26 June 2014

Carnac Is Ignorant

I've had "Applied Predictive Modeling" for a while and read through it at least a couple of times. Mentioned it a couple of times, too.

Now, the Revolution folks have posted a review, which is largely positive. But, Mr. Rickert pushes the button which irritates me about the book, and I'd assume, the cabal:
They emphasize that predictive modeling is primarily concerned with making accurate predictions and not necessarily building models that are easily interpreted.
The text contains nearly that sentence, verbatim. And it's the reason I've avoided saying much. As Your Good Mother used to teach, "if you've nothing nice to say; say nothing". Predicting from data up to time t works only if the real world is *exactly* the same at time t+1, and so on. It never is. The authors are bio-stats, more or less, so in that world God makes the rules, more or less. Not so out in finance.

But it has to be said: the reason we got The Great Recession is that quants bought the Markov principle (not the real one, of course), which is that all information about series X at time t is embedded in all observations up to time t. Which, of course, is bullshit. While it is somewhat useful in physical processes, it's bullshit when applied to human processes. Data series on human stuff is determined by policy changes over time. Decisions, and data, follow incentives. Not too surprisingly, incentives are the product of policy. Change the policy, change the incentives, and presto-chango the data takes a right turn. Thus, the whole sub-prime mess derived from loophole spelunking in law and regulation, which displayed astounding fungiblity. And was thus not predictable (and wasn't, in fact) by the quants. They didn't want to see that the rules were being gamed, so they didn't.

Read a good newspaper if you want to know where the world, or your specific concern, is headed.

No comments: