Critics have had a field day the past couple of years claiming that Economics is not up to the task because its models are too abstract. Macroeconomics has been especially affected because the trigger of the bandwagon was a macroeconomic event, the now Great Recession. I have discussed a few bad attempts at criticism, which were usually bad because ill-informed and because they could not offer any viable alternative.
The latest salvo comes from Hashem Pesaran and Ron Smith. They have several arguments. The first is that macroeconomic modeling of the DSGE brand insists too much on internal consistency and should allow more degrees of freedom to fit the data. Pesaran and Smith should first specify what the goal of the model is before criticizing the approach. If it is short-term forecasting, then go ahead with a purely statistical approach on macro data. If you want policy advice, you need something that withstands the Lucas Critique, and strong micro-foundations is then the way to go. But without a given purpose, any criticism is moot.
Pesaran and Smith, given their track record, are of course strong advocates of purely statistical methods. Throw every possible series in a regression, and see what sticks. I am not saying this cannot be useful, it allows to establish relationships in the data and I regularly report on such results, but this does not allow you to explain things. For this, you need some structure and theory provides you that. This brings me to the title of this post. There appears to be some disagreement about the meaning of the word "model." To me, model is a set of relationships established by theory that can then be used on data, for policy experiments, etc. For Pesaran and Smith, a model is a set of aggregate data series that are used in a statistical analysis of some kind (VAR, non-parametric, etc.). If we cannot agree what we are talking about, of course there will be endless and fruitless discussions.
For example, they are not the first to criticize DSGE models for failing to include housing, finance and the external sector. Well, models (the way I see them) are abstractions, and you do not want to include everything them, or you cannot understand, interpret or do something useful with. It is so across all sciences. You build a model to answer a particular question, and you give it the necessary bells and whistles. The fact that most DSGE models did not include housing and bank liquidity is not a failure of DSGE modelling, it is a failure of recognizing what questions could be important in the future and this is damn hard to do properly.
Pesaran and Smith's solution to what they call the straightjacket of DSGE is to throw all these missing variables in a regression. Essentially, they want to bypass the discipline that theory imposes by letting the data speak. Again this is OK if you want to explore and find relationships, but this is not going to be very useful if you want to explain what is going on. Specifically, they advocate using vector autoregressions (VAR). As they complain that DSGE use representative agents when heterogeneity matters, they call for the use of data from several countries in the VAR, a rather strange argument, but I suppose this is because they are limited to aggregate data (and they neglect all the DSGE models using household level data...). In a statistical sense, the big problem is now that one quickly runs out degrees of freedom, as one has only so many time periods, and every additional variable eats degrees of freedom at a quadratic rate (times the number of lags). The other problem is that interpreting the resulting errors ("shocks") becomes difficult. One is then limited to vague notions like demand and supply shocks, much like in factor analysis. But at least Pesaran and Smith acknowledge that theory can be useful in selecting, say, long-term restrictions.
PS: I gave much of the same arguments in the discussion of a paper by David Hendry. Pesaran appears to be more knowledgeable of DSGE and is more willing to use theory to guide empirics. Unfortunately, Pesaran has the same habit of abusing self-citations, 12 out of 32.