DSGE models — a total waste of time
from Lars Syll
While one can understand that some of the elements in DSGE models seem to appeal to Keynesians at first sight, after closer examination, these models are in fundamental contradiction to Post-Keynesian and even traditional Keynesian thinking. The DSGE model is a model in which output is determined in the labour market as in New Classical models and in which aggregate demand plays only a very secondary role, even in the short run.
In addition, given the fundamental philosophical problems presented for the use of DSGE models for policy simulation, namely the fact that a number of parameters used have completely implausible magnitudes and that the degree of freedom for different parameters is so large that DSGE models with fundamentally different parametrization (and therefore different policy conclusions) equally well produce time series which fit the real-world data, it is also very hard to understand why DSGE models have reached such a prominence in economic science in general.
Neither New Classical nor ‘New Keynesian’ microfounded DSGE macro models have helped us foresee, understand or craft solutions to the problems of today’s economies. But still, many young academic macroeconomists want to work with DSGE models. That certainly should be a very worrying confirmation of economics — at least from the point of view of realism and relevance — becoming more and more a waste of time.Why do these young bright guys waste their time and efforts? Besides aspirations of being published, I think maybe Frank Hahn gave the truest answer back in 2005 when interviewed on the occasion of his 80th birthday, he confessed that some economic assumptions didn’t really say anything about “what happens in the world,” but still had to be considered very good “because it allows us to get on with this job.”
Hahn’s answer reminds me of an episode when, thirty years ago, Phil Mirowski was invited to give a speech on themes from his book More Heat than Light at my economics department in Lund, Sweden. All the mainstream professors were there. Their theories were totally mangled and no one — absolutely no one — had anything to say even remotely reminiscent of a defence. Being at a nonplus, one of them, in total desperation, finally asked: “But what shall we do then?”
Yes indeed — what shall one do when the emperor has turned out to be naked?
Leave a comment
—– look inside —– $5.94 / $20.00
—– look inside —– $4.90 / $8.00
—– look inside —– $15.99
—– look inside —– $5.99 / 12.99
—– look inside —– $5.93 / $12.99
—– look inside —– $4.97 / $9.90
—— Ugarteche, Puyana and Madi ——
Gerson Lima / Maria Alejandra Madi
Edward Fullbrook and Jamie Morgan
————— Michael Hudson ————–
Maria Alejandra Madi / Jack Reardon
————- Edward Fullbrook ————-
—————— Steve Keen —————–
————— Richard Smith —————
————– Gustavo Marques————
– Victor Beker and Beniamino Moro –
————– Lars Pålsson Syll ————-
—————– Stuart Birks —————-
Edward Fullbrook and Jamie Morgan
———— Armando Ochangco ———-
Shimshon Bichler / Jonathan Nitzan
————— Mauro Gallegati ————–
————— Herman Daly —————-
————— Asad Zaman —————
—————– C. T. Kurien —————
————— Robert Locke —————-
Real-World Economics Review Blog 
One thing we could do is go back and reread Keynes’ General Theory with an open mind: As was noted by Keynes: “The ideas which are here expressed so laboriously are extremely simple and should be obvious. The difficulty lies, not in the new ideas, but in escaping from the old ones, which ramify, for those brought up as most of us have been, into every corner of our minds.” Since the 1960s economics students have been forced to read that book, and when they couldn’t find any of things they were told they were supposed to find in it they put it aside and moved on to concentrate on the nonsense they were being taught. It’s time to look at it again: https://rweconomics.com/CIKGT.pdf.
This confused matter is associated with the apparent inability to take suitable assumptions after finding the necessary axioms and accurate definitions that are the beginning parts of this subject. Rather than struggle with those prior badly biased and intuitive ways, it is better to start over again and how to do this is provided in my short working-paper SSRN 2867751 “Einstein’s Criterion Applied to Logical Macroeconomics Modelling”. It is in answer to the question of how many different KINDS of activities are there in our business social system and how are they connected? (That it is a system should be obvious–many parts of nature are so arranged.)
Part of the problem is in recognizing the different aspects of the macro-system from that of microeconomics. When examined in this newer way it becomes clear that what we are dealing with is actually a true science, and one that replaces the past pseudo-science that so many of the experts claim is all that it comprises, thereby increasing the amount of complexity and confusion.
People who think economics stopped with Keynes are almost as irritating as people who use the DSGE models, rightly castigated by Lars Syll. If we want to know how people make decisions we have to ask them and look at what they do in case studies. We then have to incorporate those insights into agent-based models that are not designed to converge on a pre-specified equilibrium. We have to check whether the models are compatible with relevant data sets by statistical analysis and then we have to simulate such models to explore their properties and see how far they provide insights into the real world. Pioneering work of that sort was done in Sweden by Gunnar Eliasson. It is too much like hard work for most economists who look for short cuts, though ABMs are receiving more attention now as the DSGE folly fades (slowly).
Lars, why not change the subject and tell us what you think are the limitations of the ABM school? We know what you think of the post-Lucas neo-classicals. And we all, largely, agree.
Agent-based models (ABM) are formal models usually constructed using mathematical programming and performing simulations and ‘artificial experiments’ with the intention of being able to (more explicitly than in conventional mainstream game theory) describe aggregate effects and dynamics of interacting individuals and socio-economic structures without standardly having to assume equilibria, non-emergence, Walrasian auctioneers, representative agents, rational expectations, etc., etc..
Agent-based models come in different degrees of realism and are usually conceptualised as different kinds of self-organising complex systems. But one thing they all have in common is reliance on mathematical formalism. In essence the agent-based modelling endeavour in macroeconomics is an attempt at providing new alternative mathematical models where many of the bizarre and ridiculous known-to-be ‘unrealistic’ assumptions in standard DSGE models are replaced with other less ‘unrealistic’ assumptions. But the idea that mathematical modelling as such is always appropriate to apply is never seriously questioned. And that’s where I find it hard to follow. One set of mathematical tractability assumptions are substituted for another. But what if the mathematical modelling in itself is the problem? What if the use of mathematical-formalistic modelling in itself biases your research efforts in specific directions? If it is the mathematical-formalistic approach in itself that is the problem, we only end up with different models based on the same unquestioned mathematical modelling strategy. From my own critical-realist perspective I can’t see that mathematical modelling is the self-evidently appropriate way to perform analyses of societies and economies. The kind of ‘closures’ demanded of the target systems for warranting the analyses, I would argue, simply often aren’t there.
As a critique of mainstream economics, yours truly fully appreciates the work done by people like Alan Kirman and other ABM oriented economists. But although their alternative agent-based models in many ways are superior to the more traditional mainstream ‘Walt Disney’ kind of models, I am not convinced that their unquestioned attachment to mathematical-formalist modelling is the right way to move forward in making economics a more realist and relevant science.
I appreciate Lars’ reply to my challenge about ABM modelling
Broader normative questions about how society should be organised involve philosophical issues that are only partly accessible to logic. I do not think many people would claim that mathematical modelling is usually appropriate or possible to such questions. But ABM models, based as they are on trying to represent how people behave in existing contexts and institutions, are not used to address such big questions so are not generally misapplied. You can argue that the big normative ones are the most important questions. That may well be but it does not follow that humbler questions are not worth posing.
If , for example, we are looking to use the usual tools of government macroeconomic policy, i.e. the extent of a government deficit or surplus or the level of policy interest rates, we need quantification. We have to calibrate and quantify the state of the macroeconomy and have some idea of the effect of policy instruments in the current state. For that mathematical modelling is necessary though it will be of a probabilistic kind and subject to error.
When we are dealing with phenomena that can be measured, however incompletely, the ability to make quantitative statements generally betokens a higher degree of understanding than the ability to make only qualitative ones. I agree with Lars that such a level of understanding is not always possible and should not be counterfeited when it is not genuinely achievable. He should agree with me that it is nonetheless the legitimate aspiration of any discipline dealing with phenomena subject to cardinal measurement, as is much economic activity.
I don’t think it is true to say ABM substitutes “one kind of mathematical tractability for another”. It assumes the possibility of measurement but then traces outcomes by numerical methods – comprehensive trial and error essentially. Computing power replaces the need for mathematical tractability.
Incidentally, the process analysis in George Blackford’s paper (see his link) is convincing. To go further and fully understand dynamics when different forces are at work empirical study is necessary. Some quantification is also necessary if we have aspirations to control the phenomena. If we are not satisfied with positing macro relationships and testing them against data, that is if we insist on microfoundations, I see no alternative to the route pioneered by Eliasson.
I would note that an examination of the of the way in which this kind of process analysis can be used in a more comprehensive model can be found in:
Mr-Keynes-and-the-Neoclassics-A-Reinterpretation.pdf (cmjpublishers.com)
In a simple model one thing follows another. In reality with many sectors and many actors, similar processes will be at work in different places and they will not be synchronised, so one thing will merge into another rather than following simply. Eddies and feedbacks may be created. It is one thing for theory to tell you the sign of a parameter but it can’t usually tell you its size and that is likely to be all-important in determining outcomes. Outcomes that looked inevitable in simple theory may turn out to be rather more contingent in reality . There is no way to sort that out from first principles. You have to try and find the relevant data and examine it. Armchair theorising is a great starting point but a very poor place to stop. One of the greatest failings of DSGE is that it is so tightly specified and so constrains reality that it just cannot accept data and evidence that is clearly relevant. People will tell you what they expect if you ask them. If you “know” their expectations are “rational” you can’t and don’t ask.
So Lars, what about ABM?