For some Friday night fun, and, caution, after a day grinding through DSGE models of credit frictions, preparing for an MSc Masters module at Bristol later this year, I thought I would comment on Noah Smith’s typically entertaining macro-smackdown.
For the uninitiated, a ‘smackdown‘ is an American wrestling term for, I guess, a decisive point score by getting your opponent on his or her back on the floor of the ring. It should be part of a US blog reading glossary for consuming the Smith, DeLong and other fare. Steve Williamson chided Noah once, indirectly, for using terms like this. But for me in conjures up the era of wrestling on Saturday afternoon TV in the UK in the 1970s, pure choreographed theatre; on the surface, violent, but actually perfectly friendly. So ‘smackdown’ seems fine to me. It’s especially appropriate for this post.
Noah’s smackdown for DSGE models is to observe that no-one uses them in the private sector. Evidence he takes for their lack of worth in society. OK, how do we go about throwing that one onto its back?
Well, one tactic is to point out that there are quite a few private sector people doing DSGE modelling. I used to work with some of them. I’m not going to name them, in case they feel shamed by being outed as DSGE-ers. But you know who you are. However, this is a pretty sterile tactic. Because the reason these people are doing it, and I speculate many are, is because they know central banks are doing it. And if everybody in the central banking community is doing it, you can see why it might pay the private sector economist to be doing it. If they think that central banks are actually using those critters to set policy, they might be able to get their clients an edge by figuring out what those DSGE models will be telling central banks. Why does that make this tactic for refuting Noah sterile?
1. Central banks could be mistaken in using these models. Though it pays the private sector to use them to try to shadow them, it doesn’t prove their worth. Tony Blair denied that he and George Bush used to pray together. But, even supposing they did, and we decided to check out the Bible for a clue as to what they were thinking, that wouldn’t prove the social utility of the Holy Book.
2. Central banks could be hoodwinking everyone into thinking they are using these models. [My comment on the BoE: they have much less of an influence than you might think. So little influence that they can ship out one model, and ship a whole new one in without anyone really noticing it in the inflation forecast profile. In fact, they can perform this feat twice in a decade]. Why would they do that? Well, many reasons. Here’s one: there is an arms race between policy institutions to try to hire smart people who like thinking about the economy. To get them, you have to tell them that these models are very important to you. Once they are in the door, you let them write a few working papers, and pay a few flights for conferences. Of course, all you want to do is ask whether real yields went up or down last month, but to get someone able to pick up the phone, open a spreadsheet, and subtract one line from another, you have to go through the whole charade.
OK, so we are done with trying to refute Noah’s thesis by pointing out that private sector people do use these models.
Well, not quite. A lot of people who wind up in the private sector have studied DSGE models. Are there any macro Masters or PhD courses that don’t cover them now? [I mean DSGE in the term I think Noah meant it, ie microfounded macro, not just direct descendents of Blanchard-Kiyotaki]. If these models are worthless, why are the private sector still hiring a disproportionate number of people who have wasted their lives committing these fairy-stories to memory, and cluttering up their computers with them? [My microfounded model Noah Smith replies: they are trapped by the same network externality. The market coordinated on this being the way to test whether someone can endure the tedium of early mornings, business casual, and Powerpoint over lunch.]
Onto other tactics.
Perhaps the worth of DSGE models is a public good. Perhaps they cement certain externalities to do with the good design of monetary and fiscal policy. The Germans managed to cement their own distaste for inflation with a hyperinflation, so they did it without the benefit of DSGE models, which were at that point only twinkles in the eye of the people who would beget the people who would eventually beget them. However, perhaps DSGE models offer another, less socially catastrophic form of institutional glue. In which case, they might not be all that useful for the private sector. Especially if they are only good for figuring out the headlines. Like: high inflation doesn’t pay in the end. Or: you should use monetary and fiscal policy actively to stabilise the business cycle. [This is what they do say, by the way, in case you are a general reader and you have, by some miracle, not clicked away in irritation by this point].
Noah argues that according to DSGE creed, you can’t do conditional projections with anything but a Lucas-Critique proof model. Because if you do, when you project conditional on a policy that looks a lot different from the one that went before, you’ll invalidate the statistical correlations encoded in your non-Lucas-Critique proof model, and the effects you hoped for won’t materialise. However, it was widely recognised that how much those coefficient changes would depend on how much the contemplated conditional projection differs from what went before. Leeper and Zha explore his in their paper on ‘modest policy interventions’. ‘Modest’ means modest enough to make it ok to use a VAR, estimated on a particular policy history, to do a conditional projection. So, perhaps the private sector nerds Noah knows never need DSGE models because they find their modest policy interventions don’t undermine their empirical model coefficients.
Aren’t unconditional distributions ok for many purposes? Here I court disaster in the form of a finance smackdown. But, don’t I sometimes just want to know the joint unconditional density of outturns for everything relevant for pricing all assets? In which case, isn’t an infinite dimensional nonlinear empirical time series representation adequate for such a piece-of-cake task? Or, failing that, say a VAR with a few choice macro finance variables and a few lags of each? If I want to price a ten year government security, perhaps I would be happy just to know what the central bank rate will be over the next ten years. In which case the details of policy, provided they don’t change, can be mashed up with the details of everything else in the VAR. This is what some of the other people I used to work with do now for a living.
WHUMPH. [Sound of 120kg wrestler thrown on ring].