Title: Using intraclass correlation coefficients to quantify spatial variability of catastrophe model errors
Authors: Baldvin Einarsson - AIR-Worldwide (United States) [presenting]
Rafal Wojcik - AIR-Worldwide (United States)
Jayanta Guin - AIR-Worldwide (United States)
Abstract: Systematic spatial errors of natural catastrophe (CAT) models are quantified using hierarchical linear models. Insurance claims are grouped into spatial bins on a regular grid, which avoids computationally expensive distance calculations when estimating spatial covariances. For insurance claims and CAT model estimates, damage ratios are used to determine the model errors. The spatial structure of claims distributions around a model estimate is determined via intraclass correlation coefficient (ICC). A methodology is introduced to incorporate all claims, which greatly enhances the usability and robustness of the statistical models. These statistical models can have a hierarchy of spatial bins nested within larger bins, and both the number of such hierarchies, as well as the sizes of the rectangular bins at each layer, are investigated. Furthermore, several validation procedures are presented using the claims data from a major earthquake. The results are obtained with the R-package lme4.