Monday, January 13, 2014

Are People Being Mean To Those Doing DSGE Models, Or Is It The Other Way Around?

There has been an outbreak recently of people picking on the usefulness of dynamic stochastic general equilibrium (DSGE) models in analyzing macroeconomies, particularly for policymaking and forecasting purposes.  This has been going on for a long time, and I confess to having been one of the guilty parties.  In the past those of us doing it have tended to emphasize the unreality of assumptions in these models, such as rational expectations and general equilibrium.  Although newer models have allowed for limited versions of heterogeneous agents, many of them have also assumed a homogeneous agent.  Positivist instrumentalism is used to justify making obviously unrealistic assumptions in order to achieve empirical understanding and especially forecastibility.  While the calibration of DSGE models have allowed them to be able to replicate past outcomes, their ability to forecast the future or provide even conditional policy advice, the latter what they are supposed to do best given their supposed ability to satisfy the Lucas Critique, has proven very poor, with 2008-09 the poster boy for this.  Supposedly front room central bankers are highly frustrated with what they are getting from their DSGE modelers in the back rooms (although they also use Klein-style quasi-Keynesian models and purely atheoretical VAR-type time-series models as well), with these complaints becoming public and many criticizing these models.

The most recent outbreak first showed up on the infamously anonymous Econ Job Market Rumors site only to get repeated by Noah Smith and some others who note that these models simply are not used by any private consultants or investment firms or, well, any private firm that anybody has been able to dredge up in lengthy comments, although one commenter noted that at least one firm uses an ability to DSGE model as a screening device for hiring people, given that this does take some quantitative skills, even though these people are not being hired to actually do any DSGE modeling.  This lack of any such use by the private sector is touted as the final bottom line proof of the ultimate worthlessness of these models.

On the other hand others argue that this is unfair and overblown, with Tyler Cowen weighing in that even if there are problems with DSGE models, people should merely "devalue" rather than "dismiss" them, the latter argued to be unreasonable and narrow-minded, .  (This has a link to Noah's post, which I seem to be unable to  link to for whatever reason.).  The comments on Tyler's post go further in defending the DSGE approach more generally on the basis of its consistency with micro optimization that is foundational for micro, and this is all about microfoundations naturally, and "we cannot throw out all our textbooks!" blah blah blah.  One defender did claim that DSGE models, "in all but name," are used by private firms, although without mentioning any specifically, and somebody on Noah's site brought up Computable General Equilibrium (CGE) models, which are used by private firms, but only for micro rather than macro purposes.

So why has all this back and forth become so heated and nasty?  Part of it is that the defenders of DSGE have been fully arrogant and have fought back and maintained their position, even as they have been ridiculed and attacked by many.  I dislike invoking what looks like conspiracy theory, but the hard fact is that the advocates of DSGE modeling in its varied forms continue to control both the top journals as well as the top departments in economics.  One of the more serious alternatives that has been put forward to DSGE models has been Agent-Based Models (ABM).  I know that not a single ABM paper has been published in a top 4 journal.  I know that when these are submitted, they are rejected on such grounds as "this is not an equilibrium model," which is a sign that failure to fit into the DSGE paradigm is sufficient for the papers to be rejected.  The defenders of DSGE trumpet this lack of such papers getting into the top journals, actually bragging about it, although most loudly on anonymouse sites like EJMR.  Rather than evidence of suppression of competing ideas, this is seen as proof positive of the rightness of their position.  It is this, and their stranglehold on the research departments of most central banks as well, which draws forth the ridicule and opposition when their models have so clearly failed to deliver the advertised goods.

Let me recognize that DSGE modelers are trying hard to make their models more useful for studying things that have occurred.  They are introducing heterogeneous agents in the form of intervals containing continua of agents who as an interval act like the homogeneous agent of the older models, although allowing some endogeneity of the distribution across the interval, such as a wealth distribution.  They are introducing financial frictions into them, thereby supposedly Minskyizing them.  Some now even have forms of bounded rationality rather than full rationality in the form of extra elements added to the stochastic shocks that are already assumed to be driving everything in these models.  T'hey have long had price and wage stickiness in them, leading to the New Keynesian forms of these models, with such ad hoc devices as the Calvo pricing mechanism fairy, although most Post Keynesians and even some other Keynesians view wage and price stickiness as far from being key to a model being "Keynesian."  Indeed the idea that wage and price flexibility will eliminate business  cycles is a serious myth, speculative bubbles showing that excessive price flexibility can cause and aggravate them, and the general failure of the DSGE models to handle those one of their biggest weaknesses.

In any case, given the combination of arrogant self-satisfaction and defensiveness by the DSGE modelers in their various redoubts and the clear annoyance and dismissal of these models, most crucially by many who should be their main customers, policymakers, with this added rub of their complete non-use by the private sector, this fight is likely to continue.  They may provide "discipline," as their advocates note, but I also know that their appeal to micro is not accepted particularly by real micro general equilibrium theorists, who pretty much scorn DSGE models.  

Barkley Rosser

Added: DSGE models should be distinguished from Computable General Equilibrium (CGE) models.  These are open to some of the same criticisms as DSGE, but not nearly as much so. These have been around for a long time and used for policy purposes, but are strictly micro.  A well-known example is their use in predicting impacts on trade flows by sector NAFTA.  Several competing models were used that gave answers with different numbers, qualitatively similar answers in terms of which sectors would be affected how, although one can argue that this qualitative matter could be determined without them.  In any case, their strictly micro focus differentiates them from DSGE models, and while flawed, are used by private sector firms and viewed as more successful than DSGE ones.


  1. Thanks for hoisting me out of the "devalue and dismiss" dilemma of whether or not to learn DSGE. In real life, I have been trying to deal with nutrition in Brazil, where virtually all the models are CGE. The Wikipedia entry for CGE explains why.

    Problem is, the standard consumer elasticity assumptions don't hold at micro level. So it is the same dilemma either way!

  2. Barkley, I should point out that there are CGE models used for macro/international analysis. Admittedly, this is a minority, but not so much because it is uncommon, but rather because the data requirements for running a credible global macro CGE model are so significant that only a few institutions with large staff are able to routinely maintain such things (we do at the World Bank, and have used such models to engage topics ranging from macro impacts of global climate change to the future evolution of global saving and investment). Add to the high startup and maintenance costs the fact that CGE models seldom get featured in top journals and we see why the cost-benefit payoff skews so heavily away from CGE modeling. Final minor point: CGE models tend to be recursively dynamic, in the sense that dynamics in these models hinge on accumulation equations that pass period-optimized equilibrium outcomes to the next period, rather than solving an infinite-horizon stochastic optimization problem as in DSGE models.