Friday, August 12, 2016

The Future of DSGE Models in Macroeconomics

One of the hardest problems in studying the macroeconomy is that time keeps advancing. You can't go back to, say, 2001 or 2009, not enact the Bush tax cuts or the Obama economic stimulus, and then re-run the economy and see what happens. Instead, researchers end up comparing effects of seemingly similar policies enacted at different times--but the policies and the circumstances are never quite identical, so room for dispute remains. Indeed, disagreements among macroeconomists are nearly proverbial. "Macroeconomists have predicted nine of the last five recessions." "Two macroeconomists, five opinions." "Economists are the experts who explain why the prediction they made yesterday didn't come true today."

I sometimes receive notes from readers asking for a sense of why macroeconomists disagree.  Olivier Blanchard opens up some of the central issues for useful discussion in a short and readable paper, "Do DSGE Models Have a Future?" written for the Peterson Institute for International Economics (Policy Brief 16-11, August 2016).

For the uninitiated, DSGE models of the macroeconomy are a method that is both well-established and the stuff of continuing controversy. DSGE stands for "dynamic stochastic general equilibrium model," which represents a broad class of macroeconomic models. In the jargon, "dynamic" means that the models show the evolution of a (hypothetical) economy over time. "Stochastic" means that the models show how the economy would respond if certain shocks occur, whether the shocks involve policy choices or economic events (like a rise or fall in the rate of productivity growth). "General equilibrium" means that these models don't look at the macroeconomy one sector at a time--say, first consumption, then investment, then foreign trade--but instead try to take all the interactions of these sectors into account.  Blanchard describes the models in this way:
"For those who are not macroeconomists, or for those macroeconomists who lived on a desert island for the last 20 years, here is a brief refresher. DSGE stands for “dynamic stochastic general equilibrium.” The models are indeed dynamic, stochastic, and characterize the general equilibrium of the economy. They make three strategic modeling choices: First, the behavior of consumers, firms, and financial intermediaries, when present, is formally derived from microfoundations. Second, the underlying economic environment is that of a competitive economy, but with a number of essential distortions added, from nominal ties to monopoly power to information problems. Third, the model is estimated as a system, rather than equation by equation in the previous generations of macroeconomic models. ... [C]urrent DSGE models are best seen as large scale versions of the New Keynesian model, which emphasizes nominal rigidities and a role for aggregate demand."
Blanchard gives four main concerns about DSGE models along with some thoughts about each one. Thus, he writes:
There are many reasons to dislike current DSGE models. First: They are based on unappealing assumptions. Not just simplifying assumptions, as any model must, but assumptions profoundly at odds with what we know about consumers and firms.  ... Second: Their standard method of estimation, which is a mix of calibration and Bayesian estimation, is unconvincing. ... Third: While the models can formally be used for normative purposes, normative implications are not convincing. ... Fourth: DSGE models are bad communication devices. A typical DSGE paper adds a particular distortion to an existing core. It starts with an algebra-heavy derivation of the model, then goes through estimation, and ends with various dynamic simulations showing the effects of the distortion on the general equilibrium properties of the model. "
You can read the details of Blanchard's responses in the paper, but I'd characterize his overall Blanchard's view of DSGE models seems to be negative, ambivalent, and  positive all at the same time. He writes: "I see the current DSGE models as seriously flawed, but they are eminently improvable and central to the future of macroeconomics." A snippet of his more detailed answer like this:
The pursuit of a widely accepted analytical macroeconomic core, in which to locate discussions and extensions, may be a pipe dream, but it is a dream surely worth pursuing. If so, the three main modeling choices of DSGEs are the right ones. Starting from explicit microfoundations is clearly essential; where else to start from? Ad hoc equations will not do for that purpose. Thinking in terms of a set of distortions to a competitive economy implies a long slog from the competitive model to a reasonably plausible description of the economy. But, again, it is hard to see where else to start from. Turning to estimation, calibrating/estimating the model as a system rather than equation by equation also seems essential. Experience from past equation-by-equation models has shown that their dynamic properties can be very much at odds with the actual dynamics of the system. 
It's worth unpacking this a bit. Blanchard's comment that the DSGE approach "may be a pipe dream, but it is a dream surely worth pursuing," is not calculated to inspire confidence in the results of such studies! This intellectual agenda involves modelling activities of real-world economic actors, including various assumptions and some combination of rational choice and behavioral economics, involves many possible choices. The selection of possible frictions like monopoly power, wages and prices which adjust in a sticky manner, the formation of expectations, the issue raised by financial markets, all adds another set of possible choices. The question of how to get a workable quantatitive number out of this model involves choosing some plausible values from other studies (that is, "calibrating" the model) and what parts of the model to estimate using data involves still more choices.

In addition, Blanchard discusses how DSGE modelling needs to be open to new insights from behavioral economics, from the use of big data, from issues about problems that can arise in financial markets, and more. He also suggests: "At one end, maximum theoretical purity is indeed the niche of DSGEs. For those models, fitting the data closely is less important than clarity of structure." This comments is not calculated to inspire confidence in the results of such studies either. He suggests that there is also a need for one set of related-but-different studies for policy purposes, and another set of related-but-different models for puruposes economic forecasting, and still other lessons that are most accessible through simpler ad hoc models (like the IS-LM model from intermediate-level macro textbooks).

In a way, what macroeconomists have been learning in the last few decades is to reach a deeper understanding how many different ingredients might be included in a macroeconomic model. But no model can look at everything at once, so macroeconomists are always trying to figure out which ingredients matter most. My own takeaway is that DSGE models will continue to matter a lot to high-powered researchers in macroeconomics, like Blanchard. But for the rest of us, the task is to keep track of how insights from those models filter down through the research literature and become practical lessons that can be explained and applied in more stripped-down contexts.