Sunday, January 15, 2017

Cracks in the anti-behavioral dam?


This is purely my impression, buttressed with some anecdotes; I don't have any systematic data to back this up. But in both papers and casual discussion, I'm seeing macro people taking behavioral ideas more seriously. 

"Behavioral" is a very squishy idea, but basically I think of it as meaning "imperfect use of information". The difficulty with labeling a model "behavioral" is that we don't really know what information is available. This is why I believe there's a fundamental equivalence between behavioral and informational models - for any "informational" model where agents don't know all of the facts, there's an observationally equivalent "behavioral" model where they do observe the facts and just don't make use of them. 

But anyway, in macro, most models use Rational Expectations, so let's think of "behavioral" as just meaning "non-RE". Actually, non-RE models have been kicking around for a long time - for example, Sargent's learning models, or Mankiw and Reis' sticky information models. What seems to be changing (slightly) is that A) younger people seem to be making non-RE models, B) people are recommending non-RE models for policy analysis, and C) the departures from RE are getting more stark. 

Some recent examples I've seen are:

1. The learning approach to New Keynesian models, promulgated by Evans et al., which seems to be solidly mainstream

2. Mike Woodford's response to Neo-Fisherism, which relies crucially on a slight departure from RE

3. Xavier Gabaix's behavioral New-Keynesian model, where consumers are short-term thinkers instead of infinitely far-ahead-looking

These are all well-established people making these models - they cut their teeth on RE models for years before daring to venture out into behavioral waters. But now I'm starting to see young people doing behavioral stuff as well. A good example, sent to me by Kurt Mitman, is this paper by Kozlowski, Veldkamp, and Venkateswaran, entitled "The Tail that Wags the Economy: Belief-Driven Business Cycles and Persistent Stagnation". 

The basic idea of the paper is that instead of knowing the true PDF of macroeconomic shocks, people re-estimate the distribution every time they see a shock. Not too crazy, right? But that seemingly small departure from RE has big business-cycle implications. 

The reason is tail events. When big shocks are rare, just one of them can change people's whole understanding of how the economy works. How many events like the Great Depression have there been in American history? Really, there are only two since we started keeping national accounts. Two! In 2008 we abruptly went from "There was that one really bad depression one time" to "Whoa, this is a thing that can happen multiple times!". To think that this would have zero impact on agents' beliefs about the economy - which is exactly what RE demands we think - seems implausible. The authors write:
No one knows the true distribution of shocks to the economy. Economists typically assume that agents in their models do know this distribution as a way to discipline beliefs. But assuming that agents do the same kind of real-time estimation that an econometrician would do is equally disciplined and more plausible. For many applications, assuming full knowledge has little effect on outcomes and offers tractability. But for outcomes that are sensitive to tail probabilities, the difference between knowing these probabilities and estimating them with real-time data can be large.
Anyway, to make a long story short, this can produce long economic stagnations, like the one we just had. Taking a gander at the literature review section, I see that these authors didn't aren't the first to use this mechanism in a theory - it looks like it can be traced back to a 2007 AER paper by Lars Hansen. The other similar papers the authors cite, however, all come from 2013 or later, showing that this sort of idea has been gaining currency recently and rapidly. 

Now, Koslowski et al. do dodge one important issue: what data set do agents use to estimate the distribution of economic shocks? The data set they use goes back to Word War 2 - they don't even include the Great Depression. But even if we go back further than that, we'll miss earlier episodes like the Panic of 1873, when good national accounts just weren't kept at all. Data availability is so recent that there's almost an observational equivalence between assuming that people use all the available data, vs assuming that people overweight data from their own lifetimes.

If the authors - or some other authors - were to assume that people overweight data from their own lifetimes, as evidence from Malmendier and Nagel suggests, it would have important implications down the line. Instead of people's expectations slowly converging to RE over the decades (centuries?), people would forget the lessons of history and continue being surprised by depressions every 50 or 100 years or so. 

For now, macroeconomists don't have to worry about this question. Authors like Koslowski et al. can frame their papers as quasi-behavioral papers, where RE is limited by data availability, instead of fully behavioral papers where RE is limited by collective forgetting. So these are still only cracks in the anti-behavioral dam, not a full torrential flood.

But my question is this: What happens when people start applying this mechanism to more complicated shock processes? What if the economy has regime switches that last decades? What if there is more than one kind of rare shock (e.g. the Great Inflation of the 70s/80s)? I've seen some people try to model stuff like this, and the end result can come out looking like practically any type of non-rational expectations you can think of. Meanwhile, empirical macro people are starting to pay more attention to survey measures of expectations. And people from behavioral finance are starting to put things extrapolative expectations into macro models, to explain macro facts. And evidence like that collected by Malmendier and Nagel continues to pile up.

And I should mention casual conversation as well. More and more young macro people that I interact with, including (even especially?) those who run in "freshwater" circles, are saying that behavioral explanations will have to be part of our understanding of how consumption works. Here's an example from a recent blog comment. 

So I wouldn't be surprised to see some more cracks in the anti-behavioral dam in the years to come. Chris House, my old macro prof, proclaimed three years ago that behaviorism was a dead end and would never have a transformative impact on macro. But seeing papers like Koslowski et al.'s, I'm thinking that his prediction now looks to have been quite ill-timed.


Updates

Just for fun, I'll post some more random behavioral macro papers I see.

"Monetary Policy, Bounded Rationality, and Incomplete Markets", by Farhi and Werning

"Explaining Consumption Excess Sensitivity with Near-Rationality: Evidence from Large Predetermined Payments", by Kueng

"YOLO: Mortality Beliefs and Household Finance Puzzles", by Heimer, Myrseth, and Schoenle

"Learning about Consumption Dynamics", by Johannes, Lochstoer, and Mou

"Understanding Uncertainty Shocks and the Role of Black Swans", by Orlik and Veldkamp

"The Liquid Hand-to-Mouth: Evidence from Personal Finance Management Software", by Olafsson and Pagel

25 comments:

  1. Good read. The Fed, amongst others, have been ahead of the curve in that respect for some time now: http://www.frbsf.org/our-district/press. However, I do take your point that behavioural econ is an agent based science and not neccesrily compatible with aggregate data. I mean where would one start? The 10-year regime switch is a good start I think; transitional probabilities are a burden in the best of cases because they typically utilise an exegenous asset price distribution such as an index to draw returns from. Not to mention CB data such as BoP, seigniorage etc as well as microfoundations such as labour productivity, wage growth then we can start talking about fiddling with behavioural aspects vis-a-vis household consumption and saving with a view to estimating how monetary policy affects wealth inr, right?

    ReplyDelete
  2. I mostly agree with you.

    However, your claim that "To think that [the 2008 crash] would have zero impact on agents' beliefs about the economy - which is exactly what RE demands we think - seems implausible."

    This in no way is demanded by RE. Maybe it is a consequence of RE in certain models but provide a model in which there are contingent facts about the operation of the economy which determine the rate of crashes and RE allows one to use the crash to update your distribution for these parameters.

    ReplyDelete
  3. Bad start to this post. Irrational and uninformed are not two sides of the same coin. People are often irrational, drawing different conclusions from identical data (think opposing interpretation of sports replays on disputed calls). Motivated reasoning. You can do better.

    ReplyDelete
    Replies
    1. I think you should consider going back and reading what I wrote, and thinking a little harder about it... :-)

      You can do better.

      Delete
    2. Ok, I'll have a go.

      You said:

      This is why I believe there's a fundamental equivalence between behavioral and informational models - for any "informational" model where agents don't know all of the facts, there's an observationally equivalent "behavioral" model where they do observe the facts and just don't make use of them.

      This seems like an unjustifiable position to me. Going back to the sports fan analogy, there is a geographic correlation in the (mis)interpretation of the replay due to hometown bias. You cannot, with equal parsimoy, model that as ignorance: where would the spatial correlation come from? How would you explain serial correlated bias in interpretation of replays favoring and disfavoring the home team?

      I could go for informational models can be recast as ignorance models and vice versa if you neglect spatial and temporal correlations. But, I think correlated decision errors would allow us to distinguish between these types of models in many cases.

      Delete
    3. You're absolutely right that some models are much more *plausible* than others. I completely agree. My point was that for every simple, elegant behavioral model we make to explain some economic phenomenon, some person who wants to keep RE can go write down a complicated, tortured, improbable-seeming model where people just aren't adequately informed. So that can be very frustrating.

      Delete
    4. Perhaps I misunderstood what you were arguing...

      Delete
    5. I would certainly agree with your point then. Thanks for elaborating.

      Delete
  4. Strange noise in the graveyard of economics
    Comment on Noah Smith on ‘Cracks in the anti-behavioral dam?’

    Economics is a failed science. More precisely, standard/orthodox economics from Jevons/Walras/Menger to DSGE/RBC/New Keynesianism is scientific rubbish or what Feynman famously called cargo cult science.

    Standard economics is built upon this set of foundational propositions, a.k.a. axioms: “HC1 economic agents have preferences over outcomes; HC2 agents individually optimize subject to constraints; HC3 agent choice is manifest in interrelated markets; HC4 agents have full relevant knowledge; HC5 observable outcomes are coordinated, and must be discussed with reference to equilibrium states.” (Weintraub, 1985)

    Methodologically, these premises are forever unacceptable: (i) The ultimate reason can be stated as an impossibility theorem: NO way leads from the explanation of individual behavior to the explanation of how the economic system works. Because of this, the microfoundations approach has already been dead in the cradle. (ii) The standard/orthodox/Walrasian axiom set contains THREE NONENTITIES: (a) constrained optimization (HC2), (b) rational expectations (HC4), (c) equilibrium (HC5). Every theory/model that contains a nonentity is A PRIORI false. So, standard/orthodox economics is axiomatically false ― it is as simple as that.

    The microfoundations speak about human behavior. As a matter of principle, human behavior is the subject matter of psychology, sociology, anthropology etcetera and NOT of economics (Hudik, 2011). The error/mistake/idiocy of half-witted critics, though, has always been in the belief that it suffices to make behavioral assumptions ‘more realistic’. The classical case is the idea to replace the assumption of perfect rationality by bounded rationality.

    What instead has to be done is to FULLY REPLACE the standard/orthodox microfoundations with methodologically correct macrofoundations, that is, to change the very definition of economics.

    OLD behavioral definitions: “It is a touchstone of accepted economics that all explanations must run in terms of the actions and reactions of individuals.” (Arrow, 1994) “Economics is the science which studies human behavior as a relationship between ends and scarce means which have alternative uses.” (Robbins, 1935)

    NEW objective-systemic definition: “Economics is the science which studies how the monetary economy works.”

    The move from the old behavioral definition to the new systemic=behavior-free definition of economics is called a paradigm shift. Nothing less will do to get economics out of its proto-scientific deadlock.

    Noah Smith has spotted a new trend in macro: “I’m seeing macro people taking behavioral ideas more seriously. ... in macro, most models use Rational Expectations, so let’s think of ‘behavioral’ as just meaning ‘non-RE’.”

    To see hope or even progress is this pathetic putting-lipstick-on-a dead-pig is as self-debunking as it can get. Standard economics is scientifically dead since 140+ years and it will NOT come to life by replacing RE with non-RE.

    Egmont Kakarot-Handtke

    ReplyDelete
    Replies
    1. String5:00 PM

      Son Goku is a fictional character and main protagonist of the Dragon Ball manga series created by Akira Toriyama. He is based on Sun Wukong, a main character in the classic Chinese novel Journey to the West. Goku is introduced in Dragon Ball chapter #1 Bulma and Son Goku, originally published in Japan's Weekly Shōnen Jump magazine on December 3, 1984, as an eccentric, monkey-tailed boy who practices martial arts and possesses superhuman strength.

      Delete
  5. This is a very important development. The RE hypothesis has led us on a wild goose chase for ever more complex and intelligent consumers who are somehow able to "intuitively" solve complex partial differential equations that pHD economists can barely formulate, all to prop up a series of unrealistic assumptions. If we are going to add "epi-cycles" to our economic models so that they better fit empirical data, it makes no sense to restrict ourselves to RE hypothesis consistent epicyces. We can begin to lay the ground work for some future economic s Galileo to at least have a productive mess to clean up.

    ReplyDelete
  6. "for any "informational" model where agents don't know all of the facts, there's an observationally equivalent "behavioral" model where they do observe the facts and just don't make use of them."

    The difference seems to be that in the "behavioral" model, people have the facts but don't know how to make use of them, and there's a policymaker who has the facts but somehow knows how the people who don't know how to make use of the facts should make use of said facts. In the private information world, some people know some stuff, other people don't, and the policymaker is about as good as anyone else in ferreting out information.

    ReplyDelete
    Replies
    1. I think bringing in a policymaker is a complicating factor! Figuring out how things work is hard enough without figuring out if there's any way to improve things. Of course, existing policy has to be taken into account when figuring out how things work, but that's a bit of a different matter...

      Delete
    2. Anonymous2:05 PM

      I don't think behavioral models have any presumption that policymakers know more than others. They're human, and have the same cognitive limitations as other humans.

      Delete
  7. I think your characterization of resistance (i.e. a "dam" that's holding back this flood of behavioral economics) is incorrect. Interest in "learning," for example, goes way back. It's an active sub-field, and has its own conferences, for example, where I'm told Sargent and Hansen show up on a regular basis. Macreconomists have dabbled in non-expected utility theory and self-control, for example. But, it's not like these things are talking over economics. A good reference is David Levine's work:

    http://www.dklevine.com/general/behavioral/doomed.htm

    ReplyDelete
    Replies
    1. I think David is being a bit melodramatic here. I also think he's probably wrong. The practice of trying to import insights directly from psych is probably doomed, at least until psych stops sucking (which may be never). But the practice of putting non-rational expectations in models may just be getting started. It all depends on what we call "behavioral".

      Delete
  8. Anonymous5:21 PM

    Using the rational expectation model makes it easier to formulate an opinion or prediction on what human event is going to occur. These models are not perfect but when you include the behavioral aspect you have constant calculus with consumption that will break any model. Humans are mapping their consumption patterns based on patterns and behavioral information (i.e. the more people learn about Prospect Theory the more people attempt to become less risk averse) and I think most are constantly performing game theory decisions (unknowingly).

    Since the cognitive sciences having become more developed by the day the constant win/lose decision to optimize their consumption based on what they are behaviorally allowed to get away with is going on. I think behavioral theory will revolutionize rational expectations but it is far too complex to model accurately given the tools, data, and methods currently available. Pharming data such as what Facebook is doing may change that if the information is asymmetric and consumers aren't allowed to readjust.

    ReplyDelete
  9. Hi Noah,
    You are absolutely right, and I can point you to many more articles from the past decade that are taking behavioral aspects more seriously. However, the ultimate proof is that for the past couple of years Andrew Caplin and Michael Woodford have been organizing an NBER group on Behavioral Macro, which meets for a day during the Summer Institute.

    ReplyDelete
  10. Normative economics, I mean behavioral economics, is a fantastic trend. People love being told by PhDs that they do not have all the facts and behave erratically. It's what people live for. Middle America is just waiting for elite economists to tell them how, when, and on what, to spend their paycheck because elite economists think that people are ill-informed. People won't think that this is junk science in a field with serious issues replicating results. They will not at all think that this is effort by left-leaning nanny-type elites to confirm and impose researchers' own biases.

    Thankfully, people never tired of being told how to live their lives, at all. I cannot wait to read in Breitbart and Drudge how left-leaning coastal economists promote ideas that obviously require more bigger government to force people to make choices consistent with New York values.


    I know, I know, none of what I wrote could possibly be true, it's just an epic rant. It's just another example of how internet commenters are ill informed, uneducated, reactionary, and out of touch.

    ReplyDelete
    Replies
    1. Anonymous10:03 AM

      Apparently *some* internet commentators are ill-informed, uneducated, reactionary, and out of touch.

      Delete
    2. Anonymous2:06 PM

      There have been elites telling others what they ought to do for a very long time. This is not something that started with behavioral economics.

      Delete
  11. RE has always seemed a bit irrational to me. Half our citizenry, if polled, would assert that FDR caused the Depression, and that assumes they knew what the Depression was and when it occurred. So, yeah, an individual's weighted lifetime of experience counts for something.

    30+ years ago I took a mandated manager training class where one of the videos was by Morris Massey. His gazillion words a minute video's point was what happened to you at around the age of 10 left a huge impression on you. At the time people who were born around 1920 were still in the workforce, and that what motivated them was different than what motivated us boomers. Well, yeah.

    Strauss and Howe's Generations talks about a cycle of 4 generations that ends in Crisis.

    Put Massey, Strauss and Howe, Krugman's comments about zombie and cockroach ideas, and calculated politcal disinformation (Social Security is bankrupt! More violent crime!) in a blender it seems that people do not and cannot possibly behave in a manner that RE can remotely model.

    Back in the 80s Kasey Kasem conducted a survey on the greatest rock group of all time, and the winner was... Queen.

    ReplyDelete
  12. It's good to see some cracks in the dam. You are definitely right about psych, though some cognitive stuff might be useful if adapted.

    The typical RE argument always seemed to be caught in a time warp, like something out of an old Leninist catechism, rather than a rational line of argument. I can almost imagine the RE interpretation of Pavlov's work with dogs. "Of course, the dog won't salivate when the bell rings, since the rational dog will realize the discount rate of blah blah blah means that eventually he'll be hungry again." Woof!

    ReplyDelete
  13. Noah, Scott Alexander is working on something, you might want to take a look:

    http://slatestarscratchpad.tumblr.com/post/156916981936/im-working-on-a-complicatedcontroversial-post

    Regards,

    Antimule

    ReplyDelete
    Replies
    1. This post is great, and is very important. I did a short summary of this stuff here, in fact: https://www.bloomberg.com/view/articles/2015-03-12/put-a-lid-on-u-s-s-excessive-costs-to-boost-growth

      I don't have Scott's email, but I think this post is spot-on.

      Delete