Model Uncertainty and the Pricing of Hurricane Risk in Florida

This paper examines how epistemic uncertainty affects the price of homeowners insurance in Florida. We use unique data on expected loss rate projections from seven hurricane risk models approved by regulators for use in Florida property insurance rate filings to quantify model uncertainty. By combining these data with newly published information on local property insurance markets, we are able to empirically test the relationship between model uncertainty and insurance premiums across Florida ZIP codes. Our review of model documentation and test deck results shows that there is significant dispersion among model loss forecasts at the level of individual ZIP codes, and that these differences can primarily be attributed to differing assumptions and judgments applied by model developers rather than to sampling error or differential access to data. Controlling for confounding variables and latent factors that may be cross-sectionally correlated with observed variables, we find strong empirical support for the hypothesis that greater dispersion among model forecasts leads to higher homeowners insurance premiums. Our findings suggest that, had model dispersion been ten percent lower than that observed 2021, a typical Florida homeowner would have saved $50 to $90 on her annual homeowners insurance premium.