it the most deadly Atlantic windstorm since 1780. In September, Hurricane Georges wreaked $3 billion in insured losses when it tore into Puerto Rico and the US Gulf Coast. The hyperactive 1998 season capped a record-setting 4-year span in which 33 hurricanes roamed the Atlantic. Insurance industry participants in the Risk Prediction Initiative (RPI) are now teaming with leading scientists to explore how their businesses can best respond to the kinds of climate variability embodied by the recent upswing in Atlantic hurricanes.

Begun in 1994 at the Bermuda Biological Station for Research, the RPI is a science-business partnership that helps businesses better understand, assess, and manage all types of climate-related risks. The RPI's continuing goal is to make the science of climate prediction understandable, usable, and relevant to the global insurance industry.

According to Dr. Tony Knap, RPI co-founder and Biological Station Director, RPI participants use a series of workshops to steer the program. At the workshops, insurers and invited scientists identify new directions for insurance-relevant climate research. The RPI uses funds contributed by its industry partners to support this research, and creates communication tools that translate the research results into a form that insurers can use.

The RPI has now disbursed more than $1 million in research funding contributed by its sponsors to climate scientists around the world. Because hurricanes cause most of the losses faced by catastrophe insurers, better understanding the hurricane hazard remains a focus of RPI research.

A current RPI research goal is to develop a public, catastrophe risk model.

"Cat'' models are software tools that insurers use to assess the risks that natural hazards pose to insured property.

Risk Prediction Initiative going public Before these tools became widely available, most insurance companies relied on past loss data to predict future losses from natural catastrophes. This actuarial approach works well for most types of insurance, such as automotive coverage, as loss patterns are unlikely to change significantly or quickly.

However, the number of hurricanes and the magnitude of the insured losses they cause change significantly through time. For instance, the average annual number of hurricanes in the Atlantic is five. But last year, the Atlantic spawned 14 named storms and 10 hurricanes. Seven of these storms hit the US, more than twice the average. The resulting damages helped to push insured catastrophe losses in 1998 to $10 billion, nearly four times higher than the $2.6 billion lost in 1997. Insured property losses during 1998 are surpassed only by the $23-billion loss in 1992-the year of Hurricanes Andrew and Iniki-and the $17-billion loss in 1994, when the Northridge earthquake hit California.

Hurricane activity also varies on longer time-scales. The recent busy spell in the Atlantic resembles another active period from the 1940s through the mid-1960s.

Relying solely on past averages to predict future hurricane activity thus provides a challenge to insurers. "Cat'' models help insurers face this challenge. The models use statistical techniques to expand the short historical hurricane record into a database with thousands of simulated storms.

RPI update The business of selling risk models to insurance companies has grown over the last few years. Why then does the RPI see a need to produce a risk model of its own? The reason, says RPI Science Program Manager Dr. Rick Murnane, is due to public suspicion of commercial risk models.

"A problem with commercial risk models,'' he says, "is that many people see them as black boxes.'' The risk model that the RPI plans to create is not designed to compete directly with the commercial models, says Dr. Murnane. For one thing, it will provide only one part of a full risk model: the component that calculates the probability of hurricane landfall. The RPI model will likely not provide the engineering data needed to estimate structural damage given a certain wind speed. Nor will it account for stipulations in an insurance policy, such as whether it covers both structural damage and damage to contents.

"The advantage of the RPI model,'' says Dr. Murnane, "is that it's public.

This means that the computer code on which the model is based is freely available." Dr. Murnane believes that the main use of the RPI model will be "to serve as a benchmark against which the commercial models can be judged. That way, if the public and commercial model differ, model users will be able to ask the modeling companies to explain the differences.'' Another advantage of the RPI model is that it incorporates data about the probability of prehistoric hurricanes. Such data are now becoming available via RPI funding of proxy studies.

They made it happen: Bermuda Insurance Symposium IV Committee members Barry Brewer, Suzie McKeegan, Carla Kaufman, chairman Robin Specer-Arscott, Nigel Clark and Madeline Joell MP. Missing: Kathryn McIntrye, Robert Mulderig and Stanley Lee.

CONFERENCE CON