01.08.2014 Views

Estimation, Evaluation, and Selection of Actuarial Models

Estimation, Evaluation, and Selection of Actuarial Models

Estimation, Evaluation, and Selection of Actuarial Models

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

3.2. MEASURES OF QUALITY 43<br />

When multiplied by 1.2, the sample median has second moment<br />

Z ∞<br />

E[(1.2Y ) 2 ] = 1.44 y 2 6 ³e −3y/θ´<br />

−2y/θ − e dy<br />

0 θ<br />

= 1.44 6 · µ −θ<br />

y 2<br />

θ 2 e−2y/θ + θ µ θ<br />

2<br />

<br />

3 e−3y/θ − 2y<br />

4 e−2y/θ − θ2<br />

9 e−3y/θ<br />

µ −θ<br />

3<br />

∞<br />

+2<br />

= 8.64<br />

θ<br />

8 e−2y/θ + θ3<br />

27<br />

¸¯¯¯¯<br />

e−3y/θ<br />

µ 2θ<br />

3<br />

<br />

8 − 2θ3 = 38θ2<br />

27 25<br />

for a variance <strong>of</strong><br />

38θ 2<br />

25 − θ2 = 13θ2<br />

25 > θ2<br />

3 .<br />

The sample mean has the smaller MSE regardless <strong>of</strong> the true value <strong>of</strong> θ. Therefore, for this problem,<br />

it is a superior estimator <strong>of</strong> θ. ¤<br />

Example 3.15 For the uniform distribution on the interval (0, θ) compare the MSE <strong>of</strong> the estimators<br />

2 ¯X <strong>and</strong> n+1<br />

n max(X 1,...,X n ). Also evaluate the MSE <strong>of</strong> max(X 1 ,...,X n ).<br />

The first two estimators are unbiased, so it is sufficient to compare their variances. For twice<br />

thesamplemean,<br />

Var(2 ¯X) = 4 4θ2<br />

Var(X) =<br />

n 12n = θ2<br />

3n .<br />

For the adjusted maximum, the second moment is<br />

0<br />

E<br />

" µn +1<br />

n<br />

Y n<br />

2<br />

#<br />

=<br />

(n +1)2 nθ 2<br />

n 2 n +2 = (n +1)2 θ 2<br />

(n +2)n<br />

for a variance <strong>of</strong><br />

(n +1) 2 θ 2<br />

(n +2)n − θ2 =<br />

θ 2<br />

n(n +2) .<br />

Except for the case n = 1 (<strong>and</strong> then the two estimators are identical), the one based on the<br />

maximum has the smaller MSE. The third estimator is biased. For it, the MSE is<br />

nθ 2<br />

(n +2)(n +1) 2 + µ nθ<br />

n +1 − θ 2<br />

=<br />

2θ 2<br />

(n +1)(n +2)<br />

which is also larger than that for the adjusted maximum. ¤<br />

Exercise 53 For the sample <strong>of</strong> size three in Exercise 51, compare the MSE <strong>of</strong> the sample mean<br />

<strong>and</strong> median as estimates <strong>of</strong> θ.<br />

Exercise 54 (*) You are given two independent estimators <strong>of</strong> an unknown quantity θ. For estimator<br />

A, E(ˆθ A ) = 1000 <strong>and</strong> Var(ˆθ A ) = 160, 000, while for estimator B, E(ˆθ B ) = 1200 <strong>and</strong><br />

Var(ˆθ B )=40, 000. Estimator C is a weighted average, ˆθ C = wˆθ A +(1− w)ˆθ B . Determine the<br />

value <strong>of</strong> w that minimizes Var(ˆθ C ).

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!