Friday, July 13, 2012

Two truly excellent modeling posts by Paul Krugman

1. Simple economics is not simplistic economics. The key, of course, is to get a version of "simple" that explains a great deal. If it can't do that, it is probably "simplistic". If it can do that, it's a good model. We are not physicists. We are modeling processes and mechanisms - we are not replicating the economy in a computer. That's impossible. I think a lot of different schools of thought have a lot of very important things to say, but the reason why I think of myself as a "Keynesian" is that a simple Keynesian model explains several orders of magnitude more about this crisis than any other model out there that I've come across, and when we approach episodes of a very different nature (like 1920-1921), there's nothing that Keynesianism says that is contradicted by those episodes. You need both of those elements for a simple model that works: the ability to explain a lot in times that it matters, and the ability to avoid contradicting the evidence when your explanation is somewhat less relevant.

2. Gadgets in models are not the heart of models. Here, Krugman is criticizing people that emphasize mere gadgets too much, but this is also what bothers me about a lot of critiques of mainstream models because of their gadgets. Modeling conventions like CES utility functions, Ricardian equivalence, and different discounting and intertemporal decision making assumptions are used to structure models. I don't believe Ricardian equivalence strictly holds. I don't think anyone does. But the question is, if we're writing up a simple model of an economic process, should our first approximation be Ricardian equivalence or not? The first approximation should absolutely be Ricardian equivalence. Now - if you're result depends on the strict truth of Ricardian equivalence then you need to recognize that you have a problem, because Ricardian equivalence is a gadget in a model - it's not the reality of things. But the point is if people have tendancy toward rational behavior it makes more sense to model them behaving rationally and then loosen assumptions than it does to abandon the formal modeling project altogether. It's a judgement call in a lot of cases, but it's very frustrating to get criticized for these modeling assumptions by people who don't really understand how mainstream economists think about modeling assumptions. And you get that sort of criticism from both the libertarians and the lefties.


  1. Should a model really take into account RE? Really? I mean, ask your average person if they structure their spending plans around the actions of the government and central bank. Ask them if they know anything about monetary and fiscal policy.

    1. Two thoughts -
      First, as Krugman often points out, Ricardian equivalence (I assume this is what you mean by RE, and not rational expectations) does not mean everything it's claimed to mean about spending decisions. RE with permanent income hypothesis (another lefty favorite!) does not imply that deficit spending is going to be canceled out by spending reductions elsewhere, even in the case of full Ricardian equivalence.

      Second - how realistic is Ricardian equivalence? It's tough to say with changes on the margin around a reasonably table debt level (yes we've swung around but we've never borrowed unsustainably for an extended period of time). If for the next ten years the government ran surpluses, I would probably expect my future tax bill to go down. Wouldn't you? If after the recession was over deficits continued to be around ten percent of GDP I would expect my future tax bill to go up. Wouldn't you? Wouldn't everyone? But any given year's deficit is of course not going to do this - it's a matter of persistent behavior. In the entire postwar period we have stably practiced functional finance. I fully expect we'll continue to. Deficit changes will be marginal so detecting behavioral responses consistent with Ricardian equivalence will be tough. But yes, I think as an approximation we respond to things rationally. We have our irrational blindspots, and since this is an intertemporal question it's important to note that I think we do engage in hyperbolic discounting to an extent. But aside from those wrinkles, I think it's reasonable to say we behave rationally.

      Let me ask you this - you and I both think we depart from rationality to some extent. Do you think we depart so radically that rational expectations modeling techniques like Ricardian equivalence are even a bad starting point (presumably you think so). If that's the case, where would you start?

    2. "If for the next ten years the government ran surpluses, I would probably expect my future tax bill to go down. Wouldn't you?"

      No. That is not an important datum for my expectation of future taxes. My expectation would depend, among other things, upon why the gov't ran surpluses. The important cognitive point is, once I take those reasons into account, it would be irrational of me also to take the fact that the gov't ran surpluses into account. It does not work the other way around. Knowing only about gov't surpluses (or deficits) does not give me enough information to predict future taxes.

      My reaction in '09, upon hearing the Ricardian equivalence argument against gov't stimulus during our depression, was two-fold. First, how could anyone over the age of 30 believe such a thing? Second, why isn't it an argument for giving money to poor people?

    3. Ricardian equivalence has been disproved by a simple example, see

  2. Daniel Kuehn: Economists may not be physicists, and our models may not be perfect. I would agree with you and Krugman that Sir John Hicks's IS/LM model still is a good pedagogical tool, but why not have more dynamic methods in the study of economics?

    You know well enough that agent-based modelling is now popular in economics research. Of course, I doubt that a computer program's model would perfectly capture the macroeconomic level.

    As for Ricardian equivalence...doesn't the work of Daniel Ellsberg technically show that people's preferences and decision-making processes demonstrate non-linearity and non-additivity? They may be "rational", but their preferences certainly aren't stable.

    1. I'm not understanding the last point. I think it's pretty reasonable to say that preferences aren't stable, aren't linear, and aren't additive (I guess... I'm not particularly clued into the implications of that last one), but Ricardian equivalence isn't really about preferences, it's about intertemporal expectations associated with fiscal decisions.

      And to clarify I don't think people strictly practice Ricardian equivalence for a lot of the reasons that unlearningecon listed above. But I do think people are rational enough that rational expectations, utility maximization, and Ricardian equivalence are excellent "gadgets" (as Krugman puts it) to start with. If we toss out rational expectations I'm not sure what I'd replace it with. Seems to make more sense to start with rational expectations and then start loosening assumptions.

    2. Daniel,
      It seems peculiar to start with a highly unrealistic set of axioms. You think that people are rational enough, but really, how rational? We certainly don't behave as 'homogeneous globules of desire'. Nor could we predict the future for it is uncertain.
      And it's not like you could not choose other axiomatics.

    3. I agree it seems peculiar to start with a highly unrealistic set of axioms.

      I don't think we behave like homogenous globules of desire, nor do I think we can predict the future (we talk about that a lot on here actually).

      I'm not sure what your point is.

      Stop commenting anonymously, please.

    4. Daniel,
      My point is that rationality and foresight are highly suspect (not possible) in everyday human life, so then laying them as axioms/assumptions in the heart of the theory will likely lead to the theory not reflecting reality.
      I do not have neither real life academic career nor web notoriety (so I might as well be nobody), but sure, if you prefer it this way.

  3. Additivity is a mathematical property.

    Since Dr. Michael Emmett Brady has been making the argument that George Boole and J.M. Keynes are the founders of non-additive probabilities, this renders Subjective Expected Utility as a special case of a more general theory of decision-making. I'm sure that the Boole-Keynes approach to probability would have implications on Ricardian equivalence, but I'd need to consult Dr. Brady again.


All anonymous comments will be deleted. Consistent pseudonyms are fine.