Exam 4/C

Howard Mahler

Frequency Distributions
The slides are in the same order as the sections of my study guide.

Section #Pages
Section Name
A 1 Introduction
2 5-15
Basic Concepts
3 16-41
Binomial Distribution
4 42-72
Poisson Distribution
B 5 73-94
Geometric Distribution
6 95-120
Negative Binomial Distribution
7 121-148
Normal Approximation
C 8 149-161
Skewness
9 162-175
Probability Generating Functions
10 176-188
Factorial Moments
11 189-210
(a, b, 0) Class of Distributions
12 211-222
Accident Profiles
D 13 223-242
Zero-Truncated Distributions
14 243-261
Zero-Modified Distributions
15 262-276
Compound Frequency Distributions
16 277-294
Moments of Compound Distributions
E 17 295-334
Mixed Frequency Distributions
18 335-345
Gamma Function
F 19 346-387
Gamma-Poisson Frequency Process
20 388-398
Tails of Frequency Distributions
21 399-406
Important Formulas and Ideas

At the end, there are some additional questions for study.

After the slides are my section of important ideas and formulas.
 Basic Concepts
A Frequency Distribution is discrete.
It has probability on the nonnegative integers.
The probabilities add to one.

Probability Probability Probability

Number Density times times Square
of Claims Function # of Claims of # of Claims
0 0.1 0 0
1 0.2 0.2 0.2
2 0 0 0
3 0.1 0.3 0.9
4 0 0 0
5 0 0 0
6 0.1 0.6 3.6
7 0 0 0
8 0 0 0
9 0.1 0.9 8.1
10 0.3 3 30
11 0.1 1.1 12.1
Sum 1 6.1 54.9
Average of X = 1st moment about the origin =
6.1 = Mean.
Average of X2 = 2nd moment about the origin = 54.9.
Variance = 54.9 - 6.12 = 17.7.
Questions 2.9 and 2.10 contrast
summing the rolls of two six-sided dice,
and multiplying a single roll by two.
Binomial Distribution:

Support: x = 0,1,2,3,...,m. Finite support.

Parameters: 1 > q > 0, m ≥ 1, integer.

m!
f(x) = qx (1-q)m-x, 0 ≤ x ≤ m.
x! (m-x)!

Mean = mq Variance = mq(1-q)

For example, for m = 5 and q = 0.1:

f(2) = 5! 0.12 0.95-2 = (10) 0.12 0.93 = 0.0729.
2! 3!

f(x) ⇔ pk.
Binomial Distribution for m = 1 is
a Bernoulli Distribution.
Bernoulli Distribution: f(1) = q, f(0) = 1 - q.

The sum of m independent, identically

distributed Bernoulli Distributions is a
Binomial Distribution.

The sum of two independent Binomials with the

same q is another Binomial:

Binom(m1, q) + Binom(m2, q) = Binom(m1 + m2, q).

TI-30XS Multiview calculator saves time doing
repeated calculations using the same formula.
For example constructing a table of the densities
of a Binomial distribution, with m = 5 and q = 0.1:

⎛ 5⎞
f(x) = ⎜ ⎟ 0.1x 0.95-x.
⎝ x⎠

table
y = (5 nCr x) * .1x * .9(5-x)
Enter
Start = 0
Step = 1
Auto
OK

x=0 y = 0.59049
x=1 y = 0.32805
x=2 y = 0.07290
x=3 y = 0.00810
x=4 y = 0.00045
x=5 y = 0.00001
Poisson Distribution:

λx e- λ
f(x) = , x ≥ 0 (infinite support)
x!

λ4 e − λ 2.54 e − 2.5
If λ = 2.5, f(4) = = = 13.36%.
4! 4!

0.25

0.2

0.15

0.1

0.05

0 2 4 6 8 10
Mean = λ Variance = λ Mean = Variance.
A Poisson Distribution ⇔
a constant independent claim intensity.
The sum of two independent variables each
of which is Poisson with parameters λ1 and λ2
is also Poisson, with parameter λ1 + λ2.
Frequency Poisson and severity independent
of frequency ⇒ the number of claims above a
certain amount (in constant dollars) is also a
Poisson.
This is called “thinning” a Poisson.

The small claims are also Poisson.

The number of small and large claims are
independent of each other.
4.6. The male drivers in the State of Grace each
have their annual claim frequency given by a
Poisson distribution with parameter equal to 0.05.
The female drivers in the State of Grace each
have their annual claim frequency given by a
Poisson distribution with parameter equal to 0.03.
You insure in the State of Grace 20 male drivers
and 10 female drivers. Assume the claim
frequency distributions of the individual drivers are
independent. What is the chance of observing 3
claims in a year?
4.6. E. The sum of the claims from 20 males is
Poisson with mean: (20)(0.05) = 1.0.
The sum of the claims from 10 females is Poisson
with mean: (10)(0.03) = 0.3.
The sum of the claims from everyone is Poisson
with mean: 1.0 + 0.3 = 1.3.
The chance of 3 claims is: (1.33) e-1.3 / 3! = 9.98%.
Next Q. 4.12.
4.12. Frequency follows a Poisson Distribution
with λ = 7.
20% of losses are of size greater than $50,000.
Frequency and severity are independent.
Let N be the number of losses of size greater than
$50,000. What is the probability that N = 3?
4.12. D. Number of large losses follows a Poisson
Distribution with λ = (20%)(7) = 1.4.

1.43 e − 1.4
f(3) = = 11.3%.
3!

Comment: Number of small losses follows a

Poisson Distribution with λ = (80%)(7) = 5.6.
Number of small losses is independent of the
number of large losses!
 Geometric Distribution
βx
f(x) = , x = 0, 1, 2, 3, ...
(1+β) x+1

Mean = β. Variance = β (1+β). Mode = 0.

f(0) = 1 .
1+β

f(n+1) = f(n) β < f(n).

1+β
f(n+1) = f(n) β < f(n).
1+β

A Geometric Distribution with β = 4:

Density
0.20

0.15

0.10

0.05

n
0 5 10 15 20

Geometric Distribution ⇔ Exponential Distribution.

5.33. 1, 5/00, Q.36. In modeling the number of
claims filed by an individual under an automobile
policy during a three-year period, an actuary
makes the simplifying assumption that for all
integers n ≥ 0, pn+1 = pn / 5, where pn represents
the probability that the policyholder files n claims
during the period. Under this assumption, what is
the probability that a policyholder files more than
one claim during the period?
1, 5/00, Q.36. A.
The densities are declining geometrically.

Therefore, this is a Geometric Distribution, with

β = 1/5. ⇒ β = 1/4.
1+β

Prob[more than one claim] = 1 - f(0) - f(1) =

β
1- 1 - = 1 - 4/5 - 4/25 = 4%.
1 + β (1 + β)2
Negative Binomial Distribution:

βx
f(x) = r (r+1) ... (r+x−1)
x! (1 + β )r+x
r+x−1⎞⎟
⎛
⎜ βx
= ⎜ ⎟ .
x ⎠ (1 + β )r+x
⎜⎜
⎝
⎟⎟

1 .
f(0) =
(1 + β)r
rβ
f(1) = .
(1 + β)r+1
β2
f(2) = r (r+1) .
2 (1 + β)r+ 2

β3
f(3) = r (r+1) (r+2)
6 (1 + β)r+ 3
Negative Binomial Distribution

Mean = rβ Variance = r β (1+β) > Mean.

Negative Binomial Distrib. with r = 2 and β = 4,
(2)(3)...(x+1) 4x
f(x) = = (x+1) (0.04) (0.8x):
x! 5x+2

Density
0.08

0.06

0.04

0.02

n
0 5 10 15 20 25 30
Negative Binomial with r = 1 is a Geometric.

For the Negative Binomial Distribution with

parameters β and r, with r integer ⇔
the sum of r independent Geometric
distributions each with mean β.

Neg Bin. ⇔ Gamma Geometric ⇔ Exponential

The sum of independent Negative Binomial

Distributions with the same β, is Negative
Binomial with the sum of the r parameters.
6.33. CAS3, 11/06, Q.23
An actuary has determined that the number of
claims follows a negative binomial distribution with
mean 3 and variance 12.
Calculate the probability that the number of claims
is at least 3 but less than 6.
CAS3, 11/06, Q.23. B.
r β = 3. r β (1+β) = 12.

⇒ 1 + β = 12/3 = 4. ⇒ β = 3. ⇒ r = 1.

33 34 35
f(3) + f(4) + f(5) = 4 + 5 + 6 = 0.244.
4 4 4

Comment: We have fit via Method of Moments.

Since r = 1, this is a Geometric Distribution.
 Normal Distribution

Standard Normal, with mean zero and standard

deviation one, which is in the table attached to the
exam: Distribution Function Φ[x]

As shown at the very beginning of the Tables:

exp[-x2 / 2]
Density function φ[x] = , -∞ < x < ∞.
2π

F(x) = Φ[ x - µ ]
σ
(x - µ)2
exp[- ]
2
f(x) = φ[ x - µ ] / σ =
2σ
.
σ σ 2π
Mean = µ Variance = σ2

Skewness = 0 (distribution is symmetric)

Standard Normal Dist, 95% Confidence Interval

2.5% 2.5%

- 1.96 1.96
 NORMAL DISTRIBUTION TABLE
(Standard Normal with µ = 0 and σ = 1)
x 0 1 2 3
0.0 0.5000 0.8413 0.9772 0.9987
0.1 0.5398 0.8643 0.9821 0.9990
0.2 0.5793 0.8849 0.9861 0.9993
0.3 0.6179 0.9032 0.9893 0.9995
0.4 0.6554 0.9192 0.9918 0.9997
0.5 0.6915 0.9332 0.9938 0.9998
0.6 0.7257 0.9452 0.9953 0.9998
0.7 0.7580 0.9554 0.9965 0.9999
0.8 0.7881 0.9641 0.9974 0.9999
0.9 0.8159 0.9713 0.9981 1.0000
Φ(z) = Pr(Z < z) z

0.800 0.842
0.850 1.036
0.900 1.282
0.950 1.645
0.975 1.960
0.990 2.326
0.995 2.576
When using the normal distribution, choose
the nearest z-value to find the probability, or
if the probability is given, chose the nearest
z-value. No interpolation should be used.
Page 123.
A Binomial Distribution with m = 10 and q = 0.3,
has mean: (10) (0.3) = 3,
and variance: (10) (0.3) (0.7) = 2.1.

A graph of this Binomial and the Normal

Distribution with the same mean and variance:

Prob.

0.25

0.2

0.15

0.1

0.05

n
2 4 6 8 10
A Binomial Distribution with m = 10 and q = 0.3.
The exact probability of one or two claims is
f(1) + f(2), the sum of the area of two rectangles:
Prob.

0.25

0.2

0.15

0.1

0.05

n
0.5 1 1.5 2 2.5 3

The rectangles go from 1 - 1/2 = 0.5

to 2 + 1/2 = 2.5.
The Normal Approximation is the area under the
Normal Distribution with the same mean and
variance, from 0.5 to 2.5.

This continuity correction is used with the Normal

Approximation for discrete variables.
More than 15 claims ⇔ At least 16 claims
15 15.5 16

|→
15.5 - µ
Prob[More than 15 claims] ≅ 1 - Φ[ ].
σ

For a frequency distribution with mean 14 and

standard deviation 2:
Prob[At least 16 claims] = Prob[ > 15 claims] ≅
15.5 - µ
1 - Φ[ ] = 1 - Φ[15.5 - 14 ] =
σ 2
1 - Φ[0.75] = 1 - 0.7734 = 22.66%.
Page 125.
µ = mean of frequency distribution
σ = standard deviation of frequency distrib.
Prob[i ≤ # claims ≤ j] ≅

Φ[ (j + 0.5) - µ ] - Φ[ (i - 0.5) - µ ].
σ σ

Less than 12 claims ⇔ At most 11 claims

⇔ 11 claims or less.

11 11.5 12
←|
11.5 - µ
Prob[Less than 12 claims] ≅ Φ[ ].
σ
7.13. You are given the following:
• Sue takes an actuarial exam with 40 multiple
choice questions, each of equal value.
• Sue knows the answers to 13 questions and
answers them correctly.
• Sue guesses at random on the remaining 27
questions, with a 1/5 chance of getting each
such question correct, with each question
independent of the others.
If 22 correct answers are needed to pass the
exam, what is the probability that Sue passed her
exam? Use the Normal Approximation.
A. 4% B. 5% C. 6% D. 7% E. 8%
7.13. D. The number of correct guesses is
Binomial with parameters: m = 27 and q = 1/5,
with mean: (1/5)(27) = 5.4
and variance: (1/5) (4/5) (27) = 4.32.
Therefore, Prob(# correct guesses ≥ 9) ≅

1 - Φ[ 8.5 - 5.4 ] = 1 - Φ[1.49] = 6.8%.

4.32

Here is a graph of Sueʼs chance of passing,

given that she knows the answers to a certain number
of questions and guesses on the rest:

Pass%
1

0.8

0.6

0.4

0.2

# Sue Knows
10 12 14 16 18 20 22
 E[ (X - E[X])3 ]
Skewness =
STDDEV3
Third Central Moment
= .
Variance1.5
Positive skewness: Poisson & Negative Binomial.

As q → 0, Binomial → Poisson.

Binomial with q < 1/2, positive skewness.

Density
0.4
m = 6, q = 0.2
0.3

0.2

0.1

n
0 1 2 3 4 5 6
Binomial with q > 1/2, negative skewness.

Density
m = 6, q = 0.75
0.35
0.30
0.25
0.20
0.15
0.10
0.05
n
0 1 2 3 4 5 6
Binomial with q = 1/2 is symmetric: 0 skewness.

m = 6, q= 0.5
Density
0.30
0.25
0.20
0.15
0.10
0.05
n
0 1 2 3 4 5 6
Probability Generating Function, p.g.f.:
∞

P(z) = Expected Value of zn ≡ Σ f(n) zn.

n=0

Poisson: P(z) = Exp[λ (z - 1)].

p.g.f. of the sum of independent frequencies is the

product of the individual p.g.f.s.

The distribution determines the probability

generating function and vice versa.
⎛
d n P(z)⎞⎟
⎜
f(n) = ⎜
n ⎟ / n!.
⎜ ⎟
⎝ dz ⎠ z=0

f(0) = P(0).

f(1) = Pʼ(0).

f(2) = Pʼʼ(0) / 2.
 Factorial Moments

nth factorial moment = µ(n)

= E[X (X-1) ... (X + 1 - n)].

µ(2) = E[X(X-1)] = E[X2] - E[X].

Poisson: µ(n) = λn.

⎛
d n P(z)⎞⎟
µ(n) =⎜
⎜
n ⎟ .
⎜ ⎟
⎝ dz ⎠ z=1

E[X] = Pʼ(1).
 (a, b, 0) Class of Distributions:

Distribution Mean Variance

Binomial, Var. < Mean mq mq(1 - q)

Poisson, Var. = Mean λ λ

Negative Binomial rβ rβ(1 + β)

Variance > Mean
 f(x+1) = a + b , x = 0, 1, 2, ...
f(x) x+1

f(x+1) = {a + b } f(x), x = 0, 1, 2, ...

x+1

f(1) = (a + b) f(0). f(2) = (a + b/2) f(1).

f(3) = (a + b/3) f(2). f(4) = (a + b/4) f(3).
pk b
= a + , k = 1, 2, 3 ...
pk -1 k

a and b are in the Tables attached to the exam:

Distribution a b
Binomial -q / (1 - q) (m + 1) q / (1 - q)
Poisson 0 λ

Negative Binomial β / (1 + β) (r - 1) β / (1 + β)
11.19. 3, 11/02, Q.28 & 2009 Sample Q.94
X is a discrete random variable with a probability
function which is a member of the (a,b,0) class of
distributions.
You are given:
(i) P(X = 0) = P(X = 1) = 0.25
(ii) P(X = 2) = 0.1875
Calculate P(X = 3).
(A) 0.120 (B) 0.125 (C) 0.130 (D) 0.135 (E) 0.140
3, 11/02, Q.28. B. For a member of the (a,b,0)
class of distributions, f(x+1) / f(x) = a + {b / (x+1)}.
f(1) / f(0) = a + b. ⇒ 0.25 / 0.25 = 1 = a + b.
f(2) / f(1) = a + b/2. ⇒ 0.1875 / 0.25 = 0.75 = a + b/2.
Therefore, a = 0.5 and b = 0.5.
f(3) = f(2) (a + b/3) = (0.1875) (0.5 + 0.5/3) = 0.125.

Alternately, once one solves for a and b,

a > 0 ⇒ a Negative Binomial Distribution.
1/2 = a = β / (1 + β). ⇒ β = 1.
1/2 = b = (r - 1) β / (1 + β). ⇒ r - 1 = 1. ⇒ r = 2.
r(r + 1)(r + 2) β3 (2)(3)(4) 13
f(3) = = = 1/8.
3! (1+β)r+3 6 25
Accident Profiles:

For a member of the (a, b, 0) class:

f(x+1) = a + b
f(x) x+1
⇔ (x+1) f(x+1) / f(x) = (x+1) a + b = a x + a + b.

It is a straight line with slope a, and intercept a + b.

Thus graphing (x+1) f(x+1) can be a useful

f(x)
method of determining whether one of these three
distributions fits the given data.
Thus graphing (x+1) f(x+1) can be a useful
f(x)
method of determining whether one of these three
distributions fits the given data.

If a straight line does seem to fit this “accident

profile”, then one should use a member of the
(a, b, 0) class.

The slope determines which of the three

distributions is likely to fit:
if the slope is close to zero then a Poisson,
if significantly negative then a Binomial, and
if significantly positive then a Negative Binomial.
12.5. You are given the following distribution of
the number of claims per policy during a one-year
period for 20,000 policies.
Number of claims per policy Number of Policies
0 6503
1 8199
2 4094
3 1073
4 128
5 3
6+ 0
Which of the following distributions would be the
most appropriate model for this data?
12.5. Calculate (x+1) f(x+1) =
f(x)
(x+1) (number of policies with x+1 claims) .
(number of policies with x claims)

Number of (x+1) f(x+1) / f(x) Differences

claims Observed
0 6,503 1.261
1 8,199 0.999 -0.262
2 4,094 0.786 -0.212
3 1,073 0.477 -0.309
4 128 0.117 -0.360
5 3

Since (x+1) f(x+1) is approximately linear,

f(x)
we probably have a member of the (a, b, 0) class.
a = slope < 0. ⇒ Binomial Distribution.
A graph of
(x+1) (number of policies with x+1 claims) :
(number of policies with x claims)

1.2
1.0
0.8
0.6
0.4
0.2
x
0 1 2 3 4
Zero-Truncated Distributions

If f is a distribution on 0, 1, 2, 3,..., then

g(x) = f(x) is a distribution on 1, 2, 3, ...
1 - f(0)

Loss Models uses the notation pT for the densities

k
of a zero-truncated distribution.

The moments of a zero-truncated distribution, g,

are given in terms of those of the corresponding
untruncated distribution, f, by:
n]
E [X
Eg[Xn] = f .
1 - f(0)
13.4. The number of vehicles involved in an
automobile accident is given by a
Zero-Truncated Binomial Distribution with
parameters q = 0.3 and m = 5.

What is the chance of observing exactly 3

vehicles involved in an accident?
13.4. D. For a non-truncated Binomial,
f(3) = 5! 0.33 0.72 = 0.1323.
3! 2!
For the zero-truncated distribution one gets the
density by dividing by 1 - f(0):
0.1323 / (1 - 0.75) = 15.9%.
Page 225

The Logarithmic Distribution with parameter β has

support equal to the positive integers:

β x 1
f(x) = for x = 1, 2, 3,...
(1 + β)x x ln(1 + β)
For the (a, b, 1) class of frequency
distributions:
f(x+1) = a + b , for x ≥ 1.
f(x) x+1

For the (a, b, 0) class of frequency distributions

we required that the same relationship hold, but
starting instead at x = 0.

The (a, b, 1) class consists of:

the (a, b, 0) class,
the zero-truncated distributions,
the logarithmic distribution,
and the zero-modified distributions.
For a Zero-Modified distribution, an arbitrary
amount of probability has been placed at zero.
The remaining probability has been spread
on x = 1, 2, 3,
proportional to some density f(x) on x = 0, 1, 2, 3,

g(0) = pM,
0
1 - pM
g(x) = f(x) 0 , x = 1, 2, 3,...
1 - f(0)
is a distribution on 0, 1, 2, 3, ....

Loss Models uses the notation pM for the

k
densities of a zero-modified distribution.

As shown in Appendix B:

pM = (1 - pM) pT = (1 - pM) pk / (1 - p0).

k 0 k 0
The moments of a zero-modified distribution g
are given in terms of those of the corresponding
unmodified distribution, f, by:

1 - pM
Eg[Xn] = Ef[Xn] 0 .
1 - f(0)

The probability generating function of the

zero-modified distribution g
are given in terms of those of the corresponding
unmodified distribution, f, by:

Pg[z] = pM + (1 - pM) Pf[z].

0 0
14.20. The number of claims per year is given by
a Zero-Modified Poisson Distribution with
parameter λ = 2.5,
and with 30% chance of zero claims.
What is the chance of observing 2 claims over the
coming year?
14.20. B. For an unmodified Poisson with λ = 2.5:
f(0) = e-2.5.
f(2) = (2.52) e-2.5 / 2! = 0.2565.

For the zero-modified distribution one gets the

densities for values other than zero by multiplying
by (1 - 0.3) and dividing by 1 - e-2.5:

(0.2565) (1 - 30%) / (1 - e-2.5) = 19.56%.

 Compound Frequency Distributions

Compound distributions are mathematically

equivalent to aggregate distributions.

The formula for their variance will be covered with

aggregate distributions.
 Mixed Frequency Distributions

Each insured has a Poisson frequency, but there

are four types:

Type Lambda Probability

Excellent 1 40%
Good 2 30%
Bad 3 20%
Ugly 4 10%

For example, for an Ugly insured the chance of 6

46 e -4
claims is: = 10.4%.
6!

For an insured picked at random of unknown type,

the probability of 6 claims is:
(.4)(.05%) + (.3)(1.20%) + (.2)(5.04%) + (.1)(10.42%)
= 2.43%.

The density function of the mixture, is the

mixture of the density functions.
Type Lambda Probability
Excellent 1 40%
Good 2 30%
Bad 3 20%
Ugly 4 10%

Number of Prob. Prob. Prob. Prob. Prob.

Claims Excellent Good Bad Ugly All
0 0.3679 0.1353 0.0498 0.0183 0.1995
1 0.3679 0.2707 0.1494 0.0733 0.2656
2 0.1839 0.2707 0.2240 0.1465 0.2142
3 0.0613 0.1804 0.2240 0.1954 0.1430
4 0.0153 0.0902 0.1680 0.1954 0.0863
5 0.0031 0.0361 0.1008 0.1563 0.0478
6 0.0005 0.0120 0.0504 0.1042 0.0243
7 0.0001 0.0034 0.0216 0.0595 0.0113
8 0.0000 0.0009 0.0081 0.0298 0.0049
9 0.0000 0.0002 0.0027 0.0132 0.0019
10
11
12
13
14 0.0000 0.0000 0.0002
0.0001
0.0000
0.0008 0.0019
0.0006
0.0002
0.0001
0.0053 0.0007
0.0002
0.0001
0.0000
SUM 1.0000 1.0000 1.0000 1.0000 1.0000

For 6 claims: (40%)(.05%) + (30%)(1.20%)

+ (20%)(5.04%) + (10%)(10.42%) = 2.43%.

This is an example of a discrete mixture.

There are also continuous mixtures.
17.14. Each insuredʼs claim frequency follows a
Negative Binomial Distribution, with r = 0.8.
There are two types of insureds as follows:
Type A Priori Probability β
A 70% 0.2
B 30% 0.5
What is the chance of an insured picked at random
having 1 claim next year?
A. 13% B. 14% C. 15% D. 16% E. 17%
 rβ
17.14. B. f(1) = .
(1 + β)r+1

For Type A: f(1) = (0.8) (0.2) = 11.52%.

1.21.8

For Type B: f(1) = (0.8) (0.5) = 19.28%.

1.51.8

(70%) (11.52%) + (30%) (19.28%) = 13.85%.

17.21. For a given value of q, the number of
claims is Binomial distributed with parameters
m = 3 and q.
In turn q is distributed uniformly from 0 to 0.4.
What is the chance that zero claims are observed?
17.21. D. Given q, we have a Binomial with
parameters m = 3 and q.
The chance that we observe 0 claims is: (1 - q)3.

π(q) = 1 / (0.4 - 0) = 2.5, for 0 ≤ q ≤ 0.4.

0. 4 0. 4
f(0) = ∫ f(0 | q) π(q) dq = ∫ (1 - q)3 (2.5) dq
0 0

q = 0.4

= (-2.5/4) (1 - q)4 ] = (-0.625) (0.64 - 14) = 0.544.

q=0
Complete Gamma Function:
∞ ∞
Γ(α) = ∫ tα-1 e-t dt = θ−α ∫ tα-1 e-t/θ dt .
0 0

For α integer, Γ(α) = (α −1) ! Γ(α) = (α−1) Γ(α−1).

Γ(1) = 1. Γ(2) = 1. Γ(3) = 2. Γ(4) = 6. Γ(5) = 24.

Incomplete Gamma Function:

x
Γ(α ; x) = ∫ tα-1 e-t dt / Γ(α).
0

Γ(α ; 0) = 0. Γ(α ; ∞) = 1.

Gamma Distribution: F(x) = Γ(α ; x / θ).

Gamma Distribution

F(x) = Γ(α; x / θ). Two parameters α and θ.

x α-1 e -x / θ
f(x) = , x > 0.
Γ(α) θ α

E[X] = α θ.

E[X2] = α (α + 1) θ2.

Var[X] = α θ2.

For the important special case α = 1,

we have an Exponential distribution:
f(x) = e-x/θ / θ, x ≥ 0.
The density of a Gamma Distrib. integrates to 1.

x α-1 e -x/θ
f(x) = , x > 0.
Γ(α) θ α

∞
∫ tα-1 e-t/θ dt = Γ(α) θα.
0

Can rewrite this with n = α - 1, and c = 1/θ:

∞
n e-c t dt = n! / cn+1.
∫ t
0
Gamma-Poisson

The number of claims a particular policyholder

makes in a year is Poisson with mean λ.

The λ values of the portfolio of policyholders

are Gamma distributed.

Lambda varies via a continuous distribution; this is

a very important example of a continuous mixture.
Prior Distribution

The λ values of the portfolio of policyholders are

Gamma distributed with α = 3 and θ = 2/3:

f(λ) = 1.6875 λ2 e−1.5λ, λ > 0.

density
0.4

0.3

0.2

0.1

lambda
1 2 3 4 5 6
Mixed Distribution

The mixed distribution is a Negative Binomial

Distribution with r = α and β = θ.

Beta rhymes with theta.

In this case, r = α = 3, and β = θ = 2/3.

In this case, r = α = 3, and β = θ = 2/3.

The chance of a policyholder chosen at random

having 6 claims is:
(3) (4) (5) (6) (7) (8) (2/3)6 = 2.477%.
6! (5/3)6+3
Density
0.25

0.20

0.15

0.10

0.05

n
0 2 4 6 8 10

Neg. Binomial is a distribution of number of claims,

while the Gamma is a distribution of parameters.
Overall Mean

Mean = E[λ] = mean of the Gamma = α θ = (3) (2/3)

=2
= (3) (2/3) = r β = mean of the Negative Binomial.
Exponential-Poisson

For the important special case α = 1, we have an

Exponential distribution of λ: f(λ) = e−λ/θ / θ, λ ≥ 0.

(See 3/11/01, Q.27.)

The mixed distribution is a Negative Binomial

Distribution with r = 1 and β = θ;

For the Exponential-Poisson,

the mixed distribution is a Geometric
Distribution with β = θ.

We will revisit the Gamma-Poisson when we

discuss Conjugate Priors.
19.61. CAS3, 5/05, Q.10. Low Risk Insurance
Company provides liability coverage to a
population of 1,000 private passenger automobile
drivers. The number of claims during a given year
from this population is Poisson distributed.
If a driver is selected at random from this
population, his expected number of claims per year
is a random variable with a Gamma distribution
such that α = 2 and θ = 1.
Calculate the probability that a driver selected at
random will not have a claim during the year.
A. 11.1% B. 13.5% C. 25.0% D. 33.3% E. 50.0%
CAS3, 5/05, Q.10. C. Gamma-Poisson.

The mixed distribution is Negative Binomial with

r = α = 2 and β = θ = 1.

1 1
f(0) = = = 1/4.
(1 + β)r (1 + 1)2

Density
0.25

0.20

0.15

0.10

0.05

n
0 2 4 6 8 10
CAS3, 5/05, Q.10. Low Risk Insurance Company
provides liability coverage to a population of 1,000
private passenger automobile drivers.
The number of claims during a given year from this
population is Poisson distributed.
If a driver is selected at random from this
population, his expected number of claims per year
is a random variable with a Gamma distribution
such that α = 2 and θ = 1.

19.62. In CAS3, 5/05, Q.10, what is the probability

that at most 265 of these 1000 drivers will not
have a claim during the year?
19.62. E. For each driver, Prob[no claim] = 1/4.
For 1000 independent drivers, the number of
drivers with no claims is Binomial with m = 1000
and q = 1/4, with mean m q = 250,
and variance m q (1 - q) = 187.5.
Prob[At most 265 claim-free drivers] ≅
Φ[ 265.5 - 250 ] = Φ[1.13] = 87.08%.
187.5
The Normal with mean 250 and variance 187.5:
density

0.025

0.020

0.015

0.010 87%
0.005

drivers
200 220 240 265.5 300
CAS3, 5/05, Q.10. Low Risk Insurance Company
provides liability coverage to a population of 1,000
private passenger automobile drivers.
The number of claims during a given year from this
population is Poisson distributed.
If a driver is selected at random from this
population, his expected number of claims per year
is a random variable with a Gamma distribution
such that α = 2 and θ = 1.

19.63. In CAS3, 5/05, Q.10, what is the probability

that these 1000 drivers will have a total of more
than 2020 claims during the year?
19.63. D.
The distribution of number of claims from a single
driver is Neg. Binomial with r = 2 and β = 1.
For the sum of 1000 independent drivers:
Neg. Binomial with r = (1000) (2) = 2000 and β = 1,
with mean rβ = 2000, variance r β (1 + β) = 4000.
Prob[more than 2020 claims] ≅

1 - Φ[ 2020.5 - 2000 ] = 1 - Φ[0.32] = 37.45%.

4000

Alternately, the mean of the sum of 1000

independent drivers is 1000 times the mean of
single driver: (1000) (2) = 2000.
The variance of the sum of 1000 independent
drivers is 1000 times the variance of single driver:
(1000) (2) (1) (1+1) = 4000.
Proceed as before.
The Normal Distribution with mean 2000 and
variance 4000, approximating the total number of
claims from all 1000 drivers:

density
0.006

0.005

0.004

0.003

0.002

0.001 37%
claims
1800 1900 2020.5 2100 2200
 Tails of Frequency Distributions

Unlikely to be asked about on your exam.

Additional Questions
• For each individual driver, the number of
accidents in a year follows a Poisson Distribution.
• For each individual driver, the mean of their
Poisson Distribution λ is the same each year.
• For each individual driver, the number of
accidents each year is independent of other years.
• The number of accidents for different drivers are
independent.
• λ varies between drivers via a Gamma
Distribution with mean 0.08 and variance 0.0032.
• Moe, Larry, and Curly are each drivers.

19.32. What is the probability that Moe has

exactly one accident next year?
A. 6.9% B. 7.1% C. 7.3% D. 7.5% E. 7.7%
19.32. B. For the Gamma, mean = α θ = 0.08,

and variance = α θ2 = 0.0032.

Thus θ = 0.04 and α = 2.

This is a Gamma-Poisson, with mixed distribution

a Negative Binomial:
with r = α = 2 and β = θ = 0.04.
rβ (2)(0.04)
f(1) = = = 7.11%.
(1 + β)r + 1 (1 + 0.04 )3

Comment: The fact that it is the next year rather

than some other year is irrelevant.
• For each individual driver, the number of
accidents in a year follows a Poisson Distribution.
• For each individual driver, the mean of their
Poisson Distribution λ is the same each year.
• For each individual driver, the number of
accidents each year is independent of other years.
• The number of accidents for different drivers are
independent.
• λ varies between drivers via a Gamma
Distribution with mean 0.08 and variance 0.0032.
• Moe, Larry, and Curly are each drivers.

19.33. What is the probability that Larry has

exactly 2 accidents over the next 3 years?
A. 2.25% B. 2.50% C. 2.75% D. 3.00% E. 3.25%
19.33. C. For one year, each insureds mean is λ,
and is distributed via a Gamma with:
θ = 0.04 and α = 2.

Over three years, each insureds mean is 3λ,

and is distributed via a Gamma with:
θ = (3)(0.04) = 0.12, and α = 2.

This is a Gamma-Poisson, with mixed distribution

a Negative Binomial:
with r = α = 2 and β = θ = 0.12.
2
r (r+1) β
f(2) =
2 (1 + β)r + 2

(2) (3) 0.122

= = 2.75%.
2 (1 + 0.12) 4
Comment: Assume a Gamma-Poisson model of
insured drivers.
Lois has a low expected annual claim frequency,
and Hi has a very high expected annual claim
frequency.

Drivers such as Lois with a low λ in one year are

assumed to have the same low λ every year.
Such good drivers have a small chance of having
a large number of claims over several years.

Drivers such as Hi with a very high λ in one year

are assumed to have the same high λ every year.
Such very bad drivers have a significant chance of
having a large number of claims over several
years.
Observe a Gamma-Poisson process for Y years,
and each insuredʼs Poisson parameter does not
change over time.

λ ~ Gamma(α, θ).

Over Y years an insured is Poisson with mean Yλ.

Yλ ~ Gamma(α, Yθ), as per inflation.

If one has a Poisson Distribution mixed by a

Gamma Distribution with parameters α and θ,
then over a period of length Y,
the mixed distribution is Negative Binomial
with r = α and β = Y θ.
• For each individual driver, the number of
accidents in a year follows a Poisson Distribution.
• For each individual driver, the mean of their
Poisson Distribution λ is the same each year.
• For each individual driver, the number of
accidents each year is independent of other years.
• The number of accidents for different drivers are
independent.
• λ varies between drivers via a Gamma
Distribution with mean 0.08 and variance 0.0032.
• Moe, Larry, and Curly are each drivers.

19.34. What is the probability that Moe, Larry, and

Curly have a total of exactly 2 accidents during
the next year?
A. 2.25% B. 2.50% C. 2.75% D. 3.00% E. 3.25%
19.34. B. For one year, each insureds mean is λ,
and is distributed via a Gamma with:
θ = 0.04 and α = 2.

This is a Gamma-Poisson, with mixed distribution

a Negative Binomial:
with r = α = 2 and β = θ = 0.04.

We add up three individual independent drivers

and we get a Negative Binomial with:
with r = α = (3)(2) = 6, and β = 0.04.
2
r (r+1) β
f(2) =
2 (1 + β)r + 2

(6) (7) 0.042

= = 2.46%.
2 (1 + 0.04) 8
Comment: The Negative Binomial Distributions
here and in the previous solution have the same
mean, however the densities are not the same.
• For each individual driver, the number of
accidents in a year follows a Poisson Distribution.
• For each individual driver, the mean of their
Poisson Distribution λ is the same each year.
• For each individual driver, the number of
accidents each year is independent of other years.
• The number of accidents for different drivers are
independent.
• λ varies between drivers via a Gamma
Distribution with mean 0.08 and variance 0.0032.
• Moe, Larry, and Curly are each drivers.

19.35. What is the probability that Moe, Larry, and

Curly have a total of exactly 3 accidents during
the next four years?
A. 5.2% B. 5.4% C. 5.6% D. 5.8% E. 6.0%
19.35. E. For one year, each insureds mean is λ,
and is distributed via a Gamma with:
θ = 0.04 and α = 2.
Over four years, each insureds mean is 4λ, and is
distributed via a Gamma with:
θ = (4)(0.04) = 0.16, and α = 2.
This is a Gamma-Poisson, with mixed distribution
a Negative Binomial:
with r = α = 2 and β = θ = 0.16.

We add up three individual independent drivers

and we get a Negative Binomial with:
with r = α = (3)(2) = 6, and β = 0.16.
r (r + 1) (r + 2) β3
f(3) =
6 (1 + β)r + 3

(6) (7) (8) 0.163

= = 6.03%.
6 (1 + 0.16) 9
V and X are each given by the result of rolling
a six-sided die.
V and X are independent of each other.
Y= V + X.
Z = 2X.
Hint: The mean of X is 3.5
and the variance of X is 35/12.

2.9. What is the standard deviation of Y?

A. less than 2.0
B. at least 2.0 but less than 2.3
C. at least 2.3 but less than 2.6
D. at least 2.9 but less than 3.2
E. at least 3.2
2.9. C. Var[Y] = Var[V + X] = Var[V] + V[X]
= (35/12) + (35/12) = 35/6 = 5.83.
Standard Deviation[Y] = 5.83 = 2.41.
V and X are each given by the result of rolling
a six-sided die.
V and X are independent of each other.
Y= V + X.
Z = 2X.
Hint: The mean of X is 3.5
and the variance of X is 35/12.

2.10. What is the standard deviation of Z?

A. less than 2.0
B. at least 2.0 but less than 2.3
C. at least 2.3 but less than 2.6
D. at least 2.9 but less than 3.2
E. at least 3.2
2.10. E. Var[Z] = Var[2X] = 22 Var[X]
= (4) (35/12) = 35/3 = 11.67.

Standard Deviation[Z] = 11.67 = 3.42.

14.28. X is a discrete random variable with a
probability function which is a member of the
(a, b, 1) class of distributions.
pk denotes the probability that X = k.
p1 = 0.1637, p2 = 0.1754, and p3 = 0.1503.
Calculate p5.
(A) 7.5% (B) 7.7% (C) 7.9% (D) 8.1% (E) 8.3%
14.28. B.
Since we have a member of the (a, b, 1) family:
p2 / p1 = a + b/2.
⇒ 2a + b = (2)(0.1754) / 0.1637 = 2.1429.
p3 / p2 = a + b/3.
⇒ 3a + b = (3)(0.1503) / 0.1754 = 2.5707.
⇒ a = 0.4278. ⇒ b = 1.2873.

p4 = (a + b/4) p3
= (0.4278 + 1.2873/4) (0.1503) = 0.1127.

p5 = (a + b/45) p4
= (0.4278 + 1.2873/5) (0.1127) = 0.0772.

Comment: Based on a zero-modified Negative

Binomial, with r = 4, β = 0.75, and pM
0 = 20%.
Important Formulas and Ideas

Here are what I believe are the most important formulas and ideas from this study guide to know for the
exam.

Basic Concepts (Section 2)

The mean is the average or expected value of the random variable.

The mode is the point at which the density function reaches its maximum.
The median, the 50th percentile, is the first value at which the distribution function is ≥ 0.5.
The 100pth percentile as the first value at which the distribution function ≥ p.
Variance = second central moment = E[(X - E[X])2 ] = E[X2 ] - E[X]2 .
Standard Deviation = Square Root of Variance.

Binomial Distribution (Section 3)

⎛m⎞ m!
f(x) = f(x) = ⎜ ⎟ qx (1- q )m-x = qx (1- q)m - x , 0 ≤ x ≤ m.
⎝x ⎠ x! (m- x)!
Mean = mq Variance = mq(1-q)
Probability Generating Function: P(z) = {1 + q(z-1)}m
The Binomial Distribution for m =1 is a Bernoulli Distribution.
X is Binomial with parameters q and m1 , and Y is Binomial with parameters q and m2 ,
X and Y independent, then X + Y is Binomial with parameters q and m1 + m2 .

Poisson Distribution (Section 4)

f(x) = λx e−λ / x!, x ≥ 0

Mean = λ Variance = λ

Probability Generating Function: P(z) = eλ(z-1) , λ > 0.

A Poisson is characterized by a constant independent claim intensity and vice versa.

The sum of two independent variables each of which is Poisson with parameters λ1 and λ2 is

also Poisson, with parameter λ1 + λ2 .

If frequency is given by a Poisson and severity is independent of frequency, then the number
of claims above a certain amount (in constant dollars) is also a Poisson.
Geometric Distribution (Section 5)

βx
f(x) = .
(1 + β) x + 1

Mean = β Variance = β(1+β)

1
Probability Generating Function: P(z) = .
1- β(z-1)
⎛ β ⎞n
For a Geometric Distribution, for n > 0, the chance of at least n claims is: ⎜ ⎟ .
⎝ 1+ β ⎠
For a series of independent identical Bernoulli trials, the chance of the first success following x failures is
given by a Geometric Distribution with mean
β = (chance of a failure) / (chance of a success).

Negative Binomial Distribution (Section 6)

r(r + 1)...(r + x - 1) βx
f(x) = . Mean = rβ Variance = rβ(1+β)
x! (1+ β )x + r

Negative Binomial for r = 1 is a Geometric Distribution.

For the Negative Binomial Distribution with parameters β and r, with r integer, can be thought
of as the sum of r independent Geometric distributions with parameter β.
If X is Negative Binomial with parameters β and r1 , and Y is Negative Binomial with parameters β and r2 ,

X and Y independent, then X + Y is Negative Binomial with parameters β and r1 + r2 .

For a series of independent identical Bernoulli trials, the chance of success number r following x failures is
given by a Negative Binomial Distribution with parameters r and
β = (chance of a failure) / (chance of a success).
Normal Approximation (Section 7)

In general, let µ be the mean of the frequency distribution, while σ is the standard deviation of
the frequency distribution, then the chance of observing at least i claims and not more than j
claims is approximately:
(j + 0.5) - µ (i - 0.5) − µ
Φ[ ] - Φ[ ].
σ σ

Normal Distribution
F(x) = Φ((x−µ)/σ)
(x - µ)2
exp[- ]
2σ 2 , -∞ < x < ∞. exp[-x2 / 2]
f(x) = φ((x−µ)/σ) /σ = φ(x) = , -∞ < x < ∞.
σ 2π 2π

Mean = µ Variance = σ2
Skewness = 0 (distribution is symmetric) Kurtosis = 3

Skewness (Section 8)

Skewness = third central moment /STDDEV3 = E[(X - E[X])3 ]/STDDEV3

= {E[X3 ] - 3 X E[X2 ] + 2 X 3 } / Variance3/2.

A symmetric distribution has zero skewness.

Binomial Distribution with q < 1/2 ⇔ positive skewness ⇔ skewed to the right.

Binomial Distribution q = 1/2 ⇔ symmetric ⇒ zero skewness.

Binomial Distribution q > 1/2 ⇔ negative skewness ⇔ skewed to the left.

Poisson and Negative Binomial have positive skewness.

Probability Generating Function (Section 9)

Probability Generating Function, p.g.f.:

∞
P(z) = Expected Value of zn = E[zn ] = ∑ f(n) zn .
n=0
The Probability Generating Function of the sum of independent frequencies is the product of the individual
Probability Generating Functions.
The distribution determines the probability generating function and vice versa.
f(n) = (dn P(z) / dzn )z=0 / n!. f(0) = P(0). Pʼ(1) = Mean.

If a distribution is infinitely divisible, then if one takes the probability generating function to any positive
power, one gets the probability generating function of another member of the same family of
distributions. Examples of infinitely divisible distributions include: Poisson, Negative Binomial,
Compound Poisson, Compound Negative Binomial, Normal, Gamma.

Factorial Moments (Section 10)

nth factorial moment = µ(n) = E[X(X-1) .. (X+1-n)].

µ(n) = (dn P(z) / dzn )z=1.

Pʼ(1) = E[X]. Pʼʼ(1) = E[X(X-1)].

(a, b, 0) Class of Distributions (Section 11)

For each of these three frequency distributions: f(x+1) / f(x) = a + {b / (x+1)}, x = 0, 1, ...
where a and b depend on the parameters of the distribution:

Distribution a b f(0)
Binomial -q/(1-q) (m+1)q/(1-q) (1-q)m
Poisson 0 λ e−λ
Negative Binomial β/(1+β) (r-1)β/(1+β) 1/(1+β)r

Distribution Mean Variance Variance Over Mean

Binomial mq mq(1-q) 1-q < 1 Variance < Mean
Poisson λ λ 1 Variance = Mean
Negative Binomial rβ rβ(1+β) 1+β > 1 Variance > Mean

Distribution Thinning by factor of t Adding n independent, identical copies

Binomial q → tq m → nm

Poisson λ → tλ λ → nλ

Negative Binomial β → tβ r → nr
For X and Y independent:
X Y X+Y
Binomial(q, m1 ) Binomial(q, m2 ) Binomial(q, m1 + m2 )
Poisson(λ 1) Poisson(λ 2) Poisson(λ 1 + λ2)

Negative Binomial(β, r1 ) Negative Bin.(β, r2 ) Negative Bin.(β, r1 + r2 )

Accident Profiles (Section 12)

For the Binomial, Poisson and Negative Binomial Distributions:

(x+1) f(x+1) / f(x) = a(x + 1) + b, where a and b depend on the parameters of the distribution.
a < 0 for the Binomial, a = 0 for the Poisson, and a > 0 for the Negative Binomial Distribution.

Thus if data is drawn from one of these three distributions, then we expect (x+1) f(x+1) / f(x) for this data
to be approximately linear with slope a; the sign of the slope, and thus the sign of a, distinguishes
between these three distributions of the (a, b, 0) class.

Zero-Truncated Distributions (Section 13)

In general if f is a distribution on 0,1,2,3,..., then g(x) = f(x) / {1 - f(0)} is a distribution on 1,2,3, ....
We have the following three examples:
Distribution Density of the Zero-Truncated Distribution

m! qx (1- q)m - x
x! (m - x)!
Binomial x = 1, 2, 3,... , m
1 - (1- q)m

e- λ λx / x!
Poisson x = 1, 2, 3,...
1 - e- λ
r(r +1)...(r + x - 1) βx
x! (1+ β)x + r
Negative Binomial x = 1, 2, 3,...
1 - 1/ (1+ β)r

The moments of a zero-truncated distribution, g, are given in terms of those of the corresponding
untruncated distribution, f, by: Eg [Xn ] = Ef[Xn ] / {1 - f(0)}.

⎛ β ⎞x
⎜ ⎟
⎝ 1+ β ⎠
The Logarithmic Distribution has support equal to the positive integers: f(x) = .
x ln(1+β)
The (a,b,1) class of frequency distributions is a generalization of the (a,b,0) class.
As with the (a,b,0) class, the recursion formula: f(x)/f(x-1) = a + b/x applies.
However, it need only apply now for x ≥ 2, rather than x ≥ 1.

Members of the (a,b,1) family include: all the members of the (a,b,0) family, the zero-truncated versions
of those distributions: Zero-Truncated Binomial, Zero-Truncated Poisson,
Extended Truncated Negative Binomial, and the Logarithmic Distribution.
In addition the (a,b,1) class includes the zero-modified distributions corresponding to these.

Zero-Modified Distributions (Section 14)

If f is a distribution on 0,1,2,3,..., and 0 < pM

0 < 1,

then g(0) = pM M
0 , g(x) = f(x){1 - p 0 } / {1 - f(0)}, x=1, 2 , 3..., is a distribution on 0, 1, 2, 3, ....
The moments of a zero-modified distribution g are given in terms of those of f by
Eg [Xn ] = (1- pM
0 ) Ef[X ] / {1 - f(0)}.
n

Compound Frequency Distributions (Section 15)

A compound frequency distribution has a primary and secondary distribution, each of which is a frequency
distribution. The primary distribution determines how many independent random draws from the
secondary distribution we sum.

p.g.f. of compound distribution = p.g.f. of primary dist.[p.g.f. of secondary dist.]

P(z) = P1 [P2 (z)].
compound density at 0 = p.g.f. of the primary at the density at 0 of the secondary.

Moments of Compound Distributions (Section 16)

Mean of Compound Dist. = (Mean of Primary Dist.)(Mean of Sec. Dist.)

Variance of Compound Dist. = (Mean of Primary Dist.)(Var. of Sec. Dist.)
+ (Mean of Secondary Dist.)2 (Variance of Primary Dist.)

In the case of a Poisson primary distribution with mean λ, the variance of the compound distribution could
be rewritten as: λ(2nd moment of Second. Dist.).
The third central moment of a compound Poisson distribution = λ(3rd moment of Sec. Dist.).
Mixed Frequency Distributions (Section 17)

The density function of the mixed distribution, is the mixture of the density function for
specific values of the parameter that is mixed.

The nth moment of a mixed distribution is the mixture of the nth moments.
First one mixes the moments, and then computes the variance of the mixture from its first and second
moments.

The Probability Generating Function of the mixed distribution, is the mixture of the probability generating
functions for specific values of the parameter.
For a mixture of Poissons, the variance is always greater than the mean.

Gamma Function (Section 18)

The (complete) Gamma Function is defined as:

∞ ∞
Γ(α) = ∫ tα - 1 e - t dt = θ−α ∫ tα - 1e - t / θ dt , for α ≥ 0 , θ ≥ 0.
0 0

Γ(α) = (α -1)! Γ(α) = (α-1)Γ(α-1)

∞
∫ tα - 1 e - t / θ dt = Γ(α) θα.
0

The Incomplete Gamma Function is defined as:

x
Γ(α ; x) = ∫ tα - 1 e- t dt / Γ(α).
0
Gamma-Poisson Frequency Process (Section 19)

If one mixes Poissons via a Gamma, then the mixed distribution is in the form of the
Negative Binomial distribution with r = α and β = θ.

If one mixes Poissons via a Gamma Distribution with parameters α and θ, then over a period of length Y,
the mixed distribution is Negative Binomial with r = α and β = Yθ.

For the Gamma-Poisson, the variance of the mixed Negative Binomial is equal to:
mean of the Gamma + variance of the Gamma.

Var[X] = E[Var[X | λ]] + Var[E[X | λ]]. Mixing increases the variance.

Tails of Frequency Distributions (Section 20)

From lightest to heaviest tailed, the frequency distribution in the (a,b,0) class are:
Binomial, Poisson, Negative Binomial r > 1, Geometric, Negative Binomial r < 1.