Wednesday, October 29, 2014

That which is not dead doth sleeping lie (RBC edition)



The other day I had dinner with two prominent New Keynesian macroeconomists (whose names shall not herein be revealed). Both of them told me that the idea of technology shocks as the main driver of business cycles is effectively dead. At the same time, a number of Twitter people (Josh Hendrickson, Tony Yates, Ryan Decker, and Pedro Serodio) asserted that NK is the dominant school of macro. Other macroeconomists, like Chris House, have made the same claim.

I just don't think this is true at all.

Now, it is true - to my knowledge, anyway - that central banks use NK-style models (like Smets-Wouters), and don't use models where tech shocks are the main driver.

Also, I've recently seen people like Patrick Kehoe, Mark Bils and Peter Klenow - who had been known for their skepticism of the sticky-price mechanism - writing some papers giving more credence to the idea of sticky prices.

But in terms of academic publishing, the hot areas in macro seem to be A) labor search models, and B) financial-friction models. In both of those, you often see tech shocks as being the main driver.

The Diamond-Mortensen-Pissarides model, when it includes business cycles, includes them by adding TFP shocks. In other words, once you leave the steady state, DMP is just RBC with a matching friction. By my count, that makes two Nobel Prizes for RBC models and zero for NK. Actually, the vast majority of labor-search models I've seen, if they try to match business cycle facts, use the RBC mechanism to produce the cycles. The big exception I've seen is Christiano, Eichenbaum and Trabandt.

For financial-friction macro it's more even, as you definitely do have models that augment New Keynesian models with financial frictions - e.g. this one by Christiano et al., or this one by Curdia and Woodford. And of course there is Bernanke-Gertler, one of the original financial-friction models. But Kiyotaki-Moore, the other canonical financial-friction macro model I know of, uses RBC-style tech shocks as the single shock. And Brunnermeier-Sannikov uses what is basically a TFP-news-shock model. So here, the literature is pretty split, but RBC is alive and well.

But there are other macro-ish areas of research besides macro! There's international finance, where the models I've seen are all RBC-type models (caveat: my teacher for the international finance field class in grad school did her PhD at Minnesota). And when macro-finance asset pricing models bother to model production, they usually - though not always - use RBC-style tech shocks to make dividends and consumption move around.

So it seems pretty clear to me that RBC - defined loosely as "models where fluctuating TFP is the main driver of the business cycle" - ain't dead by any means. What's more, it seems unlikely that it will die for a good long while. Why? At least two reasons - A) ease of use, and B) familiarity.

First of all, productivity fluctuations are just plain easy to use. You can have constant returns to scale. You don't have to think about money or the aggregate price level. If you want to focus on other hard stuff, like labor search or banks, the easiest thing to do is just to use the simplest possible mechanism for the aggregate shock.

Second of all, everyone is really familiar with RBC, because everyone learns it before they learn any other type of shock mechanism (like NK). This is because RBC came first historically, because it's relatively easy to learn and understand, and maybe because of the enormous respect still given to the original RBC guys. So everyone knows how to use tech shocks, so they're the first go-to mechanism for many people.

So I think that all the people who pronounce RBC dead should think again. And NK's dominance is far from complete. If you don't believe me, just ask the Nobel committee!

31 comments:

  1. Do empirics differ?

    ReplyDelete
    Replies
    1. Yep. But generally they reject all these models pretty easily.

      Delete
    2. Which do they reject more, and when? A large N leads to statistical significance for even tiny differences.

      Delete
  2. Anonymous10:00 PM

    also Angeletos/Collard/Dellas; Paul Beaudry; Acemoglu network origins, which is a descendant of Long/Plosser. Personally I hope the input-output macro becomes the default model, it looks the most promising, lots of work still to be done in this model

    Uncle Charlie

    ReplyDelete
    Replies
    1. Yup.

      Those are my favorite flavor of RBC models...I suppose since they're never going away, I had better pick a flavor!

      Delete
  3. Anonymous10:09 PM

    and then there's the whole endogenous vs exogenous thing. Old models like
    Goodwin's predator-prey cycle and Kaldor's nonlinear(1940) model had endogenous cycles. Maybe we should revive them. actually Steve Keen is trying to.......

    ReplyDelete
  4. economist A: Macro is the things I care about, so I make a claim X

    economist B: claim X ain't true, because macro is about other things.

    economist C: the state of macro is good when A and B ignore each other

    ReplyDelete
  5. Anonymous5:03 AM

    I've never been able to understand how RBC is something that professional people who've spent their lives studying a field can do anything but sit there and laugh at. Nobel Prizes were actually given for that idea?

    I think economics needs to be burned down and started over again, frankly. Its predictive value is basically null at this point, and ideas like TFP shocks having any real importance at all are contributing to that. We're almost at the point where the people quoting Aristotelian dogma in the 1400s had a better real world track record: even if you accepted the dubious Milton Friedman philosophy that "as long as the model works, the legitimacy of the assumptions don't matter", there's still the small matter of the model not actually working. At least the epicycles in geocentrism made the skies reasonably predictable.

    ReplyDelete
    Replies
    1. Anonymous10:48 AM

      Milton Friedman's methodological dictum isn't dubious; it's out and out deranged.

      Delete
    2. I propose that apples fall to the ground because they love to be covered in dirt. It's genetics.

      Delete
    3. Anonymous4:36 PM

      I propose that none of your opinions matter in the least.

      Delete
    4. I laughed, but I can actually see an evolutionary logic to droppping seed-containers on the ground. They could have evolved to bear fruit in places with vertical support (like pollen on a flower). But the more obvious explanation is that fruit loves to be eaten and have its seeds expelled by animals.

      Delete
  6. I don't have the references to hand - I can remember Nick Bloom writing something like this - but in an RBC setting, the distinction between a technology shock and a demand shock is not a great as might first appear, both effectively make it harder for firms to produce (if we mean produce and sell) although now I type that I worry I may have go that wrong. Anyway, perhaps the choice of TFP shock isn't so crucial.

    ReplyDelete
    Replies
    1. Could somebody give me a source (for a layman) on what a 'technology shock' is?

      Delete
    2. Not really because nobody knows what 'technology' is in this context. Look up total factor productivity for some idea.

      Delete
  7. But one thing is to use TFP to enter an exogenous shock into your stuff and a very different one is "the whole" RBC thing in motion. It include things like the magic Euler equation and all that. I agree with your two diners and expect that this "enormous respect still given to the original RBC guys" will fade away.

    ReplyDelete
  8. - I think we've moved way beyond TFP shocks? RBC model is any model in which nominal prices and wages are flexible, and you can study real quantities and price ratios independent of nominal variables. This includes the models with sunspot shocks by people like Roger Farmer, models with sentiment shocks by people like Angeletos, models with financial shocks or uncertainty shocks like in some of Gertler and Kiyotaki's papers and many people in academia and the Fed. You can also have real wage rigidity for example in RBC models: the key is that in the RBC model you cannot fool workers into accepting lower real wages via a nominal devaluation, as some people were suggesting in the Greek depression case (the unions would simply extract higher pay raises in the next negotiation round).
    - There's lots of research that was done integration labour market search frictions into New Keynesian models (look up people like Carl Walsh, Thomas Lubik and many more).
    - TFP shocks can be given aggregate demand interpretations as shocks to the efficiency of matching in product markets (once you recognise the pervasiveness of search and matching frictions in product markets). See for example papers by Rios Rull and co.

    ReplyDelete
    Replies
    1. Anonymous4:07 PM

      Please, consult the Greek wage statistics. Nominal wages are down 20/25%. Only aggravated unemployment, of course, as it disrupted the circular flow of spending.

      Delete
  9. Anonymous9:35 AM

    Noah,

    Don't NK models also use shocks to tfp growth as a source of shocks? It's just not the only source...

    ReplyDelete
    Replies
    1. Yes, many do. But the TFP shock doesn't end up accounting for much of the action. It's included either as a nice gesture to the RBC people, or as a robustness check to show that including TFP shocks doesn't change the results.

      Delete
  10. Anonymous9:52 AM

    Noah

    How long a shock must last to be considered as such ? How economist do differentiate between an evolution (Internet) and a 'revolution' (70's Oil shocks ?) Sorry if questions are trivial for trained economists.

    Ludovic Coval

    ReplyDelete
    Replies
    1. A shock can be permanent. "Shock" just means "surprise".

      Delete
  11. I wonder if this is just Occam aesthetics, for better or worse. Every model needs technology; some have sticky prices and demand shocks and other New Keynesian features as well. If you are trying to tell a clear story about job search, tradition urges you to use the simplest possible economic setting, and if you need shocks and dynamic plans that's RBC. Adding the NK features leaves it less clear what is driving results.

    I think this is a good tradition for fundamental theory, but maybe not for calibrated models. However, I can't figure out how to learn from most calibrated models in any case.

    ReplyDelete
  12. Anonymous12:25 PM

    I wouldn't say that RBC is dead. It would be more accurate to say that we've simply moved on. The NK model is probably the closest thing to a "baseline" now, but in reality there are many baseline models that people know and think capture important features of the economy, and many shocks that people work with.

    TFP shocks are standard, but mainly by convention. Most people would probably agree that TFP shocks are not the main driver of the business cycle, but they still use them in models because they're an accepted way to shock a model and examine the response. So if you're not setting out to take a stand on what exogenous shock drives recessions, TFP shocks are a nice place to start.

    ReplyDelete
  13. Anonymous1:11 PM

    What's more interesting is that the micro-foundations based DSGE framework was and has remained the leading framework in macro. The only issue is, do you have technology shocks as the primary driver or 19 other shocks also playing a role a-la Christiano-Eichenbaum, Smets-Wouters. The former is the RBC model and the later is the NK model.

    Many claimed that the DSGE approach should be replaced after the great recession. Clearly that is not happening!

    ReplyDelete
  14. It's not dead. It's just resting.

    ReplyDelete
  15. “What's more, it seems unlikely that it will die for a good long while. Why? At least two reasons - A) ease of use, and B) familiarity.”

    And I think it’s also the interesting (in a profoundly harmful way) and widespread phenomena of “Science advances one funeral at a time”. I wrote about this recently:

    I've been very interested in this quote for a long time, especially with regard to economics. But recently I've come upon and thought about some other important relevant areas.

    I'm currently reading "The Second Machine Age", by MIT business economist Erik Brynjolfsson and MIT computer scientist Andrew McAfee…

    …something recently really startled me on page 102. The authors talk about how electrification so profoundly increased productivity. It was the second great GPT (General Purpose Technology) after the steam engine. Yet, initially it did little for productivity, not until factory managers started to design factories for electrical rather than steam power. Electricity had the advantage that machines could be spread out and arranged in a horizontal, assembly line, pattern, as opposed to steam, where machines had to be placed close to the large central engine, both horizontally and vertically (on floors above and below the engine). The difference was profound, and revolutionized industrial productivity.

    The quote that struck me was this:

    “Only after thirty years -- long enough for the original managers to retire and be replaced by a new generation -- did factory layouts change.”

    ReplyDelete
    Replies
    1. Yeah, the post is at:

      http://richardhserlin.blogspot.com/2014/02/what-else-advances-one-funeral-at-time.html

      Delete
  16. Why do you assume trend growth is smooth?

    ReplyDelete
    Replies
    1. Why do you assume that trend growth originates on the supply side?

      See Japan, the euro zone, etc..

      Delete
  17. Pissarides's first edition of Equilibrium Unemployment Theory (circal 1990), has some Keynesian-flavored experiments with consumption and saving in the basic Mortensen-Pissarides framework. And his 1970 book and 1970s and early 1980s articles also have a lot of neo-Keynesian ingredients. Also, while Phelps got the Nobel for something else, his structural slump book is basically Mortensen-Pissarides with effort (what the Stiglitz generation called efficiency wages). But both the first edition of Pissarides and the only edition of Structural Slumps are a little messy, the second edition of Pissarides, without the old-style Keynesian stuff and with the technology shocks thrown in, is a lot easier to read. The second edition is the edition that people read.

    I think we're still waiting for an NK model that's as easy to read as a Prescott tract.

    ReplyDelete