Escolar Documentos
Profissional Documentos
Cultura Documentos
Exam Notes:
Make sure you have the ST exam tables on hand before you start the exam.
Total number of claims is given by a Poisson process with claims intensity (t) = 10t, t > 0.
At any time the probability of a claim being from Claim Type A is 0.6 and from Claim Type B is
0.4.
Calculate the probability that by time 0.5 there will be fewer than two claims with severity at least
equal to 1000.
What is the proability that an occuring claim, X, has severity at least equal to 1000? Compute this by
conditioning on the type of the claim.
The number of claims by time 0.5 is therefore a Poisson random variable with parameter
Z 0.5 0.5 !
10t2
(0.48) 10tdt = (0.48) = 0.6.
0 2 0
The probability that this is fewer than 2 is the probability that it is equal to zero or one, e0.6 +0.6e0.6 =
0.878098618 , which fits into the range of answer choice E.
You get a ride to work from either a bus or a taxi, whichever arrives first.
Calculate the probability you will have to wait more than 10 minutes for a ride to work.
The combination of two independent Poisson processes is a new Poisson process with rate equal to the
sum of the rates of the component processes. That means that some kind of vehicle arrives as Poisson
Process with rate 15. The probability that the first waiting time from this process is more than 10
minutes (= 1/6 of an hour) is the probability that an exponential random variable with mean 1/15 is
more than 1/6. This is
P[X > 1/6] = e(1/6)/(1/15) = 0.082084999 .
This fits into the range of answer choice B. Note that this answer also seems reasonable, since we expect
a Bus about every 12 minutes and a taxi about every 6 minutes, making it relatively unlikely that the
wait would be longer than 10 minutes for some kind of vehicle to arrive.
Calculate E[S(4)2 ]
The second moment of the process up to time 4 is the variance plus the first moment squared. We recall
that
Z 4
t 4 3.14
E[S] = e dt E[X] = (1 e ) = 1.541244447,
0 2
Z 4
(0 3.14)2
t 2 4
Var[S] = e dt E[X ] = (1 e ) = 3.226338376.
0 3
The second moment of the compound process, then, is 3.226338376 + 1.5412444472 = 5.601772821 ,
answer choice E.
A. Less than 10
B. At least 10, but less than 11
C. At least 11, but less than 12
D. At least 12, but less than 13
E. At least 13
We recognize this density as Weibull with = 2. Knowing that this is Weibull isnt that helpful unless
you have memorized the MLE for the Weibull. Compute it as usual, via the likelihood and log likelihood
functions. The sum of the squared observations is 781.
5 Y
2 2 Xi2
Xi e(1/ )
P
L() =
2
P 2
Y Xi
`() = 5 ln(2) 2 5 ln() + ln( (Xi ))
2
P 2
25 X
`0 () = 0 +0+2 3 i
781 2 5
0=2
3
0 = 2 781 2 5 2
r
781
= = 12.49799984 .
5
Calculate the value of c that minimizes the mean squared error of the estimator of .
It is a bit shorter to compute directly, but in any case you should skip this one on the first pass.
MSEcX () = E (cX )2
= E c2 X 2 2cX + 2
= c2 E[X 2 ] 2cE[X] + 2
2E[X]
This is a quadratic equation in c, so its minimium is at the vertex, where c = 2E[X 2 ]
. We need the first
and second moments of the Pareto, computing from the tables.
2 2 2
E[X] = = , E[X 2 ] = = .
51 4 (5 1)(5 2) 6
2
2 2
10
Var[X] = = .
6 4 96
If you only recall that the MSE is the variance plus the square of the bias, then the variance and bias
are
c2 c2 Var(X)
Var cX = 2 n Var(X) = ,
n n
c2 Var(X)
MSE(cX) = + (cE[X] )2
n
2
c2 102
= + c
96 15 4
2 2
c c c
= 2 + 2 +1
144 16 4
2
5c c
= 2 +1
72 2
To minimize this,
5c2
d c 10 1
+1 =c
dc 72 2 72 2
10 1
0= c
72 2
1
2 72
c= 10 = = 3.6.
72
20
X P(X = x)
0
1 2
2 1 3
A. Less than 2
B. At least 2, but less than 4
C. At least 4, but less than 6
D. At least 6, but less than 8
E. At least 8
To find the expected value of the maximum, start with the CDF.
x n
P[Xmax x] = (P[X x])n =
x n1 1 nxn1
fXmax (x) = n =
n
Z
nxn1 cn n cn n+1
Z
cn
E[cXmax ] = c x n
= n
x dx = n
=
0 0 n+1 n+1
To make the estimator unbiased, then, take c = (n + 1)/n = 101/100 = 1.01 , which fits into the range
of answer choice B.
Age Group
Gender Young Old Total
Male 20 15 35
Female 12 22 34
Total 32 37 69
A Chi-squared hypothesis test is performed to determine if gender and age group are independent.
The overall proportion of males is 35/69 = 0.507246377. Store this in the x variable of your calculator.
Now create the table of expected outcomes via the uniform distribution, if gender and age group are
independent.
Now we can compute our Chi-squared statistic, which will have (2 1)(2 1) = 1 degree of freedom.
H0 : = 4
Ha : = 7
A type II error occurs when the null is accepted given that the alternative is true.
" #
X 7 7.25 7
P[X < 7.25 | = 7] = P p <p
25/80 25/80
P [N (0, 1) < 0.447213596] = (0.45) = 0.6736
which is in the range of answer choice E, but is a bit close to the cutoff. We could use interpolation to
be more sure, but even (0.44) > 0.65, so even with interpolation we would be sure to obtain a result
above 0.65.
Samples from X and Y are taken with sizes 40 and 50, respectively
P P
Xi Yi
40 = X = 80.1 and 50 = Y = 78.8
(Xi X)2 (Yi Y )2
P P
40 = 34.8 and 50 = 25.0
H0 : x = y
H1 : x > y
For a two-sample test of means in the absence of known variances, we compute a T -statistic with
40 + 50 2 = 88 degrees of freedom,
X Y
p
2
S (1/40 + 1/50)
where
(Xi X)2 + (Yi Y )2
P P
2 34.8 40 + 25.0 50
S = = = 30.0227.
40 + 50 2 88
80.178.8 1.3
The T -statistic is = 1.162335032 = 1.118438285. Unfortunately, this means that the
30.0227(1/40+1/50)
T -table is of little help to us, since our statistic is smaller than the 0.05 level (for a one-sided test -
look in 0.1 column) critical value at all degrees of freedom from 60 on down. So we know that our
p-value is above 0.05, but not how far above it is. Our only recourse is to recall that the T -distribution
is approximated by the normal distribution for the large n case. On the normal table, a z-score of 1.118
would correspond to a probability of about 0.867, giving us an approximate p-value of 10.867 = 0.133 .
This fits into the range of answer choice C.
Territory X: A random sample of 100 claims is drawn from a normal distribution with variance
10,000, and has a sample mean of 300
Territory Y : A random sample of 70 claims is drawn from a normal distribution with variance
5,000, and has a sample mean of 250
Under the null-hypothesis, the mean from X minus the mean from Y is 30. The variance of the difference
in the sample means is
1 1
Var(X Y ) = Var(X) + Var(Y ) = 171.4285714.
100 70
The z-score is
X Y 30
z= = 1.527525232.
171.4285714
The p-value of this one-sided test, is 1 (1.527) 1 0.7(0.9370) 0.3(0.9357) = 0.06339 which fits
into the range of answer choice D.
Outcome X Y
1 94 157
2 244 168
3 135 197
4 158 136
5 128 220
6 241 122
H0 : PX = PY
H1 : PX 6= PY
Calculate the minimum significance level at which the null hypothesis will be rejected.
The proportion of successes observed for die X is 0.527, and the proportion of successes observed for die
Y is 0.478. Under the null hypothesis, the variance of each trial is the same, p(1p), where p = PX = PY .
The variance of PX PY is therefore estimated under the null as p(1 p)/1000 + p(1 p)/1000, which
0.5270.478
with an estimate of p = 0.5025 gives a z-score of = 2.19. The p-value of our
20.5025(10.5025)/1000
two-sided test is therefore 2(1 0.9857) = 0.0286 , which fits into the range of answer choice B.
For the Exponential distribution, the natural logarithm of the likelihood function evaluated at the
maximum likelihood estimate is -828.37.
For the Weibull distribution, the natural logarithm of the likelihood function evaluated at the
maximum likelihood estimate is -826.23.
Calculate the significance level at which the Weibull distribution provides a better fit than the Expo-
nential distribution.
The natural log of the likelihood ratio times negative two is a Chi-squared statistic with 1 degree of
freedom, since the exponential random variable is a Weibull random variable with specified. Since
we are given the natural logs of the likelihood functions at their maximum values, the statistic is
2(828.37 (826.23)) = 4.28. This is between the critical values in the Chi-square table with d.o.f.
1 of 0.95 and 0.975, so our p-value is between 0.025 and 0.05, which falls into the range of answer choice
D.
Students are ranked in descending order for both the number of practice exams and the score on the
final exam.
Calculate Spearmans rank correlation coefficient between the number of practice exams and the score
on final exam.
Now via the MultiView calculator, entering the practice exam rank into L1 and the Exam score rank
into L2, we find that rS = 0.6428571429 , which fits into the range of answer choice D.
While the provided table of values for the Wilcoxon Signed-rank test is only for the sum of either the
positive ranks or of the negative ranks, in either event a positive value, the actual Wilcoxon Signed-rank
statistic is the sum of the positive ranks minus the sum of the negative ranks, which can be a negative
number. The differences from the hypothesized median, 2.3, are
4 8 6 2 3 7 1 5
The positive ranks sum to 11, and the negative ranks sum to 25, so that the formal Wilcoxon Signed-
rank statistic is 11 25 = 14 . This corresponds to answer choice E.
Let Y1 < Y2 < Y3 < Y4 < Y5 be the order statistics of five independent observations from a unifonn
distribution on (0,10).
A. Less than 2
B. At least 2, but less than 4
C. At least 4, but less than 6
D. At least 6, but less than 8
E. At least 8
The density of the uniform is 1/10, and the CDF is y/10. The density of Y4 is
5!
g4 (y) = (y/10)3 (1 y/10)1 (1/10) = 2(y/10)3 (1 y/10)
3! 1!
The variable Y4 is therefore a Beta distribution with = 10, a = 4 and b = 2, From the tables,
10 4 20
E[Y4 ] = =
6 3
10245 1000
E[Y42 ] = =
67 21
2
1000 20 200
Var(Y4 ) = = = 3.1746031746 .
21 3 63
This fits into the range of answer choice B. If you dont make the connection with the Beta distribution
things are tougher.
Z 10
E[Y4 ] = y 2(y/10)3 (1 y/10)dy Let u = y/10.
0
Z 1
= 10u 2u3 (1 u)(10du)
0
Z 1
2
= 2 10 (u4 u5 )du
0
5
!
6 1
2 u u 2 1 1 20
= 2 10 = 2 10 =
5 6 0 5 6 3
Z 10
E[Y42 ] = y 2 2(y/10)3 (1 y/10))dy again, u = y/10
0
Z 1
= 2 102 u2 u3 (1 u)(10du)
0
Z 1
3
= 2 10 u5 u6 du
0
1 !
u6 u7
3 1 1 1000
= 2 10 = 2 103 = .
6 7 0 6 7 21
P[Y12 t] = t12
P[0.6 Y12 0.75] = P[Y12 0.75] P[Y12 0.6]
= 0.7512 0.612 = 0.02949957 ,
You wish to test whether the average sales prices are the same among the dealers:
H0 : A = B = C (The mean sales price between the three dealers is the same)
H1 : The mean sales price between the three dealers is not the same.
Create an ANOVA table and assume the total sum squares, SS(TO) = 81.6
We are given the sums of squares errors for each of the three dealerships, so to find the total sum of
squares errors, we add these together to get 46.78. The sum of squares treatement is the total less the
error, 81.6 46.78 = 34.82.
The F -statistic is
SSTr
31
F2,12 = SSE
= 4.466
153
Looking at the second column of the F -table, we see that this result is between the 0.05 critical value
(3.88) and the 0.01 critical value (6.93), so we will reject at the 0.05 level, but not at the 0.01 level.
Answer choice B.
0 = 2.31
1 = 1.15
0 = 0.057
1 = 0.043
H0 : 1 = 1
H1 : 1 6= 1
Determine the minimum significance level at which the null hypothesis would be rejected.
1 1 1.15 1
T62 = = = 3.48837209
1 0.043
In the T -table, the result lies between the 0.05 critical value (2.776) and the 0.02 critical value (3.747)
of the fourth row. Our T -table naturally tests two sided tests, which we want here, so we would reject
at the 0.05 level, but not at the 0.02 level, corresponding to answer choice C.
Assume that the distribution of each Xi is normal with mean i and common variance 2
A. Less than 9
B. At least 9, but less than 13
C. At least 13, but less than 17
D. At least 17, but less than 21
E. At least 21
This is a Poisson/Gamma pair with k = 27 claims observed in m = 1 period. The posterior is Gamma
with 0 = 3 + 27 = 30, and = 1/2+1
1
= 2/3. The mean of this posterior is 30 23 = 20 , which leads to
answer choice D.
X1 , . . . , X5 are a random sample from a Poisson distribution with parameter , where follows a
gamma distribution with parameters = 2 and .
The mean of this Poisson-gamma conjugate pair can be represented as a weighted average of the
maximum likelihood estimator for the mean and the mean of the prior distribution.
Calculate WMLE .
From bullet one, the prior distribution is Gamma with = 2. From bullet five, the variance of this gamma
is 8, which allows us to compute .
Var(Prior) = 2 ()2
8 = 3 22 42 = 22
= 2.
Im assuming that the mean of the conjugate pair refers to the mean of the posterior or predicitive
distribution, which are the same for the Poisson/Gamma case. Since the MLE for the mean is X, we
find that the number of exposures is n = 5 and the number of claims observed is nX = 5 1.2 = 6.
The posterior is therefore Gamma with 0 = 2 + 6 and 0 = 1/2+5
1
= 2/11. The mean of the posterior is
8 2/11 = 16/11.
To find the weight WMLE , set up the weighted average, set it equal to 16/11, and solve.
X follows the Normal distribution with parameters N (, 2 ) where 2 , the variance, is known to
be 2.
is assumed to follow a Normal distribution with parameters N (1, 4) prior to gathering a random
sample
A random sample of size 10 is collected to develop a posterior estimate of and the posterior
estimate of is equal to 1.2
This is a Normal/Normal pair. For a N (, ) with N (, a), the posterior mean is found by computing
n
z n+/a , and then using the formula
(1 z) + z(X)
n 10
Here, z n+/a = 10+2/4 = 20/21, and given that the posterior mean turns out to be 1.2,
1.2 = (1 z) + z(X)
1.2 = 1(1 20/21) + 20/21(X)
25.2 = 1 + 20X
X = 1.21
This is a Beta/Binomial pair. The number of exposures is 15 n, with k = 10 claims observed, to put
things in our usual language. The posterior is Beta with a0 = 4 + 10, b0 = 6 + 15 n 10. The Bayesian
14 14
estimate for is the mean of the posterior, 14+15n4 = 10+15n . At this point you have some thinking
to do. If the CAS meant for this to be a Bernoulli distribution instead of a Binomial, then n = 1, the
result is 0.56, and this fits into answer choice B. This seems to be the most likely interpretation for this
problem.