Networks and the Macroeconomy

That is the next NBER macro session, the authors are Acemoglu, Akcigit, and Kerr, the pdf is here, and here is the abstract:

The propagation of macroeconomic shocks through input-output and geographic networks can be a powerful driver of macroeconomic fluctuations. We first exposit that in the presence of Cobb-Douglas production functions and consumer preferences, there is a specific pattern of economic transmission whereby demand-side shocks propagate upstream (to input supplying industries) and supply-side shocks propagate downstream (to customer industries) and that there is a tight relationship between the direct impact of a shock and the magnitudes of the downstream and the upstream indirect effects. We then investigate the short-run propagation of four different types of industry-level shocks: two demand-side ones (the exogenous component of the variation in industry imports from China and changes in federal spending) and two supply-side ones (TFP shocks and variation in knowledge/ideas coming from foreign patenting). In each case, we find substantial propagation of these shocks through the input-output network, with a pattern broadly consistent with theory. Quantitatively, the network-based propagation is larger than the direct effects of the shocks, sometimes by severalfold. We also show quantitatively large effects from the geographic network, capturing the fact that the local propagation of a shock to an industry will fall more heavily on other industries that tend to collocate with it across local markets. Our results suggest that the transmission of various different types of shocks through economic networks and industry interlinkages could have first-order implications for the macroeconomy.
As I am doing today, my live-blogging will be in the comments of this post…

Comments

This is a paper about how shocks propagate through an economy. I find it notable for being outside the framework of traditional RBC comovement structures, but for me that makes it all the more interesting. There is input-output analysis and geographic networks as well. Typical Acemoglu eclecticism, and a bunch of mechanisms thrown into the bin. But do note, that will make it a little hard to see how this squares with standard macro results on transmission...

The name "Leontief" is up on the Powerpoint...the shocks are from the supply side by the way. The presentation starts with matrix algebra, rather than utility maximization, another deviation from the standard CGE approaches. Can a paper like this have a big impact on macro? Or is the field too narrow in its orientation? There are now some demand-side shocks on the board too. I wonder what is the implicit assumption to make demand and supply side shocks separable?

Speaking of networks, Cardinal Francis George just passed away, at age 78.

Will network analysis ever have a really big impact on economics? That is part of the question behind this paper.

Did neurons evolve twice?
https://www.quantamagazine.org/20150325-did-neurons-evolve-twice/

And there is some pretty nice art in this hotel, including two Philip Guston lithographs (?), just outside the room of the conference.

I'd like more of an overview of how the main mechanisms are supposed to work before seeing these details about the indexing of sectors and the like.

Orwell could not have said it better.

I give an introduction and some background to these kinds of models in my comment on Tim Harford's macroeconomics book at Cato. Here starting around 48:57

http://www.cato.org/events/undercover-economist-strikes-back-how-run-or-ruin-economy

Alex

The shocks being considered are import shocks from China, federal spending changes, TFP growth, and foreign patenting growth. I would strike the last of those. Now we move to the regressions, starting with China trade shocks. What is the propagation mechanism resulting from the lower prices paid by consumers? Does that have network effects? The underlying conceptual question is how we should sum up the full effects of a shock. I suspect I'll have the same question about foreign patenting growth.

I'd like to hear more upfront about how the size of each of these shocks is being measured/estimated. The Gestalt on how all this fits together is still not clear to me. And what are the underlying assumptions about substitutability?

And now it is Xavier Gabaix up to discuss...

If the complementarities are such as suggested by the paper, what kind of supporting micro evidence should we be able to find on cross-elasticities? Is it there?

Somewhere, @monetarypolicy8 has tweetstorms on this whole approach, do any of you know exactly where?
https://twitter.com/monetarypolicy8

The tweetstorm from A Macroeconomist (co-staring Pedro Serôdio) is collated here https://storify.com/ZacGross/networks via @zacgross

"What's the closest thing in the empirical micro literature which comes close to a disconfirming result for these macro network models?" is the question I would like to see posed up front. Do we find strong enough "multiplier" results in the micro literature?

Again, all remarks represent my own views, not those of anyone else, speaking or silent.

What do you all think of this overall approach?

Any approach that (potentially) eads to better/new uses of detailed data is a good one. DGSE models have a hit brick wall when all the results are so sensitive to primitives that cannot be measured with any reasonable degree of accuracy.

The biggest problem for me is the spacing between your comments. The model live blog update is one line of text, and the formatting of the comment section means there's an eleven line gap between each command. Even on my 27-inch Thunderbolt, I have to scroll an entire page to read ~150 words.

When an input-output model is running about, are we in some way downgrading the contribution of price theory to the results of our models?

Just asking...

Hasn't macro theory downgraded price theory forever? Unless you consider Arrow security pricing as standard price theory, but I don't remember it in the Alchian/Friedman/Stigler textbooks.

What kind of distribution for firm size helps get out more comovement out of a basic macro shock? How many independent economic sectors are there really? How many independent expectational assumptions? Exactly *where* in the economy do we wish to build in what is essentially an assumption that a "small numbers problem" is being faced?

A key question in my view.

Consider the scaling of the volatility of the New Zealand economy to the volatility of the U.S. economy. Which models is that scaling consistent with? Seems like a natural test for a lot of "small numbers" approaches. Interestingly, the smaller economies just don't seem that much more volatile, although they are much less diversified. Is that an argument for or against the "small number" approaches? Maybe the American economy isn't that diversified after all. Are the truly diversified economies based not on size, but rather on independent physical events, such as "hoof and mouth disease" vs. "how much lamb does the Middle East want to buy this year?"

And now Lawrence Christiano takes the floor...

Lots of talk of upstream vs. downstream in this paper.

The Dow is down 280 points, in part because of growing worries about Greece -- what is the best way to think about the network effect here?

This coming Monday could be quite a doozy in any number of equity markets.

Why are there no upstream effects in this model?

Christiano's comment is one of the highlights of this conference. And for the first time, the crowd laughs. Too bad I can't quote it.

And now Acemoglu with the response...

My sense of the session overall is this: there is not yet a strong enough community of researchers working on this method to pin down its strengths and weaknesses, and even how the model works in some fairly fundamental ways. The networks idea has potential, but I'd like to see some of the methodological and "what does this model really mean?" questions cleared up before proceeding further. Again, my key question is "which microeconomic results is this *least* consistent with?" This is macro, so there is always a micro part of the picture that isn't going to work out. What is that here? Inquiring minds wish to know. The implied predictions about the lack of change for upstream suppliers, following an economic shock, may be part of the answer to this question. The session has been very helpful in bringing out that point.

"The networks idea has potential, but I’d like to see some of the methodological and “what does this model really mean?” questions cleared up before proceeding further."

That is putting the cart before the horse. We know the real economy is a network. We know that information does not transmit instantaneously with 100% accuracy. We know that there are problems with composition in moving from micro foundations to macro predictions. There is nothing to be gained by putting up hurdles to starting to examine the properties and consequences of networks.

Ah, *he* speaks...

Re: "There is not enough researchers working this method [network analytics] to pin down its strengths and weaknesses..."

Are you kidding? Time to go back to school. You can enroll in Jackson's Social and Economics Networks course on Coursera. There's tons of literature, particularly on the macro and micro side. What do you think a stress test involves when you look at trading partners. You can download a free early version of Easley and Kleinberg's Networks, Crowds and Markets.

Agree with JC above.

For more reading, you can go here
https://sites.google.com/site/s2012nyunetworksmacro/

I think Acemoglu also has a syllabus in his MIT course on this as well.

You can download Acemoglu's Advanced Macro course with network analytics (syllabus and class notes) here:
http://economics.mit.edu/faculty/acemoglu/courses

Actually Bill, Tyler is correct that the networks approach has not been extensively used for macro. Sure there are some researchers in social and economic networks, but not doing what Acemoglu et al are up to.

As a personal aside, you really don't always have to be a hater.

Dan, I don't like the categorical statements "that there are not enough researchers working this method {network analytics} to pin down its strengths and weaknesses". My god, you are right: I hate that statement, because it is false. Don't confuse the statement with the author.

As for the claim that this is "new", I really beg to differ. It may be new for those who have not taken courses over the last 20 years, but for others it is not. As an adjunct, I attend graduate seminars in the business school marketing department, and get to listen to the papers presented by visiting scholars interested in getting a faculty position. Over the last ten years, there has been a constant stream of papers using network analytics in micro econ and marketing. What is interesting is to watch the faculty members who are very familiar with regressions, stats, econometrics, etc. having a difficult time critiquing the papers because they have not kept up by taking some network analytics courses. One friend--a nationally recognized econ expert on the theory of the firm--asked to borrow some network books so he could learn some of the mythology and also use it. He gave up, and went into administration.

There is a rich literature on the macro and micro network analytics. You can find it by going to the appendices to these books, to the papers, and by visiting the websites of Stanford, Caltech, MIT, Harvard, etc. and seeing what papers are being written by the faculty and graduate students. But, to say that this is nascent is just silly and unfounded in fact.

Dan, If you want to do an experiment on my claim that networks in macro is not new, just google "networks and macroeconomics" and follow the articles, dates and references, and then come back with support to say: "There is not enough researchers working this method [network analytics] to pin down its strengths and weaknesses..."

Dan, You can also go to Economides website on network in economics and read the extensive bibliography. Here is the link: http://www.stern.nyu.edu/networks/

The financial network thesis has been around at least since Fisher's and others' writings from the early 1930s explaining the Great Depression. My impression is that the new networks thesis began as an effort to somehow salvage RBC by starting with a postulated localized "productivity shock," but it has inevitably arrived to the same old financial spiral place.

Comments for this post are closed