I ran across the IMF working paper “Reading the Stars” by Peter D. Williams, Yasser Abdih, and Emanuel Kopp. I have only scanned the paper quickly, but it appears to be in line with similar papers that I previously read. My argument here is that this paper — or content very similar to it — would be the best starting chapter for a book that attempts to explain Dynamic Stochastic General Equilibrium (DSGE) macro theory to readers who have existing math skills. It is effectively a survey of the useful empirical applications of DSGE macro.
This is the introductory article in a series that will attempt such a feat. I am attempting to explain why this paper in particular is the best starting point, but without the technical arguments. However, the technical arguments will be bewildering if we do not know what the objective of the exercise is.
Why this article? It parallels exactly how I would describe my thinking on DSGE macro. My thinking over the years has not changed from a technical perspective, but how I express my thinking has changed dramatically. Furthermore, the implications I draw are different. Given that readers are aware of my disdain towards DSGE macro, the argument is that if we accept that there are fundamental methodological issues with the theory in this article, we easily understand the issues with the vast DSGE literature.
Before People Get Offended
Neoclassical economists (as I term them) go out of their way to be offended by heterodox critiques. I want to make it immediately clear that I have zero interest in almost all “economic” research, only what I would call “DSGE macro theory.” Since people cannot be expected to understand where I draw the lines dividing research, I will quickly categorise neoclassical macroeconomic research, where only the first two categories are of interest.
The first thing to note that I am discussing the theoretical contributions of papers, and not the authors, or what papers are cited. I would then divide the contributions into high level categories.
- What might be termed “teaching models” or “toy models” that use DSGE models that have surface similarities to a Real Business Cycle model core. (Note that not all DSGE models fit this bill, it is possible to have option-pricing models that are classified as DSGE models. I am interested in macro models, not option pricing.) The model is derived, and/or its properties are tied to observed economic data.
- Attempts to fit a model in the previous class of papers to real world data in a serious fashion. “Reading the Stars” fits into this category, but it can include papers where the model is laid out and fit to to data. Papers describing benchmark DSGE models fall under this category.
- Empirical/econometric papers — which are not in the category of interest. What divides these papers from the previous two is the following question: could the key contributions of this paper be applied to the analysis of a model that is not a DSGE macro model? Very simply, applied mathematics is applied mathematics.
- Everything else – also not in the categories of interest.
Unlike many heterodox economists, I have zero interest in writing about what is going in Economics 101 textbooks (other than mocking them), microeconomics, theory of competition, whatever. Not my theoretical turf. As such, I am not generalising about “mainstream economics” or even “mainstream macroeconomics.”
Finally, I will note that I am using “neoclassical” to refer to the classes of DSGE models of interest, as well as the users of the models. This may not be the best description, but I am not going to use a clunkier description just to please my two readers who are bugs on the history of economic thought. (I avoid “mainstream” since it is too vague, as it arguably applies to the musings of the more numerous bank economists, not a handful of academics.)
Why This Paper in Particular?
I chose “Reading the Stars” as a starting point as it is the closest to what I see as the key issues raised by the application of DSGE macro. There have been similar papers in the past, and my prediction is that there will be similar papers in the future. The issue is how to relate observed economic data to DSGE models (or any mathematical model of the economy). No amount of econometric wizardry — which is what is used to judge academic originality — can offset the fundamental modelling issues faced. That is, even if some other authors can advance the econometric analysis, that is just mathematics. So long as the mathematics is properly described, any trained researcher can pick up those innovations. Meanwhile, the paper seems to be the cleanest exposition available. It is clear that mainstream economic journals currently aim to maximise obfuscation.
However, I want to underline that I have just scanned the paper at the time of writing this article. I am not offering any representations about the quality — it is entirely possible that the authors botched their analysis. If that is the case, it just means is that we need to find a similar paper that did the analysis properly. Since my concern is not the particular econometric techniques used, this is not a significant setback.
What is the Paper About?
The paper discusses the techniques used to estimate the “star variables” — e.g., r* (the “natural rate of interest”) and u* (something akin to NAIRU). These are “hidden variables” that have to be estimated from observed data.
The key is that all of these variables are estimated at the same time, which is a point that is often obscured, which is one reason to target this paper.
Although it is safe to say that the vast majority of readers will be interested in the particular estimates produced, that is exactly the wrong lesson to draw from this paper. Until the existing DSGE macro approach is abandoned, there will always be new estimation techniques for these variables, and these estimates will presumably be “better” in some sense (since they would not get published otherwise). Instead, we need to look at the underlying dynamics of any estimation technique, and how it relates to the real world. (This explains why any paper in this genre is largely interchangeable with any other one for my purposes.)
Why Do We Care?
When discussing DSGE macro, people — including myself, when I was slightly younger and foolish — jump up and down discussing the flaws with the papers in what I labelled as category 1 — teaching models. This is a giant red herring. Those papers exist solely as a means of producing large numbers of articles needed to jump hoops for gaining credentials. They are only useful if the mathematical models can be related to the real world in a systematic fashion.
This is not being done. Instead, these models are used to demonstrate that an optimising model can produce some desired behavioural dynamic. For example, any shnook can observe that if the banking system is stressed, it will tend to reduce economic growth. Since the core RBC model does not have a banking system, this causes concerns. So what is done is that a DSGE model is invented that has a “banking system,” and lo and behold, when it is “stressed,” output falls. The issue is that this model is largely crippled, and can only be used to demonstrate this desired outcome.
This is how teaching models work. I cannot throw stones:
my Python SFC models framework just generates teaching models; going beyond that would require heavy user intervention (i.e., rebuilding from scratch).
I luckily avoided economics training, instead my training was in the analysis of mathematical models and how they can be applied to the real world. From my perspective, it is extremely obvious that there are an infinite number of mathematical models that are similar to the RBC core, and this infinitude of models can generate practically any behavioural responses. Creating a literature that is effectively an attempt to enumerate this infinite class of models is not the best use of one’s limited time on Earth.
For a model to go beyond usage in parables, it has to be tied to the real world. It has to make quantifiable predictions, and those predictions needed to be tested out-of-sample. That is where “category 2” papers come in, and why they are the most important within the DSGE literature.
Intractable Problems and Head-Bashing
I left academia for a number of reasons, and one of those reasons was that I specialised in the wrong area. I was working in nonlinear control theory, and I discovered first hand the issues with intractability. Roughly speaking, if we looked at the interesting things that we could do with nonlinear control theory, they all collapsed to being able to solve a certain class of partial differential inequalities. However, the “curse of dimensionality” meant that we would not be able to solve these for interesting systems. Basically, no matter where you started in nonlinear control theory, once you followed the logic, you hit the same brick wall. I either had to re-invent myself in a new area of academia — or get a job in finance. (No prizes for guessing which option I picked.)
DSGE macro theory (as I define it) faces a different brick wall (although the “curse of dimensionality” probably shows up as well as a kicker). The “stars estimation” problem shows us how neoclassicals are attempting to get through that brick wall.
If one is a fan of neoclassical macro, one can look at the ever-rising publication counts, and argue that there is progress.
However, if one is skeptical. those publication counts are just researchers moving to a slightly different position of the same wall, and smashing their heads into it. There is no reason to be concerned about the details, the wall will still be there when we get back.
The obvious thing to do is to do an end-run around the wall. It is entirely possible that this will happen, and given the ability of neoclassicals to produce a firehouse of articles on any topic, the end run might be over rather quickly. That said, doing so would require admitting that the wall exists in the first place.
To Be Continued…
Now that I have described why I want to zero in on this paper, I will now work on related articles. They are meant to be stand-alone, so there will not be a single narrative structure running through them.
Since I need to explain why the paper is useful first, my readers who like criticism of DSGE models may need to wait for a long time. Although I am critical of DSGE macro, it exists for a reason, and there are very plausible reasons for the existing structures. We need to understand the strengths before the weaknesses can be appreciated.
These articles are likely to be tied together into a chapter in Recessions: Volume II. (My MMT primer manuscript is nearly done, I am just sitting on it before the final editing pass.)