Newsletter Sign-up

By MATTHEW BRODSKY, senior editor/Web editor

San Francisco-based Andy Thompson, leader of the risk consulting practice at global engineering and consulting firm Arup, wants to know when the last time a catastrophe modeler actually designed and built a structure. He wants to know if they know how contractors in Japan build differently than contractors in Jersey. When's the last time they came up with a mitigation strategy to quake-proof a facility, then actually put those braces and other retrofits in place.

Catastrophe modelers are software designers and scientists, not engineers, is his point.

Their products serve a purpose for insurers and reinsurers with thousands of properties in their books, but they don't fulfill the needs for corporate risk managers with one, a dozen or hundreds of properties on the ground.

Ask modelers about how well their software captures the details of one of those structures. Thompson recalls how he recently investigated a coal manufacturing plant whose greatest risk was one oil pipe that was braced improperly. That pipe sent oil to the generator, and had it gone down, the generator would have gone down, and had that generator gone down, the whole plant would be kaput for three months.

"It is unlikely that a catastrophe model could capture that," said Thompson, leading us along his line of thinking like a train on its tracks.

"These models cannot capture the details that you need," agrees Tom Chan, CEO of Global Risk Miyamoto Inc. out of Lafayette, Calif.

Just a very simple thing such as a window attachment or the roofing attachment, the shape of eaves or overhangs, whether your factory's equipment is three feet above grade or below grade--"these are not details that you can capture in a computer model," said Chan. For that last one--grading--the modelers don't even ask that question, he added.

One of the biggest technical reasons that catastrophe models are not the ultimate tool for corporate risk managers is a principle called the law of large numbers. Simply put, models are based on so-called probabilistic fragility curves--the relationship between a given structure and the intensity of a hazard that results in a loss--and these curves come about by the modelers averaging out a wealth of historic data. But as Thompson explained, this data is scattered like a "shotgun" blast, and models can carve a smooth curve out of them simply because there is so many data points to support that statistical simplification.

These fragility curves work when insurers and reinsurers input their books of business into the models because off the vast number of properties. The law of large numbers kicks in for them.

"But for individual corporations with small portfolios, the chances are slim that the probabilistic fragility-curve approach, without appropriate site-specific assessment by qualified engineers who are familiar with local design and construction practice, will provide you with accurate data," Thompson explained. "It is like trying to guess how a specific neighborhood, or worse an individual, will vote based on countywide voting trends. People are trying to make decisions based on a completely skewed level of information accuracy."

"When you're a risk manager and you only have 20 or 50 properties, you are not really getting the larger numbers to average out, factor out, your risk," said Doug Frazier, a senior adviser to Arup and former chairman, CEO and co-founder of EQE.

Put simply, the models work by drawing a line among the "scatter," said Chan, who had been a partner with EQE in the past. And it's not that they don't work; it's just the more unique, the more complicated a facility that risk managers have on their books, the less likely models work for them.

Chan said models are really good, say, for reinsurers and Lloyd's guys who can't send somebody out to everything they're underwriting.

"For kind of a first pass review, models are great," he said. They'll define the risks for you, such as the return period of hazard in a particular geographic area.

"As far as what's going to happen to your site when that event occurs, that is where the models kind of fall down," he said. "Can you imagine a refinery or a plastics manufacturing plant where there's only 10 of them in a whole area and none of them have ever been through a Cat-4 (hurricane) before?"

MODELERS REBUTTAL

Modelers will consent that for highly complex facilities, their basic software programs might not cut it.

"Certainly for anything that is unusual or highly engineered, people do need to be particularly careful about what input are going into the models," said Matthew Grant, global client development at Risk Management Solutions Inc. "A generic CAT model won't be accurate for a corporate risk manager."

Said Tom Larsen, senior vice president of Eqecat Inc., your typical CAT model won't be able to pick up individual deficiencies at one site.

"Every refinery is a work of art," he said, while every auto plant is continually updated and retooled. "It is very challenging for a model."

But Tom Larsen, senior vice president at Eqecat Inc., explained how the modeler's parent company, ABSG Consulting, can send in engineers at such facilities and produce what they call a "PML Plus," which can then be inputted into a portfolio using the models.

Grant at RMS mentioned also how more detailed models are available that can provide better granularity to more fully enter a facility's characteristics into the program and that can give credit for risk mitigation. And that his firm can create "bespoke vulnerability curves" for corporate clients.

"We can basically design curves to show how well they are hardened or mitigated against CATs, or if one building is really better built than the norm," he said.

As many as 60 percent to 70 percent of his clients use these more detailed models versus the generic RMS catastrophe models, Grant estimated.

December 1, 2008

Copyright 2008© LRP Publications