In the article, Lavoie discusses how the mainstream wishes to justify Dynamic Stochastic General Equilibrium (DSGE ) modelling on the basis that they work well during an expansion. He writes:
The central line of defense of orthodox economists is that the criticisms of orthodox macroeconomic theory are misdirected because, as put forth by Thomas Sargent, its models ‘were designed to describe aggregate economic fluctuations during normal times … not during financial crises and market breakdowns’ (Rolnick, 2010). It is claimed that the main tool of central bankers, the DSGE model, was doing very well as long as the American or European economies were moving along the lines of the Great Moderation. There seems to be a consensus about the limited range of application of DSGE models. Charles Goodhart (2009, p. 352) writes that these models ‘were, by construction, fair weather models only’. [... deleted: Bernanke comments - BR] Blanchard (2014, p. 29), has also claimed that everyone knew that these models were useless when the conditions of the Great Moderation did not hold: ‘we all knew that there were “dark corners” – situations in which the economy could badly malfunction. But we thought we were far away from those corners, and could for the most part ignore them’. Blanchard (2014) still believed then that with improved models the economy will stay away from what he calls ‘the dark corners’, where mainstream models can provide no light. But this sounds like wishful thinking. There is a need for theories or models that take these dark corners as genuine components or likely possibilities.Marc Lavoie is obviously not a fan of DSGE models. However, he writes in a proper, polite, and scholarly manner, and so a casual reader could come away with too charitable a view of DSGE macro. The idea that classical economics describes "normal times," and that Keynesian economics only describes a downturn was a standard way of reducing the importance of Keynes' contribution to the creation of macroeconomics. One could mistakenly extrapolate that interpretation as applying to DSGE macro (since it is a direct descendant of classical macro).
- DSGE macros are entirely useless in describing a downturn.
- DSGE macros underperform drawing straight lines though data during an expansion.
In other words, sticking lines through time series [*] -- a generalisation of technical analysis used in financial markets -- dominates DSGE macro. The advantages of sticking straight lines through data are:
- Very few people are foolish enough to believe that we can do straight line extrapolation of economic time series forever. Extrapolating DSGE macro predictions? Not so much.
- You do not need to pretend to understand stochastic calculus in order to stick a straight line through data.
- The odds that someone can put a straight line through data properly are a lot higher than getting them to run all of the statistical mumbo jumbo needed to crank out a DSGE model.
- We will not have the situation of trained academics debating what straight lines "really mean." (See below.)
(As an example of debating what DSGE models "really mean," a good recent example is "Cheshire Cats and New Keynesian Central Banks" by Professor Nick Rowe. That article is likely to be the subject of another rant -- I think I agree with the DSGE orthodoxy in their interpretation. In any event, I just want to note that in addition to Professor Rowe, Professor Brad DeLong agrees that the standard interpretation of DSGE models make no sense.)
There is no economic regime in which the usage of DSGE modelling methodology makes sense.
Appendix: DSGE Models are Nonlinear -- Yeah, Whatever
Some DSGE model purists might object to my statement that representative agent DSGE models are dominated by linear regressions, on the basis that the models are nonlinear (and hence not a subset of linear models).
This objection relies upon taking the ridiculous claims of DSGE authors at face value. Sure, they write down really cool nonlinear models that look like they belong on a blackboard on a Far Side cartoon. The reality is that those nonlinear models are just a bait-and-switch used to make the models look less foolish than they really are. There is no chance whatsoever of solving the nonlinear system, and so the authors pretend that they can linearise them. All of the statistical techniques are applied to the linear system, and so that is the only version of the model which matters.
[*] More generally, we need to develop a regression model to generate our "straight line" model. A regression model involves mucking around with arbitrary linear models, in order to fit observed data. Since the set of linearised DSGE models is a subset of all possible linear models, the definition of "subset" tells us that we should expect that a DSGE model will have a worse "fit" (however we define "fit") than an arbitrary linear model; at best, it can only tie.
Post a Comment
Note: Posts are manually moderated, with a varying delay. Some disappear.
The comment section here is largely dead. My Substack or Twitter are better places to have a conversation.
Given that this is largely a backup way to reach me, I am going to reject posts that annoy me. Please post lengthy essays elsewhere.