One Step Ahead of Disaster.

As the catastrophe-modeling industry races to predict the next disaster, it is developing new insurance markets and charting new risks--from river floods to terrorism.

Hurricane Andrew--the single most costly natural disaster in U.S. history--was a watershed for property/casualty insurers, with its enormous bill of about $30 billion in damages, nearly $16 billion of which were in insured losses.

But this ferocious storm in 1992 also helped deliver a tremendous boost to the fledgling business of catastrophe modeling, which uses computer programs to estimate losses from potential disasters like hurricanes, earthquakes, tornadoes and windstorms. After Andrew, insurers realized the hard truth: had that hurricane made landfall just 20 miles farther north, in the heart of much-developed Miami, the effects on the insurance industry's financial health could have been devastating with upward of $75 billion in insured property losses.

As insurers began clamoring for new tools to assist them in managing their catastrophe risks, catastrophe models began multiplying in a market now dominated by three private firms: Risk Management Solutions Inc. of Menlo Park, Calif.; Applied Insurance Research Inc. of Boston; and Eqecat Inc. of New York. In the past few months alone, these companies have launched major new products that show how far the 13-year-old industry has come, from early outcomes that raised concerns about the accuracy and reliability of these models, to the latest breed of high-tech systems whose applications can range from charting the probable flow of flood waters to predicting how quickly toxic chemicals would spread floor by floor if terrorists attacked a federal building.

"It's an evolving situation, not only from the standpoint of how the models work, but from the science underlying a lot of the modeling," said Patrick Musick, assistant vice president, property/casualty, with the Alliance of American insurers. "For instance, seismologists continue to discover fault zones where they didn't know fault zones existed. They discover historic data that shows that earthquakes occurred in a fault area more often than originally thought. As the science changes, the modelers have to continually update their models to reflect the new science."

In the early days of the catastrophe modeling industry, the models weren't as accurate as today's breed. "It's actually quite easy to build a model," said Robert Muir-Wood, technical director of Risk Management Solutions. "The question at the end of it is, are the results meaningful and are they accurate? To check that, you need to do an enormous exercise of calibration."

Building Credibility

That's where the industry has been getting smarter, he said. "Instead of just thinking you build a model and put it out the door and then people live with it, it's fundamentally our job to prove to the people who are going to use it that it really can solve the kinds of questions they're going to throw at it and be demonstrably correct when it comes to some specific task."

In this fiercely competitive business, catastrophe modelers try to stay ahead of the pack by pouring funds into research and development of new models and products, as well as improving their analytical approach.

For example, modeling pioneer Applied Insurance Research recently changed the way it models windstorms and earthquakes. Traditionally, such models have been purely statistical, attempting to predict events based on storms in the past, said Karen M. Clark, president and chief executive officer of AIR, which began operations in 1987.

That was AIR's approach, too, until earlier this year when it announced it had a commercial application of a technique called numerical weather prediction, which scientists have been attempting to master for almost a century. AIR said its system follows a model developed by the U.S. National Center for Atmospheric Research. This approach is based on physical principles and "is much more objective, much more scientifically advanced and less purely statistical;' Clark said.

AIR's innovation, known as the Advanced Prediction System, can forecast European storm activity with reasonable certainty for about 10 days in advance, as well as generate real-time loss estimates for events as they occur, Clark said. The system uses past weather activity as the basis for analyzing current weather data to create hypothetical storms. It considers temperature, humidity, wind speed, atmospheric pressure, sea-surface temperature, sea ice cover and soil moisture. "We actually will initiate storms and let them evolve in a three-dimensional space," Clark said.

A supercomputer at AIR's center in Boston gathers the data continuously from satellite feeds and other sources, then updates the material every six hours. The amount of information is massive. "Two gigabytes of data a day come in," Clark said. When an estimate is needed, data is pulled for the region of interest and the system produces a collection of possible scenarios, then generates files that can be used to estimate losses. The data also is processed to generate forecasts, and a meteorological analysis is undertaken to predict the weather as much as a year in advance.

The system was tested in early December 1999, when a storm clobbered Denmark, causing an estimated 7 billion Danish kroner (about $850 million) in insured damage. At that time, clients were able to call up real-time loss estimates from AIR's Web site. The system underwent another test soon after, when windstorms Lothar and Martin battered France after Christmas.

Clark said AIR had been investing in the project for years and estimated that it was at least two years ahead of its competitors on this front.

Validate and Calibrate 

AIR's new technique for modeling earthquakes--the advanced component method--is on the market in the United States and soon will be offered in Japan. With earthquakes, modelers usually consult structural engineering experts who estimate the extent of potential damage to each building component, like wood beams or masonry, based on the magnitude of a quake.

"You start with these curves based on the judgment of experts--you validate and calibrate," Clark said. But once again, the company has gone "to a more objective and physical approach," commissioning design firms throughout the country to submit detailed design information on their projects, she said. The information on each building type then goes into standard software that will tell which components will fail under which conditions. This method was included in a new product, CLASIC/2, a desktop underwriting system for primary insurers, which the company released in January, Clark said. The product enables a large company to do a complete portfolio evaluation for all perils.

AIR claims to have more reinsurers on its subscriber, list than any other modeling company. "Now we want to utilize this new technology for primary insurers and corporations around the world," Clark said, adding that the company's marketing efforts will focus on adding clients in Europe, the United States and Japan.

Models on the Web

Another industry breakthrough occurred May 18, two weeks before the start of the annual hurricane season, as Risk Management Solutions launched an Internet-based service that allows risk professionals to track probable loss estimates as early as three to four days before a hurricane makes landfall. Subscribers to the Web site, RiskOnline, get immediate access to the latest extensive loss projections for the industry or their own portfolio, including breakdowns by state, region, line of business or business unit. The model's approach is based on a blend of meteorological forecasts and statistics on the behavior of past storms. Based on current weather conditions, the model simulates possible storm paths and intensities, providing updates as rapidly as information is processed from the National Hurricane Center.

The product was in development for more than a year, said Olivier Collignon, the director in charge of RiskOnline.

"We started last year with a small prototype that we used during hurricane season," Collignon said. "Then as Hurricane Floyd approached land, we ran simulations with the product and provided twice daily our estimates of the range of expected insurance loss." The results, he added, were fairly close to the actual insurance losses from Hurricane Floyd.

The product is designed for use by risk, reinsurance and catastrophe managers who need to monitor risk levels for their portfolios to decide if they need to transfer that risk as a hurricane bears down on the U.S. coast. It is also meant for claims managers, who can use it to access a regional distribution of losses across several counties, showing them how many claims people need to be dispatched to a given area after a hurricane strikes.

"Doing live hurricane modeling is made possible by the fact that we invested in developing these techniques, which come out of the fluid mechanical treatment of turbulence," Risk Management Solutions' MuirWood said.

Making Markets

This quarter, RMS expects to unveil a series of new catastrophe models, starting with high-resolution earthquake models for the eastern Mediterranean region, including Greece, Turkey, Cyprus and Israel. In the cases of Greece and Turkey--on the heels of devastating earthquakes in 1999--there's a major expansion under way in the earthquake insurance markets, Muir-Wood said.

"We are fundamentally of the belief that models can help make markets, that if you have the ability to quantify map--how to price risk but also how to transfer risk into the reinsurance markets--then you can actually help markets get going," he said. "So we are working with the local insurers and reinsurers to help develop these models to facilitate the growth of the market."

Additionally, RMS is completing a high-resolution earthquake model for Taiwan, again to support expansion of that local insurance market following a 1999 quake there.

River Floods

The company also is entering uncharted waters by releasing what it calls the first river flood-catastrophe analysis model. "We've developed a whole series of new techniques for looking at flood loss, not simply in terms of the risk at individual locations, but what is the probability that you can lose a certain amount of money as the result of a flood affecting several rivers at the same time" Muir-Wood said.

The initial river flood model is geared to the United Kingdom, where flood is a universal coverage for all property insurance. But Muir-Wood said the company was interested in creating similar models for other countries in Europe, as well as for Japan and the United States. In the case of the United States, RMS thinks flood insurance increasingly will be handled by private insurers and reinsurers, rather than by the Federal Emergency Management Agency.

Recently, Eqecat, a wholly owned subsidiary of EQE International, took yet another tack, issuing the Eqecat Underwriting product, a Web-based, Internet and intranet underwriting system that allows companies to easily disseminate catastrophe modeling throughout their organizations.

"Every underwriter can have this product on their desk," said Rick Clinton, vice president of product management. "They can pull it up, look at it and assess the individual risk as they're writing it. So, before it's even on the books, they can tell how much exposure they are adding to their overall catastrophe [risk] from a corporate standpoint."

The product is geared to small and medium-sized companies that may not be able to invest in a huge system or may not want the overhead of managing it. Instead, they can access the model through Eqecat's Internet site.

"We manage the server, we manage the process, we can provide them with the analysis that goes along with that if they want it, and they have every one of their underwriters accessing it just by clicking on the Internet," Clinton said. "So it's a way of getting this down to all levels of the insurance industry."

The company rolled out the Internet version of the product during the first quarter and the intranet version, where the underlying technology is housed on a server in a client's offices, in May.

Eqecat is active in modeling earthquakes and all types of wind hazard worldwide, said James J. Johnson, chairman and chief executive officer of Eqecat and chief operating officer of of EQE International. And for the past 18 months, the company has been developing the national flood-modeling system in the United States, under contract with the Federal Emergency Management Agency

In terms of man-made hazards, Eqecat provides consulting and software for managing anti-terrorist activities. More than 60 nuclear power plants and industrial chemical facilities around the world use its Midas software or the recently enhanced version, Midas-AT. Developed by PLG, another EQE subsidiary, the product also is used by the Chemical Biological Incident Response Force of the U.S. Marine Corps. Eqecat also licenses the product to federal, state and local governments' emergency-management teams, so they can model the spread of biological and chemical agents in particular areas. "There is significant demand for these types of tools," Johnson said.

Eqecat, like other catastrophe modelers, draws on a range of in-house talent to create its products. Eqecat's staff includes civil, mechanical and chemical engineers; meteorologists; environmental engineers; seismologists; geotechnical engineers; mathematicians; software developers and engineers; and actuarial and financial experts.

Johnson thinks the company has a competitive edge with insurers, reinsurers and financial institutions because its catastrophe-model development is closely linked to its parent company's management-consulting expertise.

"As EQE, the owner of Eqecat, we provide consulting services to all the industries that insurance companies write-policies to--the oil and gas industry, chemical industry, transportation, Fortune 1000 companies, the government, research institutions and commercial nuclear power facilities worldwide," he said. "So we have the underlying technology in terms of natural hazards risk and operational risk for all these types of facilities, because we're working with them day to day."

Securitization 

Catastrophe modeling has come into growing use for the securitization of risk, most commonly with catastrophe bonds. securitization is the process of transferring risk to private investors. In this realm, Eqecat broke ground in May 1999 by supplying modeling support for Oriental Land Inc. in its $200 million Tokyo Disneyland  securitization against earthquake risk. This made Tokyo Disneyland the first noninsurer to issue bonds to cover catastrophe losses ("Securitization Frontierland," Best's Review, July 1999, Property/Casualty edition).

In a market where each company claims it is the one on the cutting edge, insurers and reinsurers are left to choose which model is right for their needs. In RMS' case, company officials woo potential clients by taking them through the technical basis of how a model is built and the rigors that go into the underlying research.

"We often invite clients into our offices and get them to spend time with the engineers and scientists who built the models," Muir-Wood said. "We try and show them the degree of depth and detail which goes into this. We like to get technical. We find that the more technical it is, the more clients like it."

Clark of AIR said the recent wave of mergers and acquisitions in the insurance and reinsurance arenas has made serving clients a greater challenge.

"Our client base is becoming much more complex, with global companies that are large and complicated," Clark said. "These companies want to manage risks well. They're expanding in areas in which they can use this technology."

Applying the Models

Because of this trend, Clark expects that AIR will have to work more closely with clients in the future to integrate its technology into their daily operations. "Our support staff and our consulting staff will need to be out there, making sure clients use this technology in the ways they want to use it," Clark said.

That's an important part of the modeling equation, agreed Musick of the Alliance of American Insurers.

"To make the models work, the insurance companies have to input some data into that model, too, because the exposure base from one company to another is going to differ," he said. "So it's not only what model they choose, but how well the company's staff is able to work with those models to make them generate workable and useful results."

Information provided by catastrophe modeling has become an important part of the process of rating the financial strength of insurers.

"The models help companies identify where they are most heavily concentrated and quantify the degree to which a severe catastrophic event could weaken their surplus position," said Robert Farnam, financial analyst and actuary in the property/casualty division of A.M. Best Co. "It's helping them to be proactive in mitigating their risk."

For several years, A.M. Best has asked companies to provide indications of their probable maximum loss, which generally requires data from one or two outside modeling firms, said John Andre, assistant vice president and manager of personal-lines ratings for A.M. Best. The catastrophe-modeling firms must continue to refine their models for hurricanes and earthquakes even as they develop models for other risks, such as tornadoes and catastrophe-related workers' compensation exposures, Andre said. "We continue to observe disparities in loss indications for particular insurance companies as we compare their PML output from the various modeling firms."

State Farm, Bloomington, Ill., the leading U.S. personal-lines insurer, currently licenses catastrophe-simulation models for earthquakes and hurricanes from RMS and Eqecat. The insurer also has used AIR and E.W Blanch models in the past, said actuary Jeff McCarty.

At first, modelers concentrated their efforts on probable maximum-loss calculations, and insurers couldn't use them to measure expected hurricane loss costs for setting rates, McCarty said. In the past decade, "they've focused on measurement of expected loss that can be used on the rate side" he said.

Catastrophe models are "the best tool we have available," McCarty said. "Historical information just does not work--no matter what adjustment is made."

At American Reinsurance Corp., Princeton, N.J., six people work full time on analyzing and fine-tuning the reinsurer's portfolio using the RMS catastrophe-modeling system through a process called "portfolio optimization."

"We are making more effective use of our capital by balancing our portfolio across different geographic areas," said Carl Hedde, vice president of corporate underwriting. In the United States, that means balancing California earthquake risk, New Madrid earthquake risk in the Midwest, and hurricane risk on the East Coast. "You could even have more of a distinction between what you are writing in Northern vs. Southern California," Hedde said. "The tool really helps you to fine-tune what your portfolio really looks like."

Hedde expects modeling to expand beyond property catastrophe risks, but the benefits of modeling other perils will have to be weighed against the cost.

"Ultimately, you'll see this type of modeling used across portfolios, but we re in the very early stages for other types of risks," he said. "We're not even to maturity on hurricane and earthquake risk. We still have work to do on the science there."

As models for other risks emerge, "what the industry is going to struggle with is the cost/benefit of capturing the information," Hedde said. "The challenge is finding the next accumulation issue that presents a threat to the surplus of the companies."

Hedde suspects the next big issue will be e-commerce liability.

Modeling in the Future

Musick expects that advances in technology and information gathering will serve to further refine the art of catastrophe modeling.

"As science becomes more sophisticated, as engineering becomes more sophisticated, they're going to keep adding components, and hopefully as they keep adding new information, models will become more and more reliable," he said. "Clearly, right now, this is the wave of the future, because what woke up insurers in both Northridge [Calif., after the 1994 earthquake] and Hurricane Andrew, was that the old techniques of rate making, of building in a risk load over a long period of time, aren't going to help you when you have a hurricane like an Andrew."

Muir-Wood anticipates that in coming years, more and more catastrophe modeling will be accessible via the Internet. In fact, RMS is waiting to see the reaction of clients to its new online service. "Based on the feedback from that, it will determine how fast we move a lot of our modeling activities to be available over the Internet," he said.

The industry has come a long way in a short time, Muir-Wood said.

In the first generation of modeling, the models were created because insurers and reinsurers were fighting with problems of pricing risk in Florida, he said.

"In the second generation of modeling, you'll see that models actually help create new markets, not simply fulfill the solving of problems in existing markets," Muir-Wood said.