Monday, October 06, 2008

When climate models are treated like reality

As followers of the anthropogenic catastrophic climate scam will know, a great deal of the alarmist "proof" rests on computer models. Which is unfortunate because, as Climate Skeptic pithily reminds us, computer modelling isn't all that it's cracked up to be.
On April 28, 2004, the SEC made a significant change in policy in the regulation of large investment banks. On that day, they "decided to allow the five largest US investment banks to substitute advanced mathematical risk models for traditional capital requirements." Al Gore has touted the "success" of such models as a reason to feel confident that computer models can accurately predict long-term climate trends.

But it turns out, as everyone is discovering this week, that computer models are not reality. In fact, computer models are extraordinarily sensitive to their inputs, and small changes in their inputs, or the narrowing of models to ignore certain factors, can make them worse than useless. Computer models are also very easy to force to a preferred conclusion.

Still, the alarmists were partially correct: the end of the world is nigh: they just chose the wrong sector...

4 comments:

Anonny Mouse said...

Computer models are just that: models. They can be made completely insensitive to their inputs, if you choose. All they do is follow a logical set of principals - they allow you to see the perfectly logical conclusions of a set of assumptions. The thing is: you need to get those assumptions right first. This is why you should not trust models in the long term - because effects deemed to small to account for in the testing phase can lead to large changes given sufficient iterations.

However, there is nothing wrong with computer models, per se. Just the way they are used. Like knives, or regulation, or words, or whatever. Please don't criticise inanimate tools! Criticise the animate ones instead...

D.S.Harford said...

It's the same as that P40 model on my desk. It's a model not a real airplane.

passer by said...

..and let not forget some of the same principles are behind some of the risk models used in financial institutions to such stunning effect.

AMcGuinn said...

There is a "weakest link" problem with these models. I worked as a programmer on the firmwide risk model of a now-defunct investment bank. I worked mostly on the equity market risk part of the model, which was fairly sophisticated. But however good it was, it wouldn't have helped when mortgage risk wiped out the company.

Similarly, a climate model only has to get one thing wrong to be utterly worthless, however good the rest of it is.