Mr. Clinton and other modeling executives said they would like rating agency analysts to have a better understanding of how models work, because they are very different. “They don't produce the same numbers, and the rating agencies don't take this into account,” he said.

Capital charges used in rating agency models depend on company-supplied estimates of probable maximum losses, such as 100-year windstorms and 250-year earthquakes. “Whatever model you use” to estimate those losses “is fine with them,” Mr. Clinton reported.

“This creates a moral problem. There is good potential for gaming these models,” he said, noting that insurers could seek to use the model producing the lowest loss estimates for the purposes of a rating, while using other models for underwriting and pricing decisions.

While new versions of catastrophe models are not scheduled to be released until late spring, rating agencies began adjusting their capital requirements for catastrophic events even before last year's storms.

Steven Dreyer, managing director of Standard & Poor's in New York, said S&P changed the way it looked at catastrophe risk in response to the storms of 2004.

“We recognized that our previous standard didn't deal with aggregate risk very well, so we moved from an event-based approach to an aggregate approach,” Mr. Dreyer explained. In June 2005, S&P replaced a capital charge for reinsurers based on a one-in-100-year event standard with a one-in-250-year aggregate standard. (After the 2005 storms, S&P said it would also apply the new standard to primary insurers.)

“Our expectation, going forward, is that the modeling agencies themselves are adjusting their factors…The net result will be a more onerous capital requirement,” Mr. Dreyer said, noting that while S&P's one-in-250-year standard won't be changed, estimates of aggregate losses occurring once every 250 years will increase as modelers release new model versions.

Donald Kramer, CEO and president of Ariel Re, believes that by applying new stress tests to PMLs that will also jump as modelers rework their tools, rating agencies are “double-counting” or “compounding” the impact of catastrophes in their formulas.

“Steve Dreyer said it himself,” Mr. Kramer noted, questioning the reactions of both the rating agencies and the modeling firms to recent events.

“Is the perception of risk so great that they think a one-in-250 is going to happen every year?” he asked. Although he happily noted an opportunity for an underwriter to make money if the perception of risk is greater than the actual incidence of loss, he suggested that both rating agencies and modelers are overreacting.

“Weather has changed–is changing,” he said. “There will be a time when the Gulfstream shuts down–when Bermuda will be freezing. The question is, 'Is this cyclical? Is Katrina a once-a-year event? Are these events we can expect in 2006?'”

“No company became insolvent as a result of the hurricanes. They might have been wounded [or] diminished. Yet companies were forced out of business by the elimination of ratings,” he added.

His company, Ariel Re, bought the infrastructure of Rosemont Re, which was forced out of business when its ratings were cut. More recently, PXRE and Quanta, facing the ratings knife, said their businesses could be sold. Still, Mr. Dreyer noted there weren't a lot of downgrades from the 2005 storms, attributing the result to the revised stress test which put actual losses within S&P's expectations.

Mr. Clinton, who refused to be coaxed to describe rating agency actions as “double-counting,” did speculate that changes in modeling assumptions and rating agency formulas could, in some cases, produce 50-to-100 percent hikes in capital requirements. “That's going to have a dramatic impact on the industry–on the ability to provide capacity to the market,” he said.

Karen Clark, president and CEO of AIR Worldwide in Boston, noted that “gaming” of models extends to pricing. “There is this quaint notion that reinsurance pricing is based on the [cat] models, but the market sets the prices, not the models,” she said.

She noted, for example, that while many industry experts agree AIR's model has “the most advanced science” for estimating European windstorm losses, “many prefer other models because the indicated pricing is cheaper.”

Responding to Mr. Kramer's comments about the gap between perception and actual risk, she said, “I don't think [the perception] is high enough. I don't think people have internalized the numbers that we [the modelers] have been talking about.”

Over time, she said, “the numbers always go up,” noting that in 10 years, loss dollars will be twice as high as they are now just because of increasing property values. “One-hundred-billion-dollar events are going to cost $200 billion,” she added.

Referring to Mr. Kramer's questions about whether Katrina-like events are now expected to occur annually, she referred to the science of cool and warm cycles underlying hurricane activity. “We can be in a period now [where there is] a clustering of events. You have to be prepared for that.”

She added that “we ought to take these 250- and 100-year return periods out of the language,” noting that such terms give a false sense that an event will only happen once every 250 years. Instead, it means there's a 0.4 percent probability that it will happen this year, while a 100-year loss has a 1 percent probability, she noted. “Over 10 years, there's a 10 percent probability that a 100-year loss will take place.”

Using “return period” terms, RenaissanceRe CEO Neill Currie said he empathized with rating agencies and modelers. “When a rating agency receives something [from an insurer] that says this is my 1-in-250, and the actual loss is four times that [amount], how in the world can a rating agency give someone with that kind of data credibility?” he asked.

AIR, in an analysis published in November 2005, noted that part of the reason for such discrepancies is the fact that inaccurate data was being put into catastrophe models. AIR's analysis found, for example, that for 90 percent of commercial properties, insurers recorded replacement values much lower than values that would be set by an engineering analysis, with some values underestimated by as much as 70 percent.

“You do not want to be at the mercy of this bad data,” Ms. Clark said, advising that until data quality is improved, insurers need to benchmark their company loss estimates against industry loss estimates.

While modelers admit to some technical deficiencies in their models, Mr. Currie, Mr. Clark and others said the failure of insurers and reinsurers to use models appropriately to make intelligent underwriting decisions contributed to the surprises last year.

“If you are writing risk-excess [reinsurance] and you don't have an occurrence limit for catastrophe, I don't think that's AIR's fault,” Mr. Currie said. “Underwriting without tools doesn't work, and underwriters without experience relying just on tools doesn't work, either.”

“There is a lot of na?ve use of the models out there,” agreed Eqecat's Mr. Clinton. “I don't know how many times I've heard, 'Your model is wrong because it's not matching the market price.' That's not the intention of models,” he said, explaining that the intent is to provide the best science and engineering. “You have to determine how you're going to price around that along with uncertainty.”

Hemant Shah, president and CEO of Newark, Calif.-based RMS, said modeling should be viewed as nothing more than a systematic way of organizing knowledge. “That actually makes the challenge of underwriting that much greater,” he said. “You not only have to underwrite the account. You need to underwrite the effectiveness of that model in interpreting the risks of that account.”

Ms. Clark discussed the need for insurers to use model results based on long-term historical hurricane experience and short-term views for different types of decisions. In particular, she argued it would be unlikely primary insurers could get regulatory approval for price hikes based on short-term analyses, which will be provided for the first time in this year's updates. But they could use the short-term view to decide how much reinsurance to buy, she added.