Wednesday, May 02, 2007

Irrelevant Aside On Friedman's Methodology

You may sometimes read (neoclassical) "Economic theory shows" p. Often these claims are untrue. One can prove this falsity by constructing examples in which the assumptions of (neoclassical) economic theory hold, but not p results. So the statement you have read is logically flawed.

I sometimes encounter people who respond with the non sequitur, "Assumptions do not need to be realistic." If this response made sense, those who propound it would be able to specify which assumption is claimed to be unrealistic in these examples. Good neoclassical economists understand this logic.

Anyways, those economists that insist that assumptions do not need to be realistic - that, in fact, unrealistic assumptions are actually desirable - probably take their position from Milton Friedman. From Steve Keen's Debunking Economics, I found a response by Alan Musgrave to this idea. Musgrave, perhaps when combined with a later Uskali Mäki comment, seems to me perceptive. From these two authors, one can get a taxonomy of assumptions. Here's Musgrave's, albeit with the first reworded following Mäki's suggestion:
  • Negligibility: "The hypothesis that some factor", or combination of factors "has an effect upon" "the phenomenon under investigation" "small enough to be neglected relative to a given purpose".
  • Domain: A specification of the domain of applicability of a theory.
  • Heuristic: An assumption for developing a theory in a simple form that can be removed in later approximations.
The same wording might be used to express an assumption in all categories. For example, a model assumption might be "The government has a balanced budget". This might mean that the effects of budget imbalances have a negligible effect on the results of the model, that the model only applies when the budget is balanced, or that the model can be developed in a more complicated fashion with no assumption on the balance of the budget. Pace Friedman, assumptions, properly paraphrased, better be true ("realistic") in applications.

Update: Radek doubts some economists write "Economic theory shows..." As far as I am concerned, this is close enough:
"These observations should not be puzzling, for they are what standard economic theory predicts." -- Edward C. Prescott (1986). "Theory Ahead of Business Cycle Measurement", Quarterly Review, Federal Reserve Bank of Minneapolis (Fall): 9-22
References
  • Milton Friedman (1953). "The Methodology of Positive Economics", in Essays in Positive Economics, University of Chicago Press
  • Uskali Mäki (2000). "Kinds of Assumptions and Their Truth: Shaking an Untwisted F-Twist", Kyklos, V. 53, Fasc. 3: 317-336
  • Alan Musgrave (1981). "'Unreal Assumptions' in Economic Theory: The F-Twist Untwisted", Kyklos, V. 34, Fasc 3: 377-387

18 comments:

Gabriel Mihalache said...

If I tell you the story of Spiderman to show that "With great power comes great responsability." you'll reply that getting superpowers from the bite of a radioactive spider is absurd.

Anonymous said...

Well, if you are making policy recommendations to "improve" the economy I would like to think that the model you use is based in reality.

The classic example, of course, is Friedman's own Monetarism. With the given (unreal) assumptions, it worked fine in theory. In practice, well, it helped cause the deepest recessions since the 1930s.

Basically, if you assume enough surreal and insane assumptions you can prove anything is true. This seems to be the logic of neo-classical economics. When the absurd assumptions underlying the conclusions are pointed out, economists say they do not matter.

Hence marginal productivity theory can be used to justify rising inequality regardless of its assumptions having on bearing to reality. Which is good if you are wealthy, of course.

I'm just glad that engineers do not apply the same methodology. We would have lots of building which would work fine, assuming that gravity did not exist...

Anonymous said...

To continue the Spiderman analogy...

How to explain exploding inequality? Why are CEOs getting such high pay?

Let us assume two groups of people. One group, a minority, get bitten by a radioactive spider. These people gain superhuman levels of strength, dexterity and senses. The rest do not.

By competition, the bitten group rise up the corporate and social hierarchy due to their spider-enhanced abilities. Their high wages, therefore, reflects this. Clearly, then, CEOs today deserve their pay due to their superhuman abilities rather than, say, their position in the corporate hierarchy.

Ah, critics point out, this is absurd. They have not been bitten by radioactive spiders. To which the economist replies, the market
works "as if" they had been...

Ultimately, I think economics should be a bit more than self-serving "just-so" stories.

Iain

PS Steve Keen's book is excellent. Well worth reading, as is his webpage (www.debunkingeconomics.com)

Gabriel Mihalache said...

Iain, you're more than welcomes to offer an alternative model.

Why waste your time "debunking economics" when you could be taking your insights and coming up with a really great theory?

YouNotSneaky! said...

In regard to Prescott's statement. I agree, that's not good.

YouNotSneaky! said...

The classic example, of course, is Friedman's own Monetarism. With the given (unreal) assumptions, it worked fine in theory. In practice, well, it helped cause the deepest recessions since the 1930s.

And what exactly are these unreal assumptions? For moneterism to "work" you basically need a fairly stable money demand function. And prior to the late 70's, the money demand function had indeed been stable. So this particular assumption was quite realistic. Of course the world is under no obligation to stay constant and be nice to economic theories, so in the late 70's, due to changes in transaction technology and deregulation of the financial system, money demand became volatile. As a result the recession associated with the Volcker disinflation was more severe than anyone anticipated. On the other hand inflation came down from 13% to 3% when money supply was frozen - validating the "monetary phenomenon" aspect of monetarism.

Anonymous said...

To quote an economist, Thomas Balogh (from The Irrelevance of Conventional Economics (Weidenfield and Nicolson, London, 1982))

Friedman's assumptions have "been shown to be fallacious and the empirical evidence questionable if not totally misinterpreted." (p. 165)

"none of the assumptions which Friedman made to reach his extraordinary conclusions bears any relation to reality. They were chosen precisely because they led to the desired conclusion, that inflation is a purely monetary phenomenon, originating solely in excess monetary demand." (p. 167)

He also suggested this excellent summary of Friedmanite policies in practice, namely that governments influenced by Monetarism were "deliberately setting out to base the viability of the capitalist system on the maintenance of a large 'industrial reserve army' [of the unemployed] . . . [it is] the incomes policy of Karl Marx." (pp. 177-8)

YouNotSneaky! said...

Umm.... let me repeat myself:

What exactly are these unreal assumptions?

Gabriel Mihalache said...

It has become quite a nasty fashion, for heterodox folk, to use the following pseudo-argument when dealing with the "autistic" mainstream:

"Heterodox author X said blah, blah, blah (sundry value judgments)".

Tsk, tsk.

Steve Keen comes up a lot, esp. his point about Cobb-Douglas and the national accounts identity. -- A basic familiarity with the history of the Cobb-Douglas function will show the irrelevance and wildly-speculative nature of that argument.

Robert Vienneau said...

No, Gabriel. That argument is due to Anwar Shaikh, building on work of Franklin Fisher. I believe Herbert Simon mentioned it in his "Nobel" acceptance speech. As of 2005, Fisher, Shaikh, and others still accept the argument.

Gabriel Mihalache said...

Douglas asked Cobb for a functional form that would exhibit certain behaviors: homogeneous of degree one, constant shares, etc. It's a function, not a theory.

So when you apply it to variables of the national accounts, this is what you get: the accounting identity, because that's where you got the variables from; and constant shares and homogeneity, because that's how you picked the functional form.

The function is a mathematical expression of the stylized facts of the national accounts in Western nations.

It's not a proof of anything. People attack it like it's a proof. This is the nature of their misunderstanding.

Many theories can explain this approximate constancy of shares, by various means. There's nothing particularly neoclassical about this.

h.e. said...

It's a function, not a theory.


So what? The function by itself says exactly nothing about factor shares; you need to derive factor demands from the function before you can make any claims that the Cobb-Dogglas production function is useful for explaining factor shares. At this point in time you have theory, which is clearly neoclassical.

h.e. said...

It's not a proof of anything. People attack it like it's a proof. This is the nature of their misunderstanding.

One purpose of a theory is to provide an explanation for an observed outcome. Ability to fit the data is not the only criteria by which a theory can be evaluated. Shaikh's main point is that the ability to fit the data is not terribly surprising, given the relative stability of factor shares in the U.S., and doesn't provide terribly compelling support for the theory.

The main assaults on the neoclassical theory are theoretical and have not been refuted. Hence the reliance on an instrumentalist defense. This doesn't mean that critics don't understand the theory. It means they don't accept an instrumentalist defense of a theory that seems to be impossible to generalize outside of simple two factor models, where each factor is homogenous...

YouNotSneaky! said...

I think the main point on this is that IF you think that an aggregate production function is a useful way to describe the national economy THEN the Cobb Douglas one is the only one which fits the stylized facts - constant shares, r constant in steady state, wages growing. IF you don't think so, well, then you don't care whether it fits the data or not apriori.

It is a theory in the sense that it does provide an explanation for WHY shares and r, as it defines r, are constant in the steady state - basically beta-convergence (which itself is a consequence of the Inada conditions, not dminishing returns as is often said).

Another quick note on the CD function - even if you don't think markets are competitive, aggregate or non aggregate production functions don't exist and marginal product conditions don't hold - the CD production function could still be useful in the sense that it could faithfully represent the bargaining outcome between labor and capital, if that bargaining takes the form of the Nash solution. In that case alpha rather than being a technological coefficient measuring diminishing returns to capital, whatever that is, measures the relative bargaining power of capitalists vis a vis workers.

Sometimes "as if" arguments are perfectly valid.

Robert Vienneau said...

"The main assaults on the neoclassical theory are theoretical and have not been refuted." Thanks, H.E.

For example, I see no reason to accept a theory that contains an aggregate production function and that equates its derivative with respect to capital to the equilibrium interest rate. I do not believe Radek's response to H.E. addresses my concern.

h.e. said...

Sometimes "as if" arguments are perfectly valid.

This assertion is so vague that I don't know how to evaluate it.

Your mention of Nash bargaining reminds me of what my main concerns are with mainstream economics: the strange fetish with mathematical formulism. Nash bargaining is pretty typical of economic theory; the bargaining outcome is required to satisfy a set of axioms. It's never clear *why* bargaining solutions should have these properties, outside of having "nice" logical properties.

"Capital" in economic models demonstrates another aspect of the fetish for models: a lack of concern with what properties the "real world" variables have and no apparent concern about how they could be measured. The various indices that economists use without question are, as I understand it, constructed using index theory, which has no direct connection to economic theory, but requires that the indicies in turn satisfy a different set of axioms.

Now these axioms in no way resemble the laws of physical sciences. There is no laboratory component to economics education; the relevence of microfoundations seems to rest largely on a kind of faith.

Anyhow, I understand that models aren't always going to be realistic, but it's not clear to me that a purely deductive reasoning process can form the basis of an empirical science...

YouNotSneaky! said...

I didn't claim it did. It was an IF...THEN statement and if you don't believe the IF then the THEN doesn't follow.

Still, the empirical fact of constant factor shares has to be addressed/explained. Part of the point is that the derivative of the production function with respect to capital does not have to be equal to the interest rate. It could be generated via bargaining or some other way. But still, a Cobb Douglas production function could describe things pretty well... "however it's generated" (quoting your previous post). Of course in that case marginal productivity theory doesn't hold.

If income shares are constant, then the (neoclassical) proposition that there’s a negative monotonic relationship between r and K (or w and L) amounts to the proposition that when K goes up, Y goes down less than proportionally – in other words, diminishing returns to capital. This is independent of how shares are determined – whether in competitive markets, monopolistic markets, monopsonistic markets, by bargaining or by fiat, as long as they stay constant.
One objection to this is that no such thing as K (or L) exists. In some sense I’m sympathetic, although in this day and age I think the distinction between different types of labor is actually more important than the distinction among heterogeneous types of capital. But this isn’t enough, because the strange fact that share of whatever it is economists are measuring when they think they’re measuring K, has stayed constant. So one still needs to explain why this artificial or implausible or completely fallacious thing that the neoclassicals have been calling K is so well behaved. If we’re measuring nonsense, then why has this nonsense been getting a constant share of output for the past 100 years or so?

Your mention of Nash bargaining reminds me of what my main concerns are with mainstream economics: the strange fetish with mathematical formulism.

As opposed to the Sraffian heavy use of linear production models and linear algebra? And you're exaggerating the extent of mathematical formulism in mainstream economics. Maybe in the heydey of the 1970's (even then you had the pretty informal Friedman) but the pendelum has definetly swung the other way. See all the essentially theory-free empirical work under the heading of "Applied Micro".

Nash bargaining is pretty typical of economic theory; the bargaining outcome is required to satisfy a set of axioms. It's never clear *why* bargaining solutions should have these properties, outside of having "nice" logical properties.

No, axioms aren't about "nice logical properties" they're about reasonable starting assumptions (and yes, in many context IIA is a reasonable axiom, while in others it's a very strong one).

The various indices that economists use without question are, as I understand it, constructed using index theory, which has no direct connection to economic theory,

But this brings me back to the original point! If K is measured independently of the theory which requires it's share in national income to be constant, why has this share stayed constant for so long?

Now these axioms in no way resemble the laws of physical sciences. There is no laboratory component to economics education; the relevence of microfoundations seems to rest largely on a kind of faith.

Of course they don't resemble physical laws, why should they? But since there's no laboratory component (for obvious ethical and other reasons) what's the second best? Ummmm, inferential statistics combined with logical deduction? And IS + LD != faith, but reason and data.

Anyhow, I understand that models aren't always going to be realistic, but it's not clear to me that a purely deductive reasoning process can form the basis of an empirical science...

But it ain't purely deductive. This whole discussion was about how Kaldor noticed some facts about the real world - constant income shares, constant interest rates, wages growing at a constant rate - and people began to wonder what kind of theory/functional form would fit this data. Basically the CD function may not be the way to go, but it passes the first test of producing results that look like the data we observe. Any model that doesn't pass this test is probably not very good. Of course one could go with a model where there's enough indeterminacy to fit any kind of fact instead.

Gabriel Mihalache said...

Vernon Smith does *Experimental* Economics. See here:

http://gabriel.mihalache.name/econ/2007/04/prices.php

Mostly related to market outcomes. Helas, his research supports much of the "neoclassical orthodoxy".