Monday, December 16, 2013

Grilled Cheese Sandwich

This relates to this post, and this one. To quote myself:
If I want to find a Phillips curve relationship in the data, then through some process of specification searches, Bayesian estimation with alternative priors, or whatever, I will be able to find it. If I'm looking hard for the Virgin Mary, I can find her in a grilled cheese sandwich.
So, Paul Krugman has gone searching for the Virgin Mary and, big surprise, claims success. Of course I can take Krugman's two time series and find something else. Here is the data from October 2009 (peak unemployment) to November 2013:
In that figure, the line joins observations for October 2009 on the far right with November 2013 on the far left.

Note two things:
1. Unemployment and wage inflation are positively correlated over that four-year period.
2. Sometimes unemployment is falling when wage inflation is falling; sometimes unemployment is falling when wage inflation is rising.

Krugman seems to forget the key lesson of the 1970s Phillips curve debate, which is that you need theory - structure - to make sense of what we see in the world, in a way that is useful for policy.


  1. I think there might be something in this post, but I am having problems aligning it with the post-1968 expectations augmented Phillips curve of Friedman and Phelps (be those expectations rational, adaptive, survey-based, or financial market-based). My 11-year career of macro forecasting suggest no one would expect to see much of relationship in your overly simplistic formulation. (If your figure were right, it would make my job so much easier.)

    1. Suppose you were schooled in expectations-augmented Phillips curves. Apparently you were, and that's what Dick Lipsey taught me in the 1970s. Over the period that the chart above covers, the Fed was telling us that expectations were "well-anchored," and that seems roughly consistent with observed TIPS yields and nominal bond yields (from which we can extract a measure of anticipated inflation). So, think of anticipated inflation as roughly constant for the period we're talking about. So, what does the chart tell you?

  2. While I disagree with the notion that the Phillips curve is always vertical, i.e. that money is always neutral, I totally agree with your point about expectations and the Lucas critique.
    But I do not just disagree with the Old Keynesian perspective which ignores expectations, I also disagree with the Neoclassical perspective as I never bought ANY expectation mechanism (be it rational, adaptive or whatever).

    There is a third way besides estimating mere VARs or structural models, Romer&Romer's historical approach ( In the case of monetary policy taking a look upon how actors thought is basically more microeconomics than e.g. assuming rational expectations.

    1. "While I disagree with the notion that the Phillips curve is always vertical, i.e. that money is always neutral..."

      I'm not sure who you're disagreeing with. Even Friedman did not say that money was always neutral.

      "I never bought ANY expectation mechanism..."

      So, people don't think about the future when they make decisions?

      Romer and Romer's work is interesting, but you could question whether that is capturing "how actors thought." Basically, they read the FOMC minutes, and derive some course measures of the stance of monetary policy. You could argue whether that's measuring anything useful.

  3. Of course we should continue working with intertemporal models and expectation mechanisms.

    My point is that while e.g. rational expectations are theoretically clearly superior to the pre Lucas critique stuff, i.e. adaptive expectations or none at all, it doesn't necessarily fit the data better and it doesn't necessarily describe people's behaviour better (i.e. my point here is not that the microfoundations approach is wrong but not radical enough).

    As you mentioned Friedman, let's take his permanent income hypothesis which is basically in-between the static, Hicksian 'people consume everything in one period' and the Classical perfect consumption smoothing.
    So we have a mere hypothesis with little to no theory underneath it and yet it works better than two benchmark models.

    In other terms, fitting the data without any model can work better than deep theory. Why? Because a representative agent who can transfer consumption perfectly between periods captures no feature of the capital market, be it financial intermediation which creates a wedge in interest rates, or asymmetric information which creates incentive issues and is impossible to model with a rep.agent.

    As I pointed out in the first paragraph, I am all for microfoundations. But not when these microfoundations are utterly ridiculous and work empirically worse than some back-of-the-envelope stuff Friedman did.

    This is by the way also how some central banks work, a microfounded DSGE model at the core plus a non-microfounded periphery model.

    1. I think we agree more than you realize. First, when you say:

      "Because a representative agent who can transfer consumption perfectly between periods captures no feature of the capital market, be it financial intermediation which creates a wedge in interest rates, or asymmetric information which creates incentive issues and is impossible to model with a rep.agent."

      What you're describing is my research agenda. I work with models in which there is financial intermedation, interest rate wedges, incentive problems, etc. And no representative agents.

      Where I disagree is:

      1. Rational expectations is partly for convenience. The idea is that we're focusing attention on things other than expectations formation, because those are the things we know more about. In the meantime we can be assured that there is not something funny going on in our models that depends only the nature of learning, for example.

      2. In your last sentence, note that the fact that central banks do it doesn't make it right.

    2. Sorry for the messed up postings but as I wrote in the post below, what we have here is he interesting case of a Monetarist and a Keynesian who probably disagree on a practical level (in the Mankiwian sense of macroeconomists as engineers) on many issues but agree on how one should conduct theoretical microfounded macro.

      Hell, if I find the time one day (I am not an academic economist but try to keep up with some of the literature) I might even read your papers.

  4. Addendum: I know that there are ample of models which incorporate frictions and market imperfections.

    But who really believes that Calvo contracts (as short-way for the Blanchard-Mankiw imperfect competition plus menu costs stuff) in standard DSGE models is the real source of the problem? I don't and believe that it is rather a mixture of plenty of factors, from wage and price stickiness to coordination problems to incentive issues in finance (IMO the most serious problem as it undoes perfect future markets, impacts consumption smoothing and the credit channel through which money mainly works) that make underemployment equilibria possible.

    1. So if its not Calvo contracts, but you think wage and price stickiness are important, how do you model that?

  5. Well, the "long way" is imperfect competition plus menu costs like in Blanchard&Kiyotaki '87. I am not an expert on this literature or mainstream DSGE stuff but I think that the "long way" is not used as it is analytically burdensome and Calvo contracts are considerer to be an appropriate "short cut".

    Personally I think that the mainstream macro model, a classical market with some nominal frictions is nonsense.

    I do not know your work but you often point out that you include a variety of market imperfections in your papers, above all financial frictions. And IMO this is the way to go.
    Models in which money simply appears in the utility function of households seem dubious to me whereas models that actually explore money in the sense of credit and analyze frictions in the financial markets can actually provide some real insights. Stiglitz wrote an entire book about the subject, "Towards a New Paradigm for Monetary Economics" (not a good one and he is a bit obsessed with paradigm changes but at least he sketched out the right track).

    I am a Keynesian ... but from a theoretical perspective I see little value in the majority of Old Keynesian and New Keynesian literature.

    1. There was work on so-called state-dependent pricing models, which is essentially menu costs in a dynamic setting. Most New Keynesians went with Calvo pricing, which you can think of as a random menu cost - the cost of changing a price is either zero or infinity. King/Wolman/Khan worked on state-dependent pricing models. And there's Lucas and Golosov, though maybe you wouldn't like their result. All of that stuff basically looks like a Woodford model in most ways - representative agent, complete markets, monopolistic competition in goods markets.

    2. Noah Smith recommends this paper as a better microfounding of sticky wages. What do you think?

    3. That paper is not about sticky wages. What Christiano et al. want to do is to have a sub-model of unemployment that they can treat as a component of their large quantitative model. They don't want to use Mortensen-Pissarides, as that doesn't fit nicely with the rest of the model. But the idea basically involves a search friction, that works in a fashion that is related to search and unemployment models, but with some bells and whistles. Wages are determined by bargaining between a worker and a firm. Not fundamentally different from Mortensen-Pissarides, but it's not Nash bargaining. So, I'm glad to see that Noah got something out of the conference we both attended at the St. Louis Fed, but I'm afraid Noah took the wrong message away from Christiano's paper.

    4. I think that was my bad, not Noah's. The abstract used the term "wage inertia", and I paraphrased that as "sticky wages". Noah himself did not, he just said it had better microfoundations.

    5. Good. Noah is making progress. I'm inclined to think we should focus our attention on frictions we know something about. Search is a very useful paradigm, and well-worked-out. Some people thought about pricing in search models - e.g. Burdett and Judd ( is a model of price dispersion. And, clearly, credit friction are important. We know a lot about limited commitment, something about debt contracts, and something about collateral. If you asked me, I would think those would be the places to go looking if we want to think about the effects of monetary policy, for example.