Comments On "Business Cycle Anatomy"

George-Marios Angeletos, Fabrice Collard, and Harris Dellas released a working paper "Business Cycle Anatomy." They did some econometric analysis to decompose the movements of various economic time series, tying the results to Dynamic Stochastic General Equilibrium (DSGE) theory. I think the paper will be interesting to anyone with an advanced knowledge of economic theory.

The article may be of interest to heterodox economists in that it appears to skewer some of the concepts from older DSGE models. The following text from the Abstract provides a teaser:

Our findings support the existence of a main business-cycle driver but rule out the following candidates for this role: technology or other shocks that map to TFP [total factor productivity] movements; news about future productivity; and inflationary demand shocks of the textbook type. Prominent members of the DSGE literature also lack the propagation mechanism seen in our anatomy of the data. Models that aim at accommodating demand-driven cycles under flexible prices appear promising. 

It should be noted that this does not imply a movement towards heterodox macro in the authors' view. From the concluding section.

All in  all, our  findings  suggest that the  flexible-price  core  of the mainstream, New Keynesian framework is problematic to start with, in the sense that this core is itself unable to accommodate the kind of non-inflationary demand shock we have documented in the data. Encouragingly, there is now a growing literature that attempts to accommodate demand-driven business cycles even without nominal rigidities and Phillips curves; see the references in footnote 3. [in the original text] We hope that the characterization of the data performed in the present paper will stimulate further research in this direction and help guide macroeconomic theory. 

If one wants to go into the long-standing post-Keynesian/neoclassical debate, one needs to decide whether the new models do in fact cover up the defects that were diagnosed in earlier generations of models. I have an obvious bias, but I will largely dodge the question (until I get around to writing Volume II of Recessions).

Overview of Analysis

The authors take a variant of existing econometric techniques and attempt to isolate "shocks" associated with a number of macro variables. This allows them to work towards the Main Business Cycle (MBC) shock, which they describe as follows.

The Main Business Cycle Shock. Consider the shocks that target any of the following variables over the business-cycle frequencies: unemployment, hours worked, GDP, and investment. These shocks are interchangeable in terms of the dynamic comovements (IRFs) they produce. Furthermore, any one of them accounts for about three-quarters of the business-cycle volatility of the targeted variable and for more than one half of the business-cycle volatility in the remaining variables, and triggers strong positive co-movement in all variables. The shock that targets consumption is less tightly connected in terms of variance contributions, but still similar in terms of comovements/IRFs. 

From a technical perspective, this starts to resemble a first principal component (which they discuss).

They examine the properties of these MBC shocks, and show that they do not resemble standard culprits like productivity shocks.

Post-Keynesian Perspective

I only glanced at the article, and I am a long way from being able to replicate the techniques. However, it looked interesting, and is somewhat theory-agnostic. It could just as easily be used to compare observed data to post-Keynesian models.

The interesting point of comparison is the notion of "shocks." The paper continues the tradition of assuming that the economy will follow some steady-state path, and then we superimpose "shocks" that generate a business cycle. This is essentially a mode of thought inherited from the analysis of linear models, where we approximate nonlinear models by looking at linearisations, and we then look at how small deviations from the central model trajectory act.

I cannot sneer too much at linear modelling; in my academic area of control theory, most of the useful results are actually from linear system theory. That said, engineers are very well aware of the domain of applicability of the linear models. The problem for economic theory is that all the interesting stuff is happening well away from the "linear steady state."

Either the Business Cycle Exists, Or It Doesn't

The reason why I view these results as an interesting curiosity while not some new fundamental approach one should drop everything else to pursue is that the results are exactly one would have expected from thinking about this topic for a short period of time.

We are looking at the business cycle. The definition of a business cycle is that many important economic variables (production, employment) will all move together in some coordinated fashion. Therefore, if a business cycle exists, we have to be able to extract something resembling a first principal component that does a good job of explaining the bulk of the variation in those variables.

There is no guarantee that a business cycle exists; one could write down a model where all economic variables largely vary at random. However, it is safe to say that such a model would be rejected by the data, as business cycles are pretty obvious to even a simple visual inspection of the data.

Academic economists want to paint economics as a scientific endeavor, and so they want to use "sophisticated" statistical techniques as much as possible. "Simple visual analysis" is not in the vocabulary. Since I am in a position of not caring about such issues, I just ask myself: is it worth doing the work? In this case, I just see this as smashing a walnut with a sledgehammer. (I may add a section in my book to discuss it, although it would likely appear in Volume II.) That said, anyone in industry or academia needs to sound sophisticated to get ahead in the world, so these techniques should have some utility. 

Uncertainty vs. Randomness, Expectations vs. Forecasts

The other leg of the post-Keynesian analysis is that this once again highlights some familiar differences in perspective.

• Uncertainty vs. Randomness. Neoclassicals use random processes to model uncertainty, whereas post-Keynesians emphasise the concept of fundamental uncertainty. That is, we cannot fit a probability distribution over forecast outcomes. The post-Keynesian stance poses a lot of issues for econometric work.
• Expectations vs. Forecasts. Neoclassical models assume that transactions are happening in spot and forward markets that are simultaneously cleared. The market clearing forward price matches the mathematical expectation of the probability distribution of that future price. (As matching interest rate expectation models.) These mathematical expectations are assumed to be the same thing as "forecasts" of the future by agents. The post-Keynesian argument is that prices are not determined in this fashion, and that we cannot substitute market forwards for forecasts in reality. 

In this case, the latter factor is probably a larger concern. Although post-Keynesian mathematical models typically have simple rules for the generation of forecasts, that does not mean that post-Keynesians ignore the role of beliefs about the future. Rather, they are extremely hard to model. The argument is that business cycles are generated by animal spirits: coordinated changes in beliefs about future prospects. The feedback loop between investment and profits (as seen in the Kalecki Profit Equation, link to primer) means that such shifts are self-fulfilling.

Therefore, from a post-Keynesian perspective, it is obvious that there should be a central business cycle factor. The only issue is what drives the shifts in that factor, and basic national accounting analysis says that fixed investment is key. As a result, if one is interested in the business cycle, one needs to dig into what drives fixed investment. My argument is that base neoclassical models eliminate all the important factors driving fixed investment, and improved models typically only add one while still ignoring the rest.

In summary, although one should probably keep an idea on some of these econometric techniques, I am not holding my breath waiting for any theoretical convergence between post-Keynesians and neoclassicals.

(c) Brian Romanchuk 2019