NU Online News Service, Jan. 18, 3:09 p.m.EST

Catastrophe modelers, through their near-term models, have overshot actual insured losses for 2006 through 2010 by as much as $53 billion.

Karen Clark & Company has issued its third annual report on the performance of three near-term catastrophe models, which have "significantly overestimated" losses during the five-year period.

The near-term hurricane models were intended to be reflective of loss expectations for the five-year period, 2006-2010.

However, Karen Clark, president and chief executive of Karen Clark & Co., a modeling, risk and risk management firm, said in a statement, "It has become clear, as our previous reports have found, that a short-term horizon is not sufficient for credibly estimating insured losses from hurricanes."

"It's like doing brain surgery with a chainsaw," Ms. Clark told NU Online News Service. 

Ms. Clark developed the first hurricane model more than 20 years ago and founded Applied Insurance Research, which later became AIR Worldwide.

According to the report, modelers AIR Worldwide (AIR), EQECAT and Risk Management Solutions (RMS) each dramatically overestimated insured losses above the long-term average, even with some minor revisions to their models.

Had they been correct, insured losses for the period 2006 through 2010 would have been between $60.4 billion (predicted by AIR) and $68.2 billion (EQECAT's estimate). RMS's projection for the period was $67.2 billion.

Actual insured losses for 2006 through 2010 was $15.2 billion.

The models aren't the problem; it is how they are used to establish probable maximum losses (PMLs) for insurers, Ms. Clark said. 

"Using the information in models to pinpoint a metric (such as a 1-in-100-year PML) is not helpful," she said. "Models can get you a range—they can get you in the ballpark—but you can't narrow the range."

Each time the models change, so do the PMLs. Yet companies continue to use them to optimize portfolios and with model revisions on the horizon, PMLs for insurers may change again, Ms. Clark said.

"You can't run a business that way—'Oops, we canceled the wrong policies,'" she said.  

Ms. Clark referred to what she calls the "Hurricane Frequency Paradox"—an increase in cyclone activity does not correlate to increases in hurricane landfalls in the United States but the short term models are based on the frequency of hurricanes.

Last year illustrated the point of the "paradox" as it was one of the busiest hurricane seasons on record but no hurricanes reached U.S. land.

Landfall activity, which is what insurance companies are most interested in, Ms. Clark said, has been average or below average for seven of the 11 decades since 1900.

"There is simply no clear basis for concluding we are in a period when losses associated with hurricanes should be expected to be well above the long-term average," Ms. Clark concluded.

The modelers will say the period of five years is not enough to accurately project losses and they are right, said Ms. Clark.

"That is the whole point," she said. "[Hurricanes] are random. You can't predict when there will be an increase in losses and there is no connectivity to the frequency of storms."

Citing findings by the National Oceanic and Atmospheric Administration (NOAA), Ms. Clark's statement noted that there may not have been more tropical cyclones in the Atlantic over the last four decades. If there are, storms making landfall have decreased about 60 percent. Rather, as NOAA said, an increase in storm frequency may be attributable to better technology, leading to increased detection of storms and hurricanes. Many tropical storms and hurricanes during the 2010 season might have gone undetected in years past due to their location or duration. For instance, three storms last year were hurricanes for less than a day.

At the historical average of 1.7 hurricanes making U.S. landfall each year, it may take "quite a while," maybe a "couple of decades" to accumulate the data needed to establish a detectible trend, Ms. Clark said.

Karen Clark & Co. is focusing on primary insurers. Reinsurers can adjust because they are less regulated but primary insurers face regulation, and constant requirements for the almighty PML.

Ms. Clark said companies can continue to provide overseers with PMLs but make different business decisions using risk metrics based on fixed event sets to manage risks. The sets can be applied to entire portfolios and to individual policies, she added.