17.
3ZF2 Survival Models II I D Currie
Tutorial 7: Outline Solutions
1. (a) 2-test This will have 18 4 = 14 d.f. and value
X (d E q_ )2 = 20:1: x x x
E q_ (1 q_ )x x x x
This is not signi cant at the 5% level since 205 14 = 23:68. : ;
(b) Distribution of Standardized Deviations There is scarcely enough data to
perform this test but going through the routine and doing some grouping we nd
Deviation range Expected Actual (A E )2=E
1 to 3 0.00 0
3 to 2 0.36 1 0.436
2 to 1 2.52 3
1 to 0 6.12 7 0.126
0 to 1 6.12 6 0.002
1 to 2 2.52 1
2 to 3 0.36 0 1.227
3 to 1 0.00 0
and 2 = 1:791, not signi cant at the 5% level on 3 degrees of freedom.
(c) Sign test We have observed 7 positive and 11 negative signs. Let X B(18; 0:5).
Then Pr(X 7) = 0:2403 from NCST so there is no suggestion of any non-
randomness in the signs.
(d) Change of sign test There are 14 changes of sign and 3 non-changes. Clearly, no
test is required since we are concerned with too few changes. (Let X B(17; 0:5).
Then Pr(X 14) = 0:9988 from NCST { very large, as expected.)
(e) Cumulative deviations test We'll compute this test for the randomly chosen
ranges up to 34.5, 54.5 and 74.5, and the whole age range. We nd
qP
Age up to P(d E q_ ) x x x E q_
x (1 q_ )
x x
34.5 10 67.5
54.5 + 5 177.2
74.5 83 287.8
All 101 327.4
and there is nothing suspicious here.
This study source was downloaded by 100000778279571 from CourseHero.com on 05-18-2025 00:25:47 GMT -05:00
https://www.coursehero.com/file/54812110/solution7pdf/
(f) Grouping of signs test There are 7 +'s, 11 's, and 7 groups of +'s. This
seems very unexceptional. The expected number of + groups is
n1 (n2 + 1) 7 12
n1 + n2
= 18 = 4:7
so we have seen more groups than we'd expect under the hypothesis of random-
ness.
(g) Serial correlation test We compute the lag-1 serial correlation r1 0:168
and refer to N (0; 1=17) so the observed value of r1 has a z-score of 0:694. We
are on the look out for large positive values of r1 so the observed negative value
of r1 is of no concern. This is consistent with the result for the change of sign test
above.
Suitability for premiums and reserves
For assurance business, we want to overestimate mortality if we must err at all. For
endowment business the graduation does not show any marked tendency to lie below
the observed rates, and is satisfactory. All the deviations at ages over 85 are negative,
however, which might lead to reserves under whole of life contracts being too low at
these ages. This is not likely, in modern conditions, to have much impact, but it might
make the graduation unsuitable for a published table.
2. (a) It is diÆcult to establish degrees of freedom. One rule of thumb is to deduct 2 or
3 for each range of 10 or so age groups; this represent roughly the tting of the
level, slope and (possibly) the curvature of the curve.
(b) (i) Use the standardised deviations test. Notice that all standardised deviations
fall in the range -1 to 1 so it looks as though we have too few large deviations!
With the usual intervals we get:
Range < -3 (-3,-2) (-2,-1) (-1,0) (0,1) (1,2) (2,3) > 3
Actual 0 0 0 9 7 0 0 0
Expected 0 0.32 2.24 5.44 5.44 2.24 0.32 0
Taking the rst three and the last three cells together we get 2 = 7:9 which
is signi cant at the 5% level on 3 d.f. However, we've really detected too
good a t!
(ii) We could use the sign test, or the accumulated deviations test. For the sign
test we have 7 positive and 9 negative signs, so no test is required. The
accumulated deviations test is usually very poor over the full age range: this
is no exception. We nd
X(d E q_ ) = 13:36
X E q_ (1 q_ ) X E q_ = 782:37
x x x
x
x x x x x
x x
so the z value of the accumulated deviations is only 0:477.
This study source was downloaded by 100000778279571 from CourseHero.com on 05-18-2025 00:25:47 GMT -05:00
https://www.coursehero.com/file/54812110/solution7pdf/
(iii) You could use the change of sign test or the grouping of signs test. We observe
8 sign changes out of possible 15, so no suggestion of too few sign changes
here. With the grouping of signs test there are n1 = 7 +'s and n2 = 9 's so
the mean number of groups of positive signs is 4.4. The observed value is 4.
No alarms here!
(c) We test the smoothness by drawing up a table of third di erences. We have
(1; 000)
q_ q_ 2q_ 3q_
x x x x
1.747 0.365
2.112 0.916 0.551 0.797
3.028 0.670 0.246 0.674
3.698 1.098 0.428 1.074
4.796 0.452 0.646 1.752
5.248 1.558 1.106 0.555
6.806 2.109 0.551 0.104
8.915 2.764 0.655 4.546
11.679 7.965 5.201 4.085
19.644 9.081 1.116 1.033
28.725 11.230 2.149 15.918
39.955 29.297 18.067 5.813
69.252 41.551 12.254 16.258
110.803 70.063 28.512 12.852
180.866 111.427 41.364
292.293
Clearly the graduation is not very smooth. This explains the very small deviations
- the graduator has adhered too closely to the crude rates. Apart from that, the
other tests (for positive or negative signs and clumping) are satisfactory. An
inspection by eye, however, suggests that the graduated rates are signi cantly
too high at high ages, a point not brought out by the tests because of the small
weightings at these ages.
The graduation should certainly be rejected.
This study source was downloaded by 100000778279571 from CourseHero.com on 05-18-2025 00:25:47 GMT -05:00
https://www.coursehero.com/file/54812110/solution7pdf/
Powered by TCPDF (www.tcpdf.org)
