1

2 INSTITUTE AND FACULTY OF ACTUARIES

3
4
5
6
7
EXAMINATION
8
9 1 October 2015 (am)
10
11 Subject CT4 – Models
12 Core Technical
13
14
Time allowed: Three hours
15
INSTRUCTIONS TO THE CANDIDATE
16
17 1. Enter all the candidate and examination details as requested on the front of your answer
booklet.
18
2. You must not start writing your answers in the booklet until instructed to do so by the
19
supervisor.
20
3. Mark allocations are shown in brackets.
21
22 4. Attempt all 11 questions, beginning your answer to each question on a new page.

23 5. Candidates should show calculations where this is appropriate.

24
25 Graph paper is NOT required for this paper.

26
AT THE END OF THE EXAMINATION
27
28 Hand in BOTH your answer booklet, with any additional sheets firmly attached, and this
question paper.
29
30
In addition to this paper you should have available the 2002 edition of the Formulae
31
and Tables and your own electronic calculator from the approved list.
32
33
34
35 CT4 S2015  Institute and Faculty of Actuaries
1 List four factors, other than age and sex, by which mortality statistics are often
subdivided. [2]

2 Describe the differences between a stochastic and a deterministic model. [4]

3 (i) Define how the following forms of censoring arise in a survival investigation:

 right censoring
 type I censoring
 random censoring [3]

An experience analysis is conducted where the event of interest is the lapse of a term
assurance policy.

(ii) Explain whether each form of censoring listed in part (i) occurs in each of the
following situations. If it is not possible to state whether a form of censoring
occurs, explain why this is the case.

(a) A policyholder dies.

(b) A subset of the policies is migrated to a new administration system and

no data are provided from the new system to the experience analysis
team.

(c) A policy reaches its maturity date. [4]

[Total 7]

CT4 S2015–2
4 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 policies policies
1 Jan. 2012 1 Jan. 2013 1 Jan. 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 policies policies
1 April 2012 1 April 2013 1 April 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]

CT4 S2015–3 PLEASE TURN OVER

5 (i) Describe why models are used in actuarial work. [4]

The following diagrams illustrate sample paths for four stochastic processes.
Sample Path A Sample Path B

State
State

Time Time

Sample Path C Sample Path D

State

State

Time Time

(ii) Identify which sample path is most likely to correspond to a:

 discrete time, discrete state process.

 continuous time, discrete state process.
 discrete time, continuous state process.
 continuous time, continuous state process.
[2]

(iii) Discuss the reasons why a discrete or continuous process may be selected for a
modelling exercise. [3]
[Total 9]

CT4 S2015–4
6 (i) Describe what is meant by a proportional hazards model. [3]

A pharmaceutical company is interested in testing a new treatment for a debilitating

but non-fatal condition in cows. A randomised trial was carried out in which a sample
of cows with the condition was assigned to either the new treatment or the previous
treatment. The event of interest was the recovery of a cow from the condition. The
results were analysed using a Cox regression model.

The final model estimated the hazard, h(t , x) as:

h(t, x) = h0(t) exp(0z + 1x + 2xz),

where:

h0(t) is the baseline hazard;

z is a covariate taking the value 1 if the cow was assigned the new treatment and 0 if
the cow was assigned the previous treatment;

x is a covariate denoting the length of time (in days) for which the cow had been
suffering from the condition when treatment was started;

and t is the number of days since treatment started.

0 , 1 and 2 are parameters. Their estimated values were 0 = 0.8, 1 = 0.4

and 2 = 0.1 .

(ii) Determine the characteristics of the baseline cow. [1]

For a particular cow, the new treatment and the previous treatment have exactly the
same hazard.

(iii) Calculate the number of days for which that cow had the condition before the
initiation of treatment. [2]

Under the previous treatment, cows whose treatment began after they had been
suffering from the condition for three days had a median recovery time of 14 days
once treatment had started.

(iv) Calculate the proportion of these cows which would still have had the
condition after 14 days if they had been given the new treatment. [4]
[Total 10]

CT4 S2015–5 PLEASE TURN OVER

7 A school offers a one year course in a foreign language as an evening class. This is
divided into three terms of 13 weeks each with one lesson per week. At the end of
each lesson all the students sit a test and any that pass are awarded a qualification, and
no longer attend the course.

Last year 33 students started the course. Of these 13 dropped out before completing
the year, and 16 passed the test before the end of the year. The last lesson attended by
the students who did not stay for the whole 39 lessons is shown in the table below
along with their reason for leaving.

Number of Last lesson Reason for

students attended leaving

5 1 Dropped out
1 6 Dropped out
2 7 Passed test
2 13 Dropped out
5 14 Passed test
6 27 Passed test
4 28 Dropped out
1 30 Dropped out
3 36 Passed test

(i) Calculate the Nelson-Aalen estimate of the survival function. [5]

(ii) Sketch a graph of the Nelson-Aalen estimate of the survival function, labelling
the axes. [2]

(iii) Determine the probability that a student who starts the course passes by the
end of the year. [1]

Since only four students had not passed by the end of the year and a total of 16 had
passed, the school claims in its publicity that 80% of students are awarded the
qualification by the end of the year.

(iv) Comment on the school’s claim in light of your answer to part (iii). [2]
[Total 10]

CT4 S2015–6
8 (i) Define a Markov Jump Process. [1]

A company provides phones on contracts under which it is responsible for repairing

or replacing any phones which break down.

When a customer reports a fault with a phone, it is immediately taken to the

company’s repair shop and it is assessed whether it can be fixed (meaning fixable at
reasonable cost). Based on previous experience, it is estimated that the probability of
a phone being fixable is 0.75. If a phone is not fixable it is discarded and the
customer is provided with a new phone.

If a repaired phone breaks again the company, in line with its customer charter, will
not attempt to repair it again, and so discards the phone and replaces it with a new
one.

The status of a phone is to be modelled as a Markov Jump Process with state space
{Never Broken (NB), Repaired (R), Discarded (D)}.

The company considers the rate at which phones break down to vary according to
whether a phone has previously been repaired as follows:

Status Probability of break down in small

interval of time, dt

Never Broken 0.1dt + o(dt)

Repaired 0.2dt + o(dt)

(ii) Draw a transition diagram for the possible transitions between the states,
including the associated transition rates. [2]

Let PNB(t), PR(t) and PD(t) be the probabilities that a phone is in each state after time t
since it was provided as a new phone.

(iii) Determine Kolmogorov’s forward equations in component form for PNB(t),

PR(t) and PD(t). [2]

(iv) Solve the equations in part (iii) to obtain PNB(t) and PR(t). [4]

(v) Calculate the probability that a phone has not been discarded by time t. [1]
[Total 10]

CT4 S2015–7 PLEASE TURN OVER

9 Doctors at a health centre are carrying out an investigation to see if obesity affects the
likelihood of dying from heart disease. They propose to use a model with four states:

1. Obese
2. Not obese
3. Dead due to heart disease, and
4. Dead due to other causes

(i) Write down, defining all the terms you use, the likelihood for the transition
intensities. [3]

(ii) Derive the maximum likelihood estimator of the force of mortality from heart
disease for Obese people. [3]

The investigation has followed several thousand people aged 50–59 years for five
years and has the following data:

Waiting time in state Obese (in person-years) 14,392

Waiting time in state Not obese (in person-years) 18,109
Number of deaths due to heart disease for those persons who are Obese 178
Number of deaths due to heart disease for those persons who are Not obese 190
Number of deaths due to other causes for those persons who are Obese 89
Number of deaths due to other causes for those persons who are Not obese 53

The doctors want to promote healthy living and therefore wish to claim that Obese
people have a much higher chance, statistically, of dying from heart disease than do
people who are Not obese.

(iii) Test whether this claim is true at the 90% confidence level. [5]
[Total 11]

CT4 S2015–8
10 A profession has examination papers in two subjects, A and B, each of which is
marked by a team of examiners. After each examination session, examiners are given
the choice of remaining on the same team, switching to the other team, or taking a
session’s holiday.

In recent sessions, 10% of subject A’s examiners have elected to switch to subject B
and 10% to take a holiday. Subject B is more onerous to mark than subject A, and in
recent sessions, 20% of subject B’s examiners have elected to take a holiday in the
next session, with 20% moving to subject A.

After a session’s holiday, the profession allocates examiners equally between subjects
A and B. No examiner is permitted to take holiday for two consecutive sessions.

(i) Sketch the transition graph for the process. [2]

(ii) Determine the transition matrix for this process. [2]

(iii) Calculate the proportion of the profession’s examiners marking for subjects A
and B in the long run. [4]

The profession considers that in future, an equal number of examiners is likely to be

required for each subject. It proposes to try to ensure this by adjusting the proportion
of those examiners on holiday who, when they return to marking, are allocated to
subjects A and B.

(iv) Calculate the proportion of examiners who, on returning from holiday, should
be allocated to subject B in order to have an equal number of examiners on
each subject in the long run. [4]
[Total 12]

CT4 S2015–9 PLEASE TURN OVER

11 (i) Describe why an insurance company might want to compare the results of a
mortality investigation with previous experience. [2]

A large life insurance company has undertaken an investigation of the mortality of its
policyholders. Currently it assumes that mortality at age x, x, is equal to a standard
table. The company wishes to use the results from the investigation to see whether
the standard table is still appropriate. Below are shown some data from the
investigation.

Age x Number of Actual death Expected death

policies in force claims claims from
standard table

70 1,000 13 23.74
71 1,200 28 31.80
72 1,100 31 32.50
73 1,100 34 36.20
74 1,000 39 36.63
75 1,000 41 40.73
76 950 41 42.99
77 900 40 45.20
78 850 46 47.34
79 800 48 49.35

(ii) Perform an overall test of the hypothesis that the underlying mortality of the
company’s policyholders is, over this range of ages, represented by the
standard table. [6]

(iii) Evaluate the suitability of the standard table for use in the company’s financial
modelling by performing two additional tests for different possible
inconsistencies between the actual death rates and those represented by the
standard table. [6]

The company discovers that at age 70 years, one individual owns 25 of the policies in
the investigation, the remaining policies each being owned by different individuals.

(iv) Assess the impact of this on the variance of the number of claims at age 70
years. [4]
[Total 18]

END OF PAPER

CT4 S2015–10