Exam C Manual

An Introductory Guide in the Construction of
Actuarial Models:
A Preparation for the Actuarial Exam C/4
Marcel B. Finan
Arkansas Tech University
All
c Rights Reserved
Preliminary Draft
Last Updated
November 4, 2017
To my son
Amin
ii
Preface
This is the fifth of a series of lecture notes intended to help individuals

to pass actuarial exams. The topics in this manuscript parallel the topics
tested on Exam C of the Society of Actuaries exam sequence. As with the
previous manuscripts, the main objective of the present manuscript is to
increase users’ understanding of the topics covered on the exam.
The flow of topics follows very closely that of Klugman et al. Loss Models:
From Data to Decisions. The lectures cover designated sections from this
book as suggested by the 2012 SOA Syllabus.
The recommended approach for using this manuscript is to read each sec-
tion, work on the embedded examples, and then try ALL the problems given
in the text. An answer key is provided by request. Email:[email protected].
Problems taken from previous SOA/CAS exams will be indicated by the

symbol ‡.
This manuscript can be used for personal use or class use, but not for com-
mercial purposes. If you find any errors, I would appreciate hearing from
you: [email protected]
Marcel B. Finan
Russellville, Arkansas
February 15, 2013.
iii
iv
Contents
Preface iii
Actuarial Modeling 1
1 Understanding Actuarial Models . . . . . . . . . . . . . . . . . . 2
A Review of Probability Related Results 5

2 A Brief Review of Probability . . . . . . . . . . . . . . . . . . . . 6
3 A Review of Random Variables . . . . . . . . . . . . . . . . . . . 18
4 Raw and Central Moments . . . . . . . . . . . . . . . . . . . . . 34
5 Empirical Models, Excess and Limited Loss variables . . . . . . . 45
6 Median, Mode, Percentiles, and Quantiles . . . . . . . . . . . . . 62
7 Sum of Random Variables and the Central Limit Theorem . . . . 69
8 Moment Generating Functions and Probability Generating Func-
tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Tail Weight of a Distribution 87

9 Tail Weight Measures: Moments and the Speed of Decay of S(x) 88
10 Tail Weight Measures: Hazard Rate Function and Mean Excess
Loss Function . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
11 Equilibrium Distributions and Tail Weight . . . . . . . . . . . . 102
Risk Measures 109

12 Coherent Risk Measurement . . . . . . . . . . . . . . . . . . . . 110
13 Value-at-Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
14 Tail-Value-at-Risk . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Characteristics of Actuarial Models 125

15 Parametric and Scale Distributions . . . . . . . . . . . . . . . . 126
16 Discrete Mixture Distributions . . . . . . . . . . . . . . . . . . . 131
17 Data-dependent Distributions . . . . . . . . . . . . . . . . . . . 138
v
Generating New Distributions 143
18 Scalar Multiplication of Random Variables . . . . . . . . . . . . 144
19 Powers and Exponentiation of Random Variables . . . . . . . . 148
20 Continuous Mixing of Distributions . . . . . . . . . . . . . . . . 153
21 Frailty (Mixing) Models . . . . . . . . . . . . . . . . . . . . . . 160
22 Spliced Distributions . . . . . . . . . . . . . . . . . . . . . . . . 165
23 Limiting Distributions . . . . . . . . . . . . . . . . . . . . . . . 168
24 The Linear Exponential Family of Distributions . . . . . . . . . 172
Discrete Distributions 177

25 The Poisson Distribution . . . . . . . . . . . . . . . . . . . . . . 178
26 The Negative Binomial Distribution . . . . . . . . . . . . . . . . 183
27 The Bernoulli and Binomial Distributions . . . . . . . . . . . . 189
28 The (a, b, 0) Class of Discrete Distributions . . . . . . . . . . . . 195
29 The Class C(a, b, 1) of Discrete Distributions . . . . . . . . . . . 201
30 The Extended Truncated Negative Binomial Model . . . . . . . 206
Modifications of the Loss Random Variable 211

31 Ordinary Policy Deductibles . . . . . . . . . . . . . . . . . . . . 212
32 Franchise Policy Deductibles . . . . . . . . . . . . . . . . . . . . 220
33 The Loss Elimination Ratio and Inflation Effects for Ordinary
Deductibles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
34 Policy Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
35 Combinations of Coinsurance, Deductibles, Limits, and Inflations237
36 The Impact of Deductibles on the Number of Payments . . . . 250
Aggregate Loss Models 257

37 Individual Risk and Collective Risk Models . . . . . . . . . . . 258
38 Aggregate Loss Distributions via Convolutions . . . . . . . . . . 264
39 Stop Loss Insurance . . . . . . . . . . . . . . . . . . . . . . . . . 284
40 Closed Form of Aggregate Distributions . . . . . . . . . . . . . 294
41 Distribution of S via the Recursive Method . . . . . . . . . . . 300
42 Discretization of Continuous Severities . . . . . . . . . . . . . . 308
43 Individual Policy Modifications Impact on Aggregate Losses . . 314
44 Aggregate Losses for the Individual Risk Model . . . . . . . . . 321
45 Approximating Probabilities in the Individual Risk Model . . . 328
Review of Mathematical Statistics 333

46 Properties of Point Estimators . . . . . . . . . . . . . . . . . . . 334
47 Interval Estimation . . . . . . . . . . . . . . . . . . . . . . . . . 341
vi
48 Hypothesis Testing . . . . . . . . . . . . . . . . . . . . . . . . . 344
The Empirical Distribution for Complete Data 351

49 The Empirical Distribution for Individual Data . . . . . . . . . 352
50 Empirical Distribution of Grouped Data . . . . . . . . . . . . . 357
Estimation of Incomplete Data 363

51 The Risk Set of Incomplete Data . . . . . . . . . . . . . . . . . 364
52 The Kaplan-Meier and Nelson-Åalen Estimators . . . . . . . . . 370
53 Mean and Variance of Empirical Estimators with Complete Data379
54 Greenwood Estimate for the Variance of the Kaplan-Meier Es-
timator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
55 Variance Estimate of the Nelson-Åalen Estimator and Confi-
dence Intervals . . . . . . . . . . . . . . . . . . . . . . . . . . 393
56 Kernel Density Estimation . . . . . . . . . . . . . . . . . . . . . 400
57 The Kaplan-Meier Approximation for Large Data Sets . . . . . 409
Methods of parameter Estimation 419

58 Method of Moments and Matching Percentile . . . . . . . . . . 421
59 Maximum Likelihood Estimation for Complete Data . . . . . . 434
60 Maximum Likelihood Estimation for Incomplete Data . . . . . . 445
61 Asymptotic Variance of MLE . . . . . . . . . . . . . . . . . . . 455
62 Information Matrix and the Delta Method . . . . . . . . . . . . 463
63 Non-Normal Confidence Intervals for Parameter Estimation . . 471
64 Basics of Bayesian Inference . . . . . . . . . . . . . . . . . . . . 474
65 Bayesian Parameter Estimation . . . . . . . . . . . . . . . . . . 485
66 Conjugate Prior Distributions . . . . . . . . . . . . . . . . . . . 493
67 Estimation of Class (a, b, 0) . . . . . . . . . . . . . . . . . . . . 497
68 MLE with (a, b, 1) Class . . . . . . . . . . . . . . . . . . . . . . 506
Model Selection and Evaluation 513

69 Assessing Fitted Models Graphically . . . . . . . . . . . . . . . 514
70 Kolmogorov-Smirnov Hypothesis Test of Fitted Models . . . . . 521
71 Anderson-Darling Hypothesis Test of Fitted Models . . . . . . . 527
72 The Chi-Square Goodness of Fit Test . . . . . . . . . . . . . . . 532
73 The Likelihood Ratio Test . . . . . . . . . . . . . . . . . . . . . 540
74 Schwarz Bayesian Criterion . . . . . . . . . . . . . . . . . . . . 546
vii
Credibility Theory 551
75 Limited Fluctuation Credibility Approach: Full Credibility . . . 552
76 Limited Fluctuation Credibility Approach: Partial Credibility . 561
77 Greatest Accuracy Credibility Approach . . . . . . . . . . . . . 567
78 Conditional Distributions and Expectation . . . . . . . . . . . . 571
79 Bayesian Credibility with Discrete Prior . . . . . . . . . . . . . 581
80 Bayesian Credibility with Continuous Prior . . . . . . . . . . . 594
81 Bühlman Credibility Premium . . . . . . . . . . . . . . . . . . . 600
82 The Bühlmann Model with Discrete Prior . . . . . . . . . . . . 605
83 The Bühlmann Model with Continuous Prior . . . . . . . . . . 618
84 The Bühlmann-Straub Credibility Model . . . . . . . . . . . . . 627
85 Exact Credibility . . . . . . . . . . . . . . . . . . . . . . . . . . 637
86 Non-parametric Empirical Bayes Estimation for the Bühlmann
Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643
87 Non-parametric Empirical Bayes Estimation for the Bühlmann-
Straub Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 652
88 Semiparametric Empirical Bayes Credibility Estimation . . . . 661
Basics of Stochastic Simulation 667

89 The Inversion Method for Simulating Random Variables . . . . 668
90 Applications of Simulation in Actuarial Modeling . . . . . . . . 675
91 Estimating Risk Measures Using Simulation . . . . . . . . . . . 684
92 The Bootstrap Method for Estimating Mean Square Error . . . 686
Answer Key 691
Exam C Tables 781
BIBLIOGRAPHY 796
Index 798
viii
Actuarial Modeling
This book is concerned with the construction and evaluation of actuarial

models. The purpose of this chapter is to define models in the actuarial
setting and suggest a process of building them.
1
2 ACTUARIAL MODELING
1 Understanding Actuarial Models

Modeling is very common in actuarial applications. For example, life in-
surance actuaries use models to arrive at the likely mortality rates of their
customers; car insurance actuaries use models to work out claim probabil-
ities by rating factors; pension fund actuaries use models to estimate the
contributions and investments they will need to meet their future liabilities.
A “model” in actuarial applications is a simplified mathematical descrip-

tion of a certain actuarial task. Actuarial models are used by actuaries to
form an opinion and recommend a course of action on contingencies relating
to uncertain future events.
Commonly used actuarial models are classified into two categories:
(I) Deterministic Models. These are models that produce a unique set
of outputs for a given set of inputs such as the future value of a deposit
in a savings account. In these models, the inputs and outputs don’t have
associated probability weightings.
(II) Stochastic or Probabilistic Models. In contrast to deterministic

models, these are models where the outputs or/and some of the inputs are
random variables. Examples of stochastic models that we will discuss in
this book are the asset model, the claims model, and the frequency-severity
model.
The book in [4] explains in enormous detail the advantages and disadvan-
tages of stochastic (versus deterministic) modeling.
Example 1.1
Determine whether each of the model below is deterministic or stochastic.
(a) The monthly payment P on a home or a car loan.
(b) A modification of the model in (a) is P + ξ, where ξ is a random variable
introduced to account for the possibility of failure of making a payment.
Solution.
(a) In this model, the element of randomness is absent. This model is a
deterministic one.
(b) Because of the presence of the random variable ξ, the given model is
stochastic
1 UNDERSTANDING ACTUARIAL MODELS 3
In [1], the following process for building an actuarial model is presented.
Phases of a Good Modeling Process
A good modeling requires a thorough understanding of the problem mod-

elled. The following is a helpful checklist of a modeling process which is by
no means complete:
Choice of Models. Appropriate models are selected based on the actuary’s

prior knowledge and experience and the nature of the available data.
Model Calibration. Available data and existing techniques are used to cali-
brate a model.
Model Validation. Diagnostic tests are used to ensure the model meets its
objectives and adequately conforms to the data.
Adequacy of Models. There is a possibility that the models in the previ-

ous stage are inadequate in which case one considers other possible models.
Also, there is a possibility of having more than one adequate models.
Selection of Models. Based on some preset criteria, the best model will
be selected among all valid models.
Modification of Models. The model selected in the previous stage needs

to be constantly updated in the light of new future data and other changes.
4 ACTUARIAL MODELING
Practice Problems
Problem 1.1
After an actuary being hired, his or her annual salary progression is mod-
eled according to the formula S(t) = $45, 000e0.06t , where t is the number
of years of employment.
Determine whether this model is deterministic or stochastic.
Problem 1.2
In the previous model, a random variable ξ is introduced: S(t) = $45, 000e0.06t +
ξ.
Determine whether this model is deterministic or stochastic.
Problem 1.3
Consider a model that depends on the movement of a stock market such as
the pricing of an option with an underlying stock.
Does this model considered a deterministic or stochastic model?
Problem 1.4
Consider a model that involves the life expectancy of a policyholder.
Is this model categorized as stochastic or deterministic?
Problem 1.5
Insurance companies use models to estimate their assets and liabilities.
Are these models considered deterministic or stochastic?

A Review of Probability
Related Results
One aspect of insurance is that money is paid to policyholders upon the

occurrence of a random event. For example, a claim in an automobile in-
surance policy will be filed whenever the insured auto is involved in a car
wreck. In this chapter a brief outline of the essential material from the the-
ory of probability is given. Almost all of the material presented here should
be familiar to the reader. A more thorough discussion can be found in [2]
and a listing of important results can be found in [3]. Probability concepts
that are not usually covered in an introductory probability course will be
introduced and discussed in futher details whenever needed.
5
6 A REVIEW OF PROBABILITY RELATED RESULTS
2 A Brief Review of Probability

In probability, we consider experiments whose results cannot be predicted
with certainty. Examples of such experiments include rolling a die, flipping
a coin, and choosing a card from a deck of playing cards.
By an outcome or simple event we mean any result of the experiment.

For example, the experiment of rolling a die yields six outcomes, namely,
the outcomes 1,2,3,4,5, and 6.
The sample space Ω of an experiment is the set of all possible outcomes

for the experiment. For example, if you roll a die one time then the exper-
iment is the roll of the die. A sample space for this experiment could be
Ω = {1, 2, 3, 4, 5, 6} where each digit represents a face of the die.
An event is a subset of the sample space. For example, the event of rolling
an odd number with a die consists of three simple events {1, 3, 5}.
Example 2.1
Consider the random experiment of tossing a coin three times.
(a) Find the sample space of this experiment.
(b) Find the outcomes of the event of obtaining more than one head.
Solution.
We will use T for tail and H for head.
(a) The sample space is composed of eight simple events:
Ω = {T T T, T T H, T HT, T HH, HT T, HT H, HHT, HHH}.
(b) The event of obtaining more than one head is the set
{T HH, HT H, HHT, HHH}
The complement of an event E, denoted by E c , is the set of all possible

outcomes not in E. The union of two events A and B is the event A ∪ B
whose outcomes are either in A or in B. The intersection of two events A
and B is the event A ∩ B whose outcomes are outcomes of both events A
and B.
Two events A and B are said to be mutually exclusive if they have no

outcomes in common. Clearly, for any event E, the events E and E c are
mutually exclusive.
2 A BRIEF REVIEW OF PROBABILITY 7
Remark 2.1
The above definitions of intersection, union, and mutually exclusive can be
extended to any number of events.
Probability Axioms
Probability is the measure of occurrence of an event. It is a function Pr(·)
defined on the collection of all (subsets) events of a sample space Ω and
which satisfies Kolmogorov axioms:
Axiom 1: For any event E ⊆ Ω, 0 ≤ Pr(E) ≤ 1.

Axiom 2: Pr(Ω) = 1.
Axiom 3: For any sequence of mutually exclusive events {En }n≥1 , that is
Ei ∩ Ej = ∅ for i 6= j, we have
P∞
Pr (∪∞n=1 En ) = n=1 Pr(En ). (Countable Additivity)
If we let E1 = Ω, En = ∅ forPn > 1 then by Axioms P 2 and 3 we have

∞ ∞
1 = Pr(Ω) = Pr (∪∞ n=1 En ) = n=1 Pr(E n ) = Pr(Ω) + n=2 Pr(∅). This
implies that Pr(∅) = 0. Also, if {E1 , E2 , · · · , En } is a finite set of mutually
exclusive events, then by defining Ek = ∅ for k > n and Axiom 3 we find
n
X
Pr (∪nk=1 Ek ) = Pr(Ek ).
k=1
Any function Pr that satisfies Axioms 1-3 will be called a probability mea-
sure.
Example 2.2
Consider the sample space Ω = {1, 2, 3}. Suppose that Pr({1, 3}) = 0.3
and Pr({2, 3}) = 0.8. Find Pr(1), Pr(2), and Pr(3). Is Pr a valid probability
measure?
Solution.
For Pr to be a probability measure we must have Pr(1) + Pr(2) + Pr(3) = 1.
But Pr({1, 3}) = Pr(1) + Pr(3) = 0.3. This implies that 0.3 + Pr(2) = 1
or Pr(2) = 0.7. Similarly, 1 = Pr({2, 3}) + Pr(1) = 0.8 + Pr(1) and so
Pr(1) = 0.2. It follows that Pr(3) = 1 − Pr(1) − Pr(2) = 1 − 0.2 − 0.7 = 0.1.
It can be easily seen that Pr satisfies Axioms 1-3 and so Pr is a probability
measure
Probability Trees
For all multistage experiments, the probability of the outcome along any
path of a tree diagram is equal to the product of all the probabilities along
the path.
Example 2.3
In a city council, 35% of the members are female, and the other 65% are
male. 70% of the male favor raising city sales tax, while only 40% of the
female favor the increase. If a member of the council is selected at random,
what is the probability that he or she favors raising sales tax?
Solution.
Figure 2.1 shows a tree diagram for this problem.
Figure 2.1
The first and third branches correspond to favoring the tax. We add their
probabilities.
Pr(tax) = 0.455 + 0.14 = 0.595
Conditional Probability and Bayes Formula
Consider the question of finding the probability of an event A given that an-
other event B has occurred. Knowing that the event B has occurred causes
us to update the probabilities of other events in the sample space.
To illustrate, suppose you roll two dice of different colors; one red, and
one green. You roll each die one at time. Our sample space has 36 out-
1
comes. The probability of getting two ones is 36 . Now, suppose you were
told that the green die shows a one but know nothing about the red die.
What would be the probability of getting two ones? In this case, the answer
is 16 . This shows that the probability of getting two ones changes if you have
partial information, and we refer to this (altered) probability as a condi-

tional probability.
If the occurrence of the event A depends on the occurrence of B then the

conditional probability will be denoted by P (A|B), read as the probability
of A given B. It is given by
number of outcomes corresponding to event A and B
Pr(A|B) = .
number of outcomes of B
Thus,
n(A∩B)
n(A ∩ B) n(S) Pr(A ∩ B)
Pr(A|B) = = n(B)
=
n(B) Pr(B)
n(S)
provided that Pr(B) > 0.
Example 2.4
Let A denote the event “an immigrant is male” and let B denote the event
“an immigrant is Brazilian”. In a group of 100 immigrants, suppose 60
are Brazilians, and suppose that 10 of the Brazilians are males. Find the
probability that if I pick a Brazilian immigrant, it will be a male, that is,
find Pr(A|B).
Solution.
Since 10 out of 100 in the group are both Brazilians and male, Pr(A ∩ B) =
10 60
100 = 0.1. Also, 60 out of the 100 are Brazilians, so Pr(B) = 100 = 0.6.
Hence, Pr(A|B) = 0.1
0.6 = 6
1
It is often the case that we know the probabilities of certain events con-
ditional on other events, but what we would like to know is the “reverse”.
That is, given Pr(A|B) we would like to find Pr(B|A).
Bayes’ formula is a simple mathematical formula used for calculating Pr(B|A)

given Pr(A|B). We derive this formula as follows. Let A and B be two events.
Then
A = A ∩ (B ∪ B c ) = (A ∩ B) ∪ (A ∩ B c ).
Since the events A ∩ B and A ∩ B c are mutually exclusive, we can write
Pr(A) = Pr(A ∩ B) + Pr(A ∩ B c )

= Pr(A|B)Pr(B) + Pr(A|B c )Pr(B c ) (2.1)
Example 2.5
A soccer match may be delayed because of bad weather. The probabilities
are 0.60 that there will be bad weather, 0.85 that the game will take place
if there is no bad weather, and 0.35 that the game will be played if there is
bad weather. What is the probability that the match will occur?
Solution.
Let A be the event that the game will be played and B is the event that
there will be a bad weather. We are given Pr(B) = 0.60, Pr(A|B c ) = 0.85,
and Pr(A|B) = 0.35. From Equation (2.1) we find
Pr(A) = Pr(B)Pr(A|B)+Pr(B c )Pr(A|B c ) = (0.60)(0.35)+(0.4)(0.85) = 0.55
From Equation (2.1) we can get Bayes’ formula:

Pr(A ∩ B) Pr(A|B)Pr(B)
Pr(B|A) = = . (2.2)
Pr(A) Pr(A|B)Pr(B) + Pr(A|B c )Pr(B c )
Example 2.6
A factory uses two machines A and B for making socks. Machine A produces
10% of the total production of socks while machine B produces the remaining
90%. Now, 1% of all the socks produced by A are defective while 5% of all
the socks produced by B are defective. Find the probability that a sock
taken at random from a day’s production was made by the machine A,
given that it is defective?
Solution.
We are given Pr(A) = 0.1, Pr(B) = 0.9, Pr(D|A) = 0.01, and Pr(D|B) =
0.05. We want to find Pr(A|D). Using Bayes’ formula we find
Pr(A ∩ D) Pr(D|A)Pr(A)
Pr(A|D) = =
Pr(D) Pr(D|A)Pr(A) + Pr(D|B)Pr(B)
(0.01)(0.1)
= ≈ 0.0217
(0.01)(0.1) + (0.05)(0.9)
Formula 2.2 is a special case of the more general result:
Theorem 2.1 (Bayes’ Theorem)

Suppose that the sample space Ω is the union of mutually exclusive events
H1 , H2 , · · · , Hn with P (Hi ) > 0 for each i. Then for any event A and 1 ≤
i ≤ n we have
Pr(A|Hi )Pr(Hi )
Pr(Hi |A) =
Pr(A)
where
Pr(A) = Pr(H1 )Pr(A|H1 ) + Pr(H2 )Pr(A|H2 ) + · · · + Pr(Hn )Pr(A|Hn ).
Example 2.7
A survey is taken in Oklahoma, Kansas, and Arkansas. In Oklahoma, 50%
of surveyed support raising tax, in Kansas, 60% support a tax increase, and
in Arkansas only 35% favor the increase. Of the total population of the
three states, 40% live in Oklahoma, 25% live in Kansas, and 35% live in
Arkansas. Given that a surveyed person is in favor of raising taxes, what is
the probability that he/she lives in Kansas?
Solution.
Let LI denote the event that a surveyed person lives in state I, where I
= OK, KS, AR. Let S denote the event that a surveyed person favors tax
increase. We want to find Pr(LKS |S). By Bayes’ formula we have
Pr(S|LKS )Pr(LKS )
Pr(LKS |S) =
Pr(S|LOK )Pr(LOK ) + Pr(S|LKS )Pr(LKS ) + Pr(S|LAR )Pr(LAR )
(0.6)(0.25)
= ≈ 0.3175
(0.5)(0.4) + (0.6)(0.25) + (0.35)(0.35)
Practice Problems
Problem 2.1
Consider the sample space of rolling a die. Let A be the event of rolling
an even number, B the event of rolling an odd number, and C the event of
rolling a 2.
Find
(a) Ac , B c and C c .
(b) A ∪ B, A ∪ C, and B ∪ C.
(c) A ∩ B, A ∩ C, and B ∩ C.
(d) Which events are mutually exclusive?
Problem 2.2
If, for a given experiment, O1 , O2 , O3 , · · · is an infinite sequence of outcomes,
verify that
i
1
Pr(Oi ) = , i = 1, 2, 3, · · ·
2
is a probability measure.
Problem 2.3 ‡
An insurer offers a health plan to the employees of a large company. As
part of this plan, the individual employees may choose exactly two of the
supplementary coverages A, B, and C, or they may choose no supplementary
coverage. The proportions of the company’s employees that choose cover-
ages A, B, and C are 14 , 13 , and , 12
5
respectively.
Determine the probability that a randomly chosen employee will choose

no supplementary coverage.
Problem 2.4
A toll has two crossing lanes. Let A be the event that the first lane
is busy, and let B be the event the second lane is busy. Assume that
Pr(A) = 0.2, Pr(B) = 0.3 and Pr(A ∩ B) = 0.06.
Find the probability that both lanes are not busy.

Hint: Recall the identity
Pr(A ∪ B) = Pr(A) + Pr(B) − Pr(A ∩ B).

Problem 2.5
If a person visits a car service center, suppose that the probability that he
will have his oil changed is 0.44, the probability that he will have a tire
replacement is 0.24, the probability that he will have airfilter replacement
is 0.21, the probability that he will have oil changed and a tire replaced is
0.08, the probability that he will have oil changed and air filter changed is
0.11, the probability that he will have a tire and air filter replaced is 0.07,
and the probability that he will have oil changed, a tire replacement, and
an air filter changed is 0.03.
What is the probability that at least one of these things done to the car?
Recall that
Pr(A∪B∪C) = Pr(A)+Pr(B)+Pr(C)−Pr(A∩B)−Pr(A∩C)−Pr(B∩C)+Pr(A∩B∩C)
Problem 2.6 ‡
A survey of a group’s viewing habits over the last year revealed the following
information
(i) 28% watched gymnastics
(ii) 29% watched baseball
(iii) 19% watched soccer
(iv) 14% watched gymnastics and baseball
(v) 12% watched baseball and soccer
(vi) 10% watched gymnastics and soccer
(vii) 8% watched all three sports.
Find the probability of a viewer that watched none of the three sports during
the last year.
Problem 2.7 ‡
The probability that a visit to a primary care physician’s (PCP) office re-
sults in neither lab work nor referral to a specialist is 35% . Of those coming
to a PCP’s office, 30% are referred to specialists and 40% require lab work.
Determine the probability that a visit to a PCP’s office results in both

lab work and referral to a specialist.
Problem 2.8 ‡
You are given Pr(A ∪ B) = 0.7 and Pr(A ∪ B c ) = 0.9.
Determine Pr(A).
Problem 2.9 ‡
Among a large group of patients recovering from shoulder injuries, it is found
that 22% visit both a physical therapist and a chiropractor, whereas 12%
visit neither of these. The probability that a patient visits a chiropractor
exceeds by 14% the probability that a patient visits a physical therapist.
Determine the probability that a randomly chosen member of this group

visits a physical therapist.
Problem 2.10 ‡
In modeling the number of claims filed by an individual under an auto-
mobile policy during a three-year period, an actuary makes the simplifying
assumption that for all integers n ≥ 0, pn+1 = 15 pn , 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?
Problem 2.11
An urn contains three red balls and two blue balls. You draw two balls
without replacement. Construct a probability tree diagram that represents
the various outcomes that can occur.
What is the probability that the first ball is red and the second ball is
blue?
Problem 2.12
Repeat the previous exercise but this time replace the first ball before draw-
ing the second.
Problem 2.13 ‡
A public health researcher examines the medical records of a group of 937
men who died in 1999 and discovers that 210 of the men died from causes
related to heart disease. Moreover, 312 of the 937 men had at least one par-
ent who suffered from heart disease, and, of these 312 men, 102 died from
causes related to heart disease.
Determine the probability that a man randomly selected from this group
died of causes related to heart disease, given that neither of his parents
suffered from heart disease.
Problem 2.14 ‡
An actuary is studying the prevalence of three health risk factors, denoted
by A, B, and C, within a population of women. For each of the three fac-
tors, the probability is 0.1 that a woman in the population has only this risk
factor (and no others). For any two of the three factors, the probability is
0.12 that she has exactly these two risk factors (but not the other). The
probability that a woman has all three risk factors, given that she has A
and B, is 31 .
What is the probability that a woman has none of the three risk factors,
given that she does not have risk factor A?
Problem 2.15 ‡
An auto insurance company insures drivers of all ages. An actuary compiled
the following statistics on the company’s insured drivers:
Age of Probability Portion of Company’s
Driver of Accident Insured Drivers
16 - 20 0.06 0.08
21 - 30 0.03 0.15
31 - 65 0.02 0.49
66 - 99 0.04 0.28
A randomly selected driver that the company insures has an accident.
Calculate the probability that the driver was age 16-20.
Problem 2.16 ‡
An insurance company issues life insurance policies in three separate cate-
gories: standard, preferred, and ultra-preferred. Of the company’s policy-
holders, 50% are standard, 40% are preferred, and 10% are ultra-preferred.
Each standard policyholder has probability 0.010 of dying in the next year,
each preferred policyholder has probability 0.005 of dying in the next year,
and each ultra-preferred policyholder has probability 0.001 of dying in the
next year.
A policyholder dies in the next year.
What is the probability that the deceased policyholder was ultra-preferred?
Problem 2.17 ‡
Upon arrival at a hospital’s emergency room, patients are categorized ac-
cording to their condition as critical, serious, or stable. In the past year:
(i) 10% of the emergency room patients were critical;

(ii) 30% of the emergency room patients were serious;
(iii) the rest of the emergency room patients were stable;
(iv) 40% of the critical patients died;
(vi) 10% of the serious patients died; and
(vii) 1% of the stable patients died.
Given that a patient survived, what is the probability that the patient was
categorized as serious upon arrival?
Problem 2.18 ‡
A health study tracked a group of persons for five years. At the beginning
of the study, 20% were classified as heavy smokers, 30% as light smokers,
and 50% as nonsmokers.
Results of the study showed that light smokers were twice as likely as non-
smokers to die during the five-year study, but only half as likely as heavy
smokers.
A randomly selected participant from the study died over the five-year pe-
riod.
Calculate the probability that the participant was a heavy smoker.
Problem 2.19 ‡
An actuary studied the likelihood that different types of drivers would be
involved in at least one collision during any one-year period. The results of
the study are presented below.
Probability
Type of Percentage of of at least one
driver all drivers collision
Teen 8% 0.15
Young adult 16% 0.08
Midlife 45% 0.04
Senior 31% 0.05
Total 100%
Given that a driver has been involved in at least one collision in the past
year, what is the probability that the driver is a young adult driver?
Problem 2.20 ‡
A blood test indicates the presence of a particular disease 95% of the time
when the disease is actually present. The same test indicates the presence
of the disease 0.5% of the time when the disease is not present. One percent
of the population actually has the disease.
Calculate the probability that a person has the disease given that the test
indicates the presence of the disease.
3 A Review of Random Variables

Let Ω be the sample space of an experiment. Any function X : Ω −→ R
is called a random variable with support the range of X. The function
notation X(s) = x means that the random variable X assigns the value x
to the outcome s.
We consider three types of random variables: Discrete, continuous, and

mixed random variables.
A random variable is called discrete if either its support is finite or a count-

ably infinite. For example, in the experiment of rolling two dice, let X be
the random variable that adds the two faces. The support of X is the finite
set {2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}. An example of an infinite discrete random
variable is the random variable that counts the number of times you play
a lottery until you win. For such a random variable, the support is the set N.
A random variable is called continuous if its support is uncountable. An

example of a continuous random variable is the random variable X that
gives a randomnly chosen number between 2 and 3 inclusively. For such a
random variable the support is the interval [2, 3].
A mixed random variable is partly discrete and partly continuous. An

example of a mixed random variable is the random variable X : (0, 1) −→ R
defined by
1 − s, 0 < s < 21

X(s) = 1 1
2, 2 ≤ s < 1.
We use upper-case letters X, Y, Z, etc. to represent random variables. We
use small letters x, y, z, etc to represent possible values that the correspond-
ing random variables X, Y, Z, etc. can take. The statement X = x defines
an event consisting of all outcomes with X−measurement equal to x which
is the set {s ∈ Ω : X(s) = x}.
Example 3.1
State whether the random variables are discrete, continuous, or mixed.
(a) A coin is tossed ten times. The random variable X is the number of
heads that are noted.
(b) A coin is tossed repeatedly. The random variable X is the number of
times needed to get the first head.
(c) X : (0, 1) −→ R defined by X(s) = 2s − 1.
3 A REVIEW OF RANDOM VARIABLES 19
1
(d) X : (0, 1) −→ R defined by X(s) = 2s − 1 for 0 < s < 2 and X(s) = 1
for 12 ≤ s < 1.
Solution.
(a) The support of X is {1, 2, 3, · · · , 10}. X is an example of a finite discrete
random variable.
(b) The support of X is N. X is an example of a countably infinite discrete
random variable.
(c) The support of X is the open interval (−1, 1). X is an example of a
continuous random variable.
(d) X is continuous on (0, 21 ) and discrete on [ 21 , 1)
Because the value of a random variable is determined by the outcome of

the experiment, we can find the probability that the random variable takes
on each value. The notation Pr(X = x) stands for the probability of the
event {s ∈ Ω : X(s) = x}.
There are five key functions used in describing a random variable:
Probability Mass Function (PMF)

For a discrete random variable X, the distribution of X is described by the
probability function(pf) or the probability mass function(pmf) given
by the equation
p(x) = Pr(X = x).
That is, a probability mass function (pmf) gives the probability that a dis-
crete random variable is exactly equal to some value. Note that the domain
of the pf is the support of the corresponding random variable. The pmf can
be an equation, a table, or a graph that shows how probability is assigned
to possible values of the random variable.
Example 3.2
Suppose a variable X can take the values 1, 2, 3, or 4. The probabilities
associated with each outcome are described by the following table:
x 1 2 3 4
p(x) 0.1 0.3 0.4 0.2
Draw the probability histogram.

Solution.
The probability histogram is shown in Figure 3.1
Figure 3.1
Example 3.3
A committee of m is to be selected from a group consisting of x men and y
women. Let X be the random variable that represents the number of men
in the committee. Find p(n) for 0 ≤ n ≤ m.
Solution.
For 0 ≤ n ≤ m, we have

x y
n m−n
p(n) =
x+y
m
Note that if the support of a discrete random variable is Support = {x1 , x2 , · · · }

then
p(x) ≥ 0, x ∈ Support
p(x) = 0, x 6∈ Support
Moreover, X
p(x) = 1.
x∈Support
Probability Density Function

Associated with a continuous random variable X is a nonnegative function
f (not necessarily continuous) defined for all real numbers and having the
property that for any set B of real numbers we have
Z
Pr(X ∈ B) = f (x)dx.
B
We call the function f the probability density function (abbreviated

pdf) of the random variable X.
If we let B = (−∞, ∞) = R then

Z ∞
f (x)dx = Pr[X ∈ (−∞, ∞)] = 1.
−∞
Now, if we let B = [a, b] then

Z b
Pr(a ≤ X ≤ b) = f (x)dx.
a
That is, areas under the probability density function represent probabilities
as illustrated in Figure 3.2.
Figure 3.2
Now, if we let a = b in the previous formula we find
Z a
Pr(X = a) = f (x)dx = 0.
a
It follows from this result that
Pr(a ≤ X < b) = Pr(a < X ≤ b) = Pr(a < X < b) = Pr(a ≤ X ≤ b)
and
Pr(X ≤ a) = Pr(X < a) and Pr(X ≥ a) = Pr(X > a).
Example 3.4
Suppose that the function f (t) defined below is the density function of some
random variable X. −t
e t ≥ 0,
f (t) =
0 t < 0.
Compute P (−10 ≤ X ≤ 10).
Solution.
Z 10 Z 0 Z 10
P (−10 ≤ X ≤ 10) = f (t)dt = f (t)dt + f (t)dt
−10 −10 0
Z 10
10
= e−t dt = −e−t 0 = 1 − e−10
0
Cumulative Distribution Function

The cumulative distribution function (abbreviated cdf) F (t) of a ran-
dom variable X is defined as follows
F (t) = Pr(X ≤ t)
i.e., F (t) is equal to the probability that the variable X assumes values,
which are less than or equal to t.
For a discrete random variable, the cumulative distribution function is found

by summing up the probabilities. That is,
X
F (t) = Pr(X ≤ t) = p(x).
x≤t
Example 3.5
Given the following pmf

1, if x = a
p(x) =
0, otherwise.
Find a formula for F (x) and sketch its graph.

Solution.
A formula for F (x) is given by

0, if x < a
F (x) =
1, otherwise
Its graph is given in Figure 3.3
Figure 3.3
For discrete random variables the cumulative distribution function will al-
ways be a step function with jumps at each value of x that has probability
greater than 0. Note that the value of F (x) is assigned to the top of the jump.
For a continuous random variable, the cumulative distribution function is

given by Z t
F (t) = f (y)dy.
−∞
Geometrically, F (t) is the area under the graph of f to the left of t.
Example 3.6
Find the distribution functions corresponding to the following density func-
tions:
1
(a)f (x) = , −∞ < x < ∞.
π(1 + x2 )
e−x
(b)f (x) = , −∞ < x < ∞.
(1 + e−x )2
Solution.
(a)
Z x x
1 1
F (x) = 2
dy = arctan y
−∞ π(1 + y ) π −∞
1 1 −π 1 1
= arctan x − · = arctan x + .
π π 2 π 2
(b)
x
e−y
Z
F (x) = −y 2
dy
−∞ (1 + e )
x
1 1
= −y
=
1+e −∞ 1 + e−x
Next, we list the properties of the cumulative distribution function F (x) for
any random variable X.
Theorem 3.1
The cumulative distribution function of a random variable X satisfies the
following properties:
(a) 0 ≤ F (x) ≤ 1.
(b) F (x) is a non-decreasing function, i.e. if a < b then F (a) ≤ F (b).
(c) F (x) → 0 as x → −∞ and F (x) → 1 as x → ∞.
(d) F is right-continuous.
In addition to the above properties, a continuous random variable satisfies

these properties:
(e) F 0 (x) = f (x).
(f) F (x) is continuous.
For a discrete random variable with support {x1 , x2 , · · · , } we have
p(xi ) = F (xi ) − F (xi−1 ). (3.1)
Example 3.7
If the distribution function of X is given by

 0 x<0
1


 0≤x<1
 16

5

F (x) = 16 1 ≤ x < 2
11

 16 2 ≤ x < 3
15
3≤x<4


 16


1 x≥4
find the pmf of X.
Solution.
1
Using 3.1, we get p(0) = 16 , p(1) = 14 , p(2) = 38 , p(3) = 14 , and p(4) = 1
16 and
0 otherwise
The Survival Distribution Function of X

The survival function (abbreviated SDF), also known as a reliability
function is a property of any random variable that maps a set of events,
usually associated with mortality or failure of some system, onto time. It
captures the probability that the system will survive beyond a specified time.
Thus, we define the survival distribution function by
S(x) = Pr(X > x) = 1 − F (x).
It follows from Theorem 3.1, that any random variable satisfies the prop-
erties: S(−∞) = 1, S(∞) = 0, S(x) is right-continuous, and that S(x) is
nonincreasing.
For a discrete random variable, the survival function is given by

X
S(x) = p(t)
t>x
and for a continuous random variable, we have

Z ∞
S(x) = f (t)dt.
x
Note that S 0 (t) = −f (t).
Remark 3.1
For a discrete random variable, the survival function need not be left-
continuous, that is, it is possible for its graph to jump down. When it
jumps, the value is assigned to the bottom of the jump.
Example 3.8 ‡
For watches produced by a certain manufacturer:
(i) Lifetimes follow a single-parameter Pareto distribution with α¿ 1 and
θ = 4.
(ii) The expected lifetime of a watch is 8 years.
Calculate the probability that the lifetime of a watch is at least 6 years.
Solution.
From Table C, we have
αθ 4α
E(X) = = = 8 =⇒ α = 2.
α−1 α−1
Also, from Table C, we have

α
θ
F (x) = 1 − =⇒ F (6) = 0.555.
x
Hence, S(6) = 1 − F (6) = 1 − 0.555 = 0.444
The Hazard Rate Function

The hazard rate function, also known as the force of mortality or the
failue rate function, is defined to be the ratio of the density and the
survival functions:
f (x) f (x)
h(x) = hX (x) = = .
S(x) 1 − F (x)
Example 3.9
Show that
S 0 (x) d
h(x) = − = − [ln S(x)]. (3.2)
S(x) dx
Solution.
S 0 (x)
The equation follows from f (x) = −S 0 (x) and d
dx [ln S(x)] = S(x)
Example 3.10
Find the hazard rate function of a random variable with pdf given by f (x) =
e−ax , a > 0.
Solution.
We have
f (x) ae−ax
h(x) = = −ax = a
S(x) e
Example 3.11
Let X be a random variable with support [0, ∞). Show that
S(x) = e−Λ(x)
where Z x
Λ(x) = h(s)ds.
0
Λ(x) is called the cumulative hazard function

Solution.
Integrating equation (3.2) from 0 to x, we have
Z x Z x
d
h(s)ds = − [ln S(s)]ds = ln S(0)−ln S(x) = ln 1−ln S(x) = − ln S(x).
0 0 ds
Now the result follows upon exponentiation
Some Additional Results

For a given event A, we define the indicator of A by

1, x ∈ A
IA (x) =
0, x 6∈ A.
Let X and Y be two random variables. It is proven in advanced probability

theory that the conditional expectation E(X|Y ) is given by the formula
E(XIY )
E(X|Y ) = . (3.3)
Pr(Y )
Example 3.12
Let X and Y be two random variables. Find a formula of E[(X −d)k |X > d]
in the
(a) discrete case
(b) continuous case.
Solution.
(a) We have
− d)k p(xj )
P
k xj >d (x
E[(X − d) |X > d] = .
Pr(X > d)
(b) We have
Z ∞
k 1
E[(X − d) |X > d] = (x − d)k fX (x)dx
Pr(X > d) d
If Ω = A ∪ B then for any random variable X on Ω, we have by the double

expectation property1
E(X) = E(X|A)Pr(A) + E(X|B)Pr(B).

1
See Section 37 of [2]
Probabilities can be found by conditioning:

X
Pr(A) = Pr(A|Y = y)Pr(Y = y)
y
in the discrete case and

Z ∞
Pr(A) = Pr(A|Y = y)fY (y)dy
−∞
in the continuous case.
Weighted mean and variance

Given a set of data x1 , x2 , · · · , xn with weights w1 , w2 , · · · , wn . The weighted
mean is given by
w1 x 1 + · · · + x n wn
X=
w1 + · · · + wn
and the weighted variance is
n
1 X
Var(X) = Pn wi (xi − X)2 .
w
i=1 i − 1
i=1
Example 3.13
You are given the following information
xi 0 1 2 3 4 5
wi 512 307 123 41 11 6
Determine the weighted variance.
Solution.
The weighted mean is
0(512) + 1(307) + 2(123) + 3(41) + 4(11) + 5(6)

X= = 0.75.
512 + 307 + 123 + 41 + 11 + 6
The weighted variance is
1
Var(X) = [512(0 − 0.75)2 + 307(1 − 0.75)2 + 123(2 − 0.75)2 + 41(3 − 0.75)2
1000 − 1
+11(4 − 0.75)2 + 6(5 − 0.75)2 ] = 0.93243
Practice Problems
Problem 3.1
State whether the random variables are discrete, continuous, or mixed.
(a) In two tossing of a coin, let X be the number of heads in the two tosses.
(b) An urn contains one red ball and one green ball. Let X be the number
of picks necessary in getting the first red ball.
(c) X is a random number in the interval [4, 7].
(d) X : R −→ R such that X(s) = s if s is irrational and X(s) = 1 if s is
rational.
Problem 3.2
Toss a pair of fair dice. Let X denote the sum of the dots on the two faces.
Find the probability mass function.
Problem 3.3
Consider the random variable X : {S, F } −→ R defined by X(S) = 1 and
X(F ) = 0. Suppose that p = Pr(X = 1).
Find the probability mass function of X.
Problem 3.4 ‡
The loss due to a fire in a commercial building is modeled by a random
variable X with density function

0.005(20 − x) 0 < x < 20
f (x) =
0 otherwise.
Given that a fire loss exceeds 8, what is the probability that it exceeds 16 ?
Problem 3.5 ‡
The lifetime of a machine part has a continuous distribution on the interval
(0, 40) with probability density function f, where f (x) is proportional to
(10 + x)−2 .
Calculate the probability that the lifetime of the machine part is less than
6.
Problem 3.6 ‡
A group insurance policy covers the medical claims of the employees of a
small company. The value, V, of the claims made in one year is described
by
V = 100000Y
where Y is a random variable with density function
k(1 − y)4 0 < y < 1

f (x) =
0 otherwise
where k is a constant.
What is the conditional probability that V exceeds 40,000, given that V

exceeds 10,000?
Problem 3.7 ‡
An insurance policy pays for a random loss X subject to a deductible of
C, where 0 < C < 1. The loss amount is modeled as a continuous random
variable with density function

2x 0 < x < 1
f (x) =
0 otherwise.
Given a random loss X, the probability that the insurance payment is less
than 0.5 is equal to 0.64 .
Calculate C.
Problem 3.8
Let X be a continuous random variable with pdf
αxe−x , x > 0

f (x) =
0, x ≤ 0.
Determine the value of α.
Problem 3.9
Consider the following probability distribution
x 1 2 3 4
p(x) 0.25 0.5 0.125 0.125
Find a formula for F (x) and sketch its graph.
Problem 3.10
Find the distribution functions corresponding to the following density func-
tions:
a−1
(a)f (x) = , 0 < x < ∞, 0 otherwise.
(1 + x)a
α
(b)f (x) =kαxα−1 e−kx , k > 0, α, 0 < x < ∞, 0 otherwise.
Problem 3.11
Let X be a random variable with pmf
1 2 n

p(n) = , n = 0, 1, 2, · · · .
3 3
Find a formula for F (n).
Problem 3.12
Given the pdf of a continuous random variable X.
1 −x
5e
5 if x ≥ 0
f (x) =
0 otherwise.
(a) Find Pr(X > 10).

(b) Find Pr(5 < X < 10).
(c) Find F (x).
Problem 3.13
A random variable X has the cumulative distribution function
ex
F (x) = .
ex + 1
Find the probability density function.
Problem 3.14
Consider an age-at-death random variable X with survival distribution de-
fined by
1 1
S(x) = (100 − x) 2 , 0 ≤ x ≤ 100.
10
(a) Explain why this is a suitable survival function.

(b) Find the corresponding expression for the cumulative probability func-
tion.
(c) Compute the probability that a newborn with survival function defined
above will die between the ages of 65 and 75.
Problem 3.15
Consider an age-at-death random variable X with survival distribution de-
fined by
S(x) = e−0.34x , x ≥ 0.
Compute Pr(5 < X < 10).
Problem 3.16
Consider an age-at-death random variable X with survival distribution S(x) =
x2
1 − 100 for x ≥ 0.
Find F (x).
Problem 3.17
Consider an age-at-death random variable X. The survival distribution is
x
given by S(x) = 1 − 100 for 0 ≤ x ≤ 100 and 0 for x > 100.
(a) Find the probability that a person dies before reaching the age of 30.
(b) Find the probability that a person lives more than 70 years.
Problem 3.18
An age-at-death random variable has a survival function
1 1
S(x) = (100 − x) 2 , 0 ≤ x ≤ 100
10
and 0 otherwise.
Find the hazard rate function of this random variable.
Problem 3.19
Consider an age-at-death random variable X with force of mortality h(x) =
µ > 0.
Find S(x), f (x), and F (x).

Problem 3.20
Let x 61
F (x) = 1 − 1 − , 0 ≤ x ≤ 120.
120
Find h(40).
4 Raw and Central Moments

Several quantities can be computed from the pdf that describe simple charac-
teristics of the distribution. These are called moments. The most common
is the mean, the first moment about the origin, and the variance, the second
moment about the mean. The mean is a measure of the centrality of the
distribution and the variance is a measure of the spread of the distribution
about the mean.
The nth moment = µ0n = E(X n ) of a random variable X is also known

as the nth moment about the origin or the nth raw moment. For a
continuous random variable X we have
Z ∞
µ0n = xn f (x)dx
−∞
and for a discrete random variable we have

X
µ0n = xn p(x).
x
By contrast, the quantity µn = E[(X − E(X))n ] is called the nth central

moment of X or the nth moment about the mean. For a continuous
random variable X we have
Z ∞
µn = (x − E(X))n f (x)dx
−∞
and for a discrete random variable we have

X
µn = (x − E(X))n p(x).
x
Note that Var(X) is the second central moment of X.
Example 4.1
Let X be a continuous random variable with pdf given by f (x) = 83 x2 for
0 < x < 2 and 0 otherwise. Find the second central moment of X.
Solution.
We first find the mean of X. We have
3 4 2
Z 2 Z 2
3 3
E(X) = xf (x)dx = x dx = x = 1.5.
0 0 8 32 0
4 RAW AND CENTRAL MOMENTS 35
The second central moment is

Z 2
µ2 = (x − 1.5)2 f (x)dx
0
Z 2
3 2
= x (x − 1.5)2 dx
0 8
2
3 x5

= − 0.75x4 + 0.75x3 = 0.15
8 5 0
The importance of moments is that they are used to define quantities that
characterize the shape of a distribution. These quantities which will be dis-
cussed below are: skewness, kurtosis and coefficient of variation.
Departure from Normality: Coefficient of Skewness

The third central moment, µ3 , is called the skewness and is a measure of
the symmetry of the pdf. A distribution, or data set, is symmetric if it looks
the same to the left and right of the mean.
A measure of skewness is given by the coefficient of skewness γ1 :
µ3 E(X 3 ) − 3E(X)E(X 2 ) + 2[E(X)]3

γ1 = = 3 .
σ3 [E(X 2 ) − E(X)2 ] 2
That is, γ1 is the ratio of the third central moment to the cube of the
standard deviation. Equivalently, γ1 is the third central moment of the
standardized variable
X −µ
X∗ = .
σ
If γ1 is close to zero then the distribution is symmetric about its mean such
as the normal distribution. A positively skewed distribution has a “tail”
which is pulled in the positive direction. A negatively skewed distribution
has a “tail” which is pulled in the negative direction (see Figure 4.1).
Figure 4.1
Example 4.2
A random variable X has the following pmf:
x 120 122 124 150 167 245

1 1 1 1 1 1
p(x) 4 12 6 12 12 3
Find the coefficient of skewness of X.
Solution.
We first find the mean of X :
1 1 1 1 1 1 2027
µ = E(X) = 120× +122× +124× +150× +167× +245× = .
4 12 6 12 12 3 12
The second raw moment is
1 1 1 1 1 1 379325
E(X 2 ) = 1202 × +1222 × +1242 × +1502 × +1672 × +2452 × = .
4 12 6 12 12 3 12
Thus, the variance of X is
379325 4108729 443171

Var(X) = − =
12 144 144
and the standard deviation is
r
443171
σ= = 55.475908183.
144
The third central moment is
2027 3 1 2027 3 2027 3 1

1
µ3 = 120 − × + 122 − × + 124 − ×
12 4 12 12 12 6
3 3 3
2027 1 2027 1 2027 1
+ 150 − × + 167 − × + 245 − ×
12 12 12 12 12 3
=93270.81134.
Thus,
93270.81134
γ1 = = 0.5463016252
55.4759081833
Example 4.3
Let X be a random variable with density f (x) = e−x on (0, ∞) and 0
otherwise. Find the coefficient of skewness of X.
Solution.
Since
Z ∞ ∞
E(X) = xe−x dx = −e−x (1 + x)0 = 1
Z0 ∞
∞
E(X 2 ) = x2 e−x dx = −e−x (x2 + 2x + 2)0 = 2
Z0 ∞
∞
3
E(X ) = x3 e−x dx = −e−x (x3 + 3x2 + 6x + 6)0 = 6
0
we find
6 − 3(1)(2) + 2(1)3
γ1 = 3 =2
(2 − 12 ) 2
Coefficient of Kurtosis
The fourth central moment, µ4 , is called the kurtosis and is a measure of
peakedness/flatness of a distribution with respect to the normal distribution.
A measure of kurtosis is given by the coefficient of kurtosis:

E[(X − µ)4 ] E(X 4 ) − 4E(X 3 )E(X) + 6E(X 2 )[E(X)]2 − 3[E(X)]4
γ2 = = .
σ4 [E(X 2 ) − (E(X))2 ]2
The coefficient of kurtosis of the normal distribution is 3. The condition
γ2 < 3 indicates that the distribution is flatter compared to the normal
distribution, and the condition γ2 > 3 indicates a higher peak (relative to
the normal distribution) around the mean value.(See Figure 4.2)
Figure 4.2
Example 4.4
x 120 122 124 150 167 245

1 1 1 1 1 1
p(x) 4 12 6 12 12 3
Find the coefficient of kurtosis of X.
Solution.
We first find the fourth central moment.
2027 4 1 2027 4 2027 4 1

1
µ4 = 120 − × + 122 − × + 124 − ×
12 4 12 12 12 6
4 4 4
2027 1 2027 1 2027 1
+ 150 − × + 167 − × + 245 − ×
12 12 12 12 12 3
=13693826.62.
Thus,
13693826.62
γ2 = = 1.44579641
55.4759081834
Example 4.5
Find the coefficient of kurtosis of the random variable X with density func-
tion f (x) = 1 on (0, 1) and 0 elsewhere.
Solution.
Since Z 1
1
E(X k ) = xk dx = .
0 k+1
we obtain,
1 1
1
1
1 2 1 4

5 −4 4 2 +6 3 2 −3 2 9
γ2 = =
1 1 2 5

3 − 4
Coefficient of Variation
Some combinations of the raw moments and central moments that are also
commonly used. One such combination is the coefficient of variation,
denoted by CV (X), of a random variable X which is defined as the ratio of
the standard deviation to the mean:
σ
CV (X) = , µ = µ01 = E(X).
µ
It is an indication of the size of the standard deviation relative to the mean,

for the given random variable.
Often the coefficient of variation is expressed as a percentage. Thus, it

expresses the standard deviation as a percentage of the sample mean and
it is unitless. Statistically, the coefficient of variation is very useful when
comparing two or more sets of data that are measured in different units of
measurement.
Example 4.6
Let X be a random variable with mean of 4 meters and standard deviation
of 0.7 millimeters. Find the coefficient of variation of X.
Solution.
The coefficient of variation is
0.7
CV (X) = = 0.0175%
4000
Example 4.7
x 120 122 124 150 167 245
1 1 1 1 1 1
p(x) 4 12 6 12 12 3
Find the coefficient of variation of X.
Solution.
We know that µ = 2027 12 = 168.9166667 and σ = 55.47590818. Thus, the
coefficient of variation of X is
55.47590818
CV (X) = = 0.3284217754
168.9166667
Example 4.8
Find the coefficient of variation of the random variable X with density func-
tion f (x) = e−x on (0, ∞) and 0 otherwise.
Solution.
We have Z ∞ ∞
µ = E(X) = xe−x dx = −e−x (1 + x)0 = 1
0
and
1 1
σ = (E(X 2 ) − (E(X))2 ) 2 = (2 − 1) 2 = 1.
Hence,
CV (X) = 1
Practice Problems
Problem 4.1
Consider n independent trials. Let X denote the number of successes in n
trials. We call X a binomial random variable. Its pmf is given by
p(r) = C(n, r)pr (1 − p)n−r
where p is the probability of a success.
(a) Show that E(X) = np and E[X(X − 1)] = n(n − 1)p2 . Hint: (a + b)n =
P n k n−k .
k=0 C(n, k)a b
(b) Find the variance of X.
Problem 4.2
A random variable X is said to be a Poisson random variable with param-
eter λ > 0 if its probability mass function has the form
λk
p(k) = e−λ , k = 0, 1, 2, · · ·
k!
where λ indicates the average number of successes per unit time or space.
(a) Show that E(X) = λ and E[X(X − 1)] = λ2 .

(b) Find the variance of X.
Problem 4.3
A geometric random variable with parameter p, 0 < p < 1 has a probability
mass function
p(n) = p(1 − p)n−1 , n = 1, 2, · · · .
(a) By differentiating the geometric series ∞ n 1
P
n=0 x = 1−x twice and using
x = 1 − p is each equation, show that
P∞ n−1 = p−2 and
P∞ n−2 = 2p−3 .
n=1 n(1 − p) n=1 n(n − 1)(1 − p)
(b) Show that E(X) = p1 and E[X(X − 1)] = 2p−2 (1 − p).
(c) Find the variance of X.
Problem 4.4
A normal random variable with parameters µ and σ 2 has a pdf
1 (x−µ)2
f (x) = √ e− 2σ2 , − ∞ < x < ∞.
2πσ

Show that E(X) = µ and Var(X) = σ 2 . Hint: E(Z) = E X−µ σ = 0 where
Z is the standard normal distribution with parameters (0,1).
Problem 4.5
An exponential random variable with parameter λ > 0 is a random variable
with pdf
λe−λx if x ≥ 0

f (x) =
0 if x < 0
1 2
(a) Show that E(X) = λ and E(X 2 ) = λ2
.
(b) Find Var(X).
Problem 4.6
A Gamma random variable with parameters α > 0 and θ > 0 has a pdf
(
1 α−1 e− xθ if x ≥ 0
θ α Γ(α) x
f (x) =
0 if x < 0
where Z ∞
e−y y α−1 dy = Γ(α) = αΓ(α − 1).
0
Show:
(a) E(X) = αθ
(b) V ar(X) = αθ2 .
Problem 4.7
Let X be a continuous random variable with pdf given by f (x) = 83 x2 for
0 ≤ x ≤ 2 and 0 otherwise.
Find the third raw moment of X.
Problem 4.8
x 120 122 124 150 167 245
1 1 1 1 1 1
p(x) 4 12 6 12 12 3
Find the fourth raw moment.
Problem 4.9
x 120 122 124 150 167 245
1 1 1 1 1 1
p(x) 4 12 6 12 12 3
Find the fifth central moment of X.

Problem 4.10
Compute the coefficient of skewness of a uniform random variable, X, on
[0, 1].
Problem 4.11
Let X be a random variable with density f (x) = e−x and 0 otherwise.
Find the coefficient of kurtosis.
Problem 4.12
x 120 122 124 150 167 245
1 1 1 1 1 1
p(x) 4 12 6 12 12 3
Find the coefficient of variation.
Problem 4.13
Let X be a continuous random variable with density function f (x) = Axb e−Cx
for x ≥ 0 and 0 otherwise. The parameters A, B, and C satisfy
1 1
A = R∞ , B ≥ − , C > 0.
0 xB e−Cx dx 2
Show that
B+r
E(X n ) = E(X n−1 ).
C
Problem 4.14
1 1
A = R∞ , B ≥ − , C > 0.
0 xB e−Cx dx 2
Find the first and second raw moments.
Problem 4.15
1 1
A = R∞ , B ≥ − , C > 0.
0 xB e−Cx dx 2
Find the coefficient of skewness.

Problem 4.16
1 1
A = R∞ , B ≥ − , C > 0.
0 xB e−Cx dx 2
Find the coefficient of kurtosis.
Problem 4.17
You are given: E(X) = 2, CV (X) = 2, and µ03 = 136. Calculate γ1 .
Problem 4.18
Let X be a random variable with pdf f (x) = 0.005x for 0 ≤ x ≤ 20 and 0
otherwise.
(a) Find the cdf of X.

(b) Find the mean and the variance of X.
(c) Find the coefficient of variation.
Problem 4.19
Let X be the Gamma random variable with pdf f (x) = 1 α−1 e− xθ
θα Γ(α) x for
x > 0 and 0 otherwise. Suppose E(X) = 8 and γ1 = 1.
Find the variance of X.
Problem 4.20
Let X be a Pareto random variable in one parameter and with a pdf
a
f (x) = xa+1 , x ≥ 1 and 0 otherwise.
a
(a) Show that E(X k ) = a−k for 0 < k < a.
(b) Find the coefficient of variation of X.
Problem 4.21
For the random variable X you are given:
(i) E(X) = 4
(ii) Var(X) = 64
(iii) E(X 3 ) = 15.
Calculate the skewness of X.

Problem 4.22
Let X be a Pareto random variable with two parameters α and θ, i.e., X
has the pdf
αθα
f (x) = , α > 1, θ > 0, x > 0
(x + θ)α+1
and 0 otherwise.
Calculate the mean and the variance of X.
Problem 4.23
Let X be a Pareto random variable with two parameters α and θ, i.e., X
has the pdf
αθα
f (x) = , α > 1, θ > 0, x > 0
(x + θ)α+1
and 0 otherwise.
Calculate the coefficient of variation.
Problem 4.24
Let X be the Gamma random variable with pdf f (x) = 1 α−1 e− xθ
θα Γ(α) x for
x > 0 and 0 otherwise. Suppose CV (X) = 1.
Determine γ1 .
Problem 4.25
You are given the following times of first claim for five randomly selected
auto insurance policies observed from time t = 0 :
1 2 3 4 5
Calculate the kurtosis of this sample.

5 EMPIRICAL MODELS, EXCESS AND LIMITED LOSS VARIABLES45
5 Empirical Models, Excess and Limited Loss vari-

ables
Empirical models are those that are based entirely on data. Consider a
statistical model that results in a sample Ω of size n. Data points in the
sample are assigned equal probability of n1 . Let X be a random variable
defined on Ω. We refer to this model as an empirical model.
Example 5.1 ‡
You are given the following for a sample of five observations from a bivariate
distribution:
(i)
x y
1 4
2 2
4 3
5 6
6 4
(ii) x = 3.6 and y = 3.8.

A is the covariance of the empirical distribution Fe as defined by these
five observations. B is the maximum possible covariance of an empirical
distribution with identical marginal distributions to Fe .
Determine B − A.
Solution.
We have
A =Cov(X, Y ) = E(XY ) − E(X)E(Y )

4 + 4 + 12 + 30 + 24
= − 3.6(3.8) = 1.12.
5
Now, since E(X) and E(Y ) are fixed, we want to create a new bivariate
distribution from the given one with maximum E(XY ). Clearly, this occurs
if largest values of X are paired with largest values of Y. Hence, the following
bivariate distribution has the same marginal distributions as the original
bivariate distribution:
x y
6 6
5 4
4 4
2 3
1 2
The covariance of this distribution is

36 + 20 + 16 + 6 + 2
B= − 3.6(3.8) = 2.32.
5
The final answer is B − A = 2.32 − 1.12 = 1.2
Example 5.2 ‡
You are given the following graph of cumulative distribution functions:
Determine the difference between the mean of the lognormal model and
the mean of the data.
Solution.
The empirical distribution is given by
p(10) =F (10) − F (0) = 0.20 − 0 = 0.20

p(100) =F (100) − F (10) = 0.60 − 0.2 = 0.4
p(1000) =F (1000) − F (100) = 1 − 0.6 = 0.4.
The mean of the data is
0.2(10) + 0.4(100) + 0.4(1000) = 442.
Now, from the graph we see that the 20th and 60th percentiles of the log-
normal distribution are 10 and 100 respectively. That is,

0.2 = Φ ln 10−µ
σ and 0.6 = Φ ln 100−µ
σ
Using the table of standard normal distribution, we find

ln 10−µ ln 100−µ
−0.84 = σ and 0.25 = σ
Solving this system, we find µ = 4.0771 and σ = 2.1125. Thus, the mean of
the lognormal distribution is (Table C)
2 2)
eµ+0.5σ = e4.0771+0.5(2.1125 = 549.18.
The final answer is 549.18 − 442 = 107.18
Example 5.3
In a fitness club monthly new memberships are recorded in the table below.
January February March April May June
100 102 84 84 100 100
July August September October November December

67 45 45 45 45 93
Use an empirical model to construct a discrete probability mass function for

X, the number of new memberships per month.
Solution.
The sample under consideration has 12 data points which are the months of
the year. For our empirical model, each data point is assigned a probability
1
of 12 . The pmf of the random variable X is given by:
x 45 67 84 93 100 102
1 1 1 1 1 1
p(x) 3 12 6 12 4 12
Excess Loss Random Variable

Consider an insurance policy with a deductible. The insurer’s interest are
usually in the losses that resulted in a payment, and the actual amount
paid by the insurance. The insurer pays the insuree the amount of the loss
that was in excess of the deductible2 . Any amount of losses that are be-
low the deductible are ignored by the insurance since they do not result in
an insurance payment being made. Hence, the insurer would be consider-
ing the conditional distribution of amount paid, given that a payment was
actually made. This is what is referred to as the excess loss random variable.
Let X be the random variable representing the amount of a single loss. In

insurance terms, X is known as the loss random variable or the severity
random variable. For a given threshold d such that Pr(X > d) > 0, the
random variable

P undefined, X ≤ d
Y = (X − d|X > d) =
X − d, X>d
is called the excess loss variable, the cost per payment, or the left
truncated and shifted variable. It stands for the amount paid by the
insurance which is also known as claim amount.
We can find the k th moment of the excess loss variable as follows. For a
continuous distribution with probability density function f (x) and cumula-
tive distribution function F (x), we have3
Z ∞
k k 1
eX (d) =E[(X − d) |X > d] = (x − d)k f (x)dx
Pr(X > d) d
Z ∞
1
= (x − d)k f (x)dx
1 − F (d) d
provided that the improper integral is convergent.
For a discrete distribution with probability density function p(x) and a cu-
mulative distribution function F (x), we have
1 X
ekX (d) = (xj − d)k p(xj )
1 − F (d)
xj >d
provided that the sum is convergent.
2
The deductible is referred to as ordinary deductible. Another type of deductible is
called franchise deductible and will be discussed in Section 32.
3
See (3.3)
When k = 1, the expected value
eX (d) = E(Y P ) = E(X − d|X > d)
is called the mean excess loss function. Other names used have been
mean residual life function and complete expectation of life.
If X denotes payment, then eX (d) stands for the expected amount paid
given that there has been a payment in excess of the deductible d. If X de-
notes age at death, then eX (d) stands for the expected future lifetime given
that the person is alive at age d.
Example 5.4
Show that for a continuous random variable X, we have
Z ∞ Z ∞
1 1
eX (d) = (1 − F (x))dx = S(x)dx.
1 − F (d) d S(d) d
Solution.
Using integration by parts with u = x − d and v 0 = f (x), we have
Z ∞
1
eX (d) = (x − d)f (x)dx
1 − F (d) d
(x − d)(1 − F (x)) ∞
Z ∞
1
=− + 1 − F (d) (1 − F (x))dx
1 − F (d) d d
Z ∞ Z ∞
1 1
= (1 − F (x))dx = S(x)dx.
1 − F (d) d S(d) d
Note that
Z ∞ Z ∞
0 ≤ xS(x) = x f (t)dt ≤ tf (t)dt =⇒ lim xS(x) = 0
x x x→∞
Example 5.5
Let X be an excess loss random variable with pdf given by f (x) = 31 (1 +
2x)e−x for x > 0 and 0 otherwise. Calculate the mean excess loss function
with deductible amount x.
Solution.
The cdf of X is given by
x
e−x (2x + 3)
Z x
1 −t 1 −t
F (x) = (1 + 2t)e dt = − e (2t + 3) = 1 −
0 3 3 0 3
where we used integration by parts. The mean excess loss function is
∞ e−x (2x+5)
e−t (2t + 3)
Z
1 3 2x + 5
eX (x) = e−x (2x+3)
1− dt = e−x (2x+3)
=
x 3 2x + 3
3 3
Example 5.6 ‡
For an industry-wide study of patients admitted to hospitals for treatment
of cardiovascular illness in 1998, you are given:
(i)
Duration In Days Number of Patients

Remaining Hospitalized
0 4,386,000
5 1,461,554
10 486,739
15 161,801
20 53,488
25 17,384
30 5,349
35 1,337
40 0
(ii) Discharges from the hospital are uniformly distributed between the du-
rations shown in the table.
Calculate the mean residual time remaining hospitalized, in days, for a pa-
tient who has been hospitalized for 21 days.
Solution.
Let X denote the number of days at the hospital measured from time 0. We
are asked to find E(X − 21|X > 21) which by Example 5.4 can be expressed
as Z ∞
SX (x)
E(X − 21|X > 21) = dx.
21 S X (21)
In life contingency theory, the assumption that discharges are uniform on a

given interval means that the graph of survival function is a linear function
on that interval. Hence, the graph of SSXX(21)
(x)
consists of line segments on the
intervals [0, 5], [5, 10], etc. Thus, E(X −21|X > 21) is just the area under the
graph from 21 to 40. The area under the graph is just the sum of areas of
the trapezoids with bases [21, 25], [25, 30], [30, 35], and [35, 40]. For instance,
the area under the first trapezoid is

1 SX (21) SX (25)
(25 − 21) + .
2 SX (21) SX (21)
Again, from the theory of life contingencies (see [3]), we have
`x
SX (x) =
`0
where `0 is the number of patients at the hospital at time 0 and `x is the
expected number of patients in the hospital at time x. Thus,
SX (x) `x
= .
SX (21) `21
By linear interpolation, we have

53488 − 17384
`21 = 53, 488 − (21 − 20) = 46, 267.2.
25 − 20
Now, we go back to finding the area of the first trapezoid mentioned above,
we find

1 SX (21) SX (25) 1 17384
(25 − 21) + = (4) 1 + = 2.751.
2 SX (21) SX (21) 2 46, 267.20
We repeat the same calculation with the remaining three trapezoids, we find
E(X − 21|X > 21) = 2.751 + 1.228 + 0.361 + 0.072 = 4.412
Example 5.7
Show that
FX (y + d) − FX (d)
FY P (y) = .
1 − FX (d)
Solution.
We have
Pr(d < X ≤ y + d)
FY P (y) =Pr(Y P ≤ y) = Pr(X − d ≤ y|X > d) =
Pr(X > d)
FX (y + d) − FX (d)
=
1 − FX (d)
Left-Censored and Shifted Random Variable

Note that in the excess loss situation, losses below or at the value d are not
recorded in any way, that is, the excess loss random variable is left-truncated
and it is shifted because of a number subtracted from X. However, when
small losses at or below d are recorded as 0 then we are led to a new random
variable which we call a left-censored and shifted random variable or the
cost per loss random variable. It is defined as

0, X≤d
Y L = (X − d)+ =
X − d, X > d.
The k th -moment of this random variable is given by

Z ∞
E[(X − d)k+ ] = (x − d)k f (x)dx
d
in the continuous case and

X
E[(X − d)k+ ] = (xj − d)k p(xj )
xj >d
in the discrete case. Note the relationship between the moments of Y P and
Y L given by
E[(X − d)k+ ] = ekX (d)[1 − F (d)] = ekX (d)S(d).
Setting k = 1 and using the formula for eX (d) we see that
Z ∞
E(Y L ) = S(x)dx.
d
This expected value is sometimes called the stop loss premium.4
We can think of the excess loss random variable as of a random variable

that exists only if a payment has been made. Alternatively, the left cen-
sored and shifted random variable is equal to zero every time a loss does not
produce payment.
Example 5.8
For a house insurance policy, the loss amount (expressed in thousands), in
the event of a fire, is being modeled by a distribution with density
3
f (x) = x(5 − x), 0 < x < 4.
56
4
See Section 39.
For a policy with a deductible amount of $1,000, calculate the expected

amount per loss.
Solution.
We first calculate the survival function:
Z x
3 3 5 2 1 3
S(x) = 1 − F (x) = 1 − t(5 − t)dt = 1 − x − x .
0 56 56 2 3
Thus,
Z 4
L 3 5 2 1 3
E(Y ) = 1− x − x dx = 1325.893
1 56 2 3
Note that Y L is a mixed random variable. The discrete part is represented

by Y L = 0 for X ≤ d where
pY L (0) = Pr(Y L = 0) = Pr(X ≤ d) = FX (d) = FX (d) − FX (d− ).
The continuous part of the distribution of Y L is given by
fY L (y) = fX (y + d)
for y > 0 since Y L = y for X = y + d.
Limited Loss Variable or Policy Limits

Many insurance policies are covered up to a certain limit which we refer
to as policy limit. Let’s say the limit is u. That is, the insurer covers all
losses up to u fully but pays u for losses greater than u. Thus, if X is a loss
random variable then the amount paid by the insurer is X ∧ u. We call X ∧ u
the limited loss variable and is defined by

X, X ≤ u
X ∧ u = min(X, u) =
u, X > u.
Notice that the distribution of X is censored on the right and that is why the
limit loss variable is also known as the right-censored random variable.
The expected value of the limited loss value is E(X ∧ u) and is called the
limited expected value.
For a discrete distribution with probability density function p(xj ) and a

cumulative distribution function F (xj ) for all relevant index values j, the
k th moment of the limited loss random variable is given by
X
E[(X ∧ u)k ] = xkj p(xj ) + uk [1 − F (u)].
xj ≤u
For a continuous distribution with probability density function f (x) and

cumulative distribution function F (x), the k th moment is given by
Z u Z ∞ Z u
E[(X∧u)k ] = xk f (x)dx+ uk f (x)dx = xk f (x)dx+uk [1−F (u)].
−∞ u −∞
Using integration by parts, we can derive an alternative formula for the
k th moment:
Z u
k
E[(X ∧ u) ] = xk f (x)dx + uk [1 − F (u)]
−∞
Z0 Z u
k
= x f (x)dx + xk f (x)dx + uk [1 − F (u)]
−∞ 0
0 Z 0
k
= x F (x) − kxk−1 F (x)dx

−∞ −∞
u Z u
k
− x S(x) + kxk−1 S(x)dx + uk S(u)

0 0
Z 0 Z u
k−1
=− kx F (x)dx + kxk−1 S(x)dx.
−∞ 0
Note that for x < 0 and k odd we have
Z x Z x
k k
t f (t)dt ≤ x f (t)dt = xk F (x) ≤ 0
−∞ −∞
so that
lim xk F (x) = 0.
x→−∞
A similar argument for x < 0 and k even.
In particular, for k = 1, we obtain
Z 0 Z u
E(X ∧ u) = − F (x)dx + S(x)dx.
−∞ 0
One can easily see that

0, X≤d X, X ≤ u
(X − d)+ + X ∧ u = + = X.
X − d, X > d. u, X > u.
That is, buying one policy with a deductible d and another one with a limit
d is equivalent to purchasing full cover.
Remark 5.1
Usually the random variable X is non-negative (for instance, when X rep-
resents a loss or a time until death), and the lower integration limit −∞ is
replaced by 0. Thus, for X ≥ 0, we whave
Z u
E(X ∧ u) = S(x)dx.
0
Example 5.9
A continuous random variable X has a pdf f (x) = 0.005x for 0 ≤ x ≤ 20
and 0 otherwise. Find the mean and the variance of X ∧ 10.
Solution.
The cdf of X is

Z x Z x  0, x<0
F (x) = 0.005tdt = 0.005tdt = 0.0025x2 , 0 ≤ x ≤ 20
−∞ 0 
1, x > 20.
Thus,
Z 10 Z 10
55
E(X ∧ 10) = [1 − F (x)]dx = [1 − 0.0025x2 ]dx = .
0 0 6
Now,
Z 10
2
E[(X ∧ 10) ] = x2 (0.005x)dx + 102 [1 − F (10)] = 87.5.
0
Finally,
2
55 125
Var(X ∧ 10) = 87.5 − =
6 36
Example 5.10 ‡
The unlimited severity distribution for claim amounts under an auto liability
insurance policy is given by the cumulative distribution:
F (x) = 1 − 0.8e−0.02x − 0.2e−0.001x , x ≥ 0.
The insurance policy pays amounts up to a limit of 1000 per claim. Calculate
the expected payment under this policy for one claim.
Solution.
We are asked to find the limited expected value E(X ∧ 1000). We have
Z 1000 Z 1000
E(X ∧ 1000) = S(x)dx = [0.8e−0.02x + 0.2e−0.001x ]dx = 166.4
0 0
Example 5.11 ‡
A health plan implements an incentive to physicians to control hospitaliza-
tion under which the physicians will be paid a bonus B equal to c times the
amount by which total hospital claims are under 400 (0 ≤ c ≤ 1).
The effect the incentive plan will have on underlying hospital claims is
modeled by assuming that the new total hospital claims will follow a two-
parameter Pareto distribution with α = 2 and θ = 300.
Suppose that E(B) = 100. Calculate the value of c.
Solution.
Let X denote the number of hospital claims. We are told that

c(400 − x), x < 400 400c − cx, x < 400
B= = = 400c−X∧400.
0, x ≥ 400 400c − 400c, x ≥ 400
Thus,
100 = E(B) = E[400c − X ∧ 400] = 400c − cE(X ∧ 400).
We are told that X has a Pareto distribution with parameters α = 2 and

θ = 300. Using Table C, we have
" α−1 #
θ θ
E(X ∧ u) = 1− .
α−1 u+θ
1200
With u = 400, we find E(X ∧ 400) = 7 . Finally,
1200
100 = 400c − c =⇒ c ≈ 0.44
7
Example 5.12
Show that, for X ≥ 0, we have
E(X) − E(X ∧ d)
eX (d) = .
S(d)
Solution.
We have
Z ∞ Z ∞
E(X) = xf (x)dx = −xS(x)|∞
0 + S(x)dx
0 0
Z ∞
= S(x)dx
0
R∞ R∞ Rd
d S(x)dx S(x)dx − 0 S(x)dx
eX (d) = = 0
S(x) S(d)
E(X) − E(X ∧ d)
=
S(d)
Example 5.13 ‡
The random variable for a loss, X, has the following characteristics:
x F (x) E(X ∧ x)
0 0.0 0
100 0.2 91
200 0.6 153
1000 1.0 331
Calculate the mean excess loss for a deductible of 100.
Solution.
We are asked to find
E(X) − E(X ∧ 100)
eX (100) = .
1 − FX (100)
The only term unknown in this formula is E(X). Now, Pr(X > 1000) =
1 − F (1000) = 0. This shows that X ≤ 1000 so that X ∧ 1000 = X. It
follows that E(X) = E(X ∧ 1000) = 331.
The mean excess loss is
E(X) − E(X ∧ 100) 331 − 91
eX (100) = = = 300
1 − FX (100) 1 − 0.2
Remark 5.2
Just as in the case of a deductible, the random variable Y = X ∧ u has
a mixed distribution with continuous part fY (y) = fX (y) for y < u and a
discrete part pY (u) = 1 − FX (u).
Practice Problems
Problem 5.1
Suppose that a policy has a deductible of $500. Complete the following
table.
Amount of loss 750 500 1200
Insurance payment
Problem 5.2
Referring to Example 5.3, find the cumulative distribution function of X.
Problem 5.3
Referring to Example 5.3, find the first and second raw moments of X.
Problem 5.4
Suppose you observe 8 claims with amounts
5 10 15 20 25 30 35 40
Calculate the empirical coefficient of variation.
Problem 5.5
Let X be uniform on the interval [0, 100]. Find eX (d) for d > 0.
Problem 5.6
Let X be uniform on [0, 100] and Y be uniform on [0, α]. Suppose that
eY (30) = eX (30) + 4.
Calculate the value of α.
Problem 5.7
Let X be the exponential random variable with mean λ. Its pdf is f (x) =
λe−λx for x > 0 and 0 otherwise.
Find the expected cost per payment (i.e., mean excess loss function).
Problem 5.8
For an automobile insurance policy, the loss amount (expressed in thou-
sands), in the event of an accident, is being modeled by a distribution with
density
3
f (x) = x(5 − x), 0 < x < 4
56
and 0 otherwise.
For a policy with a deductible amount of $2,500, calculate the expected

amount per loss.
Problem 5.9
1
The loss random variable X has an exponential distribution with mean λ
and an ordinary deductible is applied to all losses.
Find the expected cost per loss.
Problem 5.10
1
The loss random variable X has an exponential distribution with mean λ
and an ordinary deductible is applied to all losses.
Find the variance of the cost per loss random variable.
Problem 5.11
The loss random variable X has an exponential distribution with mean λ1
and an ordinary deductible is applied to all losses. The variance of the cost
per payment random variable (excess loss random variable) is 25,600.
Find λ.
Problem 5.12
per payment random variable (excess loss random variable) is 25,600. The
variance of the cost per loss random variable is 20,480.
Find the amount of the deductible d.
Problem 5.13
per payment random variable (excess loss random variable) is 25,600. The
variance of the cost per loss random variable is 20,480.
Find expected cost of loss.

Problem 5.14 φ
For the loss random variable with cdf F (x) = xθ , 0 < x < θ, and 0
otherwise, determine the mean residual lifetime eX (x).
Problem 5.15
Let X be a loss random variable with pdf f (x) = (1 + 2x2 )e−2x for x > 0
and 0 otherwise.
(a) Find the survival function S(x).

(b) Determine eX (x).
Problem 5.16
Show that
SX (y + d)
SY P (y) = .
SX (d)
Problem 5.17
Let X be a loss random variable with cdf F (x) = 1 − e−0.005x − 0.004e−0.005x
for x ≥ 0 and 0 otherwise.
(a) If an ordinary deductible of 100 is applied to each loss, find the pdf
of the per payment random variable Y P .
(b) Calculate the mean and the variance of the per payment random vari-
able.
Problem 5.18
A continuous random variable X has a pdf f (x) = 0.005x for 0 ≤ x ≤ 20
and 0 otherwise.
Find the mean and the variance of (X − 10)+ .
Problem 5.19 ‡
For a random loss X, you are given: Pr(X = 3) = Pr(X = 12) = 0.5 and
E[(X − d)+ ] = 3.
Calculate the value of d.
Problem 5.20 ‡
A loss, X, follows a 2-parameter Pareto distribution with α = 2 and unspec-
ified parameter θ. You are given:
5
E(X − 100|X > 100) = E(X − 50|X > 50).
3
Calculate E(X − 150|X > 150).
Problem 5.21 ‡
For an insurance:
(i) Losses can be 100, 200 or 300 with respective probabilities 0.2, 0.2, and
0.6.
(ii) The insurance has an ordinary deductible of 150 per loss.
(iii) Y P is the claim payment per payment random variable.
Calculate Var(Y P ).
Problem 5.22 ‡
For an insurance:
(i) Losses have density function

0.02x, 0 << 10
f (x) =
0, otherwise.
(ii) The insurance has an ordinary deductible of 4 per loss.

(iii) Y P is the claim payment per payment random variable.
Calculate E[Y P ].
Problem 5.23 ‡
The loss severity random variable X follows the exponential distribution
with mean 10,000.
Determine the coefficient of variation of the excess loss variable Y = max{(X−

30000, 0)}.
6 Median, Mode, Percentiles, and Quantiles

In addition to the information provided by the moments of a distribution,
some other metrics such as the median, the mode, the percentile, and the
quantile provide useful information.
Median of a Random Variable

In probability theory, median is described as the numerical value separat-
ing the higher half of a probability distribution, from the lower half. Thus,
the median of a discrete random variable X is the number M such that
Pr(X ≤ M ) ≥ 0.50 and Pr(X ≥ M ) ≥ 0.50.
Example 6.1
Given the pmf of a discrete random variable X.
x 0 1 2 3 4 5
p(x) 0.35 0.20 0.15 0.15 0.10 0.05
Find the median of X.
Solution.
Since Pr(X ≤ 1) = 0.55 and Pr(X ≥ 1) = 0.65, 1 is the median of X
In the case of a continuous random variable X, the median is the num-

ber M such that Pr(X ≤ M ) = Pr(X ≥ M ) = 0.5. Generally, M is found
by solving the equation F (M ) = 0.5 where F is the cdf of X.
Example 6.2
1
Let X be a continuous random variable with pdf f (x) = b−a for a < x < b
and 0 otherwise. Find the median of X.
Solution. R M dx
We must find a number M such that a b−a = 0.5. This leads to the
M −a
equation b−a = 0.5. Solving this equation we find M = a+b
2
Remark 6.1
A discrete random variable might have many medians. For example,
x let X be
the discrete random variable with pmf given by p(x) = 12 , x = 1, 2, · · ·
and 0 otherwise. Then any number 1 < M < 2 satisfies Pr(X ≤ M ) =
Pr(X ≥ M ) = 0.5.
6 MEDIAN, MODE, PERCENTILES, AND QUANTILES 63
Mode of a Random Variable

The mode is defined as the value that maximizes the probability mass func-
tion p(x) (discrete case) or the probability density function f (x) (continuous
case.) In the discrete case, the mode is the value that is most likely to be
sampled. In the continuous case, the mode is where f (x) is at its peak.
Example 6.3
1 x

Let X be the discrete random variable with pmf given by p(x) = 2 , x=
1, 2, · · · and 0 otherwise. Find the mode of X.
Solution.
The value of x that maximizes p(x) is x = 1. Thus, the mode of X is 1
Example 6.4
Let X be the continuous random variable with pdf given by f (x) = 0.75(1 −
x2 ) for −1 ≤ x ≤ 1 and 0 otherwise. Find the mode of X.
Solution.
The pdf is maximum for x = 0. Thus, the mode of X is 0
Percentiles and Quantiles

In statistics, a percentile is the value of a variable below which a certain per-
cent of observations5 fall. For example, if a score is in the 85th percentile,
it is higher than 85% of the other scores. For a random variable X and
0 < p < 1, the 100pth percentile (or the pth quantile) is the number x
such
Pr(X < x) ≤ p ≤ Pr(X ≤ x).
For a continuous random variable, this is the solution to the equation F (x) =
p. The 25th percentile is also known as the first quartile, the 50th percentile
as the median or second quartile, and the 75th percentile as the third quar-
tile.
Example 6.5
A loss random variable X has the density function
(
2.5(200)2.5
x3.5
x > 200
f (x) =
0 otherwise.
Calculate the difference between the 25th and 75th percentiles of X.

5
Another term for an ”observation” in this text is ”exposure”.
Solution.
First, the cdf is given by
x
2.5(200)2.5
Z
F (x) = dt.
200 t3.5
If Q1 is the 25th percentile then it satisfies the equation

1
F (Q1 ) =
4
or equivalently
3
1 − F (Q1 ) = .
4
This leads to
2.5 ∞
∞
2.5(200)2.5 200 2.5
Z
3 200
= dt = − = .

4 Q1 t3.5 t Q1
Q1
Solving for Q1 we find Q1 = 200(4/3)0.4 ≈ 224.4. Similarly, the third quartile

(i.e. 75th percentile) is given by Q3 = 348.2, The interquartile range
(i.e., the difference between the 25th and 75th percentiles) is Q3 − Q1 =
348.2 − 224.4 = 123.8
Example 6.6
1
Let X be the random variable with pdf f (x) = b−a for a < x < b and 0
otherwise. Find the pth quantile of X.
Solution.
We have x
x−a
Z
dt
p = Pr(X ≤ x) = = .
a b−a b−a
Solving this equation for x, we find x = a + (b − a)p
Example 6.7
What percentile is 0.63 quantile?
Solution.
0.63 quantile is 63rd percentile
Practice Problems
Problem 6.1
Using words, explain the meaning of F (1120) = 0.2 in terms of percentiles
and quantiles.
Problem 6.2
Let X be a discrete random variable with pmf p(n) = (n−1)(0.4)2 (0.6)n−2 , n ≥
2 and 0 otherwise.
Find the mode of X.

Problem 6.3
Let X be a continuous random variable with density function
1
λ 9 x(4 − x), 0 < x < 3
f (x) =
0, otherwise.
Find the mode of X.
Problem 6.4
Suppose the random variable X has pmf
1 2 n

p(n) = , n = 0, 1, 2, · · ·
3 3
and 0 otherwise.
Find the median and the 70th percentile.

Problem 6.5
The time in minutes between individuals joining the line at an Ottawa Post
Office is a random variable with the density function
−2x
2e , x ≥ 0
f (x) =
0, x < 0.
Find the median time between individuals joining the line and interpret your
answer.
Problem 6.6
Suppose the random variable X has pdf
−x
e , x≥0
f (x) =
0, otherwise.
Find the 100pth percentile.
Problem 6.7 ‡
An insurance company sells an auto insurance policy that covers losses in-
curred by a policyholder, subject to a deductible of 100 . Losses incurred
follow an exponential distribution with mean 300.
What is the 95th percentile of actual losses that exceed the deductible?
Problem 6.8
Let X be a randon variable with density function
λe−λx ,

x>0
f (x) =
0, otherwise.
Find λ if the median of X is 13 .
Problem 6.9
People are dispersed on a linear beach with a density function f (y) =
4y 3 , 0 < y < 1, and 0 elsewhere. An ice cream vendor wishes to locate
her cart at the median of the locations (where half of the people will be on
each side of her).
Where will she locate her cart?
Problem 6.10 ‡
An automobile insurance company issues a one-year policy with a deductible
of 500. The probability is 0.8 that the insured automobile has no accident
and 0.0 that the automobile has more than one accident. If there is an
accident, the loss before application of the deductible is exponentially dis-
tributed with mean 3000.
Calculate the 95th percentile of the insurance company payout on this policy.
Problem 6.11
Let Y be a continuous random variable with cumulative distribution function
y≤a

0,
F (y) = − 12 (y−a)2
1−e , otherwise
where a is a constant.
Find the 75th percentile of Y.

Problem 6.12
Find the pth quantile of the exponential distribution defined by the distri-
bution function F (x) = 1 − e−x for x ≥ 0 and 0 otherwise.
Problem 6.13
A continuous random variable has the pdf f (x) = e−|x| for x ∈ R.
Find the pth quantile of X.
Problem 6.14
Let X be a loss random variable with cdf
( α
θ
1 − θ+x , x≥0
F (x) =
0, x < 0.
The 10th percentile is θ − k. The 90th percentile is 3θ − 3k.
Problem 6.15
A random variable X follows a normal distribution with µ = 1 and σ 2 = 4.
Define a random variable Y = eX , then Y follows a lognormal distribution.
It is known that the 95th percentile of a standard normal distribution is
1.645.
Calculate the 95th percentile of Y.
Problem 6.16
4x
Let X be a random variable with density function f (x) = (1+x2 )3
for x > 0
and 0 otherwise.
Calculate the mode of X.
Problem 6.17
3 5000 4

Let X be a random variable with pdf f (x) = 5000 x for x > 0 and 0
otherwise.
Determine the median of X.

Problem 6.18
Let X be a random variable with cdf

 0, x<0
F (x) = x3
, 0≤x≤3
 27
1, x > 3.
Find the median of X.
Problem 6.19
Consider a sample of size 9 and observed data
45, 50, 50, 50, 60, 75, 80.120, 230.
Using this data as an empirical distribution, calculate the empirical mode.
Problem 6.20
3
A distribution has a pdf f (x) = x4
for x > 1 and 0 otherwise.
Calculate the 0.95th quantile of this distribution.

7 SUM OF RANDOM VARIABLES AND THE CENTRAL LIMIT THEOREM69
7 Sum of Random Variables and the Central Limit

Theorem
Random variables of the form
Sn = X1 + X2 + · · · + Xn
appear repeatedly in probability theory and applications. For example, in

the insurance context, Sn can represent the total claims paid on all policies
where Xi is the ith claim. Thus, it is useful to be able to determine proper-
ties of Sn .
For the expected value of Sn , we have
E(Sn ) = E(X1 ) + E(X2 ) + · · · + E(Xn−1 ) + E(Xn ).
A similar formula holds for the variance provided that the Xi0 s are indepen-
dent6 random variables. In this case,
Var(Sn ) = Var(X1 ) + Var(X2 ) + · · · + Var(Xn ).
Example 7.1 ‡
The random variables X1 , X2 , · · · , Xn are independent and identically dis-
tributed with probability density function
1 x
f (x) = e− θ .
θ
2
Determine E[X ].
Solution.
The random variable Xi has an exponential distribution with mean θ. Thus,
6
We say that X and Y are independent random variables if and only if for any two
sets of real numbers A and B we have
P (X ∈ A, Y ∈ B) = P (X ∈ A)P (Y ∈ B).
E(Xi ) = θ and Var(Xi ) = θ2 . Thus,
E(X1 ) + · · · + E(Xn )
E[X] = =θ
n
Var(X1 ) + · · · + Var(Xn )
Var[X] =
n2
θ 2
=
n
2
E[X ] =Var[X] + E[X]2
θ2

2 n+1
= +θ = θ2
n n
The central limit theorem reveals a fascinating property of the sum of inde-
pendent random variables. It states that the CDF of the sum converges to
the standard normal CDF as the number of terms grows without limit. This
theorem allows us to use the properties of the standard normal distribution
to obtain accurate estimates of probabilities associated with sums of random
variables.
Theorem 7.1
Let X1 , X2 , · · · be a sequence of independent and identically distributed
random variables, each with mean µ and variance σ 2 . Then,
√ Z a
n X1 + X2 + · · · + Xn 1 x2
P −µ ≤a → √ e− 2 dx
σ n 2π −∞
as n → ∞.
The Central Limit Theorem says that regardless of the underlying distribu-
tion
√ of the variables Xi , so long as they are independent, the distribution of
n X1 +X2 +···+Xn
σ n − µ converges to the same, normal, distribution.
Example 7.2
The weight of a typewriter has a mean of 20 pounds and a variance of 9
pounds. Consider a train that carries 200 of these typewriters. Estimate the
probability that the total weight of typewriters carried in the train exceeds
4050 pounds.
Solution.
Label the typewriters as Typewriter 1, Typewriter 2, etc. Let Xi be the
weight of Typewriter i. Thus,

200
! P200 !
X X
i=1 √i − 200(20) 4050 − 20(200)
P Xi > 4050 =P > √
i=1
3 200 3 200
≈P (Z > 1.179) = 1 − P (Z ≤ 1.179)
=1 − Φ(1.179) = 1 − 0.8810 = 0.119
where Φ is the CDF of the standard normal distribution
Example 7.3 ‡
In an analysis of healthcare data, ages have been rounded to the nearest
multiple of 5 years. The difference between the true age and the rounded
age is assumed to be uniformly distributed on the interval from −2.5 years to
2.5 years. The healthcare data are based on a random sample of 48 people.
What is the approximate probability that the mean of the rounded ages is
within 0.25 years of the mean of the true ages?
Solution.
Let X denote the difference between true and reported age. We are given X
is uniformly distributed on (−2.5, 2.5). That is, X has pdf f (x) = 1/5, −2.5 <
x < 2.5. It follows that E(X) = 0 and
2.5
x2
Z
2 2
σX = E(X ) = dx ≈ 2.083
−2.5 5
√
so that SD(X) = 2.083 ≈ 1.443.
Now X 48 the difference between the means of the true and rounded ages,
has a distribution that is approximately normal with mean 0 and standard
deviation 1.443
√
48
≈ 0.2083. Therefore,

1 1 −0.25 X48 0.25
P − ≤ X 48 ≤ =P ≤ ≤
4 4 0.2083 0.2083 0.2083
=P (−1.2 ≤ Z ≤ 1.2) = 2Φ(1.2) − 1
≈2(0.8849) − 1 = 0.77
Example 7.4
Let X1 , X2 , X3 , X4 be a random sample of size 4 from a normal distribution
with mean 2 and variance 10, and let X be the sample mean. Determine a
such that P (X ≤ a) = 0.90.
Solution.
2
The sample mean X is normal with mean µ = 2 and variance σn = 10
= 2.5,
√ 4
and standard deviation 2.5 ≈ 1.58, so

X −2 a−2 a−2
0.90 = P (X ≤ a) = P < =Φ .
1.58 1.58 1.58
a−2
Using Excel, we get 1.58 = 1.28, so a = 4.02
Practice Problems
Problem 7.1
A shipping agency ships boxes of booklets with each box containing 100
booklets. Suppose that the average weight of a booklet is 1 ounce and the
standard deviation is 0.05 ounces.
What is the probability that 1 box of booklets weighs more than 100.4
ounces?
Problem 7.2
In the National Hockey League, the standard deviation in the distribution
of players’ height is 2 inches. The heights of 25 players selected at random
were measured.
Estimate the probability that the average height of the players in this sample
is within 1 inch of the league average height.
Problem 7.3
A battery manufacturer claims that the lifespan of its batteries has a mean
of 54 hours and a standard deviation of 6 hours. A sample of 60 batteries
were tested.
What is the probability that the mean lifetime is less than 52 hours?
Problem 7.4
Roll a dice 10 times. Estimate the probability that the sum obtained is
between 30 and 40, inclusive.
Problem 7.5
Consider 10 independently random variables each uniformly distributed over
(0,1).
Estimate the probability that the sum of the variables exceeds 6.
Problem 7.6
The Chicago Cubs play 100 independent baseball games in a given season.
Suppose that the probability of winning a game in 0.8.
What’s the probability that they win at least 90?

Problem 7.7
An insurance company has 10,000 home policyholders. The average annual
claim per policyholder is found to be $240 with a standard deviation of $800.
Estimate the probability that the total annual claim is at least $2.7 mil-
lion.
Problem 7.8
A certain component is critical to the operation of a laptop and must be
replaced immediately upon failure. It is known that the average life of this
type of component is 100 hours and its standard deviation is 30 hours.
Estimate the number of such components that must be available in stock so

that the system remains in continual operation for the next 2000 hours with
probability of at least 0.95?
Problem 7.9
An instructor found that the average student score on class exams is 74 and
the standard deviation is 14. This instructor gives two exams: One to a
class of size 25 and the other to a class of 64.
Using the Central Limit Theorem, estimate the probability that the average
test score in the class of size 25 is at least 80.
Problem 7.10
The Salvation Army received 2025 in contributions. Assuming the contri-
butions to be independent and identically distributed with mean 3125 and
standard deviation 250.
Estimate the 90th percentile for the distribution of the total contributions
received.
Problem 7.11 ‡
An insurance company issues 1250 vision care insurance policies. The num-
ber of claims filed by a policyholder under a vision care insurance policy
during one year is a Poisson random variable with mean 2. Assume the
numbers of claims filed by distinct policyholders are independent of one an-
other.
What is the approximate probability that there is a total of between 2450

and 2600 claims during a one-year period?
Problem 7.12
A battery manufacturer finds that the lifetime of a battery, expressed in
months, follows a normal distribution with mean 3 and standard deviation
1 . Suppose that you want to buy a number of these batteries with the
intention of replacing them successively into your radio as they burn out.
Assuming that the batteries’ lifetimes are independent, what is the small-
est number of batteries to be purchased so that the succession of batteries
keeps your radio working for at least 40 months with probability exceeding
0.9772?
Problem 7.13
The total claim amount for a home insurance policy has a pdf
x
1 − 1000

f (x) = 1000 e x>0
0 otherwise.
An actuary sets the premium for the policy at 100 over the expected total
claim amount.
If 100 policies are sold, estimate the probability that the insurance com-
pany will have claims exceeding the premiums collected.
Problem 7.14 ‡
A city has just added 100 new female recruits to its police force. The city
will provide a pension to each new hire who remains with the force until
retirement. In addition, if the new hire is married at the time of her re-
tirement, a second pension will be provided for her husband. A consulting
actuary makes the following assumptions:
(i) Each new recruit has a 0.4 probability of remaining with the police force
until retirement.
(ii) Given that a new recruit reaches retirement with the police force, the
probability that she is not married at the time of retirement is 0.25.
(iii) The number of pensions that the city will provide on behalf of each new
hire is independent of the number of pensions it will provide on behalf of
any other new hire.
Determine the probability that the city will provide at most 90 pensions
to the 100 new hires and their husbands.
Problem 7.15
The amount of an individual claim has a two-parameter Pareto distribution
with θ = 8000 and α = 9. Consider a sample of 500 claims.
Estimate the probability that the total sum of the claims is at least 550,000.
Problem 7.16
Suppose that the current profit from selling a share of a stock is found to
follow a uniform distribution on [−45, 72].
Using the central limit theorem, approximate the probability of making a

profit from the sale of 55 stocks.
Problem 7.17
The severities of individual claims have the Pareto distribution with param-
eters α = 38 and θ = 8000.
Use the central limit theorem to approximate the probability that the sum
of 100 independent claims will exceed 600,000.
Problem 7.18 ‡
Let X and Y be the number of hours that a randomly selected person
watches movies and sporting events, respectively, during a three-month pe-
riod. The following information is known about X and Y :
E(X) = 50
E(Y) = 20
Var(X) = 50
Var(Y) = 30
Cov (X,Y) = 10
One hundred people are randomly selected and observed for these three
months. Let T be the total number of hours that these one hundred people
watch movies or sporting events during this three-month period.
Approximate the value of P (T < 7100).
Problem 7.19 ‡
Automobile losses reported to an insurance company are independent and
uniformly distributed between 0 and 20,000. The company covers each such
loss subject to a deductible of 5,000.
Calculate the probability that the total payout on 200 reported losses is
between 1,000,000 and 1,200,000.
Problem 7.20 ‡
For Company A there is a 60% chance that no claim is made during the
coming year. If one or more claims are made, the total claim amount is
normally distributed with mean 10,000 and standard deviation 2, 000.
For Company B there is a 70% chance that no claim is made during the
coming year. If one or more claims are made, the total claim amount is
normally distributed with mean 9,000 and standard deviation 2,000.
Assume that the total claim amounts of the two companies are independent.
What is the probability that, in the coming year, Company B’s total claim
amount will exceed Company A’s total claim amount?
8 Moment Generating Functions and Probability

Generating Functions
A useful way to analyze the sum of independent random variables is to trans-
form the PDF or PMF of each random variable to a moment generating
function, abbreviated mgf.
The moment generating function of a continuous random variable X with a

density function f (x) is denoted by MX (t) and is given by
Z ∞
MX (t) = E[etx ] = etx f (x)dx.
−∞
The moment generating function of a discrete random variable X with a

probability mass function p(x) is denoted by MX (t) and is given by
X
MX (t) = E[etx ] = etx p(x).
x∈support(X)
Example 8.1
Calculate the moment generating function for the exponential distribution
with parameter λ, i.e. f (x) = λe−λx for x > 0 and 0 otherwise.
Solution.
We have
∞ ∞
λ −x(λ−t) ∞
Z Z
tx −λx −x(λ−t) λ
MX (t) = e λe dx = λe dx = − e = λ − t, t < λ
0 0 λ−t 0
Example 8.2
Let X be a discrete random variable with pmf given by the following table
x 1 2 3 4 5
p(x) 0.15 0.20 0.40 0.15 0.10
and 0 otherwise. Calculate MX (t).
Solution.
We have
MX (t) = 0.15et + 0.20e2t + 0.40e3t + 0.15e4t + 0.10e5t
As the name suggests, the moment generating function can be used to gen-
erate moments E(X n ) for n = 1, 2, · · · . The next result shows how to use
the moment generating function to calculate moments.
8 MOMENT GENERATING FUNCTIONS AND PROBABILITY GENERATING FUNCTIONS79
Theorem 8.1
For any random variable X, we have
dn

E(X n ) = MX
n (0) where M n (0) =
X dtn MX (t) t=0 .

Example 8.3
Let X be a binomial random variable with parameters n and p. Find the
expected value and the variance of X using moment generating functions.
Solution.
We can write
n
X n
X
tk k n−k
MX (t) = e C(n, k)p (1−p) = C(n, k)(pet )k (1−p)n−k = (pet +1−p)n .
k=0 k=0
Differentiating yields
d d
MX (t) = npet (pet + 1 − p)n−1 =⇒ E(X) = MX (t) |t=0 = np.
dt dt
To find E(X 2 ), we differentiate a second time to obtain
d2
MX (t) = n(n − 1)p2 e2t (pet + 1 − p)n−2 + npet (pet + 1 − p)n−1 .
dt2
Evaluating at t = 0 we find
00
E(X 2 ) = MX (0) = n(n − 1)p2 + np.
Observe that this implies the variance of X is
V ar(X) = E(X 2 ) − (E(X))2 = n(n − 1)p2 + np − n2 p2 = np(1 − p)
Example 8.4
Let X be a Poisson random variable with parameter λ. Find the expected
value and the variance of X using moment generating functions.
Solution.
We can write
∞ tn −λ n ∞ tn n ∞
X e e λ −λ
X e λ −λ
X (λet )n t t −1)
MX (t) = =e =e = e−λ eλe = eλ(e .
n! n! n!
n=0 n=0 n=0
Differentiating for the first time we find

0 t −1) 0
MX (t) = λet eλ(e =⇒ E(X) = MX (0) = λ.
Differentiating a second time we find

00 t −1) t −1) 00
MX (t) = (λet )2 eλ(e + λet eλ(e =⇒ E(X 2 ) = MX (0) = λ2 + λ.
The variance is then
V ar(X) = E(X 2 ) − (E(X))2 = λ
Example 8.5
Let X be a normal random variable with parameters µ and σ 2 . Find the
expected value and the variance of X using moment generating functions.
Solution.
First we find the moment of a standard normal random variable with pa-
rameters 0 and 1. We can write
Z ∞ Z ∞
(z 2 − 2tz)

tZ 1 2
tz − z2 1
MZ (t) =E(e ) = √ e e dz = √ exp − dz
2π −∞ 2π −∞ 2
Z ∞ Z ∞
(z − t)2 t2

1 t2 1 (z−t)2 t2
=√ exp − + dz = e 2 √ e− 2 dz = e 2
2π −∞ 2 2 2π −∞
Now, since X = µ + σZ we have
MX (t) =E(etX ) = E(etµ+tσZ ) = E(etµ etσZ ) = etµ E(etσZ )

2 2
tµ
2 2
tµ σ 2t σ t
=e MZ (tσ) = e e = exp + µt
2
By differentiation we obtain
σ 2 t2

0
MX (t) = (µ + tσ 2 )exp + µt
2
and
σ 2 t2
2 2
00 2 2 2 σ t
MX (t) = (µ + tσ ) exp + µt + σ exp + µt
2 2
and thus
0 (0) = µ and E(X 2 ) = M 00 (0) = µ2 + σ 2
E(X) = MX X
The variance of X is
V ar(X) = E(X 2 ) − (E(X))2 = σ 2

Moment generating functions are also useful in establishing the distribu-

tion of sums of independent random variables. Suppose X1 , X2 , · · · , XN are
independent random variables. Then, the moment generating function of
Y = X1 + · · · + XN is
N
Y N
Y
t(X1 +X2 +···+XN ) X1 t XN t Xk t
MY (t) = E(e ) = E(e ···e )= E(e )= MXk (t).
k=1 k=1
Another important property is that the moment generating function uniquely

determines the distribution. That is, if random variables X and Y both have
moment generating functions MX (t) and MY (t) that exist in some neigh-
borhood of zero and if MX (t) = MY (t) for all t in this neighborhood, then
X and Y have the same distributions.
Example 8.6
If X and Y are independent binomial random variables with parameters
(n, p) and (m, p), respectively, what is the pmf of X + Y ?
Solution.
We have
MX+Y (t) = MX (t)MY (t) = (pet + 1 − p)n (pet + 1 − p)m = (pet + 1 − p)n+m .
Since (pet + 1 − p)n+m is the moment generating function of a binomial

random variable having parameters m + n and p, X + Y is a binomial
random variable with this same pmf
Example 8.7
If X and Y are independent Poisson random variables with parameters λ1
and λ2 , respectively, what is the pmf of X + Y ?
Solution.
We have
t −1) t −1) t −1)
MX+Y (t) = MX (t)MY (t) = eλ1 (e eλ2 (e = e(λ1 +λ2 )(e .
t
Since e(λ1 +λ2 )(e −1) is the moment generating function of a Poisson random
variable having parameter λ1 + λ2 , X + Y is a Poisson random variable with
this same pmf
Probability Generating Function

Another useful tool for dealing with the distribution of a sum of discrete ran-
dom variables is the probability generating function or the z−transform
of the probability mass function, abbreviated pgf. For a discrete random
variable X, we define the probability generating function by
X
PX (t) = E(tx ) = tx p(x), ∀t ∈ R for which the sum converges.
x∈Support(X)
Note that PX (t) = MX [ex ln t ] and MX (t) = PX (et ). The pgf transforms a
sum into a product and enables it to be handled much more easily: Let
X1 , X2 , · · · , Xn be independent random variables and Sn = X1 + X2 + · · · +
Xn . It can be shown that
PSn (t) = PX1 (t)PX2 (t) · · · PXn (t).
Example 8.8
Find the pgf of the Poisson distribution of parameter λ.
Solution.
λx e−λ
Recall that the Poisson random variable has a pmf p(x) = x! . Hence,
∞ x e−λ ∞
X
xλ −λ
X (λt)x
PX (t) = t =e = e−λ eλt = eλ(t−1)
x! x!
x=0 x=0
The probability generating function gets its name because the power series
can be expanded and differentiated to reveal the individual probabilities.
Thus, given only the pgf PX (t) = E(tx ), we can recover all probabilities
Pr(X = x).
It can be shown that
1 dn

p(n) = PX (t) .
n! dtn
t=0
Example 8.9
Let X be a discrete random variable with pgf PX (t) = 5t (2 + 3t2 ). Find the
distribution of X.
Solution. 00 (0)
PX
We have, p(0) = PX (0) = 0; p(1) = PX0 (0) = 2
5 ; p(2) = 2! = 0, p(3) =
P X 000 (0)
3! = 35 ; and p(n) = 0, ∀n ≥ 4
Theorem 8.2
For any discrete random variable X, we have
dk

E[X(X − 1)(X − 2) · · · (X − k + 1)] = k PX (t) .
dt t=1
Example 8.10
Let X be a Poisson random variable with parameter λ. Find the mean and
the variance using probability generating functions.
Solution.
We know that the pgf of X is PX (t) = eλ(t−1) . We have
E(X) =PX0 (1) = λ

E[X(X − 1)] =λ2
E(X 2 ) =λ2 + λ
Var(X) =λ
Like moment generating functions, the probability generating function uniquely

determines the distribution. That is, if we can show that two random vari-
ables have the same pgf in some interval containing 0, then we have shown
that the two random variables have the same distribution.
Example 8.11
Let X be a Poisson random variable with parameter λ and Y is Poisson
with parameter µ. Find the distribution of X + Y, assuming X and Y are
independent.
Solution.
We have
PX+Y (t) = PX (t)PY (t) = eλ(t−1) eµ(t−1) = e(λ+µ)(t−1) .
This is the pgf of a Poisson random variable with parameter λ + µ. So, by

the uniqueness of pgfs, X + Y is a Poisson random variable with parameter
λ+µ
Practice Problems
Problem 8.1
Let X be an exponential random variable with parameter λ.
Find the expected value and the variance of X using moment generating
functions.
Problem 8.2
Let X and Y be independent normal random variables with parameters
(µ1 , σ12 ) and (µ2 , σ22 ), respectively.
Find the distribution of X + Y.
Problem 8.3 ‡
Let X and Y be identically distributed independent random variables such
that the moment generating function of X + Y is
M (t) = 0.09e−2t + 0.24e−t + 0.34 + 0.24et + 0.09e2t , − ∞ < t < ∞.
Calculate Pr(X ≤ 0).
Problem 8.4 ‡
The value of a piece of factory equipment after three years of use is 100(0.5)X
where X is a random variable having moment generating function
1
MX (t) = 1−2t for t < 21 .
Calculate the expected value of this piece of equipment after three years of
use.
Problem 8.5
Let X and Y be two independent random variables with moment generating
functions
2 +2t 2 +t
MX (t) = et and MY (t) = e3t .
Determine the moment generating function of X + 2Y.
Problem 8.6
The random variable X has an exponential distribution with parameter b.
It is found that MX (−b2 ) = 0.2.
Find b.
Problem 8.7
If the moment generating function for the random variable X is MX (t) =
1 3
t+1 , find E[(X − 2) ].
Problem 8.8
Suppose a random variable X has moment generating function
9
2 + et

MX (t) = .
3
Find the variance of X.
Problem 8.9
A random variable X has the moment generating function
1
MX (t) = .
(1 − 2500t)4
Determine the standard deviation of X.
Problem 8.10 ‡
A company insures homes in three cities, J, K, and L . Since sufficient dis-
tance separates the cities, it is reasonable to assume that the losses occurring
in these cities are independent.
The moment generating functions for the loss distributions of the cities are:
MJ (t) =(1 − 2t)−3
MK (t) =(1 − 2t)−2.5
MJ (t) =(1 − 2t)−4.5
Let X represent the combined losses from the three cities.
Calculate E(X 3 ).
Problem 8.11
Let X be a binomial random variable with pmf p(k) = C(n, k)pk (1 − p)n−k .
Find the pgf of X.
Problem 8.12
Let X be a geometric random variable with pmf p(n) = p(1 − p)n−1 , n =
1, 2, · · · , where 0 < p < 1.
Find the pgf of X.

Problem 8.13
Let X be a random variable with pgf PX (t) = eλ(t−1) . True or false: X is a
Poisson random variable with parameter λ.
Problem 8.14
Let X be a random variable and Y = a + bX. Express PY (t) in terms of
PX (t).
Problem 8.15
Let X have the distribution of a geometric random variable with parameter
p. That is, p(x) = p(1 − p)x−1 , x = 1, 2, 3, · · · .
Find the mean and the variance of X using probability generating func-
tions.
Problem 8.16
You are given a sample of size 4 with observed data
2 2 3 5 8
Using empirical distribution framework, calculate the probability generating

function.
Problem 8.17
Let X be a discrete random variable with the pmf given below
7
x −2 3 π 2
1 1 1 3
p(x) 3 6 8 8
and 0 otherwise. Find the probability generating function PX (t).
Problem 8.18
1
Suppose p(n) = 2n−2 , n = 3, 4, 5, · · · and 0 otherwise. Find the probability
generating function PX (t).
Problem 8.19
1 1 2 4

Suppose PX (t) = t 3t + 3 . Find the pmf of X.
Problem 8.20
Let X be a random variable with probability generating function PX (t) =
t(1+t)
2(3−2t) .
Using PX (t), find E(X) and Var(X).

Tail Weight of a Distribution
The (right-)tail of a distribution is the portion of the distribution corre-

sponding to large values of the random variable. Alternatively, we can define
the tail of a random variable X as the interval (x, ∞) with probability
Z ∞
Pr(X > x) = SX (x) = 1 − FX (x) = fX (x)dx
x
where SX (x) is the survival function of X.
A distribution is said to be a heavy-tailed distribution if it significantly

puts more probability on larger values of the random variable. We also say
that the distribution has a larger tail weight. In contrast, a distribution that
puts less and less probability for larger values of the random variable is said
to be light-tailed distribution. According to [1], there are four ways to
look for indication that a distribution is heavy-tailed. The purpose of this
chapter is to discuss these various ways.
87
88 TAIL WEIGHT OF A DISTRIBUTION
9 Tail Weight Measures: Moments and the Speed

of Decay of S(x)
There are four ways of measuring heavy-tailed distributions as suggested by
[1]:
• Existence of non-central moments.
• The speed for which the survival function decays to zero.
• The hazard rate function.
• The mean excess loss function.
In this section we cover the first two and the last two will be covered in the
next section.
The Existence of Moments

A distribution fX (x) is said to be light-tailed if E(xk ) < ∞ for all k > 0.
The distribution fX (x) is said to be heavy-tailed if either E(xk ) does not
exist for all k > 0 or the moments exist only up to a certain value of a
positive integer k.
Example 9.1
Show that the exponential distribution with parameter λ > 0 is light-tailed
according to the above definition. Refer to Table C.
Solution.
Using Table C, for all positive integers k, we have
Γ(k + 1)
E(X k ) = .
λk
Hence, the exponential distribution is light-tailed
Example 9.2
Show that the Pareto distribution with parameters α and θ is heavy-tailed.
Refer to Table C.
Solution.
Using Table C, we have
θk Γ(k + 1)Γ(α − k)
E(X k ) =
Γ(α)
provided that −1 < k < α. Since the moments are not finite for all positive
k, the Pareto distribution is heavy-tailed
9 TAIL WEIGHT MEASURES: MOMENTS AND THE SPEED OF DECAY OF S(X)89
Example 9.3
Let X be a continuous random variable with pdf fX (x) defined for x > 0 and
0 otherwise. Suppose that there is a constant M > 0 such that fX (x) = xCn
for all x ≥ M and 0 otherwise, where n > 1 and C = Mn−1 1−n . Show that X
has a heavy-tailed distribution.
Solution.
We have
Z M Z ∞
k k
E(X ) = x fX (x)dx + C xk−n dx
0 M
Z M ∞
k xk−n+1
= x fX (x)dx + C
0 k − n + 1 M
=∞
for all k > n − 1
Classification Based on the Speed of Decay of the Survival Function

The survival function SX (x) = P (X > x) captures the probability of the
tail of a distribution. Recall that SX (x) decreases to 0 as x → ∞. The
question is how fast the function decays to zero. If the survival function of a
distribution decays slowly to zero (equivalently the cdf goes slowly to one),
it is another indication that the distribution is heavy-tailed. In contrast,
when the survival function decays to 0 very rapidly then this is indication
that the distribution is light-tailed.
Next, we consider comparing the tail weights of two distributions with the
same mean. This is done by comparing the survival functions of the two
distributions. Algebraically, we compute the ratio of the tail probabilities
or the survival functions which we will refer to as the relative tail weight:
0 (x)
−SX
SX (x) fX (x)
lim = lim = lim ≥ 0.
x→∞ SY (x) x→∞ −S 0 (x) x→∞ fY (x)
Y
Note that in the middle limit we used L’Hôpital’s rule since limx→∞ SX (x) =
limx→∞ SY (x) = 0.
Now, if the above limit is 0, then this happens only when the numerator
is 0 and the denominator is positive. In this case, we say that the distri-
bution of X has lighter tail than Y. If the limit is finite positive number
then we say that the distributions have similar or proportional tails. If the
limit diverges to infinity, then more probabilities on large values of X are

assigned to the numerator, In this case, we say that the distribution X has
heavier tail than the distribution Y.
Example 9.4
Compare the tail weight of the inverse Pareto distribution with pdf fX (x) =
θ2
τ θ1 xτ −1 θ2α e− x
(x+θ1 )τ +1
with the inverse Gamma distribution with pdf fY (x) = xα+1 Γ(α)
where θ1 , θ2 , τ > 0 and α > 1.
Solution.
We have
fX (x) τ θ1 xτ −1 xα+1 Γ(α)
lim = lim ·
x→∞ fY (x) x→∞ (x + θ1 )τ +1 θ2
θ2α e− x
τ +1
τ θ1 Γ(α) θ2 x
= lim e x xα−1
x→∞ θ2α x + θ1
τ θ1 Γ(α) 0
= · e · ∞ = ∞.
θ2α
Thus, X has a heavier tail than Y
Example 9.5
LetX be the exponential distribution with survival function SX = e−x for
x ≥ 0 and 0 otherwise, and Y be the distribution with survival function
SY (x) = x1 for x ≥ 1 and 0 otherwise. Compare the tail weight of these
distributions.
Solution.
We have
SX (x)
lim = lim xe−x = 0.
x→∞ SY (x) x→∞
Hence, X has a lighter tail than Y
Practice Problems
Problem 9.1
Let X be a random variable with pdf fX (x) = Cxn e−bx for x > 0 and 0
R ∞ −1
otherwise, where b, n > 0 and C = 0 xn e−bx dx .
Show that X has a light tail distribution.

Problem 9.2 R∞
Suppose X has a heavy-tailed distribution. Let t > 0. Show that N etx fX (x)dx =
∞ for some N = N (t) > 0.
Problem 9.3
Suppose X has a heavy-tailed distribution. Show that MX (t) = ∞ for all
t > 0.
Problem 9.4
Determine whether the Γ distribution with parameters α > 0 and θ > 0 is
light-tailed or heavy-tailed. Refer to Table C.
Problem 9.5
Let X be the inverse Weibull random variables with parameters θ and τ.
Determine whether the distribution is light-tailed or heavy-tailed. Refer

to Table C.
Problem 9.6
αθ1α
Compare the tail weight of the Pareto distribution with pdf fX (x) = (x+θ1 )α+1
− x
xτ −1 e θ2
with the Gamma distribution with pdf fY (x) = θ2τ Γ(τ ) where θ1 , θ2 , τ > 0
and α > 1.
Problem 9.7
Compare the tail weight of the Weibull distribution with pdf fX (x) =
τ τ
τ x τ −( xθ ) τ θ τ −( xθ )

x θ e and the inverse Weibull distribution with pdf fY (x) = x x e
where τ, θ > 0.
Problem 9.8
2
Let X be a random variable with pdf fX (x) = π(1+x 2 ) for x > 0 and 0
α
otherwise. Let Y be a random variable with pdf fY (x) = (1+x) α+1 for x > 0
and 0 otherwise, where α > 1.
Compare the tail weight of these distributions.

Problem 9.9
2
Let X be a random variable with pdf fX (x) = π(1+x2 )
for x > 0 and 0
1
otherwise. Let Y be a random variable with pdf fY (x) = (1+x)2
for x > 0
and 0 otherwise.
Problem 9.10
2
Let X be a random variable with pdf fX (x) = π(1+x 2 ) for x > 0 and 0
α
otherwise. Let Y be a random variable with pdf fY (x) = (1+x) α+1 for x > 0
and 0 otherwise, where 0 < α < 1.
Problem 9.11
The distribution of X has the survival function
θxγ
SX (x) = 1 − , θ, γ > 0.
1 + θxγ
and 0 otherwise. The distribution of Y has pdf
x
xγ−1 e− θ
SY (x) = γ .
θ Γ(γ)
and 0 otherwise.
Compare the tail behavior of these distributions.
Problem 9.12
Using the criterion of existence of moments, complete the following. Refer
to Table C.
Distribution Heavy-Tail Light-Tail
Weibull
Inverse Pareto
Normal
Loglogistic
Problem 9.13
to Table C.

Paralogistic
Lognormal
Inverse Gamma
Inverse Gaussian
Problem 9.14
to Table C.

Inverse Paralogistic
Inverse Exponential
Problem 9.15
Show that the Loglogistic distribution has a heavier tail than the Gamma
distribution.
Problem 9.16
Show that the Paraloglogistic distribution has a heavier tail than the Log-
normal distribution.
Problem 9.17
Show that the inverse exponential distribution has a heavier tail than the
exponential distribution.
Problem 9.18
SX (x)
Let X and Y have similar (proportional) right tails and limx→∞ SY (x) = c.
Which of the following is a possible value of c?
(i) c = ∞ (ii) c = 0 (c) c > 0.
Problem 9.19
Let X be a Pareto distribution with parameters α = 4 and θ = 340. Let Y
be a Pareto distribution with parameters α = 6 and θ = 340.
Which of these has a heavier right tail relative to the other?
Problem 9.20
You are given the right-tails of the survival functions of three distributions
X, Y and Z. Order these distributions according to tail weight.

10 TAIL WEIGHT MEASURES: HAZARD RATE FUNCTION AND MEAN EXCESS LOSS FUNCTION95
10 Tail Weight Measures: Hazard Rate Function

and Mean Excess Loss Function
In this section we classify the tail weight of a distribution based on the haz-
ard rate function and the mean excess loss function.
Classification Based on the Hazard Rate Function

Another way to classify the tail weight of a distribution is by using the
hazard rate function:
f (x) F 0 (x) d S 0 (x)
h(x) = = = − [ln S(x)] = − .
S(x) 1 − F (x) dx S(x)
By the existence of moments, the Pareto distribution is considered heavy-
tailed. Its hazard rate function is
f (x) α
h(x) = = .
S(x) x+θ
Note that h0 (x) = − (x+θ)
α
2 < 0 so that h(x) is nonincreasing. Thus, it
makes sense to say that a distribution is considered to be heavy-tailed if the
hazard rate function is nonincreasing. Likewise, the random variable X with
pdf f (x) = xe−x for x > 0 and 0 otherwise has a light-tailed distribution
according to the existence of moments (See Problem 9.1). Its hazrad function
x
is h(x) = x+1 which is a nondecreasing function. Hence, a nondecreasing
hazard rate function is an indication of a light-tailed distribution.
Example 10.1
Let X be a random variable with survival function f (x) = x12 if x ≥ 1 and
0 otherwise. Based on the hazard rate function of the distribution, decide
whether the distribution is heavy-tailed or light-tailed.
Solution.
The hazard rate function is
S 0 (x) − 23 2
h(x) = − = − 1x = .
S(x) x2
x
Hence, for x ≥ 1,
2
h0 (x) = −<0
x2
which shows that h(x) is nonincreasing. We conclude that the distribution
of X is heavy-tailed
Remark 10.1
Under this definition, a constant hazard function can be called both non-
increasing and nondecreasing. We will refer to distributions with constant
hazard function as medium-tailed distribution. Thus, the exponential
random variable which was classified as light-tailed in Example 9.1, will be
referred to as a medium-tailed distribution.
The next result provides a criterion for testing tail weight based on the
probability density function.
Theorem 10.1
If for a fixed y ≥ 0, the function f (x+y)
f (x) is nonincreasing (resp. nondecreas-
ing) in x then the hazard rate function is nondecreasing (resp. nonincreas-
ing).
Proof.
We have
R∞ R∞ ∞
f (t)dt f (x + y)dy
Z
−1 x 0 f (x + y)
[h(x)] = = = dy.
f (x) f (x) 0 f (x)
f (x+y)
Thus, if f (x) is nondecreasing in x for a fixed y, then h(x) is a nonincreas-
f (x+y)
ing in x. Likewise, if f (x) is nonincreasing in x for a fixed y, then h(x) is
a nondecreasing in x
Example 10.2
Using the above theorem, show that the Gamma distribution with parame-
ters θ > 0 and 0 < α < 1 is heavy-tailed.
Solution.
We have
f (x + y) y α−1 − y
= 1+ e θ
f (x) x
and
d f (x + y) y(1 − α) y α−2 − y
= 1 + e θ >0
dx f (x) x2 x
for 0 < α < 1. Thus, the hazard rate function is nonincreasing and the
distribution is heavy-tailed
Next, the hazard rate function can be used to compare the tail weight of two
distributions. For example, if X and Y are two distributions with increasing

(resp. decreasing) hazard rate functions, the distribution of X has a lighter
(resp. heavier) tail than the distribution of Y if hX (x) is increasing (resp.
decreasing) at a faster rate than hY (x) for a large value of the argument.
Example 10.3
Let X be the Pareto distribution with α = 2 and θ = 150 and Y be the
Pareto distribution with α = 3 and θ = 150. Compare the tail weight of
these distributions using
(a) the relative tail weight measure;
(b) the hazard rate measure.
Compare your results in (a) and (b).
Solution.
(a) Note that both distributions are heavy-tailed using the hazard rate anal-
ysis. However, h0Y (x) = − (x+150)
2 0 1
2 < hX (x) = − (x+150)2 so that hY (x) de-
creases at a faster rate than hX (x). Thus, X has a lighter tail than X.
(b) Using the relative tail weight, we find
fX (x) 2(150)2 (x + 150)4

lim = lim · = ∞.
x→∞ fY (x) x→∞ (x + 150)2 3(150)4
Hence, X has a heavier tail than Y which is different from the result in
(a)!
Remark 10.2
Note that the Gamma distribution is light-tailed for all α > 0 and θ > 0
by the existence of moments analysis. However, the Gamma distribution is
heavy-tailed for 0 < α < 1 by the hazard rate analysis. Thus, the concept
of light/heavy right tailed is somewhat vague in this case.
Classification Based on the Mean Excess Loss Function

A fourth measure of tail weight is the mean excess loss function as introduced
in Section 5. For a loss random variable X, the expected amount by which
loss exceeds x, given that it does exceed x is
E(X) − E(X ∧ x)
e(x) = eX (x) = E[X − x|X > x] = .
1 − F (x)
In the context of life contingency models (See [3]), if X is the random vari-
able representing the age at death and if T (x) is the continuous random
variable representing time until death of someone now alive at age x then
e(x) is denoted by e̊( x) = E[T (x)] = E[X − x|X > x]. In words, for a
newborm alive at age x, e̊( x) is the average additional number of years until
death from age x, given that an individual has survived to age x. We call
e̊( x) the complete expectation of life or the residual mean lifetime.
Viewed as a function of x, an increasing mean excess loss function is an

indication of a heavy-tailed distribution. On the other hand, a decreasing
mean excess loss function indicates a light-tailed distribution.
Next, we establish a relationship between e(x) and the hazard rate func-
tion. We have
R∞ Rx
E(X) − E(X ∧ x) 0 SX (y)dy − 0 SX (y)dy
e(x) = =
1 − F (x) SX (x)
R∞ Z ∞
x SX (y)dy SX (x + y)
= = dy.
SX (x) 0 SX (x)
But one of the characteristics of the hazard rate function is that it can
generate the survival function:
Rx
SX (x) = e− 0 h(t)dt
.
Hence, we can write

R x+y
∞ ∞
e− h(u)du
Z Z
0 R x+y
e(x) = −
Rx
h(u)du
dy = e− x h(u)du
dy, y > 0.
0 e 0 0
From the above discussion, we see that for a fixed y > 0, if SX (x+y)
SX (x) is an
increasing function of x (and therefore e(x) is increasing) then the hazard
rate function is decreasing and consequently the distribution is heavy-tailed.
Likewise, if the SX (x+y)
SX (x) is a decreasing function of x (and therefore e(x) is
decreasing) then the hazard rate function is increasing and consequently the
distribution is light-tailed.
Example 10.4
2
Let X be a random variable with pdf f (x) = 2xe−x for x > 0 and 0
otherwise. Show that the distribution is light-tailed by showing SX (x+y)
SX (x) is
a decreasing function of x.
Solution. R∞ 2 2
We have SX (x) = x 2te−t dt = e−x . Thus, for a fixed y > 0, we have
SX (x + y)
= e−2xy − y 2
SX (x)
whose derivative with respect to x is

d SX (x + y) 2
= −2ye−2xy−y < 0.
dx SX (x)
SX (x+y)
That is, SX (x) is a decreasing function of x
Practice Problems
Problem 10.1
Show that the Gamma distribution with parameters θ > 0 and α > 1 is
light-tailed by showing that f (x+y)
f (x) is nonincreasing.
Problem 10.2
Show that the Gamma distribution with parameters θ > 0 and α = 1 is
medium-tailed.
Problem 10.3
Let X be the Weibull distribution with probability density function f (x) =
τ
− x
τ xτ −1 e ( θ )
θτ .
Using hazard rate analysis, show that the distribution is heavy-tailed for
0 < τ < 1 and light-tailed for τ > 1.
Problem 10.4
2
Let X be a random variable with pdf f (x) = 2xe−x for x > 0 and 0 other-
wise.
Determine the tail weight of this distributions using Theorem 10.1.
Problem 10.5
Using Theorem 10.1, show that the Pareto distribution is heavy-tailed.
Problem 10.6
Show that the hazard rate function of the Gamma distribution approaches
1
θ as x → ∞.
Problem 10.7
1
Show that limx→∞ e(x) = limx→∞ h(x) .
Problem 10.8
Find limx→∞ e(x) where X is the Gamma distribution.
Problem 10.9
Let X be the Gamma distribution with 0 < α < 1 and θ > 0. Show that
e(x) increases from αθ to θ.
Problem 10.10
Let X be the Gamma distribution with α > 1 and θ > 0. Show that e(x)
decreases from αθ to θ.
Problem 10.11
Find limx→∞ e(x) where X is the Pareto distribution with parameters α and
θ and conclude that the distribution is heavy-tailed.
Problem 10.12
1
Let X be a random variable with pdf f (x) = (1+x)2
for x > 0 and 0 other-
wise.
Find an expression of limx→∞ e(x).
Problem 10.13
Let X be a random variable with mean excess loss function e(x) = x + 1.
(a) Find S(x), f (x) and h(x).

(b) Determine the tail behavior of X by using the moment criterion for tail
weight.
Problem 10.14
Determine the tail behavior of X by using the hazard rate analysis.
Problem 10.15
Determine the tail behavior of X by using the mean excess loss function
analysis.
Problem 10.16
x 2
Let X be a random variable with cdf S(x) = e−( θ ) . Determine the tail
behavior of X by using the mean excess loss function analysis.
Problem 10.17
Let X be the single-Pareto distribution with pdf
αθα
f (x) = .
xα+1
Use Theorem 10.1, to show that X is heavy-tailed.
11 Equilibrium Distributions and Tail Weight

In this section, we shed more insight into the mean residual lifetime. Let X
be a random variable such that
R ∞ S(0) = 1. Using Example 5.4 with d = 0,
we can write e(0) = E(X) = 0 S(x)dx. We define the random variable Xe
with probability density function
S(x)
fe (x) = , x≥0
E(X)
and 0 otherwise. We call the distribution of Xe , the equilibrium distri-
bution or integrated tail distribution.
The corresponding survival function is

Z ∞ Z ∞
S(t)
Se (x) = fe (t)dt = dt, x ≥ 0.
x x E(X)
The corresponding hazard rate function is
fe (x) S(x) 1
he (x) = = R∞ = .
Se (x) x S(t)dt
e(x)
Thus, if he (x) is increasing then the distribution Xe (and thus X) is ligh-

tailed. If he (x) is decreasing then the distribution Xe (or X) is heavy-tailed.
Example 11.1
Show that the equilibrium mean is given by
E(X 2 )
E(Xe ) = .
2E(X)
Solution.
Using integration by parts , we find
Z ∞
2
E(X ) = x2 f (x)dx
0
∞
Z ∞
2
= −x S(x)0 + 2 xS(x)dx
0
Z ∞
=2 xS(x)dx
0
since Z ∞ Z ∞
2 2
0 ≤ x S(x) = x f (t)dt ≤ t2 f (t)dt
x x
11 EQUILIBRIUM DISTRIBUTIONS AND TAIL WEIGHT 103
which implies
lim x2 S(x) = 0.
x→∞
Now,
∞ ∞
E(X 2 )
Z Z
1
E(Xe ) = xfe (x)dx = xS(x)dx =
0 E(X) 0 2E(X)
Example 11.2
Show that h i
e(0) − R0x 1
dt
S(x) = e e(t)
.
e(x)
Solution.
Using e(0) = E(X), We have
S(x) =E(X)fe (x) = e(0)fe (x) = e(0)Se (x)he (x)

R xh 1
i
− dt
=e(0)he (x)e 0 e(t)
h i
e(0) − R0x e(t)
1
dt
= e
e(x)
Example 11.3
Show that
e(x) Se (x)
= .
e(0) S(x)
Solution. R∞ R∞
Since Se (x) = 1 R S(x)
E(X) x S(t)dt, we have x S(t)dt = e(0)Se (x). Since x∞ S(t)dt =
1
R∞
e(x) ,
we obtain x S(t)dt = e(x)S(x). Thus, e(x)S(x) = e(0)Se (x) or equiv-
alently
e(x) Se (x)
=
e(0) S(x)
If the mean residual life function is increasing ( implied if the hazard rate
function of X is decreasing by Section 10) then e(x) ≥ e(0) and
Se (x) ≥ S(x).
Integrating both sides of this inequality, we find

Z ∞ Z ∞
Se (x)dx ≥ S(x)dx
0 0
which implies
E(X 2 )
≥ E(X)
2E(X)
and this can be rewritten as
E(X 2 ) − [E(X)]2 ≥ [E(X)]2
which gives
Var(X) ≥ [E(X)]2 .
Also,
Var(X)
[CV (x)]2 = ≥ 1.
[E(X)]2
Example 11.4
2
Let X be the random variable with pdf f (x) = (1+x) 3 for x ≥ 0 and 0
otherwise.
(a) Determine the survival function S(x).
(b) Determine the hazard rate function h(x).
(c) Determine E(X).
(d) Determine the pdf of the equilibrium distribution.
(e) Determine the survival function Se (x) of the equilibrium distribution.
(f) Determine the hazard function of the equilibrium distribution.
(g) Determine the mean residual lifetime of X.
Solution.
(a) The survival function is
Z ∞ ∞
2dt 1 1
S(x) = 3
=− 2
= .
x (1 + t) (1 + t) x (1 + x)2
(b) The hazard rate function is

f (x) 2
h(x) = = .
S(x) 1+x
(c) We have
Z ∞ Z ∞
2x
E(X) = xf (x)dx = dx = 1.
0 0 (1 + x)3
(d) We have
S(x) 1
fe (x) = =
E(X) (1 + x)2
for x > 0 and 0 otherwise.

(e) We have
1 ∞
Z ∞
dt 1
Se (x) = 2
=− = .
x (1 + t) 1+t x
1+x
(f) We have
fe (x) 1
he (x) = = .
Se (x) 1+x
(g) We have
1
e(x) = = x + 1, x ≥ 0
he (x)
Practice Problems
Problem 11.1
2
Let X be the random variable √with pdf f (x) = 2xe−x for x > 0 and 0
R ∞ −x2
otherwise. Recall 0 e dx = 2π .
(a) Determine the survival function S(x).
(b) Determine the equilibrium distribution fe (x).
Problem 11.2
Let X be a random variable with pdf f (x) = 31 (1 + 2x)e−x for x > 0
and 0 otherwise. Determine the hazard rate function of the equilibrium
distribution. Hint: Example 5.3.
Problem 11.3
Let X be a random variable with mean excees loss function
1
e(x) = , x > 0.
1+x
Determine the survival funtion of the distribution X.
Problem 11.4
1
e(x) = , x > 0.
1+x
Determine the survival function of the equilibrium distribution.
Problem 11.5
1
e(x) = , x > 0.
1+x
Determine the mean of the equilibrium distribution.
Problem 11.6
3 2
Let X be a random variable with pdf f (x) = 8x for 0 < x < 2 and 0
otherwise.
(a) Find E(X) and E(X 2 ).
(b) Find the equilibrium mean.
Problem 11.7
Let X be a loss random variable with mean excess loss function
e(x) = 10 + 9x, x > 0.
Determine the survival function S(x).
Problem 11.8
A random variable X has an exponential distribution with parameter λ.
Calculate the equilibrium mean.
Risk Measures
Most insurance models are stochastic or probabilistic models, i.e. involve

future uncertainty. As such, insurance companies are exposed to risks. Actu-
aries and risk managers job is to try to find the degree at which the insurance
companies are subject to a particular aspects of risk. In this chapter, we pro-
vide a definition of risk measure and discuss a couple of ways of measuring
risk.
109
110 RISK MEASURES
12 Coherent Risk Measurement

In financial terms, a risk measure is the necessary capital to be put on
reserve to support future risks associated, say to a portfolio. Risk man-
agement is about understanding and managing the potential losses in the
total portfolio. One of the key tasks of risk management is to quantify the
risk of the uncertainty in the future value of a portfolio. This quantification
is usually achieved by modeling the uncertain payoff as a random variable,
to which then a certain functional is applied. Such functionals are usually
called risk measures. This functional gives a single value that is intended
to provide a magnitude of the level of risk exposure.
Mathematically, we assign a random variable, defined on an appropriate

probability space, to each portfolio loss over some fixed time interval. Let
L be the collection of all such random variables. We will assume that L is a
convex cone so that if L1 and L2 are members of L then L1 + L2 and cL1
belong to L as well, where c is a constant. We define the coherent risk
measure to be the functional ρ : L −→[0, ∞) that satisfies the following
properties:
(P1) Subadditivity: For any L1 , L2 ∈ L, we have ρ(L1 + L2 ) ≤ ρ(L1 ) + ρ(L2 ).

This property says that the risk of two positions cannot get any worse than
adding the two risks separately. This property reflects the idea that pooling
risks helps to diversify a portfolio.
(P2) Monotonicity: If L1 ≤ L2 then ρ(L1 ) ≤ ρ(L2 ).

From an economic viewpoint, this property is obvious−positions leading al-
ways to higher losses require more risk capital.
(P3) Positive homogeneity: ρ(αL) = αρ(L), α > 0.

This property reflects the fact that there are no diversification benefits when
we hold multiples of the same portfolio, L.
(P4) Translation invariance: For any real number α, ρ(L + α) = ρ(L) + α.

That property states that by adding or subtracting a deterministic quantity
α to a position leading to the loss L we alter our capital requirements by
exactly that amount.
Remark 12.1
If L is a loss random variable then ρ(L) may be interpreted as the riskiness
12 COHERENT RISK MEASUREMENT 111
of a portfolio or the amount of capital that should be added to a portfolio

with a loss given by L, so that the portfolio can then be deemed acceptable
from a risk point of view. Coherent risk measures are important when the
risk comes from separate risk-taking departments within the same company.
Example 12.1
Show that the expectation function E(·) is a coherent risk measure on L.
Solution.
The expectation function E(·) satisfies the following properties:
(P1) E(L1 + L2 ) = E(L1 ) + E(L2 ).

(P2) If L1 ≤ L2 then E(L1 ) ≤ E(L2 ).
(P3) E(αL) = αE(L), α > 0.
(P4) E(L + α) = E(L) + α
Example 12.2
Show that the variance is not a cohorent risk measure.
Solution.
Since Var(L + a) = Var(L) 6= Var(L) + a, the variance of a distribution is
not a cohorent risk measure
Example 12.3
Show that ρ(L) = E(L) + βVar(L), where β > 0, satisfies the property of
translation invariant but not positive homogeneity. We refer to this risk
measure as the variance premium principle
Solution.
We have
ρ(L + α) =E(L + α) + βVar(L + α)

=E(L) + α + βVar(L)
=ρ(L) + α
ρ(αL) =E(αL) + βVar(αL)
=αE(L) + α2 βVar(L)
6=αρ(L)
where α > 0
112 RISK MEASURES
Example 12.4
Show that ρ(L) = α1 ln [E(eαL )], where α, t > 0, satisfies the properties of
translation invariant and monotonicity. We refer to this risk measure as the
exponential premium principle
Solution.
We have
1
ρ(L + β) = ln [E(eα(L+β) )]
α
1
= ln [E(eαL eαβ )]
α
1
= ln [eαβ E(eαL )]
α
1
= [αβ + ln [E(eαL )]
α
=ρ(L) + β.
Next, suppose that L1 ≤ L2 . We have
eαL1 ≤eαL2
E(eαL1 ) ≤E(eαL2 )
ln [E(eαL1 )] ≤ ln [E(eαL2 )]
ρ(L1 ) ≤ρ(L2 )
12 COHERENT RISK MEASUREMENT 113
Practice Problems
Problem 12.1
Show that ρ(0) = 0 and interpret this result.
Problem 12.2
Show that ρ(αL + β) = αρ(L) + β, where α > 0 and β ∈ R.
Problem 12.3
Show that ρ(L) = (1 + α)E(L) is a coherent risk measure, where α ≥ 0.
This risk measure is known as the expected value premium principle.
Problem 12.4
Which of the following is an implication of the subadditivity requirement
for a coherent risk measure?
(a) If the subadditivity requirement is met, then a merger of positions cre-

ates extra risk.
(b) If the subadditivity requirement is met, then a merger of positions does
not create extra risk.
(c) If the subadditivity requirement is met, then a merger of positions does
not affect risk.
Problem 12.5
Which of the following is an implication of the monotonicity requirement
for a coherent risk measure?
(a) Increasing the value of a portfolio increases risk.

(b) Increasing the value of a portfolio reduces risk.
(c) Increasing the value of a portfolio does not affect risk.
Problem 12.6
Which of the following is an implication of the positive homogeneity require-
ment for a coherent risk measure? More than one answer may be correct.
(a) If one assumes twice the amount of risk formerly assumed, one will
need twice the capital.
(b) As the size of a position doubles, the risk stays unchanged.
(c) The risk of the position increases in a linear way with the size of the
position.
114 RISK MEASURES
Problem 12.7
Which of the following is an implication of the translation invariant require-
ment for a coherent risk measure? More than one answer may be correct.
(a) Adding a fixed amount to the initial financial position should increase
the risk by the same amount.
(b) Subtracting a fixed amount to a portfolio decreases the required risk
capital by the same amount.
(c) Getting additional capital, if it is from a risk-free source, cannot funda-
mentally alter the riskiness of a position.
Problem 12.8
Show that ρ(L) = E(L) + αE[L − E(L)] satisfies (P1), (P3), and (P4).
Problem 12.9
Find the numerical value of ρ(L − ρ(L)).
Problem 12.10 p
Show that ρ(L) = E(L) + Var(L) satisfies the properties of translation
invariant and positive homogeneity. We refer to this risk measure as the
standard deviation principle.
13 VALUE-AT-RISK 115
13 Value-at-Risk
A standard risk measure used to evaluate exposure to risk is the value-at-
risk, abbreviated VaR. In general terms, the value-at-risk measures the
potential loss of value of an asset or a portfolio over a defined time with a
high level of certainty. For example, if the VaR is $1 million at one-month,
99% confidence level, then there is 1% chance that under normal market
movements the monthly loss will exceed $1 million. Bankers use VaR to
capture the potenetial losses in their traded portfolios from adverse market
movements over a period of time; then they use it to compare with their
available capital and cash reserves to ensure that the losses can be covered
withoud putting the firm at risk.
Mathematically, the value-at-risk is found as follows: Let L be a loss random

variable. The Value-at-risk at the 100p% level, denoted by VaRp (L) or πp ,
is the 100pth percentile or the p quantile of the distribution L. That is, πp
is the solution of FL (πp ) = p or equivalently, SL (πp ) = 1 − p.
Example 13.1
Let L be an exponential loss random variable with mean λ > 0. Find πp .
Solution.
x
The pdf of L is f (x) = λ1 e− λ for x > 0 and 0 otherwise. Thus, F (x) =
x πp
1 − e− λ . Now, solving the equation F (πp ) = p, that is, 1 − e− λ = p, we
obtain πp = −λ ln (1 − p)
Example 13.2
The loss random variable L has a Pareto distribution with parameters α
and θ. Find πp .
Solution.
αθα
The pdf of L is f (x) = (x+θ) α+1 for x > 0 and 0 otherwise. The cdf is
α
θ
F (x) = 1 − x+θ . Solving the equation F (πp ) = p, we find
1
πp = θ[(1 − p)− α − 1]
Example 13.3
The loss random variable L follows a normal distribution with mean µ and
standard deviation σ. Find πp .
116 RISK MEASURES
Solution.
Let Z = L−µσ . Then Z is the standard normal distribution. The p− quantile
of Z satisfies the equation Φ(z) = p. Thus, z = Φ−1 (p). Hence,
πp = µ + σz = µ + σΦ−1 (p)
Example 13.4
Consider a sample of size 8 in which the observed data points were 3,5,6,6,6,7,7,
and 10. Find VaR0.90 (L) for this empirical distribution.
Solution.
The pmf of L is given below.
x 3 5 6 7 10
1 1 3 2 1
p(x) 8 8 8 8 8
We want to find π0.90 such that
Pr(L < π0.90 ) < 0.90 ≤ Pr(L ≤ 0.90).
Thus, π0.90 = 10
Remark 13.1
According to [1], VaRp (L) is monotone, positive homogeneous, and trans-
lation invariant but not subadditive. Thus, VaRp (L) is not a coherent risk
measure.
13 VALUE-AT-RISK 117
Practice Problems
Problem 13.1
The loss random variable L has a uniform distribution in [a, b]. Find VaRp (L).
Problem 13.2
The cdf of a loss random variable L is given by
x2
FL (x) = 4 , 0<x≤2
1, x > 2.
Find π0.90 .
Problem 13.3
You are given the following empirical distribution
3, 5, 6, 6, 6, 7, 7, 10.
The risk measure under the standard deviation principle is ρ(L) = E(L) +
ασ(L). Determine the value of α so that ρ(L) = π0.90 .
Problem 13.4
Losses represented by L are distributed as a Pareto distribution with pa-
rameters α = 2 and θ = 60. Find VaR0.75 (L).
Problem 13.5
Losses represented by L are distributed as a single Pareto distribution with
α
a pdf f (x) = xαθ
α+1 , x > θ and 0 otherwise. Find πp .
Problem 13.6
A loss random variable X has a survival function
2
θ
S(x) = , x > 0.
x+θ
Find θ given that π0.75 = 40.
Problem 13.7
Let L be a random variable with discrete loss distribution given by
x 0 100 1000 10000 100000
p(x) 0.65 0.20 0.07 0.05 0.03
Calculate the Value-at-Risk of L at the 90% level.
118 RISK MEASURES
Problem 13.8
A loss random variable L has a two-parameter Pareto distribution satisfying:
VaR0.90 (L) = 216.23 and VaR0.99 (L) = 900.
Calculate VaR0.95 (L).
Problem 13.9
Let L be a loss random variable with probability generating function
PL (x) = 0.4x2 + 0.2x3 + 0.2x5 + 0.2x8 .
Determine VaR0.80 (L).
Problem 13.10
A loss random variable L has a survival function
2
100
S(x) = , x > 0.
x + 100
Calculate VaR0.96 and interpret this result.

14 TAIL-VALUE-AT-RISK 119
14 Tail-Value-at-Risk
The quantile risk meaure discussed in the previous section provides us only
with the probability that a loss random variable L will exceed the VaRp (L)
for a certain confidence level. It does not provide any information about
how large the losses are beyond a particular percentile. The Tail-Value-
at-Risk (TVaR) measure does consider losses above a percentile. Other
names used for TVaR are Tail Conditional Expectation and Expected
Shortfall.
The Tail-Value-at-Risk of a random variable L at the 100p% security level

is defined as
T V aRp (L) = E[L|L > V aRp (L)] = E[L|L > FL−1 (p)]
where FL (x) is the distribution of L. This is the expected value of the loss,
conditional on the loss exceeding πp . Note that TVaR is also the expected
cost per payment with a franchise deductible7 of πp .
We can write TVaR as

R∞ R∞
πp xf (x)dx πp xf (x)dx
T V aRp (L) = E[L|L > V aRp (L)] = = .
1 − F (πp ) 1−p
Now, using the substitution u = F (x), we can write

R1 R1
p xf (x)dx p VaRu (L)du
TVaRp (L) = = .
1−p 1−p
TVaR can also be written as
Z ∞ Z ∞
πp 1
TVaRp (L) = f (x)dx+ (x−πp )f (x)dx = πp +E[X−πp |X > πp ].
1 − p πp 1 − p πp
Since
E(L) − E(L ∧ πp )
E[X − πp |X > πp ] = ,
1−p
we can write
E(L) − E(L ∧ πp )
TVaRp (L) = VaRp (L) + .
1−p
7
See Section 32.
120 RISK MEASURES
Remark 14.1
Unlike the VaR risk measure, TVaR risk measure is shown to be coherent.
Example 14.1
Find the Tail-Value-at-Risk of an exponential distribution with mean λ > 0.
Solution.
From Problem 5.7, we have e(πp ) = λ. This and Example 13.1 give
T V aRp (L) = λ − λ ln (1 − p)
Example 14.2
Find the Tail-Value-at-Risk of a Pareto distribution with parameters α > 1
and θ > 0.
Solution.
The survival function of the Pareto distribution is
α
θ
S(x) = , x > 0.
x+θ
Thus,
Z ∞
1
e(πp ) = S(x)dx
S(πp ) πp
Z ∞
=(πp + θ)α (x + θ)−α dx
πp
(πp + θ)α ∞
= (x + θ)1−α πp
α−1
πp + θ
= .
α−1
On the other hand, using Example 13.2, we have
1
πp = θ[(1 − p)− α − 1].
Hence,
πp + θ 1
TVaRp (L) = + θ[(1 − p)− α − 1]
α−1
Example 14.3
x2
Let Z be the standard normal distribution with pdf fZ (x) = √1 e− 2 . Find
2π
TVaRp (Z).
Solution.
Notice first that fZ (x) satisfies the differential equation xfX (x) = −fZ0 (x).
Using the Fundamental Theorem of Calculus, we can write
Z ∞
1
TVaRp (Z) = xfZ (x)dx
1 − p Φ−1 (p)
Z ∞
1
=− f 0 (x)dx
1 − p Φ−1 (p) Z
∞
1
=− fZ (x)
1−p Φ−1 (p)
1
= fZ [Φ−1 (p)]
1−p
Example 14.4
Let L be a loss random variable having a normal distribution with mean µ
and standard deviation σ. Find TVaRp (L).
Solution.
Since TVaRp (L) is a coherent risk measure, it is positive homogeneous and
translation invariant. Thus, we have
σ
TVaRp (L) = TVaRp (µ + σZ) = µ + σTVaRp (Z) = µ + fZ [Φ−1 (p)]
1−p
122 RISK MEASURES
Practice Problems
Problem 14.1
Let L be a loss random variable with uniform distribution in (a, b).
(a) Find the mean residual life e(x).

(b) Find TVaRp (L).
Problem 14.2
The cdf of a loss random variable L is given by
x2
FL (x) = 4 , 0<x≤2
1, x > 2.
(a) Find π0.90 and e(π0.90 ).

(b) Find TVaR0.90 (L).
Problem 14.3
3, 5, 6, 6, 6, 7, 7, 10.
Find π0.85 and TVaR0.85 (L).
Problem 14.4
Losses are distributed as Pareto distributions with mean of 200 and variance
of 60000.
(a) Find the values of the parameters α and β.

(b) Find e(100).
(c) Find π0.95 .
(d) Find TRaV0.95 (L).
Problem 14.5
Losses represented by the random variable L are uniformly distributed from
0 to the maximum loss. You are given that Var(L) = 62, 208.
Find TVaR0.75 (L).
Problem 14.6
Losses represented by the random variable L are uniformly distributed in
(0, 864). Determine β so that the standard deviation principle is equal to
TVaR0.75 (L).
Problem 14.7
Diabetes claims follow an exponential distribution with parameter λ = 2.
Find TVaR0.90 (L).
Problem 14.8
3, 5, 6, 6, 6, 7, 7, 10.
Let β1 be the value of β in the standard deviation principle such that

µ + β1 σ = VaR0.85 . Let β2 be the value of β in the standard deviation
principle such that µ + β2 σ = TVaR0.85 (L).
Calculate β2 − β1 .
Problem 14.9
Let L1 be a Pareto random variable with parameters α = 2 and θ = 100.
Let L2 be a random variable with uniform distribution on (0, 864). Find p
such that
TVaR0.99 (L1 ) TVaRp (L2 )
= .
VaR0.99 (L1 ) VaRp (L2 )
Problem 14.10
Let L be a random variable with discrete loss distribution given by
x 0 100 1000 10000 100000

p(x) 0.65 0.20 0.07 0.05 0.03
Calculate the Tail-Value-at-Risk of L and the 90% level.
Problem 14.11
Find TVaR0.95 (L) when L has a normal distribution with mean of 100 and
standard deviation of 10.
124 RISK MEASURES
Characteristics of Actuarial
Models
In the previous chapter, a characteristic of an actuarial model is the tail

weight of the corresponding distribution. In this chapter we look at other
factors that characterize a model from another model. One such a factor is
the number of parameters needed in the determination of the model’s distri-
bution. More parameters involved in a model means that more information
is required and in this case the model is categorized as a complex model. We
start this chapter by discussing first simple models and then move toward
more complex models.
125
126 CHARACTERISTICS OF ACTUARIAL MODELS
15 Parametric and Scale Distributions

These are considered the simplest families of actuarial models. We will
consider a model that requires less parameters than another model as less
complex.
A parametric distribution is one that is completely determined by a set

of quantities called parameters. Examples of commonly used parametric
distributions are listed below.
Name PDF Parameters
Exponential f (x) = θe−θx θ>0
αθα
Pareto f (x) = (x+θ)α+1 α > 0, θ > 0
(x−µ)2
Normal f (x) = √ 1 e− 2σ 2 µ, σ
2πσ
e−λ λx
Poisson p(x) = x! λ>0
Additional parametric distributions can be found in the Tables of Exam C.
Now, when multiplying a random variable by a positive constant and the

resulting distribution belongs to the same family of distributions of the orig-
inal random variable then we call the distribution scale distribution.
Example 15.1
Show that the Pareto distribution is a scale distribution.
Solution.
The cdf of the Pareto distribution is
α
θ
FX (x) = 1 − .
x+θ
Let Y = cX. Then
y
FY (y) =Pr(Y ≤ y) = Pr X ≤
α c
cθ
=1 − .
y + cθ
This is a Pareto distribution with parameters α and cθ
Example 15.2
Show that the Weibull distribution with parameters θ and τ is a scale dis-
tribution.
15 PARAMETRIC AND SCALE DISTRIBUTIONS 127
Solution.
The cdf of the Weibull distribution is
x τ
FX (x) = 1 − e−( θ ) .
Let Y = cX. Then

y
FY (y) =Pr(Y ≤ y) = Pr X ≤
c
x τ
−( cθ )
=1 − e .
This is a Weibull distribution with parameters cθ and τ
A parameter θ in a scale distribution X is called a scale parameter if

cθ is a parameter of cX and θ is the only changed parameter.
Example 15.3
Show that the parameter θ in the Pareto distribution is a scale parameter.
Solution.
This follos from Example 15.1
Example 15.4
Find the scale parameter of the Weibull distribution with parameters θ and
τ.
Solution.
According to Example 15.2, the scale parameter is θ
Example 15.5
The amount of money in dollars that Clark received in 2010 from his invest-
ment in Simplicity futures follows a Pareto distribution with parameters
α = 3 and θ. Annual inflation in the United States from 2010 to 2011 is
i%. The 80th percentile of the earning size in 2010 equals the mean earning
size in 2011. If Clark’s investment income keeps up with inflation but is
otherwise unaffected, determine i.
Solution.
Let X be the earning size in 2010 and Y that in 2011. Then Y is a Pareto
distribution with parameters α = 3 and (1 + i)θ. We are told that
(1 + i)θ (1 + i)θ
π0.80 = E(Y ) = (1 + i)E(X) = = .
3−1 2
Thus,
!3
(1 + i)θ (1 + i)θ θ
0.8 = Pr X < = FX =1− (1+i)θ
.
2 2 +θ
2
Solving the above equation for i, we find i = 0.42
By assigning all possible numerical values to the parameters of a partic-

ular parametric distribution we obtain a family of distributions that we call
a parametric distribution family.
Example 15.6
Show that exponential distributions belong to the Weibull distribution fam-
ily with parameters θ and τ.
Solution. τ
τ − x ( )
τ ( xθ ) eθ
Weibull distributions with parameters θ and τ have pdf fX (x) = x .
x
e− θ
Letting τ = 1, the pdf reduces to fX (x) = θ which is the pdf of an expo-
nential distribution
15 PARAMETRIC AND SCALE DISTRIBUTIONS 129
Practice Problems
Problem 15.1
Show that the exponential distribution is a scale distribution.
Problem 15.2
2
Let X be a random variable with pdf fX (x) = 2xe−x for x > 0 and 0
otherwise. Let Y = cX for c > 0.
Find FY (y).
Problem 15.3
Let X be a uniform random variable on the interval (0, θ). Let Y = cX for
c > 0.
Find FY (y).
Problem 15.4
x −α
Show that the Fréchet distribution with cdf FX (x) = e−( θ ) and parame-
ters θ and α is a scale distribution.
Problem 15.5
Show that the three-parameter Burr distribution with cdf FX (x) = 1 −
1
γ α is a scale distribution.
[1+( xθ ) ]
Problem 15.6
Find the scale parameter of the following distributions:
(a) The exponential distribution with parameter θ.

(b) The uniform distribution on (0, θ).
(c) The Fréchet distribution with parameters θ and α.
(d) The Burr distribution with parameters α, θ, and γ.
Problem 15.7
Claim severities are modeled using a continuous distribution and inflation
impacts claims uniformly at an annual rate of r.
Which of the following are true statements regarding the distribution of

claim severities after the effect of inflation?
(1) An exponential distribution will have a scale parameter of (1 + r)θ.

(2) A Pareto distribution will have scale parameters (1 + r)α and (1 + r)θ
(3) A Burr distribution will have scale parameters α, (1 + r)θ, γ.
Problem 15.8
Let X be the
lognormal
distribution with parameters µ and σ and cdf
FX (x) = Φ ln x−µ
σ .
Show that X has a scale distribution with no scale parameters.
Problem 15.9
Show that the Gamma distribution is a scale distribution. Is there a scale
parameter?
Problem 15.10
Earnings during 2012 follow a Gamma distribution with variance 2,500. For
the 2013, earnings are expected to be subject to P % inflation and the ex-
pected variance for the 2013 year is 10,000.
Determine the value of P.
Problem 15.11
The Gamma distribution with parameters α and θ has the pdf fX (x) =
x
xα−1 e− θ
Γ(α) .
Show that the exponential distributions belong to this family.
Problem 15.12
Hardy Auto Insurance claims X are represented by a Weibull distribution
with parameters α = 2 and θ = 400. It is found that the claim sizes are
inflated by 30% uniformly.
Calculate the probability that a claim will be at least 90 counting infla-

tion.
16 DISCRETE MIXTURE DISTRIBUTIONS 131
16 Discrete Mixture Distributions

In probability and statistics, a mixture distribution is the probability
distribution of a random variable whose values can be interpreted as being
derived from an underlying set of other random variables. For example, a
dental claim may be from a check-up, cleaning, filling cavity, a surgical pro-
cedure, etc.
A random variable X is a k-point mixture of the random variables X1 , · · · , Xk

if its cumulative distribution function (cdf) is given by
FX (x) = a1 FX1 (x) + a2 FX2 (x) + · · · + ak FXk (x)
where each mixing weight ai > 0 and a1 + a2 + · · · ak = 1. The mixture

X is defined in terms of its pdf or cdf and is not the sum of the random
variables a1 X1 , a2 X2 , · · · , an Xn .
The mixing weights are discrete probabilities. To see this, let Θ be the dis-
crete random variable with support {1, 2, · · · , k} and pmf Pr(Θ = i) = ai .
We can think of the distribution of Θ as a conditioning distribution where
X = Xi is conditioned on Θ = i, or equivalently, FX|Θ (x|Θ = i) = FXi (x).
In this case, X is the unconditional distribution with cdf
k
X
FX (x) = a1 FX1 (x)+a2 FX2 (x)+· · ·+ak FXk (x) = FX|Θ (x|Θ = i)Pr(Θ = i).
i=1
In actuarial and insurance terms, discrete mixtures arise in situations where

the risk class of a policyholder is uncertain, and the number of possible risk
classes is discrete.
The continuous mixture of distributions will be discussed in Section 20 of

this study guide.
Example 16.1
Let Y be a 2-point mixture of two random variables X1 and X2 with mixing
weights 0.6 and 0.4 respectively. The random variable X1 is a Pareto random
variable with parameters α = 3 and θ = 900. The random variable X2 is a
Pareto random variable with parameters α = 5 and θ = 1500. Find the pdf
of Y.
Solution.
We are given
3(900)3 5(1500)5
fX1 = (x+900)4
and fX2 (x) = (x+1500)6
.
Thus,
3(900)3 5(1500)5

fY (x) = 0.6fX1 + 0.4fX2 = 0.6 + 0.4
(x + 900)4 (x + 1500)6
Example 16.2 ‡
The random variable N has a mixed distribution:
(i) With probability p, N has a binomial distribution with q = 0.5 and
m = 2.
(ii) With probability 1 − p, N has a binomial distribution with q = 0.5 and
m = 4.
Calculate Pr(N = 2).
Solution.
We have
pN1 (N1 = 2) =C(2, 2)(0.5)2 = 0.25

pN2 (N2 = 2) =C(4, 2)(0.5)2 (0.5)2 = 0.375
pN (N = 2) =ppN1 (N1 = 2) + (1 − p)pN2 (N1 = 2) = 0.375 − 0.125p
Example 16.3
Determine the mean and second moment of the two-point mixture distribu-
tion with the cdf
α α+2
θ1 θ2
FX (x) = 1 − α − (1 − α) .
x + θ1 x + θ2
Solution.
The first part is the distribution of a Pareto random variable X1 with pa-
rameters α1 = α and θ1 . The second part is the distribution of a Pareto
random variable X2 with parameters α2 = α + 2 and θ2 . Thus,
θ1 θ1
E(X1 ) = =
α1 − 1 α−1
θ2 θ2
E(X2 ) = =
α2 − 1 α+1
θ1 θ2
E(X) =α + (1 − α)
α α+1
θ12 2! 2θ12
E(X12 ) = =
(α1 − 1)(α1 − 2) (α − 1)(α − 2)
2
θ2 2! 2θ22
E(X22 ) = =
(α2 − 1)(α2 − 2) (α)(α + 1)
2 2θ22

2 2θ1
E(X ) =α + (1 − α)
(α − 1)(α − 2) (α)(α + 1)
Next, we consider mixtures where the number of random variables in the
mixture is unknown. A variable-component mixture distribution has
a distribution function that can be written as
FX (x) = a1 FX1 (x) + a2 FX2 (x) + · · · + aN FXN (x)
where each aj > 0 and N

P
i=1 ai = 1 and N ∈ N.
In a variable-component mixture distribution, each of the mixture weights

associated with each individual FXj (x) is a parameter. Also, there are
(N − 1) parameters corresponding to the weights a1 through aN −1 . The
weight aN is not itself a parameter, since the value of aN is determined by
the value of the constants a1 through aN −1 .
Example 16.4
Determine the distribution, density, and hazard rate functions for the vari-
able mixture of exponential distributions.
Solution.
The distribution function of the variable mixture is
− θx − θx − θx
FX (x) = 1 − a1 e 1 − a2 e 2 − · · · − aN e N
where aj > 0 and N

P
i=1 ai = 1.
The density function is
a1 − θx a2 − x aN − θx
fX (x) = e 1 + e θ2 + · · · + e N
θ1 θ2 θN
and the hazard rate function is
x x x
a1 − θ1 a2 − θ2 aN − θN
θ1 e + θ2 e + ··· + θN e
hX (x) = −x −x − x
.
a1 e θ1 + a2 e θ2 + ··· + aN e θN
Note that the parameters corresponding to N = 3 are (a1 , a2 , θ1 , θ2 , θ3 ) and
those corresponding to N = 5 are (a1 , a2 , a3 , a4 , θ1 , θ2 , θ3 , θ4 , θ5 )
Example 16.5 ‡
You are given claim count data for which the sample mean is roughly equal
to the sample variance. Thus you would like to use a claim count model
that has its mean equal to its variance. An obvious choice is the Poisson
distribution.
Determine which of the following models may also be appropriate.
(A) A mixture of two binomial distributions with different means
(B) A mixture of two Poisson distributions with different means
(C) A mixture of two negative binomial distributions with different means
(D) None of (A), (B) or (C)
(E) All of (A), (B) and (C).
Solution.
Let X be a 2-point mixture of the random variables X1 and X2 with mixing
weights α and 1−α. Let Θ be the discrete random variable such that Pr(Θ =
1) = α and Pr(Θ = 2) = 1 − α. Thus, we have
E(X) =αE(X1 ) + (1 − α)E(X2 )
Var(X) =E(X 2 ) − E(X)2
=αE(X12 ) + (1 − α)E(X22 ) − [αE(X1 ) + (1 − α)E(X2 )]2
=αVar(X1 ) + (1 − α)Var(X2 ) + α(1 − α)[E(X1 ) − E(X2 )]2 .
If X1 and X2 are Poisson with means λ1 and λ2 respectively with λ1 6= λ2 ,
then
Var(X) =αλ1 + (1 − α)λ2 + α(1 − α)(λ1 − λ2 )2
>αλ1 + (1 − α)λ2 = E(X).
If X1 and X2 are negative binomial with parameters (r1 , β1 ) and (r2 , β2 )
respectively with r1 β1 6= r2 β2 , then
Var(X) =αr1 β1 (1 + β1 ) + (1 − α)r2 β2 (1 + β2 ) + α(1 − α)(r1 β1 − r2 β2 )2
>αr1 β1 + (1 − α)r2 β2 = E(X).
If X1 and X2 are binomial with parameters (m1 , q1 ) and (m2 , q2 ) respectively
with m1 q1 6= m2 q2 , then
Var(X) =αm1 q1 (1 − q1 ) + (1 − α)m2 q2 (1 − q2 ) + α(1 − α)(m1 q1 − m2 q2 )2
=E(X) + α(1 − α)(m1 q1 − m2 q2 )2 − αm1 q12 − (1 − α)m2 q22 .
The expression α(1 − α)(m1 q1 − m2 q2 )2 − αm1 q12 − (1 − α)m2 q22 can be
positive, negative, or zero. Thus, a mixture of two binomial distributions
with different means may result in the variance being equal to the mean
Example 16.6 ‡
Losses come from an equally weighted mixture of an exponential distribu-
tion with mean m1 , and an exponential distribution with mean m2 .
Determine the least upper bound for the coefficient of variation of this dis-
tribution.
Solution.
Let X be the random variable with pdf

1 1 − mx 1 − mx
f (x) = e 1 + e 2 .
2 m1 m2
We have
1
E(X) = (m1 + m2 )
2
1
E(X ) = (2m21 + 2m22 )
2
2
2
1 2 2 1
Var(X) = (2m1 + 2m2 ) − (m1 + m2 ) .
2 2
The square of coefficient of variation of X is

1
2
2 + 2m22 ) − 12 (m1 + m2 )

2 (2m1
CV 2 = 1 2
2 (m1 + m2 )
3m21 − 2m1 m2 + 3m22
=
(m1 + m2 )2
8m1 m2
=3 − .
(m1 + m2 )2
m1
Let r = m2 . Then
8r
CV 2 = 3 − .
(1 + r)2
Thaking the derivative and setting it to 0, we find
−8(1 + r)−2 + 16r(1 + r)−3 = 0 =⇒ r = 1.
Moreover, r00 (1) = 1 > 0, r(0) = r(∞) = 3 so that r = 1 is a global

minimum.
√ Hence, the least upper bound of the coefficient of variation is
3
Practice Problems
Problem 16.1
The distribution of a loss, X, is a 2-point mixture:
(i) With probability 0.6, X1 is a Pareto distribution with parameters α = 3
and θ = 900.
(ii) With probability 0.4, X2 is a Pareto distribution with parameters α = 5
and θ = 1500.
Determine Pr(X > 1000).

Problem 16.2
(i) With probability 0.5, X1 is a Burr distribution with parameters α =
1, γ = 2 and θ = 10000.5 .
and θ = 1000.
Determine the median of X.

Problem 16.3
You are given:
• X is a 2-point mixture of two exponential random variables X1 and X2
with parameters θ1 = 1 and θ2 = 3 and mixing weights 12 and 16 respectively.
• Y = 2X and Y is a mixture of two exponential random variables Y1 and Y2 .
Find E(Y1 ) and E(Y2 ).

Problem 16.4
The severity distribution function for losses from your renters insurance is
the following:
5 3
1000 3500
FX (x) = 1 − 0.3 − 0.7 .
1000 + x 3500 + x
Calculate the mean and the variance of the loss size.
Problem 16.5
Seventy-five percent of claims have a normal distribution with a mean of
3,000 and a variance of 1,000,000. The remaining 25% have a normal dis-
tribution with a mean of 4,000 and a variance of 1,000,000.
Determine the probability that a randomly selected claim exceeds 5,000.

Problem 16.6
How many parameters are there in a variable component mixture consisting
of 9 Burr distributions?
Problem 16.7
Determine the distribution, density, and hazard rate functions for the vari-
able mixture of two-parameter Pareto distribution.
Problem 16.8
A Weibull distribution has two parameters: θ and τ. An actuary is creating
variable-component mixture distribution consisting of K Weibull distribu-
tions. If the actuary chooses to use 17 Weibull distributions instead of 12,
how many more parameters will the variable-component mixture distribu-
tion have as a result?
Problem 16.9
Let X be a 2-point mixture with underlying random variables X1 and X2 .
The distribution of X1 is a Pareto distribution with parmaters α1 = 3 and
θ. The distribution of X2 is a Gamma distribution with parameters α2 = 2
and θ2 = 2000.
Given that a1 = 0.7, a2 = 0.3, and E(X) = 1340, determine the value
of θ.
Problem 16.10
Let X be a 3-point mixture of three variables X1 , X2 , X3 . You are given the
following information:
R.V. Weight Mean Standard Deviation
X1 0.2 0.10 0.15
X2 0.5 0.25 0.45
X3 0.3 0.17 0.35
Determine Var(X).
Problem 16.11 ‡
(i) With probability 0.8, X1 is a Pareto distribution with parameters α = 2
and θ = 100.
and θ = 3000.
Determine Pr(X ≤ 200).

17 Data-dependent Distributions
In Section 15, we discussed parametric distributions. In Section 16, we
introduced the k−point mixture distributions that are also known as semi-
parametric distributions. In this section, we look at non-parametric
distributions.
According to [1], a data-dependent distribution is at least as complex

as the data or knowledge that produced it, and the number of “parameters”
increases as the number of data points or the amount of knowledge increases.
We consider two-types of data-dependent distributions:
• The empirical distribution is obtained by assigning a probability of

1
n to each data point in a sample with n data points.
Example 17.1
Below are the losses suffered by policyholders of an insurance company:
49, 50, 50, 50, 60, 75, 80, 120, 230.
Let X be the random variable representing the losses incurred by the poli-
cyholders. Find the pmf and the cdf of X.
Solution.
The pmf is given by the table below.
x 49 50 60 75 80 120 130
1 1 1 1 1 1 1
p(x) 9 3 9 9 9 9 9
The cdf is defined by

FX (x) = 19 number of elements in the sample that are ≤ x.
Thus, for example,
5
FX (73) =
9
• A kernel smoothed distribution
Given an empirical distribution, we wish to create a continuous distribution
whose pdf will be a good estimation of the (discrete) empirical distribution.
The density function is given by
n
X
fX (x) = pn (x)ki (x)
i=1
17 DATA-DEPENDENT DISTRIBUTIONS 139
where pn (x) = n1 and ki (x) is the kernel smoothed density function.

We illustrate these concepts next.
Example 17.2
49, 50, 50, 50, 60, 75, 80, 120, 230.
Develop a kernel-smoothed distribution associated with this set, such that

each point x is associated with a uniform distribution that has positive prob-
ability over the interval (x − 5, x + 5). As your answer, write the probability
density function (pdf) of the kernel smoothed distribution.
Solution.
For i = 1, 2, · · · , 9, we have
1

10 , x i − 5 ≤ x ≤ xi + 5
ki (x) =
0, otherwise.
We refer to ki (x) as the uniform kernel with bandwith 5.Thus,

9
X 1
fX (x) = ki (x)
9
i=1
A futher discussion of kernel density models will be covered in Section 56 of

this book.
Practice Problems
Problem 17.1
You are given the following empirical distribution of losses suffered by poli-
cyholders Prevent Dental Insurance Company:
94, 104, 104, 104, 134, 134, 180, 180, 180, 180, 210, 350, 524.
cyholders.
Find the mean, the variance and the mode of X.
Problem 17.2
94, 104, 104, 104, 134, 134, 180, 180, 180, 180, 210, 350, 524.
Let X be the random variable representing the losses incurred by the policy-
holders. The insurance company issued a policy with an ordinary deductible
of 105.
Calculate E(X ∧ 105) and the cost per payment eX (105).
Problem 17.3
94, 104, 104, 104, 134, 134, 180, 180, 180, 180, 210, 350, 524.
cyholders.
Find the value of β in the standard deviation principle µ + βσ so that

the standard deviation principle is equal to VaR0.8 (X).
Problem 17.4
17 DATA-DEPENDENT DISTRIBUTIONS 141
94, 104, 104, 104, 134, 134, 180, 180, 180, 180, 210, 350, 524.
cyholders.
Find the empirical pmf and cdf of X.
Problem 17.5
94, 104, 104, 104, 134, 134, 180, 180, 180, 180, 210, 350, 524.
cyholders.
Find the coefficient of variation of X.
Problem 17.6
94, 104, 104, 104, 134, 134, 180, 180, 180, 180, 210, 350, 524.
cyholders.
Find the coefficient of skewness of X.
Problem 17.7
You are given the following the distribution of losses suffered by policyhold-
ers Prevent Dental Insurance Company:
94, 104, 104, 104, 134, 134, 180, 180, 180, 180, 210, 350, 524.
cyholders.
Calculate fX (x) using smoothed kernel setting with uniform kernel of band-
with 4.
Generating New
Distributions
In this chapter a collection of continuous models that are commonly used for
most actuarial modeling situations will be developed. Processes for creating
new distributions from existing ones will be introduced. We point out here
that most of the distributions that are used in actuarial modeling have
nonnegative support so that FX (0) = 0.
143
144 GENERATING NEW DISTRIBUTIONS
18 Scalar Multiplication of Random Variables

The first process that we discuss for creating new distributions from existing
ones is the multiplication of a random variable by a constant. The following
theorem sheds information about this process.
Theorem 18.1
Let X be a continuous random variable and c a positive constant. Let
Y = cX. Then fY (y) = 1c fX yc and FY (y) = FX yc . Thus, c is a scale

parameter for Y.
Proof.
We start by finding the cdf of Y and then we generate the pdf by differen-
tiation. We have
y y
FY (y) = Pr(Y ≤ y) = Pr X ≤ = FX .
c c
Now, differentiating FY (y) and using the chain rule we find
1 y
fY (y) = fX
c c
Example 18.1
Suppose that random losses are exponentially distributed with parameter θ.
Find the pdf and the cdf of the random variable Y = cX, c > 0.
Solution.
We have y y
FY (y) = Pr(Y ≤ y) = Pr X ≤ = 1 − e− cθ .
c
Thus,
d 1 y
fY (y) = FY (y) = e− cθ
dy c
Scalar Multiples of random variables are useful in actuarial modeling when
annual losses are subject to future uniform inflation. For example, if X is
the random variable representing this year’s losses and uniform losses are
known to be i% for the next year then the next year’s losses can be modeled
with the random variable Y = (1 + 0.01i)X.
Example 18.2
You are given:
(i) In 2011, losses follow a Pareto distribution with parameters α = 2 and
18 SCALAR MULTIPLICATION OF RANDOM VARIABLES 145
θ = 100.
(ii) Inflation of 3.5% impacts all losses uniformly from 2011 to 2012.
What is the probability that the losses will exceed 350 in 2012?
Solution.
Let X and Y be the random variables representing the losses in 2011 and
2012 respectively. Then Y = 1.035X. We want to find Pr(Y > 350). Recall
that the cdf of the Pareto distribution is
2
100
FX (x) = 1 − .
x + 100
Thus,
!2
350 350 100
Pr(Y > 350) = Pr X > = SX X > = 350 = 0.0521
1.035 1.035 1.035 + 100
Practice Problems
Problem 18.1
You are given:
(i) X has a Pareto distribution with parameters α = 3 and θ = 2000.
(ii) Y = cX, c > 0.
(iii) σY = 1500.
Determine the value of c.
Problem 18.2
Losses in 2011 are represented by a random variable X with pdf fX (x) = 83 x2
for 0 < x < 2 and 0 otherwise. Let Y be the random variable of losses in
2012. It is expected that losses in 2012 will go down by 50% than the current
year.
Find fY (y), FY (y), and SY (y).
Problem 18.3
Losses from auto accidents are modeled by a Pareto distribution with pa-
rameters α = 3 and θ = 2000. The insurance policy pays only 75% of any
auto accident claim.
Find the mean and the standard deviation of the claims for this policy.
Problem 18.4
Let X have cdf FX (x) = 1 − (1 + x)−α where x > 0 and α > 0. Determine
the pdf and the cdf of Y = θX.
Problem 18.5
Let Y have the lognormal distribution with parameters µ and σ. Let Z = θY.
Show that Z also have the lognormal distribution.
Problem 18.6
Losses in 1993 follow the density function fX (x) = 3x−4 , x > 1 where x is
the loss in millions of dollars. Inflation of 10% impacts all claims uniformly
from 1993 to 1994.
Determine the cdf of losses for 1994 and use it to determine the probability
that a 1994 loss exceeds 2,200,000.
18 SCALAR MULTIPLICATION OF RANDOM VARIABLES 147
Problem 18.7
You are given: √
(i) X is a loglogistic random variable with parameters γ = 2 and θ = 10 10.
(ii) Y is a Pareto distribution with parameters α = 1 and θ = 1000.
(iii) Z is a 2-point mixture of X and Y with equal mixing weights.
(iv) W = (1 + r)Z where r > 0.
Show that W is an equal mixture of a loglogistic and a Pareto distribu-

tion and determine the parameters of W.
19 Powers and Exponentiation of Random Variables

A new distribution is obtained by raising a random variable to a certain
power such as
1 1
Y = Xτ or Y = X − τ
where τ > 0. In the first case, Y is called transformed. In the second case,
assuming τ 6= 1, Y is called inverse transformed. If τ = 1, we call Y the
inverse of X.
The pdf and the cdf of the new distribution are provided by the next theo-
rem.
Theorem 19.1
Let X be a continuous random variable with pdf and cdf fX (x) and FX (x)
with FX (0) = 0. Let τ > 0. We have
1
(a) In the transformed case, Y = X τ
FY (y) = FX (y τ ) and fY (y) = τ y τ −1 fX (y τ ).
1
(b) In the inverse transformed case, Y = X − τ
FY (y) = 1 − FX (y −τ ) and fY (y) = τ y −τ −1 fX (y −τ ).
(c) In the inverse case, Y = X −1
FY (y) = 1 − FX (y −1 ) and fY (y) = y −2 fX (y −1 ).
Proof.
(a) We have
FY (y) = Pr(Y ≤ y) = Pr(X ≤ y τ ) = FX (y τ ).
Differentiating this function with respect to y and using the chain rule, we
find
fY (y) = τ y τ −1 fX (y τ ).
(b) We have
FY (y) = Pr(Y ≤ y) = Pr(X ≥ y τ ) = 1 − FX (y −τ ).
find
fY (y) = τ y −τ −1 fX (y −τ ).
19 POWERS AND EXPONENTIATION OF RANDOM VARIABLES 149
(c) We have
FY (y) = Pr(Y ≤ y) = Pr(X ≥ y −1 ) = 1 − FX (y −1 ).
find
fY (y) = y −2 fX (y −1 )
Example 19.1 √
Let X be a random variable with pdf fX (x) = x for 0 < x < 2 and 0
1
otherwise. Let Y = X 4 . Find the pdf and the cdf of Y.
Solution. √ √
The cdf of X is FX (x) = x2 for 0 ≤ x ≤ 2 and 1 for x > 2. Thus,
1 1
FY (y) = FX (y 4 ) = y 8 for 0 ≤ y ≤ 2 8 and 1 for y ≥ 2 8 . The pdf of Y is
1
fY (y) = 4y 3 fX (y 4 ) = 8y 7 for 0 < y < 2 8 and 0 otherwise
Example 19.2
Let X have the beta distribution with pdf
Γ(α + β) α−1
fX (x) = x (1 − x)β−1 , 0 < x < 1
Γ(α)Γ(β)
1
and 0 otherwise. Find the pdf of Y = X τ where τ > 0.
Solution.
We have
Γ(α + β) τ (α−1)
fY (y) = −τ y τ −1 fX (y τ ) = −τ y τ −1 y (1 − y τ )β−1 , 0 < y < 1
Γ(α)Γ(β)
and 0 otherwise
Let X be a continuous random variable and Y = eX . That is, Y has a

lognormal distribution.
Theorem 19.2
Let X be a continuous random variable with pdf fX (x) and cdf FX (x) such
that fX (x) > 0 for all x ∈ R. Let Y = eX . Then, for y > 0, we have
FY (y) = FX (ln y) and fY (y) = y1 fX (ln y).

Proof.
We have
FY (y) = Pr(Y ≤ y) = Pr(eX ≤ y) = Pr(X ≤ ln y) = FX (ln y).
Differentiating with respect to y, we find

1
fY (y) = fX (ln y)
y
Example 19.3
Let X be a normal distribution with parameters µ = 1 and σ 2 = 4. Define
the random variable Y = eX .
(a) Find E(Y ).
(b) The 95th percentile of the standard normal distribution is 1.645. Find
the 95th percentile of Y.
Solution.
(a) We have

2
Z ∞ (x−1)
1 −
ex √ e
8
E(Y ) = dx
−∞ 2 2π
Z ∞
1 1 2
=e3 √ e− 8 (x−5) dx
−∞ 2 2π
=e3 · 1 = 20.086.
Note that the last integral is the integral of the density function of the
normal distribution with parameters µ = 5 and σ 2 = 4.
(b) Let π0.95 be the 95th percentile of Y. Then Pr(Y ≤ π0.95 ) = 0.95. Thus,

X −1 ln π0.95 − 1
0.95 = Pr(Y ≤ π0.95 ) = Pr(X ≤ ln 0.95) = Pr ≤ .
2 2
Hence,
ln π0.95 − 1
= 1.645 =⇒ π0.95 = 72.97
2
19 POWERS AND EXPONENTIATION OF RANDOM VARIABLES 151
Practice Problems
Problem 19.1
Let X be the exponential distribution with parameter θ. Determine the pdf
and the cdf of the transformed, inverse transformed, and inverse exponential
distribution.
Problem 19.2
Find the cdf of the inverse of a Pareto distribution with parameters α and
θ. What’s the name of the new distribution and its parameter(s)?
Problem 19.3
Let X be a random variable with pdf fX (x) = 2x for 0 < x < 1 and 0
otherwise.
Find the pdf of Y = X −1 .
Problem 19.4
Find the pdf of the inverse of the Gamma distribution with parameters α
and θ = 1.
Problem 19.5
1
Let X have a uniform distribution in (0, b). Find the pdf of Y = X τ , with
τ > 0.
Problem 19.6
X

Let X have a Pareto distribution with parameters α and θ. Let Y = ln 1 + θ .
Determine the name of the distribution of Y and its parameter(s).
Problem 19.7
Let X have the normal distribution with parameters µ and σ 2 . Find the pdf
and cdf of Y = eX .
Problem 19.8
Let X have a uniform distribution in (0, b). Find the pdf of Y = eX .
Problem 19.9
Let X have an exponential distribution with parameter θ. Find the pdf of
Y = eX .
Problem 19.10 ‡
You are given:
(i) X has a Pareto distribution with parameters α = 2 and θ = 100.
(ii) Y = ln 1 + xθ .
Calculate the variance of Y.

20 CONTINUOUS MIXING OF DISTRIBUTIONS 153
20 Continuous Mixing of Distributions

In Section 16 we introduced the concept of discrete mixture of distributions.
In this section we will define the continuous version of mixing where the dis-
crete probabilities are replaced by the pdf of a continuous random variable.
In actuarial terms, continuous mixtures arise when a risk parameter from
the loss distribution is uncertain and the uncertain parameter is continuous.
Suppose that Λ is a continuous random variable with pdf fΛ (λ). Let X

be a continuous random variable that depends on a parameter λ. We say
that X is a mixture of distributions if its pdf is
Z ∞
fX (x) = fX|Λ (x|λ)fΛ (λ)dλ.
−∞
The distribution fX (x) is a marginal density function: Let f (x, λ) be the

joint distribution of X and Λ. Then fX (x) and fΛ (λ) are referred to as
marginal densities. Then the product fX|Λ (x|λ)fΛ (λ) is the joint den-
sity function and the marginal density function can be recovered by inte-
grating the joint density function(See Section 33 of [2].)
Theorem 20.1
For the mixture
R ∞ distribution X as defined above, we have
(a) FX (x) = −∞ FX|Λ (x|λ)fΛ (λ)dλ.
(b) E(X k ) = E[E(X k |Λ)].
(c) Var(X) = E[Var(X|Λ)] + Var[E(X|Λ)].
Proof.
(a) Assuming that the order of integration can be reversed, we have
Z x Z ∞ Z ∞Z x
FX (x) = fX|Λ (t|λ)fΛ (λ)dλdt = fX|Λ (t|λ)fΛ (λ)dtdλ
−∞ −∞ −∞ −∞
Z ∞
= FX|Λ (x|λ)fΛ (λ)dλ.
−∞
(b) We have
Z ∞ Z ∞
E(X k ) = xk fX|Λ (x|λ)fΛ (λ)dλdx
−∞ −∞
Z ∞ Z ∞
k
= x fX|Λ (x|λ)dx fΛ (λ)dλ
−∞ −∞
Z ∞
= E(X k |Λ)fΛ (λ)dλ = E[E(X k |Λ)].
−∞
(c) We have
Var(X) =E(X 2 ) − (E(X))2

=E[E(X 2 |Λ)] − {E[E(X|Λ)]}2
=E{Var(X|Λ) + [E(X|Λ)]2 } − {E[E(X|Λ)]}2
=E[Var(X|Λ)] + E{[E(X|Λ)]2 } − {E[E(X|Λ)]}2
=E[Var(X|Λ)] + Var[E(X|Λ)]
Mixture distributions tend to be heavy-tailed as seen in the next example.
Example 20.1
1
The distribution of X|Λ is exponential with parameter Λ. The distribution
of Λ is Gamma with parameters α and θ. Find fX (x).
Solution.
We have
∞
θα α−1 −λθ
Z
fX (x) = λe−λx λ e dλ
0 Γ(α)
∞
θα
Z
= λα e−λ(x+θ) dλ
Γ(α) 0
θα Γ(α + 1) αθα
= = .
Γ(α) (x + θ)α+1 (x + θ)α+1
This is the distribution of a Pareto random variable and we know that this
distribution is heavy-tailed
Example 20.2
Suppose that X|Λ has a normal distribution with parameters λ and σ1 . That
is,
(x−λ)2
1 − 2
fX|Λ (x|λ) = √ e 2σ1
, − ∞ < x < ∞.
σ1 2π
Suppose that Λ has a normal distribution with parameters µ and σ2 . That
is,
(λ−µ)2
1 − 2
fΛ (λ) = √ e 2σ2
, − ∞ < x < ∞.
σ2 2π
Determine the unconditional pdf of X.
Solution.
We first establish the following identity:
2
x−λ 2 λ−µ 2
2
σ1 + σ22 σ22 x + µσ12 (x − µ)2

+ = λ − + .
σ1 σ2 σ12 σ22 σ12 + σ22 σ12 + σ22
By completing the square, we find
x−λ 2 λ − µ 2 x2 − 2λx + λ2 λ2 − 2µλ + µ2

+ = +
σ1 σ2 σ12 σ22
2 2 σ22 x + σ12 µ σ 2 x 2 + σ 2 µ2

2 σ1 + σ2
=λ 2 2 − 2λ 2 2 + 2 2 21
σ1 σ2 σ1 σ2 σ1 σ2
2 2
σ1 + σ22 σ22 x + µσ12 (σ22 x + µσ12 )2 σ22 x2 + σ12 µ2

= λ − − +
σ12 σ22 σ12 + σ22 σ12 σ22 (σ12 + σ22 σ12 σ22
2 2
σ1 + σ22 σ22 x + µσ12 σ12 σ22 x2 − 2µxσ12 σ22 + σ12 σ22 µ2

1
= λ− + 2 2
σ12 σ22 σ12 + σ22 σ1 σ2 σ12 + σ22
2 2
σ1 + σ22 σ22 x + µσ12 (x − µ)2

= 2 2 λ − 2 2 + 2 .
σ1 σ2 σ1 + σ2 σ1 + σ22
Now, the marginal density function of X is
Z ∞ (x−λ)2 (λ−µ)2
1 − 2 1 − 2
fX (x) = √ e 2σ1
√ e 2σ2 dλ
−∞ σ1 2π σ2 2π
2 2
Z ∞ 1 x−λ λ−µ
1 − 2 σ1
+ σ2
= e dλ
2πσ1 σ2 −∞
(x−µ)2
− 2 +σ 2 ) Z ∞s 2 " 2 #
σ1 + σ22 σ12 + σ22 σ22 x + µσ12
2(σ1

e 2
=p exp − λ− dλ.
2π(σ12 + σ22 ) −∞ 2πσ12 σ22 2σ12 σ22 σ12 + σ22
The integrand in the last integral is the pdf of a normal distribution with
σ 2 x+σ 2 µ σ2 σ2
parameters 1σ2 +σ22 and σ21+σ22 so that the integral is 1. Hence,
1 2 1 2
(x−µ)2
− 2 +σ 2 )
2(σ1
e 2
fX (x) = p
2π(σ12 + σ22 )
which is the pdf of the normal distribution with parameters µ and σ12 + σ22
Example 20.3 ‡
The scores on the final exam in Ms. B’s Latin class have a normal distribu-
tion with mean θ and standard deviation equal to 8. θ is a random variable
with a normal distribution with mean equal to 75 and standard deviation

equal to 6.
Each year, Ms. B chooses a student at random and pays the student 1 times
the student’s score. However, if the student fails the exam (score ≤ 65),
then there is no payment.
Calculate the conditional probability that the payment is less than 90, given
that there is a payment.
Solution.
Let S denote the scores. Since S|Θ and Θ are normal, S is also normally
distributed with mean
E(S) = E[E(S|Θ)] = E(Θ) = 75
and variance
Var(S) = E[Var(S|Θ)] + Var[E(S|Θ)] = 64 + 36 = 100.
We want
Pr(65 < S < 90) FS (90) − FS (65)

Pr(S < 90|S > 65) = =
Pr(S > 65) 1 − FS (65)
90−75 65−75

Φ 10 − Φ 10
=
1 − Φ 65−75

10
Φ(1.5) − Φ(−1.0) Φ(1.5) − [1 − Φ(1)]
= =
1 − Φ(−1.0) Φ(1)
0.9332 − (1 − 0.8413)
= = 0.9206
0.8413
Example 20.4
Let X|Λ have a Poisson distribution with parameter λ. Let Λ have a Gamma
distribution with parameters α and β. That is,
−λ
λα−1 e β
fΛ (λ) = α .
β Γ(α)
Find the unconditional probability Pr(X = 1).

Solution.
We have8
Z ∞
Pr(X = 1) = Pr(X = 1|Λ)fΛ (λ)dλ
0
λ
Z ∞ α−1 e− β
−λ λ
= λe dλ
0 β α Γ(α)
Z ∞
1 1
α −λ 1+ β
= α λ e dλ
β Γ(α) 0
−(α+1) λ
− −1
1 + β1 Γ(α + 1) Z ∞ λ(α+1)−1 e (1+ β1 )
= dλ.
β α Γ(α) −(α+1)
0
1 + β1 Γ(α + 1)
The integrand is the pdf of a Gamma distribution with parameters α + 1

−1
and 1 + β1 so that the integral is 1. Hence,
−(α+1)
1
1+ β Γ(α + 1) αβ
Pr(X = 1) = =
β α Γ(α) (1 + β)α+1
Example 20.5 ‡
Bob is a carnival operator of a game in which a player receives a prize worth
W = 2N if the player has N successes, N = 0, 1, 2, · · · , Bob models the
probability of success for a player as follows:
(i) N has a Poisson distribution with mean Λ.
(ii) Λ has a uniform distribution on the interval (0, 4).
Calculate E[W ].
Solution.
We know that PN (z) = eλ(z−1) = E(Z N ). In particular, E(W |Λ) = PN (2) =
eλ . Thus,
Z 4
1 4 λ
Z
E(W ) = E(W |Λ)fΛ (λ)dλ = e dλ = 13.4
0 4 0
8
See P. 380 of [2]
Practice Problems
Problem 20.1
Let X be a loss random variable having a Pareto distribution with parame-
ters α and Θ. The parameter Θ is uniformly distributed in (0, b).
Find the unconditional variance of X.
Problem 20.2
Let X|Θ be the inverse exponential random variable with parameter Θ. Its
pdf is
1
fX|Θ (x|θ) = 2 θe−θx , x > 0
x
and 0 otherwise. Let Θ have the exponential distribution with mean 4.
Determine the unconditional distribution of X.
Problem 20.3
Let X|Λ have the pdf
fX|Λ (x|λ) = λe−x + (1 − λ)xe−x , x > 0
and 0 otherwise.
Find the unconditional variance of X given that E(Λ) = 0.45.
Problem 20.4
Let X|Λ have a Poisson distribution with parameter Λ. Let Λ have a Gamma
distribution with parameters α and β. That is,
−λ
λα−1 e β
fΛ (λ) = α .
β Γ(α)
Determine the expectation of Λ and then the unconditional expectation of

X.
Problem 20.5
γ
Suppose that X|Λ has a Weibull distribution with cdf FX|Λ (x|λ) = 1−e−λx
for x ≥ 0. Suppose that Λ is exponentially distributed with mean θ.
Find the unconditional cdf of X.

Problem 20.6
Let N have a Poisson distribution with mean Λ. Let Λ have a uniform dis-
tribution on the interval (0,5).
Determine the unconditional probability that N ≥ 2.
Problem 20.7
Let N |Λ have a negative binomial distribution with r = 1 and Λ. Let Λ have
a Gamma distribution with α = 1 and θ = 2.
Find the unconditional variance of N.
Problem 20.8 ‡
A claim count distribution can be expressed as a mixed Poisson distribu-
tion. The mean of the Poisson distribution is uniformly distributed over the
interval [0, 5].
Calculate the probability that there are 2 or more claims.
Problem 20.9 ‡
The length of time T , in years, that a person will remember an actuarial
statistic is modeled by an exponential distribution with mean Y1 . In a certain
population, Y has a gamma distribution with α = θ = 2.
Calculate the probability that a person drawn at random from this pop-
ulation will remember an actuarial statistic less than 12 year.
21 Frailty (Mixing) Models

A continuous mixture model that arises within the context of survival anal-
ysis is the frailty model. We will discuss mainly the mathematical com-
ponents of this model so that the focus remains on the importance of gener-
erating new distributions by mixing.
A frailty model is defined as follows: Let X|Λ be a random variable with

conditional hazard function given by
hX|Λ (x|λ) = λa(x)
where a(x) is some specified and known function of x. The frailty random
variable Λ is supposed to have positive support.
Theorem 21.1
Rx
Let A(x) = 0 a(z)dz. We have:
(a) The conditional survival function of X given Λ is
SX|Λ (x|λ) = e−λA(x) .
(b) The unconditional survival function of X is
SX (x) = MΛ (−A(x)).
Proof.
(a) Recall that a survival function can be recovered from the hazard rate
function so that we have
Rx Rx
SX|Λ (x|λ) = e− 0 hX|Λ (z|λ)dz
= e− 0 λa(z)dz
= e−λA(x) .
(b) The unconditional survival function is
Z
SX (x) = SX|Λ (x|λ)fΛ (λ)dλ = E[SX|Λ (x|λ)] = E[e−λA(x) ] = MΛ (−A(x))
λ
Remark 21.1
If X|Λ has an exponential distribution in the frailty model, a(x) will be 1,
and A(x) will be x. When X|Λ has Weibull distribution in the frailty model,
a(x) will be γxγ−1 , and A(x) will be xγ .
Example 21.1
Let X|Λ have a Weibull distribution with conditional survival function
γ
SX|Λ (x|λ) = e−λx . Let Λ have a Gamma distribution with parameters α
and θ. Find the unconditional or marginal survival function of X.
21 FRAILTY (MIXING) MODELS 161
Solution.
We first find the moment generating function of Λ. We have
Z ∞
1 x
tX
E(e ) = α etx xα−1 e− θ dx
θ Γ(α) 0
Z ∞
1 1
= α xα−1 e−x(−t+ θ ) dx
θ Γ(α) 0
(−t + 1θ )−α e−y
Z ∞ α−1
y
= dy
0 θα Γ(α)
−α
−t + 1θ Γ(α) 1
= α
= (1 − θt)−α , t < .
θ Γ(α) θ
Also, we know that A(x) = xγ . Hence,
SX (x) = MΛ (−A(x)) = (1 + θxγ )−α .
This is the survival function of a Burr distribution with parameters α and

−1
θ γ
Example 21.2
A continuous mixture is used in a frailty model with frailty random variable
1
Λ, such that a(x) = x+1 , x > 0. Find the conditional survival function of
X.
Solution.
We first find A(x) :
Z x
dt
A(x) = = ln (1 + x).
0 1+t
Thus,
1
SX|Λ (x|λ) = e−λA(x) = e−λ ln (1+x) =
(1 + x)λ
Example 21.3
Given that the marginal survival function in a frailty model is SX (x) =
x−2 , x ≥ 1. Let the frailty random variable have an exponential distribution.
Determine MΛ (x).
Solution.
Since Λ have an exponential distribution, A(x) = x. Thus, x−2 = SX (x) =
MΛ (−A(x)) = MΛ (−x). Hence, MΛ (x) = x−2 , x ≥ 1
Example 21.4
The marginal survival function in a frailty model is given to be SX (x) =
1
2 α MΛ (x). The frailty random variable Λ is a Gamma random variable with
parameters α and θ. Determine A(x).
Solution.
1
The moment generating function of Λ is Mλ (x) = (1 − θx)−α . Thus, 2 α (1 −
θx)−α = SX (x) = MΛ (−A(x)) = (1 + θA(x))−α . Hence, 2(1 − θx) = 1 +
θA(x). Solving for A(x), we find A(x) = 2−θx
θ
Example 21.5
Consider a frailty model where X|Θ has an exponential distribution with
conditional hazard rate function hX|Θ (x|θ) = θ. The frailty random variable
Θ has a uniform distribution in (1, 11). Find the conditional survival function
of X.
Solution.
We have
Rx Rx
SX|Θ (x|θ) = e− 0 hX|θ (z|θ)dz
= e− 0 θdz
= e−θx , x > 0
Example 21.6
Consider the exponential-inverse Gaussian frailty model with
θ
a(x) = √ , θ > 0.
2 1 + θx
Determine A(x).
Solution.
We have
x x √ x √
Z Z
θ
A(x) = a(t)dt = √ dt = 1 + θt = 1 + θx − 1

0 0 2 1 + θt 0
21 FRAILTY (MIXING) MODELS 163
Practice Problems
Problem 21.1
Let X|Λ have an exponential distribution with conditional survival function
SX|Λ (x|λ) = e−λx . Let Λ have a Gamma distribution with parameters α and
θ.
Find the unconditional or marginal survival function of X.
Problem 21.2
A continuous mixture is used in a frailty model with frailty random variable
1
Λ, such that a(x) = x+1 , x > 0. The frailty random variable has a uniform
distribution in (0, 1).
Find the marginal survival function of X.
Problem 21.3
Given that the marginal survival function in a frailty model is SX (x) =
x
e− θ , x ≥ 0. Let the frailty random variable have an exponential distribu-
tion.
Determine MΛ (x).
Problem 21.4
The marginal survival function in a frailty model is given to be SX (x) =
1
2 α MΛ (x). The frailty random variable Λ is a Gamma random variable with
parameters α and θ.
Determine a(x).
Problem 21.5
The probability generating function of the frailty random variable is PΛ (x) =
ex−1 . Suppose X|Λ has an exponential distribution.
Problem 21.6
Consider a frailty model where X|Θ has an exponential distribution with
conditional hazard rate function hX|Θ (x|θ) = θ. The frailty random variable
Θ has a uniform distribution in (1, 11).

Problem 21.7
Suppose that X|Λ has the Weibull distribution with conditional survival
γ
function SX|Λ (x|λ) = e−λx , x ≥ 0. The frailty random variable Λ has an
exponential distribution with mean θ.
Problem 21.8
θ
a(x) = √ , θ > 0.
2 1 + θx
Determine the conditional survival function SX|Λ (x|λ).
Problem 21.9
θ
a(x) = √ , θ > 0.
2 1 + θx
Suppose Λ has a Gamma distribution with parameters 2α and θ = 1.
Determine the marginal survival function of X.
Problem 21.10
Determine the unconditional probability density function of frailty distribu-
tion.
Problem 21.11
Determine the unconditional hazard rate function of frailty distribution.
22 SPLICED DISTRIBUTIONS 165
22 Spliced Distributions
A spliced distribution consists of different distributions one for each part
of the domain of the random variable. For example, a n-component
spliced distribution has the following pdf


 α1 f1 (x), c1 < x < c2
 α2 f2 (x),

c2 < x < c3
fX (x) = ..


 .
αn fn (x), cn−1 < x < cn

where each αj > 0, α1 + α2 + · · · + αn = 1, and each fi (x) is a legitimate

pdf with support (ci , ci+1 ).
An interpretation in insurance claims is that the distributions vary by size

of claims.
Example 22.1
Show that the pdf below is a spliced distribution of some random variable
X and then find E(X).

0.01, 0 ≤ x < 50
f (x) =
0.02, 50 ≤ x ≤ 75
and 0 otherwise.
Solution.
For the given pdf, we have f1 (x) = 0.02 for 0 ≤ x < 50; f2 (x) = 0.04
for 50 ≤ x ≤ 75; α1 = α2 = 0.5. Note that f1 (x) and f2 (x) are uniform
densities. The mean of X is
Z 75 Z 50 Z 75
E(X) = xf (x)dx = 0.01xdx + 0.02xdx = 43.75
0 0 50
Example 22.2
Let g1 (x) and g2 (x) be two nonnegative functions defined on the intervals
(c1 , c2 ) and [c2 , c3 ) respectively. Create a pdf of a 2-component spliced
distribution with support (c1 , c3 ).
Solution.
Let f1 (x) = R c g1 (x) , f2 (x) = R c g2 (x) , α1 = α > 0, α2 = 1 − α where α
2 2
c g1 (x)dx
1 c g2 (x)dx
1
is some positive constant. Then

αf1 (x), c1 < x < c2
f (x) =
(1 − α)f2 (x), c2 ≤ x < c3
Example 22.3 ‡
An actuary for a medical device manufacturer initially models the failure
time for a particular device with an exponential distribution with mean 4
years. This distribution is replaced with a spliced model whose density
function:
(i) is Uniform over [0,3]
(ii) is proportional to the initial modeled density function after 3 years
(iii) is continuous.
Calculate the probability of failure in the first 3 years under the revised
distribution.
Solution.
The two-spliced pdf is
α

f (x) = 3, 0<x<3
(1 − α)e [0.25e−0.25x ],
0.75 x≥3
Using continuity, we have

α1 α
lim = lim 0.25(1 − α)e0.75 e−0.25x =⇒ = 0.25(1 − α).
x→3− 3 x→3+ 3
Solving, we find α = 37 . Finally,

Z 3
3 dx 3
Pr(X ≤ 3) = =
0 7 3 7
22 SPLICED DISTRIBUTIONS 167
Practice Problems
Problem 22.1
Find the density function of a random variable that is uniform on (0, c) and
exponential thereafter.
Problem 22.2 ‡
Suppose a loss distribution is a two-component spliced model with:
(i) a Weibull distribution having parameters θ1 = 1500 and τ = 1 for losses
up to $4,000; and
(ii) a Pareto distribution having parameters θ2 = 12000 and α = 2 for losses
$4,000 and up.
The probability that losses are less than $4,000 is 0.60.
Calculate the probability that losses are less than $25,000.
Problem 22.3
A random variable X follows a continuous two-component spliced distribu-
tion that is uniform in (0, 3] and exponential (with mean 1) thereafter. Find
the 95th percentile of X.
Problem 22.4
Using the results of the previous problem, find E(X|X > π0.95 ).
Problem 22.5
Write the density function for a 2-component spliced model in which the
density function is proportional to a uniform density over the interval from
0 to 1000 and is proportional to an exponential density function from 1000
to ∞. Ensure that the resulting density function is continuous.
Problem 22.6
The pdf of two-component spliced distribution is given below.

0.01, 0 ≤ x < 50
f (x) =
0.02, 50 ≤ x ≤ 75
and 0 otherwise. Find the variance of X.

23 Limiting Distributions
In addition to the methods described in the previous sections, we can obtain
new distributions as limiting cases of other ones. This is accomplished by
letting the parameters go to either infinity or zero.
Example 23.1
For a Pareto distribution with parameters α and θ, let both α and θ go
to infinity with the ratio αθ → ξ held constant. Show that the result is an
exponential distribution.
Solution.
Let ξ = αθ so that α = ξθ. Substituting into the cdf of the Pareto distribution,
we find ξθ
θ
FX (x) = 1 − .
x+θ
ξθ
θ
Let w = x+θ . We have
lim ln w = lim ξθ[ln θ − ln (x + θ)]

θ→∞ θ→∞
ln θ − ln (x + θ)
=ξ lim
θ→∞ θ−1
d
dθ [ln θ − ln (x + θ)]
=ξ lim d −1
dθ (θ )
θ→∞
θ−1 − (x + θ)−1
=ξ lim
θ→∞ −θ−2
xθ
= − ξ lim = −ξx.
θ→∞ x + θ
It follows that limθ→∞ w = e−ξx and limθ→∞ FX (x) = 1 − e−ξx which is the
cdf of an exponential distribution with mean 1ξ
Example 23.2
For a transformed beta distribution with parameters α, γ and θ, let both
α and θ go to infinity with the ratio θ1 → ξ. Show that the result is a
αγ
transformed gamma distribution.
Solution.
For large α, Stirling’s formula gives
1 1
Γ(α) ≈ e−α αα− 2 (2π) 2 .
23 LIMITING DISTRIBUTIONS 169
1
θ
Also, we let ξ = 1 so that θ = ξα γ .
αγ
Using this and Stirling’s formula in the pdf of a transformed beta distribu-
tion, we find
Γ(α + τ )γxγτ −1
fX (x) =
Γ(α)Γ(τ )θγτ (1 + xγ θ−γ )γ+τ
1 1
e−α−τ (α + τ )α+τ − 2 (2π) 2 γxγτ −1
≈ 1 1 1
e−α (α)α− 2 (2π) 2 Γ(τ )(α γ )γτ (1 + xγ ξ −γ α−1 )γ+τ
1
e−τ [(α + τ )/α]α+τ − 2 γxγτ −1
= .
Γ(τ )ξ γτ [1 + (x/ξ)γ /α]α+τ
Now, let
τ α+τ − 21
w1 = 1 + .
α
We have
1 τ
lim ln w1 = lim (α + τ − ) ln 1 +
α→∞ α→∞ 2 α
τ
ln 1 + α
= lim
α→∞ (α + τ − 1 )−1
2
−2
−1
−τ α 1 + ατ
= lim
α→∞ −(α + τ − 1 )−2
2
1 2

τ −1 τ
=τ lim 1 + 1+ −
α→∞ α α 1α
=τ.
Thus, limα→∞ w1 = eτ . Now, let
α+τ
(x/ξ)γ

w2 = 1 + .
α
We have
(x/ξ)γ

lim ln w2 = lim (α + τ ) ln 1 +
α→∞ α→∞ α
h i
(x/ξ)γ
ln 1 + α
= lim
α→∞ (α + τ )−1
γ −1
h i
1 + (x/ξ) −(x/ξ)γ α−2

α
= lim
α→∞ −(α + γ)−2
(x/ξ)γ −1

γ
τ 2
=(x/ξ) lim 1 + 1+
α→∞ α α
γ
=(x/ξ) .
Hence,
γ
lim w2 = e(x/ξ) .
α→∞
Finally, γ
x
−
γxγτ −1 e ξ
lim fX (x) =
α→∞ Γ(τ )ξ γτ
which is the pdf of the transformed gamma distribution
23 LIMITING DISTRIBUTIONS 171
Practice Problems
Problem 23.1
Show:
1 h iα+τ ξ γ
α α+τ − 2 (ξ/x)γ
= eα = e( x ) .

limτ →∞ 1 + τ and limτ →∞ 1 + τ
Problem 23.2
For a transformed beta distribution with parameters α, γ and θ, let θ go to
1
infinity with the ratio θτ γ → ξ.
Show that the result is the inverse transformed Gamma distribution.
Problem 23.3
For an inverse Pareto distribution with parameters τ and θ, let θ → 0, τ →
∞ and τ θ → ξ.
Show that the result is the inverse exponential distribution.

24 The Linear Exponential Family of Distributions

A random variable X is said to belong to the linear exponential family
(LEF) if its pdf can be reformulated in terms of a parameter θ as
p(x) r(θ)x
f (x, θ) = e
q(θ)
where p(x) depends solely of x. The support of X must not depend on θ.
Other parameters of the distribution (parameters that are not θ) may occur
in the expressions p(x), q(θ), and r(θ). But they have no role in determining
whether the distribution belongs to the linear exponential family.
Example 24.1
Show that a normal distribution with parameters µ and σ 2 belongs to the
linear exponential family with θ = µ.
Solution.
The pdf of the normal distribution function can be written as
1 1 x−µ 2
f (x, µ) = √ e− 2 ( σ )
σ 2π

2 2
1 − 12 x −2µx+µ
σ2
= √ e
σ 2π
2
1 − x 2 µ2 x
= √ µ 2 e 2σ e σ .
σ 2πe 2σ2
x2
Thus, X belongs to the linear exponential family with p(x) = √1 e− 2σ2 , q(µ) =
σ 2π
µ2
µ
e 2σ 2 , and r(µ) = σ2
Example 24.2
Let X belong in the linear exponential family. Find an expression of E(X).
Solution.
Taking the logarithm of f (x, θ) we find
ln f (x, θ) = ln p(x) + r(θ)x − ln q(θ).
Now, differentiating both sides with respect to θ, we find

1 ∂f q 0 (θ) ∂f q 0 (θ)
(x, θ) = xr0 (θ)− =⇒ (x, θ) = r0 (θ)f (x, θ)x− f (x, θ).
f (x, θ) ∂θ q(θ) ∂θ q(θ)
24 THE LINEAR EXPONENTIAL FAMILY OF DISTRIBUTIONS 173
Now we integrate both sides with respect to x to obtain

Z Z Z 0
∂f 0 q (θ)
(x, θ)dx = xr (θ)f (x, θ)dx − f (x, θ)dx.
∂θ q(θ)
By the definition of the family, the support of X is independent of θ and so
is the range of x. Thus, we can write
q 0 (θ)
Z Z Z
∂
f (x, θ)dx =r0 (θ) xf (x, θ)dx − f (x, θ)dx
∂θ q(θ)
∂ q 0 (θ)
(1) =r0 (θ)E(X) − .
∂θ q(θ)
That is,
q 0 (θ)
r0 (θ)E(X) − = 0.
q(θ)
Solving for E[X] yields
q 0 (θ)
E(X) = = µ(θ)
r0 (θ)q(θ)
Example 24.3
Let X belong in the linear exponential family. Find the variance of X.
Solution.
From the previous example, we have
∂f
(x, θ) = [x − µ(θ)]r0 (θ)f (x, θ).
∂θ
Differentiating with respect to θ and using the already obtained first deriva-
tive of f (x, θ) yield
∂2f ∂f
2
(x, θ) =r00 (θ)[x − µ(θ)]f (x, θ) − µ0 (θ)r0 (θ)f (x, θ) + [x − µ(θ)]r0 (θ) (x, θ)
∂θ ∂θ
=r00 (θ)[x − µ(θ)]f (x, θ) − µ0 (θ)r0 (θ)f (x, θ) + [x − µ(θ)]2 [r0 (θ)]2 f (x, θ).
Now we integrate both sides with respect to x to obtain
Z 2
∂ f
(x, θ)dx = [r0 (θ)]2 Var(X) − r0 (θ)µ0 (θ).
∂θ2
By the definition of the family, the support of X is independent of θ and so
is the range of x. Thus, we can write
∂2
Z
v(x, θ)dx = [r0 (θ)]2 Var(X) − r0 (θ)µ0 (θ).
∂θ2
That is,
[r0 (θ)]2 Var(X) − r0 (θ)µ0 (θ) = 0.
Solving for Var(X) yields
µ0 (θ)
Var(X) =
r0 (θ)
Example 24.4
Let X be the normal distribution with θ = µ = 24 and σ = 3. Use the
formulas of this section to verify that E(X) = 24 and Var(X) = 9.
Solution.
x2 θ2
For the given distribution, we have p(x) = √1 e− 2σ2 , q(θ) = e 2σ2 , and
σ 2π
θ
r(θ) = σ2
. Thus,
θ2
θ 2σ2
q 0 (θ) σ2
e
E(X) = 0 = θ2
= 24.
r (θ)q(θ) 1 2σ2
e
σ2
Likewise, we have
µ0 (θ) 1
Var(X) = 0
= 1 =9
r (θ) 9
24 THE LINEAR EXPONENTIAL FAMILY OF DISTRIBUTIONS 175
Practice Problems
Problem 24.1
Show that the Gamma distributions belongs to the linear exponential family.
Problem 24.2
Show that the Poisson distribution with parameter λ belongs to the linear
exponential family.
Problem 24.3
Show that the binomial distribution with m trials and parameter p belongs
to the linear exponential family.
Problem 24.4
Use the formulas of this section to find the mean and the variance of the
Poisson distribution.
Problem 24.5
Use the formulas of this section to find the mean and the variance of the
binomial distribution.
Discrete Distributions
The distributions and the families of distributions that we have been dis-
cussing so far are mainly used to describe the amount of risks. Next, we turn
our attention to distributions that describe the number of risks or claims.
In this chapter, we introduce classes of counting distributions. By a count-
ing distribution we mean a discrete distribution with support a subset of
N ∪ {0}. We will adopt the following notation: If N is the random variable
representing the number of events (or claims) then the probability mass
function or the probability function Pr(N = k) will be denoted by pk .
177
178 DISCRETE DISTRIBUTIONS
25 The Poisson Distribution

The first counting distribution that we consider is the Poisson distribution.
A random variable X is said to be a Poisson random variable with pa-

rameter λ if its probability mass function has the form
e−λ λk
pk = , k = 0, 1, 2, · · · .
k!
Note that
∞ ∞
X
−λ
X λk
pk = e = e−λ eλ = 1.
k!
k=0 k=0
The Poisson random variable is most commonly used to model the number
of random occurrences of some phenomenon in a specified unit of space or
time. For example, the number of phone calls received by a telephone op-
erator in a 10-minute period or the number of typos per page made by a
secretary.
The probability generating function of N is

∞ ∞
X X (zλ)k
PN (z) = E(z N ) = pk z k = e−λ = eλ(z−1)
k!
k=0 k=0
and the moment generating function is

t −1)
MN (t) = PN (et ) = eλ(e .
Using the pgf, we can find the mean and the variance:
E(N ) =PN0 (1) = λ

E[N (N − 1)] =PN00 (1) = λ2
Var(N ) =E[N (N − 1)] + E(N ) − [E(N )]2
=λ2 + λ − λ2 = λ.
We next discuss two useful properties of Poisson distributions. The first

concerns the sum of independent Poisson random variables.
Theorem 25.1
Let N1 , N2 , · · · , Nn be n independent Poisson random variables with param-
eterd λ1 , λ2 , · · · , λn respectively. Then the random variable S = N1 + N2 +
· · · + Nn is also a Poisson distribution with parameters λ1 + λ2 + · · · + λn .
25 THE POISSON DISTRIBUTION 179
Proof.
Since N1 , N2 , · · · , Nn are independent so are etN1 , etN2 , · · · , etNm . Using the
fact that the expectation of a product of independent random variables is
the product of the individual expectations, we have
MS (t) =E(et(N1 +N2 +···+Nn ) = E[etN1 etN2 · · · etNn ]

=E[etN1 ]E[etN2 ] · · · E[etNn ]
t −1) t −1) t −1)
=eλ1 (e eλ2 (e · · · eλn (e
t −1)
=eλ(e
where λ = λ1 +λ2 +· · ·+λn . Since the moment generating function uniquely

determines the distribution, S has a Poisson random variable with parame-
ter λ
The second result is very useful when modeling insurance losses or claims
where claims have classification. This result is known as the decomposi-
tion property of Poisson distribution.
Theorem 25.2
Let the total number of events N be a Poisson random variable with mean
λ. Suppose that the events can be classified into independent types such
as Type 1, Type 2,· · · , Type m with probabilities pk , k = 1, · · · , m. Let
Nk be the random variable representing the number of events of Type
k, k = 1, · · · , m. Then N1 , N2 , · · · , Nm are mutually independent Poisson
distributions with means λp1 , λp2 , , · · · , λpm respectively.
Proof.
Given n = n1 +n2 +· · ·+nm , the conditional joint distribution of N1 , N2 , · · · , Nm
is a multinomial distribution with parameters n, p1 , p2 , · · · , pm . Its condi-
tional pmf is
n!
Pr(N1 = n1 , N2 , n2 , · · · , Nm |N = n) = pn1 pn2 · · · pnmm .
n1 !n2 ! · · · nm ! 1 2
(See p.155 of [2]).

Now since N has a Poisson distribution with mean λ, we have
e−λ λn
Pr(N = n) = .
n!
The unconditional joint distribution of N1 , N2 , · · · , Nm is

Pr(N1 = n1 , N2 = n2 , · · · , Nm = nm ) =
Pr(N1 = n1 , N2 = n2 , · · · , Nm = nm |N = n)×Pr(N = n) =
n! e−λ λn
pn1 1 pn2 2 · · · pnmm .
n1 !n2 ! · · · nm ! n!
With some algebra, we find the following
Pr(N1 = n1 , N2 = n2 , · · · , Nm = nm ) =
pn1 1 pn2 2
· · · pnmm e−λ(p1 +p2 +···+pm ) λn1 +n2 +···+nm
=
n1 !n2 ! · · · nm !
(λp1 )n1 (λp2 )n2 · · · (λpm )nm e−λn1 e−λn2 · · · e−λnm
=
n1 !n2 ! · · · nm !
m λpi
Y e (λpi )ni
.
ni !
i=1
Next, the unconditional marginal pmf of Ni , i = 1, 2, · · · , m is

∞
X
Pr(Ni = ni ) = Pr(Ni = ni |N = n)Pr(N = n)
n=ni
∞
X e−λ λn
= C(n, ni )pni (1 − p)n−ni
n=ni
n!
∞
e−λ (λpi )ni X [λ(1 − pi )n−ni
=
ni ! n=n
(n − ni )!
i
e−λ (λpi )ni λ(1−pi )

= e
ni !
e−λ (λpi )ni
= .
ni !
Thus, Ni has a Poisson distribution with mean λpi . Since the joint dis-
tribution is the product of the marginal distributions, N1 , N2 , · · · , Nm are
mutually independent
Example 25.1
The number of patients that arrive at an emergency room during an hour
has a Poisson distribution with mean 6. Of every 50 patients that arrive to
the emergency, one has an appendicitis. Calculate the probability that the
number of patients with appendicitis that arrive to the emergency room in
a 24-hour period will exceed the expected number of appendicitis.
25 THE POISSON DISTRIBUTION 181
Solution.
Let N be the number of patients with appendicitis that arrive to the emer-
gency room in the 24 hour period. Then
1
E(N ) = 6 × 24 × = 2.88.
50
Thus,
e−2.88
Pr(N > 2.88) = 1−p0 −p1 −p2 = 1−e−2.88 −2.88e−2.88 −2.882 = 0.5494
2
Example 25.2
In a portfolio of insurance, a claim can be classified as Type A, Type B, or
Type C with probabilities 0.2, 0.3, and 0.5 respectively. Suppose that the
total number of claims is a Poisson random variable with mean 10. Each
type has a Poisson distribution and these random variables are supposed to
be independent. What is the probability that out of 5 claims, 2 are of Type
A?
Solution.
We have
Pr(NA = 2, NB + NC = 3)
Pr(NA = 2|N = 5) =
Pr(N = 5)
e−2 22 e−8 83
= 2! 3!
e−10 105
= 0.2048
5!
Practice Problems
Problem 25.1
The number of students in College Algebra that go to Math Help room has
a Poisson distribution with mean 2 per hour. The number of students in
Trigonometry that go to Math Help room has a Poisson distribution with
mean 1 per hour. The number of students in Pre-calculus that go to Math
Help room has a Poisson distribution with mean 0.5 per hour.
Calculate the probability that more than 3 students (in any class) go to
Math Help room between 2:00om and 4:pm.
Problem 25.2
Let N1 and N2 be two independent Poisson random variables with mean 2
and 3 respectively.
Calculate Pr(N1 + N2 = 1).
Problem 25.3
The number of monthly car wrecks in Russellville follows a Poisson distri-
bution with mean 10. There are three possible mutually exclusive causes of
accidents. The probability of a wreck due to icy road is 0.40. The proba-
bility of a wreck due to break malfunction is 0.45 and the probability of a
wreck due to driver’s error is 0.15.
What is the probability that exactly 5 wrecks will occur this month due
to driver’s error?
Problem 25.4
Suppose that the number of cars arriving for service at a service facility in
one week has a Poisson distribution with a mean of 20. Suppose that each
car is classified as either domestic or foreign. Suppose also that each time
a car arrives for service there is a 75% chance that it domestic and a 25%
chance that it is foreign.
What is the weekly expected number of foreign cars arriving at the service
facility?
26 THE NEGATIVE BINOMIAL DISTRIBUTION 183
26 The Negative Binomial Distribution

The Poisson distribution is a one-parameter counting distribution. The neg-
ative binomial distribution is a two-parameter counting distribution.
A random variable X is a negative binomial distribution if its pdf is of

the form
r n
n+r−1 1 β
pn = Pr(N = n) =
n 1 + β) 1 + β)
where n = 0, 1, 2, · · · and r > 0, β > 0. Note that r is a positive real number
not just a positive integer.
Example 26.1
Find the probability generating function of the negative binomial distribu-
tion.
Solution.
We have
∞ r n
N
X n+r−1
n 1 β
PN (z) =E(z ) = z
n 1 + β) 1+β
n=0
∞ r n
X n+r−1 1 zβ
=
n 1+β 1+β
n=0
r −r X∞
1 zβ n+r−1
= 1− (p∗ )r (1 − p∗ )n
1+β 1+β n
n=0
r −r
1 zβ
= 1− = [1 − β(z − 1)]−r
1+β 1+β
zβ
where p∗ = 1 − 1+β
Example 26.2
Find the mean and the variance of a negative binomial distribution N.
Solution.
We have
E(N ) =PN0 (1) = rβ
E[N (N − 1)] =PN00 (1) = r(r + 1)β 2
Var(N ) =E[N (N − 1)] + E(N ) − (E(N ))2
=rβ(1 + β).
Note that since β > 0, we have Var(N ) > E(N )
Example 26.3
Show that a negative binomial distribution is a mixture of a Poisson distri-
bution with a random parameter distributed Gamma.
Solution.
Suppose that N |Λ has a Poisson distribution with parameter λ. Suppose
that Λ has a Gamma distribution with parameters α and θ. We wish to find
the distribution of N. We have
Z ∞
pn =Pr(N = n) = Pr(N = n|Λ = λ)Pr(Λ = λ)dλ
0
λ
∞ Z ∞
e−λ λn α−1 e− θ
Z
λ 1 1 1
= α
dλ = α
e−λ(1+ θ ) λn+α−1 dλ
0 n! θ Γ(α) n! θ Γ(α) 0
λ
α n Z ∞ −
e (1+1/θ)−1 λα+n−1

n+α−1 1 θ
= dλ
n 1 + θ) 1 + θ) 0 (1 + 1/θ)−(α+n) Γ(α + n)
α n
n+α−1 1 θ
= .
n 1+θ 1+θ
This shows that the mixed Poisson, with a Gamma mixing distribution is
the same as the negative binomial distribution
Example 26.4 ‡
Glen is practicing his simulation skills. He generates 1000 values of the
random variable X as follows:
(i) He generates the observed value λ from the gamma distribution with
α = 2 and θ = 1 (hence with mean 2 and variance 2).
(ii) He then generates x from the Poisson distribution with mean λ.
(iii) He repeats the process 999 more times: first generating a value λ, then
generating x from the Poisson distribution with mean λ.
(iv) The repetitions are mutually independent.
Calculate the expected number of times that his simulated value of X is 3.
Solution.
By the previous result, X is a negative binomial distribution with r = α = 2
and β = θ = 1. From Table C, we have
r(r + 1)(r + 2)β 3 (2)(3)(4)13
p3 = = = 0.125.
3!(1 + β)r+3 3!(2)5
Thus we expect 1000p3 = 125 out of 1000 simulated values to be 3
Example 26.5
Show that a negative binomial distribution with r → ∞, n → 0, and rn → ξ
results in a Poisson distribution.
Solution.
Replace β in the pgf of N by rξ . We have

ξ
lim PN (z) = lim 1 − (z − 1)
r→∞ r→∞ r

ξ
=exp lim −r ln 1 − (z − 1)
r→∞ r
( )
ln 1 − ξ(z − 1)r−1

=exp lim −
r→∞ r−1
[1 − ξ(z − 1)r−1 ]ξ(z − 1)r−2

=exp lim
r→∞ r−2

rξ(z − 1)
=exp lim = eξ(z−1) .
r→∞ r − ξ(z − 1)
This is the pgf of a Poisson distribution with parameter ξ
Example 26.6 ‡
Actuaries have modeled auto windshield claim frequencies and have con-
cluded that the number of windshield claims filed per year per driver follows
the Poisson distribution with parameter Λ, where Λ follows the Gamma
distribution with mean 3 and variance 3. Calculate the probability that a
driver selected at random will file no more than 1 windshield claim next
year.
Solution.
We are given that αθ = 3 and αθ2 = 3. Thus, α = 3 and θ = 1. On the
other hand, the number of windshield is a negative binomial distribution
with r = α = 3 and β = θ = 1. Hence,
1 3 5
Pr(N ≤ 0) = p0 + p1 = + = = 0.3125
8 16 16
For the special case r = 1, the negative binomial random variable is called
the geometric random variable. The geometric distribution, like the expo-
nential distribution, has the memoryless property: the distribution of a
variable that is known to be in excess of some value d does not depend on
d.
Example 26.7
The number of students having the flu in a particular college follows a geo-
metric distribution with β = 0.4. What is the difference between the follow-
ing two values:
A. The expected number of students having the flu in excess of 3 if it is
known that the number of students having the flu is greater than 6.
B. The expected number of students having the flu in excess of 2 if it is
known that the number of students having the flu is greater than 2.
Solution.
They are equal by the memoryless property of the geometric distribution
Practice Problems
Problem 26.1
You are modeling the frequency of events, and you need to select a distri-
bution to use. You observe that the variance of the number of events is less
than the mean number of events.
Which of the following distributions should you use?
(a) Poisson
(b) Negative binomial
(c) Geometric
(d) None of the above
Problem 26.2
Assume that a certain type of claims in one month follows a geometric dis-
tribution with β = 3.
What is the probability that there is at least 1 claim in a month?
Problem 26.3
Suppose that the number of claims N in one month follows a geometric dis-
tribution with β = 3.
Calculate E(N ) and Var(N ).
Problem 26.4
Suppose that the number of claims N in one month folloes a negative bino-
mial distribution with r = 3 and β = 2.
Problem 26.5
Let N |Λ have a negative binomial distribution with r = 2 and λ. Let Λ have
a Gamma distribution with α = 2 and θ = 3.
Find Var[N ].
Problem 26.6
Let N be a negative binomial random variable with mean 8 and variance 40.
Find the parameters r and β.

Problem 26.7
Find the probability generating function of a geometric random variable N.
Problem 26.8
Find the coefficient of variation of a negative binomial random variable with
parameters r and β.
27 THE BERNOULLI AND BINOMIAL DISTRIBUTIONS 189
27 The Bernoulli and Binomial Distributions

Binomial experiments are problems that consist of a fixed number of trials
n, with each trial having exactly two possible outcomes: Success and fail-
ure. The probability of a success is denoted by q and that of a failure by
1 − q. Also, we assume that the trials are independent, that is what happens
to one trial does not affect the probability of a success in any other trial.
The prefix ”bi” in binomial experiment refers to the fact that there are two
possible outcomes (e.g., head or tail, true or false, working or defective) to
each trial in the binomial experiment.
Let N represent the number of successes that occur in m trials. Then N is

said to be a binomial random variable with parameters (m, q) If m = 1
then N is said to be a Bernoulli random variable. The central question
of a binomial experiment is to find the probability of r successes out of m
trials. In the next paragraph we’ll see how to compute such a probability.
Now, anytime we make selections from a population without replacement,
we do not have independent trials. For example, selecting a ball from an
urn that contain balls of two different colors.
The probability of r successes in a sequence out of m independent trials

is given by q r (1 − q)m−r . Since the binomial coefficient C(m, r) counts all
the number of outcomes that have r successes and m − r failures, the prob-
ability of having r successes in any order is given by the binomial mass
function
pr = Pr(N = r) = C(m, r)q r (1 − q)m−r
where q denotes the probability of a success. Note that
m
X m
X
pk = C(m, k)q k (1 − q)m−k = (q + 1 − q)m = 1.
k=0 k=0
Example 27.1
Suppose that in a particular sheet of 100 postage stamps, 3 are defective.
The inspection policy is to look at 5 randomly chosen stamps on a sheet and
to release the sheet into circulation if none of those five is defective. Write
down the random variable, the corresponding probability distribution and
then determine the probability that the sheet described here will be allowed
to go into circulation.
Solution.
Let N be the number of defective stamps in the sheet. Then N is a binomial
random variable with probability distribution

Pr(N = k) = C(5, k)(0.03)k (0.97)5−k , k = 0, 1, 2, 3, 4, 5.
Now,
Pr(sheet goes into circulation) = Pr(N = 0) = (0.97)5 = 0.859
Example 27.2
Suppose 40% of the student body at a large university are in favor of a ban
on drinking in dormitories. Suppose 5 students are to be randomly sampled.
Find the probability that
(a) 2 favor the ban.
(b) less than 4 favor the ban.
(c) at least 1 favor the ban.
Solution.
(a) Pr(N = 2) = C(5, 2)(.4)2 (0.6)3 ≈ 0.3456.
(b) Pr(N < 4) = Pr(0) + Pr(1) + Pr(2) + Pr(3) = C(5, 0)(0.4)0 (0.6)5 +
C(5, 1)(0.4)1 (0.6)4 + C(5, 2)(0.4)2 (0.6)3 + C(5, 3)(0.4)3 (0.6)2 ≈ 0.913.
(c) Pr(N ≥ 1) = 1 − Pr(N < 1) = 1 − C(5, 0)(0.4)0 (0.6)5 ≈ 0.922
Example 27.3
A student has no knowledge of the material to be tested on a true-false exam
with 10 questions. So, the student flips a fair coin in order to determine the
response to each question.
(a) What is the probability that the student answers at least six questions
correctly?
(b) What is the probability that the student answers at most two questions
correctly?
Solution.
(a) Let N be the number of correct responses. Then N is a binomial random
variable with parameters m = 10 and q = 12 . So, the desired probability is
Pr(N ≥ 6) =Pr(N = 6) + Pr(N = 7) + Pr(N = 8) + Pr(N = 9) + Pr(N = 10)
10
X
= C(10, k)(0.5)k (0.5)10−k ≈ 0.3769.
k=6
(b) We have
2
X
Pr(N ≤ 2) = C(10, k)(0.5)k (0.5)10−k ≈ 0.0547
k=0
Theorem 27.1
Let N be binomial distribution with parameters (m, q). Then
(a) PN (z) = (zq + 1 − q)m .
(b) E(N ) = mq > and Var(N ) = mq(1 − q). Note that Var(N ) < E(N ).
Proof.
(a) The probability
Pngenerating function is found by using the binomial for-
n k
mula (a + b) = k=0 a b n−k :
m
X m
X
k k m−k
PN (z) = C(m, k)z q (1 − q) = (zq)k (1 − q)m−k = (zq + 1 − q)m .
k=0 k=0
(b) We have:
PN0 (z) =mq(zq + 1 − q)m−1

E(N ) =PN0 (1) = mq
PN00 (z) =m(m − 1)q 2 (zq + 1 − q)m−2
E[N (N − 1)] =PN00 (1) = m(m − 1)q 2
Var(N ) =E[N (N − 1)] + E(N ) − (E(N ))2
=m(m − 1)q 2 + mq − m2 q 2 = mq(1 − q)
Example 27.4
Let N be a binomial random variable with parameters (12, 0.5). Find the
variance and the standard deviation of N.
Solution.
We have m = 12 and q = 0.5. Thus, Var(N
√ ) = mq(1 − q) = 6(1 − 0.5) = 3.
The standard deviation is SD(N ) = 3
Example 27.5
An exam consists of 25 multiple choice questions in which there are five
choices for each question. Suppose that you randomly pick an answer for
each question. Let N denote the total number of correctly answered ques-
tions. Write an expression that represents each of the following probabilities.
(a) The probability that you get exactly 16, or 17, or 18 of the questions
correct.
(b) The probability that you get at least one of the questions correct.
Solution.
(a) We have
Pr(N = 16 or N = 17 or N = 18) =C(25, 16)(0.2)16 (0.8)9 + C(25, 17)(0.2)17 (0.8)8

+C(25, 18)(0.2)18 (0.8)7 .
(b) Pr(N ≥ 1) = 1 − P (N = 0) = 1 − C(25, 0)(0.8)25
A useful fact about the binomial distribution is a recursion for calculating

the probability mass function.
Theorem 27.2
Let N be a binomial random variable with parameters (m, q). Then for
k = 1, 2, 3, · · · , n
q m−k+1
p(k) = p(k − 1)
1−q k
Proof.
We have
p(k) C(m, k)q k (1 − q)m−k
=
p(k − 1) C(m, k − 1)q k−1 (1 − q)m−k+1
m! k m−k
k!(m−k)! q (1 − q)
= m! k−1 (1 − q)m−k+1
(k−1)!(m−k+1)! q
(m − k + 1)q q m−k+1
= =
k(1 − q) 1−q k
Practice Problems
Problem 27.1
You are again modeling the frequency of events, and you need to select a
distribution to use. You observe that the variance of the number of events
is less than the mean number of events.
Which of the following distributions should you use?
(a) Binomial
(b) Poisson
(c) Negative binomial
(d) Geometric
(e) None of the above are correct.
Problem 27.2
Let N be a random variable which follows a binomial distribution with pa-
rameters m = 20 and q = 0.2
Calculate E(2N ).
Problem 27.3
Suppose that the number of claims N has a binomial distribution with m = 2
and q = 0.7.
Problem 27.4
An insurance company insures 15 risks, each with a 2.5% probability of loss.
The probabilities of loss are independent.
What is the probability of 3 or more losses in the same year?
Problem 27.5
Suppose that N |Λ has a binomial distribution with parameters Λ and q =
0.4. Suppose that Λ has a probability function defined by p(1) = p(2) =
p(3) = p(4) = 0.25.
Calculate the unconditional variance of N.

Problem 27.6
An actuary has determined that the number of claims follows a binomial
distribution with mean 6 and variance 3.
Calculate the probability that the number of claims is at least 3 but less
than 5.
Problem 27.7
Let N be a binomial random variable with m = 10 and q = 0.2. Let F (m)
denote the cdf of N. Complete the following table.
m 0 1 2 3 4
pm
F (m)
Problem 27.8
Let N1 and N2 be two independent binomial random variables with respec-
tive parameters (m1 , q) and (m2 , q).
Compute the pmf of N1 + N2 .

28 THE (A, B, 0) CLASS OF DISCRETE DISTRIBUTIONS 195
28 The (a, b, 0) Class of Discrete Distributions

The Poisson, negative binomial, geometric, and binomial distributions all
satisfy the recursive equation
pk b
=a+ (28.1)
pk−1 k
for some constants a and b and for k ∈ N. We will denote the collection of
these discrete distributions by C(a, b, 0). The table below list the parameters
a and b for each distribution together with the probability function at 0.
Distributions a b p0
Poisson 0 λ e−λ
β β
Negative binomial 1+β (r − 1) 1+β (1 + β)−r
β
Geometric 1+β 0 (1 + β)−1
q q
Binomial − 1−q (m + 1) 1−q (1 − q)m
Note that (28.1) can be written as

pk
k = ak + b (28.2)
pk−1
so that the right-side is a linear function of k with slope a. Thus, if we

pk
graph k pk−1 against k, we get a straight line that has a positive slope for the
negative binomial or geometric distributions, negative slope for the binomial
distribution, and 0 slope for the Poisson distribution. This characterization
of the sign of a, helps in the selection of the distribution when (28.2) is
given. We illustrate this idea in the next example.
Example 28.1
Let N be a member of C(a, b, 0) satisfying the recursive probabilities
pk 3
k = k + 3.
pk−1 4
Identify the distribution N.
Solution.
Since a > 0 and b > 0, N is the negative binomial distribution. We have
β 3 β
1+β = 4 which implies β = 3. Also, we have (r − 1) 1+β = 3 which yields
r=5
Example 28.2 ‡
The distribution of accidents for 84 randomly selected policies is as follows:
# of Accidents # of Policies
0 32
1 26
2 12
3 7
4 4
5 2
6 1
Identify the frequency model that best represents these data.
Solution.
We have
pk
# of Accidents # of Policies k pk−1
0 32 NA
1 26 0.8125
2 12 0.9231
3 7 1.75
4 4 2.2857
5 2 2.5
6 1 3

pk
Plotting the points k, k pk−1 we find that the negative binomial distribu-
tion is the best model from C(a, b, 0) to use (slope of line is positive)
Example 28.3
The number of dental claims in a year follows a Poisson distribution with
a mean of λ. The probability of exactly six claims during a year is 40% of
the probability that there will be 5 claims. Determine the probability that
there will be 4 claims.
Solution.
Let N be the number of dental claims in a year. Since N is a member of
C(a, b, 0) we can write
pk b λ
= a + = , k = 1, 2, · · · .
pk−1 k k
We are told that

p6 λ
0.4 = = =⇒ λ = 2.4.
p5 6
Thus,
e−2.4 (2.4)4
p4 = = 0.1254
4!
Example 28.4 ‡
X is a discrete random variable with a probability function which is a mem-
ber of the C(a, b, 0) class of distributions.
You are given:
(i) Pr(X = 0) = Pr(X = 1) = 0.25
(ii)Pr(X = 2) = 0.1875
Calculate Pr(X = 3).
Solution.
Let N denote the distribution under consideration. Since N is a member of
the C(a, b, 0) class, we have the recursive relation

b
pk = a + pk−1 , k = 1, 2, · · · .
k
From (i), we obtain 0.25 = (a + b)(0.25) which implies a + b = 1. From (ii),

we have 0.1875 = 1 − 2b (0.25) which implies b = 0.5. Hence, a = 0.5 and

0.5
Pr(X = 3) = 0.5 + (0.1875) = 0.125
3
Example 28.5 ‡
A discrete probability distribution has the following properties:
1

(i) pk = c 1 + k pk−1 , k = 1, 2, · · ·
(ii) p0 = 0.5.
Solution.
This is a class C(a, b, 0) distribution with a = b = c. We will go through each
distribution in the class and see which one fits the condition of the problem.
• If the distribution is Poisson then a = b = c = 0. In this case, p0 = 0.5
X∞
and pk = 0 for k = 1, 2, · · · . Since pk = 0.5 6= 1, this distribution can
k=1
not be the answer.
• If the distribution is geometric then a = b = c = 0 and as in the previous

case, the distribution can not be geometric.
• The distribution can not be binomial since a = b. For binomial a and b
are of opposite signs.
• The only possibility remaining is that the distribution is negative binomial.
β β
In this case, a = 1+β and b = (r − 1) 1+β . We have
a 1
=1= =⇒ r = 2.
b r−1
Also,
p0 = 0.5 = (1 + β)−r =⇒ β = 0.414.
Finally,
β 0.414
c= = = 0.29
1+β 1.414
Practice Problems
Problem 28.1
pk 4 1
= − .
pk−1 k 3
Identify the distribution N.
Problem 28.2
pk 4 1
= − .
pk−1 k 3
Find E(N ) and Var(N ).
Problem 28.3 ‡
The number of claims is being modeled with an C(a, b, 0)) class of distribu-
tions. You are given:
• p0 = p1 = 0.25
• p2 = 0.1875.
Find the value of p3 .
Problem 28.4
Let N be a counting distribution in C(a, b, 0) satisfying:
• p0 = 415
pk
= c 0.25 + k1 , k = 1, 2, 3, · · · .

• pk−1
Problem 28.5
Suppose that the number of claims N has a Poisson distribution with mean
λ = 3.
p255
Calculate p254 .
Problem 28.6
Let N be a negative binomial random variable with r = 2.5 and β = 5.
pk
Find the smallest value of k such that pk−1 < 1.
Problem 28.7
Let N be a member of C(a, b, 0) such that p0 = p1 and p2 = 0.6p1 .
Determine the values of a and b.
Problem 28.8
For N in C(a, b, 0) you are given the following:
• p0 = p1 .
• p2 = 0.6p1 .
Based on this information, which of the following are true statements?
(I) N has a Poisson distribution.

(II) N has a negative binomial distribution.
(III) N has a binomial distribution.
(IV) None of the above.
Problem 28.9
Let N be a member of C(a, b, 0). You are given:
• p5 = 0.00144
• p4 = 0.006
• p3 = 0.02
Determine the value of p0 .
Problem 28.10
Let N be a member of C(a, b, 0). You are given:
• p2 = 0.1536
• p1 = 0.4096
• p0 = 0.4096
Determine E(N ).
Problem 28.11 ‡
For a discrete probability distribution, you are given the recursion relation
2
p(k) = p(k − 1), k = 1, 2, · · · .
k
Determine p(4).
29 THE CLASS C(A, B, 1) OF DISCRETE DISTRIBUTIONS 201
29 The Class C(a, b, 1) of Discrete Distributions

All the four members of C(a, b, 0) have a fixed positive probability at 0. For
some models, an adjustment at 0 is sometimes required. We would like to
be able to assign any value in the interval [0, 1) to p0 and not be bounded
by the fixed values provided by the four distributions of C(a, b, 0).
Let N be a discrete non-negative random variable with pmf pk , k = 0, 1, 2, · · · .

We can create a new random variable N ∗ with pmf p∗k such that p∗0 has a
preassigned value in the interval [0, 1). The pmf of N ∗ is given by
1 − p∗0
p∗k = pk , k = 1, 2, · · · .
1 − p0
Note that
∞ ∞ ∞
X X 1 − p∗0 X 1 − p∗0
p∗k = p∗0 + p∗k = p∗0 + pk = p∗0 + (1 − p0 ) = 1.
1 − p0 1 − p0
k=0 k=1 k=1
We say that N ∗ has a zero-modified distribution. In the spacial case when

p∗0 = 0, we say that N ∗ has a zero-truncated distribution. Note that the
zero-truncated distribution is a special case of the zero-modified distribu-
tion. We will therefore develop properties of the zero-modified distributions.
We will consider only the distributions of C(a, b, 0). The collection of all
members of C(a, b, 0) together with the associated zero-modified distribu-
tions belong to a class denoted by C(a, b, 1). Note that, for k = 1, 2, · · · , we
have
p∗k pk b
= =a+ .
p∗k−1 pk−1 k
Note that both members of C(a, b, 0) and C(a, b, 1) satisfy

b
pk = a + pk−1
k
but for the C(a, b, 0) class the k starts from 1 whereas for the C(a, b, 1) class
the k starts from 2.
The probabilities of a zero-modified distribution are denoted by pM

k where
1 − pM
pM
k =
0
pk , k = 1, 2, · · ·
1 − p0
and pM0 is an arbitrary number in [0, 1). The probabilities of a zero-truncated

distribution are denoted by pTk where
1
pTk = pk , k = 1, 2, · · ·
1 − p0
and pT0 = 0.
Theorem 29.1
Let N be in C(a, b, 0) with corresponding moment generating function MN (t).
Then the moment generating function of N M is
pM − p0 1 − pM

M
MN (t) = 0 + 0
MN (t).
1 − p0 1 − p0
Proof.
We have
∞
X
M tN
MN (t) =E(e )= etn pM
n
n=0
∞
− pM
X
1
=pM
0 + 0
etn pn
1 − p0
n=1
∞
"X #
pM

1−
=pM
0 + 0 tn
e pn − p0
1 − p0
n=0
1 − pM

=pM
0 + 0
[MN (t) − p0 ]
1 − p0
1 − pM 1 − pM

M 0 0
=p0 − p0 + MN (t)
1 − p0 1 − p0
pM − p0 1 − pM

= 0 + 0
MN (t)
1 − p0 1 − p0
Corollary 29.1
Let N be in C(a, b, 0) with corresponding probability generating function
PN (t). Then the probability generating function of N M is
pM − p0 1 − pM

PNM (t) = 0 + 0
PN (t).
1 − p0 1 − p0
Proof.
We have
pM 1 − pM

0 − p0
PNM (t) M
=MN (ln t) = + 0
MN (ln t)
1 − p0 1 − p0
pM 1 − pM

0 − p0 0
= + PN (t)
1 − p0 1 − p0
Remark 29.1
Note that
1 − pM 1 − pM

PNM (t) = 1− 0
·1+ 0
PN (t).
1 − p0 1 − p0
Thus, N M is a mixture of a degenerate distribution (i.e. adistribution with

pM

0 −p0
support consisting of a single point) with mixing weight 1−p0 and the
M
1−p
corresponding member of C(a, b, 0) with mixing weight 1−p00 .
Example 29.1
Let N be the Poisson distribution with parameter λ. Find the probability
functions of
(a) the zero-truncated distribution
(b) the zero-modified distribution with preassigned pM
0 = 0.10.
Solution.
e−λ λk
(a) We have p0 = e−λ and pk = k! . Hence,
1 1 e−λ λk
pTk = pk = , k = 1, 2, · · · .
1 − p0 1 − e−λ k!
(b) We have, with pM

0 = 0.10,
1 − pM 0.90 e−λ λk
pM
k =
0
pk = , k = 1, 2, · · ·
1 − p0 1 − e−λ k!
Example 29.2
Let N have the negative binomial distribution with r = 2.5 and β = 0.5.
(a) Determine pk , k = 0, 1, 2, 3.
(b) Determine pT1 , pT2 , pT3 .
(c) Determine pM M M M
1 , p2 , p3 given that p0 = 0.6.
Solution.
(a) From the table in the previous section, we find
p0 =(1 + β)−2.5 = 0.362887

β 0.5 1
a= = =
1+β 1 + 0.5 3
β (2.5 − 1) 1
b =(r − 1) = = .
1+β 3 2
Hence,

1 11
p1 =p0 + = 0.302406
3 21

1 11
p2 =p1 + = 0.176404
3 22

1 11
p3 =p2 + = 0.088202.
3 23
(b) We have
pT0 =0
1
pT1 = p1 = 0.474651
1 − p0
1
pT2 = p2 = 0.276880
1 − p0
1
pT3 = p3 = 0.138440.
1 − p0
(c) We have
pM
0 =0.6
1 − pM
pM
1 =
0
p1 = 0.189860
1 − p0
1 − pM
pM
2 = 0
p2 = 0.110752
1 − p0
1 − pM
pM
3 = 0
p3 = 0.055376
1 − p0
Practice Problems
Problem 29.1
Show that
pM M T
k = (1 − p0 )pk , k = 1, 2, · · ·
Problem 29.2
Show that
1 − pM

M 0
E(N ) = E(N ).
1 − p0
Problem 29.3
Let N M be the zero-modified distribution associated to N.
Find E[N M (N M − 1)] and Var(N M ).
Problem 29.4
Let N have a Poisson distribution with mean 1.
(a) Determine p0 , p1 , and p2 .

(b) Find the mean and the variance of E(N T ).
Problem 29.5
Consider the zero-modified geometric distribution:
1
pM
0 =
2
1 2 k−1

pM
k = , k = 1, 2, 3, · · · .
6 3
(a) Find the moment generating function of N M .

(b) Find the mean and the variance of N M .
Problem 29.6
1 1 2
You are given: pM M M
1 = 6 , p2 = 9 , and p3 = 27 . Find pM
0 .
30 The Extended Truncated Negative Binomial Model

The C(a, b, 1) class not only contains the members of C(a, b, 0) but also
two additional distributions the so-called the extended truncated neg-
ative binomial distribution(ETNB) and the logarithmic distribution.
The extended truncated negative binomial distribution has the pobability

function defined recursively as
p0 =0
pk b
=a + , k = 2, 3, · · ·
pk−1 k
where
β β
a= 1+β , β>0 and b = (r − 1) 1+β , r > −1, r 6= 0.
Example 30.1
Show that
k−1
β r+1 r+2 r+k−1
pk = p1 · ··· , k = 2, 3, · · · .
β+1 2 3 k
Solution.
The proof is by induction on k = 2, 3, · · · . For k = 2, we have

β r−1 β β r+1
p2 = p1 + = p1 .
1+β 2 1+β 1+β 2
Suppose that up to k, we have

k−1
β r+1 r+2 r+k−1
pk = p1 · ··· .
β+1 2 3 k
Then,
k
β r−1 β β r+1 r+2 r+k−1r+k
pk+1 = p1 + = p1 · ···
1+β k+11+β β+1 2 3 k k+1
Example 30.2
Pp1 > 0 then pk > 0 for all k = 2, 3, · · · .
(a) Show that if
(b) Show that ∞ k=1 pk < ∞.
30 THE EXTENDED TRUNCATED NEGATIVE BINOMIAL MODEL207
Solution.
(a) Since r > −1, β > 0 and k = 2, 3, · · · , from the previous example we
conclude that pk > 0 for all k = 2, 3, · · · .
(b) We have
∞ ∞ k−1
X X β r+1 r+2 r+k−1
pk = p1 · ··· .
β+1 2 3 k
k=1 k=2
Let
k−1
β r+1 r+2 r+k−1
ak = · ··· .
β+1 2 3 k
Then
ak+1 β r+k β
lim = lim = < 1.
k→∞ ak k→∞ 1 + β k + 1 1+β
Hence, by the ratio series test, the given series is convergent
Now, letting r → 0 we have

∞ ∞ k−1
X X β 1 2 k−1
1= pk = p1 · ···
1+β 2 3 k
k=1 k=2
∞ k−1
X β 1
=p1
1+β k
k=2
X ∞ k
1+β β 1
=p1
β 1+β k
k=2

1
=p1 − ln 1 −
1+β
where we used the Taylor expansion of ln (1 − x). Hence, p0 = 0 and

k
β 1
pk = , k = 1, 2, · · · .
1+β k ln (1 + β)
But this is the probability function of the logarithmic distribution with

parameter β.
Example 30.3
Find the probability generating function of the logarithmic distribution.
Solution.
We have
∞ ∞ n
X
n 1 X β 1 n
P (z) = pn z = z
ln (1 + β) 1+β n
n=0 n=1

1 zβ
= − ln 1 −
ln (1 + β) 1+β

1 1+β
= ln
ln (1 + β) 1 − β(z − 1)
ln [1 − β(z − 1)]
=1 −
ln (1 + β)
Example 30.4
Consider the extended zero-truncated negative binomial distribution with
r = −0.5 and β = 1. Calculate pT2 , and pT3 given that pT1 = 0.853553.
Solution.
We have
β 1
a= = = 0.5
1+β 1+1
β
b =(r − 1) = −0.75
1+β

T T 0.75
p2 =p1 0.5 − = 0.106694
2

T T 0.75
p3 =p2 0.5 − = 0.026674
3
Remark 30.1
The C(a, b, 1) class consists of the following distributions:
• Poisson, Negative binomial, Geometric, Binomial, and Logarithmic.
• Zero-truncated: Poisson, Binomial, Negative Binomial (ETNB), and Ge-
ometric.
• Zero-modified: Poisson, Binomial, ETNB, Geometric, and Logarithmic.
30 THE EXTENDED TRUNCATED NEGATIVE BINOMIAL MODEL209
Practice Problems
Problem 30.1
Consider the extended zero-modified negative binomial distribution with
r = −0.5 and β = 1.
Calculate pM M M T M
1 , p2 , and p3 given that p1 = 0.853553 and p0 = 0.6.
Problem 30.2
Let N denote the logarithmic distribution introduced in this section. Find
E(N ).
Problem 30.3
E[N (N − 1)].
Problem 30.4
Var(N ).
Problem 30.5
Let N T denote the zero-truncated distribution corresponding to the dis-
tribution N. Let PNT (z) be the probability generating function of N T and
PN (z) be the probability generating function of N. Show that
PN (z) − p0
pTN (z) = .
1 − p0
Problem 30.6
Find the probability generating function for the extended truncated negative
binomial distribution.
Problem 30.7
Find the mean of the extended truncated negative binomial.
Problem 30.8
Let N T denote the ETNB. Find E[N T (N T − 1)].
Problem 30.9
Let N T denote the ETNB. Find Var(N T ).
Modifications of the Loss
Random Variable
The purpose of this chapter is to relate the random variables introduced

earlier in this text to insurance applications. In this chapter, all the random
vairables have a support that is a subset of [0, ∞). For purposes of notation,
we shall refer to X as the (ground-up) loss amount random variable−before
any modifications. We shall denote the modified loss amount to be Y and
will be referred to as the claim amount paid by the insurer. We will use the
notation Y L to denote the loss-variable and the notation Y P to denote the
payment-variable. In this chapter, we will discuss three types of coverage
modifications: deductibles, policy limits, and coinsurance factors.
211
212 MODIFICATIONS OF THE LOSS RANDOM VARIABLE
31 Ordinary Policy Deductibles

For an insurance policy, in order for a claim to be paid, a threshold d must
be exceeded. That is, the ground-up loss X must exceed d. In this case, the
insurance pays the policyholder the amount X −d. For loss amount less than
d the insurance pays 0. This is referred to as ordinary deductible. The
amount paid by the insurance is called the cost-per loss9 or left censored
and shifted variable (see Section 5). It is denoted by

L 0 X≤d
Y = max{X − d, 0} = (X − d)+ =
X − d X > d.
The cost-per loss Y L is an example of a mixed random variable: For X ≤ d,

Y L is a discrete random variable with probability at 0 given by
pY L (0) = Pr(Y L = 0) = Pr(X ≤ d) = FX (d).
The continuous part of Y L is given by
fY (y) = fX (y + d), y > 0.
The cdf of Y L is given by
FY L (y) = Pr(Y L ≤ y) = Pr(X − d ≤ y) = FX (y + d), y ≥ 0.
From this, it follows that
SY L (y) = SX (y + d), y ≥ 0.
The kth moments of Y L are given by

X
E[(Y L )k ] = (x − d)k pX (x)
x>d
if X is discrete and
Z ∞
L k
E[(Y ) ] = (x − d)k fX (x)dx
d
if X is continuous. Moreover, for the continuous case, if k = 1 we have

Z ∞ Z ∞
L
E(Y ) = (x − d)fX (x)dx = [1 − FX (x)]dx.
d d
9
The random variable Y L is also referred to as the claim amount paid per-loss
event.
31 ORDINARY POLICY DEDUCTIBLES 213
It follows that E(Y L ) ≤ E(X) since

Z ∞ Z ∞
[1 − FX (x)]dx ≤ [1 − FX (x)]dx.
d 0
Note that with the presence of deductibles, the number of payments is fewer
than the losses since losses with amount less than or equal to the deductible
will result in no payments.
Example 31.1
Determine the pdf, cdf, and the sdf for Y L if the gound-up loss amount func-
tion has an exponential distribution with mean 1θ and an ordinary deductible
of d.
Solution.
Recall that
fX (x) = θe−θx , x > 0.
Thus,
1 − e−θd y = 0

fY L (y) =
θe−θ(y+d) y > 0.
1 − e−θd

y=0
FY L (y) = −θ(y+d)
1−e y > 0.
e−θd

y=0
SY L (y) =
θe−θ(y+d) y>0
In the cost per loss situation, all losses below or at the deductible level are
recorded as 0. We next examine the situation where all losses below or at
the deductible level are completely ignored and not recorded in any way.
This sitution is represented by the random variable

P L L L undefined X ≤ d
Y = (Y |Y > 0) = (Y |X > d) =
X −d X > d.
We call Y P , the excess loss variable10 , the cost per-payment, or the

left truncated and shifted variable as defined in Section 5. Also, recall
10
We refer to Y P the amount paid per-payment event
the following results from Section 5:

fX (y + d)
fY P (y) = , y>0
SX (d)
FX (y + d) − FX (d)
FY P (y) = , y>0
SX (d)
SX (y + d)
SY P (y) = , y>0
SX (d)
fX (y + d)
hY P (y) = = hX (y + d), y > 0.
SX (y + d)
Note that the excess loss random variable places no probability at 0. Also,
recall from Section 5,
E(Y L )
E(Y P ) = > E(Y L ).
1 − FX (d)
That is, the mean excess loss is larger than the expected amount paid per-
loss event.
Example 31.2
Determine the pdf, cdf, sdf, and the hazard rate function for Y P if the
gound-up loss amount function has an exponential distribution with mean
1
θ and an ordinary deductible of d.
Solution.
Recall that
fX (x) = θe−θx , x > 0.
Thus,
fY P (y) =θ−θy
FY P (y) =1 − e−θy
SY P (y) =e−θy
hY P (y) =θ
The following theorem deals with calculating the expectation of Y L and Y P

under ordinary deductibles.
Theorem 31.1
For an ordinary deductible d, we have
E(Y L ) = E(X) − E(X ∧ d)

and
E(X) − E(X ∧ d)
E(Y P ) =
1 − FX (d)
where X ∧ d is the limited loss variable (see Section 5) defined by

X, X < d
X ∧ d = min(X, d) =
d, X ≥ d
and Z d
E(X ∧ d) = SX (x)dx.
0
Proof.
Buying one policy with a deductible d and another one with a limit d is
equivalent to purchasing full cover. That is,
Y L + X ∧ d = X.
Hence,
E(Y L ) = E(X) − E(X ∧ d).
Also,
E(Y L ) E(X) − E(X ∧ d)
E(Y P ) = =
1 − FX (d) 1 − FX (d)
Example 31.3
Losses are distributed exponentially with parameter θ. Policies are subject
to ordinary deductible d. Find E(Y L ) and E(Y P ).
Solution.
We have Z d
1
E(X ∧ d) = e−θx dx = (1 − e−θd )
0 θ
and Z ∞
1
E(X) = xθe−θx dx = .
0 θ
Hence,
1 1 e−θd
E(Y L ) = − (1 − e−θd ) =
θ θ θ
and
e−θd
1
E(Y P ) = θ
=
e−θd θ
Example 31.4 ‡
The annual number of doctor visits for each individual in a family of 4 has
a geometric distribution with mean 1.5. The annual numbers of visits for
the family members are mutually independent. An insurance pays 100 per
doctor visit beginning with the 4th visit per family.
Calculate the expected payments per year for this family.
Solution.
Let Xi be the annual number of visits by member i of the family, where
i = 1, 2, 3, 4. Let Y = X1 + X2 + X3 + X4 be the annual number of visits by
the whole family. For each Xi , the probability generating function is
P (z) = [1 − 1.5(z − 1)]−1 .
Using independence,
P Y (z) = [1 − 1.5(z − 1)]−4 .
Thus, Y is has a negative binomial distribution with r = 4 and β = 1.5.

The number of visits resulting in insurance payments is
(Y − 3)+ = Y − Y ∧ 3
so that
E[(Y − 3)+ ] = E(Y ) − E(Y ∧ 3).
From Table C, the pmf of the negative binomial with r = 4 and β = 1.5 is
r(r + 1) · · · (r + k − 1)β k
Pr(Y = k) = .
k!(1 + β)r+k
Now, we have the following
E(Y ) =4E(X) = 4(1.5) = 6

E(Y ∧ 3) =1 · Pr(Y = 1) + 2 · Pr(Y = 2) + 3 · Pr(Y = 3) + 3[1 − Pr(Y ≤ 3)
=1 · Pr(Y = 1) + 2 · Pr(Y = 2) + +3Pr(Y ≥ 3)
1
Pr(Y = 0) = = 0.0256
(2.5)4
4(1.5)
Pr(Y = 1) = = 0.06144
(2.5)4+1
4(5)(1.5)2
Pr(Y = 2) = = 0.09216
2(2.5)4+2
Pr(Y ≥ 3) =1 − 0.0256 − 0.06144 − 0.09216 = 0.8208.
Thus,
E(Y ∧ 3) = 0.06144 + 2(0.09216) + 3(0.8208) = 2.71
and the expected number of visits resulting in insurance payments is
E(Y ) = 6 − 2.71 = 3.29.
The insurance pays 100 per visit so that the total expected insurance pay-
ment for the year is 3.29 × 100 = 329
Practice Problems
Problem 31.1
The cdf of a loss amount distribution is expressed as:
x 2
FX (x) = 1 − e−( 100 ) , x > 0.
The ordinary deductible for an insurance policy is 50. Find the pdf, cdf,
and the survival function of the cost per loss Y L .
Problem 31.2
The cdf of a loss amount distribution is expressed as:
x 2
FX (x) = 1 − e−( 100 ) , x > 0.
The ordinary deductible for an insurance policy is 50. Find the pdf, cdf,
survival function, and the hazard rate function of the cost per payment Y P .
Problem 31.3
Loss amounts are exponentially distributed with parameter θ = 1000. An
insurance policy is subject to deductible of 500.
Find E(Y L ).
Problem 31.4
Loss amounts are exponentially distributed with parameter θ = 1000. An
insurance policy is subject to deductible of 500.
Find E(Y P ).
Problem 31.5 ‡
Losses follow an exponential distribution with parameter θ. For an ordinary
deductible of 100, the expected payment per loss is 2000.
What is the expected payment per loss in terms of θ for an ordinary de-
ductible of 500?
Problem 31.6
Loss amounts are distributed as a single Pareto with parameters α = 4 and
θ = 90. An insurance policy is subject to an ordinary deductible of 100.
Determine Var(Y L ).
Problem 31.7
Losses are uniformly distributed in (0, b). An insurance policy is subject to
an ordinary deductible of d.
Calculate Var(Y P ).
Problem 31.8
Loss amounts have a discrete distribution with the following probabilities:
Loss Amount Probability

150 0.30
300 0.25
425 0.15
750 0.30
An insurance coverage for these losses has an ordinary deductible of 250.
(a) Calculate the expected insurance payment per loss.

(b) Calculate the expected insurance payment per payment.
Problem 31.9 ‡
Risk 1 has a Pareto distribution with parameters α > 2 and θ. Risk 2 has
a Pareto distribution with parameters 0.8α and θ. Each risk is covered by a
separate policy, each with an ordinary deductible of k.
(a) Determine the expected cost per loss for risk 1.

(b) Determine the limit as k goes to infinity of the ratio of the expected cost
per loss for risk 2 to the expected cost per loss for risk 1.
Problem 31.10
Loss amounts are uniformly distributted in (0, 10). An ordinary policy de-
ductible of d is applied.
Find d if E(Y L ) = 20.

32 Franchise Policy Deductibles

The second type of deductibles that we consider is the franchise de-
ductibles. Under this deductible policy, the loss is fully paid when the
loss exceeds the deductible.
The cost per-loss random variable is defined by

L 0, X ≤ d
Y =
X, X > d
and the cost per-payment is

P undefined, X ≤ d
Y = (X|X > d) =
X, X > d.
The key distributional functions are

FX (d), y = 0
fY L (y) =
fX (y), y > d

FX (d), 0 ≤ y ≤ d
FY L (y) =
FX (y), y>d

SX (d), 0 ≤ y ≤ d
SY L (y) =
SX (y), y>d

0, 0<y<d
hY L (y) =
hX (y), y>d
for the cost per-loss and
fX (y)
fY P (y) = , y>d
SX (d)
(
0, 0≤y≤d
FY P (y) = FX (y)−FX (d)
SX (d) , y>d
(
1, 0≤y≤d
SY P (y) = SX (y)
SX (d) , y>d

0, 0<y<d
hY P (y) =
hX (y), y>d
for the cost per-payment.
32 FRANCHISE POLICY DEDUCTIBLES 221
Example 32.1
Loss amounts are distributed exponentially with mean 1θ . Insurance policies
are subject to a franchise deductible of d. Find the pdf, cdf, sdf, and the
hazard rate function for the cost per-loss Y L .
Solution.
The pdf of X is
fX (x) = θe−θx . x > 0.
We have
1 − e−θd , y = 0

fY L (y) =
θe−θy , y > d.
1 − e−θd , 0 ≤ y ≤ d

FY L (y) =
1 − e−θy , y > d.
−θd
e , 0≤y≤d
SY L (y) =
e−θy , y > d.

0, 0 < y < d
hY L (y) =
θ, y > d
Example 32.2
Loss amounts are distributed exponentially with mean 1θ . Insurance policies
are subject to a franchise deductible of d. Find the pdf, cdf, sdf, and the
hazard rate function for the cost per-payment Y P .
Solution.
We have
fY P (y) = θe−θ(y−d) , y > d.

0, 0≤y≤d
FY P (y) =
1 − e−θ(y−d) , y > d.

1, 0≤y≤d
SY P (y) = −θ(y−d)
e , y > d.

0, 0 < y < d
hY P (y) =
θ, y > d
The following theorem deals with calculating the expectation of Y L and Y P

under franchise deductibles.
Theorem 32.1
For a franchise deductible d, we have
E(Y L ) = E(X) − E(X ∧ d) + d[1 − FX (d)]
and
E(X) − E(X ∧ d)
E(Y P ) = +d
1 − FX (d)
where X ∧ d is the limited loss variable (see Section 5) defined by

X, X < d
X ∧ d = min(X, d) =
d, X ≥ d
and Z d
E(X ∧ d) = SX (x)dx.
0
Proof.
We will prove the results for the continuous case. We have
Z ∞ Z ∞ Z ∞
E(Y L ) = xfX (x)dx = (x − d)fX (x)dx + d fX (x)dx
d d d
=E[(X − d)+ ] + d[1 − FX (d)]
=E(X) − E(X ∧ d) + d[1 − FX (d)]
E(Y L ) E(X) − E(X ∧ d)
E(Y P ) = = +d
1 − FX (d)] 1 − FX (d)
Example 32.3
Losses are distributed exponentially with parameter θ. Policies are subject
to franchise deductible d. Find E(Y L ) and E(Y P ).
Solution.
Using Example 31.3, we have
e−θd
E(Y L ) = + de−θd
θ
and
eθd
E(Y P ) = +d
θ
Example 32.4 ‡
Insurance agent Hunt N. Quotum will receive no annual bonus if the ratio of
incurred losses to earned premiums for his book of business is 60% or more
for the year. If the ratio is less than 60%, Hunt’s bonus will be a percentage
of his earned premium equal to 15% of the difference between his ratio and
60%. Hunt’s annual earned premium is 800,000.
Incurred losses are distributed according to the Pareto distribution, with
θ = 500, 000 and α = 2.
Calculate the expected value of Hunt’s bonus.
Solution.
Let L be the incurred losses and B Hunt’s bonuses. We are told that if
L L
800,000 < 0.6, that is, L ≤ 480, 000, then B = 0.15 0.6 − 800,000 (800, 000) =
0.15(480, 000 − L). This can be written as

480, 000 − L, L < 480, 000
B = 0.15 = 0.15[480, 000 − X ∧ 480, 000)].
0, L ≥ 480, 000
Its expected value is
E(B) = 0.15[480, 000 − E(X ∧ 480, 000)].
From Table C, we have

" 1 #
500, 000 500, 000
E(X ∧ 480, 000) = 1− = 244, 898.
1 500, 000 + 480, 000
Hence,
E(B) = 0.15(480, 000 − 244, 898) = 35, 265
Practice Problems
Problem 32.1
Loss amounts are uniformly distributed on (0, θ). For a franchise deductible
d < θ, find fY L (y), FY L (y), SY L (y), and hY L (y).
Problem 32.2
d < θ, find fY P (y), FY P (y), SY P (y), and hY P (y).
Problem 32.3
d < θ, find E(Y L ) and E(Y P ).
Problem 32.4
Claim amounts X have the following Pareto distribution
3
800
FX (x) = 1 − .
x + 800
An insurance policy has a franchise deductibe of 300. Find the expected
cost per-loss.
Problem 32.5 ‡
Auto liability losses for a group of insureds (Group R) follow a Pareto dis-
tribution with α = 2 and θ = 2000. Losses from a second group (Group
S) follow a Pareto distribution with α = 2 and θ = 3000. Group R has an
ordinary deductible of 500, Group S has a franchise deductible of 200.
Calculate the amount that the expected cost per payment for Group S
exceeds that for Group R.
Problem 32.6
Loss amounts are exponentially distributed with parameter θ. For a franchise
deductible d, it is given that E(Y L ) = 0.40E(Y P ). Express d in terms of θ.
Problem 32.7
Loss amounts have a discrete distribution with the following probabilities:
Loss amounts Probability
100 20%
300 70%
1000 10%
An insurance coverage for these losses has a franchise deductible of 250.
(a) Calculate the expected insurance payment per loss.

(b) Calculate the expected insurance payment per payment.
Problem 32.8
Losses in 2011 are distributed as a Pareto distribution with α = 2 and
θ = 2000. An insurance company sells a policy that covers these losses with
a franchise deductible of 500 during 2011. Losses in 2012 increase by 20%.
During 2012, the insurance company will sell a policy covering the losses.
However, instead of the franchise deductible used in 2010, the company will
implement an ordinary deductible of d. The expected value of per-loss is the
same for both years.
Find the value of d.
Problem 32.9
Losses are distributed as a Pareto distribution with α = 5 and θ = 1000.
Losses are subject to a franchise deductible of d. The expected value per
payment after the deductible is 820.
Calculate d.
Problem 32.10
Loss amounts follow a Pareto distribution with parameters α = 2 and θ. The
expected value per-payment for an ordinary deductible of 10000 is 20000.
Calculate the expected value per-loss when losses are subject to a franchise
deductible of 15000.
33 The Loss Elimination Ratio and Inflation Effects

for Ordinary Deductibles
Loss amounts X subject to an ordinary deductible d satisfy the relation

0, X≤d X, X ≤ d
X = (X − d)+ + X ∧ d = +
X − d, X > d d, X > d.
In words, this relation says that for a policy with a deductible d, losses below
d are not covered and therefore can be covered by another policy with a limit
of d. Rewriting the above relation as
X ∧ d = X − (X − d)+
we see that X ∧ d is a decrease of the overall losses and hence can be con-
sidered as savings to the policyholder in the presence of deductibles. The
expected savings (due to the deductible) expressed as a percentage of the
loss (no deductible at all) is called the Loss Elimination Ratio:
E(X ∧ d)
LER = .
E(X)
For a continuous loss amounts X we have
Rd
[1 − FX (x)]dx
LER = R 0∞ .
0 [1 − FX (x)]dx
Example 33.1
Loss amounts X follow an exponential distribution with mean θ = 1000.
Suppose that insurance policies are subject to an ordinary deductible of
500. Calculate the loss elimination ratio.
Solution.
We have
Rd R 500 − x
[1 − FX (x)]dx e 1000 dx
0
LER = R ∞ = R0∞ − x = 1 − e−0.5
0 [1 − FX (x)]dx 0 e
1000 dx
Example 33.2 ‡
You are given:
(i) Losses follow an exponential distribution with the same mean in all years.
(ii) The loss elimination ratio this year is 70%.
(iii) The ordinary deductible for the coming year is 4/3 of the current de-
ductible.
Compute the loss elimination ratio for the coming year.
33 THE LOSS ELIMINATION RATIO AND INFLATION EFFECTS FOR ORDINARY DEDUCTIBLES227
Solution.
We have
− xθ
Rd
0 e dx d
LER = R ∞ −x = 1 − e− θ =⇒ d = θ ln (1 − LER).
0 e
θ dx
Thus,
dLast year = θ ln 0.30
and
dLast year = θ ln (1 − LERNext year ).
We are told that dNext year = 43 dLast year . Thus,
4
θ ln (1 − LERNext year ) = θ ln 0.30 =⇒ LERNext year = 0.80
3
Example 33.3
An insurance company offers two types of policies: Type A and Type B.
The distribution of each type is presented below.
Type A Type B
Loss Amount Probability Loss Amount Probability
100 0.65 300 0.70
200 0.35 400 0.30
55% of the policies are of Type A and the rest are of Type B. For an ordinary
deductible of 125, calculate the loss elimination ratio and interpret its value.
Solution.
The expected losses without deductibles is
E(X) =E(X|A)Pr(A) + E(X|B)Pr(B)

=[100(0.65) + 200(0.35](0.55) + [300(0.70) + 400(0.30)](0.45)
=222.75.
The expected savings (due to the deductible) is
E(X ∧ 125) =E(X ∧ 125|A)Pr(A) + E(X ∧ 125|B)Pr(B)

=[100(0.65) + 125(1 − 0.65)](0.55) + 125(0.45) = 116.0625.
Hence,
116.0625
LER = = 0.521 = 52.1%
222.75
This is the percentage of savings in claim payments due to the presence of

the deductible 125
The effect of inflation

To see the effect of inflation, consider the following situation: Suppose an
event formerly produced a loss of 475. With a deductible of 500, the insurer
has no payments to make. Inflation of 12% will increase the loss to 532 and
thus results of insurer’s payment of 32.
Theorem 33.1
Let loss amounts be X and let Y be the loss amounts after uniform inflation
of r. That is, Y = (1 + r)X. For an ordinary deductible of d, the expected
cost per-loss is

L d
E(Y ) = (1 + r) E(X) − E X ∧ .
1+r
The expected cost per-payment is
−1
P E(Y ) − E(Y ∧ d) d d
E(Y ) = = 1 − FX (1+r) E(X) − E X ∧ .
1 − FY (d) 1+r 1+r
Proof.
We have

y y
FY (y) =Pr(Y ≤ y) = Pr X ≤ = FX
1+r 1+r

1 y
fY (y) = fX
1+r 1+r
E(Y ) =(1 + r)E(X).
Z d
E(Y ∧ d) = yfY (y)dy + d[1 − FY (d)]
0

Z d yfX y
1+r d
= dy + d 1 − FX
0 1+r 1+r
Z d
1+r d
= (1 + r)xfX (x)dx + d 1 − FX
0 1+r
(Z d )
1+r d d
=(1 + r) xfX (x)dx + 1 − FX
0 1+r 1+r

d
=(1 + r)E X ∧
1+r
Hence, the expected cost per loss is

L d
E(Y ) = E(Y ) − E(Y ∧ d) = (1 + r) E(X) − E X ∧ .
1+r
For the cost per-payment, see Theorem 31.1
Example 33.4
Determine the effect of inflation at 10% on an ordinary deductible of 500
applied to an exponential distribution with mean 1000.
Solution.
Before the inflation, We have
E(X) =1000
Z 500
x
E(X ∧ 500) = e− 1000 dx = 1000(1 − e−0.5 )
0
L
E(YBI ) =1000e−0.5
P 1000e−0.5
E(YBI )= = 1000.
e−0.5
After the inflation, we have
Z 500
500 1.1 x 1
E X∧ = e− 1000 dx = 1000(1 − e− 2.2 )
1.1 0
1 1
L
E(YAI ) =(1.1)[1000 − 1000(1 − e− 2.2 ) = 1100e− 2.2
1
P 1100e− 2.2
E(YAI )= 1 = 1100.
e− 2.2
1
Thus, the expected cost per loss increased from 1000e−0.5 to 1100e− 2.2 , an
1
1100e− 2.2 −1000e−0.5
increase of 1000e−0.5
= 15.11%. The cost per-pay increased from
1000 to 1100, an increase of 10%
Example 33.5 ‡
The graph of the density function for losses is:
Calculate the loss elimination ratio for an ordinary deductible of 20.
Solution.
We will use the formula
E[(X − 20)+
LER = 1 − .
E(X)
We have
Z 80 Z 120
E(X) = 0.01xdx + (0.03x − 0.00025x2 )dx
0 80
=50.66667
Z 20 Z 20
E[(X − 20)+ ] =E(X) − xf (x)dx − 20[1 − f (x)dx]
0 0
Z 20 Z 20
2
=E(X) − 0.01x dx − 20[1 − 0.01xdx]
0 0
=50.6667 − 2 − 20(0.8) = 32.6667
32.6667
LER =1 − = 0.3553
50.6667
Practice Problems
Problem 33.1
Loss amounts are being modeled with a distribution function expressed be-
low:
x2
FX (x) = , 0 ≤ x ≤ 50.
2500
An insurance policy comes with an ordinary deductible of 30.
Calculate the loss elimination ration and interpret its value.
Problem 33.2
Loss amounts are being exponentially distributed with the same mean for all
years. Suppose that with an ordinary deductible of d, LER(2011) = 0.75.
In 2012, the deductible is expected to increase by 45%.
Calculate LER(2012).
Problem 33.3 ‡
Losses have an exponential distribution with a mean of 1000. There is an or-
dinary deductible of 500. The insurer wants to double loss elimination ratio.
Determine the new deductible that achieves this.
Problem 33.4
Losses follow an exponential distribution with mean of θ = 1000. An in-
surance company applies an ordinary policy deductible d which results in a
Loss Elimination Ratio of 1 − e−0.5 .
Calculate d.
Problem 33.5 ‡
Losses follow a distribution prior to the application of any deductible with
a mean of 2000. The loss elimination ratio at a deductible of 1000 is 0.3.
The probability of a loss being greater than 1000 is 0.4.
Determine the average size of a loss given it is less than or equal to the
deductible of 1000, that is, find E(X|X ≤ 1000).
Problem 33.6
Loss amounts are being modeled with a distribution function expressed be-
low
x
FX (x) = 1 − e− 100 .
An insurance policy comes with a deductible of 50.
Calculate the difference in the loss elimination ratio before and after a uni-
form inflation of 30%
Problem 33.7 ‡
Losses have a Pareto distribution with α = 2 and θ = k. There is an ordi-
nary deductible of 2k.
Determine the loss elimination ration before and after 100% inflation.
Problem 33.8 ‡
Claim sizes this year are described by a 2-parameter Pareto distribution
with parameters α = 4 and θ = 1500.
What is the expected claim size per loss next year after 20% inflation and
the introduction of a $100 ordinary deductible?
Problem 33.9 ‡
Losses in 2003 follow a two-parameter Pareto distribution with α = 2 and
θ = 5. Losses in 2004 are uniformly 20% higher than in 2003. An insurance
covers each loss subject to an ordinary deductible of 10.
Calculate the Loss Elimination Ratio in 2004.

34 POLICY LIMITS 233
34 Policy Limits
If a policy has a limit u, then the insurer will pay the full loss as long as
the losses are less than or equal to u, otherwise, the insurer pays only the
amount u. Thus, the insurer is subject to pay a maximum covered loss of u.
Let Y denote the claim amount random variable for policies with limit u.
Then
X, X ≤ u
Y = min{X, u} = X ∧ u =
u, X > u.
We call Y the limited loss random variable(See Section 5).
Just as in the case of a deductible, Y has a mixed distribution, with the

discrete part
pY (u) = Pr(X > u) = 1 − FX (u)
and a continuous part given by
fY (y) = fX (y), y < u.
Moreover, fY (y) = 0 for y > u. The cdf of Y is

FX (y), y < u
FY (y) = Pr(Y ≤ y) =
1, y ≥ u.
Example 34.1 Ru
Show that: E(Y ) = 0 SX (x)dx.
Solution.
First note that
Y =u+Z
where
X − u, X ≤ u
Z=
0, X > u.
Thus,
Z u
E(Y ) =u + (x − u)fX (x)dx
0
Z u
u
=u + [(x − u)FX (x)]0 − FX (x)dx
0
Z u
= [1 − FX (x)]dx
0
Theorem 34.1
Let E(Y ) be the expected cost before inflation. Suppose that the same policy
limit applies after an inflation at rate r. Then the after inflation expected
cost is given by

u
E((1 + r)X ∧ u) = (1 + r)E X ∧ .
1+r
Proof.
See the proof of Theorem 33.1
Example 34.2
Losses follow a Pareto distribution with parameters α = 2 and θ = 1000.
For a coverage with policy limit of 2000, find the pdf and the cdf of the
limited loss random variable.
Solution.
Recall that
αθα 2(1000)2
fX (x) = =
(x + θ)α+1 (1000 + x)3
and α 2
θ 1000
FX (x) = 1 − =1− .
θ+x 1000 + x
Thus, 
1

 9, y = 2000
2(1000)2
fY (y) = (1000+y)3
, y < 2000

 0, y > 2000
and ( 2
1000
1− , y < 2000
FY (y) = 1000+x
1 y ≥ 2000
Example 34.3
Calculate the expected cost for a coverage with policy limit of 2000.
Solution.
Recall that α
θ
S(x) = .
θ+x
34 POLICY LIMITS 235
Thus,
Z 2000 Z 2000 2
1000
E(X ∧ 2000) = S(x)dx = dx = 666.67
0 0 1000 + x
Example 34.4
For a coverage with policy limit 2000 and after an inflation rate of 30%,
calculate the after inflation expected cost.
Solution.
We have
" !#
2000 2000 1000
E(1.3X∧2000) = 1.3E X ∧ = 1.3 1− = 1575.76
1.3 2−1 1000 + 2000
1.3
Example 34.5 ‡
A jewelry store has obtained two separate insurance policies that together
provide full coverage. You are given:
(i) The average ground-up loss is 11,100.
(ii) Policy A has an ordinary deductible of 5,000 with no policy limit.
(iii) Under policy A, the expected amount paid per loss is 6,500.
(iv) Under policy A, the expected amount paid per payment is 10,000.
(v) Policy B has no deductible and a policy limit of 5,000.
Given that a loss has occurred, determine the probability that the payment
under policy B is 5,000.
Solution.
Let X denote the ground-up loss random variable. By (i), E(X) = 11, 100.
By (ii) and (iii), we have
E(Y L ) = E(X) − E(X ∧ 5000) = 6, 500.
By (iv), we habe
E(X) − E(X ∧ 5000)
E(Y P ) = = 10, 000.
1 − FX (500)
Thus,
6500
= 10, 000 =⇒ FX (5000) = 0.35.
1 − FX (5000)
For Policy B, a payment of 5000 will occur if X ≥ 5000. Hence,
Pr(X ≥ 5000) = 1 − FX (5000) = 1 − 0.35 = 0.65
Practice Problems
Problem 34.1
Losses follow an exponential distribution with mean 1θ . For a coverage with
policy limit u, find fY (y) and FY (y).
Problem 34.2
policy limit u, find E(X ∧ u).
Problem 34.3
policy limit u and with an inflation at rate r, find the expected cost.
Problem 34.4
Losses are distributed uniformly between 0 and 100. An insurance policy
which covers the losses has an upper limit of 80. Find the expected cost.
Problem 34.5 ‡
An insurance company offers two types of policies: Type Q and Type R.
Type Q has no deductible, but has a policy limit of 3000. Type R has no
limit, but has an ordinary deductible of d. Losses follow a Pareto distribu-
tion with α = 3 and θ = 2000.
Calculate the deductible d such that both policies have the same expected
cost per loss.
Problem 34.6
Suppose that the ground-up losses for 2010 follow an exponential distribu-
tion with a mean of 1000. In 2011, all losses are subject to uniform inflation
of 25%. The policy in 2011 has limit u.
Determine the value of u if the expected cost in 2011 is equal to the ex-
pected loss in 2010.
Problem 34.7
Suppose that the ground-up losses for 2010 follow a Pareto distribution with
parameters α = 3 and θ = 9800. In 2011, all losses are subject to uniform
inflation of 6%. The policy limit in 2011 is 170,000.
Calculate the expected cost in 2011.

35 COMBINATIONS OF COINSURANCE, DEDUCTIBLES, LIMITS, AND INFLATIONS237
35 Combinations of Coinsurance, Deductibles, Lim-

its, and Inflations
Another coverage modification is the use of coinsurance factor. In a policy
with a coinsurance factor 0 < α < 1, the insurer portion of the loss is αX
and the insured portion is (1 − α)X. The claim amount random variable is
Y = αX and its distribution function is
y y
FY (y) = Pr(Y ≤ y) = Pr X ≤ = FX .
α α
Its density function is given by
1 y
fY (y) = fX .
α α
Now, suppose that a policy is subject to a coinsurance factor α, an ordinary
deductible d, and a policy limit u∗ . By convention, the coinsurance factor
is applied after the application of any deductible or limit. We define the
maximum covered loss to be u = u∗ + d. Thus, in the absence of a
deductible the maximum covered loss is just the policy limit. The claim
amount per loss random variable is given by

 0, X≤d
L
Y = α(X − d), d < X ≤ u
α(u − d), X > u.

In compact form, we can write

Y L = α[X ∧ u − X ∧ d].
Note that α(u − d) is the maximum payment per loss.
Example 35.1
A health insurance policy has a deductible of 200, a policy limit of 5000 and
a coinsurance factor of 80%. Calculate the expected claim amount per loss
event and the expected claim amount per payment event for this policy if
losses follow an exponential distribution with mean 1000.
Solution.
Note that u = 5000 + 200 = 5200. The expected claim amount per loss is
E(Y L ) =α[E(X ∧ u) − E(X ∧ d)]

1 −u 1 − d
=α (1 − e θ ) − (1 − e θ )
θ θ
−5.2
) − 1000(1 − e−0.2 )

=0.8 1000(1 − e
=650.57
The expected claim amount per payment is
E(Y L ) 650.57
E(Y P ) = = −0.2 = 794.61
1 − FX (200) e
For a policy subject to a coinsurance factor α, an ordinary deductible d,

policy limit u∗ , and uniform inflation at rate r applied only to losses, the
claim amount per loss is given
 d
 0, X ≤ 1+r
YL = α[(1 + r)X − d], 1+rd u
< X ≤ 1+r
u
α(u − d), X > 1+r

where u = u∗ + d. In compact form, we can write
Y L = α[(1 + r)X ∧ u − (1 + r)X ∧ d].
Theorem 35.1
The expected value of the per-loss random variable is

L u d
E(Y ) = α(1 + r) E X ∧ −E X ∧ .
1+r 1+r
The expected value of the per-payment random variable is
E(Y L )
E(Y P ) = .
d
1 − FX 1+r
Proof.
We have
E(Y L ) =α{E[(1 + r)X ∧ u] − E[(1 + r)X ∧ d]}

u d
=α (1 + r)E X ∧ − (1 + r)E X ∧
1+r 1+r

u d
=α(1 + r) E X ∧ −E X ∧ .
1+r 1+r

d
For the expected cost per payment divide the expression by 1 − FX 1+r
To find the variance of Y L , the second moment is required which is pro-

vided by the following theorem.
Theorem 35.2
The second moment for the per-loss random variable is
E[(Y L )2 ] = [α(1+r)]2 E[(X ∧ u∗ )2 ] − E[(X ∧ d∗ )2 ] − 2d∗ [E(X ∧ u∗ ) − E(X ∧ d∗ )] .

Proof.
Using
Y L = α[(1 + r)X ∧ u − (1 + r)X ∧ d] = α(1 + r)[X ∧ u∗ − X ∧ d∗ ]
we can write
2
YL

=[X ∧ u∗ − X ∧ d∗ ]2
α(1 + r)
=(X ∧ u∗ )2 − 2(X ∧ u∗ )(X ∧ d∗ ) + (X ∧ d∗ )2
=(X ∧ u∗ )2 − (X ∧ d∗ )2 − 2(X ∧ d∗ )[X ∧ u∗ − X ∧ d∗ ].
But
 d
0,  X ≤ 1+r
∗ ∗ ∗ ∗ ∗ ∗
(X∧d )[X∧u −X∧d ] = d (X − d∗ ),
∗ d u
1+r < X ≤ 1+r = d [X∧u −X∧d ].
 ∗ ∗
d (u − d∗ ), u
X > 1+r
Thus,
2
YL

= (X ∧ u∗ )2 − (X ∧ d∗ )2 − 2d∗ [X ∧ u∗ − X ∧ d∗ ].
α(1 + r)
Now, the result of the theorem follows by taking the expectation of both
sides
Example 35.2
Determine the mean and the standard deviation per loss for an exponential
distribution with mean 1000 and with a deductible of 500 and a policy limit
of 2500.
Solution.
x
Recall that E(X ∧ x) = θ(1 − e− θ ). Thus,
3000 500
E(Y L ) = E(X∧3000)−E(X∧500) = 1000(1−e− 1000 )−1000(1−e− 1000 ) = 556.74.
For the second moment of Y L we have

Z 3000
x
2
E[(X ∧ 3000) ] = x2 (0.001)e− 1000 dx + 30002 e−3 = 1, 601, 703.453
0
Z 500
x
E[(X ∧ 500)2 ] = x2 (0.001)e− 1000 dx + 5002 e−0.5 = 180, 408.0209
0
L 2
E[(Y ) ] =1, 601, 703.453 − 180, 408.0209 − 2(500)(556.74)
=864, 555.4321.
The variance of Y L is
Var(Y L ) = 864555.4321 − 556.742 = 554, 596.0045
Example 35.3 ‡
A group dental policy has a negative binomial claim count distribution with
mean 300 and variance 800.
Ground-up severity is given by the following table:
Severity Probability
40 0.25
80 0.25
120 0.25
200 0.25
You expect severity to increase 50% with no change in frequency. You decide
to impose a per claim deductible of 100.
Calculate the expected total claim payment after these changes.
Solution.
After imposing the 50% increase, the severity values are
Severity (X) Probability
60 0.25
120 0.25
180 0.25
300 0.25
Let N be the claim frequency. Then the expected total claim payment is
the expected number of losses times the expected payment per loss. That
is,
E(N )[E(X) − E(X ∧ 100)].
We have
E(N ) =300
E(X) =(60 + 120 + 180 + 300)(0.25) = 165
E(X ∧ 100) =60(0.25) + 100(1 − 0.25) = 90.
Thus, the expected total claim payment is 300(165 − 90) = 22, 500
Example 35.4 ‡
An insurer has excess-of-loss reinsurance on auto insurance. You are given:
(i) Total expected losses in the year 2001 are 10,000,000.
(ii) In the year 2001 individual losses have a Pareto distribution with
2
2000
F (x) = 1 − , x > 0.
x + 2000
(iii) Reinsurance will pay the excess of each loss over 3000.
(iv) Each year, the reinsurer is paid a ceded premium, Cyear , equal to 110%
of the expected losses covered by the reinsurance.
(v) Individual losses increase 5% each year due to inflation.
(vi) The frequency distribution does not change.
(a) Calculate C2001 .
(b) Calculate CC2001 .
2002
Solution.
(a) The reinsurance fraction per loss is given by
E(X) − E(X ∧ 3000)
E(X)
where
θ 2000
E(X) = = = 2000
α−1" 2−1
α−1 #
θ 2000
E(X ∧ 3000) = 1−
α−1 3000 + 2000
" 2−1 #
2000 2000
= 1− = 1200.
2−1 3000 + 2000
Thus,
E(X) − E(X ∧ 3000) 1200
=1− = 0.40.
E(X) 2000
Finally,
C2001 = 1.10(0.40)(10, 000, 000) = 4, 400, 000.
(b) Due to inflation, the amount per loss in 2002 is X2002 = 1.05X2001 .
Thus, E(X2002 ) = 1.05E(X2001 ) = 1.05(2000) = 2100. Now, X2002 is a
Pareto distribution with parameters α = 2 and θ = 2100) (see Section 18).
Thus,
" 2−1 #
2100 2100
E(X2002 ∧ 3000) = 1− = 1235.
2−1 3000 + 2100
Hence,

2100 − 1235
C2002 = 1.10 (10, 000, 000)(1.05) = 4, 758, 600
2100
and
C2002 4, 758, 600
= = 1.08
C2001 4, 400, 000
Example 35.5 ‡
Annual prescription drug costs are modeled by a two-parameter Pareto dis-
tribution with θ = 2000 and α = 2.
A prescription drug plan pays annual drug costs for an insured member sub-
ject to the following provisions:
(i) The insured pays 100% of costs up to the ordinary annual deductible of
250.
(ii) The insured then pays 25% of the costs between 250 and 2250.
(iii) The insured pays 100% of the costs above 2250 until the insured has
paid 3600 in total.
(iv) The insured then pays 5% of the remaining costs.
Determine the expected annual plan payment.
Solution.
Let X denote the annual drug cost and Y the insurer’s payment. What is the
first value of X where the insured total payment reaches 3600? From what
is given, if X = 2250 the insured’s payment is 250 + 0.25(2250 − 250) = 750.
After this point, the insured’s will pay 100% of the costs above 2250 until
the insured has paid 3600 in total. But this means that insured will pay
3600 − 750 = 2850 past the 2250 mark. In other words, the insured reaches
the total payment of 3600 when X = 2250 + 2850 = 5100. Having said that,
Y can be expressed as follows


 0, 0 ≤ X ≤ 250
0.75(X − 250), 250 < X ≤ 2250

Y =

 0.75(2250 − 250), 2250 < X ≤ 5100
1500 + 0.95(X − 5100), X > 5100.

Next, note the following,

 
 0, 0 ≤ X ≤ 250  X − X, 0 ≤ X ≤ 250
0.75(X − 250), 250 < X ≤ 2250 = 0.75 X − 250, 250 < X ≤ 2250
0.75(2250 − 250), 2250 < X ≤ 5100 2250 − 250, 2250 < X ≤ 5100
 
= 0.75(X ∧ 2250 − X ∧ 250).

Thus, we can express Y as
Y = 0.75(X ∧ 2250 − X ∧ 250) + 0.95(X − X ∧ 5100).
Hence,
E(Y ) =0.75[E(X ∧ 2250) − E(X ∧ 250)] + 0.95[E(X) − E(X ∧ 5100)]

2000 2000 2000 2000
=0.75 1− − 1−
2−1 2000 + 2250 2−1 2000 + 250

2000 2000 2000
+0.95 + 1−
2−1 2−1 2000 + 5100
=1163
Example 35.6 ‡
Loss amounts have the distribution function
x 2
100 , 0 ≤ x ≤ 100
F (x) =
1, x > 100
An insurance pays 80% of the amount of the loss in excess of an ordinary

deductible of 20, subject to a maximum payment of 60 per loss.
Calculate the conditional expected claim payment, given that a payment
has been made.
Solution.
The maximum covered loss is
u = 20 + 0.8(60) = 95.
Thus,
E(Y L ) =0.8[E(X ∧ 95) − E(X ∧ 20)]
Z 95 Z 20
=0.8 S(x)dx − S(x)dx
0 0
Z 95 x 2 Z 20 x 2
=0.8 1− dx − 1− dx
0 100 0 100
Z 95 x 2
=0.8 1− dx = 37.35.
20 100
The expected cost per payment is
E(Y L ) 37.35
E(Y P ) = = = 38.91
1 − F (20) 1 − 0.04
Example 35.7 ‡
For a special investment product, you are given:
(i) All deposits are credited with 75% of the annual equity index return,
subject to a minimum guaranteed crediting rate of 3%.
(ii) The annual equity index return is normally distributed with a mean of
8% and a standard deviation of 16%.
(iii) For a random variable X which has a normal distribution with mean
µ and standard deviation σ, you are given the following limited expected
values:
E(X ∧ 3%)
µ = 6% µ = 8%
σ = 12% −0.43% 0.31%
σ = 16% −1.99% −1.19%
E(X ∧ 4%)
µ = 6% µ = 8%
σ = 12% 0.15% 0.95%
σ = 16% −1.43% −0.58%
Calculate the expected annual crediting rate.
Solution.
LetY denote the annual credit rating and X the annual equity index return.
Then we have

3, 0.75X ≤ 3 3, X≤4 0, X≤4
Y = = = 3+
0.75X, 0.75X > 3 0.75X, X > 4 0.75X − 3, X > 4
That is,
Y = 3 + (0.75X − 3)+ = 3 + 0.75X − (0.75X ∧ 3).
Hence,
E(Y ) = 3 + 0.75E(X) − 0.75E(X ∧ 4).
By (ii), E(X) = 8. From the second given table, we see that E(X ∧ 4) =
−0.58. Hence,
E(Y ) = 3 + 6 − 0.75(−0.58) = 9.435%
Example 35.8 ‡
A risk has a loss amount which has a Poisson distribution with mean 3.
An insurance covers the risk with an ordinary deductible of 2. An alternative
insurance replaces the deductible with coinsurance α, which is the proportion
of the loss paid by the insurance, so that the expected insurance cost remains
the same.
Calculate α.
Solution.
The expected cost per loss with a deductible of 2 is
E[(X − 2)+ ] =E(X) − E(X ∧ 2) = 3 − [Pr(X = 1) + 2Pr(X = 2) + 2(1 − Pr(X ≤ 2))]
=3 − Pr(X = 1) + 2[1 − Pr(X = 0) − Pr(X = 1)]
=3 − 3e−3 − 2[1 − (e−3 + 3e−3 )] = 1.249.
The expected cost per loss with a coinsurance of α is αE(X) = 3α. We are
told that 3α = 1.249 so that α = 1.249
3 = 0.42
Example 35.9 ‡
Michael is a professional stuntman who performs dangerous motorcycle
jumps at extreme sports events around the world.
The annual cost of repairs to his motorcycle is modeled by a two parameter
Pareto distribution with θ = 5000 and α = 2.
An insurance reimburses Michael’s motorcycle repair costs subject to the
following provisions:
(i) Michael pays an annual ordinary deductible of 1000 each year.
(ii) Michael pays 20% of repair costs between 1000 and 6000 each year.
(iii) Michael pays 100% of the annual repair costs above 6000 until Michael
has paid 10,000 in out-of-pocket repair costs each year.
(iv) Michael pays 10% of the remaining repair costs each year.
Calculate the expected annual insurance reimbursement.
Solution.
Let X denote the annual repair cost. For 1000 ≤ X ≤ 6000, the insurance
pays 0.80(X − 1000). Thus, if cost is 6000, Michael’s share is 2000 and the
insurance share is 4000. For the next 8000, Michael’s share is the whole
amount. That is, the insurance pays nothing. But in this case, Michael’s
out-of-pocket reaches 10,000. It follows that, the insurance pays 90% of cost
over 14,000 which is 0.90(X − 14000).
The annual insurance reimbursement is
R =0.8[(X − 1000)+ − (X − 6000)+ ] + 0.9(X − 14000)+

=0.8[X − X ∧ 1000 − (X − X ∧ 6000)] + 0.9(X − X ∧ 14000)
=0.8(X ∧ 6000 − X ∧ 1000) + 0.9(X − X ∧ 14000).
The expected annual insurance reimbursement is
E(R) = 0.8[E(X ∧ 6000) − E(X ∧ 1000)] + 0.9[E(X) − E(X ∧ 14000)].
From Table C,

θ 5000x
E(X ∧ x) = θ 1 − =
θ+x x + 5000
and
θ
E(X) = = 5000.
α−1
Hence,

5000(6000) 5000(1000) 5000
E(R) = 0.8 − +0.9 5000 − = 2699.36
6000 + 5000 1000 + 5000 14000 + 5000
Practice Problems
Problem 35.1 ‡
Losses this year have a distribution such that E(X ∧ x) = −0.025x2 +
1.475x − 2.25 for x = 11, 12, · · · , 26. Next year, losses will be uniformly
higher by 10%. An insurance policy reimburses 100% of losses subject to a
deductible of 11 up to a maximum reimbursement of 11.
Determine the ratio of next year ’s reimbursements to this year’s reimburse-

ment.
Problem 35.2 ‡
Losses have an exponential distribution with mean 1000. An insurance com-
pany will pay the amount of each claim in excess of a deductible of 100.
Calculate the variance of the amount paid by the insurance company for
one claim, including the possibility that the amount paid is 0.
Problem 35.3 ‡
Losses follow a Poisson distribution with mean λ = 3. Consider two insur-
ance contracts. One has an ordinary deductible of 2. The second has no
deductible and coinsurance in which the insurance company pays α of the
loss.
Determine α so that the expected cost of the two contracts is the same.
Problem 35.4 ‡
x
You are given that e(0) = 25 and S(x) = 1 − w, 0 ≤ x ≤ w, and Y P is the
excess loss variable for d = 10.
Determine the variance of Y P .
Problem 35.5 ‡
Total hospital claims for a health plan were previously modeled by a two-
parameter Pareto distribution with α = 2 and θ = 500. The health plan
begins to provide financial incentives to physicians by paying a bonus of
50% of the amount by which total hospital claims are less than 500. No
bonus is paid if total claims exceed 500. Total hospital claims for the health
plan are now modeled by a new Pareto distribution with α = 2 and θ = K.
The expected claims plus the expected bonus under the revised model equals
expected claims under the previous model.
Calculate K.
Problem 35.6
The amount of a loss has a Pareto distribution with α = 2 and θ = 5000. An
insurance policy on this loss has an ordinary deductible of 1,000, a policy
limit of 10,000, and a coinsurance of 80%.
With a uniform inflation of 2%, calculate the expected claim amount per
payment on this policy.
Problem 35.7
Claim amounts follow a Pareto distribution with parameters αX = 3. and
θX = 2000. A policy is subject to a coinsurance rate of α. The standard
deviation of the claims for this policy is 1472.24.
Problem 35.8
The loss size distribution is exponential with mean 50. An insurance policy
pays the following for each loss. There is no insurance payment for the first
20. The policy has a coinsurance in which the insurance company pays 75%
of the loss. The maximum covered loss is 100.
Calculate E(Y P ).
Problem 35.9
The loss size distribution is a Pareto α = 2 and θ = 100. An insurance policy
pays the following for each loss. There is no insurance payment for the first
20. The policy has a coinsurance in which the insurance company pays 75%
of the loss. The policy has a limit of u∗ . The maximum covered loss is 100.
Given that E(Y P ) = 34.2857, find the maximum payment per loss for this
policy.
Problem 35.10 ‡
In 2005 a risk has a two-parameter Pareto distribution with α = 2 and
θ = 3000. In 2006 losses inflate by 20%.
An insurance on the risk has a deductible of 600 in each year. Pi , the
premium in year i, equals 1.2 times the expected claims.
The risk is reinsured with a deductible that stays the same in each year. Ri ,
the reinsurance premium in year i, equals 1.1 times the expected reinsured
claims.
Suppose R2005 /P2005 = 0.55. Calculate R2006 /P2006 .
36 The Impact of Deductibles on the Number of

Payments
Deductibles always affect the number of payments. For example, when an
imposed deductible is increased the number of payments per period is de-
creased whereas decreasing deductibles results in an increasing number of
payments per period.
Let Xj denote the jth ground-up loss and assume that there are no cov-
erage modifications. Let N L be the total number of losses. Now, suppose
that a deductible is imposed. Let v = Pr(X > d) be the probability that a
loss will result in a payment. We will let Ij be the indicator random variable
whose value is 1 if the jth loss occur (and thus results in a payment) and is 0
otherwise. Then Ij is a Bernoulli random variable such that Pr(Ij = 1) = v
and Pr(Ij = 0) = 1 − v. The corresponding pgf is PIj (z) = vz + 1 − v.
Now, let N P be the total number of payments. Then N P = I1 +I2 +· · ·+IN L .

We call N P a compound frequency model or aggregate claims model.
If I1 , I2 , · · · , Ij , NL are mutually independent and the individual claims Ij
are identically and independently distributed (iid) then N P is a compound
distribution with primary distribution N L and secondary distribu-
tion a Bernoulli distribution. For such a distribution, we have
∞
X ∞ X
X ∞
PN P (z) = Pr(N P = k)z k = Pr(N P = k|N L = n)Pr(N L = n)z k
k=0 k=0 n=0
X∞ ∞
X
= Pr(N L = n) Pr(I1 + I2 + · · · + In = k|N L = n)z k
n=0 k=0
X∞
= Pr(N L = n)[PIj (z)]n
n=0
=PN L [PIj (z)] = PN L [1 + v(z − 1)]
where we used the fact that the pgf of a sum of independent random variables
is the product of pgf of the individual random variables.
Example 36.1
Suppose that losses follow an exponential distribution all with the same
mean 100. Suppose that insurance policies are subject to an ordinary de-
ductible of 20 and that N L follows a Poisson distribution with λ = 3.
36 THE IMPACT OF DEDUCTIBLES ON THE NUMBER OF PAYMENTS251
(a) Calculate the probability that a loss will result in a payment.

(b) Find the pgf of an indicator random variable Ij .
(c) Find the pgf of the total number of losses N L .
(d) Find the pgf of the total number of payments N P .
Solution.
(a) The probability of a loss that results in a payment is
Z ∞
x
v= e− 100 dx = 100e−0.2 .
20
(b) PIj (z) = 1 − 100e−0.2 + 100e−0.2 z.

(c) PN L (z) = e3(z−1) .
−0.2 −0.2
(d) PN P (z) = e3(100e z−100e )
Remark 36.1
In general, if K is a compound distribution with primary distribution N
and secondary distribution M then
PK (z) = PN [PM (z)].
Example 36.2 ‡
An actuary has created a compound claims frequency model with the fol-
lowing properties:
(i) The primary distribution is the negative binomial with probability gen-
erating function
P (z) = [1 − 3(z − 1)]−2
(ii) The secondary distribution is the Poisson with probability generating
function
P (z) = eλ (z − 1).
(iii) The probability of no claims equals 0.067.
Calculate λ.
Solution.
From the above remark, we have
PK (z) = PN [PM (z)] = PN [eλ(z−1) ] = [1 − 3(eλ(z−1) − 1)]−2 .
From Section 8, we know that PK (0) = Pr(K = 0). Hence
0.067 = [1 − 3(e−λ − 1]−2 =⇒ λ = 3.1

Now, suppose that N L depends on a parameter θ and that its pgf PN L (z; θ)
satisfies the equation
PN L (z; θ) = B[θ(z − 1)]
where B(z) is independent of θ and both z and θ only appear in the pgf as
θ(z − 1). Then we have
PN P (z) =PN P (1 − v + vz) = B[θ(1 − v + vz − 1)]

=B[vθ(z − 1)] = PN L (z, vθ).
This shows that N L and N P are both from the same parametric family and
only the parameter θ need to be changed to vθ.
Example 36.3
Suppose that N L follows a negative binomial distribution with parameters
r and β. Find PN L (z) and B(z).
Solution.
We have PN L (z) = [1 − β(z − 1)]−r so that B(z) = (1 − z)−r . Note that β
takes on the role of θ in the above result
Example 36.4
Losses follow a Pareto distribution with α = 3 and θ = 1000. Assume that
N L follow a negative binomial distribution with r = 2 and β = 3. Find
PN P (z) when a deductible of 250 is imposed on N L .
Solution.
N P has a negative binomial distribution with r∗ = 2 and β ∗ = βv. Since
3
1000
v = 1 − FX (250) = = 0.512
1000 + 250
we have
PN P (z) = B[vβ(z − 1)] = [1 − 0.512(3)(z − 1)]−2 = [1 − 1.536(z − 1)]−2
Now, suppose that N L depends on two parameters θ and α with the pgf
satisfying the equation
B[θ(z − 1)] − B(−θ)

PN L (z) = PN L (z; θ; α) = α + (1 − α) .
1 − B(−θ)
Note that α = PN L (0) = Pr(N L = 0) so that N L is a zero-trancated

distribution. With this at hand, by letting
B(−vθ) − B(−θ)
α∗ = PN L (1 − v; θ; α) = α + (1 − α)
1 − B(−θ)
we have
PN P (z) =PN L [1 − v + vz]

B[vθ(z − 1)] − B(−θ)
=α + (1 − α)
1 − B(−θ)

B[−vθ] − B(−θ)
= α + (1 − α)
1 − B(−θ)

B[−vθ] − B(−θ) B[vθ(z − 1)] − B(−vθ)
+ 1 − α + (1 − α)
1 − B(−θ) 1 − B(−vθ)
B[vθ(z − 1)] − B(−vθ)
=α∗ + (1 − α∗ )
1 − B(−vθ)
∗
=PN L (z; vθ; α )
where
α∗ = PN P (0) = Pr(N P = 0).
Hence, if N L is zero-modified then N P is zero-modified.
Example 36.5
Losses follow a Pareto distribution with α = 3 and θ = 1000. Assume that
N L follow a zero-modified negative binomial distribution with r = 2, β = 3
and pM0 = 0.4 Find PN P (z) when a deductible of 250 is imposed on N .
L
solution.
The pgf of N L is (See Problem 30.6)
[1 − β(z − 1)]−r − (1 + β)−r

PN L (z) = pM M
0 + (1 − p0 )
1 − (1 + β)−r
where α = pM −r P
0 and B(z) = (1 − z) . Hence, N is a zero-modified negative
∗ ∗
binomial distribution with r = r2, β = vβ = 1.536 and
(1 + 1.536)−2 − (1 + 3)−2
α ∗ = pM
0 ∗ = 0.4 + (1 − 0.4) = 0.4595.
1 − (1 + 3)−2
Hence,
[1 − 1.536(z − 1)]−2 − 2.536−2

PN P (z) = 0.4595 + 0.5405
1 − 2.536−2
Now, we can obtain the pgf of N L given the pgf of N P :
PN P (z) = PN L [PIj (z)] =⇒ PN L (z) = PN P [PI−1

j
(z)] = PN P (1 − v −1 + v −1 z).
If N P is a function of some parameter θ then we have

θ
PN L (z; θ) = PN P z; .
v
Example 36.6
Suppose payments on a policy with a deductible of 250 have the zero-
modified negative binomial distribution with r∗ = 2, β ∗ = 1.536 and
pM
0 ∗ = 0.4595. Losses have the Pareto distribution with α = 3 and β = 1000.
Determine the distribution of payments when the deductible is removed.
solution.
When the deductibles are removed, the number of payments follow a zero-
∗
modified negative binomial with r = 2, β = βv = 1.536
0.512 = 3 and
∗ ∗
pM ∗ −r + (1 + β )−r − pM ∗ (1 + β )−r
0 ∗ −(1 + β ) 0
pM
0 = v v
1 − (1 + β ∗ )−r
0.4595 − 2.536−2 + 4−2 − 0.4595(4)−2
pM
0 = = 0.4
1 − 2.536−2
Practice Problems
Problem 36.1 ‡
The frequency distribution for the number of losses when there is no de-
ductible is negative binomial with r = 2 and β = 5. Loss amounts have a
Weibull distribution with r = 0.3 and θ = 1000.
Determine the expected number of payments when a deductible of 200 is

applied.
Problem 36.2
Individual losses have an exponential distribution with mean 340. With a
deductible of 200, the frequency distribution for the number of payments is
Poisson with mean λ = 0.5.
Determine the probability that a loss will result in a payment.
Problem 36.3
Find the pgf of the number of payments.
Problem 36.4
What fraction of policyholders would be expected to have at least one claim

paid to them?
Problem 36.5
Loss amounts follow a Pareto distribution with α = 3 and θ = 1000. With
a deductible of 500, the frequency distribution for the number of payments
is geometric with β = 0.4.
Find PN P (z).
Problem 36.6

zero-modified Poisson with mean λ = 0.5.
Determine pM M
0 ∗ if p0 = 0.4.
Problem 36.7
Suppose payments on a policy with deductible 200 have a zero-modified
Poisson distribution with λ∗ = 0.27756 and pM0 ∗ = 0.6304. Losses have an
exponential distribution with mean θ = 1000.
Find pM
0 if the deductible is removed.
Aggregate Loss Models
We continue our discussion of modeling losses. Keep in mind that losses

depend on the loss frequency(i.e., the number of losses) and on the loss
severity(i.e., the size or the amount of the loss). In this chapter, we will
concentrate our attention on aggregate loss models. An aggregate loss
refers to the total amount of losses in one period of time, which is often
encountered in the analysis of a portfolio of risks such as a group insurance
policy.
257
258 AGGREGATE LOSS MODELS
37 Individual Risk and Collective Risk Models

Consider a set of n insurance policies. Let Xi , i = 1, 2, · · · , n, denote the
loss amount of each policy. We assume that the Xi0 s are independent but
are not necessarily identically distributed. The individual risk model
represents the aggregate loss as a sum
S = X1 + X2 + · · · + Xn .
The Xi0 s usually have mixed distributions with a probability mass at zero
corresponding to the probability of no loss or payments.
This type of models is used in modeling group life or health insurance policy
of a group of n individuals where each individual can have different coverage
and different level of loss probabilities.
If the Xi0 s are identically distributed then the individual risk model be-
comes a special case of the so-called collective risk model which we define
next.
Let N be a random variable representing the number of claims which we

refer to as the claim count random variable or frequency and whose
distribution is called the claim count distribution or frequency distri-
bution. Let X1 , X2 , · · · , XN be the loss amounts. We refer to each Xi as a
severity and to its distribution as the severity distribution. We assume
that the Xi0 s are independent and identically distributed. We further assume
that the Xi0 s are independent of N. The collective risk model represents
the aggregate loss as the sum
S = X1 + X2 + · · · + XN .
S is an example of a compound distribution with primary distribution

the distribution of N and secondary distribution the distribution of the
losses. Also, we mention here that when S = 0 then N = 0.
There are advantages in modeling the claim frequency and claim severity
separately, and then combine them to obtain the aggregate loss distribu-
tion. For example, expansion of insurance business may have impacts on
the claim frequency but not the claim severity. In contrast, a cost increae
may a affect the claim severity with no effects on the claim frequency.
37 INDIVIDUAL RISK AND COLLECTIVE RISK MODELS 259
Example 37.1
An insurance company insures 500 individuals against accidental death.
Which of the following risk models is better suited for evaluating the risk to
the insurance company?
(a) A collective risk model.
(b) An individual risk model.
Solution.
Since the number of policies is fixed, the model that is most suited to the
insurance company is the individual risk model
Example 37.2
Which of the following statements are true?
(a) In collective risk model, the number of summands in the aggregate loss
is fixed.
(b) In individual risk model, the number of summands is a random variable.
(c) In individual risk models, the summands are independent but not nec-
essarily identically distributed.
Solution.
(a) This is false. In collective risk model, the number of summands in the
aggregate loss S is a random variable−which we refer to as the frequency
random variable.
(b) This is false. In individual risk model, the number of summands is a
fixed number.
(c) This is true
Example 37.3 ‡
You are given:
# of Claims Probability Claim Size Probability

1
0 5
3 1
1 5 25 3
2
150 3
1 2
2 5 50 3
1
200 3
Claim sizes are independent. Determine the variance of the aggregate loss.
Solution.
For N = 0, we have S = 0. For N = 1 either S = 25 or S = 150. For N = 2,
either S = 50 + 50 = 100, S = 200 + 200 = 400, or S = 50 + 200 = 250. The

probability distribution of the aggregate loss is as follows:
1
Pr(S = 0) =
5
Pr(S = 25) =Pr(S = 25|N = 1)Pr(N = 1)

1 3 1
=Pr(X = 25|N = 1)Pr(N = 1) = =
3 5 5
Pr(S = 100) =Pr(S = 100|N = 2)Pr(N = 2)

2 2 1 4
=Pr[(X1 = 50)and(X2 = 50)]Pr(N = 2) = =
3 3 5 45
Pr(S = 150) =Pr(S = 150|N = 1)Pr(N = 1)

2 3 2
=Pr(X = 150|N = 1)Pr(N = 1) = =
3 5 5
Pr(S = 250) =Pr(S = 250|N = 2)Pr(N = 2)
=[Pr[(X1 = 50)and(X2 = 200)] + Pr[(X1 = 200)and(X2 = 50)]Pr(N = 2)

2 1 1 4
=2 =
3 3 5 45
Pr(S = 400) =Pr(S = 400|N = 2)Pr(N = 2)

1 1 1 1
=Pr[(X1 = 200)and(X2 = 200)]Pr(N = 2) = = .
3 3 5 45
Thus,

1 1 4 2 4 1
E(S) =0 + 25 + 100 + 150 + 250 + 400
5 5 45 5 45 45
=105

2 2 1 2 1 2 4 2 2 2 4 2 1
E(S ) =0 + 25 + 100 + 150 + 250 + 400
5 5 45 5 45 45
=19125
Var(S) =E(S 2 ) − E(S)2 = 19125 − 1052 = 8100
In [1], it is suggested that the frequency distribution N is best to have a

distribution with the following probability generating function:
PN (z; α) = Q(z)α ,
where Q(z) is some function of z independent of α. This implies that if the

volume of business of the insurance company increases by 100r%, expected
claims will increase in a manner proportional to (1 + r)α.
Example 37.4
Show that the Poisson distribution with parameter λ has a probability gen-
erating function of the form P (z) = Q(z)α .
Solution.
The probability generating function is
P (z) = eλ(z−1) = [e(z−1) ]λ .
Thus, Q(z) = ez−1 and α = λ
Example 37.5 ‡
In order to simplify an actuarial analysis Actuary A uses an aggregate dis-
tribution S = X1 + X2 + · · · + XN , where N has a Poisson distribution with
mean 10 and Xi = 1.5 for all i.
Actuary A’s work is criticized because the actual severity distribution is
given by
Pr(Yi = 1) = Pr(Yi = 2) = 0.5, for alli,
where Yi0 s are independent.
Actuary A counters this criticism by claiming that the correlation coefficient
between S and S ∗ = Y1 + Y2 + · · · + YN is high.
Calculate the correlation coefficient between S and S ∗ .
Solution.
The coefficient of corrolation between S and S ∗ is given by
Cov(S, S ∗ )
ρ= p p .
Var(S) Var(S 8 )
We have
E(S) =E(N )E(X) = 10(1.5) = 15
Var(S) =E(N )Var(X) + Var(N )E(X)2
=10(0) + 10(1.5)2 = 22.5
E(S ∗ ) =E(N )E(Y ) = 10(1.5) = 15
Var(S ∗ ) =E(N )Var(Y ) + Var(N )E(Y )2
=10(0.25) + 10(1.5)2 = 25
E(SS ∗ ) =E[E(SS ∗ |N )] = E {E[(X1 + X2 + · · · + XN )(Y1 + Y2 + · · · + YN )|N )}

=E {E(1.5N (Y1 + Y2 + · · · + YN )|N )} = E[1.5N E(N Y )]
=E[1.5N 2 E(Y )] = E[1.5N 2 (1.5)] = 2.25E(N 2 )
=2.25[Var(N ) + E(N )2 ] = 2.25(10 + 100) = 247.5
Cov(S, S ∗ ) =E(SS ∗ ) − E(S)E(S ∗ )
=247.5 − 152 = 22.5
22.5
ρ =√ √ = 0.949
22.5 25
Practice Problems
Problem 37.1
Which of the following statements are true?
(a) In a collective risk model one policy can be modeled using a Pareto
distribution and another can be modeled with an exponential distribution.
(b) In an individual risk model, the loss amounts can have different proba-
bilities at zero.
(c) In the collective risk model, the frequency and severity of payments are
modeled separately.
(d) In the collective risk model, the number of payments affects the size of
each individual payment.
Problem 37.2
Consider the aggregate loss sum in an individual loss model:
S = X1 + X2 + · · · + Xn .
Assume that the loss amounts are identically distributed. Find the limit of
the coefficient of variation of S as n goes to infinity.
Problem 37.3
Consider a portfolio of two policies. One policy follows a Pareto distribu-
tion with parameters α = 3 and θ = 100 and the other policy follows an
exponential distribution with parameter 0.05. Assume that the two policies
are independent.
Calculate the mean and the variance of the aggregate loss S.
Problem 37.4
Determine whether the following model is individual or collective: The num-
ber of claims per day N has a geometric distribution with mean 2. The size
of each claim has an exponential distribution with mean 1000. The number
of losses and loss sizes are mutually independent.
Problem 37.5
Let N, the claim count random variable, follow a negative binomial distri-
bution with parameters r and β. Show that PN (z; α) = [Q(z)]α .
38 Aggregate Loss Distributions via Convolutions

In this section, we look at distributional quantities of the aggregate loss
random variable for the collective model. Let
S = X1 + X2 + · · · + XN
where N is the counting random variable or the frequency random variable

and the Xi0 s are independent and identically distributed. Moreover, the Xi0 s
are independent of N i.e., the claim counts are independent of the claim
sizes. Thus, S is a compound distribution with primary distribution N
and secondary distribution the common distribution of the Xi0 s. Analytical
methods for the calculation of compound distributions can be accomplished
using methods of convolutions. This section presents these methods.
The distribution of S can be accomplished via convolutions as:

∞
X ∞
X
FS (x) = Pr(S ≤ x) = Pr(S ≤ x|N = n)Pr(N = n) = Pr(N = n)FX∗n (x)
n=0 n=0
where FX∗n (x) = Pr(X1 + X2 + · · · + Xn ≤ x) is the n−th fold convolution

of FX (x) defined recursively by
Z x
∗(n−1)
FX∗n (x) = FX (x − y)fX (y)dy
0
and
1, x ≥ 0
FX∗0 (x) =
0, x < 0.
The random variable X is the common distribution of the Xi0 s. Note that
FX∗1 = FX (x). By differentiation, and assuming that differentiation and in-
tegration can be reversed, we find the pdf
Z x
∗n ∗(n−1)
fX (x) = fX (x − y)fX (y)dy, n = 2, 3, · · · .
0
∗n (x) is the probability that, given exactly n claims occur, that

In words, fX
the aggregate amount is x. Now, for a continuous X, the pdf of S is given
by
∞
X
∗n
fS (x) = Pr(N = n)fX (x)
n=1
38 AGGREGATE LOSS DISTRIBUTIONS VIA CONVOLUTIONS 265
and a discrete mass point at x = 0 given by Pr(S = 0) = Pr(N = 0).
If X has a discrete counting distribution with probabilities at 0, 1, 2, · · ·

then
x
∗(n−1)
X
FX∗n (x) = FX (x − y)fX (y), x = 0, 1, 2, · · · ; n = 2, 3, · · · .
y=0
The corresponding pdf is

x
∗(n−1)
X
∗n
fX (x) = fX (x − y)fX (y), x = 0, 1, 2, · · · ; n = 2, 3, · · · .
y=0
∗0 (0) = 1 and f ∗0 (x) = 0 for x 6= 0, we can write

By defining fX X
∞
X
∗n
fS (x) = Pr(S = x) = Pr(N = n)fX (x), x = 0, 1, 2, · · · .
n=0
Though the obtained formula is analytic, its direct calculation is difficult

because, in general, the convolution powers are not available in closed-form.
Panjer recursion , discussed in Section 41 is a very efficient numerical method
to calculate these convolutions.
Example 38.1
An insurance portfolio produces N claims, where
n Pr(N = n)
0 0.5
1 0.2
2 0.2
3 0.1
Individual claim amounts have the following distribution:
x fX (x)
1 0.9
2 0.1
Individual claim amounts and N are mutually indepedent. Complete the

following table:
x ∗0 (x)
fX ∗1 (x)
fX ∗2 (x)
fX ∗3 (x)
fX fS (x)
0
1
2
3
4
Pr(N = n)
Solution.
x ∗0 (x)
fX ∗1 (x)
fX ∗2 (x)
fX fX∗3 (x) fS (x)
0 1 0 0 0 0.5
1 0 0.9 0 0 0.18
2 0 0.1 0.81 0 0.182
3 0 0 0.09 0.729 0.0909
4 0 0 0.01 0.162 0.0182
Pr(N = n) 0.5 0.2 0.2 0.1
To find fX∗n (x) pick two columns whose superscripts sum to n. Then add all
combinations of products from these columns where the arguments sum to x.

For example, fX ∗3 (4) = f ∗1 (1)f ∗2 (3)+f ∗1 (3)f ∗2 (1)+f ∗1 (2)f ∗2 (2). To obtain,
X X X X X X
fS (x), each row of the matrix of convolutions of fX (x) is multiplied by the
probabilities from the row below the table and the products are summed
Example 38.2 ‡
The number of claims in a period has a geometric distribution with mean
4. The amount of each claim X follows Pr(X = x) = 0.25, x = 1, 2, 3, 4.
The number of claims and the claim amounts are independent. S is the
aggregate claim amount in the period.
Calculate FS (3).
Solution.
Note first that S is discrete so that
FS (3) = fS (0) + fS (1) + fS (2) + fS (3)
where
∞
X
∗n
fS (x) = Pr(N = n)fX (x), x = 0, 1, 2, · · · .
n=0
With four claims or more, i.e., n ≥ 4 the aggregate sum is at least 4. Since
∗n (x) = 0 for n ≥ 4. Next, we create
we are interested in S ≤ 3, we see that fX
the chart
x ∗0 (x)
fX ∗1 (x)
fX fX∗2 (x) ∗3 (x)
fX fS (x)
0 1 0 0 0 0.2
1 0 0.25 0 0 0.04
2 0 0.25 0.0625 0 0.048
3 0 0.25 0.125 0.0156 0.0576
Pr(N = n) 0.2 0.16 0.128 0.1024
where
4n
Pr(N = n) = , n = 1, 2, · · · .
5n+1
Hence,
FS (3) = 0.2 + 0.04 + 0.048 + 0.0576 = 0.3456
Example 38.3
Severities have a uniform distribution on [0, 100]. The frequency distribution
is given by
n Probability
0 0.60
1 0.30
2 0.10
Find the 2-fold convolution of FX (x).
Solution.
We know that FX∗0 (x) = 1 for x ≥ 0. Thus,
Z x
∗1 dt x
FX (x) = = .
0 100 100
The 2-fold convolution is

x
x2
Z
t dt
FX∗2 (x) = =
0 100 100 20000
Example 38.4
Severities have a uniform distribution on [0, 100]. The frequency distribution
is given by
n Probability
0 0.60
1 0.30
2 0.10
Find FS (x).
Solution.
We have
x x2
FS (x) = 0.60 + 0.30 + 0.10
100 20000
The pgf of S is found as follows:
∞
X
PS (x) =E(z S ) = E[z 0 ]Pr(N = 0) + E[z X1 +X2 +···+Xn |N = n]Pr(N = n)
n=1
∞
X ∞
X
E Πnj=1 z Xj Pr(N = n) = Pr(N = n)[PX (z)]n

=Pr(N = 0) +
n=1 n=1
=PN [PX (z)].
The moment generating function of S is
MS (z) = PS (ez ) = PN [PX (ez )] = PN [MX (z)].
From this, we find the following:
E(S) =MS0 (0) = PN0 [MX (0)]MX

0
(0) = PN0 (1)MX
0
(0) = E(N )E(X)
E(S 2 ) =MS00 (0) = PN00 [MX (0)][MX
0
(0)]2 + PN0 [MX (0)]MX
00
(0)
=PN00 (1)[MX
0
(0)]2 + PN0 (1)MX
00
(0)
=E[N (N − 1)][E(X)]2 + E(N )E(X 2 )
Var(S) =E[N (N − 1)][E(X)]2 + E(N )E(X 2 ) − E(N )2 E(X)2
=E(N )[E(X 2 ) − E(X)2 ] + [E(N 2 ) − E(N )2 ]E(X)2
=E(N )Var(X) + Var(N )E(X)2 .
Example 38.5
Loss frequency N follows a Poisson distribution with parameter λ. Loss
severity X follows an exponential distribution with mean θ. Find the ex-
pected value and the variance of the aggregate loss random variable.
Solution.
We have
E(S) = E(N )E(X) = λθ
and
E(S 2 ) = E[N (N − 1)][E(X)]2 + E(N )E(X 2 ) = λ2 θ2 + λ(2θ2 ).

Thus,
Var(S) = λ2 θ2 + λ(2θ2 ) − λ2 θ2 = 2λθ2
Example 38.6
Loss frequency N follows a Poisson distribution with parameter λ. Loss
severity X follows an exponential distribution with mean θ. Find an expres-
sion of the probability that the aggregate loss random variable will exceed
a value of n
(a) if S can be approximated with the standard normal distribution
(b) if S can be approximated with the lognormal distribution.
Solution.
(a) We have
!
S − E(S) n − E(S)
Pr(S > n) =Pr p > p
Var(S) Var(S)
!
n − E(S)
=Pr Z > p
Var(S)
!
n − E(S)
=1 − Φ p .
Var(S)
(b) If S is approximated with a lognormal distribution then we must have

1 2 2
E(S) = eµ+ 2 σ and E(S 2 ) = e2(µ+σ ) . Thus, µ and θ2 are the solutions to
2 2 2 ))
the equations µ + 12 σ 2 = ln (λθ) and µ + θ2 = ln (λ θ +λ(2θ
2 . Hence,

ln S − µ ln n − µ
Pr(S > n) =Pr >
σ σ

ln n − µ
=Pr Z >
σ

ln n − µ
=1 − Φ
σ
Example 38.7 ‡
You own a fancy light bulb factory. Your workforce is a bit clumsy they
keep dropping boxes of light bulbs. The boxes have varying numbers of
light bulbs in them, and when dropped, the entire box is destroyed.
You are given:
(i) Expected number of boxes dropped per month : 50
(ii) Variance of the number of boxes dropped per month: 100
(iii) Expected value per box: 200

(iv)Variance of the value per box: 400
You pay your employees a bonus if the value of light bulbs destroyed in a
month is less than 8000.
Assuming independence and using the normal approximation, calculate the
probability that you will pay your employees a bonus next month.
Solution.
Let S denote the total value of boxes destroyed in a month. Then
S = X1 + X2 + · · · + XN
where N is the number of boxes destroyed in a month and Xi is the value

of the ith box destroyed. S is a compound distribution so that
E(S) =E(N )E(X) = 50(200) = 10, 000

V ar(S) =E(N )Var(X) + Var(N )E(X)2
=50(400) + 100(200)2 = 4, 020, 000.
Using normal approximation, we have

S − 10, 000 8, 000 − 10, 000
Pr(S < 8, 000) =Pr √ < √ = P (Z < −0.9975) = 0.16
4, 020, 000 4, 020, 000
Example 38.8 ‡
The number of claims, N, made on an insurance portfolio follows the follow-
ing distribution:
n Pr(N = n)
0 0.7
2 0.2
3 0.1
If a claim occurs, the benefit is 0 or 10 with probability 0.8 and 0.2, respec-
tively. The number of claims and the benefit for each claim are independent.
Calculate the probability that aggregate benefits will exceed expected ben-
efits by more than 2 standard deviations.
Solution.
We have
E(N ) =0(0.7) + 2(0.2) + 3(0.1) = 0.7

E(N 2 ) =02 (0.7) + 22 (0.2) + 32 (0.1) = 1.7
Var(N ) =1.7 − 0.72 = 2
E(X) =0(0.8) + 10(0.2) = 2
E(X 2 ) =02 (0.8) + 102 (0.2) = 20
Var(X) =20 − 22 = 16
E(S) =E(N )E(X) = 0.7(0.2) = 1.4
=0.7(16) + 2(2)2 = 16.04
√
ST D = 16.04 = 4
Pr(S > 1.4 + 2(4)) =Pr(S > 9.4)
=1 − Pr(S = 0).
But,
Pr(S = 0) =Pr[(S = 0) ∩ (N = 0)] + Pr[(S = 0) ∩ (N = 2)] + Pr[(S = 0) ∩ (N = 3)]

=Pr(S = 0|N = 0)Pr(N = 0) + Pr(S = 0|N = 2)Pr(N = 2)
+Pr(S = 0|N = 3)Pr(N = 3)
=(1)(0.7) + (0.8)2 (0.2) + (0.8)3 (0.1) = 0.8792.
Finally,
Pr(S > 9.4) = 1 − 0.8792 = 0.1208
Example 38.9 ‡
For a collective risk model the number of losses, N, has a Poisson distribu-
tion with λ = 20.
The common distribution of the individual losses has the following charac-
teristics:
(i) E(X) = 70
(ii) E(X ∧ 30) = 25
(iii) Pr(X > 30) = 0.75
(iv) E(X 2 |X > 30) = 9000
An insurance covers aggregate losses subject to an ordinary deductible of 30
per loss.
Calculate the variance of the aggregate payments of the insurance.
Solution.
Let S denote the aggregate payments. Then S is a compound distribution
with primary distribution N and secondary distribution the payment per
loss (X − 30)+ . We are asked to find
Var(S) = E(N )E[(X − 30)2+ ] = 20E[(X − 30)2+ ].
We have
E[(X − 30)2+ ] =E[(X − 30)2+ |X > 30](1 − FX (30)) = 0.75E[(X − 30)2+ |X > 30]
=0.75E[(X − 30)2 |X > 30] = 0.75E(X 2 − 60X + 900|X > 30)
=0.75[E(X 2 |X > 30) − 60E(X|X > 30) + 900]
=0.75[E(X 2 |X > 30) − 60E(X − 30|X > 30) − 1800 + 900]
=0.75[E(X 2 |X > 30) − 60(E(X) − E(X ∧ 30))(1 − FX (30))−1 − 900]
=0.75[9000 − 60[70 − 25)(1 − 0.75)−1 − 900] = 3375
Var(S) =20(3375) = 67500
Example 38.10 ‡
The repair costs for boats in a marina have the following characteristics:
Number of Probability that Mean of repair cost Variance of repair

Boat type boats repair is needed given a repair cost given a repair
Power boats 100 0.3 300 10,000
Sailboats 300 0.1 1000 400,000
Luxury yachts 50 0.6 5000 2,000,000
At most one repair is required per boat each year. Repair costs are inde-
pendent.
The marina budgets an amount, Y, equal to the aggregate mean repair costs
plus the standard deviation of the aggregate repair costs.
Calculate Y.
Solution.
Let N be the number of boats that need repair. Then N is a binomial
distribution with mean mq and variance mq(1 − q). Let X denote the repair
cost. Let S be the aggregate cost of a type of boats. Then
E(S) =E(N )E(X)
Var(S) =E(N )Var(X) + Var(N )E(X)2 .
We have
Boat type E(N ) Var(N ) E(X) Var(X) E(S) Var(S)

Power boats 30 21 300 10,000 9,000 2,190,000
Sailboats 30 27 1000 400,000 30,000 39,000,000
Luxury yachts 30 12 5000 2,000,000 150,000 360,000,000
Thus,
p
E(Y ) = 9, 000 + 30, 000 + 150, 000 + (2.19 + 39 + 360) × 106 ≈ 209, 000
Example 38.11 ‡
For an insurance:
(i) The number of losses per year has a Poisson distribution with λ = 10.
(ii) Loss amounts are uniformly distributed on (0, 10).
(iii) Loss amounts and the number of losses are mutually independent.
(iv) There is an ordinary deductible of 4 per loss.
Calculate the variance of aggregate payments in a year.
Solution.
Let S be the aggregate claims. Then S is a compound distribution with
primary distribution N, the number of claims, and secondary distribution
Y = (X − 4)+ , the amount paid per loss, where X is the loss amount. We
want
Var(S) = E(N )Var(Y ) + Var(N )E(Y )2 = λE(Y 2 )
where
Z 10 Z 10
2 2
E(Y ) = (x − 4) fX (x)dx = (x − 4)2 (0.1)dx = 7.2.
4 4
Hence,
Var(S) = 10(7.2) = 72
Example 38.12 ‡
For an insurance portfolio:
(i) The number of claims has the probability distribution
n pn
0 0.1
1 0.4
2 0.3
3 0.2
(ii) Each claim amount has a Poisson distribution with mean 3.

(iii) The number of claims and claim amounts are mutually independent.
Calculate the variance of aggregate claims.
Solution.
Let S be the aggreagte claims, N the number of claims, and X the amount
of claim. Then S is a compound distribution with primary distribution N
and a secondary distribution X. Thus,
Var(S) = E(N )var(X) + Var(N )E(X)2
where
E(X) =λ = 3
Var(X) =λ = 3
E(N ) =0(0.1) + 1(0.4) + 2(0.3) + 3(0.2) = 1.6
E(N 2 ) =02 (0.1) + 12 (0.4) + 22 (0.3) + 32 (0.2) = 3.4
Var(N ) =3.4 − 1.62 = 0.84.
Hence,
Var(S) = 1.6(3) + 0.84(32 ) = 12.36
Example 38.13 ‡
Aggregate losses are modeled as follows:
(i) The number of losses has a Poisson distribution with λ = 3.
(ii) The amount of each loss has a Burr (Burr Type XII, Singh-Maddala)
distribution with α = 3, θ = 2, and γ = 1.
(iii) The number of losses and the amounts of the losses are mutually inde-
pendent.
Calculate the variance of aggregate losses.
Solution.
Let N denote the number of losses and X be the severity random variable.
Then
Var(S) = E(N )Var(X) + Var(N )E(X)2 = λE(X 2 ).
From Table C, we find
22 Γ(3)Γ(1)
E(X 2 ) = = 4.
Γ(3)
Hence,
Var(S) = 3(4) = 12
Example 38.14 ‡
You are the producer for the television show Actuarial Idol. Each year,
1000 actuarial clubs audition for the show. The probability of a club being
accepted is 0.20.
The number of members of an accepted club has a distribution with mean
20 and variance 20. Club acceptances and the numbers of club members are
mutually independent.
Your annual budget for persons appearing on the show equals 10 times the
expected number of persons plus 10 times the standard deviation of the
number of persons.
Calculate your annual budget for persons appearing on the show.
Solution.
Let S be the number of people appearing on the show. Then S is a compound
distribution with frequency N (number of clubs being accepted) and severity
X (number of members of an accepted club).
The frequency N has a binomial distribution with parameters m = 1000
and q = 0.20. Thus,
E(N ) =mq = 1000(0.20) = 200

Var(N ) =mq(1 − q)200(1 − 0.2) = 160.
The severity X has mean 20 and variance 20. Hence,
E(S) =E(N )E(X) = 200(20) = 4000

=200(20) + 160(20)2 = 68000.
The annual budget for persons appearing on the show is

√
10(4000) + 10 68000 ≈ 42610
Example 38.15 ‡
For an aggregate loss distribution S :
(i) The number of claims has a negative binomial distribution with r = 16
and β = 6.
(ii) The claim amounts are uniformly distributed on the interval (0, 8).
(iii) The number of claims and claim amounts are mutually independent.
Using the normal approximation for aggregate losses, calculate the premium
such that the probability that aggregate losses will exceed the premium is
5%.
Solution.
The aggregate losses S has a compound distribution with primary function
the frequency N and secondary function the severity X. We have
E(N ) =rβ = 16(6) = 96

Var(N ) =rβ(1 + β) = 16(6)(7) = 672
0+8
E(X) = =4
2
(8 − 0)2
Var(X) = = 5.333
12
E(S) =E(N )E(X) = 96(4) = 384
=96(5.333) + 672(42 ) = 11264.
Let P denote the premium such that Pr(S > P ) = 0.05. Using normal
approximation, we have

S − 384 P − 384
0.05 =Pr(S > P ) = Pr √ > √
11264 11264

P − 384
=Pr Z > √
11264

P − 384
=1 − Pr Z ≤ √
11264

P − 384
=1 − Φ √ .
11264
Hence,
P − 384 P − 384
Φ √ = 0.95 =⇒ √ = 1.645.
11264 11264
Solving this last equation, we find P = 558.59
Practice Problems
Problem 38.1
Find the third raw moment of the aggregate loss random variable S : E[S 3 ].
Problem 38.2
Find the third central moment of the aggregate loss random variable S :
E[(S − E(S))3 ].
Problem 38.3
Let S = X1 + X2 + · · · XN where the Xi0 s are independent with common
distribution the lognormal distribution with parameters µ and σ. The fre-
quency random variable N has a Poisson distribution with parameter λ.
Write the integral representing FX∗2 (x).

Problem 38.4
Find the first and the second raw moments of S in Example 38.1.
Problem 38.5
Find the variance of S in Example 38.1.
Problem 38.6 ‡
For an insured, Y is the random variable representing the total amount
of time spent in a hospital each year. The distribution of the number of
hospital admissions in a year is:
# of Admissions Probability
0 0.60
1 0.30
2 0.10
The distribution of the length of each stay for an admission follows a Gamma
distribution with α = 1 and θ = 5.
Determine E(Y ) and Var(Y ).

Problem 38.7
For an insurance company, each loss has a mean of 100 and a variance of
100. The number of losses follows a Poisson distribution with a mean of
500. Each loss and the number of losses are mutually independent.
Using the normal approximation, calculate the probability that the aggre-
gate losses will exceed 52,250.
Problem 38.8
For an insurance company, each loss has a mean of 100 and a variance of
100. The number of losses follows a Poisson distribution with a mean of
500. Each loss and the number of losses are mutually independent.
Using the lognormal approximation, calculate the probability that the ag-
gregate losses will exceed 52,250.
Problem 38.9
An insurance company offers car insurance to a group of 1000 employees.
The frequency claim has negative binomial distribution with r = 1 and
β = 1.5. Severity claims are exponentially distributed with a mean of 5000.
Assume that the number of claims and the size of the claim are independent
and identically distributed.
Using the normal distribution as an approximating distribution of aggre-

gate losses, calculate the probability that losses will exceed 8 million.
Problem 38.10 ‡
When an individual is admitted to the hospital, the hospital charges have
the following characteristics:
Charges Mean Std deviation

Room 1000 500
Other 500 300
The covariance between an individual’s room charges and other charges is

100,000. An insurer issues apolicy that reimburses 100% for room charges
and 80% for other charges. The number of hospital admissions has Poisson
distribution with parameter 4.
Determine the mean and the standard deviation of the insurer’s payout
for the policy.
Problem 38.11 ‡
Computer maintenance costs for a department are modeled as follows:
(i) The distribution of the number of maintenance calls each machine will
need in a year is Poisson with mean 3.
(ii) The cost for a maintenance call has mean 80 and standard deviation
200.
(iii) The number of maintenance calls and the costs of the maintenance calls
are all mutually independent.

The department must buy a maintenance contract to cover repairs if there
is at least a 10% probability that aggregate maintenance costs in a given
year will exceed 120% of the expected costs.
Using the normal approximation for the distribution of the aggregate main-
tenance costs, calculate the minimum number of computers needed to avoid
purchasing a maintenance contract.
Problem 38.12 ‡
A towing company provides all towing services to members of the City Au-
tomobile Club. You are given:
Towing Distance (in miles) Towing Cost Frequency

0-9.99 80 50%
10-29.99 100 40%
30+ 160 10%
(i) The automobile owner must pay 10% of the cost and the remainder is
paid by the City Automobile Club.
(ii) The number of towings has a Poisson distribution with mean of 1000
per year.
(iii) The number of towings and the costs of individual towings are all mu-
tually independent.
Using the normal approximation for the distribution of aggregate towing

costs, calculate the probability that the City Automobile Club pays more
than 90,000 in any given year.
Problem 38.13 ‡
The number of auto vandalism claims reported per month at Sunny Daze In-
surance Company (SDIC) has mean 110 and variance 750. Individual losses
have mean 1101 and standard deviation 70. The number of claims and the
amounts of individual losses are independent.
Using the normal approximation, calculate the probability that SDIC’s ag-
gregate auto vandalism losses reported for a month will be less than 100,000.
Problem 38.14 ‡
At the beginning of each round of a game of chance the player pays 12.5.
The player then rolls one die with outcome N. The player then rolls N dice
and wins an amount equal to the total of the numbers showing on the N
dice. All dice have 6 sides and are fair.
Using the normal approximation, calculate the probability that a player

starting with 15,000 will have at least 15,000 after 1000 rounds. Caution:
Use continuity correction since a discrete distribution is being approximated
by a continuous one.
Problem 38.15 ‡
A dam is proposed for a river which is currently used for salmon breeding.
You have modeled:
(i) For each hour the dam is opened the number of salmon that will pass
through and reach the breeding grounds has a distribution with mean 100
and variance 900.
(ii) The number of eggs released by each salmon has a distribution with
mean of 5 and variance of 5.
(iii) The number of salmon going through the dam each hour it is open and
the numbers of eggs released by the salmon are independent.
Using the normal approximation for the aggregate number of eggs released,
determine the least number of whole hours the dam should be left open so
the probability that 10,000 eggs will be released is greater than 95%.
Problem 38.16 ‡
You are the producer of a television quiz show that gives cash prizes. The
number of prizes, N, and prize amounts, X, have the following distributions:
x Pr(X = x)
n Pr(N = n)
0 0.2
1 0.8
100 0.7
2 0.2
1000 0.1
Your budget for prizes equals the expected prizes plus the standard devia-
tion of prizes.
Calculate your budget.
Problem 38.17 ‡
The number of accidents follows a Poisson distribution with mean 12. Each
accident generates 1, 2, or 3 claimants with probabilities 12 , 13 , 16 , respectively.
Calculate the variance of the total number of claimants.
Problem 38.18 ‡
In a clinic, physicians volunteer their time on a daily basis to provide care
to those who are not eligible to obtain care otherwise. The number of physi-
cians who volunteer in any day is uniformly distributed on the integers 1
through 5. The number of patients that can be served by a given physician
has a Poisson distribution with mean 30.
Determine the probability that 120 or more patients can be served in a day
at the clinic, using the normal approximation with continuity correction.
Problem 38.19 ‡
For an individual over 65:
(i) The number of pharmacy claims is a Poisson random variable with mean
25.
(ii) The amount of each pharmacy claim is uniformly distributed between 5
and 95.
(iii) The amounts of the claims and the number of claims are mutually in-
dependent.
Determine the probability that aggregate claims for this individual will ex-
ceed 2000 using the normal approximation.
Problem 38.20 ‡
Two types of insurance claims are made to an insurance company. For each
type, the number of claims follows a Poisson distribution and the amount
of each claim is uniformly distributed as follows:
Type of Claim Poisson Parameter λ for Range of Each Claim

Number of Claims Amount
I 12 (0, 1)
II 4 (0, 5)
The numbers of claims of the two types are independent and the claim
amounts and claim numbers are independent.
Calculate the normal approximation to the probability that the total of

claim amounts exceeds 18.
Problem 38.21 ‡
For aggregate losses, S :
(i) The number of losses has a negative binomial distribution with mean 3
and variance 3.6.
(ii) The common distribution of the independent individual loss amounts is
uniform from 0 to 20.
Calculate the 95th percentile of the distribution of S as approximated by

the normal distribution.
Problem 38.22 ‡
In a CCRC, residents start each month in one of the following three states:
Independent Living (State #1), Temporarily in a Health Center (State #2)
or Permanently in a Health Center (State #3). Transitions between states
occur at the end of the month.
If a resident receives physical therapy, the number of sessions that the res-
ident receives in a month has a geometric distribution with a mean which
depends on the state in which the resident begins the month. The num-
bers of sessions received are independent. The number in each state at the
beginning of a given month, the probability of needing physical therapy in
the month, and the mean number of sessions received for residents receiving
therapy are displayed in the following table:
State # Number in Probability of Mean number
state needing therapy of visits
1 400 0.2 2
2 300 0.5 15
3 200 0.3 9
Using the normal approximation for the aggregate distribution, calculate the
probability that more than 3000 physical therapy sessions will be required
for the given month.
Problem 38.23 ‡
You are given:
(i) Aggregate losses follow a compound model.
(ii) The claim count random variable has mean 100 and standard deviation
25.
(iii) The single-loss random variable has mean 20,000 and standard devia-
tion 5000.
Determine the normal approximation to the probability that aggregate claims

exceed 150% of expected costs.
39 Stop Loss Insurance

When a deductible d is applied to the aggregate loss S over a definite period,
then the insurance payment will be

0, S≤d
(S − d)+ = max{S − d, 0} = S − S ∧ d =
S − d, S > d.
We refer to this payment as the stop-loss insurance. Its expected cost is
called the net stop-loss premium and can be computed as
Z ∞ Z ∞ Z ∞
E[(S − d)+ ] = S(x)dx = [1 − FS (x)]dx = (x − d)fS (x)dx
d d d
in the continuous case and

X
E[(S − d)+ ] = (x − d)fS (x)
x>d
in the discrete case. Note that this is identical to the discussion of ordinary
deductible of Section 31.
Example 39.1
The distribution of aggregate losses covered under a policy of stop-loss in-
surance is given by FS (x) = 1 − x12 , x > 1. Calculate E[(S − 3)+ ].
Solution.
We have ∞
1 ∞ 1
Z
dx
E[(S − 3)+ ] = =− =
3 x2 x 3 3
The following provides a simple calculation of the net stop-loss premium in
a special case.
Theorem 39.1
Suppose that Pr(a < S < b) = 0. Then, for a ≤ d ≤ b, we have
b−d d−a
E[(S − d)+ ] = E[(S − a)+ ] + E[(S − b)+ ].
b−a b−a
Proof.
Let a ≤ x ≤ b. Since a ≤ x and FS (x) is nondecreasing, we have FS (a) ≤
FS (x). On the other hand,
Z x Z a Z x Z b
FS (x) = fS (y)dy = fS (y)dy+ fS (y)dy ≤ FS (a)+ fS (y)dy = FS (a).
0 0 a a
39 STOP LOSS INSURANCE 285
Hence, FS (x) = FS (a) for all a ≤ x ≤ b. Next, we have

Z ∞
E[(S − d)+ ] = [1 − FS (x)]dx
d
Z ∞ Z d
= [1 − FS (x)]dx − [1 − FS (x)]dx
a a
Z d
=E[(S − a)+ ] − [1 − FS (a)]dx
a
=E[(S − a)+ ] − (d − a)[1 − FS (a)].
The above is true for all a ≤ d ≤ b. In particular, if d = b, we have

E[(S − a)+ − E[(S − b)+ ]
E[(S−b)+ ] = E[(S−a)+ ]−(b−a)[1−FS (a)] =⇒ 1−FS (a) = .
b−a
Hence,
d−a
E[(S − d)+ ] =E[(S − a)+ ] − {E[(S − a)+ − E[(S − b)+ ]}
b−a
b−d d−a
= E[(S − a)+ ] + E[(S − b)+ ]
b−a b−a
Example 39.2
A reinsurer pays aggregate claim amounts in excess of d, and in return it re-
ceives a stop-loss premium E[(S − d)+ ]. You are given E[(S − 100)+ ] =
15, E[(S − 120)+ ] = 10, and the probability that the aggregate claim
amounts are greater than 100 and less than to 120 is 0. Calculate E[(S −
105)+ ].
Solution.
We have

120 − 105 105 − 100
E[(S − 105)+ ] = (15) + (10) = 13.75
120 − 100 120 − 100
More simplification result follows.
Theorem 39.2
Suppose S is discrete and Pr(S = kh) ≥ 0 for some fixed h and k =
0, 1, 2, · · · . Also, Pr(S = x) = 0 for all x 6= kh. Then, for any nonnega-
tive integer j, we have
∞
X
E[(S − jh)+ ] = h {1 − FS [(n + 1)j]} .
n=0
In particular,
E[(S − (j + 1)h)+ ] − E[(S − jh)+ ] = h[FS (jh) − 1].
Proof.
We have
X
E[(S − jh)+ ] = (x − jh)fS (x)
x>jh
X∞ ∞
X
= (kh − jh)Pr(S = kh) = h (k − j)Pr(S = kh)
k=j k=j
∞ k−j−1
X X ∞ k−j−1
X X
=h Pr(S = kh) = h Pr(S = kh)
k=j n=0 n=0 k=n+j+1
X∞
=h {1 − FS [(n + 1)j]} .
n=0
Finally, we have
∞
X
E[(S − (j + 1)h)+ ] − E[(S − jh)+ ] =h {1 − FS [(n + 1)j + n + 1]}
n=0
X∞
−h {1 − FS [(n + 1)j]}
n=0
X∞
=h {1 − FS [(n + 1)j]}
n=1
X∞
−h {1 − FS [(n + 1)j]}
n=0
=h[FS (jh) − 1]
Example 39.3
Given the following information about the distribution of a discrete aggre-
gate loss random variable:
x 0 25 50 75
FS (x) 0.05 0.065 0.08838 0.12306
Calculate E[(S − 25)+ ], E[(S − 50)+ ], E[(S − 75)+ ], and E[(S − 100)+ ] given
that E(S) = 314.50.
Solution.
We have
E[(S − 25)+ ] =314.50 − 25(1 − 0.05) = 290.75
E[(S − 50)+ ] =290.75 − 25(1 − 0.065) = 267.375
E[(S − 75)+ ] =267.375 − 25(1 − 0.08838) = 244.5845
E[(S − 100)+ ] =244.5845 − 25(1 − 0.12306) = 222.661
Example 39.4 ‡
Prescription drug losses, S, are modeled assuming the number of claims has
a geometric distribution with mean 4, and the amount of each prescription
is 40. Calculate E[(S − 100)+ ].
Solution.
Let N denote the number of prescriptions and S the aggregate losses. Then
S = 40N. We have
E[(S − 100)+ ] =E(S) − E(S ∧ 100)
=40E(N ) − 40E(N ∧ 2.5)
=40E(N ) − 40[fN (1) + 2fN (2) + 2.5(1 − FN (2))]
=40(4) − 40[0.16 + 2(0.1280) + 2.5(0.5120)]
=92.16
where
4n
fN (n) =
5n+1
Example 39.5 ‡
WidgetsRUs owns two factories. It buys insurance to protect itself against
major repair costs. Profit equals revenues, less the sum of insurance premi-
ums, retained major repair costs, and all other expenses. WidgetsRUs will
pay a dividend equal to the profit, if it is positive.
You are given:
(i) Combined revenue for the two factories is 3.
(ii) Major repair costs at the factories are independent.
(iii) The distribution of major repair costs (k) for each factory is
k Probability(k)
0 0.4
1 0.3
2 0.2
3 0.1
(iv) At each factory, the insurance policy pays the major repair costs in
excess of that factorys ordinary deductible of 1. The insurance premium is
110% of the expected claims.
(v) All other expenses are 15% of revenues.
Calculate the expected dividend.
Solution.
Let R denote the retained major repair cost for both factories. That is, this
is the portion of the major repair costs to be covered by the ownership of the
factory. Note that R is an aggregate random variable and R = 0, 1, 2. Let D
stand for the dividend. The expected claims per factory for the insurance
company is
E[(X − 1)+ ] = (2 − 1)(0.2) + (3 − 1)(0.1) = 0.4.
Thus, the insurance company charges WidgetRUs a premium of (1.10)(2)(0.4) =

0.88. Hence, the formula of the dividend can be expressed as
D = 3 − 0.88 − R − 0.15(3) = 1.67 − R.
The distribution of R is found as follows:

For R = 0, both factories have no major repair costs and in this case Pr(R =
0) = (0.4)(0.4) = 0.16. For R = 1, one factory has a major repair cost and
the other does not. In this case, Pr(R = 1) = 2(0.4)(0.6) = 0.48. Hence,
Pr(R = 2) = 1 − Pr(R = 0) − Pr(R = 1) = 0.36. The distribution of D is
D Probability(D)
1.67 0.16
0.67 0.48
0 0.36
Finally,
E(D) = 1.67(0.16) + 0.67(0.48) = 0.5888
Example 39.6 ‡
For a collective risk model:
(i) The number of losses has a Poisson distribution with λ = 2.
(ii) The common distribution of the individual losses is:
x f (x)
1 0.6
2 0.4
An insurance covers aggregate losses subject to a deductible of 3. Calculate

the expected aggregate payments of the insurance.
Solution.
We have
E[(S − 3)+ ] =E[(S − 2)+ ] + FS (2) − 1
=E[(S − 1)+ ] + FS (2) + FS (1) − 2
=E(S) + FS (0) + FS (1) + FS (2) − 3
=E(S) + 3fS (0) + 2fS (1) + fS (2) − 3
e−2 22
=2(1.4) + 3e−2 + 2e−2 (0.6) + 2e−2 (0.4) + (0.6)2 − 3
2!
=0.7365
Example 39.7 ‡
A compound Poisson claim distribution has λ = 5 and individual claim
amounts distributed as follows:
x f (x)
5 0.6
k 0.4
where k > 5. The expected cost of an aggregate stop-loss insurance subject
to a deductible of 5 is 28.03.
Calculate k.
Solution.
The stop-loss insurance with deductible 5 pays
(S − 5)+ = S − S ∧ 5.
Thus,
E[(S − 5)+ ] = E(S) − E(S ∧ 5).
We have
E(S) = E(N )E(X) = 5[5(0.6) + k(0.4)] = 15 + 2k
and
E(S ∧ 5) =5[1 − FS (5)]
=5[1 − Pr(S = 0)]
=5[1 − Pr(N = 0)]
=5(1 − e−5 ) = 4.9663.
Thus,
28.03 = E[(S − 5)+ ] = 15 + 2k − 4.9663 =⇒ k = 9
Practice Problems
Problem 39.1
You are given: E[(S − 15)+ = 0.34 and E[(S − 30)+ ] = 0.55. Calculate
FS (15).
Problem 39.2 ‡
An aggregate claim distribution has the following characteristics: Pr(S =
i) = 16 for i = 1, 2, 3, 4, 5, 6. A stop-loss insurance with deductible amount d
has an expected insurance payment of 1.5 .
Find d.
Problem 39.3 ‡
You are given:
(i) S takes only positive integer values.
(ii) E(S) = 53 .
(iii) E[(S − 2)+ ] = 16 .
(iv) E[(S − 3)+ ] = 0.
Find the value of fS (1).
Problem 39.4 ‡
For a stop-loss insurance on a three person group:
(i) Loss amounts are independent.
(ii) The distribution of loss amount for each person is:
Loss Amount Probability
0 0.4
1 0.3
2 0.2
3 0.1
iii) The stop-loss insurance has a deductible of 1 for the group.
Calculate the net stop-loss premium.
Problem 39.5
Suppose that the aggregate loss random variable is discrete satisfying Pr(S =
1
50k) = 2k+1 for k = 0, 1, 2, · · · and Pr(S = x) = 0 for all other x.
Calculate E[(S − 150)+ ].

Problem 39.6 ‡
For a certain company, losses follow a Poisson frequency distribution with
mean 2 per year, and the amount of a loss is 1, 2, or 3, each with probability
1/3. Loss amounts are independent of the number of losses, and of each
other.
An insurance policy covers all losses in a year, subject to an annual aggre-
gate deductible of 2.
Calculate the expected claim payments for this insurance policy.
Problem 39.7 ‡
For a stop-loss insurance on a three person group:
(i) Loss amounts are independent.
(ii) The distribution of loss amount for each person is:
Loss Amount (X) Probability (X)
0 0.4
1 0.3
2 0.2
3 0.1
(iii) The stop-loss insurance has a deductible of 1 for the group.
Calculate the net stop-loss premium.
Problem 39.8 ‡
The number of annual losses has a Poisson distribution with a mean of 5.
The size of each loss has a two-parameter Pareto distribution with θ = 10
and α = 2.5. An insurance for the losses has an ordinary deductible of 5 per
loss.
Calculate the expected value of the aggregate annual payments for this in-
surance.
Problem 39.9 ‡
In a given week, the number of projects that require you to work overtime
has a geometric distribution with β = 2. For each project, the distribution
of the number of overtime hours in the week is the following:
x f (x)
5 0.2
10 0.3
20 0.5
The number of projects and number of overtime hours are independent. You
will get paid for overtime hours in excess of 15 hours in the week.
Calculate the expected number of overtime hours for which you will get
paid in the week.
40 Closed Form of Aggregate Distributions

Finding the distribution of an aggregate loss random variable via the method
of convolutions is practically very difficult. Mostly, this is done through nu-
merical methods which we will discuss in the coming sections. However, it
is possible to find a closed form for the distribution of some special families
of severities which is the topic of this section.
Let S be an aggregate loss random variable such that the severities are
all independent and identically distributed with common distribution the
exponential distribution with mean θ. We assume that the severities and
the frequency distributions are independent. By independence, we have
MS (z) = [MX (z)]n = (1 − θz)−n .
Thus, S is equivalent to a Gamma distribution with parameters θ and n and

with cdf
x
Pr(X1 + X2 + · · · + XN ≤ x) = FX∗n (x) = Γ n;
θ
where Γ(α; x) is the incomplete Gamma function defined by
Z x
1
Γ(α; x) = tα−1 e−t dt, x > 0, α > 0
Γ(α) 0
and where Z ∞
Γ(α) = tα−1 e−t dt.
0
If α = n is a positive integer then we can find a closed form of Γ(n : x).
Theorem 40.1
Let n be a positive integer. Then
n−1
X xj e−x
Γ(n; x) = 1 − .
j!
j=0
Proof.
The proof is by induction on n. For n = 1, we have
Z x 1−1 j −x
−t −x
X x e
Γ(1; x) = e dt = 1 − e =1− .
0 j!
j=0
40 CLOSED FORM OF AGGREGATE DISTRIBUTIONS 295
So suppose the formula is true for 1, 2, · · · , n. Using integration by parts,

we have
Z x
1
Γ(n + 1; x) = tn e−t dt
Γ(n + 1) 0
Z x
1 n −t x
n−1 −t
= −t e 0 + nt e dt
n! 0
1
= (−xn e−x ) + Γ(n; x)
n!
n−1
−xn e−x X xj e−x
= +1−
n! j!
j=0
n
X xj e−x
=1 −
j!
j=0
Now, from Section 38, we have

∞
X x ∞
X
FS (x) = Pr(N = n)FX∗n (x) = Pr(N = 0) +
Pr(N = n)Γ n;
θ
n=0 n=1
 
∞
X n−1
X (x/θ)j e−x/θ
=Pr(N = 0) + Pr(N = n) 1 −
 
j!
n=1 j=0
∞ n−1
X X (x/θ)j e−x/θ
=1 − Pr(N = n)
j!
n=1 j=0
∞ ∞
x X (x/θ)j X
=1 − e− θ Pr(N = n)
j!
j=0 n=j+1
∞
x X (x/θ)j
=1 − e− θ Pj , x≥0
j!
j=0
where
∞
X
Pj = Pr(N = n), j = 0, 1, · · · .
n=j+1
The pdf of S can be found upon differentiating FS (x).
Example 40.1
Find FS (x) if the frequency N has a binomial distribution with parameters
n and m.
Solution.
We have P j = 0 for j = m, m + 1, · · · . Hence,
m n−1
X (x/θ)j
X m n m−n
FS (x) = 1 − q (1 − q)
n j!
n=1 j=0
When the subscripted independent random variables Xi0 s belong to a same

family of distributions and the resulting sum S belongs to the same family,
we say that the family is closed under convolution.
Example 40.2
Suppose that X1 , X2 , · · · , XN are independent normal random variables
with parameters (µ1 , σ12 ), (µ2 , σ22 ), · · · , (µN , σN
2 ). Show that S = X + X +
1 2
· · · + XN is also a normal random variable.
Solution.
By independence, we have
MS (z) =MX1 (z)MX2 (z) · · · MXN (z)
2 2 2
=eµ1 +σ1 t eµ2 +σ2 t · · · eµN +σN t
2 2 2
=e(µ1 +µ2 +···+µN )+(σ1 +σ2 +···+σN )t .
Hence, S has a normal distribution with parameters (µ1 +µ2 +· · ·+µN , σ12 +
σ22 + · · · + σN
2 )
A compound Poisson distribution is an aggregate distribution S where

the frequency random variable is Poisson.
Theorem 40.2
Suppose that S1 , S2 , · · · , Sn are compound Poisson distributions with pa-
rameters λ1 , λ2 , · · · , λn and severity distributions with cdf’s F1 (x), F2 (x), · · · , Fn (x).
Suppose that the Si0 s are independent. Then S = S1 + · · · + Sn has a com-
pound Poisson distribution with parameter λ = λ1 + λ2 + · · · + λn and
severity distribution with cdf
n
X λi
F (x) = Fj (x).
λ
j=1
Proof.
Let Mj (t) be the mgf of Fj (x), Then
MSj (t) = PSj [Mj (t)] = eλj [Mj (t)−1] .

Hence, by independence, we have

n
Y n
Y
MS (t) = MSj (t) = eλj [Mj (t)−1]
j=1 j=1
=e(λ1 M1 (t)+λ2 M2 (t)+···+λn Mn (t))−λ

λ1 λ
λ M1 (t)+ λ2 M2 (t)+···+ λλn Mn (t) −λ
=e λ
.
Thus, S is a compound Poisson distribution with parameter λ = λ1 + λ2 +

· · · + λn and a severity distribution with cdf
n
X λi
F (x) = Fj (x)
λ
j=1
Example 40.3
Let S1 , S2 , S3 , and S4 be independent compound Poisson with parameters
λ1 = 1, λ2 = 2, λ3 = 3, λ4 = 4 and severities with cdf Fj (x) = x2 for
0 ≤ x ≤ 1. Let S = S1 + S2 + S3 + S4 . Find the mgf of S.
Solution.
The moment generating function of Fj (x) is
1
et et
Z
1
M (t) = Mj (t) = etx (2x)dx = − 2 + 2.
0 t t t
Thus,   
4
 X λj 
λ  Mj (t) − 1 = 10[M (t) − 1].
 λ 
j=1
Hence, t t

10[M (t)−1] 10 et − e2 + 1
−1
MS (t) = e =e t t2
Practice Problems
Problem 40.1
Suppose that the frequency distribution N of S has a negative binomial
with integer value r and parameter β. Suppose also that the severities are
identically distributed with common distribution the exponential distribu-
tion with mean θ. Suppose that the severities are independent among each
other and with N.
Find the moment generating function. Show that the given model is equiv-
alent to the binomial-exponential model.
Problem 40.2
Show that the cdf of the model of Exercise 40.1 is given by
r n r−n
X r β 1
FS (x) = 1 − × Pn (x)
n 1+β 1+β
n=1
where
n−1 −1 (1+β)−1
X [xθ−1 (1 + β)−1 ]j e−xθ
Pn (x) = .
j!
j=0
Problem 40.3
Suppose N has a geometric distribution with parameter β. The severities
are all exponential with mean θ. The severities and frequency are indepen-
dent.
Find a closed form of FS (x).
Problem 40.4
Suppose N has a geometric distribution with parameter β. The severities
are all exponential with mean θ. The severities and frequency are indepen-
dent.
Find a closed form of fS (x).
Problem 40.5
Show that the family of Poisson distributions is closed under convolution.
Problem 40.6
Show that the family of binomial distributions is closed under convolution.
Problem 40.7
Show that the family of negative binomial distributions with the same pa-
rameter β but different r0 s is closed under convolution.
Problem 40.8
Show that the family of Gamma distributions with common paramter θ is
closed under convolution.
Problem 40.9
Let S be an aggregate loss random variable with a discrete frequency distri-
bution N defined by the table below.
n 0 1 2
Pr(N = n) 0.5 0.4 0.1
The severity claim has an exponential distribution with mean 2. Find FS (x).
Problem 40.10
Let S1 , S2 , · · · , S5 be i.i.d. compound Poisson random variables, each with
parameter λj = 7. The pdf of the severity distribution for each Sj is Fj (x) =
x2 for 0 < x < 1. For S = S1 + S2 + · · · + S5 , find MS (2).
41 Distribution of S via the Recursive Method

As pointed out in Section 38, the n−fold convolutions are often extremely
difficult to compute in practice and therefore one encounters difficulties deal-
ing with the probability distribution of S. An alternative approach is to use
various approximation techniques. In this section, we consider a technique
known as the Panjer recursive formula.
The recursion method is used to find the distribution of S when the fre-
quency distribution belongs to either the (a, b, 0) class or the (a, b, 1) class
and the severity is integer valued and nonnegative. So let N be in the (a, b, 1)
class and denote its distribution by pn = Pr(N = n). Then pn satisfies the
equation
b
pn = a + pn−1 , n = 2, 3, · · · . (41.1)
n
We have the following theorem.
Theorem 41.1
For a frequency distribution in (a, b, 1), the pdf of S is recursively defined
by
n
X b
[p1 − (a + b)p0 ]fX (n) + a + j fX (j)fS (n − j)
n
j=1
fS (n) = (41.2)
1 − afX (0)
where fS (n) = Pr(S = n).
Proof.
Rewrite (41.1) in the form
npn = (n − 1)apn−1 + (a + b)pn−1 , n = 2, 3, · · ·
Multiplying each side by [PX (z)]n−1 PX0 (z), summing over n and reindexing
yields
∞
X ∞
X
npn [PX (z)] n−1
PX0 (z) − p1 PX0 (z) =a (n − 1)pn−1 [PX (z)]n−1 PX0 (z)
n=1 n=2
∞
X
+(a + b) pn−1 [PX (z)]n−1 PX0 (z)
n=2
41 DISTRIBUTION OF S VIA THE RECURSIVE METHOD 301
∞
X ∞
X
=a npn [PX (z)]n PX0 (z) + (a + b) pn [PX (z)]n PX0 (z)
n=1 n=1
X∞ X∞
=a npn [PX (z)]n PX0 (z) + (a + b) pn [PX (z)]n PX0 (z) − p0 (a + b)PX0 (z).
n=1 n=0
P∞
Because PS (z) = n=0 pn [PX (z)]n (see similar argument in Section 36), the
previous calculation yields
PS0 (z) − p1 PX0 (z) = aPS0 (z)PX (z) + (a + b)PS (z)PX0 (z) − p0 (a + b)PX0 (z)
or
PS0 (z) = [p1 − (a + b)p0 ]PX0 (z) + aPS0 (z)PX (z) + (a + b)PS (z)PX0 (z).
Each side can be expanded in powers of z. The coefficient of z n−1 in such an
expansion must be the same on both sides of the equation. Hence, obtaining
n
X
nfS (n) =[p1 − (a + b)p0 ]nfX (n) + a (n − j)fX (j)fS (n − j)
j=0
n
X
+(a + b) jfX (j)fS (n − j) = [p1 − (a + b)p0 ]nfX (n) + anfX (0)fS (n)
j=0
n
X n
X
+a (n − j)fX (j)fS (n − j) + (a + b) jfX (j)fS (n − j)
j=1 j=1
=[p1 − (a + b)p0 ]nfX (n) + anfX (0)fS (n)

Xn n
X
+an fX (j)fS (n − j) + b jfX (j)fS (n − j).
j=1 j=1
Hence
n
X bj
[1 − afX (0)]fS (n) = [p1 − (a + b)p0 ]fX (n) + a+ fX (j)fS (n − j).
n
j=1
Finally, the result follows by dividing both sides of the last equation by
1 − afX (0)
Corollary 41.2
If N is in the (a, b, 0) class then
n
X b
a + j fX (j)fS (x − j)
x
j=1
fS (x) = Pr(S = x) = (41.3)
1 − afX (0)
Proof.
p1
If N is in the (a, b, 0) class then from p0 = a + b we find p1 − (a + b)p0 = 0
so that (41.2) reduces to (41.3)
The recursive method requires an initial starting value fS (0) which can be
found as follows:
fS (0) = Pr(S = 0) = Pr(N = 0) = PN [PX (0)] = PN [Pr(X = 0)].
Example 41.1
Develop the recursive formula for the case of compound Poisson distribution
with parameter λ.
Solution.
The initial value is
fS (0) = PN [PX (0)] = eλ(fX (0)−1) .
For the Poisson distribution, a = 0 and b = λ. The recursive method reduces

to
n
X λj
fS (n) = fX (j)fS (n − j)
n
j=1
Example 41.2 ‡
For a tyrannosaur with a taste for scientists:
(i) The number of scientists eaten has a binomial distribution with q = 0.6
and m = 8.
(ii) The number of calories of a scientist is uniformly distributed on (7000, 9000).
(iii) The numbers of calories of scientists eaten are independent, and are in-
dependent of the number of scientists eaten.
Calculate the probability that two or more scientists are eaten and exactly
two of those eaten have at least 8000 calories each.
Solution.
If X denotes the number of calories of a scientist and N the number of
scientists eaten. Then
9000 − 8000 1
Pr(8000 ≤ X ≤ 9000) = FX (8000) − FX (9000) = = .
9000 − 7000 2
Thus, half the scientists are considered heavy. Let Y denote a heavy scientist
S the total number of heavy scientist. We are asked to find Pr(S = 2) =
fS (2). Note that S is a compound distribution with primary distribution the

number of scientist (a binomial distribution) and a secondary distribution
standing for a heavy scientist (a Bernoulli distribution with a probability of
a success equals to 0.5). Using Panjer algorithm, we have
q −0.6
a=− =− = −1.5
1−q 0.4
q
b =(m + 1) = 9(1.5) = 13.5
1−q
fS (0) =PN [Pr(X = 0)] = PN (0.5)
=[1 + 0.6(0.5 − 1)8 ] = 0.057648

1 13.5
fS (1) = −1.5 + fX (1)fS (0)
1 + 1.5(0.5) 1
1
= (12)(0.5)(0.057648) = 0.197650
1.75
1 13.5
fS (2) = −1.5 + fX (1)fS (0) + (−1.5 + 13.5)fX (2)fS (1)
1 + 1.5(0.5) 2
1
= [(5.25)(0.5)(0.057648) + 12(0)(0.197650)] = 0.29647
1.75
Example 41.3 ‡
Let S have a Poisson frequency distribution with parameter λ = 5. The
individual claim amount has the following distribution:
x fX (x)
100 0.8
500 0.16
1000 0.04
Calculate the probability that aggregate claims will be exactly 600.
Solution.
In order to have exactly 600 in aggregate claims, one of following must
happen:
• N = 2, X1 = 100, and X2 = 500. In this case, the probability is
e−5 52
(0.8)(0.16) = 0.010781.
2!
• N = 2, X1 = 500, and X2 = 100. In this case, the probability is
e−5 52
(0.16)(0.8) = 0.010781.
2!
• N = 6, X1 = · · · = X6 = 100. In this case, the probability is
e−5 56
(0.8)6 = 0.038331.
6!
Summing up, we find a total probability of about 0.06
Practice Problems
Problem 41.1
Let S have a Poisson frequency distribution with parameter λ = 0.04. The
x fX (x)
1 0.5
2 0.4
3 0.1
Find fS (x), x = 1, 2, 3, 4 using Panjer recursion formula.
Problem 41.2
Let S be a compound Poisson distribution with parameter λ = 0.04 and
individual claim distribution given by
x fX (x)
1 0.5
2 0.4
3 0.1
Show that: fS (n) = n1 [0.02fS (n − 1) + 0.032fS (n − 2) + 0.012fS (n − 3)].
Problem 41.3 ‡
You are given:
• S has a compound Poisson distribution with λ = 2.
• Individual claim amounts are distributed as follows: fX (1) = 0.4 and
fX (2) = 0.6.
Determine fS (4).
Problem 41.4 ‡
Aggregate claims S has a compound Poisson distribution with parameter λ
and with discrete individual claim amount distributions of fX (1) = 13 and
fX (3) = 23 . Also, fS (4) = fS (3) + 6fS (1).
Determine the value of λ.
Problem 41.5 ‡
Aggregate claims S has a compound Poisson distribution with parameter λ
and with discrete individual claim amount distributions of fX (1) = 13 and
fX (3) = 32 . Also, fS (4) = fS (3) + 6fS (1).
Determine Var(S).
Problem 41.6 ‡
For aggregate claim S, you are given:
∞
X
∗n e−50 (50n )
fS (x) = fX (x) .
n!
n=0
Losses are distributed as follows: fX (1) = 0.4 fX (2) = 0.5, and fX (3) = 0.1.
Calculate Var(S).
Problem 41.7
Let S have a Poisson frequency distribution with parameter λ = 0.04. The
x fX (x)
1 0.5
2 0.4
3 0.1
Find FS (4).
Problem 41.8
The frequency distribution of an aggregate loss S follows a binomial dis-
tribution with m = 4 and q = 0.3. Loss amount has the distribution:
fX (0) = 0.2, fX (1) = 0.7, fX (2) = 0.1
Find the starting value fS (0) in the recursion formula.
Problem 41.9
fX (0) = 0.2, fX (1) = 0.7, fX (2) = 0.1
Calculate fS (1) and fS (2).
Problem 41.10
fX (0) = 0.2, fX (1) = 0.7, fX (2) = 0.1
Calculate Pr(S ≥ 3).
Problem 41.11
Annual aggregate losses for a dental policy follow the compound Poisson
distribution with λ = 3. The distribution of individual losses is:
Loss Probability
1 0.4
2 0.3
3 0.2
4 0.1
Calculate the probability that aggregate losses in one year do not exceed 3.
42 Discretization of Continuous Severities

The severity distribution in the recursive method of Section 41 was assumed
to be discrete. Most severity distributions are of continuous type. An analog
of Panjer recursive method with a continuous severity is
Z x
by
fS (x) = (a + b)Pr(S = 0)fX (x) + a+ fX (y)fS (x − y)dy
0 x
where the probability of the severity distribution is taken on the positive
real line. This is a Volterra integral equation of the second kind. This type
of equations is usually very difficult to solve analytically. Instead, numerical
methods are used. The two methods discussed in this section are based on
the discretization of the continuous severiy distribution. Once a continuous
severity is discretized, the recursive method can be applied.
The Method of Rounding

An arithmetic distribution is a distribution with support N = {1, 2, · · · }.
An equispaced arithmetic distribution is defined on positive multiples of a
unit of measurement h which we call the span. Transforming a continuous
distribution to an arithmetic distribution is referred to as discretizing or
arithmetizing the distribution.
Let X denote the random variable to be arithmetized and h > 0 denote
the span. The method of rounding (also known as the method of mass
dispersal) concentrates the probability one-half span either side of jh and
places it at jh :

h h
f0 =Pr X < = FX
2 2

h h
fj =Pr jh − ≤ X < jh +
2 2

h h
=FX jh + − FX jh − , j = 1, 2, · · ·
2 2
At some point, it is reasonable to halt the discretization process at some
point, say mh, once most of the probability has been accounted for. At
this terminal point, we have fm = 1 − FX mh − h2 . With this method, all
the fj sum up to 1, ensuring that the discretization leads to a legitimate
probability distribution.
Example 42.1
Let S be an aggregate random variable with a frequency distribution that
42 DISCRETIZATION OF CONTINUOUS SEVERITIES 309
has a Poisson distribution with λ = 2 and a severity distribution that has a

Pareto distribution with parameters α = 2 and θ = 1000.
(a) Use the method of rounding to approximate f0 , f1 , f2 and f3 in the
arithmetize distribution such that f3 is the last positive probability. Use
the span h = 6.
(b) Find fS (12).
Solution.
(a) The cdf of the severity distribution is
2
1000
FX (x) = 1 − .
x + 1000
Thus,
2
1000
f0 =FX (3) = 1 − = 0.006
3 + 1000
2 2
1000 1000
f1 =FX (9) − FX (3) = − = 0.0118
3 + 1000 9 + 1000
2 2
1000 1000
f2 =FX (15) − FX (9) = − = 0.0116
9 + 1000 16 + 1000
2
1000
f3 =1 − FX (15) = = 0.9707.
15 + 1000
(b) Let A denote the aggregate random variable with frequency distribution
the Poisson distribution with λ = 2 and severity distribution the arithmetize
distribution. We want to find fS (12) = fA (2). Using the recursive method,
we find
fA (0) =eλ(f0 −1) = 0.137

fA (1) =2[f1 fA (0)] = 0.0032332
fA (2) =[f1 fA (1) + 2f2 fA (0)] = 0.00322
The Method of Local Moment Matching

The rounding method has the disadvantage that the approximate distri-
bution does not preserve any moment of the exact distribution. Gerber
proposed a method that matched moments locally: The local moment
matching method. This method constructs discrete equispaced distri-
butions that matches some moments of the exact distribution. The idea
is to replace the probability of the severity distribution X over intervals

[xk , xk + ph) by point masses located at xki = xk + ih where i = 0, 1, · · · , p
and where the integer p represents the number of moments to be matched.
The method then tries to calculate the p + 1 point masses mk0 , mk1 , · · · , mkp
located at xk0 , xk1 , · · · , xkp which are solutions to the (p + 1) equations
p
X Z xk +ph
r
(xk + jh) mkj = xr dFX (x), r = 0, 1, 2, · · · , p. (42.1)
j=0 xk
In practice, the above system is solved in each of the intervals
[0, ph), [ph, 2ph), [2ph, 3ph), · · · ,
that is, xk+1 = xk + ph with x0 = 0 and therefore xk = kph. The final

probabilities of the discretized distribution are the point masses obtained as
solutions of the previous system in each interval, summed for the endpoints
of each interval:
f0 = m00 f1 = m01 f2 = m02 ···
fp = m0p + m10 fp+1 = m11 fp+2 = m12 ···
The unique solution to the system (42.1) is provided by the following theo-
rem.
Theorem 42.1
The solution to (42.1) is
 
Z xk +ph Y x − xk − ih
mkj =   dFX (x), j = 0, 1, · · · , p.
xk (j − i)h
i6=j
Proof.
Let f (x) be a polynomial and consider the set of data points
(x0 , f (x0 )), (x1 , f (x1 )), · · · , (xp , f (xp )).
Then Lagrange interpolation formula allows us to write

 
X p Y x − xi
f (x) = f (xj )  .
xj − xi
j=0 i6=j
In particular, by letting f (x) = xr and xi = xk + ih (i = 0, 1, · · · , p) we find

 
Xp Y x − xk − ih
xr = (xk + jh)r   , r = 0, 1, 2, · · ·
(j − i)h
j=0 i6=j
Integrate both sides with respect to FX (x) over the interval [xk , xk + ph) we
find
 
p p Z xk +ph Y
X X x − xk − ih 
(xk + jh)r mkj = (xk + jh)r  dFX (x).
xk (j − i)h
j=0 j=0 i6=j
By uniqueness of the solution to the system the desired result follows
Example 42.2
Suppose X has the exponential distribution with pdf fX (x) = 0.1e−0.1x . Use
the method of local moment mathing with p = 1 and a span h = 2
(a) to find the equation corresponding to r = 0 in the resulting system of
equations. Assume that x0 = 0;
(b) to find mk0 and mk1 using Theorem 42.1;
(c) to find f0 and fn .
Solution.
(a) We have
1
X Z 2k+2 Z 2k+2
0 k
(xk +jh) mj = fX (x)dx = 0.1 e−0.1x dx = e−0.2k −e−0.2k−0.2
j=0 2k 2k
which implies mk0 + mk1 = e−0.2k − e−0.2k−0.2 .

(b) Using Theorem 42.1, we find
Z 2k+2
x − 2k − 2
k
m0 = [0.1e−0.1x ]dx = 5e−0.2k−0.2 − 4e−0.2k
2k −2
and
2k+2
x − 2k
Z
mk1 = [0.1e−0.1x ]dx = −6e−0.2k−0.2 + 5e−0.2k .
2k 2
(c) We have
f0 =m00 = 5e−0.2 − 4 = 0.09365

fn =mn−1
1 + mn0 = 5e−0.2n+0.2 − 10e−0.2n + 5e−0.2n−0.2
Example 42.3
Loss amounts follow a Pareto distribution with parameters α = 3 and θ = 5.
Use the Method of Local Moment Matching with h = 3 and p = 1 to find
the system of two equations in the unknowns m00 and m01 . Solve this system.
Assume x0 = 0.
Solution.
We have
3
5
m00 + m01 = Pr(0 ≤ X < 3) = FX (3) − FX (0) = 1 − = 0.7559.
3+5
On the other hand,

3 3
3(5)3 x
Z Z
x0 m00 + x1 m01 = xfX (x)dx = dx = 0.79102.
0 0 (x + 5)4
This implies
3m01 = 0.79102 =⇒ m01 = 0.2637
and
m00 = 0.7559 − 0.2637 = 0.4922
Practice Problems
Problem 42.1
Let S be an aggregate random variable with a frequency distribution that
has a Poisson distribution with λ = 2 and a severity distribution that has a
uniform distribution in (0, 50).
(a) Use the method of rounding to approximate f0 , f1 , f2 and f3 in the
arithmetize distribution such that f3 is the last positive probability. Use
the span h = 10.
(b) Find fS (30).
Problem 42.2
Medical claims have an exponential distribution with mean θ = 500. Using
the method of rounding with a span h = 50 estimate the probability that
the claim amount is 500.
Problem 42.3
Loss amounts follow a Weibull distribution with parameters α = 2 and
θ = 1000. Using the method of rounding with a span h = 500 estimate the
probability that the loss amount is 2000.
Problem 42.4
the system in two equations in the unknowns m10 and m11 . Solve this system.
Assume x0 = 0.
Problem 42.5
f0 and f3 .
Problem 42.6
Loss amounts follow an exponential distribution θ = 500. A discrete distri-
bution is created using the Method of Local Moment Matching such that
p = 1 and h = 50. Calculate the probability to a loss amount of 500 using
Theorem 42.1.
Problem 42.7
Loss amounts follow a Pareto distribution α = 4 and θ = 10000. A discrete
distribution is created using the Method of Local Moment Matching such
that p = 1 and h = 300. Calculate the probability to a loss amount of 3000
using Theorem 42.1.
43 Individual Policy Modifications Impact on Ag-

gregate Losses
Any individual policy modifications such as deductibles, policy limits, or
coinsurance will have an impact on the aggregate loss/payment random
variable. In this section, it will be assumed that the ground-up distribution
of the individual loss amount X is unaffected by policy modifications, and
only the payments themselves are affected.
We next start by reminding the reader of the following notations: The num-
ber of losses will be denoted by N L ; the number of payments by N P , the
probability of a loss resulting in a payment by v; the amount of payment
on a per-loss basis is Y L with Y L = 0 if a loss results in no payment; and
the amount of payment on a per-payment basis is Y P where only losses
that result on a nonzero payment are considered and the losses that result
in no payment are completely ignored. Note that Pr(Y P = 0) = 0 and
Y P = Y L |Y L > 0.
The cdfs of Y L and Y P are related as follows:
FY L (y) =Pr(Y L ≤ y|Y L = 0)Pr(Y L = 0) + Pr(Y L ≤ y|Y L > 0)Pr(Y L > 0)

=1 − v + vFY P (y).
The mgfs of Y L and Y P are related as follows:

L L
MY L (t) =E(etY |Y L = 0)Pr(Y L = 0) + E(etY |Y L > 0)Pr(Y L > 0)
=(1 − v) + vMY P (t).
The pgfs of N P and N L are related as follows (see Section 36):
PN P (z) = PN (1 − v + vz).
Now back to the aggregate payments. On a per-loss basis, the total payments
may be expressed as
S = Y1L + Y2L + · · · + YNLL
with S = 0 if N L = 0 and where YjL is the payment amount on the jth loss.
Alternatively, ignoring all losses that do not result in a nonzero-payment,
the aggregate S can be expressed as
S = Y1P + Y2P + · · · + YNPP

43 INDIVIDUAL POLICY MODIFICATIONS IMPACT ON AGGREGATE LOSSES315
with S = 0 if N P = 0 and where YjP is the payment amount on the jth loss
which results in a nonzero payment. On a per-loss basis, S is a compound
distribution with primary distribution N L and secondary distribution Y L
so that
MS (t) = PN L [MY L (t)].
Likewise, on a per-payment basis, we have
MS (t) = PN P [MY P (t)].
Note that
PN L [MY L (t)] = PN L [1 − v + vMY P (t)] = PN P [MY P (t)].
Note that individual policy modifications factor either in the expression of

Y L or Y P and thus have an impact on the aggregate payment.
Example 43.1
A ground-up model of individual losses follows a Pareto distribution with
α = 4 and θ = 10. The number of losses is a Poisson distribution with λ = 3.
There is an ordinary deductible of 6, a policy limit of 18− applied before the
policy limit and the coinsurance, and coinsurance of 75%. Find the expected
value and the variance of S, the aggregate payments on a per-loss basis.
Solution.
We have
E(N L ) =3
" α−1 #
θ θ
E(X ∧ 24) = 1−
α−1 x+θ
" 4−1 #
10 10
= 1− = 3.2485
3 24 + 10
" 4−1 #
10 10
E(X ∧ 6) = 1− = 2.5195
3 6 + 10
E(Y L ) =α[E(X ∧ u) − E(X ∧ d)] = 0.75[E(X ∧ 24) − E(X ∧ 6)]
=0.75(3.2485 − 2.5195) = 0.54675.
E(S) =E(N L )E(Y L ) = 3(0.54675) = 1.64

Z 24 4
4(10)4

2 2 2 10
E[(X ∧ 24) ] = x dx + 24 = 26.379
0 (x + 10)5 24 + 10
Z 6 4
4(10)4

2 2 2 10
E[(X ∧ 6) ] = x dx + 6 = 10.5469
0 (x + 10)5 6 + 10
E[(Y L )2 ] =(0.75)2 E[(X ∧ 24)2 ] − E[(X ∧ 6)2 ]]

−2(6)E(X ∧ 24) + 2(d)E(X ∧ 6) } = 3.98481

Var(Y ) =3.98481 − 0.546752 = 3.685874438
L
Var(S) =E(N L )Var(Y L ) + E(Y L )2 Var(N L )

=3(3.685874438) + 0.546752 (3) = 11.95443
Example 43.2
α = 4 and θ = 10. The number of losses is a Poisson distribution with
λ = 3. There is an ordinary deductible of 6, a policy limit of 18− applied
before the policy limit and the coinsurance, and coinsurance of 75%. Find
the distribution of S, the aggregate payments on a per-payment basis.
Solution.
Since we are treating S on a per-payment basis, we can look at S as an aggre-
gate distribution with frequency distribution N P and severity distribution
Y P . The probability that a loss will result in a payment is
4
10
v = Pr(X > 6) = SX (6) = = 0.15259.
10 + 6
We also have (see Section 36)
E(N P ) = E(N L )E(Ij ) = λv = 3(0.15259) = 0.45776.
Let Z = X − 6|X > 6 denote the individual payment random variable with
only a deductible of 6. Then
Pr(X > z + 6)
Pr(Z > z) = .
Pr(X > 6)
Now with the 75% coinsurance, Y P = 0.75Z and the maximum payment is
0.75(24 − 6) = 13.5 so that for y < 13.5 we have
y
Pr(X > 6) − Pr X > 6 + 0.75
FY P (y) = 1 − Pr(0.75Z > y) =
Pr(X > 6)
and FY P (y) = 1 for y ≥ 13.5.

Now since Y P is continuous, we will use the method of rounding to discretize
it. We will use a span of 2.25 to obtain
Pr(X > 6 + 1.125

0.75 )
f0 =FY P (2.25/2) = 1 − = 0.30124
0.15259
f1 =FY P (3.375) − FY P (1.125) = 0.32768
.. ..
. =.
f6 =FY P (14.625) − FY P (12.375) = 0.05874
fn =1 − 1 = 0, n = 7, 8, · · · .
Now, S can be computed using the compound Poisson recursive formula

n
0.45776 X
fS (n) = yfn fS (n − j), n = 1, 2, 3, · · ·
n
j=1
with fS (0) = e0.45776(1−0.30124) = 0.72625
Example 43.3 ‡
Aggregate losses for a portfolio of policies are modeled as follows:
(i) The number of losses before any coverage modifications follows a Poisson
distribution with mean λ.
(ii) The severity of each loss before any coverage modifications is uniformly
distributed between 0 and b.
The insurer would like to model the impact of imposing an ordinary de-
ductible, d (0 < d < b), on each loss and reimbursing only a percentage,
c(0 < c ≤ 1), of each loss in excess of the deductible.
It is assumed that the coverage modifications will not affect the loss distri-
bution. The insurer models its claims with modified frequency and severity
distributions. The modified claim amount is uniformly distributed on the
interval [0, c(b − d)].
Determine the mean of the modified frequency distribution.
Solution.
Imposing the deductible will limit payments to those losses that are greater
than d. In this case, the proprotion of losses that result in a payment is
Pr(X > d) = 1 − db (remember that loss amounts are uniform in (0, b).)
Thus, the mean of the modified frequency distribution, that is, the expected
number of losses that will result in a payment being made is the product of
the total expected number of losses and the probability of a loss resulting
in a payment. That is
b
λ 1−
d
Example 43.4 ‡
A company insures a fleet of vehicles. Aggregate losses have a compound
Poisson distribution. The expected number of losses is 20. Loss amounts,
regardless of vehicle type, have exponential distribution with θ = 200.
In order to reduce the cost of the insurance, two modifications are to be
made:
(i) a certain type of vehicle will not be insured. It is estimated that this will
reduce loss frequency by 20%.
(ii) a deductible of 100 per loss will be imposed.
Calculate the expected aggregate amount paid by the insurer after the mod-
ifications.
Solution.
We want
E(S) = E(N L )E(Y L ) = E(N L )E[(X−100)+ ] = E(N L )[E(X)−E(X∧100)].
From Table C,
100
E(X ∧ 100) = 200(1 − e− 200 ).
Thus, h i
100
E(S) = (20)(0.8) 200 − 200(1 − e− 200 ) ≈ 1941
Practice Problems
Problem 43.1
There is an ordinary deductible of 6, a policy limit of 18− applied before
the policy limit and the coinsurance, and coinsurance of 75%.
Find FY L (y).
Problem 43.2
An insurance company has a policy where the amount of each payment for
losses follows an exponential distribution with mean θ = 100. The number
of losses follows a Poisson distribution with λ = 7. There is an ordinary de-
ductible of 30, a maximum covered losses of 340− applied before the policy
limit and the coinsurance, and coinsurance of 53%.
Find the expected value and variance of S, the aggregate payments on a

per-loss basis.
Problem 43.3
limit and the coinsurance, and coinsurance of 53%. Let S be the aggregate
payments on a per-payment basis.
Find the distribution of Y P .
Problem 43.4
Find f0 and fn under the method of rounding using a span of 30.
Problem 43.5

Find fS (0) and fS (n).
Problem 43.6
An insurance company has a policy where the amount of each payment
for losses on a per-payment basis follows an exponential distribution with
mean θ = 100. The probability that a loss will result in a payment is 0.74082.
Find MY L (t).
Problem 43.7
There is an ordinary deductible of 6, a policy limit of 18− applied before
the policy limit and the coinsurance, and coinsurance of 75%.
Find PN P (z).
44 AGGREGATE LOSSES FOR THE INDIVIDUAL RISK MODEL 321
44 Aggregate Losses for the Individual Risk Model

Consider a portfolio of n insurance policies or risks. Denote the loss to
the insurer, for a fixed period, for each policy i by Xi , for i = 1, 2, · · · , n.
We assume that the Xi0 s are independent but not necessarily identically
distributed. The individual risk model represents the aggregate losses
for the fixed period as a fixed sum
S = X1 + X2 + · · · + Xn .
Let qi denote the probability of claim within a fixed period of the ith pol-
icyholder and let bi denote the fixed benefit to the ith policyholder. Then
the pdf of the loss to the insurer is

1 − qi , x = 0
fXj (x) =
qi , x = bi .
On the other hand, the pgf of Xi is
PXi (z) = E(z Xi ) = 1 − qi + qi z bi .
By the independence of the Xi0 s, we can write
n
Y
PS (z) = (1 − qi + qi z bi ).
i=1
The claim amount variable Xi for each policy is usually presented as

Xi = Ii Bi , i = 1, 2, · · · , n
where random variables I1 , I2 , · · · , In , B1 , B2 , · · · , Bn are independent. The
random variable Ii is an indicator random variable such that Pr(Ii = 1) = qi
and Pr(Ii = 0) = 1 − qi . Thus, Ii indicates whether the ith policy produced
a payment. The positive random variable Bi can have an arbitrary distri-
bution and represents the amount of the payment with respect to the ith
policy given that a payment was made.
To find the mean of S, we proceed as follows:
E(Xi ) =E[E(Xi |Ii )] = E[E(Ii Bi |Ii )] = E[Ii E(Bi |Ii )]
=E[Ii E(Bi )] = E(Ii )E(Bi ) = qi E(Bi ).
Hence, by the independence of the Xi0 s we have
n
X
E(S) = qi E(Bi ).
i=1
On the other hand,

E(Xi2 ) =E[E(Xi2 |Ii )] = E[E(Ii2 Bi2 |Ii )] = E[Ii2 E(Bi2 |Ii )]
=E[Ii2 E(Bi2 )] = E(Ii )E(Bi2 ) = qi E(Bi2 ).
Hence,
Var(Xi ) = qi E(Bi2 ) − qi2 E(Bi )2 = qi Var(Bi ) + qi (1 − qi )E(Bi )2
and
n
X
Var(S) = [qi Var(Bi ) + qi (1 − qi )E(Bi )2 ].
i=1
The moment generating function of Xi is
MXi (t) =E(etIi Bi ) = E[etIi Bi |Ii = 0]Pr(Ii = 0) + E[etIi Bi |Ii = 1]Pr(Ii = 1)
=1 − qi + qi MBi (t).
By independence, the moment generating function of S is
n
Y n
Y
MS (t) = MXi (t) = [1 − qi + qi MBi (t)].
i=1 i=1
Example 44.1
For a portfolio of 5 life insurance policies, you are given:
i qi bi
1 0.32 500
2 0.10 1000
3 0.54 250
4 0.23 375
5 0.14 650
Calculate E(S) and Var(S) where S = X1 + X2 + · · · + X5 .
Solution.
The aggregate mean is
E(S) = 0.23(500) + 0.10(1000) + 0.54(250) + 0.23(375) + 0.14(650) = 527.25.
The variance of S is
Var(S) =0.32(1 − .32)(500)2 + 0.10(1 − 0.10)(1000)2
+0.54(1 − 0.54)(250)2 + 0.23(1 − 0.23)(375)2 + 0.14(1 − 0.14)(650)2
=235698.6875
Example 44.2
For a portfolio of 3 life insurance policies, you are given:
i qi bi
1 0.32 500
2 0.10 1000
3 0.54 250
Calculate PS (z).
Solution.
We have
PS (z) = (0.68 + 0.32z 500 )(0.90 + 0.10z 1000 )(0.36 + 0.54z 250 )
Example 44.3
A life insurance portfolio consists of 100 policies where each policy has a
probability of a claim of 0.20. When a claim occurs, the amount of the
claim follows a Pareto distribution with parameters α = 3 and θ = 1000.
Caculate the mean and the variance of the aggregate loss.
Solution.
For i = 1, 2 · · · , 100, we have
θ 1000
E(Bi ) = = = 500.
α−1 3−1
Hence
E(S) = nqE(B) = 100(0.20)(500) = 10, 000.
Next,
αθ2
Var(B) = = 750, 000.
(α − 1)2 (α − 2)
Therefore,
Var(S) = 100[0.20(750, 000) + 0.20(0.80)(500)2 ] = 19, 000, 000
Example 44.4 ‡
Each life within a group medical expense policy has loss amounts which fol-
low a compound Poisson process with λ = 0.16. Given a loss, the probability
1
that it is for Disease 1 is 16 .
Loss amount distributions have the following parameters:
Standard
Mean per loss Deviation per loss
Disease 1 5 50
Other Diseases 10 20
Premiums for a group of 100 independent lives are set at a level such that
the probability (using the normal approximation to the distribution for ag-
gregate losses) that aggregate losses for the group will exceed aggregate
premiums for the group is 0.24.
A vaccine which will eliminate Disease 1 and costs 0.15 per person has been
discovered.
Define:
A= the aggregate premium assuming that no one obtains the vaccine, and
B= the aggregate premium assuming that everyone obtains the vaccine and
the cost of the vaccine is a covered loss.
Calculate A/B.
Solution.
Let X1 denote the loss per person due to Disease 1 and X2 the loss per
person due to Other Diseases. Then X1 and X2 are Bernoulli with parameter
1
q = 16 .
The expected loss per perosn is
1 15 155
E(L) = (5) + (10) = = 9.6875
16 16 16
and the variance is
Var(L) = E(Var(X)) + Var[E(X)].

1
Now, E(X) is a random variable with outcomes 5 (with probability 16 ) and
5
10 (with probability 16 ). Thus,

2 1 2 15
Var(E(X)) = 5 + 10 − 9.68752 = 1.4648.
16 16
Also,
2 1 2 15
E(Var(X)) = 50 + 20 = 531.25.
16 16
Hence,
Var(L) == 531.25 + 1.4648 = 532.7148.
Let S be the aggregate losses. Then S consists of 100 independent compound

distributions where each compound distribution has a primary distribution
N that is Poisson with parameter λ = 0.16 and secondary distribution L.
Hence,
E(S) = 100(0.16)(9.6875) = 155
and
Var(S) = 100[E(N )Var(L)+Var(N )E(L)2 ] = 100(0.16)(531.25+9.688752 ) = 10025.
Now, suppose that no one gets the vaccine. Then

S − 155 A − 155 A − 155
Pr(S > A) = Pr √ > √ = Pr Z > √ = 0.24.
10025 10025 10025
Thus,
A − 155
√ = 0.7 =⇒ A = 225.09.
10025
Next, if everyone gets the vaccine then the losses due to Disease 1 are elim-
inated. In this case,

15
E(S) = 100 0.16 (10) + 0.15 = 165
16
and
2 15 2 15
Var(S) = 100(0.16) 20 + 10 = 7500.
16 16
In this case,

S − 165 B − 165 B − 165
Pr(S > B) = Pr √ > √ = Pr Z > √ = 0.24.
7500 7500 7500
Thus,
B − 165
√ = 0.7 =⇒ B = 225.62.
7500
Finally,
A 225.09
= = 0.998
B 225.62
Practice Problems
Problem 44.1
claim follows an exponential distribution with mean θ = 1000.
Caculate the mean and the variance of the aggregate loss.
Problem 44.2
Find MS (t).
Problem 44.3
Find PS (z).
Problem 44.4 ‡
A group life insurance contract covering independent lives is rated in the
three age groupings as given in the table below.
Age Number in Probability of Mean of the exponential
group age group claim per life distribution of claim amounts
18-35 400 0.03 5
36-50 300 0.07 3
51-65 200 0.10 2
(a) Find the mean and the variance of the aggregate claim.
(b) Find the 95th percentile of S.
Problem 44.5 ‡
The probability model for the distribution of annual claims per member
in a health plan is shown below. Independence of costs and occurrences
among services and members is assumed. Suppose that the plan consists of
n members.
Probability Mean of Variance of

Service of claim claim dist. claim dist.
office visits 0.7 160 4,900
Surgery 0.2 600 20,000
Other Services 0.5 240 8,100
Find the mean and the variance of the aggregate claim S.

45 Approximating Probabilities in the Individual

Risk Model
In this section, we approximate probabilities of aggregate losses in the in-
dividual risk model using either the normal distribution or the lognormal
distribution.
For large n, the normal approximation uses the Central Limit Theorem
as follows:
!
S − E(S) s − E(S)
Pr(S ≤ s) =Pr p ≤p
Var(S) Var(S)
!
s − E(S)
≈Pr Z ≤ p
Var(S)
!
s − E(S)
=Φ p
Var(S)
where Φ is the cdf of the standard normal distribution.
For the lognormal approximation, we solve the system of two equations

µ + 0.5σ 2 = ln [E(S)] and 2µ + 2σ 2 = ln [Var(S) + E(S)2 ] for µ and σ 2 . The
probability of the aggregate losses is approximated as follows:

ln S − µ ln s − µ
Pr(S ≤ s) ≈Pr ≤
σ σ

ln s − µ
=Φ .
σ
Example 45.1 ‡
A group life insurance contract covering independent lives is rated in the
three age groupings as given in the table below. The insurer prices the
contract so that the probability that the total claims will exceed the premium
is 0.05.
Age Number in Probability of Mean of the exponential
group age group claim per life distribution of claim amounts
18-35 400 0.03 5
36-50 300 0.07 3
51-65 200 0.10 2
45 APPROXIMATING PROBABILITIES IN THE INDIVIDUAL RISK MODEL329
(a) Find the mean and the variance of the aggregate claim.
(b) Using the normal approximation, determine the premium that the in-
surer will charge.
Solution.
(a) The mean is given by
E(S) = 400(0.03)(5) + 300(0.07)(3) + 200(0.10)(2) = 163.
The variance is given by
Var(S) =400[(0.03)(5)2 + (0.03)(0.97)52 ]
+300[(0.07)(3)2 + 0.07(0.93)(3)2 ] + 200[0.10(2)2 + 0.10(0.9)(2)2 ]
=1107.77.
(b) We have
Pr(S > P ) =0.05
1 − Pr(S > P ) =0.95
Pr(S ≤ P ) =0.95

P − 163
Pr Z ≤ √ =0.95
1107.77
P − 163
√ =1.645.
1107.77
Solving the last equation, we find P = 217.75
Example 45.2
Repeat the previous example by replacing the normal approximation with
the lognormal approximation.
Solution.
Solving the system µ + 0.5σ 2 − 163 and 2µ + 2σ 2 = ln 163 and 2µ + 2σ 2 =
ln (1632 + 1107.77) we find µ = 5.073 and σ 2 = 0.041. Thus,
Pr(S > P ) =0.05
1 − Pr(S > P ) =0.95
Pr(S ≤ P ) =0.95

ln P − 5.073
Pr Z ≤ √ =0.95
0.041
ln P − 5.073
√ =1.645.
0.041
Solving the last equation, we find P = 222.76
Practice Problems
Problem 45.1 ‡
The probability model for the distribution of annual claims per member in
a health plan is shown below. Independence of costs and occurrences among
services and members is assumed.
Probability Mean of Variance of
Service of claim claim dist. claim dist.
office visits 0.7 160 4,900
Surgery 0.2 600 20,000
Other Services 0.5 240 8,100
Using the normal approximation, determine the minimum number of mem-

bers that a plan must have such that the probability that actual charges
will exceed 115% of the expected charges is less than 0.10.
Problem 45.2
An insurer has a portfolio consisting of 5 one-year life insurance policies
grouped as follows:
i qi bi
1 0.32 500
2 0.10 1000
3 0.54 250
4 0.23 375
5 0.14 650
The insurer sets up an initial capital of consisting of 150% of its expected

loss to cover its future obligations.
Use a normal approximation to find the probability that the company will
not meet its obligation next year.
Problem 45.3
Repeat the previous problem by replacing the normal approximation with
the lognormal approximation.
Problem 45.4 ‡
An insurance company is selling policies to individuals with independent
future lifetimes and identical mortality profiles. For each individual, the
probability of death by all causes is 0.10 and the probability of death due
45 APPROXIMATING PROBABILITIES IN THE INDIVIDUAL RISK MODEL331
to an accident is 0.01. Each insurance policy pays a benefit of 10 for an

accidental death and 1 for non-accidental death. The company wishes to
have at least 95% confidence that premiums with a relative security load-
ing of 0.20 are adequate to cover claims. (In other words, the premium is
1.20E(S).)
Using the normal approximation, determine the minimum number of policies

that must be sold.
Problem 45.5
Consider a portfolio of 100 independent life insurance policies. It is deter-
mined that the death benefit of each insured follows a Poisson distribution
with paramaeter λ. The probability of a death is 0.1. Using the normal
approximation, it has been estimated that the probability of the aggregate
loss exceeding 25 is 0.05.
Determine the value of λ.

Review of Mathematical
Statistics
In this chapter we review some concepts of mathematical statistics which is

required in the sequel. Mainly, our attention will be focused on estimation
and hypothesis testing.
333
334 REVIEW OF MATHEMATICAL STATISTICS
46 Properties of Point Estimators

In model estimation, we are interested in estimating quantities related to a
random variable X and determine properties of those estimates.
By an estimate we mean a single value obtained when an estimation pro-

cedure is applied to a set of values.
By an estimator or statistic we mean a formula or rule that produced

the estimate. An example of an estimator is the sample mean of a random
sample of values {x1 , x2 , · · · , xn } given by
x1 + x2 + · · · + xn
x= .
n
A point estimation is an estimation procedure that results of a single es-
timate which is served as the best estimate of an unknown parameter. The
point estimator of a parameter θ will be denoted by θ̂ and is itself a random
variable.
The first property of estimators is the property of unbiasedness. An es-

timator θ̂ is said to be unbiased if E(θ̂) = θ for all θ. That is, the mean
of the estimator is just the quantity being estimated. Any estimator that is
not unbiased is called biased. The bias associated to a parameter estima-
tor is biasθ̂ (θ) = E(θ̂) − θ. That is, the difference between this estimator’s
expected value and the true value of the parameter being estimated. Note
also that E(θ̂) − θ = E(θ̂ − θ) so that the bias is the expected value of the
error.
Example 46.1
A population consists of the values 1,3,5, and 9. We want to estimate
the mean of the population µ. A random sample of two values from this
population is taken without replacement, and the mean of the sample µ̂ is
used as an estimator of the population mean µ.
(a) Find the probability distribution of µ̂.
(b) Is µ̂ an unbiased estimator?
Solution.
(a) The various samples are {1, 3}, {1, 5}, {1, 9}, {3, 5}, {3, 9}, and {5, 9} each
occurring with probability of 16 . The following table provides the sample
mean.
46 PROPERTIES OF POINT ESTIMATORS 335
Sample {1,3} {1,5} {1,9} {3,5} {3,9} {5,9}

µ̂ 2 3 5 4 6 7
(b) We have
1+3+5+9
µ= = 4.5
4
and
2+3+5+4+6+7
E(µ̂|µ) = = 4.5 = µ.
6
Hence, the estimator is unbiased
Example 46.2
Let X1 , X2 , · · · , Xn be normally distributed random variables each with
mean µ and variance σ 2 . Show that the estimator of the variance
Pn (Xi −µ̂)2 X1 +X2 +···+Xn
σ̂ 2 = i=1 n where µ̂ = n
is biased.
Solution.
The estimator can be expressed as
n
2 1X 2
σ̂ = Xi − µ̂2 .
n
i=1
Now, we have the following calculation,

" n # n 2
2 1 X
2 2 1X 2 2 σ 2
E(σ̂ ) =E Xi − µ̂ = (σ + µ ) − +µ
n n n
i=1 i=1
σ2 (n − 1) 2
=σ 2 + µ2 − − µ2 = σ 6= σ 2 .
n n
Hence, σ̂ 2 is biased
Now when an estimator exhibits a small amount of bias that vanishes as

the sample size increases without bound then we say that the estimator is
asymptotically unbiased. Symbolically, we have
lim E(θ̂n ) = θ
n→∞
for all θ.
Example 46.3
Let X1 , X2 , · · · , Xn be uniform random variables on (0, θ). Show that the
estimator
θ̂n = max{X1 , X2 , · · · , Xn }
is asymptotically unbiased.
Solution.
The cdf of θ̂n is expressed as
y n
Fθ̂n (y) = Pr(X1 ≤ y, X2 ≤ y, · · · , Xn ≤ y) = .
θ
Hence, the pdf is
ny n−1
fθ̂n (y) = , 0<y<θ
θn
and the expected value of θ̂n is
Z θ
n
E(θ̂n ) = ny n θ−n dy = θ→θ
0 n+1
as n → ∞. Thus, the estimator θ̂n is asymptotically unbiased
The next property of estimators that we consider is the property of con-

sistency: An estimator is (weakly)consistent if the probability of the error
being greater than a small amount is zero as the size of the sample goes to
infinity. Symbolically,
lim Pr(|θ̂n − θ| > δ) = 0, ∀δ > 0.

n→∞
Theorem 46.1
If limn→∞ E(θ̂n ) = θ and limn→∞ Var(θ̂n ) = 0 then θ̂n is consistent.
Proof.
Let δ > 0. Then by Chebyshev’s inequality (see p.405 of [2]) we can write
Var(θ̂n )
0 ≤ Pr(|θ̂n − E(θ̂n | ≥ 2δ > δ) ≤ .
4δ 2
Letting n → ∞ and using the squeeze rule of limits the result follows
Example 46.4
Let X1 , X2 , · · · , Xn be uniform random variables on (0, θ). Show that the
estimator
θ̂n = max{X1 , X2 , · · · , Xn }
is weakly consistent.
Solution.
We have already shown that θ̂n is asymptotically unbiased. It remains to
show that its variance goes to 0 as n goes to infinity. We have
Z θ
n 2
E(θ̂n2 ) = ny n+1 θ−n dy = θ
0 n+2
n 2 n2
Var(θ̂n ) = θ − θ2
n+2 (n + 1)2
nθ2
= →0
(n + 2)(n + 1)2
as n → ∞. That, is θ̂ is consistent
The third property of estimators that we consider is the mean-squared er-

ror: The mean-squared error (MSE) is the second moment (about the
origin) of the error. That is,
MSEθ̂ (θ) = E[(θ̂ − θ)2 ].
That is, the mean-squared error is an average of the squares of the difference
between the actual observations and those estimated. The mean-squared
error is arguably the most important criterion used to evaluate the perfor-
mance of an estimator. The MSE incorporates both the variance of the
estimator and its bias. Indeed, we have
MSEθ̂ (θ) =E[(θ̂ − θ)2 ] = E(θ̂2 − 2θθ̂ + θ2 )

=E(θ̂2 − 2θE(θ̂) + θ2
=Var(θ̂) + E(θ̂)2 − 2θE(θ̂) + θ2
=Var(θ̂) + (E(θ̂) − θ)2 = Var(θ̂) + [biasθ̂ (θ)]2 .
Since the MSE decomposes into a sum of the bias and variance of the es-
timator, both quantities are important and need to be as small as possible
to achieve good estimation performance. It is common to trade-off some
increase in bias for a larger decrease in the variance and vice-versa.
Example 46.5
Let X1 , X2 , · · · , Xn be independent and identically distributed randm vari-
ables with common distribution a normal distribution with parameters (µ, σ 2 ).
Find MSEµ̂ (µ).
Solution.
We have
MSEµ̂ (µ) =Var(µ̂) + [biasµ̂ (µ)]2 = Var(µ̂)

n
σ2

X1 + X2 + · · · + Xn 1 X
=Var = 2 Var(Xi ) =
n n n
i=1
Given two unbiased estimators θ̂ and θ̂0 of θ, we say that θ̂ is more efficient
than θ̂0 if Var(θ̂) < Var(θ̂0 ). Note that for an unbiased estimator θ̂, we have
MSEθ̂ (θ) = Var(θ̂). If the estimator θ̂ satisfies the property Var(θ̂) < Var(θ̂0 )
where θ̂0 is another unbiased estimator of θ then we call θ̂ a uniformly
minimum variance unbiased estimator(UMVUE). That is, θ̂ is the
most efficient unbiased estimator of θ.
Example 46.6
Let X1 , X2 , · · · , Xn be independent uniform random variables with param-
eters (0, θ). Consider the two unbiased estimators
n+1
θ̂a = 2µ̂ and θ̂b = n max{X1 , X2 , · · · , Xn }.
Determine which one is more efficient than the other.
Solution.
We have
4 θ2 θ2
Var(θ̂a ) = 4Var(µ̂) = =
n 12 3n
and
2 2
n+1 n+1 n
Var(θ̂b ) = Var(θ̂n ) = θ2 .
n n (n + 1)2 (n + 2)
Thus,
n−1
Var(θ̂a ) − Var(θ̂b ) = > 0, n > 1.
3n(n + 2)
Thus, θ̂b is more efficient than θ̂a
Practice Problems
Problem 46.1 ‡
You ar given:
• E(X) = θ > 0.
θ2
• Var(X) = 25 .
kX
• θ̂ = k+1 .
• MSEθ̂ (θ) = 2[biasθ̂ (θ)]2 .
Determine the value of k.
Problem 46.2 ‡
You are given two independent estimates of an unknown quantity θ :
a. Estimator A : E(θ̂A ) = 1000 and σ(θ̂A ) = 400.
b. Estimator B : E(θ̂B ) = 1200 and σ(θ̂B ) = 200.
Estimator C is a weighted average of Estimator A and Estimator B such
that
θ̂C = wθ̂A + (1 − w)θ̂B .
Determine the value of w that minimizes σ(θ̂C ).
Problem 46.3 ‡
Claim sizes are uniformly distributed over the interval [0, θ]. A sample of 10
claims, denoted by X1 , , X2 , · · · , X10 was observed and an estimate of θ was
obtained using
θ̂ = max{X1 , X2 , · · · , X10 }.
Recall that the probability density function for θ̂ is
10y 9
fθ̂ (y) = .
θ10
Calculate the mean-squared error for θ̂ for θ = 100.
Problem 46.4 ‡
A random sample, X1 , X2 , · · · , Xn is drawn from a distribution with a mean
of 2/3 and a variance of 1/18. An estimator of the distribution mean is given
by
X1 + X2 + · · · + Xn
µ̂ = .
n−1
Find MSEµ̂ (µ).
Problem 46.5
A random sample of 10 independent values were taken from a Pareto dis-
tribution with parameters α = 3 and θ. The estimator used to estimate the
mean of the distribution is given by
X1 + X2 + · · · + X10
µ̂ = .
10
It is given that MSEµ̂ (µ) = 300. Determine the value of θ.
Problem 46.6
Let X1 , X2 , · · · , Xn be uniform random variables on (0, θ). The parameter
θ is estimated using the estimator
2n
θ̂n = µ
n−1
where
X1 + X2 + · · · + Xn
µ=
n
Find the bias of θ̂.
Problem 46.7
Let θ̂ denote an estimator of a parameter θ. Suppose that E(θ̂) = 3 and
E(θ̂2 ) = 17.
Calculate MSEθ̂ (5).
Problem 46.8 ‡
Which of the following statements is true?
(A) A uniformly minimum variance unbiased estimator is an estimator such

that no other estimator has a smaller variance.
(B) An estimator is consistent whenever the variance of the estimator ap-
proaches zero as the sample size increases to infinity.
(C) A consistent estimator is also unbiased.
(D) For an unbiased estimator, the mean squared error is always equal to
the variance.
(E) One computational advantage of using mean squared error is that it is
not a function of the true value of the parameter.
47 INTERVAL ESTIMATION 341
47 Interval Estimation
The next estimation procedure that we consider is the interval estimation.
Point estimation provides a single value as an estimate to the true value.

In contrast, interval estimation provides a range of values, any of which is
likely to be the true value.
A specific type of interval estimator is the confidence interval: A level

100(1 − α)%, 0 < α < 1, confidence interval for a population param-
eter θ is an interval [L, U ] computed from a sample data such that Pr(L ≤
θ ≤ U ) ≥ 1−α for all θ. The number 1−α is called the level of confidence.
This says that 100(1 − α)% of all random samples yield intervals containing
the true value of the parameter.
Example 47.1
Let X1 , X2 , · · · , Xn be a sample from a normal distribution with a known
standard deviation σ but unknown mean µ. Find 100(1 − α)% confidence
interval of µ.
Solution.
Let X = X1 +X2n+···+Xn by an estimator of the mean. Then the random vari-
X−µ
able σ/√ has the standard normal distribution. But the standard normal
n
distribution is symmetric about 0 so we choose L = −U so that

X −µ
Pr −U ≤ √ ≤ U = 1 − α.
σ/ n
We choose U as U = z α2 where z α2 is the 100 1 − α2 th percentile of the

standard normal distribution. That is,

Z zα
2 α
φ(x)dx = 1 −
−∞ 2
where φ(x) is the pdf of the standard normal distribution. Note that z α2 is
the area defined as the value of the normal random variable Z such that the
area to its right is α2 . Hence, a confidence interval is given by
σ σ
X − z α2 √ ≤ µ ≤ X + z α2 √
n n
Example 47.2
For a 95% confidence interval, what is z α2 ?
Solution.
We have 1 − α = 0.95 so that α = 0.05. Thus, z α2 satisfies
Z zα
2
Φ(z α2 ) = φ(x)dx = 1 − 0.025 = 0.975.
0
Using the table for the standard normal distribution, we find z0.025 = 1.96
In general, confidence intervals are constructed as follows: Suppose that

θ̂ is a point estimator of the population parameter θ such that E(θ̂) ≈ θ
and Var(θ̂) ≈ v(θ) with θ̂ being approximated by a normal distribution. At
a later section, we will show that for a confidence interval 100(1 − α)%, we
have !
θ̂ − θ
1 − α ≈ Pr −z α2 ≤ p ≤ z α2 .
v(θ)
Example 47.3
Construct a 95% confidence interval for the mean of a Poisson distribution.
Solution.
Let θ = λ, the mean of the Poisson distribution. Let the point estimator
be θ̂ = X. By the Central Limit Theorem, X is normally distributed with
E(θ̂) = E(X) = θ and Var(θ̂) = Var(X)
n = nθ . Thus, we want
 
X −θ
0.95 ≈ Pr −1.96 ≤ q ≤ 1.96 .
θ
n
Thus, we want

X − θ
q ≤ 1.96
θ

n
which is equivalent to
3.8416θ
(X − θ)2 ≤
n
or
3.8416 2
θ2 − 2X + θ + X ≤ 0.
n
Solving this quadratic inequality, we find the confidence interval with end-
points  s 
X ± 1.9208 − 1 15.3664X + 3.84162 /n 
n 2 n
47 INTERVAL ESTIMATION 343
Practice Problems
Problem 47.1
Problem 47.2
Problem 47.3
You are given that Pr(L ≤ θ ≤ U ) ≥ 0.80. What is the probability that a
confidence interval will not include the population parameter?
Problem 47.4
You are given:
• a population has a normal distribution with mean µ and σ = 3.
• X = 80
• n = 10
Construct a 95% confidence interval.
Problem 47.5 ‡
A sample of 2000 policies had 1600 with no claims and 400 with one or
more claims. Using the normal approximation, determine the symmetric
95% confidence interval for the probability that a single policy has one or
more claims.
48 Hypothesis Testing
Within the context of this book, a statistical hypothesis is a claim or a
statement regarding a population parameter or a probability distribution.
To test the validity of a hypothesis with absolute certainty requires the

examination of the entire population and this is hard to do in general. In-
stead, a hypothesis testing is used in collecting evidence from a random
sample and based on the evidence one can judge whether to support the
claim or not.
Hypothesis testing is formulated based on two hypotheses:
• The null hypothesis denoted by H0 is the claim that is assumed to

be true. The conclusion of any hypothesis testing is either to reject the
null hypothesis or fail to reject because one does not have enough evidence,
based on a sample, to reject it. Strangely, one will never accept the null
hypothesis. The null hypothesis typically includes a statement of equality
such as “=”, “≥”, or “≤ ”.
• The alternative hypothesis denoted by H1 (or Ha ) is the one to be

tested and is the claim that is oppose it to the null hypothesis. The alter-
native hypothesis typically includes a statement of strict inequality or not
equal:“ > ”, “ < ”, “ 6= ”.
Example 48.1
An insurance company is reviewing its current policy rates. Originally, the
insurance company believed that the average claim amount should be $1,200.
Currently, they are suspecting that the true mean is actually higher than
this. What are the hypothesis for this problem?
Solution.
The null and alternative hypothesis for this problem are:
H0 :µ ≤ 1, 200
H1 :µ > 1, 200
The testing method used in making the decision to either reject or not to
reject the null hypothesis involves two concepts: A test statistic and a re-
jection region.
48 HYPOTHESIS TESTING 345
A test statistic is a random variable represented by a numerical value

obtained from the random sample and this value is used (see below) to de-
termine whether to reject or fail to reject the null hypothesis. The random
samples are assumed to have a large number of data so that the data values
have a normal distribution thanks to the Central Limit Theorem. Thus, a
common test statistic is the Z score.
A rejection region is the set of all values of the test statistic for which the
null hypothesis will be rejected,i.e., values that provide strong evidence in
favor of the alternative hypothesis.
The boundaries of the rejection region are called the critical values. In
the case of an alternative hypothesis with > sign the rejection region lies in
the right-tail of the distribution of the test statistic, with < the tail is left,
and with 6= sign the region is two-tailed.
Because we are making a decision based on a finite sample, there is a pos-

sibility that we will make mistakes. The possible outcomes are listed in the
chart:
H0 is true H1 is true
Do not Correct Type II
reject H0 decision error
reject H0 Type I correct
error decision
It is possible that we reject a null hypothesis in a situation where it happens

to be true. In this case, we make an error which is referred to as Type I
error. The maximum probability of making a Type I error given that the
null hypothesis is true is called the level of confidence and is denoted by
α. The lower significance level α, the less likely we are to commit a type I
error. The level of confidence is usually set in advance by the experimenter
and is often between 1% and 10%. So for example, if α = 0.05, there is a
5% chance that, when the null hypothesis is true, we will erroneously reject
it.
Example 48.2
Find the rejection region for a level of significance α = 0.01 when the alter-
native hypothesis has the 6= sign.
Solution.
We need to put half of α in the left tail, and the other half of α in the right
tail. The rejection region is the two-tailed region of the Z−distribution:

|Z| > 2.58. The critical values are −2.58 and 2.58
Example 48.3
An insurance company is reviewing its current policy rates. When originally
setting the rates they believed that the average claim amount was $1,800.
They are concerned that the true mean is actually higher than this, because
they could potentially lose a lot of money. They randomly select 40 claims,
and calculate a sample mean of $1,950. Assuming that the standard de-
viation of claims is $500, and set α = 0.025, test to see if the insurance
company should be concerned.
Solution.
The null and alternative hypothesis for this problem are:
H0 :µ ≤ 1, 800
H1 :µ > 1, 800.
The test statistic of the given sample is

x−µ 1950 − 1800
z= √ = √ = 1.897.
σ/ n 500/ 40
Since
Pr(Z > 1.96|µ ≤ 1, 800) = 0.025
the rejection region is Z > 1.96. Since 1.897 < 1.96, we fail to reject the null
hypothesis. We cannot conclude anything statistically significant from this
test, and cannot tell the insurance company whether or not they should be
concerned about their current policies
Failure to reject H0 when H1 is true is called a Type II error. The prob-

ability of committing a type II error is denoted by β.
α and β are not independent of each other−as one increases, the other de-
creases. However, increases in the sample size cause both to decrease, since
sampling error is reduced.
Example 48.4
Consider the following hypotheses:
H0 :µ ≥ 3
H1 :µ < 3.
You throw a fair die. If the face shows a 3 or less then you reject H0 . If the
face shows a number greater than 3 you fail to reject H0 or H1 is true.
(a) Calculate the level of significance α.
(b) Calculate the probability of Type II error.
Solution.
(a) We have
α =Pr(H0 is true but is rejected)

3 1
=Pr(X ≤ 3) = = .
6 2
(b) We have
β =Pr(H1 is true but fail to rejected H0 )

3 1
=Pr(X > 3) = =
6 2
The rejection region is one way to test hypotheses where you compare the
test statistic to a critical value. Another method for testing hypothesis
which allows us to draw a conclusion without calculating critical values is
the p−value method: A p−value or probability value is the probability of
a more extreme departure from the null hypothesis than the observed data.
The p−value is the area under the normal curve.
By comparing the p−value to the alpha level we can easily decide to re-
ject or fail to reject: If p−value > α, fail to reject H0 . If p−value ≤ α, reject
H0 .
How do we calculate the p−value? One method is as follows:
1. Determine the value of the test statistic z corresponding to the result

of the sampling experiment.
2. (a) If the test is one-tailed, the p−value is equal to the tail area be-
yond z in the same direction as the alternative hypothesis. Thus, if the
alternative hypothesis is of the form >, the p−value is the area to the right
of, or above the observed z value. The same is true in the case of < .
(b) If the test is two-tailed, the p−value is equal to twice the tail area
beyond the observed z value in the direction of the sign of z. That is, if z is
positive,the p−value is twice the area to the right of, or above the observed
z value. The same holds true in the case where z is negative.
Example 48.5
Find the p−value of Example 48.3.
Solution.
The p−value is 2Pr(Z > 1.897). This is equivalent to twice the area under
the standard normal curve to the right of z = 1.897. Looking in a standard
normal table we find that z = 1.897 ≈ 1.90 corresponds to 0.0287. Since
2(0.0287) = 0.0574 > α = 0.05, we fail to reject the null hypothesis
Example 48.6
You are performing a hypothesis test as follows:
H0 :X follows an exponential distribution mean 0.1

H1 :X follows an exponential distribution mean 0.2
You pick a random value of X. This is x, your test statistic. Your test
statistic in this case is 2. What is the p−value of this test?
Solution.
Recall that the p−value is the probability that, if the null hypothesis is true,
a higher value than the test statistic is observed.The sdf of X if H0 is true
is S(x) = e−10x . Thus, the p−value is S(4) = e−10(2) = e−20
Practice Problems
Problem 48.1
Suppose the current mean cost to treat a cancer patient for one month is
$18,000. Consider the following scenarios.
(a) A hospital treatment plan is implemented which hospital authorities feel
will reduce the treatment costs.
(b) It is uncertain how a new treatment plan will affect costs.
(c) A treatment plan is implemented which hospital authorities feel will in-
crease treatment costs.
Let µ represent the mean cost per patient per month after the new treat-
ment plan is implemented.
Give the research hypothesis in symbolic form for each of the above cases
Problem 48.2
Classify each of the following as a lower-tailed, upper-tailed, or two-tailed
rejection region:
(i) H0 : µ = 12 and H1 : µ 6= 12.
(ii) H0 : µ ≥ 12 and H1 : µ < 12.
(iii) H0 : µ = 12 and H1 : µ > 12.
Problem 48.3
Consider a college campus where finding a parking space is not easy. The
university claims that the average time spent in finding a parking space is at
least 30 minutes. Suppose you suspect that it takes less than that. So in a
sample of five, you found that the average time is 20 minutes. Assuming that
the time it takes to find a parking spot is normal with standard deviation
σ = 6 minutes, then perform a hypothesis test with level of significance
α = 0.01 to see if your claim is correct.
Problem 48.4
A hypothesis test has a p−value of 0.037. At which of these significance
levels would you reject the null hypothesis?
(i) 0.025 (ii) 0.074 (iii) 0.05 (iv) 0.042.
Problem 48.5
A new restaurant has opened in town. A statistician claims that the amount
spent per customer for dinner is more than $20. To verify whether his claim
is valid or not, he randomly selected a group of 49 custimers and found that
the average amount spent was $22.60. Assume that the standard deviation
is known to be $2.50.
Using α = 2%, would he conclude the typical amount spent per customer is
more than $20.00?
Problem 48.6
Suppose a production line operates with a mean filling weight of 16 ounces
per container. Since over− or under−filling can be dangerous, a quality
control inspector samples 30 items to determine whether or not the filling
weight has to be adjusted. The sample revealed a mean of 16.32 ounces.
From past data, the standard deviation is known to be 0.8 ounces.
Using a 0.10 level of significance, can it be concluded that the process is

out of control (not equal to 16 ounces)?
Problem 48.7
A dietician is trying to test the claim that a new diet plan will cause a person
to lose 10 lbs. over 4 weeks. To test her claim, she selects a random sample
of 49 overweighted individulas and found that an average weight loss of 12.5
pounds over the four weeks, with σ = 7 lbs.
Identify the critical value suitable for conducting a two-tail test of the hy-
pothesis at the 2% level of significance.
Problem 48.8
A Type II error is committed when
(a) we reject a null hypothesis that is true.
(b) we don’t reject a null hypothesis that is true.
(c) we reject a null hypothesis that is false.
(d) we don’t reject a null hypothesis that is false.
The Empirical Distribution
for Complete Data
The focus of this chapter is on estimating distributional quantities in data-

dependent models. Recall from Section 17 the two types of data-dependent
distributions: The empirical distribution and the kernel-smoothed distribu-
tion. In what follows, by a complete data we mean a set of outcomes with
known exact values. When the exact value of an outcome in not known then
the outcome belongs to an incomplete data.
351
352 THE EMPIRICAL DISTRIBUTION FOR COMPLETE DATA
49 The Empirical Distribution for Individual Data

In Section 17 we defined the empirical model as follows: The empirical
distribution is obtained by assigning a probability of n1 to each data point
in a sample with n individual data points.
The empirical distribution function(edf) is

number of elements in the sample that are ≤ x
Fn (x) = .
n
Example 49.1
49, 50, 50, 50, 60, 75, 80, 120, 130.
Let X be the random variable representing the losses incurred by the pol-
icyholders. Find the empirical distribution probability function and the
empirical distribution function of X.
Solution.
The pmf is given by the table below.
x 49 50 60 75 80 120 130
1 1 1 1 1 1 1
p(x) 9 3 9 9 9 9 9
The edf is defined by

number of elements in the sample that are ≤x
FX (x) = 9 .
Thus, for example,
5
FX (73) =
9
In order to define empirical estimates, we need to introduce some additional
notation. Because of possible duplications of values, we re-define the sample
by considering the k distinct values arranged in increasing order as follows
y1 < y2 < y3 < · · · < yk , k ≤ n.
Let si denote
Pk the number of times the value yi appears inPthe sample.
k
Clearly, s
i=1 i = n. Next, for each 1 ≤ j ≤ k, let rj = i=j i . That
s
is, rj is the number of observations greater than or equal to yj . The set
of observations greater than or equal to yj is called the risk set11 . The
11
When listing the elements of this set, repeated observations must be listed
49 THE EMPIRICAL DISTRIBUTION FOR INDIVIDUAL DATA 353
convention is to call rj also as the risk set. Using the notation of rj , we can
express the edf as follows

 0, x < y1
r
Fn (x) = 1 − nj , yj−1 ≤ x < yj , j = 2, · · · , k
1, x ≥ yk .

Example 49.2
Determine the edf of Example 49.1 using the previous paragraph.
Solution.
We have the following chart:
j yj sj rj
1 49 1 9
2 50 3 8
3 60 1 5
4 75 1 4
5 80 1 3
6 120 1 2
7 130 1 1
Using the above chart, we find

 0, x < 49
 8 1


 1 − 9 = 9, 49 ≤ x < 50
5 4



 1− 9 = 9, 50 ≤ x < 60
1 − 49 = 5

9, 60 ≤ x < 75

Fn (x) =

 1 − 39 = 2
3, 75 ≤ x < 80
1 − 29 = 7
9, 80 ≤ x < 120




1 − 19 = 8
120 ≤ x < 130

9,



x ≥ 130

1,
Since the empirical model is a discrete model, the derivative required to
create the density and hazard rate functions cannot be taken. The best one
can do is to estimate the cumulative hazard rate function defined by:
H(x) = − ln S(x) = − ln [1 − F (x)].
Note that once an estimate of H(x) is found, we can find estimates for
F (x) = 1 − e−H(x) . An estimate of the cumulative hazard rate function is
the Nelson-Åalen estimate given by:

 P 0,
 x < y1
j−1 si
Ĥ(x) = i=1 ri , yj−1 ≤ x < yj , j = 2, 3, · · · , k
 Pk si ,

x ≥ yk .
i=1 ri
Example 49.3
Determine the Nelson-Åalen estimate for Example 49.1.
Solution.
We have 
 0, x < 49
 1


 9 , 49 ≤ x < 50
1 3 35



 9 + 8 = 72 , 49 ≤ x < 50
35 1 247

+ = , 60 ≤ x < 75

Ĥ(x) = 72 5 360
247 1 337

 360 + 4 = 360 , 75 ≤ x < 80
337 1 457
360 + 3 = 360 , 80 ≤ x < 120




457 1 637
≤ x < 130



 360 + 2 = 360 , 120
637 1 997
x ≥ 130

360 + 1 = 360
Example 49.4 ‡
A portfolio of policies has produced the following claims:
100 100 100 200 300 300 300 400 500 600
Determine the empirical estimate of H(300).
Solution.
We have
j yj sj rj
1 100 3 10
2 200 1 7
3 300 3 6
4 400 1 3
5 500 1 2
6 600 1 1
Using the above chart, we find

s3 3
Ŝ(300) = = .
n 10
Thus,
Ĥ(300) = − ln [Ŝ(300)] = − ln (0.3) = 1.204
49 THE EMPIRICAL DISTRIBUTION FOR INDIVIDUAL DATA 355
Practice Problems
Problem 49.1
Twelve policyholders were monitored from the starting date of the policy to
the time of first claim. The observed data are as follows:
Time of first claim 1 2 3 4 5 6 7
Number of claims 2 1 2 2 1 2 2
Calculate p12 (x) and F12 (x).
Problem 49.2
Find the empirical mean and the empirical variance.
Problem 49.3
(a) Find the cumulative hazard function from the Nelson-Åalen estimate.
(b) Find the survival distribution function from the Nelson-Åalen esti-
mate.
Problem 49.4
49, 50, 50, 50, 60, 75, 80, 120, 230.
cyholders. Find the empirical survival function.
Problem 49.5
49, 50, 50, 50, 60, 75, 80, 120, 130.
Let X be the random variable representing the losses incurred by the pol-
icyholders. For the observation 50, find the number of elements in the
associated risk set.
Problem 49.6 ‡
You are given a random sample of 10 claims consisting of two claims of
400, seven claims of 800, and one claim of 1600. Determine the empirical
skewness coefficient.
Problem 49.7 ‡
You are given the following about 100 insurance policies in a study of time
to policy surrender:
(i) The study was designed in such a way that for every policy that was
surrendered, a new policy was added, meaning that the risk set, rj , is always
equal to 100.
(ii) Policies are surrendered only at the end of a policy year.
(iii) The number of policies surrendered at the end of each policy year was
observed to be:
1 at the end of the 1st policy year

2 at the end of the 2nd policy year
3 at the end of the 3rd policy year
..
.
n at the end of the nth policy year.
(iv) The Nelson-Åalen empirical estimate of the cumulative distribution

function at time n is F̂ (n) = 0.542.
What is the value of n?

50 EMPIRICAL DISTRIBUTION OF GROUPED DATA 357
50 Empirical Distribution of Grouped Data

Empirical distribution of grouped data is not possible. However, an estimate
of the empirical distribution can be obtained. For that purpose, denote the
grouped data by the intervals
(c0 , c1 ], (c1 , c2 ], · · · , (ck−1 , ck ], (ck , ∞)
where ck is the maximum value of the range and c0 = 0.
Now, letPnj denote the number of observations in the interval (cj−1 , cj ].

Clearly, nj=1 nj = n, where n is the total number of observations in (c0 , ck ].
For such a grouped data, the empirical distribution is determined at the
boundary points via the formula
j
1X
Fn (cj ) = ni , j = 1, · · · , k
n
i=1
where Fn (c0 ) = 0. With this notation, the empirical distribution can be

obtained by connecting the group boundaries with straight lines via the
formula
cj − x x − cj−1
Fn (x) = Fn (cj−1 ) + Fn (cj ), cj−1 ≤ x ≤ cj .
cj − cj−1 cj − cj−1
The graph of the resulting distribution is called an ogive.
Example 50.1
Given the following grouped data.
Interval Number of observations

(0, 2] 25
(2, 10] 10
(10, 100] 10
(100, 1000] 5
Find the empirical distribution function of the ogive corresponding to this

data set.
Solution.
We have c0 = 0, c1 = 2, c2 = 10, c3 = 100, and c4 = 1000. We first evaluate
the empirical distribution at the boundary points:
F50 (0) =0
25
F50 (2) = = 0.5
50
35
F50 (10) = = 0.7
50
45
F50 (100) = = 0.9
50
F50 (1000) =1.
The empirical cdf is

x

 4, 0≤x≤2
9 x

+ 40 , 2 ≤ x ≤ 10


 20
61 x
F50 (x) = 90 + 450 , 10 ≤ x ≤ 100
8 x
9 + 9000 , 100 ≤ x ≤ 1000




undef ined, x > 1000

Note that Fn0 (x) exists for all x 6= cj , j = 0, 1, · · · , k. Therefore, the density
function can be obtained via the formula
Fn (cj ) − Fn (cj−1 ) nj
fn (x) = = , cj−1 ≤ x < cj , j = 1, 2, · · · , k.
cj − cj−1 n(cj − cj−1 )
The resulting graph is called a histogram.
Example 50.2
Find the density function in Example 50.1.
Solution.
The density function is
1

 4, 0≤x<2
1

40 , 2 ≤ x < 10



1
f50 (x) = 450 , 10 ≤ x < 100
1
9000 , 100 ≤ x < 1000




undef ined, x ≥ 1000

Example 50.3
Find E(X ∧ 250) in Example 50.1
Solution.
We have
Z 250
E(X ∧ 250) = xf50 (x)dx + 250[1 − F50 (250)]
0
Z 2 Z 10 Z 100
x x x
= dx + dx + dx
0 4 2 40 10 450
Z 250
x
+ dx + 250[1 − 0.9167]
100 9000
=0.5 + 1.2 + 11 + 2.9167 + 20.825
=36.4417
Example 50.4 ‡
You are given
Claim size Number of claims

(0, 25] 25
(25, 50] 28
(50, 100] 15
(100, 200] 6
Assume a uniform distribution of claim sizes within each interval.

Estimate E(X 2 ) − E[(X ∧ 150)2 ].
Solution.
We have
Z 200 Z 150 Z 200
2 2 2 2 2
E(X ) − E[(X ∧ 150) ] = x f74 (x)dx + x f74 (x)dx + 150 f74 (x)dx
0 0 150
Z 200
= (x2 − 1502 )f74 (x)dx
150
Z 200
2 2 6
= (x − 150 ) dx
150 7400
3 200
x
− 1502 x = 337.84

=
3 150
Practice Problems
Problem 50.1 ‡
You are given
(0, 25] 30
(25, 50] 32
(50, 100] 20
(100, 200] 8
Assume a uniform distribution of claim sizes within each interval.
(a) Estimate the mean of the claim size distribution.
(b) Estimate the second raw moment of the claim size distribution.
Problem 50.2 ‡
For 500 claims, you are given the following distribution:
(0, 500] 200
(500, 1000] 110
(1000, 2000] x
(2000, 5000] y
(5000, 10000] ?
(10000, 25000] ?
(25000, ∞) ?
You are also given the following values taken from the ogive: F500 (1500) =
0.689 and F500 (3500) = 0.839. Determine y.
Problem 50.3 ‡
A random sample of payments from a portfolio of policies resulted in the
following:
(0, 50] 36
(50, 150] x
(150, 250] y
(250, 500] 84
(500, 1000] 80
(1000, ∞) 0
Suppose that Fn (90) = 0.21 and Fn (210) = 0.51. Find the value of x.
Problem 50.4 ‡
You are given the following information regarding claim sizes for 100 claims:

(0, 1000] 16
(1000, 3000] 22
(3000, 5000] 25
(5000, 10000] 18
(10000, 25000] 10
(25000, 50000] 5
(50000, 100000] 3
(100000, ∞) 1
Use the ogive to estimate the probability that a randomly chosen claim is
between 2000 and 6000.
Problem 50.5 ‡
You are given:
(i)
Interval Number of observations

(0, 50] 30
(50, 100] 36
(100, 200] 18
(200, 400] 16
(ii) Claim sizes within each interval are uniformly distributed.

b3 −a3
(iii) The second moment of the uniform distribution on [a, b) is 3(b−a) .
2
Estimate E[(X ∧ 350) ], the second moment of the claim size distribution
subject to a limit of 350.
Estimation of Incomplete
Data
Complete data means that each outcome in a random sample is observed

or known. In contrast, in an incomplete data or modified data we may
only know that some observations were above a certain value or below a
certain value. Incomplete data are modified by either truncation or censor-
ing, concepts to be introduced in Section 51. Our objective is to be able to
estimate distributional quantities of modified data.
363
364 ESTIMATION OF INCOMPLETE DATA
51 The Risk Set of Incomplete Data
An important element of the estimation is the risk set which is the subject
of this section.
What do we mean by modified or incomplete data? An observation can

be modified by truncation and/or censoring. An observation is
• left truncated at d if it is not recorded whenever its value is less than or

equal to d but when it is above d , it is recorded at its observed value.
• right truncated at u, if when it is greater than or equal to u it is not

recorded, but when it is below u it is recorded at its observed value.
• left censored at d if it is recorded as d if its value is less than or equal

to d, and recorded at its observed value otherwise.
• right censored at u if it is recorded as u if its value is greater than

or equal to u, and recorded at its observed value otherwise.
The most common modified data are the left truncated and right censored
observations. Left truncation usually occurs when a policy has an ordinary
deductible d (see Section 31). Right censoring occurs with a policy limit
(see Section 34). In what follows we will just use the term truncated to refer
to left truncated observation and we use the term censored to mean right
censored.
As pointed out earlier, an important element of the estimation procedure is

the concept of risk set. Some notations are first in order. For an individual
data, let dj denote the truncation point of the j-th observation with dj = 0
in the absence of truncation. Let uj denote the censored point of the j-th
observation. Let xj denote the uncensored value (loss amount).12 Uncen-
sored observations can be repeated (in other words, multiple loss amounts
for a particular observation). Let y1 < y2 < · · · < yk be the k P
unique (unre-
peated) values of the xi s that appear in the sample. Let sj = i I(xi = yj )
be the number of times the uncensored observation yj appears in the sample.
12
Each observation will have an assigned value of d and either (but not both) a value
of x or u.
51 THE RISK SET OF INCOMPLETE DATA 365
Now, for each unique observation yj , the risk set is given by
X X X
rj = I(xi ≥ yj ) + I(ui ≥ yj ) − I(di ≥ yj )
i i i
X X X
= I(di < yj ) − I(xi < yj ) − I(ui < yj )
i i i
where I is the indicator function.
The risk set can be interpreted as follows:
• For survival/mortality data13 , the risk set is the number of people ob-
served alive at age yj .
• For loss amount data, the risk set is the number of policies with observed
loss amounts (either the actual amount or the maximum amount due to a
policy limit) larger than or equal to yj less those with deductibles greater
than or equal to yj .
From the above formulas of rj , the following recursive formula holds
X X X
rj = rj−1 + I(yj−1 ≤ di < yj ) − I(xi = yj−1 ) − I(yj−1 ≤ ui < yj )
i i i
and r0 = 0. That is,
rj = rj−1 − (# of x0 s equal to yj−1 )+ (# of observations left-truncated at

yj−1 ) − (# of observations right-censored at yj−1 ).
Example 51.1
You are given the following mortality table:
13
In a typical mortality study, the following notation is used for an individual i : di will
denote the time the individual joined the study; ui will denote the time of withdrawal
from the study; and xi will denote the time of death of the individual.
Life Time of Entry Time of exit Reason of exit

1 0 4 End of Study
2 0 0.5 Death
3 0 1 Surrunder
4 0 4 End of Study
5 1 4 End of Study
6 1.2 2 Death
7 1.5 2 Death
8 2 3 Surrunder
9 2.5 4 End of Study
10 3.1 3.2 Death
Complete the following table.
i di xi ui
1
2
3
4
5
6
7
8
9
10
Solution.
i di xi ui
1 0 − 4
2 0 0.5 −
3 0 − 1
4 0 − 4
5 1 − 4
6 1.2 2 −
7 1.5 2 −
8 2 − 3
9 2.5 − 4
10 3.1 3.2 −
Example 51.2
Create a table summarizing yj , sj , and rj of Example 51.1
Solution.
The table is given below.
j yj sj rj
1 0.5 1 4+6−6=4
2 2 2 3+5−3=5
3 3.2 1 1+4−0=5
Practice Problems
Problem 51.1
For a ground up loss of amount X, an insurer pays the loss in excess of a
deductible d, and with a policy limit of u. Which of the following statements
is true regarding censoring and truncation.
(A) Ground up losses below the deductible amount are right-truncated.

(B) Ground up losses below the deductible amount are left-truncated.
(C) Ground up losses above amount u are right-censored.
(D) Ground up losses above amount u are left-censored.
Problem 51.2
(A) Losses of an amount in excess of a policy limit are right-truncated

losses.
(B) Losses of an amount in excess of a policy limit are left-truncated losses.
(C) Losses below a policy deductible are right-truncated losses.
(D) Losses below a policy deductible are left-truncated losses.
(E) All of A, B, C and D are false.
Problem 51.3
You are given the following mortality table:
Life Time of Entry Time of exit Reason of exit
1 0 0.2 Lapse
2 0 0.3 Lapse
3 0 0.5 Lapse
4 0 0.5 Death
5 1 0.7 Lapse
6 1.2 1.0 Death
7 1.5 2.0 Lapse
8 2 2.5 Death
9 2.5 3.0 Lapse
10 3.1 3.5 Death
11 0 4.0 Expiry of Study

19 0.5 4.0 Expiry of Study
20 0.7 1.0 Death
21 1.0 3.0 Death
22 1.0 4.0 Expiry of Study
23 2.0 2.5 Death
24 2.0 2.5 Lapse
25 3.0 3.5 Death
Complete the table similar to Example 51.1.
Problem 51.4
You are given the following
i di xi ui
1 1 − 6
2 0 4 −
3 2 4 −
4 6 8 −
5 0 − 5
Create a table summarizing yj , sj , and rj .
Problem 51.5
You are given the following
j dj xj uj
1 0 0.9 −
2 0 − 1.2
3 0 1.5 −
4 0 − 1.5
5 0 − 1.6
6 0 1.7 −
7 0 − 1.7
8 1.3 2.1 −
9 1.5 2.1 −
10 1.6 − 2.3
Create a table summarizing yj , sj , and rj .
52 The Kaplan-Meier and Nelson-Åalen Estimators

In this section, we consider the question of estimating the survival function
of incomplete data. We look at two estimators, the Kaplan-Meier product-
limit estimator and the Nelson-Åalen estimator.
In a typical mortality study, the following notation is used for an individual

i : di will denote the time when an individual has joined the study; ui will
denote the time of withdrawal from the study; and xi will denote the time of
death of the individual. The values si and ri are as defined in the previous
section.
The Kaplan-Meier (product-limit) estimate for the survival function

is given by
0 ≤ t < y1

 1,
j−1



 Y si
1− , yj−1 ≤ t < yj , j = 2, 3, · · · , k


Sn (t) = Ŝ(t) = ri
i=1
 k
si

 Y
1− t ≥ yk .


 or 0,
 ri
i=1
Example 52.1
Find the Kaplan-Meier estimate of the survival function for the data in
Example 51.1
Solution.
We have 
 1, 0 ≤ t < 0.5
1− sr11 =0.75,

0.5 ≤ t < 2



S10 (t) = s2
0.75 1 − r2 = 0.45, 2 ≤ t < 3.2



0.45 1 − sr33 = 0.36, t ≥ 3.2


The Kaplan-Meier applied to complete data is the same as the empirical

survival function defined by
number of observation ≤ x
Sn (x) = 1 − .
n
Example 52.2
49, 49, 50, 50, 50, 60, 75, 80, 120, 130.
52 THE KAPLAN-MEIER AND NELSON-ÅALEN ESTIMATORS 371
cyholders. The observations are not truncated or censored. Use a Kaplan-
Meier product-limit estimator to approximate the survival function for this
data.
Solution.
We have
0 ≤ t < 49

 1,
 2


 1 − 10 = 0.8, 49 ≤ t < 50
3



 0.8 1 − 8 = 0.5, 50 ≤ t < 60
1

0.5 1 − 5 = 0.4, 60 ≤ t < 75

S9 (t) =

 0.4 1 − 14 = 0.3, 75 ≤ t < 80
0.3 1 − 13 = 0.2, 80 ≤ t < 120




1
 0.2 1 − 2 = 0.1, 120 ≤ t < 130



0.1 1 − 11 = 0, t ≥ 130.

Note that this is the same as the empirical survival function for this data
set
The Kaplan-Meier estimate can be used to evaluate probabilities. If we

let T denote the time until death or failure (or size of loss), then Pr(T >
y1 ) = 1 − sr11 or equivalently Pr(T ≤ y1 ) = sr11 . The probability of those who
were still alive at time y1 and who are at risk of death at time y2 and survive
is Pr(T > y2 |T > y1 ) = 1 − sr22 . Thus,

s1 s2
Sn (y2 ) = Pr(T > y1 )Pr(T > y2 |T > y1 ) = 1 − 1− .
r1 r2
Likewise,
Sn (y3 ) =Pr(T > y1 )Pr(T > y2 |T > y1 )Pr(T > y3 |T > y2 )

s1 s2 s3
= 1− 1− 1−
r1 r3 r3
and so on.
Example 52.3 ‡
You are studying the length of time attorneys are involved in settling bod-
ily injury lawsuits. T represents the number of months from the time an
attorney is assigned such a case to the time the case is settled. Nine cases
were observed during the study period, two of which were not settled at
the conclusion of the study. For those two cases, the time spent up to the
conclusion of the study, 4 months and 6 months, was recorded instead. The
observed values of T for the other seven cases are as follows:
1 3 3 5 8 8 9.
Estimate Pr(3 ≤ T ≤ 5) using the Product-Limit estimator.
Solution.
We have the following charts
j dj xj uj
1 0 1 −
2 0 3 − j yj sj rj
3 0 3 − 1 1 1 9
4 0 − 4 2 3 2 8
5 0 5 − 3 5 1 5
6 0 − 6 4 8 2 3
7 0 8 − 5 9 1 1
8 0 8 −
9 0 9 −
The Kaplan-Meier estimator of S(t) is given by

 1, 0≤t<1
1 8



 1− 9 = 9 ,2 1≤t<3
 8
 2
9 1 − 8 = 3 , 3 ≤t<5
Ŝ(t) = 2 1 8

 3 1 − 5 = 15 , 5 ≤t<8
8 2 8
1 − 3 = 45 , 8 ≤ t < 9


 158


1
45 1 − 1 = 0, t ≥ 9.
Now, we have
Pr(3 ≤ T ≤ 5) =Pr(T ≥ 3) − Pr(T > 5)
=Ŝ(3− ) − Ŝ(5)
8 8
= − = 0.356
9 15
Example 52.4 ‡
The claim payments on a sample of ten policies are:
2 3 3 5 5+ 6 7 7+ 9 10+
where the ”+” indicates that the loss exceeded the policy limit.
Using the Product-Limit estimator, calculate the probability that the loss
on a policy exceeds 8.
Solution.
We have the following charts
j dj xj uj
1 0 2 −
2 0 3 − j yj sj rj Ŝ(yj )
− 1
3 0 3 1 2 1 10 1 − 10 = 0.9
− 2
4 0 5 2 3 2 9 0.9(1 − 9 ) = 0.7
5 0 − 5 3 5 1 7 0.7(1 − 71 ) = 0.6
6 0 6 − 4 6 1 5 0.6(1 − 51 ) = 0.48
7 0 7 − 5 7 1 4 0.48(1 − 14 ) = 0.36
8 0 − 7 6 9 1 2 0.36(1 − 21 ) = 0.18
9 0 9 −
10 0 − 10
Since, y5 ≤ 8 < y6 , we have Ŝ(8) = Ŝ(7) = 0.36
An alternative estimator for the survival function is to estimate the cu-

mulative hazard rate function via the Nelson-Åalen given by

 P 0,
 t < y1
j−1 si
Ĥ(t) = i=1 ri , yj−1 ≤ t < yj , j = 2, 3, · · · , k
 Pk si ,

t ≥ yk
i=1 ri
and then set Ŝ(t) = e−Ĥ(t) .

Example 52.5
Determine the Nelson-Åalen estimate of the survival function for the data
in Example 51.1.
Solution.
We first use the Nelson-Åalen estimate of the cumulative hazard rate func-
tion. 
 0, 0 ≤ t < 0.5
1

= 0.25, 0.5 ≤ t < 2

Ĥ(t) = 4
2

 0.25 + 5 = 0.65, 2 ≤ t < 3.2
1
0.65 + 5 = 0.85, t ≥ 3.2.

An estimate of the survival function is


 1, 0 ≤ t < 0.5
 −0.25

e = 0.7788, 0.5 ≤ t < 2
Ŝ(t) =

 e−0.65 = 0.5220, 2 ≤ t < 3.2
 −0.85
e = 0.4274, t ≥ 3.2
Example 52.6
For a mortality study with right-censored data, you are given:
Time Number of deaths Number at risk
tj sj rj
3 1 50
5 3 49
6 5 k
10 7 21
You are also told that the Nelson-Åalen estimate of the survival function at
time 10 is 0.575. Determine k.
Solution.
From 0.575Ŝ(10) = e−Ĥ(10) we find Ĥ(10) = − ln 0.575 = 0.5534. Thus,
X si 1 3 5 7
0.5534 = = + + + .
ri 50 49 k 21
ti ≤10
Solving this equation, we find k = 36

Example 52.7 ‡
You are given:
(i) A mortality study covers n lives.
(ii) None were censored and no two deaths occurred at the same time.
(iii) tk = time of the k th death.
(iv) A Nelson-Åalen estimate of the cumulative hazard rate function is
39
Ĥ(t2 ) = 380 .
Determine the Kaplan-Meier product-limit estimate of the survival function
at time t9 .
Solution.
We have Ĥ(t1 ) = sr11 = n1 and Ĥ(t2 ) = s1
r1 + s2
r2 = 1
n + 1
n−1 . The assumption
39
Ĥ(t2 ) = 280 leads to the equation
39n2 − 799n + 380 = 0.
Solving this quadratic equation we find the roots n = 0.487(discard) and n =
20. Thus, the Kaplan-Meier product-limit estimate of the survival function
at time t9 is
9
Y si
Sn (t9 ) = 1−
ri
i=1
19 18 11 11
= · ··· = = 0.55
20 19 12 20
Example 52.8 ‡
You are given the following times of first claim for five randomly selected
auto insurance policies observed from time t = 0 :
1 2 3 4 5
You are later told that one of the five times given is actually the time of
policy lapse (i.e., terminated), but you are not told which one.
The smallest Product-Limit estimate of S(4), the probability that the first
claim occurs after time 4, would occur at the lapse time t0 . Find t0 .
Solution.
If the time of policy lapse is at t = 1, then the at risk group at death time
2 is 4, at death time 3 is 3 and at death time 4 is 2, so that
4
Y 0 1 1 1
Ŝ(4) = = 1− 1− 1− 1− = 0.250.
5 4 3 2
i=1
4
Y 1 0 1 1
Ŝ(4) = = 1− 1− 1− 1− = 0.267.
5 4 3 2
i=1
4
Y 1 1 0 1
Ŝ(4) = = 1− 1− 1− 1− = 0.300.
5 4 3 2
i=1
4
Y 1 1 1 0
Ŝ(4) = = 1− 1− 1− 1− = 0.400.
5 4 3 2
i=1
1 is 5, at death time 2 is 4, at death time 3 is 3, and at death time 4 is 2 so
that
4
Y 1 1 1 1
Ŝ(4) = = 1− 1− 1− 1− = 0.200.
5 4 3 2
i=1
Ŝ(4) is the smallest at time t = 5, which is the policy lapse time

Practice Problems
Problem 52.1 ‡

tj sj rj
5 2 15
7 1 12
10 1 10
12 2 6
Calculate the estimate of Ŝ(12) based on the Nelson-Åalen estimate of

Ĥ(12).
Problem 52.2 ‡
You are given:
(i) The following data set:
2500 2500 2500 3617 3662 4517 5000 5000 6010 6932 7500 7500
(ii) Ĥ1 (7000) is the Nelson-Åalen estimate of the cumulative hazard rate
function calculated under the assumption that all of the observations in (i)
are uncensored.
(iii) Ĥ2 (7000) is the Nelson-Åalen estimate of the cumulative hazard rate
function calculated under the assumption that all occurrences of the values
2500, 5000 and 7500 in (i) reflect right-censored observations and that the
remaining observed values are uncensored.
Calculate |Ĥ1 (7000) − Ĥ2 (7000)|.
Problem 52.3 ‡
For a mortality study of insurance applicants in two countries, you are given:
(i)
Country A Country B
yj sj rj sj rj
1 20 200 15 100
2 54 180 20 85
3 14 126 20 65
4 22 112 10 45
(ii) rj is the number at risk over the period (yj−1 , yj ). Deaths during the
period (yj−1 , yj ) are assumed to occur at yj .
(iii) S T (t) is the Product-Limit estimate of S(t) based on the data for all
study participants.
(iv) S B (t) is the Product-Limit estimate of S(t) based on the data for coun-
try B.
Calculate |S T (4) − S B (4)|.
Problem 52.4 ‡
For observation j of a survival study:
• dj is the left truncation point
• xj is the observed value if not right-censored
• uj is the observed value if right-censored.
You are given:
j dj xj uj
1 0 0.9 −
2 0 − 1.2
3 0 1.5 −
4 0 − 1.5
5 0 − 1.6
6 0 1.7 −
7 0 − 1.7
8 1.3 2.1 −
9 1.5 2.1 −
10 1.6 − 2.3
Determine the Kaplan-Meier Product-Limit estimate of S10 (1.6).
Problem 52.5 ‡
You are given:
(i) All members of a mortality study are observed from birth. Some leave
the study by means other than death.
(ii) s3 = 1, s4 = 3.
(iii) The following Kaplan-Meier product-limit estimates were obtained:
Sn (y3 ) = 0.65, Sn (y4 ) = 0.50, Sn (y5 ) = 0.25.
(iv) Between times y4 and y5 , six observations were censored.
(v) Assume no observations were censored at the times of deaths.
Determine s5 .
Problem 52.6 ‡
In a study of claim payment times, you are given:
(i) The data were not truncated or censored.
(ii) At most one claim was paid at any one time.
(iii) The Nelson-Åalen estimate of the cumulative hazard function, H(t),
immediately following the second paid claim, was 23/132.
Determine the Nelson-Åalen estimate of the cumulative hazard function,

H(t), immediately following the fourth paid claim.
Problem 52.7 ‡
You are given:
(i) The following is a sample of 15 losses: 11, 22, 22, 22, 36, 51, 69, 69, 69,
92, 92, 120, 161, 161, 230.
(ii) Ĥ1 (x) is the Nelson-Åalen empirical estimate of the cumulative hazard
rate function.
(iii) Ĥ2 (x) is the maximum likelihood estimate of the cumulative hazard
rate function under the assumption that the sample is drawn from an expo-
nential distribution.
Calculate |Ĥ2 (75) − Ĥ1 (75)|.
Problem 52.8 ‡
For the data set
200 300 100 400 X
you are given:
(i) k = 4
(ii) s2 = 1
(iii) r4 = 1
(iv) The NelsonÅalen Estimate Ĥ(410) > 2.15.
Determine X.
53 MEAN AND VARIANCE OF EMPIRICAL ESTIMATORS WITH COMPLETE DATA379
53 Mean and Variance of Empirical Estimators with

Complete Data
In this section, we calculate the variance of empirical survival function and
empirical probability estimate for complete data.
Individual Data
We first consider finding the variance of the empirical survival function of
individual data. Suppose that the sample is of size n. Let Sn (t) be the
empirical estimate of the survival function S(x) defined by
number of observation greater than x Y

Sn (x) = =
n n
where Y is the number of observations in the sample that are greater than
x. If we regard an observation as a trial then we define a success to be the
observation that is greater than x and which occurs with probability S(x).
Then Y is a binomial random variable with parameters n and S(x) and with
mean E(Y ) = nS(x) and variance Var(Y ) = nS(x)(1 − S(x)). Thus,
E(Y ) nS(x)
E[Sn (x)] = = = S(x).
n n
This shows that the empirical estimate Sn (x) is unbiased (see Section 46).
In the same token, we have
1 S(x)(1 − S(x))
Var[Sn (x)] = Var(Y ) = →0
n2 n
as n → ∞. This shows that Sn (t) is consistent.
If S(x) is unknown, then we can estimate the variance of Sn (x) using Sn (x)
itself in the formula
d n (x)] = Sn (x)(1 − Sn (x)) .

Var[S
n
Example 53.1
Let X be a discrete random variable and p = Pr(a < X ≤ b). An estimate
of p is p̂ = Sn (a) − Sn (b). Show that p̂ is unbiased and consistent.
Solution.
We first note that p̂ = Yn where Y is a binomial random variable with
parameters n and S(a) − S(b). Thus,
1
E(p̂) = [n(S(a) − S(b))] = S(a) − S(b) = p.
n
This shows that p̂ is unbiased. Next, we show that p̂ is consistent. Indeed,
we have
1 p(1 − p)
Var(p̂) = 2
[n(S(a) − S(b))(1 − S(a) + S(b))] = →0
n n
as n → ∞. This shows that p̂ is consistent. When p is unknown, we have
p̂(1 − p̂)
Var(p̂)
d =
n
Next, we consider survival probability estimators. Recall (see [3]) that the
probability of a life aged x to attain age x + t is the conditional probability
t px = Pr(X > x + t|X > x)
where X is the age-at-death random variable.
Let nx and ny be the observed number of survivors past time x and y

respectively. The the probability of a person aged x to reach age y is
y p0
y−x px =
x p0
where S(x) = x p0 . Thus, an estimator of y−x px is
Sn (y) ny
y−x p̂x = = .
Sn (x) nx
The variance is given by
d y−x q̂x |nx ) = (nx − ny )ny .

d y−x p̂x |nx ) = Var(
Var(
n3x
Example 53.2
The following chart provides the time of death of 15 individuals under ob-
servation from time 0.
Time of Death 1 2 3 4 5
# of Deaths 1 3 2 4 5
(a) Estimate S(3) and estimate the variance of the estimator.

(b) Estimate q2 , the conditional probability that death occurs not later than
3 given survival to time 2. Estimate the variance of this estimator.
Solution.
9
(a) We have S15 (3) = 15 = 0.6 and
d 15 (3) = 0.6(0.4) = 0.016.

Var(S
15
(b) We have
S15 (3) 9 2
q̂2 = 1 − =1− =
S15 (2) 11 11
and
d 2 |n2 ) = (n2 − n3 )n3 = 0.0135
Var(q̂
n32
Example 53.3
The following random sample of 9 losses has been observed from the distri-
bution of loss random variable X :
49, 50, 50, 50, 60, 75, 80, 120, 130.
(a) Find the estimated variance of the estimate of Pr(X > 60).
(b) Find the estimated variance of the estimate of Pr(75 < X ≤ 120).
(c) Find the estimated variance of the estimate of Pr(X > 60|X > 50).
Solution. 4 4
d 1 ) = 9 (1− 9 ) = 20 .
(a) Let p1 = Pr(X > 60). Then p̂1 = 94 and Var(p̂ 9 729
(b) Let p2 = Pr(75 < X ≤ 120). Then p̂2 = S9 (75) − S9 (120) = 93 − 19 = 29
2 2
d 2 ) = 9 (1− 9 ) = 14 .
and Var(p̂ 9 729
(c) Let p3 = Pr(X > 60|X > 50) = 10 p50 . Then p̂3 = SS99 (60) 4
(50) = 5 and
d 3) = 4
Var(p̂ 125
Grouped Data
We first use the ogive to find the variance of the estimator of the survival
function and the density function. Let n observations be spread over the
grouped data
(c0 , c] , (c1 , c2 ], · · · , (ck−1 , ck ]
Let nj be the number of observations in (cj−1 , cj ]. We define the survival

estimate at cj by
n1 + n2 + · · · + nj
Sn (cj ) = 1 − .
n
If x ∈ (cj−1 , cj ] then Sn (x) can be found via the interpolation equation
Sn (x) = Sn (cj−1 ) + Sn (cj ).
Let Y be the number of observations up to cj−1 . That is, Y = n1 + n2 + · · · +

nj−1 . Let Z be the number of observations in (cj−1 , cj ]. That is, Z = nj .
Then the previous equation can be expressed as
Y (cj − cj−1 ) + Z(x − cj−1 )

Sn (x) = 1 − .
n(cj − cj−1 )
Note that Y can be regarded as a binomial random variable with parameters

(n, 1 − S(cj−1 )) whereas Z can be regarded as a binomial random variable
with parameters (n, S(cj−1 ) − S(cj )). Thus,
n[1 − S(cj−1 )](cj − cj−1 ) + n[S(cj−1 ) − S(cj )](x − cj−1 )

E[Sn (x)] =1 −
n(cj − cj−1 )
= S(cj−1 ) + S(cj ).
The variance of Sn (x) is
(cj − cj−1 )2 Var(Y ) + (x − cj−1 )2 Var(Z)

Var[Sn (x)] =
[n(cj − cj−1 )]2
2(cj − cj−1 )(x − cj−1 )Cov(Y, Z)
+
[n(cj − cj−1 )]2
where
Var(Y ) =nS(cj−1 )[1 − S(cj−1 )]

Var(Z) =n[S(cj−1 ) − S(cj )][1 − S(cj−1 ) + S(cj )]
Cov(Y, Z) = − n[1 − S(cj−1 )][S(cj−1 ) − S(cj )].
Example 53.4
Using a histogram (see Section 50), the empirical density function can be
expressed as
Z
fn (x) =
n(cj − cj−1 )
where Z = nj . Find the mean of fn (x).
Solution.
The random variable Z is a binomial random variable with parameters
(n, S(cj−1 ) − S(cj )). Thus,
E(Z) Sn (cj−1 ) − Sn (cj )

E[fn (x)] = =
n(cj − cj−1 ) cj − cj−1
Example 53.5
Find Var[fn (x)], where fn (x) is the empirical density function in the previous
example.
Solution.
We have
Var(Z) [Sn (cj−1 ) − Sn (cj ][1 − Sn (cj−1 ) + Sn (cj ]

Var[fn (x)] = 2
=
[n(cj − cj−1 )] n(cj − cj−1 )2
Example 53.6
Loss Range Number of Losses

(0, 2] 25
(2, 10] 10
(10, 100] 10
(100, 1000] 5
Estimate the probability that a loss will be no more than 90, and find the
estimated variance of that estimate.
Solution.
An estimate to the probability that a loss will be no more than 90 is
1 − S50 (90) =1 − Sn (cj−1 ) − Sn (cj )
100 − 90 15 90 − 10 5
=1 − −
100 − 10 50 100 − 10 50
1 3 8 1
=1 − −
9 10 9 10
=0.8778.
d ) =nSn (cj−1 )[1 − Sn (cj−1 )]
Var(Y
15 15
=50 [1 − ] = 10.5
50 50
Var(Z) =n[Sn (cj−1 ) − Sn (cj )][1 − Sn (cj−1 ) + Sn (cj )]
d
10 10
=50[ ][1 − ] = 8
50 50
Cov(Y, Z) = − n[1 − Sn (cj−1 )][Sn (cj−1 ) − Sn (cj )]
d
15 10
= − 50[1 − ] = −7
50 50
d − S50 (90)] =Var[S
Var[1 d 50 (90)]
(cj − cj−1 )2 Var(Y
d ) + (x − cj−1 )2 Var(Z)
d
=
[n(cj − cj−1 )]2
2(cj − cj−1 )(x − cj−1 )Cov(Y,
d Z)
+ 2
[n(cj − cj−1 )]
=0.00175
Practice Problems
Problem 53.1
Let X be a discrete random variable. In a sample of n outcomes, let Nj
denote the number of times the value xj was observed in the sample with
corresponding probability p(xj ). Nj can be regarded as a binomial random
variable with parameters (n, p(xj )).
Show that the estimator

Nj
pn (xj ) =
n
is unbiased and consistent.
Problem 53.2

(0, 2] 25
(2, 10] 10
(10, 100] 10
(100, 1000] 5
Estimate the density function of the loss random variable at x = 90, and
find the estimated variance of the estimator.
Problem 53.3
Consider the following data
Number of accidents Number of drivers

0 155
1 109
2 64
3 50
4 or more 8
Estimate the probability that a driver will have two accidents and find the
estimate of the variance of that estimator.
Problem 53.4
Estimated variances can be used to create confidence intervals for the true
probability. The formula for the endpoints of a 1 − α confidence interval is

as follows (see Section 47):
r
pn (1 − pn )
pn ± z α2
n
where Φ(z α2 ) = 1 − α2 .
Construct approximate 95% confidence inervala for p(2) of the previous ex-
ercise.
54 GREENWOOD ESTIMATE FOR THE VARIANCE OF THE KAPLAN-MEIER ESTIMATOR387
54 Greenwood Estimate for the Variance of the

Kaplan-Meier Estimator
In this section we look at estimating the variance to the Kaplan-Meier limit-
product estimator.
First, we recall the Kaplan-Meier limit-product (see Section 50) in mor-

tality context: Let 0 < y1 < y2 < · · · < yk be the unique observed death
times in the sample. The risk size rj is the number of individuals in the
sample that were alive (or at “risk”) just before time yj . Equivalently, rj , is
the number of individuals who are either alive and observed at time yj or
else who died at time yj . The number of individuals who died at time yj is
denoted by sj . The Kaplan-Meier estimate of the survival function is given
by
0 ≤ t < y1

 1,
j−1



 Y si

1− , yj−1 ≤ t < yj , j = 2, 3, · · · , k


Sn (t) = Ŝ(t) = ri
i=1
 k
si

 Y
1− t ≥ yk .


 or 0,
 ri
i=1
In what follows, y0 < y1 will denote the smallest alive or observed age in the
sample. Assume that the rj s and yj s are fixed. The number of individuals
who died at time yj is the only random quantity, which we denote by Sj .
As a random variable, Sj has a binomial distribution based on a sample of
rj and a probability of success
Pr(yj−1 < T ≤ yj ) S(yj−1 ) − S(yj )

Pr(T ≤ yj |T > yj−1 ) = =
Pr(T > yj−1 ) S(yj−1 )
where T is the time until death. For 1 ≤ j ≤ k, we have

Sj [S(yj−1 ) − S(yj )] S(yj )
E 1− =1− = (54.1)
rj S(yj−1 ) S(yj−1 )
and
Sj 1 [S(yj−1 ) − S(yj )]S(yj )
Var 1 − = 2 Var(Sj ) = .
rj rj rj S(yj−1 )2
Example 54.1
Assume that the Sj s are independent. Show that
S(yj )
E[Sn (yj )] = (54.2)
S(y0 )
Solution.
We have
" j # Yj
Y Si Si
E[Sn (yj )] =E 1− = E 1−
ri ri
i=1 i=1
j
Y S(yi ) S(yj )
= = .
S(yi−1 ) S(y0 )
i=1
It is important to keep in mind that this estimate of Sn (yj ) is conditioned

on being alive at time or age y0 .
Using Problems 54.1 and 54.2, we have

" j #
Y Sj
Var[Sn (yj )] =Var 1−
rj
i=1
j
S(yi )2 S(yj )2

Y [S(yi−1 ) − S(yi )]S(yi )
= + −
S(yi−1 )2 ri S(yi−1 )2 S(y0 )2
i=1
j
" #
Y S(yi )2 2 S(yj )2
= ri S(yi−1 )2
ri S(y i−1 ) −
[S(yi−1 ) − S(yi )]S(yi ) S(y0 )2
i=1 +
j
S(yi )2 ri S(yi ) + [S(yi−1 ) − S(yi )] S(yj )2
Y
= −
S(yi−1 )2 ri S(yi ) S(y0 )2
i=1
j j
S(yi )2 Y ri S(yi ) + [S(yi−1 ) − S(yi )] S(yj )2
Y
= −
S(yi−1 )2 ri S(yi ) S(y0 )2
i=1 i=1
( j )
S(yj )2 Y

S(yi−1 ) − S(yi )
= 1+ −1 .
S(y0 )2 ri S(yi )
i=1
If we assume ri S(yi ) to be large for i = 1, 2, · · · , j we can estimate the

product as
j j
Y S(yi−1 ) − S(yi ) X S(yi−1 ) − S(yi )
1+ =1+
ri S(yi ) ri S(yi )
i=1 i=1
and thus we obtain

j
2 X
S(yj ) S(yi−1 ) − S(yi )
Var[Sn (yj )] ≈ . (54.3)
S(y0 ) ri S(yi )
i=1
Because the survival function is usually unknown, we use (54.1) and (54.2)
to write
S(yj ) 2
h i
S(yi )
S(y0 ) ≈ [Sn (yj )]2 and S(yi−1 )
≈ 1 − srii
Using these estimates in (54.3) we obtain

j
2
X si
Var[Sn (yj )] ≈ [Sn (yj )] . (54.4)
ri (ri − si )
i=1
Equation (54.4) is known as Greenwood’s approximation.
Remark 54.1
For non-death ages, the convention is to take the sum up to the last death
age that is less than or equal to to the age under consideration.
Example 54.2 ‡
For a survival study with censored and truncated data, you are given:
Number at risk Failures at

Time (t) at time t time t
1 30 5
2 27 9
3 32 6
4 25 5
5 20 4
The probability of failing at or before Time 4, given survival past Time 1, is

3 q1 . Calculate Greenwood’s approximation of the variance of the estimator
3 q̂1 .
Solution.
Let 3 p1 be the probability that an individual which has survived past time
1 will also survive past time 4. Then 3 p1 = 1 − 3 q1 which implies
Var(
d 3 q̂1 ) = Var(
d 3 p̂1 ).
Now, by (54.3), we have

S(4)
3 p̂1 ≈ ≈ S5 (4).
S(1)
Since we are concerned with the individuals who survived time 1, then y0 = 1
so that the next death time is 2. Hence, the calculation in Greenwood’s
approximation must start from time 2 and not time 1. Having this in mind,
we have
9 6 5 13
3 p̂1 = 1 − 1− 1− =
27 32 25 30
and
2
13 9 6 5
Var(
d 3 q̂1 ) = + + = 0.0067
30 27(27 − 9) 32(32 − 6) 25(25 − 5)
Example 54.3 ‡
The following is a sample of 10 payments:
4 4 5+ 5+ 5+ 8 10+ 10+ 12 15
where + indicates that a loss exceeded the policy limit. Determine Green-
wood’s approximation to the variance of the product-limit estimate S10 (11).
Solution.
We first create the following table summarizing yj , rj , and sj .
yj rj sj
4 10 2
8 5 1
12 2 1
15 1 1
Since 11 is an uncensored value (loss), the largest loss less than or equal to
11 is 8. Thus,
Var(S
d 10 (11)) = Var(Sd 10 (8)).
The Kaplan-Meier estimate is

2 1 16
S10 (8) = 1 − 1− = .
10 5 25
Hence
2
16 2 1
Var(S
d 10 (11)) = + = 0.03072
25 10(10 − 2) 5(5 − 1)
Practice Problems
Problem 54.1
Let X1 , X2 , · · · , Xn be indepedent random variable. Show that
n
Y n
Y
Var(X1 X2 · · · Xn ) = (µ2i + σi2 ) − µ2i .
i=1 i=1
Problem 54.2
Show that if 0 < ai << 1 for i = 1, 2, · · · , n then
(1 + a1 )(1 + a2 ) · · · (1 + an ) ≈ 1 + a1 + a2 + · · · + an .
Problem 54.3

tj sj rj
5 2 15
7 1 12
10 1 10
12 2 6
Find the approximate variance of the Product-Limit estimate of Ŝ(10).
Problem 54.4

tj sj rj
5 2 15
7 1 12
10 1 10
12 2 6
Find the approximate variance of the Product-Limit estimate of 7 p̂5 .
Problem 54.5 ‡
For 200 auto accident claims you are given:
(i) Claims are submitted t months after the accident occurs, where t =
0, 1, 2, · · · .
(ii) There are no censored observations.
(iii) Ŝ(t) is calculated using the Kaplan-Meier product limit estimator.

(iv) c2S (t) = Var(Ŝ(t))
d
, where Var(
d Ŝ(t)) is calculated using Greenwood’s ap-
Ŝ(t)2
proximation.
(v) Ŝ(8) = 0.22, Ŝ(9) = 0.16, c2S (9) = 0.02625, c2S (10) = 0.04045.
Determine the number of claims that were submitted to the company 10

months after an accident occurred.
55 VARIANCE ESTIMATE OF THE NELSON-ÅALEN ESTIMATOR AND CONFIDENCE INTERVALS39
55 Variance Estimate of the Nelson-Åalen Estima-

tor and Confidence Intervals
We start this section by deriving an estimate formula for the variance of the
Nelson-Åalen estimator. We assume ri to be fixed so that the random quan-
tity is si . We also assume that Si has a Poisson distribution with parameter
si . In this case, we can write
j j j
!
X S i
X Var(S
d i) X si
Var(
d Ĥ(yj )) = Var
d = 2 = 2 (55.1)
ri ri ri
i=1 i=1 i=1
where we assume that the Sj s are independent.
Example 55.1
For a survival study with censored and truncated data, you are given:
Number at risk Failures at
Time (t) at time t time t
1 30 5
2 27 9
3 32 6
4 25 5
5 20 4
Estimate the variance of Ĥ(4).
Solution.
The estimated variance is given by
4
X si 5 9 6 5
Var(
d Ĥ(y4 )) =
2 = 2 + 2 + 2 + 2 = 0.0318
r 30 27 32 25
i=1 i
We next look at computing the confidence intervals (see Section 47) of both
the Kaplan-Meier estimator and the Nelson-Åalen estimator. But first we
consider the following example.
Example 55.2
You are given:
• The Kaplan-Meier estimator: Ŝ(3) = 0.8667.
• The Greenwwod approximation: Var(d Ŝ(3)) = 0.0077.
• Ŝ(t) is approximated by a normal distribution.
Find the 95% confidence interval of Ŝ(3).
Solution.
Using normal approximation, the 95% confidence interval can be expressed
as √
0.8667 ± 1.96 0.0077
or in interval notation as
(0.6947, 1.0387).
Confidence intervals of this type are referred to as linear confidence in-
tervals
A disadvantage of the above estimate is that the computed confidence in-

terval may fall outside the range (0, 1) which is not desirable for confidence
intervals involving probabilities and survival functions. One way to remedy
this drawback is to use the so-called log-transformed confidence inter-
vals which we state without justification: A 100(1 − α)% confidence interval
of Sn (t) is given by
1
(Sn (t) U , Sn (t)U )
where  q 
Var(
d Ŝn (t))
U = exp z α2 .
Sn (t) ln Sn (t)
Log-transformed confidence intervals never result in probabilities outside

the range from 0 to 1, which is extremely desirable for confidence intervals
involving probabilities and survival functions.
Example 55.3
Obtain the log-transformed confidence interval to Ŝ(3) as in Example 55.2.
Solution.
We have " √ #
0.0077
U = exp 1.96 ≈ 0.2498.
0.8667 ln 0.8667
Thus, the confidence interval is
1
(0.8667 0.2498 , 0.86670.2498 ) = (0.564, 0.9649)
Similar results are available for the Nelson-Åalen estimators. We define the
linear (1−α) confidence interval for the cumulative hazard rate function
by q
Ĥ(t) ± z α Var(
2
d Ĥ(yj )), yj ≤ t < yj+1 .
The corresponding log-transformed confidence interval is given by

!
Ĥ(t)
, Ĥ(t)U
U
where q 
Var(
d Ĥ(yj ))
U = exp z α2 .
Ĥ(t)
Example 55.4
You are given:
yj rj sj
1 50 4
2 53 5
3 32 9
4 45 11
5 20 2
(i) Find Ĥ(5) and Var(

d Ĥ(5)).
(ii) Obtain the 95% linear confidence interval to H(5).
(iii) Obtain the 95% log-transformed confidence interval to H(5).
Solution.
(i) We have
5
X sj
Ĥ(5) = Ĥ(y5 ) = = 0.8
rj
j=1
and
5
X sj
Var(
d Ĥ(5)) = Var(
d Ĥ(y5 )) = = 0.0226.
rj
j=1
(ii)The linear confidence interval is

√ √
(0.8 − 1.96 0.0226, 0.8 + 1.96 0.0226) = (0.5053, 1.09467).
(iii) The log-transformed confidence interval is

√ √
0.0266 0.0266
−1.96 1.96
(0.8e 0.8 , 0.8e 0.8 ) = (0.5365, 1.1562)
Example 55.5 ‡
A survival study gave (1.63, 2.55) as the 95% linear confidence interval for
the cumulative hazard function H(t0 ).
Calculate the 95% log-transformed confidence interval for H(t0 ).
Solution.
The interval (1.63, 2.55) has endpoints that can be written as 2.09 ± 0.46.
Thus, Ĥ(t0 ) = 2.09 and 1.96σ̂ = 0.46. Hence, σ̂ = 0.2347. For the log-
transformed confidence interval, we first find
0.2347
U = e1.96( 2.09 ) = 1.2462.
Thus, the confidence interval is

!
Ĥ(t) 2.09
, Ĥ(t)U = , 2.09(1.2462) = (1.68, 2.60)
U 1.2462
Example 55.6 ‡
For a survival study, you are given:
(i) Deaths occurred at times y1 < y2 < · · · < y9 .
(ii) The Nelson-Åalen estimates of the cumulative hazard function at y3 and
y4 are
Ĥ(y3 ) = 0.4128 and Ĥ(y4 ) = 0.5691
(iii) The estimated variances of the estimates in (ii) are:
Var(
d Ĥ(y3 )) = 0.009565 and Var(
d Ĥ(y4 )) = 0.014448.
Determine the number of deaths at y4 .

Solution.
The number of deaths at time y4 is s4 . From the Nelson-Åalen estimate
fiormula, we have
r4
= Ĥ(y4 ) − Ĥ(y3 ) = 0.5691 − 0.4128 = 0.1563.
s4
From Equation (55.1), we have
s4 d Ĥ(y4 )) − Var(
d Ĥ(y3 )) = 0.014448 − 0.009565 = 0.004883.
= Var(
r42
Thus,
2
s4
r4 0.15632
s4 = s4 = =5
r42
0.004883
Example 55.7 ‡
You are given:
(i) Eight people join an exercise program on the same day. They stay in the
program until they reach their weight loss goal or switch to a diet program.
(ii) Experience for each of the eight members is shown below:
Time at which...
Member Reach Weight Loss Goal Switch to Diet Program
j xj uj
1 4
2 8
3 8
4 12
5 12
6 12
7 22
8 36
(iii) The variable of interest is time to reach weight loss goal.
Using the Nelson-Åalen estimator, calculate the upper limit of the symmetric
90% linear confidence interval for the cumulative hazard rate function H(12).
Solution.
Reaching weight loss goal is equivalent to death in mortality theory and
switching to a dieting program is considered a censored observation. We
have the following chart
j yj sj rj
1 8 1 7
2 12 2 5
3 22 1 2
4 36 1 1
By the Nelson-Åalen estimation, we have
s1 s2 1 2
Ĥ(12) = + = + = 0.5429.
r1 r2 7 5
The estimated variance of the Nelson-Åalen estimate is
d Ĥ(12)] = s1 + s2 = 1 + 2 = 0.1004.
Var[
r12 r22 72 52
The upper limit of the symmetric 90% linear confidence interval of Ĥ(12) is
√
0.5429 + 1.645 0.1004 = 1.06
Practice Problems
Problem 55.1 ‡
You are given the following information
yj rj sj
1 30 5
2 27 9
3 32 6
4 25 5
5 20 4
Find Ĥ(3) and Var(

d Ĥ(3)).
Problem 55.2 ‡
Obtain the 95% linear confidence interval and the 95% log-transformed con-
fidence interval in Problem 55.1
Problem 55.3 ‡
The interval (0.357, 0.700) is a 95% log-transformed confidence interval for
the cumulative hazard rate function at time t, where the cumulative hazard
rate function is estimated using the Nelson-Åalen estimator.
Determine the value of the Nelson-Åalen estimate of S(t).
Problem 55.4 ‡
the time of the first claim. The observed data are as follows.
Using the Nelson-Åalen estimator, calculate the 95% linear confidence in-
terval for the cumulative hazard rate function H(4.5).
Problem 55.5 ‡
For a survival study, you are given:
(i) The Product-Limit estimator Sn (t0 ) is used to construct confidence in-
tervals for S(t0 ).
(ii) The 95% log-transformed confidence interval for S(t0 ) is (0.695,0.843).
Determine Sn (t0 ).
Problem 55.6 ‡
Obtain the 95% log-transformed confidence interval for H(3) in Example
55.1, based on the Nelson-Åalen estimate.
Problem 55.7 ‡
Fifteen cancer patients were observed from the time of diagnosis until the
earlier of death or 36 months from diagnosis. Deaths occurred during the
study as follows:
Time in Months Number of

Since Diagnosis Deaths
15 2
20 3
24 2
30 d
34 2
36 1
The Nelson-Åalen estimate Ĥ(35) is 1.5641.
Calculate the Åalen estimate of the variance of Ĥ(35).

56 Kernel Density Estimation

We have had a first encounter with kernel density models in Section 17.
The objective of this section is to be able to create a continuous distribution
whose pdf will approximate a given (discrete) empirical distribution consist-
ing of a complete individual data set such as {y1 , y2 , · · · , yk }. This process
is referred to as kernel smoothing.
The process of kernel smoothing goes as follows: For each observation

yj , j = 1, 2, · · · , k, we let p(yj ) be the probability assigned by the em-
pirical distribution. We also create a continuous random variable Xj with
mean yj and with a pdf denoted by kyj (x) and which we call the kernel
function. The corresponding cdf will be denoted by Kyj (x).
Next, we define the kernel density estimator, also known as the ker-
nel smoothed estimate, of the distribution function by
k
X
F̂ (x) = p(yj )Kyj (x)
j=1
and the kernel density estimator of the density function by
k
X
fˆ(x) = p(yj )kyj (x).
j=1
In [1], three types of kernels are only considered: uniform, triangular, and
Gamma. The uniform kernel with bandwith b is given by

 0, x < y − b,
1
ky (x) = , y − b ≤ x ≤ y + b,
 2b
0, x > y + b.
The corresponding cdf is given by


 0, x < y − b,
x−y+b
Ky (x) = 2b , y − b ≤ x ≤ y + b,
1, x > y + b.

Note that kyj (x) is the pdf of a uniform distribution in the interval [yj −
b, yj + b].
56 KERNEL DENSITY ESTIMATION 401
Example 56.1
You are given the following ages at time of death of 10 individuals:
25 30 35 35 37 39 45 47 49 55.
Use a uniform kernel of bandwidth 10 to estimate the probability density

function at x = 40.
Solution.
With b = 10, we have that ky (40) = 0 for y = 25 and y = 55. Thus, we have
1
k30 (40) = k35 (40) = k37 (40) = k39 (40) = k45 (40) = k47 (40) = k49 (40) = .
20
Hence,
fˆ(40) =p(30)k35 (40) + p(35)k35 (40) + p(37)k37 (40) + p(39)k39 (40)

+p(45)k45 (40) + p(47)k47 (40) + p(49)k49 (40)

1 1 2 1
=6 +
10 20 10 20
1
=
25
The triangular kernel with bandwith b is given by

x < y − b,

 0,
 x−y+b

b2
, y − b ≤ x ≤ y,
ky (x) = y+b−x
, y ≤ x ≤ y + b,
 b2


0, x > y + b.
The corresponding cdf is given by


 0, x < y − b,
 (x−y+b)2

2b2
, y − b ≤ x ≤ y,
Ky (x) = (y+b−x)2


 1 − 2b2 , y ≤ x ≤ y + b,
1, x > y + b.

Example 56.2
25 30 35 35 37 39 45 47 49 55.
Using a triangular kernel with bandwith 10, find the kernel smoothed density
estimate fˆ(40).
Solution.
The triangular kernel with bandwith 10 is

 0, x < y − 10,
 x−y+10

100 , y − 10 ≤ x ≤ y,
ky (x) = y+10−x
, y ≤ x ≤ y + 10,
 100


0, x > y + 10.
We first create the following chart:
y−b y y+b
15 25 35
20 30 40
25 35 45
27 37 47
29 39 49
35 45 55
37 47 57
39 49 59
45 55 65
We have
fˆ(40) =p(30)k30 (40) + p(35)k35 (40) + p(37)k37 (40) + p(39)k39 (40)

+p(45)k45 (40) + p(47)k47 (40) + p(49)k49 (40)
1 0 2 5 1 7 1 9
= + + +
10 100 10 100 10 100 10 100
1 5 1 3 1 1
+ + +
10 100 10 100 10 100
=0.035
y
The Gamma kernel has a Gamma distribution with parameters α and α.
That is,
− xα
xα−1 e y
ky (x) = .
(y/α)α Γ(α)
Example 56.3
25 30 35 35 37 39 45 47 49 55.
Using a triangular kernel with α = 1, find the kernel smoothed density

estimate fˆ(40).
Solution.
With α = 1, the kernel is expressed as follows:
−x
e y
ky (x) = .
y
Thus,
fˆ(40) =p(25)k25 (40) + p(30)k30 (40) + p(35)k35 (40) + p(37)k37 (40)

+p(39)k39 (40) + p(45)k45 (40) + p(47)k47 (40)
+p(49)k49 (40) + p(55)k55 (40)
40 40 40
1 e− 25 1 e− 30 2 e− 35
= + +
10 25 10 30 10 35
40
− 37 − 40 40
1 e 1 e 39 1 e− 45
+ + +
10 37 10 29 10 45
40
− 47 − 40 40
1 e 1 e 49 1 e− 55
+ + +
10 47 10 49 10 55
=0.0491
Example 56.4 ‡
You are given the kernel:
2
p
π 1 − (x − y)2 , y − 1 ≤ x ≤ y + 1
ky (x) =
0, otherwise.
You are also given the following random sample:
1 3 3 5
Determine which of the following graphs shows the shape of the kernel den-
sity estimator.
Solution.
We are given that y1 = 1, y2 = 3, and y3 = 5. The empirical probabilities
are p(y1 ) = p(y3 ) = 0.25 and p(y2 ) = 0.5. The kernel densities are
2
p
π 1 − (x − 1)2 , 0 ≤ x ≤ 2
k1 (x) =
0, otherwise.
2
p
π 1 − (x − 3)2 , 2 ≤ x ≤ 4
k3 (x) =
0, otherwise.
2
p
π 1 − (x − 5)2 , 4 ≤ x ≤ 6
k5 (x) =
0, otherwise.
The graphs of k1 (x), k3 (x), and k5 (x) are shown below. The kernel density
estimator is
3
X
fˆ(x) = p(yi )kyi (x) = 0.25k1 (x) + 0.5k3 (x) + 0.25k5 (x).
i=1
We see that the middle curve has double the coefficient as the curves on the
left and right, so the middle curve is doubled.
Thus, the answer is (D)
Example 56.5 ‡
You are given:
(i) The sample:
1 2 3 3 3 3 3 3 3 3
(ii) F̂1 (x) is the kernel density estimator of the distribution function using
a uniform kernel with bandwidth 1.
(iii) F̂2 (x) is the kernel density estimator of the distribution function using
a triangular kernel with bandwidth 1.
Determine the interval where F̂1 (x) = F̂2 (x).
Solution.
1
The empirical distribution of the data set is given by: p(1) = 10 = 0.1, p(2) =
1 8
10 = 0.1, and p(3) = 10 = 0.8.
The kernel density estimator of the distribution function using a uniform
kernel with bandwidth 1 is
F̂1 (x) = 0.1K1u (x) + 0.1K2u (x) + 0.8K3u (x)
where
( (
0, x<0 0, x<1 0, x<2

K1u (x) = x
2
, 0≤x≤2 , K2u (x) = x−1
2
, 1≤x≤3 , K3u (x) = x−2
2
, 2≤x≤4
1, x>2 1, x>3 1, x > 4.
Thus, 

 0, x<0



 0.05x, 0≤x≤1
0.1x − 0.05, 1≤x≤2

F̂1 (x) =

 0.45x − 0.75, 2≤x≤3
0.4x − 0.6, 3≤x≤4





1, x>4
The kernel density estimator of the distribution function using a triangular
kernel with bandwidth 1 is
F̂2 (x) = 0.1K1t (x) + 0.1K2t (x) + 0.8K3t (x)
where
 

 0, x<0 
 0, x<1
 x2
 (x−1)2

2 ,0≤x≤1 
, 1≤x≤2
K1t (x) = (2−x)2 , K2t (x) = 2
(3−x) 2
 1− 2 , 1≤x≤2
 
 1− 2 , 2≤x≤3
 
 1, x>2 
1, x>3
and 

 0, x<2
 (x−2)2
, 2≤x≤3

K3t (x) = 2
(4−x)2


 1 − 2 , 3≤x≤4
1, x > 4.

Thus, 
 0, x<0
0.05x2 ,




 0≤x≤1
0.1x − 0.05, 1≤x≤2

F̂2 (x) =

 0.35x2 − 1.3x + 1.35, 2≤x≤3
−0.4x2 + 3.2x − 5.4, 3≤x≤4





1, x>4
It follows that F̂1 (x) = F̂2 (x) for all 1 ≤ x ≤ 2

Practice Problems
Problem 56.1 ‡
25 30 35 35 37 39 45 47 49 55.
Using a uniform kernel of bandwidth 10, determine the kernel density esti-
mate of the probability of survival to age 40.
Problem 56.2 ‡
From a population having distribution function F, you are given the follow-
ing sample:
2.0 3.3 3.3 4.0 4.0 4.7 4.7 4.7
Calculate the kernel density estimate of F (4) using the uniform kernel with
bandwidth 1.4.
Problem 56.3 ‡
You use a uniform kernel density estimator with b = 50 to smooth the
following workers compensation loss payments:
82 126 161 294 384.
If F̂ (x) denotes the estimated distribution function and F5 (x) denotes the
empirical distribution function, determine |F̂ (150) − F5 (150)|.
Problem 56.4 ‡
You study five lives to estimate the time from the onset of a disease to death.
The times to death are:
2 3 3 3 7
Using a triangular kernel with bandwidth 2, estimate the density function
at 2.5.
Problem 56.5 ‡
You are given:
(i) The sample: 1 2 3 3 3 3 3 3 3 3.
(ii) F̂1 (x) is the kernel density estimator of the distribution function using
a uniform kernel with bandwidth 1.
(iii) F̂2 (x) is the kernel density estimator of the distribution function using
a triangular kernel with bandwidth 1.
Determine the interval(s) where F̂1 (x) = F̂2 (x).

Problem 56.6 ‡
You study five lives to estimate the time from the onset of a disease to death.
The times to death are:
2 3 3 3 7
Using a triangular kernel with bandwidth 2, estimate the density function
at 2.5.
57 THE KAPLAN-MEIER APPROXIMATION FOR LARGE DATA SETS409
57 The Kaplan-Meier Approximation for Large Data

Sets
In this section, we consider modifying the Kaplan-Meier approximation of
survival function for small data sets to sets with considerably a large number
of data.
Following the notation of [1], suppose the sample data can be split into
k intervals with boundary points c0 < c1 < · · · < ck . Let dj denote the
number of observations that are left-truncated at some value within the in-
terval [cj , cj+1 ). In mortality terms, dj is the number of lives that were first
observed at an age in the given range. Let uj be the number of observations
that are right censored (individual leaving the study for reason other than
death) at some value within the interval (cj , cj+1 ]. Note that the intervals
for dj and uj differ by which endpoints are included and which are omitted.
This is because left-truncation is not possible at the right end of a closed
interval, while right-censoring is not possible at the left end of a closed in-
terval. Let xj be the number of uncensored observations (observed deaths)
within the interval (cj , cj+1 ]. With these notation, the sample size can be
axpressed as n = k−1
P Pk−1
j=1 dj = j=1 (uj + xj ).
In developping the Kaplan-Meier approximation, the following assumptions

are made:
(1) All truncated values occur at the left-endpoint of the intervals and all
censored values occur at the right-endpoints.
(2) None of the uncensored values fall at the endpoints of the intervals.
(3) Ŝ(c0 ) = 1.
With these assumptions, the number at risk14 for the first interval is r0 =
d0 = all the new entrants for the first interval. For this interval, Ŝ(c1 ) =
1 − xd00 .
For the second interval, the number at risk is
r1 = d0 + d1 − x0 − u0 .
14
Recall that the risk set is the number of observations available at a given time that
could produce an uncensored observation at that time.
This is the survivors from the first interval (d0 − x0 − u0 ) plus the new
entrants d1 in the second interval. For this interval,

x0 x1
Ŝ(c2 ) = 1 − 1− .
r0 r1
The general formulas are:
r0 =d0
j
X j−1
X
rj = di − (xi + ui )
i=0 i=0
Ŝ(c0 ) =1
j−1
Y
xi
Ŝ(cj ) = 1−
ri
i=0
where j = 1, 2, · · · , k. Note that the survival function estimate is valued at

the endpoints. For non-endpoints, a linear interpolation between endpoints
is used.
We can use the above developed approximation to estimate the probability

that, given someone is alive at age cj , that person does not survive past age
cj+1 . This estimation is given by
Ŝ(cj ) − Ŝ(cj+1 )
q̂j ≈Pr(T ≤ cj+1 |T > cj ) ≈
Ŝ(cj )
j−1 j
Y xi Y xi
1− − 1−
ri ri
= i=0 j−1
i=0
Y xi

1−
ri
i=0

xj xj
=1 − 1 − = .
rj rj
That is,
number of deaths in time period
q̂j = number of lives considered during that time period .
Remark 57.1
The reader needs to be aware of the difference of notations for xj , dj , uj , and
rj used in this section and the ones used in Section 52. See Problem 57.4.
Example 57.1
Below is the data for a 5-year mortality study.
Age Number of persons Number of persons withdrawing Number of

joining the study at or before the next age deaths
j cj at this age (dj ) uj xj
0 45 800 85 10
1 46 50 65 8
2 47 65 55 6
3 48 45 35 4
4 49 30 25 2
Assume that the activities of withdrawing and joining occur at integral

ages. Use the Kaplan-Meier approximation for large data sets to estimate
the probability of survival to age 47.
Solution.
We first create the following table:
j cj dj uj xj rj
0 45 800 85 10 800
1 46 50 65 8 755
2 47 65 55 6 747
3 48 45 35 4 731
4 49 30 25 2 722
Thus,

x0 x1 x2
Ŝ(2) = 2 p̂45 = 1− 1− 1− = 0.9692
r0 r1 r2
Example 57.2 ‡
Loss data for 925 policies with deductibles of 300 and 500 and policy limits
of 5,000 and 10,000 were collected. The results are given below:
Loss range 300 deductible Policy type (II)

(300,500] 50 −
(500,1000] 50 75
(1000,5000] 150 150
(5000,10000] 100 200
at 5000 40 80
at 10000 10 20
Total 400 525
The ground-up loss distribution for both types of policy is assumed to be the
same. Using the Kaplan-Meier approximation for large data sets to estimate
F (5000).
Solution.
The boundaries of the intervals are: c0 = 300, c1 = 500, c2 = 1000, c3 =
5000 and c4 = 10000. Recall that dj is the number observations that are
left-truncated at some value in [cj , cj+1 ); uj is the number of observations
that are right censored at some value in (cj , cj+1 ]; and xj is the number of
uncensored observations in (cj , cj+1 ). We have the following chart
j dj uj xj rj
0 400 − 50 400
1 525 − 125 875
2 0 120 300 750
3 0 30 300 330
4 0 − − 0
Using the Kaplan-Meier approximation we find

x0 x1 x2
Ŝ(5000) = Ŝ(c3 ) = 1− 1− 1− = .45
r0 r1 r2
Hence,
F̂ (5000) = 1 − 0.45 = 0.55
Example 57.3 ‡
The following table was calculated based on loss amounts for a group of
motorcycle insurance policies:
Pj−1
cj dj ui xj Pj = i=0 (di − ui − xi )
250 6 0 1 0
500 6 0 2 5
1000 7 1 4 9
2750 0 1 7 11
5500 0 1 1 3
6000 0 0 1 1
10,000 0 0 0 0
Estimate the probability that a policy with a deductible of 500 will have a
claim payment in excess of 5500.
Solution.
First note that the risk set is rj = Pj + dj with P0 = 0. The insurance will
pay over 5500 if the insured claim is above 6000 (because of the deductible).
Thus, we are asked to estimate Pr(X > 6000|X > 500) which by Bayes’
theorem is
Pr(X > 6000) S(6000)
Pr(X > 6000|X > 500) = = .
Pr(X > 500) S(500)
We have
x0
Ŝ(500) =Ŝ(c1 ) = 1 −
r0
1
=1 − = 0.83333
6
4
Y xj
Ŝ(6000) =Ŝ(c5 ) = 1−
rj
j=0

5 9 12 4 2
=
6 11 16 11 3
=0.12397.
Hence,
0.12397
Pr(X > 6000|X > 500) ≈ = 0.14876
0.83333
In life table applications, q̂j is a an example of a single-decrement prob-
abilities. When there are multiple causes of decrement, we can express life
table functions pertaining to all causes of decrement with a right superscript
(τ ) 0 (i)
such as qj . For a single cause of decrement we will use the notation qj .
Suppose that there are n causes for a decrement, then the following is true:
n
(τ )
Y 0 (i)
qj = 1 − (1 − pj ).
i=1
Hence, Kaplan-Meier approximation of a survival function can be applied

in the case of multiplt-decrement probabilities.
Example 57.4 ‡
For a double-decrement study, you are given:
(i) The following survival data for individuals affected by both decrements
(1) and (2):
(τ )
j cj qj
0 0 0.100
1 20 0.182
2 40 0.600
3 60 1.000
0 (2)
(ii) qj = 0.05 for all j.
(iii) Group A consists of 1000 individuals observed at age 0.
(iv) Group A is affected by only decrement (1).
Determine the Kaplan-Meier multiple-decrement estimate of the expected
number of individuals in Group A that survive to be at least 40 years old.
Solution.
(τ )
First, the notation qj stands for the probability that a person at age cj
will departs due to some decrement by time cj+1 . Also,
(τ ) 0 (1) 0 (2)
qj = 1 − (1 − qj )(1 − qj ).
From these equations, we find

(τ ) 0 (1) 0 (2) 0 (1) 0 (1)
0.1 = q0 = 1−(1−q0 )(1−q0 ) = 1−(1−q0 )(1−0.05) =⇒ 1−q0 = 0.9474.
Likewise,
(τ ) 0 (1) 0 (2) 0 (1) 0 (1)
0.182 = q1 = 1−(1−q1 )(1−q1 ) = 1−(1−q1 )(1−0.05) =⇒ 1−q1 = 0.8611.
Since Group A is affected only by decrement 1, the survival probability to

age 40= c2 for Group A is
0 (1) 0 (1)
(1 − q0 )(1 − q0 ) = (0.9474)(0.8611) = 0.8158.
The expected number of survivors to age 40 from 1000 Group A individuals

observed at 0 is 1000(0.8158) ≈ 816
Life Table Approach

A second approach for estimating the survival function that is mentioned in
[1] is the life table approach. In this approach, assumptions (1)-(2) are
replaced by the following assumptions:
(1)’ Truncation points and censoring points occur uniformly in each interval.
(2)’ Uncensored observations all occur at the midpoint of the interval.
For such an approach, the risk set ri for an interval (ci , ci+1 ] is found by
the formula
ri = number of observations with value each greater than ci1 + number of

uncensored observations in (ci , ci+1 ] + half the total number of censored
and truncated observations in (ci , ci+1 ].
We illustrate this approach in the next example.
Example 57.5 ‡
You are given the following information about a group of 10 claims:
Claim Size Number of Claims Number of Claims

Interval in Interval Censored in Interval
(0 − 15, 000] 1 2
(15, 000 − 30, 000] 1 2
(30, 000 − 45, 000] 4 0
Assume that claim sizes and censorship points are uniformly distributed
within each interval.
Estimate, using the life table methodology, the probability that a claim
exceeds 30,000.
Solution.
We are asked to find Ŝ(30, 000) which is given by

x0 x1
Ŝ(30, 000) = 1− 1−
r0 r1
where x0 = x1 = 1 (uncensored observations). By the life table methodology,

we have
r0 =7 + 1 + 1 = 9
r1 =4 + 1 + 1 = 6.
Thus,
1 1
Ŝ(3000) = 1 − 1− = 0.741
9 6
Practice Problems
Problem 57.1
Below is the data for a 5-year mortality study.
Age Number of persons Number of persons withdrawing Number of

joining the study at or before the next age deaths
j cj at this age (dj ) uj xj
0 45 800 85 10
1 46 50 65 8
2 47 65 55 6
3 48 45 35 4
4 49 30 25 2
Assume that the activities of withdrawing and joining occur at integral ages.
(a) What is the sample size?
(b) What is the probability that a person does not survive past age 47, given
that he was alive at age 45?
Problem 57.2
Show that rj = rj−1 + dj − (xj−1 + uj−1 ).
Problem 57.3 ‡
(300,500] 50 −
(500,1000] 50 75
(1000,5000] 150 150
(5000,10000] 100 200
at 5000 40 80
at 10000 10 20
Total 400 525
The ground-up loss distribution for both types of policy is assumed to be
the same.
Using the Kaplan-Meier approximation for large data sets to estimate S(1000).
Problem 57.4 ‡

(300,500] 50 −
(500,1000] 50 75
(1000,5000] 150 150
(5000,10000] 100 200
at 5000 40 80
at 10000 10 20
Total 400 525
The ground-up loss distribution for both types of policy is assumed to be

the same.
Estimate the probability that a loss will be greater than 3000 using a Kaplan-
Meier type approximation for large data sets.
Methods of parameter
Estimation
The purpose of this chapter is to discuss methods for the estimation of pa-
rameters in parametric models. Below we present an example of a parameter
estimation.
Example 58.1 ‡
For a sample of 15 losses, you are given:
(i)
Observed number
Interval of Losses
(0, 2] 5
(2, 5] 5
(5, ∞) 5
(ii) Losses follow the uniform distribution on (0, θ).

3
X (Ei − Oi )2
Estimate θ by minimizing the function where Ei is the ex-
Oi
i=1
pected number of losses in the irmth interval and Oj is the observed number
of losses in the ith interval.
Solution.
Since there are losses in (5, ∞), we must have θ > 5. Since there are 15 losses
and the probability of a loss to be in (0, 2] is 2−0
θ , the expected number of
losses in that interval is E1 = 2θ (15) = 30θ . Likewise, E2 = 45θ and E3 =
75
15 − θ . Hence, the formula given in the problem reduces to
" 2 2 #
1 30 45 75 2
f (θ) = −5 + − 5 + 10 − .
5 θ θ θ
419
420 METHODS OF PARAMETER ESTIMATION
Taking the derivative and setting it 0, we find

1
−2(30θ−1 − 5)(30θ−2 ) − 2(45θ−1 − 5)(45θ−2 ) + 2(10 − 75θ−1 )(75θ−2 ) = 0

5
Multiply both sides by − 52 θ3 and simplify the resulting equation to obtain
8550 − 1125θ = 0 =⇒ θ = 7.60.
Thus, f (θ) has only one critical value. Moreover, f 00 (θ) = 10260θ−4 and
f 00 (7.60) > 0 so that f (θ) is minimized at θ = 7.60. That is, θ̂ = 7.60
58 METHOD OF MOMENTS AND MATCHING PERCENTILE 421
58 Method of Moments and Matching Percentile

In this section, we want to estimate parameters in a parametric model by
using finite random samples taken from the underlying distribution. We will
look at two methods of estimation: the method of moments and the method
of percentile matching.
Let X be a random variable with parameters θ1 , θ2 , · · · , θp . From the distri-

bution of X, select randomly n independent observations. We will denote
the cdf of the distribution by
F (x|θ) where θT = (θ1 , θ2 , · · · , θp ).
We will also denote the kth raw moment of X by µ0k (θ) = E(xk |θ). For a
sample of n independent observations {x1 , x2 , · · · , xn } from this distribution
0 1 Pn k
we let µ̃k = n i=1 xi be the empirical estimate of the kth moment.
A method-of-moments estimate of the vector θ is any solution to the

following system of p equations
µ0k (θ) = µ̃0k , k = 1, 2, · · · , p.
Example 58.2
For a normal distribution, derive expressions for the method of moment
estimators for the parameters µ and σ 2 .
Solution.
We have to solve the following system of two equations
n
1X
µ = E(X) = xi = X
n
i=1
and
n
1X 2
xi = E(X 2 ) = σ 2 + µ2 .
n
i=1
Thus,
2
µ̃ = X and σ˜2 = 1 Pn 2
n i=1 xi −X
Example 58.3
For a Gamma distribution with a shape parameter α and a scale parameter
θ, derive expressions for their method of moment estimators.
Solution.
n
1X
αθ = E(X) = xi
n
i=1
and
n
1X
α(α + 1)θ = E(X 2 ) = xi .
n
i=1
We have
1 Pn 2 1 Pn 2
α(α + 1)θ2 n i=1 xi n i=1 xi
= =⇒ θ̃ = −X
αθ X X
and " P #−1
1 n 2
n i=1 xi
α̃ = X −X
X
Example 58.4
For a Pareto distribution, derive expressions for the method of moment
estimators for the parameters α and θ.
Solution.
θ
=X
α−1
and
n
2θ2 1X 2
= xi .
(α − 1)(α − 2) n
i=1
We have −2 Pn
1 2
2θ2

θ n i=1 xi
= 2 .
(α − 1)(α − 2) α−1 X
Solving this last equation, we find
" n #" n
#−1
1X 2 2 1 X 2 2
α̃ = xi − X xi − X .
n 2n
i=1 i=1
Also, " 
n
#" n
#−1
 1X 1 X 2 
2 2
θ̃ = X x2i − X xi − X −1
 n 2n 
i=1 i=1
Let 100gth percentile be denoted by πg (θ) where F (πg (θ)|θ) = g. Let π̂g
denote the smoothed empirical estimate of the 100gth percentile.
A percentile matching estimate of θ is any solution of the p equations
πgk (θ) = π̂gk , k = 1, 2, · · · , p
or
F (π̂gk |θ) = gk , k = 1, 2, · · · , p
where g1 , g2 , · · · , gp are arbitrarily chosen percentiles. In this book, the pos-
sible p values are either 1 or 2.
The smoothed empirical estimate π̂g of the 100gth percentile is found in

the following way:
(i) order the sample values from smallest to largest: x(1) , x(2) , · · · , x(n) ;
(ii) find the integer p such that
p p+1
≤g≤ ;
n+1 n+1
(iii) π̂g is found by linear interpolation
π̂g = [p + 1 − (n + 1)g]x(p) + [(n + 1)g − p]x(p+1) .
Example 58.5 ‡
A random sample of 20 observations has been ordered as follows:
12 16 20 23 26 28 30 32 33 35
36 38 39 40 41 43 45 47 50 57
Determine the 60th sample percentile using the smoothed empirical estimate.
Solution.
We want an integer p such that
p ≤ 0.6(21) ≤ p + 1 =⇒ p = 12.
Thus,
π̂0.6 = [12 + 1 − (20 + 1)(0.6)](38) + [(20 + 1)(0.6) − 12](39) = 38.6

Example 58.6 ‡
You are given:
(i) Losses follow a Burr distribution with parameters γ, θ and α = 2.
(ii) A random sample of 15 losses is
195 255 270 280 350 360 365 380 415 450 490 550 575 590 615.
Use the smoothed empirical estimates of the 30th and 65th percentiles match-
ing to estimate the parameters γ and θ.
Solution.
The cdf of the Burr distribution is
2
1
F (x|θ) = 1 − .
1 + (x/θ)γ
We want an integer p0.3 such that
p0.3 ≤ 0.3(16) ≤ p0.3 + 1 =⇒ p0.3 = 4.
Likewise, we want an integer p0.65 such that
p0.65 ≤ (0.65)(16) ≤ p0.65 + 1 =⇒ p0.65 = 10.
Hence,
π̂0.3 = [5 − 16(0.3)](280) + [16(0.3) − 4](350) = 336
and
π̂0.65 = [11 − 16(0.65)](450) + [16(0.65) − 10](490) = 466.
Hence, we have
2 γ
1 336
1− = 0.3 =⇒ = 0.1952
1 + (336/θ)γ θ
and 2 γ
1 466
1− = 0.65 =⇒ = 0.6903.
1 + (466/θ)γ θ
Hence,
γ γ −γ
466 466 336
= = 3.5364.
336 θ θ
Hence, γ̃ = ln 3.5364[ln 466 − ln 336]−1 = 3.86 and θ̃ = 512.96
Example 58.7 ‡
For a sample of dental claims X1 , X2 , · · · , X10 , you are given:
10
X X10
(i) Xi = 3860 and Xi2 = 4, 574, 802.
i=1 i=1
(ii) Claims are assumed to follow a lognormal distribution with parameters
µ and σ.
(iii) µ and σ are estimated using the method of moments.
Calculate E(X ∧ 500) for the fitted distribution.
Solution.
We have
10
X
Xi
µ+0.5σ 2 i=1
e = E(X) = = 386 =⇒ µ + 0.5σ 2 = 5.9558
10
and
10
X
Xi2
2 i=1
e2µ+2σ = E(X 2 ) = = 457, 480.2 =⇒ 2µ + 2σ 2 = 13.0335.
10
Solving this system of equations, we find µ̂ = 5.3949 and σ̂ 2 = 1.1218.
Next, using A.5.1.1 in Table C, we have

ln 500 − 5.3949 − 1.1218
E(X ∧ 500) =e5.3949+0.5(1.1218) Φ √
1.1218

ln 500 − 5.3949
+500 1 − Φ √
1.1218
5.3949+0.5(1.1218)
=e Φ(−0.2852) + 500[1 − Φ(0.7739)
=e5.3949+0.5(1.1218) (0.3877) + 500(1 − 0.7805)
=259.4.
Note that the values of the standard normal distribution were obtained using
Excel
Example 58.8 ‡
You are given:
(i) Losses follow an exponential distribution with mean θ.
(ii) A random sample of losses is distributed as follows:

(0 − 100] 32
(100 − 200] 21
(200 − 400] 27
(400 − 750] 16
(750 − 1000] 2
(1000 − 1500] 2
Estimate θ by matching at the 80th percentile.
Solution.
The empirical distribution of the grouped data (see Section 50) is expressed
as follows
c 100 200 400 750 1000 1500
F (c) 0.32 0.53 0.8 0.96 0.98 1.00
Thus,
400
0.8 = F (400) = 1 − e− θ =⇒ θ̂ = 248.53
Example 58.9 ‡
You are given the following sample of claim counts:
0 0 1 2 2
You fit a binomial (m, q) model with the following requirements:

(i) The mean of the fitted model equals the sample mean.
(ii) The 33rd percentile of the fitted model equals the smoothed empirical
33rd percentile of the sample.
Determine the smallest estimate of m that satisfies these requirements.
Solution.
Let N denote the binomial random variable. Then (i) yields
0+0+1+2+2
mq = E(N ) = X = = 1.
5
For the 33rd percentile of the sample, we seek p such that
p ≤ 0.33(6) ≤ p + 1 =⇒ p = 1.
Since the first and the second terms in the sample are 0, we conclude that
the 33rd percentile of the sample is 0. By (ii) and the definition of the
percentile introduced in Section 6, we must have p0 = Pr(N = 0) ≥ 0.33

where
1 m

p0 = 1 − .
m
We now have
m p0
1 0
2 0.25
3 0.2963
4 0.3154
5 0.3277
6 0.3349
Hence, the smallest value of m is m = 6
Example 58.10 ‡
You are given:
(i) Losses on a certain warranty product in Year i follow a lognormal distri-
bution with parameters µi and σi .
(ii) σi = σ for i = 1, 2, 3.
(iii) The parameters µi vary in such a way that there is an annual inflation
rate of 10% for losses.
(iv) The following is a sample of seven losses:
Year 1: 20 40 50
Year 2: 30 40 90 120.
Using trended losses, determine the method of moments estimate of µ3 .

Solution.
Counting inflation in both Year 1 and Year 2, the amount of the 7 losses in
Year 3 are:
Year 1 Year 2 Year 3
20 20(1.1) 20(1.1)2 = 24.2
40 40(1.1) 40(1.1)2 = 48.4
50 50(1.1) 50(1.1)2 = 60.5
− 30 30(1.1) = 33
− 40 40(1.1) = 44
− 90 90(1.1) = 99
− 120 120(1.1) = 132
We have
24.2 + 48.4 + 60.5 + 33 + 44 + 99 + 132
µ01 = = 63.014
7
24.22 + 48.42 + 60.52 + 332 + 442 + 992 + 1322
µ02 = = 5252.64
7
2
µ1 =eµ3 +0.5σ3
2
µ2 =e2µ3 +2σ3 .
By the method of moments, we have the following system of equations

2
eµ3 +0.5σ3 = 63.014 =⇒ µ3 + 0.5σ32 = ln 63.014
2
e2µ3 +2σ3 = 5252, 64 =⇒ 2µ3 + 2σ32 = ln 5252.64.
Solving this system, we find
4 ln 63.014 − ln 5252.64
µ3 = = 4.00
2
Practice Problems
Problem 58.1 ‡
The 20th and 80th percentiles of a sample are 5 and 12. Using the percentile
matching method, estimate S(8) assuming the population has a Weibull
distribution.
Problem 58.2 ‡
You are given the following information about a sample of data:
(i) Mean = 35,000
(ii) Standard deviation = 75,000
(iii) Median = 10,000
(iv) 90th percentile = 100,000
(v) The sample is assumed to be from a Weibull distribution.
Determine the percentile matching estimate of the parameter τ.
Problem 58.3 ‡
You are given the following sample of five claims:
4 5 21 99 421
You fit a Pareto distribution using the method of moments. Determine the
95th percentile of the fitted distribution.
Problem 58.4 ‡
In year 1 there are 100 claims with an average size of 10,000, and in year 2
there are 200 claims with an average size of 12,500. Inflation increases the
size of all claims by 10% per year. A Pareto distribution with α = 3 and θ
unknown is used to model the claim size distribution.
Estimate θ for year 3 using the method of moments.
Problem 58.5 ‡
The following 20 wind losses (in millions of dollars) were recorded in one
year:
1 1 1 1 1 2 2 3 3 4
6 6 8 10 13 14 15 18 22 25
Determine the 75th sample percentile using the smoothed empirical estimate.
Problem 58.6 ‡
You are given:
(i) Losses follow a loglogistic distribution with cumulative distribution func-
tion:
(x/θ)γ
F (x) = .
1 + (x/θ)γ
(ii) The sample of losses is:
10 35 80 86 90 120 158 180 200 210 1500
Calculate the estimate of θ by percentile matching, using the 40th and 80th
empirically smoothed percentile estimates.
Problem 58.7 ‡
You are given:
(i) A sample x1 , x2 , · · · , , x10 is drawn from a distribution with probability
density function:

1 1 −x 1 −x
f (x) = e + e
θ σ , x > 0.
2 θ σ
(ii) θ > σ.
10
X 10
X
(iii) xi = 150 and x2i = 5000.
i=1 i=1
Estimate θ by matching the first two sample moments to the corresponding
population quantities.
Problem 58.8 ‡
You are given the following claim data for automobile policies:
200 255 295 320 360 420 440 490 500 520 1020
Calculate the smoothed empirical estimate of the 45th percentile.
Problem 58.9 ‡
You are given:
x 0 1 2 3
Pr(X = x) 0.5 0.3 0.1 0.1
The method of moments is used to estimate the population mean µ, and

4
X
2 2 1
variance σ , by X nad s4 = 4 (Xi − X)2 , respectively.
i=1
Calculate the bias of s24 .
Problem 58.10 ‡
You are given:
(i) Claim amounts follow a shifted exponential distribution with probability
density function:
1 x−δ
f (x) = e− θ , x > δ.
θ
(ii) A random sample of claim amounts X1 , X2 , · · · , X10
5 5 5 6 8 9 11 12 16 23.
10
X 10
X
(iii) xi = 100 and xi = 1306.
i=1 i=1
Estimate δ using the method of moments.
Problem 58.11 ‡
You are given the following random sample of 13 claim amounts:
99 133 175 216 250 277 651 698 735 745 791 906 947
Determine the smoothed empirical estimate of the 35th percentile.
Problem 58.12 ‡
The parameters of the inverse Pareto distribution
τ
x
F (x) =
x+θ
are to be estimated using the method of moments based on the following

data:
15 45 140 250 560 1340
Estimate θ by matching k th moments with k = −1 and k = −2.
Problem 58.13 ‡
You are given:
(i) Losses are uniformly distributed on (0, θ) with θ > 150.
(ii) The policy limit is 150.

(iii) A sample of payments is:
14 33 72 94 120 135 150 150
Estimate θ by matching the average sample payment to the expected pay-
ment per loss.
Problem 58.14 ‡
You are given the following data:
0.49 0.51 0.66 1.82 3.71 5.20 7.62 12.66 35.24

You use the method of percentile matching at the 40th and 80th percentiles
to fit an Inverse Weibull distribution to these data.
Determine the estimate of θ.
Problem 58.15 ‡
The following claim data were generated from a Pareto distribution:
130 20 350 218 1822
Using the method of moments to estimate the parameters of a Pareto dis-
tribution, calculate the limited expected value at 500.
Problem 58.16 ‡
A random sample of claims has been drawn from a Burr distribution with
known parameter α = 1 and unknown parameters θ and γ. You are given:
(i) 75% of the claim amounts in the sample exceed 100.
(ii) 25% of the claim amounts in the sample exceed 500.
Estimate θ by percentile matching.
Problem 58.17 ‡
A random sample of observations is taken from a shifted exponential distri-
bution with probability density function:
1 (x−δ)
f (x) = e− θ , δ < x < ∞.
θ
The sample mean and median are 300 and 240, respectively.
Estimate δ by matching these two sample quantities to the corresponding

population quantities.
Problem 58.18 ‡
For a portfolio of policies, you are given:
(i) Losses follow a Weibull distribution with parameters θ and τ.
(ii) A sample of 16 losses is :
54 70 75 81 84 88 97 105 109 114 122 125 128 139 146 153
(iii) The parameters are to be estimated by percentile matching using the

20th and 70th smoothed empirical percentiles.
Calculate the estimate of θ.
Problem 58.19 ‡
You are modeling a claim process as a mixture of two independent distribu-
tions A and B. You are given:
(i) Distribution A is exponential with mean 1.
(ii) Distribution B is exponential with mean 10.
(iii) Positive weight p is assigned to distribution A.
(iv) The standard deviation of the mixture is 2.
Determine p using the method of moments.
Problem 58.20 ‡
You are given the following information about a study of individual claims:
(i) 20th percentile = 18.25
(ii) 80th percentile = 35.80
Parameters µ and σ of a lognormal distribution are estimated using per-
centile matching.
Determine the probability that a claim is greater than 30 using the fitted
lognormal distribution.
59 Maximum Likelihood Estimation for Complete

Data
Maximum likelihood is a relatively simple method of constructing an esti-
mator for an unknown parameter θ in a parametric model.
In order to define the maximum likelihood estimator, we start with the

following notations: Let the data set be grouped into events A1 , A2 , · · · , An
where Aj is whatever observed for the jth observation. The event Aj can
be a single point (single individual data) or an interval (grouped or cen-
sored data). Furthermore, the event Aj results from observing the random
variable Xj . We assume that the random variables X1 , X2 , · · · , Xn are in-
dependent and their distributions depend on the same parameter that need
to be estimated. The method of maximum likelihood is to estimate the pa-
rameter that maximizes the probability or the likelihood of getting the data
we observed. This is done by maximizing the likelihood function
n
Y
L(θ) = Pr(Xj ∈ Aj |θ).
i=1
We use the symbol “|” to indicate that the distribution also depends on a
parameter θ, where θ could be a real-valued unknown parameter or a vector
of parameters.
In the case of a complete individual data, consider a random sample with ob-
served values X1 = x1 , X2 = x2 , · · · , Xn = xn . Then the likelihood function
is
n
Y
L(θ) = f (x1 , x2 , · · · , xn |θ) = f (xi |θ)
i=1
where f (x1 , x2 , · · · , xn |θ) is a joint density if the the Xi s are continuous or

a joint mass function if the Xi s are discrete.
A maximum likelihood estimate(MLE) is the value of θ that maximizes

L(θ).
Example 59.1
Consider the following discrete random variable X whose pmf is given below.
X 0 1 2 3
θ 1−θ 2θ 2(1−θ)
p(x|θ) 3 3 3 3
59 MAXIMUM LIKELIHOOD ESTIMATION FOR COMPLETE DATA435
where 0 ≤ θ ≤ 1. Consider the following random sample of 7 observations

taken from this distribution {2, 3, 0, 1, 1, 2, 3}. Determine L(θ).
Solution.
We have
L(θ) =p(2|θ)p(3|θ)p(0|θ)p(1|θ)p(1|θ)p(2|θ)p(3|θ)
2
2(1 − θ) 2 1 − θ 2 θ

2θ
=
3 3 3 3
Clearly, the likelihood function L(θ) is not easy to maximize. But maximiz-
ing L(θ) is equivalent to maximizing ln [L(θ)] since ln [L(θ)] is an increasing
function of θ. We define the loglikelihood function as
n
X
`(θ) = ln [L(θ)] = f (xi |θ).
i=1
Example 59.2
Find the maximum likelihood estimate of θ in the previous example.
Solution.
Let us look at the log likelihood function
`(θ) = ln [L(θ)]

2θ 2(1 − θ) 1−θ θ
=2 ln + 2 ln + 2 ln + ln .
3 3 3 3
Using calculus, we have
d` 3
= 0 =⇒ θ = .
dθ 7
Also,
d2 `

343
=− < 0.
dθ2 θ= 3 49
7
3
Hence, θ̂ = 7
Example 59.3
A random sample of 5 claims obtained from an exponential distribution with
parameter θ is given as follows:
15 10 7 8 20.
Find the maximum likelihood estimate of θ.

Solution.
1 − xθ
The density function of the distribution is f (x|θ) = θe . The likelihood
function is given by
1 − 1 (x1 +x2 +···+x5 )
L(θ) = e θ .
θ5
The loglikelihood function is
1
`(θ) = − (x1 + x2 + · · · + x5 ) − 5 ln θ.
θ
Let the derivative with respect to θ be zero:
x1 + x2 + · · · + x5
`0 (θ) = 0 =⇒ θ = = 12.
5
Moreover,
d2 `

= −5 < 0.
dθ2 θ=12
Thus, the MLE is
x1 + x2 + · · · + x5
θ̂ = = 12
5
For a complete and grouped data, the process of finding the naximum like-
lihood estimate goes as follows: Arrange the unique observation values in
increasing order
c0 < c1 < · · · < ck
where c0 is the smallest possible observation (often zero) and ck is the largest
possible observation (often infinity). For j = 1, 2, · · · , k, let nj the number
of observations in Aj = (cj−1 , cj ]. The likelihood contribution of each value
in the jth observation is
Pr(X ∈ Aj ) = Pr(cj−1 < x ≤ cj ) = F (cj |θ) − F (cj−1 |θ).
Thus, the likelihood function is

k
Y
L(θ) = [F (cj |θ) − F (cj−1 |θ)]nj
j=1
and the loglikelihood function is

k
X
`(θ) = nj ln [F (cj |θ) − F (cj−1 |θ)].
j=1
Example 59.4 ‡
Suppose that a group of 20 losses resulted in the following
Loss Range # of observations

(0, 10] 9
(10, 25] 6
(25, ∞) 5
Losses follow the distribution F (x) = 1 − xθ , where x > θ. Calculate the

maximum likelihood estimate of θ.
Solution.
The likelihood function is
L(θ) =[F (10)]9 [F (25) − F (10)]6 [1 − F (25)]5

=(1 − 0.1θ)9 (0.06θ)6 (0.04θ)5 .
`(θ) = 9 ln (1 − 0.1θ) + 6 ln 0.06 + 6 ln θ + 5 ln 0.04 + 5 ln θ.
Taking its derivative and setting it to zero, we find

0.9 11
`0 (θ) = − + = 0 =⇒ θ = 5.5.
1 − 0.1θ θ
Furthermore,
d2 `

= −0.8081 < 0.
dθ2 θ=5.5
Hence, θ̂ = 5.5
It is possible that a data set contains a combination of individual and

grouped data as suggested in the following example.
Example 59.5 ‡
The random variable X has survival function:
θ4
SX (x) = .
(θ2 + x2 )2
Two values of X are observed to be 2 and 4. One other value exceeds 4.

Calculate the maximum likelihood estimate of θ.
Solution.
L(θ) = fX (2|θ)fX (4|θ)SX (4|θ)
where
4xθ4
fX (x) = .
(θ2 + x2 )3
`(θ) = ln fX (2|θ)+ln fX (4|θ)+ln SX (2|θ) = 7 ln 2+12 ln θ−3 ln (4 + θ2 )2 −5 ln (16 + θ2 )2 .
Taking the derivative to obtain
−4θ4 + 104θ2 + 768

`0 (θ) = .
θ(θ2 + 4)(θ2 + 16)
Thus,
√
`0 (θ) = 0 =⇒ −4θ4 + 104θ2 + 768 = 0 =⇒ θ2 = 32 =⇒ θ = 4 2.
√ √
We leave it to the reader to check that `00 (4 2) < 0 so that θ̂ = 4 2
Example 59.6 ‡
You have observed the following three loss amounts:
186 91 66
Seven other amounts are known to be less than or equal to 60. Losses follow
an inverse exponential with distribution function
θ
F (x) = e− x , x > 0.
Calculate the maximum likelihood estimate of the population mode.
Solution.
θ
The pdf of the inverse exponential distribution is given by f (x) = θx−2 e− x .
L(θ) =f (186)f (91)f (66)F (60)7

θ θ θ θ
=θ186−2 e− 186 θ91−2 e− 91 θ66−2 e− 66 (e− 60 )7
−1 +91−1 +66−1 +7(60−1 )
=(186 · 91 · 66)−2 θ3 e−θ(186 .
ln(θ) = −2 ln (186 · 91 · 66) + 3 ln θ − θ(186−1 + 91−1 + 66−1 + 7(60−1 ).
Taking the first derivative and setting it to 0, we find

3
`0 (θ) = − (186−1 + 91−1 + 66−1 + 7(60−1 ) = 0 =⇒ θ̂ = 20.25.
θ
From Table C, the mode of the inverse exponential function is
θ̂ 20.25
= = 10.125
2 2
Example 59.7 ‡
You are given:
(i) The distribution of the number of claims per policy during a one-year
period for 10,000 insurance policies is:
Number of Claims per Policy Number of Policies

0 5000
1 5000
2 or more 0
(ii) You fit a binomial model with parameters m and q using the method of
maximum likelihood.
Determine the maximum value of the loglikelihood function when m = 2.
Solution.
The pdf of the binomial distribution with m = 2 is

2
f (x) = q x (1 − q)2−x .
x

5000 5000
5000 5000 2 2 2
L(q) =f (0) f (1) = (1 − q) q(1 − q)
0 1
=25000 q 5000 (1 − q)15000 .
`(q) = 5000 ln 2 + 5000 ln q + 15000 ln (1 − q).

Taking the derivative and setting it to zero, we find

5000 15000
`0 (q) = − = 0 =⇒ q̂ = 0.25.
q 1−q
Finally,
`(0.25) = 5000 ln 2 + 5000 ln 0.25 + 15000 ln 0.75 = −7780.97

Practice Problems
Problem 59.1
Suppose X1 , X2 , · · · , Xn are i.i.d random variables with a Gamma distribu-
tion with α = 2 and θ.
Problem 59.2
Suppose X1 , X2 , · · · , Xn are i.i.d random variables with a uniform distribu-
tion is (0, θ).
Problem 59.3 ‡
You are given the following three observations:
0.74 0.81 0.95
You fit a distribution with the following density function to the data:
f (x) = (p + 1)xp , 0 < x < 1, p > −1.
Determine the maximum likelihood estimate of p.
Problem 59.4 ‡
The proportion of allotted time a student takes to complete an exam, X, is
described by the following distribution:
f (x) = (θ + 1)xθ , 0 ≤ x ≤ 1, θ > −1.
A random sample of five students produced the following observations:
Student Proportion of Allotted Time

1 0.92
2 0.79
3 0.90
4 0.65
5 0.86
Using the sample data, calculate the maximum likelihood estimate of θ.

Problem 59.5 ‡
Let X1 , X2 , · · · , Xn be a random sample from the following distribution with
pdf −x+θ
e , θ < x, − ∞ < θ < ∞
f (x) =
0, otherwise.

Problem 59.6 ‡
You are given:
(i) Losses follow a Single-parameter Pareto distribution with density func-
tion:
α
f (x) = α+1 , x > 1, 0 < α < ∞.
x
(ii) A random sample of size five produced three losses with values 3, 6 and
14, and two losses exceeding 25.
Determine the maximum likelihood estimate of α.

Problem 59.7 ‡
You are given:
(i) Low-hazard risks have an exponential claim size distribution with mean
θ.
(ii) Medium-hazard risks have an exponential claim size distribution with
mean 2θ.
(iii) High-hazard risks have an exponential claim size distribution with mean
3θ.
(iv) No claims from low-hazard risks are observed.
(v) Three claims from medium-hazard risks are observed, of sizes 1, 2 and
3.
(vi) One claim from a high-hazard risk is observed, of size 15.
Determine the maximum likelihood estimate of θ.

Problem 59.8 ‡
A random sample of three claims from a dental insurance plan is given below:
225 525 950
Claims are assumed to follow a Pareto distribution with parameters θ = 150
and α.

Problem 59.9 ‡
You are given:
(i) Losses follow an exponential distribution with mean θ.
(ii) A random sample of 20 losses is distributed as follows:
Loss Range # of observations

(0, 1000] 7
(1000, 2000] 6
(2000, ∞) 7
Problem 59.10 ‡
You fit an exponential distribution to the following data:
1000 1400 5300 7400 7600
Determine the coefficient of variation of the maximum likelihood estimate

of the mean, θ.
Problem 59.11 ‡
Losses come from a mixture of an exponential distribution with mean 100
with probability p and an exponential distribution with mean 10,000 with
probability 1 − p. Losses of 100 and 2000 are observed.
Determine the likelihood function of p.
Problem 59.12 ‡
Let x1 , x2 , · · · , xn and y1 , y2 , · · · , ym denote independent random samples of
losses from Region 1 and Region 2, respectively. Single-parameter Pareto
distributions with θ = 1, but different values of α, are used to model losses
in these regions.
Past experience indicates that the expected value of losses in Region 2 is 1.5
times the expected value of losses in Region 1. You intend to calculate the
maximum likelihood estimate of α1 for Region 1, using the data from both
regions.
d
Find dα1 `(α1 ).
Problem 59.13 ‡
You have observed the following claim severities:
11.0 15.2 18.0 21.0 25.8
You fit the following probability density function to the data:

1 (x−µ)2
f (x) = √ e− 2x , x > 0, µ > 0.
2πx
Determine the maximum likelihood estimate of µ.
Problem 59.14 ‡
Phil and Sylvia are competitors in the light bulb business. Sylvia advertises
that her light bulbs burn twice as long as Phil’s. You were able to test 20
of Phil’s bulbs and 10 of Sylvia’s. You assumed that the distribution of the
lifetime (in hours) of a light bulb is exponential, and separately estimated
Phil’s parameter as θP = 1000 and Sylvia’s parameter as qθS = 1500 using
maximum likelihood estimation.
Determine θ∗ , the maximum likelihood estimate of θP restricted by Sylvia’s

claim that θS = 2θP .
60 MAXIMUM LIKELIHOOD ESTIMATION FOR INCOMPLETE DATA445
60 Maximum Likelihood Estimation for Incomplete

Data
If some of the data in the sample has been right-censored, then each observa-
tion in the sample censored at u contributes a factor of S(u|θ) = 1 − F (u|θ)
to the likelihood function. This is because, Pr(Xj ∈ Aj ) = Pr(Xj > u) =
1 − F (u|θ). Recall that when a ground-up loss random variable has a policy
limit of u, every loss less than or equal to u will be recorded with its observed
value while all losses above u will be recorded as u.
Example 60.1
A ground-up loss random variable X has a policy limit of 30. The following
is a random sample of 6 insurance payment amounts:
20 25 27 28 30 30.
If X is assumed to have an exponential distribution, apply maximum likeli-

hood estimation to estimate the mean of X.
Solution.
L(θ) =fX (20|θ)fX (25|θ)fX (27|θ)fX (28|θ)[SX (30|θ)]2

1 1 30×2
= 4 e− θ (20+25+27+28) e− θ
θ
1 1
= 4 e− θ (20+25+27+28+2×30)
θ
1 160
= 4 e− θ .
θ
160
`(θ) = − − 4 ln θ
θ
and
160 4 160
`0 (θ) = − = 0 =⇒ θ = = 40.
θ2 θ 4
Furthermore,

00 320 4
` (θ)(40) = − 3 + 2 = −0.0025 < 0.
θ θ θ=40
Finally, θ̂ = 40
If some of the data in the sample has been left-truncated, for example a
policy with an ordinary deductible d, then each observation in the sam-
ple truncated at d contributes a factor of f (y + d|θ)[1 − F (d|θ)] where y is
recorded after the deductible d is applied. For example, suppose the jth
observation is yj , the loss amount after a deductible of d is applied. Then
f (xj |θ) f (yj + d|θ)
Pr(Xj ∈ Aj ) = Pr(Xj |Xj > d) = = , xj = yj + d.
1 − F (d|θ) 1 − F (d|θ)
Example 60.2
A ground up loss X has a deductible of 7 applied. A random sample of 6
insurance payments (after deductible is applied) is given
3 6 7 8 10 12.
If X is assumed to have an exponential distribution, apply maximum likeli-

hood estimation to estimate the mean of X.
Solution.
The loss amounts before the deductible is applied are:
10 13 14 15 17 19.

f (10|θ)f (13|θ)f (14|θ)f (15|θ)f (17|θ)f (19|θ)
L(θ) =
[1 − F (7|θ)]6
1 1
= 6 e− θ (10+13+14+15+17+19−6×7)
θ
1 46
= 6 e− θ .
θ
46
`(θ) = − − 6 ln θ
θ
and
46 6 46
`0 (θ) = 2
− = 0 =⇒ θ = = 7.6667.
θ θ 6
Furthermore,

00 92 6
` (θ)(7.667) = − 3 + 2 = −0.1021 < 0.
θ θ θ=7.6667
Finally, θ̂ = 7.6667
Instead of using the ground up losses in the above estimation process, one
can use instead the payments after the deductible is applied (i.e. cost per
payment). In this case, the problem reduces to the case of a complete indi-
vidual data. We illustrate this point in the next example.
Example 60.3
A ground up loss X has a deductible of 7 applied. A random sample of 6
insurance payments (after deductible is applied) is given
3 6 7 8 10 12.
(a) If X is assumed to have a uniform distribution in (0, θ), apply maximum

likelihood estimation to estimate the mean of X.
(b) Assuming the cost per payment Y has a uniform distribution in (0, θ),
estimate the mean of Y.
Solution.
(a) Note that the condition xi < θ is equivalent to max{x1 , x2 , · · · , xn } < θ.
Hence, an estimate of θ is θ̂ = max{10, 13, 14, 15, 17, 19} = 19 so that an
estimate of the mean of X is 192 = 9.5.
(b) Similar to (a), an estimate of θ is θ̂ = max{3, 6, 7, 8, 10, 12} = 12 so that
an estimate of the mean is 122 =6
Now, if a policy has an ordinary deductible and maximum loss cover u

then the policy limit is u − d. If the cost per payment y is less than u − d
then the ground up loss amount is x = y + d and this contributes a factor
f (x|θ)[1 − F (d|θ)]−1 to the likelihood function. If the cost per payment is
y = u − d, then the ground up loss amount is at least u and this contributes
a factor [1 − F (u|θ)][1 − F (d|θ)]−1 to the likelihood function.
Example 60.4
A policy has a deductible of d = 3 and a maximum covered loss of u = 14.
A random sample of 6 insurance payments is given:
1 3 7 9 11 11.
Suppose that the ground up loss distribution is exponential with parameter

θ. Find the maximum likelihood estimate of θ.
Solution.
The policy limit is u − d = 14 − 3 = 11. The likelihood function is
f (4|θ)f (6|θ)f (10|θ)f (12|θ) 1 − F (14|θ) 2

L(θ) =
[1 − F (3|θ)]4 1 − F (3|θ)
1 1
= 4 e− θ (4+6+10+12+2×14−6×3)
θ
1 42
= 4 e− θ .
θ
42
`(θ) = − − 4 ln θ
θ
and
42 4
`0 (θ) = − = 0 =⇒ θ = 10.5.
θ2 θ
Moreover,
00 84 4
` (10.5) = − 3 + 2 = −0.0363.
θ θ θ=10.5
Hence, θ̂ = 10.5
Example 60.5 ‡
You are given:
(i) The number of claims follows a Poisson distribution with mean λ.
(ii) Observations other than 0 and 1 have been deleted from the data.
(iii) The data contain an equal number of observations of 0 and 1.
Determine the maximum likelihood estimate of λ.
Solution.
Let N be the number of claims. Notice that we are trying to estimate λ so
the data in our sample represent number of claims. We are told that N is
right truncated at 1. Thus, we have the following
Pr(N = 0) Pr(N = 0) e−λ 1

Pr[N = 0|N ≤ 1] = = = −λ −λ
=
FN (1) Pr(N = 0) + Pr(N = 1) e + λe 1+λ
Pr(N = 1) λe −λ λ
Pr[N = 1|N ≤ 1] = = −λ = .
Pr(N = 0) + Pr(N = 1) e + λe−λ 1+λ
By (ii), the data contain an equal number of observations of 0 and 1. Let’s

say that our sample has n data points of 0’s and 1’s. Then the likelihood
function is
n/2 n/2 n/2
1 λ λ
L(λ) = = .
1+λ 1+λ (1 + λ)2
n
`(λ) = ln λ − n ln (1 + λ).
2
Differentiating and setting to 0, we find
n n
`0 (λ) = − = 0 =⇒ λ̂ = 1
2λ 1 + λ
Example 60.6 ‡
You are given:
(i) At time 4 hours, there are 5 working light bulbs.
(ii) The 5 bulbs are observed for p more hours.
(iii) Three light bulbs burn out at times 5, 9, and 13 hours, while the re-
maining light bulbs are still working at time 4 + p hours.
(iv) The distribution of failure times is uniform on (0, ω).
(v) The maximum likelihood estimate of ω is 29.
Determine p.
Solution.
Let T be the time to failure random variable. T has a uniform distribution
on (0, ω). Its pdf is f (t) = ω1 and its sdf is S(t) = 1 − ωt . The likelihood
function is
f (5) f (9) f (13) S(4 + p) 2

L(ω) =
S(4) S(4) S(4) S(4)
2
1 1 1 4+p
ωωω 1− ω
= 5
1 − ω4
(ω − 4 − p)2
= .
(ω − 4)5
`(ω) = 2 ln (ω − 4 − p) − 5 ln (ω − 4).
Taking the derivative and setting it to zero, we find

2 5 2 5
`0 (ω) = − = 0 =⇒ = .
ω−4−p ω−4 ω̂ − 4 − p ω̂ − 4
But ω̂ = 29. Hence,

2 5
= =⇒ p = 15
29 − 4 − p 29 − 4
Practice Problems
Problem 60.1 ‡
You observe the following five ground-up claims from a data set that is
truncated from below at 100:
125 150 165 175 250
You fit a ground-up exponential distribution using maximum likelihood es-
timation. Determine the mean of the fitted distribution.
Problem 60.2 ‡
You are given:
(i) A sample of losses is: 600 700 900
(ii) No information is available about losses of 500 or less.
(iii) Losses are assumed to follow an exponential distribution with mean θ.

Problem 60.3 ‡
A policy has an ordinary deductible of 100 and a policy limit of 1000. You
observe the following 10 payments:
15 50 170 216 400 620 750 900 900 900.
An exponential distribution is fitted to the ground up distribution function,
using the maximum likelihood estimate.
Determine the estimated parameter θ̂.

Problem 60.4
A ground-up loss random variable X has a policy limit of 2000. The following
is a random sample of 3 insurance payment amounts:
300 1000 2000.
If X is assumed to have a uniform distribution in (0, θ), apply maximum
likelihood estimation to estimate θ.
Problem 60.5
A ground-up loss random variable X has a policy limit of 1000. A random
sample of two payments is obtained as follows: 200 and 1000. The loss X
has a Pareto distribution with θ = 2500.
Calculate the maximum likelihood estimate for α.

Problem 60.6 ‡
For a dental policy, you are given:
(i) Ground-up losses follow an exponential distribution with mean θ.
(ii) Losses under 50 are not reported to the insurer.
(iii) For each loss over 50, there is a deductible of 50 and a policy limit of
350.
(iv) A random sample of five claim payments for this policy is:
50 150 200 350+ 350+
where + indicates that the original loss exceeds 400.
Determine the likelihood function
Problem 60.7 ‡
You are given:
(i) An insurance company records the following ground-up loss amounts,
which are generated by a policy with a deductible of 100:
120 180 200 270 300 1000 2500
(ii) Losses less than 100 are not reported to the company.
(iii) Losses are modeled using a Pareto distribution with parameters θ = 400
and α.
Use the maximum likelihood estimate of α to estimate the expected loss

with no deductible.
Problem 60.8 ‡
You are given the following information about a group of policies:
Claim Payment Policy Limit

5 50
15 50
60 100
100 100
500 500
500 1000
Determine the likelihood function.

Problem 60.9 ‡
You are given the following 20 bodily injury losses (before the deductible is
applied):
Loss Number of Deductible Policy Limit

Losses
750 3 200 ∞
200 3 0 10,000
300 4 0 20,000
>10,000 6 0 10,000
400 4 300 ∞
Past experience indicates that these losses follow a Pareto distribution with
parameters α and θ = 10, 000.

Problem 60.10 ‡
Personal auto property damage claims in a certain region are known to
follow the Weibull distribution:
x 0.2
F (x) = 1 − e−( θ ) , x > 0.
A sample of four claims is:
130 240 300 540
The values of two additional claims are known to exceed 1000.

Problem 60.11 ‡
(i) There is no deductible and the policy limit varies by policy.
(ii) A sample of ten claims is:
350 350 500 500 500+ 1000 1000+ 1000+ 1200 1500
where the symbol + indicates that the loss exceeds the policy limit.
(iii) Ŝ1 (1250) is the product-limit estimate of S(1250).
(iv) Ŝ2 (1250) is the maximum likelihood estimate of S(1250) under the as-
sumption that the losses follow an exponential distribution.
Determine the absolute difference between Ŝ1 (1250) and Ŝ2 (1250).
Problem 60.12 ‡
You are given a sample of losses from an exponential distribution. However,
if a loss is 1000 or greater, it is reported as 1000. The summarized sample
is:
Reported Loss Number Total Amount
Less than 1000 62 28,140
1000 38 38,000
Total 100 66,140
Determine the maximum likelihood estimate of θ, the mean of the exponen-

tial distribution.
Problem 60.13 ‡
You are given the following claims settlement activity for a book of auto-
mobile claims as of the end of 1999:
Number of Claims Settled
Year Year Settled
Reported 1997 1998 1999
1997 Unknown 3 1
1998 5 2
1999 4
L = (Year Settled Year Reported) is a random variable describing the time

lag in settling a claim. The probability function of L is
fL (`) = p(1 − p)` , ` = 0, 1, 2, · · · .
Determine the maximum likelihood estimate of the parameter p.
Problem 60.14 ‡
You are given the following information about a random sample:
(i) The sample size equals five.
(ii) The sample is from a Weibull distribution with τ = 2 and unknown θ.
(iii) Two of the sample observations are known to exceed 50, and the re-
maining three observations are 20, 30 and 45.
61 ASYMPTOTIC VARIANCE OF MLE 455
61 Asymptotic Variance of MLE

Estimators can always be regarded as random variables themselves. In par-
ticular, the maximum likelihood estimator is a random variable. Calculating
the variance of mle is not an easy task. Instead, methods of approximating
the variance exist. The goal of this section is to discuss one such a method.
Our method uses the following concept found in the statistics literature:
Let X be a random variable whose distribution depends on a parameter θ
and denote its pdf by f (x|θ). We define the Fisher information I(X|θ) in
X by Z
I(X|θ) = E[`0 (x|θ)2 ] = [`0 (x|θ)]2 f (x|θ)dx
where `(x|θ) = ln f (x|θ). The prime symbol stands for the differentiation
with respect to θ. We assume that we can exchange the order of differenti-
ation and integration, then
Z Z
0 d d
f (x|θ)dx = f (x|θ)dx = (1) = 0.
dθ dθ
Similarly,
d2 d2
Z Z
f 00 (x|θ)dx =
2
f (x|θ)dx = 2 (1) = 0.
dθ dθ
With these properties of f, we have
Z Z 0 Z
f (x|θ)
E[`0 (x|θ)] = `0 (x|θ)f (x|θ)dx = f (x|θ)dx = f 0 (x|θ)dx = 0.
f (x|θ)
Hence,
I(X|θ) = Var[`0 (x|θ)].
Also, notice that
f 00 (x|θ)f (x|θ) − f 0 (x|θ)2 f 00 (x|θ)
`00 (x|θ) = = − [`0 (θ)]2
f (x|θ)2 f (x|θ)
so that Z
00
E[` (x|θ)] = f 00 (x|θ)dx − E[`0 (x|θ)2 ] = −I(x|θ).
From this, we obtain another way for calculating I(θ) :

Z 2
∂
I(X|θ) = − ln f (x|θ) f (x|θ)dx.
∂θ2
Example 61.1
Let X be a normal random variable with parameters µ and σ 2 . Suppose that
σ 2 is knwon but µ is the unknown parameter. Find the Fisher information
I(X|µ) in X.
Solution.
The pdf of X is
(x−µ)2
1 −
2σ 2
f (x|µ) = √ e .
2πσ
Thus,
1 (x − µ)2
`(x|µ) = − ln (2πσ 2 ) − .
2 2σ 2
Hence
(x−µ)
`0 (x|µ) = σ2
and `00 (x|µ) = − σ12 .
Thus,
1
I(X|µ) = −E[`00 (x|µ)] =
σ2
Now, suppose we have a random sample X1 , X2 , · · · , Xn coming from a
distribution for which the pdf is f (x|θ) where the value of the parameter θ
is unknown. Assuming independence, the joint pdf is given by
n
Y
L(θ) = fn (x1 , x2 , · · · , xn |θ) = f (xi |θ).
i=1

n
X n
X
`(θ) = `(xi |θ) = lnf (xi |θ)
i=1 i=1
and
fn0 (x1 , x2 , · · · , xn |θ)
`0 (θ) = .
fn (x1 , x2 , · · · , xn |θ)
We define the Fisher information I(θ) in the random sample X1 , X2 , · · · , Xn
as
Z Z
I(θ) = E[`0 (θ)2 ] = · · · `0n (θ)2 fn (x1 , x2 , · · · , xn |θ)dx1 dx2 · · · dxn .
We assume that we can exchange the order of differentiation and integration,

then
Z Z
0 d d
f (x1 , x2 , · · · , xn |θ)dx1 dx2 · · · dxn = f (x1 , x2 , · · · , xn |θ)dx1 dx2 · · · dxn = (1) = 0.
dθ dθ
Similarly,
d2 d2
Z Z
f 00 (x1 , x2 , · · · , xn |θ)dx1 dx2 · · · dxn = 2 f (x1 , x2 , · · · , xn |θ)dx1 dx2 · · · dxn = 2 (1) = 0.
dθ dθ
With these properties of f, we have
Z
E[`0 (θ)] = `0n (θ)f (x1 , x2 , · · · , xn |θ)dx1 dx2 · · · dxn
Z 0
f (x1 , x2 , · · · , xn |θ)
= f (x1 , x2 , · · · , xn |θ)dx1 dx2 · · · dxn
f (x1 , x2 , · · · , xn |θ)
Z
= f 0 (x1 , x2 , · · · , xn |θ)dx1 dx2 · · · dxn = 0.
Hence,
I(θ) = Var[`0 (θ)].
Also, notice that
E[`00 (θ)] = −I(θ).
and
Xn n
X
I(θ) = −E[ `00 (xi |θ)] = − E[`00 (xi |θ)] = nI(X|θ).
i=1 i=1
In other words, the Fisher information in a random sample of size n is simply
n times the Fisher information in a single observation.
Example 61.2
Let X1 , X2 , · · · , Xn be a random sample from N (µ, σ 2 ) where σ 2 is known
but µ is unknown. Find the Fisher information of this random sample.
Solution.
We have
n
I(µ) = nI(X|µ) =
σ2
Now, let θ̂ denote an arbitrary estimator of θ. It is proven in statistics that
m0 (θ)
Var(θ̂) ≥
I(θ)
where E(θ̂) = m(θ). If θ̂ is unbiased then m(θ) = θ and the above inequality
becomes
1
Var(θ̂) ≥ .
I(θ)
The right-hand side of the inequality is known as the Cramér-Rao lower
bound. It is shown in Statistics that under certain conditions, no other
unbiased estimator of the parameter θ based on an i.i.d. sample of size n
can have a variance smaller than the Cramér-Rao lower bound.
Example 61.3
Let X1 , X2 , · · · , Xn be a random sample from N (µ, θ) where µ is known and
θ is unknown. Calculate the Cramér-Rao lower bound of variance for any
unbiased estimator,
Solution.
The pdf of N (µ, θ) is given by
1 (x−µ)2
f (x|θ) = √ e− 2θ .
2πθ
Thus,
(x − µ)2 1 1
`(x|θ) = − − ln (2π) − ln θ
2θ 2 2
and
(x−µ)2 2
`0 (x|θ) = 2θ2
− 1
2θ and `00 (x|θ) = − (x−µ)
θ3
+ 1
2θ2
.
Hence,
1 1 1
I(X|θ) = −Eθ [`00 (θ)] = 3
E[(X − µ)2 ] − 2 = 2
θ 2θ 2θ
and
n
I(θ) = nI(X|θ) = .
2θ2
2θ2
The Cramér-Rao lower bound is n
The following result whose proof is omitted is a direct application of the

Central Limit Theorem.
Theorem 61.1
Let X1 , X2 , · · · , Xn be a random sample fron a distribution with pdf f (x|θ)
and unknown parameter θ. Let θ denote the true value of the parameter and
p
θ̂ the MLE estimator of θ. Then the probability distribution of I(θ)(θ̂ − θ)
approaches the standard normal distribution as n → ∞. That is,

1
θ̂ ∼ N θ, .
I(θ)
Since this is merely a limiting result, which holds as the sample size tends to
infinity, we say that the MLE is asymptotically unbiased and refer to the
variance of the limiting normal distribution as the asymptotic variance
of the MLE.
Example 61.4
Let X1 , X2 , · · · , Xn be a random sample from an exponential distribution
with mean θ. Find the asymptotic distribution of θ̂.
Solution.
We have
x
`(x|θ) = − − ln θ
θ
x 1
`0 (x|θ) = 2 −
θ θ
00 2x 1
` (x|θ) = − 3 + 2
θ θ
2X 1
I(x|θ) = − E[`00 (x|θ)] = E[
3
− 2
θ θ
2 1 1
= 3 E(X) − 2 = 2
θ θ θ
n
I(θ) = 2 .
θ
Hence,
θ2

θ̂ ∼ N θ,
n
It follows that
θ2
Var(θ̂) =
n
and since we don’t know the exact value θ, we have
2
d θ̂) = θ̂ .
Var(
n
Example 61.5
For an exponential distribution, the parameter θ is estimated via the method
of maximum likelihood by analyzing data from the following sample:
7 12 15 19.
Find the estimated variance of the parameter estimate.
Solution.
From Example 59.3, we know that θ̂ = X. Hence,
2 2 2
d θ̂) = θ̂ = X = [(7 + 12 + 15 + 19)/4] = 43.890625
Var(
n n 4
Since θ̂ is approximately normal for large sampling distribution, we can find

confidence intervals for the true value θ. A 100(1 − α)% confidence interval
for θ is given by
q q
θ̂ − z α2 Var(θ̂), θ̂ + z α2 Var(θ̂) .
d d
Example 61.6
For an exponential distribution, the parameter θ is estimated via the method
of maximum likelihood by analyzing data from the following sample:
7 12 15 19.
Find the 95% confidence interval for θ.
Solution.
We have θ̂ = X = 7+12+15+19
4 = 13.25 and Var(
d θ̂) = 43.890625. Hence, the
95% confidence interval is
√ √
[13.25 − 1.96 43.890625, 13.25 + 1.96 43.890625] = [0.265, 26.2350]
Practice Problems
Problem 61.1
Let X have a Pareto distribution with parameter α and θ = 20. The pa-
rameter α is estimated via the method of maximum likelihood by analyzing
data from the following sample:
12 15 17 19.
Find `0 (α)2 , the square of the first partial derivative of the loglikelihood
function.
Problem 61.2
12 15 17 19.
Find `00 (α), the second partial derivative of the loglikelihood function.
Problem 61.3
12 15 17 19.
Find the Fisher information associated with the maximum likelihood esti-
mator.
Problem 61.4
12 15 17 19.
Estimate the asymptotic variance of the maximum likelihood estimator.
Problem 61.5
12 15 17 19.
Find the 95% confidence interval for α.
Problem 61.6 ‡
The information associated with the maximum likelihood estimator of a
parameter θ is 4n, where n is the number of observations. Calculate the
asymptotic variance of the maximum likelihood estimator of 2θ.
Problem 61.7 ‡
You fit an exponential distribution to the following data:
1000 1400 5300 7400 7600
Determine the coefficient of variation of the maximum likelihood estimate

of the mean, θ.
Problem 61.8 ‡
A random sample of size n is drawn from a distribution with probability
density function:
θ
f (x) = < 0 < x < ∞, 0 < θ < ∞.
(θ + x)2
Determine the asymptotic variance of the maximum likelihood estimator of

θ.
Problem 61.9 ‡
You are given:
(i) The distribution of the number of claims per policy during a one-year
period for a block of 3000 insurance policies:
# of claims per policy # of policies
0 1000
1 1200
2 600
3 200
4+ 0
(ii) You fit Poisson model to the number of claims per policy using the
method maximum likelihood.
(iii) You construct the large-sample 90% confidence interval for the mean of
the underlying Poisson model that is symmetric around the mean.
Determine the lower end-point of the confidence interval.

62 INFORMATION MATRIX AND THE DELTA METHOD 463
62 Information Matrix and the Delta Method

In this section, we consider the Fisher information for a multiple parameter
vector θ = (θ1 , θ2 , · · · , θn )T . Each θi has an asymptotic normal distribution.
In this case, the Fisher information is a n × n square matrix I(θ) whose ijth
entry is given by
∂2

∂ ∂
I(θ)ij = E `(θ) `(θ) = −E `(θ) .
∂θi ∂θj ∂θj ∂θi
We refer to I(θ) as the Fisher information matrix.
Example 62.1
Find the Fisher information matrix of the MLE for the lognormal distribu-
tion.
Solution.
Recall the pdf of the lognormal distribution
1 (ln x−µ)2
f (x) = √ e− 2σ 2 .
xσ 2π
Thus, the likelihood function is
n (ln xi −µ)2
1
e−
Y
L(µ, σ) = √ 2σ 2
i=1
xi σ 2π
and its loglikelihood function is
n
" #
1 ln xi − µ 2

X 1
`(µ, σ) = − ln xi − ln σ − ln (2π) − .
2 2 σ
i=1
Taking first derivatives, we find
n n
∂`
X ln xi − µ ∂`
X (ln xi − µ)2
∂µ = and ∂σ = − nσ + .
σ2 σ3
i=1 i=1
Taking second derivatives, we find
∂2` n
2
=− 2
∂µ σ
2 n
∂ ` X ln xi − µ
=−2
∂σ∂µ σ3
i=1
n
∂2` n X (ln xi − µ)2
= − 3 .
∂σ 2 σ 2 σ4
i=1
Taking expectations, and recalling that ln Xi ∼ N (µ, σ 2 ), we have

2
∂ ` n
E 2
=− 2
∂µ σ
2
∂ `
E =0
∂σ∂µ
2
∂ ` 2n
E 2
=− 2.
∂σ σ
Hence, the Fisher information matrix is

n
0
I(θ) = σ2 2n
0 σ2
The inverse matrix I(θ)−1 , referred to as the covariance matrix, has the
variance of the individual random variables on the main diagonal and co-
variances in the off-diagonal positions.
Example 62.2
Find the covariance matrix of Example 62.1.
Solution.
The covariance matrix is
" #
n σ2

−1 1 σ2
0 n 0
I(θ) = 2n = σ2
det[I(θ)] 0 σ2 0 2n
In many cases, taking both the derivatives of the loglikelihood function and
the corresponding expectations are not always easy. A way to avoid this
problem is to simply not take the expected value. So instead of taking
the values that result from the expectation, we can just use the observed
data points. This method will result in the observed information. We
illustrate this concept in the next example.
Example 62.3
You model a loss process using a lognormal distribution with parameters µ
and σ. You are given:
(i) The maximum likelihood estimates of µ and σ are µ̂ = 4.4654 and σ̂ 2 =
0.3842.
(ii) The following five observations: 27 82 115 126 155.
Estimate the covariance matrix using the observed information.
Solution.
Substituting the observations into the second derivatives, we find
∂2` n 5
2
=− 2 =− 2
∂µ σ σ
2 n
∂ ` X ln xi − µ 22.3272 − 5µ
=−2 3
= −2
∂σ∂µ σ σ3
i=1
n
∂2` n X (ln xi − µ)2 5 101.6219 − 44.6544µ + 5µ2
= − 3 = − 3 .
∂σ 2 σ 2 σ4 σ2 σ4
i=1
Using µ̂ = 4.4654 and σ̂ 2 = 0.3842 in the previous expressions, we find

13.0141 0
I(θ) =
0 26.0307
and
d σ̂) = [I(θ)]−1 = 0.0768 0
Var(µ̂,
0 0.0384
The Delta Method

The delta method is a method for estimating the variance of a function of
estimators. The method is based on Taylor series expansions.
Let X be a distribution depending on a parameter θ. Suppose f (x) is a

differentiable function of x. The Taylor approximation of order one around
θ is
f (θ̂)) ≈ f (θ) + (θ̂ − θ)f 0 (θ).
Hence,
Var[f (θ̂)] ≈ [f 0 (θ)]2 Var(θ̂).
Example 62.4
Consider a random sample of size n from an exponential distribution with
mean θ.
(a) Find the variance of the estimated distribution variance.
(b) Construct the approximate 95% confidence interval for Pr(X > m).
Solution.
(a) We have Var(X)
d ≈ θ̂2 . Hence,
1 θ2 4θ4
Var(θ̂2 ) = ([θ2 ]0 )2 Var(θ̂) = 4θ2 · = 4θ2 · = .
I(θ) n n
m
(b) The estimated probability is e− θ̂ . Thus,
2 2m 2 2
d e− θ̂ = m e− θ̂ θ̂ = m e− θ̂ .
h mi 2m
Var
θ̂ 4 n nθ̂ 2
An approximate 95% confidence interval is

m m m
e− θ̂ ± 1.96 √ e− θ̂
nθ̂
This method can be extended to a pair of related maximum likelihood es-
timates. Let X be a distribution depending on two parameters θ1 and θ2 .
Let f (s, t) be a differentiable function in the variables s and t. The delta
method asserts that

d (θ̂1 , θ̂2 )] = fθ fθ
Var( θ̂ 1 ) Cov( θ̂ 1 , θ̂ 2 ) fθ1
Var[f 1 2 .
Cov(θ̂1 , θ̂2 ) Var(θ̂2 ) fθ2
Example 62.5 ‡
You model a loss process using a lognormal distribution with parameters µ
and σ. You are given:
(i) The maximum likelihood estimates of µ and σ are µ̂ = 4.215 and σ̂ =
1.093.
(ii) The estimated covariance matrix of µ̂ and σ̂ is

0.1195 0
0 0.0597
σ2
(iii) The mean of the lognormal distribution is eµ+ 2 .
Estimate the variance of the maximum likelihood estimate of the mean of
the lognormal distribution, using the delta method.
Solution.
σ2
Let f (µ, σ) = eµ+ 2 . We have
σ̂ 2
" #" #
σ̂ 2
h
µ̂+ σ̂2
2
µ̂+ σ̂2
2
i
n 0 eµ̂+ 2
Var[f
d (µ̂, σ̂)] = e σ̂e σ̂ 2 σ̂ 2
0 2n σ̂eµ̂+ 2

0.1195 0 123.02
= 123.02 134.46
0 0.0597 134.46

123.02
= 14.70089 8.027262
134.46
=2887.85
Example 62.6 ‡
You are given:
(i) Fifty claims have been observed from a lognormal distribution with un-
known parameters µ and σ.
(ii) The maximum likelihood estimates are µ̂ = 6.84 and σ̂ = 1.49.
(iii) The covariance matrix of µ̂ and σ̂ is

0.0444 0
0 0.0222
(iv) The partial derivatives of the lognormal cumulative distribution function
are:
∂F
∂µ = − φ(z)
σ and ∂F
∂σ = − zφ(z)
σ .
z2
where φ(z) = √12π e− 2 .
(v) An approximate 95% confidence interval for the probability that the
next claim will be less than or equal to 5000 is: [PL , PH ].
Determine PL .
Solution.
Let
ln5000 − µ
F (µ, σ) = Pr(X ≤ 5000) = Φ .
σ
The point estimate is

ln5000 − 6.84
F̂ (6.84, 1.49) = Φ = Φ(1.125) = 0.87.
1.49
For the delta method, we need
∂F φ(1.125)
=− = −0.1422
∂µ 1.49
∂F 1.125φ(1.125)
=− = −0.16.
∂σ 1.49
Thus,

0.0444 0 −0.1422
Var[F
d (µ̂, σ̂)] = −0.1422 −0.16
0 0.0222 −0.16
=0.001466.
The lower limit of the 95% confidence interval is
√
PL = 0.87 − 1.96 0.001466 = 0.79496
Practice Problems
Problem 62.1
For a Pareto distribution, the parameters α and θ are both estimated via the
method of maximum likelihood by analyzing data from a random sample of
n observations.
Find the loglikelihood function for this distribution.
Problem 62.2
n observations.
Find the second partial derivatives of the loglikelihood function.
Problem 62.3
n observations.
Find the information matrix.
Problem 62.4
n observations.
Find the covariance matrix.
Problem 62.5
You model a loss process using a Pareto distribution with parameters α and
θ = 10. You are given:
(i) The maximum likelihood estimate of α : α̂ = 1.26.
(ii) The following eight observations: 3 4 8 10 12 18 22 35.
Estimate I(α, θ) using the observed information.
Problem 62.6
You model a loss process using a Pareto distribution with parameters α and
θ = 10. You are given:
(i) The maximum likelihood estimate of α : α̂ = 1.26.

(ii) The following eight observations: 3 4 8 10 12 18 22 35.
Estimate the covariance matrix using the observed information.
Problem 62.7 ‡
You are given:
(i) Loss payments for a group health policy follow an exponential distribu-
tion with unknown mean.
(ii) A sample of losses is: 100 200 400 800 1400 3100.
Use the delta method to approximate the variance of the maximum like-
lihood estimator of S(1500).
Problem 62.8 ‡
The time to an accident follows an exponential distribution. A random sam-
ple of size two has a mean time of 6. Let Y denote the mean of a new sample
of size two.
(a) Using moment generating functions, show that the sum of two inde-
pendent exponential random variables is a Gamma random variables.
(b) Determine the maximum likelihood estimate of Pr(Y > 10).
(c) Use the delta method to approximate the variance of the maximum like-
lihood estimator of FY (10).
Problem 62.9 ‡
A survival study gave (0.283, 1.267) as the symmetric linear 95% confidence
interval for H(5).
Using the delta method, determine the symmetric linear 95% confidence
interval for S(5).
Problem 62.10 ‡
You have modeled eight loss ratios as Yt = α + βt + t , t = 1, 2, · · · , 8, where
Yt is the loss ratio for year t and t is an error term.
You have determined:
α̂ 0.5
=
β̂ 0.02

α̂ 0.00055 −0.00010
Var =
β̂ −0.00010 0.00002
Estimate the standard deviation of the forecast for year 10, Ŷ10 = α̂ + 10β̂,
using the delta method.
Problem 62.11 ‡
A sample of ten observations comes from a parametric family f (x, y, θ1 , θ2 )
with loglikelihood function
10
X
ln [L(θ1 , θ2 )] = ln [f (xi , yi , θ1 , θ2 )] = −2.5θ12 − 3θ1 θ2 − θ22 + 5θ1 + 2θ1 + k
i=1
where k is a constant.
Determine theestimated covariance matrix of the maximum likelihood es-

θˆ1

timator, .
θˆ2
63 NON-NORMAL CONFIDENCE INTERVALS FOR PARAMETER ESTIMATION471
63 Non-Normal Confidence Intervals for Parameter

Estimation
Up to this point, confidence intervals for the maximum likelihood estima-
tors have been based on the assumption that the estimators are normally
distributed. Such an assumption is true for large sampling but may not
be true for small or even moderate samples. In this section, we consider a
method for finding confidence intervals that does not require the normality
assumption.
We first recall the reader of the chi-square distribution. The chi-square

distribution with k degrees of freedom is the distribution of the aggregate
random variable
Q = Z12 + Z22 + · · · + Zk2
where the Zi0 s are standard normal distributions. The chi-square, denoted
by χ2 , is a one parameter distribution with parameter k.
A generalization of a confidence interval is a confidence region. We define

the 100(1 − α)% confidence region to be the set of values of a parameter
θ such that
1
`(θ) ≥ c = `(θ̂) − χ21−α
2
2
where χ1−α is the 1 − α percentile from the chi-square distribution with
degrees of freedom equal to the number of estimated parameters.
Example 63.1
Write the inequality that will result in the 95% non-normal confidence region
for the mean θ of the exponential distribution.
Solution.
n
X xi
`(θ) = −n ln θ − .
θ
i=1
The maximum likelihood estimator is

x1 + x2 + · · · + xn
θ̂ = x = .
n
Thus,
`(θ̂) = −n − n ln x.
Now, from the table of chi-square distribution, χ20.95 = 3.841. Hence, the
confidence region is
nx
−n ln θ − ≥ −n − n ln x − 1.9205
θ
which must be evaluated numerically
Remark 63.1
Since numerical methods are needed for determining most of the non-normal
confidence intervals and the exam candidates are not provided with the tools
(such as Excel) for solving these problems, it is very unlikely to see this type
of questions on the exam.
Example 63.2
Suppose that n = 20 and x = 1424.40 in the previous example. Use Excel’s
Solver in finding the confidence interval by solving the confidence region
28488
− − 20 ln θ ≥ −167.15.
θ
Solution.
Using Excel’s Solver, one solves the equation
−28488 − 20θ ln θ + 167.15θ = 0.
The two roots are 946.85 and 2285.05. Hence, the confidence interval is
[946.85, 2285.05]
63 NON-NORMAL CONFIDENCE INTERVALS FOR PARAMETER ESTIMATION473
Practice Problems
Problem 63.1
You are given:
• `(θ) = 5 ln θ − 2.55413(θ − 1).
• θ̂ = 1.95762.
Find the value of c to be used in the construction of a 95% non-normal

confidence region.
Problem 63.2
Use Excel’s Solver in finding the confidence interval by solving the confidence
region
5 ln θ − −2.55413(θ − 1) ≥ −1.0077.
Problem 63.3
You are given:
• `(θ) = −θ2 + 11θ − 24.
• θ̂ = 3.
Find a 95% non-normal confidence interval for θ.

64 Basics of Bayesian Inference

There are two approaches to parameter estimation. The frequentest ap-
proach assumes that there is an unknown but fixed parameter θ. It chooses
the value of θ that maximizes the likelihood of observed data. In other
words, making the available data as likely as possible. A common example
is the maximum likelihood estimator. In the frequentest approach, probabil-
ities are defined as the frequency of successful trials over the total number of
trials in an experiment. To elaborate, suppose that the parameter to be esti-
mated is the probability of getting a head when flipping a coin. we flip a fair
coin 100 times and it comes out head 30 times and tail 70 times. In this case,
we say that the probability of getting a head is 30% and that of a tail is 70%.
In contrast, the Bayesian approach allows probability to represent sub-

jective uncertainty or subjective belief. In the coin example, our subjective
belief tells us that if head and tail are equally likely to occur than the prob-
ability of each to occur must be 50%. This approach fixes the data and
instead assumes possible values of θ. In other words, Bayesians treat the
unknown model parameters as random variables and assign probabilities to
the subsets of the parameter space.
We next introduce some Bayesian inference terminology: A prior distri-

bution of a parameter θ is the probability distribution over the space of
possible values of the parameter. It is denoted by π(θ) and represents your
uncertainty about the parameter before the current data are examined.
A prior distribution π(θ) is said to be improper if π(θ) ≥ 0 and
Z
π(θ)dθ = ∞.
Example 64.1
Show that the uniform prior distribution on the real line π(θ) = 1 for −∞ <
θ < ∞ is improper.
Solution.
This follows from Z ∞
1dθ = ∞
−∞
Let the observable random variables X1 , X2 , · · · , Xn be i.i.d and denote

X = (X1 , X2 , · · · , Xn )T . The probability distribution of X given a particular
64 BASICS OF BAYESIAN INFERENCE 475
value of a parameter is called the model distribution and is given by

fX|Θ (x|θ) = fX|Θ (x1 |θ)fX|Θ (x2 |θ) · · · fX|Θ (xn |θ)
where x = (x1 , x2 , · · · , xn )T .
The joint distribution of X and Θ is given by

fX,Θ (x, θ) = fX|Θ (x|θ)π(θ).
The marginal distribution of X is given by
Z
fX (x) = fX|Θ (x|θ)π(θ)dθ.
The posterior distribution is the conditional probability distribution of

the parameters, given the observed data. It is denoted πΘ|X (θ|x) and is
given by Bayes Theorem
fX|Θ (x|θ)π(θ)
πΘ|X (θ|x) = .
fX (x)
The predictive distribution is the conditional probability distribution of
a new observation y, given the data x. It is denoted fY |X (y|X) and is given
by Z
fY |X (y|X) = fY |Θ (y|θ)πΘ|X (θ|x)dθ.
Example 64.2 ‡
You are given:
(i) A portfolio consists of 100 identically and independently distributed risks.
(ii) The number of claims for each risk follows a Poisson distribution with
mean λ.
(iii) The prior distribution of λ is:
(50λ)4 e−50λ
π(λ) = , λ > 0.
6λ
During Year 1, the following loss experience is observed:
# of claims # of risks
0 90
1 7
2 2
3 1
Total 100
(a) Find the model distribution.

(b) Find the joint density of X and Λ.
(c) Find the marginal density at x.
(d) Find the posterior distribution of Λ.
(e) Find the predictive distribution.
Solution.
(a) The model distribution is
90 7 2
e−λ λ0 e−λ λ1 e−λ λ2 e−λ λ3 e−100λ λ14

fX|Λ (x|λ) = = .
0! 1! 2! 3! 24
(b) The joint density is
e−100λ λ14 (50λ)4 e−50λ 390625 −150λ 17

fX,λ (x, λ) = = e λ .
24 6λ 9
(c) The marginal density of X is
λ 18
− 1

λ
390625
Z ∞
390625 Γ(18)
Z ∞ e 150 1
fX (x) = e−150λ λ17 dλ = 150
dλ
9 0 9 15018 0 λΓ(18)
390625 Γ(18)
= .
9 15018
(d) The posterior distribution of Λ is
15018 −150λ 17
πΛ|X (λ|x) = e λ .
17!
Note that the posterior distribution is a Gamma distribution with α = 18
1
and θ = 150 .
(e) The predictive distribution is
∞ ∞
e−λ λy 15018 −150λ 17 15018
Z Z
fY |X (y|x) = e λ dλ = λ17+y e−151λ dλ
0 y! 17! y!17! 0
18
150 (18 + y)!
=
y!17! 15118+y
Example 64.3 ‡
You are given:
(i) Losses on a company’s insurance policies follow a Pareto distribution

with probability density function:
θ
f (x|θ) = , x > 0.
(x + θ)2
(ii) For half of the company’s policies θ = 1, while for the other half θ = 3.
For a randomly selected policy, losses in Year 1 were 5. Determine the
posterior probability that losses for this policy in Year 2 will exceed 8.
Solution.
Let Xn denote the losses in Year n. We are asked to find Pr(X2 > 8|X1 = 5).
Recall from Section 3, that probabilities can be found by conditioning. Thus,
we can write
Pr(X2 > 8|X1 = 5) =Pr(X2 > 8|θ = 1)Pr(θ = 1|X1 = 5)

+Pr(X2 > 8|θ = 3)Pr(θ = 3|X1 = 5).
Since Θ|X is discrete, we have
f (θ = 1|X1 = 5)π(1)
Pr(θ = 1|X1 = 5) =π(θ = 1|x = 5) =
f (θ = 1|X1 = 5)π(1) + f (θ = 3|X1 = 5)π(3)
1 1
36 2 16
= 1 1 3 1 = .
36 2 + 64 2
43
Thus,
16 27
Pr(θ = 3|X1 = 5) = 1 − Pr(θ = 1|X1 = 5) = 1 − = .
43 43
On the other hand,
Z ∞
dx 1
Pr(X2 > 8|θ = 1) = = .
8 (x + 1)2 9
Likewise, Z ∞
3 3
Pr(X2 > 8|θ = 3) = 2
dx = .
8 (x + 3) 11
Finally,

1 16 3 27
Pr(X2 > 8|X1 = 5) = + = 0.2126
9 43 11 43
Example 64.4 ‡
You are given:
(i) Each risk has at most one claim each year.
(ii)
i Type of Risk(θi ) π(θi ) Annual claim probability

1 I 0.7 0.1
2 II 0.2 0.2
3 III 0.1 0.4
One randomly chosen risk has three claims during Years 1-6. Determine the
posterior probability of a claim for this risk in Year 7.
Solution.
Let Xn denote the number of claims in year n. Then Xn |Θ is a Bernoulli
X6
random vraible with probability of a claim (success) p. Let S = Xi .
i=1
Then S|Θ is a binomial distribution with 6 trials (number of years) and
probability of a claim p. Note that, S|T heta = 3 is the number of claims in
6 years. We are asked to find Pr(X7 = 1||(S = 3) for a given θ which by
Bayes theorem is
Pr[(X7 = 1) ∩ (S = 3)]
Pr(X7 = 1|S = 3) = .
Pr(S = 3)
We have, for a given θ,

6
Pr(S = 3|θ1 ) = (0.1)3 (0.9)3 = 0.01458
3

6
Pr(S = 3|θ2 ) = (0.2)3 (0.8)3 = 0.08192
3

6
Pr(S = 3|θ3 ) = (0.4)3 (0.6)3 = 0.27648
3
Pr(S = 3) =Pr(S = 3|θ1 )π(θ1 ) + Pr(S = 3|θ2 )π(θ2 ) + Pr(S = 3|θ3 )π(θ3 )
=0.01458(0.7) + 0.08192(0.2) + 0.27648(0.1) = 0.054238.
Now, (X7 = 1) ∩ (S = 3) is the intersection of two independent events that

depend on θ. Thus, we have
Pr[(X7 = 1) ∩ (S = 3)] =Pr[(X7 = 1) ∩ (S = 3)|θ1 ]π(θ1 )
+Pr[(X7 = 1) ∩ (S = 3)|θ2 ]π(θ2 )
+Pr[(X7 = 1) ∩ (S = 3)|θ3 ]π(θ3 ))
=Pr(X7 = 1|θ1 )Pr(S = 3|θ1 )π(θ1 )
+Pr(X7 = 1|θ2 )Pr(S = 3|θ2 )π(θ2 )
+Pr(X7 = 1|θ3 )Pr(S = 3|θ3 )π(θ3 )
=(0.1)(0.01458)(0.7) + (0.2)(0.08192)(0.2) + (0.4)(0.27648)(0.1)
=0.0153566.
Finally,
0.0153566
Pr(X7 = 1|S = 3) = = 0.283
0.054238
Example 64.5 ‡
You are given:
(i) For Q = q, the random variables X1 , X2 , · · · , Xm are independent, iden-
tically distributed Bernoulli random variables with parameter q.
(ii) Sm = X1 + X2 + · · · + Xm .
(iii) The prior distribution of Q is beta with a = 1, b = 99, and θ = 1.
Determine the smallest value of m such that the mean of the marginal dis-
tribution of Sm is greater than or equal to 50.
Solution.
Since X1 , X2 , · · · , Xm are independent and identically distributed with com-
mon distribution a Bernoulli distribution with parameter q, Sm |Q is a bino-
mial distribution with parameters (m, q). The mean of the marginal distri-
bution is (using the double expectation property, Theorem 78.2) is
m
E(Sm ) = E[E(SM |Q)] = E(mQ) = mE(Q) = = 0.01m.
1+b
The inequality E(Sm ) ≥ 50 implies m ≥ 5000
Example 64.6 ‡
You are given:
(i) The number of claims made by an individual in any given year has a
binomial distribution with parameters m = 4 and q.
(ii) The prior distribution of q has probability density function
π(q) = 6q(1 − q), 0 < q < 1.
(iii) Two claims are made in a given year. Determine the mode of the
posterior distribution of q.
Solution.
We have
f (2, q) =f (2|q)π(q) = 36q 3 (1 − q)3

Z 1
f (2) = 36q 3 (1 − q)3 dq
0
1 3
π(q|2) = q (1 − q)3 .
f (2)
The mode is the value of q that maximizes the posterior distribution. Taking
the derivative and setting it to zero, we find
3q 2 (1 − q)3 − 3q 3 (1 − q)2 = 0 =⇒ q = 0.5.
Moreover,
π 00 (2|0.5) = −6(0.5)4 /f (2) < 0
so that the posterior distribution is maximizex when q = 0.5. That is, the
mode of the posterior distribution is 0.5
Example 64.7 ‡
The observation from a single experiment has distribution:
Pr(D = d|G = g) = g (1−d) (1 − g)d , ford = 0, 1.
The prior distribution of G is:
Pr G = 51 = 3 1
= 25 .

5 and Pr G = 3
Calculate Pr G = 13 |D = 0 .

Solution.
We have
Pr D = 0|G = 13 Pr G = 13

1
Pr G = |D = 0 =
Pr D = 0|G = 13 Pr G = 13 + Pr D = 0|G = 15 Pr G = 15

3
(2/3)(1/5) 10
= =
(2/3)(1/5) + (1/5)(3/5) 19
Practice Problems
Problem 64.1
Show that the prior distribution π(θ) = 1θ , 0 < θ < ∞ is improper.
Problem 64.2 ‡
You are given:
(i) In a portfolio of risks, each policyholder can have at most two claims per
year.
(ii) For each year, the distribution of the number of claims is:
Number of claims Probability

0 0.10
1 0.90 − q
2 q
(iii) The prior density is:
q2
π(q) = , 0.2 < q < 0.5.
0.039
A randomly selected policyholder had two claims in Year 1 and two claims
in Year 2. For this insured, determine
(a) the model distribution
(b) the joint distribution
(c) the marginal distribution
(d) the posterior distribution.
Problem 64.3 ‡
You are given:
(i) The annual number of claims for a policyholder follows a Poisson distri-
bution with mean λ.
(ii) The prior distribution of Λ is Gamma with probability density function:
(2λ)5 e−2λ
f (λ) = , λ > 0.
24λ
An insured is selected at random and observed to have x1 = 5 claims during
Year 1 and x2 = 3 claims during Year 2.
Determine the posterior distribution.

Problem 64.4 ‡
You are given:
(i) Annual claim frequencies follow a Poisson distribution with mean λ.
(ii) The prior distribution of Λ has probability density function:
0.4 − λ 0.6 − λ
π(λ) = e 6 + e 12 , λ > 0.
6 12
Ten claims are observed for an insured in Year 1.
Determine the posterior distribution.
Problem 64.5 ‡
You are given:
(i) In a portfolio of risks, each policyholder can have at most one claim per
year.
(ii) The probability of a claim for a policyholder during a year is q.
q3
(iii) The prior density is π(q) = 0.07 0.6 < q < 0.8.
A randomly selected policyholder has one claim in Year 1 and zero claims
in Year 2.
For this policyholder, determine the posterior probability that 0.7 < q < 0.8.
Problem 64.6 ‡
You are given:
(i) The probability that an insured will have exactly one claim is θ.
(ii) The prior distribution of Θ has probability density function:
√
π(θ) = 1.5 θ, 0 < θ < 1.
A randomly chosen insured is observed to have exactly one claim.
Determine the posterior probability that Θ is greater than 0.60.
Problem 64.7 ‡
You are given:
(i) The prior distribution of the parameter Θ has probability density func-
tion:
1
π(θ) = 2 , θ > 1.
θ
Given Θ = θ, claim sizes follow a Pareto distribution with parameters α = 2
and θ.
A claim of 3 is observed.
Calculate the posterior probability that Θ exceeds 2.
Problem 64.8 ‡
You are given:
(i) The number of claims observed in a 1-year period has a Poisson distri-
bution with mean θ.
(ii) The prior density is:
e−θ
π(θ) = , 0 < θ < k.
1 − e−k
(iii) The unconditional probability of observing zero claims in 1 year is 0.575.
Determine k.
Problem 64.9 ‡
You are given: (i) Conditionally, given β, an individual loss X follows the
exponential distribution with probability density function:
1 − βx
f (x|β) = e , 0 < x < ∞.
β
(ii) The prior distribution of β is inverse gamma with probability density
function:
c2 − c
π(β) = 3 e β , 0 < β < ∞.
β
R ∞ 1 −a
(iii) 0 yn e y dy = (n−2)!
an−1
, n = 2, 3, 4, · · · .
Given that the observed loss is x, calculate the mean of the posterior distri-
bution of β.
Problem 64.10 ‡
You are given:
(i) Annual claim counts follow a Poisson distribution with mean λ.
(ii) The parameter λ has a prior distribution with probability density func-
tion:
1 λ
f (λ) = e− 3 .
3
Two claims were observed during the first year.
Determine the variance of the posterior distribution of λ.

Problem 64.11 ‡
You are given:
(i) For Q = q, the random variables X1 , X2 , · · · , Xm are independent, iden-
tically distributed Bernoulli random variables with parameter q.
(ii) Sm = X1 + X2 + · · · + Xm .
(iii) The prior distribution of Q is beta with a = 1, b = 99, and θ = 1.
Determine the variance of the marginal distribution of S101 .
Problem 64.12 ‡
You are given the following information about workers compensation cover-
age:
(i) The number of claims for an employee during the year follows a Poisson
distribution with mean (100 − p)/100, where p is the salary (in thousands)
for the employee.
(ii) The distribution of p is uniform on the interval (0, 100].
An employee is selected at random. No claims were observed for this em-
ployee during the year.
Determine the posterior probability that the selected employee has salary
greater than 50 thousand.
Problem 64.13 ‡
Prior to observing any claims, you believed that claim sizes followed a Pareto
distribution with parameters θ = 10 and α = 1, 2 or 3 , with each value
being equally likely.
You then observe one claim of 20 for a randomly selected risk.
Determine the posterior probability that the next claim for this risk will
be greater than 30.
65 BAYESIAN PARAMETER ESTIMATION 485
65 Bayesian Parameter Estimation

Among the objectives of Bayesian inference is the estimation of a model
parameter. In this section, we consider the approach discussed in [1].
A Bayes estimate of a parameter θ is that value that minimizes the ex-

pected loss function given the posterior distribution of θ. The three most
commonly used loss functions are:
• Squared-error loss : L(θ̂, θ) = (θ̂ − θ)2 .

• Absolute loss: L(θ̂, θ) = |θ̂ − θ|.
• Zero-one loss:
0, θ̂ = θ,
L(θ̂, θ) =
1, otherwise
The following theorem provides the Bayes estimate of each of the above loss
functions.
Theorem 65.1
For squared-error loss, the Bayes estimate is the mean of the posterior dis-
tribution; for absolute loss, it is a median; for zero-one loss, it is a mode.
Example 65.1
1 − xθ
The posterior distribution of θ is given by πΘ|X (θ|x) = xe . Determine
the Bayes estimate of θ using
(a) the squared-error loss function;
(b) the absolute loss function;
(c) the zero-one loss function.
Solution.
(a) θ̂ = x.
(b) The median of the posterior distribution is the number M such that
R M 1 −θ
0 xe
x dθ = 0.5. Solving this equation, we find M = x ln 2. Thus, θ̂ =
x ln 2.
(c) The value that maximizes the posterior distribution is θ = 0. That is,
the mode of the posterior distribution is 0. Hence, θ̂ = 0
The expected value of the predictive distribution provides a point-estimate

of the (n + 1)st observation if the first n observations and the prior distri-
bution are known. The expected value of the predictive distribution can be
expressed as follows:
Z Z Z
E(Y |x) = yfY |X (y|x)dy = y fY |Θ (y|θ)πΘ|X (θ|x)dθdy
Z Z Z
= πΘ|X (θ|x) yfY |Θ (y|θ)dydθ = E(Y |θ)πΘ|X (θ|x)dθ.
Example 65.2 ‡
You are given:
(i) The annual number of claims for a policyholder has a binomial distribu-
tion with probability function:

2
p(x|q) = q x (1 − q)2−x , x = 0, 1, 2.
x
(ii) The prior distribution is:
π(q) = 4q 3 , 0 < q < 1.
This policyholder had one claim in each of Years 1 and 2. Determine the
Bayesian estimate of the number of claims in Year 3.
Solution.
Let X be the previously observed data and let Y be the number of claims
in Year 3. Then E(Y |q) = nq = 2q. The joint distribution is
fX,Q (x, q) = 4q 2 (1 − q)2 (4q 3 ) = 16q 5 (1 − q)2 .
The marginal distribution is

Z 1
16
fx (x) = 16q 5 (1 − q)2 dq = .
0 168
The posterior distribution is
πQ|X (q|x) = 168q 5 (1 − q)2 .
Thus,
Z 1 Z 1
4
E(Y |x) = E(Y |q)πQ|X (q|x)dq = (2q)[168q 5 (1 − q)2 ]dq =
0 0 3
Example 65.3 ‡
For a group of insureds, you are given:
(i) The amount of a claim is uniformly distributed but will not exceed a
certain unknown limit θ.
(ii) The prior distribution of Θ is
500
π(θ) = , θ > 500.
θ2
Two independent claims of 400 and 600 are observed.
Determine the probability that the next claim will exceed 550.
Solution.
We are asked to find Pr(X3 > 550|X1 , X2 ). We have
Z ∞ Z ∞Z
Pr(X3 > 550|X1 , X2 ) = fX3 |X (x3 |x)dx3 = fX3 |Θ (x3 |θ)πΘ|X (θ|x)dθdx3
Z550 Z 550 Z
= πΘ|X (θ|x) fX3 |Θ (x3 |θ)dx3 dθ = Pr(X3 > 550|Θ)πΘ|X (θ|x)dθ.
550
Since X3 |Θ has a uniform distribution on (0, θ), we have

550 − θ
Pr(X3 > 550|Θ) = .
θ
The model distribution is
1
f (x|θ) = , θ > 600.
θ2
The joint distribution is
500
f (x|θ) = f (x|θ)π(θ) = , θ > 600.
θ4
Z ∞
500 1
f (x) = 4
dθ = .
600 θ 3(600)2
3(600)3
πΘ|X (θ|x) = .
θ4
Hence,
∞
3(600)3

550 − θ
Z
Pr(X3 > 550|X1 , X2 ) = dθ = 0.3125
600 θ θ4
A 100(1 − α)% credibility interval for θ is an interval [a, b] such that the
posterior probability
Pr(a ≤ Θ ≤ b|x) ≥ 1 − α.
It is possible to have many 100(1 − α)% credibility interval for a parameter

θ. The following result provides the shortest such interval.
Theorem 65.2
If the posterior random variable Θ|X is continuous and unimodal, then the
100(1 − α)% credibility interval with smallest width b − a is the unique
solution to the following system of equations:
Z b
πΘ|X (θ, x)dθ =1 − α
a
πΘ|X (a, x) =πΘ|X (b, x).
Example 65.4
You are given:
(i) The probability that an insured will have at least one loss during any
year is p.
(ii) The prior distribution for p is uniform on [0, 0.5].
(iii) An insured is observed for 8 years and has at least one loss every year.
Develop a non-zero width 95% credibility interval for the posterior proba-
bility that the insured will have at least one loss during Year 9.
Solution.
In Problem 65.2, we find
πP |X (p, x) = 4608p8 and p = 0.45.
Thus, according to Theorem 65.2, one of the equations of the system that
we will need to solve is
Z b Z b
πΘ|X (θ, x)dθ = 1 − α =⇒ 4608p8 dp = 0.95.
a a
The other equation is
πΘ|X (a, x) = πΘ|X (b, x) =⇒ 4608a8 = 4608b8 .

The second equation can only be solved either such that a = b or that
a = −b. The solution a = b would give a width of zero. The solution a = −b
implies the following:
Z b
4608p8 dp = 1024b9 = 0.95 =⇒ b = 0.4603.
−b
Thus, the credibility interval is [−0.4603, 0.4603]
The interval in Theorem 65.2 is a special case of a highest posterior density

(HPD) credibility set: For a given posterior distribution, the 100(1 − α)%
HPD credibility set for a parameter θ is a set defined by
C = {θ : πΘ|X (θ, x) ≥ k}
where k is the largest number satisfying
Z
πΘ|X (θ, x)dθ = 1 − α.
C
Thus, C is the region with the highest posterior density. The HPD credible
interval may not be unique and may in fact be a union of intervals. Also, it
can be shown that the 100(1 − α)% is the shortest among all 100(1 − α)%
credible intervals. HPD regions are difficult to determine analytically and
we shall not pursue this topic any further.
Now, for large sampling, there is a version of the Central Limit Theorem
which we call the Bayesian Central Limit Theorem:
If π(θ) (the prior distribution) and fX|Θ (x|θ) (the model distribution) are
both twice differentiable in the elements of θ and other commonly satisfied
assumptions hold, then the posterior distribution of Θ given X = x is asymp-
totically normal.
Example 65.5
Redo Example 65.4 using the Bayesian Central Limit Theorem.
Solution.
Both the prior distribution function and the model distribution are twice
differentiable so that we can assume that the posterior distribution is ap-
proximetely normal. We have
E(p) =0.45
Var(p) =E(p2 ) − E(p)2
Z 0.5
= 4608p10 dp − 0.452 = 0.002.
0
Thus, the credibility interval is

√ √
[0.45 − 1.96 0.002, 0.45 + 1.96 0.002] = [0.3623, 0.5377]
Practice Problems
Problem 65.1
The true value of a parameter is θ = 50. Suppose that the Bayes estimate of
θ using the squared-error loss function is θ̂ = 52. Find the absolute value of
the difference between the squared-error loss function and the absolute loss
function.
Problem 65.2 ‡
You are given:
year is p.
Determine the posterior probability that the insured will have at least one
loss during Year 9.
Problem 65.3
You are given:
year is p.
Using the Bayesian Central Limit Theorem, estimate the posterior prob-
ability that p > 0.6.
Problem 65.4 ‡
You are given:
(i) The amount of a claim, X, is uniformly distributed on the interval [0, θ].
(ii) The prior density of θ is π(θ) = 500
θ2
, θ > 500.
Two claims, x1 = 400 and x2 = 600, are observed. You calculate the
posterior distribution as:
6003

πΘ|X (θ|x) = 3 , θ > 600.
θ4
Calculate E(X3 |X).
Problem 65.5 ‡
You are given:
(i) In a portfolio of risks, each policyholder can have at most two claims per
year.
(ii) For each year, the distribution of the number of claims is:
Number of claims Probability

0 0.10
1 0.90 − q
2 q
(iii) The prior density is:
q2
π(q) = , 0.2 < q < 0.5.
0.039
A randomly selected policyholder had two claims in Year 1 and two claims
in Year 2.
For this insured, determine the Bayesian estimate of the expected number
of claims in Year 3.
Problem 65.6 ‡
You are given:
(i) The annual number of claims for each policyholder follows a Poisson
distribution with mean θ.
(ii) The distribution of θ across all policyholders has probability density
function:
f (θ) = θe−θ , θ > 0.
R∞
(iii) 0 θe−nθ dθ = n12 .
A randomly selected policyholder is known to have had at least one claim
last year.
Determine the posterior probability that this same policyholder will have
at least one claim this year.
66 CONJUGATE PRIOR DISTRIBUTIONS 493
66 Conjugate Prior Distributions

In this section, we look at models where the prior and posterior distributions
belong to the same family of distributions.
If a prior distribution combined with a model distribution result in a pos-

terior distribution that belongs to the same family of functions as the prior
distribution (perhaps with different parameters) then we call the prior dis-
tribution a conjugate prior distribution.
Example 66.1
You are given:
• The prior distribution Θ is Gamma with parameters α and β.
• The model distribution is exponential with parameter θ.
Show that Θ has a conjugate prior distribution.
Solution.
Pn
fX|Θ (x|θ) = θn e−θ i=1 xi
.

θ
α−1 e− β
xi θ
Pn
n −θ
fX,Θ (x, θ) = θ e i=1 .
β α Γ(α)

P
n
∞ −θ x +1
θn+α−1 e i=1 i β
Z
fX (x) = dθ
0 β α Γ(α)
P −(n+α)
n 1
i=1 xi + β Γ(n + α)
=
βα Γ(α)
θ
− P −1
Z ∞ n+α−1 ( n xi + 1 )
θ e i=1 β
× P −(n+α) dθ
n
0
+ β1
i=1 xi Γ(α + n)
P −(n+α)
n 1
i=1 x i + β Γ(n + α)
= α
.
β Γ(α)

θ
− −1
( ni=1 xi + β1 )
P
θn+α−1 e
πΘ|X (θ, x) = −(n+α) .
Pn 1
x
i=1 i + β Γ(α + n)
Pn −1
1
Hence, Θ|X ∼ Γ(α + n, β + i=1 xi )
Example 66.2
You are given:
• The prior distribution Λ is Gamma with parameters α and θ.
• The model distribution X|Λ is Poisson with parameter λ.
Show that Λ has a conjugate pair distribution.
Solution.
Pn
e−nλ λ i=1 xi
fX|Λ (x|λ) = Qn .
i=1 xi !

Pn λ
e−nλ λ i=1 xi λα−1 e− θ
fX,Λ (x, λ) = Qn .
i=1 xi ! θα Γ(α)
Pn
xi +α−1 e−λ(n+ θ1 )
Pni=1 xi +α
∞
Γ( ni=1 xi + α)
Z P
λ i=1 θ
fX (x) = Qn dλ = Qn .
0 θα Γ(α) i=1 xi ! nθ + 1 α
i=1 xi !θ Γ(α)

Pn λ(nθ+1)
λ i=1 xi +α−1 e− θ
πΛ|X (λ, x) = Pni=1 xi +α P .
θ n
nθ+1 Γ( i=1 xi + α)
P
n θ
Hence, Λ|X ∼ Gamma i=1 xi + α, nθ+1
Example 66.3
You are given:
• The prior distribution Λ is normal with mean µ and variance a2 .
• The model distribution X|Λ is normal with mean λ and variance σ 2 .
66 CONJUGATE PRIOR DISTRIBUTIONS 495
Solution.
Pn 2
1 i=1 (xi −λ)
fX|Λ (x|λ) = p e− 2σ 2 .
σ n (2π)n

Pn 2
1 i=1 (xi −λ) 1 (λ−µ)2
fX,Λ (x, λ) = p e− 2σ 2 √ e− 2a2 .
σ n (2π)n a 2π
The joint distribution can be simplified as follows:
Pn 2
(λ − µ)2
P 2
xi − 2λ xi + nλ2 λ2 − 2λµ + µ2
P
i=1 (xi − λ)
+ = +
σ2 a2 σ2 a2 P
x2i µ2
P
n 1 2 xi µ
= + λ − 2λ + + +
σ 2 a2 σ2 a2 σ2 a2
"
1 −1
P
n 1 2 xi µ n
= + λ − 2λ + 2 +
σ 2 a2 σ2 a σ 2 a2
#
1 −1
P 2
µ2

xi n
+ + 2 +
σ2 a σ 2 a2
2
n + 1 −1
P
xi µ
λ−
σ2
+ 2
a σ 2 a2
( )
−1 P
xi
2 P 2
xi 2 − −1
1 n
+ 12 + µ2 − − µ2 2 n2 + 12
( )
e 2 σ2 a σ2 a 2σ 2 2a e σ a
= √
n 1 1/2 n 1 −1/2
p
σ2
+ a2
aσ (2π)n 2π σ2
+ a2

−1 P 2 P 2

xi xi 2
∞
1 n
+ 12 + µ2 − − µ2
σ2 σ2 2σ 2
Z
e 2 a a 2a
fX (x) = fX,Λ (x, λ)dλ = .
n 1 1/2
p
0
σ2
+ a2 aσ (2π)n

2
n + 1 −1
P
xi µ
λ−
σ2
+ 2
a σ 2 a2
( )
− −1
2 n2 + 12( )
e σ a
πΛ|X (λ, x) = √ .
n 1 −1/2

2π σ2
+ a2
P
xi µ n 1 −1

Thus, Λ|X has a normal distribution with mean σ2
+ a2 σ2
+ a2
−1
and variance σn2 + a12
Practice Problems
Problem 66.1
You are given:
• The prior distribution Q is beta with parameters a, b and 1.
• The model distribution X|Q is binomial with parameters m and q.
Show that Q has a conjugate pair distribution.
Problem 66.2
You are given:
• The prior distribution Λ is Gamma with parameters α and θ.
• The model distribution X|Λ is inverse exponential with parameters λ.
Problem 66.3
You are given:
• The prior distribution Λ is inverse Gamma with parameters α and θ.
• The model distribution X|Λ is exponential with mean λ.
Problem 66.4
You are given:
• The prior distribution Λ is a single parameter Pareto with parameter α
α
and with pdf f (λ) = λαθ
α+1 , λ > θ.
• The model distribution X|Λ is uniform in [0, λ].

67 ESTIMATION OF CLASS (A, B, 0) 497
67 Estimation of Class (a, b, 0)

In this section we use the methods of moments and the method of maximum
likely estimate for the estimation of parmeters in (a, b, 0) class.
Let N be a random variable in the class of (a, b, 0) with probability function

pk = Pr(N = k), k = 0, 1, 2, · · · . Let nk be the number of observations for
which N = k and n be the total number of observations or the sample size.
X∞
Note that n = nk .
k=0
Example 67.1
Let N be a Poisson random variable with parameter λ. Then E(N ) = λ and
Var(N ) = λ.
(a) Estimate the Poisson parameter using the method of moments.
(b) Estimate the Poisson parameter using the method of maximum likeli-
hood.
(c) Calculate E(λ̂) and Var(λ̂).
(d) Find the asymptotic variance of λ̂.
(e) Construct a 95% confidence interval of the true value of λ.
Solution.
(a) The Poisson distribution parameter estimate by the method of moments
is
X∞
knk P∞
k=1 knk
λ̂ = x = ∞ = k=1 .
X n
nk
k=0
(b) The likelihood function is
∞ −λ k nk
Y e λ
L(λ) = .
k!
k=0

∞ ∞ ∞ ∞
e−λ λk
X X X X
`(λ) = nk ln = −λ nk + knk ln λ − nk ln (k!).
k!
k=0 k=0 k=1 k=0
The MLE of λ is found from

∞
X
∞
knk
d 1X k=1
`(λ) = −n + knk = 0 =⇒ λ̂ = .
dλ λ n
k=1
(c) We have
E(λ̂) =E(N ) = λ
Var(N ) λ
Var(λ̂) = = ,
n n
Thus, λ̂ is unbiased and consistent.

(d) The asymptotic variance is found as follows:
−λ N
∂2

e λ
I(λ) =nI(N |λ) = −nE 2
ln
∂λ N!

N n
=nE 2
=
λ λ
λ
Var(λ̂) = .
n
q q
(e) The confidence interval is (λ̂ − 1.96 Var(λ̂), λ̂ + 1.96 Var(λ̂))
The following example describes the process of finding the likelihood func-
tion in the case of incomplete data.
Example 67.2
Number of Accidents Number of Policies

0 31
1 20
2 12
3+ 7
Total 70
Let N denote the number of accidents. Suppose N ∼ Poisson(λ). Find the

likelihood function.
Solution.
The likelihood function is given by
L(λ) = p31 20 12
0 p1 p2 (1 − p0 − p1 − p2 )
7
where
e−λ λk
pk =
k!
Example 67.3
Estimate the negative binomial parameters by the method of moments.
Solution.
By the method of moments, we have the following system of two equations
∞
X
knk
k=1
rβ = =x
n
and 2
∞
X
 ∞
X
2
k nk  knk 
k=1  k=1  2
rβ(1 + β) = −
 n  =s .

n  
Solving this system for r and β we find

x s2
r̂ = and β̂ = x − 1.
β̂
Note that, if s2 < x then β̂ < 0 which is an indication that the negative
binomial model is not a good representation for the data
Example 67.4
Estimate the negative binomial parameters by the method of maximum
likelihood.
Solution.
Let r k
k+r−1 1 β
pk = , k = 0, 1, 2, · · · .
k 1+β 1+β
Y
L(r, β) = pk
k
and the loglikelihood function is

∞
X k+r−1
`(r, β) = nk ln − r ln (1 + β) + k ln β − k ln (1 + β) .
k
k=0
Taking first derivatives with respect to both r and β, we find

∞
∂` X k r+k
= nk −
∂β β 1+β
k=0
∞ ∞
∂` X X ∂ (r + k − 1) · · · r
=− nk ln (1 + β) + nk ln
∂r ∂r k!
k=0 k=0
∞ k−1
X ∂ Y
= − n ln (1 + β) + nk ln (r + m)
∂r
k=1 m=0
∞ k−1
X ∂ X
= − n ln (1 + β) + nk ln (r + m)
∂r
k=1 m=0
∞
X k−1
X
= − n ln (1 + β) + nk (r + m)−1 .
k=1 m=0
Setting `β (r, β) = 0, we find

∞
X
knk
k=0
r̂β̂ = =x
n
and setting `r (r, β) = 0 we find
∞ k−1
X X x
n ln (1 + β̂) = nk (r̂ + m)−1 = n ln 1 + .
r̂
k=1 m=0
The above equations are usually solved using numerical methods such as the
Newton-Raphson method. Note that if r is given then β̂ = xr
Example 67.5
Let N be a binomial random variable with parameters m and q. Then
E(N ) = mq and Var(N ) = mq(1 − q).
(a) Estimate q using the method of moments, assuming that m is known.
(b) Estimate q using the method of moments, assuming that m is unknown.
Solution.
(a) We have
∞
X ∞
X
knk knk
k=1 k=1
mq̂ = =⇒ q̂ = .
n nm
(b) We have to solve the system of two equations
∞ ∞  ∞ 2
X X X
2
knk k nk  knk 
k=1 k=1
 k=1 
m̂q̂ = n and m̂q̂(1 − q̂) = n −
 n  .

 
We obtain
∞
X ∞
X ∞
X
k 2 nk knk knk
k=1 k=1 k=1
q̂ = 1 − ∞ + n and m̂ = nq̂
X
knk
k=1
Example 67.6
Let N be a binomial random variable with parameters m and q. Then
E(N ) = mq and Var(N ) = mq(1 − q).
Estimate q using the method of maximum likelihood, assuming that m is
known.
Solution.
Let
m
pk = Pr(N = k) = q k (1 − q)m−k .
k
m
Y
L(m, q) = pnk k .
k=0
m
X
`(m, q) = nk ln pk
k=0
m
X m
= nk ln + k ln q + (m − k) ln (1 − q) .
k
k=0
Since m known, we just need to maximize `(m, q) with respect to q :

m m
∂` 1X 1 X
= knk − (m − k)nk .
∂q q 1−q
k=1 k=0
Setting this expression to 0, we obtain

∞
X
knk
k=1
q̂ =
nm
Remark 67.1
If m and q are both unknown, the estimation analysis is complex and hence
is omitted.
Example 67.7 ‡
You are given:
(i) A hospital liability policy has experienced the following numbers of claims
over a 10-year period:
10 2 4 0 6 2 4 5 4 2
(ii) Numbers of claims are independent from year to year.

(iii) You use the method of maximum likelihood to fit a Poisson model.
Determine the estimated coefficient of variation of the estimator of the Pois-
son parameter.
Solution.
Recall that
10 + 2 + 4 + 0 + 6 + 2 + 4 + 5 + 4 + 2
λ̂ = X = = 3.9.
10
We have
E(λ̂) =E(X) = λ
λ
Var(λ̂) =
n
q
λ
n 1
CV = =√
λ λn
1
= √ = 0.16
39
Practice Problems
Problem 67.1

0 32
1 26
2 12
3 7
4 4
5 2
6 1
Total 84
(a) Estimate the Poisson parameter using the method of moments.

(b) Estimate the Poisson parameter using the method of maximum likeli-
hood.
(c) Let N denote the number of accidents. Suppose N ∼ Poisson(λ). Cal-
culate E(λ̂) and Var(λ̂).
(d) Find the asymptotic variance of λ̂.
(e) Construct a 95% confidence interval of the true value of λ.
Problem 67.2 ‡
You are given the following observed claim frequency data collected over a
period of 365 days:
# of claims/day observed number of days

0 50
1 122
2 101
3 92
4+ 0
Fit a Poisson distribution to the above data, using the method of maximum
likelihood.
Problem 67.3
period of 365 days:
# of claims/day observed number of days

0 50
1 122
2 101
3+ 92
Find the likelihood function of λ.
Problem 67.4
You are given:
(i)
k 0 1 2 3 4
nk 30 35 20 10 5
(ii) The model is negative binomial.
Find the estimates of r and β using the method of moments.
Problem 67.5
You are given:
(i)
k 0 1 2 3 4
nk 30 35 20 10 5
(ii) The model is negative binomial.

Find the equation H(r̂) = 0, where r̂ = MLE(r).
Problem 67.6 ‡
The number of claims follows a negative binomial distribution with param-
eters β and r, where β is unknown and r is known. You wish to estimate β
based on n observations, where x is the mean of these observations.
Determine the maximum likelihood estimate of β.
Problem 67.7 ‡

0 6
1 17
2 26
3 21
4 18
5 7
6 5
Total 100
Estimate the binomial distribution parameters m and q using the method

of moments.
Problem 67.8 ‡
You are given the following data for the number of claims during a one-year
period:
Number of Claims Number of Policies

0 157
1 66
2 19
3 4
4 2
5+ 0
Total 248
A geometric distribution is fitted to the data using maximum likelihood es-

timation.
Let P = probability of zero claims using the fitted geometric model.
A Poisson distribution is fitted to the data using the method of moments.
Let Q = probability of zero claims using the fitted Poisson model.
Calculate |P − Q|.
68 MLE with (a, b, 1) Class

In this section15 , we estimate the parameters of the (a, b, 1) class. The esti-
mation procedure is similar to the one used for the (a, b, 0) class.
Let N be a zero-modified random variable with probability function pk =

Pr(N = k), k = 0, 1, · · · . Let nk be the number of observations for which
N = k and n be the total number of observations or the sample size. Recall
that the probabilities of N are given by
1 − pM
pM
k =
0
pk = (1 − pM T
0 )pk , k = 1, 2, 3, · · ·
1 − p0
where
b
pk = a+ pk−1 , k = 2, 3, · · · .
k
and pM M
0 = α, 0 ≤ α < 1. The parameters to be estimated are a, b, and p0 .

∞
Y ∞
Y
L(a, b, pM M n0
0 ) = (p0 ) (pM
k )
nk
= (pM
0 )
n0
[(1 − pM T nk
0 )pk ] .
k=1 k=1

∞
X
`(a, b, pM M
0 ) =n0 ln p0 + nk [ln (1 − pM T
0 ) + ln pk ]
k=1
∞
X ∞
X
=n0 ln pM
0 + nk ln (1 − pM
0 ) + nk [ln pk − ln (1 − p0 )] = `0 + `1
k=1 k=1
where
∞
X
`0 =n0 ln pM
0 + nk ln (1 − pM
0 )
k=1
∞
X
`1 = nk [ln pk − ln (1 − p0 )].
k=1
15
This section has not appeared in any of the C exams.
68 MLE WITH (A, B, 1) CLASS 507
The MLE of pM 0 is found by setting the first partial derivative of ` with

M
respect to p0 to zero:
∞
∂` ∂`0 n0 X nk n0 n − n0
M
= M
= M
− M
= M − =0
∂p0 ∂p0 p0 k=1
1 − p0 p0 1 − pM
0
resulting in
n0
pˆ0 M =
,
n
the proportion of observations that equal 0. This result is true for any
zero-modified distribution.
Example 68.1
Let pM0 be the zero-modified geometric probability function. Find the MLE
for pM
0 and β.
Solution.
Recall that
βk 1
pk = , p0 = .
(1 + β)k+1 1+β
The MLE of pM
0 is
n0
p̂M
.0 =
n
For finding the MLE of β, we first find `1 :
∞
X
`1 = nk [ln pk − ln (1 − p0 )]
k=0
∞
X
= nk [k ln β − (k + 1) ln (1 + β) − ln β − ln (1 + β)]
k=0
∞
X ∞
X
= ln β (knk − nk ) − ln (1 + β) [(k + 1)nk − nk ]
k=0 k=0
∞ ∞
∂`1 1 X 1 X
= nk (k − 1) − knk .
∂β β 1+β
k=0 k=0
Setting this last equation to zero, we find

P∞
knk nx
β̂ = k=0 −1= −1
n − n0 n − n0
Example 68.2
Find `1 for the zero-modified Poisson distribution.
Solution.
We have
∞ −λ k
X e λ
`1 = nk ln − ln (1 − e−λ )
k!
k=1
∞
!
X
= − (n − n0 )λ + knk ln λ − (n − n0 ) ln (1 − e−λ ) + c
k=1
= − (n − n0 )[λ + ln (1 − e−λ )] + nx ln λ + c
where
∞
X
c=− nk ln k!
k=1
and
∞
X
knk
k=1
x=
n
To find the estimate of λ in Example 68.2, we set the first derivative of `1
with respect to λ to zero:
∂`1 n − n0 nx
=− −λ
+ =0
∂λ 1−e λ
resulting in
n − n0
x(1 − e−λ ) = λ
n
which is solved numerically for λ. Note also that this last equation can be
expressed as
1 − p̂M
0
x= λ.
1 − p0
Example 68.3
Find `1 for the zero-modified binomial distribution.
Solution.
We have
m
X m k m−k m
`1 = nk ln q (1 − q) − ln [1 − (1 − q) ]
k
k=1
m ∞
!
X X
= knk ln q + (m − k)nk ln (1 − q)
k=1 k=1
m
X
− nk ln [1 − (1 − q)m ] + c
k=1
=nx ln q + m(n − n0 ) ln (1 − q) − nx ln (1 − q)
−(n − n0 ) ln [1 − (1 − q)m ] + c
where
m
X m
c= nk ln
k
k=1
The equation
∂` nx m(n − n0 ) nx (n − n0 )m(1 − q)m−1

0= = − + −
∂q q 1−q 1−q 1 − 1(1 − q)m
results in
1 − p̂M
0
x= mq
1 − p0
where p0 = (1 − q)m . If m is known then the MLE of q is obtained by solving
the above equation.
Practice Problems
Problem 68.1
You have the following observations of a discrete random variable.
Frequency (k) nk
0 9048
1 905
2 45
3 2
4+ 0
You are to fit these to a zero-modified geometric distribution using the

maximum likelihood. Find the fitted values of pM
0 and β.
Problem 68.2
Frequency (k) nk
0 9048
1 905
2 45
3 2
4+ 0
You are to fit these to a zero-modified Poisson distribution using the maxi-
mum likelihood. Find the MLE for pM 0 and λ.
Problem 68.3
Frequency (k) nk
0 10
1 6
2 4
You are to fit these to a zero-truncated binomial distribution using the

maximum likelihood. Find the MLE for pM 0 and q when m = 3.
Problem 68.4
Frequency (k) nk
0 10
1 6
2 4
You are to fit these to a zero-modified negative binomial distribution. Find

`1 when r = 2 and β = 3.
Model Selection and
Evaluation
The goal of this chapter is to evaluates models that fit the best to a given
data set and compare competing models.
513
514 MODEL SELECTION AND EVALUATION
69 Assessing Fitted Models Graphically

Given a sample of size n, we let Fn (x) and fn (x) denote the cumulative and
the density functions of the empirical distribution. The distribution and the
density functions of the estimated model to which we are trying to fit the
data are denoted by F ∗ (x) and f ∗ (x). In this section, we assess how well an
estimated model fit the original data graphically.
One method is to compare the plot of the empirical distribution to the

distribution of the estimated model. A good fit is when the plot of F ∗ (x)
alternates above and below the histogram of Fn (x).
One of the difficulties of the previous method is that the distinction is dif-
ficult to make when the heights between the two graphs is small. We con-
sider two ways for magnifying small changes to better interpret the good-
ness of fit. The first method consists of graphing the difference function
D(x) = Fn (x) − F ∗ (x), known as the D(x)−plot. A fit is considered good,
if the graph of D(x) is close to the horizontal axis.
The second method is to create a probability plot, also known as p-

p plot. The plot is created by ordering the observations in increasing
order x1 ≤ x2 ≤ · · · ≤ xn . For each j = 1, 2, · · · , we plot the point
(Fn (xj ), F ∗ (xj )). According to [1], a model is a good fit if these points are
close to the line y = x. This requires a new definition of Fn (xj ). If F denotes
the correct model and Xj is the j−th order statistic (i.e., the j−th smallest
j
value) then it can be shown that E[F (Xj )] = n+1 . Thus, it makes sense to
define
j
Fn (xj ) = .
n+1
Now, examining tails in the p-p plot: on the left-hand tail (near y = 0),
if a point in the p-p plot is above the line y = x, then F ∗ (x) > Fn (x) for
all x to the left of that point. That is, the fitted distribution is thicker on
the left than the empirical distribution. In probability terms, there is more
probability on the fitted than the empirical. On the right-end tail (near
y = 1), if a point in the p-p plot is above the line y = x, then F ∗ (x) > Fn (x)
for all x to the right of that point which implies 1 − F ∗ (x) < 1 − Fn (x) for all
x to the right of that point. That is, to the right of that point there is less
probability on the fitted than the empirical or in other words the fitted is
thinner than the empirical. What about points not near end tails? Taking
the j−th and the (j + k)−th points on the p − p plot, the slope of the line
69 ASSESSING FITTED MODELS GRAPHICALLY 515
crossing these points is

F (Xj+k ) − F (Xj )
k/n + 1
where F is the correct model, has an expected value of 1. So if this slope

is greater than 1, the model puts more probability between those two order
statistics. See Problem 69.2.
Example 69.1
You are given:
(i) The following observed data: 2, 3, 3, 3, 5, 8, 10, 13, 16.
(ii) An exponential distribution is fit to the data using the maximum likeli-
hood to estimate the mean of the exponential distribution.
(a) Plot F8 (x) and F ∗ (x) in the same window.
(b) Plot D(x).
(c) Create a p − p plot.
Solution.
The empirical mass function is
x 2 3 5 8 10 13 16
1 1 1 1 1 1 1
p(x) 9 3 9 9 9 9 9
The empirical cumulative distribution is


 0, x<2
 1


 9, 2≤x<3
4


 , 3≤x<5
 59


9 , 5≤x<8
F9 (x) = 2

 3, 8 ≤ x < 10
7
, 10 ≤ x < 13



 9
8
, 13 ≤ x < 16


 9


1, x ≥ 16.
By the maximum likelihood method, θ̂ = x = 7. The fitted distribution is

given by
x
F ∗ (x) = 1 − e− 7 .
(a) The plots of both F9 (x) and F ∗ (x) are shown in Figure 69.1.
Figure 69.1
The plot indicates that the fitted model is a reasonable one.
(b) The difference function is
 −x
 e 7 − 1, x<2
− x7

8
e − 9, 2 ≤ x < 3



−x

 5
 e x7 − 9 , 3 ≤ x < 5


 −7
e − 49 , 5 ≤ x < 8

D(x) = F9 (x) − F ∗ (x) = x
 e− 7 − 13 , 8 ≤ x < 10
x

e− 7 − 29 , 10 ≤ x < 13



 x
e− 7 − 19 , 13 ≤ x < 16




 x
e− 7 , x ≥ 16.

The graph of D(x) is shown in Figure 69.2.
Figure 69.2
(c) We first create the points on the graph
j
j xj F9 (xj ) = 9 F ∗ (xj )
1 2 0.1 0.249
2 3 0.2 0.349
3 3 0.3 0.349
4 3 0.4 0.349
5 5 0.5 0.510
6 8 0.6 0.681
7 10 0.7 0.760
8 13 0.8 0.844
9 15 0.9 0.883
The p − p plot is shown in Figure 69.3
Figure 69.3
Example 69.2
You are given:
(i) The following are observed claim amounts:
400 1000 1600 3000 5000 5400 6200.
(ii) An exponential distribution with θ = 3300 is hypothesized for the data.
(iii) The data are left-truncated at 500.
Write the formula for F ∗ (x) and f ∗ (x).
Solution.
The distribution function using a truncation of 500 is
(
∗
0, x < 500
F (x) = F (x)−F (500) (x−500)
− 3300
1−F (500) = 1 − e , x ≥ 500
and the pdf is

(
∗
0, x < 500
f (x) = f (x) 1 −
(x−500)
1−F (500) = 3300 e

3300 , x ≥ 500
Practice Problems
Problem 69.1 ‡
You are given:
400 1000 1600 3000 5000 5400 6200.

(iii) The goodness of fit is to be assessed by a p − p plot and a D(x) plot.
Let (s, t) be the coordinates of the p − p plot for a claim amount of 3000.
Determine (s − t) − D(3000).
Problem 69.2 ‡
The graph below shows a p − p plot of a fitted distribution compared to a
sample.
Which of the following is true?
(A) The tails of the fitted distribution are too thick on the left and on
the right, and the fitted distribution has less probability around the median
than the sample.
(B) The tails of the fitted distribution are too thick on the left and on the
right, and the fitted distribution has more probability around the median
than the sample.

(C) The tails of the fitted distribution are too thin on the left and on the
right, and the fitted distribution has less probability around the median than
the sample.
(D) The tails of the fitted distribution are too thin on the left and on the
right, and the fitted distribution has more probability around the median
than the sample.
(E) The tail of the fitted distribution is too thick on the left, too thin on
the right, and the fitted distribution has less probability around the median
than the sample.
Problem 69.3 ‡
You are given the following p − p plot:
The plot is based on the sample:
1 2 3 15 30 50 51 99 100
Which of the following is a possible fitted model underlying the p − p plot?
(A) F (x) = 1 − x−0.25 , x ≥ 1

x
(B) F (x) = 1+x ,x > 0
(C) Uniform on [1, 100]
(D) Exponential with mean 10 (E) Normal with mean 40 and standard de-
viation 40
70 KOLMOGOROV-SMIRNOV HYPOTHESIS TEST OF FITTED MODELS521
70 Kolmogorov-Smirnov Hypothesis Test of Fitted

Models
In this and the coming sections, we consider hypothesis tests of how well a
fitted model fits the data. As such, consider the following hypothesis test
problem:
H0 : The data comes from the estimated model

Ha : The data does not come from the estimated model
The test statistic is usually a measure of how close the fitted distribution
function is to the empirical distribution function. When the null hypothesis
completely specifies the model (i.e. the parameters of the fitted distribution
are given), critical values are well-known. In contrast, when the parameters
of the distribution function in H0 are to be estimated from the data, the test
statistic tends to be smaller than it would be have been had the parameter
values been prespecified. The estimation method tries to choose parameters
that produce a distribution that is close to the data and this decreases the
probability that the null hypothesis be rejected.
In this section, we look at the Kolmogorov-Smirnov statistic(KS) for

testing whether an empirical distribution fits a hypothesized distribution
well.
Consider a random sample of size n with order statistics x1 ≤ x2 ≤ · · · ≤ xn

so that x1 is the smallest observation and xn is the largest observation.
Let Fn (x) denote the empirical distribution and F ∗ (x) be the hypothesized
distribution. The Kolmogorov-Smirnov statistic is defined by
D= max |Fn (x) − F ∗ (x)|.

x1 ≤x≤xn
But Fn (x) is right-continuous and increasing step function and F ∗ (x) is

continuous and increasing so we only need to compare the differences at the
observed data points. Moreover, at each order statistic xi , one compares
the differences |Fn (xi−1 ) − F ∗ (xi )| and |Fn (xi+1 ) − F ∗ (xi )|. Hence, D can
be written as
D= max {|Fn (xi−1 ) − F ∗ (xi )|, |Fn (xi+1 ) − F ∗ (xi )|}

i=1,2,··· ,n
where Fn (x0 ) = 0.
Example 70.1
You are given:
200 400 1000 1600 3000 5000 5400 6200
Find the value of D.
Solution.
We create the following chart.
xj F8 (xj−1 ) F8 (xj ) F ∗ (xj ) Maximum of difference
200 0 0.125 0.05881 0.06619
400 0.125 0.25 0.1142 0.1358
1000 0.25 0.375 0.2614 0.1136
1600 0.375 0.5 0.3842 0.1158
3000 0.5 0.625 0.5971 0.0971
5000 0.625 0.75 0.7802 0.177
5400 0.75 0.875 0.8053 0.1552
6200 0.875 1.0 0.8472 0.1528
x
where F ∗ (x) = 1 − e− 3300 . Hence, D = 0.177
When F ∗ (x) is completely specified with no unknown parameters, the crit-

ical values of D for some selected values of α are given as follows
Level of confidence α 0.10 0.05 0.01
1.22 1.36 1.63
Critical value √
n
√
n
√
n
A value of D greater than the critical value will result in rejection of the
null hypothesis.
Example 70.2
Determine whether the testing in Example 70.1 will result in rejection of the
null hypothesis for a 10% level of confidence.
Solution.
For α = 0.10, the critical value is 1.22
√
8
= 0.431 which is smaller than D.
Hence, the null hypothesis is rejected
Example 70.3
Suppose the data in Example 70.1 is right censored at 5100 and the estimated
exponential mean is θ̂ = 3100. Find the KS statistic.
Solution.
x
With censoring, we have F ∗ (x) = 1 − e− 3000 for x ≤ 5100 and F ∗ (x) = 1 for
x > 5100. Hence,
xj F8 (xj−1 ) F8 (xj ) F ∗ (xj ) Maximum of difference
200 0 0.125 0.0623 0.0623
400 0.125 0.25 0.1211 0.0039
1000 0.25 0.375 0.2757 0.0257
1600 0.375 0.5 0.4032 0.0282
3000 0.5 0.625 0.6201 0.1201
5000 0.625 0.75 0.8007 0.1757
5100 0.75 0.75 0.8070 0.057
Hence, D = 0.1757
Remark 70.1
According to [1], if the data is right-censored then the critical values should
be smaller because there is less opportunity for the difference to become
large.
Practice Problems
Problem 70.1 ‡
You are given:
(i) A sample of claim payments is:
29 64 90 135 182
(ii) Claim sizes are assumed to follow an exponential distribution.

(iii) The mean of the exponential distribution is estimated using the method
of moments.
Calculate the value of the Kolmogorov-Smirnov test statistic.
Problem 70.2 ‡
You are given a random sample of observations:
0.1 0.2 0.5 0.7 1.3
You test the hypothesis that the probability density function is:
4
f (x) = , x > 0.
(1 + x)5
Determine the KS test statistic.
Problem 70.3 ‡
You are given:
(i) A random sample of five observations:
0.2 0.7 0.9 1.1 1.3
(ii) You use the Kolmogorov-Smirnov test for testing the null hypothesis,
H0 , that the probability density function for the population is:
4
f (x) = , x > 0.
(1 + x)5
(iii) Critical values for the Kolmogorov-Smirnov test are:
Level of comfidence 0.10 0.05 0.025 0.01

1.22 1.36 1.48 1.63
Critical value √
n
√
n
√
n
√
n
Determine the result of the test.

Problem 70.4 ‡
The size of a claim for an individual insured follows an inverse exponential
distribution with the following probability density function:
θ
θe− x
f (x|θ) = , x > 0.
x2
For a particular insured, the following five claims are observed:
1 2 3 5 13
Determine the value of the Kolmogorov-Smirnov statistic to test the good-

ness of fit of f (x|θ = 2).
Problem 70.5
You are given:
200 400 1000 1600 3000 5000 5400 6200.

(iii) The data are left-truncated at 500.
You conduct a Kolmogorov-Smirnov test at the 0.10 significance level. Do
you reject or fail to reject the null hypothesis that the data came from a pop-
ulation having the left-truncated exponential distribution above? Perform
the test and justify your answer.
Problem 70.6
You use the Kolmogorov-Smirnov goodness-of-fit test to assess the fit of
the natural logarithms of n = 200 losses to a distribution with distribution
function F ∗ .
You are given:
(i) The largest value of |F ∗ (x) − Fn (x)| occurs for some x between 4.26 and
4.42.
(ii)
x F ∗ (x) Fn (x−) Fn (x)
4.26 0.584 0.505 0.510
4.30 0.599 0.510 0.515
4.35 0.613 0.515 0.520
4.36 0.621 0.520 0.525
4.39 0.636 0.525 0.530
4.42 0.638 0.530 0.535
Commonly used large-sample critical values for this test are

1.22 1.36 1.48 1.63
Critical value √
n
√
n
√
n
√
n

71 ANDERSON-DARLING HYPOTHESIS TEST OF FITTED MODELS527
71 Anderson-Darling Hypothesis Test of Fitted Mod-

els
The Anderson-Darling test is another test of the null hypothesis of the
previous section. Like the Kolmogorov-Smirnov test, the test does not work
for grouped data but rather on a complete individual data.
Consider a random sample {x1 , x2 , · · · , xn }. Some of the x0i s may be re-

peated. Suppose there are k distinct values which we arrange in increasing
order
y0 < y1 < y2 < · · · < yk < yk+1
where y0 = 0 if there is no truncation and yk+1 = ∞ if there is no censoring.
If there is a left-truncation of d, we set y0 = d. If there is right-censoring of
u, we set yk+1 = u.
The Anderson-Darling test statistic, denoted by A2 , is defined by

 
k ∗
X k ∗
X 1 − F (yj ) F (y j+1 )
A2 = −nF ∗ (u)+n  [1 − Fn (yj )]2 ln ∗ (y
+ Fn (yj )2 ln ∗ (y )
.
j=0
1 − F j+1 ) j=1
F j
When F ∗ (x) is completely specified with no unknown parameters (see Re-

mark below), the critical values of A2 for some selected values of α are given
as follows
Critical value 1.933 2.492 3.857
A value of A2 greater than the critical value will result in rejection of the
null hypothesis.
Example 71.1
You are given:
200 400 1000 1600 3000 5000 5400 6200

(a) Find the value of A2 .
(b) Determine the result of the test at the 10% level of confidence.
Solution.
x
We have d = 0 and u = ∞. Also, F ∗ (x) = 1 − e− 3300 . Let
1−F ∗ (yj ) F ∗ (yj+1 )

a = [1 − Fn (yj )]2 ln 1−F ∗ (yj+1 ) and b = Fn (yj )2 ln F ∗ (yj ) .
We create the following chart.

yj F8 (yj ) F ∗ (yj ) a b
0 0 0 0.0606 0
200 0.125 0.0588 0.0464 0.0104
400 0.25 0.1142 0.1022 0.0517
1000 0.375 0.2614 0.0710 0.0541
1600 0.5 0.3842 0.1061 0.1102
3000 0.625 0.5971 0.0852 0.1045
5000 0.75 0.7802 0.0076 0.0178
5400 0.875 0.8053 0.0038 0.0388
6200 1.0 0.8472 0 0.1658
Total − − 0.4829 0.5533
Hence,
A2 = −8 + 8(0.4829 + 0.5533) = 0.2896.
(b) At the 10% level of confidence, the critical value is 1.933 > A2 so the we
fail to reject the null hypothesis
Example 71.2
You are given:
(i) Four policy claims:
300 300 800 1500
(ii) Policy limit u = 2000.
(iii) The claims are hypothezised by uniform distribution on [0, 2500].
Calculate the Anderson-Darling test statistic.
Solution.
We have: d = 0, u = 2000, and F ∗ (x) = x
2500 . We create the following chart.
yj F8 (yj ) F ∗ (yj ) a b
0 0 0 0.1278 0
300 0.5 0.12 0.0644 0.2452
800 0.75 0.32 0.0332 0.3536
1500 1.0 0.6 0 0.2877
Total − − 0.2254 0.8865
Hence,
A2 = −4(0.8) + 4(0.2254 + 0.8865) = 1.2476
Remark 71.1
Keep in mind that the critical values for Kolomogorv-Smirnov test and the
Anderson-Darling test are correct ONLY when the null hypothesis com-
pletely specifies the model16 .
16
See [1], page 450.
Practice Problems
Problem 71.1
Batonic Inc. has workers’ compensation claims during a month of:
100 350 550 1000
An actuary who works for the company believes that the claims are dis-
tributed exponentially with mean 500.
Determine the Anderson-Darling test statistic.
Problem 71.2
Using the previous problem, complete the following chart.

Test Result
Problem 71.3 ‡
(A) For a null hypothesis that the population follows a particular distri-
bution, using sample data to estimate the parameters of the distribution
tends to decrease the probability of a Type II error.
(B) The Kolmogorov-Smirnov test can be used on individual or grouped
data.
(C) The Anderson-Darling test tends to place more emphasis on a good fit
in the middle rather than in the tails of the distribution.
(D) None of the above.
Problem 71.4 ‡
Which of the following is false?
(A) For the Kolmogorov-Smirnov test, when the parameters of the distribu-
tion in the null hypothesis are estimated from the data, the probability of
rejecting the null hypothesis decreases.
(B) For the Kolmogorov-Smirnov test, the critical value for right censored
data should be smaller than the critical value for uncensored data.
(C) The Anderson-Darling test does not work for grouped data.
(D) None of the above is true.
Problem 71.5
If the null hypothesis completely spacifies the fitted distribution, the critical
values tend to zero as the sample size goes to infinity for which test?
(A) The Kolmogorov-Smirnov test.

(B) The Anderson-Darling test.
(C) None of the above.
(D) True for both tests.
72 The Chi-Square Goodness of Fit Test

The K-S test and the A-D are mostly applicable for complete individual
data. In contrast, the Chi-square goodness-of-fit test is applicable to
grouped data. To this end, data can be grouped into k intervals such as
(c0 , c1 ], (c1 , c2 ], · · · .(ck−1 , ck ], where ck may be infinity in which case the last
interval is written (ck−1 , ∞). We let nj denote the number of observations
in the interval (cj−1 , cj ]. Thus, if the total number of observations is n then
n = n1 + n2 + · · · + nk .
Also, we allow data to be grouped into categories as in the chart below.
Frequencey (k) Number of Observations (nk )

0 n0
1 n1
2 n2
.. ..
. .
For the j−th category, let pj denote the probability that an observation
falls in the interval (cj−1 , cj ]. Then p̂j = F ∗ (cj ) − F ∗ (cj−1 ). Let pnj =
F (cj ) − F (cj−1 ) be the same probability according to the empirical distribu-
tion. Thus, the expected number of observations based on F ∗ (·) is Ej = np̂j
and that based on Fn (·) is Oj = npnj = nj .
The Chi-square test statistic is given by

 
k 2 k 2 k 2
X (Ej − Oj ) X n(p̂j − pnj ) X nj
χ2 = = =  − n.
Ej p̂j Ej
j=1 j=1 j=1
If F ∗ (x) is completely specified with no unknown parameters, the critical

values for this test come from the Chi-square distribution with degrees of
freedom equal to k−r−1 where r is the number of parameters that have been
estimated in the model distribution, and k is the number of categories. If α
is the level of significance or confidence, then the critical value χ2k−r−1,1−α
is the 100(1 − α)th percentile of the Chi-square distribution with degrees of
freedom k − r − 1. If χ2 ≥ χ2k−r−1,1−α then the null hypothesis is rejected.
In such a case, the model does not fit the data well.
Example 72.1 ‡
You test the hypothesis that a given set of data comes from a known distri-
bution with distribution function F (x). The following data were collected:
72 THE CHI-SQUARE GOODNESS OF FIT TEST 533
Interval F (cj ) Number of Observations

x<2 0.035 5
2≤x<5 0.130 42
5≤x<7 0.630 137
7≤x<8 0.830 66
8≤x 1.000 50
Total 300
where cj is the upper endpoint of each interval.
(a) Find the Chi-square statistic.
(b) Test the null hypothesis at the 5% level of significance.
(c) Test the null hypothesis at the 2.5% level of significance.
Solution.
We have
j p̂j Ej Oj (Ej − Oj )2
1 0.035 10.5 5 30.25
2 0.095 28.5 42 182.25
3 0.5 150 137 169
4 0.2 60 66 36
5 0.17 51 50 1
(a) The Chi-square statistic is
30.25 182.25 169 36 1
χ2 = + + + + = 11.02.
10.5 28.5 150 60 51
(b) Since no mention of any parameter estimation, we assume r = 0. Thus,
χ24,0.95 = 9.488 < χ2 so reject the null hypothesis.
(c) We have χ24,0.975 = 11.143 > χ2 so fail to reject
Example 72.2 ‡
1000 workers insured under a workers compensation policy were observed
for one year. The number of work days missed is given below:
Number of Days Number of Workers
of Work Missed
0 818
1 153
2 25
3+ 4
Total 1000
Total Number of Days Missed 230
The Chi-square goodness-of-fit test is used to test the hypothesis that the
number of work days missed follows a Poisson distribution where:
(i) The Poisson parameter is estimated by the average number of work days
missed.
(ii) Any interval in which the expected number is less than one is combined
with the previous interval.
(a) Find the Chi-square statistic.
(b) Complete the following table.
Level of Significance χ2k−r−1,1−α Test result

10%
5%
2.5$
1%
Solution.
230
For the Poisson distribution, the mean is estimated as λ̂ = 1000 = 0.23. The
probability that a worker missed 0 days is p0 = e−0.23 = 0.7945. We can
create the following table of information.
j p̂j Ej Oj (Ej − Oj )2
0 0.7945 794.5 818 552.25
1 0.1827 182.7 153 882.09
2 0.210 21.0 25 16
3+ 0.0017 1.7 4 5.29
Note that p̂4 = 1 − 0.210 − 0.1827 − 0.7945 = 0.0017.

(a) The Chi-square statistic is
552.25 882.09 16 5.29

χ2 = + + + = 9.397.
794.5 182.7 21 1.7
(b) We have

10% 4.605 Reject
5% 5.991 Reject
2.5$ 7.378 Reject
1% 9.210 Reject
where the degrees of freedom is k − r − 1 = 4 − 1 − 1 = 2

Example 72.3 ‡
You are given:
(i) A computer program simulates n = 1000 pseudo−U (0, 1) variates.
(ii) The variates are grouped into k = 20 ranges of equal length.
20
X
(iii) Oj2 = 51, 850.
j=1
(iv) The Chi-square goodness-of-fit test for U (0, 1) is performed.
Solution.
The hypothesized cdf is
F ∗ (x) = x.
The 20 ranges are of equal length each of length 0.05. Thus, p̂nj = F ∗ (cj ) −
F ∗ (cj−1 ) = cj − cj−1 = 0.05 and Ej = np̂nj = 1000(0.05) = 50. The Chi-
square statistic is
20
X (Ej − Oj )2
χ2 =
Ej
j=1
 
X20 20
X
=0.02  Oj2 − 100 Oj + 20(502 )
j=1 j=1
=0.02[51, 850 − 100(1000) + 50, 000] = 37.

There are k−r−1 = 20−0−1 = 19 degrees of freedom. Note that r = 0 since
no model parameters are estimated. From the table of chi- square, we find
χ219,0.99 = 36.191 < χ2 so the null hypothesis is rejected at the 1% lelvel of
confidence. At the 0.5% level of confidence, we have χ219,0.995 = 38.582 > χ2 ,
the null hypothesis is not rejected
Example 72.4 ‡
During a one-year period, the number of accidents per day was distributed
as follows:
# of Accidents Days
0 209
1 111
2 33
3 7
4 3
5 2
You use a chi-square test to measure the fit of a Poisson distribution with
mean 0.60.
The minimum expected number of observations in any group should be 5.
The maximum possible number of groups should be used.
Determine the chi-square statistic.
Solution.
j
We have Ej = 365e−0.6 0.6
j! . We create the following table:
j Ej Oj (Ej − Oj )2 /Ej
0 200.32 209 0.38
1 120.19 111 0.70
2 36.06 33 0.26
3 7.21 7 1.52∗
4 1.08 3
5 0.13 2
∗ We are told that the minimum expected number of observations for any
group should be 5. Therefore, we combine groups 3, 4 and 5 to obtain
(Ej − Oj )2 (12 − 8.42)2

= = 1.52.
Ej 8.42
The chi-square test-statistic is
χ2 = 0.38 + 0.7 + 0.26 + 1.52 = 2.86

Practice Problems
Problem 72.1 ‡
A particular line of business has three types of claims. The historical prob-
ability and the number of claims for each type in the current year are:
Historical Number of Claims
Type Probability in Current Year
A 0.2744 112
B 0.3512 180
C 0.3744 138
You test the null hypothesis that the probability of each type of claim in
the current year is the same as the historical probability.
Calculate the Chi-square goodness-of-fit test statistic.
Problem 72.2 ‡
period of 365 days:
Number of Claims per Day Observed Number of Days
0 50
1 122
2 101
3 92
4+ 0
(a) Fit a Poisson distribution to the above data, using the method of maxi-
mum likelihood.
(b) Regroup the data, by number of claims per day, into four groups:
0 1 2 3+
Apply the Chi-square goodness-of-fit test to evaluate the null hypothesis
that the claims follow a Poisson distribution. Determine the result of the
chi-square test.
Problem 72.3 ‡
You are investigating insurance fraud that manifests itself through claimants
who file claims with respect to auto accidents with which they were not
involved. Your evidence consists of a distribution of the observed number of
claimants per accident and a standard distribution for accidents on which
fraud is known to be absent. The two distributions are summarized below:
Number of Claimants Observed Number

per Accident Standard Probability of Accidents
1 0.25 235
2 0.35 335
3 0.24 250
4 0.11 111
5 0.04 47
6+ 0.01 22
Total 1.00 1000
Determine the result of a Chi-square test of the null hypothesis that there
is no fraud in the observed accidents.
Problem 72.4 ‡
You are given the following random sample of 30 auto claims:
54 140 230 560 600 1,100 1,500 1,800 1,920 2,000

2,450 2,500 2,580 2,910 3,800 3,800 3,810 3,870 4,000 4,800
7,200 7,390 11,750 12,000 15,000 25,000 30,000 32,300 35,000 55,000
You test the hypothesis that auto claims follow a continuous distribution
F (x) with the following percentiles:
x 310 500 2,498 4,876 7,498 12,930

F (x) 0.16 0.27 0.55 0.81 0.90 0.95
You group the data using the largest number of groups such that the ex-
pected number of claims in each group is at least 5.
Calculate the Chi-square goodness-of-fit statistic.
Problem 72.5 ‡
(A) For a null hypothesis that the population follows a particular distri-
bution, using sample data to estimate the parameters of the distribution
tends to decrease the probability of a Type II error.
(B) The Kolmogorov-Smirnov test can be used on individual or grouped
data.
(C) The Anderson-Darling test tends to place more emphasis on a good fit
in the middle rather than in the tails of the distribution.
(D) For a given number of cells, the critical value for the Chi-square goodness-
of-fit test becomes larger with increased sample size.
(E) None of (A), (B), (C) or (D) is true.
Problem 72.6 ‡
Which of statements (A), (B), (C), and (D) is false?
(A) The Chi-square goodness-of-fit test works best when the expected num-
ber of observations varies widely from interval to interval.
(B) For the Kolmogorov-Smirnov test, when the parameters of the distribu-
tion in the null hypothesis are estimated from the data, the probability of
rejecting the null hypothesis decreases.
(C) For the Kolmogorov-Smirnov test, the critical value for right censored
data should be smaller than the critical value for uncensored data.
(D) The Anderson-Darling test does not work for grouped data.
(E) None of (A), (B), (C) or (D) is false.
73 The Likelihood Ratio Test

A likelihood ratio test is a statistical test used to compare the fit of two
models, one of which (the null model) is a special case of the other (the
alternative model). For example, the null model is the exponential model
while the alternative model is the Gamma model. The hypothesis test prob-
lem can be expressed as follows:
H0 : The data comes from a population with distribution A.

Ha : The data does not come from a population with distribution A.
The likelihood ratio test is conducted as follows: Let L(θ) denote the likeli-
hood function. Let Θ0 denote the set of all possible values of θ as specified
in the null hypothesis. Suppose that the maximum of L(θ) occurs at some
value θ0 ∈ Θ0 with maximum value L0 = L(θ0 ). Likewise, let Θa denote
the set of all possible values of θ as specified in the alternative hypothesis.
Suppose that the maximum of L(θ) occurs at some value θa ∈ Θa with max-
imum value La = L(θa ). Note that θa = M LE(θ).
The likelihood ratio statistic, denoted by T, is defined as

L1
T = 2 ln = 2[ln L1 − ln L0 ].
L0
Regarding T as a random variable, it is shown that, for large sampling,

T is approximated by a Chi-square distribution with degrees of freedom r
equals the number of free parameters under the alternative hypothesis minus
the number of free parameters under the null hypothesis. Let cα denote
the critical value at a significance level α. That is, cα is the 100(1 − α)th
percentile from the Chi-square distribution with degrees of freedom r. Then
the null hypothesis is rejected if T > cα .
Example 73.1
A random sample of n = 8 values from an exponential random variable X
is given:
3 3 4 6 7 8 10 25.
You are performing the following hypothesis test:
H0 : θ = 8.
H1 : θ 6= 8.
73 THE LIKELIHOOD RATIO TEST 541
(a) Find the likelihood ratio statistic for this hypothesis.

(b) Test the hypothesis at the 5% level of significance.
Solution.
(a) We have Θ0 = {8} so that θ0 = 8. The likelihood function is
8
Y 1 xi 1 − 66
L(θ) = e− θ = e θ.
θ θ8
i=1
66
Hence, L0 = L(8) = 818 e− 8 .
66
On the other hand, Θa = {θ > 0 : θ 6= 8} and θ1 = M LE(θ) = x = 8 =
1 − 66
8.25 and therefore L1 = 8.25 8 e 8.25 . The test statistic is
" 8 #
8 66
− 8.25 + 66
T = 2 ln e 8 = 0.00765.
8.25
(b) The number of free parameters in the null hypothesis is 0 (θ is specified

as 8) while it is 1 in the alternate hypothesis (θ is freely chosen to maximize
L(θ)). That is, r = 1. From the table of Chi-square distribution we find
c0.05 = 3.84. Since T < c0.05 , do not reject the null hypothesis
Example 73.2 ‡
You are given:
(i) A random sample of losses from a Weibull distribution is:
595 700 789 799 1109
(ii) You use the likelihood ratio test to test the hypothesis:
H0 : τ = 2.
6 2.
Ha : τ =
(iii) When τ = 2, the maximum likelihood estimate of θ is 816.7.

(a) Find ln L0 .
(b) Find the degrees of freedom.
(c) Complete the following table:
α 10% 5% 2.5% 1%
cα
Solution.
(a) The likelihood function is
x τ
j
−
5
Y τ xτj −1 e θ
L(τ, θ) = .
θτ
j=1

5 h x τ i
j
X
`(τ, θ) = ln τ + (τ − 1) ln xj − τ ln θ − .
θ
j=1
Thus,
ln L0 =5 ln 2 + ln (595 · 700 · 789 · 799 · 1109) − 10 ln 816.7

5952 + 7002 + 7892 + 7992 + 11092

− = −35.28.
816.72
(b) The free parameter in the null hypothesis is θ. The free parameters in
the alternate hypothesis is τ and θ. Hence, r = 2 − 1 = 1.
(c) Using the Chi-square distribution with one degree of freedom, we find
α 10% 5% 2.5% 1%
cα 2.706 3.841 5.024 6.635
Example 73.3 ‡
You are given:
(i) Twenty claim amounts are randomly selected from a Pareto distribution
with α = 2 and an unknown θ.
(ii) The
P maximum likelihood estimate of θ is 7.0
(iii) P ln(xi + 7.0) = 49.01
(iv) ln(xi + 3.1) = 39.30
You use the likelihood ratio test to test the hypothesis that θ = 3.1.
Determine the result of the test at the 1% level.
Solution.
20
Y αθα
L(α, θ) =
(xi + θ)α+1
j=1
and its loglikelihood function is

20
X
`(α, θ) = 20 ln α + 20α ln θ − (α + 1) ln(xj + θ).
j=1
The loglikelihood value under the null hypothesis (α = 2 and θ = 3.1) is
ln L0 = 20 ln 2 + 20(2) ln 3.1 − 3(39.30) = −58.781.
The loglikelihood value under the alternative hypothesis (α = 2 and θ = 7.0)

is
ln L1 = 20 ln 2 + 20(2) ln 7.0 − 3(49.01) = −55.331.
There is one degree of freedom (the hypothesized distribution has no pa-
rameters estimated versus one estimated parameter in the alternative hy-
pothesis). The test statistic for the Chi-square test is
2(ln L1 − ln L0 ) = 2(−55.331 + 58.781) = 6.9.
In the Chi-square table with 1 degree of freedom, we see that χ21,0.99 =

6.636 < 6.9 so the null hypothesis is rejected at this level
Practice Problems
Problem 73.1
Consider the following hypothesis test problem:
H0 : The data came from a Pareto distribution with α = 1.5 and θ = 7.8.
6 1.5 and θ 6= 7.8.
Ha : The data came from a Pareto distribution with α =
Find the degrees of freedom associated with the likelihood ratio test.
Problem 73.2
Consider the following hypothesis test problem:
H0 : The data came from a Pareto distribution with α = 1.5 and θ = 7.8.
6 1.5 and θ 6= 7.8.
Ha : The data came from a Pareto distribution with α =
Using the likelihood ratio test, complete the following table:
α 5% 2.5% 1% 0.5%
cα
Problem 73.3 ‡
You fit a Pareto distribution to a sample of 200 claim amounts and use the
likelihood ratio test to test the hypothesis that α = 1.5 and θ = 7.8.
You are given:
(i) The maximum likelihood estimates are α̂ = 1.4 and θ̂ = 7.6..
(ii) The natural logarithm of the likelihood function evaluated at the maxi-
mumPlikelihood estimates is −817.92.
(iii) ln(xi + 7.8) = 607.64.
Problem 73.4 ‡
You are given:
(i) A random sample of losses from a Weibull distribution is:
595 700 789 799 1109
(ii) At the maximum likelihood estimates of θ and τ, 5i=1 ln f (xi ) = −33.05.

P
(iii) When τ = 2, the maximum likelihood estimate of θ is 816.7.
(iv) You use the likelihood ratio test to test the hypothesis:
H0 : τ = 2.
6 2.
Ha : τ =
Problem 73.5 ‡
During a one-year period, the number of accidents per day was distributed
as follows:
# of Accidents Days
0 209
1 111
2 33
3 7
4 3
5 2
For these data, the maximum likelihood estimate for the Poisson distribu-
tion is λ̂ = 0.60, and for the negative binomial distribution, it is r̂ = 2.9 and
β̂ = 0.21.
The Poisson has a nega tive loglikelihood value of 385.9, and the negative
binomial has a negative loglikelihood value of 382.4.
Determine the likelihood ratio test statistic, treating the Poisson distribu-
tion as the null hypothesis.
74 Schwarz Bayesian Criterion

When selecting a model, one has to keep two things in mind:
• Principle of Parsimony: When selecting a model, a simpler model

is always preferred unless there is considerable evidence to do otherwise. A
reason of this preference is that a complex model may do a great job of
matching the data but may fail in matching the population from which the
data were sampled.
• Reduce, if possible, the universe of potential models.
In [1], two approaches for model selection are considered. The first one is
judgement-based approach where the modeler’s experience is critical. This
is a feature of the second point mentioned above. The other approach is
a score-based approach where a numerical value (a score) is assigned to a
model and the model selected is the one with best score.
In this section, we consider a score-based selection method, known as the

Schwarz Bayesian Criterion, denoted by SBC, and is defined by
r
SBC = ln L(θ̂) − ln n
2
where ln L(θ̂) is the maximum loglikelihood value, r is the number of param-

eters being estimated in the model, and n is the sample size. The quantity
r
2 ln n is referred to as the Schwarz Bayesian adjustment.
With multiple models, the SBC of each model is computed. The model
preferred is the one with highest SBC.
Example 74.1
You are given that a particular model has a maximum loglikelihood value
of −412. You also know that the model uses 2 parameters and is used to
interpret a sample of 260 data points. Find the SBC of this model.
Solution.
The SBC is given by
2
SBC = −412 − ln 260 = −417.56
2
74 SCHWARZ BAYESIAN CRITERION 547
Example 74.2
What is the limit of the Schwarz Bayesian adjustement as the sample size
increases without bound?
Solution.
We have
r
lim ln n = ∞
n→∞ 2
Example 74.3
Four models are fitted to a sample of n = 200 observations with the following
results:
Model # of Parameters Loglikelihood
I 3 −180.2
II 2 −181.4
III 2 −181.6
IV 1 −183
Determine the model favored by the Schwarz Bayesian criterion.
Solution.
We have the following
Model # of Parameters Loglikelihood SBC

I 3 −180.2 −188.1
II 2 −181.4 −186.7
III 2 −181.6 −186.9
IV 1 −183 −185.6
Hence, the favored model is IV

Practice Problems
Problem 74.1
Which of the following statements is false?
(i) The principle of parsimony states that a more complex model is bet-
ter because it will always match the data better.
(ii) In judgment-based approaches to determining a model, a modeler’s ex-
perience is critical.
(iii) In score-based approaches, one assigns scores to the potential models.
Problem 74.2 ‡
If the proposed model is appropriate, which of the following tends to zero
as the sample size goes to infinity?
(A) Kolmogorov-Smirnov test statistic

(B) Anderson-Darling test statistic
(C) Chi-square goodness-of-fit test statistic
(D) Schwarz Bayesian adjustment
(E) None of (A), (B), (C) or (D)
Problem 74.3 ‡
Five models are fitted to a sample of n = 260 observations with the following
results:

I 1 −414
II 2 −412
III 3 −411
IV 4 −409
V 6 −409
Problem 74.4 ‡
You are given:
(i) Sample size = 100
(ii) The negative loglikelihoods associated with five models are:
74 SCHWARZ BAYESIAN CRITERION 549

Generalized Pareto 1 −414
Burr 2 −412
Pareto 3 −411
Lognormal 4 −409
Inverse exponential 6 −409

Credibility Theory
A framework of credibility problem can be described as follows: We consider

a block of insurance policies, referred to as a risk group. This risk group is
covered by an insurer over a period of time upon the payment of a premium.
The value of the premium is decided based on a rate specified in the manual
rate and on the specific risk characteristics of the group. An actuary studies
the recent claim experience of the risk group to decide whether a revised
premium for the next period is required. Credibility theory concerns the
finding of this premium for the next period using the recent claim experience
and the manual rate.
We will consider the following three approaches to credibility: The limited
fluctuation credibility approach17 , the Bayesian approach, and the
Bühlmann’s approach.
17
Also known as classical credibility.
551
552 CREDIBILITY THEORY
75 Limited Fluctuation Credibility Approach: Full

Credibility
Let X1 , X2 , · · · , Xn represent the total claims/losses18 (or other entity) ex-
perienced by a policyholder in the past n years. That is, Xi is being experi-
enced at time i. In the literature, an observed value of Xi is also referred to
as an exposure19 or exposure unit of Xi . The average claim severity is
X1 + X2 + · · · + Xn
X= .
n
The statistic X is the average of past experience for the given group20 .
We make the assumptions:
E(Xj ) =ξ
Var(Xj ) =σ 2
for all j = 1, 2, · · · , n. That is ξ is the mean, which is the premium to be

charged if known, across the members of the group−presumed to be stable
over time. If the Xi0 are independent, then we can write
E(X) =ξ
σ2
Var(X) = .
n
An insurer’s goal is to decide on the value of ξ. The choice can be done in
one of three ways:
• Ignore past experience or data (no credibility) and charge the manual
premium21 M.
• Use only past data (full credibility) and charge X (the observed pure
premium).
• Use a combination of M and X (partial credibility).

18
Recall that the distribution of the number of claims is a frequency distribution and
that of the size of a claim is a severity distribution
19
In general, an exposure for a particular random variable is one observation of that
random variable.
20
In the literature, this is also referred to as pure premium
21
A manual premium is a premium that comes from a manual (book) of premiums.
75 LIMITED FLUCTUATION CREDIBILITY APPROACH: FULL CREDIBILITY553
In general, insurers prefer to choose X because of its stability over time

and is a good indicator of future results. However, in the presence of more
variability, a manual premium is more suitable.
By assigning X to ξ (full credibility) we require that X be stable

over time.
X−ξ
In statistical terms, this means that the relative error ξ is small with
high probability. That is, there are two numbers r > 0 ( with r close to 0,
a common choice is r = 0.05) and 0 < p < 1(with p close to 1, a common
choice is p = 0.9). Mathematically,

X − ξ
Pr ≤ r ≥ p.
ξ
As n → ∞, by the Central Limit theorem, we have
σ2

X −ξ
X ∼ N ξ, → σ ∼ N (0, 1).
n √
n
In this case, we have the equivalence

X − ξ rξ √n
!
X − ξ
Pr ≤ r ≥ p ⇐⇒ Pr σ ≤ ≥ p.

ξ √n σ
Now, let
X − ξ
yp = inf Pr ≤y ≥p .
y ξ
If X is continuous then yp satisfies

X − ξ
Pr ≤ yp = p.
ξ
Accordingly, the condition of full credibility is met when

√
rξ n
≥ yp .
σ
y 2
Letting λ0 = rp , we obtain
√ r
rξ n σ n
≥ yp ⇐⇒ ≤ .
σ ξ λ0
σ
This says that full credibility is assigned if the coefficient of variation ξ of
q
Xj is no larger than λn0 .
Alternatively, full credibility occurs if
σ2 ξ2
Var(X) = ≤ .
n λ0
The number of exposure units required for full credibility is
2
σ
n ≥ λ0 . (75.1)
ξ
We next describe how to find yp when X is continuous. Recall that the cdf
of the standard normal distribution Z is denoted by Φ. We have
p =Pr(|Z| ≤ yp )
=Pr(−yp ≤ Z ≤ yp )
=Φ(yp ) − Φ(−yp )
=Φ(yp ) − 1 + Φ(yp )
=2Φ(yp ) − 1.

Thus, Φ(yp ) = 1+p
2 so that yp is the 100
1+p
2 −th percentile of the standard
normal distribution.
Example 75.1
Determine yp and λ0 if p = 0.9 and r = 0.05. Also, determine the full-
credibility standard.
Solution.
yp 2 1.645 2

The standard normal tables give y0.9 = 1.645. Thus, λ0 = r = 0.05 =
1082.41. The full credibility standard is
2
σ
n ≥ 1082.41
ξ
The right hand-side of Equation (75.1) is known as the standard of full

credibiltiy when measuring it in terms of exposure units. If some other unit
is desired, such as the expected number of claims or the expected amount
of claims, it is usually sufficient to multiply both sides of (75.1) by an ap-
propriate factor.
Example 75.2
Let N1 , N2 , · · · , Nn the number of claims in the past n years for a policy.
That is, Ni is the total number of claims of year i. Let X1 , X2 , · · · , Xn . be
the losses in the past n years. That is, Xi is the total losses in year i. We
assume that the Xi0 and the Ni0 s are independent with the Ni0 s being iid
with common distribution the Poisson distribution with mean λ.
Let Yij denote the j th claim in year i. We assume that the Yij are iid with
mean θY and variance σY2 . Then we have
Ni
X
Xi = Yij .
j=1
Find the standard of full credibility based on

(a) the number of exposure units;
(b) the expected total number of claims;
(c) the expected total amount of claims.
Solution.
Xi is a compound Poisson distribution so that E(Xi ) = E(Ni )E(Yij ) = λθY
and Var(Xi ) = Var(Ni )[E(Yij )]2 + E(Ni )Var(Yij ) = λ(θY2 + σY2 ).
(a) The standard of full credibility based on the number of exposure units
is "
2 2 #
σXi λ0 σY
λ0 = 1+ exposures.
E(Xi ) λ θY
(b) The expected total number of claims is ni=1 E(Ni ) = nE(Ni ) = nλ.
P
The standard of full credibility based on the expected total number of claims
is "
2 2 #
σXi σY
E(Ni )λ0 = λ0 1 + claims.
E(Xi ) θY
Pn
(c) The expected total amount of claims is i=1 E(Xi ) = nE(Xi ). The
standard of full credibility based on the expected total amount of claims is
2 2
σX σY2

σXi i
E(Xi )λ0 = λ0 = λ0 θY + dollars
E(Xi ) E(Xi ) θY
Example 75.3
Repeat the previous example by replacing the Poisson distribution with a
binomial distribution with parameters (50, p0 ).
Solution.
Xi is a compound binomial distribution so that E(Xi ) = E(Ni )E(Yij ) =
50p0 θY and Var(Xi ) = Var(Ni )[E(Yij )]2 + E(Ni )Var(Yij ) = λ(θY2 + σY2 ) =
50p0 [(1 − p0 )θY2 + σY2 ].
(a) The standard of full credibility based on the number of exposure units
is 2 " 2 #
σXi λ0 σY
λ0 = 0
(1 − p0 ) + .
E(Xi ) 50p θY
(b) The expected total number of claims is ni=1 E(Ni ) = nE(Ni ) = 50np0 .
P
The standard of full credibility based on the expected total number of claims
is " 2 #
σY
λ0 (1 − p0 ) + .
θY
Pn
(c) The expected total amount of claims is i=1 E(Xi ) = nE(Xi ). The
standard of full credibility based on the expected total amount of claims is
2 2
σX σY2

σXi i 0
E(Xi )λ0 = λ0 = λ0 (1 − p )θY +
E(Xi ) E(Xi ) θY
Example 75.4 ‡
You are given:
(i) The number of claims follows a negative binomial distribution with pa-
rameters r and β = 3.
(ii) Claim severity has the following distribution:
Claim Size Probability

1 0.4
10 0.4
100 0.2
(iii) The number of claims is independent of the severity of claims.
Determine the expected number of claims needed for aggregate losses to be
within 10% of expected aggregate losses with 95% probability.
Solution.
We are asked to find the full credibility standard based on the number of
claims of aggregate losses. If N denote the number of claims then the full
credibility standard is the right side of the inequality
p !2
Var(S)
n(3r) ≥ (3r)λ0
E(S)
where S is the aggregate loss and

1.96 2

λ0 = = 384.16.
0.1
We have
E(S) =E(N )E(X) = 3r[1(0.4) + 10(0.4) + 100(0.2)] = 3r(24.4) = 73.2r
=3r(12 (0.4) + 102 (0.4) − 1002 (0.2) − 24.42 ) + 3r(1 + 3)24.42
=11, 479.44r.
Thus, the full credibility standard based on the number of claims of aggre-
gate losses
11, 479.44
3r(384.16) = 2469.06.
(73.2r)2
Hence, the expected number of claims needed is at least 2470
Example 75.5 ‡
For an insurance portfolio, you are given:
(i) For each individual insured, the number of claims follows a Poisson dis-
tribution.
(ii) The mean claim count varies by insured, and the distribution of mean
claim counts follows a gamma distribution.
(iii) For a random sample of 1000 insureds, the observed claim counts are
as follows:
Number Of Claims, n 0 1 2 3 4 5
Number Of Insureds, fn 512 307 123 41 11 6
P P 2
nfn = 750 and n fn = 1494.
(iv) Claim sizes follow a Pareto distribution with mean 1500 and variance
6,750,000.
(v) Claim sizes and claim counts are independent.
(vi) The full credibility standard is to be within 5% of the expected aggregate
loss 95% of the time.
Determine the minimum number of insureds needed for the aggregate loss
to be fully credible.
Solution.
We want the smallest positive integer n such that
Var(S)
n ≥ λ0
E(S)2
where 2
1.96
λ0 = = 1536.64.
0.05
We have
750
ξ =E(S) = E(N )E(X) = (1500) = 1125
1000
=0.75(6, 750, 000) + Var(N )(1500)2
fi (ni − 0.75)2
P
Var(N ) = = 0.93243
1000 − 1
Var(S) =0.75(6, 750, 000) + 0.93243(1500)2
=7, 160, 468.
Thus,
Var(S) 7, 160, 468
λ0 = 1536.64 = 8693.77
E(S)2 11252
so that n = 8694
Practice Problems
Problem 75.1 ‡
You are given:
(i) The number of claims has a Poisson distribution.
(ii) Claim sizes have a Pareto distribution with parameters θ = 0.5 and
α = 6.
(iii) The number of claims and claim sizes are independent.
(iv) The observed pure premium should be within 2% of the expected pure
premium 90% of the time.
Determine the expected number of claims needed for full credibility.
Problem 75.2 ‡
You are given the following information about a commercial auto liability
book of business:
(i) Each insured’s claim count has a Poisson distribution with mean λ, where
λ has a gamma distribution with α = 1.5 and θ = 0.2.
(ii) Individual claim size amounts are independent and exponentially dis-
tributed with mean 5000.
(iii) The full credibility standard is for aggregate losses to be within 5% of
the expected with probability 0.90.
Using classical credibility, determine the expected number of claims required

for full credibility.
Problem 75.3 ‡
You are given:
(i) The number of claims has probability function:

m
p(x) = q x (1 − q)m−x , x = 0, 1, · · · , m.
x
(ii) The actual number of claims must be within 1% of the expected number
of claims with probability 0.95.
(iii) The expected number of claims for full credibility is 34,574.
Determine q.
Problem 75.4 ‡
You are given:
(i) The number of claims follows a Poisson distribution.

(ii) Claim sizes follow a gamma distribution with parameters α (unknown)
and θ = 10, 000.
(iii) The number of claims and claim sizes are independent.
(iv) The full credibility standard has been selected so that actual aggregate
losses will be within 10% of expected aggregate losses 95% of the time.
Using limited fluctuation (classical) credibility, determine the expected num-

ber of claims required for full credibility.
Problem 75.5 ‡
A company has determined that the limited fluctuation full credibility stan-
dard is 2000 claims if:
(i) The total number of claims is to be within 3% of the true value with
probability p.
(ii) The number of claims follows a Poisson distribution.
The standard is changed so that the total cost of claims is to be within 5%
of the true value with probability p, where claim severity has probability
density function:
1
f (x) = , 0 ≤ x ≤ 10, 000.
10, 000
Using limited fluctuation credibility, determine the expected number of

claims necessary to obtain full credibility under the new standard.
76 LIMITED FLUCTUATION CREDIBILITY APPROACH: PARTIAL CREDIBILITY561
76 Limited Fluctuation Credibility Approach: Par-

tial Credibility
When the premium charged under full credibility is not approapriate reflec-
tion of the actual premium, an alternative remedy is the partial credibility.
Under partial credibility the premium is a weighted average of the past

experience X and the manual premium M > 0. Mathematically,
Pc = ZX + (1 − Z)M, 0 ≤ Z ≤ 1
where Z is called the credibility factor and Pc is the credibility pre-

mium. Note that when Z = 1, the partial credibility coincides with the full
credibility.
Example 76.1
(a) Find E(Pc ) and Var(Pc ).
(b) Find P√c −E(Pc ) .
Var(Pc )
Solution.
(a) We have
E(Pc ) = E[ZX + (1 − Z)M ] == Zξ + (1 − Z)M
and
σ2
Var(Pc ) = Var[ZX + (1 − Z)M ] = Var(ZX) = Z 2 Var(X) = Z 2 .
n
(b) We have
Pc − E(Pc ) ZX − Zξ X −ξ
p = √ = √
Var(Pc ) Zσ/ n σ/ n
How do we find Z? There are several ways for finding Z and all of them
lead to the same result. We will follow
an approach similar to Section 75:
ZX−Zξ
We want to choose Z such that ξ is small with high probability.
Mathematically,
ZX − Zξ
Pr ≤ r ≥ p.

ξ
As n → ∞, by the Central Limit theorem, we have
2

2σ ZX − Zξ
ZX ∼ N Zξ, Z → ∼ N (0, 1).
n Z √σn
In this case, we have the equivalence

X − ξ rξ √n
!
ZX − Zξ
Pr ≤ r ≥ p ⇐⇒ Pr σ ≤

≥ p.
ξ √n Zσ
Now, let
X − ξ
yp = inf Pr
√ ≤y ≥p .
y σ/ n
Accordingly, the condition of partial credibility is met when
√
rξ n
≥ yp .
Zσ
y 2
Letting λ0 = rp , we obtain
√ r
rξ n ξ n
≥ yp ⇐⇒ Z ≤ .
Zσ σ λ0
We choose Z such that

r
ξ n
Z = min ,1 .
σ λ0
Letting nF = √ξ be the standard full credibility measured in terms of ex-

σ λ0q
posures then Z = nnF can be interpreted as the square root of the number
of observations (exposures) available divided by the number of observations
(exposures) required for full credibility.
Based on the total number of claims, Z is the square root of the number
of available claims divided by the total number of claims required for full
credibility. Based on total amout of claims, Z is the square root of the total
amount of available claims divided by the total amount of claims required
for the standard credibility.
Example 76.2
You are given:
(i) The number of claims per exposure follows a Poisson distribution with a
mean of 10.
(ii) Claim size follows a Pareto distribution with parameters α = 3 and
θ = 1.
(iii) The number of claims per exposure and claim sizes are independent.
(iv) The method of limited fluctuation credibility is used, and the full cred-
ibility standard has been selected so that total claim dollars per exposure
will be within 10% of expected total claim dollars per exposure 95% of the
time.
Find the credibility factor
(a) based on 45 exposures
(b) based on a total claim number of 120
(c) based on a total claim amount of 600.
Solution.
(a) The standard for full credibility based on the number of exposures is
σY2 1.96 2 1 0.752

λ0
1+ 2 = 1+ = 124.852.
λ θY 0.1 10 0.52
The credibility factor based on 45 exposures is
r
45
Z= = 0.6004.
124.852
(b) The standard for full credibility based on the total number of claims is
σY2 1.96 2 0.752

λ0 1 + 2 = 1+ = 1248.52.
θY 0.1 0.52
The credibility factor based on 120 claims is
r
120
Z= = 0.31.
1248.52
(c) The standard for full credibility based on the total amount of claims is
σY2 1.96 2 0.752

λ 0 θY + = 0.5 + = 624.26.
θY 0.1 0.5
The credibility factor based on a total amount of 2500 is
r
600
Z= = 0.9804
624.26
Example 76.3
You are given:
(i) 350 claims with a total of 300,000.
(ii) The manual premium of M = 1000.
(iii) The credibility factor Z = 0.809.
Determine the credibility premium.
Solution.
The partial credibility pure premium is
300000
X= = 857.14.
350
The credibility premium is
Pc = ZX + (1 − Z)M = 0.809(857.14) + (1 − 0.809)(1000) = 884.43

Practice Problems
Problem 76.1 ‡
You are given the following information about a general liability book of
business comprised
PNi of 2500 insureds:
(i) Xi = j=1 Yij , is a random variable representing the annual loss of the
ith insured.
(ii) N1 , N2 , · · · , N2500 , are independent and identically distributed random
variables following a negative binomial distribution with parameters r = 2
and β = 0.2.
(iii) Yi1 , Yi2 , · · · , YiNi are independent and identically distributed random
variables following a Pareto distribution with α = 3.0 and θ = 1000.
(iv) The full credibility standard is to be within 5% of the expected aggre-
gate losses 90% of the time.
Using classical credibility theory, determine the partial credibility of the

annual loss experience for this book of business.
Problem 76.2 ‡
You are given:
(i) Xpartial = pure premium calculated from partially credible data.
(ii) µ = E[Xpartial ].
(iii) Fluctuations are limited to ±kµ of the mean with probability P. (iv)
Z = credibility factor.
Which of the following is equal to P?
(A) Pr(|Xpartial − µ| ≤ kµ)

(B) Pr(|ZXpartial − Zµ| ≤ k)
(C) Pr(|ZXpartial − Zµ| ≤ µ)
(D) Pr(|ZXpartial + (1 − Z)µ − 1| ≤ k)
(E) Pr(|ZXpartial + (1 − Z)µ − µ| ≤ kµ).
Problem 76.3
You are given:
(i) 50 claim amounts.
(ii) Claim size is uniform in [0, θ].
(iii) The full credibility standard is to be within 5% of the expected claim
amount 90% of the time.
Using partial credibility theory, calculate the credibility factor Z.

Problem 76.4 ‡
You are given:
(i) Claim counts follow a Poisson distribution.
(ii) Claim sizes follow a lognormal distribution with coefficient of variation
3.
(iii) Claim sizes and claim counts are independent.
(iv) The number of claims in the first year was 1000.
(v) The aggregate loss in the first year was 6.75 million.
(vi) The manual premium for the first year was 5.00 million.
(vii) The exposure in the second year is identical to the exposure in the first
year.
(viii) The full credibility standard is to be within 5% of the expected aggre-
gate loss 95% of the time.
Determine the limited fluctuation credibility net premium (in millions) for
the second year.
Problem 76.5 ‡
You are given:
(i) Claim counts follow a Poisson distribution.
(ii) claim size follows a Pareto distribution with parameters α = 3 and θ = 1.
(iii) A full credibility standard is established so that the actual number of
claims will be within 5% of the expected number of claims 95% of the time.
Determine the number of expected claims needed for 30% partial credibility
for the distribution of number of claims.
Problem 76.6 ‡
You are given:
(i) The full credibility standard is 100 expected claims.
(ii) The square-root rule is used for partial credibility.
You approximate the partial credibility formula with a Bühlmann credibility
formula by selecting a Bühlmann k value that matches the partial credibility
formula when 25 claims are expected.
Determine the credibility factor for the Bühlmann credibility formula when
100 claims are expected.
77 GREATEST ACCURACY CREDIBILITY APPROACH 567
77 Greatest Accuracy Credibility Approach

Greatest accuracy credibility theory is a credibility theory based on
Bayesian inference. In this section, we consider this model-based approach
to the solution of the following credibility problem:
Let X1 , X2 , · · · , Xn be the abserved losses or claims in the past n years of

a particular policyholder. Past experience had shown that the manual rate
µ is not an appropriate measure of the premium for the next year (µ 6= X).
This leads to the question of whether the net premium of next year should
be based on µ alone, X alone, or a combination of µ and X.
Greatest accuracy credibility theory addresses the following two questions:
(1) Is the policyholder is different than what was assumed in figuring out
the manual rate µ?
(2) Has it been random chance that has been responsible in the difference
between µ and X?
Now when an individual applies for an insurance policy, there is usually

a strict underwriting process by the insurer. In this process, a policyholder
is rated and placed into a rating or risk class. One thinks that there is ho-
mogeneity in the rating process among the members of the same risk class.
Unfortunately, reality shows otherwise. That is, there is heterogeneity in
risk ratings amongst policyholders of the same rating class.
This assumption that a policyholder is different from members of the same

rating class leads to the question of how an underwriter chooses what is an
appropriate rate for the policyholder. In order to answer this question, we
make the following assumptions:
(1) Every policyholder is characterized by a risk level θ within the rating

class.
(2) The value of θ varies amongst policyholders in the same risk class. This
assumption allows us to quantify the difference between policyholders with
respect to the risk characteristics.
(3) θ can be viewed as a random variable on the set of risk levels with
a probability function π(θ) and cumulative distribution function Π(θ). In

words, Π(θ) represents the probability that a policyholder picked at random
from the risk class has a risk parameter less than or equal to θ.
While the risk parameter θ associated with an individual policyholder is

not known, we will assume that π(θ) is known.
Because risk level varies in the population, the experience of a policyholder

picked at random from the population arises from a two-stage process:
(1) The risk parameter θ is selected from the distribution π(θ) (prior distri-
bution).
(2) Claims or losses are selected from the conditional distribution fX|Θ (x|θ).
(model distribution)
Example 77.1
The amount of a claim X|Θ has the exponential distribution with parameter
1
θ . The risk parameter Θ has a gamma distribution with parameters α and
β. Provide a mathematical description of this model.
Solution.
For the risk parameter, we have
−θ
θα−1 e β
π(θ) = α .
β Γ(α)
For the claims, we have
1 x
fX|Θ (x|θ) = e− θ
θ
Example 77.2
The amount of a claim X|Λ has the inverse exponential distribution with pa-
rameter λ The risk parameter Λ has a gamma distribution with parameters
α and θ. Provide a mathematical description of this model.
Solution.
For the risk parameter, we have
λ
λα−1 e− θ
π(λ) = α .
θ Γ(α)
77 GREATEST ACCURACY CREDIBILITY APPROACH 569

λ −λ
fX|Λ (x|λ) = e x
x2
Practice Problems
Problem 77.1
The amount of a claim X|Λ has the Poisson distribution with parameter λ.
The risk parameter Λ has a gamma distribution with parameters α and β.
Provide a mathematical description of this model.
Problem 77.2
The amount of a claim X|Θ has the normal distribution with parameters θ
and σ12 . The risk parameter Θ has a normal distribution with parameters µ
and σ22 .
Problem 77.3
The amount of a claim X|Q has the Binomial distribution with parameters
(m, q) where m is known. The risk parameter Q has a beta distribution
with parameters (a, b, 1).
Problem 77.4
The amount of a claim X|Λ has the exponential distribution with parameter
1
λ . The risk parameter Λ has an inverse Gamma distribution with parame-
ters α and θ.
Problem 77.5
The amount of a claim X|Λ has the uniform distribution in [0, λ]. The risk
parameter Λ s single parameter Pareto distribution with α and θ.

78 CONDITIONAL DISTRIBUTIONS AND EXPECTATION 571
78 Conditional Distributions and Expectation

The credibility model of the previous section together with the credibility
models to follow in the coming sections require a good understanding of
conditional distributions and conditional expectation. In this section, a dis-
cussion of these topics are presented.
Suppose X and Y are two continuous random variables with joint density
fXY (x, y). Let fX|Y (x|y) denote the probability density function of X given
that Y = y. The conditional density function of X given Y = y is
fXY (x, y)
fX|Y (x|y) =
fY (y)
provided that fY (y) > 0.

Compare this definition with the discrete case where
pXY (x, y)
pX|Y (x|y) = .
pY (y)
The marginal distribution of X is obtained by integrating out the joint

distribution,
Z ∞
fX (x) = fXY (x, y)dy
−∞
or Z ∞
fX (x) = fX|Y (x|y)fY (y)dy.
−∞
Note that
Z ∞ Z ∞
fXY (x, y) fY (y)
fX|Y (x|y)dx = dx = = 1.
−∞ −∞ fY (y) fY (y)
Example 78.1
Suppose X and Y have the following joint density
1

2 |X| + |Y | < 1
fXY (x, y) =
0 otherwise
(a) Find the marginal distribution of X. That is, fX (x).

(b) Find the conditional distribution of Y given X = 12 .
Solution.
(a) Clearly, X only takes values in (−1, 1). So fX (x) = 0 if |x| ≥ 1. Let
−1 < x < 1,
Z ∞ Z 1−|x|
1 1
fX (x) = dy = dy = 1 − |x|.
−∞ 2 −1+|x| 2
(b) The conditional density of Y given X = 12 is then given by
f ( 12 , y) 1 − 12 < y < 12

fY |X (y|x) = =
fX ( 12 ) 0 otherwise
Thus, fY |X follows a uniform distribution on the interval − 12 , 12

Theorem 78.1
Continuous random variables X and Y with fY (y) > 0 are independent if
and only if
fX|Y (x|y) = fX (x).
Proof.
Suppose first that X and Y are independent. Then fXY (x, y) = fX (x)fY (y).
Thus,
fXY (x, y) fX (x)fY (y)
fX|Y (x|y) = = = fX (x).
fY (y) fY (y)
Conversely, suppose that fX|Y (x|y) = fX (x). Then fXY (x, y) = fX|Y (x|y)fY (y) =
fX (x)fY (y). This shows that X and Y are independent
Example 78.2
Let X and Y be two continuous random variables with joint density function

c 0≤y<x≤2
fXY (x, y) =
0 otherwise
(a) Find fX (x), fY (y) and fX|Y (x|1).

(b) Are X and Y independent?
Solution.
(a) We have Z x
fX (x) = cdy = cx, 0 ≤ x ≤ 2
0
Z 2
fY (y) = cdx = c(2 − y), 0 ≤ y ≤ 2
y
and
fXY (x, 1) c
fX|Y (x|1) = = = 1, 1 ≤ x ≤ 2.
fY (1) c
(b) Since fX|Y (x|1) 6= fX (x), X and Y are dependent
Example 78.3
Show that
fY |X (y|x)fX (x)
fX|Y (x|y) = .
fY (y)
Solution.
We have
fX|Y (x|y)fY (y) = fY |X (y|x)fX (x)(= fXY (x, y)).
Dividing by sides of this equation by fY (y), the result follows
We now turn our attention to conditional expectation. Let X and Y be

random variables. We define conditional expectation of X given that
Y = y by
Z ∞
E(X|Y = y) = xfX|Y (x|y)dx
−∞
where
fXY (x, y)
fX|Y (x|y) = .
fY (y)
Example 78.4
Suppose that the joint density of X and Y is given by
−x
y e−y
e
fXY (x, y) = , x, y > 0.
y
Compute E(X|Y = y).

Solution.
The conditional density is found as follows
fXY (x, y)
fX|Y (x|y) =
fY (y)
fXY (x, y)
=R ∞
−∞ fXY (x, y)dx
−x
y e−y
(1/y)e
=R ∞ −x
(1/y)e y e−y dx
0
−x
(1/y)e y
=R ∞ −x
0 (1/y)e y dx
1 −x
= e y
y
Hence,
Z ∞ ∞ Z ∞ x
x − xy −x −
E(X|Y = y) = e dx = − xe y − e y dx

0 y 0 0
i∞
−x −x
h
= − xe + ye
y y =y
0
Notice that if X and Y are independent then fX|Y (x|y) = fX (x) so that
E(X|Y = y) = E(X).
Theorem 78.2 (Double Expectation Property)
E(X) = E(E(X|Y ))
Proof.
We give a proof in the case X and Y are continuous random variables.
Z ∞
E(E(X|Y )) = E(X|Y = y)fY (y)dy
−∞
Z ∞ Z ∞
= xfX|Y (x|y)dx fY (y)dy
−∞ −∞
Z ∞Z ∞
= xfX|Y (x|y)fY (y)dxdy
Z−∞
∞
−∞
Z ∞
= x fXY (x, y)dydx
−∞ −∞
Z ∞
= xfX (x)dx = E(X)
−∞
Example 78.5
Suppose that X|Θ has a Poisson distribution with parameter θ and Θ has
a Gamma distribution with parameters α and β. Find E(X).
Solution.
We have
E(X) = E[E(X|Θ)] = E(Θ) = αβ
Now, for any function g(x, y), the conditional expected value of g given
Y = y is, in the continuous case,
Z ∞
E(g(X, Y )|Y = y) = g(x, y)fX|Y (x|y)dx
−∞
if the integral exists.
Example 78.6
Show that
E[E(g(X, Y )|Y )] = E[g(X, Y )].
Solution.
We have
Z ∞
E[E(g(X, Y )|Y )] = E(g(X, Y )|Y = y)fY (y)dy
−∞
Z ∞ Z ∞
= g(x, y)fX|Y (x|y)dx fY (y)dy
Z−∞
∞ Z ∞
−∞
= g(x, y)fX|Y (x|y)fY (y)dxdy

−∞ −∞
Z ∞ Z ∞
= g(x, y) fXY (x, y)dydx
−∞ −∞
Z ∞
= g(x, y)fX (x)dx = E[g(X, Y )]
−∞
The Conditional Variance

Next, we introduce the concept of conditional variance. Just as we have
defined the conditional expectation of X given that Y = y, we can define
the conditional variance of X given Y as follows
Var(X|Y = y) = E[(X − E(X|Y ))2 |Y = y].
Note that the conditional variance is a random variable since it is a function

of Y.
Proposition 78.1
Let X and Y be random variables. Then
(a) Var(X|Y ) = E(X 2 |Y ) − [E(X|Y )]2
(b) E(Var(X|Y )) = E[E(X 2 |Y ) − (E(X|Y ))2 ] = E(X 2 ) − E[(E(X|Y ))2 ].
(c) Var(E(X|Y )) = E[(E(X|Y ))2 ] − (E(X))2 .
(d) Law of Total Variance: Var(X) = E[Var(X|Y )] + Var(E(X|Y )).
Proof.
(a) We have
Var(X|Y ) =E[(X − E(X|Y ))2 |Y ]

=E[(X 2 − 2XE(X|Y ) + (E(X|Y ))2 |Y ]
=E(X 2 |Y ) − 2E(X|Y )E(X|Y ) + (E(X|Y ))2
=E(X 2 |Y ) − [E(X|Y )]2
(b) Taking E of both sides of the result in (a) we find
E(Var(X|Y )) = E[E(X 2 |Y ) − (E(X|Y ))2 ] = E(X 2 ) − E[(E(X|Y ))2 ].
(c) Since E(E(X|Y )) = E(X) we have
Var(E(X|Y )) = E[(E(X|Y ))2 ] − (E(X))2 .
(d) The result follows by adding the two equations in (b) and (c)
Example 78.7
Suppose that X and Y have joint distribution
( 2
3y
x3
0<y<x<1
fXY (x, y) =
0 otherwise
Find E(X), E(X 2 ), V ar(X), E(Y |X), V ar(Y |X), E[V ar(Y |X)], V ar[E(Y |X)],
and V ar(Y ).
Solution.
First we find marginal density functions.
Z x 2
3y
fX (x) = 3
dy = 1, 0 < x < 1
0 x
Z 1 2
3y 3
fY (y) = 3
dx = (1 − y 2 ), 0 < y < 1
y x 2
Now, Z 1
1
E(X) = xdx =
0 2
Z 1
1
E(X 2 ) = x2 dx =
0 3
Thus,
1 1 1
V ar(X) = − = .
3 4 12
Next, we find conditional density of Y given X = x
fXY (x, y) 3y 2
fY |X (x|y) = = 3 , 0<x<y<1
fX (x) x
Hence,
x
3y 3
Z
3
E(Y |X = x) = 3
dx = x
0 x 4
and
x
3y 4
Z
2 3
E(Y |X = x) = dx = x2
0 x3 5
Thus,
3 9 3
V ar(Y |X = x) = E(Y 2 |X = x) − [E(Y |X = x)]2 = x2 − x2 = x2
5 16 80
Also,

3 9 9 1 3
V ar[E(Y |X)] = V ar x = V ar(X) = × =
4 16 16 12 64
and
3 2 3 3 1 1
E[V ar(Y |X)] = E X = E(X 2 ) = × = .
80 80 80 3 80
Finally,
19
V ar(Y ) = V ar[E(Y |X)] + E[V ar(Y |X)] =
320
Example 78.8 ‡
An actuary for an automobile insurance company determines that the dis-
tribution of the annual number of claims for an insured chosen at random is
modeled by the negative binomial distribution with mean 0.2 and variance
0.4.
The number of claims for each individual insured has a Poisson distribution
and the means of these Poisson distributions are gamma distributed over
the population of insureds.
Calculate the variance of this gamma distribution
Solution.
Let N be the annual number of claims. We are given that E(N ) = E(N |Γ) =
E(Γ) = 0.2. By the law of total variance, we have
0.4 = Var(N ) = E[Var(N |Γ)] + Var(E(N |Γ)) = E(Γ) + Var(Γ).
Solving for Var(Γ) we find Var(Γ) = 0.4 − 0.2 = 0.2

Practice Problems
Problem 78.1
Suppose that X is uniformly distributed on the interval [0, 1] and that, given
X = x, Y is uniformly distributed on the interval [1 − x, 1].
(a) Determine the joint density fXY (x, y).
(b) Find the probability P (Y ≥ 21 ).
Problem 78.2
The joint density of X and Y is given by
15
fXY (x, y) = 2 x(2 − x − y) 0 ≤ x, y ≤ 1
0 otherwise
Compute the conditional density of X, given that Y = y for 0 ≤ y ≤ 1.
Problem 78.3
The joint density function of X and Y is given by
( −x
e y e−y
fXY (x, y) = y x ≥ 0, y ≥ 0
0 otherwise
Compute P (X > 1|Y = y).
Problem 78.4
Let Y be a random variable with a density fY given by
α−1
yα y>1
fY (y) =
0 otherwise
where α > 1. Given Y = y, let X be a random variable which is Uniformly

distributed on (0, y).
(a) Find the marginal distribution of X.
(b) Calculate E(Y |X = x) for every x > 0.
Problem 78.5

8xy 0 < x < y < 1
fXY (x, y) =
0 otherwise
Find E(X|Y ) and E(Y |X).

Problem 78.6
21 2 2
fXY (x, y) = 4 x y x <y <1
0 otherwise
Find E(Y ) in two ways.
Problem 78.7
The stock prices of two companies at the end of any given year are modeled
with random variables X and Y that follow a distribution with joint density
function
2x 0 < x < 1, x < y < x + 1
fXY (x, y) =
0 otherwise
What is the conditional variance of Y given that X = x?
Problem 78.8
Let X be a random variable with mean 3 and variance 2, and let Y be
a random variable such that for every x, the conditional distribution of Y
given X = x has a mean of x and a variance of x2 .
What is the variance of the marginal distribution of Y ?
Problem 78.9
The number of stops X in a day for a delivery truck driver is Poisson with
mean λ. Conditional on their being X = x stops, the expected distance
driven by the driver Y is Normal with a mean of αx miles, and a standard
deviation of βx miles.
Give the mean and variance of the numbers of miles she drives per day.
79 BAYESIAN CREDIBILITY WITH DISCRETE PRIOR 581
79 Bayesian Credibility with Discrete Prior
Let’s recall the credibility problem in Section 77: For a particular policy-
holder, we have the observed past losses X1 , X2 , · · · , Xn and we are inter-
ested in setting the premium to cover the loss of the next exposure unit
(next year) Xn+1 . We assume that the risk parameter θ (which is unknown)
associated with the policyholder comes from a prior distribution π(θ) and
that the losses X1 , X2 , · · · , Xn+1 are conditionally independent, that is the
Xi |Θ are independent, but not necessarily identically distributed.
In order to predict the claim in the (n+1) period we normally condition on θ,

but θ for the n+1 period is unknown so we condition on Xn+1 . The resulting
distribution is the predictive distribution fXn+1 |X (xn+1 (xn+1 |x) (intro-
duced in Section 64) where X = (X1 , X2 , · · · , Xn )T and x = (x1 , x2 , · · · , xn )T .
Using the terms discussed in Section 64, the model distribution of X

given Θ is
n
Y
fX|Θ (x|θ) = fXj |θ (xj |θ).
j=1
The joint distribution of X given Θ is

 
n
Y
fX,Θ (x, θ) =  fXj |θ (xj |θ) π(θ).
j=1
The marginal distribution of X is obtained by integrating the joint dis-

tribution with respect to θ to obtain
 
Z n
Y
fX (x) =  fXj |θ (xj |θ) π(θ)dθ.
j=1
The posterior distribution is by Bayes’ Theorem

 
n
fX|Θ (x|θ) 1  Y
πΘ|X (θ|x) = = fXj |θ (xj |θ) π(θ).
fX (x) fX (x)
j=1
Thus, the predictive distribution can be expressed as

 
Z n+1
1 Y
fXn+1 |X (xn+1 |x) =  fXj |θ (xj |θ) π(θ)
fX (x)
j=1
 
Yn
 fXj |θ (xj |θ) π(θ)
Z
j=1
= fXn+1 |Θ (xn+1 |θ) dθ
fX (x)
Z
= fXn+1 |Θ (xn+1 |θ)πΘ|X (θ|x)dθ.
The mean of the predicitive distribution, also known as the Bayesian pre-
mium or Bayesian estimate, is what we would charge to cover the loss
Xn+1 . It is given by
Z
E[Xn+1 |X = x] = xn+1 fXn+1 |X (xn+1 |x)dxn+1
Z Z
= xn+1 fXn+1 |Θ (xn+1 |θ)πΘ|X (θ|x)dθ dxn+1
Z Z
= xn+1 fXn+1 |Θ (xn+1 |θ)dxn+1 πΘ|X (θ|x)dθ
Z
= µn+1 (θ)πΘ|X (θ|x)dθ
where
Z
µn+1 (θ) = E(Xn+1 |Θ = θ) = xn+1 fXn+1 |Θ (xn+1 |θ)dxn+1
is the hypothetical mean, the premium we charge if we knew θ. Taking

the expectation of the previous equation, we find
E(Xn+1 ) = E[E(Xn+1 |Θ)] = E[µn+1 (Θ)].
This is the premium we charge if we knew nothing about the policyholder.

It is called the pure or collective premium.. Note that this premium
is independent of the risk parameter θ and does not depend on the data
collected from the policyholder.
Remark 79.1
In the case Θ is discrete, the integrals above are replaced by sums.
Example 79.1 ‡
You are given the following for a dental insurer:
(I) Claim counts for individual insureds follow a Poisson distribution.
(ii) Half of the insureds are expected to have 2.0 claims per year.
(iii) The other half of the insureds are expected to have 4.0 claims per year.
A randomly selected insured has made 4 claims in each of the first two policy
years. Determine the Bayesian estimate of this insured’s claim count in the
next (third) policy year.
Solution.
Let X be the claim count for an individual. We are given that conditional
claim count X|Θ is Poisson with mean Θ. Let Xn be the number of claims
in year n. We want to find E(X3 |X1 , X2 ) where x1 = x2 = 4.
The prior distribution is: π(2) = 0.5 and π(4) = 0.5.
fX (4, 4) =fX1 |Θ (4|θ = 2)fX2 |Θ (4|θ = 2)π(2) + fX1 |Θ (4|θ = 4)fX2 |Θ (4|θ = 4)π(4)
−2 4 2 −4 4 2
e 2 e 4
= (0.5) + (0.5)
4! 4!
=0.02315.
The posterior distribution is:
2
e−2 24
fX|Θ (x|2)π(2) 4! (0.5)
πΘ|X (2|x) = = = 0.1758
fX (x) 0.02315
and
πΘ|X (4|x) = 1 − 0.1758 = 0.8242.
Finally, the Bayesian premium is
E(X3 |X1 , X2 ) =E(X3 |Θ = 2)πΘ|X (2|x) + E(X3 |Θ = 4)πΘ|X (4|x)
=2(0.1758) + 4(0.8242) = 3.6484
Example 79.2 ‡
You are given:
Class Number of Claim Count Probabilities

Insureds 0 1 2 3 4
1 1 1
1 3000 3 3 3 0 0
2 2000 0 16 23 16 0
1 2 1
3 1000 0 0 6 3 6
A randomly selected insured has one claim in Year 1. Determine the ex-
pected number of claims in Year 2 for that insured.
Solution.
Let Xn denote the number of claims in Year n. We are asked to find
E(X2 |X1 = 1). The parameter θ stands for the class. The prior distribution
is π(1) = 3000 1 1 1
6000 = 2 , π(2) = 3 , and π(3) = 6 . The marginal distribution
evaluated at x1 = 1 is
f (1) =f (1|1)π(1) + f (1|2)π(2) + f (1|3)π(3)
1 1 1 1 2
= · + · = .
3 2 6 3 9
1
f (1|1)π(1) 6 3
π(1|1) = = 2 = 4
f (1) 9
1
f (1|2)π(2) 18 1
π(2|1) = = 2 = 4
f (1) 9
f (3|1)π(1) 0
π(3|1) = = 2 = 0.
f (1) 9
Thus,
E(X2 |X1 = 1) =E(X2 |1)π(1|1) + E(X2 |2)π(2|1) + E(X2 |3)π(3|1)

1 1 3 1 2 1 1
= 1 +2 + 1 +2 +3
3 3 4 6 3 6 4
=1.25
Example 79.3 ‡
You are given the following information about six coins:
Coin (θ) Probability of Heads

1-4 0.50
5 0.25
6 0.75
A coin is selected at random and then flipped repeatedly. Let Xi denote the
outcome of the ith flip, where ”1” indicates heads and ”0” indicates tails.
The following sequence is obtained:
S = {X1 , X2 , X3 , X4 } = {1, 1, 0, 1}.
Determine E(X5 |S) using Bayesian analysis.
Solution.
The prior distribution is
4 1
π(θ1 ) = 6 π(θ2 ) = 6 π(θ3 ) = 61 .
f (S|θ1 ) =(0.5)4 = 0.0625

f (S|θ2 ) =(0.25)2 (0.75)(0.25) = 0.011719
f (S|θ3 ) =(0.75)2 (0.25)(0.75) = 0.105469.

4
f (S, θ1 ) = (0.0625) = 0.04167
6
1
f (S, θ2 ) = (0.011719) = 0.00195
6
1
f (S, θ3 ) = (0.105469) = 0.01758.
6
0.04167
f (θ1 |S) = = 0.68088
0.04167 + 0.00195 + 0.01758
0.00195
f (θ2 |S) = = 0.03186
0.04167 + 0.00195 + 0.01758
0.01758
f (θ3 |S) = = 0.28726.
0.04167 + 0.00195 + 0.01758
Thus,
E(X5 |S) =E(X3 |θ1 )f (θ1 |S) + E(X3 |θ2 )f (θ2 |S) + E(X3 |θ3 )f (θ3 |S)
=0.5(0.68088) + 0.25(0.03186) + 0.75(0.28726) = 0.5639
Example 79.4 ‡
For a particular policy, the conditional probability of the annual number of
claims given Θ = θ, and the probability distribution of Θ are as follows:
Number of claims 0 1 2
Probability 2θ θ 1 − 3θ
θ 0.10 0.30
Probability 0.80 0.20
One claim is observed in Year 1. Calculate the Bayesian credibility estimate

of the number of claims in Year 2.
Solution.
The marginal distribution is Let N denote the annual number of claims. We
have
fN (1) = f (1|0.10)π(0.10)+f (1|0.30)π(0.30) = (0.10)(0.80)+(0.30)(0.20) = 0.14.
0.80(0.10) 4
πΘ|N (0.10|1) = =
0.14 7
0.20(0.30) 3
πΘ|N (0.30|1) = = .
0.14 7
The Bayesian credibility estimate of the number of claims in Year 2 is
E(N2 |N1 ) =E(N2 |Θ = 0.10)πΘ|N (0.10|1) + E(N2 |Θ = 0.30)πΘ|N (0.30|1)

=[0(2)(0.10) + 1(0.10) + 2(1 − 3 × 0.10)](4/7)
+[0(2)(0.30) + 1(0.30) + 2(1 − 3 × 0.30)](3/7) = 1.071
Example 79.5 ‡
You are given:
(i) The claim count and claim size distributions for risks of type A are:
Number of Claims Probabilities

Claim Size Probabilities
0 4/9
500 1/3
1 4/9
1235 2/3
2 1/9
(ii) The claim count and claim size distributions for risks of type B are:

0 1/9
250 2/3
1 4/9
328 1/3
2 4/9
(iv) Claim counts and claim sizes are independent within each risk type.
A randomly selected risk is observed to have total annual losses of 500.
Determine the Bayesian premium for the next year for this same risk.
Solution.
The prior parameter represents the type of risk so that either Θ = A or
Θ = B. The prior distribution is Pr(A) = Pr(B) = 0.5. For each of the two
classes, total annual loss L has a compound distribution. We want
E(L2 |L1 = 500) = E(L2 |A)Pr(A|L1 = 500) + E(L2 |B)Pr(B|L1 = 500).
We have
E(L|A) =E(N |A)E(X|A) = [0(4/9) + 1(4/9) + 2(1/9)][500(1/3) + 1235(2/3)] = 660

E(L|B) =E(N |B)E(X|B) = [0(1/9) + 1(4/9) + 2(4/9)][250(2/3) + 328(1/3)] = 368
Pr(L = 500|A)Pr(A)
Pr(A|L = 500) =
Pr(L = 500|A)Pr(A) + Pr(L = 500|B)Pr(B)
Pr(L = 500|A) =Pr(N = 1)Pr(X) = 500)
=(4/9)(1/3) = 4/27
Pr(L = 500|B) =Pr(N = 2)[Pr(X) = 250)]2
=(4/9)(2/3)2 = 16/81
(4/27)(1/2)
Pr(A|L = 500) = = 3/7
(4/27)(1/2) + (16/81)(1/2)
Pr(B|L = 500) =1 − 3/7 = 4/7.
Thus,
E(L2 |L1 = 500) =E(L2 |A)Pr(A|L1 = 500) + E(L2 |B)Pr(B|L1 = 500)

=660(3/7) + 368(4/7) = 493
Example 79.6 ‡
Two eight-sided dice, A and B, are used to determine the number of claims
for an insured. The faces of each die are marked with either 0 or 1, repre-
senting the number of claims for that insured for the year.
Die Pr(Claims=0) Pr(Claims=1)
A 1/4 3/4
B 3/4 1/4
Two spinners, X and Y, are used to determine claim cost. Spinner X has
two areas marked 12 and c. Spinner Y has only one area marked 12.
Spinner Pr(Cost=12) Pr(Ccost=c)
X 1/2 1/2
Y 1 0
To determine the losses for the year, a die is randomly selected from A and
B and rolled. If a claim occurs, a spinner is randomly selected from X and
Y and spun. For subsequent years, the same die and spinner are used to
determine losses.
Losses for the first year are 12. Based upon the results of the first year, you
determine that the expected losses for the second year are 10.
Calculate c.
Solution.
The prior parameter Θ can be one of AX, BX, AY, and BY. The prior
distribution is
1
π(AX) = π(AY ) = π(BX) = π(BY ) =
4
We have
10 =E(L2 |L1 = 12) = E(L2 |AX)Pr(AX|L1 = 12) + E(L2 |AY )Pr(AY |L1 = 12)
+E(L2 |BX)Pr(BX|L1 = 12) + E(L2 |BY )Pr(BY |L1 = 12)

1 3 1 3
E(L2 |AX) =12 +c
2 4 2 4
3
= (12 + c)
8
3 3
E(L2 |AY ) =12(1) + c(0) =9
4 4

1 1 1 1
E(L2 |BX) =12 +c
2 4 2 4
1
= (12 + c)
8
1 1
E(L2 |BY ) =12(1) + c(0) =3
4 4
Pr(L1 = 12) =Pr(L1 = 12|AX)Pr(AX) + Pr(L1 = 12|AY )Pr(AY )
=Pr(L1 = 12|BX)Pr(BX) + Pr(L1 = 12|BY )Pr(BY )

3 1 1 3 1 1 1 1 1 1
= + (1) + + (1)
4 2 4 4 4 4 2 4 4 4
3
=
8
3 1 1

Pr(L1 = 12|AX)Pr(AX) 4 2 4 1
Pr(AX|L1 = 12) = = 3 =
Pr(L1 = 12) 8
4
3 1

Pr(L1 = 12|AY )Pr(AY ) 4 (1) 4 1
Pr(AY |L1 = 12) = = 3 =
Pr(L1 = 12) 8
2
1 1 1

Pr(L1 = 12|BX)Pr(BX) 4 2 4 1
Pr(BX|L1 = 12) = = 3 =
Pr(L1 = 12) 8
12
1 1

Pr(L1 = 12|BY )Pr(BY ) 4 (1) 4 1
Pr(BY |L1 = 12) = = 3 = .
Pr(L1 = 12) 8
6
Thus,

3 1 1 1 1 1
(12 + c) +9 + (12 + c) +3 = c.
8 4 2 8 12 6
Solving this equation, we find c = 36
Example 79.7 ‡
For a risk, you are given:
(i) The number of claims during a single year follows a Bernoulli distribution
with mean p.
(ii) The prior distribution for p is uniform on the interval [0, 1].
(iii) The claims experience is observed for a number of years.
(iv) The Bayesian premium is calculated as 1/5 based on the observed claims.
Which of the following observed claims data could have yielded this calcu-
lation?
(A) 0 claims during 3 years
(B) 0 claims during 4 years
(C) 0 claims during 5 years
(D) 1 claim during 4 years
(E) 1 claim during 5 years
Solution.
Let xi be the number of claims in year i where i = 1, 2, · · · , n and xi = 0, 1.
Let x = x1 + · · · + xn be the number of claims in n years. We have that
Xi |p is a Bernoulli distribution with probability function
f (Xi |p) = pxi (1 − p)1−xi , xi = 0, 1.
The Bayesian premium is
Z 1 Z 1
E(Xn+1 |x1 , x2 , · · · , xn ) = E(Xn+1 |p)f (p|x1 , x2 , · · · , xn )dp = pf (p|x1 , x2 , · · · , xn )dp.
0 0
We have
n
Y
f (x1 , x2 , · · · , xn |p) = pxi (1 − p)1−xi
i=1
=p (1 − p)n−x
x
f (x1 , x2 , · · · , xn |p)π(p)
f (p|x1 , x2 , · · · , xn ) =
f (x1 , x2 , · · · , xn )
px (1 − p)n−x
=
f (x1 , x2 , · · · , xn )
Z 1
f (x1 , x2 , · · · , xn ) = px (1 − p)n−x dp
0
Γ(x + 1)Γ(n − x + 1) 1
Z
Γ(n + 2) 1
= px+1 (1 − p)n−x+1−1 dp
Γ(n + 2) 0 Γ(x + 1)Γ(n − x + 1) p
Γ(x + 1)Γ(n − x + 1)
=
Γ(n + 2)
Γ(n + 2) 1
f (p|x1 , x2 , · · · , xn ) = px+1 (1 − p)n−x+1−1 .
Γ(x + 1)Γ(n − x + 1) p
Thus, the posterior distribution is a beta distribution with a = x + 1, b =

n − x + 1, and θ = 1. Moreover, E(Xn+1 |x1 , x2 , · · · , xn ) is the mean of this
distribution which implies
x+1 1
= .
n+2 5
This is satisfied for x = 0 and n = 3 so the answer is (A)
Practice Problems
Problem 79.1
Drivers are classified as good (G), average (A), or bad (B).
• Good drivers make up 70% of the population and for a driver in this class,
the probability of having 0 claim in one year is 0.65, 1 claim is 0.25, and 2
claims is 0.10.
• Average drivers make up 20% of the population and for a driver in this
class, the probability of having 0 claim in one year is 0.40, 1 claim is 0.40,
and 2 claims is 0.20.
• Bad drivers make up 10% of the population and for a driver in this class,
claims is 0.20.
For a policyholder, the risk parameter is the classification of the individual
as G, A, or B. For a particular policyholder, it has been observed that x1 = 1
and x2 = 2.
(a) Write the prior distribution of this model.
(b) Find the model distribution of x = (1, 2)T .
Problem 79.2
In Problem 79.1, answer the following questions:
(a) Find the marginal probability of X.
(b) Find the joint distribution of X1 , X2 , X3 given x = (1, 2)T .
Problem 79.3
(a) Find the predictive distribution given x = (1, 2)T .
(b) Find the posterior probabilities.
Problem 79.4
(a) Determine the hypothetical means.
(b) Determine the pure of the collective premium.
Problem 79.5
(a) Determine the Bayesian premium without using the hypothetical means.
(b) Determine the Bayesian premium by using the hypothetical means.
Problem 79.6 ‡
In a certain town the number of common colds an individual will get in a
year follows a Poisson distribution that depends on the individual’s age and
smoking status. The distribution of the population and the mean number
of colds are as follows:
Proportion of population Mean number of colds

Children (C) 0.30 3
Adult Non-Smokers (ANS) 0.60 1
Adult Smokers (AS) 0.10 4
Calculate the conditional probability that a person with exactly 3 common

colds in a year is an adult smoker.
Problem 79.7 ‡
You are given:
(i) The annual number of claims on a given policy has the geometric distri-
bution with parameter β.
(ii) One-third of the policies have β = 2, and the remaining two-thirds have
β = 5.
A randomly selected policy had two claims in Year 1.
Calculate the Bayesian expected number of claims for the selected policy in
Year 2.
Problem 79.8 ‡
An insurance company sells three types of policies with the following char-
acteristics:
Type of Policy Proportion of Total Annual Claim
Policies Frequency
I 5% Poisson with λ = 0.25
II 20% Poisson with λ = 0.50
III 75% Poisson with λ = 1.00
A randomly selected policyholder is observed to have a total of one claim

for Year 1 through Year 4.
For the same policyholder, determine the Bayesian estimate of the expected
number of claims in Year 5.
Problem 79.9 ‡
You are given:
(i) Claim sizes follow an exponential distribution with mean θ.
(ii) For 80% of the policies, θ = 8.
(iii) For 20% of the policies, θ = 2.

A randomly selected policy had one claim in Year 1 of size 5.
Calculate the Bayesian expected claim size for this policy in Year 2.
Problem 79.10 ‡
You are given:
(i) Two classes of policyholders have the following severity distributions:
Probability of claim Probability of claim

Amount of claim amount for Class 1 amount for Class 2
250 0.5 0.7
2500 0.3 0.2
60000 0.2 0.1
(ii) Class 1 is twice as likely to be observed as Class 2.

Determine the Bayesian estimate of the expected value of a second claim
from the same policyholder.
Problem 79.11 ‡
You are given:
(i) An individual automobile insured has annual claim frequencies that fol-
low a Poisson distribution with mean λ.
(ii) An actuary’s prior distribution for the parameter λ has probability den-
sity function:
1 λ
f (λ = 0.5[5e−5λ + e− 5 ].
5
(iii) In the first policy year, no claims were observed for the insured.
Determine the expected number of claims in the second policy year.
80 Bayesian Credibility with Continuous Prior
In this section, we consider the Bayesian credibility approach to models

where the prior distribution is continuous.
Example 80.1
Claim amount is assumed to be exponential with mean Θ1 . The prior distri-
bution Θ is assumed to be Gamma with parameters α = 5 and β = 0.0005.
Suppose a person has claims in the amount of $2000,$1000, and $3000.
(a) Provide a mathematical description of this model.
(b) Determine the predictive distribution of the fourth claim.
(c) Determine the posterior distribution of Θ.
(d) Determine the Bayesian premium without using the hypothetical means.
(e) Determine the Bayesian premium by using the hypothetical means.
Solution.
(a) The claims amount distribution (model distribution) is given by
fX|Θ (x|θ) = θe−θx .
The risk parameter (prior) distribution is given by
20005 θ4 e−2000θ
π(θ) = .
24
(b) The marginal density at the observed values is
∞
θ1 20002 e−2000θ
Z
f (2000, 1000, 3000) = (θe−2000θ )(θe−1000θ )(θe−3000θ ) dθ
0 4!
20005 ∞ 7 −8000θ
Z
= θ e dθ
4! 0
20005 7! ∞ θ7 80008 e−8000θ
Z
= dθ
80008 4! 0 Γ(8)
| {z }
1
7! 20005
= .
4! 80008
80 BAYESIAN CREDIBILITY WITH CONTINUOUS PRIOR 595
Similarly,
∞
θ1 20002 e−2000θ
Z
f (2000, 1000, 3000, x4 ) = (θe−2000θ )(θe−1000θ )(θe−3000θ )(θe−x4 θ ) dθ
0 4!
20005 ∞ 8 −θ(8000+x4 )
Z
= θ e dθ
4! 0
20005 8! ∞ (8000 + x4 )9 θ8 e−θ(8000+x4 )
Z
= dθ
(8000 + x4 )9 4! 0 Γ(9)
| {z }
1
8! 20005
= .
4! (8000 + x4 )9
The predictive distribution is given by
8! 20005
4! (8000+x4 )9 8(80008 )
f (x4 |2000, 1000, 3000) = 7! 20005
=
(8000 + x4 )9
4! 80008
which is a type 2 Pareto distribution with parameters α8 and β = 8000.

(c) The posterior distribution of Θ is given by
f (2000, 1000, 3000, θ)

π(θ|2000, 1000, 3000) =
f (2000, 1000, 3000)
1 20002 e−2000θ
(θe−2000θ )(θe−1000θ )(θe−3000θ ) θ 4!
= 7! 20005
4! 80008
20005 7 −8000θ
θ e
= 4! 7! 20005
4! 80008
80008 7 −8000θ
= θ e .
7!
(d) We have
∞
8(80008 )
Z
E[X4 |2000, 1000, 3000] = x4 dx4
(8000 + x4 )9
|0 {z }
mean of Pareto
8000 8000
= = .
8−1 7
(e) We have
Z ∞
E[X4 |2000, 1000, 3000] = µ5 (θ)π(θ|2000, 1000, 3000)dθ
0
∞
1 80008 7 −8000θ
Z
= θ e
0 θ 7!
Z ∞
80008 6! 80007 θ6 e−8000θ
= dθ
7! 80007 0 Γ(7)
| {z }
1
8000
=
7
Example 80.2 ‡
You are given:
(i) The number of claims for each policyholder has a binomial distribution
with parameters m = 8 and q.
(ii) The prior distribution of q is beta with parameters a (unknown), b = 9,
and θ = 1.
(iii) A randomly selected policyholder had the following claims experience:
Year Number of Claims

1 2
2 k
(iv) The Bayesian credibility estimate for the expected number of claims in
Year 2 based on the Year 1 experience is 2.54545.
(v) The Bayesian credibility estimate for the expected number of claims in
Year 3 based on the Year 1 and Year 2 experience is 3.73333.
Determine k.
Solution.
By 0
Pn Problem 0
66.1, Q|N has
Pna beta distribution with parameters a = a +
i=1 xi , b = b + nm − i=1 xi and θ = 1, where x1 , x2 , · · · , xn are past
data.
By (iv), we have n = 1 and x1 = 2 so that a0 = a + 2 and b0 = 9 + 8 − 2 = 15.
The Bayesian estimate is

Z ∞
Γ(a + 17)
E(N2 |N1 ) = mq q a+2 (1 − q)14 dq
0 Γ(a + 2)Γ(15)
m(a + 2 ∞
Z
Γ(a + 18)
= mq q a+3 (1 − q)14 dq
a + 17 0 Γ(a + 3)Γ(15)
m(a + 2
= .
a + 17
Thus,
8(a + 2)
= 2.54545 =⇒ a ≈ 5.
a + 17
By Problem 66.1, Q|N has a beta distribution with parameters a0 = a +
P n 0
P n
i=1 xi , b = b + nm − i=1 xi and θ = 1, where x1 , x2 , · · · , xn are past
data.
By (v), we have n = 2 and x1 = 2 and x2 = k. Thus, a0 = a + 2 + k and
b0 = 9 + 16 − 2 − k = 23 − k. The Bayesian estimate is
Z ∞
Γ(a + 25)
E(N2 |N1 ) = mq q a+2+k (1 − q)22−k dq
0 Γ(a + 2 + k)Γ(23 − k)
m(a + 2 + k ∞
Z
Γ(a + 26)
= mq q a+3+k (1 − q)22−k dq
a + 25 0 Γ(a + 3 + k)Γ(23 − k)
m(a + 2 + k
= .
a + 25
Thus,
8(5 + 2 + k
= 3.73333 =⇒ k ≈ 7
25 + 5
Practice Problems
Problem 80.1
Suppose an individual’s claim amounts are given by an exponential distri-
bution with mean Λ where Λ is an inverse Gamma with parameters α = 2
and θ = 15. Last year claim was $12.
(a) Provide a mathematical description of this model.
(b) Determine the predictive distribution of next year claim.
(c) Determine the posterior distribution of Λ.
(d) Determine the Bayesian premium.
Problem 80.2 ‡
You are given:
(i) The annual number of claims for a policyholder follows a Poisson distri-
bution with mean Λ.
(ii) The Prior distribution of Λ is gamma with probability density function:
(2λ)5 e−2λ
π(λ) = , λ > 0.
24λ
An insured is selected at random and observed to have x1 = 5 claims during
Year 1 and x2 = 3 claims during Year 2. Determine E(Λ|x1 = 5, x2 = 3).
Problem 80.3
You are given:
(i) X|P is a binomial distribution with parameters (10, p).
(ii) The prior distribution of P is
π(p) = 2p, 0 < p < 1.
Find the posterior distribution of P given X1 = 4.
Problem 80.4
(i) Show that the negative binomial distribution with parameters r and β
can be expressed in the form
Γ(r + k) r
pk = q (1 − q)k , k = 0, 1, 2, · · · .
Γ(r)Γ(k)
(ii) Suppose that X|Q is negative binomial with parameters r and q. Suppose
also that Q is beta with (a, b, 1). Find the posterior distribution of Q.
Problem 80.5
The amount of a claim X|Λ has the normal distribution with mean θ and
known variance σ12 . The risk parameter Λ has a normal distribution with
mean µ and and variance σ22 . Find the posterior distribution of Θ.
Problem 80.6 ‡
You are given:
(i) The parameter Λ has an inverse gamma distribution with probability
density function:
10
g(λ) = 500λ−4 e− λ . λ > 0.
(ii) The size of a claim has an exponential distribution with probability
density function:
x
f (x|λ) = λ−1 e− λ , x > 0, λ > 0.
For a single insured, two claims were observed that totaled 50.
Determine the expected value of the next claim from the same insured.
81 Bühlman Credibility Premium

The purpose of this section is to find an estimate of the hypothetical mean
µn+1 (θ) = E(Xn+1 |Θ). The estimation procedure disussed here is the one
developed by Bühlmann.
The idea is to estimate µn+1 (θ) with a linear combination of the past data:
α0 + α1 X1 + · · · + αn Xn
where α0 , α1 , · · · , αn are to be determined.
The estimation is done with the linear least squares regression where the
square of the distance between the µn+1 (θ) and the estimator is to be min-
imized. That is, we want to find the αi0 s that minimize the function
n
X
Q = Q(α0 , α1 , · · · , αn ) = E[(µn+1 (Θ) − α0 − αi Xi )2 ].
i=1
To minimize Q, we take the first derivatives of Q with respect to the αi and

set them to zero. Thus,
n
∂Q X
= E[2(µn+1 (Θ) − α0 − αi Xi )(−1)] = 0
∂α0
i=1
which leads to
n
X
E[µn+1 (Θ)] = α0 + αi E(Xi ). (81.1)
i=1
Equation (81.1) is known as the unbiased equation since
E[µn+1 (Θ)] = E[E(Xn+1 |Θ)] = E(Xn+1 ).
Thus, we can write

n
X
E(Xn+1 ) = α0 + αi E(Xi ). (81.2)
i=1
Next, we have
n
∂Q X
= E[2(µn+1 (Θ) − α0 − αi Xi )(−Xi )] = 0
∂αi
i=1
81 BÜHLMAN CREDIBILITY PREMIUM 601
which leads to
n
X
E[µn+1 (Θ)Xi ] = α0 E(Xi ) + αj E(Xj Xi ).
j=1
However, we have
E[µn+1 (Θ)Xi ] =E{E[Xi µn+1 (Θ)|Θ]}

=E[µn+1 (Θ)E(Xi |Θ)]
=E[E(Xn+1 |Θ)E(Xi |Θ)]
=E[E(Xn+1 Xi |Θ)] (by independence)
=E(Xn+1 Xi ).
Hence,
n
X
E(Xn+1 Xi ) = α0 E(Xi ) + αj E(Xj Xi ). (81.3)
j=1
Multiplying (81.2) by E(Xi ) and subtracting the resulting equation from

(81.3) to obtain
n
X
Cov(Xi , Xn+1 ) = αj Cov(Xi , Xj ), i = 1, 2, · · · , n. (81.4)
j=1
Equations (81.2) and (81.4) are known as the normal equations. Solving
these n + 1 equations to yield the credibility premium
n
X
α̂0 + α̂i Xi . (81.5)
i=1
Example 81.1
You are given:
(i) E(Xi ) = 2 and Var(Xi ) = 3 for i = 1, 2, · · · , 20.
(ii) Cov(Xi , Xj ) = 1.5 for all i 6= j.
Determine the credibility premium.
Solution.
The unbiasedness equation yields
20
X
α̂0 + 2 α̂j = 2.
j=1
This implies
20
X α̂0
α̂j = 1 − .
2
j=1
On the other hand, for i = 1, 2, · · · , 20, we have

20
X
α̂j (1.5) + 3α̂i = 1.5
j=1
j 6= i
or equivalently
20
X
α̂j (1.5) + 1.5α̂i = 1.5.
j=1
Solving this last equation for α̂i we find

α̂0 α̂0
α̂i = 1 − 1 − = .
2 2
Summation over i from 1 to 20 yields

20
X
α̂i = 10α̂0
1=1
or equivalently
α̂0
1− = 10α̂0 .
2
Solving this equation for α̂0 we find
2
α̂0 = .
21
Hence,
1
α̂i =
21
and the credibility premium is
20
X 2
α̂0 + α̂i Xi = (1 + 10X)
21
i=1
81 BÜHLMAN CREDIBILITY PREMIUM 603
Remark 81.1
It is easy to check that the values α̂0 , α̂1 . · · · , α̂n also minimize
" #2 
 Xn 
Q1 = E E(Xn+1 |X) − α0 − αi Xi
 
i=1
and
n
X
Q2 = E[(Xn+1 − α0 − αi Xi )2 ].
i=1
That is, the credibility premium (81.5) is the best linear estimator of each
the hypothetical mean E(Xn+1 |Θ), the Bayesian premium (Xn+1 |X), and
Xn+1 .
Example 81.2 ‡
You are given the following information about a credibility model:
Bayesian estimate
First observation (T ) Pr(X1 = T ) E(X2 |X1 = T )
1 1/3 1.50
2 1/3 1.50
3 1/3 3.00
Determine the Bühlmann credibility estimate of the second observation,
given that the first observation is 1.
Solution.
Let X1 be the outcome of the first observation. By Problem 81.5, the
Bühlmann credibility estimate is of the form ZX1 +(1−Z)µ where X1 = 1, 2,
or 3.By Remark 8.1, the Bühlmann estimate is the least squares approxima-
tion to the Bayesian estimate. Thus, Z and µ are the minimizers of
1 1 1
f (Z, µ) = [1.50−Z−(1−Z)µ]2 + [1.50−2Z−(1−Z)µ]2 + [3.00−3Z−(1−Z)µ]2 .
3 3 3
Taking the derivative with respect to µ and setting it to zero, we find µ = 2.
Next, taking the derivative of f with respect to Z and setting it to zero, we
find
2(−Z + 0.5)(−1) + 2(0.5)(0) + 2(Z − 1)(1) = 0 =⇒ Z = 0.75.
Thus, the Bühlmann credibility estimate of the second observation, given
that the first observation is 1, is
ZX1 + (1 − Z)µ = 0.75(1) + (1 − 0.75)(2) = 1.25
Practice Problems
Problem 81.1
You are given:
(i) E(Xj ) = µ and Var(Xj ) = σ 2 for all j = 1, 2, · · · , n.
(ii) Cov(Xi , Xj ) = ρσ 2 for all i 6= j, where ρ is the coefficient of correlation.
Use the unbiasedness equation to show that
n
X α̂0
α̂j = 1 − .
µ
j=1
Problem 81.2
With the assumptions of Problem 81.1, show that the n equations (81.4)
lead to
Xn
ρ=ρ α̂j + α̂i (1 − ρ), i = 1, 2, · · · , n.
j=1
Problem 81.3
With the assumptions of Problem 81.1, show that
ρα̂0
α̂i = .
µ(1 − ρ)
Problem 81.4
(1−ρ)µ ρ
α̂0 = 1−ρ+nρ and α̂i = 1−ρ+nρ
Problem 81.5
n
X
α̂0 + α̂j Xj = (1 − Z)µ + ZX
j=1
where Z to be determined.
82 THE BÜHLMANN MODEL WITH DISCRETE PRIOR 605
82 The Bühlmann Model with Discrete Prior

The Bühlmann assumes that the past losses X1 , X2 , · · · , Xn have the same
mean and variance and that Xi |Θ are iid. We define the following:
• The hypothetical mean: µ(θ) = E[Xj |Θ = θ].
• The process variance: v(θ) = Var(Xj |Θ = θ).
• The expected value of the hypothetical mean22 : µ = E[µ(Θ)].
• The expected value of the process variance: v = E[v(Θ)].
• The variance of the hypothetical mean: a = Var[µ(Θ)] = E[µ2 (Θ)] −
E[µ(Θ)]2 .
Example 82.1
Find the (a) mean, (b) variance and (c) covariance of Xi .
Solution.
(a) The mean is given by
E(Xi ) = E[E(Xi |Θ)] = E[µ(Θ)] = µ.
(b) The variance is given by
Var(Xi ) =E[Var(Xi |Θ)] + Var[E(Xi |Θ)]

=E[v(Θ)] + Var[µ(Θ)] = v + a.
(c) For i 6= j, The covariance of Xi and Xj is given by
Cov(Xi , Xj ) =E(Xi , Xj ) − E(Xi )E(Xj )

=E[E(Xi Xj |Θ)] − E[µ(Θ)]2
=E[E(Xi |Θ)E(Xj |Θ)] − E[µ(Θ)]2 (by independence)
=E[µ2 (Θ)] − E[µ(Θ)]2
=Var[µ(Θ)] = a
a
Using Problem 81.5 with σ 2 = v + a and ρ = v+a , the Bühlmann credi-
bility factor is given by
nρ n
Z= =
1 − ρ + nρ n+k
22
Also known as the collective premium. It is the value that is used if there is no
information about past claims data.
where
v E[Var(Xj |Θ)]
k= = .
a Var[E(Xj |Θ)]
The Bühlmann credibility premium is given by
BP = ZX + (1 − Z)µ.
Some of the features of the Bühlmann model:

• As more data is gathered, i.e., n → ∞, the value of Z tends to 1 and the
weighting average tends toward the sample mean rather than the collective
premium, a feature that agrees with intuition.
• If the population is homogenepus with respect to Θ, then µ(Θ) are very
close in values so that they have very small variability. That is, a is very
small which implies a very large k and consequently Z approaches 0.
• If the population is heterogeneous more weight will be given to the sample
mean as it is better indicator of an individual’s future claims.
Example 82.2
Drivers are classified as good (G), average (A), or bad (B).
• Good drivers make up 70% of the population and for a driver in this class,
claims is 0.10.
• Average drivers make up 20% of the population and for a driver in this
class, the probability of having 0 claim in one year is 0.40, 1 claim is 0.40,
and 2 claims is 0.20.
• Bad drivers make up 10% of the population and for a driver in this class,
claims is 0.20.
For a policyholder, the risk parameter is the classification of the individual
as G, A, or B. For a particular policyholder, it has been observed that x1 = 1
and x2 = 2.
Determine the Bühlmann premium.
Solution.
By Problem 79.4(a), we have
µ(G) = E(Xi |G) = 0.45, µ(A) = E(Xi |A) = 0.80 µ(B) = E(Xi |B) = 0.70
π(G) = 0.70, π(A) = 0.20, π(B) = 0.10.
Hence,
X
µ= µ(θ)π(θ) = 0.45(0.70) + 0.80(0.20) + 0.70(0.10) = 0.545
θ
X
2
E[µ (Θ)] = µ(θ)2 π(θ) = 0.452 (0.70) + 0.802 (0.20) + 0.702 (0.10) = 0.31875
θ
a =E[µ2 (Θ)] − µ2 = 0.31875 − 0.5452 = 0.021725.
For the process variance, we have
v(G) =Var(Xi |G) = 02 (0.65) + 12 (0.25) + 22 (0.10) − 0.452 = 0.4475

v(A) =Var(Xi |A) = 02 (0.40) + 12 (0.40) + 22 (0.20) − 0.802 = 0.56
v(B) =Var(Xi |B) = 02 (0.50) + 12 (0.30) + 22 (0.20) − 0.702 = 0.61
X
v= v(θ)π(θ) = 0.4475(0.70) + 0.56(0.20) + 0.61(0.10) = 0.48625
θ
v 0.48625
k= = = 22.3820
a 0.021725
n 2
Z= = = 0.0820.
n+k 2 + 22.3820
Hence, the Bühlmann premium is
BP = ZX + (1 − Z)µ = 0.0820(1.5) + (1 − 0.0820)(0.545) = 0.62331.
This is the best linear approximation to the Bayesian premium of 0.6297

found in Problem 79.5
Example 82.3 ‡
You are given the following information on claim frequency of automobile
accidents for individual drivers:
Busienss use Pleasure use

Expected claims Claim variance Expected claims Claim variance
Rural 1.0 0.5 1.5 0.8
Urban 2.0 1.0 2.5 1.0
Total 1.8 1.06 2.3 1.12
You are also given:
(i) Each driver’s claims experience is independent of every other driver’s.
(ii) There are an equal number of business and pleasure use drivers.
Determine the Bühlmann credibility factor Z for a single driver.
Solution.
Let Θ be risk parameter with designations BR= Business Rural, BU= busi-
ness urban, PR = Pleasure Rural, and PU = Pleasure Urban. By double
expectation, we have
1.8 = E(B) = E[E(B|R, U )] = E(B|R)Pr(R)+E(B|U )Pr(U ) = Pr(R)+2Pr(U ).
Likewise,
2.3 = 1.5Pr(R) + 2.5Pr(U ).
Solving these two equations, we find Pr(R) = 0.2 and Pr(U ) = 0.8.
π(BR) =π(P R) = (0.5)(0.2) = 0.10

π(BU ) =π(P U ) = (0.5)(0.8) = 0.40.
Let X represent the claim frequency of auto accidents of a randomly selected

driver. Then, we have the following
µ(BR) =E(X|BR) = 1.0

µ(BU ) =E(X|BU ) = 2.0
µ(P R) =E(X|P R) = 1.5
µ(P U ) =E(X|P U ) = 2.5.
Thus,
X
µ= µ(θ)π(θ) = (0.10)(1.0) + (0.40)(2.0) + (0.10)(1.5) + (0.40(2.5) = 2.05
θ
and
X
a= µ(θ)2 π(θ) − µ2
θ
=(0.10)(1.0)2 + (0.40)(2.0)2 + (0.10)(1.5)2 + (0.40(2.5)2 − 2.052 = 0.2225.
For the process variance, we have
v(BR) =0.5
v(BU ) =1.0
v(P R) =0.8
v(P U ) =1.0.
Hence,
X
v= v(θ)π(θ) = 0.5(0.10) + 1.0(0.40) + 0.8(0.10) + 1.0(0.40) = 0.93.
θ
It follows that
v 0.93
k= = = 4.18
a 0.2225
and the credibility factor is
n 1
Z= = = 0.193
n+k 1 + 4.18
Example 82.4 ‡
For a particular policy, the conditional probability of the annual number of
claims given Θ = θ, and the probability distribution of Θ are as follows:
Probability 2θ θ 1 − 3θ
θ 0.05 0.30
Probability 0.80 0.20
Two claims are observed in Year 1. Calculate the Bühlmann credibility
estimate of the number of claims in Year 2.
Solution.
We have
E(Θ) =0.05(0.80) + 0.30(0.20) = 0.1
E(Θ2 ) =0.052 (0.80) + 0.302 (0.20) = 0.02
µ(θ) =E(N |Θ) = 0(2θ) + 1(θ) + 2(1 − 3θ) = 2 − 5θ
µ =E(2 − 5Θ) = 2 − 5E(Θ) = 2 − 5(0.1) = 1.5
a =Var(2 − 5Θ) = 25Var(Θ) = 25(0.02 − 0.12 ) = 0.25
v(θ) =Var(N |Θ) = 02 (2θ) + 12 (θ) + 22 (1 − 3θ) − (2 − 5θ)2 = 9θ − 25θ2
v =E(9Θ − 25Θ2 ) = 9(0.1) − 25(0.02) = 0.4
v 0.4
k= = = 1.6
a 0.25
1 5
Z= = .
1+k 13
The required estimate is

5 5
2Z + (1 − Z)µ = 2 + 1− (1.5) = 1.6923
13 13
Example 82.5 ‡
For a group of policies, you are given:
(i) The annual loss on an individual policy follows a gamma distribution
with parameters α = 4 and θ.
(ii) The prior distribution of θ has mean 600.
(iii) A randomly selected policy had losses of 1400 in Year 1 and 1900 in
Year 2.
(iv) Loss data for Year 3 was misfiled and unavailable.
(v) Based on the data in (iii), the Bühlmann credibility estimate of the loss
on the selected policy in Year 4 is 1800.
(vi) After the estimate in (v) was calculated, the data for Year 3 was located.
The loss on the selected policy in Year 3 was 2763.
Calculate the Bühlmann credibility estimate of the loss on the selected policy
in Year 4 based on the data for Years 1, 2 and 3.
Solution.
Let the annual loss be denoted by X. Then X|Θ has a Gamma distribution
with parameters α = 4 and θ. We have
µ(θ) =E(X|Θ) = αθ = 4θ
v(θ) =Var(X|Θ) = αθ2 = 4θ2
µ =E[µ(Θ)] = 4E(Θ) = 4(600) = 2400
v =E[v(Θ)] = 4E(Θ2 )
a =Var[µ(Θ)] = 16Var(Θ).
Based on Year 1 and Year 2, the average loss is
1400 + 1900
X= = 1650.
2
Based on (iv), we have
2
1650Z + (1 − Z)(2400) = 1800 =⇒ Z = 0.8 = =⇒ k = 0.5.
2+k
By (vi), we have
1400 + 1900 + 2763
X= = 2021
3
and
3 6
Z= = .
3 + 0.5 7
Thus, the Bühlmann credibility estimate of the loss on the selected policy
in Year 4 based on the data for Years 1, 2 and 3

6 6
(2021) + 1 − (2400) = 2075.14
7 7
Example 82.6 ‡
You are given:
(i) The annual number of claims for an individual risk follows a Poisson
(ii) For 75% of the risks, λ = 1.
(iii) For 25% of the risks, λ = 3.
A randomly selected risk had r claims in Year 1. The Bayesian estimate of
this riskfs expected number of claims in Year 2 is 2.98.
Determine the Bühlmann credibility estimate of the expected number of
claims for this risk in Year 2.
Solution.
Let X be the annual amount of claims and Λ be the prior parameter. Then
X|Λ has a Poisson distribution with mean λ. The prior distribution is:
π(1) = 0.75 and π(3) = 0.25.
e−1
f (r|1)π(1) r! (0.75)
π(1|r) = = e−1 e−3 3r
f (r|1)π(1) + f (r|3)π(3)
r! (0.75) + r! (0.25)
0.2759
=
0.2759 + 3r (0.1245)
0.2759 3r (0.1245)
π(3|r) =1 − = .
0.2759 + 3r (0.1245) 0.2759 + 3r (0.1245)
Thus,
2.98 =E(X2 |X1 = r) = E(X2 |Λ = 1)π(1|r) + E(X2 |Λ = 3)π(3|r)

0.2759 3r (0.1245)
=(1) + (3)
0.2759 + 3r (0.1245) 0.2759 + 3r (0.1245)
0.2759 + 0.3735(3 )r
= .
0.2759 + 3r (0.1245)
Cross multiply to obtain
0.82218 + 0.037103(3r ) = 0.2759 + 0.03735(3r ).

Rearranging the terms and solving, we find
0.54628 = 0.00025(3r ) =⇒= r = 7.
Using Bühlmann theory, we find
µ(Λ) =E(X|Λ) = λ
v(Λ) =Var[X|Λ) = λ
µ =E(Λ) = 0.75(1) + 0.25(3) = 1.5
v =E(Λ) = 1.5
a =Var[Λ) = 0.75(12 ) + 0.25(32 ) − 1.52 = 0.75
v 1.5
k= = =2
a 0.75
1 1
Z= = .
1+2 3
The Bühlmann credibility estimate of the expected number of claims for this
risk in Year 2 is
1 2
(7) + (1.5) = 3.33
3 3
Example 82.7 ‡
You are given:
(i) The claim count and claim size distributions for risks of type A are:

0 4/9
500 1/3
1 4/9
1235 2/3
2 1/9
(ii) The claim count and claim size distributions for risks of type B are:

0 1/9
250 2/3
1 4/9
328 1/3
2 4/9
(iv) Claim counts and claim sizes are independent within each risk type.
(v) The variance of the total losses is 296,962.
A randomly selected risk is observed to have total annual losses of 500.
Determine the Bühlmann premium for the next year for this same risk.
Solution.
Let L denote annual losses. The prior parameter represents the type of
risk so that either Θ = A or Θ = B. The prior distribution is Pr(A) =
Pr(B) = 0.5. For each of the two classes, total annual loss L has a compound
distribution.
We want
ZX + (1 − Z)µ.
Since there is a single observation (n = 1), the sample mean is X = 500.
We have
µ(A) =E(L|A) = E(N |A)E(X|A)

=[0(4/9) + 1(4/9) + 2(1/9)][500(1/3) + 1235(2/3)] = 660
µ(B) =E(L|B) = E(N |B)E(X|B)
=[0(1/9) + 1(4/9) + 2(4/9)][250(2/3) + 328(1/3)] = 368
µ =E[µ(Θ)] = 660(1/2) + 368(1/2) = 514
a =Var[µ(Θ)] = 6602 (1/2) + 3682 (1/2) − 5142 = 21, 316
Var(L) =Var[E(L|Θ)] + E[Var(L|Θ)]
296, 962 =Var[µ(Θ)] + E[v(Θ)
296, 962 =21, 316 + v
v =275, 646
1
Z= 275,646 = 0.0718.
1 + 21,316
The Bühlmann premium is
0.0718(500) + (1 − 0.0718)(514) ≈ 513

Practice Problems
Problem 82.1 ‡
You are given:
(i) Two risks have the following severity distributions:
Probability of claim Probability of claim
Amount of claim amount for Risk 1 amount for Risk 2
250 0.5 0.7
2500 0.3 0.2
60000 0.2 0.1
(ii) Risk 1 is twice as likely to be observed as Risk 2.
Determine the Bühlmann credibility estimate of the second claim amount
from the same risk.
Problem 82.2 ‡
You are given the following joint distribution:
Θ
X 0 1
0 0.4 0.1
1 0.1 0.2
2 0.1 0.1
P10
For a given value of Θ and a sample of size 10 for X : i=1 xi = 10.
Determine the Bühlmann credibility premium.
Problem 82.3 ‡
An insurer writes a large book of home warranty policies. You are given
the following information regarding claims filed by insureds against these
policies:
(i) A maximum of one claim may be filed per year.
(ii) The probability of a claim varies by insured, and the claims experience
for each insured is independent of every other insured.
(iii) The probability of a claim for each insured remains constant over time.
(iv) The overall probability of a claim being filed by a randomly selected
insured in a year is 0.10.
(v) The variance of the individual insured claim probabilities is 0.01.
An insured selected at random is found to have filed 0 claims over the past
10 years.
Determine the Bühlmann credibility estimate for the expected number of

claims the selected insured will file over the next 5 years.
Problem 82.4 ‡
You are given:
(i) Claim size, X, has mean µ and variance 500.
(ii) The random variable µ has a mean of 1000 and variance of 50.
(iii) The following three claims were observed: 750, 1075, 2000
Calculate the expected size of the next claim using Buhlmann credibility
Problem 82.5 ‡
You are given:
(i) A portfolio of independent risks is divided into two classes.
(ii) Each class contains the same number of risks.
(iii) For each risk in Class 1, the number of claims per year follows a Poisson
distribution with mean 5.
(iv) For each risk in Class 2, the number of claims per year follows a binomial
distribution with m = 8 and q = 0.55.
(v) A randomly selected risk has three claims in Year 1, r claims in Year 2
and four claims in Year 3.
The Bühlmann credibility estimate for the number of claims in Year 4 for
this risk is 4.6019. Determine r.
Problem 82.6 ‡
For a portfolio of independent risks, the number of claims for each risk in a
year follows a Poisson distribution with means given in the following table:
Mean Number of
Class Claims per risk Number of Risks
1 1 900
2 10 90
3 20 10
You observe x claims in Year 1 for a randomly selected risk.

The Bühlmann credibility estimate of the number of claims for the same
risk in Year 2 is 11.983.
Determine x.
Problem 82.7 ‡
An insurance company sells two types of policies with the following charac-
teristics:
Type of Policy Proportion of Total Policies Poisson Annual Claim Frequency

I θ λ = 0.50
II 1−θ λ = 1.50
A randomly selected policyholder is observed to have one claim in Year 1.
For the same policyholder, determine the Bhlmann credibility factor Z for
Year 2.
Problem 82.8 ‡
You are given:
(i) Losses in a given year follow a gamma distribution with parameters α
and θ, where θ does not vary by policyholder.
(ii) The prior distribution of α has mean 50.
(iii) The Buhlmann credibility factor based on two years of experience is
0.25.
Calculate Var(α).
Problem 82.9 ‡
For a portfolio of independent risks, you are given:
(i) The risks are divided into two classes, Class A and Class B.
(ii) Equal numbers of risks are in Class A and Class B.
(iii) For each risk, the probability of having exactly 1 claim during the year
is 20% and the probability of having 0 claims is 80%. (iv) All claims for
Class A are of size 2.
(v) All claims for Class B are of size c, an unknown but fixed quantity.
One risk is chosen at random, and the total loss for one year for that risk is
observed. You wish to estimate the expected loss for that same risk in the
following year.
Determine the limit of the Bühlmann credibility factor as c goes to infinity.
Problem 82.10 ‡
An insurance company writes a book of business that contains several classes
of policyholders. You are given:
(i) The average claim frequency for a policyholder over the entire book is
0.425.
(ii) The variance of the hypothetical means is 0.370.
(iii) The expected value of the process variance is 1.793.
One class of policyholders is selected at random from the book. Nine poli-
cyholders are selected at random from this class and are observed to have
produced a total of seven claims. Five additional policyholders are selected
at random from the same class.

Determine the Bühlmann credibility estimate for the total number of claims
for these five policyholders.
83 The Bühlmann Model with Continuous Prior

In this section, we consider Bühlmann models with continuous prior.
Example 83.1
Let X1 , X2 , · · · , Xn be past claim amounts. Suppose that Xi |Θ are indepen-
dent and identically exponentially distributed with mean Θ and Θ is Gamma
distributed with parameters α and β. Determine the Bühlmann premium.
Solution.
The hypothetical mean is
µ(θ) = E(Xi |Θ) = θ.
Thus,
µ = E[µ(Θ)] = E(Θ) = αβ
and
a = Var[µ(Θ)] = Var(Θ) = αβ 2 .
Likewise,
v(θ) = Var(Xi |Θ) = θ2
and
v = E[v(Θ)] = E[Θ2 ] = Var(Θ) + E(Θ)2 = αβ 2 + α2 β 2 .
Hence,
v αβ 2 + α2 β 2
k= = =1+α
a αβ 2
and
n n
Z= = .
n+k n+α+1
The Bühlmann premium is
n α+1
BP = X+ (αβ)
n+α+1 n+α+1
Example 83.2 ‡
You are given:
(i) Claim counts follow a Poisson distribution with mean θ.
(ii) Claim sizes follow an exponential distribution with mean 10θ.
(iii) Claim counts and claim sizes are independent, given θ.
(iv) The prior distribution has probability density function:
5
π(θ) = , θ > 1.
θ6
Calculate Bühlmann’s k for aggregate losses.
83 THE BÜHLMANN MODEL WITH CONTINUOUS PRIOR 619
Solution.
Let N be the Poisson claim count variable, let X be the claim size variable,
and let S be the aggregate loss variable. Note that S|Θ is a compound Pois-
son distribution with primary distribution N |Θ and secondaru distribution
X|Θ.
µ(θ) = E(N |Θ)E(X|Θ) = θ(10θ) = 10θ2
with expected value

Z ∞
2 2 5 50
µ = E[10Θ ] = 10θ dθ = .
1 θ6 3
The variance of the hypothetical mean is

Z ∞ 2
2 2 5 50
a= (10θ ) dθ − = 222.2222.
1 θ6 3
The process variance is
v(θ) =Var(S|Θ) = E(N |Θ)Var(X|Θ) + Var(N |Θ)E(X|Θ)2

=(θ)((10θ)2 + θ(10θ)2
=200θ3 .
and its expected value is

Z ∞
3 5
v = E[v(Θ)] = 200θ dθ = 500.
1 θ6
Hence,
v 500
k= = = 2.25
a 222.2222
Example 83.3 ‡
You are given:
(i) The number of claims made by an individual insured in a year has a
Poisson distribution with mean λ.
(ii) The prior distribution for Λ is Gamma with parameters α = 1 and
θ = 1.2.
Three claims are observed in Year 1, and no claims are observed in Year 2.
Using Bühlmann credibility, estimate the number of claims in Year 3.
Solution.
µ(λ) = E(X|Λ) = λ
with expected value

µ = E(Λ) = αθ = 1.2.
The variance of the hypothetical mean is
a = Var[µ(Λ)] = Var(Λ) = αθ2 = 1.44.
The process variance is
v(λ) = Var(X|Λ) = λ
and its expected value is
v = E[v(λ)] = E(Λ) = 1.2.
Hence,
v 1.2 1
k= = =
a 1.44 1.2
and the credibility factor is
n 2
Z= = 1 = 0.706.
n+k 2 + 1.2
The expected the number of claims in Year 3 is
BP = ZX + (1 − Z)µ = 0.706(1.5) + (1 − 0.706)(1.2) = 1.4118
Example 83.4 ‡
You are given:
(i) The number of claims in a year for a selected risk follows a Poisson dis-
tribution with mean λ.
(ii) The severity of claims for the selected risk follows an exponential distri-
bution with mean θ.
(iii) The number of claims is independent of the severity of claims.
(iv) The prior distribution of λ is exponential with mean 1.
(v) The prior distribution of θ is Poisson with mean 1.
(vi) A priori, λ and θ are independent.
Using Bühlmann’s credibility for aggregate losses, determine k.
Solution.
We have
E(S|λ, θ) =E(N )E(X) = λθ

Var(S|λ, θ) =E(N )E(X 2 ) = λ(2θ2 )
a =Var(λθ) = E(λ2 )E(θ2 ) − [E(λ)E(θ)]2
=2(12 + 1) − [(1)(1)]2 = 3
v =E(2λθ2 ) = 2E(λ)E(θ2 ) = 2(1)(12 + 1) = 4
v 4
k= =
a 3
Example 83.5 ‡
You are given:
(i) The annual number of claims on a given policy has a geometric distribu-
tion with parameter β.
(ii) The prior distribution of β has the Pareto density function
α
π(β) = , 0 < β < ∞,
(β + 1)α+1
where α is a known constant greater than 2.
A randomly selected policy had x claims in Year 1. Determine the Bühlmann
credibility estimate of the number of claims for the selected policy in Year
2.
Solution.
Note that the prior distribution is a Pareto distribution with parameters α
and 1. We have
µ(β) =E(X|β) = β
1
µ =E(β) =
α−1
2 1 α
a =Var(β) = − =
(α − 1)(α − 2) (α − 1)2 (α − 1)2 (α − 2)
v(β) =Var(X|β) = β(β + 1)
1 2 α
v =E[β(β + 1)] = + =
α − 1 (α − 1)(α − 2) (α − 1)(α − 2)
v
k = =α−1
a
1 1
Z= = .
1+k α
The required estimate is

x 1 x+1
xZ + (1 − Z)µ = + =
α α α
Example 83.6 ‡
For five types of risks, you are given:
(i) The expected number of claims in a year for these risks ranges from 1.0
to 4.0.
(ii) The number of claims follows a Poisson distribution for each risk.
During Year 1, n claims are observed for a randomly selected risk.
For the same risk, both Bayes and Bhlmann credibility estimates of the
number of claims in Year 2 are calculated for n = 0, 1, 2, · · · , 9.
Which graph represents these estimates?
Solution.
If X is the number of claims in Year 1, then the Bühlmann estimate is
ZX + (1 − Z)µ which is a linear function of X. This implies that (E) cannot
be the answer. The Bayes estimate is given by
Z 4
E(X2 |X1 = n) = λπ(λ|X1 = n)dλ
1
where 1.0 ≤ λ ≤ 4 by (i). Thus,
π(λ|X1 = n) ≤ λπ(λ|X1 = n) ≤ 4λπ(λ|X1 = n) =⇒ 1 ≤ E(X2 |X1 = n) ≤ 4.
The graph (B) can not be the answer since E(X2 |X1 = 8) and E(X2 |X1 = 9)
are greater than 4. Likewise, the graph (D) cannot be the answer since
E(X2 |X1 = 0) and E(X2 |X1 = 1) are less than 1. Now, by Remark 81.1, the
Bühlmann estimates are the linear least squares approximation to the Bayes
estimates. We see from graph (C) that the Bayes estimates are consistently
higher than the Bühlmann estimates and so it can not be the answer. Hence,
(A) is the most appropriate answer for the problem
Practice Problems
Problem 83.1
Let X1 , X2 , · · · , Xn be past claim amounts. Suppose that Xi |Θ are inde-
pendent and identically Poisson distributed with mean Θ and Θ is Gamma
distributed with parameters α and β. Determine the Bühlmann premium.
Problem 83.2 ‡
You are given:
(i) Annual claim frequency for an individual policyholder has mean λ and
variance σ 2 .
(ii) The prior distribution for λ is uniform on the interval [0.5, 1.5].
(iii) The prior distribution for σ 2 is exponential with mean 1.25.
A policyholder is selected at random and observed to have no claims in Year
1.
Using Bühlmann credibility, estimate the number of claims in Year 2 for the
selected policyholder.
Problem 83.3 ‡
You are given the following information about a book of business comprised
of 100 insureds:
(i) Xi = N
P i
j=1 Yij is a random variable representing the annual loss of the
irmth insured.
(ii) N1 , N2 , · · · , N100 are independent random variables distributed accord-
ing to a negative binomial distribution with parameters r (unknown) and
β = 0.2.
(iii) Unknown parameter r has an exponential distribution with mean 2.
(iv) Yij are independent random variables distributed according to a Pareto
distribution with α = 3.0 and θ = 1000.
Determine the Bühlmann credibility factor, Z, for the book of business.
Problem 83.4 ‡
You are given:
(i) The annual number of claims for an insured has probability function:

3
p(x) = q x (1 − q)3−x , x = 0, 1, 2, 3.
x
(ii) The prior density is π(q) = 2q, 0 < q < 1.
A randomly chosen insured has zero claims in Year 1.
Using Bühlmann credibility, estimate the number of claims in Year 2 for the
selected insured.
Problem 83.5 ‡
You are given:
(i) Claims are conditionally independent and identically Poisson distributed
with mean Θ.
(ii) The prior distribution function of Θ is:
2.6
1
π(θ) = 1 − , θ > 0.
1+θ
Five claims are observed. Determine the Bühlmann credibility factor.
Problem 83.6 ‡
You are given:
(i) Claim counts follow a Poisson distribution with mean λ.
(ii) Claim sizes follow a lognormal distribution with parameters µ and σ.
(iii) Claim counts and claim sizes are independent.
(iv) The prior distribution has joint probability density function:
f (λ, µ, σ) = 2σ, 0 < λ < 1, 0 < µ < 1, 0 < σ < 1.
Calculate Bühlmann k for aggregate losses.
Problem 83.7
(i) The annual claim amount on a policy has probability density function:
2x
f (x|θ) = , 0 < x < θ.
θ2
(ii) The prior distribution of θ has density function:
π(θ) = 4θ3 , 0 < θ < 1.
(iii) A randomly selected policy had claim amount 0.1 in Year 1.

Determine the Bühlmann credibility estimate of the claim amount for the
selected policy in Year 2.
Problem 83.8 ‡
You are given the following information about workers compensation cover-
age:
(i) The number of claims for an employee during the year follows a Poisson
distribution with mean (100 − p)/100, where p is the salary (in thousands)
for the employee.
(ii) The distribution of p is uniform on the interval (0, 100].

An employee is selected at random. During the last 4 years, the employee
has had a total of 5 claims.
Determine the Bühlmann credibility estimate for the expected number of
claims the employee will have next year.
84 THE BÜHLMANN-STRAUB CREDIBILITY MODEL 627
84 The Bühlmann-Straub Credibility Model

The standard Bühlmann credibility model assumes that the conditional
losses Xi |Θ, where i = 1, 2, · · · , n, of a policyholder in the past n years
are independent and identically distributed with respect to each past year.
A drawback of this assumption is that it does not allow for variations in
exposure or size. For example, what happens if claims experience in year 1
for a policyholder reflected only a portion of a year? For a group insurance,
what happens if the size of the group changed over time?
These issues can be resolved by modifying the Bühlmann model as fol-

lows: Let mi be the number of exposure units in year i with exposure
units Xi1 , Xi2 , · · · , Ximi such that Xi1 |Θ, Xi2 |Θ, , · · · , Ximi |Θ are indepen-
dent with the same mean µ(θ) and variance v(θ). Let Xi be the average of
the claims in year i. That is,
Xi1 + Xi2 + · · · + Ximi
Xi = .
mi
Note that E(Xi |Θ) = µ(θ) (the hypothetical mean) and Var(Xi |θ) = v(θ) mi
(the process variance). It is still assumed that there is independence from
one period (or group) to another.
As in the Bühlmann model, we let

µ = E[µ(Θ)], v = E[v(Θ)], a = Var[µ(Θ)].
We then have the following
E(Xi ) =E[E(Xi |Θ)] = E[µ(Θ)] = µ
=E[E(Xi Xj |Θ)] − E[µ(Θ)]2
=E[µ2 (Θ)] − E[µ(Θ)]2
=Var[µ(Θ)] = a, i 6= j

v(Θ)
=E + Var[µ(Θ)]
mi
v
= +a
mi
To find the credibility premium, we denote the total exposure by m =

P n
i=1 mi . Using the normal equations, we find
n n
X X α̂0
E(Xn+1 ) = µ = α̂0 + α̂i µ =⇒ α̂i = 1 − . (84.1)
µ
i=1 i=1
For i = 1, 2, · · · , n, (81.4) becomes

n X n
X v v α̂i
a= α̂j a + α̂i a + = α̂j a + .
mi mi
j=1
j=1
j 6= i
Solving this equation for α̂i , we find
 
n
a X a α̂0
α̂i = mi 1 − α̂j  = mi . (84.2)
v v µ
j=1
By (84.1) and (84.2), we find

n n
α̂0 X X a α̂0 a α̂0
1− = α̂j = mj = m
µ v µ v µ
j=1 j=1
which leads to
v/a
α̂0 = µ
m + v/a
and
mi
α̂i = .
m + v/a
Letting k = av , the credibility premium can be expressed as
n
X
α̂0 + α̂j Xj = ZX + (1 − Z)µ
j=1
where
m m1 X1 +m2 X2 +···+mn Xn
Z= m+k and X = m .
If Xi is interpreted to be the average loss/claims experienced by the mi
group members in year i, then mi Xi is the total loss/claims of the mi group
members in year i. Also, X is the overall loss/claims per group member over
the n years. The credibility premium to be charged to the group in year
n + 1 is mn+1 [ZX + (1 − Z)µ] for the mn+1 members in the next year. Keep
in mind that ZX + (1 − Z)µ is the credibility premium per exposure unit
(i.e., per occurrence of an individual Xij ).
Remark 84.1
If mi = 1 for i = 1, 2, · · · , n then the Bühlmann-Straub model coincides
with the original Bühlmann model.
Example 84.1
You are given:
(i) In year j, there are Nj claims for mj policies.
(ii) An individual policy has a Poisson distribution with mean Λ.
(iii) Λ has a Gamma distribution with parameters α and β.
(a) Determine the Bühlmann-Straub estimate of the number of claims for
one policyholder in year n + 1.
(b) Determine the Bühlmann-Straub estimate of the number of claims in
year n + 1 if there will be mn+1 policies.
Solution.
Ni
We let Xi = m i
. Because Ni has a Poisson distribution with mean mi λ then
Xi |Λ has a Poisson distribution with mean λ. Thus,
1 mi λ v(λ)
Var(Xi |Λ) = 2 Var(Ni ) = 2 = =⇒ v(λ) = λ.
mi mi mi
We have
µ(Λ) =E(Xi |Λ) = λ

µ =E[µ(Λ)] = E(Λ) = αβ
a =Var[µ(Λ)] = Var(Λ) = αβ 2
v =E[v(Λ)] = E(Λ) = αβ
v αβ 1
k= = =
a αβ 2 β
m mβ
Z= = .
m + 1/β mβ + 1
(a) The Bühlmann-Straub estimate of the number of claims for one policy-
holder in year n + 1 is

mβ 1
Pc = X+ (αβ)
mβ + 1 mβ + 1
1 P
where X = m i=1 nmi Xi .
(b) The Bühlmann-Straub estimate of the number of claims if there are
mn+1 policies in year n + 1 is mn+1 Pc
Example 84.2 ‡
You are given:
(i) The number of claims incurred in a month by any insured has a Poisson
(ii) The claim frequencies of different insureds are independent.
(iii) The prior distribution is Gamma with probability density function:
(100λ)6 e−100λ
π(λ) = .
120λ
(iv)
Month Number of Insureds Number of Claims
1 100 6
2 150 8
3 200 11
4 300 ?
Determine the Bühlmann-Straub credibility estimate of the number of claims
in Month 4.
Solution.
N
Let Xj = mjj . Note that
1 mi λ v(λ)
Var(Xi |Λ) = Var(Ni ) = 2 = =⇒ v(λ) = λ.
m2i mi mi
We have
µ(λ) =E(Xi |Λ) = λ
µ =E[µ(Λ)] = E(Λ) = αβ = 0.06
a =Var[µ(Λ)] = Var(Λ) = αβ 2 = 0.0006
v =E[v(Λ)] = E(Λ) = 0.06
v 0.06
k= = = 100
a 0.0006
m 450 9
Z= = =
m+k 450 + 100 11
N1 + N2 + N3 6 + 8 + 11 25
X= = = .
m 450 450
The credibility estimate of the expected number of claims for one insured in
month 4 is
9 25 2
Pc = ZX + (1 − Z)µ = + (0.06) = 0.056364.
11 450 11
For 300 insureds, the expected number of claims in Month 4 is 300(0.056364) =

16.9
Example 84.3 ‡
For a portfolio of insurance risks, aggregate losses per year per exposure
follow a normal distribution with mean θ and standard deviation 1000, with
θ varying by class as follows:
Class θ Percent of Risks in Class

X 2000 60%
Y 3000 30%
Z 4000 10%
A randomly selected risk has the following experience over three years:
Year Number of Exposures Aggregate Losses

1 24 24,000
2 30 36,000
3 26 28,000
Calculate the Bühlmann-Straub estimate of the mean aggregate losses per

year per exposure in Year 4 for this risk.
Solution.
Let Xi be the aggregate losses per exposure in year i. That is, the average
over all exposures of the total losses in year i. Thus,
24, 000
X1 = = 1, 000
24
36, 000
X2 = = 1, 200
30
28, 000 14, 000
X1 = = .
26 13
We are given: m1 = 24, m2 = 30, m3 = 26. The risk parameter Θ is the
mean of the normal distribution. The prior distribution is
π(2000) =0.6
π(3000) =0.3
π(4000) =0.1.
We are given that Xi is the average of mi (independent) random variables

each normally distributed with mean θ and variance 10002 , conditional on
Θ = θ. Hence, we have the following
µ(θ) =E(Xi |Θ) = θ

v(θ) =1, 000, 000
µ =E(Θ) = 2000(0.6) + 3000(0.3) + 4000(0.1)
=2500
v =1, 000, 000
a =Var(Θ) = 20002 (0.6) + 30002 (0.3) + 40002 (0.1) − 25002
=450, 000
v 1, 000, 000
k= =
a 450, 000
80
Z= = 0.97297
80 + 1,000,000
450,000
3
1 X
X= mi Xi
m
i=1

1 14, 000
= 24(1000) + 30(1200) + 26
80 13
=1, 100.
Hence, the The Bühlmann-Straub credibility estimate for the loss per expo-
sure in Year 4 is
ZX + (1 − Z)µ = 0.97297(1, 100) + (1 − 0.97297)(2500) = 1137.84

Practice Problems
Problem 84.1
Let X1 , X2 , · · · , Xn be losses in the past n years. Suppose that they all have
the same risk parameter θ. We assume that Xi |Θ are idependent with mean
µ(θ) = E(Xi |Θ and variance Var(Xi |Θ) = w(θ) + v(θ) mi . Such a credibility
model is knwon as Hewitt’s model.
(a) Show that
m2i + m2j

mi Xi + mj Xj v(θ)
Var Θ = 2
w(θ) + .
mi + mj (mi + mj ) mi + mj
(b) Show that µ = E(Xi ), Cov(Xi , Xj ) = a for i 6= j and Var(Xi ) = w + mvj

where v = E[v(Θ) and w = E[w(Θ)].
Problem 84.2
Consider the Hewitt’s model introduced above. Let α̂0 + nj=1 α̂j Xj be the
P
credibility premium. Using normal equations, show that
aα̂0 /µ µ
α̂i = w+v/mi and α̂0 = 1+am∗
where
n
X mj
m∗ = .
v + wmj
j=1
Problem 84.3
Consider the Hewitt’s model introduced above. Show that
n
X
α̂0 + α̂j Xj = ZX + (1 − Z)µ
j=1
where
n
am∗ 1
X mj
Z= 1+am∗ and X = m∗ Xj .
v + wmj
j=1
Problem 84.4 ‡
You are given four classes of insureds, each of whom may have zero or one
claim, with the following probabilities:
Class Number of claims

0 1
I 0.9 0.1
II 0.8 0.2
III 0.5 0.5
IV 0.1 0.9
A class is selected at random (with probability 0.25), and four insureds are
selected at random from the class. The total number of claims is two.
If five insureds are selected at random from the same class, estimate the
total number of claims using Bühlmann-Straub credibility.
Problem 84.5 ‡
You are given the following data on large business policyholders:
(i) Losses for each employee of a given policyholder are independent and
have a common mean and variance.
(ii) The overall average loss per employee for all policyholders is 20.
(iii) The variance of the hypothetical means is 40.
(iv) The expected value of the process variance is 8000.
(v) The following experience is observed for a randomly selected policy-
holder:
Average loss per Number of
Year Employee Employees
1 15 800
2 10 600
3 5 400
Determine the Bühlmann-Straub credibility premium per employee for this
policyholder.
Problem 84.6 ‡
Members of three classes of insureds can have 0, 1 or 2 claims, with the
following probabilities:
Number of Claims
Class 0 1 2
I 0.9 0.0 0.1
II 0.8 0.1 0.1
III 0.7 0.2 0.1
A class is chosen at random, and varying numbers of insureds from that
class are observed over 2 years, as shown below:
Year Number of Insureds Number of Claims

1 20 7
2 30 10

in Year 3 for 35 insureds from the same class.
Problem 84.7 ‡
You are given:
(i) The number of claims incurred in a month by any insured follows a
Poisson distribution with mean λ.
(ii) The claim frequencies of different insureds are independent.
(iii) The prior distribution of Λ is Weibull with θ = 0.1 and τ = 2.
(iv) Some values of the gamma function are:
Γ(0.5) = 1.77245, Γ(1) = 1, Γ(1.5) = 0.88623, Γ(2) = 1.
(v)
Month Number of Insureds Number of claims

1 100 10
2 150 11
3 250 14

in the next 12 months for 300 insureds.
Problem 84.8 ‡
For each policyholder, losses X1 , · · · , Xn , conditional on Θ, are indepen-
dently and identically distributed with mean,
µ(θ) = E(Xj |Θ = θ), j = 1, 2, · · · , n
and variance,
v(θ) = Var(Xj |Θ = θ), j = 1, 2, · · · , n.
You are given:
(i) The Bühlmann credibility assigned for estimating X5 based on X1 , · · · , X4
is Z = 0.4.
(ii) The expected value of the process variance is known to be 8.
Calculate Cov(Xi , Xj ), i 6= j.
Problem 84.9 ‡
You are given n years of claim data originating from a large number of
policies. You are asked to use the Bühlmann-Straub credibility model to
estimate the expected number of claims in year n + 1.
Which of conditions (A), (B), or (C) are required by the model?
(A) All policies must have an equal number of exposure units.
(B) Each policy must have a Poisson claim distribution.
(C) There must be at least 1082 exposure units.
(D) Each of (A), (B), and (C) is required.
(E) None of (A), (B), or (C) is required.
Problem 84.10 ‡
You are given the following information about a single risk:
(i) The risk has m exposures in each year.
(ii) The risk is observed for n years.
(iii) The variance of the hypothetical means is a.
v
(iv) The expected value of the annual process variance is w + m .
Determine the limit of the Bhlmann-Straub credibility factor as m ap-
proaches infinity.
85 EXACT CREDIBILITY 637
85 Exact Credibility
The term exact credibility refers to the case when the credibility premium
is equal to the Bayesian premium. Exact credibility arises in Bühlmann and
Bühlmann-Struab models specifically in situations involving the linear ex-
ponential family members and their conjugate priors (see Sections 24 and
66) which we demonstrate next.
Let Xi |Θ be a member of the linear exponential family. Then its pdf can be
expressed as
p(x)er(θ)x
fXi |Θ (x|θ) =
q(θ)
and where the support of Xi |Θ does not depend on θ.
For a member in the linear exponential family, the mean is given by (see
Example 24.2)
q 0 (θ)
µ(θ) = E(Xi |Θ) = .
r0 (θ)q(θ)
The variance (see Example 24.3) is given by
µ0 (θ)
Var(Xi |Θ) = .
r0 (θ)
Let the pdf of Θ (the prior distribution) be given by
[q(θ)]−k eµkr(θ) r0 (θ)

π(θ) = , θ0 < θ < θ 1 .
c(µ, k)
We will show that the posterior distribution is of the same type as π(θ) so
that the prior distribution is a conjugate prior distribution. Indeed, we have
h i
Z Qn p(xj ) er(θ) nj=1 xj
P
[q(θ)]−k eµkr(θ) r0 (θ)
Z
j=1
fX|Θ (x|θ)π(θ)dθ = × dθ
[q(θ)]n c(µ, k)
µk+ n
P
  j=1 xj
n r(θ) (k+n)
[q(θ)]−(k+n) e r0 (θ)
Y Z k+n
=  p(xj )  dθ
c(µ, k)
j=1
 
n ∗ ∗ ∗
[q(θ)]−k er(θ)µ k r0 (θ)
Y Z
= p(xj )
c(µ, k)
j=1
 
Yn
= p(xj )
j=1
where
µk+ n
P
j=1 xj
µ∗ = k+n and k ∗ = k + n.
Pn
Qn r(θ) j=1 xj
[ j=1 p(xj )]e [q(θ)]−k eµkr(θ) r0 (θ)
[q(θ)]n × c(µ,k)
π(θ|x) = hQ i
n
j=1 p(xj )
µk+ n
P
j=1 xj
r(θ) (k+n)
[q(θ)]−(k+n) e r0 (θ)
k+n
=
c(µ, k)
∗ ∗ ∗
[q(θ)]−k er(θ)µ k r0 (θ)
= .
c(µ, k)
From the expression of π(θ|x), we can write

π(θ|x)
ln = −k ∗ ln [q(θ)] + µ∗ k ∗ r(θ) − ln [c(µ, k)].
r0 (θ)
Differentiating with respect to θ yields
[π(θ|x)/r0 (θ)]0 0
∗ q (θ)
= −k + µ∗ k ∗ r0 (θ)
π(θ|x)/r0 (θ) q(θ)
which can be rearranged as

d π(θ|x)
= −k ∗ [µ(θ) − µ∗ ]π(θ|x).
dθ r0 (θ)
Integrating over the interval (θ0 , θ1 ) yields
π(θ1 |x) π(θ0 |x)

− 0 = −k[E(Xn+1 |X) − µ∗ ]
r0 (θ1 ) r (θ0 )
which yields the Bayesian premium
π(θ0 |x) π(θ1 |x)
E(Xn+1 |X = µ∗ + − 0 .
kr0 (θ0 ) kr (θ1 )
Assuming
π(θ0 |x) π(θ1 |x)
= 0
r0 (θ0 ) r (θ1 )
we see that E(Xn+1 |X = µ∗ = a0 + nj=1 aj Xj .
P
Now recall that the Bühlmann premium is the minimizer of (see Remark
81.1) "
n
#2 
 X 
Q1 = E E(Xn+1 |X) − α0 − αi Xi .
 
i=1
By letting, αj = aj for j = 0, 1, · · · , n we see that the credibility premium

coincides with the Bayesian premium and credibility is exact.
The following table list some models where the Bayesian premium equals
the credibility premium:
fX|Θ (x|θ) π(θ) π(θ|x)
Poisson Gamma Gamma
Normal Normal Normal
Bernoulli Beta Beta
Exponential Inverse Gamma Inverse Gamma
Example 85.1
You are given:
(i) The model distribution X|Λ is Poisson with parameter Λ.
(ii) The prior distribution of Λ is Gamma with parameters α and θ.
Show that this model satisfies exact credibility.
Solution.
Example 66.2 shows that the posterior distribution is a Gamma distribution
β
with parameters α0 = α + nX and θ0 = nβ+1 . The hypothetical mean is
µ(λ) = E(Xi |Λ) = λ

and the Bayesian premium is

Pn λ(nθ+1)
∞
λ i=1 xi +α−1 e− θ
Z
E(Xn+1 |X) = λ Pni=1 xi +α P dλ
0 θ n
nθ+1 Γ( i=1 xi + α)
Pn λ(nθ+1)
∞
λ i=1 xi +α+1−1 e− θ
Z
θ
=(nX + α) Pni=1 xi +α+1 P dλ
nθ + 1 0

θ n
nθ+1 Γ( x
i=1 i + α + 1)
θ
=(nX + α)
nθ + 1
Note that the Bayesian premium is a linear function of X1 , X2 , · · · , Xn .
Next, we find the Bühlmann credibility. We have
µ(λ) =λ
µ =E(Λ) = αθ
v(λ) =Var(Xi |Λ) = λ
v =E(Λ) = αθ
a =Var(Λ) = αθ2
v 1
k= =
a θ
n nθ
Z= =
n+k nθ + 1
Pc =ZX + (1 − Z)µ
nθ 1
= X+ (αθ)
nθ + 1 nθ + 1
θ
=(nX + α) .
nθ + 1
Thus, the credibility premium equals the Bayesian premium
Example 85.2 ‡
You are given:
(i) The number of claims per auto insured follows a Poisson distribution
with mean λ.
(ii) The prior distribution for Λ has the following probability density func-
tion:
(500λ)50 e−500λ
f (λ) = .
λΓ(50)
(iii) A company observes the following claims experience:
Year 1 Year 2
Number of claims 75 210
Number of autos insured 600 900
The company expects to insure 1100 autos in Year 3. Determine the ex-
pected number of claims in Year 3.
Solution.
The model distribution is Poisson with mean λ and the prior distribution is
1
Gamma with α = 500 and θ = 500 . Thus, this model satisfies exact credi-
bility.
The Bayesian Approach Solution

Let Xn denote the number of claims in Year n. We are asked to find
E(X3 |X1 , X2 ). We have E(X3 |Λ) = λ. The model distribution is Poisson
1
and the prior distribution is Gamma with parameters α = 50 and θ = 500 .
By Example 66.2, the posterior distribution is Gamma with parameters
(1/500)
α0 = 75 + 210 + 50 = 335 and θ0 = nθ+1 θ
= 1500(1/500)+1 = 0.0005. Thus,
E(X3 |X1 , X2 ) = E(Λ|X) = α0 θ0 = (335)(0.0005) = 0.1675. This is the ex-
pected number of claims per policy. The expected number of claims in the
next year is E(X3 |X1 , X2 ) = 1100(0.1675) = 184.25.
The Bühlmann Approach Solution

We have
µ(λ) =E(X|Λ) = λ
50
µ =E(Λ) = αθ = = 0.1
500
50
a =Var(Λ) = αθ2 = = 0.0002
5002
v(λ) =Λ
v =E(Λ) = 0.1
v 0.1
k= = = 500
a 0.0002
n 1500
Z= = = 0.75
n+k 1500 + 500
75 + 210
X= = 0.19.
600 + 900
Thus,
E(X3 |X1 , X2 ) = 1100[ZX+(1−Z)µ] = 1100[0.19(0.75)+0.25(0.1)] = 184.25
Practice Problems
Problem 85.1
You are given:
(i) The model distribution X|Λ is exponential with mean Λ.
(ii) The prior distribution of Λ is inverse Gamma with parameters α and β.
Problem 85.2
You are given:
• The model distribution X|Q is binomial with parameters m and q.
• The prior distribution Q is beta with parameters a, b and 1.
Problem 85.3
You are given:
• The model distribution X|Λ is normal with mean Λ and variance σ 2 .
• The prior distribution Λ is normal with mean µ and variance a2 .
Problem 85.4 ‡
You are given:
(i) The conditional distribution of the number of claims per policyholder is
Poisson with mean λ.
(ii) The variable λ has a gamma distribution with parameters α and θ.
(iii) For policyholders with 1 claim in Year 1, the credibility estimate for the
number of claims in Year 2 is 0.15.
(iv) For policyholders with an average of 2 claims per year in Year 1 and
Year 2, the credibility estimate for the number of claims in Year 3 is 0.20.
Determine θ.
86 NON-PARAMETRIC EMPIRICAL BAYES ESTIMATION FOR THE BÜHLMANN MODEL643
86 Non-parametric Empirical Bayes Estimation for

the Bühlmann Model
The credibility models introduced earlier assume that the conditional distri-
bution and the prior distribution are known to be of a certain form. More
precisely, they are of parametric form. However, practical models necessi-
tate the choice of parameters in such a way to ensure a close agreement
between the model and reality.
What we will do next is to look at situations where the conditional dis-

tribution and the prior distribution are unknown. We refer to this situation
as the nonparametric case. The unknown quantities of interest that are
related to the prior distribution such as µ, v, and a (the ones to be estimated)
are called structural parameters. The estimation process will be based on
data at hand. We refer to this process as the empirical Bayes estimation.
The format of the data for our analysis has the following structure: Suppose
we have r ≥ 1 policyholders or groups of policyholders. For policyholder
i there are ni years of claim experience/observed exposure units. Let Xij
be the average number of losses/claims for policyholder i in year j. Let the
claim vector for average number of losses/claims for policyholder i over all
years be:
Xi = (Xi1 , · · · , Xini )T , i = 1, · · · , r.
We assume that X1 , · · · , Xr are independent, this is reasonable to think
that different groups claims will be independent of each other.
Let θi denote the risk parameter for the ith policyholder. We assume that
the Θi are independent and identically distributed. We also assume that
Xij |Θi are independent for j = 1, · · · , ni .
Let mij denote the number of exposure units for policyholder i in year
j. Then the total number of exposure units over all the years is
ni
X
mi = mij , i = 1, · · · , r.
j=1
The total exposure units for all policyholders over all the years is
r
X
m= mi .
i=1
The past average loss experience of policyholder i over all the years is
ni
1 X
Xi = mij Xij , i = 1, · · · , r.
mi
j=1
The overall average losses are

r
1 X
X= mi X i .
m
i=1
Similary to the previous credibility premiums we can define the structural

parameters:
µ = E[µ(Θi )], v = E[v(Θi )], a = Var[µ(Θi )]
where µ(θi ) = E(Xij |Θi ) and v(θi ) = Var(Xij Θi ).

These parameters will be estimated from the data at hand. Let µ̂, v̂, and â
be the estimators of µ, v, and a respectively. Then the credibility premium
estimate is
Ẑi X i + (1 − Ẑi )µ̂i
where
mi v̂
Ẑi = , k̂ = .
mi + k̂ â
The credibility premium to cover all mi,ni +1 exposure units for policyholder
i in the next year would be
mi,ni +1 [Ẑi X i + (1 − Ẑi )µ̂i ], i = 1, · · · , r.
Our goal is to estimate the parameters µ, v and a in the Bühlmann and

Bühlmann-Straub models given the data. In this section, we cover the
Bühlmann model.
For Bühlmann model, we assume that mij = 1 and ni = n for i = 1, · · · , r
and j = 1, · · · , n. Thus, mi = n and m = nr. Moreover,
n
1X
Xi = Xij , i = 1, · · · , r
n
j=1
and
r r n
1X 1 XX
X= Xi = Xij .
r nr
i=1 i=1 j=1
For an estimator of µ, we let µ̂ = X. The expected value of µ̂ is

r n r n
1 XX 1 XX
E(û) = E(Xij ) = E[E(Xij |Θi )
rn rn
i=1 j=1 i=1 j=1
r X n r X n
1 X 1 X
= E[µ(Θi )] = µ = µ.
rn rn
i=1 j=1 i=1 j=1
Hence, µ̂ is an unbiased estimator of µ.

In order to estimate v, we first need the following result.
Theorem 86.1
Let Y1 , Y2 , · · · , Yk be independent random variables with the P same mean
µ = E(Yi ) and the same variance σ 2 = Var(Yi ). Let Y = k1 ki=1 Yi . Then
(a) E(Y
h ) =P µ i
(b) E k−1 ki=1 (Yi − Y )2 = σ 2 . That is, k−1
1 1 Pk 2
i=1 (Yi − Y ) is an unbiased
estimator of the variance of Yi .
Proof.
(a) We have
k
1X kµ
E(Y ) = E(Yi ) = = µ.
k k
i=1
(b) We have
k
X k
X
2
(Yi − Y ) = [(Yi µ) + (µ − Y )]2
i=1 i=1
Xk
= [(Yi − µ)2 + 2(Yi − µ)(µ − Y ) + (µ − Y )2 ]
i=1
k
X k
X k
X
2
= (Yi − µ) + 2(µ − Y )) (Yi − µ) + (Y ) − µ)2
i=1 i=1 i=1
Xk
= (Yi − µ)2 + 2(µ − Y ))(kY − kµ) + k(Y ) − µ)2
i=1
Xk
= (Yi − µ)2 − 2k(Y − µ)2 + k(Y ) − µ)2
i=1
Xk
= (Yi − µ)2 − k(Y − µ)2 .
i=1
Taking the expectation of both sides yields
k k
" #
X X
2
E (Yi − Y ) = E[(Yi − µ)2 ] − kE[(Y − µ)2 ]
i=1 i=1
Xk
= Var(Yi ) − kVar(Y )
i=1
σ2

=kσ 2 − k = (k − 1)σ 2 .
k
Hence,
k
" #
1 X
E (Yi − Y )2 = σ 2
k−1
i=1
By the above theorem, an unbiased estimator to Var(Xij |Θi ) = v(θi ) is

n
1 X
v̂i = (Xij − X i )2 .
n−1
j=1
Since
E(v̂i ) = E[E(v̂i |Θi )] = E[v(Θi )] = v
and
r
!
1X
E v̂i =v
r
i=1
an unbiased estimator of v is
r r n
1X 1 XX
v̂ = v̂i = (Xij − X i )2 .
r r(n − 1)
i=1 i=1 j=1
It remains to find an estimator of a. We begin with

n n
1X 1X
E(X i |Θi ) = E(Xij |Θi ) = µ(θi ) = µ(θi ).
n n
j=1 j=1
unconditionally, we have
E(X i ) = E[E(X i |Θi )] = E[µ(Θi )] = µ.

Moreover,
Var(X i ) =Var[E(X i |Θi )] + E[var(X i |Θi )]

v(Θi ) v
=Var[µ(Θi )] + E =a+ .
n n
Let â be an estimator of a, v̂ be an estimator of v. An unbiased estimator

1 Pr 2
of Var(X i ) is r−1 i=1 (X i − X) . Thus, we have
r
1 X v̂
X i = â +
r−1 n
i=1
which implies
r
1 X v̂
â = (X i − X)2 −
r−1 n
i=1
r r n
1 X 1 XX
= (X i − X)2 − (Xij − X i )2 .
r−1 rn(n − 1)
i=1 i=1 j=1
Remark 86.1
Due to the subtraction in the formula for â, it is possible that â ≤ 0. When
this happens, it is customary to set â = Ẑ = 0.
Example 86.1
You are given the losses of two policyholders over a period of three years:
Policyholder Year 1 Year 2 Year 3

1 3 5 7
2 6 12 9
Determine the Bayes estimate of the Bühlmann premium for each policy-
holder for Year 4.
Solution.
We have
3+5+7
X1 = =5
3
6 + 12 + 9
X2 = =9
3
5+9
µ̂ =X = =7
2
3
1 X
v̂1 = (X1j − X 1 )2
3−1
j=1
1
= [(3 − 5)2 + (5 − 5)2 + (7 − 5)2 ] = 4
3−1
3
1 X
v̂2 = (X2j − X 2 )2
3−1
j=1
1
= [(6 − 9)2 + (12 − 9)2 + (9 − 9)2 ] = 9
3−1
v̂1 + v̂2 9+4
v̂ = = = 6.5
2 2
r
1 X v̂
â = (X i − X)2 −
r−1 n
i=1
1 6.5 35
= [(5 − 7)2 + (9 − 7)2 ] − =
2−1 3 6
13
v̂ 2 39
k̂ = = 35 =
â 6
35
3 35
Ẑ = 39 = 48 .
3 + 35
The estimated Bühlmann premium for policyholder 1 in year 4 is:

35 35 133
ẐX 1 + (1 − Ẑ)µ̂ = (5) + 1 − (7) = .
48 48 24
The estimated Bühlmann premium for policyholder 2 in year 4 is:

35 35 203
ẐX 2 + (1 − Ẑ)µ̂ = (9) + 1 − (7) =
48 48 24
Remark 86.2
Due to the subtraction in the formula for â, it is possible that â ≤ 0. When
this happens, it is customary to set â = Ẑ = 0.
Example 86.2 ‡
Survival times are available for four insureds, two from Class A and two
from Class B. The two from Class A died at times t = 1 and t = 9. The two
from Class B died at times t = 2 and t = 4.
Nonparametric Empirical Bayes estimation is used to estimate the mean
survival time for each class. Unbiased estimators of the expected value of
the process variance and the variance of the hypothetical means are used.
Estimate Z, the Bühlmann credibility factor.
Solution.
We have: r = n = 2, X11 = 1, X12 = 9, X21 = 2, X22 = 4, X 1 = 5 and
X 2 = 3. Thus,
r n
1 XX
v̂ = (Xij − X i )2
r(n − 1)
i=1 j=1
1
= [(1 − 5)2 + (9 − 5)2 + (2 − 3)2 + (4 − 3)2 ] = 17
2(2 − 1)
and
r r n
1 X 1 XX
â = (X i − X)2 − (Xij − X i )2
r−1 rn(n − 1)
i=1 i=1 j=1
1 17
= [(5 − 4)2 + (3 − 4)2 ] − = −6.5.
2−1 2
Since â < 0, we obtain Ẑ = 0

Practice Problems
Problem 86.1 ‡
An insurer has data on losses for four policyholders for 7 years. The loss
from the ith policyholder for year j is Xij .
You are given:
4 X
X 7
(Xij − X i )2 = 33.60
i=1 j=1
and
4
X
(X i − X)2 = 3.30.
i=1
Using nonparametric empirical Bayes estimation, calculate the Bühlmann

credibility factor for an individual policyholder.
Problem 86.2 ‡
You are given total claims for two policyholders:
Year
Policyholder 1 2 3 4
X 730 800 650 700
Y 655 650 625 750
Using the nonparametric empirical Bayes method, determine the Bühlmann

credibility premium for Policyholder Y.
Problem 86.3 ‡
Three individual policyholders have the following claim amounts over four
years:
Policyholder Year 1 Year 2 Year 3 Year 4

X 2 3 3 4
Y 5 5 4 6
Z 5 5 3 3
Using the nonparametric empirical Bayes procedure, calculate the estimated

variance of the hypothetical means.
Problem 86.4 ‡
Three policyholders have the following claims experience over three months:
Policyholder Month 1 Month 2 Month 3 Mean Variance

I 4 6 5 5 1
II 8 11 8 9 3
III 5 7 6 6 1
Nonparametric empirical Bayes estimation is used to estimate the credibility

premium in Month 4.
Calculate the credibility factor Z.
87 Non-parametric Empirical Bayes Estimation for

the Bühlmann-Straub Model
In this section, we turn our attention to estimating the credibility factors

µ, v, and a for the Bühlmann-Straub Model.
Recall that for this model, the hypothetical mean is
E(Xij |Θi ) = µ(θi )
and the process variance is
mij Var(Xij |Θi ) = v(θi ).
From
r
1 X
X= mi X i
m
i=1
we find
r r
1 X 1 X
E(X) = mi E(X i ) = mi E[E(X i |Θi )])
m m
i=1 i=1
 
r ni
1 X X m ij
= mi E  E(Xij |Θi )
m mi
i=1 j=1
 
r ni r
1 X X m ij 1 X
= mi E  µ(θi ) = mi E[µ(θi )]
m mi m
i=1 j=1 i=1
r
1 X
= mi µ = µ.
m
i=1
Hence, µ̂ = X is an unbiased estimator of µ.

Next, we turn our attention to estimating v. We begin with
ni
1 X
E(X i |Θi ) = mij E(Xij |Θi ) = µ(θi ).
mi
j=1
87 NON-PARAMETRIC EMPIRICAL BAYES ESTIMATION FOR THE BÜHLMANN-STRAUB MODEL6
The conditional variance of X i is

ni
1 X
Var(X i |Θi ) = m2 Var(Xij |Θi )
m2i j=1 ij
ni
1 X v(θi )
= 2 m2ij
mi j=1 mij
v(θi )
= .
mi
Similar to the argument in Section 86, one can show that

ni
X ni
X
mij (Xij − X i )2 = mij (Xij − µ(θi ))2 − mi (µ(θi ) − X i )2 .
j=1 j=1
Taking the expectations of both sides, we find

 
Xni ni
X
2
E  mij (Xij − X i ) = mij E[(Xij − µ(θi )2 ] − mi E[(µ(θi ) − X i )2 ]
j=1 j=1
ni
X
= mij Var(Xij |Θi ) − mi Var(X i |Θi )
j=1
ni
X v(θi ) v(θi )
= mij − mi
mij mi
j=1
=ni v(θi ) − v(θi ) = (ni − 1)v(θi ).
Thus,  
ni
1 X
E mij (Xij − X i )2  = v(θi ).
ni − 1
j=1
Thus, an unbiased estimator of v(θi ) is

in
1 X
v̂i = mij (Xij − X i )2 , i = 1, 2, · · · , r.
ni − 1
j=1
By the double expectation formula, we have
E(v̂i ) = E[E(v̂i |Θi )] = E[v(θi )] = v

so that
Pr v̂i is an unbiased
Pr estimator of v. Another unbiased estimator of v is
v̂ = i=1 wi v̂i where i=1 wi − 1, One choice of the wi is
ni − 1
wi = Pr .
i=1 (ni − 1)
Hence, an unbiased estimator to v is

Pr Pni
i=1 j=1 mij (Xij − X i )2
v̂ = P r .
i=1 (ni − 1)
We now turn to estimation of a. we first note that
Var(X i ) =Var[E(X i |Θi )] + E[Var(X i |Θi )]

v(Θi ) v
=Var[µ(Θi )] + E =a+ .
mi mi
Again, one can show
r
X r
X
2
mi (X i − X) = mi (X i − µ)2 − m(X − µ)2 .
i=1 i=1
Taking the expectations of both sides, we find

" r # r
X X
2
E mi (X i − X) = mi E[(X i − µ)2 ] − mE[(X − µ)2 ]
i=1 i=1
r
X
= mi Var(X i ) − mVar(X)
i=1
r r
X v 1 X 2 v
= mi a + −m· 2 mi a +
mi m mi
i=1 i=1
r
!
1 X
=ma + rv − a m2i + mv
m
i=1
r
!
1 X 2
=a m − mi + v(r − 1).
m
i=1
Thus,
 !−1 " #
r r
1 X 2 X 
E m− mi mi (X i − X)2 − v(r − 1) =a
 m 
i=1 i=1
so that an unbiased estimator of a is
r
!−1 " r
#
1 X 2 X
2
â = m− mi mi (X i − X) − v̂(r − 1) .
m
i=1 i=1
With these estimators, for the policyholder i, we have
v̂ mi
k̂ = â and Ẑi =
mi +k̂
Thus, the premium for each member of the policy is:
Ẑi X i + (1 − Ẑi )µ̂
and the credibility premium to cover all mi,ni +1 exposure units for policy-
holder i in the next year would be
mi,ni +1 [Ẑi X i + (1 − Ẑi )µ̂i ], i = 1, · · · , r.
Remark 87.1
Note that the above equations provide unbiased estimators of µ, v, and a
respectively. They are nonparametric, requiring no distributional assump-
tions. Also, due to the subtraction in the formula for â, it is possible that
â ≤ 0. When this happens, it is customary to set â = Ẑ = 0.
Example 87.1
You are given:
Policy Group Year 1 Year 2 Year 3 Year 4

Total Losses I ? 750 600 ?
Number in Group ? 3 2 4
Total Losses II 975 1200 900 ?
Number in Group 5 6 4 5
(a) Calculate the unbiased estimates for µ, v, and a in the Bühlmann-Straub

model.
(b) Determine the Bühlmann-Straub premium for each policyholder in Year
4.
Solution.
(a) We have
750 + 600
X1 = = 270
3+2
975 + 1200 + 900
X2 = = 205
5+6+4
5 15
X= X1 + X2
200 20
5 15
= (270) + (205) = 221.25
200 20
µ̂ =X = 221.25
2 2
3 750
3 − 270 + 2 600
2 − 270
v̂1 = = 3000
2−1
2 2 2
5 975
5 − 205 + 6 1200
6 − 205 + 4 900
4 − 205
v̂2 = = 1125
3−1
(2 − 1)v̂1 + (3 − 1)v̂2 3000 + 2(1125)
v̂ = = = 1750
(2 − 1) + (3 − 1) 3
1 2 2
â = 2 2 [5(270 − 221.25) + 15(205 − 221.25) − (2 − 1)(1750)] = 1879.1667.
20 − 5 +15
20
(b) We have
v̂ 1750
k̂ = = = 0.9313
â 1879.1667
m1 5
Ẑ1 = = = 0.843
m1 + k̂ 5 + 0.9313
15
Ẑ2 = = 0.9415.
15 + 0.9313
The premium in year 4 for a policyholder in Group I is
4[Ẑ1 X 1 + (1 − Ẑ1 )µ̂] = 4[0.8413(270) + (1 − 0.8413)(221.25)] = 1049.38.
The premium in year 4 for a policyholder in Group II is
5[Ẑ2 X 2 + (1 − Ẑ2 )µ̂] = 5[0.9415(205) + (1 − 0.9415)(221.25)] = 1029.75
Now, the total losses (TL) on all policyholders in observed units is

r
X
TL = mi X i .
i=1
The total premium (TP) for all policyholders in year n + 1 is:

r
X
TP = mi [Ẑi X i + (1 − Ẑi )µ̂].
i=1
We can express (TP) in terms of (TL) as follows

r
X r
X
TP = mi (1 − Ẑi )(µ̂ − X i ) + mi X i .
i=1 i=1
| {z }
TL
It is often desirable for TP to equal TL. This leads to the following calcula-
tion
Xr
0= mi (1 − Ẑi )(µ̂ − X i )
i=1
r
X
0= k̂ Ẑi (µ̂ − X i )
i=1
r
X r
X
µ̂ Ẑi = Ẑi X i
i=1 i=1
Pr
i=1 Ẑi X i
µ̂ = P r
.
i=1 Ẑi
Hence, we have another estimator of µ̂. We refer to this process of estimat-
ing µ̂ as the credibility weighted average method or the method of
preserving total losses/claims (See Problem 87.4).
Example 87.2
Redo Example 87.1(b) if µ is estimated by credibility weighted average.
Solution.
We have
Ẑ1 X 1 + Ẑ2 X 2 0.843(270) + 0.9415(205)
µ̂ = = = 235.7061.
Ẑ1 + Ẑ2 0.843 + 0.9415
The premium in year 4 for a policyholder in Group I is
4[Ẑ1 X 1 + (1 − Ẑ1 )µ̂] = 4[0.8413(270) + (1 − 0.8413)(235.7061)] = 1058.44.
The premium in year 4 for a policyholder in Group II is
5[Ẑ2 X 2 + (1 − Ẑ2 )µ̂] = 5[0.9415(205) + (1 − 0.9415)(235.7061)] = 1033.98
Practice Problems
Problem 87.1 ‡
You are given the following commercial automobile policy experience:
Company Year 1 Year 2 Year 3
Total Losses I 50,000 50,000 ?
Number of Automobiles 100 200 ?
Total Losses II ? 150,000 150,000
Number of Automobiles ? 500 300
Total Losses II 150,000 ? 150,000
Number of Automobiles 50 ? 150
Determine the nonparametric empirical Bayes credibility factor, Z, for Com-
pany III.
Problem 87.2 ‡
You are given:
Group Year 1 Year 2 Year 3 Total
Total Claims 1 10,000 15,000 25,000
Number in Group 50 60 110
Average 200 250 227.27
Total Claims 2 16,000 18,000 34,000
Number in Group 100 90 190
Average 160 200 178.95
Total Claims 59,000
Number in Group 300
Average 196.67
You are also given â = 651.03..
Use the nonparametric empirical Bayes method to estimate the credibility
factor for Group 1.
Problem 87.3 ‡
You are given the following data:
Year 1 Year 2
Total losses 12,000 14,000
Number of Policyholders 25 30
The estimate of the variance of the hypothetical means is 254.
Determine the credibility factor for Year 3 using the nonparametric empirical
Bayes method.
Problem 87.4 ‡
You are making credibility estimates for regional rating factors. You observe
that the Bühlmann-Straub nonparametric empirical Bayes method can be
applied, with rating factor playing the role of pure premium.
Xij denotes the rating factor for region i and year j, where i = 1, 2, 3 and
j = 1, 2, 3, 4.
Corresponding to each rating factor is the number of reported claims, mij ,
measuring exposure.
You are given:
4
X 4
X 4
X
i mi = mi Xi = 1
mi mij Xij v̂i = 1
3 mij (Xij − X i )2 mi (X i − X)2
i=1 j=1 j=1
1 50 1.406 0.536 0.887
2 300 1.298 0.125 0.191
3 150 1.178 0.172 1.348
Determine the credibility estimate of the rating factor for region 1 using the
X3
method that preserves mi X i .
i=1
Problem 87.5 ‡
You are given the following experience for two insured groups:
Group Year
1 2 3 Total
1 Number of members 8 12 5 25
Average loss per member 96 91 113 97
2 Number of members 25 30 20 75
Average loss per member 113 111 116 113
Total Number of members 100
Average loss per member 109
2 X
X 3 2
X
mij (Xij − X i )2 = 2020 and mi (X i − X)2 = 4800.
i=1 j=1 i=1
Determine the nonparametric Empirical Bayes credibility premium for group

1, using the method that preserves total losses.
Problem 87.6 ‡
You are given:
(i) A region is comprised of three territories. Claims experience for Year 1
is as follows:
A 10 4
B 20 5
C 30 3
(ii) The number of claims for each insured each year has a Poisson distribu-
tion.
(iii) Each insured in a territory has the same expected claim frequency.
(iv) The number of insureds is constant over time for each territory.
Determine the Bühlmann-Straub empirical Bayes estimate of the credibility
factor Z for Territory A.
Problem 87.7 ‡
You are given the following information on towing losses for two classes of
insureds, adults and youths:
Exposures
Exposure
Year Adult Youth Total
1996 2000 450 2450
1997 1000 250 1250
1998 1000 175 1175
1999 1000 125 1125
Total 5000 1000 6000
Pure Premium
Year Adult Youth Total
1996 0 15 2.755
1997 5 2 4.400
1998 6 15 7.340
1999 4 1 3.667
Weighted Average 3 10 4.167
You are also given that the estimated variance of the hypothetical means is
17.125.
Determine the nonparametric empirical Bayes credibility premium for the
youth class.
88 SEMIPARAMETRIC EMPIRICAL BAYES CREDIBILITY ESTIMATION661
88 Semiparametric Empirical Bayes Credibility Es-

timation
In this section, we look at semiparamteric methods where parametric as-
sumptions concerning X|Θ are made, while the prior distribution of the risk
parameter Θ remains unspecified and nonparametric (and thus the name
semiparametric).
Let Xij denote the average number of claims for policyholder i in year
j where i = 1, 2, · · · , r. Suppose that the number of claims, given Θi ,
mij Xij |Θi has a Poisson distribution with parameter mij θi . Under these
assumptions, we have
1 mij θi
µ(θi ) =E[Xij |Θi ) = E(mij Xij |Θi ) = = θi
mij mij
1 mij θi
v(θi ) =mij Var(Xij |Θi ) = Var(mij Xij |Θi ) = = θi .
mij mij
It follows that
µ = v = E(Θi )
and an unbiased estimator for both µ and v is X. That is, µ̂ = v̂ = X.
Next, we turn our attention to estimating a. We first note that
Var(Xij ) =Var[E(Xij |Θi )] = E[Var(Xij |Θi )]

=Var[µ(Θi )] + E[v(Θi )] = a + v.
Now, if we assume ni = mij = 1, then (see Section 87)

r
X
E[ (Xi1 − X)2 ] = a(r − 1) + v(r − 1)
i=1
r
" #
1 X
E (Xi1 − X)2 = a + v.
r−1
i=1
1 Pr
Hence, r−1 i=1 (Xi1 − X)2 is an unbiased estimator of a + v and therefore
r
1 X
â = (Xi1 − X)2 − v̂.
r−1
i=1
Example 88.1 ([1])

In the past year, the distribution function of automobile insurance policy-
holders the number of claims is given by:
# of claims # of insureds
0 1563
1 271
2 32
3 7
4 2
Total 1875
Estimate the credibility premium for the number of claims next year for
a policyholder with 2 claims. Assume a conditional Poisson distribution
function of the number of claims for each policyholders.
Solution.
We have: r = 1875, ni = 1, mi1 = 1 and Xi1 |Θi is Poisson with parameter
θi . The estimators of µ, v, and a are found as follows:
1875
1 X 0(1563) + 1(271) + 2(32) + 3(7) + 4(2)
X= Xi1 = = 0.194
1875 1875
i=1
µ̂ =v̂ = 0.194
1875
X
(Xi1 − X)2 =1563(0 − 0.194)2 + 271(1 − 0.194)2 + 32(2 − 0.194)2
i=1
+7(3 − 0.194)2 + 2(4 − 0.194)2 = 423.3355
423.3355
â = − 0.194 = 0.032
1874
0.194
k̂ = = 6.06
0.032
1
Ẑ = = 0.14.
1 + 6.06
The estimated credibility premium for the number of claims for each poli-
cyholder is
0.14Xi1 + (0.86)(0.194).
For a policyholder with two claims, Xi1 = 2 so that the premium is
0.14(2) + (0.86)(0.194) = 0.44684

Practice Problems
Problem 88.1
Suppose that mij Xij |Θi has a binomial distribution with parameters (mij , θi ).
Express a in terms of µ and v.
Problem 88.2 ‡
You are given:
(i) During a single 5-year period, 100 policie s had the following total claims
experience:
Number of claims Number
in Year 1 through of
Year 5 Policies
0 46
1 34
2 13
3 5
4 2
(ii) The number of claims per year follows a Poisson distribution.
(iii) Each policyholder was insured for the entire period.
A randomly selected policyholder had 3 claims over the period.
Using semiparametric empirical Bayes estimation, determine the Bhlmann
estimate for the number of claims in Year 6 for the same policyholder.
Problem 88.3 ‡
You are given:
(i) During a 2-year period, 100 policies had the following claims experience:
Number of claims Number
in Year 1 through of
Year 2 Policies
0 50
1 30
2 15
3 4
4 1
(ii) The number of claims per year follows a Poisson distribution.
(iii) Each policyholder was insured for the entire 2-year period.
A randomly selected policyholder had one claim over the 2-year period.
Using semiparametric empirical Bayes estimation, determine the Bühlmann

estimate for the number of claims in Year 3 for the same policyholder.
Problem 88.4 ‡
You are given:
(i) Over a three-year period, the following claim experience was observed
for two insureds who own delivery vans:
Year
Insured 1 2 3
A Number of Vehicles 2 2 1
B Number of Vehicles N/A 3 2
Number of claims N/A 2 3
(ii) The number of claims for each insured each year follows a Poisson dis-
tribution.
Determine the semiparametric empirical Bayes estimate of the claim fre-
quency per vehicle for Insured A in Year 4.
Problem 88.5 ‡
For a group of auto policyholders, you are given:
(i) The number of claims for each policyholder has a conditional Poisson
distribution.
(ii) During Year 1, the following data are observed for 8000 policyholders:
Number of claims Number of Policies
0 5000
1 2100
2 750
3 100
4 50
5+ 0
A randomly selected policyholder had one claim in Year 1.
Determine the semiparametric empirical Bayes estimate of the number of
claims in Year 2 for the same policyholder.
Problem 88.6 ‡
For a portfolio of motorcycle insurance policyholders, you are given:
(i) The number of claims for each policyholder has a conditional Poisson
distribution.
(ii) For Year 1, the following data are observed:
# of claims # of insureds
0 2000
1 600
2 300
3 80
4 20
Total 3000
Determine the credibility factor, Z, for Year 2.
Problem 88.7 ‡
The following information comes from a study of robberies of convenience
stores over the course of a year:
(i) Xi is the number of robberies of the ith store, with i = 1, 2, · · · , 500.
500
X
(ii) Xi = 50.
i=1
500
X
(iii) Xi2 = 220.
i=1
(iv) The number of robberies of a given store during the year is assumed to
be Poisson distributed with an unknown mean that varies by store.
Determine the semiparametric empirical Bayes estimate of the expected
number of robberies next year of a store that reported no robberies dur-
ing the studied year.
Basics of Stochastic
Simulation
In this chapter we look at some stochastic simulation procedures to imitate

or simulate financial and insurance problems. The term ”stochastic” is used
so that to emphasize the step of the simulation procedure where values are
randomnly chosen from probability distributions.
667
668 BASICS OF STOCHASTIC SIMULATION
89 The Inversion Method for Simulating Random

Variables
A procedure of simulation can consist of the following four steps:
Step 1. Select the appropriate model X.

Step 2. For j = 1, 2, · · · , n generate simulated( also known as pseudoran-
dom) values x1 , x2 , · · · , xn .
Step 3. An approximation of the cdf of X is the cdf of the empirical dis-
tribution Fn (s) generated by x1 , x2 , · · · , xn .
Step 4. Statistical quantities such as mean, variance, percentiles, etc. are
found using the empirical cdf Fn (s).
For such procedure, two important questions are in place: The first question
deals with finding the pseudorandom values. The second question is the size
of n.
One method for finding the pseudorandom values is the inversion method23
which we discuss next.
Let X be a continuous random variable with cdf FX (x). Since F is strictly

increasing, it has an inverse F −1 defined on the interval (0, 1). Thus, for
0 < u < 1, the equation FX (x) = u has a unique solution x. Let U be the
uniform distribution on (0, 1). Then
FF −1 (U ) (x) = Pr(FX−1 (U ) ≤ x) = Pr(U ≤ FX (x)) = FU (FX (x)) = FX (x).
X
Hence, FX−1 (U ) has the distribution function FX . Thus, we can simulate a

realization of X by simulating a realization of FX−1 (U ).
Example 89.1
A random number generated from a uniform distribution on (0, 1) is 0.3102.
Using the inversion method, calculate the simulated value of X assuming X
to have the Pareto distribution with parameters α = 3 and θ = 1000.
Solution.
We are seeking x such that FX (x) = 0.3102. That is, x satisfies the equation
3
1000
1− = 0.3012.
1000 + x
23
This procedure is sometimes referred to with the phrase small uniform random num-
bers correspond to small simulated values.
89 THE INVERSION METHOD FOR SIMULATING RANDOM VARIABLES669
Solving this equation, we find x = 131.772
Example 89.2 ‡
You wish to simulate a value, Y, from a two point mixture. With probability
0.3, Y is exponentially distributed with mean 0.5. With probability 0.7, Y
is uniformly distributed on [−3, 3]. You simulate the mixing variable where
low values correspond to the exponential distribution. Then you simulate
the value of Y, where low random numbers correspond to low values of Y.
Your uniform random numbers from [0, 1] are 0.25 and 0.69 in that order.
Calculate the simulated value of Y.
Solution.
The statement ”low random numbers correspond to low values of Y ” implies
that the inversion method is to be used. The value 0.25 is used to simulate
the mixture and the number 0.69 is used to simulate the value of Y. We are
told that for low simulating mixing variable, the exponential distribution
must be used. Since 0.25 < 0.3, this satisfies the criterion. That is, the
exponential distribution is to be used. By the inversion method, the simu-
lated value of Y is the number y such that Pr(Y ≤ y) = 0.69. That is, y is
y
the solution to the equation 1 − e− 0.5 = 0.69. Solving this equation, we find
y = 0.5855
The inversion method can also be used to generate discrete or mixed-type

variables. In the case of a mixed-type distribution, suppose FX (x) has a
jump at x = c. Let FX (c− ) = a and FX (c) = b > a. For any uniform num-
ber a ≤ u < b, the equation FX (x) = u has no solution. In this case, we
choose c as the simulated value.
Example 89.3

0.3x, 0≤x<1
FX (x) =
0.3 + 0.35x, 1 ≤ x ≤ 2.
Determine the simulated values of x resulting from the uniform numbers:

0 ≤ u1 < 0.3, 0.3 ≤ u2 < 0.65, and 0.65 ≤ u3 ≤ 1.
Solution.
To find the simulated value of 0 ≤ u1 < 0.3, we solve the equation 0.3x1 = u1
u1
obtaining x1 = 0.3 . For any uniform number 0.3 ≤ u2 < 0.65, the simulated
value is x2 = 1. Note that Pr(0.3 ≤ U < 0.65) = 0.65 − 0.3 = 0.35 =
Pr(X = 1). Finally, to find the simulated value of 0.65 ≤ u3 ≤ 1, we solve

−0.3
the equation 0.3 + 0.35x3 = u3 obtaining x3 = u30.35
It is possible that FX (x) is constant for a ≤ x ≤ b, say FX (x) = c in

that interval. In this case, every value in a ≤ x ≤ b is a simulated value of
c. One chooses b as the simulated value corresponding to u = c. See Remark
89.1.
Example 89.4
Suppose that

 0.5x, 0 ≤ x < 1.2
FX (x) = 0.6, 1.2 ≤ x < 2.4
0.5x − 0.6, 2.4 ≤ x ≤ 3.2.

Determine the simulated values of x resulting from the uniform numbers

0.5, 0.6, and 0.8.
Solution.
For 0 ≤ x < 1.2, we have 0 ≤ FX (x) < 0.6. Since 0.5 is in that range, the
simulad value resulting from 0.5 is found by solving the equation 0.5x1 = 0.5
which implies x1 = 1. Next, we see that FX (x) = 0.6 for 1.2 ≤ x ≤ 2.4, For
thia case, x2 = 2.4. Finally, 0.6 ≤ 0.8 ≤ 1, so the corresponding simulated
value is found by solving the equation 0.5x3 −0.6 = 0.8 resulting in x3 = 2.8
Recall that a discrete distribution has jumps at the possible values of the
random variable and is constant in between, two features covered in the
previous two examples.
Example 89.5
Simulate values from a binomial distribution with m = 2 and q = 0.3 using
uniform numbers.
Solution.
[x]
X m
FX (x) = q i (1 − q)m−i
i
i=0
where [x] is the greatest integer less than or equal to x. Thus, for m = 2
and q = 0.3, we have

 0,
 x < 0,
0.49, 0 ≤ x < 1

FX (x) =

 0.91, 1 ≤ x < 2.
1, x ≥ 2.

For 0 ≤ u < 0.49, the simulated value is x = 0. For 0.49 ≤ u < 0.91, the
simulated value is x = 1. For 0.91 ≤ u < 1, the simulated value is x = 2
Remark 89.1
Note that FX (x) = 0.49 for all 0 ≤ x < 1. But Pr(0.49 ≤ U < 0.91) =
0.91 − 0.49 = Pr(X = 1). This is the motivation for choosing the largest
value in an interval where the cdf is constant.
The second question that we look at is the number of simulated values

needed to achieve a certain objective such as estimating the mean of X. We
illustrate this question in the next example.
Example 89.6 ([1])

Suppose X has the Pareto distribution with parameters α = 3 and θ = 1000.
Use simulation to estimate the mean of X. Stop the simulation when you
are 95% confident that the simulated mean is within 1% of the population
mean. Assume that the central limit theorem is applicable.
Solution.
The empirical estimate of µ = E(X) is x. The central limit theorem tells us
that X n is approximately normal so that we can write
0.95 =Pr(|X n − µ| ≤ 0.01µ)

=Pr(0.99µ ≤ X n ≤ 1.01µ)

0.01µ Xn − µ 0.01µ
=Pr − √ ≤ √ ≤ √
σ/ n σ/ n σ/ n

0.01µ 0.01µ
=Pr − √ ≤ Z ≤ √ .
σ/ n σ/ n
Our goal is achieved when
0.01µ 38, 416σ 2

√ ≥ 1.96 =⇒ n ≥ .
σ/ n µ2
Since we do not know σ 2 and µ2 , we estimate them with the sample variance
and mean. Thus, we cease simulation when
38416s2
n≥
x2
We can apply a similar sort of idea to estimating a probability.
Example 89.7 ([1])

Suppose X has the Pareto distribution with parameters α = 3 and θ = 1000.
Use simulation to estimate FX (1000). Stop the simulation when you are 95%
confident that the simulated mean is within 1% of the population mean.
Assume that the central limit theorem is applicable.
Solution.
Let Pnn be the empirical estimator of FX (1000) where Pn is the number of
values at or below 1000 after n simulations (see Sections 49). The central
limit theorem tells us that Pnn is approximately normal with mean FX (1000)
and variance FX (1000)[1 − FX (1000)]/n. (See Section 53). Using Pnn as an
estimator of FX (1000) and arguing as in the previous example, we arrive at

n − Pn
n ≥ 38416
Pn
Practice Problems
Problem 89.1 ‡
To estimate E(X), you have simulated X1 , X2 , X3 , X4 , and X5 with the
following results:
1 2 3 4 5.
You want the standard deviation of the estimator of E(X) to be less than
0.05. Estimate the total number of simulations needed.
Problem 89.2 ‡
A company insures 100 people age 65. The annual probability of death for
each person is 0.03. The deaths are independent.
Use the inversion method to simulate the number of deaths in a year. Do
this three times using:
u1 =0.20
u2 =0.03
u3 =0.09.
Calculate the average of the simulated values.
Problem 89.3 ‡
You simulate observations from a specific distribution F (x), such that the
number of simulations N is sufficiently large to be at least 95 percent con-
fident of estimating F (1500) correctly within 1 percent.
Let P represent the number of simulated values less than 1500. Determine
which of the following could be values of N and P.
(A) N = 2000 P = 1890
(B) N = 3000 P = 2500
(C) N = 3500 P = 3100
(D) N = 4000 P = 3630
(E) N = 4500 P = 4020
Problem 89.4 ‡
You are planning a simulation to estimate the mean of a non-negative ran-
dom variable. It is known that the population standard deviation is 20%
larger than the population mean.
Use the central limit theorem to estimate the smallest number of trials
needed so that you will be at least 95% confident that the simulated mean
is within 5% of the population mean.
Problem 89.5 ‡
Simulation is used to estimate the value of the cumulative distribution func-
tion at 300 of the exponential distribution with mean 100.
Determine the minimum number of simulations so that there is at least a
99% probability that the estimate is within ±1% of the correct value.
Problem 89.6 ‡
You are simulating a compound claims distribution:
(i) The number of claims, N, is binomial with m = 3 and mean 1.8.
(ii) Claim amounts are uniformly distributed on {1, 2, 3, 4, 5}.
(iii) Claim amounts are independent, and are independent of the number of
claims.
(iv) You simulate the number of claims, N, then the amounts of each of those
claims, X1 , X2 , · · · , XN . Then you repeat another N, its claim amounts, and
so on until you have performed the desired number of simulations.
(v) When the simulated number of claims is 0, you do not simulate any
claim amounts.
(vi) All simulations use the inverse transform method, with low random
numbers corresponding to few claims or small claim amounts.
(vii) Your random numbers from (0, 1) are
0.7 0.1 0.3 0.1 0.9 0.5 0.5 0.7 0.3 0.1
Calculate the aggregate claim amount associated with your third simulated
value of N.
90 APPLICATIONS OF SIMULATION IN ACTUARIAL MODELING675
90 Applications of Simulation in Actuarial Model-

ing
In this section, we look at how simulation is used in solving actuaruial
models.
Example 90.1 ‡
Unlimited claim severities for a warranty product follow the lognormal dis-
tribution with parameters µ = 5.6 and σ = 0.75.
You use simulation to generate severities. The following are six uniform
(0, 1) random numbers:
0.6179 0.4602 0.9452 0.0808 0.7881 0.4207.
Using these numbers and the inversion method, calculate the average pay-
ment per claim for a contract with a policy limit of 400.
Solution.
Let X be the lognormal random variable with µ = 5.6 and σ = 0.75. Its cdf
is given by
ln x − 5.6
F (x) = Φ
0.75
where Φ is the cdf of the standard normal distribution. Using the table of
the standard normal distribution, we find

ln x1 − 5.6 ln x1 − 5.6
Φ = 1−0.6179 = 0.3821 =⇒ = 0.3 =⇒ x1 = 338.66.
0.75 0.75
In a similar manner, we find

ln x2 − 5.6 ln x2 − 5.6
Φ = 1−0.4602 = 0.5398 =⇒ = −0.1 =⇒ x3 = 250.89.
0.75 0.75

ln x3 − 5.6 ln x3 − 5.6
Φ = 1−0.9452 = 0.0548 =⇒ = 1.6 =⇒ x3 = 897.85.
0.75 0.75

ln x4 − 5.6 ln x4 − 5.6
Φ = 1−0.0808 = 0.9192 =⇒ = −1.4 =⇒ x4 = 94.63.
0.75 0.75

ln x5 − 5.6 ln x5 − 5.6
Φ = 1−0.7881 = 0.2119 =⇒ = 0.8 =⇒ x5 = 492.75.
0.75 0.75

ln x6 − 5.6 ln x1 − 5.6
Φ = 1−0.4207 = 0.5793 =⇒ = −0.2 =⇒ x6 = 232.76.
0.75 0.75
Because of the policy limits, the payments are
338.66 250.89 400 94.63 400 232.76.
The average payment per claim is

338.66 + 250.89 + 400 + 94.63 + 400 + 232.76
= 286.16
6
Example 90.2 ‡
Total losses for a group of insured motorcyclists are simulated using the
aggregate loss model and the inversion method.
The number of claims has a Poisson distribution with λ = 4. The amount
of each claim has an exponential distribution with mean 1000.
The number of claims is simulated using u = 0.13. The claim amounts are
simulated using u1 = 0.05, u2 = 0.95, and u3 = 0.10 in that order, as needed.
Determine the total losses.
Solution.
Finding the first three probabilities of the Poisson distribution, we obtain
e−4 (40 )
p0 = = 0.0183
0!
e−4 (41 )
p1 = = 0.0733
1!
e−4 (42 )
p2 = = 0.1463.
2!
Thus, FX (x) = 0.0183 for 0 ≤ x < 1, FX (x) = 0.0916 for 1 ≤ x < 2
and FX (x) = 0.2381 for 2 ≤ x < 3. Since u = 0.13 falls in the interval
(0.0916, 0.2381), the simulated number of claims is 2.
For the simulated amount of claim corresponding to u1 = 0.05, we have
x1
1 − e− 1000 = 0.05 =⇒ x1 = 51.29.
Likewise, x2
1 − e− 1000 = 0.95 =⇒ x2 = 2995.73.
Since the simulated number of claims is 2, there is no need to consider u3 .
In conclusion, the total losses are 51.29 + 2995.73 = 3047.02
Example 90.3 ‡
Losses for a warranty product follow the lognormal distribution with under-
lying normal mean and standard deviation of 5.6 and 0.75 respectively.
You use simulation to estimate claim payments for a number of contracts

with different deductibles.
The following are four uniform (0, 1) random numbers:
0.6217 0.9941 0.8686 0.0485.
Using these numbers and the inversion method, calculate the average pay-
ment per loss for a contract with a deductible of 100.
Solution.
Let X be the lognormal random variable with µ = 5.6 and σ = 0.75. Its cdf
is given by
ln x − 5.6
F (x) = Φ
0.75
where Φ is the cdf of the standard normal distribution. Using the table of
the standard normal distribution, we find

ln x1 − 5.6 ln x1 − 5.6
Φ = 1−0.6217 = 0.3783 =⇒ = 0.31 =⇒ x1 = 341.21.
0.75 0.75
In a similar manner, we find

ln x2 − 5.6 ln x2 − 5.6
Φ = 1−0.9941 = 0.0059 =⇒ = 2.52 =⇒ x2 = 1790.05.
0.75 0.75

ln x3 − 5.6 ln x3 − 5.6
Φ = 1−0.8686 = 0.1314 =⇒ = 1.12 =⇒ x3 = 626.41.
0.75 0.75

ln x4 − 5.6 ln x4 − 5.6
Φ = 1−0.0485 = 0.9515 =⇒ = −1.66 =⇒ x4 = 77.87.
0.75 0.75
The amounts after the deductible are
241.21 1690.05 526.41 0.
The average payment per loss is

241.21 + 1690.05 + 526.41 + 0
= 614.42
4
Example 90.4 ‡
You are the consulting actuary to a group of venture capitalists financing a
search for pirate gold.
It’s a risky undertaking: with probability 0.80, no treasure will be found,
and thus the outcome is 0.

The rewards are high: with probability 0.20 treasure will be found. The
outcome, if treasure is found, is uniformly distributed on [1000, 5000].
You use the inverse transformation method to simulate the outcome, where
large random numbers from the uniform distribution on [0, 1] correspond to
large outcomes.
Your random numbers for the first two trials are 0.75 and 0.85. Calculate
the average of the outcomes of these first two trials.
Solution.
Let X denote the outcome of the hunt. We first find the cdf of X. We are
told that F (0) = 0.80. Also, for 1000 ≤ x ≤ 5000, we have
0.2
F (x) = 0.8 + (x − 1000).
5000 − 1000
Note the presence of 0.2 in the second term of F (x). Without it, F (5000) =
1.8 > 1 which contradicts the definition of F (x) (i.e. 0 ≤ F (x) ≤ 1). So,
the cdf of X can be expressed as


 0, x<0
0.8, 0 ≤ x < 1000

F (x) =

 0.75 + 0.00005x, 1000 ≤ x ≤ 5000
1, x > 5000

By the inversion method, for u = 0.75 we find F (0.75) = 0. For u = 0.85,

we find F (0.85) = 2000. Thus, the average of the two outcomes is 0+2000
2 =
1000
Example 90.5 ‡
You are simulating the gain/loss from insurance where:
(i) Claim occurrences follow a Poisson process with λ = 32 per year.
(ii) Times between successive claims follow an exponential distribution with
mean 1.5.
(iii) Each claim amount is 1, 2 or 3 with p(1) = 0.25, p(2) = 0.25, and
p(3) = 0.50.
(iv) Claim occurrences and amounts are independent. Successive time claims
are independent.
(v) The annual premium equals expected annual claims plus 1.8 times the
standard deviation of annual claims.
(vi) i = 0.
You use 0.25, 0.40, 0.60, and 0.80 from the unit interval and the inversion
method to simulate time between claims.

You use 0.30, 0.60, 0.20, and 0.70 from the unit interval and the inversion
method to simulate claim size.
Calculate the gain or loss from the insurer’s viewpoint during the first 2
years from this simulation.
Solution.
Let N be the number of claims, X the size of a claim, and S the total annual
gain/loss. We have
E(X) =1(0.25) + 2(0.25) + 3(0.50) = 2.25

E(X 2 ) =12 (0.25) + 22 (0.25) + 32 (0.50) = 5.75
Var(X) =5.75 − 2.252 = 0.6875
2
E(S) =E(N )E(X) = (2.25) = 1.5
3
2 2
Var(S) = (0.6875) + (2.25)2 = 3.8333.
3 3
The annual premium is
3 √
+ 1.8 3.8333 = 5.0242
2
and the 2-year premium is 2(5.0242) = 10.05.
The time between successive claims has an exponential distribution with
mean 1.5. The simulated inter-claim times are ti where ui = 1 − e−1.5ti . The
table below lists the times between claims.
u 0.25 0.40 0.60 0.80
t 0.43 0.77 1.37 2.41
We see that 2 claims occur before time 2. First claim at time 0.43 and
second claim at time 1.2.
The cdf of X is defined by


 0, x≤0
0.25, 0 < x ≤ 1

FX (x) =

 0.50, 1 < x ≤ 2
1, x≥1

Let u1 = 0.30. Since 0.25 < 0.30 ≤ 0.5, we have x1 = 2. Let u2 = 0.60.
Since 0.5 < 0.6 ≤ 1, we find x2 = 3. Finally, the gain to the insurer is
10.05 − (2 + 3) = 4.95
Practice Problems
Problem 90.1 ‡
A dental benefit is designed so that a deductible of 100 is applied to annual
dental charges. The reimbursement to the insured is 80% of the remaining
dental charges subject to an annual maximum reimbursement of 1000.
You are given:
(i) The annual dental charges for each insured are exponentially distributed
with mean 1000.
(ii) Use the following uniform (0, 1) random numbers and the inversion
method to generate four values of annual dental charges:
0.30 0.92 0.70 0.08.
Calculate the average annual reimbursement for this simulation.
Problem 90.2 ‡
For a warranty product you are given:
(i) Paid losses follow the lognormal distribution with µ = 13.294 and σ =
0.494.
(ii) The ratio of estimated unpaid losses to paid losses, y, is modeled by
y = 0.801x0.851 e−0.747x
where
x = 2006 − contract purchase year.
The inversion method is used to simulate four paid losses with the following
four uniform (0,1) random numbers:
0.2877 0.1210 0.8238 0.6179.
Using the simulated values, calculate the empirical estimate of the average
unpaid losses for purchase year 2005.
Problem 90.3 ‡
You are given:
(i) The cumulative distribution for the annual number of losses for a poli-
cyholder is:
n FN (n)
0 0.125
1 0.312
2 0.500
3 0.656
4 0.773
5 0.855
.. ..
. .
(ii) The loss amounts follow the Weibull distribution with θ = 200 and τ = 2.
(iii) There is a deductible of 150 for each claim subject to an annual maxi-
mum out-of-pocket of 500 per policy.
The inversion method is used to simulate the number of losses and loss
amounts for a policyholder:
(a) For the number of losses use the random number 0.7654.
(b) For loss amounts use the random numbers: 0.2738 0.5152 0.7537 0.6481 0.3153.
Use the random numbers in order and only as needed.
Based on the simulation, calculate the insurer’s aggregate payments for this
policyholder.
Problem 90.4 ‡
The price of a non dividend-paying stock is to be estimated using simula-
tion. It is known that: h i
2
(i) The price St follows the lognormal distribution: ln SS0t ∼ N α − σ2 t, σ 2 t .
(ii) S0 = 50, α = 0.15, and σ = 0.30.
Using the following uniform (0, 1) random numbers and the inversion method,
three prices for two years from the current date are simulated.
0.9830 0.0384 0.7794
Calculate the mean of the three simulated prices.
Problem 90.5 ‡
You are given:
(i) For a company, the workers compensation lost time claim amounts follow
the Pareto distribution with α = 2.8 and θ = 36.
(ii) The cumulative distribution of the frequency of these claims is:
n FN (n)
0 0.5556
1 0.8025
2 0.9122
3 0.9610
4 0.9827
5 0.9923
.. ..
. .
(iii) Each claim is subject to a deductible of 5 and a maximum payment of

30.
Use the uniform (0, 1) random number 0.981 and the inversion method to
generate the simulated number of claims.
Use as many of the following uniform (0, 1) random numbers as necessary,
beginning with the first, and the inversion method to generate the claim
amounts.
0.571 0.932 0.303 0.471 0.878
Calculate the total of the company’s simulated claim payments.
Problem 90.6 ‡
N is the random variable for the number of accidents in a single year. N
follows the distribution:
Pr(N = n) = 0.9(0.1)n−1 , n = 1, 2, · · · .
Xi is the random variable for the claim amount of the ith accident. Xi
follows the distribution:
g(xi ) = 0.01e−0.01xi , xi > 0, i = 1, 2, · · · .
Let U and V1 , V2 , · · · be independent random variables following the uni-

form distribution on (0, 1). You use the inverse transformation method with
U to simulate N and Vi to simulate Xi with small values of random numbers
corresponding to small values of N and Xi .
You are given the following random numbers for the first simulation:
u v1 v2 v3 v4
0.05 0.30 0.22 0.52 0.46
Calculate the total amount of claims during the year for the first simulation.
Problem 90.7 ‡
Annual dental claims are modeled as a compound Poisson process where the
number of claims has mean 2 and the loss amounts have a two-parameter
Pareto distribution with θ = 500 and α = 2.
An insurance pays 80% of the first 750 of annual losses and 100% of annual
losses in excess of 750.
You simulate the number of claims and loss amounts using the inverse trans-
form method with small random numbers corresponding to small numbers
of claims or small loss amounts.
The random number to simulate the number of claims is 0.8. The random
numbers to simulate loss amounts are 0.60, 0.25, 0.70, 0.10 and 0.80.
Calculate the total simulated insurance claims for one year.
91 Estimating Risk Measures Using Simulation
In this section, we use simulation to estimate the risk measures VaR and
TVaR. Let y1 ≤ y2 ≤ · · · ≤ yn be a simulated sample of size n of a random
variable. For a percentile p, let k = [pn] + 1, where [x] is the largest integer
less than or equal to x. The estimators of VaR and TVaR are
n
X
1
VaR
d p (X) = yk and TVaR
\ p (X) =
n−k+1 yi .
i=k
\ p (X) is the mean of the sample {yk , yk+1 , · · · , yn }. The

Note that TVaR
variance of this sample is
n
1 X
s2p = \ p (X))2 .
(yi − TVaR
n−k
i=k
It has been shown that an asymptotically unbiased estimator of the variance

of TVaR
\ p (X) is given by
s2p + p[TVaR d p (X)]2

\ p (X) − VaR
Var(
d TVaR
\ p (X)) = .
n−k+1
Example 91.1
Consider the following sample of simulated values
45 107 210 81 153 189.
Find VaR
d p (X), TVaR
\ p (X), and Var(
d TVaR
\ p (X)) for p = 0.6.
Solution.
Rearrnaging the values in increasing order to obtain
45 81 107 153 189 210.

91 ESTIMATING RISK MEASURES USING SIMULATION 685
We have the following sequence of calculation
k =[pn] + 1 = [3.6] + 1 = 4
VaR
d p (X) =153
n
1 X
TVaR
\ p (X) = yi
n−k+1
i=k
1
= (153 + 189 + 210) = 184
6−4+1
n
1 X
s2p = \ p (X))2
(yi − TVaR
n−k
i=k
1
= [(153 − 184)2 + (189 − 184)2 + (210 − 184)2 ] = 831
6−4
s2 + p[TVaR
\ p (X) − VaRd p (X)]2
Var( \ p (X)) = p
d TVaR
n−k+1
831 + 0.6(184 − 153)2
= = 469.20
6−4+1
Practice Problems
Problem 91.1
Consider the following sample of simulated values
93 109 120 123 150 153 189 190 195 200.
Find VaR
d p (X), TVaR
\ p (X), and Var(
d TVaR
\ p (X)) for p = 0.3.
92 The Bootstrap Method for Estimating Mean Square

Error
Let θ be a quantity related to a distribution X. This quantity can be the
mean of the distribution, the variance, quantile, etc. Let θ̂ denote an esti-
mator of θ. The mean square error of θ̂ is defined by
M SE(θ̂) = E[(θ̂ − θ)2 ].
The bootstrap method is a way of estimating M SE(θ̂). This is how the
method work in terms of simulation: We select a random sample of size n
from the distribution X and we create from this sample the empirical distri-
bution. Using this distribution, we calculate the estimator θ̂. The next step
is to ”resample” from the original sample. Since the size of the original sam-
ple is n, there are nn possible bootstrap samples(with repetition allowed).
For each bootstrap sample we calculate θ̂i , i = 1, 2, · · · , nn . The next step
is to calculate the square deviation (θ̂i − θ̂)2 . The bootstrap is the average
of these deviations, that is an estimate to M SE(θ̂) is
n n
1 X
M
\ SE(θ̂) = n (θ̂i − θ̂)2 .
n
i=1
Example 92.1
A sample of size 2 contains the values x1 = 2 and x2 = 4. Calculate the
MSE of the unbiased estimator of the population mean using the bootstrap
method.
Solution.
The original sample mean is x = 2+4
2 = 3. Since the original sample is of size
2
2, there are 2 = bootstrap samples. The table below provides the various
samples along with their mean and square deviation.
Sample Xi (X i − X)2
2,2 2 (2 − 3)2 = 1
2,4 3 (3 − 3)2 = 0
4,2 3 (3 − 3)2 = 0
4,4 4 (4 − 3)2 = 1
Total 2
Hence, the bootstrap estimate is given by
1+0+0+1
M
\ SE(µ̂) = = 0.5
4
92 THE BOOTSTRAP METHOD FOR ESTIMATING MEAN SQUARE ERROR687
Example 92.2 ‡
You are given a random sample of two values from a distribution function
F : x1 = 1 and x2 = 3.
You estimate θ(F ) = Var(X) using the estimator g(X1 , X2 ) = 21 2i=1 (Xi −
P
X)2 where X = X1 +X2

2
.
Determine the bootstrap approximation to the mean square error.
Solution.
The estimator for the original sample is g0 = g(1, 3) = 1. We have the
following table.
Sample X1 X2 Xi gi = g(X1 , X2 ) (gi − g0 )2
1 1 3 2 1 0
2 1 1 1 0 1
3 3 1 2 1 0
4 3 3 3 0 1
Total 2
Hence, the bootstrap estimate is given by
2
M
\ SE(g) = = 0.5
4
Example 92.3 ‡
A sample of claim amounts is {300, 600, 1500}. By applying the deductible
to this sample, the loss elimination ratio for a deductible of 100 per claim
is estimated to be 0.125. You are given the following simulations from the
sample:
Simulation Claim
1 600 600 1500
2 1500 300 1500
3 1500 300 600
4 600 600 300
5 600 300 1500
6 600 600 1500
7 1500 1500 1500
8 1500 300 1500
9 300 600 300
10 600 600 600
Determine the bootstrap approximation to the mean square error of the
estimate.
Solution.
We have
Simulation X1 X2 X3 LER (LER − 0.125)2
1 600 600 1500 0.111111 0.000193
2 1500 300 1500 0.090909 0.001162
3 1500 300 600 0.125000 0.000000
4 600 600 300 0.200000 0.005625
5 600 300 1500 0.125000 0.000000
6 600 600 1500 0.111111 0.000193
7 1500 1500 1500 0.066667 0.003403
8 1500 300 1500 0.090909 0.001162
9 300 600 300 0.250000 0.015625
10 600 600 600 0.166667 0.001736
Total 0.029099
0.029099
The bootstrap estimate to the mean sqaure error is 10 = 0.0029099
92 THE BOOTSTRAP METHOD FOR ESTIMATING MEAN SQUARE ERROR689
Practice Problems
Problem 92.1 ‡
You are given a random sample of two values from a distribution function
F : x1 = 1 and x2 = 3.
You estimate θ(F ) = Var(X) using the estimator g(X1 , X2 ) = 2i=1 (Xi −
P
X)2 where X = X1 +X2
2
.
Determine the bootstrap approximation to the mean square error.
Problem 92.2 ‡
With the bootstrapping technique, the underlying distribution function is
estimated by which of the following?
(A) The empirical distribution function
(B) A normal distribution function
(C) A parametric distribution function selected by the modeler
(D) Any of (A), (B) or (C)
(E) None of (A), (B) or (C).
Problem 92.3 ‡
Three observed values of the random variable X are:
1 1 4
You estimate the third central moment of X using the estimator:
3
1X
g(X1 , X2 , X3 ) = (Xi − X)3 .
3
i=1
Determine the bootstrap estimate of the mean-squared error of g.
Problem 92.4 ‡
For a policy that covers both fire and wind losses, you are given:
(i) A sample of fire losses was 3 and 4.
(ii) Wind losses for the same period were 0 and 3.
(iii) Fire and wind losses are independent, but do not have identical distri-
butions.
Based on the sample, you estimate that adding a policy deductible of 2 per
wind claim will eliminate 20% of the insured loss.
Determine the bootstrap approximation to the mean square error of the
estimate.
Problem 92.5 ‡
The random variable X has the exponential distribution with mean θ. Cal-
culate the mean-squared error of X 2 as an estimator of θ2 .
Answer Key
Section 1
1.1 Deterministic
1.2 Stochastic
1.3 Stochastic
1.4 Stochastic
1.5 Mostly stochastic
Section 2
2.1 (a) Ac = B, B c = A and C c = {1, 3, 4, 5, 6}

(b)
A ∪ B = {1, 2, 3, 4, 5, 6}
A ∪ C = {2, 4, 6}
B ∪ C = {1, 2, 3, 5}
(c)
A∩B =∅
A ∩ C = {2}
B∩C =∅
(d) A and B are mutually exclusive as well as B and C
691
692 ANSWER KEY
2.2 Note that Pr(E) > 0 for any event E. Moreover, if S is the sample
space then
∞ ∞
i
X 1X 1 1 1
Pr(S) = Pr(Oi ) = = · 1 =1
2 2 2 1− 2
i=1 i=0
Now, if E1 , E2 , · · · is a sequence of mutually exclusive events then
∞ X
X ∞ ∞
X
Pr(∪∞
n=1 Ei ) = Pr(Onj ) = Pr(En )
n=1 j=1 n=1
where En = {On1 , On2 , · · · }. Thus, Pr defines a probability function.
2.3 0.5
2.4 0.56
2.5 0.66
2.6 0.52
2.7 0.05
2.8 0.6
2.9 0.48
2.10 0.04
2.11 The probability is given is the figure below.

693
The probability that the first ball is red and the second ball is blue is
PR(RB) = 0.3.
2.12
The probability that the first ball is red and the second ball is blue is
PR(RB) = 6/25.
2.13 0.173
2.14 0.467
2.15 0.1584
2.16 0.0141
694 ANSWER KEY
2.17 0.29
2.18 0.42
2.19 0.22
2.20 0.657
Section 3
3.1 (a) Discrete (b) Discrete (c) Continuous (d) Mixed.
3.2
x 2 3 4 5 6 7 8 9 10 11 12
1 2 3 4 5 6 5 4 3 2 1
p(x) 36 36 36 36 36 36 36 36 36 36 36
3.3

 p, x=1
p(x) = 1 − p, x=0
0, x 6= 0, 1.

3.4 1/9
3.5 0.469
3.6 0.132
3.7 0.3
3.8 α = 1.
3.9


 0 x<1
0.25 1 ≤ x<2



F (x) = 0.75 2 ≤ x < 3
0.875 3≤x<4




1 4≤x

695
3.10 (a) (
0, x<0
F (x) = 1
1− (1+x)a−1
, x ≥ 0.
(b)

0, x<0
F (x) = −kxα
1−e , x ≥ 0.
3.11
n
X
F (n) =P (X ≤ n) = P (X = k)
k=0
n
1 2 k
X
=
3 3
k=0
2 n+1

11− 3
=
3 1 − 32
n+1
2
=1 −
3
3.12 (a) 0.135 (b) 0.233 (c)

x
1 − e− 5

x≥0
F (x) =
0 x<0
696 ANSWER KEY
ex
3.13 f (x) = F 0 (x) = (1+ex )2
1
3.14 (a) We have that S(0) = 1, S 0 (x) = − 201
(100 − x)− 2 ≤ 0, s(x) is
right continuous, and S(100) = 0. Thus, S satisfies the properties of a sur-
vival function.
1
1
(b) F (x) = 1 − S(x) = 1 − 10 (100 − x) 2 .
(c) 0.092
3.15 0.149
x2
3.16 F (x) = 1 − S(x) = 100 , x≥0
3.17 (a) 0.3 (b) 0.3

0
3.18 h(x) = − SS(x)
(x)
= 12 (100 − x)−1
3.19 S(x) = e−µx , F (x) = 1 − e−µx , and f (x) = F 0 (x) = µe−µx .
3.20 1/480
Section 4
4.1 (b) np(1 − p)
4.2 (b) λ
4.3 (c) (1 − p)p−2
1
4.5 (c) λ2
697
4.6 (a)
Z ∞
1 x
E(X) = xe− θ xα−1 dx
θΓ(α) 0
Z ∞
θ 1 −x α
= e θ x dx
Γ(α) 0 θ
θΓ(α + 1)
=
Γ(α)
=αθ
(b)
Z ∞ α α−1
1 x θ
E(X ) =2
x2 e − θ dx
Γ(α) 0 x
Z ∞
1 1 x
= xα+1 α e− θ dx
Γ(α) 0 θ
2 Z ∞
θ Γ(α + 2) xα+1 x
= α+2
e− θ dx
Γ(α) 0 θ Γ(α + 2)
2
θ Γ(α + 2)
=
Γ(α)
where the last integral is the integral of the pdf of a Gamma random variable
with parameters (α + 2, θ). Thus,
θ2 Γ(α + 2) θ2 (α + 1)Γ(α + 1)
E(X 2 ) = = = θ2 α(α + 1).
Γ(α) Γ(α)
Finally,
V ar(X) = E(X 2 ) − (E(X))2 = θ2 α2 (α + 1) − α2 θ2 = αθ2
4.7 4
4.8 1,417,708,752
4.9 730,182,499.20
4.10 0
4.11 9
698 ANSWER KEY
4.12 0.3284
4.13 We have
Z ∞
µ0n = xn f (x)dx
0
Z ∞
=A xB+n e−Cx dx
0
∞
xB+n e−Cx B + n ∞ B+n−1 −Cx
Z
=A − + x e dx
C
0 C 0
Z ∞
B+n
= AxB+n−1 e−Cx dx
C 0
B+n
= E(X n−1 ).
C
(B+1)(B+2)
4.14 µ = B+1
C and µ02 = C2
4.15 √2
B+1
3(B+3)
4.16 B+1
4.17 0.5

Rx Rx  0, x<0
4.18 (a) F (x) = −∞ 0.005tdt = 0 0.005tdt = 0.0025x2 , 0 ≤ x ≤ 20
1, x > 20

40 200
(b) The mean is 3 and the variance is 9
(c) 0.354
4.19 16
R∞ ∞
axk a a
4.20 (a) E(X k ) = 1 xa+1
dx = k−1 xk−a
= a−k , 0 < k < a.
1
(b) √ 1
a(a−2)
4.21 −1.596
θ αθ2
4.22 The mean is α−1 and the variance is (α−1)2 (α−2)
q
α
4.23 α−2
699
4.24 2
4.25 1.7
Section 5
5.1
Amount of loss 750 500 1200

Insurance payment 250 0 700
5.2
4 1
Pr(X ≤ 45) = =
12 3
5
Pr(X ≤ 67) =
12
7
Pr(X ≤ 84) =
12
8
Pr(X ≤ 93) =
12
11
Pr(X ≤ 100) =
12
Pr(X ≤ 102) =1.
5.3 µ01 = 75.8333 and µ02 = 6312.8333
5.4 0.509175
5.5 12 (100 − d) for 0 < d < 100 and 0 otherwise.
5.6 108
1
5.7 λ
5.8 308,8728
e−λd
5.9 λ
700 ANSWER KEY
1 −λd
5.10 λ2
e (2 − e−λd )
1
5.11 160
5.12 94.84
5.13 88.4
θ xφ+1
5.14 θ − φ+1 −x− θφ (φ+1)
5.15
Z ∞
S(x) =Pr(X > x) = (1 + 2t2 )e−2t dt
x
∞
= −(1 + t + t2 )e−2t x = (1 + x + x2 )e−2x
for x ≥ 0 and 0 otherwise.

1+x+ 21 x2 )
(b) 1+x+x 2
5.16 We have
1 − SX (y + d) − [1 − SX (d)]
SY P (y) =1 − FY P (y) = 1 −
Sx (d)
SX (y + d) + SX (d) SX (y + d)
=1 − = .
SX (d) SX (d)
5.17 (a)
f (y + 100) (0.001 + 0.00002(y + 100))e−0.005(y+100)

fY P (y) = =
1 − F (100) 1.4e−0.5
(0.0003 + 0.00002y)e−0.005y
=
1.4
3 1
= + y e−0.005y , y > 0
1400 70000
and 0 otherwise.
(b) E(Y P ) = 2200 P
7 and Var(Y ) =
3560000
49
425
5.18 E[(X − 10)+ ] = and Var[(X − 10)+ ] = 36
5.19 d = 6.
701
5.20 175
5.21 1875
5.22 3.43
5.23 6.259
Section 6
6.1 we can either say 1120 is the twentieth percentile or 1120 is the one-fifth
quantile
6.2 3
6.3 2
6.4 the median is 1 and the 70th percentile is 2
6.5 The median is M = 0.3466. This means that half the people get in
line less than 0.3466 minutes (about 21 seconds) after the previous person,
while half arrive more than 0.3466 minutes later
6.6 0.693
6.7 998.72
6.8 3 ln 2
6.9 The median is 0.8409
6.10 3659
√
6.11 a + 2 ln 2
6.12 − ln (1 − p)
6.13 − ln [2(1 − p)]

702 ANSWER KEY
6.14 2
6.15 72.97
6.16 0.4472
6.17 6299.61
6.18 2.3811
6.19 50
6.20 2.71
Section 7
7.1 0.2119
7.2 0.9876
7.3 0.0094
7.4 0.692
7.5 0.1367
7.6 0.0088
7.7 0
7.8 23
7.9 0.0162
7.10 6,342,637.5
7.11 0.8185
7.12 16
703
7.13 0.1587
7.14 0.9887
7.15 0.0244
7.16 0.9985
7.17 0.1056
7.18 0.8413
7.19 0.8201
7.20 0.224
Section 8
2 1
8.1 E(X) = λ2
and V ar(X) = λ2
8.2 A normal random variable with mean µ1 + µ2 and variance σ12 + σ22
8.3 0.70
8.4 41.9
2 +4t
8.5 e13t
8.6 4
8.7 −28
8.8 2
8.9 5000
8.10 10560
8.11 (tp + 1 − p)n

704 ANSWER KEY
tp
8.12 1−t(1−p) provided that |t| < (1 − p)−1
8.13 True
8.14 ta PX (tb )
1 1−p
8.15 E(X) = p and Var(X) = p2
8.16 0.4t2 + 0.2t3 + 0.2t5 + 0.2t8

7
8.17 13 t−2 + 61 t3 + 18 tπ + 83 t 2
t3
8.18 2−t
8.19
x −1 0 1 2 3
16 32 24 8 1
p(x) 81 81 81 81 81
8.20 E(X) = 3.5 and Var(X) = 6.25
Section 9
9.1 Let m > 0. Then there is M > 0 such that ebx ≥ xm+1 which is
equivalent to saying that xm e−bx ≤ x1 for x ≥ M. By the comparison test of
improper integrals we find that
Z ∞
xm e−bx < ∞.
0
Since E(X k ) is an integral of the above form, we conclude that E(X k ) < ∞
for all k > 0. That is, the distribution of X is light-tailed.
R∞
9.2 Since X is heavy-tailed, we have E(X k ) = 0 xk fX (x)dx = ∞ for
some k > 0. Now, let t > 0. Let N be large enough so that etx ≥ xk for all
x ≥ N. Hence,
Z N Z ∞ Z N Z ∞ Z ∞
k tx k k
x fX (x)dx+ e fX (x)dx ≥ x fX (x)dx+ x fX (x)dx = xk fX (x)dx = ∞.
0 N 0 N 0
RN R∞
Since 0 xk fX (x)dx < ∞, we conclude that N etx fX (x)dx = ∞
705
9.3 We have
Z ∞ Z ∞
tx
MX (t) = e fX (x)dx ≥ etx fX (x)dx = ∞
0 N
9.4 From Table C, we have
θk Γ(α + k)
E(X k ) =
Γ(α)
for all k > 0. Hence, the Gamma distribution is light-tailed.
9.5 From Table C/ Exam 4, we have

k
E(X k ) = θk Γ 1 −
τ
provided that k < τ. Since E(X k ) is only valid for k < τ, the distribution is
heavy-tailed.
9.6 The Pareto distribution has a more heavy-tailed than the Gamma dis-
tribution,
9.7 The Weibull distribution has a lighter tail than the inverse Weibull
distribution.
9.8 X is more heavy-tailed than Y
9.9 X and Y have similar or proportional tails.
9.10 X is more light-tailed than Y
9.11 X has a heavier tail than Y
9.12
Weibull X
Inverse Pareto X
Normal X
Loglogistic X
706 ANSWER KEY
9.13
Paralogistic X
Lognormal X
Inverse Gamma X
Inverse Gaussian X
9.14
Inverse Paralogistic X
Inverse Exponential X
SX (x)
9.15 limx→∞ SY (x) =∞
SX (x)
9.16 limx→∞ SY (x) =∞
SX (x)
9.17 limx→∞ SY (x) =∞
9.18 c > 0
9.19 X has a heavier tail than Y
9.20 The tail of X is heavier than that of Y which in turn is heavier than
the tail of Z
Section 10
h i
d f (x+y) y(1−α) y α−2 − yθ
10.1 dx f (x) = x2
1+ x e < 0 for α > 1.
10.2 For α = 1, the Gamma distribution is just the exponential function

which has a constant hazard rate
10.3 We have
f (x) τ xτ −1
h(x) = = .
S(x) θτ
707
Hence,
τ (τ − 1)xτ −2
h0 (x) = .
θτ
Thus, h(x) is increasing (light-tailed distribution) for τ > 1 and decreasing
(heavy-tailed distribution) for 0 < τ < 1
10.4 X is light-tailed
10.5 We have
α+1
αθα (x + θ)α+1

x+θ
H(x) = α+1
· = .
(x + y + θ) αθα x+y+θ
Hence, α
0 x+θ y
H (x) = (α + 1) > 0.
x+y+θ (x + y + θ)2
Thus, H(x) is increasing and by Theorem 10.1, h(x) is decreasing which
shows that the Pareto distribution is heavy-tailed
10.6 We have
f (x) f 0 (x)
lim h(x) = lim = lim
x→∞ x→∞ S(x) x→∞ −f (x)
d d h xi
= − lim [ln (f (x))] = − lim (α − 1) ln x −
x→∞ dx x→∞ dx θ

1 α−1 1
= lim − = .
x→∞ θ x θ
10.7 We have
R∞
x SX (t)dt −SX (x) 1
lim e(x) = lim = lim = lim .
x→∞ x→∞ S(x) x→∞ −fX (x) x→∞ h(x)
10.8 θ
10.9 Since 0 < α < 1, the hazard rate function is decreasing and hence
e(x) is increasing. The result follows from the fact that e(0) = αθ and
e(∞) = θ.
708 ANSWER KEY
10.10 Since α > 1, the hazard rate function is inecreasing and hence e(x)
is decreasing. The result follows from the fact that e(0) = αθ and e(∞) = θ
αθα θα
10.11 For this distribution we have fX (x) = (x+θ)α+1
and SX (x) = (x+θ)α .
fx (x) α
Hence, h(x) = SX (x) = x+θ . Thus,
1
lim e(x) = = ∞.
x→∞ limx→∞ h(x)
This shows that e(x) is increasing and hence the distribution is heavy-tailed
10.12 limx→∞ e(x) = ∞
1 2 2
10.13 (a)S(x) = (x+1)2 , f (x) = (x+1)3 , and h(x) = x+1 .
(b) E(X) = 1 and E(X 2 ) = ∞ so that X is heavy-tailed.
10.14 Since the hazard rate is nonicreasing, X is a heavy-tailed
10.15 Since e(x) is nondecreasing, X is a heavy-tailed
10.16 S(x+y)
S(x) is nonincreasing so that e(x) is nonincreasing and therefore
X is light-tailed
10.17 f (x+y)
f (x) is nondecreasing so that h(x) is nonincreasing. Thus, X is
heavy-tailed
Section 11
2 ∞
R∞
2 2
11.1 (a) S(x) = x 2te−t dt = −e−t = e−x .
0
S(x) 2
(b) fe (x) = = √2 e−x for x > 0 and 0 otherwise.
E(X) π
1 2x+3
11.2 he (x) = e(x) = 2x+5
x2
Rx 1
e(0) − 0 e(t) dt
11.3 S(x) = e(x) e = (1 + x)e−x− 2 , x>0
e(x) x2
11.4 Se (x) = e(0) S(x) = e−x− 2 , x>0
11.5 0.6559
709
24 4
11.6 (a) E(X) = 3 and E(X 2 ) = 5 (b) 5
10
10 9
11.7 S(x) = 10+9x
11.8 λ
Section 12
12.1 By (P3), we have ρ(0) = ρ(α0) = αρ(0). Choosing α 6= 1, we con-

clude that ρ(0) = 0. Since there are no losses, no capital is required to
support the risk
12.2 This follows from (P3) and (P4)
12.3 Check the properties (P1)-(P4)
12.4 (b)
12.5 (a)
12.6 (a) and (c)
12.7 All three are correct
12.8 Simple algebra
12.9 0
12.10 We have
p
ρ(L + α) =E(L + α) + β Var(L + α)
p
=E(L) + α + β Var(L)
=ρ(L) + α
p
ρ(αL) =E(αL) + β Var(αL)
=αE(L) + αβVar(L)
=αρ(L)
where α > 0.
710 ANSWER KEY
Section 13
13.1 a(1 − p) + pb
13.2 1.8974
13.3 2.0227
13.4 60
1
13.5 θ(1 − p)− α
13.6 40
13.7 10,000
13.8 347.21
13.9 5
13.10 VaR0.96 = 400. This says that there is 4% chance the losses will
exceed 400
Section 14
14.1 (a) e(x) = 21 (b − x) (b) TVaRp (L) = 21 [(a + b) + p(b − a)]
14.2 (a) π0.90 =≈ 1.8974 and e(1.8974) = 0.0051 (b) TVaR0.90 (L) = 1.9025
14.3 π0.85 = 7 and TVaR0.85 (L) = 10.25
14.4 (a) θ = 1000 (b) e(100) = 220 (c) π0.95 = 647.55 (d) TRaV0.95 (L) =
867.55
14.5 TVaR0.75 (L) = 756
14.6 1.3
14.7 1.651
711
14.8 2.02
14.9 0.82
14.10 100,000
14.11 120.62
Section 15
15.1 We have
x x x x
FcX (x) = Pr(cX ≤ x) = Pr(X ≤ ) = FX = 1 − e−λ cθ = FX ( ).
c c c
This is an exponential distribution with parameter cθ
y 2
15.2 FY (y) = 1 − e−( c )
y−cθ
15.3 FY (y) = cθ
15.4 Let Y = cX, c > 0. We have

y
FY (y) =Pr(Y ≤ y) = Pr X ≤
c
y −α
=e−( cθ ) .
This is a Fréchet distribution with parameters cθ and α.
15.5 Let Y = cX, c > 0. We have

y
FY (y) =Pr(Y ≤ y) = Pr X ≤
c
1
=1 − y γ α
1 + cθ
This is a Burr distribution with parameters α, cθ, and γ
15.6 (a) θ (b) θ (c) θ (d) θ
15.7 (1) is the only correst answer

712 ANSWER KEY
15.8 Let Y = cX. Then

y
FY (y) =Pr(Y ≤ y) = Pr X ≤
! c
ln yc − µ

=Φ
σ

ln y − ln cµ
=Φ .
ln cσ
This is a lognormal distribution with parameters ln cµ and ln cσ and conse-
quently has no scale paramter
15.9 The Gamma distribution with parameters α and θ has a cdf FX (x) =
1
R xθ α−1 −t
Γ(α) 0 t e dt. Let Y = cX. Then
y
FY (y) =Pr(Y ≤ y) = Pr X ≤
c
Z y
1 cθ
= tα−1 e−t dt.
Γ(α) 0
This is a Gamma distribution with parameters α and cθ
15.10 100
x
e− θ
15.11 Letting α = 1, we obtain fX (x) = θ which is the pdf of an expo-
nential distribution
15.12 0.0295
Section 16
16.1 0.0949
16.2 100
16.3 E(Y1 ) = 2 and E(Y2 ) = 6
16.4 E(X) = 1300 and Var(X) = 6935000
16.5 0.0568
713
16.6 35
α 1 α 2 α N
θ1 θ2 θN
16.7 FX (x) = 1 − a1 x+θ 1
− a2 x+θ2 − · · · − aN x+θN where
PN
aj > 0 and i=1 ai = 1.
α α α
α θ 1 α θ 2 α θ N
h i h i h i
fX (x) = a1 (x+θ1 )1α1 +1 + a2 (x+θ2 )2α2 +1 + · · · + aN (x+θN )NαN +1
1 2 N
α α α
α1 θ1 1 α2 θ 2 2 αN θ N

a1 +a +···+a N
2 N
(x+θ1 )α1 +1 (x+θ2 )α2 +1 (x+θN )αN +1
hX (x) =
θ1
α1
θ2
α2
θN
α
N
a1 x+θ +a2 x+θ +···+aN x+θ
1 2 N
16.8 15
16.9 400
16.10 0.146
16.11 0.7566
Section 17
2478 2283152
17.1 E(X) = 13 , Var(X) = 169 and the mode is 10
1351 1127
17.2 E(X ∧ 105) = 13 and eX (x) = 9
17.3 0.1659
17.4
x 94 104 134 180 210 350 524

1 3 2 4 1 1 1
pX (x) 13 13 13 13 13 13 13
The cdf is defined by
1
FX (x) = 13 number of elements in the sample that are ≤ x
17.5 0.61
17.6 17,566,092.92
714 ANSWER KEY
17.7  1 1 1
 8 · 13 = 104 , 90 ≤ x ≤ 98
1 3 3



 8 · 13 = 104 , 100 ≤ x ≤ 108
1 2 2

· = 104 , 130 ≤ x ≤ 138


 8 13
1 4 4
fX (x) = 8 · 13 = 104 , 176 ≤ x ≤ 184
1 1 1
· = 104 , 206 ≤ x ≤ 214



 8 13
1 1 1
· = 104 , 346 ≤ x ≤ 354


8 13


1 1 1
· 520 ≤ x ≤ 528

8 13 = 104 ,
and 0 otherwise.
Section 18
18.1 0.75
18.2 We have that Y = 0.5X. Thus,
fY (y) = 2fX (2y) = 3y 2
for 0 < y < 1 and 0 otherwise.

Thus,
FY (y) = y 3 , 0 ≤ y ≤ 1
and FY (y) = 1 for y > 1. Also,
SY (y) = 1 − y 3 , 0 ≤ y < 1
and 0 for y > 1

0.75(2000)
18.3 E(Y ) = 0.75E(X) = 3−1 = 750 and σY = 1299.04
18.4 We have
α
y y −α θ
FY (y) = FX =1− 1+ =1− .
θ θ y+θ
This is the cdf of a Pareto distribution with parameters α and θ. The pdf is
αθα
fY (y) = .
(y + θ)α+1
18.5 We have
ln y − µ
FY (y) = Φ .
σ
715
Thus,
!
z

−µ

z ln θ ln z − (µ + ln θ)
FZ (z) = FY =Φ =Φ .
θ σ σ
Hence, Z is a lognormal distribution with parameters µ + ln θ and σ.

Rx
18.6 The cdf of X is FX (x) = 1 3t−4 dt = 1 − x−3 and Pr(Y > 2.2) =
1 − (1 − 2−3 ) = 81 = 0.125
18.7 We have

w
FW (w) =FZ
1+r
1 w2 1 w
= 2 2
+ .
2 w + 1000(1 + r) 2 w + 1000(1 + r)
Thus, W is an equal
√ mixture of a loglogistic distribution with parameters
γ = 2 and θ = 10 10(1 + r) and a Pareto distribution with parameters
α = 1 and θ = 1000(1 + r)
Section 19
19.1 In the transformed case, we have

τ
yτ

− yθ −
FY (y) = 1 − e and fY (y) = τθ y τ −1 e θ
.
In the inverse transformed case, we have

y −τ −τ
− − yθ
and fY (y) = τθ y −τ −1 e
θ
FY (y) = e .
In the inverse case, we have

1 1
− yθ 1 − yθ
FY (y) = e and fY (y) = y2 θ
e
19.2 We have
!α !α
θ y
FY (y) = 1 − FX (y −1 ) = 1 = 1 .
y +θ y+ θ
1
Y has the inverse Pareto distribution with parameters α and θ
716 ANSWER KEY
19.3 fY (y) = y −2 fX (y −1 ) = 1
y2
· 2
y = 2
y3
for y > 1 and 0 otherwise
1
−y
1 y 1−α e
19.4 fY (y) = 1
f (y −1 )
y2 X
= y2 Γ(α) and 0 otherwise
τ y τ −1 1
19.5 fY (y) = τ y τ −1 fX (y τ ) = b for 0 ≤ y ≤ b τ and 0 otherwise
19.6 Y has an exponential distribution with parameter α
19.7 We have
ln y − µ
FY (y) = Φ
σ
and
1 ln y − µ 1 1 ln y−µ 2
fY (y) = fZ = √ e− 2 ( σ )
y σ yσ 2π
where Z is the standard normal distribution
19.8 fY (y) = y1 fX (ln y) = 1

2by for 1 < t < eb and 0 otherwise
19.9 We have
1
1 1 1 − ln y y −( θ +1)
fY (y) = fX (ln y) = e θ =
y yθ θ
for y > 1 and 0 otherwise
19.10 0.25
Section 20
5α−2
20.1 Var(X) = 12(α−1)2 (α−2)
b2
4
20.2 fX (x) = (4+x)2
for x > 0 and 0 otherwise
20.3 1.7975
20.4 E(Λ) = αβ = E(X)
θxγ
20.5 FX (x) = 1+θxγ
20.6 0.6094
717
20.7 14
20.8 0.61
20.9 0.75
Section 21
21.1 SX (x) = Mλ (−x) = (1 + θx)−α
x
21.2 SX (x) = (1+x) ln (1+x) , x>0
x
21.3 MΛ (x) = e θ , x ≥ 0
21.4 a(x) = −1
−x −1
21.5 SX (x) = MΛ (−x) = ee
21.6 SX (x) = 1 −x − e−11x )

10x (e
21.7 SX (x) = MΛ (−A(x)) = (1 + θxγ )−1

√
21.8 SX|Λ (x|λ) = e−λ( 1+θx−1)
21.9 SX (x) = MΛ (−A(x)) = (1 + θx)−α
0 (x) = −M 0 (−A(X))(−A(x))0 = a(x)M 0 [−A(x)]

21.10 fX (x) = −SX Λ Λ
a(x)MΛ0 [−A(x)]
fX (x)
21.11 hX (x) = SX (x) = MΛ (−A(x))
Section 22
22.1
1

f (x) = α, 0<x<c
(1 − α)θe−θx , x > c.
22.2 0.9252
22.3 3 + ln 5
718 ANSWER KEY
22.4 5.61
22.5 (
1
1000+θ , 0 < x ≤ 1000
f (x) = 1 1000
− xθ
1000+θ e , x > 1000
θ
and 0 otherwise.
22.6 461.78
Section 23
23.1 We have
ln 1 + ατ

lim ln w1 = lim
τ →∞ τ →∞ (α + τ − 1 )−1
2
−1
1 + ατ (− τα2

= lim
τ →∞ −(α + τ − 1/2)−2
1 2

α −1 α
= lim 1 + α +1−
τ →∞ τ τ 2τ
=α.
Let
(ξ/x)γ α+τ

w2 = 1 + .
τ
We have
(ξ/x)γ

lim ln w1 = lim (α + τ ) ln 1 +
τ →∞ τ →∞ τ
h γ
i−1
1 + (ξ/x)
τ (−τ −2 )(ξ/x)γ
= lim
τ →∞ −(α + τ )−2
(ξ/x)γ −1 ξ γ

α 2
= lim 1 + 1+
τ →∞ τ x τ
γ
ξ
= .
x
23.2 For large α, Stirling’s formula gives

1 1
Γ(α) ≈ e−α αα− 2 (2π) 2 .
719
1
−1
Also, we let ξ = θτ γ so that θ = ξτ γ .
Using this and Stirling’s formula in the pdf of a transformed beta distribu-
tion, we find
Γ(α + τ )γxγτ −1
fX (x) =
Γ(α)Γ(τ )θγτ (1 + xγ θ−γ )γ+τ
1 1
e−α−τ (α + τ )α+τ − 2 (2π) 2 γxγτ −1
≈ 1 1
Γ(α)e−τ (τ )τ − 2 (2π) 2 ξ γτ τ −τ (1 + xγ ξ −γ τ )γ+τ
α+τ − 1 −γα−1
e−α 1 + ατ 2
γx
= −α−τ −γα −γ(τ
Γ(α)τ ξ γ(τ +α) ξ x +α) (1 + xγ ξ −γ τ )γ+τ
α+τ − 1 −γα−1
e−α 1 + ατ 2
γx
= h γ α+τ
i
Γ(α)ξ −γα 1 + (ξ/x) τ
Let
α α+τ − 21

w1 = 1 + .
τ
From the previous problem, limτ ∞ w1 = eα . Now, let
(ξ/x)γ α+τ

w2 = 1 + .
τ
Then ξ γ
lim w2 = e( x ) .
τ →∞
Hence,
γξ γα
lim fX (x) = ξ γ
τ →∞
Γ(α)xγα+1 e( x )
which is the pdf of an inverse transformed Gamma distribution
23.3 Let ξ = τ θ. Then

τ θxτ −1
finverser Pareto (x) =
(x + θ)τ +1
ξ
θ −1 θ −θ

ξ
= 2 1+ 1+
x x x
ξ
ξe− x
→
x2
720 ANSWER KEY
which is the pdf of an inverse exponential distribution with parameter ξ
Section 24
24.1 The pdf of the Gamma distribution can be written as

1 x
f (x, θ) = −α
xα−1 e− θ .
Γ(α)θ
xα−1
We have: q(θ) = θ−α , p(x) = Γ(α) , and r(θ) = − 1θ
24.2 The pdf of X is
e−λ λx e−λx ex ln λ
f (x, λ) = = .
x! x!
1
Thus, p(x) = x! , q(λ) = eλ , and r(λ) = ln |lambda
24.3 The pdf of X is

p
x ln
p C(m, x)e 1−p
f (x, p) = C(m, x)px (1−p)m−x = C(m, x) (1−p)m = .
1−p (1 − p)−m

p
Thus, p(x) = C(m.x), q(p) = (1 − p)−m , and r(p) = ln 1−p
24.4 E(X) = λ and var(X) = λ
24.5 E(X) = mp and var(X) = mp(1 − p)
Section 25
25.1 0.91873
25.2 5e−5
25.3 0.1412
25.4 5 cars per week
Section 26
721
26.1 For the Poisson distribution the variance is equal to the mean. For
the negative binomial and geometric the variance exceeds the mean. So the
answer is (d)
26.2 0.75
26.3 E(N ) = 3 and Var(N ) = 12
26.4 E(N ) = 6 and Var(N ) = 18
26.5 192
26.6 r = 2 and β = 4
26.7 PN (z) = [1 − β(z − 1)]−1

q
26.8 CV (N ) = 1+β rβ
Section 27
27.1 The Poisson distribution has a variance equal to the mean. The neg-
ative binomial and geometric distributions have a varaince exceeding the
mean. The binomial distribution has a variance less than the mean. Thus,
the answer is (a)
27.2 38.34
27.3 E(N ) = 1.4 and Var(N ) = 0.42
27.4 0.0057
27.5 6.2784
27.6 0.172
27.7
m 0 1 2 3 4
pm 0.1074 0.2684 0.3020 0.2013 0,0881
F (m) 0.1074 0.3758 0.6778 0.8791 0.9672
722 ANSWER KEY
27.8 N1 represents the number of successes in m1 independent trials, each

of which results in a success with probability q. Similarly, N2 represents the
number of successes in m2 independent trials, each of which results in a
success with probability q. Hence, as N1 and N2 are assumed to be indepen-
dent, it follows that N1 + N2 represents the number of successes in m1 + m2
independent trials, each of which results in a success with probability q. So
N1 + N2 is a binomial random variable with parameters (m1 + m2 , q)
Section 28
28.1 For the given recursice equation, we find a = − 31 < 0 and b = 4.

q
Thus, N is a binomial distribution. From − 1−q = − 13 we find q = 14 . From
q
(m + 1) 1−q = 4 we find m = 11
28.2 E(N ) = 2.75 and Var(N ) = 2.065
28.3 0.125
28.4 3
28.5 0.0118
28.6 8
28.7 a = 0.2 and b = 0.8
28.8 (III)
28.9 0.3012
28.10 0.8
28.11 0.09
Section 29
723
29.1 We have
1−pM

pM 1−p0
0
pk
k
= 1 = 1 − pM
0
pTk 1−p0 pk
29.2 We have
1 − pM 1 − pM

M
E(N ) = [PNM ]0 (1) = 0
PN0 (1) = 0
E(N )
1 − p0 1 − p0
1−pM

29.3 E[N M (N M − 1)] = 0
1−p0 E[N (N − 1)] and
2
1 − pM 1 − pM 1 − pM

M 0 0 0
Var(N ) = E[N (N −1)]+ E(N )− E(N )
1 − p0 1 − p0 1 − p0
29.4
p0 =e−λ = e−1 = 0.3679

a =0
b =1
1
p1 =p0= 0.3679
1
1
p2 =p1 = 0.1839
2
E(N ) 1
E(N T ) = = = 1.582
1 − p0 1 − 0.3679

T T 1 1
E[N (N − 1)] = E[N (N − 1)] = = 1.582
1 − p0 1 − 0.3679
Var(N T ) =1.582 + 1.582 − 1.5822 = 0.661276

1 z
29.5 (a) PNM (z) = 2 1+ 3−2z (b) E(N M ) = 1.5 and Var(N M ) = 2.25
29.6 0.5
Section 30
724 ANSWER KEY
30.1
β 1
a= = = 0.5
1+β 1+1
β
b =(r − 1) = −0.75
1+β

T T 0.75
p2 =p1 0.5 − = 0.106694
2

0.75
pT3 =pT2 0.5 − = 0.026674
3
pM T
1 =(1 − p0 M )p1 = 0.341421
pM T
2 =(1 − p0 M )p2 = 0.042678
pM T
3 =(1 − p0 M )p3 = 0.010670
β
30.2 E(N ) = ln (1+β)
β 2
30.3 E[N (N − 1)] = PN00 (1) = ln (1+β)
h i
β β
30.4 Var(N ) = ln (1+β) 1 + β − ln (1+β)
30.5 We have
∞
X ∞
X
PNT (z) = pTn z n = pTn z n
n=0 n=1
∞
1 X
= pn z n
1 − p0
n=1
∞
1 X p0
= pn z n −
1 − p0 1 − p0
n=0
PN (z) − p0
=
1 − p0
[1−β(z−1)]−r −(1+β)−r
30.6 PNT (z) = 1−(1+β)−r
0 (1)
PN rβ
30.7 E(N T ) = [PNT ]0 (1) = 1−p0 = 1−(1+β)−r
725
00 (1)
PN r(r+1)β 2
30.8 E[N T (N T − 1)] = [PNT ]00 (1) = 1−p0 = 1−(1+β)−r
h i2
r(r+1)β 2 rβ rβ
30.9 Var(N T ) = 1−(1+β)−r
+ 1−(1+β)−r
− 1−(1+β)−r
Section 31
31.1 (
1 − e−0.25 y=0
fY L (y) = −( y+50
2
0.0002(y + 50)e 100 ) y > 0.
(
1 − e−0.25 y=0
FY L (y) = y+50 2
1 − e−( 100 ) y > 0.
(
e−0.25 y=0
SY L (y) = y+50 2
e−( 100 ) y > 0.
31.2
2 −0.01y
fY P (y) =0.0002(y + 50)e−0.0001y
2 −0.01y
FY P (y) =1 − e−0.0001y
2 −0.01y
SY P (y) =e−0.0001y
hY P (y) =0.0002(y + 50).
31.3 E(Y L ) = 1 −500,000

1000 e
31.4 E(Y P ) = 0.001
31.5 2000e−400θ
31.6 1708.70
(b−d)2
31.7 12
31.8 (a) 188.75 (b) 269.64
θα
31.9 (a) (α−1)(θ+α)α−1
(b) ∞
726 ANSWER KEY
31.10 30
Section 32
32.1
d

fY L (y) = θ, y=0
1
θ, y > d.
d

θ, 0≤y≤d
FY L (y) = y
θ, y > d.
1 − dθ , 0 ≤ y ≤ d

SY L (y) =
1 − yθ , y > d.

0, 0<y<d
hY L (y) = 1
θ−y , y > d.
32.2
1
fY P (y) = , y > d.
θ−d

0, 0≤y≤d
FY P (y) = y−d
θ−d , y > d.

1, 0≤y≤d
SY P (y) = θ−y
θ−d y > d.

0, 0<y<d
hY P (y) = 1
θ−y , y > d.
θ2 −d2 θ+d
32.3 E(Y L ) = 2θ and E(Y P ) = 2
32.4 340.83
32.5 900
ln 0.40
32.6 d = −θ
32.7 (a) 310 (b) 387.5
32.8 320.83
32.9 456
727
32.10 6400
Section 33
33.1 LER = 79.2%. The is the percentage of savings in claim payments

due to the presence of the deductible 30.
33.2 86.6%
33.3 1546
33.4 500
33.5 333.33
33.6 0.07418
33.7 0.5
33.8 510.16
33.9 0.625
Section 34
34.1 We have  −θu

 e , y=u
fY (y) = θe−θy , y < u
0, y>u

and
1 − e−θy , y < u

FY (y) =
1 y≥u
Ru
e−θx dx = 1θ 1 − e−θu

34.2 E(X ∧ u) = 0
θu

− 1+r
34.3 E((1 + r)X ∧ u) = (1+r)
θ 1 − e
34.4 48
728 ANSWER KEY
34.5 182.18
34.6 2011.80
34.7 5176.78
Section 35
35.1 1.115
35.2 990,938.89
35.3 0.4163
35.4 133
35.5 353.55
35.6 3031.06
35.7 85%
35.8 29.93
35.9 109.4
35.10 0.583
Section 36
36.1 8.0925
36.2 0.5553
36.3 The pgf of N L is PN L (z) = e0.5(z−1) . Hence,

PN P (z) = PN L [1 + v(z − 1)] = e0.5(0.5553)(z−1) = e0.27756(z−1) .
Note that N P is a Poisson distribution with mean λ − 0.27756
36.4 0.242444
729
36.5 PN P (z) = (1 − 0.11852(z − 1))−1
36.6 0.6304
36.7 0.4424
Section 37
37.1 (a) This is false, In a collective risk model, all the loss amounts are
identically distributed.
(b) This is true since the loss amounts need not all have the same distribu-
tion.
(c) This is true. In the collective risk model, N (the frequency random
variable) is determined independently of each of the subscripted Xs, the
severity random variables.
(d) This is false. If frequency is independent of severity, as it is in the col-
lective risk model, then this implies that the number of payments does not
affect the size of each individual payment
37.2 0
37.3 E(S) = 70 and Var(S) = 7900
37.4 This is a collective loss model
37.5 The probability generating function of N is given by
PN (z) = [1 − β(z − 1)]−r
so that the pgf can be expressed in the form
PN (z; α) = Q(z)α
where α = r and Q(z) = [1 − β(z − 1)]−1
Section 38
38.1 E(S 3 ) = E(N 3 )E(X)2 −3E(N 2 )E(X)2 +2E(N )E(X)2 +3E(N 2 )E(X)E(X 2 )−
3E(N )E(X)E(X 2 ) + E(N )E(X 3 )
730 ANSWER KEY
38.2 E[(S−E(S))3 ] = E(N )E[(X−E(X))3 ]+3E[(N −E(N ))2 ]E(X)E[(X−

E(X))2 ] + E[(N − E(N ))3 ]E(X)3
( ln y−µ )
Rx φ
ln (x−y)−µ
38.3 FX∗2 (x) = 0 Φ σ σy
σ
dy
38.4 We have
E(S) =0.18 + 0.182(2) + 0.0909(3) + 0.0182(4) = 0.8895

E(S 2 ) =0.18 + 0.182(22 ) + 0.0909(32 ) + 0.0182(42 ) = 2.0173
38.5 1.226
38.6 E(Y ) = 2.5 and Var(Y ) = 23.75
38.7 0.1587
38.8 0.1637
38.9 0.0681
38.10 E(S) = 5600 and Var(S) = 9, 710, 400
38.11 100
38.12 0.1003
38.13 0.242
38.14 0.1230
38.15 24
38.16 518
38.17 40
38.18 0.2483
731
38.19 0.0039
38.20 0.37
38.21 65.3
38.22 0.4207
38.23 0.0233
Section 39
39.1 1.014
39.2 2.25
39.3 1/3
39.4 2.064
39.5 25/16
39.6 2.3608
39.7 2.064
39.8 18.15
39.9 18.81
Section 40
732 ANSWER KEY
40.1 The mgf of S is

MS (z) =PN [MX (z)] = PN [(1 − θz)−1 ]
−r
= 1 − β[(1 − θz)−1 − 1]

r
1 − θz
=
[1 − θ(1 + β)z]
1 − θ(1 + β)z + β r

1
=
1+β 1 − θ(1 + β)z
r
1 β
= 1+
1+β 1 − θ(1 + β)z
r
β β
= 1+ −
(1 + β)[1 − θ(1 + β)z] 1 + β
r

β −1
= 1+ [1 − θ(1 + β)z] − 1
1+β
=PN∗ [MX ∗
(z)]
where r
β
PN∗ (z) = 1+ (z − 1)
1+β
β
is the pgf of the binomial distribution with parameters r and 1+β and
∗ −1
MX (z) = [1 − θ(1 + β)z] is the mgf of the exponential distribution with
mean θ(1 + β). Thus, the negative binomial-exponential model is equivalent
to the binomial exponential model.
40.2 This follows from Exercise 40.1 and

∞ n−1
X X (x/θ)j e−x/θ
FS (x) = 1 − Pr(N = n)
j!
n=1 j=0
x
β − θ(1+β)
40.3 FS (x) = 1 − 1+β e , x≥0
x
β −θ(1+β) , x ≥ 0. Note that S is a mixed distribution
40.4 fS (x) = θ(1+β)2e
with the discrete part Pr(S = 0) = FS (0) = (1 + β)−1 and the continuous
part is the exponential distribution with mean θ(1 + β)
40.5 Suppose that X1 , X2 , · · · , XN are independent Poisson random vari-

ables with parameters λ1 , · · · , λN . Then
t −1) t −1) t −1)
MS (t) = eλ1 (e · · · eλN (e = e(λ1 +···+λN )(e .
733
Hence, S is a Poisson random variable with parameter λ1 + · · · + λN
40.6 Suppose that X1 , X2 , · · · , XN are independent binomial random vari-

ables with parameters (n1 , p), (n2 , p), · · · , (nN , p). Then
MS (t) = (q + pet )n1 · · · (q + pet )nN = (q + pet )n1 +n2 +···nN .
Hence, S is a binomial random variable with parameters (n1 + · · · + nN , p).
40.7 Suppose that X1 , X2 , · · · , XN are independent negative binomial ran-

dom variables with parameters (r1 , q), (r2 , q), · · · , (rN , q). Then
PS (t) = [1−β(z−1)]−r1 [1−β(z−1)]−r2 · · · [1−β(z−1)]−rN = [1−β(z−1)]−(r1 +r2 +···+rN ) .
Hence, S is a negative binomial random variable with parameters (r1 + · · · +

rN , β). In particular, the family of geometric random variables is closed un-
der convolution.
40.8 Suppose that X1 , X2 , · · · , XN are independent gamma random vari-

ables with parameters (α1 , θ), (α2 , θ), · · · , (αN , θ). Then
MS (t) = (1 − θt)−α1 (1 − θt)−α2 · · · (1 − θt)−αN = (1 − θt)−(α1 +α2 +···+αN ) .
Hence, S is a gamma random variable with parameters (α1 + · · · + αN , θ).

In particular, the family of exponential random variables is closed under
convolution.
40.9 We have
∞ ∞
x X (x/2)j X
FS (x) =1 − e− 2 Pr(N = n)
j!
j=0 n=j+1
1 −x x
=1 − e 2 1+ .
2 10
40.10 MS (2) = 3.6 × 1048
Section 41
734 ANSWER KEY
41.1 fS (0) = fN (0) = e−0.04 = 0.9608 and
fS (1) =0.04fX (1)fS (0) = 0.04(0.5)(0.9608) = 0.019216

fS (2) =0.02[fX (1)fS (1) + 2fX (2)fS (0)]
=0.02[0.5(0.019216) + 2(0.4)(0.9608)] = 0.01556496
0.04
fS (3) = [fX (1)fS (2) + 2fX (2)fS (1) + 3fX (3)fS (0)]
3
=0.0041487371
fS (4) =0.01[fX (1)fS (3) + 2fX (2)fS (2) + 3fX (3)fS (1) + 4fX (4)fS (0)]
=0.00037586.
41.2 We have
0.04
fS (n) = [fX (1)fS (n − 1) + fX (2)fS (n − 2) + fX (3)fS (n − 3)]
n
0.04
= [0.5fS (n − 1) + 0.4fS (n − 2) + 0.1fS (n − 3)]
n
1
= [0.02fS (n − 1) + 0.016fS (n − 2) + 0.004fS (n − 3)].
n
41.3 0.15172
41.4 12
41.5 76
41.6 165
41.7 1.0001
41.8 0.3336
41.9 fS (1) = 0.3687 and fS (2) = 0.2055
41.10 0.0921
41.11 0.2883
735
Section 42
42.1 (a)
5
f0 =FX (5) = = 0.1
50
15 5
f1 =FX (15) − FX (5) = − = 0.2
50 50
25 15
f2 =FX (25) − FX (15) = − = 0.2
50 50
25
f3 =1 − FX (25) = 1 − = 0.5.
50
(b) 0.1935
42.2 0.0368
42.3 0.0404
42.4 We have
3 3
5 5
m10 + m11 = FX (6) − FX (3) = − = 0.150226
8 11
and
6
3(5)3 x
Z
3m10 + 6m11 = dx = 0.62897.
3 (x + 5)4
Solving this system we find m10 = 0.090796 and m11 = 0.05943.
42.5 We have f0 = m00 = 0.4922 and f3 = m01 + m10 = 0.2637 + 0.090796 =

0.3545.
42.6 0.03682
42.7 0.03236
Section 43
736 ANSWER KEY
43.1
FY L (y) =1 − v + vFY P (y)

 y

Pr(X>6)−Pr(X>6+ 0.75 )
1−v+v , y < 13.5

= Pr(X>6)
2 − v, y ≥ 13.5

where v = 0.15259
43.2 E(S) = 262.4621 and Var(S) = 487, 269, 766.1

y
Pr(X>30)−Pr(X>30+ 0.53 )
43.3 FY P (y) = 1−Pr(0.53Z > y) = Pr(X>30) and FY P (y) = 1
for y ≥ 164.30
43.4
f0 =FY P (15) = 0.2465

f1 =FY P (45) − FY P (15) = 0.3257
f2 =FY P (75) − FY P (45) = 0.1849
f3 =FY P (105) − FY P (75) = 0.0778
f4 =FY P (135) − FY P (105) = 0.0322
f5 =FY P (165) − FY P (135) = 0.058
fn =1 − 1 = 0, n = 6, 7, · · ·
43.5
fS (0) =e7(1−0.3257) = 112.179

n
5.18574 X
fS (n) = fn fS (n − j)
n
j=1
43.6 MY L (t) = 0.25918 + 0.74082((1 − 100t)−1
43.7 PN P (z) = e1.06813(z−1)
Section 44
737
44.1 E(S) = 100, 000 and Var(S) = 1.6002 × 1011

100
44.2 MS (t) = 0.80 + 0.20(1 − 0.001t)−1
100
44.3 PS (z) = MS [ln z] = 0.80 + 0.20(1 − 0.001 ln z)−1
44.4 (a) The mean is E(S) = 163 and the variance is Var(S) = 1107.77. (b)
217.75
44.5 E(S) = 352n and Var(S) = 88856n
Section 45
45.1 53
45.2 0.29
45.3 0.18
45.4 1975
45.5 λ = 1.52 or λ = 5.45
Section 46
46.1 5
46.2 0.2
46.3 151.52
n+8
46.4 18(n−1)2
√
46.5 20 10
1
46.6 n−1 θ
46.7 12
46.8 (D)
738 ANSWER KEY
Section 47
47.1 1.64
47.2 2.58
47.3 0.2
47.4 78.14 ≤ µ ≤ 81.86
47.5 0.1825 ≤ p ≤ 0.2175
Section 48
48.1 (a) H0 : µ = 18, 000 and H1 : µ < 18, 000

(b) H0 : µ = 18, 000 and H1 : µ 6= 18, 000
(c) H0 : µ = 18, 000 and H1 : µ > 18, 000
48.2 (i) Two-tailed (ii) Left-tailed (iii) Right-tailed
48.3 The null and alternative hypotheses are
H0 :µ ≥ 30
H1 :µ < 30.
The test statistic for the given sample is
20 − 30
z= √ = −3.727.
6/ 5
The rejection region is Z < −1.28. Since −3.727 < −1.28, so we reject the
null hypothesis in favor of the alternative.Thus, the mean time to find a
parking space is less than 30 minutes.
48.4 We reject the null hypothesis when the level of confidence is greater
than or equal to the p−value. Thus, the answer to the question is:(ii), (iii),
and (iv).
739
H0 :µ = 20
H1 :µ > 20.
The test statistic is

22.60 − 20
z= √ = 7.28.
2.50/ 49
The rejection region corresponding to α = 0.02 is Z > 2.06. Since 7.28 >
2.06, we reject H0 and conclude that the typical amount spent per customer
is more than $20.
H0 :µ = 16
H1 :µ 6= 16.
The test statistic is

16.32 − 16
z= √ = 2.19.
0.8/ 30
The rejection region corresponding to α = 0.10 is |Z| > 1.645. Since 2.19 >
1.645, we reject H0 and conclude that the process is out of control.
H0 :µ = 10
H1 :µ 6= 10.
We have z α2 = z0.01 = 2.33. Thus, the critical values are −2.33 and 2.33.
48.8 (d)
Section 49
49.1
x 1 2 3 4 5 6 7
1 1 1 1 1 1 1
p12 (x) 6 12 6 6 12 6 6
740 ANSWER KEY

 0, x<1
 1


 6 1≤x<2
1


 2≤x<3
 45


12 3≤x<4
F12 (x) = 7

 12 4≤x<5
2
5≤x<6



 3
5
6≤x<7


6


x ≥ 7.

1,
41 1331
49.2 The empirical mean is X = 12 and the empirical variance is 144 .
49.3 (a) We have


 0, x<1
 1


 6, 1≤x<2
4



 15 , 2≤x<3
22

45 , 3≤x<4

Ĥ(x) = 244

 315 , 4≤x<5
307
315 , 5≤x<6




929
6≤x<7

630 ,



1559
x ≥ 7.

630 ,
(b) We have


 1, x<1



 0.8465, 1 ≤ x<2



 0.7659, 2 ≤ x < 3
0.6133, 3 ≤ x < 4

Ŝ(x) =

 0.4609, 4 ≤ x < 5
 0.3773, 5 ≤ x < 6



0.2289, 6 ≤ x < 7




x ≥ 7.

0.0842,
49.4 
 1, x < 49
8
= 91 , 49 ≤ x < 50



 9
5
= 94 , 50 ≤ x < 60



 9
4
= 95 , 60 ≤ x < 75


Sn (x) = 9
3

 9 = 32 , 75 ≤ x < 80
2
= 97 , 80 ≤ x < 120



 9
1
= 98 , 120 ≤ x < 130


9


x ≥ 130.

0,
49.5 6
741
49.6 1.291
49.7 12
Section 50
50.1 (a) 47.50 (b) 3958.33
50.2 81
50.3 120
50.4 0.396
50.5 20,750
Section 51
51.1 (A) and (D) are false. (B) and (C) are true
51.2 Losses above a policy limit are right-censored and losses below a policy
deductible are left-truncated. The answer is (D)
51.3
742 ANSWER KEY
Life di xi ui
1 0 − 0.2
2 0 − 0.3
3 0 − 0.5
4 0 0.5 −
5 1 − 0.7
6 1.2 1.0 −
7 1.5 − 2.0
8 2 2.5 −
9 2.5 − 3.0
10 3.1 3.5 −
11 0 − 4.0
12 0 − 4.0
13 0 − 4.0
14 0 − 4.0
15 0 − 4.0
16 0 − 4.0
17 0 − 4.0
18 0 − 4.0
19 0.5 − 4.0
20 0.7 1.0 −
21 1.0 3.0 −
22 1.0 − 4.0
23 2.0 2.5 −
24 2.0 − 2.5
25 3.0 3.5 −
51.4
j yj sj rj
1 4 1 2+2−1=3
2 8 1 1+0−0=1
51.5
743
j yj sj rj
1 0.9 1 5+5−3=7
2 1.5 1 4+4−2=6
3 1.7 1 3+2−0=5
4 2.1 2 2+1−0=3
Section 52
52.1 0.52
52.2 0.583
52.3 0.067
52.4 0.7143
52.5 2
52.6 0.385
52.7 0.112
52.8 100
Section 53
53.1 We have

Nj E(Nj ) np(xj )
E[pn (xj )] = E = = = p(xj ).
n n n
This shows that the estimator is unbiased. Finding the variance of pn (xj )
we have
np(xj )[1 − p(xj )] p(xj )[1 − p(xj )]
Var[pn (xj )] = 2
= →0
n n
as n → ∞. This shows that the estimator is consistent
53.2 3.951 × 10−7

744 ANSWER KEY
53.3 We have
64
p386 (2) == 0.1658
386
d 386 (2)] = p386 (2)[1 − p386 (2)]
Var[p
n
64 322
= 386 386 = 3.58 × 10−4
386
53.4 The endpoints of the interval are:

r
0.1658(1 − 0.1658)
0.1658 ± 1.96 =⇒ (0.1287, 0.2029)
386
Section 54
54.1 Since X1 , X2 , · · · , Xn are independent so are X12 , X22 , · · · , Xn2 . We have
Var(X1 X2 · · · Xn ) =E[(X1 X2 · · · Xn )2 ] − E(X1 X2 · · · Xn )2

=E(X12 )E(X22 ) · · · E(Xn2 ) − E(X1 )2 E(X2 )2 · · · E(Xn )2
Yn Y n
2 2
= (µi + σi ) − µ2i
i=1 i=1
54.2 We have
(1 + a1 )(1 + a2 ) · · · (1 + an ) = 1 + a1 + a2 + · · · + an + prodcuts of ai s.
But the ai s are given small so that a product of ai s is even smaller. Ignoring
all the product terms, we obtain the desired result
54.3 0.0148
54.4 0.03086
54.5 10
Section 55
745
55.1 Ĥ(3) = 0.6875 and Var(

d Ĥ(3)) = 0.02376
55.2 The 95% linear confidence interval is

√ √
(0.6875 − 1.96 0.02376, 0.6875 + 1.96 0.02376) = (0.3854, 0.9896).
The 95% log-transformed confidence interval is

√ √
0.02376 0.02376
(0.6875e−1.96 0.6875 , 0.6875e1.96 0.6875 ) = (0.4431, 1.0669)
55.3 0.607
55.4 (0.189, 1.361)
55.5 0.779
55.6 (0.443, 1.067)
55.7 0.2341
Section 56
56.1 0.485
56.2 0.53125
56.3 0.026
56.4 0.3
56.5 1 ≤ x ≤ 2
56.6 0.3
Section 57
57.1 (a) 990 (b) 0.0080

746 ANSWER KEY
57.2 We have
j
X j−1
X
rj = dj − (xj + uj )
i=0 i=1
j−1
X j−2
X
=[ dj − (xj + uj )] + dj − (xj−1 + uj−1 )
i=0 i=1
=rj−1 + dj − (xj−1 + uj−1 )
57.3 0.75
57.4 0.6
Section 58
58.1 0.52490
58.2 0.52
58.3 369
58.4 26,400
58.5 13.75
58.6 107.8
58.7 20
58.8 384
58.9 −0.24
58.10 4.468
58.11 246.6
58.12 17.55
747
58.13 208.3
58.14 1.614
58.15 296.21
58.16 224
58.17 104.4
58.18 118.32
58.19 0.983
58.20 0.345
Section 59
x1 +x2 +···+xn
59.1 θ̃ = 2n
59.2 θ̂ = max{x1 , x2 , · · · , xn }
59.3 4.3275
59.4 3.97
59.5 θ̂ = min{x1 , x2 , · · · , xn }
59.6 0.2507
59.7 2
59.8 0.6798
59.9 1996.90
59.10 0.447
748 ANSWER KEY

pe−1 1−p −0.01 p −2 1−p −0.2
59.11 L(p) = 100 + 10,000 e 100 e + 10,000 e
59.12
n m
X 2m 6 X
`0 (α1 ) = n
α1 − ln xi + − ln yi
α1 (2 + α1 ) (2 + α1 )2
i=1 i=1
59.13 16.74
59.14 916.7
Section 60
60.1 73
60.2 233.333
60.3 703
60.4 3000
60.5 2.41877
1100
e− θ
60.6 L(θ) = θ3
60.7 471
60.8 f (50)f (15)f (60)f (500)[1 − F (100)][1 − F (500)]
60.9 3.089
60.10 3,325.67
60.11 0.09
60.12 1067
60.13 3/8
60.14 52.68
749
Section 61
61.1 `0 (α)2 = 16
α2
− 18.501
α + 5.3481
61.2 `00 (α) = − α42
16
61.3 I(α) = 1.732
= 5.346
61.4 0.1871
61.5 [0.8822, 2.5778]
1
61.6 n
61.7 0.447
3θ2
61.8 n
61.9 0.97
Section 62
62.1
n
X
`(α, θ) = [ln α + α ln θ − (α + 1) ln (xi + θ)]
i=1
62.2
∂2` n
=− 2
∂α2 α
n
∂2` n X 1
= −
∂θ∂α θ xi + θ
i=1
n
∂2` nα X (α + 1)
= − +
∂θ2 θ2 (xi + θ)2
i=1
62.3
− αn2 n nα
" #
θ − (α+1)θ
I(θ) = n
θ − nα
(α+1)θ − nα
θ2
+ nα(α+1)
(α+2)θ2
750 ANSWER KEY
62.4 " #
−1 1 − nα
θ2
+ nα(α+1)
(α+2)θ2
− nθ + (α+1)θ
nα
I(θ) = n nα n
det[I(θ)] − θ + (α+1)θ − α2
62.5
5.0391 −0.4115
I(α, θ) =
−0.4115 −0.0524
62.6

1 0.0524 −0.4115 −0.1209 0.9495
Var(α̂,
d θ) = =
(5.0391)(−0.0524) − 0.41152 −0.4115 −5.0391 0.9495 11.6274
62.7 0.0187
62.8 (a) We have
MY (etY ) = MY (etX1 etX2 ) = (1 − θt)−2 .
Thus, Y is a Gamma distribution with parameters α = 2 and θ.

(b) 0.15
(c) 0.079
62.9 (0.23, 0.69)
62.10 0.02345
62.11
2 −3
−3 5
Section 63
63.1 −1.00774
63.2 [0.70206, 4.20726]
63.3 [2.641591352, 8.358408648]

751
Section 64
64.1 This follows from

Z ∞
dθ
= ln θ|∞
0 =∞
0 θ
64.2 (a) The model distribution is
fX|Q (x|q) = (q)(q) = q 2 .
(b) The joint distribution is
q2 q4
fX,Q (x, q) = q 2 = .
0.039 0.039
(c) The marginal distribution in X is
Z 0.5 4
q
fX (x) = dq = 0.15862.
0.2 0.039
(d) The posterior distribution is
q4
πQ|X (q|x) =
0.006186
413 −4λ 12
64.3 πΛ|X (λ|x) = 12! e λ
7λ 13λ
λ10 (0.8e− 6 +0.6e− 12 )
64.4 πΛ|X (λ|x) = 0.395536(10!)
64.5 0.5572
64.6 0.721
64.7 0.64
64.8 1.90
x+c
64.9 2
27
64.10 16
752 ANSWER KEY
64.11 1.9899
64.12 0.622
64.13 0.148
Section 65
65.1 2
65.2 0.45
65.3 0.000398
65.4 450
65.5 1.319
65.6 0.8148
Section 66
66.1 The model distribution is

n
Y m
fX|Q (x|q) = q xi (1 − q)m−xi .
xi
i=1
n
Y m Γ(a + b) a−1
fX,Q (x, q) = q xi (1 − q)m−xi q (1 − q)b−1 .
xi Γ(a)Γ(b)
i=1
n
Γ(a + b) 1 a+Pni=1 xi −1
Z
Y m Pn
fX (x) = q (1 − q)b+nm− i=1 xi −1 dq
xi Γ(a)Γ(b) 0
i=1
n
Y m Γ(a + b)Γ(a + Pn xi )Γ(b + nm − Pn xi )

i=1 i=1
= .
xi Γ(a)Γ(b)Γ(a + b + nm)
i=1
Γ(a + b + nm) Pn Pn
πQ|X (q, x) = Pn Pn q a+ i=1 xi −1 (1−q)b+nm− i=1 xi −1 .
Γ(a + i=1 xi )Γ(b + nm − i=1 xi )
753
Pn
Hence,
Pn Q|X has a beta distribution with parameters a + i=1 xi , b + nm −
i=1 xi , and 1.

Pn 1
−λ
λn e i=1 xi
fX|Λ (x|λ) = Qn 2 .
i=1 xi

Pn
1 1
−λ +
λn+α−1 e θ i=1 xi
fX,Λ (x, λ) = .
θα Γ(α) ni=1 x2i
Q

1 Pn
∞ −λ + i=1 x1
λn+α−1 e
Z θ i Γ(n + α)
fX (x) = Qn dλ = Pn .
θα Γ(α) 2
θα ( 1θ 1 n+α
Γ(α) ni=1 x2i
Q
0 i=1 xi + i=1 xi )

1 Pn
−λ + i=1 x1
λn+α−1 e θ i
πΛ|X (θ, x) = −(n+α) .
1 Pn 1
θ + i=1 xi Γ(n + α)
Pn −1
1 1
Hence, Λ|X has a Gamma distribution with parameters α+n and θ + i=1 xi

Pn
1 i=1 xi
fX|Λ (x|λ) = n e− λ .
λ
θ
θ α e− λ
Pn
1 i=1 xi
fX,Λ (x, λ) = n e− λ .
λ λα+1 Γ(α)
Z ∞ α − 1 (Pni=1 xi +θ)
θ e λ θα Γ(α + n)
fX (x) = dλ = .
λα+n+1 Γ(α) (θ + ni=1 xi )α+n Γ(α)
P
0

1 Pn
(θ +
Pn α+n e− λ ( i=1 xi +θ)
i=1 xi )
πΛ|X (λ, x) = .
λα+n+1 Γ(α + n)
754 ANSWER KEY
Hence, Λ|X has an inverse Gamma distribution with parameters α + n and

P n
i=1 xi + θ

1
fX|Λ (x|λ) = .
λn
αθα
fX,Λ (x, λ) = , λ>M
λn+α+1
where M = max{x1 , x2 , · · · , xn , θ}. The marginal distribution is
Z ∞
αθα αθα
fX (x) = n+α+1
dλ = .
M λ (n + α)M n+α
(n + α)M n+α
πΛ|X (λ, x) = .
λn+α−1
Hence, Λ|X has a single-parameter Pareto distribution with α0 = n + α and
θ0 = M
Section 67
67.1 (a) Let nk be the number of policies un which frequency of exactly

k accidrnts occurred. The Poisson distribution parameter estimate by the
method of moments is
P6
knk 103
λ̂ = x = Pk=1
6 = = 1.2262.
k=0 nk
84
(b) The likelihood function is

6 −λ k nk
Y e λ
L(λ) = .
k!
k=0

6 6 6 6
e−λ λk
X X X X
`(λ) = nk ln = −λ nk + knk ln λ − nk ln (k!).
k!
k=0 k=0 k=1 k=0
755
The MLE of λ is found from

6 6 P6
d X 1X knk
`(λ) = − nk + knk = 0 =⇒ λ̂ = Pk=1
6 = 1.2262.
dλ λ k=0 nk
k=0 k=1
(c) We have
E(λ̂) =E(N ) = λ
Var(N ) λ
Var(λ̂) = = ,
n n
where n = 6k=0 nk . Thus, λ̂ is unbiased and consistent.

P
(d) The asymptotic variance is found as follows:
2 −λ N
∂ e λ
I(λ) =nI(N |λ) = −nE 2
ln
∂λ N!

N n
=nE =
λ2 λ
λ
Var(λ̂) = .
n
(e) The confidence interval is
p p
1.2262 − 1.96 1.2262/84, 1.2262 + 1.96 1.2262/84 = (0.9894, 1.463)
67.2 1.6438
e−λ λk
67.3 pk = k!
67.4 β̂ = 0.03 and r̂ = 41.6667
67.5
X 4 3
1.25 X
H(r̂) =100 ln 1 + − nk (r̂ + m)−1
r̂
k=1 m=0

1.25 70 35 15 5
=100 ln 1 + − + + +
r̂ r̂ r̂ + 1 r̂ + 2 r̂ + 3
756 ANSWER KEY
x
67.6 β̂ = r
67.7 q̂ = 0.1472 and m̂ = 18.27
67.8 0.06
Section 68
68.1 We have
n0 9048
p̂M
0 = = = 0.9048
n 10000
and P∞
k=0 knk nx
β̂ = −1= − 1 = 0.05147
n − n0 n − n0
68.2 We have
n0 9048
p̂M
0 = = = 0.9048
n 10000
and
n − n0
x(1 − e−λ ) = λ =⇒ 0.1001(1 − e−λ ) = 0.0952λ =⇒ λ̂ = 0.1012
n
68.3 We have
n0 10
p̂M
0 = = = 0.5
n 20
and
1 − p̂M
0
x= mq =⇒ 0.7 = 3q =⇒ q̂ = 0.2333
1 − p0
68.4 −22.5547
Section 69
69.1 −0.0714
69.2 For x < 0.3, we have F ∗ (x) > Fn (x) so that the fitted distribu-
tion is thicker on the left than the empirical distribution. For x > 0.85,
F ∗ (x) > Fn (x) which implies S ∗ (x) < Sn (x). That is, the fitted distribution
is thinner on the right than the empirical distribution. Also, note that near
the median x = 0.5, the slope is less than 1. That is, less probability on the
757
fitted than the empirical. Hence, the answer is (E)
69.3 Let’s choose x5 = 30. Than F (30) ≈ 0.6 but smaller than 0.6. If
X is uniform in [1, 100] then its cdf is Fu (x) = x−199 and Fu (30) ≈ 0.29
so that (C) is eliminated. If X is exponential with mean 10 then Fe =
1 − e−0.1x and Fe (30) ≈ 0.95. Thus, (D) is eliminated. If F (x) = x+1
x
then
30
F (30) 31 ≈ 0.97 so that (B) is eliminated. IF X is normal with mean 40
and standard deviation 40 then Fn (30) = Φ 30−40 40 = Φ(−0.25) = 0.40
so that (E) is eliminated. Note that with the function in (A) we have
F (30) = 1 − 30−0.25 ≈ 0.57. Hence, the answer is (A)
Section 70
70.1 0.2727
70.2 0.4025
70.3

Critical value 0.5456 0.6082 0.6619 0.729
Test Result Reject Reject Reject Fail to reject
70.4 0.1679
70.5 Fail to reject the null hypothesis
70.6

Critical value 0.0863 0.0962 0.1075 0.1152
Test Result Reject Reject Reject Fail to reject
Section 71
71.1 0.252
71.2
758 ANSWER KEY

Test Result Fail to reject Fail to reject Fail to reject
71.3 (A) Using sample data gives a better than expected fit and there-
fore a test statistic that favors the null hypothesis, thus increasing the Type
II error probability.
(B) The K-S test works only on individual data and so B is false.
(C) The Anderson-Darling test emphasizes the tails so that (C) is false.
Hence, the answer is (D)
71.4 (A), (B), and (C) are all correct. Thus, the answer is (D)
71.5 (A)
Section 72
72.1 9.151
72.2 (a) λ̂ = 1.6438

(b) The Chi-square statistic is
421.4809 36.7236 32.49 76.9129

χ2 = + + + = 7.558.
70.53 115.94 95.30 83.23
We have
10% 4.605 Reject
5% 5.991 Reject
2.5$ 7.378 Reject
1% 9.210 Do not reject
where the degrees of freedom is k − r − 1 = 4 − 1 − 1 = 2
72.3 The Chi-square statistic is
225 225 100 1 49 144

χ2 = + + + + + = 17.594.
250 350 240 110 40 10
We have
759

10% 9.236 Reject
5% 11.070 Reject
2.5$ 12.833 Reject
1% 15.086 Reject
The degrees of freedom is 6 − 1 = 5
72.4 χ2 = 6.659
72.5 A is false. Using sample data gives a better than expected fit and
therefore a test statistic that favors the null hypothesis, thus increasing the
Type II error probability. The K-S test works only on individual data and
so B is false. The A-D test emphasizes the tails, thus C is false. D is false
because the critical value depends on the degrees of freedom which in turn
depends on the number of cells, not the sample size. So the answer is (E)
72.6 (A)
Section 73
73.1 We have 0 degrees of freedom in the null hypothesis, since both param-
eters are specified, and 2 degrees of freedom in the alternative hypothesis,
since both parameters are freely chosen in maximizing L(α, θ). We thus have
2 degrees of freedom overall
73.2
α 5% 2.5% 1% 0.5%
cα 5.991 7.378 9.210 10.597
73.3
α 5% 2.5% 1% 0.5%
cα 5.991 7.378 9.210 10.597
Test Result Reject Reject Do not reject Do not reject
73.4
760 ANSWER KEY
α 10% 5% 2.5% 1%
cα 2.706 3.841 5.024 6.635
Test Result Reject Reject Do not reject Do not reject
73.5 7
Section 74
74.1 (i)
74.2 (A)
74.3 (I)
74.4 Generalized Pareto
Section 75
75.1 16,913
75.2 2,381
75.3 0.10
α+1

75.4 384.16 α
75.5 960
Section 76
76.1 0.47
76.2 (E)
76.3 0.3723
76.4 5,446,250
76.5 138
761
76.6 0.8
Section 77
77.1 For the risk parameter, we have

−λ
λα−1 e β
π(λ) = α .
β Γ(α)
e−λ λx
fX|Λ (x|λ) =
x!

(θ−µ) 2
1 − 2
π(θ) = √ e 2σ2 .
σ2 2π
(x−θ)2
1 − 2
fX|Θ (x|θ) = √ e 2σ1
σ1 2π

Γ(a + b) a−1
π(q) = q (1 − q)b−1 , 0 < q < 1.
Γ(a)Γ(b)

m
fX|Q (x|q) = q x (1 − q)m−x , x = 0, 1, · · · , m
x

θ
θα e− λ
π(λ) = .
λα+1 Γ(α)
1 −x
fX|Λ (x|λ) = e λ
λ
762 ANSWER KEY
αθα
π(λ) = , λ > θ.
λα+1
1
fX|Λ (x|λ) = , 0≤x≤λ
λ
Section 78
1
78.1 (a) fXY (x, y) = fX (x)fY |X (y|x) = x, 0 < x < 1, 1 − x < y < 1 (b)
1+ln 2
2
6x(2−x−y)
78.2 fX|Y (x|y) = 4−3y
− y1
78.3 e
78.4 (a) Observe that X only takes positive values, thus fX (x) = 0, x ≤ 0.
For 0 < x < 1 we have
Z ∞ Z ∞
α−1
fX (x) = fXY (x, y)dy = fXY (x, y)dy =
−∞ 1 α
For x ≥ 1 we have
∞ ∞
α−1
Z Z
fX (x) = fXY (x, y)dy = fXY (x, y)dy =
−∞ x αxα
(b) For 0 < x < 1 we have
fXY (x, y) α
fY |X (y|x) = = α+1 , y > 1.
fX (x) y
Hence, Z ∞ Z ∞
yα dy α
E(Y |X = x) = dy = α = .
1 y α+1 1 y α α−1
If x ≥ 1 then
fXY (x, y) αxα
fY |X (y|x) = = α+1 , y > x.
fX (x) y
763
Hence,
∞
αxα
Z
αx
E(Y |X = x) = α+1
y
dy =
x y α−1
3

78.5 E(X|Y = y) = 23 y and E(Y |X = x) = 23 1−y
1−y 2
78.6 The marginal density functions are

Z 1
21 2 21
fX (x) = x ydy = x2 (1 − x4 ), − 1 < x < 1
x2 4 8
Z √y
21 2 7 5
fY (y) = √ x ydx = y 2 , 0 < y < 1.
− y 4 2
Thus, a first way for finding E(Y ) is
Z 1 Z 1
7 5 7 7 7
E(Y ) = y y 2 dy = y y 2 dy = .
0 2 0 2 9
For the second way, we use the double expectation result
Z 1 Z 1
2 1 − x6 21 2

7
E(Y ) = E(E(Y |X)) = E(Y |X)fX (x)dx = 4
x (1−x6 ) =
−1 −1 3 1−x 8 9
1
78.7 12
78.8 Var(Y ) = 13
78.9 We have E(X) = λ = Var(X), E(Y |X = x) = αx, and Var(Y |X =

x) = β 2 x2 . Thus,
E(Y ) = E[E(Y |X)] = E(αX) = αE(X) = αλ
and
Var(Y ) =E[Var(Y |X)] + Var[E(Y |X)]

=E(β 2 X 2 ) + Var(αX)
=β 2 E(X 2 ) + α2 Var(X)
=β 2 (λ + λ2 ) + α2 λ
764 ANSWER KEY
Section 79
79.1 (a) The prior distribution is
π(G) =0.70
π(A) =0.20
π(B) = 0.10.
(b) The model distribution is
fX|Θ (x|G) =(0.25)(0.10) = 0.025

fX|Θ (x|A) =(0.40)(0.20) = 0.08
fX|Θ (x|B) =(0.30)(0.20) = 0.06
79.2 (a) The marginal probability is

X
fX (1, 2) = fX1 |Θ (1|θ)fX2 |Θ (2|θ)π(θ)
θ
=(0.25)(0.10)(0.7) + (0.4)(0.2)(0.2) + (0.3)(0.2)(0.10)
=0.0395.
(b) The joint distribution is

X
fX,X3 (1, 2, 0) = fX1 |Θ (1|θ)fX2 |Θ (2|θ)fX3 |Θ (0|θ)π(θ)
θ
=(0.25)(0.10)(0.65)(0.70) + (0.4)(0.2)(0.4)(0.2) + (0.3)(0.2)(0.5)(0.10)
=0.020775
X
θ
=(0.25)(0.10)(0.25)(0.70) + (0.4)(0.2)(0.4)(0.2) + (0.3)(0.2)(0.3)(0.10)
=0.012575
X
θ
=(0.25)(0.10)(0.10)(0.70) + (0.4)(0.2)(0.2)(0.2) + (0.3)(0.2)(0.2)(0.10)
=0.00615
765
79.3 (a) The predictive distribution is

0.020775
fX3 |X (0|1, 2) = = 0.5259
0.0395
0.012575
fX3 |X (1|1, 2) = = 0.3183
0.0395
0.00615
fX3 |X (2|1, 2) = = 0.1557.
0.0395
(b) The posterior probabilities are
f (1|G)f (2|G)π(G) (0.25)(0.10)(0.70)

π(G|1, 2) = = = 0.4430
f (1, 2) 0.0395
f (1|A)f (2|A)π(A) (0.40)(0.20)(0.20)
π(A|1, 2) = = = 0.4051
f (1, 2) 0.0395
f (1|B)f (2|B)π(B) (0.30)(0.20)(0.10)
π(B|1, 2) = = = 0.1519
f (1, 2) 0.0395
79.4 (a) The hypothetical means are
µ3 (G) =0(0.65) + 1(0.25) + 2(0.10) = 0.45

µ3 (A) =0(0.40) + 1(0.40) + 2(0.20) = 0.80
µ3 (B) =0(0.50) + 1(0.30) + 2(0.20) = 0.70.
(b) The pure premium is
µ3 = E(X3 ) = 0.45(0.70) + 0.80(0.20) + 0.70(0.10) = 0.545
79.5 (a) Without using the hypothetical means, we have
E(X3 |X] = 0(0.5259) + 1(0.3183) + 2(0.1557) = 0.6297.
(b) The Bayesian premium, using hypothetical means is
E(X3 |X] = (0.45)(0.4430) + (0.80)(0.4051) + (0.70)(0.1519) = 0.62976
79.6 0.158
766 ANSWER KEY
79.7 3.83
79.8 0.6794
79.9 7.202
79.10 10,322
79.11 0.278
Section 80
80.1 (a) The model distribution is
1 −x
f (x|λ) = e λ.
λ
15
225e− λ
π(λ) = , λ > 0.
λ3
(b) The joint density of x and λ is
(x+15)
225e− λ
f (x, λ) = f (x|λ)π(λ) = , λ > 0.
λ4
For x = 12, the joint density is
27
225e− λ
f (12, λ) = , λ > 0.
λ4
The marginal density of x is
(x+15)
∞
225e−
Z
λ
f (x) = dλ
0 λ4
767
and
27
∞
225e− λ
Z
f (12) = dλ
0 λ4
Z ∞ 3 − 27
225 27 e λ
= 3 Γ(3) dλ
27 λ4 Γ(3)
|0 {z }
1
450
= 3 .
(27 )
Similarly,
15
∞
1 − (12+x2 ) 225e− λ
Z
f (12, x2 ) = e λ dλ
0 λ2 λ3
Z ∞
1 − (27+x2 )
=225 e λ dλ
0 λ5
Z ∞ (27+x2 )
225 (27 + x2 )4 e− λ
= Γ(4) dλ
(27 + x2 )4 0 λ5 Γ(4)
| {z }
1
1350
= .
(27 + x2 )4
The predictive distribution is
1350
(27+x2 )4 3(273 )
f (x2 |12) = 450 =
(273 )
(27 + x2 )4
which is a type 2 Pareto distribution with parameters α = 3 and θ = 27.

(c) The posterior distribution of Λ is
27
273 e− λ
π(λ|12) = .
2 λ4
(d) E(X2 |12) = 13.5
80.2 3.25
80.3 The posterior distribution of P is

10!
5
2 4!6! p (1 − p)6 Γ(13) 5
π(p|4) = Γ(10)Γ(7)
= p (1 − p)6
2 Γ(4)Γ(13) Γ(6)Γ(7)
768 ANSWER KEY
which is a beta distribution with a = 6, b = 7, and θ = 1
1
80.4 (i) Letting q = 1+β we can write
r k
r(r + 1) · · · (r + k − 1) 1 β
pk =
k! 1+β 1+β
r k
(r − 1)!r(r + 1) · · · (r + k − 1) 1 1
= 1−
k!(r − 1)! 1+β 1+β
Γ(r + k)
= q r (1 − q)k .
Γ(r)Γ(k + 1)
(ii) The model distribution is
Γ(r + x)
f (x|q) = q r (1 − q)x .
Γ(r)Γ(x + 1)
Γ(a + b) a−1
π(q) = q (1 − q)b−1 .
Γ(a)Γ(b)
The joint distribution of X and Q is
Γ(r + x) Γ(a + b) a−1
f (x, q) =f (x|q)π(q) = q r (1 − q)x q (1 − q)b−1
Γ(r)Γ(x + 1) Γ(a)Γ(b)
Γ(r + x) Γ(a + b) a+r−1
= q (1 − q)b+k−1 .
Γ(r)Γ(x + 1) Γ(a)Γ(b)
Γ(r + x) Γ(a + b) 1 a+r−1
Z
f (x) = q (1 − q)b+k−1 dq
Γ(r)Γ(x + 1) Γ(a)Γ(b) 0
Γ(r + x) Γ(a + b) Γ(a + r)Γ(b + k) 1 Γ(a + b + k + r) a+r−1
Z
= q (1 − q)b+k−1 dq
Γ(r)Γ(x + 1) Γ(a)Γ(b) Γ(a + b + k + r) 0 Γ(a + r)Γ(b + k)
| {z }
1
Γ(r + x) Γ(a + b) Γ(a + r)Γ(b + k)
= .
Γ(r)Γ(x + 1) Γ(a)Γ(b) Γ(a + b + k + r)
Γ(r+x) Γ(a+b) a+r−1
Γ(r)Γ(x+1) Γ(a)Γ(b) q (1 − q)b+k−1
π(q|x) = Γ(r+x) Γ(a+b) Γ(a+r)Γ(b+k)
Γ(r)Γ(x+1) Γ(a)Γ(b) Γ(a+b+k+r)
Γ(a + b + k + r) a+r−1
= q (1 − q)b+k−1
Γ(a + r)Γ(b + k)
769
which is a beta distribution with a0 = a + r and b0 = b + k
80.6 15
Section 81
81.1 The unbiasedness equation yields

n
X
α̂0 + µ α̂j = µ
j=1
which implies
n
X α̂0
α̂i = 1 −
2
j=1
81.2 For i = 1, 2, · · · , n, we have

n
X
α̂j ρσ 2 + σ 2 α̂i = ρσ 2
j=1
j 6= i
or equivalently
n
X
α̂j + α̂i (1 − ρ) = ρ
j=1
81.3 Problem 81.2 followed by Problem 81.1, we find
ρ(1 − nj=1 α̂j )

P
ρα̂0
α̂i = =
1−ρ µ(1 − ρ)
81.4 From Problem 81.3, we find

n
X nρα̂0
α̂j = .
µ(1 − ρ)
j=1
This combined with Problem 81.1 yield the equation
α̂0 nρα̂0
1− = .
µ µ(1 − ρ)
770 ANSWER KEY
Solving this equation, we find
(1 − ρ)µ
α̂0 = .
1 − ρ + nρ
Plugging this into Problem 81.3, we find
ρ
α̂i =
1 − ρ + nρ
81.5 The credibility premium is

n n
X (1 − ρ)µ X ρXj
α̂0 + α̂j Xj = +
1 − ρ + nρ 1 − ρ + nρ
j=1 j=1
=(1 − Z)µ + ZX
where
nρ
Z=
1 − ρ + nρ
Section 82
82.1 10,622
82.2 0.85651
82.3 0.22
82.4 1063.47
82.5 3
82.6 14
θ−θ2
82.7 1.5−θ2
82.8 8.33
1
82.9 9
82.10 3.27
771
Section 83
nβ 1
83.1 nβ+1 X + nβ+1 (αβ)
83.2 0.9375
83.3 0.905
83.4 1
83.5 0.93
83.6 8.69
83.7 0.428
83.8 0.8
Section 84
84.1 (a) We have
2 2
mi Xi + mj Xj mi mj
Var Θ = Var(Xi |Θ) + Var(Xj |Θ)
mi + mj mi + mj mi + mj
2 2
mi v(θ) mj v(θ)
= w(θ) + + w(θ) +
mi + mj mi mi + mj mj
2
mi + mj2
v(θ)
= w(θ) + .
(mi + mj )2 mi + mj
772 ANSWER KEY
(b) We have
E(Xi ) =E[E(Xi |Θ)] = E[µ(Θ)] = µ
=E[E(Xi Xj |Θ)] − E[µ(Θ)]2
=E[µ2 (Θ)] − E[µ(Θ)]2
=Var[µ(Θ)] = a

v(Θ)
=E w(θ) + + Var[µ(Θ)]
mi
v
=w + +a
mi
84.2 The unbiasedness equation is

n n
X X α̂0
E(Xn+1 ) = µ = α̂0 + α̂i µ =⇒ α̂i = 1 − .
µ
i=1 i=1
For i = 1, 2, · · · , n, (81.4) becomes

n
X v
a= α̂j a + α̂i a+ +w
mi
j=1
j 6= i
n
X v
= α̂j a + α̂i w +
mi
j=1

α̂0 v
=a 1 − + α̂i w + .
µ mi
Solving this equation for α̂i , we find
aα̂0 /µ
α̂i = .
w + v/mi
Summing both sides from 1 to n, we find
n
aα̂0 X mj α̂0
=1− .
µ v + wmj µ
j=1
773
Solving this equation, we find

1 µ
α̂0 = n = α̂0 =
X mj 1 1 + am∗
(a/µ) +
v + wmj µ
j=1
84.3 We have
n n
X amj
X µ 1
α̂0 + α̂j Xj = ∗
+ Xj
1 + am v + wmj 1 + am∗
j=1 j=1
n
µ a X mj
= ∗
+ ∗
Xj
1 + am 1 + am v + wmj
j=1
=ZX + (1 − Z)µ.
84.4 2.4
84.5 12
84.6 11.13
84.7 257.11
4
84.8 3
84.9 (A) is false. This is true for Bhlmann. The Bhlmann-Straub model
allows the variation in size and exposure.
(B) is false. The model is valid for any type of distributions.
(C) is false. There is no cap on the number of exposure.
Thus, the answer to the problem is (E)
n
84.10 n+ w
a
Section 85
85.1 Problem 66.3 shows that the posterior distribution

P is an inverse Gamma
distribution with parameters α0 = α + n and θ0 = ni=1 xi + θ. The hypo-
thetical mean is
µ(λ) = E(Xi |Λ) = λ
774 ANSWER KEY
1 Pn
Z ∞
(θ +
Pn α+n e− λ ( i=1 xi +θ)
i=1 xi )
E(Xn+1 |X) = λ dλ
0 λα+n+1 Γ(α + n)
1 Pn
∞
(θ + ni=1 xi )α+n−1 e− λ ( i=1 xi +θ)
Z P
(nX + θ)
= dλ
α+n−1 0 λα+n−1+1 Γ(α + n − 1)
(nX + θ)
= .
α+n−1

µ(λ) =E(Xi |Λ) = λ

θ
µ =E(Λ) =
α−1
θ2
v(λ) =Var(Xi |Λ) = λ2 v = E(Λ2 ) =
(α − 1)(α − 2)
θ2
a =Var(Λ) =
(α − 1)2 (α − 2)
v
k= =α−1
a
n n
Z= =
n+k n+α−1
Pc =ZX + (1 − Z)µ
n α−1 θ
= X+
n+α−1 n+α−1α−1
=(nX + θ)n + α − 1.
Thus, the Bühlmann credibility premium equals the Bayesian premium
85.2 Problem 66.1 shows that the posterior distribution hasPa beta dis-
tribution with parameters a = a + i=1 xi , b0 = b + nm − ni=1 xi . The
0
Pn
hypothetical mean is
µ(q) = E(Xi |Q) = mq

775

Z ∞
Γ(a + b + nm) Pn Pn
E(Xn+1 |X) = mq Pn Pn q a+ i=1 xi −1 (1 − q)b+nm− i=1 xi −1 dq
0 Γ(a + i=1 xi )Γ(b + nm − i=1 xi )
m(a + nx) ∞
Z
Γ(a + b + nm + 1) Pn Pn
= Pn Pn q a+ i=1 xi +1−1 (1 − q)b+nm− i=1 xi −1
a + b + mn 0 Γ(a + i=1 xi + 1)Γ(b + nm − i=1 xi )
m(a + nx)
= .
a + b + mn

µ(q) =E(Xi |Q) = mq

ma
µ =E(mQ) = mE(Q) =
a+b
v(q) =Var(Xi |Q) = mq(1 − q)
mab
v =E[mQ(1 − Q)] =
(a + b)(a + b + 1)
m2 ab
a =Var[mQ] = m2 Var(Q) =
(a + b)2 (a + b + 1)
v a+b
k= =
a m
n nm
Z= =
n+k nm + a + b
Pc =ZX + (1 − Z)µ
m(a + nx)
= .
a + b + nm
85.3 By Example
P 66.3 the posterior distribution has normal distribution
xi −1 −1
with mean σ2
+ aµ2 σn2 + a12 and variance σn2 + a12 . The hypo-
thetical mean is
µ(λ) = E(Xi |Λ) = λ

P −1
xi µ n 1
E(Xn+1 |X) = E[Λ|X] = + 2 + 2 .
σ2 a σ 2 a
776 ANSWER KEY

µ(λ) =λ
µ =E(Λ)
v(λ) =Var(Xi |Λ) = σ 2
v =E(σ 2 ) = σ 2
Var(Λ) =a2
v sigma2
k= =
a a2
n na2
Z= =
n+k na2 + σ 2
Pc =ZX + (1 − Z)µ
na2 x µσ 2
= 2 2
+ 2
na + σ na + σ 2
1 −1
P
xi µ n
= + 2 + .
σ2 a σ 2 a2
85.4 0.0182
Section 86
86.1 0.818
86.2 687.375
86.3 0.78
86.4 0.8718
Section 87
87.1 0.3682
87.2 0.4987
777
87.3 0.852
87.4 1.351
87.5 98.26
87.6 0.323
87.7 7.56
Section 88
88.1 a = µ − v − µ2
88.2 0.221
88.3 0.3928
88.4 0.6333
88.5 0.5747
88.6 0.2659
88.7 0.023209
Section 89
89.1 1000
89.2 1
89.3 (D)
89.4 2212.76
89.5 3477.81
89.6 7
778 ANSWER KEY
Section 90
90.1 522.13
90.2 228,503
90.3 224.44
90.4 88.75
90.5 41.897
90.6 35.7
90.7 630.79
Section 91
91.1 We have the following sequence of calculation

k =[pn] + 1 = [3] + 1 = 4
VaR
d p (X) =123
n
1 X
TVaR
\ p (X) = yi
n−k+1
i=k
1
= (123 + 150 + 153 + 189 + 190 + 195 + 200) = 171.43
10 − 4 + 1
n
1 X
s2p = \ p (X))2
(yi − TVaR
n−k
i=k
1
= [(123 − 171.43)2 + (150 − 171.43)2 + (153 − 171.43)2 + (189 − 171.43)2 +
10 − 4
s2 + p[TVaR
\ p (X) − VaRd p (X)]2
Var( \ p (X)) = p
d TVaR
n−k+1
861.61 + 0.3(123 − 171.43)2
= = 223.52
10 − 4 + 1
Section 92
779
92.1 1
92.2 (A)
44
92.3 9
92.4 0.0131
92.5 21θ4
780 ANSWER KEY
Exam C Tables
781
782 EXAM C TABLES
783
784 EXAM C TABLES
Appendix A
An Inventory of Continuous
Distributions
A.1 Introduction
The incomplete gamma function is given by
Z x
1
Γ(α; x) = tα−1 e−t dt, α > 0, x > 0
Γ(α) 0
Z ∞
with Γ(α) = tα−1 e−t dt, α > 0.
0
Also, define Z ∞
G(α; x) = tα−1 e−t dt, x > 0.
x
At times we will need this integral for nonpositive values of α. Integration by parts produces the relationship
xα e−x 1
G(α; x) = − + G(α + 1; x)
α α
This can be repeated until the first argument of G is α + k, a positive number. Then it can be evaluated
from
G(α + k; x) = Γ(α + k)[1 − Γ(α + k; x)].
The incomplete beta function is given by
Z
Γ(a + b) x a−1
β(a, b; x) = t (1 − t)b−1 dt, a > 0, b > 0, 0 < x < 1.
Γ(a)Γ(b) 0
785
APPENDIX A. AN INVENTORY OF CONTINUOUS DISTRIBUTIONS 2
A.2 Transformed beta family

A.2.2 Three-parameter distributions
A.2.2.1 Generalized Pareto (beta of the second kind)–α, θ, τ
Γ(α + τ ) θα xτ −1 x
f (x) = F (x) = β(τ , α; u), u=
Γ(α)Γ(τ ) (x + θ)α+τ x+θ
θk Γ(τ + k)Γ(α − k)
E[X k ] = , −τ < k < α
Γ(α)Γ(τ )
θk τ (τ + 1) · · · (τ + k − 1)
E[X k ] = , if k is an integer
(α − 1) · · · (α − k)
θk Γ(τ + k)Γ(α − k)
E[(X ∧ x)k ] = β(τ + k, α − k; u) + xk [1 − F (x)], k > −τ
Γ(α)Γ(τ )
τ −1
mode = θ , τ > 1, else 0
α+1
A.2.2.2 Burr (Burr Type XII, Singh-Maddala)–α, θ, γ
αγ(x/θ)γ 1
f (x) = F (x) = 1 − uα , u=
x[1 + (x/θ)γ ]α+1 1 + (x/θ)γ
θk Γ(1 + k/γ)Γ(α − k/γ)
E[X k ] = , −γ < k < αγ
Γ(α)
VaRp (X) = θ[(1 − p)−1/α − 1]1/γ
θk Γ(1 + k/γ)Γ(α − k/γ)
E[(X ∧ x)k ] = β(1 + k/γ, α − k/γ; 1 − u) + xk uα , k > −γ
Γ(α)
µ ¶1/γ
γ−1
mode = θ , γ > 1, else 0
αγ + 1
A.2.2.3 Inverse Burr (Dagum)–τ , θ , γ
τ γ(x/θ)γτ (x/θ)γ
f (x) = F (x) = uτ , u=
x[1 + (x/θ)γ ]τ +1 1 + (x/θ)γ
θk Γ(τ + k/γ)Γ(1 − k/γ)
E[X k ] = , −τ γ < k < γ
Γ(τ )
VaRp (X) = θ(p−1/τ − 1)−1/γ
θk Γ(τ + k/γ)Γ(1 − k/γ)
E[(X ∧ x)k ] = β(τ + k/γ, 1 − k/γ; u) + xk [1 − uτ ], k > −τ γ
Γ(τ )
µ ¶1/γ
τγ − 1
mode = θ , τ γ > 1, else 0
γ+1
786 EXAM C TABLES
A.2.3 Two-parameter distributions

A.2.3.1 Pareto (Pareto Type II, Lomax)–α, θ
µ ¶α
αθα θ
f (x) = F (x) = 1 −
(x + θ)α+1 x+θ
θk Γ(k + 1)Γ(α − k)
E[X k ] = , −1 < k < α
Γ(α)
θk k!
(α − 1) · · · (α − k)
VaRp (X) = θ[(1 − p)−1/α − 1]
θ(1 − p)−1/α
TVaRp (X) = VaRp (X) + , α>1
α−1
" µ ¶α−1 #
θ θ
E[X ∧ x] = 1− , α=
6 1
α−1 x+θ
µ ¶
θ
E[X ∧ x] = −θ ln , α=1
x+θ
µ ¶α
k θk Γ(k + 1)Γ(α − k) θ
E[(X ∧ x) ] = β[k + 1, α − k; x/(x + θ)] + xk , all k
Γ(α) x+θ
mode = 0
A.2.3.2 Inverse Pareto–τ , θ

µ ¶τ
τ θxτ −1 x
f (x) = F (x) =
(x + θ)τ +1 x+θ
θk Γ(τ + k)Γ(1 − k)
E[X k ] = , −τ < k < 1
Γ(τ )
θk (−k)!
E[X k ] = , if k is a negative integer
(τ − 1) · · · (τ + k)
VaRp (X) = θ[p−1/τ − 1]−1
Z x/(x+θ) ∙ µ ¶τ ¸
k k τ +k−1 −k k x
E[(X ∧ x) ] = θ τ y (1 − y) dy + x 1 − , k > −τ
0 x+θ
τ −1
2
A.2.3.3 Loglogistic (Fisk)–γ, θ
γ(x/θ)γ (x/θ)γ
f (x) = F (x) = u, u=
x[1 + (x/θ)γ ]2 1 + (x/θ)γ
E[X k ] = θk Γ(1 + k/γ)Γ(1 − k/γ), −γ < k < γ
VaRp (X) = θ(p−1 − 1)−1/γ
E[(X ∧ x)k ] = θk Γ(1 + k/γ)Γ(1 − k/γ)β(1 + k/γ, 1 − k/γ; u) + xk (1 − u), k > −γ
µ ¶1/γ
γ−1
mode = θ , γ > 1, else 0
γ+1
787
A.2.3.4 Paralogistic–α, θ
This is a Burr distribution with γ = α.
α2 (x/θ)α 1
f (x) = F (x) = 1 − uα , u=
x[1 + (x/θ)α ]α+1 1 + (x/θ)α
θk Γ(1 + k/α)Γ(α − k/α)
E[X k ] = , −α < k < α2
Γ(α)
VaRp (X) = θ[(1 − p)−1/α − 1]1/α
θk Γ(1 + k/α)Γ(α − k/α)
E[(X ∧ x)k ] = β(1 + k/α, α − k/α; 1 − u) + xk uα , k > −α
Γ(α)
µ ¶1/α
α−1
mode = θ , α > 1, else 0
α2 + 1
A.2.3.5 Inverse paralogistic–τ , θ

This is an inverse Burr distribution with γ = τ .
2
τ 2 (x/θ)τ (x/θ)τ
f (x) = F (x) = uτ , u=
x[1 + (x/θ)τ ]τ +1 1 + (x/θ)τ
θk Γ(τ + k/τ )Γ(1 − k/τ )
E[X k ] = , −τ 2 < k < τ
Γ(τ )
VaRp (X) = θ(p−1/τ − 1)−1/τ
θk Γ(τ + k/τ )Γ(1 − k/τ )
E[(X ∧ x)k ] = β(τ + k/τ , 1 − k/τ ; u) + xk [1 − uτ ], k > −τ 2
Γ(τ )
mode = θ (τ − 1)1/τ , τ > 1, else 0
A.3 Transformed gamma family

A.3.2 Two-parameter distributions
A.3.2.1 Gamma–α, θ
(x/θ)α e−x/θ
f (x) = F (x) = Γ(α; x/θ)
xΓ(α)
θk Γ(α + k)
M (t) = (1 − θt)−α , t < 1/θ E[X k ] = , k > −α
Γ(α)
E[X k ] = θk (α + k − 1) · · · α, if k is an integer
θk Γ(α + k)
E[(X ∧ x)k ] = Γ(α + k; x/θ) + xk [1 − Γ(α; x/θ)], k > −α
Γ(α)
= α(α + 1) · · · (α + k − 1)θk Γ(α + k; x/θ) + xk [1 − Γ(α; x/θ)], k an integer
mode = θ(α − 1), α > 1, else 0
788 EXAM C TABLES
A.3.2.2 Inverse gamma (Vinci)–α, θ
(θ/x)α e−θ/x
f (x) = F (x) = 1 − Γ(α; θ/x)
xΓ(α)
θk Γ(α − k) θk
E[X k ] = , k<α E[X k ] = , if k is an integer
Γ(α) (α − 1) · · · (α − k)
θk Γ(α − k)
E[(X ∧ x)k ] = [1 − Γ(α − k; θ/x)] + xk Γ(α; θ/x)
Γ(α)
θk Γ(α − k)
= G(α − k; θ/x) + xk Γ(α; θ/x), all k
Γ(α)
mode = θ/(α + 1)
A.3.2.3 Weibull–θ, τ
τ
τ (x/θ)τ e−(x/θ) τ
f (x) = F (x) = 1 − e−(x/θ)
x
E[X k ] = θk Γ(1 + k/τ ), k > −τ
VaRp (X) = θ[− ln(1 − p)]1/τ
τ
E[(X ∧ x)k ] = θk Γ(1 + k/τ )Γ[1 + k/τ ; (x/θ)τ ] + xk e−(x/θ) , k > −τ
µ ¶1/τ
τ −1
τ
A.3.2.4 Inverse Weibull (log Gompertz)–θ, τ
τ
τ (θ/x)τ e−(θ/x) τ
f (x) = F (x) = e−(θ/x)
x
E[X k ] = θk Γ(1 − k/τ ), k < τ
VaRp (X) = θ(− ln p)−1/τ
h τ
i
E[(X ∧ x)k ] = θk Γ(1 − k/τ ){1 − Γ[1 − k/τ ; (θ/x)τ ]} + xk 1 − e−(θ/x) , all k
h τ
i
= θk Γ(1 − k/τ )G[1 − k/τ ; (θ/x)τ ] + xk 1 − e−(θ/x)
µ ¶1/τ
τ
mode = θ
τ +1
789
A.3.3 One-parameter distributions

A.3.3.1 Exponential–θ
e−x/θ
f (x) = F (x) = 1 − e−x/θ
θ
M (t) = (1 − θt)−1 E[X k ] = θk Γ(k + 1), k > −1
k
E[X k ] = θ k!, if k is an integer
VaRp (X) = −θ ln(1 − p)
TVaRp (X) = −θ ln(1 − p) + θ
E[X ∧ x] = θ(1 − e−x/θ )
E[(X ∧ x)k ] = θk Γ(k + 1)Γ(k + 1; x/θ) + xk e−x/θ , k > −1
= θk k!Γ(k + 1; x/θ) + xk e−x/θ , k an integer
mode = 0
A.3.3.2 Inverse exponential–θ
θe−θ/x
f (x) = F (x) = e−θ/x
x2
k k
E[X ] = θ Γ(1 − k), k<1
VaRp (X) = θ(− ln p)−1
E[(X ∧ x)k ] = θk G(1 − k; θ/x) + xk (1 − e−θ/x ), all k
mode = θ/2
A.5 Other distributions

A.5.1.1 Lognormal–μ,σ (μ can be negative)
1 ln x − μ
f (x) = √ exp(−z 2 /2) = φ(z)/(σx), z = F (x) = Φ(z)
xσ 2π σ
k 2 2
E[X ] = exp(kμ + k σ /2)
µ ¶
k 2 2 ln x − μ − kσ 2
E[(X ∧ x) ] = exp(kμ + k σ /2)Φ + xk [1 − F (x)]
σ
mode = exp(μ − σ2 )
790 EXAM C TABLES
A.5.1.2 Inverse Gaussian–μ, θ
µ ¶1/2 µ ¶
θ θz 2 x−μ
f (x) = exp − , z=
2πx3 2x μ
" µ ¶ # µ ¶ " µ ¶1/2 #
1/2
θ 2θ θ x+μ
F (x) = Φ z + exp Φ −y , y=
x μ x μ
" Ã r !#
θ 2tμ2 θ
M (t) = exp 1− 1− , t < 2, E[X] = μ, Var[X] = μ3 /θ
μ θ 2μ
" µ ¶ # µ ¶ " µ ¶1/2 #
1/2
θ 2θ θ
E[X ∧ x] = x − μzΦ z − μy exp Φ −y
x μ x
A.5.1.3 log-t–r, μ, σ (μ can be negative)

Let Y have a t distribution with r degrees of freedom. Then X = exp(σY + μ) has the log-t distribution.
Positive moments do not exist for this distribution. Just as the t distribution has a heavier tail than the
normal distribution, this distribution has a heavier tail than the lognormal distribution.
µ ¶
r+1
Γ
2
f (x) = " ,
³r´ µ ¶2 #(r+1)/2
√ 1 ln x − μ
xσ πrΓ 1+
2 r σ
µ ¶
ln x − μ
F (x) = Fr with Fr (t) the cdf of a t distribution with r d.f.,
σ
⎧ ⎡ ⎤
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1 ⎢ ⎢r 1 r ⎥
⎥
⎪
⎪ β ⎢ , ; µ ¶2 ⎥, 0 < x ≤ eμ ,
⎪
⎪ 2 ⎣ 2 2 ln x − μ ⎦
⎪
⎨ r+
⎡ σ ⎤
F (x) =
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1 ⎢ ⎢r 1 r ⎥
⎥
⎪
⎪ 1 − β ⎢ , ; µ ¶ 2⎥ , x ≥ eμ .
⎪
⎪ 2 ⎣ 2 2 ln x − μ ⎦
⎪
⎩ r+
σ
A.5.1.4 Single-parameter Pareto–α, θ
αθα
f (x) = , x>θ F (x) = 1 − (θ/x)α , x>θ
xα+1
αθ(1 − p)−1/α
VaRp (X) = θ(1 − p)−1/α TVaRp (X) = , α>1
α−1
αθk αθk kθα
E[X k ] = , k<α E[(X ∧ x)k ] = − , x≥θ
α−k α − k (α − k)xα−k
mode = θ
Note: Although there appears to be two parameters, only α is a true parameter. The value of θ must be
set in advance.
791
A.6 Distributions with finite support

For these two distributions, the scale parameter θ is assumed known.
A.6.1.1 Generalized beta–a, b, θ, τ
Γ(a + b) a τ
f (x) = u (1 − u)b−1 , 0 < x < θ, u = (x/θ)τ
Γ(a)Γ(b) x
F (x) = β(a, b; u)
θk Γ(a + b)Γ(a + k/τ )
E[X k ] = , k > −aτ
Γ(a)Γ(a + b + k/τ )
θk Γ(a + b)Γ(a + k/τ )
E[(X ∧ x)k ] = β(a + k/τ , b; u) + xk [1 − β(a, b; u)]
Γ(a)Γ(a + b + k/τ )
A.6.1.2 beta–a, b, θ
Γ(a + b) a 1
f (x) = u (1 − u)b−1 , 0 < x < θ, u = x/θ
Γ(a)Γ(b) x
F (x) = β(a, b; u)
θk Γ(a + b)Γ(a + k)
E[X k ] = , k > −a
Γ(a)Γ(a + b + k)
θk a(a + 1) · · · (a + k − 1)
(a + b)(a + b + 1) · · · (a + b + k − 1)
θk a(a + 1) · · · (a + k − 1)
E[(X ∧ x)k ] = β(a + k, b; u)
(a + b)(a + b + 1) · · · (a + b + k − 1)
+xk [1 − β(a, b; u)]
792 EXAM C TABLES
Appendix B
An Inventory of Discrete
Distributions
B.1 Introduction
The 16 models fall into three classes. The divisions are based on the algorithm by which the probabilities are
computed. For some of the more familiar distributions these formulas will look diﬀerent from the ones you
may have learned, but they produce the same probabilities. After each name, the parameters are given. All
parameters are positive unless otherwise indicated. In all cases, pk is the probability of observing k losses.
For finding moments, the most convenient form is to give the factorial moments. The jth factorial
moment is μ(j) = E[N (N − 1) · · · (N − j + 1)]. We have E[N ] = μ(1) and Var(N ) = μ(2) + μ(1) − μ2(1) .
The estimators which are presented are not intended to be useful estimators but rather for providing
starting values for maximizing the likelihood (or other) function. For determining starting values, the
following quantities are used [where nk is the observed frequency at k (if, for the last entry, nk represents
the number of observations at k or more, assume it was at exactly k) and n is the sample size]:
∞ ∞
1X 1X 2
μ̂ = knk , σ̂ 2 = k nk − μ̂2 .
n n
k=1 k=1
When the method of moments is used to determine the starting value, a circumflex (e.g., λ̂) is used. For
any other method, a tilde (e.g., λ̃) is used. When the starting value formulas do not provide admissible
parameter values, a truly crude guess is to set the product of all λ and β parameters equal to the sample
mean and set all other parameters equal to 1. If there are two λ and/or β parameters, an easy choice is to
set each to the square root of the sample mean.
The last item presented is the probability generating function,
P (z) = E[z N ].
B.2 The (a, b, 0) class

B.2.1.1 Poisson–λ
e−λ λk
p0 = e−λ , a = 0, b=λ pk =
k!
E[N ] = λ, Var[N ] = λ P (z) = eλ(z−1)
9
793
APPENDIX B. AN INVENTORY OF DISCRETE DISTRIBUTIONS 10
B.2.1.2 Geometric–β
1 β βk
p0 = , a= , b=0 pk =
1+β 1+β (1 + β)k+1
E[N ] = β, Var[N ] = β(1 + β) P (z) = [1 − β(z − 1)]−1 .
This is a special case of the negative binomial with r = 1.
B.2.1.3 Binomial–q, m, (0 < q < 1, m an integer)
q (m + 1)q
p0= (1 − q)m , a = − , b=
1−q 1−q
µ ¶
m k
pk = q (1 − q)m−k , k = 0, 1, . . . , m
k
E[N ] = mq, Var[N ] = mq(1 − q) P (z) = [1 + q(z − 1)]m .
B.2.1.4 Negative binomial–β, r
β (r − 1)β
p0 = (1 + β)−r , a= , b=
1+β 1+β
r(r + 1) · · · (r + k − 1)β k
pk =
k!(1 + β)r+k
E[N ] = rβ, Var[N ] = rβ(1 + β) P (z) = [1 − β(z − 1)]−r .
B.3 The (a, b, 1) class

To distinguish this class from the (a, b, 0) class, the probabilities are denoted Pr(N = k) = pM k or Pr(N =
k) = pTk depending on which subclass is being represented. For this class, pM 0 is arbitrary (that is, it is a
parameter) and then pM T
1 or p1 is a specified function of the parameters a and b. Subsequent probabilities are
obtained recursively as in the (a, b, 0) class: pM M
k = (a + b/k)pk−1 , k = 2, 3, . . ., with the same recursion for pk
T
There are two sub-classes of this class. When discussing their members, we often refer to the “corresponding”
member of the (a, b, 0) class. This refers to the member of that class with the same values for a and b. The
notation pk will continue to be used for probabilities for the corresponding (a, b, 0) distribution.
B.3.1 The zero-truncated subclass

The members of this class have pT0 = 0 and therefore it need not be estimated. These distributions should
only be used when a value of zero is impossible. The first factorial moment is μ(1) = (a + b)/[(1 − a)(1 − p0 )],
where p0 is the value for the corresponding member of the (a, b, 0) class. For the logarithmic distribution
(which has no corresponding member), μ(1) = β/ ln(1+β). Higher factorial moments are obtained recursively
with the same formula as with the (a, b, 0) class. The variance is (a+b)[1−(a+b+1)p0 ]/[(1−a)(1−p0 )]2 .For
those members of the subclass which have corresponding (a, b, 0) distributions, pTk = pk /(1 − p0 ).
794 EXAM C TABLES
B.3.1.1 Zero-truncated Poisson–λ
λ
pT1 = , a = 0, b = λ,
eλ −1
λk
pTk = ,
k!(eλ − 1)
E[N ] = λ/(1 − e−λ ), Var[N ] = λ[1 − (λ + 1)e−λ ]/(1 − e−λ )2 ,
λ̃ = ln(nμ̂/n1 ),
eλz − 1
P (z) = .
eλ − 1
B.3.1.2 Zero-truncated geometric–β
1 β
pT1 = , a= , b = 0,
1+β 1+β
β k−1
pTk = ,
(1 + β)k
E[N ] = 1 + β, Var[N ] = β(1 + β),
β̂ = μ̂ − 1,
[1 − β(z − 1)]−1 − (1 + β)−1
P (z) = .
1 − (1 + β)−1
This is a special case of the zero-truncated negative binomial with r = 1.
B.3.1.3 Logarithmic–β
β β β
pT1 = , a= , b=− ,
(1 + β) ln(1 + β) 1+β 1+β
βk
pTk = ,
k(1 + β)k ln(1 + β)
β[1 + β − β/ ln(1 + β)]
E[N ] = β/ ln(1 + β), Var[N ] = ,
ln(1 + β)
nμ̂ 2(μ̂ − 1)
β̃ =− 1 or ,
n1 μ̂
ln[1 − β(z − 1)]
P (z) = 1 − .
ln(1 + β)
This is a limiting case of the zero-truncated negative binomial as r → 0.

795
B.3.1.4 Zero-truncated binomial–q, m, (0 < q < 1, m an integer)
m(1 − q)m−1 q q (m + 1)q

pT1 = m
, a=− , b= ,
1 − (1 − q) 1−q 1−q
¡m¢ k m−k
k q (1 − q)
pTk = , k = 1, 2, . . . , m,
1 − (1 − q)m
mq
E[N ] = ,
1 − (1 − q)m
mq[(1 − q) − (1 − q + mq)(1 − q)m ]
Var[N ] = ,
[1 − (1 − q)m ]2
μ̂
q̃ = ,
m
[1 + q(z − 1)]m − (1 − q)m
P (z) = .
1 − (1 − q)m
B.3.1.5 Zero-truncated negative binomial–β, r, (r > −1, r 6= 0)
rβ β (r − 1)β
pT1 = , a= , b= ,
(1 + β)r+1 − (1 + β) 1+β 1+β
µ ¶k
r(r + 1) · · · (r + k − 1) β
pTk = ,
k![(1 + β)r − 1] 1+β
rβ
E[N ] = ,
1 − (1 + β)−r
rβ[(1 + β) − (1 + β + rβ)(1 + β)−r ]
V ar[N ] = ,
[1 − (1 + β)−r ]2
σ̂2 μ̂2
β̃ = − 1, r̃ = 2 ,
μ̂ σ̂ − μ̂
[1 − β(z − 1)]−r − (1 + β)−r
P (z) = .
1 − (1 + β)−r
This distribution is sometimes called the extended truncated negative binomial distribution because the
parameter r can extend below 0.
B.3.2 The zero-modified subclass

A zero-modified distribution is created by starting with a truncated distribution and then placing an arbitrary
amount of probability at zero. This probability, pM 0 , is a parameter. The remaining probabilities are
adjusted accordingly. Values of pM k can be determined from the corresponding zero-truncated distribution
as pM M T M M
k = (1 − p0 )pk or from the corresponding (a, b, 0) distribution as pk = (1 − p0 )pk /(1 − p0 ). The same
recursion used for the zero-truncated subclass applies.
The mean is 1 − pM 0 times the mean for the corresponding zero-truncated distribution. The variance is
1 − pM M M
0 times the zero-truncated variance plus p0 (1 − p0 ) times the square of the zero-truncated mean. The
probability generating function is P (z) = p0 + (1 − pM
M M
0 )P (z), where P (z) is the probability generating
function for the corresponding zero-truncated distribution.
The maximum likelihood estimator of pM 0 is always the sample relative frequency at 0.
796 EXAM C TABLES
Bibliography
[1] S.A. Klugman , H.H. Panjer, and G.E. Willmot Loss Models from Data
to Decisions, 3rd Edition (2008), Wiley.
[2] M.B. Finan, A Probability Course for the Actuaries: A Preparation for
Exam P/1, 2007, Arkansas Tech University.
[3] M.B. Finan, A Reading in the Theory of Life Contingency Models: A

preparation for Exam MLC/3L, 2011, Arkansas Tech University.
[4] SOA/CAS, Exam C Sample Questions.
[5] IAA, Stochastic Modeling: Theory and Reality from an Actuarial

Perspective, 2010.
797
Index
C(a, b, 0), 195 Claim count distribution, 258

C(a, b, 1), 201 Claim count random variable, 258
D(x)−plot, 514 Claim experience, 551
p − p plot, 514 Classical credibility, 551
Coefficient of kurtosis, 37
Absolute loss, 485 Coefficient of skewness, 35
Aggregate claims model, 250 Coefficient of variation, 38
Aggregate loss, 257 Coherent risk measure, 110
Alternative hypothesis, 344 Coinsurance, 237
Anderson-Darling test, 527 Collective premium, 582
Arithmetic distribution, 308 collective premium, 605
Arithmetization, 308 Complement, 6
Asymptotic variance of the MLE, 459 Complete data, 351, 363
Asymptotically unbiased, 335, 459 Complete expectation of life, 49, 98
Bühlmann credibility factor, 605 Compound distribution, 250, 258
Bühlmann credibility premium, 606 Compound frequency model, 250
Bayes estimate, 485 Compound Poisson distribution, 296
Bayes Theorem, 475 Conditional hazard function, 160
Bayes’ formula, 10 Conditional probability, 9
Bayesian approach, 474 Confidence interval, 341
Bayesian Central Limit Theorem, 489 Confidence region, 471
Bayesian premium, 582 Conjugate prior distribution, 493
Bernoulli r.v., 189 Consistent estimator, 336
Biased, 334 Continuous r.v., 18
Binomial random variable, 40, 189 Convex cone, 110
Bootstrap method, 686 Convolutions, 264
Cost per loss, 52
Central limit theorem, 70 Cost per-payment, 213
Central moment, 34 Cost-per loss, 212
Chi square distribution, 471 Countable additivity, 7
Chi-square goodness-of-fit test, 532 Counting distribution, 177
Claim amount, 48, 211 Covariance matrix, 464
798
INDEX 799
Cramér-Rao lower bound, 458 First quartile, 63

Credibility factor, 561 Fisher information, 455
Credibility interval, 488 Fisher information matrix, 463
Credibility premium, 561 force of mortality, 26
Credibility theory, 551 Frailty model, 160
credibility weighted average method, Franchise deductible, 220
657 Frequency distribution, 258
Critical values, 345 Frequentest approach, 474
Cumulative distribution function, 22
Cumulative hazard function, 26 Gamma kernel, 402
Cumulative hazard rate function, 353 Gamma random variable, 41
Geometric, 185
Data-dependent distribution, 138 Geometric random variable, 40
Decomposition property, 179 Greatest accuracy credibility theory,
Delta method, 465 567
Deterministic model, 2 Greenwood’s approximation, 389
Discrete r.v., 18
Hazard rate function, 26
Empirical Bayes estimation, 643 Heavy-tailed distribution, 87, 88
Empirical distribution, 138, 352 Hewitt’s model, 633
Empirical model, 45 Histogram, 358
Equilibrium distribution, 102 Hypothesis testing, 344
Estimate, 334 Hypothetical mean, 582, 605
Estimator, 334
Event, 6 Improper prior distribution, 474
Exact credibility, 637 Incomplete data, 351, 363
Excess loss variable, 48, 213 incomplete Gamma function, 294
Expected shortfall, 119 Independent random variables, 69
Expected value of the hypothetical Individual risk model, 258, 321
mean, 605 Interquartile range, 64
Expected value of the process vari- Intersection, 6
ance, 605 interval estimation, 341
Expected value premium principle, 113 Inverse distribution, 148
Experiment, 6 Inverse transformed distribution, 148
Exponential premium principle, 112 Inversion method, 668
Exponential random variable, 41
Exposure, 63, 552 Joint density function, 153
Extended truncated negative binomial Joint distribution, 581
distribution, 206
k-point mixture, 131
Failure rate, 26 Kapler-Meier estimator, 370
800 INDEX
Kernel density estimator, 400 Mass dispersal method, 308

Kernel function, 400 Maximum covered loss, 237
Kernel smoothed density function, 139 Maximum likelihood estimate, 434
Kernel smoothed distribution, 138 mean excess loss function, 49
Kernel smoothing, 400 mean residual life function, 49
Kolmogorov axioms, 7 Mean square error, 686
Kolmogorov-Smirnov statistic, 521 Mean-squared error, 337
Kurtosis, 37 Median, 62
Medium-tailed distribution, 96
Left censored, 364 Memoryless property, 185
Left truncated, 364 Method of Rounding, 308
Left truncated and shifted variable, Method-of-moments estimate, 421
213 Mixed random variable, 18
left truncated and shifted variable, Mixing weight, 131
48 Mixture distribution, 131
Left-censored and shifted, 52 Mode, 63
Level of confidence, 341, 345 Model distribution, 475, 581
Life table approach, 415 Modified data, 363
Light tailed distribution, 88 Moment, 34
Light-tailed distribution, 87 Moment generating function, 78
Likelihood function, 434 Monotonicity, 110
Likelihood ratio test, 540 Mutually exclusive, 6
Limited expected value, 53
Limited loss random variable, 233 Negative binomial distribution, 183
Limited loss variable, 53 Nelson-Åalen, 373
linear confidence intervals, 394 Nelson-Åalen estimate, 353
Linear exponential family, 172 Net stop-loss premium, 284
Local moment matching method, 309 Non-parametric distributions, 138
Log-transformed confidence intervals, Nonparametric Estimation, 643
394 Normal equations, 601
Normal random variable, 40
Logarithmic distribution, 206, 207
Null hypothesis, 344
Loglikelihood function, 435
Loss Elimination Ratio, 226 Observed information, 464
Loss frequency, 257 Observed pure premium, 552
Loss random variable, 48 Ogive, 357
Loss severity, 257 Ordinary deductible, 212
Outcome, 6
Manual rate, 551
Marginal density, 153 p-value, 347
Marginal distribution, 581 Panjer Recursive formula, 300
INDEX 801
Parameters, 126 Right truncated, 364

Parametric distribution, 126 Right-censored r.v., 53
Parametric distribution family, 128 Risk group, 551
Pareto random variable, 43, 44 Risk management, 110
Partial credibility, 561 Risk measure, 110
Percentile, 63 Risk set, 352, 365
Percentile matching estimate, 423
Point estimation, 334 Sample space, 6
Poisson r.v., 178 Scale distribution, 126
Poisson random variable, 40 Scale parameter, 127
Policy limit, 53 Schwarz Bayesian adjustment, 546
Positive homogeneity, 110 Schwarz Bayesian Criterion, 546
Posterior distribution, 581 Secondary distribution, 250, 258
Predictive distribution, 475, 581 Semi-parametric distributions, 138
Primary distribution, 250, 258 semiparametric, 661
Principle of Parsimony, 546 Severity distribution, 258
Prior distribution, 474, 581 Severity random variable, 48, 258
Probability, 7 Simple event, 6
Probability axioms, 7 Single-decrement, 413
Probability density function, 21 Skewness, 35
Probability function, 19, 177 Smoothed empirical estimate, 423
Probability generating function, 82 Span, 308
Probability mass function, 19 Spliced distribution, 165
Probability measure, 7 Squared-error loss, 485
Probability plot, 514 standard deviation principle, 114
Probability tree, 8 Standard of full credibiltiy, 554
Process variance, 605 Statistic, 334
Pure premium, 552 Statistical hypothesis, 344
Stochastic model, 2
Quantile, 63 Stop loss premium, 52
Stop-loss insurance, 284
Railty random variable, 160 structural parameters, 643
Random variable, 18 Subadditivity, 110
Raw moment, 34 Support, 18
Rejection region, 345 Survival function, 25
Relative tail weight, 89
Reliability function,, 25 Tail conditional expectation, 119
Residual mean lifetime, 98 Tail-Value-at-Risk, 119
Right censored, 364 Test statistic, 345
Right tail, 87 Transformed distribution, 148
802 INDEX
Translation invariance, 110

Triangular kernel, 401
Type I error, 345
Type II error, 346
Unbiased, 334
Unbiased equation, 600
Uniform kernel, 139, 400
Uniformly minimum variance unbi-
ased estimator, 338
Union, 6
Value at risk, 115

Variable-component mixture distribu-
tion, 133
Variance of the hypothetical mean,
605
Variance premium principle, 111
Zero-modified, 201
Zero-one loss, 485
Zero-truncated, 201

Exam C Manual

Uploaded by

Copyright:

Exam C Manual

Uploaded by

Document Information

Original Description:

Copyright

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Exam C Manual

Uploaded by

Copyright:

An Introductory Guide in the Construction of

This is the fifth of a series of lecture notes intended to help individuals

Problems taken from previous SOA/CAS exams will be indicated by the

A Review of Probability Related Results 5

Tail Weight of a Distribution 87

Risk Measures 109

Characteristics of Actuarial Models 125

Discrete Distributions 177

Modifications of the Loss Random Variable 211

Aggregate Loss Models 257

Review of Mathematical Statistics 333

The Empirical Distribution for Complete Data 351

Estimation of Incomplete Data 363

Methods of parameter Estimation 419

Model Selection and Evaluation 513

Basics of Stochastic Simulation 667

Answer Key 691

Exam C Tables 781

This book is concerned with the construction and evaluation of actuarial

1 Understanding Actuarial Models

A “model” in actuarial applications is a simplified mathematical descrip-

Commonly used actuarial models are classified into two categories:

(II) Stochastic or Probabilistic Models. In contrast to deterministic

In [1], the following process for building an actuarial model is presented.

Phases of a Good Modeling Process

A good modeling requires a thorough understanding of the problem mod-

Choice of Models. Appropriate models are selected based on the actuary’s

Adequacy of Models. There is a possibility that the models in the previ-

Modification of Models. The model selected in the previous stage needs

Determine whether this model is deterministic or stochastic.

Determine whether this model is deterministic or stochastic.

Does this model considered a deterministic or stochastic model?

Is this model categorized as stochastic or deterministic?

Are these models considered deterministic or stochastic?

One aspect of insurance is that money is paid to policyholders upon the

2 A Brief Review of Probability

By an outcome or simple event we mean any result of the experiment.

The sample space Ω of an experiment is the set of all possible outcomes

Ω = {T T T, T T H, T HT, T HH, HT T, HT H, HHT, HHH}.

{T HH, HT H, HHT, HHH}

The complement of an event E, denoted by E c , is the set of all possible

Two events A and B are said to be mutually exclusive if they have no

Axiom 1: For any event E ⊆ Ω, 0 ≤ Pr(E) ≤ 1.

If we let E1 = Ω, En = ∅ forPn > 1 then by Axioms P 2 and 3 we have

partial information, and we refer to this (altered) probability as a condi-

If the occurrence of the event A depends on the occurrence of B then the

provided that Pr(B) > 0.

Bayes’ formula is a simple mathematical formula used for calculating Pr(B|A)

Pr(A) = Pr(A ∩ B) + Pr(A ∩ B c )

Pr(A) = Pr(B)Pr(A|B)+Pr(B c )Pr(A|B c ) = (0.60)(0.35)+(0.4)(0.85) = 0.55

From Equation (2.1) we can get Bayes’ formula:

Theorem 2.1 (Bayes’ Theorem)

Pr(A) = Pr(H1 )Pr(A|H1 ) + Pr(H2 )Pr(A|H2 ) + · · · + Pr(Hn )Pr(A|Hn ).

Determine the probability that a randomly chosen employee will choose

Find the probability that both lanes are not busy.

Pr(A ∪ B) = Pr(A) + Pr(B) − Pr(A ∩ B).

Determine the probability that a visit to a PCP’s office results in both

Determine the probability that a randomly chosen member of this group

Under this assumption, what is the probability that a policyholder files

Calculate the probability that the driver was age 16-20.

What is the probability that the deceased policyholder was ultra-preferred?

(i) 10% of the emergency room patients were critical;

Calculate the probability that the participant was a heavy smoker.