Thread by @vectornomist

#econtwitter I was thinking a lot about rationality, again! Most empirical evidence points out that individually we are not rational in the classical sense of the word. Individuals routinely break the consistency conditions implied by utility maximization. However,

when we go to aggregate setups, that is the study of market shares, or conditional expectations, rationality "as if" the population, not the individual, behaves as controlled by a random utility distribution is usually an OK description of behavior. In fact, most macro, IO,

end up using models of aggregate behavior, and are reasonably successful at it. I have yet to see an individual theory of behavior that explains the data as well as a simple Cobb-Douglas model explaining average shares of consumptions in many markets and experiments.

All of this is actually very reasonable, individual errors, biases and so on may "average out" and give us more predictable aggregate behavior. However, econ micro theory and macro "microfoundations" seems to have the wrong idea, in fact it has it upside down.

Take the MWG textbook, and its classical Chapter 4 on aggregation of consumers. We start with a population of rational individuals and then we show that they generically will produce average demand that is not rational! This is a shocking result, but more important it is at odds

with evidence. Most individuals are not consistent with the empirical implications of utility maximization, but in many setups random utility is a correct description of population behavior, and in particular, average demand is morel likely to respect the law of demand that

individual demand. Becker classically argued that consumer randomly (uniformly) choosing from linear budgets could produce demand that "looks rational" in the aggregate. But his insight has been seen more of as an oddity or the fact that testing rationality in the aggregate is

not meaningful. However, he is making a point, made in many sciences, that aggregation creates order when there was chaos. In stark contrast, micro theory in the late 60s was obsessed with mathematical purity rather than data driven theorization and got lost! They study the

wrong question in a way, what happens if we aggregate rational consumers instead of what kind of imperfect behavior could lead to rational behavior. More interestingly perhaps, and useful is the question of what models of aggregate behavior are enough for the purposes of Econ.

Yes, I know, Lucas critique, but there is no reason why we can have smart agents driving the behavior of a population with flexible heterogeneity and expectations, but at the same time only predictable in terms of the market share behavior. See dynamic discrete choice with random

utility. Luckily, many theorist are working on cool stuff right now on this topics, so very hopeful of what is going to come up out of this. Data-inspired theory is the best theory.

@JRehbeck @gereltecon @mlkovach1 @DrewFudenberg

@mogens_fosgerau as well cool work