Catastrophe modeling experts at the World Insurance Forum in Bermuda repeatedly made the point that results from different model vendors are "very different."
How different is "very different?"
When modelers aired their differences, arguing over one particular question, Brian O'Hara, CEO of XL Capital, likened the exchange to a "modeling war breaking out."
The question, posed to RMS President and CEO Hemant Shah, was whether Katrina was a "tail event"–one with a low probability of occurrence, plotted at the far right or tail-end of curves, known as exceedance probability curves.
Mr. Shah said that Katrina's roughly $60 billion price tag made it a tail event in terms of loss size, adding that outside influences, such as the levee failure, contributed to the high-dollar amount. Meteorologically, he said, it's not a tail event, noting that as a hurricane, a Category 4 hitting the Gulf isn't particularly unusual.
Karen Clark, president and CEO of AIR Worldwide, seemed to disagree with both assessments, emphatically stating that Katrina "was not a tail event–and we haven't really seen one yet." AIR's largest U.S. wind event is over $200 billion before demand surge, she said, adding that Katrina-sized losses have about a 3 percent annual probability or a 30-year return period.
She also said Katrina was an extreme meteorological event because the size of the radius of the storm–at 30 miles–was far greater than previous storms of similar intensity. Katrina is among the most intense storms to ever make landfall, she noted, listing Hurricanes Andrew and Camille among the others. But those storms were more tightly wound, she said, noting that Andrew had a radius of about 10 miles.
Ms. Clark said this finding will be incorporated in the next version of AIR's models to be released in late May.
Asked separately by NU to give some quantitative examples of model differences, she said there could be geographical areas where the results are different by only 10- or 20 percent, while in other areas, results could differ "by a factor of five or more. It really depends on the peril and the geography."
With such differences possible, EQECAT's president, Richard Clinton, warned against using a "best-of-breed-type approach," in which an insurer or reinsurer uses the results of a single vendor model rather than comparing multiple models.
"That's a very dangerous path to go down," he said. "You're banking your whole company, your rating, on the results of one model."
© Arc, All Rights Reserved. Request academic re-use from www.copyright.com. All other uses, submit a request to TMSalesOperations@arc-network.com. For more information visit Asset & Logo Licensing.