Friday, September 21, 2012

Thomas Sargent Abandons Rational Expectations, Converts to Post Keynesianism

From Keynes (1937), "The General Theory of Employment":
We have, as a rule, only the vaguest idea of any but the most direct consequences of our acts. Sometimes we are not much concerned with their remoter consequences, even though time and chance may make much of them. But sometimes we are intensely concerned with them... The whole object of the accumulation of wealth is to produce results, or potential results, at a comparatively distant, and sometimes indefinitely distant, date. Thus the fact that our knowledge of the future is fluctuating, vague and uncertain, renders wealth a peculiarly unsuitable subject for the methods of the classical economic theory. ... 
By ‘uncertain’ knowledge, let me explain, I do not mean merely to distinguish what is known for certain from what is only probable. The game of roulette is not subject, in this sense, to uncertainty... Even the weather is only moderately uncertain. The sense in which I am using the term is that in which the prospect of an European war is uncertain, or the price of copper and the rate of interest twenty years hence, or the obsolescence of a new invention, or the position of private wealth-owners in the social system in 1970. About these matters there is no scientific basis on which to form any calculable probability whatever. We simply do not know.
Better late than never, I suppose...

UPDATE: OK, ok, Sargent has written a lot about uncertainty and expectations.  Harry Konstantidis points out this 1972 article Rational Expectations and the Term Structure of Interest Rates, which is ... really interesting, actually. It's an empirical test of whether the behavior of interest rates over time is consistent with bond prices being set by a process of rational expectations. The conclusions are surprisingly negative:
The evidence summarized above implies that it is difficult to maintain both that only expectations determine the yield curve and that expectations are rational in the sense of efficiently incorporating available information. The predictions of the random walk version of the model are fairly decisively rejected by the data, particularly for forward rates with less than five years term to maturity. ... 
It is clear that our conclusions apply with equal force to the diluted form of the expectations hypothesis that allows forward rates to be determined by expectations plus time-invariant liquidity premiums. ... On the other hand, it would clearly be possible to determine a set of time-dependent "liquidity premiums" that could be used to adjust the forward rates so that the required sequences would display "white" spectral densities. ... While this procedure has its merits in certain instances, it is essentially arbitrary, there being no adequate way to relate the "liquidity premiums" so derived to objective characteristics of markets... 
An alternative way to "save" the doctrine that expectations alone determine the yield curve in the face of empirical evidence like that presented above is to abandon the hypothesis that expectations are rational. Once that is done, the model becomes much freer, being capable of accommodating all sorts of ad hoc, plausible hypotheses about the formation of expectations. Yet salvaging the expectations theory in that way involves building a model of the term structure that ... permits expectations to be formed via a process that could utilize available information more efficiently and so enhance profits. That seems ... extremely odd.
In other words: Observed yields are inconsistent with a simple version of rational expectations. They could be made consistent, but only by trvializing the theory by adding ad hoc adjustments that could fit anything. We might conclude that expectations are not rational, but that's too scary. So ... who knows.

One unambiguous point, though, is that under rational expectations interest rates should follow a random walk, and your best prediction of interest rates on a given instrument at some future date should be the rate today. Just saying "I don't now" is not consistent with rational expectations -- for one thing, it gives you no basis for deciding whether a product like the one being advertised here is priced fairly. Sargent's "No" is consistent, though, with fundamental uncertainty in the Keynes sense. In that case the decision to buy or not buy is based on nonrational behavioral rules -- like, say, looking at endorsements of recognized authorities. (Keynes: "Knowing that our individual judgment is worthless, we endeavour to fall back on the judgment of the rest of the world which is perhaps better informed.") So I stand by my one-liner.


  1. Sargent's known for his research into uncertainty

  2. OK, fair enough.

    Do you think his "No" here is consistent with rational expectations?

  3. What's wrong with "rational expectations" is that it is properly a modeling strategy. It's great virtue is that it makes for a solvable model. Somehow, when it leaps over the threshold to theory, description of the world as it is, and the rationalization of political desiderata, it becomes something of an article of faith.

    A solvable model is not likely to be a consistently accurate description of the world, for the very basic reason, which Keynes alludes to: people acting in the real world often face problems, for which they have no ready "model" and/or solution. A descriptive model of actual circumstances might be better as an approximation in critical respects, if it was not, in certain important respects, solvable, or if it, at least, acknowledged, that the solution, which the world will crank out in real time, cannot be anticipated, in quality or quantity.

    People like Sargent understand this, more or less, and feel their intellectual conceits are best expressed, by answering, "no", when asked if they can "predict" market outcomes, quantitatively. Their politics, though, rest on the parallel conceit that other people cannot predict quantitatively, or qualitatively. Is the interest rate, now, the "right" interest rate? Will the interest rate two years from now be the "right" interest rate? What is in theoretical and political dispute is what can be meant by "right", in relation to market interest rates (and, I suppose, related policy rates).

    I don't know what Sargent would say, if asked about the "right" interest rate. Does he think in terms of a goofy construct like a "natural" interest rate? I doubt it, but don't know.

  4. I'm trying to build a chain of reasoning, which will explain why I do not think it productive, as a criticism, to simply ridicule "rational expectations" or the related "efficient markets hypothesis" (or the general political worldview(s), in support of which they are sometimes invoked as authoritative economic insight) as naive or silly.

    Like Quiggin's assault on the EMH zombie, I think the critique by ridicule fails as an instrument of demolition, because it isn't placed close enough to structural flaws in the argument, to leave something superior standing. That might seem like an odd requirement: to leave something standing, but, in this case, it is hard to see how Sargent saying, "no", exposes his intellectual commitments -- even those commitments, which might be said to be at odds with Keynes.

    Where Lucas and Rational Expectations (and I presume without personal knowledge, Sargent) are at odds with Keynes is in their use of RE to resurrect and rationalize the classical theory of a self-equilibrating economy. In this, rational expectations is a kind of card trick, where the dealer palms cards and deals from the bottom of the deck. RE, by supplying a way of making solvable models subtly elides the problem of accurately describing or analyzing an economy confronting insolvable problems. The empirical evidence, as is well-known, tells against Lucasian RE, as a way of thinking about (descriptive analysis and hypotheses) the behavior of the actual economy, but it doesn't matter, because the invested effort in modeling solvable problems leads the New Keynesians to think about the actual economy as self-equilibrating, but slow and imperfect compared to the ideal, classical (RE) model. The point of management -- fiscal or monetary policy (or, I would add [but, tellingly, I'm not sure the New Keynesians would], banking/financial sector regulation) -- is to improve outcomes, by speeding adjustments. And, thus, we come to the ripe horror of post-Friedman economic policy "debate" in which the division of opinion is between those, who, in Mark Thoma's phrasing, think the economy needs help, and those, who simply doubt that we can know much of anything, and prefer almost-blind policy (Scott Sumner and his NGDP targeting) or classic laissez-faire or, like Mankiw or Taylor, feel perfectly free to lend credibility to partisan talking points or confidence fairies and the like.

    I started these comments, with my point shimmering like a mirage tantalizingly clear in the distance, and I fear, I may never get there. Oh, dear. [deep breath]

    Sargent is locked in a dialectic with the Krugman's, Thoma's (and Mankiw's), which seriously, seriously handicaps the thinking of the latter, well-intended albeit weak-minded handmaidens to plutocracy, or whatever you want to call our neo-liberals. The aspect of Sargent's intellectual edifice, which is worth destroying, is that dialectic that locks the Thoma's (and Quiggin's?) into an unproductive kind of orthodox league play. The latter already enjoy the conceit that they better acknowledge the messy reality in which, say, financial markets develop bubbles. It is just part of the established, and unproductive dialogue, in which Lucas and Sargent et alia have set an agenda, by dealing off the bottom of the deck methodologically speaking. Rational expectations is a kind of methodological crooked dealing, but how to expose it?

    The structure to blow up has to be a part that locks Krugman, Thoma, et alia into this idea of an (imperfectly) self-equilibrating economy, in relation to which the legitimate debate is about whether admittedly clumsy policy can be counted on to improve outcomes, or is likely to interfere with poorly understood "natural" processes of self-equilibration (re-calculation or whatever).

    The plain ridicule critique of RE doesn't touch Sargent (or Lucas, or Fama or Scott Sumner) on their own terms, and just confirms the New Keynesians in their conceits.

  5. OK, I admit I'm exhausted, and no closer to my mirage than when I started. I'll throw out a couple of thoughts, to justify my long-winded slogging, even I admit I didn't get where I thought I was going.

    A key difference between Rational Expectations and Keynes is that Keynes understood that you need a difference of opinion to make a horse race, or a financial market. RE collapses differences of opinion, as a way of arriving at a solution; the hidden agenda is making space for a know-nothing conservatism, which plausibly denies what can be observed. Financial markets, which only function when there is a persistent difference of opinion, cannot be modeled in a scheme, where everyone shares the same model and ends up, in equilibrium, sharing the same expectation. That's idiotic and self-contradictory. And, as you pointed out not long ago, as we approach the zero-bound (or whatever term you wish to employ for a context of especially low policy rates) financial markets cease to function reliably precisely because expectations do tend to converge on the conviction that rates must either stay the same or rise. Expectations are more rational and convergent than ever, and that's a bad thing. And, Ally bank, in the video, is playing to that telling common sense implication, whether Sargent endorses it or not.

    A more realistic model of financial markets would make use of conflicting (not rational or converging) expectations, and particularly, the hedging, which many agents engage in. Economists, who use RE to model their idea of the yield curve and explain why long rates are moving as they are relative to short rates, fail to grasp the ways in which agents manage uncertainty with hedges, with bets on relative rates. The usual central activity in any currency -- the carry trade of borrowing short and lending long -- is a hedge, in which the expectational calculation of the agent incorporates uncertainty and, critically, a difference of opinion, which contains some idea about just how "efficient" the market is, and (this is critical!) some strategic awareness about whether the bet will tend to move the market in a more or less "efficient" direction. (In plain terms, lots of agents are perfectly aware of opportunities to cheat the market; it's called arbitrage, and contra RE/EMH thinking, it doesn't necessarily make markets self-defeating arenas for strategically-aware agents.)

    The right thing to do, I think, is to accept the idea that financial markets are, or ought to be, efficient, but recast it so that efficiency is in a framework, in which it is possible to ask, and to assess from observation, just how efficient. That's not the same as simply dismissing the assertion of absolute efficiency as ridiculous, tempting as that course may be. In a priori analysis, a perfect financial market is absolutely efficient in the same sense that all deductive analysis is absolute (or categorical, as Kant might say). [That's the point Williamson made contra Quiggin, in relation to the EMH zombie.] But, an actual financial market is a mechanism, and just as a mechanical heat engine can be measurably efficient and its design analyzed to explain that degree of efficiency, so a financial market should be thought of as measurably efficient, its observable design features and norms, available to explain how efficient.

    You can take the view that the price matrix is seriously out-of-whack, and policy is necessary to put in back in a range of relative values in which the price martix can efficiently coordinate a decentralized economy. But, you have to have, not an analytical model, per se, but an interpretative modeling strategy, that can make use of observable prices and market design and functioning. In this, the New Keynesians are up the creek without a paddle.

    Not exactly where I thought I was headed, but better than nothing.

  6. i looked at the Sargent paper
    and I didn't understand a word of it; my guess, 3 years full time, i might (might) be able to understand some of the math.
    However, i have a certain rule of thumb: when the equations get to complex, with to many variables, you can always fit them to data, esp if the data are limited time series.

    I had this feeling about the Sargent paper; I'm sure he can find an equation that gives an R^2 value (or whatever is used in this sort of math in place of linear regression coefficients) that is >0.99995 (excel actually returned that yesterday on some real data from a chemisry experiment...all my coworkers are teasing me for massaging the data)