personal lines insurance. Focusing on homeowners insurance, this presentation provides a systematic comparison of several predictive generalized linear models. We compare pure premium (Tweedie) and frequency/severity models based on single perils as well as multiple perils. With multiple perils, we also introduce instrumental variable models that account for interdependencies among perils. We calibrate these models using a database of detailed individual policyholder experience. We then evaluate the many alternative models in several ways.
We also review how loss trends can be affected by perils trending at different rates, causing changes in peril mix over time. Using homeowners data we show an example of how multivariate trend analysis can be used to analyze homeowners loss trends by peril. Economic conditions may affect all perils simultaneously, leading to correlation in loss trend for different perils. There are times, however, when some perils exhibit trends that are significantly different from other perils. Multivariate stepwise regression can be used to detect data that have significant changes in peril mix or independent trends by peril. This type of analysis can lead to questions of interest for the actuary in regards to trend factor selection, and for the product manager in terms underwriting, product design, or both.

Moderator/Panelist:
Hernan L. Medina, Senior Research Associate, Insurance Services Office

Panelist:
James Calton, Director of Analytics, ISO Innovative Analytics