Institute of Actuaries of India

Subject CT6 Statistical Methods

May 2015 Examinations

INDICATIVE SOLUTIONS

Introduction
The indicative solution has been written by the Examiners with the aim of helping candidates.
The solutions given are only indicative. It is realized that there could be other points as valid
answers and examiner have given credit for any alternative approach or interpretation which they
consider to be reasonable.

IAI

CT6-0515

Solution 1:
(i) Let Y denotes the net claim payable by the insurer.
Hence
Now the net premium income by the insurer is given by,

Now the net claim payable by the insurer is =

The insurer will not make any loss if

Or
So,

[5]

(ii)

Page 2 of 18

IAI

CT6-0515

[3]
(iii) Equation of adjustment coefficient is

Or
Or
Or
Or

[3]
(iv) Using M = 50 and = 0.01 and

we get

Solving for R and taking smallest +ve root we get R = 0.01059

[3]
(v) Here M = 0, then we get the revised equation as

Solving we get, R = 0.01, taking smallest +ve root

[2]
(vi) Initial surplus U = 100
Using Lundbergs Inequality
Hence, maximum ruin probability under (d) = exp(-0.01059*100) = 34.68%

Page 3 of 18

IAI

CT6-0515

And maximum ruin probability under (e) = exp(-0.01 * 100) = 36.78%

We can see that if the retention level reduces then the ruin probability increases.

[2]
[17 Marks]

Solution 2:
(i) I : Inox, P: PVR, C: Cinemax, F: Fun Cinemas, T: Watches Theatre
P(I) = 2/6, P(P) = 1/6, P(C) = 1/6, P(F) = 2/6
P(T|I) = 7/10, P(T|P) = 3/10, P(T|C) = 5/10, P(T|F) = 8/10

P(T) = P(I)P(T | I) + P(P)P(T | P) + P(C)P(T | C) + P(F)P(T | F)

2 7 1 3 1 5 2 8 38 19

6 10 6 10 6 10 6 10 60 30

P(T )

11
30

Now
2
7
(1 )
P( IT ) P( I ) P(T | I ) 6
10 6
P( I | T )

11
P(T )
P(T )
22
30
Similarly
7
22
5
P (C | T )
22
4
P( F | T )
22
P( P | T )

So it is most likely that he has been to PVR.

[4]

(ii) The number of Sundays on which he watches theatre is a random variable X having
Binomial distribution B(3, 19/30).
Hence the required probability =P(X = 2) + P(X = 3)
= 3*(19/30)2 * (1 19/30) + (19/30)3 = 0.695

[2]

Page 4 of 18

IAI

CT6-0515

(iii) The probability that they meet on a given Sunday in the theatre
(

2 7 2 1 3 2 1 5 2 2 8 2 27
) (
) (
) (
)
6 10
6 10
6 10
6 10
200

Hence the probability that they fail to meet over two weekends = (1 27/200)2
So, the probability that they meet at least once = 1 (1 27/200)2 = 0.251775

[3]
[9 marks]

Solution 3:
(i) The lowest (worst) profit under each of the three possibilities of masala dosa, noodles
and Pizza are 850, 800 and 500 respectively.
The highest (best) profit among them is 850.
Hence the best of the worst possible case criterion solution is to choose the strategy to sell
masala dosa.
The maximax solution is to choose a strategy which maximizes the maximum profit under the
three scenarios.
The maximum profit under each of the three possibilities of masala dosa, noodles and
Pizza are 1200, 1500 and 1400 respectively. Hence the strategy which will maximize the
maximum profit is to sell noodles.
[2]
(ii) Under Bayes Criterion we will select the strategy which gives the maximum expected profit.
Probability (Low footfall) =1/4
Probability (normal footfall) =1/2
Probability (High footfall) =1/4
The expected profit under each of the three strategies is:
Expected Profit (Masala Dosa) = 850*1/4 + 950*1/2 + 1200*1/4 = 987.5
Expected Profit (Noodles) = 900*1/4 + 800*1/2 + 1500*1/4 = 1000
Expected Profit (Pizza) = 500*1/4 + 875*1/2 + 1400*1/4 = 912.5
Thus the strategy selected under Bayes Criterion is to choose to sell noodles.

[2]
[4 Marks]
Page 5 of 18

IAI

CT6-0515

Solution 4:
The aggregate claim (S) from an individual health policy can take any one of the following
values over the coming year:
0, 100000, 200000, 300000, 400000
We need to compute Probability (S<=s) where s can take any of the values mentioned above
Thus,
P (S<=0) = P (Number of claims = 0) = P (N=0) = 0.7
P (S<=100000) = P(S=0) + P (S=100000) = 0.7 + 0.2*0.7 = 0.84
P (S<=200000) = P(S=0) + P(S=100000) + P(S=200000) = P(S<=100000) + P(S=200000)
= 0.84 + 0.1*0.7*0.7 + 0.2*0.3
= 0.84 + 0.109 = 0.949
P (S<=300000) = P(S=0) + P(S=100000) + P(S=200000) + P(S=300000) = P(S<=200000) +
P(S=300000)
= 0.949 + 2*0.1*0.7*0.3
= 0.949 + 0.042
= 0.991
Similarly,
P (S<=400000) = P(S<=300000) + P(S=400000)
= 0.991 + 0.1*0.3*0.3
= 0.991 +0.009
=1

[7 Marks]

Solution 5:
(i)
(a) The sample median of our claims data is the 6th observation out of the 11 observed values.
Thus the median based on the sample data is 80000.

Page 6 of 18

IAI

CT6-0515

Assume M to be the population median.

Then M will satisfy the equation
1 exp(-M) =

(i)

Substituting the value of the sample median in equation (i) we have

1 exp(-80000) =
exp(-80000) =
Taking log on both sides and solving we get

-80000 = ln(1/2)
= - ln(1/2)/80000
= -1*-0.000008664
= 0.000008664

Hence the estimate of using the method of percentiles equals 0.000008664

[3]

(b) The likelihood of observing the 7 known claims and 4 unknown claims greater than 100000
is
L () = f(x1).f(x2).f(x3)..f(x7) * P(X>100000)4
= exp(-x1). exp(-x2). exp(-x3). exp(-x7). * (exp(-100000)4)
= 7exp(-xi) . exp(-400000)
= 7exp(-386645) . exp(-400000)
= 7exp(-786645)
By taking logarithm on both sides we get
Log L () = 7 log - 786645

(i)

To find out the mle we need to differentiate equation (i) w.r.t .

Thus we get
d Log L ()/d = 7/ - 786645

(ii)

To find out maxima or minima we set the derivative to zero.

Hence we get
Page 7 of 18

IAI

CT6-0515

7/ = 786645
= 7/786645 = 0.000008899
To find out if this estimator is a maximum we differentiate equation (ii) again with respect to
and observe that
D2 Log L ()/d2 = -7 / 2
which is less than zero.
Thus the maximum likelihood estimate of is 0.000008899

[5]

(ii)
(a) Let X denote the random variable representing the gross claim amount which follows a
pare to (,150000) distribution.
Let Y be the random variable which represents the amount paid by the insurer on a claim.
As a policy excess of 10000 is in force the claim payments made by the insurer follow a
conditional distribution.
The probability density function of Y is
s(y) = f(x)/(P(X>10000), y >0 & x = y + 10000
Now Probability (X>10000) = (150000/(150000+10000)) , since X ~ pareto(,150000)
P(X>10000) = (150000/160000) .. (i)
Now the numerator is f(x) = f(y+10000)
= * 150000 /(150000 + y + 10000)+1
= * 150000 /(160000 + y)+1
Thus s(y) = * 150000 /(160000 + y)+1 (150000/160000)
= * 160000 /(160000 + y)+1
which is the pdf of the conditional claim distribution and whose form is of a pareto distribution
with parameters (, 160000)
[4]

Page 8 of 18

IAI

CT6-0515

(b) The likelihood function L() = s(y1). s(y2). s(y3). s(y4). . s(y10).
Thus L() = * 160000 /(160000 + y1)+1 . * 160000 /(160000 + y2)+1 . * 160000
/(160000 + y3)+1.. * 160000
/(160000 + y10)+1
L() = * 160000/ /(160000 + yi)+1
= 10 * 16000010/ (160000 + yi)+1
Taking logarithm on both sides we get,
Log L() = log(10 * 16000010/ (160000 + yi)+1)
= 10 log + 10 * log 160000 (+1) (log(160000+ yi)
To find out the mle we differentiate the above equation with respect to
dlog L()/d = 10/ + 10 log 160000 - (log(160000+ yi)
and set it equal to zero.
Thus,
0 = 10/ + 10 log 160000 - (log(160000+ yi)
(log(160000+ yi) = 10/ + 10 log 160000
(log(160000+ yi) - 10 log 160000 = 10/
= 10 / ( (log(160000+ yi) - 10 log 160000 ) = 10 / (125.0217 119.8292) = 1.925
To find out if the estimator found out is a maximum we take a second derivative of the log
likelihood function with respect to .
Thus we get
d2log L()/d2 = -10/2 < 0 = max
Hence the maximum likelihood estimate of is 1.925 (approx 2).

[5]
[17 Marks]

Page 9 of 18

IAI

CT6-0515

Solution 6:
(i) Adjusting past inflation for claim paid amount we get,
Development Year
Incidence
Year
2012
2013
2014

0
1
2
49,612.50 33,993.75 24,375.00
84,288.75 57,000.00
72,800.00

Now Cumulative claim paid is given by,

Development Year
Incidence
Year
2012
2013
2014

0
1
2
49,612.50 83,606.25 107,981.25
84,288.75 141,288.75
72,800.00

Now cumulative no of claims is given by,

Development Year
Incidence
Year
2012
2013
2014

1
50
95
80

2
85
155

110

So the average cost per claim is given by,

Development Year
Incidence
Year
2012
2013
2014

0
992.25
887.25
910.00

1
983.60
911.54

2
981.65

Now using the development factors we have completed the lower triangle for the above two table
as given below:

Page 10 of 18

IAI

CT6-0515

No of claims:
Development Year
Incidence
Year
2012
2013
2014

0
50.00
95.00
80.00

1
85.00
155.00
132.41

2
110.00
200.59
171.36

DF for Year 1 = (85+155)/(50+95) = 1.65517, for Year 2 = 110/85 = 1.294118

Average cost per claim
Development Year
Incidence
Year
2012
2013
2014

0
992.25
887.25
910.00

1
983.60
911.54
917.57

2
981.65
909.73
915.75

Development factor for Year 1 = 1.00832, and for Year 2 = 0.99801

So total claim cost per claim can be calculated using the below table (after adjusting future
inflation):
Average Cost per
claim
Total No of claims
981.65
110.00
1,000.70
200.59
1,108.06
171.36

Total
Claim
107,981.25
200,728.89
189,875.67
498,585.80

Claims already paid = 72,800 + 1,41,288.75 + 1,07,981.25 = 3,22,070

Hence outstanding reserve requirement = 498,585.80 3,22,070 = 1,76,515.80

[7]

(ii) Assumptions:

First year fully run-off

The average cost per claim in each development year is a constant proportion in monetary
terms of the ultimate average cost per claim for each incidence year
Page 11 of 18

IAI

CT6-0515

The number of claims in each development year is a constant proportion of the ultimate
no of claims for each incidence year
[1]

(iii) We will project the no of claims and average cost per claim using the calculated
development factors.
No of claims projected:
Development Year
Incidence
Year
2012
2013
2014

0
50.00
95.00
80.00

1
82.76
157.24

2
107.10

Average cost per claim projected:

Development Year
Incidence
Year
2012
2013
2014

0
1
992.25 1,000.51
887.25
894.63
910.00

2
998.52

So, total cumulative claim paid is given by,

Development Year
Incidence
Year
2012
2013
2014

0
1
2
49,612.50 82,800.71 106,940.86
84,288.75 140,673.59
72,800.00

So the non-cumulative claim paid is given by,

Development Year
Incidence
Year
2012
2013
2014

0
49,612.50
84,288.75
72,800.00

1
33,188.21
56,384.84

2
24,140.15

Page 12 of 18

IAI

CT6-0515

Now taking the difference between the predicted and actual claim paid we get:
Development Year
Incidence
Year
2012
2013
2014

0
-

1
(805.54)
(615.16)

2
(234.85)

Expressing the above difference as % of actual claim paid we get,

Development Year
Incidence
Year
2012
2013
2014

0
0.00%
0.00%
0.00%

1
-2.37%
-1.08%

2
-0.96%

We can conclude that the model is reasonably fit.

The differences are not very high with a maximum absolute difference between predicted and
actual is less than 2.5%.
[4]
(iv) Each development factor is log-normally distributed. Hence their product is also lognormally distributed.
The development factor to ultimate for Incidence Year 2014 is log-normally distributed with
parameters as given below:
Average cost per claim:
No of claims:
Equation for outstanding claim for Incidence Year 2014 can be written as:
910*1.12 * exp(0.00335 + Z * 0.0765)*80*exp(0.38045 + Z * 0.9095) 72,800
So the required probability can be found from below equation:
1,50,000 88,088 exp(0.3838 + Z * 0.986) 72,800 2,00,000
Or, 0.55186 Z 0.7572
Hence, Z = 6.61%

[6]
[18 Marks]
Page 13 of 18

IAI

CT6-0515

Solution 7:
(i)

X t At Bt
At 0.5 At 1 0.5Bt 1 et( A)
Bt 0.7 At 1 0.7 At 2 et( B )
Using matrix notation we get,
( A)
At 0.50.5 At 1 0 0 At 2 et

( B )

Yt

Bt 0.70 Bt 1 0.70 Bt 2 et

e ( A)
0.50.5
0 0
Yt 1
Yt 2 t( B )
Or, Yt
e
0.70
0.70
t
The Eigen values of first matrix are given by the following equation,

Hence,
Similarly for second matrix,
As all the Eigen values are less than 1, hence the process Yt is stationary.

[3]

(ii)
a)
Or,
Or,
So Xt is ARIMA(2,1,0) process if it is I(1)
Now,
The characteristic equation is,
To meet stationary condition,
Hence Xt is ARIMA(2,1,0) process with

[2]
Page 14 of 18

IAI

CT6-0515

b) Now Cov(Yt , et ) = Cov(et , et) = 2

.(1)
Taking co-variances with Yt-1 , Yt-2 and Yt-k we ge
(2)
.(3)

From (2),

(4)

Substituting (3) in (1) we get,

Or,
Or,
Hence,
For k 2,
Now

.(5)

follows the below equation

(6)

Substituting k by k-1 and k-2 we get

Substituting the above two equation in eqn (5) we get,

Or,
Or,
Or,
Page 15 of 18

IAI

CT6-0515

Which is the original equation. Hence

follows equation (6)

Now putting k = 0 we get,

Putting k = 1 we get,
[11]

c) = 0.04, hence

Since x1, x2, , x50 are observed values

So the forecasted values are

and
Where

And

[2]
[18 Marks]

Solution 8:
(i)
The Likelihood function based on previous years claims data is
L(q) =

. qp. (1-q)3500-p , as the policies are independent and maximum permissible claim

under a single policy is 1

Taking logarithm on both sides we get ,
Log L(q) = constant + p log q + (3500-p). log(1-q)
Differentiating with respect to q we get,
Page 16 of 18

IAI

CT6-0515

d Log L(q)/ dq = p/q (3500-p)/(1-q)

Equating the above equation to zero and rearranging we have
p/q = (3500-p)/(1-q)

p(1-q) = q .(3500 p)
p pq = 3500 q pq
p = 3500 q
q = p/3500

which gives the estimate of q using the method of mle.

Posterior distribution of q
We know that the posterior distribution is proportional to prior*likelihood (i)
The prior distribution of q is Beta (,)
Thus we have f(q) = (+)* q-1.(1-q)-1 / (() ())
Also Likelihood function is represented by
. qp. (1-q)3500-p

L(q) =

Thus equation (i) becomes

Posterior distribution is
f(q|x)

. qp. (1-q)3500-p . (+)* q-1.(1-q)-1 / (() ())

(+)* q+p-1.(1-q)+3500-p-1 / (() ())

This is Beta distribution with parameters ( p+, p +3500)

[5]

(ii) Bayesian Estimate under quadratic loss function is the mean of the posterior distribution
Since the given posterior distribution is Beta with parameters ( p+, p +3500) the Bayesian
Estimate is equal to
(p+) / ( + + 3500)
For p = 500, = 1, = 4 , the Bayesian Estimate under quadratic loss function is
= (500+1)/(1+4+3500) = 501/3505 = .1429

[2]

Page 17 of 18

IAI

CT6-0515

(iii) Let us find out if p+ / ( + + 3500) can be written in the form of a credibility estimate
Thus (p+) / ( + + 3500) = p / ( + + 3500) + / ( + + 3500)
(p+) / ( + + 3500) = 3500/( + + 3500)*p/(3500) + ( + )/ ( + + 3500)* /( +
)
which is in the form of a credibility estimate, Z*x + (1-Z) *
where x = p/(3500) is the sample mean and = /( + )is the population mean(from the prior)
and Z = 3500/( + + 3500)
Thus it can be represented in the form of a credibility estimate.
And Z(credibility factor) = 3500/3505 = 0.998573 for p = 500, = 1, = 4

[3]
[10 Marks]

***************************************

Page 18 of 18