A statistical modeling approach for car insurance pricing with telematics data

We analyze telematics data from a Belgian portfolio of young drivers who underwrote a pay-as-you-drive insurance product in between 2010 and 2014. Using installed black box devices, telematics data are collected on how many kilometers are driven, during which time slots and on which type of roads. Car insurance is traditionally priced based on self-reported information from the policyholder, most importantly: age, license age, postal code, engine power, use of the vehicle, and claims history. However, these observable risk factors are only indirect indicators of the accident risk and don’t reflect the real driving behavior. By constructing predictive models for the claim frequency, we compare the performance of different sets of predictor variables (e.g. traditional vs purely telematics) and discover the relevance and impact of adding telematics insights. In particular, we contrast the use of time and distance as exposure-to-risk, the basic rating unit underlying the insurance premium. We show how to incorporate the divisions of the distance driven by road type and time slots as compositional predictors in the model and how to interpret their effect on the average claim frequency. We find that the telematics variables increase the predictive power and render the use of gender as a discriminating rating variable redundant.status: publishe