Epistemic uncertainty in catastrophe models—A base level examination

Abstract

AbstractIn this paper, we evaluate sources of variation in the output from catastrophe models with emphasis on the epistemic uncertainty in modeled expected losses. Using building data from the 34 buildings that comprised the California Northridge campus at the time of the Northridge earthquake, we explore the sensitivity of estimated average annual losses obtained from a cat model to the quality of model input. Namely, we consider how changes in four key model assumptions—building locations, building height, construction type, and the event catalog—affect cat model loss estimates. We find that accurate information on some input variables is critical (e.g., all steel construction) and the interaction between input variables should not be discounted. Our results have important implications for insurer decisions that are informed by the output of catastrophe models—product pricing, portfolio diversification and underwriting decisions, negotiations and discussions with regulators and similar activities with capital market participants. The financial impact of improving data quality and targeting data related to key model inputs for that insurer when at scale is not trivial. As such, this paper provides an impetus for establishing and improving benchmarks for model inputs.