Couret and Venter

Question 1

Key Idea of Multi-Dimensional Credibility

Answer 1

Frequency of different injury types correlated

Can weight between class and hazard group by injury type for more accurate estimate

Question 2

V, W, X, Y

Answer 2

V = Fatal to TT Ratio
W = Perm to TT
X = Major Partial to TT
Y = Minor Partial to TT

Question 3

Quintiles Test vs SSE

Answer 3

Quintiles is better
SSE impacted by random noise at class level
quintiles has more cred since grouped into quintiles

Question 4

Quintiles Test Steps

Answer 4

1. Sort by predicted ratio
2. Group quintiles by TT counts
3. Calc Actual (Holdout) ratio, raw Training ratio, and predicted ratio
4. Make these relativities to each total and include all 1s for HG total
5. Calc SSE to Actual for 3 Expecteds

Question 5

Data Split for Testing (Multi-Dim Credibility)

Answer 5

Even vs Odd years so emerging trends present in both datasets

Question 6

Purpose of Holdout Sample

Answer 6

Test model on real data not used to fit model.

Gives estimate of predictive power.

Question 7

Other Things to Look for in Quintile Test (Couret/Venter)

Answer 7

prediction vs holdout

if not as steep as holdout -> understating credibility
if too steep -> overstating credibility