The poverty of rationality
Steve Williamson has written a much longer critique of Zombie Economics. It’s a lot more temperate in tone than the blog post I criticised here, and there are some valid points. Nevertheless, the new version exhibits the same fundamental confusion I pointed out last time, trying to claim that rationality assumptions are both important and unfalsifiable.
I’m criticising it again because, in making this mistake, Williamson is not exactly Robinson Crusoe. The same confusion is evident among a great many economists, and even more among proponents of rational choice models in political science and other social sciences. This, despite the fact that the key error was skewered by William Hazlitt nearly two centuries ago, writing on self-love and benevolence.
Before starting, I’ll make a brief, purely mathematical point. Any consistent pattern of choice among objects (of any kind) that we can observe, can be represented as optimization, that is, as the maximization of a function. The classic version of this result was proved by Cantor, who gave us the modern idea of a function as a mapping between sets, and cleared up a lot of the technical puzzles about continuity and so on. Even choices that are inconsistent in various ways can be represented by more general notions of optimization. So, it makes no sense either to claim (as a lot of economists do) that the fact that we can represent action as the maximization of some “objective” function proves anything positive about the way people think or to object (as a lot of non-economists do) to representing choices in terms of optimization. To (ab)use an apposite quote – this isn’t class warfare, it’s math.
Williamson invokes the Cantor result in support of rationality assumptions saying
If the phenomenon can be described, and we can find some regularity in it, then it can also be described as the outcome of rational behavior. Behavior looks random only when one does not have a theory to make sense of it, and explaining it as the result of rational behavior is literally what we mean by “making sense of what we are seeing.
and in response to my criticisms, that I offer
”“ the usual list of complaints, for which there are standard defenses. (i) We can observe economic agents behaving irrationally, so what is all this rational agent stuff about? Answer: If you think you are observing irrational behavior, you just have the wrong model. Think harder.
So far, our disagreement is essentially semantic. Williamson wants to use the term ‘rational’ to describe optimization with respect to any function whatsover. In this includes the kind of behavior displayed by an agent (not necessarily an individual) in a model, any model. So, I can present whatever model I like, and the behavior in it is necessarily rational, and any rational behavior involves optimising something or other. Provided my model exhibits some regularity in the behavior of agents, they must be optimising something – working out what is the kind of problem normally given to sharp grad students.
By contrast, I normally use‘rational’ to refer to the kind of behavior found in the simplest form of the DSGE models: farsighted, and purely egoistic, agents maximizing the expected utility of stochastic consumption streams over time. Most of the time, at least when no-one is challenging them on it, this is the way neoclassical economists use the term themselves.
And this applies to Williamson himself. At the beginning of his defense of modern macro he writes “A second key principle in the post-1970 macroeconomic research program is adherence to optimization – a key organizing principle in all of economics.”
But we’ve already seen that, according to Williamson, any possible behavior involves optimization. That includes the behavior described by Keynesian macro models, not to mention Marxist, institutionalist and even Freudian models. So, this “key principle” is, on Williamson’s account, entirely devoid of content.
In reality of course, Williamson wants to have his cake and eat it. Most of the time he wants to help himself to the strong implications of rationality as represented in standard micro texts, and to demand that macro be built on this basis. But, when this model is challenged on empirical grounds, he retreats to a concept of rationality that is tautologically true. This is a classic example of John H’s “two-step of terrific triviality”.
To quote my own favorite bon mot on this
most rational actor models assume that “rationality” can be represented as “maximization of self-interest”. This assumption is either false or vacuous. Those committed to egoistic rationality tend, when challenged, to oscillate between the two definitions, in much the manner of the function sin (1/x) as x approaches zero.’
The one appeal to empirical evidence in his entire defence of modern macro is, unsurprisingly, the observation that the Keynesian models of the 1960s ran into big problems and that, at least arguably, this reflected the fact that they failed to take adequate account of the way in which workers and firms would rationally respond to higher inflation. Of course, I described that process in Zombie Economics and went on to show how the demand for rational microfoundations led to DSGE macro which failed in its turn. It’s in responding to this failure that Williamson relies on the non-falsifiability of his preferred group of models.
More on similar lines from Noah Smith
fn1. Our profession’s favorite representative individual