Catastrophe Modeling Software Has Evolved In The Age Of Terrorism

While revolutionary for their time, these methods were not very accurate, and companies began to research new methods of risk assessment and management. Advancements in computers led to the development of much more sophisticated models.

It started as a set of deterministic storms that were modeled against a portfolio of business. Next came a larger stochastic (or random) event set and the important development of adding "probabilities" to the events.

The main hurdle facing these new models was that more accurate and detailed information needed to be captured and collected by insurance companies in order to take advantage of these new tools. It was very costly to rewrite a large number of legacy systems, and a large majority of companies did not feel it was necessary to incur the expense.

Then Hurricane Andrew struck, and overnight the computerized catastrophe model came of age. Companies quickly invested the money necessary to revamp systems to collect whatever data the modelers requested.

The computer models and their incorporation of event probabilities gave insurance companies the ability to answer with more credibility two questions: "How high is high?" and "What should I charge to cover my catastrophe risks?"

The models answered the first question by estimating losses on a severity basis, commonly known as "PML," for probable maximum loss. A loss curve was calculated that allowed a company to view the probability that they would experience an event exceeding a given loss amount. The reciprocal of this probability is commonly expressed as a "return time."

The models also used probabilities to estimate pure premium for the given peril, commonly known as average annual loss, or AAL. These estimates allowed an insurer to begin to understand if it was collecting enough premium dollars to pay for its expected catastrophe losses, as well as to determine how to protect the company from the possibility of severe losses.

The models were also able to analyze different reinsurance strategies, which helped in determining appropriate reinsurance structures. Another important use was the review of "key drivers"using the model to determine which policies, concentrations of business, etc. were driving the loss estimates. With key driver information, a company could determine what new underwriting guidelines needed to be implemented, if certain policies needed to be canceled, etc. The models were quite good at determining where a company should not write more business.

Rating agencies also quickly discovered the models and incorporated the loss estimates into their rating criteria. This was another major push toward the acceptance of catastrophe models.

As companies became more accustomed to using models to answer the question of how high is high, another question began to be asked: If a model can tell me how risky my portfolio is, can it tell me how to improve my writings? Where shouldnt I be writing? Where can I write more business without assuming much more risk?

In the last few years, models have been used by some companies in these kinds of portfolio optimization analyses. These analyses have enabled insurers to carefully control growth to levels that they can manage internally or through carefully thought-out risk transfer strategies. Indeed, these models are now being used in developing entirely new business plans taking into account new marketing plans and target markets.

Until fairly recently, catastrophe models have focused on natural perils and property business. After the events of September 11, computer models are now being asked to branch out into relatively new territoryestimating loss potential for non-property lines from both natural perils and human peril, i.e., terrorism.

The major modelers had already been working on workers compensation models for the peril of earthquake. However, it was eerily reminiscent of the days prior to catastrophe modeling: reinsurance was cheap and nobody collected the data necessary for modeling.

Due to these problems, the workers compensation models had not yet gathered much support. And without support, development was relatively slow. Sept. 11 changed that viewpoint, however, much as Hurricane Andrew revolutionized the natural perils models 10 years ago. Today, there is extraordinary demand, and insurers are expending great effort and resources to capture the data required to make these new models useful.

Insurers are looking for software that can offer them some idea of frequency and severity of this newly recognized peril. Insurers now want to know how high is high in terms of terrorism. Modeling terrorism presents many challenges, such as what lines of business may be affected and correlated, what types of events are possible, and what are the probabilities of various types of events happening?

In the case of frequency, software developers differ on the accuracy of probabilistic models. Because of the human element, many experts say predicting a terrorist act is pure guesswork. However, others with access to terrorist manuals and communications note that terrorist groups do follow patterns and procedures which, over time, can be used to assign risk values to targets.

In recent months, two general methodologies have evolved. One is based on the Delphi Method developed and used by the Rand Corp. in nuclear-era war gaming. Educated guesses on targets by informed terrorism experts are assigned a numerical value and the potential losses are calculated.

The other method, also developed during the Cold War, is based on game theory, and takes a more real-time, intuitive, action/reaction approach. Both have been developed with the help of terrorism experts from the Department of Defense, the FBI and the CIA.

Both methods analyze and assign a value to threats posed by domestic extremists, state-sponsored terrorist organizations, and international terror groups like al-Qaida. Both also take into account a range of weapons and severity of possible attack modes: fuel-laden airliners, conventional bombs, big car bombs, as well as chemical, biological, nuclear and radiological threats and the effect they would have on a geographic area.

Because of the symbolism attached to past terrorist attacks, both methods pay special attention to landmarks and the buildings that surround them. Thus far, more than 300,000 potential terror attack targets ranging from national monuments to town halls have been cataloged.

Terrorism presents new challenges to the development of useful catastrophic risk assessment software. The industry is up to the challenge, however, and though we are still in the early stages and there is much learning to be done, models promise to play an ever-expanding role in estimating future catastrophic losses, from both natural and human perils.

Paul Budde is vice president of the Benfield Group, based in London.

Reproduced from National Underwriter Property & Casualty/Risk & Benefits Management Edition, October 14, 2002. Copyright 2002 by The National Underwriter Company in the serial publication. All rights reserved.Copyright in this article as an independent work may be held by the author.