Monday, November 29, 2010

The Practice of Macroeconomics - A Few Rules

There have been three good posts recently from Paul Krugman, Arnold Kling, and Steve Horwitz on how to do macroeconomics (or in Horwitz's case, Austrian economics in general). Krugman makes the point I often do that you can't assume that all macroeconomic episodes are created equal or that they should be responded to in the same way. It's a point that Jefferson and Keynes have both made emphatically, and one that Tom Woods and Bob Murphy would do well to digest. Arnold Kling makes several points on here about identification problems in macroeconomics. He also notes the relationship between model building and a person's macroeconomic priors. I think this is very important, and a good introduction to the way that empiricism can be used to pursue truth (and the ways in which it cannot). Steve Horwitz wrote a very thoughtful post with a lot of very thoughtful comments on what Austrian economics is - emphasizing for the most part the Peter Boettke's ten tenets of Austrian economics and Steve's own concerns about the tendancy to mesh positive and normative insights.

These all got me thinking more concretely about a few thoughts/rules of my own on the practice of economics and macroeconomics in particular that I've been thinking about for a while. I think they broadly relate to these three posts insofar as they emphasize how we have to approach macroeconomics pragmatically rather than dogmatically, and how to keep our approach objective despite the obvious judgement calls and normative concerns.

Rule # 1 - The economy is complex so you cannot successfully build it up deductively. This is not to say, of course, that deductive logic is useless. It is useful - but as a tool to be employed when necessary, and not as a totalizing method. The problem with building a macroeconomics out of strict deductionism is that there are simply too many factors to account for - we are faced with what is essentially a knowledge problem. Deduction can certainly illuminate specific processes that go on in the economy, but one cannot hope to get an adequate picture because of the very real risk that the necessary and true axioms identified are insufficient for determining the system. In other words, deductive logic should not lead you astray if you do it right and if it remains circumscribed (both of which are very big "ifs"), but it will lead you astray if you make the mistake of thinking you can rely on your deductions for a sufficient picture of the macroeconomy.

Rule # 2 - Start with theory, whether it is deduced, induced, or inherited. We need some way to conceptually organize what we empirically observe, and if we want an understanding of the macroeconomy we need some framework for understanding macroeconomic processes. So you need a story or an explanation - a theory. Where you get this theory isn't terribly important as long as you understand that what you have is a theory and not a statement of exhaustive truth. Deductive logic is very useful in theory construction, of course. But induction can be a useful approach as well. If we observe, for example, that cyclical unemployment is primarily determined by movements in hiring rates rather than separation rates, that observation can be used to think up a story about why that might be the case. It's also fine to acknowledge that you're relatively new to macroeconomics, but everything that some older, wiser scholar said has the ring of truth, and so you adopt that perspective as a theoretical starting point. The key is to understand that you are not claiming a truth - you are constructing a way of understanding the world. Pitfalls still exist for deduction (I describe these above), induction (the pitfalls of induction should be abundantly obvious), and inheritance ("arguing from authority" is never good, and hero-worship is always a risk). But these problems and fallacies are only really problems if we think of theory-building as truth-claiming. It's not. And I can't emphasize this enough. Epistemological insights are useful for science, but science is not the search for "truth". It is the search for useful approximate knowledge of the world. Theories are ways of organizing knowledge and information - they are not "truth".

Rule # 3 - History is simply past behavior of the human species and as such it is essential to the scientific study of the human species - once you have a theoretical framework your first task is to corroborate it with history. Verification of theory is extremely hard. We think it is only possible through deductive verification (which we established as futile in Rule # 1) or falsification, but falsification never provides definitive proof of truth (only definitive evidence of un-truth), and what falsification can provide us with is very hard to come by because true falsification tests are hard to arrange. However, corroboration (which is necessary, but not sufficient, for establishing truth) is comparatively easier, and therefore we should first exhaust our options for corroboration of theory with evidence. In other words, to establish the truth of a statement about the macroeconomy, we would like to uniquely map our theorization space onto our observational space (i.e. "what we theorize uniquely implies X", or "we would expect to see X given our theory") and uniquely map our observational space onto our theoretical space ("what we observe uniquely implies Y understanding of the economy"). The latter is very hard to do, but the former is somewhat easier. If you cannot corroborate your theory with what happens in the real world in every episode that comes up, you're in trouble, because you know your theoretical space cannot map onto your observational space. The fix may be relatively easy - you just recognize that you've identified one economic process among many. That's no reason to abandon your theory - that's a reason to expand your understanding of the economy. You may also have to modify the theory itself. This happened when people realized the Phillip's Curve needed to take expectations into account. This should illustrate why corroboration is very fruitful. At the very least it gives us a pragmatic theory that we can say seems to fit a lot of circumstances (and so should be a decent guide for future circumstances), but it also allows us to start the task of weeding out or fixing bad theories. It doesn't give us definitive proof, but it is a very important task. This is how I see the work I've done (and continue to do) with 1920-21. It's quite obvious how 2007-2010 or 1929-1933 corroborates the Keynesian story. 1920-21 is more of a head scratcher for some people, so it was worth some attention. It didn't take that long to realize that it actually doesn't provide an obstacle to Keynesianism at all. It's perfectly consistent with Keynesianism - a Keynesian would expect to see 1920-21 play out exactly how it did. So my theory maps uniquely onto my observation. The problem is, a couple other theories seem to be corroborated by the episode as well. ABCT and monetarism also map onto the observational space we are presented with in 1920-21. So the question is - of the theories which uniquely map onto the experience of 1920-21, which theory or combination of theories does 1920-21 uniquely map onto? That is a tougher question.

Rule # 4 - Theorization is the task of elaborating economic processes of which many could be true. From Rule #'s 1 and 2 another point starts to emerge - that our brains are inadequate to rule out additional theories deductively or inductively so any story-telling we do about the economy cannot be assumed to be exhaustive. What we are doing is theorizing economic processes and hoping to understand and integrate enough of the important ones to have a useful (not exhaustive or strictly "true") description of the way the economy works. I think this point is especially easy to grasp for someone coming from a New Keynesianish perspective. New Keynesianism as a theoretical project was largely the cobbling together of lots of different processes and market failures that could explain the idiosyncrasies of the observed economy. We had credit rationing, asymmetric information, wage rigidity, money illusion, efficiency wages, irrationality, bounded rationality, myopic discounting, frictions, etc. to explain all the funny stuff that could go on. It was an exercise in economic process identification, not economic truth proclamation.

Rule # 5 - Try to uniquely map observations onto theories, but don't hold your breath. You should always look for good opportunities to actually try to verify theories with data. This is very hard for macroeconomics because of how sparse data is. It is especially hard when we are interested in particularly rare phenomena (like depressions). Ideally we would want to have two cases that are exactly identical except one case implements the desired amount of fiscal or monetary stimulus and one doesn't. We could compare the two and get a legitimate test of various theories' implications about fiscal or monetary policy. The problem is, (1.) it's hard to establish that two historical circumstances are exactly the same, much less close enough for comparison, (2.) most of the time some intervention is tried (so there is no counter-factual), but often it's not of a magnitude that anyone is happy with, and (3.) it's hard to get sufficient statistical power even when we do have a case to look at. Macroeconomic theory testing is extremely hard for these reasons, as I mentioned recently in another post where I noted the fact that I came to macroeconomics via labor economics (which has been able to perform much more rigorous and plausible empirical analyses and has quite a high bar for satisfactory identification). I've essentially concluded that this sort of strict theory-testing empirical work is practically impossible in macroeconomics. There are a few good examples of it - like Barro's work on the multiplier. Those are always great to have. But for the most part you're grasping at straws. Empirical macroeconomics mostly has to stick to (1.) providing parameters to plug into theoretical models, and (2.) corroboration/checking for consistency with observation.

Where does this lead, in a nutshell?:

1. Less "schools of thought"
2. More openness to the operation of mutliple processes
3. More history in macroeconomics
4. Less fretting about microfoundations. They're nice and I'm not saying they're bad, but notice they don't feature very prominently in my schema.


  1. Rule #1 highlights one of the main problems that I've had with praxeology. It isn't rigourous enough, and if it were the proofs would be infinitely long.

    On another note:
    "most of the time some intervention is tried (so there is no counter-factual), but often it's not of a magnitude that anyone is happy with"

    Now, I'm not that familiar with Public Choice theory, but I think that's a problem that would be worthwhile to look at from that perspective.

  2. nick - wait for my post on public choice theory I have planned for tomorrow morning, and in the meantime read this and follow the links:

    I think you're EXACTLY on target, and I'd like to see public choice theorists doing more of this rather than using it as a rhetorical device for libertarian politics (which is largely what public choice theory amounts to today, I think).

  3. Hrm, interesting DeLong article, but there's one thing that I don't like. He seems to imply that the power of lobbying has diminished since more people are able to vote. I don't think it has at all.

    So I'd extend his first point (about rich people with an incentive to lobby for stable prices, regardless of unemployment) can be extended to today's economy, and say that there are rich people with an incentive to lobby for stable/rising asset prices, regardless of consumer prices.


All anonymous comments will be deleted. Consistent pseudonyms are fine.