Poisson Process
Noha Youssef
The American University in Cairo
nayoussef@[Link]
1 / 34
Table of Contents
Counting Processes
Poisson Process
Non-Homogeneous Poisson Process
Compound Poisson Process
Conditional Arrival Times
2 / 34
Definitions and Notations
I Sn = Arrival time of the n -th event, n = 1, 2, · · ·
I T1 = S1 = Time until the 1st event occurs
I Tn = Sn − Sn−1 = time elapsed between the (n − 1)st and
n-th event, n = 2, 3, · · ·
3 / 34
Important Distributions
Poisson Distribution
A Poisson variable is the number of successes or occurrences over
an interval of time or space. The Poisson pms has the following
form
e−λ λx
P (X = x) = , x = 0, 1, 2, · · ·
x!
where λ is the rate of successes over this period of time.
The Exponential Distribution
An exponential variable is the time until the first success. It has
the following pdf
f (x) = λe−λx , x > 0,
where λ is called the rate parameter.
4 / 34
Counting Process
A stochastic process {N (t) : t ≥ 0} is said to be a counting pro-
cess if N (t) represents the total number of events (e.g. arrivals,
departures) that have occurred up to time t and must satisfy:
I N (t) ∈ N0 , the set of Natural numbers , t ≥ 0;
I N (s) ≤ N (t), 0 ≤ s < t;
I N (t) − N (s) corresponds to the number of events that have
occurred in the interval (s, t], 0 ≤ s < t.
5 / 34
Poisson Process
A continuous-time stochastic process {N (t), t ≥ 0} is a Poisson
process with rate λ > 0 if
1. N (0) = 0,
2. N (t) counts the number of events that have occurred up to
time t (i.e. it is a counting process).
3. The times between events are independent and identically
distributed with an Exponential(λ) distribution.
4. The number of events in any interval of length t is Poisson
distributed with mean λt. That is, for all s, t ≥ 0
(λt)n
Pr(N (t + s) − N (s) = n) = e−λt .
n!
6 / 34
Examples
I Let N (t) be the number of persons who enter a specific store
at or prior to time t. Then {N (t) : t ≥ 0} is a counting
process in which an event corresponds to a person entering
the store.
I Let N (t) be the number of children born by time t in a
maternity. Then {N (t) : t ≥ 0} is a counting process in which
an event occurs whenever a child is born.
7 / 34
Example 1
Let N (t) represent now the number of persons in a store at time t.
Is {N (t) : t ≥ 0} a counting process?
Solution
No, {N (t) : t ≥ 0} does not satisfy N (s) ≤ N (t), 0 ≤ s ≤ t.
8 / 34
Counting Process
Definition 1
A counting process is said to possess independent increments if the
numbers of events that occur in disjoint time intervals are
independent. Independent increment means that N (t) is
independent of N (t + s) − N (t).
Definition 2
A counting process is said to possess stationary increments if the
distribution of the number of events that occur in any interval of
time depends only on the length of the time interval.
9 / 34
Poisson Process
Proposition 1
If N (t) is a Poisson process with rate λ then the inter-arrival times
T1 , T2 , are independent and Ti ∼ Exponential(λ), for i = 1, 2, · · · .
Consequently, as the distribution of the sum of n i.i.d Exp(λ) is
Gamma(n, λ), the arrival time of the nth event is
n
X
Sn = Ti ∼ Gamma(n, λ).
i=1
10 / 34
Proposition 1: Proof
P (Tn+1 > t|T1 = t1 , T2 = t2 , . . . , Tn = tn )
=P (0 event in (sn , sn + t] |T1 = t1 , T2 = t2 , . . . , Tn = tn )
=P (0 event in (sn , sn + t]) ( by indep increment )
=P (N (sn + t) − N (sn ) = 0)
=e−λt
stationary increment, and this 1-F (.) of the Exponential
distribution
11 / 34
Example 2
The number of events in any interval of length t is Poisson dis-
tributed with mean λt where λ = 5. That is, for all s, t ≥ 0. find,
1. the probability P r[N (5) = 6|N (3) = 2]
2. the probability P [N (5) = 6, N (3) = 2]
3. the covariance Cov(N (3), N (5)).
12 / 34
Example 2: Solution
1.
Pr[N (5) = 6|N (3) = 2] = Pr[N (5), N (3) = 2]/ Pr(N (3) = 2)
= Pr(N (5) − N (3) = 4|N (3) = 2)
= Pr[N (5) − N (3) = 4] Ind. Increments
= Pr(N (2) = 4) interval length is 2
(2 × 5)4 (2 × 5)4
= e−2λ = e−2×5
4! 4!
= 0.018916637.
2.
Pr[N (5) = 6, N (3) = 2] = Pr[N (5) = 6|N (3) = 2] Pr[N (3) = 2]
= Pr[N (5) − N (3) = 4] Pr[N (3) = 2]
4 2
−2λ (2λ) −3λ (3λ)
= e e
4! 2!
3.
Cov(N (3), N (5)) = Cov(N (3), N (5) − N (3) + N (3))
= Cov(N (3), N (5) − N (3)) + Cov(N (3), N (3))
= 0 + V (N (3)) 13 / 34
Remarks
I The interarrival distribution for a Poisson process with rate λ
is Exponential(λ).
I We can replace the Exponential interarrival time distribution
with any distribution on (0, t]), to obtain a large class of
counting processes. Such processes (when the interarrival
time distribution is general) are called Renewal Processes and
the area of their study is called Renewal Theory.
I If the interarrival time is not Exponential, then the process
will not have stationary and independent increments. That is,
the Poisson process is the only Renewal process with
stationary and independent increments.
I A random variable X possesses the memoryless property if
Pr(X > 0) = 1,(i.e., X is a positive rv) and, for every x ≥ 0
and t ≥ 0, Pr{X > t + x|X > t} = Pr{X > x}.
14 / 34
Superposition
The sum of two independent Poisson processes with respective rates
λ1 and λ2 , called the superposition of the processes, is again a Pois-
son process but with rate λ1 +λ2 . Te mark: By repeated application
of the above arguments the superposition of k independent Poisson
processes with rates λ1 , · · · , λk ,λn is again a Poisson process with
rate λ1 + λ2 + · · · , λn .
15 / 34
Thinning
Consider a Poisson process with rate λ. At each arrival of events,
it is classified as a Type 1 event with probability p or Type 2 event
with probability 1 − p, independently of all other events.
Let Ni (t) = # of type i events occurred during [0, t], i = 1, 2. Note
that N (t) = N1 (t) + N2 (t).
Proof
First observe that given N (t) = n+m, (N1 (t), N2 (t)) ∼ Binomial(n+
m, p, (1 − p)). Thus
P (N1 (t) = n, N2 (t) = m)
= P (N1 (t)= n, N2 (t) = m|N (t) = n + m) P(N (t) = n + m)
n+m n+m
= pn (1 − p)m e−λt (λt)
(n+m)! .
n
n m
= e−λtp (λpt)
n! e
−λt(1−p) (λ(1−p)t)
m! Note:λ = λ1 + λ2 = pλ + (1 − p)λ
= P (N1 (t) = n) P (N2 (t) = m)
16 / 34
Cont’d
This proves the independence of N1 (t) and N2 (t) and that N1 (t) ∼
P oisson(λpt), N2 (t) ∼ P oisson(λ(1 − p)t). Both {N1 (t)} and
{N2 (t)} inherit the stationary and independent increment properties
from {N (t)}, and hence are both Poisson processes.
17 / 34
Example 3
If immigrants to area A arrive at a Poisson rate of ten per week,
1
and if each immigrant is of English descent with probability 12 ,
then what is the probability that no people of English descent will
emigrate to area A during the month of February?
Solution The number of Englishmen emigrating to area A during
the month of February is Poisson distributed with mean λ1 = λtp =
10×4 10 −10/3 .
12 = 3 , the probability required is e
18 / 34
Example 4
Customers arrive at a store at a rate of 10 per hour. Each is either
male or female with probability 0.5. Assume that you know that
exactly 10 women entered within some hour (say, 10 to 11am).
a) Compute the probability that exactly 10 men also entered.
b) Compute the probability that at least 20 customers have
entered.
Solution
Male and women arrivals are independent Poisson processes, with
parameter is (0.5)(10) = 5. It follows that
a) Pr(exactly 10 men also entered|exactly 10 women entered ) =
10
e−5 510!
b)
Pr(at least 20 customers have entered | exactly 10 women entered )
−5 λk
P∞
k=10 e k!
19 / 34
Poisson Process
Proposition 2
Assume that you have two independent Poisson processes, N1 (t)
with rate λ1 and N2 (t) with rate λ2 . The probability that n events
occur in the first process before m events occur in the second
process is
n+m−1
X k n+m−1−k
n+m−1 λ1 λ2
.
k λ1 + λ2 λ1 + λ2
k=n
20 / 34
Example 5
Assume that λ1 = 5 and λ2 = 1. Find
1. Pr (5 events in the first process before 1 in the second).
2. Pr (5 events in the first process before 2 in the second).
Solution
5 5
1. Pr(5 events in the first process before 1 in the second)= 6
2. Pr(5 events
in
the first process before
2 in the
6 5
5 1
1 6 5 6
second)= 6 6 + 6 .
5 6
21 / 34
Example 6
You have three friends, each will call you after an exponential amount
of time with expectation 30 minutes, 1 hour and 2 hours respectively.
You will go out with the first friend that calls. What is the probability
that you go out with A?
Solution
Each call is the first event in one of the three Poisson processes with
rates λ1 = 2, λ2 = 1, λ3 = 0.5. It follows that
λ1 2
Pr(A is the first to call) = = = 0.5714.
λ1 + λ2 + λ3 3.5
22 / 34
Non-Homogeneous Poisson Process
A counting process {N (t), t ≥ 0} is a nonhomogeneous Poisson
process with rate λ(t) if:
(1) N (0) = 0.
(2) N (t) has independent increments.
Rs
(3) N (t) − N (s) ∼ Poisson( t λ(u))du
Remark A non-homogeneous Poisson process is non-stationary
except when λ(t) = λ for all t. If λ(t) = λ then the process
becomes a homogeneous Poisson process. The mean value
function is defined as
Z t
m(t) = λ(s)ds.
0
23 / 34
Example 7
Consider a shop with working hours 8:00 am to 5:00 pm. It is
assumed that the number of clients is following a NHPP. We are
given From 8 to 11 am: clients arrive at an affine increasing rate
function starting at 5 per hour with maximum 20 per hour. From
11 am to 1 pm: the rate function is 20 per hour. From 1 to 5 pm:
the rate function decreases linearly by 2 to reach its minimum 12
per hour.
1. Give the expression of the mean value function.
2. What is the probability to have no clients in this shop between
8:30 and 9:30.
3. What is the expected number of clients between 1:30 and
3:30 pm.
24 / 34
Example 7: Solution
1. The rate function λ(t) is given by
5 + 5t if 0 ≤ t ≤ 3
λ(t) = 20 if 3 ≤ t ≤ 5
30 − 2t if 5 ≤ t ≤ 9
we find the mean valued function
5t + 2.5t2 if 0 ≤ t ≤ 3
m(t) = 20t − 22.55 if 3 ≤ t ≤ 5
30t − t2 − 47.5 if 5 ≤ t ≤ 9
2. The probability to have no clients in this shop between 8:30
and 9:30 is given by
P (N (1.5) − N (0.5) = 0) = e(m(1.5)−m(0.5)) = e−10
3. The expected number of clients between 1:30 and 3:30 pm is
given by m(7.5) − m(5.5) = 34.
25 / 34
Compound Poisson Process
”When employers provide health insurance to their employees, they
are of course concerned about claim frequency, the random num-
ber of claims filed. And they’re concerned about claim severity,
the random size of each claim. But they are especially concerned
about aggregate claims, the sum total of all the claims. This is
the sum of a random number of random variables, and as such is
extremely complicated to analyze; such a probability distribution is
called a compound distribution. If frequency is assumed to follow a
Poisson process and the severities are independent and all have the
same probability distribution, the result is a compound Poisson
process.”
26 / 34
Compound Poisson Process; Cont’D
Definition 3
A compound Poisson process S has the following properties:
PN (t)
(1) For t > 0,S(t) = j=1 Xj .
(2) N is a Poisson process with rate function λ.
(3) All the random variables Xj have the same distribution as a
single random variable X
(4) for all t, the random variables N (t) and all the Xj form an
independent set; and
(5) if N (t) equals 0 for a particular value of t, then the empty
sum S(t) is taken to equal [Link], in particular, S(0) = 0.
27 / 34
Compound Poisson Process; Cont’D
Example
Suppose that health claims are filed with a health insurer at the
Poisson rate λ = 20 per day, and that the independent severities X
of each claim are Exponential random variables with mean θ = 500.
Then the aggregate S of claims is a compound Poisson process.
28 / 34
Compound-Poisson-Process Expectation and Variance
The following formulas hold for computing the expected value and
variance of the compound-Poisson-process values S(t)
1. E[S(t)] = E[N (t)]E[X]; and
2. Var[S(t)] = E[N (t)] Var[X] + Var[N (t)]E[X]2 = E[N (t)]E X 2
29 / 34
Compound-Poisson-Process Expectation and Variance:
Proof
For the expectation:
N (t)
X
E(S(t)) = E(E[ X(t)|N (t)])
j=1
where
N (t)
X
E[ X(t)|N (t)] = N (t)E(X(t))
j=1
It follows that
E(S(t)) = E(N (t)E(X(t))) = E(N (t))E(X(t)).
30 / 34
Compound-Poisson-Process Expectation and Variance:
Proof
For the variance;
N (t) N (t)
X X
V (S(t)) = V [E[ X(t)|N (t)]] + E[V [ X(t)|N (t)]]
j=1 j=1
N (t)
X
V [E[ X(t)|N (t)]] = V [N (t)E(X(t))] = V (N (t))(E(X(t)))2
j=1
N (t)
X
E[V [ X(t)|N (t)]] = E(N (t)V (X(t))) = E(N (t))V (X(t))
j=1
and
V (N (t)) = E(N (t))
Then
V (S(t)) = E(N (t))(E(X))2 + E(N (t))V (X) = E(N (t))E X 2 .
31 / 34
Example 8
Suppose that health claims are filed with a health insurer at the
Poisson rate λ = 20 per day and that the independent severities Y
of each claim are Exponential random variables with mean 500. Let
X(10) be the aggregate claim during the first 10 days.
1. Find the mean and the variance of X(10)
2. Suppose that X(10) can be approximated by a Normal
random variable. Estimate Pr(X(10) > 120000).
Solution
1. The mean and the variance of X(10) are as follows
E(X(10)) = m(10) × E(Y ) = 200 × 500 = 100000 and
V (X(10)) = m(10) × E(Y 2 ) = 200 × 2 × 5002 = 400 × 5002
2. Using the normal approximation, we
120000−100000
find Pr(X(10) > 120000) u Pr(Z > √
400×5002
) = P (Z >
2) = 0.0228
32 / 34
Conditional Arrival Times
If N (t) is a rate λ Poisson process then, conditional on N (t) = n,
the arrival times S1 , · · · , Sn have the same distribution as U (1), · · · , U (n),
the order statistics for U1 , · · · , Un ∼ iidU [0, t].
If a Poisson process contains exactly n events in an interval [0, t],
then the unordered locations, or times, of those events are uniformly
distributed on the interval. Consider the distribution of the time of
the first arrival, conditional on there being one arrival by time t. For
0 ≤ s ≤ t,
P (S1 ≤ s, Nt = 1) P (Ns = 1, Nt = 1)
P (S1 ≤ s|Nt = 1) = =
P (Nt = 1) P (Nt = 1)
P (Ns = 1, Nt − Ns = 0)
=
P (Nt = 1)
e−λs λse−λ(t−s) s
= −λt
=
e λt t
which is the cumulative distribution function of the uniform distri-
bution on [0, t].
33 / 34
Conditional Arrival Times
Given that N (t) = n, the n arrival times S1 · · · , Sn have the
following conditional density function
f (s1 , . . . , sn |n) = tn!n
.
for 0 < s1 < . . . < sn < t.
No proof required.
34 / 34