Tuesday, October 8, 2013

"Model-free Forecasting"?

Wouldn't that be something like language-free writing?

Each day I receive dozens of messages from SSRN, alerting me to new working papers in Law, Economics, and Political Science. A couple times a week, those announcements are supplemented by others from PNAS, NBER and others. This morning, I happened across a paper in the PNAS Early Edition purporting to be about "model-free forecasting." I didn't read the article, but I am puzzled by the phrase, wondering what kind of forecasting could possible be "model-free."

With the possible exception of tiny infants (who probably do not engage in much forecasting of future conditions) or those whose brains are seriously damaged, we all operate with mental models about the external world, models which are largely formed by the (for lack of a better word) cultures in which our brains and bodies develop. Those models not only help us conform to the social norms of the societies to which we belong; without them, it would be literally impossible for us to make sense of the external world or even function (that is to say, live) as part of it. In this basic sense of the word, even animals can be said to engage in mental modeling, albeit without the kind of self-reflection and Bayesian updating of which humans are capable.

Perhaps I should read the article to see what the authors are actually on about. I find it hard to believe that they could really believe in anything as absurd as "model-free" forecasting (I would say the same about claims of "value-free" or "theory-free" forecasting). Perhaps they just mean that forecasting may be free of explicit models, which at least is conceivable.

No comments:

Post a Comment

I actively moderate comments for spam, advertisements, and abusive or offensive language.