CK1 Booklet 1 PDF

Exclusive use Batch 3a
Subject CS2
Revision Notes
For the 2019 exams
Exposed to risk
Booklet 5
covering
Chapter 9 Exposed to risk
The Actuarial Education Company

CONTENTS
Contents Page
Links to the Course Notes and Syllabus 2

Overview 3
Core Reading 4
Past Exam Questions 12
Solutions to Past Exam Questions 29
Factsheet 75
Copyright agreement
All of this material is copyright. The copyright belongs to Institute and

Faculty Education Ltd, a subsidiary of the Institute and Faculty of Actuaries.
The material is sold to you for your own exclusive use. You may not hire
out, lend, give, sell, transmit electronically, store electronically or photocopy
any part of it. You must take care of your material to ensure it is not used or
copied by anyone at any time.
Legal action will be taken if these terms are infringed. In addition, we may
seek to take disciplinary action through the profession or through your
employer.
These conditions remain in force after you have finished using the course.
© IFE: 2019 Examinations Page 1

LINKS TO THE COURSE NOTES AND SYLLABUS
Material covered in this booklet
Chapter 9 Exposed to risk
These chapter numbers refer to the 2019 edition of the ActEd course notes.
Syllabus objectives covered in this booklet
4.4 Estimate transition intensities dependent on age (exact or census).
4.4.1 Explain the importance of dividing the data into

homogeneous classes, including subdivision by age and
sex.
4.4.2 Describe the principle of correspondence and explain its

fundamental importance in the estimation procedure.
4.4.3 Specify the data needed for the exact calculation of a

central exposed to risk (waiting time) depending on age
and sex.
4.4.4 Calculate a central exposed to risk given the data in 4.4.3.
4.4.5 Explain how to obtain estimates of transition probabilities.
4.4.6 Explain the assumptions underlying the census

approximation of waiting times.
4.4.7 Explain the concept of the rate interval.
4.4.8 Develop census formulae given age at birthday where the

age may be classified as next, last, or nearest relative to
the birthday as appropriate, and the deaths and census
data may use different definitions of age.
4.4.9 Specify the age to which estimates of transition intensities

or probabilities in 4.4.8 apply.
Page 2 © IFE: 2019 Examinations

OVERVIEW
This booklet covers Syllabus objective 4.4, which relates to the estimation of
transition intensities (eg mortality rates) depending on age.
This involves calculating the number of transitions (eg deaths) and dividing
this by an appropriate exposed to risk. The data are subdivided according to
a specific age definition, enabling us to estimate mortality rates for specific
ages. The process is summarised by the formula:
dx
mˆ x =
E xc
Note that:
 mˆ x is the estimate of m x + f , the transition rate at a particular age x + f
(where f is a value dependent upon the age definition of the data)
 d x (also sometimes written as q x ) is the number of transitions with the

given age definition, and
 E xc is the central exposed to risk for the given age definition.
The main issues are:

 Determining the age x + f to which the estimated rates apply: this
involves using the concept of the rate interval implied by the definition of
deaths.
 Deriving a formula for the central exposed to risk. This can either be
done exactly – by measuring observed waiting times for each individual
– or approximately using census formulae.
 In deriving the exposed to risk, careful attention has to be paid towards
satisfying the principle of correspondence – that the exposure and
deaths relate to the same age definition – and this sometimes leads to
problems in identifying exactly which census data are needed for the
formulae.
Questions on exposed to risk are largely preoccupied with one or several of

the above issues. Questions can also involve a discussion of why data need
to be subdivided into homogeneous classes for analysis and the factors by
which data are often divided.

CORE READING
All of the Core Reading for the topics covered in this booklet is contained in
this section.
We have inserted paragraph numbers in some places, such as 1, 2, 3 …, to

help break up the text. These numbers do not form part of the Core
Reading.
The text given in Arial Bold font is Core Reading.
The text given in Arial Bold Italic font is additional Core Reading that is not
directly related to the topic being discussed.
____________
Chapter 9 – Exposed to risk
Calculating the exposed to risk
We have seen how exposed to risk arises in a probabilistic model of

mortality.
In this booklet we consider some problems of a computational nature,

concerning the approximation of exposed to risk from incomplete
exposure data.
The central exposed to risk is a very natural quantity, intrinsically

observable even if observation may be incomplete in practice – that is,
just record the time spent under observation by each life. Note that
this is so even if lives are observed for only part of the year of age
[ x , x + 1] , for whatever reason.
The central exposed to risk carries through unchanged to arbitrarily

complicated multiple-decrement or multiple-state models. As we shall
see, it can easily be approximated in terms of the kind of incomplete
observations that are typically available in insured lives investigations.
____________

1 The problem of heterogeneity
The multiple state and Poisson models analyses are based on the
assumption that we can observe groups of identical lives (or at least
lives whose mortality characteristics are the same). In practice, this is
never possible.
____________
2 The solution
We can subdivide our data according to characteristics known, from

experience, to have a significant effect on mortality. This ought to
reduce the heterogeneity of each class so formed, although much will
probably remain.
____________
3 Among the factors in respect of which life insurance mortality statistics

are often sub-divided are:
(a) Sex
(b) Age (as we have assumed throughout)
(c) Type of policy (which often reflects the reason for insuring)
(d) Smoker/non-smoker status
(e) Level of underwriting
(f) Duration in force.
____________
4 Others that might be used are:

(g) Sales channel
(h) Policy size
(i) Occupation of policyholder
(j) Known impairments
(k) Postcode/geographical location
(l) Marital status.
____________

5 Two key points are:
(a) Sub-division cannot be carried out unless the relevant information

is collected, generally on the proposal form. Sometimes factors
for which there is strong external evidence of an effect on mortality
cannot be used because (for example) proposal forms have been
kept short for marketing or administrative reasons.
(b) Even in quite large investigations, sub-division using many factors

results in much smaller populations in each class, making the
statistics more difficult. A balance must be struck between
obtaining more and more homogeneity, and retaining large enough
populations to make analysis possible.
____________
The principle of correspondence
Mortality investigations based on estimation of m x + ½ at individual

ages must bring together two different items of data: deaths and
exposures. It is self-evident that these should be defined consistently,
or their ratios are meaningless. Care is sometimes needed, however,
because these data are often obtained from different sources in the life
office. For example, death data might be obtained from the claims file,
while exposure data might be obtained from the premium collection
file. There is no guarantee that these use the same definition of the
policyholders’ ages.
A precise statement of what we mean by ‘defined consistently’ is given

by the principle of correspondence.
____________
6 A life alive at time t should be included in the exposure at age x at

time t if and only if, were that life to die immediately, he or she would
be counted in the death data d x at age x .
____________
This seems almost a triviality, but it is very important and useful.

____________

Working with complete data
7 The procedure for the exact calculation of E cx is obvious:
(a) record all dates of birth

(b) record all dates of entry into observation
(c) record all dates of exit from observation
(d) compute E cx .
If we add to the data above the cause of the cessation of observation,

we have d x as well, and we have finished.
____________
Working with incomplete data
All of the remainder of the Core Reading in this booklet is about

approximate procedures when the data above have not been recorded.
We will deal with two questions:
(a) What happens when the dates of entry to and exit from observation
have not been recorded?
(b) What happens if the definition of age does not correspond exactly
to the age interval x to x + 1 (for integer x )?
____________
The available data
8 Suppose that we have death data of the form:
d x = total number of deaths age x last birthday during calendar

years K , K + 1,..., K + N
That is, we have observations over N + 1 calendar years of all deaths

between ages x and x + 1 .

However, instead of the times of entry to and exit from observation of

each life being known, we have instead only the following census data:
Px ,t = Number of lives under observation aged x last birthday at

time t where t = 1 January in calendar years K , K + 1 , …,
K +N , K +N +1
This is in fact similar to the way in which data are submitted to the CMI.
It is often quite convenient for companies to submit a total of policies
in force on a date such as 1 January.
____________
The census approximation to E cx
9 Now define Px ,t to be the number of lives under observation, aged x

last birthday, at any time t . Note that:
K + N +1
E cx = Ú Px ,t dt
K
We have the values of Px ,t only if t is a 1 January (a census date), so
we must estimate E cx from the given census data.
The problem reduces to estimating an integral, given the integrand at

just a few points (in this case, integer spaced calendar times). This is a
routine problem in numerical analysis.
____________
10 The simplest approximation, and the one most often used, is that Px ,t
is linear between census dates, leading to the trapezium
approximation.
____________
K + N +1 K +N
11 E cx = Ú Px ,t dt ª Â ½(Px ,t + Px ,t + 1)
K t =K
____________

This is the method used by the CMI. It is easily adapted to census data
available at more or less frequent intervals, or at irregular intervals.
____________
Deaths classified using different definitions of age
12 Earlier, we used a definition of age ‘x last birthday’, which identifies the

year of age [ x , x + 1] .
Other definitions could be used, for example
d x(2) = total number of deaths age x nearest birthday during

calendar years K , K + 1,..., K + N
d x(3) = total number of deaths age x next birthday during calendar

years K , K + 1,..., K + N
Each of these identifies a different year of age, called the rate interval.
Consequently, estimates of m , or q based on m , obtained from these

data ( d x(2) and d x(3) ) will not be estimates of m x + ½ or q x , but will be
estimates of m and q at other ages.
____________
13 We summarise the possibilities as follows:
Definition of x Rate interval m̂ estimates q̂ estimates
Age last birthday [ x , x + 1] m x +½ qx
Age nearest birthday [ x - ½, x + ½] mx q x ½
Age next birthday [ x - 1, x ] mx -½ q x 1
____________

14 Once the rate interval has been identified (from the age definition used
in d x ) the rule is that:
 the crude m̂ estimates m in the middle of the rate interval, and
 the crude q̂ estimates q at the start of the rate interval.

____________
Consistency between census data and death data
We must ensure that the census data are consistent with the death
data. We invoke the principle of correspondence; we must check the
following:
____________
15 The census data Px ,t are consistent with the death data d x if and only
if, were any of the lives counted in Px ,t to die on the census date, he or
she would be included in d x .
____________
16 The definition of census data corresponding to the rate interval

[ x - ½, x + ½] is:
Px(2)
,t = Number of lives under observation, age x nearest birthday
at time t , where t = 1 January in calendar year K , K + 1 ,
…, K + N , K + N + 1
and the definition of census data corresponding to the rate interval

[ x - 1, x ] is:
Px(3)
,t = Number of lives under observation, age x next birthday at
time t , where t = 1 January in calendar years K , K + 1 ,
…, K + N , K + N + 1
___________

17 In the event that the death data and the census data use different
definitions of age, we must adjust the census data. Unless it is
unavoidable, we never adjust the death data, since that ‘carries most
information’ when rates of mortality are small. Hence, it is always the
death data that determines what rate interval to use.
___________
For example, the CMI uses the definition ‘age nearest birthday’ in its
work; that is, death data as in d x(2) . However, some life offices
contribute census data classified by ‘age last birthday’, because that is
what is available from their records. The latter must be adjusted in
some way.
____________
18 For example, if we define:
Px¢ ,t = ½(Px -1,t + Px ,t )
we can see that Px¢ ,t approximates Px(2)

,t .
____________

PAST EXAM QUESTIONS
This section contains all the relevant exam questions from 2008 to 2017 that
are related to the topics covered in this booklet.
Solutions are given after the questions. These give enough information for
you to check your answer, including working, and also show you what an
outline examination answer should look like. Further information may be
available in the Examiners’ Report, ASET or Course Notes. (ASET can be
ordered from ActEd.)
We first provide you with a cross-reference grid that indicates the main
subject areas of each exam question. You can use this, if you wish, to
select the questions that relate just to those aspects of the topic that you
may be particularly interested in reviewing.
Alternatively, you can choose to ignore the grid, and attempt each question
without having any clues as to its content.

9
8
7
6
5
4
3
2
1
24
23
22
21
20
19
18
17
16
15
14
13
12
11
10
Question
Problem of




heterogeneity









Subdividing data
Principle of






correspondence
Cross-reference grid
© IFE: 2019 Examinations

ETR – exact



calculation
ETR – census













 method









Estimating rates
Age to which the




 rate applies
Question attempted
Page 13
1 Subject CT4 April 2008 Question 1
List four factors in respect of which life insurance mortality statistics are often
subdivided. [2]
2 Subject CT4 September 2009 Question 3
(i) List the data needed for the exact calculation of a central exposed to
risk depending on age. [2]
An investigation studied the mortality of persons aged between exact ages

40 and 41 years. The investigation began on 1 January 2008 and ended on
31 December 2008. The following table gives details of 10 lives involved in
the investigation.
Life Date of 40th birthday Date of death

1 1 March 2007 –
2 1 May 2007 1 October 2008
3 1 July 2007 –
4 1 October 2007 –
5 1 December 2007 1 February 2008
6 1 February 2008 –
7 1 April 2008 –
8 1 June 2008 1 November 2008
9 1 August 2008 –
10 1 December 2008 –
Persons with no date of death given were still alive when the investigation
ended.
(ii) Calculate a central exposed to risk using the data for the 10 lives in the
sample. [3]
(iii) (a) Calculate the maximum likelihood estimate of the hazard of death at
age 40 last birthday.
(b) Hence, or otherwise, estimate q40 . [2]
[Total 7]

(i) In the context of mortality investigations describe the principle of

correspondence and give an example of a situation in which it may be
hard to adhere to this principle. [2]
On 1 January 2005 a country introduced a comprehensive system of death

registration, which classified deaths by age last birthday on the date of
death.
The government of the country wishes to obtain estimates of the force of

mortality, m x , by single years of age x for the period between 1 January
2005 and 1 January 2008. Annual population censuses have been taken on
30 June each year since 2004, which classify the population by age last
birthday. However the only copy of the data from the population census of
30 June 2006 was lost when the computer disc on which it was stored was
being transferred between government departments.
Let the population aged x last birthday on 30 June in year t be denoted by

the symbol Px,t , and the number of deaths during the period of investigation
of persons aged x be denoted by the symbol d x .
(ii) Derive an expression in terms of Px,t and d x which may be used to

estimate m x . [6]
[Total 8]
List four factors often used to subdivide life insurance mortality statistics. [2]

An oil company has discovered a vast deposit of oil in an equatorial swamp.

The area is extremely unhealthy and inhabited by venomous spiders. There
is an antidote to bites from these spiders but it is expensive. The antidote
acts instantly but does not provide future immunity. The company
commissions a study to estimate the rate of being bitten by the spiders
among its employees, in order to determine the amount of antidote to
provide.
Employees of the company are posted to the swamp for six month tours of
duty starting on 1 January, 1 April, 1 July or 1 October. The first employees
to be posted arrived on 1 January 2008. The swamp is so inaccessible that
no employees are allowed to leave before their six month tours of duty are
completed.
Accidental deaths are common in this dangerous location.
The table below gives some data from the study.
Quarter Number of new Number of Number of

beginning arrivals at start accidental spider bites
of quarter deaths during during quarter
quarter
1 January 2008 90 10 15
1 April 2008 80 8 25
1 July 2008 114 10 30
1 October 2008 126 13 40
(i) Estimate the quarterly rate of being bitten by a spider for each quarter of
2008, stating any assumptions you make. [7]
(ii) Suggest reasons why the assumptions you made in (i) might not be
valid. [1]
[Total 8]

Two neighbouring small countries have for many years taken annual
censuses of their populations on 1 January in which each inhabitant must
give his or her age. Country A uses an ‘age last birthday’ definition of age,
whereas Country B uses an ‘age nearest birthday’ definition. Each country
has also operated a system in which deaths are recorded on an ‘age nearest
birthday at date of death’ basis.
On 30 June 2009 Country A invaded Country B and the two countries

became one state. The new government wishes to estimate a single set of
age-specific death rates, m x , for the new unified state using the census data
taken in the years before the invasion.
Derive a formula which the new government may use to estimate m x in

terms of the recorded number of deaths in each country, and the population
of each country recorded as being aged x in the censuses. State any
assumptions you make. [8]
(i) State the principle of correspondence as it applies to the estimation of

mortality rates. [1]
(ii) Explain why it might be difficult to ensure the principle of

correspondence is adhered to, and give a specific example of an
investigation where this may be the case. [2]
An actuary was asked to investigate the mortality of lives in a particular

geographical area. Data are available of the population of this area,
classified by age last birthday, on 1 January in each year. Data on the
number of deaths in this area in each calendar year, classified by age
nearest birthday at death, are also available.
(iii) Derive a formula which would allow the actuary to estimate the force of
mortality at age x + f , m x + f , in a particular calendar year, in terms of
the available data, and derive a value for f . [6]
(iv) List four factors other than geographical location which a government
statistical office might use to subdivide data for national mortality
analysis. [2]
[Total 11]

(i) Explain the reasons why data are subdivided when conducting mortality
investigations. [2]
(ii) Describe the problems which can arise with subdividing data. [2]
[Total 4]
(i) List four factors other than age and smoker status by which life
insurance mortality statistics are often subdivided. [2]
Two offices in different towns of the same life insurance company write
25-year term assurance policies. Below are data from these two offices
relating to policyholders of the same age. Both deaths and policies in force
are on an age last birthday basis.
Gasperton Great Hawking

Policies in force on 1 January 2009 2,000 1,770
Policies in force on 1 January 2010 2,100 1,674
Deaths in calendar year 2009 25 21
(ii) Calculate the central death rate for the calendar year 2009 at this age
for the offices in Gasperton and Great Hawking. [2]
A detailed examination of the records shows that 50% of the policyholders in

Gasperton at both censuses were smokers, and 20% of policyholders in
Great Hawking at both censuses were smokers. National death rates at this
age for smokers in 2009 were 40% higher than those for non-smokers.
(iii) Estimate the central death rates for smokers and non-smokers in
Gasperton and Great Hawking. [4]
The life insurance company charges policyholders in Gasperton and Great

Hawking the same premiums for the 25-year term assurance policies.
It charges smokers in both towns 40% more than non-smokers.
(iv) Comment on the company’s pricing structure in the light of your results
from parts (ii) and (iii) above. [3]
[Total 11]

(i) State the principle of correspondence as it applies to mortality rates. [1]
A life insurance company has the following data:
Number of policies in force on

Age last 1 January 1 January 1 July 1 January
birthday 2009 2010 2010 2011
49 2,000 2,100 2,300 2,500
50 2,100 2,200 2,300 2,400
51 2,300 2,400 2,500 2,600
Number of deaths classified by age next birthday and calendar year
Age next birthday 2009 2010
49 175 200
50 200 225
51 225 235
(ii) Estimate, using these data, the force of mortality at age 50 next birthday
for the period 1 January 2009 to 1 January 2011. [5]
(iii) State the exact age to which your answer to part (ii) relates. [1]
[Total 7]

Population censuses in a certain country are taken each year on the

President’s birthday, provided that the President’s astrological advisor
deems the taking of a census favourable. Censuses record the age of every
inhabitant in completed years (that is, curtate age). Deaths in this country
are registered as they happen, and classified according to age nearest
birthday at the time of death.
Below are some data from the three most recent censuses.
Age in
Population 2006 Population 2009 Population 2010
completed
(thousands) (thousands) (thousands)
years
64 300 320 350
65 290 310 330
66 280 300 320
Between the censuses of 2006 and 2009 there were a total of 3,000 deaths
to inhabitants aged 65 nearest birthday, and between the censuses of 2009
and 2010 there were a total of 1,000 deaths to inhabitants aged 65 nearest
birthday.
(i) Estimate, stating any assumptions you make, the death rate at age 65
years for each of the following periods:
 the period between the 2006 and 2009 censuses
 the period between the 2009 and 2010 censuses. [6]
(ii) Explain the exact age to which your estimates apply. [1]
[Total 7]
Data are often subdivided when investigating mortality statistics.
(i) Explain why this is done. [2]
(ii) Discuss one potential problem with subdividing mortality data. [2]
(iii) List four factors which are commonly used to subdivide mortality data. [2]
[Total 6]

(i) Explain why data are subdivided into homogeneous groups when
mortality investigations are conducted. [2]
(ii) List four factors, other than age and sex, by which mortality statistics are
often subdivided. [2]
[Total 4]
(i) State the principle of correspondence as it relates to mortality

investigations. [1]
Two small countries conduct population censuses on an annual basis.

Country A records its population on 1 February every year based on an age
definition of age last birthday. Country B records its population on every
1 August using a definition of age nearest birthday. Each country records
deaths as they happen based on age next birthday.
Below are some data from the last few years.
Country A
Age last Population Population Population
birthday 1 February 2011 1 February 2012 1 February 2013
44 382,000 394,000 401,000
45 374,000 381,000 385,000
46 354,000 372,000 375,000
Country B
Age nearest Population Population Population
birthday 1 August 2011 1 August 2012 1 August 2013
44 382,000 394,000 401,000
45 374,000 381,000 385,000
46 354,000 372,000 375,000
In the combined lands of Countries A and B in the calendar year 2012 there
were 4,800 deaths of those aged 46 next birthday and 4,500 deaths of those
aged 45 next birthday.
The two countries decide to form an economic union, after which it will be
mandatory to offer the same rates for life insurance to residents of each
country.

(ii) Estimate the death rate at age 45 years last birthday for the two
countries combined. [6]
(iii) Explain the exact age to which your estimate relates. [1]
[Total 8]
(i) Explain the census approximation for calculating the exposed to risk
between any two census dates. [2]
A mortality investigation bureau has collected the following information on

number of policies in-force each year from different companies.
Age and
Year Company A Company B Company C
year
Age 54 2011 3,400 1,250 5,780
2012 3,350 1,450 5,500
2013 3,000 1,500 6,010
Age 55 2011 3,250 1,190 6,000

2012 3,390 1,300 5,960
2013 3,100 1,440 6,030
Age 56 2011 3,270 1,150 5,950

2012 3,020 1,300 5,980
2013 2,950 1,500 5,990
 Company A has provided in-force policy data as at the beginning of

each calendar year using age nearest birthday.
 Company B has provided in-force policy data as at the financial year

closing date (which was 31 March in each year) using age last birthday.
 Company C has provided in-force policy data as at the end of each

calendar year using age next birthday.
(ii) Calculate the contribution to central exposed to risk for lives aged 55
last birthday for the calendar year 2012 for each of the companies. [6]
[Total 8]

(i) State the principle of correspondence as it applies to death rates. [1]
A nightclub opens at 10:00pm and closes at 2:00am. It admits only people

aged over 21 years on the production of an identity card giving date of birth.
The table below shows the number of people entering in various intervals
between 10:00pm and 2:00am on 30 June 2013. No one was admitted after
1:00am, and you may assume that all those who enter the premises stay
until 2:00am.
Year of birth 10:00 – 11:30pm 11:30 – 12:00pm 12:00pm –

1:00am
1989 100 300 200
1990 200 400 350
1991 150 400 300
1992 100 250 200
During the period of opening, 40 people aged 22 last birthday required

medical attention for heat exhaustion.
(ii) Calculate the rate per person-hour at which those attending the
nightclub aged 22 last birthday required medical attention for heat
exhaustion, stating any assumptions you make. [6]
[Total 7]
(i) State why it is important to divide data into homogeneous classes when
undertaking mortality investigations. [2]
(ii) List four factors, apart from smoking behaviour, by which mortality data
are often classified by life insurance companies. [2]
In a particular life insurance market, it has for many years been the practice
for all companies to charge smokers higher premiums than non-smokers for
the same term assurance policy. Suppose one company decides to switch
to charging smokers and non-smokers the same premiums for term
assurance policies. The other companies retain differential pricing for
smokers and non-smokers.
(iii) Discuss the likely implications for the company making the switch. [4]
[Total 8]

List four factors, other than age and sex, by which mortality statistics are
often subdivided. [2]
Company A and Company B are two small insurance companies which have
recently merged to form Company C. Company C is reviewing its premium
rates for a whole of life product and so is conducting an analysis of mortality
rates experienced.
Company A recorded the number of policies in force every 1 January using a

definition of age next birthday whereas Company B recorded the number of
policies in force every 1 April using an age definition of age last birthday.
Both companies recorded deaths as they happened using an age definition
of age last birthday.
These are the data for the most recent years.
Company A
Age next Number of Number of Number of
birthday policies 1 Jan policies 1 Jan policies 1 Jan
2012 2013 2014
51 8,192 6,421 8,118
52 7,684 8,298 7,187
53 9,421 8,016 9,026
Company B
Age last Number of Number of Number of
birthday policies 1 April policies 1 April policies 1 April
2012 2013 2014
51 4,496 3,817 4,872
52 5,281 5,218 3,812
53 4,992 5,076 5,076
In the calendar year 2013 Company A recorded 28 deaths of those aged 52

last birthday and Company B recorded 17 deaths of those aged 52 last
birthday.

(i) Estimate the force of mortality for the combined company for age 52 last
birthday, stating all assumptions that you make. [6]
(ii) Explain the exact age to which your estimate applies. [1]
[Total 7]
Write down the information required to compute the exact exposed to risk in
an investigation of mortality. [3]
(i) State the principle of correspondence in the context of a mortality

investigation. [1]
A mortality investigation collects the following data:
nx (t ) = total number of policies under which death claims are made when
the policyholder is aged x last birthday for each calendar year t .
Px (t ) = number of in-force policies where the policyholder was aged x

nearest birthday on 1 January in year t .
(ii) (a) Derive an expression, in terms of Px (t ) , for the central exposed to

risk, E xc , corresponding to the claims data which may be used to
estimate the force of mortality in year t at each age x , m x .
(b) State any assumptions that you make, indicating at which point in
your derivation each assumption is relevant. [5]
[Total 6]

An analysis of the number of term assurance policies in force for three

companies has revealed the following information:
Year Company A Company B Company C
2013 6,728 2,643 4,132

Age 50 2014 6,189 2,548
2015 5,962 2,496 4,630
2013 5,987 2,333 4,012

Age 51 2014 6,002 2,417
2015 5,056 2,213 4,500
2013 5,359 2,155 3,895

Age 52 2014 5,600 1,992
2015 4,906 2,006 4,367
 Company A has reported the number of policies in force on 1 January

each year using age nearest birthday.
 Company B has reported the number of policies in force on 1 November
each year using age last birthday.
 Company C has reported the number of policies in force on 31
December each year using age next birthday, but failed to provide data
for 2014.
(i) Calculate the contribution to the central exposed to risk for lives age 51
last birthday for the calendar year 2014 for each company individually. [5]
(ii) (a) State the assumptions you have made in order to perform
your calculations.
(b) Explain why these assumptions were required. [6]

[Total 11]

A large company operates a health benefits scheme which pays a sickness

benefit to any employee who is unable to work through ill-health and a death
benefit to any employee who dies.
(i) Draw a transition diagram with three states which could be used to
analyse data from this scheme. [2]
(ii) Give the likelihood of the data, defining all the terms you use. [4]
(iii) Derive the maximum likelihood estimator of the rate of falling sick. [3]
The company records data on 1 January each year, classified by age last
birthday on:
 the total number of employees (including those in receipt of sickness
benefit).
 the number of employees in receipt of a sickness benefit.
Some recent data are given in the table below:
Age last Total number of employees on Number of employees in

birthday 1 January in year receipt of sickness benefit on
1 January in year
2014 2015 2016 2014 2015 2016
51 148 162 180 12 20 8
52 146 148 160 10 18 7
53 140 144 146 8 20 6
The company wishes to estimate the rates of falling sick and recovery at age
52 years nearest birthday over the two-year period consisting of the calendar
years 2014 and 2015.
(iv) Determine suitable exposed-to-risks for calculating the rates of falling

sick and recovery. [6]
[Total 15]

A study was conducted into the mortality of persons aged between exact
ages 85 and 86 years. The study took place from 1 April 2015 to 31 March
2016. The following table shows information on 10 lives observed in the
study.
Life number Date of 85th birthday Date of death
1 1 August 2014 –
2 1 November 2014 –
3 1 January 2015 1 February 2016
4 1 February 2015 –
5 1 March 2015 –
6 1 April 2015 1 January 2016
7 1 June 2015 1 November 2015
8 1 July 2015 –
9 1 September 2015 1 March 2016
10 1 January 2016 –
(i) Calculate a central exposed to risk for the 10 lives in the sample,
working in months. [3]
(ii) Give the maximum likelihood estimate of the mortality hazard at age 85
last birthday. [1]
(iii) Estimate q85 . [1]

[Total 5]

SOLUTIONS TO PAST EXAM QUESTIONS
The solutions presented here are just outline solutions for you to use to
check your answers. See ASET for full solutions.
Mortality statistics are often subdivided by:

 age
 sex
 postcode
 policy type.
Also see Core Reading Paragraphs 3 and 4.
(i) Data items needed
For each life we require:

 the date of birth
 the date of entry into the investigation
 the date of exit from the investigation.

(ii) Calculation of central exposed to risk
We have:
Central exposed to risk

Life Start Stop
(in months)
1 1 Jan 08 1 Mar 08 2
4 (died aged 41 last
2 1 Jan 08 1 May 08
birthday)
3 1 Jan 08 1 Jul 08 6
4 1 Jan 08 1 Oct 08 9
5 1 Jan 08 1 Feb 08
birthday)
6 1 Feb 08 31 Dec 08 11
7 1 Apr 08 31 Dec 08 9
8 1 Jun 08 1 Nov 08
birthday)
9 1 Aug 08 31 Dec 08 5
10 1 Dec 08 31 Dec 08 1
Adding the central exposed to risk values for all the lives gives a total of 53
months, or 4 years and 5 months.
(iii)(a) MLE of hazard of death
We have observed 2 deaths at age 40 last birthday. So:
2
mˆ = = 0.45283
53 / 12
(iii)(b) Estimate of q40
Using the MLE of the force of mortality from part (iii)(a), we have:
qˆ 40 = 1 - e - m = 1 - e -0.45283 = 0.36417
ˆ

(i) Principle of correspondence and example of when it may be hard to

adhere to
The principle of correspondence says that a life alive at time t should be

included in the exposure at age x at time t if and only if, were that life to
die immediately, he or she would be counted in the death data d x at time t .
It may be difficult to adhere to this principle if the age classification used for
the death data is different from the age classification used for the exposure
data. This can occur when the death data and exposure data come from
different sources.
(ii) Estimate of force of mortality
Assuming that the force of mortality is constant between the exact ages of x
and x + 1 , it can be estimated by:
dx
mˆ x =
E xc
where E xc is the central exposed to risk at age x last birthday.
Let Px (t ) denote the number of lives at time t aged x last birthday, where
time is measured in years from 1 January 2005. (With this definition,
Px ( -½) = Px,2004 , Px (½) = Px,2005 , Px (2½) = Px,2007 , Px (3½) = Px,2008 .)
Then:
3
E xc = Ú0 Px (t ) dt
Since we only know the values of Px (t ) when t = -½, ½, 2½ and 3½, we

need to derive a formula for E xc in terms of these values only.
We can use the information at time -½ and time ½ to estimate Px (0) .

Assuming that Px (t ) varies linearly between time -½ and time ½:
1
Px (0) =
2
(Px ( -½) + Px (½))

We can similarly use the information at time 2½ and time 3½ to

estimate Px (3) . Assuming that Px (t ) varies linearly between time 2½ and
time 3½:
1
Px (3) =
2
(Px (2½) + Px (3½))
So we now know (or have estimated values of) Px (0) , Px (½) , Px (2½) and
Px (3) . If we further assume that Px (t ) varies linearly between time ½ and
time 2½, then:
1 1 1 1 1
E xc = ¥ ÈP (0) + Px (½)˘˚ + 2 ¥ ÈÎPx (½) + Px (2½)˘˚ + ¥ ÈÎPx (2½) + Px (3)˘˚
2 2Î x 2 2 2
1 È1 ˘
= Í (Px ( -½) + Px (½)) + Px (½)˙ + ÎÈPx (½) + Px (2½)˚˘
4 Î2 ˚
1È 1 ˘
+ P (2½) + (Px (2½) + Px (3½)) ˙
4 ÍÎ x 2 ˚
1 11 11 1
= Px ( -½) + Px (½) + Px (2½) + Px (3½)
8 8 8 8
1 11 11 1
= Px,2004 + Px,2005 + Px,2007 + Px,2008
8 8 8 8
Since the rate interval starts at exact age x and ends at exact age x + 1 ,
mˆ x estimates m x +½ .

 age
 sex
 postcode
 policy type.

(i) Estimate the quarterly rate
We will assume that all employees who are bitten by spiders are given the
antidote and make an immediate recovery and that the only deaths that
occur are accidental deaths.
First quarter
The first quarter started on 1 January with 90 employees in the first cohort
and ended with 80 ( = 90 - 10 ) on 31 March. If we assume that the deaths
occurred uniformly over the quarter, this gives a total employee exposed-to-
risk for this period of 21 (90 + 80) = 85 quarters.
Since there were 15 spider bites during this period, the rate of spider bites is
15
= 17.6% per quarter.
85
Second quarter
The second quarter started on 1 April with the remaining 80 employees from
the first cohort plus the 80 new arrivals in the second cohort, making a total
of 160, and ended with 152 ( = 160 - 8 ) on 30 June. If we assume that the
deaths occurred uniformly over the quarter, this gives a total employee
exposed-to-risk for the second quarter of 21 (160 + 152) = 156 quarters.
At the start of the second quarter there were equal numbers (80 of each) of
employees from the first and second cohorts. If we assume that the
proportion of deaths during each quarter was the same for each cohort, we
would expect that the 8 deaths during the second quarter were equally split
between the two groups. So 4 of the employees from the first cohort would
go home on 30 June and 4 employees from the second cohort would have
been killed.
Third quarter
Based on the assumption of equal proportions of deaths for each cohort, the
third quarter started on 1 July with the remaining 76 ( = 80 - 4 ) employees
from the second cohort plus the 114 new arrivals from the third cohort,
making a total of 190, and ended with 180 ( = 190 - 10 ) on 30 September.
This gives a total employee exposed-to-risk for the third quarter of
1 (190 + 180) = 185 quarters.
2

Again, if we assume that the deaths during the third quarter were in
proportion to the numbers at the start of the quarter, then the 10 deaths
76
during the third quarter would consist of ¥ 10 = 4 from the second
190
cohort and 6 from the third cohort.
Fourth quarter
The fourth quarter started on 1 October with the remaining 108 ( = 114 - 6 )
employees from the third cohort plus the 126 new arrivals from the fourth
cohort, making a total of 234, and ended with 221 ( = 234 - 13 ) on 31
December. This gives a total employee exposed-to-risk for the fourth
quarter of 21 (234 + 221) = 227.5 quarters.
Again, if we assume that the deaths during the fourth quarter were in
proportion to the numbers at the start of the quarter, then the 13 deaths
108
during the fourth quarter would consist of ¥ 13 = 6 from the second
234
cohort and 7 from the third cohort.
The results of these calculations are summarised in the following table:
Quarter Number Number Central Spider Estimated

at start at end ETR bites rate
1st 90 80 85 15 17.6%
2nd 160 152 156 25 16.0%
3rd 190 180 185 30 16.2%
4th 234 221 227.5 40 17.6%
(ii) Why the assumption might not be valid
We have assumed that the probability of dying does not vary by duration, the
length of time the employees have been in the region.
However, we might expect that each cohort of employees will become more
experienced by the second half of their tour and will be less likely to be
killed. For the same reason, the assumption of uniform deaths over each
quarter may not be valid.
The question only considers accidental deaths. However, there may also be
deaths from other causes, eg tropical diseases.

Period of investigation
Suppose that:
 the period of investigation covers an n -year period and starts on a 1st
January.
 time is measured in years since the start of the investigation.
Death data
Death data are classified according to age nearest birthday for both
countries. Let:
 d xA = the number of deaths aged x nearest birthday in Country A
 d xB = the number of deaths aged x nearest birthday in Country B
 d x = d xA + d xB = total number of deaths aged x nearest birthday.
Census data
The census data are classified according to age last birthday for Country A
and age nearest birthday for Country B. So the census data and death data
match for Country B but not for Country A.
Exposed to risk for Country A
Let PxA (t ) denote the number of lives in Country A at time t aged x last
birthday. The values of PxA (0) , PxA (1) , ..., PxA (n ) are known for all x .
However, since the census data don’t match the death data, we define
another function PxA * (t ) to be the number of lives in Country A at time t
aged x nearest birthday. (This function does match the death data.)
Then the central exposed to risk for Country A is:
Ú0 Px (t ) dt
n
E xc,A = A*

Assuming that PxA * (t ) varies linearly between the census dates:
1 È A* 1
E xc,A = P (0) + PxA * (1)˘ +  + ÈPxA * (n - 1) + PxA * (n )˘
2Î x ˚ 2Î ˚
1 n -1
= Â ÈPxA * (k ) + PxA * (k + 1)˘
2 k =0 Î ˚
Assuming that birthdays are uniformly distributed over the calendar year:
1È A
PxA * (t ) = Px -1 (t ) + PxA (t )˘
2Î ˚
for all times t . So:
1 n -1 È A
E xc,A = Â Px -1 (k ) + PxA (k ) + PxA-1 (k + 1) + PxA (k + 1)˘˚
4 k =0 Î
Exposed to risk for Country B
Let PxB (t ) denote the number of lives in Country B at time t aged x last
birthday. The values of PxB (0) , PxB (1) , ..., PxB (n ) are known for all x and
the central exposed to risk for Country B is given by:
Ú0 Px (t ) dt
n
E xc,B = B
Assuming that PxB (t ) varies linearly between the census dates, we have:
1 n -1 È B
E xc,B = Â Px (k ) + PxB (k + 1)˘˚
2 k =0 Î
Total exposed to risk
The combined central exposed to risk for the unified state is then:
E xc = E xc,A + E xc,B

Estimate of the force of mortality
Assuming that the force of mortality is constant over the rate interval
‘ x nearest birthday’, it can be estimated by:
dx
mˆ x =
E xc
The rate interval starts at exact age x - ½ and ends at exact age x + ½ , so
the average age of the rate interval is x . Hence mˆ x is an estimate of m x .
(i) Principle of correspondence
The principle of correspondence states that a particular day should be

included in the exposed to risk for an individual life if and only if, were the
person to die on that day, they would be included in the corresponding death
count.
(ii) Difficulties of ensuring the principle of correspondence
The main reason why it might be difficult to ensure that the principle of
correspondence is adhered to is that the data used for the numerator and
the denominator may come from different sources.
For example, at an insurance company, the deaths data might be provided

by the claims department whereas the exposed to risk data might come from
the customer services department. These two departments might use
different computer systems, different databases and different conventions,
such as the age definition adopted.
(iii) Formula
We have been given information about the population numbers. So we can

let Px (0) and Px (1) denote the number of lives aged x last birthday at the
start and end of the calendar year we are considering.
We are also given information about the number of deaths. So we can let
q x denote the number of deaths aged x nearest birthday that occurred
during the calendar year we are considering.

The estimate of m x + f will be calculated as:
qx
mˆ x =
E xc
where E xc is the central exposed to risk corresponding to q x .
We can use the census method, which tells us that, for a one-year study:
1
E xc = *
Ú0 Px (t )dt
where Px* (t ) is the number of lives at time t who are aged x nearest
birthday (ie according to the age definition used in q x for the deaths).
If we assume that the population numbers vary linearly during the calendar
year we are considering, we can approximate this integral using the
trapezium rule:
E xc ª 1 ÈP * (0) + P * (1)˘
2 Î x x ˚
Px* (0) is a count of the people who are aged x nearest birthday at the start
of the year. Since we are not given this information directly, we will need to
approximate it. If a person is aged x nearest birthday on a particular date,
then they must either be x - 1 last birthday or x last birthday. If we assume
that birthdays for these lives are spread uniformly over the calendar year in
question, there will be half of each type. So:
Px* (0) ª 1
2
ÈÎPx -1(0) + Px (0)˘˚
Using similar logic:
Px* (1) ª 1
2
ÈÎPx -1(1) + Px (1)˘˚
So, using the census method, we can approximate E xc as:
E xc ª 1
2 { 1
2
ÈÎPx -1(0) + Px (0)˘˚ + 1
2
ÈÎPx -1(1) + Px (1)˘˚}
= 1
4 {Px -1(0) + Px (0) + Px -1(1) + Px (1)}

We also need to determine the value of the age adjustment f by

considering the rate interval.
Here the rate interval is the year when a life is aged x nearest birthday.
This runs from exact age x - 21 to exact age x + 21 . We are assuming that
the force of mortality is constant over the year. So the age to which the
estimate relates will be the average age in the middle of the rate interval,
which is x . So f = 0 .
So the final formula is:
qx
mˆ x =
E xc
where:
E xc = 1
4 {Px -1(0) + Px (0) + Px -1(1) + Px (1)}
(iv) Factors for subdividing the data
Factors that a government statistical office might use to subdivide the data
include:
 sex
 marital status
 nationality or ethnic group
 employment status.
(i) Reasons for subdividing the data
Mortality rates vary significantly between different categories of lives, eg

males and females.
To reduce this heterogeneity, we can subdivide the data based on factors

such as age, sex, policy type and smoker status, to obtain groups that are
more homogeneous.

Insurance companies and other users of the data may require mortality rates
for specific categories of lives, eg a 50-year-old male smoker with a term
assurance policy.
Subdividing data enables appropriate premiums to be charged for different

individuals, which can help to avoid anti-selection.
Many models of mortality assume that the lives involved all have the same
rates of mortality. Subdividing the data makes this assumption more
reasonable.
(ii) Problems with subdividing the data
In practice it is not possible to make the subdivided groups completely

homogeneous.
Some of the subdivisions may contain little or no data, eg 20-year-olds with

pension policies. Results for these groups will therefore be unreliable.
It may not be possible to collect accurate information (eg because people do

not give honest answers or it is prohibited by law to ask them) on certain
characteristics that may affect mortality such as ethnic origin, results of
genetic tests and sex.
To ensure correspondence, it is important that, as far as possible, the same

subdivisions are used for recording deaths and exposure. This could be a
problem if, for example, somebody gives up smoking during the study.
(i) Factors by which life insurance mortality statistics are often subdivided
Factors include:
 sex
 policy type
 duration (ie how long ago the policyholder took out their policy)
 ‘first class’ versus ‘impaired’ lives (ie whether the policyholder has any
significant known health issues).

(ii) Central death rates for calendar year 2009
The central rate of mortality for the rate interval labelled x is estimated
as q x E xc , where q x denotes the number of deaths during the investigation
of lives aged x last birthday.
The period of investigation is one year. So:
1
E xc = Ú0 Px (t )dt
where Px (t ) is the number of policyholders aged x last birthday at time t

and time is measured in years from 1 January 2009.
If we assume that the population numbers varied linearly over the calendar
year, then using the census method:
E xc = 1
2 ÎÈPx (0) + Px (1)˚˘
For Gasperton, we have:
q x = 25
E xc = 1 (2,000 + 2,100)
2
= 2,050
ˆ Gasp =
qx 25
ﬁ m = = 0.0122
E xc 2,050
Similarly, for Great Hawking:
q x = 21
E xc = 1 (1,770 + 1,674)
2
= 1,722
ˆ GHawk =
qx 21
ﬁ m = = 0.0122
E xc 1,722
So the estimated central rate of mortality is 0.0122 for each town.

(iii) Central death rates for smokers and non-smokers
S
Let mGasp and m NS
Gasp be the central mortality rates for smokers and
non-smokers in Gasperton respectively. 50% of the policyholders in
Gasperton are smokers and 50% are non-smokers. So:
0.5 mSGasp + 0.5 mNS

Gasp = 0.0112
But mSGasp = 1.4m NS

Gasp , assuming the same relationship as applies
nationally. So:
0.5 ¥ 1.4mNS NS
Gasp + 0.5 m Gasp = 0.0112
and hence:
0.0112
mNS
Gasp = = 0.0106
1.2
mSGasp = 1.4 ¥ 0.0106 = 0.0142
Similarly:
0.2 mSGHawk + 0.8 mNS

GHawk = 0.0112
and again assuming the same relationship as applies nationally,

mSGHawk = 1.4mNS
GHawk .
So:
NS NS
0.2 ¥ 1.4 mGHawk + 0.8 mGHawk = 0.0112
and hence:
0.0112
mNS
GHawk = = 0.0113
1.08
S
mGHawk = 1.4 ¥ 0.0113 = 0.0158

(iv) Comment
The insurer currently charges the same premiums for both towns. However,
our crude estimates indicate that the underlying rates for both smokers and
non-smokers are higher for Great Hawking. This would suggest that the
company might be under-charging policyholders in Great Hawking and
over-charging them in Gasperton.
If the company does not differentiate on the basis of geographical area in its
prices, it may lose business in Gasperton to a rival company that does
differentiate. Conversely, in Great Hawking it may attract new business from
rival companies, but will be under-charging and hence may risk its life
assurance fund becoming insolvent.
Our calculations also assume that the national differential of 40% in mortality
rates for smokers applies in both Gasperton and Great Hawking, whereas
the differentials may actually be different.
The insurer currently charges 40% higher premiums for smokers. However,
the theoretical premiums for a 25-year term assurance policy are not directly
proportional to the mortality rate for a single age. So a 40% higher mortality
rate does not imply a 40% higher premium rate. (This is because the
calculation of the premium is affected by other factors such as interest rates
and expenses.)
The data we have been given only allows us to estimate the crude rates for
a single age, whereas the premium calculation will depend on the mortality
rates over a 25-year age range.
(i) The principle of correspondence as it applies to mortality rates
The principle of correspondence states that a particular day should be

included in the exposed to risk for an individual life if and only if, were the
person to die on that day, they would be included in the corresponding death
count.

(ii) Force of mortality at age 50
The estimate of the force of mortality m will be calculated as:
qx
mˆ x + f =
E xc
where E xc is the central exposed to risk corresponding to q x .
We have been given information about the population numbers. So we can

let Px (t ) denote the number of lives aged x last birthday at time t .
We are also given information about the number of deaths. So we can let
q x (t ) denote the number of deaths aged x next birthday in calendar year t
and Px* (t ) denote the number of lives aged x next birthday at time t .
So:
Px* (t ) = Px -1(t )
We can use the census method, which tells us that, for a two-year study:
2
E xc = *
Ú0 Px (t )dt
We can rewrite this as:
2
E xc = *
Ú0 Px (t )dt
1 * 1.5 2
= Ú0 Px (t )dt + Ú1 Px* (t )dt + Ú P * (t )dt
1.5 x
Using the trapezium rule on each integral, we have:
1È * 1 1 1 1
E xc ª Px (0) + Px* (1)˘ + ¥ ÈPx* (1) + Px* (1.5)˘ + ¥ ÈPx* (1.5) + Px* (2)˘
2Î ˚ 2 2Î ˚ 2 2Î ˚
Here we are assuming that the population varies linearly between census
dates.

Simplifying and using the fact that Px* (t ) = Px -1(t ) , this becomes:
1 * 3 1 1
E xc ª P (0) + Px* (1) + Px* (1.5) + Px* (2)
2 x 4 2 4
1 3 1 1
= Px -1(0) + Px -1(1) + Px -1(1.5) + Px -1(2)
2 4 2 4
From the data in the question, we get:
q 50 (2009) + q 50 (2010)
mˆ 50 + f =
1 3 1 1
P (0) + P49 (1) + P49 (1.5) + P49 (2)
2 49 4 2 4
200 + 225
=
1 3 1 1
¥ 2,000 + ¥ 2,100 + ¥ 2,300 + ¥ 2,500
2 4 2 4
425
= = 0.097701
4,350
(iii) Exact age to which this relates
The force of mortality m corresponds to the age in the middle of the rate
interval. q x denotes the number of deaths aged x next birthday. So, any
life dying must be between the ages of x - 1 and x and the rate interval
is ÈÎ x - 1, x ) . So the age to which our estimate of the force of mortality m
applies is 49.5.
(i) Estimate the death rate at age 65
Let Px* (t ) denote the number of inhabitants aged x nearest birthday at

time t , measured in years from the President’s birthday in 2006.
The central exposed to risks for age x for each of the two periods are then:
3 4
E xc (1) = *
Ú0 Px (t )dt and E xc (2) = *
Ú3 Px (t )dt

If we now assume that the population numbers vary linearly between the
census dates, we can apply the trapezium rule to approximate the integrals:
E xc (1) ª 3 ¥ 1
2 {P (0) + P (3)}
*
x
*
x and E xc (2) ª 1
2 {P (3) + P (4)}
*
x
*
x
Let Px (t ) denote the number of inhabitants aged x last birthday at time t ,

measured in years from the President’s birthday in 2006.
If we assume that birthdays are distributed uniformly over each calendar

year, then, of the Px* (t ) inhabitants aged x nearest birthday, half will be
aged x - 1 last birthday and half will be aged x last birthday.
So:
Px* (t ) = 1
2 {Px -1(t ) + Px (t )}
The census formulae then become:
E xc (1) ª 3 ¥ 1
2 { 1
2 (Px -1(0) + Px (0)) + 21 (Px -1(3) + Px (3))}
and:
E xc (2) ª 1
2 { 1
2 (Px -1(3) + Px (3)) + 21 (Px -1(4) + Px (4))}
Substituting the numerical values (expressed in thousands) when x = 65
gives:
c
E65 (1) ª 3 ¥ 1
2 {
(P64 (0) + P65 (0)) + 21 (P64 (3) + P65 (3))}
1
2
= 3 ¥ 21 21 (300 + 290) + 21 (320 + 310)}

{
= 3 ¥ 21 {295 + 315}
= 915

Similarly:
c
E65 (2) ª 1
2 {
(P64 (3) + P65 (3)) + 21 (P64 (4) + P65 (4))}
1
2
= 21 21 (320 + 310) + 21 (350 + 330)}

{
= 21 {315 + 340}
= 327.5
The corresponding estimates of the force of mortality are then obtained by

dividing the observed numbers of deaths (again expressed in thousands) by
the corresponding exposed to risk figures:
q 65 (1) 3
mˆ65 (1) = c
= = 0.00328
E65 (1) 915
and:
q 65 (2) 1
mˆ65 (2) = c
= = 0.00305
E65 (2) 327.5
This assumes that the force of mortality is constant over each year of age.
(ii) Age to which the estimates apply
These estimates have been derived for the rate interval labelled as 65,
which is the period when the lives were aged 65 nearest birthday. This runs
from exact age 64½ to exact age 65½. In the middle of this period the lives
were aged exactly 65. So the estimate of the force of mortality applies to
exact age 65.
(i) Why subdivide mortality data
In human populations mortality risk varies between individuals for many

reasons.
However, mortality models assume that we are dealing with groups of

people who have the same mortality characteristics (‘identical lives’).

We can subdivide the data in order to try to achieve these homogeneous

groups.
This will reduce much of the heterogeneity present.
(ii) One problem
One problem with subdividing data is that some of the subgroups may be
very small, containing only a few individuals.
Estimates of mortality rates derived from the small groups will be unreliable,
as it will be difficult to pin down the true underlying rates with any certainty.
The other main problem is incomplete data. You can discuss this instead
here.
(iii) Factors for subdividing the data
Factors that are commonly used to subdivide mortality data include:

 sex
 age
 smoker / non-smoker
 occupation / employment status.
Mortality rates vary significantly between different categories of lives,

eg males and females.

believed to have a significant effect on mortality, such as age, sex, policy
type and smoker status, to obtain groups that are more homogeneous.
assurance policy.


reasonable.
(ii) Four factors
Apart from age and sex, mortality statistics are often subdivided by:
 smoker status
 policy type (for an insurance company)
 known medical conditions
 location / postcode.
The principle of correspondence states that a life should be included in the

exposed to risk at age x at any given time if and only if, were they to die at
that time, they would be counted as a death at age x .
(ii) Estimate the death rate at age 45 last birthday
To estimate the death rate at age 45 last birthday for the calendar year 2012
based on the two countries combined, we need to divide the total number of
deaths aged 45 last birthday by the total central exposed to risk at age 45
last birthday.
The total number of deaths aged 45 last birthday is the same as the total
number of deaths aged 46 next birthday, which we are told is 4,800 for
calendar year 2012.

Country A
The central exposed to risk at age 45 last birthday for Country A for the
calendar year 2012 is:
c 1 A
E45 ( A) = Ú0 P45 (t )dt
A
where P45 (t ) denotes the number of lives in Country A aged 45 last birthday
t years after 1 January 2012. This corresponds to the shaded area in the
graph below.
To approximate this area, we can divide the shaded area into two
trapeziums:
 one covering the period from 1 Jan 12 to 1 Feb 12, which is of length 1
month
 one covering the period from 1 Feb 12 to 1 Jan 13, which is of length
11 months.
So:
1
c A 1 A
E45 ( A) = Ú012 P45 (t )dt + Ú 121 P45 (t )dt
Country A
1 Jan 13 ?
1 Jan 12 ?
385,000
381,000
374,000
1 Feb 11 1 Feb 12 1 Feb 13

t=0 t=1

Assuming that the population numbers vary linearly between the census
dates, we can approximate these integrals using the trapezium rule to get:
c
E45 ( A) ª 1
12
¥ 1
2 (P A A 1
( )) + 1211 ¥ 21 (P45A ( 121 ) + P45A (1))
45 (0) + P45 12
A
We can find the value of P45 ( 121 ) , ie at 1 Feb 12, directly from the table:
A
P45 ( 121 ) = 381,000
A A
To find the values of P45 (0) and P45 (1) , ie the January figures, we need to
interpolate between the February figures:
A 1 11 ¥ 381,000 = 380,417
P45 (0) ª 12
¥ 374,000 + 12
A 1 11 ¥ 385,000 = 384,667
P45 (1) ª 12
¥ 381,000 + 12
We then have:
c
E45 ( A) ª 1
12
¥ 1
2 (380, 417 + 381, 000) + 12 ¥ 2 (381, 000 + 384, 667)
11 1
= 382, 656
Country B
To match up the age definition for Country B, we can first find

approximations for the population numbers based on age 45 last birthday,
assuming that birthdays are uniform over the calendar year. Since a person
aged 45 last birthday must be either 45 nearest birthday (if they are in the
range (45,45½) ) or 46 nearest birthday (if they are in the range (45½,46) ),
we can average the rows for 45 and 46 to get the required approximations
for age 45 last birthday:
1 Aug 2011: 1
2(374,000 + 354,000) = 364,000
2( )
1 Aug 2012: 1 381,000 + 372,000 = 376,500
2( )
1 Aug 2013: 1 385,000 + 375,000 = 380,000

The central exposed to risk at age 45 last birthday for Country B for the
c 1 B
E45 (B) = Ú0 P45 (t )dt
B
where P45 (t ) denotes the number of lives in Country B aged 45 last birthday
t years after 1 January 2012. This corresponds to the shaded area in the
graph below.
Country B (using age last birthday)
1 Jan 13 ?
380,000
1 Jan 12 ? 376,500
364,000
1 Aug 11 1 Aug 12 1 Aug 13

t=0 t=1
Again, assuming that the population numbers vary linearly between the
census dates, we can approximate this as:
7
c B 1 B
E45 ( B) = Ú012 P45 (t )dt + Ú 127 P45 (t )dt
ª 7
12
¥ 1
2 (P
B B 7
( )) + 125 ¥ 21 (P45B ( 127 ) + P45B (1))
45 (0) + P45 12
B
We calculated the value of P45 ( 127 ) , ie at 1 August 2012, above:
B
P45 ( 127 ) = 376,500

B B
interpolate between the August figures:
B 7 5 ¥ 376,500 = 369,208
P45 (0) ª 12
¥ 364,000 + 12
B 7 5 ¥ 380,000 = 377,958
P45 (1) ª 12
¥ 376,500 + 12
We then have:
c
E45 ( B) ª 7
12
¥ 1
2 (369,208 + 376,500) + 125 ¥ 21 (376,500 + 377,958)
= 374,677
So the estimate of the death rate for age 45 last birthday is:
4,800
= 0.006338
382,656 + 374,677
(iii) Age to which the estimate relates
We have estimated a mortality rate for the rate interval when lives were aged
45 last birthday (assuming that the force of mortality is constant over this
period). This period runs from exact age 45 to exact age 46. In the middle
of this period the lives were aged 45½. So the estimate applies to age 45½.
(i) Census approximation
Census approximation is a method of estimating the central exposed to risk

for a population based only on population counts made at certain key dates.
If Px (t ) denotes the number in the population classified as aged x at

time t , the central exposed to risk between censuses taken at times t1 and
t2
t2 is equal to E xc = Út1 Px (t ) dt .
If we assume that the population number varies linearly over this period, this
can be approximated using the trapezium rule to give:
t2
E xc = Út1 Px (t ) dt ª (t2 - t1 ) ¥ 1
2
ÈÎPx (t1 ) + Px (t2 )˘˚

(ii) Calculation of the central exposed to risk
We need to find the central exposed to risk for age 55 last birthday for the
calendar year 2012, ie for the period from 1 January 2012 to 31 December
2012 (which we can assume is the same as 1 January 2013).
Company A
A life aged 55 last birthday must be either 55 or 56 nearest birthday. So, if

we assume that birthdays are uniform over the calendar year, we can
approximate the numbers aged 55 last birthday by averaging the numbers
age 55 and 56 nearest birthday given in the data.
So the number aged 55 last birthday on 1 January 2012 was approximately:
1 (3,390 + 3,020) = 3,205

2
and on 1 January 2013:
1 (3,100 + 2,950) = 3,025

2
We can use the trapezium approximation to find the exposed to risk:
c 1
E55 ( A) ª 2
(3,205 + 3,025) = 3,115
Company B
The central exposed to risk at age 55 last birthday for Company B for the
c 1 B
E55 (B ) = Ú0 P55 (t ) dt
B
where P55 (t ) denotes the number of lives in Company B aged 55 last
birthday t years after 1 January 2012. This corresponds to the shaded area
in the graph below.

1 Jan 13 ?
1 Jan 12 ?
1,440
1,300
1,190
31 Mar 11 31 Mar 12 31 Mar 13

t=0 t=1
This graph is not to scale. It shows the population numbers assuming they
vary linearly between the census dates (31 March).
trapeziums:
 one covering the period from 1 Jan 12 to 31 Mar 12, which is of length 3
months
 one covering the period from 31 Mar 12 to 1 Jan 13, which is of length
9 months.
So:
3
c B 1 B
E55 (B ) = Ú012 P55 (t ) dt + Ú123 P55 (t ) dt
c
E55 (B ) ª 3
12
¥ 1
2 (P
B B 3
55 (0) + P55 12 ( )) + 9
12
¥ 1
2 (P ( ) + P (1))
B 3
55 12
B
55
B
P55 ( ) is the number on 31 Mar 12, which we know is 1,300.
3
12

B
P55 (0) is the number on 1 Jan 12, which we can find by interpolation
between the 31 Mar 11 and 31 Mar 12 figures:
B 1 3
P55 (0) = 4
¥ 1,190 + 4
¥ 1,300 = 1,272.5
B
P55 (1) is the number on 1 Jan 13, which we can find by interpolation
between the 31 Mar 12 and 31 Mar 13 figures:
B 1 3
P55 (1) = 4
¥ 1,300 + 4
¥ 1, 440 = 1, 405
So, finally, we can estimate the exposed to risk for Company B as:
c
E55 (B ) ª 3
12
¥ 1
2 (1,272.5 + 1,300) + 129 ¥ 21 (1,300 + 1, 405) = 1,335.94
Company C
A life aged 55 last birthday on 1 January 2012 was aged 55 last birthday at
the end of the 2011 calendar year (the day before), or equivalently, aged 56
next birthday at the end of the 2011 calendar year.
So the number aged 55 last birthday on 1 January 2012 was 5,950.
Similarly, the number aged 55 last birthday on 1 January 2013 was 5,980.
We can then use the trapezium approximation to find the exposed to risk:
c 1 (5,950 + 5,980)
E55 (C ) ª 2
= 5,965


(ii) Rate of occurrence of heat exhaustion
To estimate the rate of occurrence of heat exhaustion at age 22 last

birthday, we need to divide the total number of occurrences of heat
exhaustion at age 22 last birthday (ie 40 people) by the total central exposed
to risk at age 22 last birthday.
Individuals aged 22 last birthday on 30th June 2013 were born between 1st
July 1990 (23 years earlier) and 30th June 1991 (22 years earlier). So,
assuming that births were uniform over this period, half were born in 1990
and half in 1991.
We can calculate the exposed to risk by using a census approach based on

the formula:
c 4
E22 = Ú0 P22 (t ) dt
where P22 (t ) is the number of people in the nightclub aged 22 last

birthday t hours after 10:00pm.
From the data provided, we have:
P22 (0) = 0
1
P22 (1.5) = 2
(200 + 150) = 175
P22 (2) = P22 (1.5) + 1 (400 + 400) = 575

2
P22 (3) = P22 (4) = P22 (2) + 21 (350 + 300) = 900

To approximate this integral, we need to calculate the areas of the four

trapeziums in the graph below.
900
575
175
10:00 11:30 12:00 1:00 2:00

t=0 t = 1.5 t=2 t=3 t=4
c 1.5 2 3 4
E22 = Ú0 P22 (t ) dt + Ú P (t ) dt
1.5 22
+ Ú P22 (t ) dt + Ú P22 (t ) dt
2 3
times, ie that people enter the nightclub uniformly over each period, we can
approximate these integrals using the trapezium rule to get:
c
E22 ª 1.5 ¥ 1
2 (P22 (0) + P22 (1.5) ) + 0.5 ¥ (P (1.5) + P (2))
1
2 22 22
+1 ¥ 1
2 (P (2) + P (3)) + 1¥ (P (3) + P (4))
22 22
1
2 22 22
= 1.5 ¥ 1 (0 + 175) + 0.5 ¥ 1 (175 + 575)

2 2
1 1
+1 ¥ 2
(575 + 900) + 1 ¥ 2
(900 + 900)
= 1.5 ¥ 87.5 + 0.5 ¥ 375 + 1 ¥ 737.5 + 1 ¥ 900

= 1, 956.25

Alternatively, we could calculate the exposed to risk by considering the

average exposure time for each of the three cohorts of people entering the
nightclub.
Average Average Number in cohort

Cohort entry exposure aged 22 last Exposure
time time birthday
1 (200 + 150) = 175
1 10:45 3.25 2 568.75
1 (400 + 400) = 400
2 11:45 2.25 2 900
1 (350 + 300) = 325
3 12:30 1.5 2 487.5
Total 1,956.25
This method assumes that the entry times for the individuals within each
cohort are distributed uniformly over the relevant time period.
The estimated rate of occurrence of heat exhaustion at age 22 last birthday

is then:
40
= 0.02045 per hour
1, 956.25
Mortality rates vary significantly between different categories of lives,

eg males and females.

believed to have a significant effect on mortality, such as age, sex, policy
type and smoker status, to obtain groups that are more homogeneous.
assurance policy.


reasonable.
(ii) Four factors

 age
 sex
 postcode
 policy type.
(iii) Likely implications
If the company opts to charge a premium that is intermediate between the

rates for smokers and non-smokers, then smokers will be able to obtain
cheaper premium rates with the company, whereas the rates for
non-smokers will be higher.
This will result in the company attracting more smokers and fewer
non-smokers. The company will tend to make losses on the smokers and
lose the opportunity to make profits on the non-smokers. This could
eventually lead to bankruptcy.
If the company opts to apply the higher smoker rates to everyone, then
smokers will be paying a premium that is consistent with the market,
whereas the non-smoker rates will be uncompetitive.
As a result, there will be fewer policies sold to non-smokers, but policies sold
to smokers should be unaffected.

Apart from age and sex, mortality statistics are often subdivided by:
 smoker status
 policy type (for an insurance company)
 known medical conditions
 location / postcode.
(i) Estimate the force of mortality for age 52 last birthday
To estimate the death rate at age 52 last birthday for the calendar year 2013
based on the two companies combined, we need to divide the total number
of deaths aged 52 last birthday (ie 28 + 17 = 45 ) by the total central exposed
to risk for age 52 last birthday. This assumes that the force of mortality is
constant over the year of age.
Company A
The central exposed to risk at age 52 last birthday for Company A for the
c 1 A
E52 ( A) = Ú0 P52 (t ) dt
A
where P52 (t ) denotes the number of Company A policyholders aged 52 last
birthday t years after 1 January 2013.
dates (ie between 1 Jan 2013 and 1 Jan 2014), we can approximate this
integral using the trapezium rule to get:
c
E52 ( A) ª 1
2 (PA A
52 (0) + P52 (1))
The population numbers for Company A are recorded using age next
birthday. We can find the corresponding numbers age last birthday by
noting that age x last birthday is equivalent to age x + 1 next birthday.

From the table provided:
A A
P52 (0) = 8,016 and P52 (1) = 9,026
So:
c
E52 ( A) ª 1
2 (8,016 + 9,026) = 8,521
Company B
The central exposed to risk at age 52 last birthday for Company B for the
c 1 B
E52 (B ) = Ú0 P52 (t ) dt
B
where P52 (t ) denotes the number of Company B policyholders aged 52 last
birthday t years after 1 January 2013. This corresponds to the shaded area
in the graph below.
Company B
5,281 1 Jan 13 ?
5,218
1 Jan 14 ?
3,812
1 Apr 12 1 Apr 13 1 Apr 14

t=0 t=1

trapeziums:
 one covering the period from 1 Jan 13 to 1 Apr 13, which is of length
3 months
 one covering the period from 1 Apr 13 to 1 Jan 14, which is of length
9 months.
So:
3
c B 1 B
E52 (B ) = Ú012 P52 (t ) dt + Ú123 P52 (t ) dt
c
E52 (B ) ª 3
12
¥ 1
2 (PB B 3
( )) +
52 (0) + P52 12
9
12
¥ 1
2 (P ( ) + P (1))
B 3
52 12
B
52
B B
interpolate between the April figures:
B 3 9
P52 (0) ª 12
¥ 5,281 + 12 ¥ 5,218 = 5,233.75
B 3 9
P52 (1) ª 12
¥ 5,218 + 12 ¥ 3,812 = 4,163.5
We then have:
c
E52 (B ) ª 3
12
¥ 1
2 (5,233.75 + 5,218) + 129 ¥ 21 (5,218 + 4,163.5) = 4,824.53
So the estimated combined force of mortality is:
28 + 17 45
= = 0.00337
8,521 + 4,824.53 13,345.53
(ii) Age to which the estimate relates
We have estimated a mortality rate for the rate interval when lives were aged
52 last birthday (assuming that the force of mortality is constant over this
period). This period runs from exact age 52 to exact age 53. In the middle
of this period the lives were aged 52½. So the estimate applies to age 52½.

For each individual in the study, we need:

 date of birth
 date of entry
 date of exit.

(ii)(a) Expression for the central exposed to risk
We can calculate the central exposed to risk by using a census approach

based on the formula:
1
E xc = *
Ú0 Px (s ) ds
Here Px* (s ) is the number of in-force policies where the policyholder was
aged x last birthday at time s years after the start of year t .
To approximate this integral, we can apply the trapezium rule, which

assumes that the population numbers vary linearly between the census
times, ie the number of in-force policies where the policyholder was aged x
last birthday varies linearly over year t . (Assumption 1)
Under this assumption:
1È *
E xc = Px (0) + Px* (1)˘
2Î ˚
However, the population numbers provided in the data are based on age
nearest birthday. So we need to express P * in terms of P .
We assume that the birthdays of the individuals in the study are distributed
uniformly over the calendar year. (Assumption 2)

We would then expect that:
Px* (s ) =
1
2
{
Px (t + s ) + Px +1(t + s )}
To help get the ages in this relationship correct, you can imagine that we
want to know how many people are aged 25 (say) last birthday. Such
people must be either 25 or 26 nearest birthday and, under Assumption 2,
this would be an equal split.
So:
*
P25 () =
1
2
{
P25 () + P26 () }
For the times, note that our notation Px* (0) refers to time 0 after the start of
year t , ie 1st January in year t . This is the same time point as in the
symbol Px (t ) given in the question. Similarly, Px* (1) refers to time 1 after the
start of year t , ie 31st December in year t , or 1st January in year t + 1 .
This is the same time point as in the symbol Px (t + 1) given in the question.
So the central exposed to risk is given by:
E xc =
1 È1
2 ÍÎ 2
{
Px (t ) + Px +1(t ) +} {
1
2
} ˘
Px (t + 1) + Px +1(t + 1) ˙
˚
1È
= P (t ) + Px +1(t ) + Px (t + 1) + Px +1(t + 1)˘
4Î x ˚
(ii)(b) Assumptions
This formula depends on the two assumptions indicated above. If the

assumptions do not hold, then the formula is at best an approximation.

(i) Central exposed to risk for age 51 last birthday
Company A
We can approximate the number of policies for lives aged 51 last birthday on
1 January 2014 as:
1
2 (PA A
51(2014) + P52 (2014) )= 1
2 (6,002 + 5,600) = 5,801
1 January 2015 as:
1
2 (PA A
51(2015) + P52 (2015) )= 1
2 (5,056 + 4,906) = 4,981
So the contribution to the exposed to risk for lives aged 51 last birthday for
the calendar year 2014 for Company A is:
c
E51( A) ª 1
2 (5,801 + 4,981) = 5,391
Company B
1 January 2014 as:
10 B 2 B 10 2
12
¥ P51(2013) + 12 ¥ P51(2014) = 12
¥ 2,333 + 12 ¥ 2, 417 = 2,347
1 January 2015 as:
10 B 2 ¥ P B (2015) = 10 2 ¥ 2,213 = 2,383

12
¥ P51(2014) + 12 51 12
¥ 2, 417 + 12
the calendar year 2014 for Company B is:
c
E51(B ) = 10
12
¥ 1
2 (2,347 + 2, 417) + 122 ¥ 21 (2, 417 + 2,383) = 2,385

Company C
Lives aged 51 last birthday on 1 January 2014 were aged 52 next birthday
on that date (or equivalently on 31 December 2013).
So the number of policies for lives aged 51 last birthday on 1 January 2014
is:
C
P52 (2013) = 3,895
Lives aged 51 last birthday on 1 January 2015 were aged 52 next birthday
on that date (or equivalently on 31 December 2014).
This figure is not provided in the data, but we can approximate it as:
1
2 (P
C C
52 (2013) + P52 (2015) )= 1
2 (3,895 + 4,367) = 4,131
the calendar year 2014 for Company C is:
c
E52 (C ) ª 1
2 (3,895 + 4,131) = 4,013
(ii)(a) Assumptions
Company A
For Company A we assumed that birthdays were distributed uniformly over

the calendar year so that the number of lives aged x last birthday can be
approximated as the average of the numbers aged x and x + 1 nearest
birthday.
We have also assumed that the number of policies in force varies linearly
over the calendar year 2014.
Company B
We have assumed that the number of policies in force varies linearly over
each year.
We have also assumed that each calendar month is of equal length (or more
specifically that 1 November is 10/12ths of the way through the calendar
year).

Company C
We have assumed that the number of policyholders on 31 December is the

same as the number the day after (on 1 January).
We have also assumed that the number of policies in force varies linearly
over the period from 1 January 2013 to 31 December 2014.
(ii)(b) Why these assumptions are required
The assumptions relating to the ages are needed to ensure that the exposed
to risk is based on the same age definition as is used for recording the
deaths.
The assumptions relating to linearity are required to allow us to apply the

trapezium rule to approximate the integral defining the central exposed to
risk.
The linearity assumption also allows us to approximate the figures at a

different date in the year for Company B.
A stronger linearity assumption is required for Company C to allow us to

approximate the missing data for the middle year.
The assumption of equivalence of 31 December and 1 January for

Company C is needed to allow correctly for the year end.

(i) Transition diagram
The transition diagram looks like this:

H S

 
The states used here are:

State H : Employees who are able to work (= Healthy)
State S : Employees who are unable to work due to ill-health (= Sick)
State D : Employees who are dead.
(ii) Likelihood
If ti denotes the observed waiting time in state i and nij denotes the
observed number of transitions from state i to state j , the likelihood is:
L( m,s ,n , r ) = e -( m +s )tH e -(n + r )tS m nHD s nHS n nSD r nSH ¥ C
where C is a constant.
(iii) Maximum likelihood estimator
To derive the maximum likelihood estimator for s , we start from the

likelihood and take logs to find the log-likelihood:
ln L = -( m + s )tH - (n + r )tS + nHD ln m + nHS ln s + nSD lnn

+ nSH ln r + ln C

Differentiating with respect to the parameter we wish to estimate gives:
∂ n
ln L = -tH + HS
∂s s
Equating this to zero and rearranging gives:
nHS
s =
tH
We can check this is consistent with a maximum by evaluating the second

derivative:
∂2 nHS
2
ln L = - <0
∂s s2
So this is indeed a maximum.

nHS
Hence the maximum likelihood estimate of s is sˆ = . The maximum
tH
NHS
likelihood estimator of s is s = , where NHS is the random variable
TH
representing the number of transitions from State H to State S and TH is the
random variable representing the total waiting time in State H.
(iv) Exposed to risk
We can calculate the central exposed to risk using a census approach based
on the formula:
2
E xc = *
Ú0 Px (t ) dt
where Px* (t ) is the number of employees aged x nearest birthday at time t

years after 1 January 2014.
To approximate this integral, we can apply the trapezium rule, which

assumes that the population numbers vary linearly between the census
dates, ie the number of employees aged x nearest birthday varies linearly
over each calendar year.

Under this assumption:
1È * 1
E xc = P (0) + Px* (1)˘ + ÈPx* (1) + Px* (2)˘ = 1 *
P (0) + Px* (1) + 21 Px* (2)
2Î x ˚ 2Î ˚ 2 x
However, the population numbers provided in the data are based on age last
birthday. So we need to express P * in terms of Px (t ) , the number of
employees at time t who are aged x last birthday.
If we assume that the birthdays of the individuals in the study are distributed
uniformly over the calendar year, we would then have:
Px* (t ) = 1
2 { Px -1(t ) + Px (t )}
So:
E xc = 1
2
¥ {P1
2 x -1(0) + Px (0) } + {P 1
2 x -1(1) + Px (1) }
+ 21 ¥ {P 1
2 x -1 (2) + P (2)}
x
When x = 52 , the exposed to risk for all employees is:
c
E52 (all ) = 1
2
¥ {P (0) + P (0)} + {P
1
2 51 52
1
2 51(1) + P52 (1) }
+ 21 ¥ {P (2) + P (2)}
1
2 51 52
1
= 2
¥ 21 (148 + 146) + 21 (162 + 148) + 1
2
¥ 21 (180 + 160)
= 1 ¥ 147 + 155 + 1 ¥ 170

2 2
= 313.5

The denominator in the calculation of the recovery rate is the exposed to risk
for employees receiving sickness benefit. We can calculate this by just
considering the sick employees:
c
E52 (sick ) = 1
2
¥ {P (0) + P (0)} + {P
1
2
S
51
S
52
1
2
S S
51(1) + P52 (1) }
+ 1
2
¥ {P (2) + P (2)}
1
2
S
51
S
52
1 1
= 2
¥ 2
(12 + 10) + 21 (20 + 18) + 1
2
¥ 1
2
(8 + 7)
1
= 2
¥ 11 + 19 + 21 ¥ 7.5
= 28.25
The denominator in the calculation of the sickness rate is the exposed to risk
for healthy employees who are not receiving sickness benefit. By
subtraction, this is:
c c c
E52 ( healthy ) = E52 (all ) - E52 (sick ) = 313.5 - 28.25 = 285.25

(i) Central exposed to risk
The dates shown in bold in the table below are the dates on which we
started and finished observing each life at age 85 last birthday during the
observation period. The number of months exposed to risk is then
calculated by subtraction.
Start Date End Date

85th Start of 86th End of
Life Death
birthday study birthday study
1 1 Aug 14 1 Apr 15 – 1 Aug 15 1 Apr 16
2 1 Nov 14 1 Apr 15 – 1 Nov 15 1 Apr 16
3 1 Jan 15 1 Apr 15 1 Feb 16 1 Jan 16 1 Apr 16
4 1 Feb 15 1 Apr 15 – 1 Feb 16 1 Apr 16
5 1 Mar 15 1 Apr 15 – 1 Mar 16 1 Apr 16
6 1 Apr 15 1 Jan 16 1 Apr 16 1 Apr 16
7 1 Jun 15 1 Nov 15 1 Jun 16 1 Apr 16
8 1 Jul 15 – 1 Jul 16 1 Apr 16
9 1 Sep 15 1 Mar 16 1 Sep 16 1 Apr 16
10 1 Jan 16 – 1 Jan 17 1 Apr 16
Total
So the number of months contributed to the exposed to risk by each life is:
4, 7, 9, 10, 11, 9, 5, 9, 6, 3
Summing these, we find that the central exposed to risk is 73 months or

1 years.
6 12
(ii) Maximum likelihood estimate
The number of deaths at age 85 last birthday in the sample is 3 (ie Lives 6, 7
and 9).

So the MLE of the force of mortality at age 85 last birthday is:
d85 3
mˆ = c
= 1
= 0.49315
E85 6 12
(iii) Estimate of q85
This is calculated using the estimated force of mortality as follows:
qˆ85 = 1 - e - m = 1 - e -0.49315 = 0.38930

ˆ

FACTSHEET
This factsheet summarises the main methods, formulae and information

required for tackling questions on this topic.
How to calculate E cx using the exact method
For each life in the investigation:

 use the date of entry and date of birth data to work out the date at which
the life was first observed between the x th and ( x + 1) th birthdays
 use the date of exit and date of birth data to work out the date at which
the life was last observed between the x th and ( x + 1) th birthdays
 calculate the length of time between the two above dates (expressed in
years), to give the individual’s contribution to E xc .
Finally sum all contributions over all lives in the investigation: this is E xc .
How to obtain the exact integral census formula for E cx
 define a variable Px, t that corresponds exactly (at time t) to the age
definition of deaths
 write down the integral of Px, t over [a, b] , where a and b are the start
and end dates of the investigation period respectively
b
ie E xc = Úa Px, t dt

How to obtain a census formula for E cx using the trapezium

approximation
 split the investigation period into suitable time intervals (eg intervals
between successive census dates)
 for each interval, use the trapezium rule approximation to calculate the
value of the integral, ie:
[average value of Px, t over the interval] ¥ [length of interval]
 match the census data available with those required to ensure

correspondence
 sum over all intervals in the investigation period to obtain E xc .
Rate intervals
A rate interval is a period of one year during which a person has a particular
age label. They are determined by the age definition of the deaths.
 ‘x last birthday’ means ‘year starting at the xth birthday’.
 ‘x nearest birthday’ means ‘year with the xth birthday in the middle’.
 ‘x next birthday’ means ‘year ending at the xth birthday’.
How to work out the age for a m -type rate
 work out the rate interval that is implied by the definition of deaths
 work out the age in the middle of the rate interval.
How to work out the age for a q -type rate
 work out the rate interval that is implied by the definition of deaths
 work out the age at the start of the rate interval.

NOTES

NOTES

NOTES

NOTES

NOTES

NOTES

CK1 Booklet 1 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CK1 Booklet 1 PDF

Uploaded by

Copyright:

Available Formats

Exclusive use Batch 3a

Chapter 9 Exposed to risk

The Actuarial Education Company

Links to the Course Notes and Syllabus 2

All of this material is copyright. The copyright belongs to Institute and

© IFE: 2019 Examinations Page 1

LINKS TO THE COURSE NOTES AND SYLLABUS

Material covered in this booklet

Chapter 9 Exposed to risk

Syllabus objectives covered in this booklet

4.4 Estimate transition intensities dependent on age (exact or census).

4.4.1 Explain the importance of dividing the data into

4.4.2 Describe the principle of correspondence and explain its

4.4.3 Specify the data needed for the exact calculation of a

4.4.4 Calculate a central exposed to risk given the data in 4.4.3.

4.4.5 Explain how to obtain estimates of transition probabilities.

4.4.6 Explain the assumptions underlying the census

4.4.7 Explain the concept of the rate interval.

4.4.8 Develop census formulae given age at birthday where the

4.4.9 Specify the age to which estimates of transition intensities

Page 2 © IFE: 2019 Examinations

 d x (also sometimes written as q x ) is the number of transitions with the

 E xc is the central exposed to risk for the given age definition.

The main issues are:

Questions on exposed to risk are largely preoccupied with one or several of

© IFE: 2019 Examinations Page 3

We have inserted paragraph numbers in some places, such as 1, 2, 3 …, to

The text given in Arial Bold font is Core Reading.

Chapter 9 – Exposed to risk

Calculating the exposed to risk

We have seen how exposed to risk arises in a probabilistic model of

In this booklet we consider some problems of a computational nature,

The central exposed to risk is a very natural quantity, intrinsically

The central exposed to risk carries through unchanged to arbitrarily

Page 4 © IFE: 2019 Examinations

1 The problem of heterogeneity

We can subdivide our data according to characteristics known, from

3 Among the factors in respect of which life insurance mortality statistics

4 Others that might be used are:

© IFE: 2019 Examinations Page 5

5 Two key points are:

(a) Sub-division cannot be carried out unless the relevant information

(b) Even in quite large investigations, sub-division using many factors

The principle of correspondence

Mortality investigations based on estimation of m x + ½ at individual

A precise statement of what we mean by ‘defined consistently’ is given

6 A life alive at time t should be included in the exposure at age x at

This seems almost a triviality, but it is very important and useful.

Page 6 © IFE: 2019 Examinations

Working with complete data

7 The procedure for the exact calculation of E cx is obvious:

(a) record all dates of birth

If we add to the data above the cause of the cessation of observation,

Working with incomplete data

All of the remainder of the Core Reading in this booklet is about

The available data

8 Suppose that we have death data of the form:

d x = total number of deaths age x last birthday during calendar

That is, we have observations over N + 1 calendar years of all deaths

© IFE: 2019 Examinations Page 7

However, instead of the times of entry to and exit from observation of

Px ,t = Number of lives under observation aged x last birthday at

The census approximation to E cx