Theorem 71 For a loglink function a normal asymptotic approximation gives π i Φ

# Theorem 71 for a loglink function a normal asymptotic

• 9

This preview shows page 7 - 9 out of 9 pages.

Theorem 7.1. For a log–link function, a normal asymptotic approximation gives π i = Φ( ln(1 + r ) s i ) - Φ( ln(1 - r ) s i ) , (7.8) where Φ is the standard normal cdf and s i are as in (7.7). *Proof: see Garrido and Zhou (2009). square For the case of a general link function g , let Q 1 = g [(1 - r ) μ i ] - g ( μ i ) , (7.9) Q 2 = g [(1 + r ) μ i ] - g ( μ i ) . (7.10) Theorem 7.2. For a general link function g , an approximate normal asymp- totic result gives π i = Φ( Q 2 s i ) - Φ( Q 1 s i ) , (7.11) where Φ is the standard normal cdf and s i as above. For a GLM model, these ˆ π i can be used as a criteria to choose between different possible link functions, say g 1 and g 2 . Corollary 7.2. If ˆ π ( g 1 ) i < ˆ π ( g 2 ) i , we say that the estimator given under the link function g 1 is less credible than the estimator given under g 2 , that is g 2 is better than g 1 . Example 7.3. (Car Insurance Claims) The SAS Technical Report P–243 (1993) gives the following illustrative dataset of a car insurance portfolio (also reproduced in Schmitter, 2004).

Subscribe to view the full document.

76 CHAPTER 7. CREDIBILITY FOR GLM’S Table 7.2: Car Insurance Data Class # of Risks # of Claims Car Type Age Group i = 1 500 42 small 1 2 1200 37 medium 1 3 100 1 large 1 4 400 101 small 2 5 500 73 medium 2 6 300 14 large 2 Let x i be Poisson and g be a log–link function. Here the covariates are Y i = ( y i 0 , . . . , y i 3 ) prime , where y i 0 = 1 , y i 1 = braceleftbigg 1 if car type is large 0 otherwise , y i 2 = braceleftbigg 1 if car type is medium 0 otherwise , y i 3 = braceleftbigg 1 if age group is 1 0 otherwise . In this notation Y 4 = (1 , 0 , 0 , 0) prime defines the base premium E ( X 4 ) = e β 0 for a small car type in age group 2 . The matrix of variance–covariance Σ in Proposition 7.3 is computed with SAS for weights w 1 = . . . = w 6 = 1, since here φ = 1, g ( μ ) = ln μ and V ( μ ) = μ : Σ = 0 . 008150 - 0 . 007772 - 0 . 006344 - 0 . 004623 - 0 . 007772 0 . 07418 0 . 006556 0 . 003113 - 0 . 006344 0 . 006556 0 . 01645 - 0 . 002592 - 0 . 004623 0 . 003113 - 0 . 002592 0 . 01847 . Let the tolerance level r = 0 . 1 and Y 3 = (1 , 1 , 0 , 1) prime for the third class of drivers, with a large car type in age group 1 . Then the asymptotic value in (7.7) for s 2 3 = Y prime 3 Σ Y 3 = 0 . 082236 and from (7.8) we get π 3 = Φ( ln(1+ r ) s 3 ) - Φ( ln(1 - r ) s 3 ) = 0 . 273533. Clearly, the current experience produces GLM estimators that are not credible in this class with only one claim, as s 2 3 = 0 . 082236 > 0 . 00410 = s 2 * , for r = 0 . 1 and π = 90%.
7.3. CREDIBILITY FOR GLMS 77 By contrast, if Y 1 = (1 , 0 , 0 , 1) prime gives s 2 1 = 0 . 01737 and π 1 = 0 . 55314, which indicates a more credible GLM estimator for small cars in age group 1 than for large cars in age group 1 , although not sufficient for full credibility s 2 1 = 0 . 017374 > 0 . 00410 = s 2 * . Table 7.3 reports the asymptotic variances s 2 i = V ( Y prime ij β ) Y prime i Σ Y i and the credibility probabilities π i for all 6 classes.
• Fall '09
• Dr.D.Dryanov

### What students are saying

• As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

Kiran Temple University Fox School of Business ‘17, Course Hero Intern

• I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

Dana University of Pennsylvania ‘17, Course Hero Intern

• The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

Jill Tulane University ‘16, Course Hero Intern

Ask Expert Tutors You can ask 0 bonus questions You can ask 0 questions (0 expire soon) You can ask 0 questions (will expire )
Answers in as fast as 15 minutes