Hurricane Andrew was a brusque wake-up call for the insurance industry and helped catalyze the accelerated development of catastrophe models to assess the risks posed by large hurricanes.
The industry sought improved tools that could measure and quantify hurricane risk, according to experts involved with the early models.
Hurricane preparedness had waned after a protracted period of modest hurricane activity. The Miami-Dade area had not been hit by a major hurricane since the 1960s. In the intervening years the area had undergone substantial growth and development, said Robert Muir-Wood, chief research officer in London for Risk Management Solutions Inc., a Menlo Park, California-based catastrophe modeler.
Much of the growth and the increased exposures that came with it was not closely tracked by insurers and reinsurers, said Karen Clark, co-founder and CEO of Boston-based Karen Clark & Co., a catastrophe modeling company established in 2007. In 1987, Ms. Clark founded the first catastrophe modeling company, Applied Insurance Research, which became AIR Worldwide and was acquired in 2002 by Insurance Services Office Inc.
Underestimating their exposures led to shock and surprise in the wake of Andrew’s $16 billion in insured damages, and insurers and reinsurers turned to catastrophe modeling to gain a better grasp of their potential losses.
More clients for the modelers gave them more resources with which to expand, Ms. Clark said.
“The development resources went into widening the space,” she said. For example, models for hail, European windstorm and Japanese earthquakes were developed.
Newly formed reinsurers using models as a key element of their business plans emerged to address the shortage of capacity in the wake of Andrew (see related story).
“The Bermuda reinsurers relied on risk modeling as opposed to having a long historical record of their own,” Mr. Muir-Wood said, providing another avenue of growth for the modeling industry.
Cat models use computer simulations to assess how windstorms of various strengths would affect property in a geographic area.
Justin Davies, head of Europe, the Middle East and Africa for Xceedance Inc., said the increased use of modeling also led to hiring and structural changes at insurers and reinsurers.
“You can’t just license a catastrophe model. You’ve got to have processes in place, people to prepare and input the data, and you’ve got to understand what the output means,” he said. “Andrew made people believe there was a big potential risk in Florida and on the Eastern seaboard and they started properly creating catastrophe modeling teams.”
More recently, clients have asked whether models are accounting for climate change, said Peter Sousounis, vice president, director of climate change research, in Boston for Verisk Analytics Inc.
This led Verisk to create a data catalog of years in which ocean temperatures were above normal against which models could be run to give some idea of what activity might be like during years in which sea surface temperatures are above normal, something clients use as “a proxy” for climate change, Mr. Sousounis said.
Similar data catalogs have been developed for so-called El Nino and La Nina years to reflect differences in the warm and cool phases of the recurring climate pattern across the tropical Pacific, he said.
Making catastrophe models more flexible to incorporate more variables faster is a feature that continues to evolve, said Ms. Clark, whose company can send clients daily hail and tornado footprints automatically.
Improvements in technology and data capture have provided an improved foundation for catastrophe modeling. “People are feeding the models with much more accurate data,” Mr. Davies said.
In addition, advances in computer power have enabled more accurate and detailed modeling.
“Advances in computer power helped enormously to improve and enhance the reliability and realistic nature of the models,” Ms. Clark said. Resolution has also improved from five-digit zip code levels to individual properties being geocoded with exact locations, she said.
“The idea for catastrophe models was set out in the 1970s, but at that time the cost of computation was too great to make it practical for insurers,” Mr. Muir-Wood said. “It took until the 1980s until the computational resources were available. Computational ability has increased so markedly.”
When Hurricane Andrew ripped through southern Florida in 1992, tearing up buildings and killing more than two dozen people, few had witnessed such a destructive natural catastrophe in the United States.