Katrina exposes fallibility of catastrophe models.

Much like the tour major hurricanes that hit Florida in 2004, Hurricane Katrina has exposed shortfalls in the use of catastrophe models, leading some to question their use, others to defend them and many to say there are lessons to be learned.

"Everyone's a little bit disappointed with the fact that the modeled losses were significantly smaller than the actual losses, and as a result of this, there are people saying the models aren't reliable and we shouldn't use them," said Matthias Weber, a member of Swiss Reinsurance Co.'s extended management board, Americas division, during an interview with Best's Review at the Property Casualty Insurers Association of America's annual conference in Chicago. "Personally, I think there's no way for industry to step down and say we can't use the models at all. But what industry needs to do is demand improvement of computer models, and I think it's going to happen."

Paul VanderMarck, executive vice president of cat modeler Risk Management Solutions Inc., said while there would be further refining of risk models, insurers themselves also need to make improvements. A report by RMS on Katrina found insurers' exposure data to be "outdated, incomplete, poorly resolved and miscoded," leading to the underestimation of risk. While many insurers have improved the quality of their data, "significant improvement" still is needed, VanderMarck said. Without "high-resolution, timely, accurate" exposure data, the old computer axiom "garbage in, garbage out" applies to the use of models, he said.

Rolf Tolle, franchise performance director at Lloyd's, also said from a reinsurance gathering in Baden-Baden, Germany, that the answer may lie in refining the information that is put into the models. He suggested, for instance, systems may not have accounted fully for the big movement of population in the United States to the coastlines in recent years or the increased value of home contents.

"So there are a lot of parameters, which I think need to be revisited in order to calibrate them correctly," Tolle said.

Karen Clark, chief executive officer of cat modeler AIR Worldwide, said AIR is working with insurers to improve their data on commercial-lines exposure.

"The residential data is higher quality than the commercial data," Clark said. "It's more complete and accurate."

The problem, she said, is incomplete data from the insurers--especially when it comes to replacement values. To help companies validate their exposure data, AIR is using an industrywide database it's assembled over the past 20 years that contains detailed information on more than 50 million residential and 6 million commercial properties.

"I think the issue is as much how we use the model as the models themselves," said Andrew Carrier, an active underwriter for the Kiln syndicates at Lloyd's, during a reinsurance forum held Oct. 23 at the PCI conference. Modeling "is too easy to become the comfort blanket of underwriters, and they become less comfortable to use their own judgment."

VanderMarck agreed. "Risk management needs more than just model use," he said. Models remain a vital part of risk management but must be "supplemented by thoughtful analysis of risk," he said.

Tom Larsen, president of cat modeler Eqecat Inc., said the 2005 hurricanes have shown gaps in modeling for correlated perils, flood and offshore exposures.

"What really changed in Katrina was the flooding of downtown New Orleans, which triggered a lot of commercial flood," Larsen said. "We didn't include the flood aspect prior. We really struggled to come up with a flood exposure. It's really uncertain. We have much better wind exposure information. We would still struggle if there's another event. We have to try to come up with better estimates."

Dick Luedke, a spokesman for State Farm Group, said State Farm doesn't "expect hurricane models to tell us what is going to happen over the short term. But we do expect them to be very useful over the long term."

1-in-100-Year Event: What It Means

* A one-in-100-year event means there's a 1% annual probability. A 1-in-500-year event means there's a 0.2% annual chance. Allstate Corp. recently said it didn't buy reinsurance in Louisiana because models showed a "1-in-500-year" risk of a Katrina striking Louisiana.

* While a 1-in-1,000-year event has a minus-cule probability of happening in any given year, over a 10-year period, the chance of its occurring rises to 1%.

About 150 insurance professionals
were surveyed during the A.M. Best
Co. Insurance and Technology
Conference.

What did Hurricane Katrina show
about catastrophe modeling?

It is modestly useful      47.7%

It is highly useful        23.1%

It isn't ready for
prime time                 29.2%

Rick Cornejo, Meg Green and Robert O'Connor contributed to this article