DISCRETE DISTRIBUTIONS

Actuarial Modeling Course

Week 10
Ahmad Fuad Zainuddin
[email protected]
Discrete Distributions
Outlines
1. The Poisson Distribution
2. The Negative Binomial Distribution
3. The Binomial Distribution
Introduction
We now formalize some of the notation used for models for discrete
phenomena. The probability function (pf) 𝑝𝑘 denotes the probability that
exactly 𝑘 events (such as claims or losses) occur. Let 𝑁 be a random variable
representing the number of such events.
Then,
𝑝𝑘 = Pr 𝑁 = 𝑘 ; 𝑘 = 0, 1, 2, …

As a reminder, the probability generating function (pgf) of a discrete random

variable 𝑁 with pf 𝑝𝑘 is
∞
𝑃 𝑧 = 𝑃𝑁 𝑧 = 𝐸 𝑧 𝑁 = ෍ 𝑝𝑘 𝑧 𝑘
𝑘=0
Exercise
The moment generating function (mgf) for discrete variables is defined
as
∞
𝑀𝑁 𝑧 = 𝐸 𝑒 𝑧𝑁 = ෍ 𝑝𝑘 𝑒 𝑧𝑘
𝑘=0

Demonstrate that 𝑃𝑁 𝑧 = 𝑀𝑁 ln 𝑧 .
𝑘
Use the fact that 𝐸 𝑁 𝑘 = 𝑀𝑁 0 to show that 𝑃′ 1 = 𝐸 𝑁 and
𝑃′′ 1 = 𝐸 𝑁 𝑁 − 1 .
The Poisson Distribution
The pf for the Poisson distribution is
𝑒 −𝜆 𝜆𝑘
𝑝𝑘 = ; 𝑘 = 0, 1, 2, …
𝑘!
The pgf is
𝑃 𝑧 = 𝑒 𝜆 𝑧−1 ; 𝜆 > 0
The mean and variance can be computed from the pgf as follows
𝐸 𝑁 = 𝑃′ 1 = 𝜆
𝐸 𝑁 𝑁 − 1 = 𝑃′′ 1 = 𝜆2
2
𝑉𝑎𝑟 𝑁 = 𝐸 𝑁 𝑁 − 1 + 𝐸 𝑁 − 𝐸 𝑁
= 𝜆2 + 𝜆 − 𝜆2 = 𝜆
The Poisson Distribution
Let 𝑁1 , 𝑁2 , … , 𝑁𝑛 be independent Poisson variables with parameters 𝜆1 , 𝜆2 , … , 𝜆𝑛 .
Then 𝑁 = 𝑁1 + 𝑁2 + ⋯ + 𝑁𝑛 has a Poisson distribution with parameter 𝜆1 + 𝜆2 +
⋯ + 𝜆𝑛 .
The pgf
𝑛 𝑛
𝑃𝑁 𝑧 = ෑ 𝑃𝑁𝑗 𝑧 = ෑ exp 𝜆𝑗 𝑧 − 1
𝑗=1 𝑗=1

𝑛
= exp ෍ 𝜆𝑗 𝑧 − 1 = 𝑒𝜆 𝑧−1
𝑗=1

Where 𝜆 = 𝜆1 + 𝜆2 + ⋯ + 𝜆𝑛
Just as is true with mgf, the pgf is unique and, therefore, 𝑁 must have a Poisson
distribution with parameter 𝜆.
The Negative Binomial Distribution
The pf of the negative binomial distribution is given by
𝑟 𝑘
𝑘+𝑟−1 1 𝛽
Pr 𝑁 = 𝑘 = 𝑝𝑘 =
𝑘 1+𝛽 1+𝛽
𝑘 = 0, 1, 2, … ; 𝑟 > 0 ; 𝛽 > 0

The binomial coefficient is to be evaluated as

𝑥 𝑥 𝑥−1 … 𝑥−𝑘+1
=
𝑘 𝑘!

While 𝑘 must be an integer, 𝑥 may be any real number. When 𝑥 > 𝑘 − 1, it can also be
written as
𝑥 Γ 𝑥+1
=
𝑘 Γ 𝑘+1 Γ 𝑥−𝑘+1
The Negative Binomial Distribution
It is not difficult to show that the pgf for the negative binomial
distribution is
𝑃 𝑧 = 1 − 𝛽 𝑧 − 1 −𝑟

From this it follows that the mean and variance of the negative
binomial distribution are
𝐸 𝑁 = 𝑟𝛽
𝑉𝑎𝑟 𝑁 = 𝑟𝛽 1 + 𝛽

Because 𝛽 is positive, the variance of the negative binomial distribution

exceeds the mean.
The Binomial Distribution
The binomial distribution is another distribution that arises naturally in
claim number modeling. It possesses some properties different from
the Poisson and the negative binomial distribution that make is
particularly useful.

Then the number of claims for a single person follows a Bernoulli

distribution, a distribution with probability 1 − 𝑞 at 0 and probability 𝑞
at 1. The pgf of the number of claims per individuals is then given by

𝑃 𝑧 = 1 − 𝑞 𝑧 0 + 𝑞𝑧1 = 1 + 𝑞 𝑧 − 1
The Binomial Distribution
Now if there are 𝑚 such independent individuals, then the pgf can be multiplied together
to give the pgf of the total number of claims arising from the group of 𝑚 individuals.
The pgf is
𝑃 𝑧 = 1+𝑞 𝑧−1 𝑚; 0<𝑞 <1

Then from this it is easy to show that the probability of exactly 𝑘 claims from the group is
𝑚 𝑘 𝑚−𝑘
𝑝𝑘 = Pr 𝑁 = 𝑘 = 𝑞 1−𝑞 ; 𝑘 = 0, 1, 2, … , 𝑚
𝑘

The pf for a binomial distribution with parameters 𝑚 and 𝑞. The mean and variance of the
binomial distribution are given by

𝐸 𝑁 = 𝑚𝑞 ; 𝑉𝑎𝑟 𝑁 = 𝑚𝑞 1 − 𝑞
Exercise
Let 𝑁|Λ have a Poisson distribution with mean Λ. Let Λ have a gamma
distribution with parameters 𝛼 and 𝜃. Determine the unconditional
probability of 𝑁.
Let 𝑁|Λ have a Poisson distribution with mean Λ. Let Λ have a gamma distribution with parameters 𝛼 and 𝜃.
The unconditional probability 𝑁 is given by
∞
𝑝𝑘 = Pr 𝑁 = 𝑘 = න Pr 𝑁 = 𝑘|Λ = 𝜆 𝑢Λ 𝜆 𝑑𝜆
0

𝜆
∞ − ∞
𝑒 −𝜆 𝜆𝑘 𝛼−1
𝜆 𝑒 𝜃 1 1 𝑘+𝛼−1 𝑒 −𝜆 1+
1
=න 𝑑𝜆 = න 𝜆 𝜃 𝑑𝜆
0 𝑘! 𝜃𝛼 Γ 𝛼 𝑘! 𝜃 𝛼 Γ 𝛼 0

1 1 𝜃 𝑘+𝛼 Γ 𝑘 + 𝛼 Γ 𝑘+𝛼 𝜃𝑘
= =
𝑘! 𝜃 𝛼 Γ 𝛼 1 + 𝜃 𝑘+𝛼 𝑘! Γ 𝛼 1+𝜃 𝑘+𝛼

𝛼 𝑘
𝑘+𝛼−1 1 𝜃
= ; 𝑘 = 0, 1, 2, … ; 𝛼 > 0 ; 𝜃 > 0
𝑘 1+𝜃 1+𝜃

The unconditional probability 𝑁 follows the negative binomial with parameters r = 𝛼 and 𝛽 = 𝜃.