Actuarial pricing, capital modelling and reserving

Pricing Squad

Issue 5 -- August 2016

Welcome back to Pricing Squad!

Pricing Squad is a newsletter for fellow pricing practitioners and actuaries in general insurance. Enjoy, and let me know your comments and ideas for future issues.

Today's issue is about classifying rating factors, such as occupation, line of business, car make etc. These granular risk factors tend to have hundreds of levels which often need to be classified into a few robust groups.

Classify with class

Recently I was asked to finalise a loss-cost GLM model which had taken ages to get through underwriting approval. The bottleneck in the process was the inclusion of hundreds of occupation levels. They needed to be organised into a dozen or so neat groups.

Perhaps you've come across this problem too when classifying areas or vehicle makes in personal insurance, or lines of business in commercial insurance?

Triage logic

In the end, the following common-sense and time-tested triage logic saved everyone a lot of time, and eliminated subjectivity.

First, I tabularised model residuals by the factor being classified and by another independent factor, namely the underwriting year. I used GLM prediction divided by capped actual claims as residuals.

Based on the number of underwriting years available (five), I decided that we needed four out of five years' data behaving consistently to claim statistical significance, a figure based on a binomial distribution and a 95% confidence interval.


... want to continue reading?

Please provide your email to log in or to subscribe to Pricing Squad: