SMM048 Insurance Risk Modelling

Coursework Description
Instructions
The coursework for SMM048 consists of 4 questions and you are required to answer all of
them. Note that this is a group assignment and that you are expected to work in groups
of no more than 4 members.

For all questions, you are expected to produce analytical solutions and typeset your an-
swers with Microsoft Word or Latex. Answers to analytical questions should be presented
in detail. Hence, you are advised not to miss stages and present a clear exposition of
your derivations and proofs. Any transgressions will result in a reduction of marks.

Deadline: 22 March 2024 at 4pm.

Submission Method: Online via Moodle.

Question 1
(i) A Markov chain X = {Sn : n ≥ 0} on state space S = {1, 2, 3, 4, 5, 6, 7} has
transition matrix
 
0.2 0.5 0 0 0 0.3 0
 
0 0 0 0.8 0 0 0.2
 
0 0 0.2 0 0.4 0 0.4
 
P =  0 0.5 0.5 0 0 0 0 .
 
 
0 0 0.4 0 0.5 0 0.1
 
0.7 0 0 0 0 0.3 0 
 
0 0 0.1 0 0.1 0 0.8
(a) Draw a transition diagram. [2 marks]
(b) Divide the state space S into communicating classes, giving reasons for your
answer. Indicate whether each class is recurrent or transient, periodic or
aperiodic. [4 marks]
(c) Suppose there is a probability vector π = (π1 π2 . . . π7 ) such that π T P = π T .
1. Show that, if X0 is chosen randomly with distribution P(X0 = i) = πi ,
then P(Xn = i) = πi for all n > 0. [3 marks]
2. Is it possible to deduce that P(Xn = j|X0 = i) → πj as n → ∞ for every
fixed starting point i ∈ S? Give reasons for your answer. [3 marks]
(d) Find the probability vector π satisfying π T P = π T . [3 marks]

(ii) A company runs a health insurance scheme for its employees. The level of contri-
butions made by the healthy employees and the level of benefits received by those
who are sick have not been reviewed for several years. The scheme is currently
running at a loss. The company is interested in formulating a stochastic model to
help to determine appropriate levels of contributions and benefits.
The company decides to use a time-homogeneous model with states: H (healthy),
S (sick) and R (retired, resigned or no longer employed by the company). The
generator matrix of the fitted Markov jump chain model, with states in the order
H, S, R and with time measuredin years, is 
−0.12 0.1 0.02
Q =  3.9 −4 0.1 
 

0 0 0
Let pij (t) denote the probability that the process is in state j at time t, conditional
on being in state i at time 0.
(a) Derive the equation for pSS (t) in terms of pHS if the integrated form of the
Kolmogorov Backward Equation for pHS (t) is [2 mark]
Z t
pHS (t) = 0.1e−0.12u pSS (t − u)du
0
(b) Find the probability that an employee is healthy for at least 10 years then
experiences a period of sickness, but leaves the company permanently.
[2 marks]
(c) Whilst in state H, employees pay contributions at rate c per unit time into the
insurance scheme; whilst in state S, employees receive benefits at rate b per
unit time from the scheme. Assuming that you are provided with the solutions
pHH (t), pHS (t), pSH (t) and pSS (t) of the Kolmogorov differential equations for
all t, outline how you would determine values of b and c which would ensure
that the scheme remained solvent in the long term. [3 marks]

[Total: 22 marks]
Question 2
A quality control process in the car manufacturing industry requires that car parts are to
tested in a controlled environment in order to identify whether their lifetime has improved
compared to older versions. It is assumed that, for a given part, the time taken until it
malfunctions can be modelled with a hazard function of the form

µx = 0.00018 × 1.06x

where x is the length of time (in hours) that the part has been tested for.

i. Determine the probability that a new part continuous to function properly after 60
hours of testing. [2 marks]

ii. Calculate the probability that a part that is functioning properly after 20 hours of
testing continuous to function properly after 50 hours of testing. [2 marks]

iii. Calculate the length of time when 50% of new parts are expected to malfunction.
[2 marks]

In reality, one of the hazards of car parts is that they deteriorate depending on the driving
style or the quality of the road in the area where the car is driven.

iv. Suggest a different form of the hazard function that would allow for this real world
hazard i.e. one function that models all the hazard rather than just an additional
function. [2 marks]

It has been suggested that it is not possible to claim the duration of these parts because
it is not possible to replicate how parts perform in the real world. To investigate the
latter, a sample of drivers are asked to record the length of time that the part was used
before it malfunctioned.

A Cox regression model was used to fit the data as detailed below:

hi (t) = h0 (t) × exp {−0.12(xi − 3) + 0.3 × gi − 0.2 × ci }

where:

xi = average time of individual session where the car is driven

gi = 0 if the car was driven only in the city

1 if the car was driven in both the city and rural areas

ci = 0 if the car is serviced at the designated time intervals

1 if the car was not serviced at the designated time intervals

v. State the baseline hazard function. [2 marks]

 vi. What does the model say about the survival function of the following two parts:

– a part where the car is driven only in the city, the average driving session is 2
hours and the car is not properly maintained;
– compared to a part where the car is driven within both the city and rural
areas, the average driving session lasts 4 hours and the car is not maintained.
[2 marks]

The model appears to have a high level of standard error. To further improve the model
it has been suggested that the data should be gathered on the type driving style.

vii. Comment on why the current model may have a high level of standard error and
discuss the reasons for the suggestion to include driving style as a parameter and
the problems that may occur when trying to fit this parameter. [4 marks]

viii. Suggest two other covariates that you could add to the model to try and improve
the model’s predictability. For each covariate explain the reasoning for its inclusion
and any problems that the covariate may cause. [4 marks]

[Total 20 marks]
Question 3
A country is split into five separate districts – North, South, East, West and Central.
The government actuaries department of the country produces mortality tables for the
whole population but in addition each of these districts have their own mortality tables
to help forecast future populations.

These mortality tables are constructed by breaking the population down into four differ-
ent age groups: 0-25; 26-64; 65-88; and 89+

The first three age groups are graduated using graduation by parametric formula with a
different function fitted to each age range. The final group is graduated by reference to
the country’s standard mortality table.

i. Explain why each district needs to calculate its own mortality rates. [2 marks]

ii. Discuss why the districts have used the above method to create their mortality
tables. [2 marks]

The data collected over the last three years relating to 65 to 88 year olds for one of the
districts is given below. To calculate the mortality rates the crude data is graduated
using a function with four parameters.

Age Initial Exposed to Risk Actual Deaths Graduated mortality rates

65 136120 1579 0.012000
66 134462 1792 0.013730
67 132598 2208 0.015832
68 130302 2278 0.017782
69 127933 2552 0.020248
70 125279 2886 0.022906
71 122277 3165 0.025780
72 118986 3490 0.029195
73 115356 3767 0.032977
74 111438 4162 0.037150
75 107110 4481 0.041738
76 102450 4790 0.046766
77 97468 5055 0.052256
78 92211 5296 0.058227
79 86703 5661 0.064701
80 80816 5811 0.071691
81 74772 5908 0.079210
82 68628 6012 0.087269
83 62375 5982 0.094871
84 56154 5886 0.103017
85 50033 5421 0.111702
86 44395 5505 0.121917
87 38670 5180 0.131647
88 33282 4837 0.143869

iii. (a) Carry out the following tests on the above data:
 – Chi-squared Test
– Standardised Deviations Test
– Signs Test
– Grouping of Signs Test
– Cumulative Deviation test over the whole range, half range and quarter
ranges
For each test state the reason for carrying out the test and draw a brief con-
clusion. [5 marks]
(b) For each test, confirm the result using either Excel or any other software of
your choice. [5 marks]

iv. You have also been asked to carry out a smoothness test on the graduated rates
for this age range. As well as carrying out the test, explain why a smoothness test
is not normally required and why it has been requested in this case. [3 marks]

v. Using your results from parts iii) and iv) comment the suitability of the graduation
including commenting on actions that can be taken to improve the results, and the
consequences of using these standard mortality rates to project demand for services
(not including pensions) used by the elderly. [3 marks]

[Total 20 marks]
Question 4
A benefit scheme makes payments when an employee is long-term sick during their work-
ing life, pays a pension when the person retires and a lump sum on death if this occurs
before retirement. An employee retires normally at age 65 though they may retire earlier
if they retire through ill-health. The company sometimes offers early retirement to their
staff who are over 60 though only as part of the company’s overall strategy and these
offers are only for a short period of time (usually the opportunity only lasts for three
months). The company is involved in heavy industry and both the mortality rates and the
sickness rates for members of the scheme are higher than the average rates in the country.

The scheme is open to employees over the age of 25. The scheme is to be modeled using
the following five-state model where the arrows indicate the possible transitions:

1: Active 2: Sick

3: Retired 5: Dead 4: Withdrawal

You are given the following definitions with respect to this model:
- µij
x+t is the force of transition from State i to State j at exact age (x + t), where
i ̸= j;

- t pij
x is the probability that a life in State i at exact age x is in State j at exact age
(x + t) where i = 1, 2; j = 1, 2, 3, 4, 5; i ̸= j and 1 > t ≥ 0.

i. Derive a solution for t p11

x in terms of the transitions above (where t < 1) stating
your assumptions. [5 marks]

ii. Derive a differential equation for t p22

x in terms of the transitions above (where t < 1)
and state the boundary condition. [2 marks]

iii. A student has queried whether there is an error in the model as there is no transition
possible from State 2 to State 4. Explain why this is not an error. [1 marks]

iv. Comment on the suitability of the Markov assumption for this model. [2 marks]

v. It has been suggested that the calculations would be easier if forces of transition
were assumed to be constant for each year of age x. Comment on the suitability of
this suggestion. [4 marks]
It is decided that a constant force of transitions can be assumed. The data relating to
50 year olds in the scheme was collected (over a three year period) and is given below:

39 transitions from state 1 to state 2

0 transition from state 1 to state 3
10 transitions from state 1 to state 4
6 transitions from state 1 to state 5

28 transitions from state 2 to state 1

6 transitions from state 2 to state 3
3 transitions from state 2 to state 5

The time spent in the two states was 610 years for state 1, and 93 years for state 2.

vi. State the likelihood function for these data. [2 marks]

vii. Derive the maximum likelihood estimate for µ15

50 . [2 marks]

viii. Calculate an estimate of the standard error of µ̂15 15

50 where µ̂50 is the maximum
likelihood estimator of µ15
50 . [2 marks]

[Total 20 Marks]