Robust Generalized Additive Models for Telematics-Based Auto Insurance Ratemaking

Abstract

In this thesis, we introduce a different P&C ratemaking strategy using telematics where complexity of telematics data is often seen as a challenge for traditional Generalized Linear Modeling. Generalized Additive Model with its flexible model structure is outlined and recent applications in the insurance industry are discussed and analyzed. A robust version of the Generalized Additive Model is then discussed where the modified penalized likelihood is able to reduce the influence of outliers present in the data. With an application on a synthetic dataset, it is shown that our results coincide with the referenced paper of Dr. Jean-Philippe Boucher titled "Exposure as Duration and Distance in Telematics Motor Insurance Using Generalized Additive Models" (2017) and our model with the added telematics variable shows significant improvements. When outliers are introduced to the dataset, non-robust models quickly deteriorate and thus produce a poor fit whereas robust counterparts are able to maintain a similar level of model accuracy and as a result extreme risks are better identified from such policyholders. Actuaries can now utilize the added benefit of robust Generalized Additive Model for better risk classification such that a more fair pricing scheme is made possible.