STAT3603

Jeff Yao

STAT3603 Stochastic Processes

Jeff Yao
STAT3603

Jeff Yao
Part I
Introduction
to
Probability
theory

Random Probability tools

variables

Conditional
Probabili-
ties and
Conditional
Expectation
1 Introduction to Probability theory

2 Random variables

3 Conditional Probabilities and Conditional Expectation

STAT3603

Jeff Yao
Part II
Introduction

Chapman-
Kolmogorov
Equation

Classification
Markov Chains
of states

Limiting
Probabili- 1 Introduction
ties

Mean Time
Spent in 2 Chapman-Kolmogorov Equations
Transient
States

Applications
3 Classification of States

4 Limiting Probabilities

5 Mean Time Spent in Transient States

6 Applications
STAT3603

Jeff Yao Part III

The Poisson
Process

Interarrival
and Waiting
The Poisson Process
Time Distri-
butions

Conditional
Distribution 1 The Exponential Distribution
of the
Arrival
Times 2 Independent exponential variables
Further
Properties
of Poisson 3 The Poisson Process
Process

Compound
Poisson
4 Interarrival and waiting time distributions
Process

Non- 5 Conditional Distribution of the Arrival Times

homogeneous
Poisson
process 6 Further Properties of Poisson Processes

7 Compound Poisson Process

 §3.6 Further Properties of Poisson Process

STAT3603

Jeff Yao

The Poisson
Process

Interarrival
and Waiting
Time Distri-
Colouring theorem of Poisson distribution
butions

Conditional
Theorem 7 Assume that a counting variable N has Poisson distribution P(λ);
Distribution
of the
that the counted bowls are red with probability p and blue with probability
Arrival q = 1 − p. Let NR and NB be the counts of red and blue bowls, respectively.
Times
Then, NR ∼ P(λp), NB ∼ P(λq) and they are independent.
Further
Properties
of Poisson
Process

Compound
Poisson
Process Two types of bowls
Non-
homogeneous
Poisson
process
 Proof of the colouring theorem

STAT3603

Jeff Yao
I Clearly, given N = n, NR ∼ B(n, p) and NB = n − NR . So
The Poisson
Process !
Interarrival n k n−k
and Waiting P(NR = k, NB = n − k | N = n) = p q .
Time Distri- k
butions

Conditional
Distribution
of the
Arrival I Therefore, for k, ` ≥ 0,
Times

Further
P(NR = k, NB = `)
Properties
of Poisson = P(NR = k, NB = ` | N = k + `) · P(N = k + `)
Process !
k + ` k ` −λ λk+`
Compound
Poisson
= p q ·e
Process
k (k + `)!
Non-
(λp)k (λq)`
homogeneous
Poisson
= e −λp · e −λq .
process
k! `!

That is, NR ∼ P(λp), NB ∼ P(λq) and they are independent.

 Proof of the colouring theorem

STAT3603

Jeff Yao
I Clearly, given N = n, NR ∼ B(n, p) and NB = n − NR . So
The Poisson
Process !
Interarrival n k n−k
and Waiting P(NR = k, NB = n − k | N = n) = p q .
Time Distri- k
butions

Conditional
Distribution
of the
Arrival I Therefore, for k, ` ≥ 0,
Times

Further
P(NR = k, NB = `)
Properties
of Poisson = P(NR = k, NB = ` | N = k + `) · P(N = k + `)
Process !
k + ` k ` −λ λk+`
Compound
Poisson
= p q ·e
Process
k (k + `)!
Non-
(λp)k (λq)`
homogeneous
Poisson
= e −λp · e −λq .
process
k! `!

That is, NR ∼ P(λp), NB ∼ P(λq) and they are independent.

 Colouring or decomposition of a Poisson process

STAT3603

Jeff Yao I Let Nt be a Poisson process with rate λ;

The Poisson
Process I Suppose that the events can be classified into two types: type I with
Interarrival
and Waiting
probability p (red bowls!) and type II with probability q = 1 − p (blue
Time Distri- bowls!), independently of all other events.
butions

Conditional
Distribution I Let Nt1 and Nt2 be the numbers of type I and II events occurring in [0, t]
of the
Arrival respectively. Note that Nt = Nt1 + Nt2 .
Times

Further
Properties
of Poisson
Process
Two types of bowls
Compound
Poisson
Process I For example, suppose that customers arrive at a store in accordance with
Non-
homogeneous
a Poisson process having rate λ;
Poisson
process
I suppose that each arrival is male with probability 12 and female with
probability 21 . Then type I event would correspond to a male arrival and a
type II event to a female arrival.
 Colouring or decomposition of a Poisson process

STAT3603

Jeff Yao I Let Nt be a Poisson process with rate λ;

The Poisson
Process I Suppose that the events can be classified into two types: type I with
Interarrival
and Waiting
probability p (red bowls!) and type II with probability q = 1 − p (blue
Time Distri- bowls!), independently of all other events.
butions

Conditional
Distribution I Let Nt1 and Nt2 be the numbers of type I and II events occurring in [0, t]
of the
Arrival respectively. Note that Nt = Nt1 + Nt2 .
Times

Further
Properties
of Poisson
Process
Two types of bowls
Compound
Poisson
Process I For example, suppose that customers arrive at a store in accordance with
Non-
homogeneous
a Poisson process having rate λ;
Poisson
process
I suppose that each arrival is male with probability 12 and female with
probability 21 . Then type I event would correspond to a male arrival and a
type II event to a female arrival.
 Colouring theorem of Poisson process

STAT3603

Jeff Yao Theorem 8. The counting processes {Nt1 , t ≥ 0} and {Nt2 : t ≥ 0} of type I
and II events are both Poisson processes having respectively rates pλ and qλ.
The Poisson
Process
Furthermore, they are independent. Proof (steps).
Interarrival
and Waiting
Time Distri-
I Take t1 < · · · < tn , define ∆tj = tj − tj−1 ,
butions

Conditional
∆Ntj = Ntj − Ntj−1 = ∆Nt1j + ∆Nt2j ,
Distribution
of the
Arrival
where ∆Ntj (respectively, ∆Nt1j and ∆Nt2j ) counts the number of events on
Times (tj−1 , tj ] (resp. events of type I and II).
Further
Properties
of Poisson I By assumption, the ∆Ntj ’s are independent P(λ∆tj );
Process

Compound
Poisson I By Theorem 4 (colouring theorem), ∆Nt1 and ∆Nt2 are independent and
j j
Process
distributed as P(pλ∆tj ) and P(pλ∆tj ) respectively.
Non-
homogeneous Then Nt1 = Nt1 − N01 and Nt2 = Nt2 − N02 and two indepedent Poisson processes with
Poisson
process rate pλ and qλ, respectively.

I Note. Both colouring theorems can be extended to more than two colours.
Exercise. Write-out these general colouring theorems.
 Colouring theorem of Poisson process

STAT3603

Jeff Yao Theorem 8. The counting processes {Nt1 , t ≥ 0} and {Nt2 : t ≥ 0} of type I
and II events are both Poisson processes having respectively rates pλ and qλ.
The Poisson
Process
Furthermore, they are independent. Proof (steps).
Interarrival
and Waiting
Time Distri-
I Take t1 < · · · < tn , define ∆tj = tj − tj−1 ,
butions

Conditional
∆Ntj = Ntj − Ntj−1 = ∆Nt1j + ∆Nt2j ,
Distribution
of the
Arrival
where ∆Ntj (respectively, ∆Nt1j and ∆Nt2j ) counts the number of events on
Times (tj−1 , tj ] (resp. events of type I and II).
Further
Properties
of Poisson I By assumption, the ∆Ntj ’s are independent P(λ∆tj );
Process

Compound
Poisson I By Theorem 4 (colouring theorem), ∆Nt1 and ∆Nt2 are independent and
j j
Process
distributed as P(pλ∆tj ) and P(pλ∆tj ) respectively.
Non-
homogeneous Then Nt1 = Nt1 − N01 and Nt2 = Nt2 − N02 and two indepedent Poisson processes with
Poisson
process rate pλ and qλ, respectively.

I Note. Both colouring theorems can be extended to more than two colours.
Exercise. Write-out these general colouring theorems.
 Colouring theorem of Poisson process

STAT3603

Jeff Yao Theorem 8. The counting processes {Nt1 , t ≥ 0} and {Nt2 : t ≥ 0} of type I
and II events are both Poisson processes having respectively rates pλ and qλ.
The Poisson
Process
Furthermore, they are independent. Proof (steps).
Interarrival
and Waiting
Time Distri-
I Take t1 < · · · < tn , define ∆tj = tj − tj−1 ,
butions

Conditional
∆Ntj = Ntj − Ntj−1 = ∆Nt1j + ∆Nt2j ,
Distribution
of the
Arrival
where ∆Ntj (respectively, ∆Nt1j and ∆Nt2j ) counts the number of events on
Times (tj−1 , tj ] (resp. events of type I and II).
Further
Properties
of Poisson I By assumption, the ∆Ntj ’s are independent P(λ∆tj );
Process

Compound
Poisson I By Theorem 4 (colouring theorem), ∆Nt1 and ∆Nt2 are independent and
j j
Process
distributed as P(pλ∆tj ) and P(pλ∆tj ) respectively.
Non-
homogeneous Then Nt1 = Nt1 − N01 and Nt2 = Nt2 − N02 and two indepedent Poisson processes with
Poisson
process rate pλ and qλ, respectively.

I Note. Both colouring theorems can be extended to more than two colours.
Exercise. Write-out these general colouring theorems.
 Example 3.14

STAT3603

Jeff Yao

The Poisson
Process If immigrants to area A arrive at a Poisson rate of ten per week, and if each
1
Interarrival immigrant is of English descent with probability 12 , then what is the
and Waiting
Time Distri- probability that no people of English descent will emigrate to area A during
butions
the month of February?
Conditional
Distribution
of the
Arrival
Times

Further
Properties
of Poisson
Process

Compound
Poisson
Process

Non-
homogeneous
10
Poisson
process Solution: e − 3 .
 Example 3.14

STAT3603

Jeff Yao

The Poisson
Process If immigrants to area A arrive at a Poisson rate of ten per week, and if each
1
Interarrival immigrant is of English descent with probability 12 , then what is the
and Waiting
Time Distri- probability that no people of English descent will emigrate to area A during
butions
the month of February?
Conditional
Distribution
of the
Arrival
Times

Further
Properties
of Poisson
Process

Compound
Poisson
Process

Non-
homogeneous
10
Poisson
process Solution: e − 3 .
 Marriage proposals

STAT3603

Jeff Yao

The Poisson
Process Boyfriends appear to Sarah as a Poisson process with rate 1/10 (one per ten
Interarrival month). However only 1/4 of the boyfriends are willing to marry her. Assume
and Waiting
Time Distri- that Sarah accepts the first marriage proposal she receives.
butions
What is the distribution of T , the waiting time for the marriage of Sarah ?
Conditional
Distribution And E [T ] ?
of the
Arrival
Times

Further
Properties
of Poisson
Process

Compound
Poisson
Process

Non-
homogeneous
Poisson
process Solution: T ∼ E(1/40) and E [T ] = 40 (months).
 Marriage proposals

STAT3603

Jeff Yao

The Poisson
Process Boyfriends appear to Sarah as a Poisson process with rate 1/10 (one per ten
Interarrival month). However only 1/4 of the boyfriends are willing to marry her. Assume
and Waiting
Time Distri- that Sarah accepts the first marriage proposal she receives.
butions
What is the distribution of T , the waiting time for the marriage of Sarah ?
Conditional
Distribution And E [T ] ?
of the
Arrival
Times

Further
Properties
of Poisson
Process

Compound
Poisson
Process

Non-
homogeneous
Poisson
process Solution: T ∼ E(1/40) and E [T ] = 40 (months).
 A technical lemma

STAT3603

Jeff Yao

The Poisson
Process
Definition and Lemma. Let X be a nonnegative r.v. (lifetime) with density
Interarrival
function f (x).
and Waiting
Time Distri-
butions
For a > 0, the restriction Xa of X to [a, ∞) is denoted as X |X > a; this r.v.
Conditional
has the density
Distribution
of the 1
Arrival fX |X >a (x) = f (x)1{x>a} .
Times F (a)
Further
Properties
of Poisson
Process
This is the survival distribution after time a.
Compound
Poisson
Its expectation equals
Process Z ∞
1 E [X 1{X >a} ]
Non-
homogeneous
E [Xa ] = xf (x)dx = = E [X |X > a].
Poisson F (a) a P(X > a)
process
 Example 3.15

STAT3603

Jeff Yao Assume that

The Poisson
I Nonnegative offers to buy an item that you want to sell arrive according
Process
to a P.P. with rate λ;
Interarrival
and Waiting
Time Distri- I each offer is the value of a continuous r.v. X having density function
butions
f (x);
Conditional
Distribution
of the
Arrival
I Once the offer is presented to you, you must either accept it or reject it
Times
and wait for the next offer;
Further
Properties
of Poisson I you incur costs at a rate c per unit time until the item is sold;
Process

Compound
Poisson
I your objective is to maximise your expected total return, where the total
Process
return is equal to the amount received minus the total cost incurred;
Non-
homogeneous
Poisson I you employ the policy of accepting the first offer that is greater than
process
some specified value y . (Such a type of policy, which we call a y -policy,
can be shown to be optimal.)
Question: What is the best value of y ?
STAT3603
Cont’d. Solution:
Jeff Yao
I Red offers: “> y ” with probability F = P{X > y };
The Poisson
Process
I Blue offers: “≤ y ” with probability F (y ) = P{X ≤ y };
Interarrival
and Waiting
Time Distri- I Red offers occur according to a P.P. with rate λF (y );
butions

Conditional
Distribution
I the time until an offer is accepted is an exponential random variable T1
of the
Arrival
with rate λF (y ).
Times

Further
Waiting time T1
Properties
of Poisson 0 First “>y” offer
Process
I Amount received: X |X > y ; Costs: cT1 .
Compound
Poisson
Process
I Let R(y ) denote the total return from this y -policy. Then,
Non-
homogeneous E [R(y )] = E (X |X > y ) − E (cT1 )
Poisson Z ∞
process 1 c
= xf (x)dx −
F (y ) y λF (y )
Z ∞ 
1 c
= xf (x)dx − .
F (y ) y λ
STAT3603
Cont’d. Solution:
Jeff Yao
I Red offers: “> y ” with probability F = P{X > y };
The Poisson
Process
I Blue offers: “≤ y ” with probability F (y ) = P{X ≤ y };
Interarrival
and Waiting
Time Distri- I Red offers occur according to a P.P. with rate λF (y );
butions

Conditional
Distribution
I the time until an offer is accepted is an exponential random variable T1
of the
Arrival
with rate λF (y ).
Times

Further
Waiting time T1
Properties
of Poisson 0 First “>y” offer
Process
I Amount received: X |X > y ; Costs: cT1 .
Compound
Poisson
Process
I Let R(y ) denote the total return from this y -policy. Then,
Non-
homogeneous E [R(y )] = E (X |X > y ) − E (cT1 )
Poisson Z ∞
process 1 c
= xf (x)dx −
F (y ) y λF (y )
Z ∞ 
1 c
= xf (x)dx − .
F (y ) y λ
STAT3603
Differentiation yields that
Jeff Yao
d
The Poisson E (R(Y )) = 0
Process dy
Z ∞
Interarrival c
and Waiting ⇔ −y F (y )f (y ) + xf (x)dx − f (y ) = 0
Time Distri- y λ
butions Z ∞
c
Conditional
Distribution
⇔ (x − y )f (x)dx =
of the y λ
Arrival c
Times ⇔ E [(X − y )+ ] = (∗). a+ = max(a, 0)
Further
λ
Properties
of Poisson
Process

Compound
As y 7→ φ(y ) := E [(X − y )+ ] is decreasing,
Poisson
Process 1 If φ(0) = E [X ] < λc , then φ(y ) ≤ φ(0) < λc for all y , and E (R(Y )) is
Non- always decreasing &.
homogeneous
Poisson
Maximum at y ∗ = 0: → accept any offer .
process

2 Otherwise, φ(0) = E [X ] ≥ λc , then E (R(Y )) is increasing % on [0, y ∗ ),

where φ(y ∗ ) = 0, then decreasing & on (y ∗ , ∞).
The optimal value is y ∗ , solution of (*).
STAT3603
Differentiation yields that
Jeff Yao
d
The Poisson E (R(Y )) = 0
Process dy
Z ∞
Interarrival c
and Waiting ⇔ −y F (y )f (y ) + xf (x)dx − f (y ) = 0
Time Distri- y λ
butions Z ∞
c
Conditional
Distribution
⇔ (x − y )f (x)dx =
of the y λ
Arrival c
Times ⇔ E [(X − y )+ ] = (∗). a+ = max(a, 0)
Further
λ
Properties
of Poisson
Process

Compound
As y 7→ φ(y ) := E [(X − y )+ ] is decreasing,
Poisson
Process 1 If φ(0) = E [X ] < λc , then φ(y ) ≤ φ(0) < λc for all y , and E (R(Y )) is
Non- always decreasing &.
homogeneous
Poisson
Maximum at y ∗ = 0: → accept any offer .
process

2 Otherwise, φ(0) = E [X ] ≥ λc , then E (R(Y )) is increasing % on [0, y ∗ ),

where φ(y ∗ ) = 0, then decreasing & on (y ∗ , ∞).
The optimal value is y ∗ , solution of (*).
 Superposition of two independent P.P

STAT3603

Jeff Yao

The Poisson
Process

Interarrival
and Waiting
Time Distri-
butions
Theorem 9 Let {Nt1 , t ≥ 0} and {Nt2 , t ≥ 0} be two independent P.P having
Conditional
Distribution respective rates λ1 and λ2 . Then their sum, {Nt1 + Nt2 , t ≥ 0}, the process of
of the
Arrival
counts of both events, is a Poisson process with rates λ1 + λ2 .
Times

Further
Properties
of Poisson
Process Proof. Use the definition of a P.P.
Compound
Poisson
Process

Non-
homogeneous
Poisson
process
 An application

STAT3603

Jeff Yao

The Poisson
What is the probability that n events occur in one P.P. before m events have
Process occurred in a second and independent P.P.?
Interarrival
and Waiting
Time Distri- Let
butions
I {Nt1 , t ≥ 0} and {Nt2 , t ≥ 0} be two independent P.P having respective
Conditional
Distribution rates λ1 (red events) and λ2 (blue events)
of the
Arrival
Times I Sn1 denote the time of the nth red event;
Further
Properties
of Poisson I Sm1 denote the time of the mth blue event.
Process

Compound
Poisson
We seek P(Sn1 < Sm2 ).
Process
I Consider first the special case: n = m = 1.
Non-
homogeneous Since S11 and S22 are independent exponentials with rates λ1 and λ2 ,
Poisson
process
λ1
P(S11 < S12 ) = .
λ1 + λ2
(Probability having a red event before a blue one)
 An application

STAT3603

Jeff Yao

The Poisson
What is the probability that n events occur in one P.P. before m events have
Process occurred in a second and independent P.P.?
Interarrival
and Waiting
Time Distri- Let
butions
I {Nt1 , t ≥ 0} and {Nt2 , t ≥ 0} be two independent P.P having respective
Conditional
Distribution rates λ1 (red events) and λ2 (blue events)
of the
Arrival
Times I Sn1 denote the time of the nth red event;
Further
Properties
of Poisson I Sm1 denote the time of the mth blue event.
Process

Compound
Poisson
We seek P(Sn1 < Sm2 ).
Process
I Consider first the special case: n = m = 1.
Non-
homogeneous Since S11 and S22 are independent exponentials with rates λ1 and λ2 ,
Poisson
process
λ1
P(S11 < S12 ) = .
λ1 + λ2
(Probability having a red event before a blue one)
 An application

STAT3603

Jeff Yao

The Poisson
What is the probability that n events occur in one P.P. before m events have
Process occurred in a second and independent P.P.?
Interarrival
and Waiting
Time Distri- Let
butions
I {Nt1 , t ≥ 0} and {Nt2 , t ≥ 0} be two independent P.P having respective
Conditional
Distribution rates λ1 (red events) and λ2 (blue events)
of the
Arrival
Times I Sn1 denote the time of the nth red event;
Further
Properties
of Poisson I Sm1 denote the time of the mth blue event.
Process

Compound
Poisson
We seek P(Sn1 < Sm2 ).
Process
I Consider first the special case: n = m = 1.
Non-
homogeneous Since S11 and S22 are independent exponentials with rates λ1 and λ2 ,
Poisson
process
λ1
P(S11 < S12 ) = .
λ1 + λ2
(Probability having a red event before a blue one)
STAT3603
Cont’d.
Jeff Yao
I Indeed, by viewing both events together as a P.P with rate λ1 + λ2 , we
The Poisson
Process
see that
Interarrival
and Waiting
each event that occurs is going to be an event of the N·1 process with
Time Distri- probability λ1 /(λ1 + λ2 ) or an event of the N·2 process with probability
butions
λ2 /(λ1 + λ2 ), independent of all that has previously occurred.
Conditional
Distribution
of the
Arrival
I In other words, the probability that the N·1 process reaches n before the
Times
N·2 process reaches m is just the probability that n heads will appear
Further
Properties
before m tails if one flips a coin having probability p = λ1 /(λ1 + λ2 ) of a
of Poisson
Process
head appearing.
Compound
Poisson I But by noting that this event will occur if and only if the first n + m − 1
Process
tosses result in n or more heads, we see that our desired probability is
Non-
homogeneous given by
Poisson
process !
n+m−1
X n+m−1
1 2
P(Sn < Sm ) = p k (1 − p)n+m−1−k .
k=n k
STAT3603
Cont’d.
Jeff Yao
I Indeed, by viewing both events together as a P.P with rate λ1 + λ2 , we
The Poisson
Process
see that
Interarrival
and Waiting
each event that occurs is going to be an event of the N·1 process with
Time Distri- probability λ1 /(λ1 + λ2 ) or an event of the N·2 process with probability
butions
λ2 /(λ1 + λ2 ), independent of all that has previously occurred.
Conditional
Distribution
of the
Arrival
I In other words, the probability that the N·1 process reaches n before the
Times
N·2 process reaches m is just the probability that n heads will appear
Further
Properties
before m tails if one flips a coin having probability p = λ1 /(λ1 + λ2 ) of a
of Poisson
Process
head appearing.
Compound
Poisson I But by noting that this event will occur if and only if the first n + m − 1
Process
tosses result in n or more heads, we see that our desired probability is
Non-
homogeneous given by
Poisson
process !
n+m−1
X n+m−1
1 2
P(Sn < Sm ) = p k (1 − p)n+m−1−k .
k=n k
 §3.7 Compound Poisson Process

STAT3603

Jeff Yao
Compound Poisson distribution
The Poisson
Process
I A random variable X is a compound Poisson random variable if
Interarrival
and Waiting
Time Distri-
N
X
butions X = Yi , t≥0
Conditional i=1
Distribution
of the
Arrival where N ∼ P(λ) is a Poisson random variable, {Yi , i ≥ 1} is a family of
Times
i.i.d. r.v. which are independent of N.
Further
Properties
of Poisson
Process
I This is a special case of a random sum.
Compound
Poisson I We have, by the conditioning method (review it!),
Process

Non-
homogeneous
(∗) E[X ] = λE [Y1 ];
Poisson
process (∗∗) Var(X ) = λE (Y12 ).
 §3.7 Compound Poisson Process

STAT3603

Jeff Yao
Compound Poisson distribution
The Poisson
Process
I A random variable X is a compound Poisson random variable if
Interarrival
and Waiting
Time Distri-
N
X
butions X = Yi , t≥0
Conditional i=1
Distribution
of the
Arrival where N ∼ P(λ) is a Poisson random variable, {Yi , i ≥ 1} is a family of
Times
i.i.d. r.v. which are independent of N.
Further
Properties
of Poisson
Process
I This is a special case of a random sum.
Compound
Poisson I We have, by the conditioning method (review it!),
Process

Non-
homogeneous
(∗) E[X ] = λE [Y1 ];
Poisson
process (∗∗) Var(X ) = λE (Y12 ).
 Compound Poisson Process

STAT3603

Jeff Yao

The Poisson
Process
Definition: {X (t) : t ≥ 0} is called a compound Poisson process if
Interarrival
and Waiting
Time Distri- Nt
X
butions
Xt = Yi , t≥0
Conditional
Distribution i=1
of the
Arrival
Times
where Nt is a Poisson process, {Yi , i ≥ 1} is a family of independent and
Further identically distributed random variables which are also independent of
Properties
of Poisson
{Nt : t ≥ 0}.
Process

Compound
Poisson
Examples
Process
I If Yi = 1 for all i, then Xt = Nt , and we have the usual Poisson process.
Non-
homogeneous Hence Poisson process is a special case of compound Poisson process.
Poisson
process
STAT3603
Cont’d.
Jeff Yao
I Suppose that buses arrive at a sporting event in accordance with a
The Poisson
Process
Poisson process, and suppose that the numbers of fans in each bus are
Interarrival
assumed to be i.i.d.
and Waiting
Time Distri-
Then {Xt , t ≥ 0} is a compound Poisson process, where Nt represents
butions the number of buses arriving at the sporting event by time
P t, Yi
Conditional represents the number of fans in the ith bus, and Xt = N t
i=1 i denotes
Y
Distribution
of the the number of fans who have arrived by time t.
Arrival
Times

Further
Properties
of Poisson
Process

Compound
Poisson
Process I Suppose customers leave a supermarket in accordance with a Poisson
Non-
homogeneous process. If Yi , the amount spent by the ith customer, i = 1, 2, . . ., are
Poisson
process
i.i.d.
Then {Xt , t ≥ 0} is a compound Poisson process, where Nt denotes the
number P oft customers leaving the supermarket by time t and
Xt = Ni=1 Yi denotes the total amount of money spent by time t.
STAT3603
Cont’d.
Jeff Yao
I Suppose that buses arrive at a sporting event in accordance with a
The Poisson
Process
Poisson process, and suppose that the numbers of fans in each bus are
Interarrival
assumed to be i.i.d.
and Waiting
Time Distri-
Then {Xt , t ≥ 0} is a compound Poisson process, where Nt represents
butions the number of buses arriving at the sporting event by time
P t, Yi
Conditional represents the number of fans in the ith bus, and Xt = N t
i=1 i denotes
Y
Distribution
of the the number of fans who have arrived by time t.
Arrival
Times

Further
Properties
of Poisson
Process

Compound
Poisson
Process I Suppose customers leave a supermarket in accordance with a Poisson
Non-
homogeneous process. If Yi , the amount spent by the ith customer, i = 1, 2, . . ., are
Poisson
process
i.i.d.
Then {Xt , t ≥ 0} is a compound Poisson process, where Nt denotes the
number P oft customers leaving the supermarket by time t and
Xt = Ni=1 Yi denotes the total amount of money spent by time t.
 Moments of (Xt )

STAT3603

Jeff Yao

The Poisson
Process

Interarrival
and Waiting
Time Distri-
butions
I For any fixed t, Xt is just a simple Compound Poisson random variable;
Conditional
Distribution
of the I so that
Arrival
Times
E (X (t)) = λtE (Y1 ) ,
Further
Properties Var(Xt ) = λtE (Y12 ) .
of Poisson
Process

Compound
Poisson
Process

Non-
homogeneous
Poisson
process
 Example 5.28: migration again

STAT3603
Suppose that
Jeff Yao
I families migrate to an area at a Poisson rate λ = 2 per week;
The Poisson
Process

Interarrival
I the number of people in each family is independent and takes on the
and Waiting
Time Distri-
values of 1, 2, 3, 4 with respective probabilities 61 , 31 , 13 , 16 .
butions
Question.
Conditional
Distribution
of the
1 What is the expected value and variance of the number of individuals
Arrival migrating to this area during a fixed five-week period?
Times

Further
Properties 2 Find the approximate probability that at least 240 people migrate to the
of Poisson
Process
area within the next 50 weeks.
Compound
Poisson
Process

Non-
homogeneous
Solution: λ = 2, E [Y1 ] = 5/2, E [Y12 ] = 43/6,
Poisson
215
process So that E (X5 ) = 25, Var(X5 ) = 3
.
p
Approximate probability: P{X50 ≥ 240} ≈ Φ( 30/215 = 0.3735) ≈ 0.6456.
 Example 5.28: migration again

STAT3603
Suppose that
Jeff Yao
I families migrate to an area at a Poisson rate λ = 2 per week;
The Poisson
Process

Interarrival
I the number of people in each family is independent and takes on the
and Waiting
Time Distri-
values of 1, 2, 3, 4 with respective probabilities 61 , 31 , 13 , 16 .
butions
Question.
Conditional
Distribution
of the
1 What is the expected value and variance of the number of individuals
Arrival migrating to this area during a fixed five-week period?
Times

Further
Properties 2 Find the approximate probability that at least 240 people migrate to the
of Poisson
Process
area within the next 50 weeks.
Compound
Poisson
Process

Non-
homogeneous
Solution: λ = 2, E [Y1 ] = 5/2, E [Y12 ] = 43/6,
Poisson
215
process So that E (X5 ) = 25, Var(X5 ) = 3
.
p
Approximate probability: P{X50 ≥ 240} ≈ Φ( 30/215 = 0.3735) ≈ 0.6456.
STAT3603
Solution:
Jeff Yao

The Poisson
Process

Interarrival
and Waiting
Time Distri-
butions

Conditional
Distribution
of the
Arrival
Times

Further
Properties
of Poisson
Process

Compound
Poisson
Process

Non-
homogeneous
Poisson
process
 §3.8. Non-homogeneous Poisson process

STAT3603 Remarks:
Jeff Yao I A standard P.P.(λ) has the memoryless property; in particular the
The Poisson interarrival times are i.i.d. E(λ);
Process

Interarrival I This is of cause a very strong assumption and we need more

and Waiting
Time Distri- general/flexible models without this memoryless property.
butions

Conditional
Distribution
of the
Definition
Arrival
Times A counting process {Nt : t ≥ 0} is a non-homogeneous Poisson process with a
Further continuous intensity function λ(t) : t ≥ 0, if
Properties
of Poisson 1 N(0) = 0;
Process

Compound
Poisson 2 {Nt , t ≥ 0} has independent increments;
Process

Non- 3 For small positive h > 0,

homogeneous
Poisson
process
• P{Nt+h − Nt ≥ 2} = o(h);

• P{Nt+h − Nt = 1} = λ(t)h + o(h).

[In consequence, P{Nt+h − Nt = 0} = 1 − λ(t)h + o(h).]

 §3.8. Non-homogeneous Poisson process

STAT3603 Remarks:
Jeff Yao I A standard P.P.(λ) has the memoryless property; in particular the
The Poisson interarrival times are i.i.d. E(λ);
Process

Interarrival I This is of cause a very strong assumption and we need more

and Waiting
Time Distri- general/flexible models without this memoryless property.
butions

Conditional
Distribution
of the
Definition
Arrival
Times A counting process {Nt : t ≥ 0} is a non-homogeneous Poisson process with a
Further continuous intensity function λ(t) : t ≥ 0, if
Properties
of Poisson 1 N(0) = 0;
Process

Compound
Poisson 2 {Nt , t ≥ 0} has independent increments;
Process

Non- 3 For small positive h > 0,

homogeneous
Poisson
process
• P{Nt+h − Nt ≥ 2} = o(h);

• P{Nt+h − Nt = 1} = λ(t)h + o(h).

[In consequence, P{Nt+h − Nt = 0} = 1 − λ(t)h + o(h).]

 Distribution of the increments

STAT3603 Z t
Jeff Yao Define for t ≥ 0, mt = λ(u)du..
0
The Poisson
Process
Theorem 10
Interarrival
and Waiting For a Poisson P.P. with intensity function λ(t), we have for 0 ≤ s < t,
Time Distri-
butions Z t 
Conditional
Distribution
Nt − Ns ∼ P (m(t) − m(s)) = P λ(u)du .
of the s
Arrival
Times

Further
Properties
of Poisson Comments
Process

Compound
I The increment distribution depends on both the starting time s and the
Poisson
Process
final time t; not only on the between length t − s;
Non-
homogeneous I The increments are then non stationary (a process is stationary if its
Poisson
process distribution is invariant under time shift)

I In particular, Nt ∼ P(m(t)): the function m(t) is called the mean value

function of the non-homogeneous Possion process.
 Distribution of the increments

STAT3603 Z t
Jeff Yao Define for t ≥ 0, mt = λ(u)du..
0
The Poisson
Process
Theorem 10
Interarrival
and Waiting For a Poisson P.P. with intensity function λ(t), we have for 0 ≤ s < t,
Time Distri-
butions Z t 
Conditional
Distribution
Nt − Ns ∼ P (m(t) − m(s)) = P λ(u)du .
of the s
Arrival
Times

Further
Properties
of Poisson Comments
Process

Compound
I The increment distribution depends on both the starting time s and the
Poisson
Process
final time t; not only on the between length t − s;
Non-
homogeneous I The increments are then non stationary (a process is stationary if its
Poisson
process distribution is invariant under time shift)

I In particular, Nt ∼ P(m(t)): the function m(t) is called the mean value

function of the non-homogeneous Possion process.
 Proof of Theorem 10 (using Poisson approximation)

STAT3603

Jeff Yao

The Poisson
Process

Interarrival
and Waiting
Time Distri-
butions

Conditional Poisson Approximation lemma

Distribution
of the
Arrival Let (Xi )1≤i≤n be a sequence of independent Bernoulli b(pi ) r.v’s with
Times
respective success probability (pi ). Assume that
Further
Properties
of Poisson
when n → ∞, p1 + · · · + pn → λ > 0.
Process

Compound
Then the sum X1 + · · · + Xn converges in distribution to P(λ) as n → ∞.
Poisson
Process

Non-
homogeneous
Poisson
process
 Proof of Theorem 10 (using Poisson approximation)

STAT3603
I Divide the interval (s, t] in n equal bins of size h = (t − s)/n,
Jeff Yao

The Poisson
s = a0 < a1 < · · · < an = t, ai = s + ih ;
Process

Interarrival Therefore
and Waiting
Time Distri- n
X n
X
butions
Nt − Ns = [Nai − Nai−1 ] = Xi ;
Conditional
Distribution i=1 i=1
of the
Arrival
Times

Further I By the very definition of the process, the increments Xi ’s are

Properties
of Poisson independent, approximately Bernoulli with parameter pi = hλ(ai );
Process

Compound I When n → ∞, we have pi → 0 and

Poisson
Process
n
X Z t
Non-
homogeneous p1 + · · · + pn = hλ(ai ) → λ(y )dy ;
Poisson s
process i=1

Pn
I By Poisson approximation, i=1 Xi has Poisson distribution with mean
R t+s
t
λ(y )dy .
 Example 5.24

STAT3603
Siegbert runs a hot dog stand that opens at 8 a.m. and
Jeff Yao
I From 8 until 11 a.m. customers seem to arrive, on the average, at a steadily
The Poisson increasing rate that starts with an initial rate of 5 customers per hour at 8 a.m.
Process and reaches a maximum of 20 customers per hour at 11 a.m.
Interarrival
and Waiting
Time Distri- I From 11 a.m. until 1 p.m. the (average) rate seems to remain constant at 20
butions
customers per hour.
Conditional
Distribution
of the I However, the (average) arrival rate then drops steadily from 1 p.m. until closing
Arrival
Times time at 5 p.m. at which time it has the value of 12 customers per hour.
Further 
Properties
of Poisson
20 
 5 + 5t, t ∈ [0, 3],

Process




Compound
Poisson
λ(t) = 20, t ∈ [3, 5],
12 
Process 



Non-

20 − 2(t − 5), t ∈ [5, 9].

homogeneous
Poisson 5
process

o o
0 1 2 3 4 5 6 7 8 9
Z 1.5 Z 1.5
8:00am 11:00am 1:00pm 5:00pm λ(t)dt = (5 + 5t)dt = 10.
0.5 0.5
 Questions:

STAT3603

Jeff Yao

The Poisson
Process
I If we assume that the numbers of customers arriving at Siegbert’s stand
Interarrival
and Waiting during disjoint time periods are independent, then what is a good
Time Distri-
butions
probability model for the preceding?
Conditional
Distribution I What is the probability that no customers arrive between 8:30 a.m. and
of the
Arrival 9:30 a.m. on Monday morning?
Times

Further
Properties
I What is the expected number of arrivals in this period?
of Poisson
Process

Compound
Poisson
Process

Non- Solutions. Consider N1.5 − N0.5 .

homogeneous
Poisson
process Solution: e −10 and 10.
 Questions:

STAT3603

Jeff Yao

The Poisson
Process
I If we assume that the numbers of customers arriving at Siegbert’s stand
Interarrival
and Waiting during disjoint time periods are independent, then what is a good
Time Distri-
butions
probability model for the preceding?
Conditional
Distribution I What is the probability that no customers arrive between 8:30 a.m. and
of the
Arrival 9:30 a.m. on Monday morning?
Times

Further
Properties
I What is the expected number of arrivals in this period?
of Poisson
Process

Compound
Poisson
Process

Non- Solutions. Consider N1.5 − N0.5 .

homogeneous
Poisson
process Solution: e −10 and 10.
 A more involved example

STAT3603
(Question from May 2017 exam paper)
Jeff Yao

The Poisson
Process

Interarrival
and Waiting
Time Distri-
butions

Conditional
Distribution
of the
Arrival
Times

Further
Properties
of Poisson
Process

Compound
Poisson
Process
Solution:
Non- Z ∞
8
−4
homogeneous (a) p(s) = P(Y > cs) = 24(2 + y ) dy = .
Poisson cs (2 + cs)3
process (b) We show that Nt is a time-inhomogeneous PPP with intensity function λ(t) = λp(t) (t > 0).
Then Nt is PoissonZdistributed with
Z ∞
t 8 λn 4 o
parameter mt = λ(s)ds = λ ds = 1− .
0 0 (2 + cs)3 c (2 + ct)2
−m(t) −λ/c
(c) P(Nt = 0) = e −m(t) till time t. For ever: let t → ∞, P(Nt = 0, for all t) = lim e =e .
t→∞
STAT3603

Jeff Yao Part IV

Normal
Distribution

BM Brownian Motion and Applications

Hitting
Time &
Maximum
Variable
1 Normal Distribution
Variations
on BM

Pricing
2 Brownian Motion
Stock
Options
3 Hitting Time, Maximum Variable, and the Gambler’s Ruin Problem

4 Variations on Brownian Motion

(i) Brownian Motion with Drift

(ii) Geometric Brownian Motion

5 Pricing Stock Options∗

STAT3603 Review
Jeff Yao

Review