Not Risking It

Citing concerns about catastrophe (cat) modeling technology, which insurers use to assess the risk and predict the losses of catastrophic events, Allstate (Northbrook, Ill.; $27 billion in 2005 property liability premium) recently announced it will drop earthquake coverage for 352,000 of its policyholders. The move illustrates one way P&C insurers are trying to reduce their exposure to a variety of risks, including earthquakes and hurricanes, and highlights the concerns some carriers have about the performance of cat modeling systems.

"Allstate no longer offers earthquake coverage [in certain markets] because of the unpredictable risk presented by this type of low-frequency, high-severity natural catastrophe," relates Mike Siemienas, a spokesman for Allstate. "Current modeling is based on more theory than event-driven data. We also have a concern that there are still undiscovered fault lines not taken into account," he explains. The carrier declines to reveal which cat modeling vendors it has been using.

Allstate's move affects policyholders in all states except for California (where 14 percent of homeowners have earthquake coverage through the California Earthquake Authority), Connecticut, Florida, Kentucky, New Hampshire, New York, Pennsylvania, Rhode Island and Hawaii. Allstate's decision comes in the context of a high-profile campaign by CEO Ed Liddy to generate support for a federal disaster insurance program.

In addition, Allstate took a hard hit last year with net income losses of $1.5 billion, largely due to the turbulent 2005 hurricane season, according to Siemienas. As a result, Allstate also has continued to scale back its homeowner's coverage in Florida. "Our customers are our top priority, [but] we are making sure we are not overexposed when catastrophes hit," says Siemienas.

The performance of risk management-related tools - including cat modeling systems, mapping and predictive analytics - has been under scrutiny in the wake of the 2005 hurricane season, as insurers analyze which technologies worked and which need improvement. Industry consensus has been that the existing catastrophe models fell somewhat short. "The capability of the application itself fell short last year," says Celent (Boston) senior analyst Donald Light. As a result, "There has been a great awakening among insurers to improve the tool."

Overconfidence in Models

The availability of the models had generated overconfidence among insurers, which, after the past hurricane season, are realizing the technology's limitations, according to Paul VanderMarck, EVP of products for Risk Management Solutions (RMS), a Newark, Calif.-based cat model vendor. "Cat models are essential to the industry to quantify risk, but there is a significant uncertainty in the models themselves," says VanderMarck. "As clients started to integrate the systems, they got dependent on the technology. Sometimes the pendulum swings too far and sometimes insurers put too much trust in the model."

RMS executives emphasize that models don't predict catastrophes - they can only represent a range of potential losses to help an insurer better understand its portfolios. "The cat model will never replicate the loss of any specific event," asserts Don Windler, senior models manager at RMS. "Accuracy is difficult. These models are developed from data collected for understanding mean behavior and mean losses." That is why RMS and other cat modeling vendors, such as AIR Worldwide (Boston) and EQECAT (Oakland, Calif.) are working to improve the technology by introducing more data for more-accurate results.

Looking ahead, Celent's Light believes the industry will continue to rely on modeling to reduce risk exposure. At the same time, he adds, there will be more withdrawals from catastrophe-prone areas, such as Florida. "Allstate's decision is not an indictment of the cat model, but it is a different assessment of the exposures the carrier has," says Light. "In theory, insurance companies should make decisions based only on the data available."