Can we weather proof our insurance?

The last few years were particularly volatile for the insurance industry in North America and Europe, bringing a record number of claims due to severe weather. According to a 2013 World Bank study, annual average losses from natural disasters have increased from $50 billion in the 1980s to about $200 billion nowadays. Adaptation to such changes requires early recognition of vulnerable areas and the extent of the future risk due to weather factors. Despite the well-documented impact of climate change on the insurance sector, there exists a relatively limited number of studies addressing the effect of the so-called "normal" extreme weather (i.e., higher frequency and lower individual but high cumulative impact events) on the insurance dynamics. To reduce financial repercussions of such weather events, we develop a nonlinear attribution analysis of integer-valued insurance claims and atmospheric variables. Using data-driven nonparametric procedures, we identify triggering thresholds, or tipping points, leading to an increase in the number of claims. We develop a new data-adaptive method to compare tails of observed and projected weather variables and employ its outcomes to assess future dynamics of insurance claims. We illustrate our approach by application to modeling and forecasting of flood-related home insurance claims in Norway.