Blowing in the Wind

There wasn't much risk modeling being done in the insurance industry prior to Hurricane Andrew. But after that devastating storm struck Florida in 1992, insurers realized the old way of doing business had to go and new tools were needed for the industry to manage catastrophe exposure. Today, as the models mature, insurance carriers and reinsurers are becoming increasingly reliant on the models, particularly for use in hurricane and earthquake coverages. The software assists underwriters with difficult decisions 20th century underwriters had to make relying on market knowledge and gut instinct.

Vanguard Fire and Casualty writes only Florida homeowners, a volatile market where reinsurers play a major role in protecting policyholders and insurers. “Insurers and reinsurers have a lot on the line,” says Martin Bobek, chief underwriting officer for Vanguard. “We are very close to [the reinsurers] and, to some extent, at their mercy in terms of having to underwrite business that makes [the reinsurers] comfortable doing business with us.” The reinsurers depend heavily on risk models, and to varying degrees, Bobek believes, some are dependent on the software completely.

The models represent the best estimate the industry has for storms, according to Werner Kruck, vice president of personal lines underwriting for the Southeast division of North Pointe Insurance. “If not the models, then what?” he asks. “The models capture all the knowledge we have. It may not be enough, but it's all we have.”

While catastrophe models have evolved as tools to manage a carrier's portfolio, Robert Stevenson, syndicate exposure manager for London-based Wellington Underwriting Agencies Limited, concedes models are not the only tools to manage an insurer's book. Nevertheless, “they certainly are a big contributor to the way we manage our risks, buy our reinsurance, and how our company operates,” he maintains.

The expectation within the industry is modeling tools will continue to evolve, indicates Gail McGiffin, the partner responsible for underwriting solutions in the consultancy practice for Accenture. As the world pays increased attention to global warming, she explains, updates to models must incorporate the weather-related conditional changes that likely would impact the models. Another modification will allow insurers the ability to model for a multitude of perils across multiple lines of business. “When you look at 9/11, that was the first time the industry recognized an event was not just a property loss but a loss in every line of business–life insurance, workers' comp, general liability, automobile,” she says.

Insurers need to examine the implications of occurrences such as hurricane and tornado activity as a result of climatological changes, McGiffin asserts. “The implications for the models are the severity of damage and the frequency and magnitude of hurricanes,” she says. “What are the implications for those elements that are incorporated into these models as a result of global warming?” Companies aren't modeling global warming specifically, observes McGiffin, but they are modeling the implications of global warming in the increasing frequency and severity of weather-relatedevents. “The models will continue to focus on the insurable events and determine portfolio exposure, but they will apply the latest insights as to how the climatological changes fit into the model,” she says.

Pluses and Minuses

Models have become a big part of underwriting, growth strategies, and reinsurance, but insurers sometimes are left scratching their heads, notes Bobek. “If you spend a lot of time putting together what the models call a 'good book,' yet the models are in a constant state of flux and adjustment, you can see from an underwriting standpoint for a company such as Vanguard, it becomes fairly difficult,” he says. “The rules change as the models change.”

Vanguard's philosophy, states Bobek, is the model never can be the sole driver for what the company does. “While we are forced to give a great deal of weight to modeling, we also reserve in some corners of the book a common-sense principle we believe in,” he explains. Although most Florida-based insurers have a strong knowledge of their market, Bobek feels that isn't always the case with reinsurers. “A lot of times you are talking about smart people who understand a lot but may not be experts on the Florida market and constantly have to fall back on what the modeling output is,” he says. “An old CEO I used to work for said we have to strike a balance between the art and the science of underwriting, and that's very true.”

That approach can be forgotten, though, when there's a lot of money on the line. “People like to gravitate toward the things they can point to,” continues Bobek. “Gut feelings don't have a lot of pull after you just lost $50 billion in a storm in South Florida. But [underwriting officers] always can go back to their board of directors and say they did what the industry says they should do–they used all the models. That's easier to sell at the end of the day.”

Delivering Data

While Bobek attests to the value of risk models, he emphasizes the need to obtain consistent and viable data if the models are going to be accurate. “Over the last couple of years, there has been some indication the data we've used has not been the best,” he says. “It's caused us to miscalculate to some degree and caused some problems with our results.”

In addition, the insurance industry's technology shortcomings have come into play in the collection and use of data, Bobek suggests. “The ability of insurance infrastructures to collect and maintain data, mine data, and sift through it hasn't been all that great until recently when it became apparent data is a key component of being successful in the insurance industry, especially if you are operating in CAT-exposed areas,” he asserts. “If [modelers] had the opportunity to do it again, they possibly would create better data standards and look under some of those rocks to make sure what they were getting was viable data. When the models turn out to be off–and, in some cases, by very large measures–it probably is the industry's fault it doesn't have good data. But for modelers, if that's your bread and butter, you better be sure your data is good. There should be some accountability on their end, as well.”

Stevenson, whose company uses models from Risk Management Solutions, believes the frequency of storms in the U.S. has changed the view of risk and therefore has altered storm models considerably. “[Modeling vendors] are updating their hurricane models this summer, and there certainly is a change to how hurricane risk is perceived,” he points out.

Because there is more experience built into the hurricane models, Stevenson contends they are more robust products. “The more claims that are available, [the more modelers] can refine the models, particularly the different constructions,” he says. In his view, the models work best on the law of large numbers. “Dealing with millions of locations within a book of business, [the models] tend to be more robust as far as giving loss estimates,” he says. “When you get down to small books of business, that's where the real volatility can start to exist.”

Reliant on Others

Vanguard does not own any modeling software. Instead, the carrier, like many midsize insurers, relies on its reinsurance partners, which license RiskMeter Online for that service. “We've kicked around the idea of modeling in-house, but the brokers are so efficient in this they haven't left any gaps, so we don't feel the need,” says Bobek. “It's an additional set of capabilities and experience [the carriers don't have to pay for directly], and it takes a certain degree of maintenance to keep the models up and running full time. It's one less thing for us to worry about.”

Unlike many other smaller insurers for which such an expense is prohibitive, Sunshine State Insurance licenses the modeling software directly from AIR Worldwide. Mike Weis, assistant vice president of underwriting special programs for Sunshine, reports the software is easy to use once users have the opportunity to spend some time with it. The key for the success of the modeling is compiling accurate data, comments Weis. “On a monthly basis, we take a snapshot of what our book of business looks like and configure the data so we can import it into the modeling software,” he says. “Using the parameters we've chosen, we take a look at our business at that point in time.”

Sunshine is different than other similar-size companies, Weis believes, because the carrier turns to an outsourcing partner to maintain its data. “[The data partner] has a full team of programmers and software engineers who are familiar with how to structure the data so it will be easy to import,” says Weis. “I never receive raw data I have to try to manipulate to make compatible with the AIR software.”

Room for Improvement

In general, while modeling software is getting easier to use, Kruck remarks it still remains a complicated activity. “You do have to use accurate data to get the file structure right,” he asserts. “Then you have to make sure you set your parameters correctly. The large intermediaries and the large reinsurers have experts who know this stuff backward and forward. For small companies, to license the software as well as learn how to use it is prohibitive.”

So, although the purpose of models has been to manage exposure and predict costs, investing decisions must involve more than just the model in the short term, observes Kruck. “You are looking at making an investment where you could lose everything,” he says. “The model is just a tool. At some point you have to ask, 'How much of a risk do I want to take with my capital?' I think we're making adjustments in the market, and in a couple of years, it will be a pretty attractive market.”

The model used by Sunshine predicts both the frequency of future events and geographic location, Weis reports. “The purpose of the model is to give an indication of where the highest probability is of a storm hitting,” he says. “I think the models provide us with accurate information. It's not an exact science, but it's given us a good prediction of what to expect.” Recent claims activity has mirrored the results of the model, he adds. “The modeling gives us an edge from the standpoint of looking at our business and where we might be vulnerable from an exposure standpoint,” affirms Weis.

The modeling continues to refine itself, both in terms of the software and the data, agrees Kruck, who points out one of the best things the models have going for them in Florida unfortunately is all the hurricanes. “When we have hurricanes, that gives [the modelers] the opportunity to correlate their data,” he says. “The models have become more accurate in terms of how the damage will occur and at what level from an individual storm. I would guess [the models] probably are not far off on the frequency over a 100-year period of time. Right now, the issue is their frequency over the next 10 years.”

Modeling software could be improved to deal with some of the complexities of the policies written, Stevenson suggests, adding he believes the modeling companies have such improvements on their agenda. “The market has matured quite quickly to keep up with the demand,” he says. “These are complex pieces of software. To change things [quickly] could have a huge effect up the chain. The software functionality certainly could be improved, but it's not something that could be turned around in a month and everyone's happy.”

CAT models, particularly after Hurricane Katrina, came in for some bad press, notes Stevenson. “There have been some absolute roaring successes, but there have been some things we've had to learn–sometimes a bit too painfully,” he says. “But we've learned and moved forward.”

Key Measurements, New Uses

One of the many measurements the models take is the ratio of probable maximum loss (PML) to premium. The number compares the amount of premium brought in by the company with the potential maximum loss the book of business might suffer, Bobek explains. The lower the ratio–if the models are right–the better off the company and the reinsurer are. “We'll take a book of business and our projections, growth plans, and underwriting plans and run them through the models to try to determine whether our growth strategy would lead to an acceptable PML to premium ratio for the reinsurer and our company,” he says. This allows carriers to demonstrate the book of business will be good for the reinsurers in the long term.

A recent addition to the models targets terrorism. “Probabilistic terrorism modeling could be useful,” according to Stevenson. “[The models] are in their infancy as opposed to the property storm and earthquake models that have been around for a while. The terrorism models have come a long way in a reasonably short period of time.” More models also need to be refined to examine flood, fire, and tsunami catastrophes, he indicates.

Taking stock of the big picture, McGiffin confirms terrorism is considered another catastrophe category by the industry, although the models are not as well established as those for hurricane, flood, and earthquake. “[Insurers] also are trying to do better risk assessment upfront to identify their distance to exposure,” she says. Property insurers in Washington, D.C., for example, need to know how close their insured property is to other landmarks that might be terrorism targets. “Part of it is modeling, and part of it is looking at the point to point of the radial exposures your insured might be exposed to,” she says. “There are spatial analytics deployed even beyond the models themselves.”

As for other risks, McGiffin views the ability to model exposures in brush areas for fires as still advancing. Pandemic events are another area where she is seeing modeling interest. McGiffin also asserts modeling needs to take on a broader definition. “If you look at the lessons learned from Katrina, it's not only what insurers were exposed to and lost, but it's the ability to settle claims and to access the areas to deploy claims adjusters intelligently and in a timely manner to get to the people who lost their homes,” she says. “That's a whole new area where people are going to adjust their response and be more proactive–the need to centralize communications so customers can find [the insurer] when it's hard to find the customers.”

In Bobek's opinion, models continue to improve. Although some industry people worry when they see models with continuous updates, Bobek points out when modeling something as complex and unpredictable as weather, it's unrealistic to expect modelers to come out of the chute with a perfect model. “As things change–weather patterns, building standards in Florida–these changes are going to cause future [modifications] to the models,” he says. “As long as you don't become dependent solely on modeling to do business, then it's a very useful tool. But it can't be the only arrow in the quiver. There's a lot more to underwriting than running data through a model.”