Further Thoughts on DSGE Models
A number of economists have recently written about the pros and cons of dynamic stochastic general equilibrium models (DSGEs). (Among them, and in addition to my own piece, are Narayana Kocherlakota, Simon Wren-Lewis, Paul Romer, Steve Keen, Anton Korinek, Paul Krugman, Noah Smith, Roger Farmer, and Brad Delong.)
Here are my reactions to the debate:
I believe that there is wide agreement on the following three propositions; let us not discuss them further, and move on:
- Macroeconomics is about general equilibrium.
- Different types of general equilibrium models are needed for different purposes. For exploration and pedagogy, the criterion should be transparency and simplicity, and for that, toy models are the right vehicles. For forecasting, the criterion should be forecasting accuracy, and purely statistical models may, for the time being, be best. For conditional forecasting, i.e. to look for example at the effects of changes in policy, more structural models are needed, but they must fit the data closely and do not need to be religious about micro foundations.
- Partial equilibrium modelling and estimation are essential to understanding the particular mechanisms of relevance to macroeconomics. Only when they are well understood does it become essential to understand their general equilibrium effects. Not every macroeconomist should be working on general equilibrium models (there is such a thing as division of labor).
I see two propositions as more controversial:
- The specific role of DSGEs in the panoply of general equilibrium models is to provide a basic macroeconomic Meccano set, i.e. a formal, analytical platform for discussion and integration of new elements: for example, as a base from which to explore the role of bargaining in the labor market, the role of price-setting in the goods markets, the role of banks in intermediation, the role of liquidity constraints on consumption, the scope for multiple equilibria, etc. Some seem to think that it is a dangerous and counterproductive dream, a likely waste of time, or at least one with a high opportunity cost. I do not. I believe that aiming for such a structure is desirable and achievable.
- The only way in which DSGEs can play this role is if they are built on explicit micro foundations. This is not because models with micro foundations are holier than others, but because any other approach makes it difficult to integrate new elements and have a formal dialogue. For those who believe that there are few distortions (say, the believers in the original real business cycle view of the world), this is a straightforward and natural approach. For those of us who believe that there are many distortions relevant to macroeconomic fluctuations, this makes for a longer and messier journey, the introduction of many distortions, and the risk of an inelegant machine at the end. One wishes there were a short cut and a different starting point. I do not think either exists.
The Way Forward
If one does not accept the two propositions above, then DSGEs are clearly not the way to go—end of discussion. There are plenty of other things to do in life.
If, however, one does accept them (even if reluctantly), then wholesale dismissal of DSGEs is not an option. The discussion must be about the nature of the micro foundations and the distortions current models embody, and how we can do better.
Do current DSGEs represent a basic Meccano set, a set that most macroeconomists are willing to use as a starting point? (For Meccano enthusiasts, the Meccano set number is 10. The Meccano company actually has also gone through major crises, having to reinvent itself a few times—a lesson for DSGE modelers.) I believe the answer to this is no. (I have presented my objections in my first piece on the topic.) So, to me, the research priorities are clear:
First, can we write down a basic model most of us would be willing to take as a starting point, the way the IS-LM model was a widely accepted starting point earlier in time? Given technological progress and the easy use of simulation programs, such a model can and probably must be substantially larger than the IS-LM but still remain transparent. To me, this means starting from the New-Keynesian model, but with more realistic consumption- and price-setting equations, and adding capital and thus investment decisions.
Second, can we then explore serious improvements in various dimensions? Can we have a more realistic model of consumer behavior? How can we deviate from rational expectations, while keeping the notion that people and firms care about the future? What is the best way to introduce financial intermediation? How can we deal with aggregation issues (which have been largely swept under the rug)? How do we want to proceed with estimation (which is seriously flawed at this point)? And, in each case, if we do not like the way it is currently done, what do we propose as an alternative? These are the discussions that must take place, not grand pronouncements on whether we should have DSGEs or not, or on the usefulness of macroeconomics in general.