Investigating the effect of climate-related hazards on claim frequency prediction in motor insurance with incomplete data

A climate-related dataset provided by a Greek insurance company is analysed to quantify the risks that weather-related hazards, driven by climate change, pose to motor insurance. However, accurately modelling the relationship between these hazards and claim frequencies is challenging, largely because the available records are incomplete. Specifically, they capture only those storm events that result in at least one claim while omitting unreported events. To address this limitation, we introduce a novel class of compound frequency models for the joint analysis of storm occurrences and the corresponding claim frequencies with accurate predictive power. These models are specifically designed to recover the joint distribution of actual storm events and underlying claim processes even when faced with incomplete data. Additionally, we incorporate geospatial covariates to evaluate their influence on both storm occurrences and claim frequencies. Given Greece’s vulnerability to extreme weather due to its geographical position, understanding the influence of climate change on insurance risks is critical. Notably, our findings reveal a negative intrinsic dependence between actual storm counts and per-storm claim frequencies, suggesting potential diversification benefits for insurers as climate change leads to more frequent weather-related hazards.