INSTITUTE AND FACULTY OF ACTUARIES

EXAMINATION

1 October 2013 (pm)

Subject CT6 – Statistical Methods

Core Technical

Time allowed: Three hours

INSTRUCTIONS TO THE CANDIDATE

1. Enter all the candidate and examination details as requested on the front of your answer
booklet.

2. You must not start writing your answers in the booklet until instructed to do so by the
supervisor.

3. Mark allocations are shown in brackets.

4. Attempt all 10 questions, beginning your answer to each question on a separate sheet.

5. Candidates should show calculations where this is appropriate.

Graph paper is NOT required for this paper.

AT THE END OF THE EXAMINATION

Hand in BOTH your answer booklet, with any additional sheets firmly attached, and this
question paper.

In addition to this paper you should have available the 2002 edition of the Formulae
and Tables and your own electronic calculator from the approved list.

CT6 S2013 © Institute and Faculty of Actuaries

1 An insurance company has a portfolio of n policies. The probability of a claim in a
given year on each policy is p independently from policy to policy, and the possibility
of more than one claim can be ignored. Prior beliefs about p are specified by a Beta
distribution with parameters α and β. In one year the insurance company has a total
of k claims on the portfolio.

Calculate the posterior estimate of p under all or nothing loss and show that it can be
written in the form of a credibility estimate. [5]

[You may use without proof the fact that the mode of a Beta distribution with
α −1
parameters α and β is .]
α +β−2

2 Claim amounts on a certain type of insurance policy follow an exponential

distribution with mean 100. The insurance company purchases a special type of
reinsurance policy so that for a given claim X the reinsurance company pays

0 if 0 < X < 80;

0.5 X − 40 if 80 ≤ X < 160;
X − 120 if X ≥ 160.

Calculate the expected amount paid by the reinsurance company on a randomly

chosen claim. [6]

3 Andy is a famous weight lifter who will be competing at the Olympic Games. He has
taken out special insurance which pays out if he is injured. If the injury is so serious
that his career is ended the policy pays $1m and is terminated. If he is injured but
recovers the insurance payment is $0.1m and the policy continues.

The insurance company’s underwriters believe that the probability of an injury in any
year is 0.2, and that the probability of more than one injury in a year can be ignored.
If Andy is injured, there is a 75% chance that he will recover.

Annual premiums are paid in advance, and the insurance company pays claims at the
end of the year. Assume that this is the only policy that the insurance company
writes, and that it has an initial surplus of $0.1m.

(i) Define what is meant by ψ ($0.1m,1) and ψ ($0.1m). [2]

(ii) Calculate the annual premium charged assuming the insurance company uses a
premium loading of 30%. [2]

(iii) Determine ψ ($0.1m, 2). [4]

[Total 8]

CT6 S2013–2
4 The table below shows the probability distribution of a discrete random variable X.

Value 1 2 3
Probability 0.3 0.3 0.4

(i) Construct an algorithm to generate random samples from X. [2]

The random variable Z takes values from X with probability 0.2 and values from an
exponential distribution Y with probability 0.8. The upper quartile point of the
distribution of Y is 2.5.

(ii) Calculate the expected value of Y. [3]

(iii) Extend the algorithm in part (i) to generate random samples from Z. [4]
[Total 9]

5 An insurance company has a portfolio of life insurance policies for 2,000 workers at a
factory. The policies pay out £5,000 if a worker dies in an industrial accident and
£2,000 if a worker dies for any other reason. For each worker, the probability of
death in any year is 0.02 and 25% of deaths are the result of industrial accidents. The
insurance company charges an annual premium of £74.25 per worker.

(i) Calculate the premium loading used by the insurance company. [2]

The insurance company is considering adopting one of the following three approaches
to reinsurance:

A None.

B 30% proportional reinsurance at a cost of £27 per worker.

C Individual excess of loss reinsurance with retention £3,000 and a

premium of £15 per worker.

(ii) Find the optimal decision under the Bayes criterion. [4]

(iii) Find the optimal decision under the minimax criterion. [2]

(iv) Comment on your answer to part (iii). [2]

[Total 10]

CT6 S2013–3 PLEASE TURN OVER

6 The tables below show cumulative data for the number of claims and the total claim
amounts arising from a portfolio of insurance policies.

Claim Numbers Total Claim Amounts

Development Year Development Year
0 1 2 0 1 2

2010 87 132 151 2010 43,290 87,430 126,310

2011 117 156 2011 68,900 125,290
2012 99 2012 74,250

Claims are fully run off after two development years.

Estimate the outstanding claims using the average cost per claim method with
grossing up factors. [10]

7 An insurance company offers dental insurance to the employees of a small firm. The
annual number of claims follows a Poisson process with rate 20. Individual loss
amounts follow an exponential distribution with mean 100. In order to increase the
take-up rate, the insurance company has guaranteed to pay a minimum amount of £50
per qualifying claim. Let S be the total claim amount on the portfolio for a given
year.

(i) Show that the mean and variance of S are 2,213.06 and 413,918.40
respectively. [7]

∞
[You may use without proof the result that if I n = ∫y
n
λe −λy dy
M
n
then I n = M n e −λM + I n −1 ]
λ

(ii) (a) Fit a log-normal distribution for S using the method of moments.

(b) Estimate the probability that S is greater than 4,000.

[3]

Sarah, the insurance company’s actuary, has instead approximated S by a Normal

distribution.

(iii) Explain, without performing any further calculations, whether the probability
that she calculates that S exceeds 4,000 will be greater or smaller than the
calculation in part (ii). [2]
[Total 12]

CT6 S2013–4
8 The number of claims per month Y arising on a certain portfolio of insurance policies
is to be modelled using a modified geometric distribution with probability density
given by

α y −1
p ( y α) = y = 1, 2,3, …
(1 + α) y

where α is an unknown positive parameter. The most recent four months have
resulted in claim numbers of 8, 6, 10 and 9.

(i) Derive the maximum likelihood estimate of α. [5]

(ii) Show that Y belongs to an exponential family of distributions and suggest its
natural parameter. [5]
[Total 10]

9 (i) State the three main stages in the Box-Jenkins approach to fitting an ARIMA
time series model. [3]

(ii) Explain, with reasons, which ARIMA time series would fit the observed data
in the charts below. [2]

ACF PACF

Now consider the time series model given by

X t = α1 X t −1 + α 2 X t −2 +β1et −1 + et

where et is a white noise process with variance σ2 .

(iii) Derive the Yule-Walker equations for this model. [6]

(iv) Explain whether the partial auto-correlation function for this model can ever
give a zero value. [2]
[Total 13]

CT6 S2013–5 PLEASE TURN OVER

10 The number of service requests received by an IT engineer on any given day follows a
Poisson distribution with mean μ . Prior beliefs about μ follow a gamma distribution
with parameters α and λ . Over a period of n days the actual numbers of service
requests received are x1 , x2 ,…, xn .

(i) Derive the posterior distribution of μ . [3]

(ii) Show that the Bayes estimate of μ under quadratic loss can be written as a
credibility estimate and state the credibility factor. [2]

Now suppose that α = 10 , λ = 2 and that the IT worker receives 42 requests in

6 days.

(iii) Calculate the Bayes estimate of μ under quadratic loss. [1]

Three quarters of requests can be resolved by telling users to restart their machine,
and the time taken to do so follows a Pareto distribution with density

3 × 203
f ( x) = for x>0.
(20 + x)4

One quarter of requests are much harder to resolve, and the time taken to resolve these
follows a Weibull distribution with density

0.5
f ( x ) = 0.4 × 0.5 x −0.5e −0.4 x for x>0.

(iv) (a) Calculate the probability that a randomly chosen request takes more
than 30 minutes to resolve.

(b) Calculate the average time spent on each request.

(c) Calculate the expected total amount of time the IT worker spends
dealing with service requests each day, using the estimate of µ from
part (iii).
[5]

The IT worker’s line manager is carefully considering his staffing requirements. He

decides to model the time taken on each request approximately using an exponential
distribution.

(v) (a) Fit an exponential distribution to the time taken per request using the
method of moments.

(b) Calculate the probability that a randomly chosen request takes more
than 30 minutes to resolve using this approximation.

(c) Comment briefly on your answer to part (v)(b). [2]

CT6 S2013–6
 The IT engineer needs to devote more of his time to a separate project, so his firm
have hired an assistant to help him. The assistant is just as fast at dealing with the
straightforward requests, and the time taken to resolve these still follows the Pareto
distribution given above. He is significantly slower at dealing with the difficult
requests, and the time taken to resolve these now follows a Weibull distribution with
density:

0.5
f ( x ) = c × 0.5 x −0.5e − cx for x>0

where c is a positive parameter. The line manager again fits an exponential

distribution as an approximation to the time taken to service each request using the
method of moments. His approximation results in an estimate that the probability that
a random service request takes longer than 30 minutes to resolve is 10%.

(vi) Determine the value of c. [4]

[Total 17]

END OF PAPER

CT6 S2013–7