Smoothing Techniques for Insurance Ratemaking Models

• Activity Number: 666
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Health Policy Statistics
• Author(s): Billie Anderson*+ and Susan Haller
• Companies: SAS Institute Inc. and SAS Institute Inc.
• Keywords: ratemaking ; insurance ; generalized additive models

Abstract

Insurance companies gain a competitive advantage by offering better rates and services to attract and retain the best customers. Generalized linear models (GLMs) have become popular and proven techniques for ratemaking and actuarial work over the past decade. The ability to smooth the parameter estimates that the GLMs produce is a very important aspect of ratemaking. Actuaries use the smoothed estimates from the GLMs in the calculation of premiums. The question aries, which smoothing method does one use? Generalized additive models (GAMs) lend themselves nicely to this ratemaking scenario because a GAM models the effect of explanatory variables in terms of non-parametric functional forms. Recently, generalized additive models for location, scale, and shape(GAMLISS) have been attracting attention in the insurance arena (Jong & Heller 2008) because they have all the properties of a GAM, but are more suited to model overdispersed count data which is a real problem in insurance ratemaking models. This paper will compare a GAM with GAMLISS as smoothing techniques using real world insurance data and conclude which modeling method is appropriate.