Lessons learned from Katrina and beyond

The 2006 Atlantic hurricane season came and went with barely a blip — despite initial projections by some that there would be two to five major hurricanes that year.

How could insurance industry predictions be so wrong?

They weren’t, according to Christopher Dineen, director in the insurance claims service practice at PricewaterhouseCoopers LLP. Robert Hair, senior state manager for Allstate Insurance Co., and David LaLonde, senior vice president for AIR Worldwide Corp., agree with him.

Speaking at a recent Casualty Actuarial Society annual meeting in San Francisco, these experts said companies weren’t necessarily that wrong in projecting catastrophe losses in the past few years. Nevertheless, they noted there are lessons that can be learned from the 2005 and 2006 experiences.

According to LaLonde, the surprise in unexpected Katrina losses in 2005 can be traced to insufficient scrutiny of the models pre-event and the fact that companies did not appropriately account for non-modeled losses. There were also problems with exposure data entered into the models.

Insurance companies don’t have an understanding of what is and isn’t included in the exposures that they’re using to analyze, LaLonde indicated. He recommended companies pay better attention to the numbers in the models.

“You need to run it to look at which policies are contributing to your large loss potential, which areas are causing geographic losses to be high, or what your key geographic regions are,” he said. “You can use the model to test underwriting strategies, to look at the impact of risk transfer. You don’t just run the model once a year and say, ‘that’s my cat risk.’ You’ve got to use the models to understand the science behind the models and use it on a day-to-day basis.”

Once companies provide correct exposure inputs for the models, they should understand that the models provide an estimate of the relative risk in making underwriting decisions. “They’re not always there to give you the absolute level of risk,” LaLonde said. For example, AIR’s models estimate losses to uninsured property, including buildings, permanent structures and outbuildings, contents, additional living expenses or direct business interruption, but they do not include hazardous waste cleanup, loss adjustment expense, etc.

Following Katrina, AIR conducted studies with companies that represent more than 50 percent of the U.S. market, to explain the differences between a company’s actual loss and modeled loss data. What the company found, LaLonde said, was that nine out of 10 commercial properties analyzed had replacement values less than the amount that would be estimated using a standard engineering base cost estimations approach.

“When you look at a policy and you have a replacement value based on the square footage of a building at $25 per square foot, you know it’s wrong,” he said, “but that was in a lot of databases.”

In the study, AIR found more than 50 percent of companies lacked construction or occupancy information for more than one-third of their policies. “When you put unknown construction into the model, it assumes the general construction category reflective of the area it’s in,” LaLonde said. However, depending whether the structure is steel frame or reinforced masonry, the loss estimates could be 50 to 80 percent more, he noted.

In some cases, multi-location policies only used one address, yet multi-location policies should have an address for each location. “If you only have one address and you’re analyzing the risk at that one location, it just isn’t going to give you the right answer,” he said.

Increase every year
In preparing for future catastrophes, LaLonde said insurers should be aware that the loss estimates computed by catastrophe models are going to increase every year, and that companies should prepare themselves for a $100 billion loss.

“This isn’t just an inflationary adjustment to the exposures,” he said of the increase in model loss estimates, “but the average house size itself is getting larger, the architectural details of homes are becoming more complex, the interior features are more opulent and the average cost per square foot of replacing a residential home and similarly commercial properties is escalating,” he said.

Loss estimates have nearly doubled from 1985 to 1995, and doubled from 1995 to 2005, he said, indicating that underwriters should prepare for increased exposures every year. “In the next 10 years, we’re going to see another doubling of estimates of loss potential coming out of the models … solely due to the fact that you’re getting bigger homes, more crowded homes, and homes in places that are along the coastline.”

Hurricane activity may have been milder than initially expected in 2006 because of cooler than expected sea surface temperatures and stronger than expected wind sheer in the Atlantic. Nevertheless, LaLonde said the standard approach to hurricane modeling is still the most credible.

He noted that the U.S. doesn’t need an active hurricane year to have a big loss. “Losses bigger than $50 billion and $100 billion are possible all over the country,” he said. “With exposure growth doubling in the next 10 years, there’s about a 20 percent probability that you’re going to see a $100 billion event in the next 20 years.

“Catastrophe models are tools,” he emphasized. “Users need to understand what they do and don’t include. Adjust the model results as appropriate, and understand and account for growth in property against exposures.”

Re-evaluate risk management
Hair suggested Katrina and 2006 were lessons to insurance companies to evaluate their risk management strategies. “When developing a cat management strategy, you need to consider risk tolerance, rating agency requirements, return on capital, impairment/solvency, efficiency and stability of strategy, and volatility of returns,” he said. “That creates some complications because companies make different decisions based on those factors.”

However, preparing for a catastrophe is similar to how consumers spend money, Hair said. “Some people like to buy big homes, some buy expensive cars, some take exotic vacations,” he explained. “It’s how people want to spend their money, and that’s what organizations do. Spending it differently can lead to perceived differences in the marketplace that causes people to question an organization’s intent.”

For example, Hair said Allstate’s strategy is “to provide shareholders with an acceptable return on the risk assumed in its property business, and to reduce the variability of the earnings while providing protection to its customers.

“That short statement leads to some different applications,” Hair said. For example, in addressing the return on risk, Allstate decided that it couldn’t get the return on risk it wanted in capital devoted to earthquakes. As a result, the company has been exiting the earthquake and “will be out of the earthquake business except for just a few states” by 2007, he said.

On the other hand, another multi-line carrier that groups home and auto customer together might decide it is able to earn an earthquake return, so that company might make a different underwriting decision, he said.

The key lesson for insurance companies following the 2005 and 2006 catastrophe seasons, Hair said, is to use the information provided by the models, but to make decisions after understanding risk to rate segmentation, and risk to residual markets. Companies should recognize that reinsurance can help to transfer risk exposure as well.

“Rating agencies are starting to require companies to keep more capital to maintain their ratings and for the need for quick action,” he said. For example, A.M. Best has a new cat risk stress test to evaluate companies when rating. Thus, he said, Allstate is looking at its reinsurance options.

Overall, insurers need to understand that exposure is increasing in risk-prone areas, risk is affected by climatological influences, and then determine their strategy, Hair recommended. Future “$200 billion and $300 billion loss events will challenge us as an industry,” he said. “We want to show customers we care, but there are organizational challenges in cultural change, structure and doing the work. Getting the organization aligned and keeping the message and communication strategy in sync. That determines where we go over the next several years.”

Curtailing claims
Dineen said another lesson from 2005’s catastrophes is that commercial claims are more complicated than homeowners claims.

“Going back to Hurricane Katrina, about 90 percent of homeowners claims have been resolved, but most commercial claims are still open,” he said.

Many commercial claims are still up in the air because of issues regarding wind versus flood, deductibles, demolition, the increase in the cost of construction, actual costs or repair/replacement.

“These issues come to a head a little bit more so in the commercial field in that there’s not the uniformity in policies as there is in residential,” Dineen explained.

With wind versus flood, for example, in commercial policies, the definition of the named storm may be different. One policy may include rain or water, whether the rain or water is driven by wind or not. “Does the wind include the rain that was pushing it?” he asked. “It’s a little bit closer to the gray area than many people would like.”

With deductibles, there also are various approaches to claims. Some have a percentage of the total insured value reported. Some are a percentage of values at the time of loss. Some are percentage per unit of insured, Dineen said.

“If you have a minimum deductible per location and you have 20 of your major customers that were affected, you potentially have a situation where you have 20 different deductibles,” he said.

With regard to business interruption, the business interruption period after a catastrophe may be different from what was projected because the economy goes south, he added.

The lesson with regard to claims, Dineen summarized, is that reconstruction costs are likely to be higher than initially predicted.

While adjusters and claims people may want to hash things out to code claims correctly, it might not matter in view of the big picture.