Science, Catastrophe Risk Models and Insurance
Richard J. Murnane

Most people, scientists included, think of homeowners’ insurance only when they receive their annual policy renewal statement. Recent changes in catastrophe risk models and the losses from the 2004 and 2005 hurricane seasons, however, caused many current and potential homeowners and businesses to worry anew over insurance. And for good reason.

The cost of homeowners’ insurance for regions exposed to hurricane landfall has in many cases dramatically increased. Some insurers stopped issuing new policies, particularly in coastal regions with high property values, forcing state governments to play a larger role as an insurer of last resort. In light of these changes, geoscientists, climatologists and meteorologists should be interested in how their research might influence property insurance bills. An appreciation of how scientific research is used in the insurance industry’s catastrophe risk models provides some insight on the relationship between geoscience and insurance.

Risk modeling companies have developed catastrophe risk models (cat models) for many different hazards, including earthquakes, tropical cyclones, wildfires, tornados, hail, and others, and for locations throughout the world where there is an insurance market. The modeling companies incorporate the best scientific, engineering, financial and actuarial knowledge into their products and tune the models to reproduce past catastrophic events. Cat models have become more complex and realistic as computational power has increased, and as information has become digitally accessible.

The four major pieces in a cat model account for exposure, hazard, vulnerability and loss. The exposure piece provides information on the structures and material that is potentially exposed to a hazard. The hazard piece describes the characteristics of the potential hazard: This is where geoscience research has the greatest impact. For earthquakes, this might be the probability, magnitude and movement of an earthquake occurring at a specific location and the response of the ground surface to the earthquake. For hurricanes, this would include the size of a storm, its maximum winds, the direction and speed of its motion, and how it weakens over land.

Among other factors, the vulnerability piece accounts for how a structure responds to the forces produced by a hazard — for example, ground motion from earthquakes or winds from a hurricane — as well as the subsequent damage to a structure’s contents, and potential follow-on events such as fire following an earthquake. The loss piece estimates the total loss from an event and accounts for policy and reinsurance details such as deductibles, limits and reinsurance treaties.

Cat models continue to evolve as knowledge advances and catastrophic events occur. This evolution can have a dramatic impact on cat model results, particularly in response to a change in our understanding of a hazard — and this is where scientific research reverberates through the insurance industry. The turmoil in the world of insurance after the 2004 and 2005 hurricane seasons is a case in point.

In some ways, the 2004 and 2005 hurricane seasons set up the “perfect storm” for the property catastrophe insurance industry. The 2004 season was an exceptional year for tropical cyclones. Florida was the first state to experience four hurricanes in a single year since 1886, when Texas was repeatedly battered. Three of the hurricanes that struck Florida — Charley, Ivan and Jeanne — were classified as major hurricanes of Category 3 or greater, with winds exceeding 177 kilometers per hour (110 miles per hour). The fourth hurricane to hit Florida, Hurricane Frances, struck as a Category-2 storm. On the other side of the world, 10 storms with winds of tropical storm strength or greater struck Japan and “blew away” the previous record of six storms, set in 1990.

The 2005 Atlantic hurricane season broke numerous records including the largest number of named storms (28), hurricanes (15), Category-5 hurricanes (four), tropical cyclone landfalls in the United States (nine), and intense hurricane landfalls in the United States (five). Hurricane Katrina also produced the largest insured loss, estimated at more than $45 billion, a record for the property catastrophe industry.

Hurricane landfalls in 2004 and 2005 caused more than $75 billion in insured property losses in the United States alone. This series of large losses attracted significant attention, and despite the lack of justification, produced a fair amount of discussion asserting that global climate change was behind the large losses.

These two active seasons were followed by several business-related phenomena, which caused the price of insurance to increase. First, in the spring of 2006, a leading risk-modeling firm released a new version of a U.S. hurricane cat model, which suggested that, relative to earlier estimates, average annual losses could be up to 40 percent higher for residential properties and 70 percent higher for commercial properties.

These increased losses were related both to changes in model vulnerability algorithms made after extensive surveys of building performance during the previous two hurricane seasons, and to an increase in the modeled probability of hurricane landfall, related to multidecadal variations in hurricane activity. Instead of using an average of hurricane activity over the past 100 years, the hurricane probability represents an expected average over the next five years. This five-year outlook attempts to balance our scientific understanding of multidecadal variability in the ocean and atmosphere — we are currently in an era with hurricane activity above the long-term average — and of shorter-term climate variability driven by phenomena such as the El Niño Southern Oscillation (El Niños tend to suppress hurricane genesis in the Atlantic) against the insurance industry’s business requirements.

Second, both actual hurricane losses and the increase in model-estimated losses caused rating agencies to change the capital requirements for insurers and reinsurers. To maintain their credit rating, insurers and reinsurers needed to limit their exposure to potentially huge losses. Thus, insurers needed to buy more property catastrophe reinsurance and reinsurers needed to sell less.

Third, the large losses of 2004 and 2005 and the expectation of additional events prompted many insurers to reduce their exposure in high-risk and high-value coastal areas. For example, some insurance companies have stopped issuing new policies in areas, such as Long Island, where they think the risk of a large loss from a hurricane is too high. Fourth, Florida changed the details of their state-backed insurance fund in an effort to maintain solvency. And finally, the National Flood Insurance Program borrowed more than $20 billion, yet still failed to meet its obligations from the 2005 hurricane season. The net effect of these factors was a dramatic increase in the cost of property insurance and a reduction in its availability.

This cycle of hurricane-related events involving both the storms and the business decisions has many potential parallels for geoscientists working on earthquake hazards. The results from research efforts to improve ground-acceleration probabilities and algorithms for seismic attenuation coefficients will be incorporated into the hazard component of future earthquake cat models. After a significant earthquake event, our understanding of a structure’s vulnerability to ground acceleration often changes, and this will be reflected in the vulnerability component of a cat model. And, as for Florida after Hurricane Andrew in 1992, building codes are often revised in response to building performance during an extreme event. For example, during the magnitude-6.7 Northridge, Calif., earthquake in 1994, steel-frame buildings performed more poorly than expected and building codes were changed. In addition, the large insured losses from the Northridge earthquake caused many insurers to stop selling or renewing homeowners’ insurance in California. In response, the state of California formed the California Earthquake Authority, a privately financed, publicly managed organization that offers earthquake insurance.

Catastrophe models have become a nearly indispensable tool for the property insurance industry. Nevertheless, we should not naively believe cat model results. As the loss experience from Hurricane Katrina demonstrates, actual losses can differ significantly from early model estimates. The preliminary estimates of insured losses ranged from approximately $10 billion to $25 billion shortly after Hurricane Katrina struck New Orleans. Revised insured loss estimates range from $40 billion to $60 billion. The large increase is due in part to the extensive flooding caused by failure of the levees and pumps. Such unanticipated, threshold-type events could occur in the future as well. Despite these potential shortcomings, cat risk models are valuable tools for sophisticated users and they will likely play an increasingly important role in insurance, emergency management and societal decisions.

Geoscientists’ research will continue to provide critical information for improving cat models. An example of a well-known community effort with direct impact on cat models involves defining the probabilities of peak ground acceleration throughout the United States. Other topics of importance include research on paleoseismicity, seismic attenuation, and intraplate seismicity. Model developers will incorporate advances from these and other topics in new models. And although no one looks forward to it, the next large earthquake will no doubt provide surprises and additional insights that will be important for geoscientists and cat modelers alike.

Back to top