Posted on 19 May 2010
Hurricane Katrina did more than cause billions of dollars in insurance claims. The most costly hurricane on record triggered a lot of analysis of catastrophe modeling.
Losses, in many cases, exceeded modeled estimates, leading to an inevitable backlash against the cat modelers. However, the modelers not only weathered the storm of criticism, they seized the opportunity to improve the models, which are now more frequently used than before.
"There was quite a bit of model bashing in the wake of Katrina," said Karen Clark, president and chief executive officer, Karen Clark & Co. "Many companies were very surprised by the size of their Katrina loss and they felt the models had kind of let them down, because the models had not indicated that a storm like Katrina was possible."
Clark, a cat model pioneer who founded modeler AIR Worldwide, said the criticism was "partly justified. The model loss estimates in the Gulf region were too low, most notably for certain types of construction."
The problem, the cat modelers said, was data.
"If you feed a model data that is inaccurate, you will not get an accurate result," said Peter Dailey, director of atmospheric science at AIR. "The quality of data is important."
While Katrina spurred insurers to ensure the valuations being used in the models were correct, modelers would invest heavily into improving their products.
But Clark cautioned, "While the models have certainly improved since Katrina, they shouldn't be thought of as being fixed because the magnitude of the uncertainty around the loss estimates is pretty much just as wide now as before the storm."
Tom Larsen, senior vice president with modeler Eqecat, agreed the models are imperfect. "They are models. They do not create reality," he said. "They don't give you an answer, they give you a range of answers."
Since Katrina, cat models have been integrated more into the operations of companies, Larsen said. "People are more comfortable with them and are making bigger decisions with them."
State Farm, the top U.S. writer of homeowners multiperil, has found "the models really do well for our book of business," said Jeff McCarty, assistant vice president.
"We have always been a multiple model user, even before Katrina," McCarty said. "We want as much information as possible. We don't want any one model influencing what our estimates are."
Each model they use provides a reasonable estimate of exposure and State Farm blends the results. There is also "more post-model analysis," where the company uses internal expertise to weigh the results. "You're trying to pull together all the information and the models are just one piece," he said.
Modelers are always striving to improve, and insurers and cat modelers are working more closely now, Larsen said. "They are actively seeking a better understanding of the workings of the model, so they can use it better," he said. "They can be more effective if they understand what goes into it."
Katrina's scope allowed the modelers to examine the wide range of damage a hurricane can produce, Dailey said. "It's really from actual storms that we get the best data," he said.
Modelers deploy teams of engineers to the field to evaluate damage after a hurricane, Dailey said. First-hand data "really allows us to improve" the model. "In the case of Katrina, we deployed multiple teams multiple times," he said. Dailey said collecting actual wind speeds helped validate the hazard component of the models.
Specialized observation data sets "will be the wave of the future," Dailey said, noting how a hurricane evolves as it moves from sea to land is very useful in modeling. The complexity of hurricane wind field structure and evolution can lead to very different losses often in the same location, he said.
Modelers "improved vulnerability functions in" building code information, flood risks, what policy forms cover, offshore energy and business interruption "to be more correct," Larsen said. For instance, they look to incorporate not just what a house is made of, but when and how it was built, Dailey noted.
"There's more of a focus to get all of the risk in, not just some of it," Larsen said. "There is more comfort that you can build a credible model."
Dailey said modelers are not only evaluating the level of damage but how the insurance terms -- such as limits and deductibles -- come into play. There is an "end-to-end evaluation of just trying to improve the quality of the data," he said.
Among the data sets unaccounted for in many models prior to Katrina were all the unseen factors that could drive up the cost of a claim. Claims can be inflated not only by rising building material costs, but when people have put off repairs. Other factors that lead to claims inflation include instances where an insurer overwhelmed with claims settles some at a higher cost and where political or other pressures cause an insurer to cover losses normally excluded, said Robert Muir-Wood, chief research officer for modeler Risk Management Solutions.
"We devised a new methodology for thinking about these loss amplification factors," Muir-Wood said. Now modelers have a "much more sophisticated understanding of loss amplification."
Dailey and Muir-Wood noted short-term activity forecasts have grown in use, with input coming from leading climatologists. All the modelers agree the Atlantic basin remains in a state of heightened hurricane activity. However, uncertainty remains.
"Natural annual variability is very large," Dailey said. A cluster of active years can be followed by one or two inactive years, because of factors such as El Nino -- which suppresses hurricane formation, even in warm seas. At the same time, a cooler-than-average year can become a hugely devastating catastrophe year if the right storm hits the right area -- such as 1992's Hurricane Andrew.
Because of this, Daily said AIR has two model catalogues: one long-term that does not include a warm ocean and one that does. Both have value and provide more information that just one view, he said.