Usage-Based Vehicle Insurance: Driving Style Factors of Accident Probability and Severity
This work addresses the need for more accurate and efficient usage-based insurance pricing for insurers, but it appears incremental as it builds on existing telematics methods without introducing major breakthroughs.
The paper tackles the problem of using telematics data to predict vehicle accident probability and severity by analyzing device types, processing data formats, and developing classification and modeling approaches, resulting in estimated models with in-sample and out-of-sample quality assessments.
The paper introduces an approach to telematics devices data application in automotive insurance. We conduct a comparative analysis of different types of devices that collect information on vehicle utilization and driving style of its driver, describe advantages and disadvantages of these devices and indicate the most efficient from the insurer point of view. The possible formats of telematics data are described and methods of their processing to a format convenient for modelling are proposed. We also introduce an approach to classify the accidents strength. Using all the available information, we estimate accident probability models for different types of accidents and identify an optimal set of factors for each of the models. We assess the quality of resulting models using both in-sample and out-of-sample estimates.