Creator of catastrophe risk model broadens focus

Oct 6, 2008

Karen Clark created the first computer model to help insurers better gauge the losses they might face after a major catastrophe. Now she is working with insurers to use the models more wisely.

Miami Herald--October 6, 2008

The woman who created the first catastrophe risk model in the mid-1980s is now in the business of evaluating them.

Karen Clark, trained as an economist, started a new company last year to help home insurers evaluate the computer risk models they use to gauge the losses they might face from a massive storm or other natural disaster.

Clark says there’s an over-reliance on these models today.

“They are an important tool, but they are just one tool. I want the industry to take it to the next level and use models sensibly. They should use more than the models because there’s a lot of uncertainty around the models.”

The uncertainty stems from the data the models use. Every time the information a model uses changes, the basis for an insurer’s decision-making would change as well.

Her new firm, Karen Clark & Company, launched the KC Hurricane Damage Scale in mid-August to help insurers get a better picture of the amount of wind damage they could expect from a tropical storm or hurricane that makes landfall.

Unlike the Saffir-Simpson Hurricane Scale, which ranks storms in categories from 1 to 5 based on wind speeds, the KC damage scale provides an indication of the possible damage a storm could cause to buildings, contents and business interruption.

Clark says damage can occur well before wind speeds reach the 74 mph where the Saffir-Simpson begins counting.

The KC scale runs from KC 0, where winds are under 40 mph and cause negligible insured damage, to KC 7, where winds are above 160 mph and all but the best-fortified buildings would be destroyed.

A few weeks ago, Clark chatted with The Miami Herald about the hurricane damage scale and the roles the computer risk models play in the insurance world today.

Q: How does the hurricane damage scale work with the computer models already out there?

A: It should help [insurance] executives get a better feel for their risk. They should go hand in hand. We’re not providing the company with any loss estimates. They will still get that from a model.

The [KC damage scale] should give insurers a vivid picture of the types of claims they could see. This helps more with planning claims-adjusting activities and knowing what kind of claims they might have versus just knowing they could have a $50 million loss. Hopefully, [the models and the damage index] won’t be saying totally opposite things.

Our mission is trying to help the industry understand catastrophe risk better. We don’t have any intention of charging for it.

Q: Florida is committed to mitigation as possibly the best way to protect homes and contain insurance costs. But figuring out how to set mitigation credits is difficult.

How do the models value mitigation?

A: There are many aspects of the models where we don’t have a lot of hard data for. This is definitely true for many of the mitigation devices. Many of the companies don’t even collect that information.

Right now, mitigation credits are based on engineering judgements. All the modelers have highly qualified structural engineers on staff and consult with outside experts.

Nobody knows for sure if [a mitigation measure is worth] 20 percent or is it 5 percent or is it 15 percent?

That’s an engineering judgment. Hopefully, in the future companies will start to collect that data and we will have more detailed information on those mitigation devices.

But there is very good data after the 2004 and 2005 hurricane seasons. There’s a lot of claims data on structures built based on the newest Florida building code, with all the new fortifications built into them. So we know mitigation works.

Q: Since 2004 and 2005, many models have changed the time frame they consider when analyzing hurricane activity and potential losses. Most have introduced models that consider a much shorter time span on activity. Is that wise?

A: To be able to predict what will happen over a short time period, while it would be nice if we could do that, it really increases the uncertainty and volatility around the hurricane loss estimate. That’s really the problem. It is much more difficult to predict what will happen over one year or two years or even five years.

Q: Insurers are obviously the modeling companies’ biggest clients. What kind of influence do insurers have over the factors that modelers stress or choose not to emphasize?

A: For the most part, the catastrophe modelers base their model assumptions on scientifically defensible information that comes from their own internal staff or external experts. They don’t base their model assumptions on what insurance companies say.

But the problem is that actual hurricane data are sparse. Much of the scientific data that is available isn’t objective or factual. It’s very subjective and open to interpretation. Always keep in mind that models are based on few facts and a lot of scientific judgments.

Q: Were the models modified after the 2004 and 2005 storms at the request of insurers?

A: I wouldn’t say the models were modified at the request of insurers. But given the industry criticism that the models underestimated both the 2004 and 2005 losses, every modeling company has had to review their loss estimates.

Every modeling company has made upward adjustments. But there has been a wide variation in how much the different modeling companies have increased their hurricane loss estimates.