Seismic loss modeling of two building portfolios using generalized Pareto model and copula

The catastrophic upper tail of probability distribution of seismic loss for multiple buildings and infrastructure is a serious concern for insurers and reinsurers, which is caused by physical features of spatiotemporally correlated ground motions. It is a routine practice to ignore some details of earthquake risk exposures to different insured properties when portfolio aggregation is conducted. However, this simplification might result in biased risk assessment of the combined portfolio. Therefore, insurers and reinsurers desire to have an effective tool to probabilistically combine two insurance portfolios into one. To deal with these problems, extreme value theory for peak-over-threshold data (i.e., generalized Pareto model) together with copula modeling is employed. The proposed method is applied to analyze seismic loss data for existing wood-frame houses in south-western British Columbia, generated from an earthquake-engineering-based (EQEng) seismic loss model. The proposed approach demonstrates good modeling capability with respect to the EQEng model, and can be useful for more complex risk management problems.