Pioneer In Field Says Insurers Are Too Reliant On Catastrophe Models

The problem was exacerbated, she suggested, by rating agency demands for modeling numbers in trying to assess insurer exposure to one-in-100-year storms and one-in-50-year storms.

But Ms. Clark–who founded the first catastrophe modeling company, Applied Insurance Research, which later became AIR Worldwide Corp. after its acquisition by the Insurance Services Office in 2002–said models are only generalized best estimates, and are not designed to replace underwriters, nor to be the final word on which risks are acceptable.

She noted that models can change, and if an insurer's business is based solely off modeling results, that insurer's business model will have to keep changing as well.

"We've become a modeling society," Ms. Clark said, contending that model results should not replace common sense.

She pointed out that some insurance company executives have admitted to her that model results have looked wrong, but the company cancelled risks based on the numbers anyway.

The key to using such technology effectively is to understand models and use them as a tool, rather than an absolute truth, said Ms. Clark, now a risk management consultant on catastrophe exposures.

She also pointed out that models can contain uncertainties, limitations and inaccuracies, adding that unknown factors can greatly alter a model's results.

As an example, Ms. Clark said the current subprime/credit crisis happened because of an overreliance on models.

Many smart people, she said, developed models showing how money can be made off of mediocre loans, and it worked well until the market experienced a factor the models did not account for–the housing market slowdown. Once this occurred, she said, the models and the business decisions based on them fell "like a house of cards."

Ultimately, Ms. Clark said, there is much about atmospheric conditions that scientists and modelers do not understand, and that has an influence on modeling results. Insurers, she noted, have to figure out how to make decisions despite uncertainties.

In response, industry representatives said insurers are balancing the use of cat models with underwriting, and that carriers actually have a much better understanding of how to use models than they did in the days before Katrina. They did, however, acknowledge a shift away from relying heavily on historical data to write catastrophe risks since Katrina.

"I think there is a clear consensus that the old ways of evaluating catastrophe risks based on historical claims simply are inadequate," said Debra Ballen, executive vice president of public policy management for the American Insurance Association.

Bob Detlefsen, vice president for public policy at the National Association of Mutual Insurance Companies, said that after Katrina, insurers may have drawn the conclusion that "relying on historical loss data is something that has to be supplemented by a technique like modeling that takes into account future trends."

But he added he does not see an industrywide overuse of cat models.

He said that while some insurers may possibly be relying too heavily on them, "from what I hear, it's a matter of how they weigh the results that the models produce relative to those more traditional underwriting factors. It's not that they've abandoned them in favor of models, but for some insurers the models are becoming more prominent in their calculations than was the case before Katrina."

He noted that models help insurers account for prospective factors other than weather, including increased concentration of population along the coast.

Mr. Detlefsen said models can tell an insurer how many people will be living in an area five- or 10 years from now, what structures may look like, and other key factors that can't be captured relying on historical loss data.

Mr. Detlefsen and Ms. Ballen also agreed there is pressure from rating agencies and reinsurers to incorporate cat modeling, while Mr. Detlefsen added that some insurers may also be in a "risk-averse mood, given the losses they experienced as a result of the 2005 hurricanes, and that may be affecting their willingness to rely more heavily on models, or their desire to rely more heavily on models."

Ms. Ballen pointed out that insurers also have a better understanding of how to use models than they did in the past, and decisions are not made solely because a model reported certain results.

"My understanding is that the actuarial standards of practice don't allow a company to…pick up a model and pluck it into its ratemaking without [understanding] what the model is and also understanding what the appropriate risk is. So the comments that [Ms. Clark] made, I think to some extent are being addressed at the individual company level through the actuarial standards of practice" companies are using.

Matthew Grant, chief markets officer at Risk Management Solutions, a modeling company, said that before 2005, communication between people using the models and senior management at insurance companies was worse than it is today. He said senior management didn't challenge those using the models to see where there may be gaps in the results.

Today, he said, "the quality of people using models continues to get better and better." The users are sophisticated now, and more educated in how to use and apply model data. Senior management, too, is now asking the right questions, he added.

Regarding whether the industry is too reliant on cat model results, Mr. Grant said companies are "making commercial decisions based on a number of different reasons," and a model "may be one of those, but it would seem strange that somebody would choose, especially in a softening market, not to write a risk only based on the model output."

He added that there may be situations where the model tells them something about the risk that could influence their decision.

Ms. Ballen and Mr. Grant agreed with Ms. Clark's comments about the importance of properly understanding models.

Ms. Ballen said there needs to be consistency with respect to exactly how models are used, "so that there isn't disconnect between, for example, a reinsurer's evaluation of what the model means and the insurer's evaluation, because that isn't going to work."

Mr. Grant said models are sophisticated, adding it is important for users to understand what level of uncertainty there is in the results. The output, he said, is "not simply black and white," and different levels of uncertainty exist depending on varying factors.