Thursday, December 20, 2007

CGE: Is There a Defense?

I’ve been asked to review an article using computable general equilibrium (CGE) methodology. I’m likely to decline, but before I do I want to ask the vast universe that follows this blog: is there any defense against the argument that CGE and its offspring (DSGE) are simply bad economics?

There are two arguments actually:

1. CGE is an attempt to implement empirically a model that has been blown away theoretically. For 30 years we have known in precise terms why representative agent GE models are hogwash. We also know that the conditions for unique solutions are impossibly restrictive. Finally, while we have models that can translate modern, post-utility understanding of economic behavior into functional form, to do this throughout an economy, in every nook and sector, would be a gargantuan, and probably pointless, project. To put it bluntly, CGE modelers take as their starting point refuted theory.

2. CGE is false empiricism. It claims to generate results based on real-world data, but testing is nil, and I mean nil. Is there any literature out there I have missed in which past models are examined retrospectively against actual economic outcomes? If not, where is the falsifiability?

I will wait to send my rejection email. Maybe one of you can convince me that it is worth a few hours of my time to promote “better” CGE work.


Sandwichman said...

- "is there any defense against the argument that CGE and its offspring (DSGE) are simply bad economics?"

- "CGE is an attempt to implement empirically a model that has been blown away theoretically."

But what if "economics" itself has been blown away?

Laurent GUERBY said...

FYI wikipedia does not present any "negative" view on CGE and DGSE.

YouNotSneaky! said...

The person to ask may be John Quiggin.

And I don't think I'd put CGE and DSGE in the same camp.

YouNotSneaky! said...

Also, if you're going to ask people to step up and defend something that around these parts is probably pretty unpopular, and as a result come in for a good bit of hatin' from your commentators, it would be nice if the relevant standard or grading scale was made explicit.

What alternatives to CGE are we comparing to?

The reason I ask for this is because I tend to be of the opinion that many of the problems (at least on the theoretical side) one runs in with GE, CGE and even DSGE are not particular to these approaches but rather are of the 'aggregation is a tricky business' variety. Which means they tend to be shared, in some form or guise, by the alternatives.

Econoclast said...

I'm no expert on CGE, but what I've seen indicates that it mostly involves "garbage in/garbage out." One makes assumptions and then cranks out the results. The assumptions are never tested. The CGE model simply imposes the "ideal market" political framework on the assumptions.

Jim D. said...


I would agree that the various beasts are not identical. DSGE models are more prone to aggregation problems, and the related problem of representative agent with rational expectations. This is a serious flaw.

Broader CGE problems that are not specifically DSGE relate to GE itself. Does it exist? What about increasing returns to scale and externalities? Might not various disequilibrium dynamics models perform better? Some have been around, although more at the macro level than the micro CGE level.

It is always wise to take any of these models with large grains of salt. Nevertheless, they can sometimes be useful as benchmarks from which to start an analysis, assuming one caveats carefully.

Econoclast is right that twiddling with the assumptions (changing the garbage in) can alter what comes out. So, if one really wants to use these babies, one should look at more than one, including both CGEs and DSGEs. They can be all over the map, for example with Menzie Chinn over at econbrowser showing a wide set of competing outcomes for DSGE models in terms of estimating effects of fiscal budget balance changes on current accounts. The CGE models predicting outcomes of NAFTA were generally off by noticeable margins, generally being too optimistic, although the scare story tellers also tended to overdo their tales as well.


Peter Dorman said...

I think by focusing on the first criticism, you might miss the larger point. It is true that other applied fields in economics rely on dubious theory, but they have the potential to mitigate this by empirical testing. (There are problems with the way economists typically undertake such testing, but that's a point for another day.) Over time, the theory bends to accommodate the facts. But if there is no empirical testing of CGE and similar models -- no attempt to discover how well their output corresponds to the real world, and what modeling aspects have contributed to better or worse correspondence -- then theoretical problems go unchecked.

Incidentally, is there any private sector demand for CGE forecasting? Are the only buyers institutions with weak performance incentives? said...


I do not know the answer to your question about who is buying CGE models. There are enough of them around that I would not be surprised if some simple ones were being used by some of the larger companies. DSGE models are strictly for central bankers and other outfits of that sort. said...

BTW, one other potential problem with these is that of multiple equilibria, although this is probably more of a problem for models of foreign exchange rates coming out of one of these than for your more garden variety CGEs.