Hurricane Loss Estimation Models: Opportunities for Improving the State of the Art
Bulletin of The American Meteorological Society - (BAMS)
N umerical hurricane loss models have become widely used in the insurance industry as a tool for determining loss costs. Loss cost is defined as the annualized dollar amount of loss a given exposure will suffer over time, in other words, how much money must be set aside each year to offset losses for a given exposure. Loss costs are used as the basis for establishing the premiums to be paid by the consumer. The models presently in use in the insurance industry are proprietary, which raises
... , which raises difficult issues for state insurance regulators charged with assuring that rates are fair, reasonable, and nondiscriminatory. There is a need to establish the limitations and performance of these models in an objective manner to provide users of loss-costs data with an understanding of the technology, especially given that the inner workings of the models are not available to general users (those who have not executed confidentiality agreements). Moreover, the sheer complexity of the models makes it difficult even for a sophisticated user to accomplish a proper evaluation. This paper reports the results of a comprehensive study of loss costs conducted under the sponsorship of the North Carolina Department of Insurance (Watson and Johnson 2003, available online at www.methaz.com/ncdoi/). The objectives of the study were to create an assessment of the "state of the art" of loss modeling, create a dataset of losses for North Carolina, and to create a method for evaluating individual model results as might be received in insurance rate filings. The basic approach was to identify nine wind models, four surface friction models, and nine damage models drawn from the published literature (meteorology, engineering, and insurance) leading to 324 combinations of models. Each of these combinations was assessed against hurricane losses reported by a major insurance company. Annual loss costs were then computed using these 324 combinations of models for both North Carolina and Florida, and compared with publicly available proprietary model results in Florida. As is shown here, there is a considerable need to improve these models. Although this study did not formally establish a baseline reference model (Pielke et al. 1999) , the "simple" models (such as based on the Rankine Vortex wind model) with no adjustments for terrain performed as well as more complex combinations. Hurricane loss models, in particular, the wind models and historical hurricane parameters, must be improved before users and regulators can apply these models with confidence.