**John Cochrane** has an excellent article on Bob Lucas’ contributions to macroeconomics. Here is a point that I also keep coming back to:

The Fed often asks economists for advice, “should we raise the funds rate?” The Post Lucas macroeconomists reply that this is not a well-posed question. It’s like saying “should we cry wolf?” The right question is, should we start following a rule, a scheme, should we create an institution that steadily and reliably raises interest rates in a situation like the current situation? Decisions do not live in isolation. They create expectations and reputations. Needless to say, this fundamental reality has not penetrated political institutions. And this answer (which I’ve tried in Fed advisory meetings) leads to glassy eyes. John Taylor’s Rule has been making progress for 30 years trying to bridge this conceptual gap, with some success.

Lucas is the economist who launched the revolution of rational expectations. Much of the skepticism about “rational expectations” stems from a lack of understanding of what the assumption actually means. Here is Cochrane:

But “rational expectations” are really just a condition of humility. It says, don’t write models where the model predictions are different from the model expectations. If you do, if your model is correct, people will read the model and understand, and the model won’t work anymore. Don’t assume that your economist (or Fed Chairman) is so much less behavioral than the people in your model. Do not base your policy on an attempt to deceive small farmers again and again. He’s not saying that people are big, super-rational calculating machines. It just says that they eventually understand.

I’d like to illustrate the problem with a hypothetical example involving a large glass jar of candy. You may recall a famous example cited in the “wisdom of crowds” literature, where an MBA class was asked to estimate the number of candies in a large jar. Most of the guesses were off the mark, but the median guess was surprisingly close, say 1% or 2%. In this case, how would I model the audience’s candy estimates?

The least bad approach might be to estimate the actual number of candies and then assume that this number was also the audience’s estimate. This approach wouldn’t work perfectly, but it’s hard to see an alternative that would be better. Would you assume that the average guess is only 60% true? How about 150%? If yes, why?

Now suppose I ask a mathematician how many 9mm long and 6mm wide ellipsoids can fit in a cylinder 8 inches high and 5 inches in diameter. The mathematician provides an equation that looks quite complicated to the average person. Does it make sense to assume that the average person uses this equation to estimate the number of candies? Obviously not. But this equation gives you a good estimate of the actual number of candies, and while we have no reason to assume that the audience estimates are biased, it also provides the best model for the audience estimate.

Rational expectations models in macroeconomics are often full of frightening equations. The modeler then assumes that the public’s forecasts of variables such as inflation are “consistent” with the model. So, if the model is predicting 7% inflation, we don’t assume the public is predicting 3% or 13% inflation – why would we? We assume that the public also expects inflation of 7%. This may not be correct, but it seems to be the least bad approach unless we have precise knowledge that the audience overestimates or underestimates the variable in question. (Unfortunately, this is difficult to test, because inflation is ill-defined. Public estimates that appear in places like the Michigan survey likely reflect a definition of inflation that does not include hedonic adjustments, and is therefore somewhat higher than the government’s estimate of inflation.)

Many people dismiss rational expectations because it seems to suggest that the public is made up of super-intelligent calculating machines. But that’s not what it means at all. Bennett McCallum suggested that it would have been better to call the concept “consistent expectations”. The demand is actually quite modest. All the rational expectations hypothesis says is that if your model specifically implies that X is true; don’t assume that the public believes X is wrong, at least not without evidence to support that claim.