Mapping the frequent storms.

In 2003, catastrophe losses totaled about $12 billion in the United States, of which only about $2.5 billion were due to hurricanes and/or earthquakes. The remaining losses were primarily caused by tornado/hail, winter storm and wildfire, said Richard Clinton, president of Eqecat Inc. In fact, 67% of all catastrophe losses and seven out of 10 of the largest catastrophe losses in the United States since 1997 have been due to events other than hurricanes and earthquakes.

Historically, that's not an unusual pattern, he noted. "Companies pay a lot of attention to managing their earthquake and hurricane loss exposures, because they are the perils that have the potential of causing the blockbuster or devastating type of losses," Clinton said. "But in reality, tornado/hail, winter storm and to a lesser degree, wildfire, also need to be aggressively managed because they have impacted the companies' operating results as much, if not more, than the earthquake and hurricane losses."

Recognizing this, catastrophe modelers that long have marketed earthquake and hurricane applications, now have expanded their modeling tools to include these more frequently occurring events.

For example, AIR Worldwide Corp., a pioneer in developing catastrophe models in the late 1980s, has released a physical-based model capable of simulating European windstorms in three dimensional space using the actual physics of the atmosphere, said Karen Clark, president and chief executive officer of the Boston-based modeling firm.

AIR soon will release a U.S. winter storm model, the first in this category, using the same technology, she said. The model will map nor'easters, Pacific storms, Santa Ana winds and other types of windstorms that until now have challenged modelers because of their complex structures and their changeability in three dimensional space, Clark said.

"It's very hard to model them in the way that a hurricane is modeled, for example," she said. Using a dozen of so variables, researchers can capture a hurricane's structure, which tends to be symmetrical with winds flowing counterclockwise around the storm and the highest winds always some distance from its center.

But a winter storm, sometimes also called an ex-tropical cyclone, tends to have multiple centers of low pressure and multiple points of peak winds. Also, its winds do not blow counterclockwise everywhere in the storm, but demonstrate a very complex wind pattern, Clark said. "So it is extremely difficult if not impossible to use a traditional catastrophe modeling approach for these types of storms," she said.

After years of investing heavily in its computer infrastructure, its meteorological staff and its data collection capability, AIR now has the capability to collect and archive more than eight gigabytes of global atmospheric data every day. "We need that level of data to run these physical models" Clark said.

AIR said its winter storm model also will be the first catastrophe model to explicitly model precipitation as a separate component. In winter storms, snow and ice can cause considerable damage, especially in the collapse of roofs. While catastrophe models currently account for damage from precipitation--the rain that falls during a hurricane, for example--that is included in the damage calculation, but is not explicitly modeled, Clark said.

Eventually, the physical modeling approach will extend to the company's other models, including hurricane, she said.

AIR also is developing a model for industrial accidents, which traditionally have not been viewed as catastrophe losses typically involving multiple policies. "But these industrial accidents--explosions at refineries or other types of major accidents--can still cause enormous losses," Clark said. "There's a lot of interest on the parts of insurers in taking some of the catastrophe modeling techniques and applying them to industrial accidents."

Snow and Hail

In response to clients' requests, Eqecat is developing an advanced product for modeling tornado/hail. "This is a model that we feel really raises the bar within the industry for tornado/hail modeling," Clinton said. "We are building a new model from the ground up incorporating the latest in science and engineering technology and validating it with claims data from several large companies."

Eqecat thinks its tornado/hail model will be the best one out there because of the robust modeling approach it has taken. The Eqecat tornado/hall stochastic event set relies on more than 750,000 events.

"The characteristics of tornado/hail events require that a large number of simulations be used to properly quantify the risk," Clinton said. "In addition, we have taken both spatial and temporal clustering into account in defining events. In this way, we capture the very significant catastrophe losses that can occur when a weather system spawns dozens of tornados and hundreds of hail streaks over a two- or three-day period."

The objective was to build a model that is refined enough for use in underwriting and rate making as well as accumulation management, he said.

Eqecat also is developing a model to analyze the kinds of winter storms that have been causing snow- and ice-related losses for years in the upper Midwest and Northeast. The tornado/hail and winter storm models are expected to be available in April for U.S. clients, Clinton said.

Earthquakes and Hurricanes

In February 2003, Risk Management Solutions, Newark, Calif., released its next generation models for western U.S. earthquake, as well as Gulf and Atlantic Coast hurricane risk.

"A major focus in those models was not only significant advances in science but in the resolution of those models in their ability to distinguish risk on a very site-specific or high-resolution basis," said Paul VanderMarck, executive vice president of products at RMS. "That fit in directly with the agenda of many of our clients who want to deploy the models and use them for underwriting individual risks."

Typically, RMS follows five-to six-year upgrade cycles for its models, said Kyle Beatty, a manager of technical marketing at RMS. This can vary, however, ff a major catastrophe occurs and essentially accelerates the learning curve, he said.

"If something like a Hurricane Andrew were to happen next year, there would be a lot of new scientific and insurance data to look at, and we might learn something quicker than is normal," Beatty said. Usually, however, advancements take several years to be fully understood in the scientific community to the point where they can be incorporated with confidence into modeling technology, he said.

But one such significant scientific advancement--the understanding of a hurricane's transition as it moves northward--did unfold after RMS' release of its second generation hurricane model in the mid-1990s, which was incorporated into the 2003 release, Beatty said.

As this transition occurs, hurricanes become less symmetrical and their advance cloud shields expand significantly in area because they are moving into an environment very different from that of the low latitudes.

"They begin to interact with the jet stream, for example, and other types of low, near-surface air that can be very cold or relatively cold compared to what they'd seen in the tropics," Beatty said. "That's an environment more similar to what you would see from a low pressure system that might affect someone on the Great Lakes and move off the Eastern seaboard."

The asymmetry of these transitioning storms can translate to rainfall focused on the left side of the hurricane track, with the most intense winds on the right side of that track. The winds also tend to expand, meaning a broader area can be affected by winds at peak speeds somewhat below those in a pure hurricane with the same central pressure, Beatty said.

Over the past two years, RMS has put a great deal of time into applying the same modeling concepts and approaches used for its natural catastrophe events to terrorism events, VanderMarck said.

"We've done very advanced modeling--for example, computational fluid dynamics models of the pressure wave propagation from bomb blasts in urban environments--to model the expected damage from various types of attacks you would expect from terrorists using both conventional and unconventional methods," he said. Clients now use those models to manage their exposure and understand the correlation between different events, VanderMarck said.

Paul Budde, senior rice president of reinsurance intermediary Benfield, gave all three major vendors high marks for developing basic terrorism models that have a good handle on the damage that, say, detonation of a 2-ton truck bomb would have on a structure.

"But the huge thing they have to work on, and I'm not sure if it's surmountable, is that so much of the terrorism risk depends on the geopolitical situation--what our government is doing and what other governments are doing," he said. "Quantifying the probability associated with each terrorist attack is the most difficult part of the calculation." While modelers have a sense of long-term average frequency and severity of earthquakes and hurricanes, they're nowhere near that level with terrorist acts, Budde said.