Simon Wren-Lewis, Paul Krugman, and Robert Waldmann have posts up discussing whether macroeconomists always need to use "microfoundations" - in other words, whether macro models should always start with some sort of individual optimization. I'll get to their ideas in a minute, but first, some general thoughts.
Why are microfoundations useful? The usual answer is that "microfoundations make models immune to the Lucas Critique." The idea is that the rules of individual behavior don't change when policy changes, so basing our models purely on the rules of individual behavior will allow us to predict the effects of government policies. Actually, I'm not sure this really works. For example, most microfounded models rely on utility functions with constant parameters - these are the "tastes" that Bob Lucas and other founders of moden macro believed to be fundamental and unchanging. But I'd be willing to bet that different macro policies can change people's risk aversion. If that's the case, then using microfoundations doesn't really answer the Lucas Critique.
A better reason to use microfoundations, in my opinion, is that they probably lead to better models. "Better," of course, means "more useful for predicting the future." If our models predict future aggregate macro variables (GDP, etc.) based solely on the past values of those variables, we'll almost certainly be using less information than is available; if we figure out how economic actors are making their decisions, we will have a lot more information. More information = better model. And there are all kinds of ways to observe and model individual behavior - survey data, lab experiments, etc.
So why would we want to use models that don't have microfoundations? Here is Simon Wren-Lewis' answer:
[S]uppose there is in fact more than one valid microfoundation for a particular aggregate model. In other words, there is not just one, but perhaps a variety of particular worlds which would lead to this set of aggregate macro relationships. (We could use an analogy, and say that these microfoundations were observationally equivalent in aggregate terms.) Furthermore, suppose that more than one of these particular worlds was a reasonable representation of reality. (Among this set of worlds, we cannot claim that one particular model represents the real world and the others do not.) It would seem to me that in this case the aggregate model derived from these different worlds has some utility beyond just one of these microfounded models. It is robust to alternative microfoundations.
In these circumstances, it would seem sensible to go straight to the aggregate model, and ignore microfoundations...I don't really like this answer. Presumably, there is some set of sets of microfoundations that leads to Aggregate Relationship A and some other set of sets of microfoundations that leads to Aggregate Relationship B. How do you choose which set is better? Well, you could look at survey data and lab experiments to figure out which microfoundations are really in effect. But if you can do that, why do you care about "robustness to alternative microfoundations" in the first place? And if you can't choose which microfoundations are better, why does "robustness to alternatives" matter?
Pretty much any model, in economics or physics or whatever, has a bunch of possible microfoundations that could give rise to it. That fact alone does not make microfoundations less important, since presumably some microfoundations are actually happening, and others aren't!
Here are Paul Krugman's answers to why we might not need microfoundations:
1. Even in microeconomics, we don’t insist on using models built up from maximizing behavior all the time. Exhibit A: supply and demand!...
2. Relatedly, as a practical matter intellectual scratch-pads — approximate version of what we really believe, but stripped down to be tractable — are what one uses for applied economic analysis all the time.If I want to ask what the effects of some shock will be, it rarely makes sense to demand that the analysis always go all the way back to the intertemporal choices of optimizing agents.Hmm. I think that incorporating microfoundations into a model is different than starting from microfoundations when applying that model.
3. In the hard sciences, when dealing with complex systems people have often used higher-level, aggregative concepts that seem to work empirically long before they have a full derivation of effects from the underlying laws of physics...Why, then, do some economists think that concepts like the IS curve or the multiplier are illegitimate because they aren’t necessarily grounded in optimization from the ground up?I think this is where the Lucas Critique comes in. The Phillips Curve is the famous example of why aggregate relationships might not be useful without understanding the microfoundations. That doesn't make aggregate-only models useless, but it should make people cautious about using them.
4. And when making such comparisons between economics and physical science, there’s yet another point: what we call “microfoundations” are not like physical laws. Heck, they’re not even true. Maximizing consumers are just a metaphor, possibly useful in making sense of behavior, but possibly not. The metaphors we use for microfoundations have no claim to be regarded as representing a higher order of truth than the ad hoc aggregate metaphors we use in IS-LM or whatever; in fact, we have much more supportive evidence for Keynesian macro than we do for standard micro.I think that this is the real argument against microfoundations as they are currently used in macro. Basically, Krugman is saying that the "microfoundations" we now use really deserve to have quotes around them, because they actually don't describe individual behavior.
In other words, our current microfoundations are mostly just garbage.
If this is true - and I think that the evidence overwhelmingly says that it is! - it means that our modern "microfounded" macro models are no more useful than aggregate-only models. The logic should be obvious. Using wrong descriptions of how people behave may or may not yield aggregate relationships that really do describe the economy. But the presence of the incorrect microfoundations will not give the aggregate results a leg up over models that simply started with the aggregates.
In other words, if you put garbage in, you may or may not get garbage out, but why bother putting the garbage in in the first place?
(Note: if you started to angrily type out the reply "But all models are wrong!", please refer to my 2nd Principle for Arguing With Economists. You are wrong.)
When I look at the macro models that have been constructed since Lucas first published his critique in the 1970s, I see a whole bunch of microfoundations that would be rejected by any sort of empirical or experimental evidence (on the RBC side as well as the Neo-Keynesian side). In other words, I see a bunch of crappy models of individual human behavior being tossed into macro models. This has basically convinced me that the "microfounded" DSGE models we now use are only occasionally superior to aggregate-only models. Macroeconomists seem to have basically nodded in the direction of the Lucas critique and in the direction of microeconomics as a whole, and then done one of two things: either A) gone right on using aggregate models, while writing down some "microfoundations" to please journal editors, or B) drawn policy recommendations directly from incorrect models of individual behavior.
Brad DeLong puts this rather more pithily:
I now have the most bizarre image in my mind:
A seminar at the Library of Alexandria in 300 A.D., with an astronomer trying to provide micro foundations in the form of calculations of how large their wings must be and how fast their wings must beat for the angels to push the planets on their tracks through the quintessential spheres…Thus it seems to me that the microfoundations revolution has not really gotten us very far yet. I would be willing, of course, to be convinced otherwise.
So what to do? The answer is clear: macroeconomists should continue using aggregate relationships for now, and try to check these against the best microfoundations available. But in the meantime, recognize that these aggregate models will have severe limitations until microeconomists come up with better explanations of individual behavior. Which, of course, they are working on.
But note that there is also a political danger here. Macroeconomists who desire a certain policy conclusion - for example, that fiscal stimulus never works - may be tempted to continue to use bad microfoundations that support that conclusion, even when microeconomists have found something better. This is sometheing the profession should work to avoid, by actively recognizing that microfoundations that fit the micro data are inherently preferable to those that do not.
Update: Paul Krugman, commenting, has a good point about what kind of predictions we should expect from economic models. Big qualitative predictions ("quantitative easing will cause runaway inflation") are more important than precise quantitative ones ("GDP growth will be 1.7% next quarter"). I agree, of course. When I say models should "predict the future," this is really what I mean.
Update 2: Richard Serlin points out that aggregation is a huge challenge for microfounded models, since complex systems often have chaotic properties. Very true. But that doesn't mean that just observing the aggregate will give you more information! What you need to handle complexity and chaos is a ton of computing power and some agent-based modeling, as is done in weather forecasting. This can provide a very important check on non-agent-based models that make simplifying assumptions in order to aggregate individual agents.
Update 3: Peter Dorman is of the opinion that the reason our current microfoundations are crappy is that the entire framework by which microeconomics is now done - equilibrium analysis and optimizing behavior - does not describe reality. I'm not willing to go that far (and besides, what about game theory?), but if he's right, it would certainly strengthen my case substantially.
Update 4: Andrew Gelman and Peter Dorman are basically on the "our current microfoundations suck, and we should get better ones" bandwagon.