Simplify!

The following may or may not be factually accurate. It all happened a long time ago. But it is absolutely 100% correct in spirit.

Twenty or so years ago I was browsing through the library of Imperial College, London, when I happened upon a book called something like The Treasury’s Model of the UK Economy. It was about one inch thick and full of difference equations. Seven hundred and seventy of them, one for each of 770 incredibly important economic variables. There was an equation for the rate of inflation, one for the dollar-sterling exchange rate, others for each of the short-term and long-term interest rates, there was the price of fish, etc. etc. (The last one I made up. I hope.) Could that be a good model with reliable forecasts?

[Hint: How good are economic forecasts generally?]

Consider how many parameters must be needed in such a model, every one impossible to measure accurately, every one unstable. I can’t remember whether these were linear or non-linear difference equations, but every undergrad mathematician knows that you can get chaos with a single non-linear difference equation so think of the output you might get from 770.

Putting myself in the mind of the Treasury economists I think “Hmm, maybe the results of the model are so bad that we need an extra variable. Yes, that’s it, if we can find the 771st equation then the model will finally be perfect.”

No, gentlemen of the Treasury, that is not right. What you want to do is throw away all but the half dozen most important equations and then accept the inevitable, that the results won’t be perfect.

A short distance away on the same shelf was the model of the Venezuelan economy. This was a much thinner book with a mere 160 equations. Again I can imagine the Venezuelan economists saying to each other, “Amigos, one day we too will have as many equations as those British cabrones, no?” No, what you want to do is strip down the 160 equations you’ve got to the most important. In Venezuela maybe it’s just a few equations, for the price of oil, inflation, and maybe how much it costs to buy a politician.

We don’t need more complex economics models. Nor do we need that fourteenth stochastic variable in finance. We need simplicity and robustness. We need to accept that the models of human behaviour will never be perfect. We need to accept all that, and then build in a nice safety margin in our forecasts, prices and measures of risk.