Modeling after the big one: applying lessons from the 2004 and 2005 hurricane seasons, catastrophe modelers have changed their products and insurers have changed how they use them.

Catastrophe models were put to possibly their most stringent test ever when Hurricane Katrina blew across the Gulf Coast in August 2005. Some insurers, such as State Farm, said their modeling estimates of damages caused by the storm were nearly on par, but others criticized the models for dramatically underestimating damages, resulting in high losses for insurers.

They "missed the mark," said some insurers, caused in part by the storm's unexpected flooding and storm surge. Cat modelers disagreed. "Right after Katrina people were looking for a scapegoat and focused on models. But that discussion quickly became more constructive and it's clear that modeling remains an essential part of the industry's ability to manage catastrophe risks," said Paul VanderMarck, executive vice president of Risk Management Solutions.

Cat models date back to the mid-1980s, but it wasn't until Hurricane Andrew hit southern Florida in August 1992 that models gained widespread attention.

Then came Katrina. Although some of the critics now say the modeling picture maybe wasn't as bleak as they initially thought, cat modelers aren't taking any chances. Using lessons from the 2004 and 2005 hurricane seasons, they've made changes to their models. Addition of several new perils and enhanced technology are among recent model improvements designed to help insurers identify risk and gain an accurate picture of potential losses from future catastrophes.

From Challenges to Change

"Modelers have spent the last eight to 10 months trying to learn lessons from prior loss history and other research to examine the underlying model parameters. This year, they took a look at hundreds of thousands of storm claims from 2004 and 2005 from multiple companies and came up with a number of different findings," said Jim Coyle, senior vice president of catastrophe planning for Travelers.

Changes that RMS made to its models included reflecting climate variability in the modeling of event frequency, increasing the differentiation between individual risks, and driving toward models that are as comprehensive as possible, VanderMarck said.

"Our clients want to use the models to capture and analyze all lines of business they have exposures to in given events," VanderMarck said. "One thing the last two seasons reinforced is the importance for us to minimize what's generally referred to as nonmodeled loss, in order to provide our clients as complete a picture as possible of their overall risk."

Another major modeler, Eqecat, also made changes to WORLDCATenterprise--its modeling software platform that enables clients to assess and quantify potential damages and losses from wind, earthquake, flood, wildfire, terrorism, among other perils. "Eqecat added a full suite of all the perils and is able to break down any one loss, such as hurricanes into flood components, wind components, or offshore energy," said Tom Larsen, senior vice president.

Hurricane Katrina prompted some modelers to add flooding and storm surges to models. With AIR Worldwide Corp's recently enhanced storm-surge subcomponent to its model, damage from winds and storm surge are estimated for every hurricane it simulates, said Dr. Jayanta Guin, vice president of research and modeling.

"The industry was surprised by two very active seasons back-to-back and having sustained very large losses, the immediate reaction was to understand whether we're in the regime of higher hurricane activity" said Guin. The company responded by introducing three "catalogs," or ultimate models, to its product offerings, including its standard model based on more than 100 years of data and the company's own research from the past two decades, a near-term view of risks over the next five years, and a seasonal catalog that's a reflection of likely outcomes for the upcoming season.

Insurers also are seeing a change to the duration of modeling. "Two of the major modelers now incorporate with recent releases a short-term frequency perspective of the hurricane exposure. They recognize hurricanes in the current warm cycle and allow users to look at loss exposure based on short-term or long-term frequency," said Jeffrey McCarty, assistant vice president and actuary for State Farm. "That's driven a pretty sizable change in loss estimates for the short term."

Demand surge, or loss amplification, also has been introduced into some models. Demand surge is the increased cost of labor and materials that occurs following a catastrophe. That's then built into a model and, the larger the storm, the larger the percentage of demand surge, said Lew Augustine, chief risk officer for Safeco. "We saw real impact on amplification and shortages in certain building materials both locally and across the country, and when we can get them, at greatly inflated prices--direct results of Katrina loss."

A few insurers aren't as optimistic about changes to models. "The only positive outcome of the past two years' devastating seasons was that it was the first opportunity we've had to calibrate the assumptions made in the models. Models are built off past data and during a period of time when they didn't have the level of detail that's now available. For instance, construction improved during the past 15 years and that was never really tested," said Andrew Castaldi, senior vice president of catastrophe perils for Swiss Re.

After the 2004 and 2005 hurricane seasons, he said, there was suddenly a wealth of data available to start calibrating models. "Now we can see how residential and commercial lines perform as the vulnerabilities were understated in many cases."

Insurers' Implications

In some cases, the changes are in the way insurers use models. "I don't know if it's fair to say modeling has changed a lot in the past two years, but we're certainly paying more attention to models than ever before," said Alan Ledbetter, senior vice president and chief underwriting officer, personal market, for Liberty Mutual. When the company gets a model, he said, it's run against other recent known storms to see how closely it predicts the damage caused by the storms.

Lessons learned over the past two years have allowed Travelers to write accounts that perform better based on an improved awareness of the impact of various risk characteristics on weather-related loss potential, Coyle said. "That also allows us to better assist clients with loss prevention strategies and techniques, including business continuity planning--all things that are important not only from a catastrophe loss standpoint but from any loss standpoint" he said.

Companies have to manage their capital more carefully, said Liberty Mutual's Ledbetter. "Using models helps us assess potential damage from new business we write, which allows us to maintain as broad a market as we can within the constraints of prudent capital management. That's a different twist on use of models than we may have had in the past. We use them solely for the purpose of constructing reinsurance portfolios."

Safeco's Augustine said one impact of the new use of models will be an overall rise in reinsurance costs. "There'll be a fairly significant increase based on new modeling results because reinsurers are using the same models. Companies will need to find ways to build those costs into their direct rates. In some cases where reinsurance costs are significant, companies will have to find unique ways to lower them or suffer the consequences of reduced returns or possibly minimized writings."

Safeco recently decided to bring a model in-house to help achieve profitable growth. "We have to be looking at modeling results to try to get optimal mix of business--both coastal and noncoastal--where we can get premium to cover risks. We'll have to balance all those concerns and do it more quickly, and bringing modeling in-house will do that" said Augustine.

Understanding Data

The biggest changes in how insurers use models "are focused on improving the quality of the input data on exposures, interpreting the output and understanding its completeness and sensitivities, and focusing on driving use of modeling further into the underwriting decision-making process," said VanderMarck.

He said RMS spent much time over the past year helping clients better understand uncertainties and sensitivities in results, based on both data quality and inherent model uncertainties. "We're helping them understand how to sensitivity-test their results relative to input data quality or to understand for certain types of underwriting decisions what key assumptions are in the model," said VanderMarck.

"Users need to remember that models are only as good as the quality of data put into it," said Travelers' Coyle. He and his team work closely with clients through producer partnerships to improve the quality of information received.

RMS recently gathered more than $10 billion of location-, address- and construction-specific claims data to develop its recently-released Version 6 update to its model. The new version substantially increases the amount of differentiation in model results within general risk classes. "Clients now can use the model to differentiate much more precisely between one risk and another based on occupancy type, year of construction, construction type, etc." said VanderMarck.

Looking Ahead

Models likely will become even broader in scope, said Swiss Re's Castaldi. For instance, he said, in addition to wind peril of a hurricane, there also may be inland flood, and "we'll adjust for local conditions like the ground being saturated from previous storms that could mean more tree damage." He said the industry will continue to get stronger in technology but "before that improves we'll have to invest some time and effort in improving our data collection and that includes claims data."

Travelers' Coyle said modelers will continue to work on several areas, including closing the gap on unmodeled exposures, additional refinement of newer products, probability models for events such as fire and infectious diseases, and potentially moving into closing the gap on data quality.

Modelers said they'll continue to take lessons learned and add enhancements to meet clients' needs. "Our clients are making underwriting decisions everyday with our models. When we learn something material that changes our understanding of risk, we'll then reflect that in the models as soon as possible,' said RMS' VanderMarck.

Key Points

* The addition of several new perils and enhanced technology are among improvements modelers have made to catastrophe models following the past two years' hurricane seasons.

* Hurricane Katrina prompted some modelers to add flooding and storm surges.

* New modeling results may cause an increase in reinsurance rates.

* Some insurers are bringing modeling in-house.

What's New in Cat Models

In the past year, some models have been enhanced to include the following capabilities:

* Consider climate variability in modeling event frequency

* Offer increased differentiation between individual risks

* Capture and analyze all lines of business in which an insurer has exposures in given events

* Break down any one loss into components, such as hurricanes into flood and wind components

* Model flooding and storm surges

* Provide a short-term frequency perspective of hurricane exposure

* Consider demand surge in figuring potential losses

Learn More

Liberty Mutual Insurance Cos.

A.M. Best Company # 00060

Distribution: Direct, independent agents and brokers, captive agents

Safeco Insurance Cos.

A.M. Best Company # 00078

Distribution: Independent agents

St. Paul Travelers Cos.

A.M. Best Company # 58470

Distribution: Independent agents

State Farm Group

A.M. Best Company # 00088

Distribution: Exclusive agencies

Swiss Re

A.M. Best Company # 85010

Distribution: Reinsurance brokers

For ratings and other financial strength information about these companies, visit www.ambest.com.

A Useful Tool

Nothing's perfect, and insurers and reinsurers have to remember that models are tools--not the answer, said Jeffrey McCarty, assistant vice president and actuary for State Farm.

"Modelers are trying to inform users that models will sometimes be wrong, but the degree to which they're wrong is an issue," said McCarty. He recommends insureds use multiple models to gain different perspectives for their portfolios.

Andrew Castaldi, senior vice president of catastrophe perils for Swiss Re, fears too many people are becoming overly dependent on models. "They believe whatever comes out of the model rather than using some common sense. Katrina was a good wake-up call, and people such as underwriters began thinking that models are helpful tools but they can't be overly reliant on the tools," he said.

Companies should apply new models to any storm that they had exposure from the previous year and see how the model results compare to their actual loss experience. "I would bet about 90% of companies would show that their losses were much greater than the modeled losses," Castaldi said.

The biggest weakness of models is in contents and business interruption vulnerability, Castaldi said. This information is "based on how buildings will react to certain storm conditions. That becomes compounded when they do a lot of their benchmarking on the models. There's much to be done as we're still in the formative stage of capturing details necessary to adjust models appropriately."

"Models are like foreign oil--you can't totally abandon their use because they're important, but you have to make sure we have additional risk analysis tools to use as well," said Jim Coyle, senior vice president of catastrophe planning for Travelers. He said the industry can't afford to over-react or under-react to model changes. "Either direction is problematic for us ... if we over-react, we could lose man), of our customers, and if we under-react we could lose lots of money," he said.

"Our focus is not merely limiting our interest to the number that comes out of the model, recognizing that the number always will have a certain amount of fragility to it. It's also important to try to understand the underlying knowledge contained in the model because ignoring that is missing a valuable source of information that helps us 'get it right.' So all the lessons learned--about types of construction, occupancies, distance-to-coast, etc.--are important when we go forward in analyzing our exposures," said Coyle.