by David Colander
http://d.repec.org/n?u=RePEc:mdl:mdlpap:0912&r=dge
This paper argues that the DSGE approach to macroeconometrics is the dominant approach because it meets the institutional needs of the replicator dynamics of the profession, not because it is necessarily the best way to do macroeconometrics. It further argues that this “DSGE-theory first” approach is inconsistent with the historical approach that economists have advocated in the past and that the alternative European CVAR approach is much more consistent with economist’s historically used methodology, correctly understood. However, because the European CVAR approach requires explicit researcher judgment, it does not do well in the replicator dynamics of the profession. The paper concludes with the suggestion that there should be an increase in dialog between the two approaches.
Is Colander correct in calling for approaches that are consistent with historic approaches to be privileged? Scientific progress is made by trying new approaches, and if they allow old ones to be rejected, then be it. But can we reject these old approaches?
NEP-DGE Blog 
Colander builds a model of ideas in macroeconomics and uses it to critique DSGE methodology. Its a bit ironic that Colander builds a theory before he tackles the data. And then when he tackles the data, he doesn’t test every assumption of his model. This are two of his main critiques of the DSGEs!
Its clear that Colander thinks CVARs are better than DSGEs for stylistic reasons. But I’m not sure why his, or anybody’s, aesthetic opinion should be given any weight when deciding on methods.
He explicitly ignores the issue of whether or not the “European” methodology produces better and more scientific results than the “American” methodology. He is “not enough of an econometrician to make a conclusive judgment on this issue.” To me, however, this is the key stylized fact to know *before* we try to explain the behavior of macroeconomists.
If the “American” methods produce more and better results, then there’s a much more parsimonious explanation for macroeconomists behavior and the dominance of DSGE. It gets us closer to the truth.
Thanks for the comment.
In the paper I refer to articles that look at the empirical work done using DSGE models and the CVAR approach. The bottom line is that there is enormous debate about the empirical work and how to judge the empirical work, but that the models such an England’s which do empirical work based on the acceptance of the DSGE model don’t seriously take the model to the data. Please look at Katarina’s Juselius’s paper in the same issue of the journal. I justify not looking at the empirical work with the argument that the paper was in special issue of the journal and a division of labor makes sense.
In my previous comment, I didn’t realize that the reference was to the working paper and not the published paper. The article has been published in the Economics Ejournal which is an open access journal that I recommend to scholars both for reading and as an outlet for your work. Here is the citation to the work.
David Colander (2009). Economists, Incentives, Judgment, and the European CVAR Approach to Macroeconometrics. Economics: The Open-Access, Open-Assessment E-Journal, Vol. 3, 2009-9. http://www.economics-ejournal.org/economics/journalarticles/2009-9
I have quite a different reading than Colander what the DSGE approach is about. I do not think it is appropriate to call it “theory-first.” Indeed, any theory, and it applies to DSGE theories, are a consequence of observations. In this case, it is standard practice to look at “stylized facts” or “empirical regularities,” which may by the way result from a VAR, and thus motivate the quest for a theory.
Also, Colander should know that model are always abstractions and will always be simplifications of the real world. This applies to DSGE models as much as CVAR models. In particular, CVAR models are largely atheoritical and thus do not any way to discipline the data. This is important if you want to make inference about policy. Just think of the Lucas Critique.
This brings me to the major shortcoming of CVAR: it cannot say anything about policy in a meaningful way. How are we to think about the impact of a policy that is not present in the data? Or even if was tried before, how to identify in the data from a myriad of other effects. You need structure that a CVAR cannot offer, but DSGE can.
That said, the CVAR approach is still useful. To describe the data. Then DSGE takes over.
Thanks for the comment.
I think most people, including supporters see it as a theory first approach. How a deductive theory develops
in relation to general observation is a complicated question, but the DSGE model is not developed from stylized facts. It is a model that concentrates on getting the logic of the intertemporal reasoning right.
In the paper I argue CVAR models are not atheoretical–implicit theory goes into the process of selecting variables. In the paper I argue that it is best to see cointegrated vector auto regression as a process–not a model–and a process that necessarily uses judgment. That judgment reflects theoretical understanding.
Let’s not kid ourselves. There is a lot of data mining going on in CVAR, and theory clearly plays second fiddle, implicitly or explicitly.
I fully agee that there is data mining going on. But when you have a system as complex as the macro economy, where formal theory does not provide much guidance, one needs to find the best patterns in the data that one can, using the best statistical techniques and combine that with one’s best understanding of theory. It is an art, and it involves judgment, which is what my paper argues is the case.
Re: policy, I would suggest it’s even worse than what’s been said, for example Dr. Juselius treats events off the “economic calendar” as structural breaks and splits the sample, so we don’t even get the “wrong”, reduced-form, Lucas-criticable answer.
OTOH, the GE-RBC tradition has always acknowledged that we don’t have a full description of the underlying process but that we can use sample statistics to assess the promise of various theoretical mechanisms, while allowing for a (perhaps limited) policy response endogeneity. (NK-tradition models who add dubious shocks to achieve fit have abandoned most if not all of the principles that initially motivated DSGE work.)
The GE-RBC people have also stressed that theory must not only explain the data but also be internally consistent. — Using (potentially unstated) theory to select variables and look for cointegration (as Dr. Colander describes the use of theory in CVAR practice) opens up the possibility that the implicit theory doesn’t “add up”. Unless you state, close and solve a formal model, how would you know?
Finally, perhaps the most difficult problem for the CVAR approach is that it hasn’t delivered any important result, it has not provided a high-profile insight.
P.S. I second M.H. point regarding (C)VARs and DSGEs as being complements rather than substitutes in the macroeconomist’s toolkit. — Beyond the anti-DSGE rhetoric, the Juselius blue book is an otherwise awesome textbook.
I totally agree that judgements have to be made with the CVAR approach as they must be made with every approach. When you have a complex system, data do not provide a final answer. What I like about CVAR is that it makes it implicit without using a prior. The DSGE approach tends to hide the judgements, and I think transparent judgements are preferrable–then others can decide whether they believe they are reasonable. When the judgements are hidden, they are harder to judge whether they are reasonable.
I am really puzzled how David Colander seems to insist the DSGE approach is hiding its judgements, whereas CVAR is not. With a DSGE model, you explicit all assumptions you are making. This is the great thing about it, it is falsifiable. CVAR, however, hides everything because it is essentially reduced form.
Look at Katarina’s and Soren Johanansen’s exploration of Peter Ireland’s “Bringing the Model to the Data.” They show that when one is careful about the specificaiton, he has not brought the model to the data at all. Most of the work that is practial just tries to use the DSGE model as a prior and then combines it with VAR analysis. That is not bringing the model to the data–and is not falsifable.
This is the last response from me for a while. I will be travelling for the next month.
Thanks everyone for their comments, and I encourage you to read the papers by Soren and Katarina in Post Walrasian Economics book I edited. (Cambridge University Press)