654 – Risk Model Selection and Extreme Values
The Generalized Pareto Distribution and Threshold Analysis of Normalized Hurricane Damage in the United States Gulf Coast
Sponsor: Section on Risk Analysis
Keywords: Extreme Value Theory, Generalized Pareto Distribution, Mean Excess Plot, Excess over Threshhold, Normalized hurricane data
Kumer Pial Das
Lamar University
Gui Li Liu
Lamar University
This study concerns the probability distribution of the most damaging hurricanes to strike the United States. Economic damage is normalized to adjust for temporal shifts in societal vulnerability. Consistent with the extreme value theory, a generalized Pareto (GP) distribution is fitted to the excess in damage over a high threshold. The focus of the statistical analysis is primarily on diagnostics to determine an appropriate threshold. Conclusive evidence is provided that such data have a heavy tail (i.e., a GP distribution with a positive shape parameter).
