Ultimate Loss Confidence Interval

[hippickles]

What's the best way to get a confidence interval around ultimate loss estimates when using "a weighted average based on assigned weights to the various methods" as described in Friedland?

I understand there are different methods which produce confidence intervals (Mack, bootstrap as in England/Verrall, GLMs), but I'm not sure of the best way to estimate a confidence interval when selecting more than one method.

[my_alt]

Originally Posted by hippickles

What's the best way to get a confidence interval around ultimate loss estimates when using "a weighted average based on assigned weights to the various methods" as described in Friedland?

I understand there are different methods which produce confidence intervals (Mack, bootstrap as in England/Verrall, GLMs), but I'm not sure of the best way to estimate a confidence interval when selecting more than one method.

Do it the same way you deal with picking any other estimate. Run multiple models, compare the results they give you, and pick the number that you think is best.

Since your selected expected level ultimates/reserves won't correspond to the expected level ultimates/reserves from any model, you'll want to look at % changes instead of raw dollars. Other than that, it's like anything else you do.