In a recent post, Steve Williamson writes:
The tools of modern macroeconomics are no more the tools of right-wingers than of left-wingers. These are not Republican tools, Libertarian tools, Democratic tools, or whatever. These are the tools of Economic Science[.]I've thought about this for a long time, and I'm not sure that Steve is right. I think there is a case to be made that the methodology of modern macroeconomics has the effect of biasing the field toward conservative policy recommendations.
Let me explain why.
Modern models of the business cycle generally rely on one of two techniques: 1) Dynamic stochastic general equilibrium models (DSGEs), or 2) Structural vector autoregressions (SVARs). The former is by far the more popular and well-accepted (although Chris Sims won the Nobel Prize for inventing the latter), so when I talk about "the methodology of modern macro," I'm going to talk about DSGEs.
One of the main features of DSGEs is that they are microfounded; that is, they try to explain macroeconomic phenomena in terms of individual decisions. Another feature is that they are based on optimization, which means that the individual decisions are modeled using the calculus of variations.
Explaining macro phenomena based on individual optimization is hard to do. Individuals may take many things into account when making their decisions; in math terms, this means you can easily have a large "state space." Also, the thing that people optimize (their "objective function") may be very complicated; in principle, it can include all manner of weird things like non-rational expectations, learning, dynamic inconsistency, habits, overconfidence, reference points and framing effects, etc. Finally, aggregating a whole bunch of individual decisions into one giant macroeconomic outcome is in principle a very hard thing to do; it's even harder if you include things like firms and governments.
So, unsurprisingly, making a DSGE model is a lot easier if you make some simplifying assumptions. Here are some simplifying assumptions that make a DSGE pretty easy to solve:
1. The assumption that the economy can be modeled with a representative agent; in other words, that the macroeconomy behaves as if there's only one person in it.
2. The assumption that government doesn't exist, or exists only to transfer income from one person to another.
3. The assumption that prices are fully flexible.
4. The assumption that firms are simple profit-maximizers and make zero profits in equilibrium.
5. The assumption that individuals have rational expectations.
6. The assumption that risk preferences can be entirely modeled using people's utility of consumption, and that this utility can be modeled using a small number of parameters that do not change over time.
7. The assumption that labor markets clear.
8. The assumption that "technology" is represented by the Solow residual, and that technology is exogenous and evolves according to a simple time-series process (for example, an AR(1)).
9. The assumption that the business cycles we observe represent small enough fluctuations that the model that describes them can be linearized around its steady state.
If you make all of these simplifying assumptions (and a few more), you end up with something like the first DSGE business-cycle model: the "Real Business Cycle" model of Edward Prescott and Finn Kydland, first published in 1982. This model, and the approach it pioneered, won a Nobel Prize for its authors.
Now, if the above assumptions seem unrealistic to you, that's because they are! And if you think that this makes the RBC model unlikely to fit the data, well, you're right. It doesn't.
(Side note: If Kydland and Prescott's model didn't fit the data, then you may ask, why was it awarded with a Nobel Prize? The answer is "nobody knows the mind of the Nobel Prize committee," but it is probably because this model was the first business-cycle model to try to answer the Lucas critique. The Lucas critique says that models should only contain "deep structural" parameters - i.e., parameters that won't change when government policy changes. Kydland and Prescott's model bases everything on "tastes" and "technology," which most economists at the time - and many even now - were willing to accept as "structural." Thus, it seemed to many people at the time that Kydland and Prescott had invented a modeling approach that had a good shot at one day explaining the business cycle in a way that wouldn't change when policy changed. Many macroeconomists still believe this, as evidenced by the dominance of the DSGE modeling approach in the macro literature.)
So, the DSGE model that is easiest to make (RBC) doesn't do a great job of describing the business cycle, much less predicting it. What would be better?
Fast-forward to 2007, and have a look at the Smets-Wouters model of the business cycle. This "New Keynesian" model is currently considered the "best" DSGE model in terms of forecasting performance. Which is to say, it performs ever so slightly better than the judgment-based forecasts of well-informed individuals. Consequently, some variant of the Smets-Wouters model is used by most central banks as their DSGE model of choice (which they use as a complement to other types of models, such as SVARs, reduced-form models, and judgment-based forecasts). Of course, the fact that Smets-Wouters is the "best" DSGE model does not mean it is very "good." Its forecasts are basically useless more than one quarter into the future.
Of course, this slight improvement on the original Kydland-Prescott model comes at a high cost in terms of the complexity of the model. Instead of one or two "shocks" (exogenous factors that are postulated to drive the business cycle), Smets-Wouters has seven. And there is a lot of doubt that all of these shocks are "structural" in the sense of the Lucas critique - in other words, there seems to be a pretty big chance that the parameters of the Smets-Wouters model would change if policymakers changed their policies (thus begging the question of why Smets and Wouters bothered to use a microfounded DSGE modeling approach in the first place).
Now realize this: It took 25 years to go from Kydland-Prescott's RBC model to Smets-Wouters. That is comparable to the time it took physicists to develop quantum mechanics.
And yet, despite being so complex, and despite making heroic assumptions about the "structural-ness" of certain parameters, and despite being 25 years in the making, the Smets-Wouters model does not come even close to capturing all of the "frictions" that people believe are at work in the macroeconomy. It does not include the financial frictions that many people believe caused the 2008 financial crisis. It does not include behavioral effects like habit formation, hyperbolic discounting, etc. It does not include learning. It does not include limited enforcement of debt contracts. It does not include hysteresis in labor markets. It does not include income or wealth heterogeneity among households or firms. And this is not even close to an exhaustive list of the relevant things that it doesn't include. To include all those things in one model is prohibitively difficult with current technology; the state space of the model explodes, and you would need a supercomputer to solve it if it could be solved at all.
So what this illustrates is that it's really hard to make a DSGE model with even a few sort-of semi-realistic features. As a result, it's really hard to make a DSGE model in which government policy plays a useful role in stabilizing the business cycle. By contrast, it's pretty easy to make a DSGE model in which government plays no useful role, and can only mess things up. So what ends up happening? You guessed it: a macro literature where most papers have only a very limited role for government.
In other words, a macro literature whose policy advice is heavily tilted toward the political preferences of conservatives.
Is that bad? Not necessarily. If the facts had a well-known conservative bias - i.e., if the models that fit the data best were the models that implied no role for government - then that would just be too bad for liberals! Liberals would have to accept that their ideas were contradicted by the best scientific evidence available.
But I contend that in the case of DSGE models, conservative policy recommendations don't emerge because they come from the best models, but only because they come from the easiest models. Thus, the conservative slant of modern macro comes not from the weight of evidence, but from the combination of publication bias and the inherent unwieldiness of the DSGE framework.
Now here's something else that might be worth mentioning. The DSGE framework was invented in large part by Ed Prescott, a man with deeply conservative political beliefs. The insistence that microfounded models with individual optimization were the only believable "structural" models - i.e., the only models that could answer the Lucas critique - came mostly from people with deeply conservative political beliefs (including Robert Lucas himself). And the criticism of alternative modeling approaches - in particular, of SVARs - seems to be much louder from economists with deeply conservative political beliefs.
That by itself proves nothing. (Maybe they're conservative because they believe the results of their models! Maybe conservatives are more scientifically honest!) But it seems like circumstantial evidence against the alleged political neutrality of modern macro methods.
Was DSGE created as an intentional conspiracy by conservatives to force macroeconomists onto a playing field tilted toward laissez-faire policy conclusions? Almost certainly not. Have conservative-minded macroeconomists been privately pleased with the publication dominance of models that tend to vindicate their prior beliefs? Almost certainly yes. Do I have a better alternative modeling approach handy? No (I'm not brave or foolish enough to mount a spirited defense of SVARs).
The real question, though, is: Has the "conservative publication bias" of DSGE made macroeconomists more complacent than they should be about searching for alternative modeling approaches, even in light of the extremely limited usefulness of DSGE models three decades after their creation? I don't know the answer. But if the answer is "Yes," then the claim that DSGE is a politically neutral tool of economic science is not quite right...
Update: It's worth pointing out that Thomas Sargent, one of the pioneers of both DSGE and Rational Expectations, and one of the three Nobel Prize winners in the photo at the top of the post, is actually a Democrat (though it's also worth pointing out that he left the Rational Expectations paradigm and started working on learning-based models, which have proven to be a lot harder to work with!).