SOCIETY OF ACTUARIES/CASUALTY ACTUARIAL SOCIETY

EXAM C CONSTRUCTION AND EVALUATION OF ACTUARIAL MODELS

EXAM C SAMPLE SOLUTIONS

Copyright 2008 by the Society of Actuaries and the Casualty Actuarial Society

Some of the questions in this study note are taken from past SOA/CAS examinations.

C-09-08 PRINTED IN U.S.A.

Question #1
Key: E

The 40th percentile is the .4(12) = 4.8th smallest observation. By interpolation it is .2(86)
+.8(90) = 89.2. The 80th percentile is the .8(12) = 9.6th smallest observation. By
interpolation it is .4(200) +.6(210) = 206.

The equations to solve are

(89.2 / θ )γ (206 / θ )γ
.4 = and .8 = .
1 + (89.2 / θ )γ 1 + (206 / θ )γ
2
Solving each for the parenthetical expression gives = (89.2 / θ )γ and 4 = (206 / θ )γ .
3
Taking the ratio of the second equation to the first gives 6 = (206 / 89.2)γ which leads to
γ = ln(6) / ln(206 / 89.2) = 2.1407 . Then 41/ 2.1407 = 206 / θ for θ = 107.8 .

Question #2
Key: E

⎛ 1.645 ⎞ ⎛ Var ( X ) ⎞
2

The standard for full credibility is ⎜ ⎟ ⎜1 + ⎟ where X is the claim size

⎝ .02 ⎠ ⎝ E ( X )2 ⎠
2(.5) 2
variable. For the Pareto variable, E ( X ) = .5 / 5 = .1 and Var ( X ) = − (.1) 2 = .015 .
5(4)
2
⎛ 1.645 ⎞ ⎛ .015 ⎞
Then the standard is ⎜ ⎟ ⎜1 + 2 ⎟ = 16,913 claims.
⎝ .02 ⎠ ⎝ .1 ⎠

Question #3
Key: B

The kernel is a triangle with a base of 4 and a height at the middle of 0.5 (so the area is
1). The length of the base is twice the bandwidth. Any observation within 2 of 2.5 will
contribute to the estimate. For the observation at 2, when the triangle is centered at 2,
the height of the triangle at 2.5 is .375 (it is one-quarter the way from 2 to the end of the
triangle at 4 and so the height is one-quarter the way from 0.5 to 0). Similarly the points
at 3 are also 0.5 away and so the height of the associated triangle is also .375. Each
triangle height is weighted by the empirical probability at the associated point. So the
estimate at 2.5 is (1/5)(3/8) + (3/5)(3/8) + (1/5)(0) = 12/40.

C-09-08 -1-
Question #4
Key: A

x
The distribution function is F ( x) = ∫ α t −α −1dt = − t −α = 1 − x −α . The likelihood function is
x

1 1

L = f (3) f (6) f (14)[1 − F (25)] 2

= α 3−α −1α 6−α −1α14−α −1 (25−α ) 2

∝ α 3 [3(6)(14)(625)]−α .
Taking logs, differentiating, setting equal to zero, and solving:
ln L = 3ln α − α ln157,500 plus a constant
(ln L)′ = 3α −1 − ln157,500 = 0
αˆ = 3 / ln157,500 = .2507.

Question #5
Key: C

π (q |1,1) ∝ p (1| q ) p(1| q)π (q) = 2q(1 − q )2q(1 − q )4q 3 ∝ q 5 (1 − q) 2

1
∫ q (1 − q) dq = 1/168, π (q |1,1) = 168q 5 (1 − q ) 2 .
5 2
0

The expected number of claims in a year is E ( X | q) = 2q and so the Bayesian estimate

is
1
E (2q |1,1) = ∫ 2q (168)q 5 (1 − q) 2 dq = 4 / 3.
0

The answer can be obtained without integrals by recognizing that the posterior
distribution of q is beta with a = 6 and b = 3. The posterior mean is
E (q |1,1) = a /(a + b) = 6 / 9 = 2 / 3. The posterior mean of 2q is then 4/3.

Question #6
Key: D

For the method of moments estimate,

386 = e μ +.5σ , 457, 480.2 = e 2 μ + 2σ
2 2

5.9558 = μ + .5σ 2 , 13.0335 = 2 μ + 2σ 2

μˆ = 5.3949, σˆ 2 = 1.1218.
Then

C-09-08 -2-
 ⎛ ln 500 − 5.3949 − 1.1218 ⎞ ⎡ ⎛ ln 500 − 5.3949 ⎞ ⎤
E ( X ∧ 500) = e5.3949+.5(1.1218) Φ ⎜ ⎟ + 500 ⎢1 − Φ ⎜ ⎟⎥
⎝ 1.1218 ⎠ ⎣ ⎝ 1.1218 ⎠⎦
= 386Φ (−.2853) + 500[1 − Φ (.7739)]
= 386(.3877) + 500(.2195) = 259.
Note-these calculations use exact normal probabilities. Rounding and using the normal table that
accompanies the exam will produce a different numerical answer but the same letter answer.

Question #7
DELETED

Question #8
Key: C

Let N be the Poisson claim count variable, let X be the claim size variable, and let S be
the aggregate loss variable.
μ (θ ) = E ( S | θ ) = E ( N | θ ) E ( X | θ ) = θ 10θ = 10θ 2
v(θ ) = Var ( S | θ ) = E ( N | θ ) E ( X 2 | θ ) = θ 200θ 2 = 200θ 3
∞
μ = E (10θ 2 ) = ∫ 10θ 2 (5θ −6 )dθ = 50 / 3
1
∞
EPV = E (200θ ) = ∫ 200θ 3 (5θ −6 )dθ = 500
3
1
∞
VHM = Var (10θ 2 ) = ∫ (10θ 2 ) 2 (5θ −6 )dθ − (50 / 3) 2 = 222.22
1

k = 500 / 222.22 = 2.25.

Question #9
Key: A

c = exp(.71(1) + .20(1)) = 2.4843 . Then Sˆ (t0 ; z ) = Sˆ0 (t0 )c = .652.4843 = .343 .

Question #10
Key: E

Y and X are linear combinations of the same two normal random variables, so they are
bivariate normal. Thus E(Y|X) = E(Y) + [Cov(Y,X)/Var(X)][X – E(X)]. From the definitions
of Y and X, E(Y) = a, E(X) = d, Var(X) = e2 + f2, and Cov(Y,X) = be + cf.

Question #11

C-09-08 -3-
Key: D

f (5 | θ = 1) Pr(θ = 1)
Pr(θ = 1| X = 5) =
f (5 | θ = 1) Pr(θ = 1) + f (5 | θ = 3) Pr(θ = 3)
(1/ 36)(1/ 2)
= = 16 / 43
(1/ 36)(1/ 2) + (3 / 64)(1/ 2)
Pr( X 2 > 8 | X 1 = 5) = Pr( X 2 > 8 | θ = 1) Pr(θ = 1| X 1 = 5) + Pr( X 2 > 8 | θ = 3) Pr(θ = 3 | X 1 = 5)
= (1/ 9)(16 / 43) + (3 /11)(27 / 43) = .2126.
∞
For the last line, Pr( X > 8 | θ ) = ∫ θ ( x + θ ) −2 dx = θ (8 + θ ) −1 is used.
8

Question #12
Key: C

The sample mean for X is 720 and for Y is 670. The mean of all 8 observations is 695.
(730 − 720) 2 + (800 − 720) 2 + (650 − 720) 2 + (700 − 720) 2
+ (655 − 670) 2 + (650 − 670) 2 + (625 − 670) 2 + (750 − 670) 2
vˆ = = 3475
2(4 − 1)
(720 − 695) 2 + (670 − 695) 2 3475
aˆ = − = 381.25
2 −1 4
kˆ = 3475 / 381.25 = 9.1148
4
Zˆ = = .305
4 + 9.1148
Pc = .305(670) + .695(695) = 687.4.

Question #13
Key: B

There are 430 observations. The expected counts are 430(.2744) = 117.99, 430(.3512) =151.02,
430(.3744) = 160.99. The test statistic is
(112 − 117.99) 2 (180 − 151.02) 2 (138 − 160.99) 2
+ + = 9.15.
117.99 151.02 160.99

Question #14
Key: B

From the information, the asymptotic variance of θˆ is 1/4n. Then

Var (2θˆ) = 4Var (θˆ) = 4(1/ 4n) = 1/ n.

C-09-08 -4-
Note that the delta method is not needed for this problem, although using it leads to the same
answer.

Question #15
Key: A

The posterior probability density is

π ( p |1,1,1,1,1,1,1,1) ∝ Pr(1,1,1,1,1,1,1,1| p)π ( p ) ∝ p8 (2) ∝ p8 .
p8 p8
π ( p |1,1,1,1,1,1,1,1) = .5
= = 9(.5−9 ) p 8 .
∫
0
p8 dp (.59 ) / 9
.5
Pr( X 9 = 1|1,1,1,1,1,1,1,1) = ∫ Pr( X 9 = 1| p)π ( p |1,1,1,1,1,1,1,1)dp
0
.5
= ∫ p9(.5−9 ) p8 dp = 9(.5−9 )(.510 ) /10 = .45.
0

Question #16
Key: A

18 26 20 13
3 pˆ1 = = . Greenwood’s approximation is
27 32 25 30
⎛ 13 ⎞ ⎛ 9 5 ⎞
2
6
⎜ ⎟ ⎜ + + ⎟ = .0067.
⎝ 30 ⎠ ⎝ 18(27) 26(32) 20(25) ⎠

C-09-08 -5-
Question #17
Key: D

Ĥ (3) = 5/30 + 9/27 + 6/32 = 0.6875

ˆ ( Hˆ (3)) = 5/(30)2 + 9/(27)2 + 6/(32)2 = 0.02376
Var

The 95% log-transformed confidence interval is:

⎛ 1.96 .02376 ⎞
Hˆ (3)U , where U = exp ⎜⎜ ± ⎟⎟ = exp(±0.43945)
⎝ .6875 ⎠
The confidence interval is:
[0.6875 exp(−0.43945), 0.6875 exp(0.43945)] = [0.443, 1.067].

Question #18
Key: D

The means are .5(250) + .3(2,500) + .2(60,000) = 12,875 and .7(250) + .2(2,500) + .1(60,000) =
6,675 for risks 1 and 2 respectively.
The variances are .5(250)2 + .3(2,500)2 + .2(60,000)2 – 12,8752 = 556,140,625 and .7(250)2 +
.2(2,500)2 + .1(60,000)2 – 6,6752 = 316,738,125 respectively.

The overall mean is (2/3)(12,875) + (1/3)(6,675) = 10,808.33 and so

EPV = (2/3)(556,140,625) + (1/3)(316,738,125) = 476,339,792 and
VHM = (2/3)(12,875)2 + (1/3)(6,675)2 – 10,808.332 = 8,542,222. Then,
k = 476,339,792/8,542,222 = 55.763 and Z = 1/(1 + 55.763) = .017617.
The credibility estimate is .017617(250) + .982383(10,808.33) = 10,622.

Question #19
Key: D

The first two sample moments are 15 and 500, and the first two population moments are
E ( X ) = .5(θ + σ ) and E ( X 2 ) = .5(2θ 2 + 2σ 2 ) = θ 2 + σ 2 . These can be obtained either
through integration or by recognizing the density function as a two-point mixture of
exponential densities. The equations to solve are 30 = θ + σ and 500 = θ 2 + σ 2 . From
the first equation, σ = 30 − θ and substituting into the second equation gives
500 = θ 2 + (30 − θ ) 2 = 2θ 2 − 60θ + 900 . The quadratic equation has two solutions, 10 and
20. Because θ > σ the answer is 20.

C-09-08 -6-
Question #20
Key: D

There are four possible samples, (5,5), (5,9), (9,5), and (9,9). For each, the estimator g must be
calculated. The values are 0, 4, 4, and 0 respectively. Assuming a population in which the
values 5 and 9 each occur with probability .5, the population variance is
.5(5 − 7) 2 + .5(9 − 7) 2 = 4 . The mean square error is approximated as
.25[(0 − 4) 2 + (4 − 4) 2 + (4 − 4) 2 + (0 − 4) 2 ] = 8 .

Question #21
Key: B

From the Poisson distribution, μ (λ ) = λ and v(λ ) = λ . Then,

μ = E (λ ) = 6 /100 = .06, EPV = E (λ ) = .06, VHM = Var (λ ) = 6 /1002 = .0006 where the
various moments are evaluated from the gamma distribution. Then,
k = .06 / .0006 = 100 and Z = 450 /(450 + 100) = 9 /11 where the 450 is the total number of
insureds contributing experience. The credibility estimate of the expected number of claims for
one insured in month 4 is (9/11)(25/450) + (2/11)(.06) = .056364. For 300 insureds the expected
number of claims is 300(.056364) = 16.9.

Question #22
Key: C

αθ α
The likelihood function is L(α ,θ ) = ∏ j =1
200
and its logarithm is
( x j + θ )α +1
l (α , θ ) = 200 ln(α ) + 200α ln(θ ) − (α + 1)∑ i =1 ln( xi + θ ) . When evaluated at the hypothesized
200

values of 1.5 and 7.8, the loglikelhood is −821.77. The test statistic is 2(821.77−817.92) = 7.7.
With two degrees of freedom (0 free parameters in the null hypothesis versus 2 in the
alternative), the test statistic falls between the 97.5th percentile (7.38) and the 99th percentile
(9.21).

C-09-08 -7-
Question #23
Key: E

Assume that θ > 5 . Then the expected counts for the three intervals are
15(2 / θ ) = 30 / θ , 15(3 / θ ) = 45 / θ , and 15(θ − 5) / θ = 15 − 75 / θ respectively. The quantity to
minimize is
1
⎡⎣ (30θ −1 − 5) 2 + (45θ −1 − 5) 2 + (15 − 75θ −1 − 5) 2 ⎤⎦ .
5
Differentiating (and ignoring the coefficient of 1/5) gives the equation
−2(30θ −1 − 5)30θ −2 − 2(45θ −1 − 5)45θ −2 + 2(10 − 75θ −1 )75θ −2 = 0 . Multiplying through by θ 3
and dividing by 2 reduces the equation to
−(30 − 5θ )30 − (45 − 5θ )45 + (10θ − 75)75 = −8550 + 1125θ = 0 for a solution of
θˆ = 8550 /1125 = 7.6 .

Question #24
Key: E

1
∫θ
1
π (θ |1) ∝ θ (1.5θ .5 ) ∝ θ 1.5 . The required constant is the reciprocal of 1.5
dθ = θ 2.5 / 2.5 = .4
0
0

and so π (θ |1) = 2.5θ . The requested probability is

1.5

1
Pr(θ > .6 |1) = ∫ 2.5θ 1.5 dθ = θ 2.5
1
= 1 − .62.5 = .721.
.6 .6

Question #25
Key: A
k knk / nk −1
0
1 0.81
2 0.92
3 1.75
4 2.29
5 2.50
6 3.00

Positive slope implies that the negative binomial distribution is a good choice.
Alternatively, the sample mean and variance are 1.2262 and 1.9131 respectively. With
the variance substantially exceeding the mean, the negative binomial model is again
supported.

C-09-08 -8-
Question #26
Key: B

e −1/(2θ ) e −2 /(2θ ) e −3/(2θ ) e −15 /(3θ ) e −8/ θ

The likelihood function is ⋅ ⋅ ⋅ = . The loglikelihood
2θ 2θ 2θ 3θ 24θ 4
function is − ln 24 − 4 ln(θ ) − 8 / θ . Differentiating with respect to θ and setting the result
4 8
equal to 0 yields − + 2 = 0 which produces θˆ = 2 .
θ θ

Question #27
Key: E

The absolute difference of the credibility estimate from its expected value is to be less than or
equal to k μ (with probability P). That is,
[ ZX partial + (1 − Z ) M ] − [ Z μ + (1 − Z ) M ] ≤ k μ
− k μ ≤ ZX partial − Z μ ≤ k μ .
Adding μ to all three sides produces answer choice (E).

Question #28
Key: C

In general,
200 150 200 200
E ( X 2 ) − E[( X ∧ 150) 2 ] = ∫ x 2 f ( x)dx − ∫ x 2 f ( x)dx − 1502 ∫ f ( x) dx = ∫ ( x 2 − 1502 ) f ( x) dx.
0 0 150 150
Assuming a uniform distribution, the density function over the interval from 100 to 200 is 6/7400
(the probability of 6/74 assigned to the interval divided by the width of the interval). The answer
is
200
200 6 ⎛ x3 ⎞ 6
∫ ( x − 150 ) dx = ⎜ − 1502 x ⎟ = 337.84.
2 2
150 7400 ⎝ 3 ⎠ 7400 150

Question #29
Key: B

The probabilities are from a binomial distribution with 6 trials. Three successes were
observed.
⎛6⎞ ⎛6⎞
Pr(3 | I) = ⎜ ⎟ (.1)3 (.9)3 = .01458, Pr(3 | II) = ⎜ ⎟ (.2)3 (.8)3 = .08192,
⎝ 3⎠ ⎝ 3⎠
⎛6⎞
Pr(3 | III) = ⎜ ⎟ (.4)3 (.6)3 = .27648
⎝ 3⎠

C-09-08 -9-
The probability of observing three successes is .7(.01458) + .2(.08192) + .1(.27648) =
.054238. The three posterior probabilities are:
.7(.01458) .2(.08192) .1(.27648)
Pr(I | 3) = = .18817, Pr(II | 3) = = .30208, Pr(III | 3) = = .50975.
.054238 .054238 .054238
The posterior probability of a claim is then
.1(.18817) + .2(.30208) + .4(.50975) = .28313.

Question #30
Key: E

.542 = Fˆ (n) = 1 − e − H ( n ) , Hˆ (n) = .78. The Nelson-Aalen estimate is the sum of successive s/r
ˆ

values. From the problem statement, r = 100 at all surrender times while the s-values follow the
pattern 1, 2, 3, …. Then,
1 2 n n(n + 1)
.78 = + + + = and the solution is n = 12.
100 100 100 200

Question # 31
Answer: C

g = [12(.45)] = [5.4] = 5; h = 5.4 − 5 = 0.4.

πˆ.45 = .6 x(5) + .4 x(6) = .6(360) + .4(420) = 384.

Question # 32
Answer: D

N is distributed Poisson(λ)
μ = E (λ ) = αθ = 1(1.2) = 1.2.
v = E (λ ) = 1.2; a = Var (λ ) = αθ 2 = 1(1.2) 2 = 1.44.
1.2 5 2 12
k= = ; Z= = .
1.44 6 2 + 5 / 6 17
Thus, the estimate for Year 3 is
12 5
(1.5) + (1.2) = 1.41.
17 17

Note that a Bayesian approach produces the same answer.

C-09-08 - 10 -
Question # 33
Answer: C

At the time of the second failure,

1 1 23
Hˆ (t ) = + = ⇒ n = 12.
n n − 1 132

At the time of the fourth failure,

1 1 1 1
Hˆ (t ) = + + + = .3854.
12 11 10 9

Question # 34
Answer: B

The likelihood is:

r (r + 1) (r + x j − 1) β
n xj n
L=∏ ∝ ∏ β j (1 + β )
x −r −x j
r+xj
.
j =1 x j !(1 + β ) j =1

The loglikelihood is:

n
l = ∑ ⎡⎣ x j ln β − (r + x j ) ln(1 + β ) ⎤⎦
j =1

n
⎡ x r + xj ⎤
l = ∑⎢ j −
′ =0
j =1 ⎣ β 1 + β ⎥⎦
n n
0 = ∑ ⎡⎣ x j (1 + β ) − (r + x j ) β ⎤⎦ =∑ x j − rnβ
j =1 j =1

0 = nx − rnβ ; βˆ = x / r.

Question # 35
Answer: C

The Bühlmann credibility estimate is Zx + (1 − Z ) μ where x is the first observation. The

Bühlmann estimate is the least squares approximation to the Bayesian estimate.
Therefore, Z and μ must be selected to minimize

1 1 1
[ Z + (1 − Z ) μ − 1.5]2 + [2Z + (1 − Z ) μ − 1.5]2 + [3Z + (1 − Z ) μ − 3]2 .
3 3 3

C-09-08 - 11 -
Setting partial derivatives equal to zero will give the values. However, it should be clear
that μ is the average of the Bayesian estimates, that is,

1
μ = (1.5 + 1.5 + 3) = 2.
3

The derivative with respect to Z is (deleting the coefficients of 1/3):

2(− Z + .5)(−1) + 2(.5)(0) + 2( Z − 1)(1) = 0

Z = .75.

The answer is

.75(1) + .25(2) = 1.25.

Question # 36
Answer: E

The confidence interval is ( Sˆ (t0 )1/θ , Sˆ (t0 )θ ).

Taking logarithms of both endpoints gives the two equations

1
ln .695 = −.36384 = ln Sˆ (t )
θ 0

ln .843 = −.17079 = θ ln Sˆ (t0 ).

Multiplying the two equations gives

.06214 = [ln Sˆ (t0 )]2

ln Sˆ (t ) = −.24928
0

Sˆ (t0 ) = .77936.
The negative square root is required in order to make the answer fall in the interval
(0,1).

C-09-08 - 12 -
Question # 37
Answer: B

The likelihood is:

α150α α150α α150α
L=
(150 + 225)α +1 (150 + 525)α +1 (150 + 950)α +1
α 31503α
= .
(375i675i1100)α +1

The loglikelihood is:

l = 3ln α + 3α ln150 − (α + 1) ln(375i675i1100)

3 3
l ′ = + 3ln150 − ln(375i675i1100) = − 4.4128
α α
αˆ = 3 / 4.4128 = .6798.

Question # 38
Answer: D

For this problem, r = 4 and n = 7. Then,

33.60 3.3 1.4

vˆ = = 1.4 and aˆ = − = .9.
4(7 − 1) 4 −1 7

Then,

1.4 14 7 63
k= = ; Z= = = .82.
.9 9 7 + (14 / 9) 77

Question # 39
Answer: B

X is the random sum Y1 + Y2 +...+YN .

. and β = θ = 0.2 .
N has a negative binomial distribution with r = α = 15

E ( N ) = r β = 0.3
Var ( N ) = r β (1 + β ) = 0.36

bg
E Y = 5000
Var bY g = 25,000,000

C-09-08 - 13 -
E ( X ) = 0.3 × 5000 = 1500
Var ( X ) = 0.3 × 25, 000, 000 + 0.36 × 25, 000, 000 = 16,500, 000

Number of exposures (insureds) required for full credibility

nFULL = (1.645 / 0.05) 2 × 16,500, 000 / (1500 ) = 7937.67 .
2

Number of expected claims required for full credibility

E ( N ) × nFULL = 0.3 × 7937.67 = 2381 .

Question # 40
Answer: E

X bg
Fn x d i
Fn x − F0 ( x ) bg bg
Fn x − F0 x d i bg
Fn x − − F0 x
29 0.2 0 0.252 0.052 0.252
64 0.4 0.2 0.473 0.073 0.273
90 0.6 0.4 0.593 0.007 0.193
135 0.8 0.6 0.741 0.059 0.141
182 1.00 0.8 0.838 0.162 0.038

where:
θˆ = x = 100 and F0 ( x) = 1 − e − x /100 .
The maximum value from the last two columns is 0.273.

Question # 41
Answer: E

μ = E (λ ) = 1; v = E (σ 2 ) = 1.25; a = Var (λ ) = 1/12.

1 1
k = v / a = 15; Z = = .
1 + 15 16

Thus, the estimate for Year 2 is

1 15
(0) + (1) = .9375.
16 16

C-09-08 - 14 -
Question # 42
DELETED

Question # 43
Answer: E

The posterior density, given an observation of 3 is:

2θ 2 1
f (3 | θ )π (θ ) (3 + θ )3 θ 2
π (θ | 3) = ∞
= ∞
∫1
f (3 | θ )π (θ )dθ ∫
1
2(3 + θ ) −3 dθ
2(3 + θ ) −3
= = 32(3 + θ ) −3 , θ > 1.
−2 ∞
−(3 + θ )
1

Then,

∞ ∞ 16
Pr(Θ > 2) = ∫ 32(3 + θ )−3 dθ = −16(3 + θ ) −2 = = .64.
2 2 25

Question # 44
Answer: B

L = F (1000)7 [ F (2000) − F (1000)]6 [1 − F (2000)]7

= (1 − e −1000 /θ )7 (e−1000 / θ − e−2000 /θ )6 (e −2000 /θ )7
= (1 − p )7 ( p − p 2 )6 ( p 2 )7
= p 20 (1 − p )13

where p = e −1000 / θ . The maximum occurs at p = 20/33 and so

θˆ = −1000 / ln(20 / 33) = 1996.90.

Question # 45
Answer: A

E ( X | θ ) = θ / 2.

∞
∞ θ 6003 3(6003 ) θ −2
∞
E ( X 3 | 400, 600) = ∫ E ( X | θ ) f (θ | 400, 600)dθ = ∫ 3 4 dθ =
600 600 2 θ 2 −2 600
3 −2
3(600 )(600 )
= = 450.
4

C-09-08 - 15 -
Question # 46
Answer: D

The data may be organized as follows:

T Y d Sˆ (t )
2 10 1 (9/10) = .9
3 9 2 .9(7/9) = .7
5 7 1 .7(6/7) = .6
6 5 1 .6(4/5) = .48
7 4 1 .48(3/4) = .36
9 2 1 .36(1/2) = .18

Because the product-limit estimate is constant between observations, the value of Sˆ (8)
is found from Sˆ (7) = .36 .

Question # 47
Answer: C

The maximum likelihood estimate for the Poisson distribution is the sample mean:

50(0) + 122(1) + 101(2) + 92(3)

λˆ = x = = 1.6438.
365

The table for the chi-square test is:

Number of days Probability Expected* Chi-square

0 e −1.6438 = .19324 70.53 5.98
1 1.6438e −1.6438 = .31765 115.94 0.32
2 1.64382 e −1.6438 95.30 0.34
= .26108
2
3+ .22803** 83.23 0.92

*365x(Probability) **obtained by subtracting the other probabilities from 1

The sum of the last column is the test statistic of 7.56. Using 2 degrees of freedom (4
rows less 1 estimated parameter less 1) the model is rejected at the 2.5% significance
level but not at the 1% significance level.

C-09-08 - 16 -
Question # 48
Answer: D

.4(0) + .1(1) + .1(2) .1(0) + .2(1) + .1(2)

μ (0) = = .5; μ (1) = =1
.6 .4
μ = .6(.5) + .4(1) = .7
a = .6(.52 ) + .4(12 ) − .7 2 = .06
.4(0) + .1(1) + .1(4) 7 .1(0) + .2(1) + .1(4) 2
v(0) = − .52 = ; v(1) = − 1 = .5
.6 12 .4
v = .6(7 /12) + .4(.5) = 11/ 20
10 60
k = v / a = 55 / 6; Z = =
10 + 55 / 6 115
60 10 55
Bühlmann credibility premium = + (.7) = .8565 .
115 10 115

Question # 49
Answer: C

μ = .5(0) + .3(1) + .1(2) + .1(3) = .8

σ 2 = .5(0) + .3(1) + .1(4) + .1(9) − .64 = .96
n −1 2 3
E ( Sn2 ) = σ = (.96) = .72
n 4
bias = .72 − .96 = −.24.

Question # 50
Answer: C

The four classes have means .1, .2, .5, and .9 respectively and variances .09, .16, .25,
and .09 respectively.

Then,

C-09-08 - 17 -
μ = .25(.1 + .2 + .5 + .9) = .425
v = .25(.09 + .16 + .25 + .09) = .1475
a = .25(.01 + .04 + .25 + .81) − .4252 = .096875
k = .1475 / .096875 = 1.52258
4
Z= = .7243
4 + 1.52258

The estimate is [.7243(2 / 4) + .2757(.425) ] ⋅ 5 = 2.40 .

Question # 51
DELETED

Question # 52
Answer: A

The distribution used for simulation is given by the observed values.

Question # 53
Answer: B

First obtain the distribution of aggregate losses:

Value Probability
0 1/5
25 (3/5)(1/3) = 1/5
100 (1/5)(2/3)(2/3) = 4/45
150 (3/5)(2/3) = 2/5
250 (1/5)(2)(2/3)(1/3) = 4/45
400 (1/5)(1/3)(1/3) = 1/45

μ = (1/ 5)(0) + (1/ 5)(25) + (4 / 45)(100) + (2 / 5)(150) + (4 / 45)(250) + (1/ 45)(400) = 105
σ 2 = (1/ 5)(02 ) + (1/ 5)(252 ) + (4 / 45)(1002 ) + (2 / 5)(1502 )
+ (4 / 45)(2502 ) + (1/ 45)(4002 ) − 1052 = 8,100.

Question # 54
Answer: A

Loss Range Cum. Prob.

0 – 100 0.320
100 – 200 0.530
200 – 400 0.800

C-09-08 - 18 -
400 – 750 0.960
750 – 1000 0.980
1000 – 1500 1.000

− 400
At 400, F(x) = 0.8 = 1 − e θ ; solving gives θ = 248.53 .

Question # 55
Answer: B

(1/ 2)(1/ 3) 3
Pr(class1|1) = =
(1/ 2)(1/ 3) + (1/ 3)(1/ 6) + (1/ 6)(0) 4
(1/ 3)(1/ 6) 1
Pr(class 2 |1) = =
(1/ 2)(1/ 3) + (1/ 3)(1/ 6) + (1/ 6)(0) 4
(1/ 6)(0)
Pr(class3 |1) = =0
(1/ 2)(1/ 3) + (1/ 3)(1/ 6) + (1/ 6)(0)

because the prior probabilities for the three classes are 1/2, 1/3, and 1/6 respectively.

The class means are

μ (1) = (1/ 3)(0) + (1/ 3)(1) + (1/ 3)(2) = 1

μ (2) = (1/ 6)(1) + (2 / 3)(2) + (1/ 6)(3) = 2.

The expectation is

E ( X 2 |1) = (3 / 4)(1) + (1/ 4)(2) = 1.25.

Question # 56
Answer: E

The first, second, third, and sixth payments were observed at their actual value and
each contributes f(x) to the likelihood function. The fourth and fifth payments were paid
at the policy limit and each contributes 1 – F(x) to the likelihood function. This is answer
(E).

Question #57
Answer is E

C-09-08 - 19 -
For an interval running from c to d, the uniform density function is f ( x) = g /[n(d − c)]
where g is the number of observations in the interval and n is the sample size. The
contribution to the second raw moment for this interval is:
d
g gx 3 g (d 3 − c3 )
∫
d
x2
dx = = .
c n( d − c ) 3n(d − c) c 3n(d − c)

For this problem, the second raw moment is:

1 ⎡ 30(253 − 03 ) 32(503 − 253 ) 20(1003 − 503 ) 8(2003 − 1003 ) ⎤
+ + + = 3958.33.
90 ⎢⎣ 3(25 − 0) 3(50 − 25) 3(100 − 50) 3(200 − 100) ⎥⎦

Question #58
Answer is B

Because the Bayes and Bühlmann results must be identical, this problem can be solved
either way. For the Bühlmann approach, μ (λ ) = v(λ ) = λ . Then, noting that the prior
distribution is a gamma distribution with parameters 50 and 1/500, we have:
μ = E (λ ) = 50 / 500 = 0.1
v = E (λ ) = 0.1
a = Var (λ ) = 50 / 5002 = 0.0002
k = v / a = 500
Z = 1500 /(1500 + 500) = 0.75
75 + 210
X= = 0.19.
600 + 900

The credibility estimate is 0.75(0.19) + 0.25(0.1) = 0.1675. For 1100 policies, the expected
number of claims is 1100(0.1675) = 184.25.

For the Bayes approach, the posterior density is proportional to (because in a given
year the number of claims has a Poisson distribution with parameter λ times the number
of policies)
e −600 λ (600λ )75 e −900 λ (900λ ) 210 (500λ )50 e −500 λ
∝ λ 335e −2000 λ which is a gamma density with
75! 210! λΓ(50)
parameters 335 and 1/2000. The expected number of claims per policy is 335/2000 =
0.1675 and the expected number of claims in the next year is 184.25.

Question #59
Answer is E

C-09-08 - 20 -
The q-q plot takes the ordered values and plots the jth point at j/(n+1) on the horizontal
axis and at F ( x j ;θ ) on the vertical axis. For small values, the model assigns more
probability to being below that value than occurred in the sample. This indicates that
the model has a heavier left tail than the data. For large values, the model again
assigns more probability to being below that value (and so less probability to being
above that value). This indicates that the model has a lighter right tail than the data. Of
the five answer choices, only E is consistent with these observations. In addition, note
that as you go from 0.4 to 0.6 on the horizontal axis (thus looking at the middle 20% of
the data), the q-q plot increases from about 0.3 to 0.4 indicating that the model puts only
about 10% of the probability in this range, thus confirming answer E.

Question #60
Answer is C

The posterior probability of having one of the coins with a 50% probability of heads is
proportional to (.5)(.5)(.5)(.5)(4/6) = 0.04167. This is obtained by multiplying the
probabilities of making the successive observations 1, 1, 0, and 1 with the 50% coin
times the prior probability of 4/6 of selecting this coin. The posterior probability for the
25% coin is proportional to (.25)(.25)(.75)(.25)(1/6) = 0.00195 and the posterior
probability for the 75% coin is proportional to (.75)(.75)(.25)(.75)(1/6) = 0.01758. These
three numbers total 0.06120. Dividing by this sum gives the actual posterior
probabilities of 0.68088, 0.03186, and 0.28726. The expected value for the fifth toss is
then (.68088)(.5) + (.03186)(.25) + (.28726)(.75) = 0.56385.

Question #61
Answer is A

Because the exponential distribution is memoryless, the excess over the deductible is
also exponential with the same parameter. So subtracting 100 from each observation
yields data from an exponential distribution and noting that the maximum likelihood
estimate is the sample mean gives the answer of 73.

Working from first principles,

f ( x1 ) f ( x2 ) f ( x3 ) f ( x4 ) f ( x5 ) θ −1e −125/θθ −1e −150 /θθ −1e−165/θθ −1e−175/θθ −1e−250 /θ
L(θ ) = =
[1 − F (100)]5 (e −100 /θ )5
= θ −5e −365 /θ .

Taking logarithms and then a derivative gives

l (θ ) = −5ln(θ ) − 365 / θ , l '(θ ) = −5 / θ + 365 / θ 2 = 0.

The solution is θˆ = 365 / 5 = 73.

C-09-08 - 21 -
Question #62
Answer is D

The number of claims for each insured has a binomial distribution with n = 1 and q
unknown. We have
μ (q) = q, v(q) = q(1 − q )
μ = E (q) = 0.1, given in item (iv)
a = Var (q) = E (q 2 ) − E (q ) 2 = E (q 2 ) − 0.01 = 0.01, given in item (v)
Therefore, E (q 2 ) = 0.02
v = E (q − q 2 ) = 0.1 − 0.02 = 0.08
10
k = v / a = 8, Z = = 5 / 9.
10 + 8

Then the expected number of claims in the next one year is (5/9)(0) + (4/9)(0.1) = 2/45
and the expected number of claims in the next five years is 5(2/45) = 2/9 = 0.22.

Question #63
DELETED

Question #64
Answer is E

The model distribution is f ( x | θ ) = 1/ θ , 0 < x < θ . Then the posterior distribution is

proportional to
1 1 500
π (θ | 400, 600) ∝ ∝ θ −4 , θ > 600.
θθ θ2

It is important to note the range. Being a product, the posterior density function is non-
zero only when all three terms are non-zero. Because one of the observations was
equal to 600, the value of the parameter must be greater than 600 in order for the
density function at 600 to be positive. Or, by general reasoning, posterior probability
can only be assigned to possible values. Having observed the value 600 we know that
parameter values less than or equal to 600 are not possible.

∞ 1
The constant is obtained from ∫600
θ −4 dθ =
3(600)3
and thus the exact posterior density is

π (θ | 400, 600) = 3(600)3θ −4 , θ > 600. The posterior probability of an observation

exceeding 550 is

C-09-08 - 22 -
 ∞
Pr( X 3 > 550 | 400, 600) = ∫ Pr( X 3 > 550 | θ )π (θ | 400, 600)dθ
600
∞ θ − 550
=∫ 3(600)3 θ −4 dθ = 0.3125
600 θ
where the first term in the integrand is the probability of exceeding 550 from the uniform
distribution.

Question #65
Answer is C

E ( N ) = r β = 0.40
Var ( N ) = r β (1 + β ) = 0.48
E (Y ) = θ /(α − 1) = 500
Var (Y ) = θ 2α / ⎡⎣(α − 1) (α − 2) ⎤⎦ = 750, 000
2

Therefore,
E ( X ) = 0.40(500) = 200
Var ( X ) = 0.40(750, 000) + 0.48(500) 2 = 420, 000

2
⎛ 1.645 ⎞ 420, 000
The full credibility standard is n = ⎜ = 11,365 and then
⎝ 0.05 ⎟⎠ 2002
Z = 2500 /11,365 = 0.47.

Question #66
Answer is E

(1 − 3) 2 + (2 − 3) 2 + (3 − 3) 2 + (4 − 3) 2 + (5 − 3) 2
The sample variance is s 2 = = 2.5. The
4
estimator of E[X] is the sample mean and the variance of the sample mean is the
variance divided by the sample size, estimated here as 2.5/n. Setting the standard
deviation of the estimator equal to 0.05 gives the equation 2.5 / n = 0.05 which yields n
= 1000.

Question #67
Answer is E

μ (r ) = E ( X | r ) = E ( N ) E (Y ) = r βθ /(α − 1) = 100r
v(r ) = Var ( X | r ) = Var ( N ) E (Y ) 2 + E ( N )Var (Y )
= r β (1 + β )θ 2 /(α − 1) 2 + r βαθ 2 /[(α − 1) 2 (α − 2)] = 210, 000r.

C-09-08 - 23 -
v = E (210, 000r ) = 210, 000(2) = 420, 000
a = Var (100r ) = (100) 2 (4) = 40, 000
k = v / a = 10.5
Z = 100 /(100 + 10.5) = 0.905.

Question #68
Answer is B

⎛ 35 ⎞ ⎛ 74 ⎞ ⎛ 34 ⎞ ⎛ 32 ⎞
Using all participants, S T ( 4 ) = ⎜1 − ⎟ ⎜1 − ⎟ ⎜1 − ⎟ ⎜1 − ⎟ = 0.41667.
⎝ 300 ⎠ ⎝ 265 ⎠ ⎝ 191 ⎠ ⎝ 157 ⎠
⎛ 15 ⎞ ⎛ 20 ⎞ ⎛ 20 ⎞ ⎛ 10 ⎞
Using only Country B, S B ( 4 ) = ⎜1 − ⎟ ⎜1 − ⎟ ⎜1 − ⎟ ⎜ 1 − ⎟ = 0.35.
⎝ 100 ⎠ ⎝ 85 ⎠ ⎝ 65 ⎠ ⎝ 45 ⎠
The difference is, S T ( 4 ) − S B ( 4 ) = 0.41667 − 0.35 = 0.0667 = 0.07.

Question #69
Answer is B

For an exponential distribution the maximum likelihood estimate of the mean is the
sample mean. We have
E ( X ) = E ( X ) = θ , Var ( X ) = Var ( X ) / n = θ 2 / n.
cv = SD( X ) / E ( X ) = [θ / n ] / θ = 1/ n = 1/ 5 = 0.447.

If the above facts are not known, the loglikelihood function can be used:
−Σx / θ
L(θ ) = θ − n e j , l (θ ) = −n ln θ − nX / θ , l '(θ ) = − nθ −1 + nX θ −2 = 0 ⇒ θˆ = X .
l "(θ ) = nθ −2 − 2nX θ −3 , I (θ ) = E[−nθ −2 + 2nX θ −3 ] = nθ −2 .
Then, Var (θˆ) = θ 2 / n.

Question #70
Answer is D

Because the total expected claims for business use is 1.8, it must be that 20% of
business users are rural and 80% are urban. Thus the unconditional probabilities of
being business-rural and business-urban are 0.1 and 0.4 respectively. Similarly the
probabilities of being pleasure-rural and pleasure-urban are also 0.1 and 0.4
respectively. Then,

C-09-08 - 24 -
μ = 0.1(1.0) + 0.4(2.0) + 0.1(1.5) + 0.4(2.5) = 2.05
v = 0.1(0.5) + 0.4(1.0) + 0.1(0.8) + 0.4(1.0) = 0.93
a = 0.1(1.02 ) + 0.4(2.02 ) + 0.1(1.52 ) + 0.4(2.52 ) − 2.052 = 0.2225
k = v / a = 4.18
Z = 1/(1 + 4.18) = 0.193.

Question #71
Answer is A

No. claims Hypothesize Observe Chi-square

d d
1 250 235 152/250 =
0.90
2
2 350 335 15 /350 =
0.64
2
3 240 250 10 /240 =
0.42
2
4 110 111 1 /110 = 0.01
5 40 47 72/40 = 1.23
6+ 10 22 122/10 =
14.40

The last column sums to the test statistic of 17.60 with 5 degrees of freedom (there
were no estimated parameters), so from the table reject at the 0.005 significance level.

Question #72
Answer is C

In part (ii) you are given that μ = 20 . In part (iii) you are given that a = 40 . In part (iv)
you are given that v = 8, 000 . Therefore, k = v / a = 200 . Then,
800(15) + 600(10) + 400(5) 100
X= =
1800 9
1800
Z= = 0.9
1800 + 200
Pc = 0.9(100 / 9) + 0.1(20) = 12.

C-09-08 - 25 -
Question #73
Answer is C

⎛ 1 ⎞⎛ 1 ⎞
Pr( X > 30, 000) = S (30, 000) = ⎜ 1 − ⎟ ⎜1 − ⎟ = 20 / 27 = 0.741.
⎝ 10 − 2 / 2 ⎠ ⎝ 7 − 2 / 2 ⎠

Question #74
DELETED

Question #75
Answer is D

∞ x ∞ y +δ
E( X ) = ∫ e − ( x −δ ) /θ dx = ∫ e− y /θ dx = θ + δ
δ θ 0 θ
∞ x 2
∞ y 2 + 2 yδ + δ 2
E( X 2 ) = ∫ e − ( x −δ ) /θ dx = ∫ e− y /θ dx = 2θ 2 + 2θδ + δ 2 .
δ θ 0 θ
Both derivations use the substitution y = x − δ and then recognize that the various
integrals are requesting moments from an ordinary exponential distribution. The
method of moments solves the two equations
θ + δ = 10
2θ 2 + 2θδ + δ 2 = 130.6
producing δˆ = 4.468.

It is faster to do the problem if it is noted that X = Y + δ where Y has an ordinary

exponential distribution. Then E ( X ) = E (Y ) + δ = θ + δ and Var ( X ) = Var (Y ) = θ 2 .

Question #76
Answer is D

The posterior density is proportional to the product of the probability of the observed
value and the prior density. Thus, π (θ | N > 0) ∝ Pr( N > 0 | θ )π (θ ) = (1 − e −θ )θ e−θ .

∞ 1 1
The constant of proportionality is obtained from ∫
0
θ e−θ − θ e−2θ dθ = − = 0.75.
12 22
The posterior density is π (θ | N > 0) = (4 / 3)(θ e − θ e −2θ ). −θ

Then,

C-09-08 - 26 -
 ∞ ∞
Pr( N 2 > 0 | N1 > 0) = ∫ Pr( N 2 > 0 | θ )π (θ |N1 > 0)dθ = ∫ (1 − e −θ )(4 / 3)(θ e −θ − θ e −2θ )dθ
0 0

4 ∞ −θ 4⎛ 1 2 1 ⎞
= ∫
3 0
θ e − 2θ e−2θ + θ e−3θ dθ = ⎜ 2 − 2 + 2 ⎟ = 0.8148.
3 ⎝1 2 3 ⎠

Question #77
Answer is E

The interval is centered at 2.09 and the plus/minus term is 0.46 which must equal 1.96σˆ
and so σˆ = 0.2347. For the log-transformed interval we need
φ = e1.96(0.2347) / 2.09 = 1.2462 . The lower limit is 2.09/1.2462 = 1.68 and the upper limit is
2.09(1.2462) = 2.60.

Question #78
Answer is B

From item (ii), μ = 1000 and a = 50. From item (i), v = 500. Therefore, k = v/a = 10 and
Z = 3/(3+10) = 3/13. Also, X = (750 + 1075 + 2000) / 3 = 1275. Then
Pc = (3 /13)(1275) + (10 /13)(1000) = 1063.46.

Question #79
Answer is C

1 − x /100 1
f ( x) = p e + (1 − p) e − x /10,000
100 10, 000
L(100, 200) = f (100) f (2000)
⎛ pe −1 (1 − p)e−0.01 ⎞ ⎛ pe−20 (1 − p)e−0.2 ⎞
=⎜ + ⎟⎜ + ⎟
⎝ 100 10, 000 ⎠ ⎝ 100 10, 000 ⎠

Question #80
Key: C

Model Solution:

For a binomial random variable with n = 100 and p = q70 = 0 .03318, simulate number of
deaths:

b g
i = 0: 1 − p
100
bg bg
= 0.03424 = f 0 = F 0

C-09-08 - 27 -
 bg
. > F 0 , continue
Since 018

i = 1: f b1g = f b0gbngb pg / b1 − pg
= b0.03424gb100gb0.03318g / b0.96682g
= 011751
.

bg b g bg
F 1 = F 0 + f 1 = 0.03424 + 011751
. = 015175
.
. > F b1g, continue
Since 018

b g b gb g b g b g
i = 2: f 2 = f 1 n − 1 / 2 p / 1 − p
= b011751
. gb99 / 2gb0.03318 / 0.96682g
= 019962
.

b g bg b g
F 2 = F 1 + f 2 = 015175
. + 019962
. = 0.35137
. < F b2g, number of claims = 2 , so claim amount = 20.
Since 018

Question # 81
Answer: C

Which distribution is it from?

0.25 < 0.30, so it is from the exponential.

Given that Y is from the exponential, we want

b g bg
Pr Y ≤ y = F y = 0.69
−y
1 − e θ = 0.69
1 − e − 0.5 = 0.69 since mean = 0.5
y

−y
0.5
b g
= ln 1 − 0.69 = −1171
.
y = 0.5855

Question #82
Key: B

If you happen to remember this distribution from the Simulation text (example 4d in third
edition), you could use:

C-09-08 - 28 -
 ⎛ log (1 − u ) ⎞ log 0.95
n = Int ⎜ ⎟ + 1 = Int +1 = 0 +1 = 1
⎝ log q ⎠ log 0.1

For mere mortals, you get the simulated value of N from the definition of the inverse
transformation method:
f(1) = F(1) = 0.9
0.05 ≤ 0.9 so n = 1

log(
1 1−v1 ) 1
x1 = =− log 0.7 = 35.67
λ 0.01

The amount of total claims during the year = 35.67

Question #83
Key: B

F(0) = 0.8
F(t) = 0.8 + 0.00005(t-1000), 1000 ≤ t ≤ 5000

0.75 ⇒ 0 found since F ( 0 ) ≥ 0.75

0.85 ⇒ 2000 found since F ( 2000 ) = 0.85
Average of those two outcomes is 1000.

Question #84
Key: A

⎧c ( 400 − x ) x < 400

B=⎨
⎩ 0 x ≥ 400
100 = E ( B ) = ci400 − cE ( X ∧ 400 )
⎛ 300 ⎞
= ci400 − ci300 ⎜1 − ⎟
⎝ 300 + 400 ⎠
⎛ 4⎞
= c ⎜ 400 − 300i ⎟
⎝ 7⎠
100
c= = 0.44
228.6

Question #85

C-09-08 - 29 -
Key: C

Let N = # of computers in department

Let X = cost of a maintenance call
Let S = aggregate cost

Var ( X ) = ⎡⎣Standard Deviation ( X ) ⎤⎦ = 2002 = 40,000

2

( )
E X 2 = Var ( X ) + ⎡⎣ E ( X ) ⎤⎦
2

= 40,000 + 802 = 46, 400

E ( S ) = N × λ × E ( X ) = N × 3 × 80 = 240 N

( )
Var ( S ) = N × λ × E X 2 = N × 3 × 46, 400 = 139, 200 N

We want 0.1 ≥ Pr ( S > 1.2 E ( S ) )

⎛ S − E (S ) 0.2 E ( S ) ⎞ 0.2 × 240 N
≥ Pr ⎜ > ⎟⇒ ≥ 1.282 = Φ ( 0.9 )
⎝ 139, 200 N 139, 200 N ⎠ 373.1 N
2
⎛ 1.282 × 373.1 ⎞
N ≥⎜ ⎟ = 99.3
⎝ 48 ⎠

Question #86
Key: D

The modified severity, X*, represents the conditional payment amount given that a
payment occurs. Given that a payment is required (X > d), the payment must be
uniformly distributed between 0 and c ⋅ (b − d ) .

The modified frequency, N*, represents the number of losses that result in a payment.
b−d
The deductible eliminates payments for losses below d, so only 1 − Fx ( d ) = of
b
losses will require payments. Therefore, the Poisson parameter for the modified
b−d
frequency distribution is λ ⋅ . (Reimbursing c% after the deductible affects only the
b
payment amount and not the frequency of payments).

Question #87
Key: E

C-09-08 - 30 -
 f ( x) = 0.01, 0 ≤ x ≤ 80
= 0.01 − 0.00025( x − 80) = 0.03 − 0.00025 x, 80 < x ≤ 120

( 0.03x − 0.00025x ) dx
80 120
E ( x) = ∫ 0.01x dx + ∫ 2
0 80

0.01x 2 80 0.03x 2 120 0.00025 x3 120

= + −
2 0 2 80 3 80

= 32 + 120 − 101.33 = 50.66667

E ( X − 20) + = E ( X ) − ∫ x f ( x ) dx − 20(1 − ∫ f ( x )dx)
20 20
0 0

= 50.6667 −
0.01x 2 20
2 0
(
− 20 1 − 0.01x
20
0
)
= 50.6667 − 2 − 20(0.8) = 32.6667

32.6667
Loss Elimination Ratio = 1 − = 0.3553
50.6667

Question #88
Key: B

First restate the table to be CAC’s cost, after the 10% payment by the auto owner:

Towing Cost, x p(x)

72 50%
90 40%
144 10%

Then E ( X ) = 0.5*72 + 0.4*90 + 0.1*144 = 86.4

( )
E X 2 = 0.5*722 + 0.4*902 + 0.1*1442 = 7905.6
Var ( X ) = 7905.6 − 86.42 = 440.64
Because Poisson, E ( N ) = Var ( N ) = 1000
E ( S ) = E ( X ) E ( N ) = 86.4*1000 = 86, 400
Var( S ) = E ( N )Var( X ) + E ( X ) 2 Var( N ) = 1000* 440.64 + 86.42 *1000 = 7,905,600
⎛ S − E ( S ) 90,000 − 86, 400 ⎞
Pr( S > 90,000) + Pr ⎜ > ⎟⎟ = Pr( Z > 1.28) = 1 − Φ (1.28) = 0.10
⎜ Var( S )
⎝ 7,905,600 ⎠

Since the frequency is Poisson, you could also have used

( )
Var ( S ) = λ E X 2 = (1000 )( 7905.6 ) = 7,905,600
That way, you would not need to have calculated Var ( X ) .

C-09-08 - 31 -
Question #89
Key: C

LER =
E ( X ∧ d ) θ 1− e
=
(
− d /θ
)
= 1 − e− d /θ
E(X ) θ
Last year 0.70 = 1 − e− d /θ ⇒ − d = θ log 0.30
Next year: − d new = θ log(1 − LER new )
4
Hence θ log (1 − LER new ) = − d new = θ log 0.30
3
log (1 − LER new ) = −1.6053
(1 − LER new ) = e−1.6053 = 0.20
LER new = 0.80

Question # 90
Answer: E

The distribution of claims (a gamma mixture of Poissons) is negative binomial.

b g d c hi b g
E N = E∧ E N Λ = E∧ Λ = 3
Var b N g = E dVar c N Λ hi + Var d E c N Λ hi
∧ ∧

= E b Λ g + Var b Λ g = 6
∧ ∧
rβ = 3
rβ b1 + β g = 6
b1 + β g = 6 / 3 = 2 ; β = 1
rβ = 3
r=3

b g = 0125
p0 = 1 + β
−r
.
rβ
p1 =
b g = 01875
1+ β
r +1
.

Pr obbat most 1g = p + p0 1
= 0.3125

C-09-08 - 32 -
Question # 91
Answer: A

bg b g b g
E S = E N × E X = 110 × 1101
, = 121110
,

Var b S g = E b N g × Var b X g + E b X g × Var b N g

2

= 110 × 702 + 11012 × 750

= 909,689,750

bg
Std Dev S = 30,161

b g c b
Pr S < 100,000 = Pr Z < 100,000 − 121110
, g
/ 30,161 where Z has standard normal h
distribution
= Pr b Z < −0.70g = 0.242

Question # 92
Answer: C

Let N = number of prescriptions then S = N × 40

n fN n bg FN nbg bg
1− FN n
0 0.2000 0.2000 0.8000
1 0.1600 0.3600 0.6400
2 0.1280 0.4880 0.5120
3 0.1024 0.5904 0.4096

∞
b g
E N = 4 = ∑ 1− F j
j =0
c b gh
∞
b
E S − 80 g + b
= 40 × E N − 2 g +
b gh
= 40 × ∑ 1 − F j
j =2
c
L ∞ O
= 40 × M∑ c1 − F b j gh − ∑ c1 − F b j ghP
1

MNj =0 j =0 PQ
= 40b4 − 144
. g = 40 × 2.56 = 102.40

C-09-08 - 33 -
 ∞
b
E S − 120 g + b
= 40 × E N − 3 g + c b gh
= 40 × ∑ 1 − F j
j =3

L ∞ O
= 40 × M∑ c1 − F b j gh − ∑ c1 − F b j ghP
2

MN j =0 j =0 PQ
= 40b4 − 1952
. g = 40 × 2.048 = 8192 .

Since no values of S between 80 and 120 are possible,

b
E S − 100 g = b120 − 100g × E bS − 80g 120
+
+ b100 − 80g × E b S − 120g
+ +
= 92.16

Alternatively,
∞
b
E S − 100 g + b g bg bg
= ∑ 40 j − 100 f N j + 100 f N 0 + 60 f N 1 + 20 f N 2
j =0
bg bg
b g
(The correction terms are needed because 40 j − 100 would be negative for j = 0, 1, 2;
we need to add back the amount those terms would be negative)

∞ ∞

j=0
bg j=0
b g b gb
= 40∑ j × f N j − 100∑ f N j + 100 0.200 + 016
. 60 + 0128
. 20 g b gb g b gb g
b g
= 40 E N − 100 + 20 + 9.6 + 2.56
= 160 − 67.84 = 92.16

Question #93
Answer: E

Method 1:

In each round,
N = result of first roll, to see how many dice you will roll
X = result of for one of the N dice you roll
S = sum of X for the N dice

b g b g
E X = E N = 35.
b g b g
Var X = Var N = 2.9167

E b S g = E b N g * E b X g = 12.25
Var b S g = E b N gVar b X g + Var b N g E b X g
2

C-09-08 - 34 -
 b gb
= 35
. 2.9167 + 2.9167 35
.g b gb g 2

= 45.938

Let S1000 = the sum of the winnings after 1000 rounds

b g
E S1000 = 1000 *12.25 = 12,250
Stddevb S g = sqrt b1000 * 45.938g = 214.33
1000

After 1000 rounds, you have your initial 15,000, less payments of 12,500, plus winnings
of S1000 .

Since actual possible outcomes are discrete, the solution tests for continuous outcomes
greater than 15000-0.5. In this problem, that continuity correction has negligible impact.

b
Pr 15000 − 12500 + S1000 > 14999.5 = g
cb g b
= Pr S1000 − 12250 / 214.33 > 14999.5 − 2500 − 12250 / 214.33 = g h
= 1 − Φ 117 b g
. = 012
.

Method 2

Realize that you are going to determine N 1000 times and roll the sum of those 1000
N’s dice, adding the numbers showing.

Let N1000 = sum of those N’s

b g b g b gb g
E N1000 = 1000 E N = 1000 35
. = 3500
b g
Var N1000 = 1000Var ( N ) = 2916.7
b g b g b g b gb g
E S1000 = E N1000 E X = 3500 35
. = 12.250
Var b S g = E b N g Var b X g + Var b N g E b X g
2
1000 1000 1000

= b3500gb2.9167g + b2916.7gb35
. g = 45.938
2

b g
Stddev S1000 = 214.33

Now that you have the mean and standard deviation of S1000 (same values as method
1), use the normal approximation as shown with method 1.

C-09-08 - 35 -
Question #94
Answer: B

FG IJ b
H
pk = a +
K k
pk −1

0.25 = ba + bg × 0.25 ⇒ a + b = 1

F bI F bI
= G a + J × 0.25 ⇒ G1 − J × 0.25 = 01875
01875
.
H 2K H 2K .

b = 0.5

a = 0.5

FG 0.5 IJ
H
p3 = 0.5 +
3
× 01875
.
K= 0125
.

Question #95
Answer: E

β = mean = 4; b g
pk = β k / 1 + β
k +1

n b
P N =n g x bg
f b1g x bg
f b2g x bg
f b 3g x
0 0.2 0 0 0 0
1 0.16 1 0.25 0 0
2 0.128 2 0.25 0.0625 0
3 0.1024 3 0.25 0.125 0.0156

bg
f b k g x = probability that, given exactly k claims occur, that the aggregate amount is x.
f b g b x g = f b x g; the claim amount distribution for a single claim
1

f b g b x g = ∑ e f b g b j gj x f b x − j g
x
k k −1

j =0

bg b
bg g
f s x = ∑ P N = k × f b k g x ; upper limit of sum is really ∞ , but here with smallest
x

k =0

possible claim size 1, f b g b x g = 0 for k > x k

C-09-08 - 36 -
fsb0g = 0.2
fsb1g = 016
. * 0.25 = 0.04
fsb2g = 016
. * 0.25 + 0128
. * 0.0625 = 0.048
fsb3g = 016
. * 0.25 + 0128
. * 0125
. + 01024
. * 0.0156 = 0.0576

F b3g = 0.2 + 0.04 + 0.048 + 0.0576 = 0.346

s

C-09-08 - 37 -
Question #96
Answer: E

Let L = incurred losses; P = earned premium = 800,000

FG L IJ
Bonus = 015
H
. × 0.60 −
P K
× P if positive

= 015 b g
. × 0.60 P − L if positive
= 015 b g
. × 480,000 − L if positive
. × c480,000 − b L ∧ 480,000gh
= 015
E (Bonus) = 0.15 (480,000– E c L ∧ 480,000gh
From Appendix A.2.3.1
= 0.15{480,000 – [500,000 × (1 – (500,000 / (480,000+500,000))]}
= 35,265

Question # 97
Key: D

Model Solution:

Severity Severity
after increase after increase and
deductible
60 0
120 20
180 80
300 200

Expected payment per loss = 0.25 × 0 + 0.25 × 20 + 0.25 × 80 + 0.25 × 200

= 75

Expected payments = Expected number of losses × Expected payment per loss

= 75 × 300
= 22,500

Question # 98
Key: A

Model Solution:

E (S) = E (N) E (X) = 50 × 200 = 10,000

C-09-08 - 38 -
 b g b g b g b g Varb Ng
Var S = E N Var X + E X
2

= b50gb400g + d200 ib100g 2

= 4,020,000

b
Pr S < 8,000 = Pr Z < g FG 8,000 − 10,000IJ
H 4,020,000 K
= Prb Z < −0.998g ≅ 16%

Question #99
Key: B

Model Solution:

Let S denote aggregate losses before deductible.

E S = 2 × 2 = 4, since mean severity is 2.

e −2 20
bg
fS 0 =
0!
= 01353
. , since must have 0 number to get aggregate losses = 0.

b g FGH e 1!2 IJK FGH 13IJK = 0.0902, since must have 1 loss whose size is 1 to get aggregate losses = 1.
−2
fS 1 =

E b S ∧ 2g = 0 × f b0g + 1 × f + 2 × c1 − f b0g − f b1gh

S S S S

= 0 × 01353 . + 1 × 0.0902 + 2 × b1 − 01353

. − 0.0902g
= 16392
.

b
E S −2 g +
= E S − E S ∧2
= 4 − 16392
.
= 2.3608

Question #100
Key: C

Model Solution:

Limited expected value =

z c b gh zd
1000 1000
1 − F x dx = i d
0.8e−0.02 x + 0.2e−0.001x dx = −40e−0.02 x − 200e−0.001x i 1000

0
= 40 + 126.4
0 0

C-09-08 - 39 -
 = 166.4

Question #101
Key: B

Model Solution:

Mean excess loss =

b g b g
E X − E X ∧ 100
b g
1 − F 100
331 − 91
= = 300
0.8

b g b g b g
E X = E X ∧ 1000 since F 1000 = 10
.

Question #102
Key: E

Model Solution:

Expected insurance benefits per factory = E X − 1 + b g

= 0.2 × 1 + 01
. × 2 = 0.4.

Insurance premium = (1.1) (2 factories) (0.4 per factory) = 0.88.

Let R = retained major repair costs, then

bg
f R 0 = 0.42 = 016
.
f b1g = 2 × 0.4 × 0.6 = 0.48
R

f b2g = 0.6 = 0.36

R
2

Dividend = 3 − 0.88 − R − 015 b gb g

. 3 , if positive
= 167
. − R , if positive

b g b gb
E Dividend = 016 . − 0 + 0.48 167
. 167 g b gb g b gb g
. − 1 + 0.36 0 = 0.5888

b g
. − R is
[The (0.36)(0) term in the last line represents that with probability 0.36, 167
negative so the dividend is 0.]

C-09-08 - 40 -
Question #103
Key: A

Model Solution:

αθ 4α
E X = = = 8 ⇒ 4α = 8α − 8
α −1 α −1
α =2

Fθ I
F b6g = 1 − G J
α
F 4I
=1− G J
2

H 6K H 6K
= 0.555

bg bg
s 6 = 1 − F 6 = 0.444

Test Question: 104

Key: C

Ε Ν = ΕΛ Ε Ν Λ = ΕΛ Λ = 2
Var Ν = Ε Λ Var Ν Λ + VarΛ E N Λ
= Ε Λ Λ + VarΛ Λ = 2 + 2 = 4

Distribution is negative binomial (Loss Models, 3.3.2)

Per supplied tables

mean = rβ = 2
b g
Var = rβ 1 + β = 4
b1 + β g = 2
β =1
rβ = 2
r=2

From tables

p3 =
b gb g
r r +1 r + 2 β3
=
b gb gb g
2 3 4 13
=
4
= 0125
b g
3! 1 + β
r +3
3! 2 5
32
.

1000 p3 = 125

C-09-08 - 41 -
Test Question: 105
Key: A

Using the conditional mean and variance formulas:

E N = EΛ N Λ c h
d c hj
Var N = VarΛ E N Λ + E Λ Var N Λ d c hj
Since N, given lambda, is just a Poisson distribution, this simplifies to:

E N = EΛ Λ bg
bg
Var N = VarΛ Λ + E Λ Λ bg
We are given that E N = 0.2 and Var N = 0.4 , subtraction gives Var Λ = 0.2 bg
Test Question: 106
Key: B

N = number of salmon
X = eggs from one salmon
S = total eggs.
E(N) = 100t
Var(N) = 900t

bg b gb g
E S = E N E X = 500t
Var b S g = E b N gVar b X g + E b X gVar b N g = 100t ⋅ 5 + 25 ⋅ 900t = 23,000t
2

Pb S > 10,000g = PG
F S − 500t > 10,000 − 500t IJ =.95 ⇒
H 23,000t 23,000t K
10,000 − 500t = −1645
. ⋅ 23000 t = −250 t
40 − 2t = − t

2 d ti 2
− t − 40 = 0
1 ± 1 + 320
t= = 4.73
4
t = 22.4
round up to 23

C-09-08 - 42 -
Test Question: 107
Key: C

X = losses on one life

b gb g b gb g b gb g
E X = 0.3 1 + 0.2 2 + 01
. 3
=1

S = total losses
E S = 3E X = 3
b g c b gh
E S − 1 + = E S − 1 1 − Fs 0
= E S − b1gc1 − f b0gh
s

= 3 − b1gd1 − 0.4 i
3

= 3 − 0.936
= 2.064

Test Question: 108

Key: C

bg
pk =
2
k
b g
p k −1

LM OP b g
2
= 0+ p k −1
N Q k

Thus an (a, b, 0) distribution with a = 0, b = 2.

Thus Poisson with λ = 2 .

e −2 2 4
bg
p4 =
4!
= 0.09

Test Question: 109

Key: B

By the memoryless property, the distribution of amounts paid in excess of 100 is still
exponential with mean 200.

With the deductible, the probability that the amount paid is 0 is

b g
F 100 = 1 − e −100/ 200 = 0.393.

Thus the average amount paid per loss is (0.393) (0) + (0.607) (200) = 121.4
The expected number of losses is (20) (0.8) = 16.
The expected amount paid is (16) (121.4) = 1942.

C-09-08 - 43 -
Test Question: 110
Key: E

b gb g b gb g
E N = 0.8 1 + 0.2 2 = 12
.
E N = b0.8g1 + b0.2gb4g = 16
2
.
Var b N g = 16
. − 12
. = 016
2
.
E X = 70 + 100 = 170
bg
Var X = E X 2 − E X b 2
g
= 7000 + 100,000 − 1702 = 78,100
E S = E N E X = 12 . b170g = 204
VarbSg = E N Varb X g + E X Varb N g = 12
. b78,100g + 170 b016
. g = 98,344
2 2

Std dev b S g = 98,344 = 313.6

So B = 204 + 314 = 518

Test Question: 111

Key: E

Treat as three independent Poisson variables, corresponding to 1, 2 or 3 claimants.

rate1 = 6
LM= 1 × 12OP
N 2 Q
rate 2 = 4
rate 3 = 2

Var1 = 6
Var2 = 16 = 4 × 22
Var3 = 18

total Var = 6 + 16 + 18 = 40, since independent.

Alternatively,

12 2 2 32 10
d i
E X2 =
2
+ + =
3 6 3

For compound Poisson, Var S = E N E X 2

b gFGH 103IJK = 40
= 12

C-09-08 - 44 -
Test Question: 112
Key: A

N = number of physicians E(N) = 3 Var (N) = 2

X = visits per physician E(X) = 30 Var (X) = 30
S = total visits

E(S) = E (N) E(X) = 90

Var(S) = E(N) Var(X) + E2 X Var(N) = bg
= 3 ⋅ 30 + 900 ⋅ 2 = 1890
Standard deviation (S) = 43.5

FG S − 90 > 119.5 − 90IJ = 1 − Φb0.68g Course 3: November 2000

Pr(S>119.5) = Pr
H 435. 435
. K
Test Question: 113
Key: E

b g
E N =.7
b g
Var N = 4×.2 + 9×.1−.49 = 121
.
b g
E X =2
b g
Var X = 100×.2 − 4 = 16
bg
E S = 2×.7 = 14
.
Var b S g = Eb N g Var b X g + E b X g Var b N g =
2

=.7 × 16 + 4 × 121
. = 16.04
Standard Devb S g = 4
Eb S g + 2 × Standard Devb S g = 14 . + 2 × 4 = 9.4
Since there are no possible values of S between 0 and 10,
Prb S > 9.4g = 1 − Prb S = 0g
= 1−.7−.2×.82 −.1×.83 =.12

Test Question: 114

Key: A

bg
P0 = 1
z
5 −λ
5 0 e dλ = 1
5 d− e i −λ
5

0
= 1
5 d1 − e i = 01987
−5
.

bg
P1 =
5
z −λ
5 0 λe dλ
1
= 1
5 d − λe −λ
− e− λ i
5

0
= 1
5 d1 − 6e i = 01919
−5
.

b g
P N ≥ 2 = 1−.1987−.1919 =.6094

C-09-08 - 45 -
Test Question: 115
Key: D

Let X be the occurrence amount, Y = max(X-100, 0) be the amount paid.

E[X] = 1,000
b
Var[X] = 1,000
2
g
P(X>100) = exp(-100/1,000) = .904837

The distribution of Y given that X>100, is also exponential with mean 1,000
(memoryless property).

So Y is
RS 0 with prob .095163
Texponential mean 1000 with prob .904837
E Y =.095163 × 0+.904837 × 1,000 = 904.837
b
E Y 2 =.095163 × 0+.904837 × 2 × 1,000 = 1,809,675 g 2

b g
Var Y = 1,809,675 − 904.837 = 990,944
2

Alternatively, think of this as a compound distribution whose frequency is Bernoulli with

p = .904837, and severity is exponential with mean 1,000.
b gb
Var = Var N × E X + Var X × E N = p 1 − p 1,000,000 + p 1,000,000
2
g b g
Test Question: 116
Key: C

Expected claims under current distribution = 500

θ = parameter of new distribution
X = claims
b g
E X =θ
bonus =.5 × 500 − X ∧ 500
FG
E (claims + bonus) = θ +.5 500 − θ 1 −
FG θ IJ IJ = 500
H H 500 + θ K K

θ−
θFG 500 IJ = 250
2 H 500 + θ K
2b500 + θ gθ − 500θ = 250b500 + θ g ⋅ 2
1000θ + θ 2 ⋅ 2 − 500θ = 2 × 250 × 500 + 500θ
θ = 250 × 500 = 354

C-09-08 - 46 -
Test Question: 117
Key: A

b gb g b gb g
UDD ⇒ l21 = 0.8 53,488 + 0.2 17,384 = 46,267.2

z z bb g
∞ ∞
S 21 + t
b g
Mrl 21 = e21 = t p21
S 21 g dt
0 0

= ∑ areas
= 2.751 + 1228
. + 0.361 + 0.072
= 4.412
b gb g
S X i S 21

1.000

0.375731
.115611
.028897 Ο
1 Ο2
Ο
3 Ο
4
xi
21 25 30 35 40

Test Question: 118

Key: D

μ ′n1 = E N = 25
μ x 2 = Var X = 675
μ N 2 = Var N = 25
E X = 50

E S = E X E N = 25 × 50 = 1250

2
Var S = E N Var X + Var N E X
= 25 × 675 + 25 × 2500 = 79,375

C-09-08 - 47 -
Standard Deviation S = 79,375 = 28174
.

b g b g
Pr S > 2000 = Pr S − 1250 / 28174 b
. > 2000 − 1250 / 28174 g b g
. = 1 − Φ 2.66

Test Question: 119

Key: C

bgbg
E X = 2000 1! / 1! = 2000

F 2000IJ × LM1 − 2000 OP = 2000 × FG1 − 2IJ = 2000 × 3 = 1200

E X ∧ 3000 = G
H 1 K N b3000 + 2000g Q H 5K 5

So the fraction of the losses expected to be covered by the reinsurance

2000 − 1200
is = 0.4.
2000
The expected ceded losses are 4,000,000 ⇒ the ceded premium is 4,400,000.

Test Question: 120

Key: E

X 2002 = 105
. × X 2001

F x IJ = 1 − LM
FG
2000 OP 2

so:
H 105
. K Nbx gQ
2002
. + 2000
2002 / 105

= 1− M
L 2100 OP 2

Nx
2002 + 2100 Q
This is just another Pareto distribution with α = 2,θ = 2100.
E X 2002 = 2100.

and

C-09-08 - 48 -
 FG 2100IJ × LM1 − FG 2100 IJ OP
E X 2002 ∧ 3000 =
H 1 K MN H b3000 + 2100gK PQ
= 2100 × M
L 3000 OP = 1235
N 5100 Q
So the fraction of the losses expected to be covered by the reinsurance is
2100 − 1235
= 0.412.
2100

The total expected losses have increased to 10,500,000, so

C2002 = 11
. × 0.412 × 10,500,000 = 4,758,600

C2002 4,758,600
And = = 108
.
C2001 4,400,000

Question #121
Key: B

⎛ω ⎞
2
For Ω, 0.4 = F (ω ) = ⎜ ⎟
⎝ 80 ⎠
ω
0.6325 =
80
ω = 50.6

t t
For T(0) using De Moivre, 0.7 = F ( t ) = =
ω 50.6
t = ( 0.7 )( 50.6 ) = 35.42

Question #122
Key: C

1.8
E [ N ] = mq = 1.8 ⇒ q = = 0.6
3
x fN ( x) FN ( x )
0 0.064 0.064
1 0.288 0.352
2 0.432 0.784
3 0.216 1.000

C-09-08 - 49 -
First: 0.432 < 0.7 < 0.784 so N = 2. Use 0.1 and 0.3 for amounts
Second: 0.064 < 0.1 < 0.352 so N = 1 Use 0.9 for amount
Third: 0.432 < 0.5 < 0.784 so N = 2 Use 0.5 and 0.7 for amounts

Discrete uniform ⇒ FX ( x ) = 0.2 x, x = 1, 2,3, 4,5

0.4 < 0.5 < 0.6 ⇒ x1 = 3
0.6 < 0.7 < 0.8 ⇒ x2 = 4

Aggregate claims = 3+4 = 7

Question #123
Key: C

α −1 ⎤
θ ⎡ ⎛ θ ⎞ 2000 x
E ( X ∧ x) = ⎢1 − ⎥=
α − 1 ⎢⎣ ⎜⎝ x + θ ⎟⎠ ⎥⎦ x + 2000

X E ( X ∧ x)
∞ 2000
250 222
2250 1059
5100 1437

0.75 ( E ( X ∧ 2250 ) − E ( X ∧ 250 ) ) + 0.95 ( E ( X ) − E ( X ∧ 5100 ) )

0.75 (1059 − 222 ) + 0.95 ( 2000 − 1437 ) = 1162.6

The 5100 breakpoint was determined by when the insured’s share reaches 3600:

3600 = 250 + 0.25 (2250 – 250) + (5100 – 2250)

Question #124
Key: D

Since each time the probability of a heavy scientist is just half the probability of a
success, the distribution is binomial with q = 0.6 × 0.5 = 0.3 and m = 8.

f ( 2 ) = ( 8 × 7 / 2 ) × ( 0.3 ^ 2 ) × ( 0.7 ^ 6 ) = 0.30

C-09-08 - 50 -
Question #125
Key: A

Let N1, N 2 denote the random variable for # of claims for Type I and II in 2 years
X 1, X 2 denote the claim amount for Type I and II
S1 = total claim amount for type I in 2 years
S2 = total claim amount for Type II at time in 2 years
S = S1 + S2 = total claim amount in 2 years
{S1} → compound poisson λ1 = 2 × 6 = 12 X 1 ∼ U ( 0, 1)
{S2} → compound poisson λ2 = 2 × 2 = 4 X 2 ∼ U ( 0, 5 )

E ( N1 ) = Var ( N1 ) = 2 × 6 = 12
E ( S1 ) = E ( N1 ) E ( X 1 ) = (12 )( 0.5 ) = 6
Var ( S1 ) = E ( N1 )Var ( X 1 ) + Var ( N1 ) ( E ( X 1 ) )
2

= (12 )
(1 − 0 ) +
(12 )( 0.5)2
12
=4

E ( N 2 ) = Var ( N 2 ) = 2 × 2 = 4
With formulas corresponding to those for S1 ,
5
E ( S2 ) = 4 × = 10
2

Var ( S2 ) = 4×
( 5 − 0)
2
+4
⎛5⎞
2
= 33.3
⎜ ⎟
12 ⎝2⎠
E ( S ) = E ( S1 ) + E ( S2 ) = 6 + 10 = 16
Since S1 and S2 are independent,
Var ( S ) = Var ( S1 ) + Var ( S 2 ) = 4 + 33.3 = 37.3
⎛ S − 16 2 ⎞
Pr ( S > 18 ) = Pr ⎜ > = 0.327 ⎟
⎝ 39.3 37.3 ⎠
Using normal approximation
Pr ( S > 18 ) = 1 − Φ ( 0.327 )
= 0.37

C-09-08 - 51 -
Question #126
Key: C

Let X be the loss random variable,

So ( X − 5 )+ is the claim random variable.
10
E(X ) = = 6.6
2.5 − 1
⎛ 10 ⎞ ⎡ ⎛ 10 ⎞
2.5−1
⎤
E ( X ∧ 5) = ⎜ ⎟ ⎢1 − ⎜ ⎟ ⎥
⎝ 2.5 − 1 ⎠ ⎢⎣ ⎝ 5 + 10 ⎠ ⎥⎦
= 3.038

E ( X − 5)+ = E ( X ) − E ( X ∧ 5)
= 6.6 − 3.038
= 3.629

Expected aggregate claims = E ( N ) E ( X − 5 )+

= ( 5 )( 3.629 )
= 18.15

Question #127
Key: B

A Pareto (α = 2, θ = 5 ) distribution with 20% inflation becomes Pareto with

α = 2, θ = 5 ×1.2 = 6
6
In 2004, E ( X ) = =6
2 −1
6 ⎛ ⎛ 6 ⎞ ⎞
2 −1
E ( X ∧ 10 ) = ⎜1 − ⎜ ⎟ ⎟ = 3.75
2 − 1 ⎜⎝ ⎝ 10 + 6 ⎠ ⎟⎠
E ( X − 10 )+ = E ( X ) − E ( X ∧ 10 )
= 6 − 3.75 = 2.25
E ( X − 10 )+ 2.25
LER = 1 − = 1− = 0.625
E(X ) 6

C-09-08 - 52 -
Question #128
Key: A

Let X = annual claims

E ( X ) = ( 0.75 )( 3) + ( 0.15 )( 5 ) + ( 0.1)( 7 )
= 3.7
⎛ 4⎞
π = Premium = ( 3.7 ) ⎜ ⎟ = 4.93
⎝ 3⎠
Change during year = 4.93 − 3 = +1.93 with p = 0.75
= 4.93 − 5 = −0.07 with p = 0.15
= 4.93 − 7 = −2.07 with p = 0.10

Since we start year 1 with surplus of 3, at end of year 1 we have 4.93, 2.93, or 0.93
(with associated probabilities 0.75, 0.15, 0.10).

We cannot drop more then 2.07 in year 2, so ruin occurs only if we are at 0.93 after 1
and have a drop of 2.07.

Prob = ( 0.1)( 0.1) = 0.01

Question #129
Key: E

Q ≥ P since in Q you only test at intervals; surplus below 0 might recover before the
next test. In P, ruin occurs if you are ever below 0.
R ≥ P since you are less likely to have surplus below 0 in the first N years (finite
horizon) than forever.

Add the inequalities

Q + R ≥ 2P

Also (why other choices are wrong)

S ≥ Q by reasoning comparable to R ≥ P . Same testing frequency in S and Q, but Q
tests forever.
S ≥ R by reasoning comparable to Q ≥ P . Same horizon in S and R; R tests more
frequently.
S ≥ P P tests more frequently, and tests forever.

C-09-08 - 53 -
Question #130
Key: E

1 ∞
4
⎡1 ⎤
E (W ) = ∫ ∑ 2i Pr ( N = i λ ) d λ ⎢ 4 is the density of λ on [ 0, 4] .⎥
4 0 i =0 ⎣ ⎦
4
= ∫ P (2 λ ) dλ
1
[see note]
40
4
1 λ ( 2−1)
[ using formula from tables for the pgf of the Poisson ]
4 ∫0
= e dλ

=
1 λ 4 1 4
e = e −1
4 0 4
( )
= 13.4

∞
Note: the probability generating function (pgf) is P ( Z ) = ∑ pk Z k so the integrand is
k =0

P ( 2 ) , or in this case P ( 2 λ ) since λ is not known.

Alternatively,

1 ∞ i
4
E (W ) = ∫ ∑ 2 Pr ( N = i λ ) d λ
4 0 i =0
1 4 ∞ 2i e− λ λ i
=
4 ∫0 ∑
i =0 i!
dλ

e − λ ( 2λ )
∞ i
1
∫0 ∑
4
= dλ
4 i =0 i!
e −2λ ( 2λ ) e −2λ ( 2λ )
∞ i i
We know ∑ i!
= 1 since
i!
is f ( i ) for a Poisson with mean Z λ
i =0

e − λ ( 2λ ) e−λ
∞ i
so ∑ = −2λ = eλ
i =0 i! e
1
Thus E (W ) = ∫ eλ d λ
4

4 0

1 λ 1 4
( )
4
= e = e −1
4 0 4
= 13.4

C-09-08 - 54 -
Question #131
Key: E

E ( S ) = λ E [ X ] = 2 / 3 (1/ 4 + 2 / 4 + 3/ 2 ) = 2 / 3 × 9 / 4 = 3/ 2
Var ( S ) = λ E ⎡⎣ X 2 ⎤⎦ = 2 / 3 (1/ 4 + 4 / 4 + 9 / 2 ) = 23/ 6

( )
So cumulative premium to time 2 is 2 3/ 2 + 1.8 23/ 6 = 10 , where the expression in
parentheses is the annual premium

Times between claims are determined by −(1/ λ ) log (1 − u ) and are 0.43, 0.77, 1.37,
2.41
So 2 claims before time 2 (second claim is at 1.20; third is at 2.57)

Sizes are 2, 3, 1, 3, where only the first two matter.

So gain to the insurer is 10-(2+3) = 5

Question #132
Key: C

To get number of claims, set up cdf for Poisson:

x f(x) F(x)
0 0.135 0.135
1 0.271 0.406
2 0.271 0.677
3 0.180 0.857

0.80 simulates 3 claims.

F ( x ) = 1 − ( 500 / ( x + 500 ) ) = u , so x = (1 − u )
2 −1/ 2
500 − 500

0.6 simulates 290.57

0.25 simulates 77.35
0.7 simulates 412.87

So total losses equals 780.79

Insurer pays ( 0.80 )( 750 ) + ( 780.79 − 750 ) = 631

Question #133
Key: C

C-09-08 - 55 -
E ( X | q) = 3q, Var ( X | q) = 3q (1 − q)
1
μ = E (3q) = ∫ 3q 2qdq = 2q 3 0 = 2
1

0
1
v = E[3q (1 − q )] = ∫ 3q (1 − q )2qdq = 2q 3 − 1.5q 4 = 0.5
1

0 0
1
a = Var (3q ) = E (9q 2 ) − μ 2 = ∫ 9q 2 2qdq − 22 = 4.5q 4 − 4 = 4.5 − 4 = 0.5
1

0 0

k = v/a = 0.5/0.5 = 1
1
Z= = 0.5
1+1
The estimate is
0.5(0) + 0.5(2) = 1.

Question #134
Key: D

0.35(14) = 4.9
πˆ 0.35 = 0.1(216) + 0.9(250) = 246.6

Question #135
Key: E

At y1 = 0.9 the risk set is r1 = 7 and s1 = 1.

At y2 = 1.5 the risk set is r2 = 6 and s2 = 1.
65
Then, S10 (1.6) = = 0.7143.
76

Question #136
Key: B

Pr(claim = 250 | class1) Pr(class1)

Pr ( class1| claim = 250 ) =
Pr(claim = 250 | class1) Pr(class1) + Pr(claim = 250 | class 2) Pr(class 2)
0.5(2 / 3) 10
= = .
0.5(2 / 3) + 0.7(1/ 3) 17
E (claim | class1) = 0.5(250) + 0.3(2,500) + 0.2(60, 000) = 12,875.
E (claim | class 2) = 0.7(250) + 0.2(2,500) + 0.1(60, 000) = 6, 675.
10 7
E (claim | 250) = (12,875) + (6, 675) = 10,322.
17 17
Question #137
Key: D

C-09-08 - 56 -
L( p ) = f (0.74) f (0.81) f (0.95)
= ( p + 1)0.74 p ( p + 1)0.81p ( p + 1)0.95 p
= ( p + 1)3 (0.56943) p
l ( p ) = ln L( p ) = 3ln( p + 1) + p ln(0.56943)
3
l '( p ) = − 0.563119 = 0
p +1
3
p +1 = = 5.32747
0.563119
p = 4.32747.

Question #138
Key: E

The sample mean is 1 and therefore mq = 1.

For the smoothed empirical 33rd percentile, (1/3)(5 + 1) = 2 and the second smallest
sample item is 0. For the 33rd percentile of the binomial distribution to be 0, the
probability at zero must exceed 0.33. So, (1 − q ) m > 0.33 and then (1 − m −1 ) m > 0.33 . Trial
and error gives m = 6 as the smallest value that produces this result.

Question #139
Key: C

Let X be the number of claims.

E ( X | I ) = 0.9(0) + 0.1(2) = 0.2
E ( X | II ) = 0.8(0) + 0.1(1) + 0.1(2) = 0.3
E ( X | III ) = 0.7(0) + 0.2(1) + 0.1(2) = 0.4
Var ( X | I ) = 0.9(0) + 0.1(4) − 0.22 = 0.36
Var ( X | II ) = 0.8(0) + 0.1(1) + 0.1(4) − 0.32 = 0.41
Var ( X | III ) = 0.7(0) + 0.2(1) + 0.1(4) − 0.42 = 0.44.
μ = (1/ 2)(0.2 + 0.3 + 0.4) = 0.3
v = (1/ 3)(0.36 + 0.41 + 0.44) = 0.403333
a = (1/ 3)(0.22 + 0.32 + 0.42 ) − 0.32 = 0.006667
k = 0.403333 / 0.006667 = 60.5
50
Z= = 0.45249.
50 + 60.5
For one insured the estimate is 0.45249(17/50) + 0.54751(0.3) = 0.31810.
For 35 insureds the estimate is 35(0.31810) = 11.13.
Question #140
Key: A

C-09-08 - 57 -
For the given intervals, based on the model probabilities, the expected counts are 4.8,
3.3, 8.4, 7.8, 2.7, 1.5, and 1.5. To get the totals above 5, group the first two intervals
and the last three. The table is
Interval Observed Expected Chi-square
0—500 3 8.1 3.21
500—2498 8 8.4 0.02
2498—4876 9 7.8 0.18
4876—infinity 10 5.7 3.24
Total 30 30 6.65

Question #141
Key: E

Let Hˆ = Hˆ (t ) and vˆ = Var ˆ ( Hˆ (t )) . The confidence interval is ĤU where

U = exp(± z vˆ / Hˆ ) . Multiplying the two bounds gives
α /2

0.7(0.357) = Hˆ 2 for Hˆ = 0.49990 . Then,

Sˆ = exp(−0.49990) = 0.60659.

Question #142
Key: C

k
0.575 = Pr( N = 0) = ∫ Pr( N = 0 | θ )π (θ )dθ
0
k
k e −θ e −2θ e−2 k 1
=∫ e−θ
−k
dθ = − −k
= − +
0 1− e 2(1 − e ) 0 2(1 − e ) 2(1 − e − k )
−k

1 − e −2 k 1 + e− k
= =
2(1 − e − k ) 2
e − k = 2(0.575) − 1 = 0.15
k = 1.90.

C-09-08 - 58 -
Question #143
Key: C

1⎛ 1 1 1 1 1 1 ⎞
The sample -1 moment is ⎜ + + + + + ⎟ = 0.017094. The sample -
6 ⎝ 15 45 140 250 560 1340 ⎠
1⎛ 1 1 1 1 1 1 ⎞
2 moment is ⎜ 2 + 2 + + + + ⎟ = 0.00083484.
6 ⎝ 15 45 140 250 560 13402 ⎠
2 2 2

Then the equations are

1
0.017094 = ,
θ (τ − 1)
2
0.00083484 = 2 .
θ (τ − 1)(τ − 2)
Divide the square of the first equation by the second equation to obtain
τ −2
0.35001 = which is solved for τ = 4.33356. From the first equation,
2(τ − 1)
1
θ= = 17.55.
3.33356(0.017094)

Question #144
Key: A

For each simulation, estimate the LER and then calculate the squared difference from
the estimate, 0.125.
Simulation First claim Second Third claim LER Squared
claim difference
1 600 600 1500 0.111111 0.000193
2 1500 300 1500 0.090909 0.001162
3 1500 300 600 0.125000 0.000000
4 600 600 300 0.200000 0.005625
5 600 300 1500 0.125000 0.000000
6 600 600 1500 0.111111 0.000193
7 1500 1500 1500 0.066667 0.003403
8 1500 300 1500 0.090909 0.001162
9 300 600 300 0.250000 0.015625
10 600 600 600 0.166667 0.001736
The last column has an average of 0.002910 which is the bootstrap estimate.

Question #145
Key: B

The subscripts denote the three companies.

C-09-08 - 59 -
 50, 000 50, 000 150, 000
xI 1 = = 500, xI 2 = = 250, xII 1 = = 300
100 200 500
150, 000 150, 000 150, 000
xII 2 = = 500, xIII 1 = = 3, 000, xIII 2 = = 1, 000
300 50 150
100, 000 300, 000 300, 000 700, 000
xI = = 333.33, xII = = 375, xIII = = 1,500, x = = 538.46
300 800 200 1,300
100(500 − 333.33) 2 + 200(250 − 333.33) 2 + 500(300 − 375) 2 + 300(500 − 375) 2
+ 50(3, 000 − 1,500) 2 + 150(1, 000 − 1,500) 2
vˆ =
(2 − 1) + (2 − 1) + (2 − 1)
= 53,888,888.89
300(333.33 − 538.46) 2 + 800(375 − 538.46) 2 + 200(1,500 − 538.46) 2 − 53,888,888.89(3 − 1)
aˆ =
3002 + 8002 + 2002
1,300 −
1,300
= 157, 035.60
53,888,888.89 200
k= = 343.1635, Z = = 0.3682
157, 035.60 200 + 343.1635

Question #146
Key: D

⎛ n α ⎞⎛ m α ⎞
Let αj be the parameter for region j. The likelihood function is L = ⎜ ∏ α 11+1 ⎟ ⎜ ∏ α 22+1 ⎟ .
⎝ i =1 xi ⎠ ⎝ i =1 yi ⎠
α2 α 3α1
The expected values satisfy = 1.5 1 and so α 2 = . Substituting this in the
α2 −1 α1 − 1 2 + α1
likelihood function and taking logs produces
n
⎛ 3α1 ⎞ 2 + 4α1 m
l (α1 ) = ln L(α1 ) = n ln α1 − (α1 + 1)∑ ln xi + m ln ⎜ ⎟− ∑ ln yi
i =1 ⎝ 2 + α1 ⎠ 2 + α1 i =1
n m
n 2m 6
l '(α1 ) = − ∑ ln xi + − ∑ ln y = 0.
α1 i =1 α1 (2 + α1 ) (2 + α1 ) 2 i =1
i

C-09-08 - 60 -
Question #147
Key: D

Let K y ( x) be the contribution at x of the data point at y. It is

⎧0, x < y − 1.4
⎪ x − y + 1.4
⎪
K y ( x) = ⎨ , y − 1.4 ≤ x ≤ y + 1.4
⎪ 2.8
⎪⎩1, x > y + 1.4.
For the particular points,
4 − 3.3 + 1.4
K 2 (4) = 1, K 3.3 (4) = = 0.75, K 4 (4) = 0.5, K 4.7 (4) = 0.25. The kernel estimate is the
2.8
1 2 2 3
weighted average (1) + (0.75) + (0.5) + (0.25) = 0.53125.
8 8 8 8

Question #148
Key: B

The mean is mq and the variance is mq(1 − q) . The mean is 34,574 and so the full
credibility standard requires the confidence interval to be ±345.74 which must be 1.96
standard deviations. Thus,
345.74 = 1.96 mq (1 − q ) = 1.96 34,574 1 − q
1 − q = 0.9, q = 0.1.

Question #149
Key: A

Only the Kolmogorov-Smirnov test statistic tends toward zero as the sample size goes
to infinity. As a consequence, the critical value for the K-S statistic has the square root
of the sample size in
the denominator. For the Anderson-Darling and the Chi-square goodness-of-fit test
statistics,
the sample size appears in the numerator of the test statistics themselves. The
Schwarz Bayesian Criterion involves an adjustment to the likelihood function, which
does not go to zero as the
sample size goes to infinity.

C-09-08 - 61 -
Question #150
Key: E

The sample average is (14 + 33 + 72 + 94 + 120 + 135 + 150 + 150)/8 = 96. The model
150 1 θ 1 1502 θ − 150 11, 250
average is E ( X ∧ 150) = ∫ x dx + ∫ 150 dx = + 150 = 150 − . The
0 θ 150 θ 2θ θ θ
11, 250 11, 250 11, 250
equation to solve is 150 − = 96, = 54, θ = = 208.3.
θ θ 54

Question #151
Key: C

E ( N |1) = 5, E ( N | 2) = 8(0.55) = 4.4, μ = 0.5(5) + 0.5(4.4) = 4.7

Var ( N |1) = 5, Var ( N | 2) = 8(0.55)(0.45) = 1.98, v = 0.5(5) + 0.5(1.98) = 3.49
a = 0.5(5) 2 + 0.5(4.4) 2 − 4.7 2 = 0.09, k = 3.49 / 0.09 = 38.7778
3 7+r
Z= = 0.0718, 4.6019 = 0.0718 + 0.9282(4.7)
3 + 38.7778 3
The solution is r = 3.

Question #152
Key: A

These observations are truncated at 500. The contribution to the likelihood function is
f ( x) θ −1e − x /θ
= −500 /θ . Then the likelihood function is
1 − F (500) e
θ −1e−600 /θ θ −1e−700 /θ θ −1e −900 / θ
L(θ ) = = θ −3e −700 /θ
(e )
−500 / θ 3

l (θ ) = ln L(θ ) = −3ln θ − 700θ −1

l '(θ ) = −3θ −1 + 700θ −2 = 0
θ = 700 / 3 = 233.33.

C-09-08 - 62 -
Question #153
Key: A

For group A let the hazard rate function be the baseline function, h0(x). For group B let
the hazard rate function be h0(x)eβ. Then the partial likelihood function is
1 eβ eβ 1 e2 β
L= = . Taking logarithms and differentiating leads
11 + e β 1 + 2e β 2 + 2e β 2(1 + 2e β )(1 + e β ) 2
to
l = 2 β − ln(2) − ln(1 + 2e β ) − 2 ln(1 + e β )
2e β 2e β
l'= 2− − =0
1 + 2e β 1 + e β
2(1 + 2e β )(1 + e β ) = 2e β (1 + e β ) + 2e β (1 + 2e β )
1 + 3e β + 2e 2 β = e β + e 2 β + e β + 2e 2 β
0 = e2 β − eβ − 1
1± 1+ 4
eβ = = 1.618
2
where only the positive root can be used. Because the value is greater than 1, group B
has a higher hazard rate function than group A. Therefore, its cumulative hazard rate
must also be higher.

Question #154
Key: E

For a compound Poisson distribution, S, the mean is E ( S | λ , μ , σ ) = λ E ( X ) = λ e μ + 0.5σ

2

and the variance is Var ( S | λ , μ , σ ) = λ E ( X 2 ) = λ e 2 μ + 2σ . Then,

2

E ( S ) = E[ E ( S | λ , μ , σ )] = ∫ ∫ ∫ λe
1 1 1
μ + 0.5σ 2
2σ d λ d μ dσ
0 0 0

=∫ ∫ σ d μ dσ = ∫ (e − 1)e
1 1 1
e μ + 0.5
σ 0.5σ 2
σ dσ
2

0 0 0

= (e − 1)(e0.5 − 1) = 1.114686
v = E[Var ( S | λ , μ , σ )] = ∫ ∫ ∫ λe
1 1 1
2 μ + 2σ 2
2σ d λ d μ dσ
0 0 0

=∫ ∫ e 2 μ + 2σ σ d μ dσ = ∫ 0.5(e 2 − 1)e 2σ σ dσ
1 1 2 1 2

0 0 0

= 0.5(e 2 − 1)0.25(e 2 − 1) = 0.125(e 2 − 1) 2 = 5.1025

C-09-08 - 63 -
a = Var[ E ( S | λ , μ , σ )] = ∫ ∫∫λe
1 1 1
2 2 μ +σ 2
2σ d λ d μ dσ − E ( S ) 2
0 0 0

2 2 μ +σ 2
1 1 11
=∫ ∫ σ d μ dσ − E ( S )2 = ∫ (e 2 − 1)eσ σ dσ − E ( S )2
2
e
0 0 3 0 3

1 1
= (e 2 − 1) (e − 1) − E ( S ) 2 = (e2 − 1)(e − 1) / 6 − E ( S ) 2 = 0.587175
3 2
5.1025
k= = 8.69 .
0.587175

Question #155
Key: D
τ τ
The equations to solve are 0.4 = e − (θ /1.82) , 0.8 = e − (θ /12.66) . Taking logs yields
0.91629 = (θ /1.82)τ , 0.22314 = (θ /12.66)τ . Taking the ratio of the first equation to the
second equation gives 4.10635 = (12.66 /1.82)τ = 6.95604τ . Taking logs again,
1.41253 = 1.93961τ and then τ = 0.72825. Returning to the first (logged) equation,
0.91629 = (θ /1.82)τ , 0.88688 = θ /1.82, θ = 1.614.

Question #156
Key: C

There are n/2 observations of N = 0 (given N = 0 or 1) and n/2 observations of N = 1

(given N = 0 or 1). The likelihood function is
n/2 n/2
⎛ e− λ ⎞ ⎛ λ e− λ ⎞ λ n / 2e − nλ λ n/2
L = ⎜ −λ −λ ⎟ ⎜ −λ −λ ⎟
= = . Taking logs, differentiating
⎝ e + λe ⎠ ⎝ e + λe ⎠ (e − λ + λ e− λ ) n (1 + λ ) n
and solving provides the answer.
l = ln L = (n / 2) ln λ − n ln(1 + λ )
n n
l'= − =0
2λ 1 + λ
n(1 + λ ) − n 2λ = 0
1 − λ = 0, λ = 1.

Question #157
Key: D

The posterior density function is proportional to the product of the likelihood function
and prior density. That is, π (q |1, 0) ∝ f (1| q ) f (0 | q )π (q ) ∝ q (1 − q)q 3 = q 4 − q 5 . To get the
exact posterior density, integrate this function over its range:

C-09-08 - 64 -
 0.8
q5 q6 q 4 − q5
∫0.6 q − q dq = 5 − 6
0.8
4 5
= 0.014069 and so π (q |1, 0) = . Then,
0.6
0.014069
0.8 q 4 − q5
Pr(0.7 < q < 0.8 |1, 0) = ∫ dq = 0.5572.
0.7 0.014069

Question #158
Key: B

The cumulative hazard function for the exponential distribution is H(x) = x/θ. The
maximum likelihood estimate of θ is the sample mean, which equals (1227/15) = 81.8.
Therefore Hˆ 2 (75) = (75/81.8) = 0.917.
To calculate Hˆ (75) use the following table.
1

j 1 2 3 4 5 6
yj 11 22 36 51 69 92
sj 1 3 1 1 3 2
rj 15 14 11 10 9 6

Therefore,
1 3 1 1 3
Hˆ 1 (75) = + + + + = 0.805.
15 14 11 10 9

Thus, Hˆ 2 (75) − Hˆ 1 (75) = 0.917 − 0.805 = 0.112.

Question #159
Key: A

0(2000) + 1(600) + 2(300) + 3(80) + 4(20)

The sample mean is = 0.5066667 = μˆ = vˆ and the
3000
sample variance is
2000(0 − μˆ ) 2 + 600(1 − μˆ ) 2 + 300(2 − μˆ ) 2 + 80(3 − μˆ ) 2 + 20(4 − μˆ ) 2
= 0.6901856. Then,
2999
0.5066667
aˆ = 0.6901856 − 0.5066667 = 0.1835189, k = = 2.760842 and
0.1835189
1
Z= = 0.2659.
1 + 2.760842

C-09-08 - 65 -
Question #160
Key: E

x 1
The cdf is F ( x) = ∫ 4(1 + t ) −5 dt = −(1 + t ) −4 = 1 −
x
.
0 0 (1 + x) 4
Observation (x) F(x) compare to: Maximum
difference
0.1 0.317 0, 0.2 0.317
0.2 0.518 0.2, 0.4 0.318
0.5 0.802 0.4, 0.6 0.402
0.7 0.880 0.6, 0.8 0.280
1.3 0.964 0.8, 1.0 0.164

K-S statistic is 0.402.

Question #161
Key: D

This follows from the formula MSE (θˆ) = Var (θˆ) + [bias(θˆ)]2 . If the bias is zero, then the
mean-squared error is equal to the variance.

Question #162
Key: B

E ⎡⎣ x − d x > d ⎤⎦ is the expected payment per payment with an ordinary deductible of d

It can be evaluated (for Pareto) as
α −1
θ θ ⎡ ⎛ θ ⎞ ⎤
− ⎢1 − ⎜ ⎟ ⎥
E ( x ) − E ( x ∧ d ) α − 1 α − 1 ⎣⎢ ⎝ d + θ ⎠ ⎦⎥
=
1− F (d ) ⎡ ⎛ θ ⎞α ⎤
1 − ⎢1 − ⎜ ⎟ ⎥
⎣⎢ ⎝ d + θ ⎠ ⎦⎥
α −1
θ ⎛ θ ⎞
α − 1 ⎜⎝ d + θ ⎟⎠
= α
⎛ θ ⎞
⎜ ⎟
⎝ d +θ ⎠
d +θ
=
α −1
= d + θ in this problem, since α = 2

C-09-08 - 66 -
E ⎡⎣ x − 100 x > 100 ⎤⎦ = 5 3 E ⎡⎣ x − 50 x > 50⎤⎦
100 + θ = 5 3 ( 50 + θ )
300 + 3θ = 250 + 5θ
= θ = 25
E ⎡⎣ x − 150 x > 150 ⎤⎦ = 150 + θ
= 150 + 25
= 175

Question #163
Key: D

Let S = score
( )
E ( S ) = E E ( S θ ) = E (θ ) = 75

Var ( S ) = E ⎡⎣Var ( S θ ) ⎤⎦ + Var ⎡⎣ E ( S θ ) ⎤⎦

( )
= E 82 + Var (θ )
= 64 + 62
= 100

S is normally distributed (a normal mixture of normal distributions with constant variance

is normal; see Example 4.30 in Loss Models for the specific case, as we have here, with
a normally distributed mean and constant variance)

F ( 90 ) − F ( 65 )
Pr ob ⎡⎣ S < 90 S > 65⎤⎦ =
1 − F ( 65 )
⎛ 90 − 75 ⎞ ⎛ 65 − 75 ⎞
Φ⎜ ⎟ − Φ⎜ ⎟
= ⎝
10 ⎠ ⎝ 10 ⎠
⎛ 65 − 75 ⎞
1− Φ ⎜ ⎟
⎝ 10 ⎠

Φ (1.5 ) − Φ ( −1.0 ) 0.9332 − (1 − 0.8413) 0.7745

= = = 0.9206
1 − Φ ( −1.0 ) 1 − (1 − 0.8413) 0.8413

Note that (though this insight is unnecessary here), this is equivalent to per payment
model with a franchise deductible of 65.

C-09-08 - 67 -
Question #164
Key: B

(Referring to the number of losses, X, was a mistake. X is the random variable for the
loss amount, the severity distribution).

Losses in excess of the deductible occur at a Poisson rate

of λ * = (1 − F ( 30 ) ) λ = 0.75 × 20 = 15

70 − 25
E ( X − 30 X > 30 ) =
45
= = 60
0.75 0.75
Var ( S ) = λ * × E (( X − 30) 2
X > 30 )
( ) (
= 15E X 2 − 60 X + 900 X > 30 = 15E X 2 − 60 ( X − 30 ) − 900 X > 30 )
= 15 ( 9,000 − 60 × 60 − 900 )
= 67,500

Question #165
Key: A

S ( S − 3) + E ⎡⎣( S − 3)+ ⎤⎦ = E [ S ] − 3 + 3 f S ( 0 ) + 2 f S (1) + 1 f S ( 2 )

0 0 E [ S ] = 2 × [ 0.6 + 2 × 0.4] = 2.8
1 0 f S ( 0 ) = e−2
2 0 f S (1) = e−2 × 2 × ( 0.6 ) = 1.2e−2
3 0 e−2 22
f S ( 2 ) = e−2 × 2 ( 0.4 ) + × ( 0.6 ) = 1.52 e−2
2

2
4 1
5 2
6 3

E ⎡⎣( S − 3)+ ⎤⎦ = 2.8 − 3 + 3 × e −2 + 2 × 1.2 e −2 + 1× 1.52 e−2

= −0.2 + 6.92 e −2
= 0.7365

C-09-08 - 68 -
Question #166
Key: C

pk c
Write (i) as =c+
pk −1 k
This is an (a, b, 0) distribution with a = b = c.
Which?
1. If Poisson, a = 0, so c = 0 and b = 0
p1 = p2 = ... = 0
p0 = 0.5
pk ' s do not sum to 1. Impossible. Thus not Poisson
2. If Geometric, b = 0, so c = 0 and a = 0
By same reasoning as #1, impossible, so not Geometric.
3. If binomial, a and b have opposite signs. But here a = b, so not binomial.
4. Thus negative binomial.
a β / (1 + β ) 1
1= = =
b ( r − 1) β / (1 − β ) r − 1
so r = 2
p0 = 0.5 = (1 + β )
−r
= (1 + β )
−2

1 + β = 2 = 1.414
β = 2 − 1 = 0.414
c = a = β / (1 + β ) = 0.29

Question #167
Key: B

The variance calculation assumes independence, which should have been explicitly
stated.
E (S ) = E (N ) E ( X )
Var ( S ) = E ( N ) Var( X ) + E 2 ( X ) Var ( N )
E(N) Var ( N ) E(X ) Var ( X ) E(S) Var ( S )
P.B 30 21 300 10,000 9,000 2.19 × 106
S.B 30 27 1000 400,000 30,000 39 × 106
L.Y 30 12 5000 2,000,000 150,000 360 × 106
189,000 400 × 106
(rounded)
Standard deviation = 400 × 106 = 20,000

189,000 + 20,000 = 209,000

C-09-08 - 69 -
Question #168
Key: B

S X (150 ) = 1 − 0.2 = 0.8

f X ( y + 150 ) 0.2
fY p ( y ) = So fY p ( 50 ) = = 0.25
S X (150 ) 0.8
0.6
fY p (150 ) = = 0.75
0.8
( )
E Yp = ( 0.25 )( 50 ) + ( 0.75 )(150 ) = 125

E ⎡(Y ) ⎤ = ( 0.25 ) ( 50 ) + ( 0.75 )(150 )

2 2
p 2
= 17,500
⎢⎣ ⎥⎦

( ) ( ) ( )
Var Y p = E ⎡ Y p ⎤ − ⎡ E Y p ⎤ = 17,500 − 1252 = 1875
2 2

⎢⎣ ⎥⎦ ⎣ ⎦

Slight time saver, if you happened to recognize it:

( )
Var Y p = Var Y p − 50 ( )
since subtracting a constant does not change
variance, regardless of the distribution
But Y − 50 takes on values only 0 and 100, so it can be expressed as 100 times
p

a binomial random variable with n = 1, q = 0.75

( )
Var = 1002 (1)( 0.25 )( 0.75 ) = 1875

Question #169
Key: A

Model Solution:

X denotes the loss variable.

X1 denotes Pareto with α = 2 ; X 2 denotes Pareto with α = 4

FX (200) = 0.8 FX1 (200) + 0.2 FX 2 (200)

⎡ ⎛ 100 ⎞2 ⎤ ⎡ ⎛ 3000 ⎞ 4 ⎤
= 0.8 ⎢1 − ⎜ ⎟ ⎥ + 0.2 ⎢1 − ⎜ ⎟ ⎥
⎢⎣ ⎝ 200 + 100 ⎠ ⎥⎦ ⎢⎣ ⎝ 3000 + 200 ⎠ ⎥⎦

2 4
⎛1⎞ ⎛ 15 ⎞
= 1 − 0.8 ⎜ ⎟ − 0.2 ⎜ ⎟
⎝ 3⎠ ⎝ 16 ⎠

= 0.7566

C-09-08 - 70 -
Question #170
Key: B

Let c denote child; ANS denote Adult Non-Smoker; AS denote Adult Smoker.

33 e−3
P (3 c ) P (c ) = × 0.3 = 0.067
3!
1e −1
P ( 3 ANS ) P ( ANS ) = × 0.6 = 0.037
3!
43 e −4
P ( 3 AS ) P ( AS ) = × 0.1 = 0.020
3!

P ( AS N = 3) =
0.020
= 0.16
( 0.067 + 0.037 + 0.020 )

Question #171
Key: C

E [ S ] = E [ N ] E [ X ] = 3 × 10 = 30
Var ( S ) = E [ N ]Var ( X ) + E [ X ] Var ( N )
2

400
= 3× + 100 × 3.6 = 100 + 360 = 460
12
For 95th percentile, E [ S ] + 1.645 Var ( S ) = 30 + 460 × 1.645 = 65.28

Question #172
Key: D

1
The CDF is F ( x) = 1 −
(1 + x) 4
Observation (x) F(x) compare to: Maximum
difference
0.2 0.518 0, 0.2 0.518
0.7 0.880 0.2, 0.4 0.680
0.9 0.923 0.4, 0.6 0.523
1.1 0.949 0.6, 0.8 0.349
1.3 0.964 0.8, 1.0 0.164

The K-S statistic is the maximum from the last column, 0.680.

C-09-08 - 71 -
Critical values are: 0.546, 0.608, 0.662, and 0.729 for the given levels of significance.
The test statistic is between 0.662 (2.5%) and 0.729 (1.0%) and therefore the test is
rejected at 0.025 and not at 0.01.

Question #173
Key: E

For claim severity,

μ S = 1(0.4) + 10(0.4) + 100(0.2) = 24.4,
σ S2 = 12 (0.4) + 102 (0.4) + 1002 (0.2) − 24.42 = 1, 445.04.
For claim frequency,
μ F = r β = 3r , σ F2 = r β (1 + β ) = 12r.
For aggregate losses,
μ = μ S μ F = 24.4(3r ) = 73.2r ,
σ 2 = μ S2σ F2 + σ S2 μ F = 24.42 (12r ) + 1, 445.04(3r ) = 11, 479.44r.
For the given probability and tolerance, λ0 = (1.96 / 0.1) 2 = 384.16.
The number of observations needed is
λ0σ 2 / μ 2 = 384.16(11, 479.44r ) /(73.2r ) 2 = 823.02 / r.
The average observation produces 3r claims and so the required number of claims is
(823.02 / r )(3r ) = 2, 469.

Question #174
Key: A

1 1 2n − 1 39
Hˆ (t2 ) = + = = => 39n 2 − 799n + 380 = 0 => n = 20, n = 0.487 .
n n − 1 n(n − 1) 380
Discard the non-integer solution to have n = 20.
The Kaplan-Meier Product-Limit Estimate is:
19 18 11 11
Sˆ (t9 ) = ... = = 0.55.
20 19 12 20

Question #175
Key: E

There are 27 possible bootstrap samples, which produce four different results. The
results, their probabilities, and the values of g are:

C-09-08 - 72 -
Bootstrap
Sample Prob g
1, 1, 1 8/27 0
1, 1, 4 12/27 2
1, 4, 4 6/27 -2
4, 4, 4 1/27 0

The third central moment of the original sample is 2. Then,

⎡8 12 6 1 2⎤ 44
MSE = ⎢ ( 0 − 2 ) + ( 2 − 2 ) + ( −2 − 2 ) + ( 0 − 2 ) ⎥ =
2 2 2
.
⎣ 27 27 27 27 ⎦ 9

Question #176
Key: A

Pick one of the points, say the fifth one. The vertical coordinate is F(30) from the model
and should be slightly less than 0.6. Inserting 30 into the five answers produces 0.573,
0.096, 0.293, 0.950, and something less than 0.5. Only the model in answer A is close.

Question #177
Key: C, E

The distribution of Θ is Pareto with parameters 1 and 2.6. Then,

1 2
v = EPPV = E (Θ) = = 0.625, a = VHM = Var (Θ) = − 0.6252 = 1.6927,
2.6 − 1 1.6(0.6)
5
k = v / a = 0.625 /1.6927 = 0.3692, Z = = 0.9312.
5 + 0.3692

Question #178
Key: E

At 300, there are 400 policies available, of which 350 survive to 500. At 500 the risk set
increases to 875, of which 750 survive to 1,000. Of the 750 at 1,000, 450 survive to
350 750 450
5,000. The probability of surviving to 5,000 is = 0.45. The distribution
400 875 750
function is 1 – 0.45 = 0.55.

Alternatively, the formulas in Loss Models could be applied as follows:

J Interval (c j , c j +1 ] dj uj xj Pj rj qj Fˆ (c j )
0 (300, 500] 400 0 50 0 400 0.125 0
1 (500, 1,000] 525 0 125 350 875 0.142857 0.125
2 (1,000, 5,000] 0 120 300 750 750 0.4 0.25
3 (5,000, 10,000] 0 30 300 330 330 0.90909 0.55

C-09-08 - 73 -
where
dj = number of observations with a lower truncation point of cj,
uj = number of observations censored from above at cj+1,
xj = number of uncensored observations in the interval (c j , c j +1 ] .
Then, Pj +1 = Pj + d j − u j − x j , rj = Pj + d j (using α = 1 and β = 0 ), and q j = x j / rj .
j −1
Finally, Fˆ (c j ) = 1 − ∏ (1 − qi ) .
i =0

Question #179
Key: D

For an exponential distribution, the maximum likelihood estimate of θ is the sample

mean, 6. Let Y = X 1 + X 2 where each X has an exponential distribution with mean 6.
The sample mean is Y/2 and Y has a gamma distribution with parameters 2 and 6.
−x/6
∞ xe
Pr(Y / 2 > 10) = Pr(Y > 20) = ∫ dx
20 36
Then ∞
xe − x / 6 −x/6 20e −20 / 6
=− −e = + e −20 / 6 = 0.1546.
6 20
6

Question #180
Key: A

10e −10 / θ
From Question 9, F (10) = 1 − − e −10 /θ = 1 − e −10 /θ (1 + 10θ −1 ) = g (θ ).
θ
20 20 400e −20 / θ
g '(θ ) = − e −20 / θ (1 + 20θ −1 ) + e−20 /θ =− .
θ2 θ2 θ3
At the maximum likelihood estimate of 6, g '(6) = −0.066063.
The maximum likelihood estimator is the sample mean. Its variance is the variance of
one observation divided by the sample size. For the exponential distribution the
variance is the square of the mean, so the estimated variance of the sample mean is
36/2 = 18. The answer is (−0.066063) 2 (18) = 0.079.

C-09-08 - 74 -
Question #181
Key B

μ (λ ,θ ) = E ( S | λ ,θ ) = λθ ,
v(λ , θ ) = Var ( S | λ , θ ) = λ 2θ 2 ,
v = EVPV = E (λ 2θ 2 ) = 1(2)(1 + 1) = 4,
a = VHM = Var (λθ ) = E (λ 2 ) E (θ 2 ) − [ E (λ ) E (θ )]2 = 2(2) − 1 = 3,
k = v / a = 4 / 3.

Question #182
Key: B

The distribution is binomial with m = 100 and q = 0.03. The first three probabilities are:
p0 = 0.97100 = 0.04755, p1 = 100(0.97)99 (0.03) = 0.14707,
100(99)
p2 = (0.97)98 (0.03) 2 = 0.22515.
2
Values between 0 and 0.04755 simulate a 0, between 0.04755 and 0.19462 simulate a
1, and between 0.19462 and 0.41977 simulate a 2. The three simulated values are 2,
0, and 1. The mean is 1.

Question #183
Key: A

A mixture of two Poissons or negative binomials will always have a variance greater
than its mean. A mixture of two binomials could have a variance equal to its mean,
because a single binomial has a variance less than its mean.

Question #184
Key: D

The posterior distribution can be found from

e − λ λ 10 ⎛ 0.4 − λ / 6 0.6 − λ /12 ⎞
⎟ ∝ λ ( 0.8e + 0.6e −13λ /12 ) .
−7 λ / 6
π (λ |10) ∝ ⎜ e + e 10

10! ⎝ 6 12 ⎠
The required constant is found from
λ 10 ( 0.8e −7 λ / 6 + 0.6e−13λ /12 ) d λ = 0.8(10!)(6 / 7)11 + 0.6(10!)(12 /13)11 = 0.395536(10!).
∞
∫0
The posterior mean is

C-09-08 - 75 -
 λ 11 ( 0.8e −7 λ / 6 + 0.6e−13λ /12 ) d λ
1 ∞
E (λ |10) =
0.395536(10!) ∫0

0.8(11!)(6 / 7)12 + 0.6(11!)(12 /13)12

= = 9.88.
0.395536(10!)

Question #185
Key: A

2 1 2 2
Hˆ (4.5) = + + + = 0.77460. The variance estimate is
12 10 9 7
2 1 2 2
2
+ 2 + 2 + 2 = 0.089397. The confidence interval is
12 10 9 7
0.77460 ± 1.96 0.089397 = 0.77460 ± 0.58603. The interval is from 0.1886 to 1.3606.

Question #186
Key: D

k θ
bias = E (θˆ) − θ = θ −θ = − ,
k +1 k +1
⎛ kθ ⎞ k 2θ 2
Var (θˆ) = Var ⎜ ⎟ = ,
⎝ k + 1 ⎠ 25(k + 1)
2

k 2θ 2 θ2
MSE = Var (θˆ) + bias 2 = + ,
25(k + 1) 2 (k + 1) 2
2θ 2
MSE = 2bias 2 = .
(k + 1) 2
Setting the last two equal and canceling the common terms gives
k2
+ 1 = 2 for k = 5.
25

Question #187
Key: D

For the geometric distribution μ ( β ) = β and v( β ) = β (1 + β ) . The prior density is Pareto

with parameters α and 1. Then,

C-09-08 - 76 -
 1
μ = E (β ) = ,
α −1
1 2 α
v = EVPV = E[ β (1 + β )] = + = ,
α − 1 (α − 1)(α − 2) (α − 1)(α − 2)
2 1 α
a = VHM = Var ( β ) = − = ,
(α − 1)(α − 2) (α − 1) 2
(α − 1) 2 (α − 2)
1 1
k = v / a = α − 1, Z = = .
1+ k α
The estimate is
1 ⎛ 1⎞ 1 x +1
x + ⎜1 − ⎟ = .
α ⎝ α ⎠ α −1 α

Question #188
Key: C

1
The hazard rate function is h( x) = h = e β1z1 + β2 z2 , a constant. For the four observations,
θ
1 .1(0)+.01(20) 1
e = 0.06786, e.1(0) +.01(30) = 0.07449,
18 18
those constants are
1 .1(1) +.01(30) 1
e = 0.08288, e.1(1) +.01(40) = 0.09160.
18 18
Each observation is from an exponential distribution with density function he − hx and
survival function e − hx . The contributions to the loglikelihood function are ln(h) − hx and
− hx , respectively. The answer is,
ln(0.06786) − 0.06786(3) − 0.07499(6) + ln(0.08288) − 0.08288(7) − 0.09160(8) = −7.147.

Question #189
Key: E

A is false. Using sample data gives a better than expected fit and therefore a test
statistic that favors the null hypothesis, thus increasing the Type II error probability. The
K-S test works only on individual data and so B is false. The A-D test emphasizes the
tails, thus C is false. D is false because the critical value depends on the degrees of
freedom which in turn depends on the number of cells, not the sample size.

Question #190
Key: B

C-09-08 - 77 -
E (θ ) = 0.05(0.8) + 0.3(0.2) = 0.1,
E (θ 2 ) = 0.052 (0.8) + 0.32 (0.2) = 0.02,
μ (θ ) = 0(2θ ) + 1(θ ) + 2(1 − 3θ ) = 2 − 5θ ,
v(θ ) = 02 (2θ ) + 12 (θ ) + 22 (1 − 3θ ) − (2 − 5θ ) 2 = 9θ − 25θ 2 ,
μ = E (2 − 5θ ) = 2 − 5(0.1) = 1.5,
v = EVPV = E (9θ − 25θ 2 ) = 9(0.1) − 25(0.02) = 0.4,
a = VHM = Var (2 − 5θ ) = 25Var (θ ) = 25(0.02 − 0.12 ) = 0.25,
1 5
k = v / a = 0.4 / 0.25 = 1.6, Z = = ,
1 + 1.6 13
5 8
P = 2 + 1.5 = 1.6923.
13 13

Question #191
Key: B

e − λ λ e − λ λ 3 25 λ 5e−2 λ
f (λ | 5,3) ∝ ∝ λ 12 e −4 λ . This is a gamma distribution with parameters
5! 3! 24λ
13 and 0.25. The expected value is 13(0.25) = 3.25.

Alternatively, if the Poisson-gamma relationships are known, begin with the prior
parameters α = 5 and β = 2 where β = 1/ θ if the parameterization from Loss Models is
considered. Then the posterior parameters are α ' = 5 + 5 + 3 = 13 where the second 5
and the 3 are the observations and β ' = 2 + 2 = 4 where the second 2 is the number of
observations. The posterior mean is then 13/4 = 3.25.

Question #192
Key: D

Because the kernel extends one unit each direction there is no overlap. The result will
be three replications of the kernel. The middle one will be twice has high due to having
two observations of 3 while only one observation at 1 and 5. Only graphs A and D fit
this description. The kernel function is smooth, which rules out graph A.

Question #193
Key: C

The two moment equations are

C-09-08 - 78 -
 θ 2θ 2
508 = , 701, 401.6 = . Dividing the square of the first equation into the
α −1 (α − 1)(α − 2)
701, 401.6 2(α − 1)
second equation gives = 2.7179366 = . The solution is α = 4.785761.
508 2
α −1
From the first equation, θ = 1,923.167. The requested LEV is
1,923.167 ⎡ ⎛ 1,923.167 ⎞ ⎤
3.785761

E ( X ∧ 500) = ⎢1 − ⎜ ⎥ = 296.21.
3.785761 ⎣⎢ ⎝ 1,923.167 + 500 ⎟⎠ ⎦⎥

Question #194
Key: B

50(200 − 227.27) 2 + 60(250 − 227.27) 2 + 100(160 − 178.95) 2 + 90(200 − 178.95) 2

vˆ = EVPV =
1+1
= 71,985.647,
kˆ = 71,985.647 / 651.03 = 110.57,
110
Zˆ = = 0.499.
110 + 110.57

Question #195
Key: B

F100 (1, 000) = 0.16, F100 (3, 000) = 0.38, F100 (5, 000) = 0.63, F100 (10, 000) = 0.81,
F100 (2, 000) = 0.5(0.16) + 0.5(0.38) = 0.27,
F100 (6, 000) = 0.8(0.63) + 0.2(0.81) = 0.666.
Pr(2, 000 < X < 6, 000) = 0.666 − 0.27 = 0.396.

Question #196
Key: C

3 4
⎡ f (750) ⎤ ⎡ f (400) ⎤
L=⎢ ⎥ f (200)3 f (300) 4 [1 − F (10, 000)]6 ⎢ ⎥
⎣1 − F (200) ⎦ ⎣1 − F (300) ⎦
3 3 4 6 4
⎡ α10, 200α ⎤ ⎡ α10, 000α ⎤ ⎡ α10, 000α ⎤ ⎡ 10, 000α ⎤ ⎡ α 10,300α ⎤
=⎢ α +1 ⎥ ⎢ α +1 ⎥ ⎢ α +1 ⎥ ⎢ α ⎥ ⎢ α +1 ⎥
⎣ 10, 750 ⎦ ⎣ 10, 200 ⎦ ⎣ 10,300 ⎦ ⎣ 20, 000 ⎦ ⎣ 10, 400 ⎦
= α 1410, 200−310, 00013α10,300−410, 750−3α −3 20, 000−6α10, 400−4α − 4
∝ α 1410, 00013α10, 750−3α 20, 000−6α10, 400−4α ,
ln L = 14 ln α + 13α ln(10, 000) − 3α ln(10, 750) − 6α ln(20, 000) − 4α ln(10, 400)
= 14 ln α − 4.5327α .
The derivative is 14 / α − 4.5327 and setting it equal to zero gives αˆ = 3.089.

C-09-08 - 79 -
Question #197
Key: C

30 + 30 + 12 + 4
vˆ = EVPV = x = = 0.76.
100
50(0 − 0.76) 2 + 30(1 − 0.76) 2 + 15(2 − 0.76) 2 + 4(3 − 0.76) 2 + 1(4 − 0.76) 2
aˆ = VHM = − 0.76
99
= 0.090909,
0.76 1
kˆ = = 8.36, Zˆ = = 0.10684,
0.090909 1 + 8.36
P = 0.10684(1) + 0.89316(0.76) = 0.78564.
The above analysis was based on the distribution of total claims for two years. Thus
0.78564 is the expected number of claims for the next two years. For the next one year
the expected number is 0.78564/2 = 0.39282.

Question #198
Key: A

/θ /θ
For members of Class A, h( x) = 2 x / θ , S A ( x) = e − x , f A ( x) = (2 x / θ )e− x
2 2
and for members
of Class B, h( x) = 2 xe β / θ = 2 xγ / θ , S B ( x) = e − x γ /θ , f B ( x) = (2 xγ / θ )e− x γ / θ where γ = e β .
2 2

The likelihood function is

L(γ ,θ ) = f A (1) f A (3) f B (2) f B (4)
= 2θ −1e −1/θ 6θ −1e−9 / θ 4γθ −1e−4γ /θ 8γθ −1e−16γ / θ
∝ θ −4γ 2 e − (10+ 20γ ) /θ .
The logarithm and its derivatives are
l (γ ,θ ) = −4 ln θ + 2 ln γ − (10 + 20γ )θ −1
∂l / ∂γ = 2γ −1 + 20θ −1
∂l / ∂θ = −4θ −1 + (10 + 20γ )θ −2 .
Setting each partial derivative equal to zero and solving the first equation for θ gives
θ = 10γ . Substituting in the second equation yields
4 10 + 20γ
− + = 0, 400γ 2 = 100γ + 200γ 2 , γ = 0.5. Then, β = ln(γ ) = ln(0.5) = −0.69.
10γ 100γ 2

C-09-08 - 80 -
Question #199
Key: E

0.2 x −0.8
The density function is f ( x) = e − ( x /θ ) . The likelihood function is
0.2

θ 0.2

L(θ ) = f (130) f (240) f (300) f (540)[1 − F (1000)]2

0.2(130) −0.8 0.2(240) −0.8 0.2(300) −0.8 0.2(540) −0.8
= e − (130 / θ ) e − (240 / θ ) e − (300 / θ ) e− (540 /θ ) e− (1000 /θ ) e− (1000 /θ )
0.2 0.2 0.2 0.2 0.2 0.2

θ 0.2
θ 0.2
θ 0.2
θ 0.2

−0.8 −θ −0.2 (1300.2 + 2400.2 + 3000.2 + 5400.2 +10000.2 +10000.2 )

∝θ e ,
l (θ ) = −0.8ln(θ ) − θ −0.2
(130 + 240 0.2 0.2
+ 300 0.2
+ 5400.2 + 10000.2 + 10000.2 )
= −0.8ln(θ ) − 20.2505θ −0.2 ,
l '(θ ) = −0.8θ −1 + 0.2(20.2505)θ −1.2 = 0,
θ −0.2 = 0.197526, θˆ = 3,325.67.

Question #200
Key: A

Buhlmann estimates are on a straight line, which eliminates E. Bayes estimates are
never outside the range of the prior distribution. Because graphs B and D include
values outside the range 1 to 4, they cannot be correct. The Buhlmann estimates are
the linear least squares approximation to the Bayes estimates. In graph C the Bayes
estimates are consistently higher and so the Buhlmann estimates are not the best
approximation. This leaves A as the only feasible choice.

Question #201
Key: C

The expected counts are 300(0.035) = 10.5, 300(0.095) = 28.5, 300(0.5) = 150,
300(0.2) = 60, and 300(0.17) = 51 for the five groups. The test statistic is
(5 − 10.5) 2 (42 − 28.5) 2 (137 − 150) 2 (66 − 60) 2 (50 − 51) 2
+ + + + = 11.02.
10.5 28.5 150 60 51
There are 5 – 1 = 4 degrees of freedom. From the table, the critical value for a 5% test
is 9.488 and for a 2.5% test is 11.143. The hypothesis is rejected at 5%, but not at
2.5%.

Question #202
Key: A

To simulate a lognormal variate, first convert the uniform random number to a standard
normal. This can be done be using the table of normal distribution values. For the six
values given, the corresponding standard normal values are 0.3, -0.1, 1.6, -1.4, 0.8, and
-0.2. Next, multiply each number by the standard deviation of 0.75 and add the mean of

C-09-08 - 81 -
5.6. This produces random observations from the normal 5.6, 0.752 distribution. These
values are 5.825, 5,525, 6.8, 4.55, 6.2, and 5.45. To create lognormal observations,
exponentiate these values. The results are 339, 251, 898, 95, 493, and 233. After
imposing the policy limit of 400, the mean is (339 + 251 + 400 + 95 + 400 + 233)/6 =
286.

Question #203
Key: C

β2
For the geometric distribution, Pr( X 1 = 2 | β ) = and the expected value is β.
(1 + β )3
Pr( X 1 = 2 | β = 2) Pr( β = 2)
Pr( β = 2 | X 1 = 2) =
Pr( X 1 = 2 | β = 2) Pr( β = 2) + Pr( X 1 = 2 | β = 5) Pr( β = 5)
4 1
= 27 3 = 0.39024.
4 1 25 2
+
27 3 216 3
The expected value is then 0.39024(2) + 0.60976(5) = 3.83.

Question #204
Key: D

The following derives the general formula for the statistic to be forgotten by time x. It
would work fine, and the equations would look simpler, if you immediately plugged in
1 1
x = , the only value you want. Then the x + becomes 1.
2 2
Let X be the random variable for when the statistic is forgotten. Then FX ( x y ) = 1 − e − xy
For the unconditional distribution of X, integrate with respect to y

∞ 2

(
FX ( x ) = ∫ 1 − e − xy
) 1 ⎛ y⎞ − y2
⎜ ⎟ e dy
Γ ( 2) y ⎝ 2 ⎠
0
∞
= 1 − ∫ y e ( 2 )d y
1 − y x+ 1
40
1
= 1−
4 ( x + 12 )
2

1
F ( 12 ) = 1 − = 0.75
4 ( 12 + 12 )
2

There are various ways to evaluate the integral in the second line:

C-09-08 - 82 -
1. Calculus, integration by parts
⎛ 1⎞
∞ ⎛ 1 ⎞ − y⎜ x + ⎟
2. Recognize that ∫0 y ⎜ x + ⎟ e ⎝ 2 ⎠ dy
⎝ 2⎠
1
is the expected value of an exponential random variable with θ =
1
x+
2
2 ⎛ 1⎞
− y⎜ x + ⎟
⎛ 1⎞
3. Recognize that Γ ( 2 ) ⎜ x + ⎟ y e ⎝ 2 ⎠ is the density function for a Gamma
⎝ 2⎠
1
random variable with α = 2 and θ = , so it would integrate to 1.
1
x+
2
1
(Approaches 2 and 3 would also work if you had plugged in x = at the start. The
2
resulting θ becomes 1).

Question #205
Key: D

State# Number Probability Mean E(N) Var(N) Var(X) E(S) Var(S)

of Number
needing of visits
Therapy E(X)
1 400 0.2 2 80 64 6 160 736
2 300 0.5 15 150 75 240 2,250 52,875
3 200 0.3 9 60 42 90 540 8,802
2,950 62,413

Std Dev ( S ) = 62413 = 250

⎛ S − 2950 50 ⎞
Pr ( S > 3000 ) = Pr ⎜ > ⎟ = 1 − Φ ( 0.2 ) = 0.42
⎝ 250 250 ⎠
The Var ( X ) column came from the formulas for mean and variance of a geometric
distribution.

Using the continuity correction, solving for Pr ( S > 3000.5 ) , is theoretically better but
does not affect the rounded answer.

C-09-08 - 83 -
Question #206
Key: B

Frequency is geometric with β = 2 , so

p0 = 1/ 3, p1 = 2 / 9, p2 = 4 / 27

Convolutions of f X ( x) needed are

X f f *2
5 0.2 0
10 0.3 0.04

so f S ( 0 ) = 1/ 3, f S ( 5 ) = 2 / 9 ( 0.2 ) = 0.044 , f S (10 ) = 2 / 9 ( 0.3) + 4 / 27 ( 0.04 ) = 0.073

E ( X ) = ( 0.2 )( 5 ) + ( 0.3)(10 ) + ( 0.5 )( 20 ) = 14
E [ S ] = 2 E ( X ) = 28
E [ S − 15]+ = E [ S ] − 5 (1 − F ( 0 ) ) − 5 (1 − F ( 5 ) ) − 5 (1 − F (10 ) )
= 28 − 5 (1 − 1/ 3) − 5 (1 − 1/ 3 − 0.044 ) − 5 (1 − 1/ 3 − 0.044 − 0.073)
= 28 − 3.33 − 3.11 − 2.75 = 18.81

Alternatively,

E [ S − 15]+ = E [ S ] − 15 + 15 f S ( 0 ) + 10 f S ( 5 ) + 5 f S (10 )
⎛1⎞
= 28 − 15 + (15 ) ⎜ ⎟ + 10 ( 0.044 ) + 5 ( 0.073)
⎝ 3⎠
= 18.81

Question #207
Key: E

S X ( 4 ) = 1 − ∫ f X ( x ) dx = 1 − ∫ 0.02 x dx
4 4
0 0
4
= 1 − 0.01 x 2
0
= 0.84
f ( y + 4 ) 0.02 ( y + 4 )
fY p ( y ) = X = 0.0238 ( y + 4 )
2
=
S X ( 4) 0.84

C-09-08 - 84 -
 ⎛ y3 4 y 2 ⎞
( )
E Y p = ∫ y ( 0.0238 ( y + 4 ) ) dy = 0.0238 ⎜ +
6 6
⎟
⎝ 3 2 ⎠
0 0

= 3.4272

Question #208
Key: E

By Theorem 4.51 (on page 93 of the second edition of Loss Models), probability of zero
claims = pgf of negative binomial applied to the probability that Poisson equals 0.

For the Poisson, f ( 0 ) = e − λ

( ) ( )
−r −2
So 0.067 = ⎡1 − β e − λ − 1 ⎤ = ⎡1 − 3 e − λ − 1 ⎤
⎣ ⎦ ⎣ ⎦
Solving gives λ = 3

Question #209
Key: D

For any deductible d and the given severity distribution

E ( X − d )+ = E ( X ) − E ( X ∧ d )
⎛ 3000 ⎞
= 3000 − 3000 ⎜1 − ⎟
⎝ 3000 + d ⎠
⎛ 3000 ⎞
= ( 3000 ) ⎜ ⎟
⎝ 3000 + d ⎠

⎛ 3000 ⎞
So P2005 = (1.2 )( 3000 ) ⎜ ⎟ = 3000
⎝ 3600 ⎠

The following paragraph just clarifies the notation in the rest of the solution:

Let r denote the reinsurer’s deductible relative to losses (not relative to reinsured
claims). Thus if r = 1000 (we are about to solve for r), then on a loss of 4000, the
insured collects
4000 – 600 = 3400, the reinsurer pays 4000 – 1000 = 3000, leaving the primary insurer
paying 400.

Another way, exactly equivalent, to express that reinsurance is that the primary
company pays the insured 3400. The reinsurer reimburses the primary company for its
claims less a deductible of 400 applied to claims. So the reinsurer pays 3400 – 400 =
3000, the same as before.

Expected reinsured claims in 2005

C-09-08 - 85 -
 ⎛ 3000 ⎞ 9,000,000
= ( 3000 ) ⎜ ⎟=
⎝ 3000 + r ⎠ 3000 + r

⎛ 9,000,000 ⎞
R2005 = (1.1) ⎜ ⎟ = ( 0.55 ) P2005
⎝ 3000 + r ⎠
9,900,000
= ( 0.55 )( 3000 ) = 1650
3000 + r
r = 3000

In 2006, after 20% inflation, losses will have a two-parameter Pareto distribution with
α = 2 and θ = (1.2 )( 3000 ) = 3600 .
The general formula for claims will be
⎛ 3600 ⎞ 12,960,000
E ( X − d )+ = ( 3600 ) ⎜ ⎟=
⎝ 3600 + d ⎠ 3600 + d
⎛ 12,960,000 ⎞
P2006 = 1.2 ⎜ ⎟ = 3703
⎝ 3000 + 600 ⎠

⎛ 12,960,000 ⎞
R2006 = 1.1⎜ ⎟ = 2160
⎝ 3600 + 3000 ⎠

R2006 / P2006 = 0.5833

[If you applied the reinsurer’s deductible to the primary insurer’s claims, you would solve
that the deductible is 2400, and the answer to the problem is the same].

Question #210
Key: C

Consider Disease 1 and other Diseases as independent Poisson processes with

⎛1⎞ ⎛ 15 ⎞
respective λ ' s = ( 0.16 ) ⎜ ⎟ = 0.01 and ( 0.16 ) ⎜ ⎟ = 0.15 respectively. Let S1 =
⎝ 16 ⎠ ⎝ 16 ⎠
aggregate losses from Disease 1; S2 = aggregate losses from other diseases.

E ( S1 ) = 100 × 0.01× 5 = 5

( )
Var ( S1 ) = 100 × 0.01× 502 + 52 = 2525
E ( S 2 ) = 100 × 0.15 × 10 = 150
( )
Var ( S2 ) = 100 × 0.15 × 202 + 102 = 7500

If no one gets the vaccine:

C-09-08 - 86 -
E ( S ) = 5 + 150 = 155
Var ( S ) = 2525 + 7500 = 10,025
Φ ( 0.7 ) = 1 − 0.24
A = 155 + 0.7 10,025 = 225.08

If all get the vaccine, vaccine cost = (100 )( 0.15 ) = 15

No cost or variance from Disease 1

B = 15 + 150 + 0.7 7500 = 225.62

A / B = 0.998

Question #211
Key: A

For current model f x = bg 1

4e
− 4x

Let g(x) be the new density function, which has

(i) g(x) = c, 0≤ x≤3
(ii) g ( x ) = ke −x/4
, x > 3*
(iii) c = ke −3/ 4 , since continuous at x = 3

Since g is density function, it must integrate to 1.

∞ −x −3 −3

∫3
1 = 3c + ke 4 d x = 3ke 4 + 4ke 4 = 3c + 4c ⇒ c = 1
7

31
F ( 3) = ∫ cd x = ∫
3
d x = 3 7 = 0.43
0 07

⎛1 ⎞
*This could equally well have been written g ( x ) = d × ⎜ e− x / 4 ⎟ , then let k = d/4, or even
⎝4 ⎠
carry the d/4 throughout.

Question #212
Key: C

Since loss amounts are uniform on (0, 10), 40% of losses are below the deductible (4),
and 60% are above. Thus, claims occur at a Poisson rate λ * = ( 0.6 )(10 ) = 6 .

Since loss amounts were uniform on (0, 10), claims are uniform on (0, 6).

C-09-08 - 87 -
Let N = number of claims; X = claim amount; S = aggregate claims.

E ( N ) = Var ( N ) = λ * = 6
E ( X ) = ( 6 − 0) / 2 = 3
Var ( X ) = ( 6 − 0 ) /12 = 3
2

Var ( S ) = E ( N )Var ( X ) + Var ( N ) ⎡⎣ E ( X ) ⎤⎦

2

= 6*3 + 6*32
= 72

Question #213
Key: E

N pn n × pn n 2 × pn
0 0.1 0 0
1 0.4 0.4 0.4
2 0.3 0.6 1.2
3 0.2 0.6 1.8
E [ N ] = 1.6 E ⎡⎣ N 2 ⎤⎦ = 3.4

Var ( N ) = 3.4 − 1.62 = 0.84

E[X ] = λ = 3
Var ( X ) = λ = 3

Var ( S ) =
= E [ N ]Var ( X ) + E ( X ) × Var ( N )
2

= 1.6 ( 3) + ( 3) ( 0.84 )
2

= 12.36

Question #214
Key: D

Sˆ (300) = 3 /10 (there are three observations greater than 300)

Hˆ (300) = − ln[ Sˆ (300)] = − ln(0.3) = 1.204 .

C-09-08 - 88 -
Question #215
Key: A

E ( X | λ ) = Var ( X | λ ) = λ
μ = v = E (λ ) = αθ ; a = Var (λ ) = αθ 2 ; k = v / a = 1/ θ
n nθ
Z= =
n + 1/ θ nθ + 1
θ 1 θ +μ
0.15 = (1) + μ=
θ +1 θ +1 θ +1
2θ 1 4θ + μ
0.20 = (2) + μ=
2θ + 1 2θ + 1 2θ + 1
From the first equation,
0.15θ + 0.15 = θ + μ and so μ = 0.15 − 0.85θ
Then the second equation becomes
0.4θ + 0.2 = 4θ + 0.15 − 0.85θ
0.05 = 2.75θ ; θ = 0.01818

Question #216
Key: E

1 1
0.75 = ; 0.25 =
1 + (100 / θ ) γ
1 + (500 / θ )γ
(100 / θ )γ = 1/ 3; (500 / θ )γ = 3
Taking the ratio of these two equalities produces 5γ = 9 . From the second equality,
9 = [(500 / θ )2 ]γ = 5γ ; (500 / θ ) 2 = 5; θ = 223.61

C-09-08 - 89 -
Question #217
Key: E

Begin with
y 350 500 1000 1200 1500
s 2 2 1 1 1
r 10 8 5 2 1
8 641
Then Sˆ1 (1250) = = 0.24
10 8 5 2
The likelihood function is
2 2 2
L(θ ) = ⎡⎣θ −1e −350 /θ ⎤⎦ ⎡⎣θ −1e−500 /θ ⎤⎦ e−500 /θ θ −1e −1000 /θ ⎡⎣ e −1000 /θ ⎤⎦ θ −1e −1200 / θ θ −1e−1500 / θ
= θ −7 e −7900 / θ
7900 7 7900
l (θ ) = −7 ln θ − ; l '(θ ) = − + = 0; θˆ = 7900 / 7
θ θ θ2
Sˆ2 (1250) = e −1250(7) / 7900 = 0.33
The absolute difference is 0.09.

Question #218
Key: E

4 xθ 4
f ( x) = − S '( x) =
(θ 2 + x 2 )3
4(2)θ 4 4(4)θ 4 θ4 128θ 12
L(θ ) = f (2) f (4) S (4) = =
(θ 2 + 22 )3 (θ 2 + 42 )3 (θ 2 + 42 ) 2 (θ 2 + 4)3 (θ 2 + 16)5
l (θ ) = ln128 + 12 ln θ − 3ln(θ 2 + 4) − 5ln(θ 2 + 16)
12 6θ 10θ
l '(θ ) = − 2 − 2 = 0;12(θ 4 + 20θ 2 + 64) − 6(θ 4 + 16θ 2 ) − 10(θ 4 + 4θ 2 ) = 0
θ θ + 4 θ + 16
0 = −4θ 4 + 104θ 2 + 768 = θ 4 − 26θ 2 − 192
26 ± 262 + 4(192)
θ2 = = 32; θ = 5.657
2

C-09-08 - 90 -
Question #219
Key: A

θ 2x 2θ θ 2x 4θ 2 θ 2 4θ 2 θ 2
E( X | θ ) = ∫ x dx = ; Var ( X | θ ) = ∫ x 2 2 dx − = − =
0 θ2 3 0 θ 9 2 9 18
1
μ = (2 / 3) E (θ ) = (2 / 3) ∫ 4θ 4 dθ = 8 /15
0
1
EVPV = v = (1/18) E (θ 2 ) = (1/18) ∫ 4θ 5 dθ = 1/ 27
0

VHM = a = (2 / 3) Var (θ ) = (4 / 9) ⎡⎣ 4 / 6 − (4 / 5) 2 ⎤⎦ = 8 / 675

2

1/ 27 1
k= = 25 / 8; Z = = 8 / 33
8 / 675 1 + 25 / 8
Estimate is (8 / 33)(0.1) + (25 / 33)(8 /15) = 0.428.

Question #220
Key: D

From the Poisson(4) distribution the probabilities at 0, 1, and 2 are 0.0183, 0.0733, and
0.1463. The cumulative probabilities are 0.0183, 0.0916, and 0.2381. Because 0.0916
< 0.13 < 0.2381 the simulated number of claims is 2. Claim amounts are simulated
from solving
u = 1 − e− x /1000 for x = −1000 ln(1 − u ) . The two simulated amounts are 51.29 and 2995.73
for a total of 3047.02

Question #221
Key: B

It may be easiest to show this by graphing the density functions. For the first function
the three components are each constant. One is of height 1/20 from 0 to 2
(representing the empirical probability of 1/10 at 1, one is height 1/20 from 1 to 3 and
one is height 8/20 from 2 to 4. The following figure shows each of them and their sum,
the kernel density estimator.

C-09-08 - 91 -
 0.5
0.45
0.4
0.35 f1
0.3
f2
0.25
f3
0.2
0.15 f
0.1
0.05
0
0 1 2 3 4

The triangular one is similar. For the triangle from 0 to 2, the area must be 1/10. With a
base of 2, the height is 1/10. the same holds for the second triangle. The third has
height 8/10. When added they look as follows;

0.9
0.8
0.7
0.6 f1
0.5 f2
0.4 f3
0.3 f
0.2
0.1
0
0 1 2 3 4

The question asks about cumulative probabilities. From 0 to 1 the first is linear and the
second is quadratic, but by x = 1 both have accumulated 0.05 of probability. Because
the cumulative distribution functions are the same at 1 and the density functions are
identical from 1 to 2, the distribution functions must be identical from 1 to 2.

C-09-08 - 92 -
Question #222
Key: D and E

For the Poisson distribution, the mean, λ, is estimated as 230/1000 = 0.23.

# of Days Poisson Expected # of Observed # of χ2

Probability Workers Workers
0 0.794533 794.53 818 0.69
1 0.182743 182.74 153 4.84
2 0.021015 21.02 25 0.75
3 or more 0.001709 1.71 4 3.07
Total 1000 9.35

The χ2 distribution has 2 degrees of freedom because there are four categories and the
Poisson parameter is estimated (d.f. = 4 – 1 – 1 = 2).

The critical values for a chi-square test with two degrees of freedom are shown in the
following table.

Significance Level Critical Value

10% 4.61
5% 5.99
2.5% 7.38
1% 9.21

9.35 is greater than 9.21 so the null hypothesis is rejected at the 1% significance level.

Question #223
Key: D

25(480 − 472.73) 2 + 30(466.67 − 472.73) 2

EVPV = vˆ = = 2423.03 where 480 = 12,000/25,
2 −1
466.67 = 14,000/30, and 472.73 = 26,000/55.
55
k = 2423.03 / 254 = 9.54; Z = = 0.852
55 + 9.54

Question #224
Key: C

Relative risk = e− β1 − β 2
which has partial derivatives −e−0.2 at βˆ1 = 0.05 and βˆ2 = 0.15
Using the delta method, the variance of the relative risk is
⎛ 2 1 ⎞ ⎛ −e
− 0.2 ⎞
7e − 0.4
1
10, 000
( )
−e − 0.2 − e− 0.2 ⎜ ⎟⎜ ⎟ =
⎝ 1 3 ⎠ ⎜⎝ −e − 0.2 ⎟⎠ 10, 000
= 0.000469

C-09-08 - 93 -
Std dev = 0.0217
upper limit = e − 0.2 + 1.96 ( 0.0217 )
= 0.8613

Alternatively, consider the quantity β1 + β 2 . The variance is

1 ⎛ 2 1 ⎞⎛ 1⎞ 7
(1 1) ⎜ ⎟⎜ ⎟ = = 0.0007 . The lower limit for this quantity is
10, 000 ⎝1 3 ⎠⎝ 1⎠ 10, 000
0.2 − 1.96 0.0007 = 0.1481 and the upper limit for the relative risk is e−0.1481 = 0.8623 .

Question #225
Key: C

⎛ ln 5000 − μ ⎞
The quantity of interest is P = Pr( X ≤ 5000) = Φ ⎜ ⎟ . The point estimate is
⎝ σ ⎠
⎛ ln 5000 − 6.84 ⎞
Φ⎜ ⎟ = Φ (1.125) = 0.87 .
⎝ 1.49 ⎠
For the delta method:
∂P −φ (1.125) ∂P −1.125φ (1.125) 1 − z2 / 2
= = −0.1422; = = −0.1600 where φ ( z ) = e .
∂μ 1.49 ∂σ 1.49 2π
Then the variance of P̂ is estimated as (−0.1422) 2 0.0444 + (−0.16) 2 0.0222 = 0.001466 and
the lower limit is PL = 0.87 − 1.96 0.001466 = 0.79496 .

Question #226
Key: A

Pr( X 1 = 1| θ = 0.1) Pr(θ = 0.1)

Pr(θ = 0.1| X 1 = 1) =
Pr( X 1 = 1| θ = 0.1) Pr(θ = 0.1) + Pr( X 1 = 1| θ = 0.3) Pr(θ = 0.3)
0.1(0.8) 4
= =
0.1(0.8) + 0.3(0.2) 7
Then,
E ( X 2 | θ = 0.1) = 0(0.2) + 1(0.1) + 2(0.7) = 1.5
E ( X 2 | θ = 0.3) = 0(0.6) + 1(0.3) + 2(0.1) = 0.5
4 3
E ( X 2 | X 1 = 1) = (1.5) + (0.5) = 1.071
7 7

Question #227
Key: D

The requirement is that

C-09-08 - 94 -
 Fˆ (1500) Sˆ (1500)
0.01Fˆ (1500) ≥ 1.96
N
P 2
P( N − P)
0.0001 2 ≥ 3.8416
N N3
NP
≥ 38, 416.
N −P
For the five answer choices, the left hand side is 34,364, 15,000, 27,125, 39,243, and
37,688. Only answer D meets the condition.

Question #228
Key: D

s4 ^ ^
= H ( y 4 ) − H ( y 3 ) = 0.5691 − 0.4128 = 0.1563 .
r4
s4 ^ ^ ^ ^
2
= V [ H ( y 4 )] − V [ H ( y 3 )] = 0.014448 − 0.009565 = 0.004883 .
r4
( s4 / r4 ) 2 0.15632
Therefore, s4 = = = 5.
s4 / r42 0.004833

Question #229
Key: A

ln f ( x) = ln θ − 2 ln(θ + x)
∂ ln f ( x) 1 2
= −
∂θ θ θ+x
∂ ln f ( x)
2
1 2
=− 2 +
∂θ 2
θ (θ + x) 2
∞
⎡ ∂ 2 ln f ( x) ⎤ 1 ∞ 2θ 1 ⎡ 2θ ⎤ 1 2 1
E⎢ ⎥ = − 2 + ∫0 dx = − 2 + ⎢ − 3⎥
=− 2 + 2 =− 2
⎣ ∂θ θ (θ + x) θ ⎣ 3(θ + x) ⎦ 0 θ 3θ 3θ
2 4
⎦
n 3θ 2
I (θ ) = ; Var =
3θ 2 n

Question #230
Key: B

C-09-08 - 95 -
μ = E[ E ( X | λ )] = E (λ ) = 1(0.9) + 10(0.09) + 20(0.01) = 2
EVPV = v = E[Var ( X | λ )] = E (λ ) = 2
VHM = a = Var[ E ( X | λ )] = Var (λ ) = 1(0.9) + 100(0.09) + 400(0.01) − 22 = 9.9
1
Z= = 0.83193; 11.983 = 0.83193 x + 0.16807(2); x = 14
1 + 2 / 9.9

Question #231
Key: A

The given interval for H can be written as 0.775 ± 1.96 0.063 and therefore the estimated
variance of Ĥ is 0.063. To apply the delta method,
dS
S = e− H ; = −e − H ; Var
ˆ ( Sˆ ) (−e− H ) 2 Var ( Hˆ ) = (−e−0.775 ) 2 (0.063) = 0.134 .
ˆ

dH
The point estimate of S is e −0.775 = 0.4607 and the confidence interval is
0.4607 ± 1.96 0.0134 = 0.2269 or (0.23, 0.69).

Question #232
Key: B

The first step is to trend the year 1 data by 1.21 and the year 2 data by 1.1. The
observations are now 24.2, 48.4, 60.5, 33, 44, 99, and 132.
The first two sample moments are 63.014 and 5262.64. The equations to solve are
63.014 = e μ + 0.5σ ; 4.14336 = μ + 0.5σ 2
2

5262.64 = e 2 μ + 2σ ; 8.56839 = 2 μ + 2σ 2 .
2

Taking four times the first equation and subtracting the second gives 2μ and therefore
4(4.14336) − 8.56839
μ= = 4.00 .
2

Question #233
Key: A

μˆ = x = 12 / 60 = 0.2, EVPV = vˆ = x = 0.2

10(0.4 − 0.2) 2 + 20(0.25 − 0.2) 2 + 30(0.1 − 0.2) 2 − (3 − 1)(0.2)
VHM = aˆ = = 0.009545
102 + 202 + 302
60 −
60
10
kˆ = 20.9524; Z = = 0.323
10 + 20.9524
Question #234
Key: B and C

C-09-08 - 96 -
For males, c j = 1 and for females, c j = e0.27 = 1.31 . Then,

= 0.6965 and Sˆ female (20) = ( e −0.6965 ) = 0.402 .

1 1 1 1.31
Hˆ (20) = + +
3 + 2(1.31) 2 + 2(1.31) 2 + 1.31
.

Question #235
Key: C

5 5
l (τ , θ ) = ∑ ln f ( x j ) = ∑ ln τ + (τ − 1) ln x j − τ ln θ − ( x j / θ )τ . Under the null hypothesis it is
j =1 j =1
5
l (2, θ ) = ∑ ln 2 + ln x j − 2 ln θ − ( x j / θ ) 2 . Inserting the maximizing value of 816.7 for θ
j =1

gives −35.28 . The likelihood ratio test statistic is 2(−33.05 + 35.28) = 4.46 . There is one
degree of freedom. At a 5% significance level the critical value is 3.84 and at a 2.5%
significance level it is 5.02.

Question #236
Key: C

4
It is given that n = 4, v = 8, and Z = 0.4. Then, 0.4 = which solves for a = 4/3. For
8
4+
a
the covariance,
Cov( X i , X j ) = E ( X i X j ) − E ( X i ) E ( X j )
= E[ E ( X i X j | θ )] − E[ E ( X i | θ )]E[ E ( X j | θ )]
= E[ μ (θ ) 2 ] − E[ μ (θ )]2 = Var[ μ (θ )] = a = 4 / 3.

Question #237
Key: A

U z x lognormal with deductible

0.6217 0.31 5.8325 341.21 241.21
0.9941 2.52 7.49 1790.05 1690.05
0.8686 1.12 6.44 626.41 526.41
0.0485 − 1.66 4.355 77.87 0
Average 614.42

The value of z is obtained by inversion from the standard normal table. That is,
u = Pr( Z ≤ z ) . The value of x is obtained from x = 0.75 z + 5.6 . The lognormal value is
obtained by exponentiating x and the final column applies the deductible.

C-09-08 - 97 -
Question #238
Key: B

MSE = E[( X 2 − θ 2 ) 2 ] = E ( X 4 − 2 X 2θ 2 + θ 4 )
= 24θ 4 − 2(2θ 2 )θ 2 + θ 4 = 21θ 4

Question #239
Key: C

157(0) + 66(1) + 19(2) + 4(3) + 2(4)

The sample mean of = 0.5 is the maximum likelihood
248
estimate of the geometric parameter β as well as the method of moments estimate of
the Poisson parameter λ. Then, P = (1 + 0.5) −1 = 0.6667 and Q = e −0.5 = 0.6065 . The
absolute difference is 0.0602.

Question #240
Key: D

5000(0) + 2100(1) + 750(2) + 100(3) + 50(4)

x= = 0.5125 and
8000
5000(0.5125) 2 + 2100(0.4875) 2 + 750(1.4875) 2 + 100(2.4875) 2 + 50(3.4875) 2
s =
2
= 0.5874 .
7999
Then, μˆ = vˆ = x = 0.5125 and aˆ = s 2 − x = 0.0749 . The credibility factor is
1
Z= = 0.1275 and the estimate is 0.1275(1) + 0.8725(0.5125) = 0.5747.
1 + 0.5125 / 0.0749

Question #241
Key: B

s = Fn (3000) = 4 / 8 = 0.5 because for the p-p plot the denominator is n+1.
t = F (3000) = 1 − e −3000 / 3300 = 0.59711 . For the difference plot, D uses a denominator of n
and so D = 4 / 7 − 0.59711 = −0.02568 and the answer is 0.5 − 0.59711 + 0.02568 = −0.071.

C-09-08 - 98 -
Question #242
Key: B

0.5
π (q | 2, 2) ∝ f (2 | q ) f (2 | q)π (q) = q(q)(q 2 / 0.039) ∝ q 4 . Because ∫
0.2
q 4 dq = 0.006186 ,
π (q | 2, 2) = q 4 / 0.006186 . Given q, the expected number of claims is
E ( N | q) = 0(0.1) + 1(0.9 − q ) + 2q = 0.9 + q . The Bayesian estimate is
0.5 q4
E ( N | 2, 2) = ∫ (0.9 + q) dq = 1.319.
0.2 0.006186

Question #243
Key: E

⎛ 200 + 110 310 + x ⎞

0.689 = F500 (1500) = 0.5 F500 (1000) + 0.5 F500 (2000) = 0.5 ⎜ + ⎟ ⇒ x = 69
⎝ 500 500 ⎠
⎛ 310 + 69 379 + y ⎞
0.839 = F500 (3500) = 0.5 F500 (2000) + 0.5 F500 (5000) = 0.5 ⎜ + ⎟ ⇒ y = 81
⎝ 500 500 ⎠

Question #244
Key: A

A is false because the test works best when the expected number of observations is
about the same from interval to interval. B is true (Loss Models, 427-8), C is true (Loss
Models, 428), and D is true (Loss Models, 430).

Question #245
Key: E

⎡ ⎛ σ ⎞2 ⎤
nλ ≥ λ0 ⎢1 + ⎜ Y ⎟ ⎥ ; θY = αθ = 10, 000α ; σ Y2 = αθ 2 = 108 α
⎢⎣ ⎝ θY ⎠ ⎥⎦
⎡ 108 α ⎤
2
⎛ 1.96 ⎞
nλ ≥ ⎜ ⎟ ⎢1 + 108 α 2 ⎥ = 384.16(1 + α )
−1

⎝ 0.1 ⎠ ⎣ ⎦

Because α is needed, but not given, the answer cannot be determined from the
information given.

C-09-08 - 99 -
Question #246
Key: E

With n + 1 = 16, we need the 0.3(16) = 4.8 and 0.65(16) = 10.4 smallest observations.
They are 0.2(280) + 0.8(350) = 336 and 0.6(450) + 0.4(490) = 466.
The equations to solve are:
2 2
⎛ θγ ⎞ ⎛ θγ ⎞
0.3 = 1 − ⎜ γ γ ⎟
and 0.65 = 1 − ⎜ γ γ ⎟
⎝ θ + 336 ⎠ ⎝ θ + 466 ⎠
(0.7) −1/ 2 = 1 + (336 / θ )γ and (0.35) −1/ 2 = 1 + (466 / θ )γ
(0.7)-1/2 − 1 (336 / θ )γ
=
(0.35) −1/ 2 − 1 (466 / θ )γ
0.282814 = (336 / 466)γ
ln(0.282814) = γ ln(336 / 466)
γ = 3.8614.

Question #247
Key: D

Let E be the even of having 1 claim in the first four years. In four years, the total
number of claims is Poisson(4λ).
Pr( E | Type I ) Pr(Type I ) e −1 (0.05) 0.01839
Pr(Type I | E ) = = = = 0.14427
Pr( E ) Pr( E ) Pr( E )
e −2 (2)(0.2) 0.05413
Pr(Type II | E ) = = = 0.42465
Pr( E ) Pr( E )
e −4 (4)(0.75) 0.05495
Pr(Type III | E ) = = = 0.43108
Pr( E ) Pr( E )
Note : Pr( E ) = 0.01839 + .05413 + .05495 = 0.12747

The Bayesian estimate of the number of claims in Year 5 is:

0.14427(0.25) + 0.42465(0.5) + 0.43108(1) = 0.67947.

C-09-08 - 100 -
Question #248
Key: B

The sample mean is 0.2(400) + 0.7(800) + 0.1(1600) = 800.

The sample variance is 0.2(400 − 800) 2 + 0.7(800 − 800) 2 + 0.1(1600 − 800) 2 = 96, 000 .
The sample third central moment is
0.2(400 − 800)3 + 0.7(800 − 800)3 + 0.1(1600 − 800)3 = 38, 400, 000 .

The skewness coefficient is 38, 400, 000 / 96, 0001.5 = 1.29 .

Question #249
Key: C

Because 0.656 < 0.7654 < 0.773, the simulated number of losses is 4. To simulate a
loss by inversion, use
τ
F ( x) = 1 − e − ( x / θ ) = u
τ
1 − u = e −( x / θ )
ln(1 − u ) = −( x / θ )τ
x = θ (− ln(1 − u ))1 / τ = 200(− ln(1 − u ))1 / 2
u1 = 0.2738, x1 = 113.12
u 2 = 0.5152, x2 = 170.18
u3 = 0.7537, x3 = 236.75
u 4 = 0.6481, x 4 = 204.39

With a deductible of 150, the first loss produces no payments and 113.12 toward the
500 limit. The second loss produces a payment of 20.18 and the insured is now out-of-
pocket 263.12. The third loss produces a payment of 86.75 and the insured is out
413.12. The deductible on the fourth loss is then 86.88 for a payment of 204.29 – 86.88
= 117.51.

The total paid by the insurer is 20.18 + 86.75 + 117.51 = 224.44.

C-09-08 - 101 -
Question #250
Key: A

The density function is f ( x) = θ x −2 e −θ / x and the likelihood function is

L(θ ) = θ (186−2 )e −θ /186θ (91−2 )e−θ / 91θ (66−2 )e−θ / 66 (e −θ / 60 )7
∝ θ 3e −0.148184θ
l (θ ) = ln L(θ ) = 3ln(θ ) − 0.148184θ
l ′(θ ) = 3θ −1 − 0.148184 = 0
θ = 3 / 0.148184 = 20.25.
The mode is θ / 2 = 20.25 / 2 = 10.125 .

Question #251
Key: D

We have μ (θ ) = 4θ and μ = 4E(θ ) = 4(600) = 2400 . The average loss for Years 1 and 2 is
1650 and so 1800 = Z (1650) + (1 − Z )(2400) which gives Z = 0.8. Because there were two
years, Z = 0.8 = 2 /(2 + k ) which gives k = 0.5.

For three years, the revised value is Z = 3 /(3 + 0.5) = 6 / 7 and the revised credibility
estimate (using the new sample mean of 2021), (6 / 7)(2021) + (1/ 7)(2400) = 2075.14 .

Question #252
Key: B

The uncensored observations are 4 and 8 (values beyond 11 are not needed). The two
r values are 10 and 5 and the two s values are 2 and 1. The Kaplan-Meier estimate is
⎛ 2 1 ⎞
Sˆ (11) = (8 /10)(4 / 5) = 0.64 and Greenwood’s estimate is (0.64) 2 ⎜ + ⎟ = 0.03072 .
⎝ 10(8) 5(4) ⎠

C-09-08 - 102 -
Question #253
Key: E

S m | Q ~ bin(m, Q) and Q ~ beta(1,99) . Then

1
E ( Sm ) = E[ E ( Sm | Q)] = E (mQ) = m = 0.01m . For the mean to be at least 50, m must
1 + 99
be at least 5,000.

Question #254
Key: D

The posterior distribution is

λ 4e −50λ
π (λ | data) ∝ (e − λ )90 (λ e− λ )7 (λ 2 e− λ ) 2 (λ 3e− λ ) = λ 17 e−150 λ which is a gamma distribution
λ
with parameters 18 and 1/150. For one risk, the estimated value is the mean, 18/150.
For 100 risks it is 100(18)/150 = 12.

Alternatively,

The prior distribution is gamma with α = 4 and β = 50. The posterior will be continue to
be gamma, with α/ = α + no. of claims = 4 + 14 = 18 and β/ = β + no. of exposures = 50
+ 100 = 150. Mean of the posterior = α / β = 18/150 = 0.12.
Expected number of claims for the portfolio = 0.12 (100) = 12.

Question #255
Key: E

0.95 = Pr(0.95μ < X < 1.05μ )

X ~ N ( μ , σ 2 / n = 1.44μ 2 / n)
⎛ 0.95μ − μ 1.05μ − μ ⎞
0.95 = Pr ⎜ <Z< ⎟
⎝ 1.2 μ / n 1.2μ / n ⎠
0.95 = Pr(−0.05 n /1.2 < Z < 0.05 n /1.2)
0.05 n /1.2 = 1.96
n = 2212.76.

C-09-08 - 103 -
Question #256
Key: B
5000 5000
⎡⎛ 2 ⎞ ⎤ ⎡⎛ 2 ⎞ ⎤
L(q) = ⎢⎜ ⎟ (1 − q) 2 ⎥ ⎢⎜ ⎟ q (1 − q) ⎥ = 25000 q 5000 (1 − q)15000
⎣⎝ ⎠
0 ⎦ ⎣⎝ ⎠1 ⎦
l (q) = 5000 ln(2) + 5000 ln(q) + 15000 ln(1 − q)
l ′(q ) = 5000q −1 − 15000(1 − q) −1 = 0
qˆ = 0.25
l (0.25) = 5000 ln(2) + 5000 ln(0.25) + 15000 ln(0.75) = −7780.97.

Question #257
Key: C

The estimate of the overall mean, μ, is the sample mean, per vehicle, which is 7/10 =
0.7.
With the Poisson assumption, this is also the estimate of v = EPV. The means for the
two insureds are 2/5 = 0.4 and 5/5 = 1.0. The estimate of a is the usual non-parametric
estimate,

5(0.4 − 0.7) 2 + 5(1.0 − 0.7) 2 − (2 − 1)(0.7)

VHM = aˆ = = 0.04
1
10 − (25 + 25)
10
(The above formula: Loss Models page 596, Herzog page 116, Dean page 25)

Then, k = 0.7/0.04 = 17.5 and so Z = 5/(5+17.5) = 2/9. The estimate for insured A is
(2/9)(0.4) + (7/9)(0.7) = 0.6333.

Question #258
Key: A

Item (i) indicates that X must one of the four given values.
Item (ii) indicates that X cannot be 200
Item (iii) indicates that X cannot be 400.
First assume X = 100. Then the values of r are 5, 3, 2, and 1 and the values of s are 2,
2 1 1 1
1, 1, and 1. Then Hˆ (410) = + + + = 2.23 and thus the answer is 100. As a check,
5 3 2 1
if X = 300, the r values are 5, 4, 3, and 1 and the s values are 1, 1, 2, and 1. Then,
1 1 2 1
Hˆ (410) = + + + = 2.12 .
5 4 3 1

C-09-08 - 104 -
Question #259
Key: B

The estimator of the Poisson parameter is the sample mean. Then,

E (λˆ ) = E ( X ) = λ
Var (λˆ ) = Var ( X ) = λ / n
c.v. = λ / n / λ = 1/ nλ
It is estimated by 1/ nλ = 1/ 39 = 0.1601.

Question #260
Key: E

Pr( X 1 = 5 | θ = 8) Pr(θ = 8)
Pr (θ = 8 | X 1 = 5 ) =
Pr( X 1 = 5 | θ = 8) Pr(θ = 8) + Pr( X 1 = 5 | θ = 2) Pr(θ = 2)
0.125e −5(0.125) (0.8)
= = 0.867035.
0.125e −5(0.125) (0.8) + 0.5e −5(0.5) (0.2)
Then,
E ( X 2 | X 1 = 5) = E (θ | X 1 = 5) = 0.867035(8) + 0.132965(2) = 7.202.

Question #261
Key: D

1 − q (T ) 1 − q (T ) q (T ) − 0.05
We have q (T ) = 1 − (1 − q′(1) )(1 − q′(2) ) and so q′(1) = 1 − = 1 − = .
1 − q′(2) 1 − 0.05 0.95
Then, 20 q0′(1) = 0.05 / 0.95 = 0.05263, ′(1)
20 q20 = 0.132 / 0.95 = 0.1389 , and

40p0′(1) = 0.9474(0.8611) = 0.8158 . Out of 1000 at age 0, 816 are expected to survive to
age 40.

C-09-08 - 105 -
Question #262
Key: D

1 1 1 ⎛ω −4− p ⎞
2

ω ω ω ⎜⎝ ω ⎟
⎠ = (ω − 4 − p)
2
L(ω ) =
⎛ω −4⎞
5
(ω − 4)5
⎜ ⎟
⎝ ω ⎠
l (ω ) = 2 ln(ω − 4 − p) − 5ln(ω − 4)
2 5
l ′(ω ) = − =0
ω −4− p ω −4
2 5
0 = l ′(29) = −
25 − p 25
p = 15.
The denominator in the likelihood function is S(4) to the power of five to reflect the fact
that it is known that each observation is greater than 4.

Question #263
Key: B

μ (λ ) = v (λ ) = λ
μ = v = E (λ ) = 0.1Γ(1 + 1/ 2) = 0.088623
VHM = a = Var (λ ) = (0.1) 2 Γ(1 + 2 / 2) − 0.0886232 = 0.002146
500
Z= = 0.92371.
500 + 0.088623 / 0.002146
The estimate for one insured for one month is
0.92371(35 / 500) + 0.07629(0.088623) = 0.07142. For 300 insureds for 12 months it is
(300)(12)(0.07142) = 257.11.

Question #264
Key: D

With no censoring the r values are 12, 9, 8, 7, 6, 4, and 3 and the s values are 3, 1, 1, 1,
2, 1, 1 (the two values at 7500 are not needed). Then,
3 1 1 1 2 1 1
Hˆ 1 (7000) = + + + + + + = 1.5456.
12 9 8 7 6 4 3
With censoring, there are only five uncensored values with r values of 9, 8, 7, 4, and 3
and all five s values are 1. Then,
1 1 1 1 1
Hˆ 2 (7000) = + + + + = 0.9623. The absolute difference is 0.5833.
9 8 7 4 3

C-09-08 - 106 -
Question #265
Key: A

The simulated paid loss is exp[0.494Φ −1 (u ) + 13.294] where Φ −1 (u ) is the inverse of the
standard normal distribution function. The four simulated paid losses are 450,161,
330,041, 939,798, and 688,451 for an average of 602,113. The multiplier for unpaid
losses is 0.801(2006 − 2005)0.851 e −0.747(2006− 2005) = 0.3795 and the answer is 0.3795(602,113)
= 228,502

Question #266
Key: A

The deduction to get the SBC is (r / 2) ln(n) = (r / 2) ln(260) = 2.78r where r is the number
of parameters. The SBC values are then -416.78, -417.56, -419.34, -420.12, and -
425.68. The largest value is the first one, so model I is to be selected.

Question #267
Key: E

Pr( X 1 = r | λ = 1) Pr(λ = 1)
Pr ( λ = 1| X 1 = r ) =
Pr( X 1 = r | λ = 1) Pr(λ = 1) + Pr( X 1 = r | λ = 3) Pr(λ = 3)
e −1
(0.75)
r! 0.2759
= −1 −3 r
= .
e e 3 0.2759 + 0.1245(3r )
(0.75) + (0.25)
r! r!
Then,
0.2759 0.1245(3r )
2.98 = (1) + (3)
0.2759 + 0.1245(3r ) 0.2759 + 0.1245(3r )
0.2759 + 0.3735(3r )
= .
0.2759 + 0.1245(3r )
Rearrange to obtain
0.82218 + 0.037103(3r ) = 0.2759 + 0.03735(3r )
0.54628 = 0.00025(3r )
r = 7.
Because the risks are Poisson, (μ = EPV, a = VHM):
μ = v = E (λ ) = 0.75(1) + 0.25(3) = 1.5
a = Var (λ ) = 0.75(1) + 0.25(9) − 2.25 = 0.75
1
Z= = 1/ 3
1 + 1.5 / 0.75
and the estimate is (1/3)(7) + (2/3)(1.5) = 3.33.

C-09-08 - 107 -
Question #268
Key: E

The uniform kernel spreads the probability of 0.1 to 10 units each side of an
observation. So the observation at 25 contributes a density of 0.005 from 15 to 35,
contributing nothing to survival past age 40. The same applies to the point at 30. For
the next 7 points:
35 contributes probability from 25 to 45 for 5(0.005) = 0.025 above age 40.
35 contributes probability from 25 to 45 for 5(0.005) = 0.025 above age 40.
37 contributes probability from 27 to 47 for 7(0.005) = 0.035 above age 40.
39 contributes probability from 29 to 49 for 9(0.005) = 0.045 above age 40.
45 contributes probability from 35 to 55 for 15(0.005) = 0.075 above age 40.
47 contributes probability from 37 to 57 for 17(0.005) = 0.085 above age 40.
49 contributes probability from 39 to 59 for 19(0.005) = 0.095 above age 40.
The observation at 55 contributes all 0.1 of probability. The total is 0.485.

Question #269
Key: A

X ~ Exp(θ )
n

∑X
i =1
i ~ Γ(n,θ )

X ~ Γ(n,θ / n)
E ( X 2 ) = (θ / n) 2 (n)(n + 1) = (n + 1)θ 2 / n.
The second line follows because an exponential distribution is a gamma distribution with
α = 1 and the sum of independent gamma random variables is gamma with the “α”
parameters added. The third line follows because the gamma distribution is a scale
distribution. Multiplying by 1/n retains the gamma distribution with the “θ” parameter
multiplied by 1/n.

Question #270
Key: C

The sample means are 3, 5, and 4 and the overall mean is 4. Then,
1+ 0 + 0 +1+ 0 + 0 +1+1+1+1+ 1+1 8
vˆ = =
3(4 − 1) 9
(3 − 4) 2 + (5 − 4) 2 + (4 − 4) 2 8 / 9 7
aˆ = − = = 0.78.
3 −1 4 9

C-09-08 - 108 -
Question #271
Key: C

The ordered values are:

22t, 25t, 27h, 28t, 31h, 33t, 35h, 39t, 42h, and 45h where t is a traditional car and h is a
hybrid car. The s values are all 1 because there are no duplicate values. The c values
are 1 for traditional cars and e −1 for hybrid cars. Then
1 1 1 1 1
Hˆ 0 (32) = −1
+ −1
+ −1
+ −1
+ = 1.0358.
5 + 5e 4 + 5e 3 + 5e 3 + 4e 2 + 4e−1

Question #272
Key: C

π (q | 2) = 6q 2 (1 − q) 2 6q(1 − q) ∝ q 3 (1 − q)3
The mode can be determined by setting the derivative equal to zero.
π ′(q | 2) ∝ 3q 2 (1 − q)3 − 3q 3 (1 − q ) 2 = 0
(1 − q) − q = 0
q = 0.5.

Question #273
Key: B

For the severity distribution the mean is 5,000 and the variance is 10,0002/12. For
credibility based on accuracy with regard to the number of claims,
2
⎛ z ⎞
2000 = ⎜ ⎟ , z = 1.8
2

⎝ 0.03 ⎠
where z is the appropriate value from the standard normal distribution. For credibility
based on accuracy with regard to the total cost of claims, the number of claims needed
is
z 2 ⎛ 100002 /12 ⎞
⎜1 + ⎟ = 960.
0.052 ⎝ 50002 ⎠

C-09-08 - 109 -
Question #274
Key: C

Sˆ (10) = e − H (10) = 0.575

ˆ

1 3 5 7
Hˆ (10) = − ln(0.575) = 0.5534 = + + + .
50 49 k 12
The solution is k = 36.

Question #275
Key: A

The annual dental charges are simulated from

u = 1 − e − x /1000
x = −1000 ln(1 − u ).
The four simulated values are 356.67, 2525.73, 1203.97, and 83.38. The
reimbursements are 205.34 (80% of 256.67), 1000 (the maximum), 883.18 (80% of
1103.97), and 0. The total is 2088.52 and the average is 522.13.

Question #276
Key: B

⎛ θ ⎞ ⎛θ θ ⎞ ⎛θ ⎞
9 6 5

L(θ ) = ⎜1 − ⎟ ⎜ − ⎟ ⎜ ⎟ ∝ (10 − θ )9 θ 11
⎝ 10 ⎠ ⎝ 10 25 ⎠ ⎝ 25 ⎠
l (θ ) = 9 ln(10 − θ ) + 11ln(θ )
9 11
l ′(θ ) = − + =0
10 − θ θ
11(10 − θ ) = 9θ
110 = 20θ
θ = 110 / 20 = 5.5.

C-09-08 - 110 -
Question #277
Key: A

The maximum likelihood estimate is θˆ = x = 1000 . The quantity to be estimated is

S (θ ) = exp(−1500 / θ ) and S ′(θ ) = 1500θ −2 exp(−1500 / θ ) . For the delta method,
Var[ S (θˆ)] ≅ [ S ′(θˆ)]2Var (θˆ)
= [1500(1000) −2 exp(−1500 /1000)]2 (10002 / 6)
= 0.01867.
This is based on Var (θˆ) = Var ( X ) = Var ( X ) / n = θ 2 / n .

Question #278
Key: A

Based on the information given

36 0.4 x
0.21 = +
n n
36 x 0.6 y
0.51 = + +
n n n
n = 200 + x + y.
Then,
0.21(200 + x + y ) = 36 + 0.4 x
0.51(200 + x + y ) = 36 + x + 0.6 y
and these linear equations can be solved for x = 119.37.

Question #279
Key: B

Pays 80% of loss over 20, with cap of payment at 60, hence cap of loss of
60
+ 20 = 95 = u
0.8

Th 5.13 E(Y per loss) = α ⎡⎣ E ( X ∧ 95 ) − E ( X ∧ 20 ) ⎤⎦

= 0.8 ⎡ ∫ S ( x ) dx − ∫ S ( x ) dx ⎤
95 20
⎢⎣ 0 0 ⎥⎦
95 ⎛ x2 ⎞ ⎛ x3 ⎞
S ( x ) dx = 0.8∫ ⎜ 1 − ⎟ = ( 0.8 )( 46.6875 ) = 37.35
95
= 0.8∫
95
⎟dx = 0.8 ⎜ x −
⎝ 10,000 ⎠ ⎝ 30,000 ⎠
20 20 20

E (Y per loss ) 37.35

E(Y per payment) = = = 38.91
1 − F ( 20 ) 0.96

C-09-08 - 111 -
Question #280
Key: D

Let S = aggregate claims, I 5 = claims covered by stop loss

S I5
0 0
5 0
x x- 5 (i.e., for any S > 5, I 5 = S − 5 )

E [ I 5 ] = E [ S ] − 5 + 5Pr ( 0 claims )
E [ S ] = 5 × ( 0.6 × 5 + 0.4 × k ) = 15 + 2k
Pr ( 0 claims ) = e −5
E [ I 5 ] = 15 + 2k − 5 + 5e −5 = 28.03
10.034 + 2k = 28.03
2k = 18
k =9

Question #281
Key: B

Let R = Equity index return. R ∼ N ( μ = 8%, σ = 16% )

X = 0.75 R X ~ N ( μ = 6%, σ = 12% )
Y = crediting rate = Max (R, 3%)
Y = X + 3% − ( X ∧ 3% )
(If X < 3%, X ∧ 3% = X and Y = X + 3% − X = 3%)
(If X > 3% X ∧ 3% = 3% and Y = X + 3% − 3% = X )
E (Y ) = E ( X + 3% − ( X ∧ 3%))
= E ( X ) + 3% − E ( X ∧ 3% )
= 6% + 3% − (−0.43%)
= 9.43%

Note that E ( X ∧ 3%) is a table lookup in the given information.

C-09-08 - 112 -
Question #282
Key: A

Let S = aggregate losses, X = severity

Since the frequency is Poisson,
( )
Var ( S ) = λ E X 2
22 Γ ( 3) Γ (1)
( )
E X2 =
Γ ( 3)
= 4 (table lookup)

Var ( S ) = 3 × 4 = 12
You would get the same result if you used
Var ( S ) = E ( N )Var ( X ) + Var ( N )( E ( X )) 2

Question #283
Key: D

−1
For each member P ( z ) = ⎡⎣1 − 1.5 ( z − 1) ⎤⎦
−4
so for family of 4 P ( z ) = ⎡⎣1 − 1.5 ( z − 1) ⎤⎦ negative binomial with β = 1.5 r = 4

k pk
0 0.026
1 0.061
2 0.092
3+ 0.821

E ( N ∧ 3) = 0 × 0.026 + 1× 0.061 + 2 × 0.092 + 3 × 0.821 = 2.71

E ( N ) − E ( N ∧ 3) = 6 − 2.71 = 3.29
3.29 ×100 per visit = 329

Alternatively, without using probability generating functions, a geometric distribution is a

special case of the negative binomial with r = 1.

Summing four independent negative binomial distributions, each with β = 1.5 and r = 1
gives a negative binomial distribution with β = 1.5 and r = 4. Then continue as above.

C-09-08 - 113 -
Question #284
Key: E

E ( X ∧ 2 ) = 1× f (1) + 2 (1 − F (1) )
= 1× f (1) + 2 (1 − f ( 0 ) − f (1) )

(
= 1× 3e −3 + 2 1 − e−3 − 3e−3 )
= 2 − 5e −3
= 1.75

Cost per loss with deductible = E ( X ) − E ( X ∧ 2 )

= 3 – 1.75
= 1.25

Cost per loss with coinsurance = α E ( X ) = 3α

Equating cost, 3α = 1.25
α = 0.42

Question #285
Key: A

Let N be the number of clubs accepted

X be the number of members of a selected club
S be the total persons appearing

N is binomial with m = 1000 q = 0.20

E ( N ) = (1000 )( 0.20 ) = 200
Var ( N ) = (1000 )( 0.20 )( 0.80 ) = 160

E ( S ) = E ( N ) E ( X ) = ( 200 )( 20 ) = 4000
Var ( S ) = E ( N ) Var ( X ) + Var ( N ) ⎡⎣ E ( X ) ⎤⎦
2

= ( 200 )( 20 ) + (160 )( 20 )
2

= 68,000

Budget = 10 × E ( S ) + 10 × Var ( S )
= 10 × 4000 + 10 × 68,000
= 42,610

C-09-08 - 114 -
Question #286
Key: C

Insurance pays 80% of the portion of annual claim between 6,000 and 1,000, and
90% of the portion of annual claims over 14,000.

The 14,000 breakpoint is where Michael reaches 10,000 that he has paid:
1000 = deductible
1000 = 20% of costs between 1000 and 6000
8000 = 100% of costs between 14,000 and 6,000

⎛ θ ⎞ 5000 x
E [ X ∧ x ] = θ ⎜1 − ⎟=
⎝ x +θ ⎠ x + 5000

x E [ X ∧ x]
1000 833.33
6000 2727.27
14000 3684.21
∞ 5000

0.80 ⎡⎣ E [ X ∧ 6000] − E [ X ∧ 1000]⎤⎦ + 0.90 ⎡⎣ E [ X ⎤⎦ − E ⎡⎣ X ∧ 14000]⎤⎦

⇒ 0.80 [ 2727.27 − 833.33] + 0.90 [5000 − 3684.21]
= 1515.15 + 1184.21 = 2699.36
⇒ 2700

Question #287
Key: D

We have the following table:

Item Dist E( ) Var ( )

Number NB (16, 6 ) 16 × 6 = 96 16 × 6 × 7 = 672
claims
Claims U ( 0, 8 ) 82 /12 = 5.33
8/ 2 = 4
amounts
Aggregate 4 × 96 = 384 96 × 5.33 + 672 × 42 = 11, 264

Premium = E ( S ) + 1.645* Sqrt (Var ( S ) ) =

= 384 + 1.645* Sqrt (11, 264 )
= 559

1.645 is the 95th percentile of the standard normal distribution.

C-09-08 - 115 -
Question #288
Key: E

With probability p, Prob ( N = 2 ) = 0.52 = 0.25

With probability (1 − p ) , Prob ( N = 2 ) = Combin ( 4, 2 ) *0.54 = 0.375

Prob ( N = 2 ) = p × 0.25 + (1 − p ) × 0.375

0.375 − 0.125 p

Question #289
Key: D

600 can be obtained only 2 ways, from 500 + 100 or from 6 × 100 .

Since λ = 5 and p (100 ) = 0.8, p ( 500 ) = 0.16 ,

e −5 56
p ( 6 claims for 100 ) = ( 0.8)6 = 0.03833 or 3.83%
6!

2 claims
e −5 52 ⎡
p ( 500 + 100 ) = ( 0.8 ) ( 0.16 ) ( 2 ) ⎤ = 0.02156 = 2.16%
1 1

2! ⎣ ⎦

The factor of 2 inside the bracket is because you could get a 500 then 100 or you
could
get a 100 then 500.

Total = 3.83% + 2.16% = 5.99%

C-09-08 - 116 -