Predictive modeling for usage-based auto insurance

Usage-based auto insurance, also known as UBI, involves analyzing data collected from policyholders’ vehicles via telematics to help determine premium rates. Behavioral information considered includes vehicles’ speeds, maneuvers, routes, mileage, and times of day of operation. UBI has been described as a potentially significant advancement over traditional techniques that rely on information such as policyholders’ ages as proxies for how riskily they drive. However, because data collected via telematics are volatile and voluminous, particular care must be taken by actuaries and data scientists when applying predictive modeling techniques to avoid overfitting or nonconvergence and to improve predictive power. In this chapter, we use a case study to evaluate how modeling techniques perform in a UBI environment and how various challenges may be addressed.