Lecture 7

Poisson Process II

1 / 25
Outline

Thinning, Superposition

Compound Poisson Process

Nonhomogeneous Poisson Process

Conditional Poisson Process

2 / 25
Thinning, Superposition

3 / 25
Thinned Poisson Process

Conceptual Example

• Each baby born is either male or female with probability (p, 1 p).

• Consider the birth process (Nt )t 0 with rate . We have

Nt = Mt + Ft ,

where
Mt : #{male births} by time t,
Ft : #{female births} by time t.

• Q: What is the joint probability distribution of (Mt , Ft )?

4 / 25
Thinned Poisson Process N-m-((pm(e p) -
+

· IIIIIIIIIII
P (Mt = m, Ft = f )
= P (Mt = m, Ft = f | Nt = m + f ) P (Nt = m + f )
✓ ◆
m+f e t ( t)m+f
= p m (1 p)f ⇥
m (m + f )!
p m (1 p)f e t(p+(1 p)) ( t)m+f
=
m!f !
e pt ( pt)m e (1 p)t ( (1 p)t)f
= ⇥
m! f!

! For all t, Mt and Ft are independent Poisson random variables with

rates pt and (1 p)t respectively.

5 / 25
Thinned Poisson Process

• Let (Nt )t 0 be a Poisson process with rate .

• Assume that each arrival, independent of other arrivals, is marked as
a type k event with probability pk , for k = 1, . . . , n, where
p1 + · · · + pn = 1.

Definition
⇣ ⌘
(k) (k)
Let Nt be the number of type k events in [0, t]. Then, Nt is a
t 0
Poisson process with parameter pk , and the processes
⇣ ⌘ ⇣ ⌘
(1) (n)
Nt , . . . , Nt
t 0 t 0

are independent. Each process is called a thinned Poisson process.

6 / 25
Thinned Poisson Process

Example
Consider the male and female birth processes. Assume that the worldwide
sex ratio at birth is 108:100 boys to girls, and that births occur on at the
average rate of 2 births per hour.
a) On an 8-hour shift, what is the expectation and standard deviation of
the number of female births?
b) Find the probability that only girls were born between 2 and 5 p.m.
c) Assume that five babies were born yesterday. Find the probability that
two are boys.

7 / 25
 108

9) N + ~ pois (2t) pm ,
=
100 + 108:0 . 32

PF = 1 - 0 32
.
= 0 48.

F+ + F+ v F8 (0 96 x8)
8 -
pris
-

E(Ful = 0 96xO
.

,
SD(F8) = 0 96x8
.

b) P)Mg- Mc =
0) =
PCMs 0 =

=
Plpris (3 x 1 .
04)
=
0)

2) P(M2p = 2 1 Nex = 5)

P(Man = 2, Fra =
3) P(M2y =
2) P(Fza 3) =

= _

P(N2a =
5) P(N2a =
3)

= (2) 0 522 0 483

,
·

,
Superposition Poisson process

Related to the thinned process is the superposition process obtained by

merging independent Poisson processes.

I' but hi tren can process

but
Lay
Definition how
Rissur variables
In thi
independent vs

Superposition Process M ⇣ - >⌘ ⇣T ⌘

(1) (n)
Given n independent Poisson processes Nt , . . . , Nt with
t 0 t 0
respective parameters 1, . . . , n. Let
(1) (n)
Nt = Nt + · · · + Nt , t 0.

Then (Nt )t 0 is a Poisson process with rate = 1 + ··· + n.

no-pors
(ax+AtT (
= Xt
8 / 25
Superposition Poisson process

O A

St S2 +1 Fu

.
1 No =
Nol + No + No = 0 Ns . -
Nse = 1 Nsc" -

Ns . ")
Ns,)(
spendent
+ e < +2
+
(Ns " -

2 . S1 <S2] n

+
(Nsc) -
ise')
Nsc- Use A N + +- Nte Nte - N+= 1 Nic" -

N..
")
3 .
+s( + : N+ =
As -Pois(X(t s)) - + (N .." - N
+ (4)
+
(N + 2) -
N+ 1) 9 / 25
Superposition Poisson process

Example
A company launches three lines of insurance products, where the number
of claims for each line follows a Poisson process with respective rate 1/2,
1/3 and 2/5 per day.
a) Find the probability that there are 5 claims in the first 3 days.
b) Suppose there are 3 claims in a week. Calculate the probability that
each product line has had a claim.

10 / 25
 N += N" N +* N'"
~pris(t)
+
9) +

f y =
+ +

P(Nz = 3)

b) D(N" = 1
,
Nyl = 1
, M" = 1 (N) =
3)
"
P (N !" = 1) P(N + = 1) PINy) =
1
=

P(N =
3)
~
(2)(i) (n) pex pe + ps
Compound Poisson Process

11 / 25
Compound Poisson Process

Let (Nt )t 0 be a Poisson process with rate , and

Y1 , Y2 , . . . ⇠ O
iid
Y,
=>

which are also independent of (Nt )t 0.

Definition
The stochastic process (Xt )t 0 defined by

Xt =
Nt
X

i=1
Yi
-
GEMEINHEY
=>
NE(YY
is called a Compound Poisson Process.

12 / 25
EX) =

IE(Yi (N +
=

k)p(N +
= 4)

k)
=
[ kE(y)p)N +
=

treneis

Ely) E(N +
condition
=
↑

v(x ) +
= E(v(X (N+ ))+ + V(E(X + (N+ ))

⑦ ELY : (N += 1)

+
V(kE(y)) = E(9) ? V (N ) +

i
Rt
·
v ( [y 1 N+ = k) = [ V(y (N1 = k)

=
kr(y)
-
Elk Var(y)) = Var(y) ETN ) +

- x (X+ ) =
V(Y) EINH) + ETY)" VINA)
un
N

= N Ely4
Compound Poisson Process
Proposition
The mean and variance of a compound Poisson process are

E(Xt ) = t · E (Y ) ,
Var(Xt ) = t · E Y 2 .

Example
O
Visitors walk into a casino in Las Vegas according to a Poisson process
-
with a rate of 50 per hour. Assume that G -
10% of them will not gamble at
all, and others
O will owhich
lose independently a random number of dollars
-
we assume has a Uniform(0, $1500) distribution.

a) What is the casino’s -

expected gain during a 12-hour period?

b) Estimate the probability that the casino’s expected gain during a

month is greater than $25, 000, 000.

13 / 25
Ye id Uni 10
,
1500 ( Xi
i = 1
=
Xi + Xe +... +
X

Nt :
thinned pp-rate :
Np
=
50x0, 9 = 43 ETXi) =
45 .
1 750 .

9)
N12
N+ +
V(Xi) = 45 1
. .
...

N + + 12

Approximate by N10 , 1)
~

E(X ++ 12
-
X+ ) =
E(X 12)
↑ D)
IXExi) ND
=
12xp .
750 =...
2) P(X0 5 .
> 10,000
120

b) %
P([Xi / 25 10 .
Xow =
X1 + X2
1000
+
... +
X1000

Np2l Pois (i) P([X : >1000)

1
M10 ~ =
-
=
. 1) i= 1

Me -
No He N2 -

N1 chia who logian id

=
M2 N1-

X1 Y , X
= El Xi) = 95. 150
V(Xi) = ...
Nonhomogeneous Poisson Process

14 / 25
Nonhomogeneous Poisson Process

• In a (homogeneous) Poisson process, arrivals occur at a constant rate,

independent of time.

• However, for many applications this is an unrealistic assumption.

E.g, the rate of customers entering a supermarket between 11-12
noon is higher than between 7-8 am.

• In a nonhomogeneous Poisson process, the rate = (t) varies in

time. - la function of
1
+

! (t) is called the intensity function.

15 / 25
Nonhomogeneous Poisson Process

Definition
A counting process (Nt )t 0 is a nonhomogeneous Poisson process with
intensity (t), if
1. N0 = 0.
2. Independent increments

Nt2 Nt1 ?? Ns2 N s1 , 8 0  s 1 < s 2  t1 < t2 .

3. For all t > 0, Nt has a Poisson distribution with mean

Z t
E (Nt ) = (s)ds.
0

16 / 25
P(N8 =
200 , No -

No - 300
=
PINO =
COO) .
P(N1-No = 800)

D(pois)10x(sids) 200).
= =

p(pois()xsids) -> 800

Nonhomogeneous Poisson Process

Remark
• A nonhomogeneous Poisson process has independent increments, but
not necessarily stationary increments.

• It can be shown that for 0 < s < t,

✓Z t ◆
Nt Ns ⇠ Pois (x)dx .
s

• If (t) = is constant, we obtain the homogeneous Poisson process

with parameter .

17 / 25
Nonhomogeneous Poisson Process

Example
Throughout the day, arrivals of phone calls to a doctor’s office can be
modeled as a nonhomogeneous Poisson process with the intensity rate
8
>
<10, for 9 : 00  s  12 : 00
(s) = 5, for 12 : 00 < s  14 : 00
>
:
8, for 14 : 00 < s  17 : 00.

a) Find the probability that there will be at least 15 phone calls between
11:00 and 13:00.

b) Find the average number of phone calls per day.

18 / 25
NH) -

10

Y
' is in in

a) D (N13 -Ney (13) At

=
P( pars(n( 15 ,

=
P(pois (15) is 15)

b) E(N2a) =
E(pois (Jude)
=
El pris (69)) =
69.
Conditional Poisson Process

19 / 25
Conditional Poisson process

Definition
A counting process (Nt )t 0 is called a conditional (or mixed) Poisson
process if 8t > 0,
Nt ⇠ Pois(⇤),
where ⇤ is a random variable with distribution f⇤ (x).

! The rate ⇤ is referred to as a random intensity rate.

20 / 25
Conditional Poisson process

• The pmf for the conditional Poisson process is specified as a

conditional probability. For all s, t 0,

( t)n t
P(Ns+t Ns = n | ⇤ = ) = e , n = 0, 1, 2, . . .
n!

• Hence, the marginal distribution of Nt is given by

P(Ns+t Ns = n) = 10 P(N My n(n x)yx(x)dx

=+ s
-
= =

Z 1
( t)n
= e t · f⇤ ( )d , n = 0, 1, 2, . . .
0 n!

21 / 25
Conditional Poisson process
Remark:
• As shown, a conditional Poisson process has stationary increments.
• However, the increments are not independent. Indeed,

P(Ns = n, Nt Ns = m)
Z 1
= P(Ns = n, Nt Ns = m | ⇤ = )f( )d
Z0 1
= P(Ns = n | ⇤ = ) ⇥ P(Nt Ns = m | ⇤ = )f( )d
0
Z 1
6= P(Ns = n | ⇤ = )f( )d ⇥
0
Z 1
⇥ P(Nt Ns = m | ⇤ = )f( )d
0
= P(Ns = n) ⇥ P(Nt Ns = m).

22 / 25
Conditional Poisson process

Proposition
• Let N(t) be a conditional Poisson process with random rate ⇤.
For all t 0, the mean and variance

E(Nt ) = t E(⇤)
Var(Nt ) = t 2 Var(⇤) + tE(⇤).

• For all 0  s < t,

Cov(Ns , Nt Ns ) = s(t s) Var(⇤)

23 / 25
EIN 1 ) =
1 E(N (n x) f(x) di
+
=

= + (oxfx(x)dx
=
t . E(x)

V (N +
Conditional Poisson process

Example (⇤ discrete)
• Assume that ⇤ in a conditional Poisson process has the pmf

P (⇤ = 0) = p, P (⇤ = 1) =1 p.

• The marginal distribution of Nt is >

- binomial poisson mis

n n
( 0 t) 0t
( 1 t) 1t
P(Nt = n) = e p+ e (1 p) ,
n! n!
for all n = 0, 1, 2, . . .

! This is a mixture of two Poisson processes with rates 0 and 1

and the mixture weights p and 1 p.

24 / 25
Conditional Poisson process

• As a mixture of two Poisson processes, Nt has mean

E(Nt ) = 0 tp + 1t (1 p)
=t( 0p + 1 (1 p)) = tE(⇤),

and variance
⇣ ⌘ ⇣ ⌘
2 2
Var(Nt ) = p 0t + ( 0 t) + (1 p) 1t + ( 1 t)

( 0 tp + 1t (1 p))2
⇣ ⌘
= t2 2
0p + 2
1 (1 p) ( 0p + 1 (1 p))2
+t( 0p + 1 (1 p))
2
= t Var(⇤) + tE(⇤).

25 / 25