Computer Models Reassess Fla. Hurricane Risk Following Busy 2004 Season

New computer models show higher chances of hurricane strikes to Florida, and a 3.9 percent increase in predicted losses. In parts of the state, particularly Escambia County, the program boosts predicted losses as much as 25 percent.

If approved by the regulatory commission that reviews the models at its next meeting in June, those models will be the engines that generate higher insurance premiums in the next billing cycle.

“That has more impact than anything we do,” Rep. Dennis Ross, chairman of the House Insurance Committee told the Tallahassee News-Press. The committee has rolled out the first big package of hurricane insurance legislation.

The post-2004 climb in hurricane losses contradicts statisticians, who say the state’s four-storm year should not change the risk insurers see in Florida. Multiple strikes have always been part of the equation.

On average, Florida experiences 1.7 hurricanes a year. Four “was unusual but not unpredicted,” Sneh Gulati, a statistics expert from Florida International University told the News-Press.

Although “black box” models are submitted for approval to a state commission on which Gulati is a member, its members can’t see the assumptions and formulas used. The commission must hire private reviewers who sign confidentiality pledges. They decide if the models meet 35 state standards but cannot disclose other details.

Actuaries and statisticians on the commission are comfortable with the process.

The public often is not. “That right there tells you, it’s smoke and mirrors,” Jim Bledsoe, a Pensacola businessman told the News-Press. Bledsoe’s 32534 zip code, according to the computers, now has a 22 percent increase in future hurricane losses. Across the street, predicted losses for zip code 32514 rise by 19 percent.

Those numbers and the complicated formulae used to derive them don’t convince Bledsoe that insurance rate hikes are justified.

“The bottom line is, did you make money? Yes? How much?” he said. “How much more money do you want to make?”

Florida insurance regulators look critically at hurricane loss projection models used to justify rates. Insurance Commissioner Kevin McCarty has led a national push to crack open the so-called “black box” models, as well as effort at home to build a public version.

“There is much regulators still do not know about how insurers use the secret models,” OIR actuary Howard Eaglefeld said. For instance, insurers need not disclose if they have run several models and are using the version that produces the greatest losses in order to justify the highest possible rates.

That is an issue because hurricane loss projection models disagree greatly.

The 2005 versions given to the commission this month predict Florida’s average storm losses as low as $1.8 billion and as high as $3.6 billion.

When state-run Citizens Property Insurance ran several models to estimate damage from last year’s actual storms, the amounts were off by millions of dollars.

One model predicted Hurricane Charley losses in Lee County for Citizens of $799 million. Six months later, the company had paid out only $154 million in claims.

The models are best at predicting long-run losses, across large geographic areas.

“You can’t look at a single storm and say, ‘Here’s what the answer is,’ ” said Marty Simons, former chief actuary for South Carolina and one of the professionals hired by the commission to review models.

AIR Worldwide, whose model is used by half the residential property insurance market in Florida and 85 percent of the companies that underwrite insurers, calculates projections over storms across a 50,000-year span.

The models contain so many variables, it takes that many “trials” to approach a predictable average. Flipping a coin, for instance, has only two variables. It takes 62,500 “trials” — coin flips — to get a result within a 2 percent margin of error.

The more variables, the more trials required to get a dependable outcome.

Applied Research Associates, another modeling expert, this year lengthened its computer modeling from 100,000 to 300,000 years to get results within an acceptable margin of error. Conversely, they are very sensitive to changes at small geographic levels.

Miriam Perkins, an AIR senior risk consultant, said adding 2003 and 2004 hurricanes to the company’s storm database causes some of its increased hurricane probabilities in regions such as Northwest Florida.

She said strike probabilities also rise because the company has “re-sampled” the central air pressure of past storms.