Arnold Kling  

Macroeconometrics: The Lost History

The Stimulus and Black Swans... "The President Believes"...

A first version is here. I was thinking of submitting something like this to the AEA's new Journal Macroeconomics. Suggestions for improvement are welcome.

Comments and Sharing

COMMENTS (12 to date)
Orlando Roncesvalles writes:

Why are the disputes so "bitter"? I suspect it's a case of the young Turks thinking that if the old fogies can't do the new math, they have no business criticizing the new modern macro. And what is the new math? Apparently it is the Markov chains, the Bayesian approach, etc. Like you, I've tried to make sense of their papers. I'm no math geek, but I suspect I'm smart enough to think they are simply pulling a "tribal dance" of a cabal. I've decided there's no point trying to convince them they have the burden of convincing the rest of the world that they have the better mousetrap. In law, the plaintiff has the burden of proof. In science, the new kid has the same. In both cases, the outside reality (called evidence aliunde by the law books) is still the final arbiter, not their priors nor theory alone.

Greg Ransom writes:

Thanks for writing this.

pushmedia1 writes:

Professor, you're criticism of structural models IS the Lucas Critque.

You mis-identify the Lucas Critique in your paper. Its not a refutation of adaptive expectations; rather its your observation that structural changes happen and relationships between macro variables change. This implies macro models should be built around constant, or relatively constant, deep parameters (like risk preferences or whatever).

sohaib writes:

Your writing is a breath of fresh air. I'm an undergrad and I frequent a large number of economics blogs. This post and your posts on lessons in macroeconomics have been inspiring. It saddens me that none of my professors of macroeconomics classes or econometrics ever talk about the unbelievable amount of uncertainty in macro models. And I go to a pretty good school.

dearieme writes:

"It saddens me that none of my professors ... ever talk about the unbelievable amount of uncertainty in macro models.

"And I go to a pretty good school."

Which of those is true?

Lee Kelly writes:

Basic logic lesson.

Let P be the conjunction of a macroeonomic theory with some initial conditions, and let Q be some consequence of P. Suppose that P is true, and infer that Q is true using modus ponens. We can write our deduction as follows:

A. P → Q, P ⊢ Q
If A is true, then the following deduction is also valid:
B. P → Q, P ⊢ P, Q
But here we get an interesting result, because B's conclusions are semantically equivalent to its premises. In other words, B is not just a deduction, but also an equation. This should be clear when the validity of the following deduction is understood:
C. Q ⊢ P → Q
A conditional formula can only be false when its consequent is false; under all other semantic interpretations it is true. It follows that if Q is a premise, and therefore assumed to be true, then P → Q cannot possibly be false, since that would require Q to also be false. Therefore, C is valid.

The problem. Most people, like economists, would try and argue for the truth of Q using P, that is, using their macroeconomic theories, models, hypotheses, etc. in conjunction with some initial conditions. But according to the above result, the following two deductions are semantically equivalent (i.e. alternative ways of expressing the same propositions):

1. P → Q, P ⊢ Q
2. P, Q ⊢ Q

Nobody would accept the premises in the second deduction to provide any reason whatever to accept the conclusion. Yet most people, like economists, do expect exactly that when they present an argument in the form of the first deduction. But both are actually equivalent, and nobody who does not accept the second should accept the first either.

This result is very general. It does not only apply to marcoeconomics, but any field of study whatever.

The fact that other fields of study are so successful despite the above only means that most of them are not actually doing what they think they are doing--they can't be, because it's not logically possible. Disputes and controversies are being settled in other ways instead.

pushmedia1 writes:

Lee Kelly, are you being ironic?

Let P="Macroeconomists don't understand logic" and
Q="Macroeconomists argue as described by LK".

Its seems like you're arguing for Q using P.

I say show me evidence for Q.

Jeff Hallman writes:


This was my comment on your previous post about macro econ. I wonder if you read it, and what you think about it.

There are a couple of ways you can approach macro econometrics. One is to think of economic theory as implying restrictions on an otherwise unconstrained vector autoregression (VAR). You estimate a VAR with and without the restrictions, calculate some statistics, and discover whether or not the data reject the restrictions. If they do, your theory is clearly wrong. If not, it doesn't mean your theory is correct, but at least it isn't obviously wrong.

If the restrictions aren't rejected, you might go further and ask whether or not imposing them improves predictions from the model. If not, in what sense is your theory knowledge? You can quantify "improvements" by comparing information criteria computed with in-sample data (such as the weak AIC or stronger BIC), or by comparing out-of-sample forecast errors. The latter is a better test.

Clive Granger and others showed in the early 1970's that simple univariate time series models were better at forecasting inflation and income than the big, hundreds-of-equations economic models. Not long afterward, Robert Litterman and Chris Simms showed that a Bayesian VAR (BVAR) with random walk priors outperformed both the big economic models and univariate time series models.

Most "structural" modeling today starts with a theory model, derives from it some testable restrictions on a VAR, and then tries to see if the data reject the restrictions. But for macro variables, the small sample sizes usually mean that the VAR with or without the theory restrictions does not forecast as well as a BVAR would. So you have an ordering:

VAR < VAR with theory < BVAR without theory

Just how that is supposed to convince anyone that macro is scientifically worthwhile is beyond me.

It seems the only thing macro theory is actually good for is helping you tell internally consistent stories about how you think the economy works. Keynesian stories don't even meet that requirement. But just because a story is internally consistent is no reason to believe it.

Lee Kelly writes:


I am not being ironic, although I understand why others might think that.

The mistake is in thinking that I am arguing for anything (to whatever extent I am, it is not in the way which is more common). Instead, I am merely proposing some premises and exploring their consequences. I think my above argument is true, and others who cannot fault my reasoning may agree. But I do not suppose that my own argument can achieve exactly what my own argument deems impossible. This, however, does not preclude the possibility that it is true.

"Deep" or "constant" parameters are probably an article of faith. That's why they're called parameters. How do you prove their depth or constancy?

von Mises once said something to the effect that "wishing" is the "father of faith."

fundamentalist writes:

I really liked Dr. Kling's paper and don't have anything to add, but I was wondering about the possibility of testing models by having them use similar data sets and make out of sample forecasts. Compare the results with RMS or something similar. I have seen Fair's macro model compared to others this way. Of course, someone would have to come up with an Austrian macro model, but that shouldn't be hard. What are the pitfalls to this approach that I don't see?

Greg Ransom writes:

Hayek explains by macroeconometrics must fail here:

Comments for this entry have been closed
Return to top