September 15, 2016
The Microfoundations Hoax
https://meansquarederrors.blogspot.it/2016/09/the-microfoundations-hoax.html
Demolition work on the rotten edifice of "modern macroeconomics" continues apace.
The emperor, it turns out, is not merely without clothes. Upon closer
inspection, he appears to be simply an empty cardboard box with the
words "Emperor Inside" scrawled across its surface in felt-tip pen.
Paul Romer's devastating critique really
deserves to be the final word on the matter. But even Paul leaves one
stone unturned, an element of modern macro so transparently
intellectually dishonest that it may properly be termed a hoax: its
so-called "microfoundations."
No modern macro model is complete without a pean to the virtues of its own microfoundations. It seems that the word "microfoundations" is not allowed to appear unaccompanied by at least one self-congratulatory adjective -- "careful," "well-specified," even (shudder) "rigorous." But, as diligent readers of George Orwell will recall, war is not peace, freedom is not slavery, ignorance is not strength, and representative agent models are not rigorously microfounded.
But let's back up a step. What is this "microfoundations" business anyway, and why should anyone not currently seeking a tenure-track appointment in econ care even a tiny bit? Here's a short version of the very long story:
Forty years ago, the name of the game in macroeconomics wasn't theory at all; it was forecasting. And it wasn't particularly successful. In retrospect, the lack of success isn't surprising. Models were typically estimated by running regressions on a handful aggregate data series representing the experience of a single country over a very short (and rather placid) period of time (1). Moreover, in macroeconomic data, everything is pretty highly correlated with everything else. So you could put pretty much whatever you liked into your regressions and get a really good fit with in-sample data. Then history would happen, new data would arrive to contradict the model's predictions, and you'd either re-estimate the model (and watch the coefficients bounce around more or less at random) or you'd declare the latest data to be some kind of special case and "adjust" for it.
So when critics denigrated the models of the early '70's as "ad hoc," they had a pretty serious point.
But what was the solution to all of this ad hoc-ery? Where were we to look for the all-important virtue of discipline? Ideally, in social science as in physical science, the source of discipline is data. If you want to tell the difference between a true theory and a false one, you ask reality to settle the question. But that was the heart of the problem: with so little data, all the models looked equally good in-sample, and no model looked especially good out-of-sample. Discipline, if there was to be any, would have to come from theory instead. And "microfoundations" was put forward as one form of theoretical discipline (2).
The idea certainly sounded good: rather than simply making up relationships between aggregate variables like interest rates, output, unemployment, and inflation, we should show how those relationships arise from the behavior of individuals. Or, failing that, we should at least restrict the relationships in our macro models to those which are consistent with our understanding of individual behavior. For surely our standard assumptions about individual behavior (basically: people do the best they can under the circumstances they find themselves in) must imply restrictions on how the system behaves in the aggregate.
Sadly, this intellectual bet was lost even before it was placed. If we take Lucas (1976) as the beginning of the microfoundations movement, we may note with some puzzlement that the premise was mathematically proven false two years earlier, in Debreu (1974) and Mantel (1974).
It is sometimes said that modern macro suffers from too much math. But the problem is not "too much," but rather that its use of math is strangely selective. In particular, the idea that microfoundations per se can impose "discipline" on aggregate models represents deliberate ignorance of one of the most important results in mathematical economics. Debreu, Mantel, and Hugo Sonnenschein had shown conclusively that, for any macro behavior you care to invent, there exists a set of classically well-behaved rational utility optimizing agents that will, collectively, exhibit the desired behavior.
Put another way, the classical assumptions about individual behavior impose no limits whatsoever on the behavior of aggregate models. Oops.
The specious pretense, then, that one's preferred models are "better" microfounded than the competition (when, in fact, all models are equally micro-foundable) is part one of the microfoundations hoax. But it gets better. (Or worse, depending.)
The models which preen themselves most ostentatiously in their "rigorous" microfoundations are invariably based on so-called "representative agents." Now, every economist, at some point in their first year of graduate school, learns the mathematically necessary conditions for the existence of a representative agent corresponding to a collection of individual agents. In the lingo of the field, we say that a representative agent exists only if (a) all of the individual agents have identical preferences; and (b) if those preferences are [jargon] quasi-homothetic [/jargon]. "Quasi-homothetic" is a fancy way of saying that 10,000 households whose resources added together equal those of Bill Gates will buy exactly what Bill Gates will buy.
Neither of these conditions are remotely plausible, and nobody believes that they are true, including macroeconomists. And if either of those conditions fails to hold, an economy which behaves as if it posessed a representative agent cannot be derived from classical microeconomic foundations.
Let that sink in for a moment: of all the macro models that have been floated over the last century or so, only the so-called microfounded models are completely and demonstrably incompatible with classical microfoundations. And this is (or should be) obvious to anyone who didn't sleep through the first year of graduate microeconomic theory.
So when I call "microfoundations" a hoax, I'm not kidding around. The only question is, what proportion of macroeconomists have perpetrated this hoax upon themselves, and what proportion has known this all along.
(1) Bear in mind that the entire apparatus for gathering and reporting economic statistics in the U.S. was basically created in 1947. Pity the macroeconomist circa 1965, trying to understand the most complex social system in history based on n < 20 observations. Yikes.
(2) "Rational expectations" was another. The "rational expectations revolution" business probably deserves a separate post. For now, just know that the name is a kind of mathematical pun, and that neither "rational" nor "expectations" means what you probably think they mean. Aren't we clever?
No modern macro model is complete without a pean to the virtues of its own microfoundations. It seems that the word "microfoundations" is not allowed to appear unaccompanied by at least one self-congratulatory adjective -- "careful," "well-specified," even (shudder) "rigorous." But, as diligent readers of George Orwell will recall, war is not peace, freedom is not slavery, ignorance is not strength, and representative agent models are not rigorously microfounded.
But let's back up a step. What is this "microfoundations" business anyway, and why should anyone not currently seeking a tenure-track appointment in econ care even a tiny bit? Here's a short version of the very long story:
Forty years ago, the name of the game in macroeconomics wasn't theory at all; it was forecasting. And it wasn't particularly successful. In retrospect, the lack of success isn't surprising. Models were typically estimated by running regressions on a handful aggregate data series representing the experience of a single country over a very short (and rather placid) period of time (1). Moreover, in macroeconomic data, everything is pretty highly correlated with everything else. So you could put pretty much whatever you liked into your regressions and get a really good fit with in-sample data. Then history would happen, new data would arrive to contradict the model's predictions, and you'd either re-estimate the model (and watch the coefficients bounce around more or less at random) or you'd declare the latest data to be some kind of special case and "adjust" for it.
So when critics denigrated the models of the early '70's as "ad hoc," they had a pretty serious point.
But what was the solution to all of this ad hoc-ery? Where were we to look for the all-important virtue of discipline? Ideally, in social science as in physical science, the source of discipline is data. If you want to tell the difference between a true theory and a false one, you ask reality to settle the question. But that was the heart of the problem: with so little data, all the models looked equally good in-sample, and no model looked especially good out-of-sample. Discipline, if there was to be any, would have to come from theory instead. And "microfoundations" was put forward as one form of theoretical discipline (2).
The idea certainly sounded good: rather than simply making up relationships between aggregate variables like interest rates, output, unemployment, and inflation, we should show how those relationships arise from the behavior of individuals. Or, failing that, we should at least restrict the relationships in our macro models to those which are consistent with our understanding of individual behavior. For surely our standard assumptions about individual behavior (basically: people do the best they can under the circumstances they find themselves in) must imply restrictions on how the system behaves in the aggregate.
Sadly, this intellectual bet was lost even before it was placed. If we take Lucas (1976) as the beginning of the microfoundations movement, we may note with some puzzlement that the premise was mathematically proven false two years earlier, in Debreu (1974) and Mantel (1974).
It is sometimes said that modern macro suffers from too much math. But the problem is not "too much," but rather that its use of math is strangely selective. In particular, the idea that microfoundations per se can impose "discipline" on aggregate models represents deliberate ignorance of one of the most important results in mathematical economics. Debreu, Mantel, and Hugo Sonnenschein had shown conclusively that, for any macro behavior you care to invent, there exists a set of classically well-behaved rational utility optimizing agents that will, collectively, exhibit the desired behavior.
Put another way, the classical assumptions about individual behavior impose no limits whatsoever on the behavior of aggregate models. Oops.
The specious pretense, then, that one's preferred models are "better" microfounded than the competition (when, in fact, all models are equally micro-foundable) is part one of the microfoundations hoax. But it gets better. (Or worse, depending.)
The models which preen themselves most ostentatiously in their "rigorous" microfoundations are invariably based on so-called "representative agents." Now, every economist, at some point in their first year of graduate school, learns the mathematically necessary conditions for the existence of a representative agent corresponding to a collection of individual agents. In the lingo of the field, we say that a representative agent exists only if (a) all of the individual agents have identical preferences; and (b) if those preferences are [jargon] quasi-homothetic [/jargon]. "Quasi-homothetic" is a fancy way of saying that 10,000 households whose resources added together equal those of Bill Gates will buy exactly what Bill Gates will buy.
Neither of these conditions are remotely plausible, and nobody believes that they are true, including macroeconomists. And if either of those conditions fails to hold, an economy which behaves as if it posessed a representative agent cannot be derived from classical microeconomic foundations.
Let that sink in for a moment: of all the macro models that have been floated over the last century or so, only the so-called microfounded models are completely and demonstrably incompatible with classical microfoundations. And this is (or should be) obvious to anyone who didn't sleep through the first year of graduate microeconomic theory.
So when I call "microfoundations" a hoax, I'm not kidding around. The only question is, what proportion of macroeconomists have perpetrated this hoax upon themselves, and what proportion has known this all along.
(1) Bear in mind that the entire apparatus for gathering and reporting economic statistics in the U.S. was basically created in 1947. Pity the macroeconomist circa 1965, trying to understand the most complex social system in history based on n < 20 observations. Yikes.
(2) "Rational expectations" was another. The "rational expectations revolution" business probably deserves a separate post. For now, just know that the name is a kind of mathematical pun, and that neither "rational" nor "expectations" means what you probably think they mean. Aren't we clever?
Nessun commento:
Posta un commento