which minimizes the mean square error
R(t) = E[(t t )2 ].
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 267
This means that for any other square integrable random variable Y , which is
measurable with respect to the eld FtQ , we have the inequality
E[(t t )2 ] E[(t Y )2 ].
It turns out that the best estimator t coincides with the conditional expecta-
tion of t given the information induced by Qs , 0 s t, namely,
t = E[t |FtQ ].
with
(t) 2 (t)
V (t) = R(t), U (t) = a(t) R(t),
2 (t) 2 (t)
and R(t) satisfying the deterministic Riccati equation
dR(t) 2 (t)
= b2 (t) + 2a(t)R(t) 2 R2 (t), R(0) = 2 .
dt (t)
lim R(t) = 0,
t
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 268
where for the second identity we used the properties of Wiener integrals.
Now, equation (11.9.46) can be solved as a linear equation. After multi-
plying by the factor t
(t)1 = exp{ U (s) ds}
0
the equation becomes
dQt = dt + dBt ,
Qs = s + Bs , 0 s t.
dt = 0
dQt = dt + dBt .
Since in this case (t) = 1, (t) = , a(t) = 0, and b(t) = 0, the Riccati
equation becomes
dR(t) 1
= 2 R2 (t),
dt
2
with the solution R(t) = depending on the parameter C. Using the
t C 2
initial condition R(0) = 2 , we obtain
2 2
R(t) = .
2 t + 2
Then the coecients V (t) and U (t) take the following form
1 2
V (t) = R(t) =
2 2 t + 2
1 2
U (t) = 2 R(t) = 2 .
t + 2
Equation (11.9.46) becomes
2 2
dt + t dt =
dQt .
2 t + 2 2 t + 2
Multiplying by the integrating factor
t 2
0 2 s+ 2 ds 2 t + 2
e =
2
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 270
2 2
t = 0 +
Qt
2t + 2 2 t + 2
1
= ( 2 + 2 Qt ).
t + 2
2
It is worth noting that the foregoing formula implies that the best estimate
of the random variable , given the continuous observations Qs , 0 s t,
depends only on the last observation Qt . In the case of discrete observations,
the best estimate will depend on each observation, see the next example.
S 1 = + 1
S 2 = + 2
S n = + n ,
It makes sense to look for the best estimator as an element L(S) such
that the distance E[( )2 ] is minimized. This occurs when the estimator
is the orthogonal projection of on the space L(S). This means we have to
determine the constants 0 , cj such that the following n + 1 conditions are
satised
= E[]
E[]
E[( )Sj ] = 0, j = 1, , n.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 271
Let = 0 + c1 S1 + + cn Sn . Since
= 0 + (c1 + + cn )
= E[]
it follows that 0 = c0 , with
c0 + c1 + c2 + cn = 1. (11.9.48)
Therefore, belongs to the convex hull of {, S1 , , Sn }, i.e.
= c0 + c1 S1 + + cn Sn .
A computation provides
j ] E[Sj ]
E[( )Sj ] = E[S
n
= c0 E[Sj ] + ck E[Sk Sj ] E[( + j )]
k=1
n
= c0 2 + ck E[( + k )( + j )] E[ 2 ] E[]E[j ]
k=1
n
n
2 2 2
= c0 + E[ ] ck + m ck kj E[ 2 ]
k=1 k=1
2 2
= c0 + E[ ](1 c0 ) + m2 cj E[ 2 ]
= c0 (2 E[ 2 ]) + m2 cj
= c0 2 + m2 cj .
The orthogonality condition E[( )Sj ] = 0 implies
c0 2
cj = , j = 1, , n. (11.9.49)
m2
Substituting in (11.9.48) provides an equation for c0 , which has the solution
m2
c0 = ,
m2 + n 2
and hence (11.9.49) becomes
2
cj = .
m2 + n 2
In conclusion, the best estimator of , given n observations S1 , , Sn , is given
by
n
= c0 + ck Sk
k=1
m2 2
n
= + Sk .
m2 + n 2 m2 + n 2
k=1
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 272
ij = Cov(k , j ) = E[k j ]
which is a linear system in ck . Under the assumption that the matrix (kj ) is
non-singular, let (kj ) be its inverse matrix, so the solution of the aforemen-
tioned linear system is
n
ck = c0 2 kj , k = 1, , n. (11.9.50)
j=1
Example 11.9.4 Consider that the observed state is the blood cholesterol
level, Ct , which is given by the stochastic equation (11.7.36)
dQt = Ct dt + dBt , Q0 = 0.
dt = at dt + dWt .
b
Let Zt = Qt C0 + E t. We note that the -algebras induced by Zt and
a
Qt are the same, FtZ = FtQ , and hence we can consider Zt as an observation
process instead of Qt , satisfying
dZt = t dt + dBt , Z0 = 0.
The Riccati equation associated with the ltering problem having the input
process t and observations Zt is given by
dR(t) 1
= 2aR(t) 2 R2 (t) + 2 , R(0) = 0,
dt
1 (1 e2Kt )
R(t) = ,
1 12 e2Kt
where
1 = a 2 a2 2 + 2
2 = a 2 + a2 2 + 2
K = a2 + (/)2 .
Using that
lim R(t) = 2 ,
t
we obtain the following long run behavior for the estimation
t
b 2
lim t = EeKt + 2 eKt eKs dZs .
t a 0
1t and Qt as follows
This can be transformed in terms of C
t
1t = C0 + (1 eKt ) b E 2 (C0 + b E) + 2 eKt
lim C eKs dQs .
t a K 2 a 2 0
Chapter 12
Here we give the hints and solutions to selected exercises. The reader should
be able to derive the solutions to the rest based on what he has learnt from
the examples in the chapters.
Chapter 2
Exercise 2.9.7 (a) For any random variables A,B, and variable R, inte-
grating in the inequality
2
A() + B() 0,
implies 2
A() + B() dP () 0.
After expanding and collecting the powers of , this can be written as
2 2
A () dP () + 2 A()B() dP () + B 2 () dP () 0.
Substituting
a= A2 () dP (), b = 2 A()B() dP () , c = B 2 () dP (),
a2 + b + c 0, R.
277
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 278
with = + , = .
2 2 /2
Exercise 2.10.5 (a) Making t = n yields E[Y n ] = E[enX ] = en+n .
(b) Let n = 1 and n = 2 in (a) to get the rst two moments and then use the
formula of variance.
Exercise 2.12.7 The tower property
E E[X|G]|H = E[X|H], HG
is equivalent to
E[X|G] dP = X dP, A H.
A A
E[(Xn X)2 ] 0
|E[X(X Xn )]| |X(X Xn )| dP
1/2 1/2
X 2 dP (X Xn )2 dP
= E[X 2 ]E[(X Xn )2 ] 0.
as n 0.
Exercise 2.16.7 The integrability of Xt follows from
E[|Xt |] = E |E[X|Ft ]| E E[|X| |Ft ] = E[|X|] < .
Exercise 2.16.9 In general the answer is no for both (a) and (b). For instance,
if Xt = Yt the process Xt2 is not a martingale, since the Jensens inequality
2
E[Xt2 |Fs ] E[Xt |Fs ] = Xs2
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 281
is not necessarily an identity. For instance Bt2 is not a martingale, with Bt the
Brownian motion process.
Exercise 2.16.10 It follows from the identity
k
k
E[Yn+k |Fn ] = Yn + E[Xn+j |Fn ] E[E[Xn+j ]|Fn ]
j=1 j=1
k
k
= Yn + E[Xn+j ] E[Xn+j ]
j=1 j=1
= Yn .
E[Zn+k Zn |Fn ] = 0.
so
since E[U ] = 0.
Exercise 2.16.12 Let Fn = (Xk ; k n). Using the independence
Chapter 3
Exercise 3.1.4 Bt starts at 0 and is continuous in t. By Proposition 3.1.2
Bt is a martingale with E[Bt2 ] = t < . Since Bt Bs N (0, |t s|), then
E[(Bt Bs )2 ] = |t s|.
Exercise 3.1.9 It is obvious that Xt = Wt+t0 Wt0 satises X0 = 0 and
that Xt is continuous in t. The increments are normal distributed Xt Xs =
Wt+t0 Ws+t0 N (0, |t s|). If 0 < t1 < < tn , then 0 < t0 < t1 + t0 <
< tn + t0 . The increments Xtk+1 Xtk = Wtk+1 +t0 Wtk +t0 are obviously
independent and stationary.
Exercise 3.1.10 Let s < t. Then we have
1 1
Xt Xs = (Wt Ws ) N 0, (t s) = N (0, t s).
The other properties are obvious.
Exercise 3.1.11 Apply Property 3.1.7.
Exercise 3.1.12 Using the moment generating function, we get E[Wt3 ] = 0,
E[Wt4 ] = 3t2 .
Exercise 3.1.13 (a) Let s < t. Then
E[(Wt2 t)(Ws2 s)] = E E[(Wt2 t)(Ws2 s)]|Fs
= E (Ws2 s)E[(Wt2 t)]|Fs
= E (Ws2 s)2 = E Ws4 2sWs2 + s2
= E[Ws4 ] 2sE[Ws2 ] + s2 = 3s2 2s2 + s2 = 2s2 .
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 283
(d) Since
Yt Ys = (t s)(W1/t W0 ) s(W1/s W1/t )
E[Yt Ys ] = (t s)E[W1/t ] sE[W1/s W1/t ] = 0,
and
1 1 1
V ar(Yt Ys ) = E[(Yt Ys )2 ] = (t s)2 + s2 ( )
t s t
(t s)2 + s(t s)
= = t s.
t
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 284
E[(Wt Ws )3 |Fs ] = E[Wt3 |Fs ] 3Ws E[Wt2 ] + 3Ws2 E[Wt |Fs ] Ws3
= E[Wt3 |Fs ] 3(t s)Ws Ws3 ,
so
E[Wt3 |Fs ] = 3(t s)Ws + Ws3 ,
since
E[(Wt Ws )3 |Fs ] = E[(Wt Ws )3 ] = E[Wts
3
] = 0.
e(t+s)/2 .
t+3s
= e 2
2 2 1 2 x2
E[e2Wt ] = e2x t (x) dx = e2x e 2t dx
2t
1 14t 2 1
= e 2t x dx = ,
2t 1 4t
1 4t > 0.
if
Otherwise, the integral is innite. We used the standard integral
2
eax = /a, a > 0.
Exercise 3.3.4 (a) It follows from the fact that Zt is normally distributed;
(b) Dierentiate the moment generating function and evaluate it at u = 0.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 286
= e 2 es = e 2 emin{s,t} .
u+s u+s
t t
Exercise 3.3.8 (a) E[Xt ] = 0 E[eWs ] ds = 0 E[es/2 ] ds = 2(et/2 1)
(b) Since V ar(Xt ) = E[Xt2 ] E[Xt ]2 , it suces to compute E[Xt2 ]. Using
Exercise 3.3.7 we have
t t t t
2
E[Xt ] = E e ds
Wt
e du = E
Wu
eWs eWu dsdu
0 0 0 0
t t t t
E[eWs +Wu ] dsdu = e 2 emin{s,t} dsdu
u+s
=
0 0 0 0
u+s u+s
s
= e 2 e duds + e 2 eu duds
D
1 D2
1
4 3
e2t 2et/2 +
u+s
= 2 e 2 eu duds = ,
D2 3 2 2
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 287
T t T t
= E Wu2 du + 2Wt E Wu du + Wt2 (T t)
0 0
1
= (T t)2 + Wt2 (T t).
2
(b) Using part (a) we have
T 1
E YT |Ft = E Ws2 ds T WT2 + T 2 |Ft
t 2
T 1
= E Ws2 ds|Ft T E WT2 |Ft + T 2
t 2
t
1 1
= Ws2 ds + (T t)2 + Wt2 (T t) T Wt2 T (T t) + T 2
0 2 2
t
1
= Ws2 ds + tWt2 + t2 = Ys .
0 2
Exercise 3.4.1
t T
E[VT |Ft ] = E e 0 Wu du+ t Wu du |Ft
t T
= e 0 Wu du E e t Wu du |Ft
t T
= e 0 Wu du E e t (Wu Wt ) du+(T t)Wt |Ft
T
= Vt e(T t)Wt E e t (Wu Wt ) du |Ft
T
= Vt e(T t)Wt E e t (Wu Wt ) du
T t
= Vt e(T t)Wt E e 0 W d
3
1 (T t)
= Vt e(T t)Wt e 2 3 .
Exercise 3.6.1
E[Xt ] 2t
E[Xt ] = = = 0, t ;
t 2t 2t
1 2
V ar(Xt ) = V ar(Rt ) = (1 ) 0, t .
t2 t 4
By Proposition 2.14.1 we get Xt 0, t in mean square.
Exercise 3.8.2
P (Nt Ns = 1) = (t s)e(ts)
= (t s) 1 (t s) + o(t s)
= (t s) + o(t s).
then
E[Nt2 |Fs ] = E[(Nt Ns )2 |Fs ] + Ns E[Nt Ns |Fs ] + E[Nt t|Fs ]Ns + tNs
= E[(Nt Ns )2 ] + Ns E[Nt Ns ] + E[Nt t]Ns + tNs
= (t s) + 2 (t s)2 + tNs + Ns2 sNs + tNs
= (t s) + 2 (t s)2 + 2(t s)Ns + Ns2
= (t s) + [Ns + (t s)]2 .
Hence E[Nt2 |Fs ] = Ns2 and hence the process Nt2 is not an Fs -martingale.
Exercise 3.8.7 (a)
mNt (x) = E[exNt ] = exk P (Nt = k)
k0
k tk
= exk et
k!
k0
= et ete = et(e
x x 1)
.
Nt
t2 t2
(b) E Sk = E[tNt Ut ] = tt = .
2 2
k=1
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 291
Exercise 3.12.6 Use Doobs inequalityfor the submartingales Wt2 and |Wt |,
and use that E[Wt2 ] = t and E[|Wt |] = 2t/, see Exercise 3.1.15 (a).
Exercise 3.12.7 Divide by t in the inequality from Exercise 3.12.6 part (b).
Exercise 3.12.10 Let = n and = n + 1. Then
N 2 4(n + 1)
t
E sup
ntn=1 t n2
Chapter 4
Exercise 4.1.2 We have
, if c t
{; () t} =
, if c > t
1 1
Exercise 4.1.4 Let Km = [a + m, b m ]. We can write
{; t} = {; Xr
/ Km } Ft ,
m1 r<t,rQ
since {; Xr
/ Km } = {; Xr K m } Fr Ft .
Exercise 4.1.6 (a) No. The event A = {; Wt () has a local maximum at time t}
is not in {Ws ; s t} but is in {Ws ; s t + }.
M () = Mt () + M () Mt ().
2 x2 /(2t) 2t
E[Xt ] = xp(x) dx = xe dy =
0 2t 0
2 2 4t 2
E[Xt2 ] = x2 ex /(2t) dx = y 2 ey dy
2t 0 0
2 1 1/2 u 1
= u e du = (3/2) = t
2t 2 0 2t
2
V ar(Xt ) = E[Xt2 ] E[Xt ]2 = t 1 .
2
2 e 4t
x
Exercise 4.3.11 (b) h(x) = t
2N x
2t
1 .
Exercise 4.3.16 It is recurrent since P (t > 0 : a < Wt < b) = 1.
Exercise 4.4.4 Since
1 1 t1
P (Wt > 0; t1 t t2 ) = P (Wt =
0; t1 t t2 ) = arcsin ,
2 t2
using the independence
1 t1 2
P (Wt1 > 0, Wt2 ) = P (Wt1 > 0)P (Wt2 > 0) = 2 arcsin .
t2
The probability
for Wt = (Wt1 , Wt2 ) to be in one of the quadrants is equal to
4 t1 2
arcsin .
2 t2
Exercise 4.5.2 (a) We have
Exercise 4.5.3
E[ecMT T (e
c c1)
] = E[X0 ] = 1
E[ecaT f (c) ] = 1
E[eT f (c) ] = eac .
so E[T ] = .
(d) The inverse Laplace transform L1 ea(s) cannot be represented by
elementary functions.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 296
Chapter 5
Exercise 5.2.3 (a) Use either the denition or the moment generation func-
tion to show that E[Wt4 ] = 3t2 . Using stationarity, E[(Wt Ws )4 ] = E[Wts
4
]=
2
3(t s) .
T
Exercise 5.4.1 (a) E[ dWt ] = E[WT ] = 0.
T 0
1 1
(b) E[ Wt dWt ] = E[ WT2 T ] = 0.
0 2 2
T
T
1 1 1 T2
(c) V ar( Wt dWt ) = E[( Wt dWt )2 ] = E[ WT2 + T 2 T WT2 ] = .
0 0 4 4 2 2
T
Exercise 5.6.3 X N (0, 1 1t dt) = N (0, ln T ).
T
Exercise 5.6.4 Y N (0, 1 tdt) = N 0, 12 (T 2 1)
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 297
Exercise
t 5.6.5 Normally distributed with zero mean and variance
1
e2(ts) ds = (e2t 1).
0 2
Exercise 5.6.6 Using the property of Wiener integrals, both integrals have
7t3
zero mean and variance .
3
Exercise 5.6.7 The mean is zero and the variance is t/3 0 as t 0.
Exercise 5.6.8 Since it is a Wiener integral, Xt is normally distributed with
zero mean and variance
t
bu 2 b2
a+ du = (a2 + + ab)t.
0 t 3
b2
Hence a2 + 3 + ab = 1.
t
Exercise 5.6.9 Since both Wt and 0 f (s) dWs have the mean equal to zero,
t t t t
Cov Wt , f (s) dWs = E[Wt , f (s) dWs ] = E[ dWs f (s) dWs ]
0 0 0 0
t t
= E[ f (u) ds] = f (u) ds.
0 0
Nt 2
Nt
Nt
f (Sk ) = f 2 (Sk ) + 2 f (Sk )f (Sj ).
k=1 k=1 k=j
Chapter 6
t
Exercise 6.1.6 Let Xt = 0 eWu du. Then
X tdX X dt teWt dt Xt dt 1 Wt
t t t
dGt = d = = = e Gt dt.
t t2 t2 t
Exercise 6.2.4
(a) eWt (1 + 12 Wt )dt + eWt (1 + Wt )dWt ;
(b) (6Wt + 10e5Wt )dWt + (3 + 25e5Wt )dt;
2 2
(c) 2et+Wt (1 + Wt2 )dt + 2et+Wt Wt dWt ;
(d) n(t + Wt )n2 (t + Wt + n12 )dt + (t + W t )dW t ;
1 1 t
(e) Wt Wu du dt;
t t 0
t
1
(f ) eWt eWu du dt.
t t 0
Exercise 6.2.5
d(tWt2 ) = td(Wt2 ) + Wt2 dt = t(2Wt dWt + dt) + Wt2 dt = (t + Wt2 )dt + 2tWt dWt .
and obtain
t
E[Mt2 |Fs ] = Ms2 + 2E[ Mu dMu |Fs ] + E[Nt |Fs ] Ns
s
= Ms2 + E[Mt + t|Fs ] Ns
= Ms2 + Ms + t Ns
= Ms2 + (t s).
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 299
f f W1 W2 1
dRt = dWt1 + dWt2 + f dt = t dWt1 + t dWt2 + dt.
x y Rt Rt 2Rt
Chapter 7
Exercise 7.2.4 (a) Use integration formula with g(x) = tan1 (x).
T T T
1 1 1 1 2Wt
dWt = (tan ) (Wt ) dWt = tan WT + dt.
0 1 + Wt2 0 2 0 (1 + Wt )2
1 T
(b) Use E 2 dW t = 0.
0 1 + Wt
x
(c) Use Calculus to nd minima and maxima of the function (x) = .
(1 + x2 )2
Exercise 7.2.5 (a) Use integration by parts with g(x) = ex and get
T T
1
Wt
e dWt = e WT
1 eWt dt.
0 2 0
1 T Wt
E[WT e WT
] = E[e ] 1 +
WT
E[e ] + E[Wt eWt ] dt
2 0
T
1
= eT /2 1 + et/2 + E[Wt eWt ] dt.
2 0
1
d(t, Wt ) = (t (t, Wt ) + x2 (t, Wt ))dt + x (t, Wt )dWt
2
= G(t)dt + f (t, Wt ) dWt .
Chapter 8
t
Exercise 8.2.2 (a) Xt = 1 + sin t 0 sin s dWs , E[Xt ] = 1 + sin t, V ar[Xt ] =
t 2 1
0 (sin s) ds = 2
t
4 sin(2t);
t 2
(b) Xt = e 1 + 0 s dWs , E[Xt ] = et 1, V ar[Xt ] = t2 ;
t
4
(c) Xt = 1 + 12 ln(1 + t2 ) + 0 s3/2 dWs , E[Xt ] = 1 + 12 ln(1 + t2 ), V ar(Xt ) = t4 .
t
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 301
so Xt = tWt + Wt2 .
(b) We have
1 1
dXt = (2t Wt )dt + dWt
t2 t
1 1
= 2tdt + d( Wt ) = d(t2 + Wt ),
t t
so Xt = t2 + 1t Wt 1 W1 .
1
(c) dXt = et/2 Wt dt + et/2 dWt = d(et/2 Wt ), so Xt = et/2 Wt .
2
(d) We have
2
Then t Xt = X0 and hence Xt = X0 eWt 2
t
.
2
Wt + 2 t
(b) t = e , d(t Xt ) = t Xt dt, dYt = Yt dt, Yt = Y0 et , t Xt = X0 et ,
2
(1 2 )t+Wt
Xt = X0 e .
Exercise 8.8.3 t dAt = dXt At dt, E[At ] = 0, V ar(At ) = E[A2t ] =
1 t 2 X02
t2 0 E[Xs ] ds = t .
t t
Exercise 8.10.1 Integrating yields Xt = X0 + 0 (2Xs + e2s ) ds + 0 b dWs .
Taking the expectation we get
t
E[Xt ] = X0 + (2E[Xs ] + e2s ) ds.
0
Dierentiating we obtain f (t) = 2f (t) + e2t , where f (t) = E[Xt ], with f (0) =
X0 . Multiplying by the integrating factor e2t yields (e2t f (t)) = 1. Inte-
grating yields f (t) = e2t (t + X0 ).
Exercise 8.10.6 (a) Using product rule and Itos formula, we get
1
d(Wt2 eWt ) = eWt (1 + 2Wt + Wt2 )dt + eWt (2Wt + Wt2 )dWt .
2
Integrating and taking expectations yields
t
1
E[Wt2 eWt ] = E[eWs ] + 2E[Ws eWs ] + E[Ws2 eWs ] ds.
0 2
Since E[eWs ] = et/2 , E[Ws eWs ] = tet/2 , if let f (t) = E[Wt2 eWt ], we get by
dierentiation
1
f (t) = et/2 + 2tet/2 + f (t), f (0) = 0.
2
Chapter 9
n
Exercise 9.2.2 A = 12 = 1
2
2
k=1 k .
Exercise 9.2.4 (a) We have
X1 (t) = x01 + W1 (t)
t
0
X2 (t) = x2 + X1 (s) dW2 (s)
0
t
0 0
= x2 + x1 W2 (t) + W1 (s) dW2 (s).
0
Chapter 10
Exercise 10.1.7 Apply Example 10.1.6 with u = 1.
t t
Exercise 10.1.8 Xt = 0 h(s) dWs N (0, 0 h2 (s) ds). Then eXt is log-
1 1
t
h(s)2 ds
normal with E[eXt ] = e 2 V ar(Xt ) = e 2 0 .
Exercise 10.1.9 (a) Using Exercise 10.1.8 we have
t t
u(s) dWs 12 u(s)2 ds
E[Mt ] = E[e 0 e 0 ]
t t t t
12 u(s)2 ds 1
u(s)2 ds 1
u(s)2 ds
= e 0 E[e 0
u(s) dWs
] = e 2 0 e2 0 = 1.
Integrating yields
t
et/2
cos Wt = 1 es/2 sin Ws dWs ,
0
Exercise 10.4.15 (c) We evaluate EQ [Xt ] in two ways. On one side EQ [Xt ] =
0, because Xt is a Q-Brownian motion. On the other side, using Girsanov
theorem
t2 T 2 T 3
EQ [Xt ] = EP [Xt MT ] = EP + Wt e 0 s dWs e 6
2
t2 2
6 T 3
t T
= +e EP [Wt e 0 s dWs
]EP [e t s dWs
]
2
t2 t t3
= + EP [Wt e 0 s dWs ]e 6 .
2
Equating to zero yields
t t2 t3
EP [Wt e 0
s dWs
]= e 6 .
2
Chapter 11
Exercise 11.2.2 (a) Integrating yields
t
vt = gt + vs dWs ,
0
so E[vt ] = gt.
(b) at = g + vt Nt , E[at ] = g.
1 2
(c) Multiply by the integrating factor t = eWt + 2 t and obtain the exact
equation
d(t vt ) = t g dt.
Integrating we get
t
Wt 12 2 t 1 2
vt = ge eWs + 2 s ds.
0
and obtain N (0) = 12 E[eh ]N (0), which implies the desired result.
(b) Jensens inequality for the random variable h becomes E[eh ] eE[h] .
This can be written as 2 eE[h] .
Exercise 11.3.2 (a) Use that N (t) = N (0)et . (b) t = (ln 0.9)/.
Exercise 11.4.1 (a) Q0 (t) = c1 et + c2 e2t . (c) Qt = Qpt + Q0 (t).
Exercise 11.4.2 (a) Let Zt = (Xt , Yt )T and write the equation as dZt =
AZt + KdWt and solve it as a linear equation. (b) Use substitutions Xt = t ,
Yt = t .
Exercise 11.5.1 (a) Pt is log-normally distributed, with
t 2
P (Pt x) = P P0 e 0 a(s)ds t/2+Bt
1 x t 1 t
= P Bt ln + a(s)ds
P0 2 0
1 t
x t 1
= FBt ln + a(s)ds ,
P0 2 0
2
1 e 2t
u
where FBt = 2t
.
Bibliography
[10] O. Calin, D.C. Chang, K. Furutani, and C. Iwasaki. Heat Kernels for
Elliptic and Sub-elliptic Operators. Birkhauser, Applied and Numerical
Harmonic Analysis, 2011.
309
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 310
[14] J.L. Doob. Stochastic Processes. John Wiley and Sons, 1953.
[21] P.R. Halmos. Measure Theory. Van Nostrand Company, Inc., 1950.
[24] K. Ito. Stochastic Integral. Proc. Imp. Acad. Tokyo, 20, 1944, pp. 519-524.
[25] K. Ito. On a stochastic integral equation. Proc. Imp. Acad. Tokyo, 22,
1946, pp. 32-35.
Bibliography 311
[39] F.A.A. Postali and P. Picchetti. Geometric Brownian Motion and struc-
tural breaks in oil prices: A quantitative analysis. Energy Economics, 28,
pp. 506-522, 2006.
[43] S. M. Ross. Stochastic Processes. Second ed., John Wiley & Sons, Inc.,
1996.
[46] D.V. Widder. The Heat Equation. Academic Press, London, 1975.
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 312
Index
313
May 15, 2015 14:45 BC: 9620 An Informal Introduction to Stochastic Calculus Driverbook page 314
Index 315