12 August 2009

Karen Clark on Cat Modeling

Karen Clark, founder and formerly of AIR (a cat modeling firm) and now president of Karen Clark and Co., says some smart things about the use of cat models: (thanks BP)
"Using a range, rather than a point estimate, is very important," Clark said. "The models give good guidance as to what a good range is, but trying to pinpoint a number within that range is a fruitless exercise, and in fact, what leads to a lot of the bad business decisions, because every time that number changes -- which they always do, when your exposures change or the models change -- you have to radically change your business decision, and that just doesn't make sense."

"A lot of times the model updates are just moving that PML around within the band of uncertainty. It may be scary, but it's a fact, and that's really what companies have to get used to. They have to get used to managing risk with the uncertainty, looking at the ranges of large losses rather than expecting a scientist or model will give them the answer, or the number. That is not possible"
My own research reinforces these views. I have a paper through peer review and now in press with the journal Environmental Hazards for October, 2009 publication. Its title is:

United States Hurricane Landfalls and Damages:
Can One to Five Year Predictions Beat Climatology?

The paper begins as follows:

The answer to the question posed in the subtitle is, unfortunately, no. This paper explains why skillful prediction of United States hurricane landfalls and damages is not possible in the short term, defined here as a time period of one to five years. A “skillful” prediction is one that improves upon expectations derived from the statistics of the long-term historical record.

If you’d like an advance copy, please send me an email, pielke@colorado.edu, and I'll get you a copy of the corrected page proofs as soon as I get them.