Catastrophe model (cat model)

August 24, 1992: Hurricane Andrew makes landfall in Florida, and its insured losses exceed what rates built on loss history had implied; the market then turns to catastrophe models. Such a model estimates losses from rare, severe events where the loss history is too thin to price on. It simulates instead of counting, in four successive modules. The hazard module generates a catalog of thousands of simulated events (earthquakes, hurricanes, floods) with their frequency and spatial intensity. The exposure module locates insured properties and their characteristics. The vulnerability module turns intensity into a damage ratio through fragility functions. The financial module applies deductibles, limits and co-insurance to get the insured loss. The output is a loss exceedance curve (OEP/AEP), then the average annual loss (AAL), the 100- or 200-year loss, or the PML. The two dominant vendors are AIR Worldwide (Verisk) and RMS (Moody's); PERILS sells no model and publishes industry loss indices instead. These models are used to price upper layers of catastrophe reinsurance, track accumulation, structure cat bonds and calculate the Solvency II catastrophe risk module. They are only as good as their assumptions: two competing models can diverge materially, and calibration on stationary data ages badly as the climate shifts. In cyber, the discipline is in its infancy, since the hazard is dynamic and vulnerability depends on human and technical behavior that keeps changing.