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.

128 APPENDIX A. SOLUTIONS TO EXERCISES<br />

Exercise 90 The loglikelihood function for an exponential distribution is<br />

l(θ) =ln<br />

nY<br />

θ −1 e −xi/θ =<br />

i=1<br />

nX<br />

− ln θ − x i /θ = −n ln θ − n¯xθ −1 .<br />

i=1<br />

Under the null hypothesis that Sylvia’s mean is double that <strong>of</strong> Phil’s, the maximum likelihood<br />

estimates <strong>of</strong> the mean are 916.67 for Phil <strong>and</strong> 1833.33 for Sylvia. The loglikelihood value is<br />

l null = −20 ln 916.67 − 20000/916.67 − 10 ln 1833.33 − 15000/1833.33 = −241.55.<br />

Under the alternative hypothesis <strong>of</strong> arbitrary parameters the loglikelihood value is<br />

l alt = −20 ln 1000 − 20, 000/1000 − 10 ln 1500 − 15, 000/1500 = −241.29.<br />

The likelihood ratio test statistic is 2(−241.29 + 241.55) = 0.52 which is not significant (the critical<br />

value is 3.84 with one degree <strong>of</strong> freedom). To add a parameter, the SBC requires an improvement<br />

<strong>of</strong> ln(30)/2 =1.70. Both procedures indicate that there is not sufficient evidence in the data to<br />

dispute Sylvia’s claim.<br />

Exercise 91 With separate estimates, all but the female-nonsmoker group had 1 surrender out<br />

<strong>of</strong> a risk set <strong>of</strong> 8 lives (For example, the female-smoker group had seven observations with left<br />

truncation at zero. The remaining three observations were left truncated at 0.3, 2.1, <strong>and</strong> 3.4. At<br />

the time <strong>of</strong> the first year surrender, 0.8, the risk set had eight members.). For those three groups,<br />

we have Ĥ(1) = 1/8, <strong>and</strong>Ŝ(1) = exp(−1/8) = 0.88250. For the female-nonsmoker group, there<br />

were no surrenders in the first year, so the estimate is Ŝ(1) = 1. For the proportional hazards<br />

models, let z 1 =1for males (<strong>and</strong> 0 for females) <strong>and</strong> let z 2 =1for smokers (<strong>and</strong> 0 for nonsmokers).<br />

The maximum likelihood estimates for the exponential are ˆβ 1 =0.438274, ˆβ 2 =0.378974, <strong>and</strong><br />

ˆθ =13.7843. Letting the subscript indicate the z-values, we have<br />

Ŝ 00 (1) = exp(−1/13.7843) = 0.930023<br />

Ŝ 01 (1) = exp(−e 0.378974 /13.7843) = 0.899448<br />

Ŝ 10 (1) = exp(−e 0.438274 /13.7843) = 0.893643<br />

Ŝ 11 (1) = exp(−e 0.438274+0.378974 /13.7843) = 0.848518.<br />

For the data-dependent approach, the two estimates are ˆβ 1 =0.461168 <strong>and</strong> ˆβ 2 =0.374517. The<br />

survival function estimates are<br />

Ŝ 00 (1) = Ŝ0(1) = 0.938855<br />

Ŝ 01 (1) = 0.938855 exp(0.374517) =0.912327<br />

Ŝ 10 (1) = 0.938855 exp(0.461168) =0.904781<br />

Ŝ 11 (1) = 0.938855 exp(0.461168+0.374517) =0.864572.<br />

Exercise 92 For this model, S(t) =exp(−e βz t/θ) where z is the index <strong>of</strong> industrial production.<br />

This is an exponential distribution with a mean <strong>of</strong> θe −βz <strong>and</strong> a median <strong>of</strong> θe −βz ln 2. Thetw<strong>of</strong>acts<br />

imply that 0.2060 = θe −10β <strong>and</strong> 0.0411 = θe −25β ln 2. Dividing the first equation by the second<br />

equation (after dividing the second equation by ln 2) produces 3.47417 = e 15β for β =0.083024 <strong>and</strong><br />

then θ =0.47254. Then, S 5 (1) = exp[−e 0.083024(5) /0.47254] = 0.04055.

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

Saved successfully!

Ooh no, something went wrong!