Response to Scott Moss
Chris Auld
entertainment of the gentle readers of sci.econ. ]
This note is in reply to Scott Moss's inaugural address, available on the web
at <http://www.cpm.mmu.ac.uk/cpmrep56.html>. In the course of private
correspondence with Scott regarding this essay he suggested this list as a
venue for a public response. I apologize if you think this list is an
inappropriate venue, although it is at least tangentially relevant since the
essay's point is to justify social simulation modeling.
I do not wish to argue against such modeling: I have done such modeling
myself. My point here is to take issue with Moss's serious, sweeping charges
against economists in general. Moss claims that economists are
"intellectually dishonest" "bad scientists" who should be "ignored by serious
social scientists." Moss gives several examples to demonstrate his claims. I
will briefly address each.
The first is "the Humbug Production Function." This addresses the validity
of Robert Solow's method for separating changes in national output caused by
changes in the supply of factors of production from changes in technology.
The details are irrelevant to Moss's argument: Moss wishes to show that
Solow's technique has been shown to completely wrong, even disavowed by
Solow, and is yet used anyways, thus showing the "intellectual dishonesty" of
the economics profession. Moss is very much mistaken.
The central issue revolves around a critique of Solow published in 1974 by
Shaikh. Moss tells his readers that Shaikh's critique demolishes Solow's
technique, a result Solow himself "accepted in substance" and which has
"never been successfully refuted." Moss proceeds to provide citation counts
showing that Solow is still oft cited whereas Shaikh is largely ignored,
allegedly demonstrating economists ignore criticism which undermines accepted
methods. In reality, Solow provided an absolutely devastating reply to
Shaikh:
Solow, R. (1974) "Law of Production and Laws of Algebra: the Humbug
Production Function: Comment," Review of Economics and Statistics
56:1, p121.
It is common practice when providing such a comment to thank the author,
comment on the thoughtfulness of the piece, and so on. Solow provides no
such banalities and goes straight for the jugular, opening his one-page reply
with:
Mr. Shaikh's article is based on misconception, pure and simple.
He proceeds to very clearly explain why the "Humbug Production Function"
Shaikh and Moss refer to is "nonsense." He closes his reply with "The humbug
seems to be on the other foot." Solow comes as close as one can in academic
journal to simply asserting Shaikh doesn't have the foggiest idea what he's
talking about. Solow certainly, unambiguously, forcefully and entirely
rejects Shaikh's critique.
Readers interested in the details of the issue are invited to read the
articles. The methodology is not the issue, the issue is whether Moss has
provided evidence to bolster his bold claims about economists. He certainly
has not: Moss claims that Solow "accepted in substance" what Shaikh had to
say, that Shaikh's "critique was never successfully refuted," whereas that is
clearly not the case. It is difficult to interpret Moss's claim as anything
other than rather spectacular intellectual dishonesty itself, which is
signally ironic.
Moss's second example bolstering his case is the response to a 1968 article
by Roy Radner. Quoting Moss:
Consequently, general equilibrium cannot exist unless individuals
have unlimited computational capacities. This would seem to be an
important result since it states formally that the theoretical basis of
economists' views of markets requires buyers and sellers in all markets to
have unlimited computational capacities. Though by no means as influential
as the Solow paper discussed above, the Social Sciences Citation Index
records 68 citations of the Radner paper since 1981. None of those papers
address Radner's conclusion that unlimited computational capacity is a
necessary condition for equilibrium when spot trading takes place over
time.
We are apparently supposed to believe that Radner's result is being ignored
because of its serious consequences, thus again revealing that economists
are dishonest and unscientific. People do not have infinite computational
ability, therefore general equilibrium theory must be wrong, therefore we
must ignore Radner (Moss's readers first must ignore the fact that Radner's
article was published in a popular mainstream journal.)
In fact, economics, like all scientific disciplines, uses models which
simplify reality. Making rather implausible assumptions about computational
abilities is just one such assumption. In this case, relaxing that
assumption forms both a research agenda looking into its consequences and is
often an issue in various models. Moss's attack here is no more substantive
than someone leaping up at the back of a lecture hall and complaining that
people don't actually have infinite lives, that there are more than two
countries in the world, or that there does not exist a continuum of
consumers. Moss is also mistaken when he asserts that no one has ever
addressed Radner's point; this point was previously made in slightly
different form by Savage (1954) in an oft-cited paper and Radner's specific
point has been discussed by Mongin and Walliser (1988), Smith (1991) and
Lipman (1991). More broadly, many economists have investigated issues in
computation and their implications. Moss is simply wrong, and even if he
were correct that no one has "addressed" this issue, he still would not have
demonstrated that economists are all "intellectually dishonest" "bad
scientists."
Moss's third example is more than a little vague. He briefly explains what
rational expectations means, then provides what he claims is the methodology
used by RE modelers:
1. Write down a rational expectations model.
2. Determine the equilibrium configuration of that model.
3. Replace the rational expectations agent with multiple agents represented
by genetic algorithms.
4. Simulate the model devised in step 3.
5. If the simulation converges to the corresponding rational expectations
equilibrium, write up the results and send to a journal.
6. If the simulation does not converge to the corresponding rational
expectations equilibrium, revise the model and/or the genetic algorithm
and go to step 2.
Moss's evidence for this questionable methodology is:
Steps 5 and 6 of this procedure have been specified inductively on the
basis of questions asked by me at seminars and workshops where such papers
have been read.
This is, of course, less than compelling. First, it is simply not the case
that all RE models are solved via genetic algorithms, indeed, that is to the
best of my knowledge not even a common solution technique for this type of
model (linearizing the marginal conditions around a steady state is one
popular solution method). And even if Moss's claim were correct, what of it?
If someone tries a numerical technique to find an equilibrium point in a
model and that attempt fails to converge to the equilibrium, prompting the
modeler to try again or use a different technique, why on earth is that
evidence the modeler is acting dishonestly or unscientifically?
Moss proceeds to claim that the fact economists never make simplifying
assumptions to make the model "more suitable for analyzing the problem" is
also evidence of his charges. This is unremarkable: simplifying assumptions
are made to make a complex reality less so and thus amenable to analysis.
All models (even verbal "models") in every discipline are of this form; one
makes simplifying assumptions because one has to. _Of course_ such
simplifications make the model less like reality.
Finally, Moss claims that economists never employ anecdotal evidence, which
we are supposed to believe is a serious fault. Moss misconstrues his
colleague's comment that he "does not believe anecdotal evidence." Lots of
economic ideas and theories are have a genesis in anecdotal observations
about how the world works. The comment decries use of such evidence to test
and measure such theories. _That_ would be unscientific: anecdotal evidence
is, of course, subject to massive selection biases amongst myriad other
problems. One can only guess that if economists did regularly use anecdotal
evidence to test theories and measure relationships, Moss would have instead
chastised them as unscientific for doing so.
In short, Moss has not supported his claims. I would strongly suggest to
Professor Moss that he make his case for the superiority of his simulation
methods on scientific grounds: show how they outperform methods in the
received literature. The end-run approach of trying to show that all of
economics is not only wrong but dishonest, and therefore Moss's own methods
are better on general principles, leaves much to be desired scientifically
and otherwise.
--
Chris Auld (403)220-4098
Economics, University of Calgary <mailto:au...@ucalgary.ca>
Calgary, Alberta, Canada <URL:http://jerry.ss.ucalgary.ca/>
Robert Vienneau
some simple questions:
1. The bulk of this work can be cited as an instance in which (a version
of) neoclassical theory has passed potentially falsifying empirical
tests.
True or False?
2. What answer would Anwar Shaikh give to the first question?
3. What answer would Robert Solow give to the first question?
4. Are your answers to questions 2 and 3 different?
--
r a Whether strength of body or of mind, or wisdom,
v c p or virtue, are found in proportion to the
i s e power or wealth of a man is a question fit
e . . perhaps to be discussed by slaves in the
n m c hearing of their masters, but highly unbecoming
@ a o to reasonable and free men in search of the
d e m truth.
r -- Rousseau
Don Dale
news:rvien-A302C6....@news.dreamscape.com...
> Solow's growth accounting framework has been widely applied. Here are
> some simple questions:
>
> 1. The bulk of this work can be cited as an instance in which (a version
> of) neoclassical theory has passed potentially falsifying empirical
> tests.
>
> True or False?
Evidence in support of the Solow model per se is mixed; see Baumol AER '86
and DeLong AER '88 for a discussion of unconditional convergence, and Young
QJE '95 for a sort of case-study approach. More complex models using the
Solow framework as a starting point have fared quite well; see
Mankiw/Romer/Weil QJE '92, Barro/Sala-i-Martin JPE '92, and Hall/Jones QJE
'99. The amazing thing about the Solow model is not that it doesn't
perfectly describe reality; the amazing thing is that a model of such
incredible parsimony and abstraction even begins to describe some aspects of
reality.
> 2. What answer would Anwar Shaikh give to the first question?
I don't know.
> 3. What answer would Robert Solow give to the first question?
I think he'd probably echo my thoughts above, though no doubt more
eloquently and with a few more insights.
> 4. Are your answers to questions 2 and 3 different?
Yes.
Don
Chris Auld
Rob doesn't understand the issues and misses the point the point of my
reply to Moss. The key issues are 1) whether Solow "agreed in substance"
with Shaikh and 2) whether Solow's position is strong enough such that
anyone failing to agree with Shaikh is inherently being "intellectually
dishonest." Anyone who bothers to read Solow can plainly see he
absolutely did not agree with Shaikh and his argument is very compelling.
I would be interested to know why Rob apparently thinks Shaikh "falsified"
Solow. That's not what Shaikh argued, nor what his example is supposed
to show. So the answers to Rob's questions are: false, false, false, no.
Robert Vienneau <rv...@see.sig.com> wrote:
>Solow's growth accounting framework has been widely applied. Here are
>some simple questions:
>
>1. The bulk of this work can be cited as an instance in which (a version
>of) neoclassical theory has passed potentially falsifying empirical
>tests.
>
> True or False?
>
>2. What answer would Anwar Shaikh give to the first question?
>
>3. What answer would Robert Solow give to the first question?
>
>4. Are your answers to questions 2 and 3 different?
>
Chris Auld
>Evidence in support of the Solow model per se is mixed; see Baumol AER '86
>and DeLong AER '88 for a discussion of unconditional convergence, and Young
Don,
The issue here is Solow's 1957 decomposition (the Solow residual and all
that), not Solow's 1956 growth model. The two are of course related in
many ways, but one could reasonably argue that the decomposition is still
useful even if one rejected the growth model.
Robert Vienneau
> "Robert Vienneau" <rv...@see.sig.com> wrote in message
> > Solow's growth accounting framework has been widely applied. Here are
> > some simple questions:
> > 1. The bulk of this work can be cited as an instance in which (a
> > version
> > of) neoclassical theory has passed potentially falsifying empirical
> > tests.
> >
> > True or False?
> Evidence in support of the Solow model per se is mixed; see Baumol AER
> '86
> and DeLong AER '88 for a discussion of unconditional convergence, and
> Young
> QJE '95 for a sort of case-study approach. More complex models using the
> Solow framework as a starting point have fared quite well; see
> Mankiw/Romer/Weil QJE '92, Barro/Sala-i-Martin JPE '92, and Hall/Jones
> QJE
> '99.
I'm going to read Don's answer as "True, more or less." I am only, at
best, generally aware of this literature. I suggest though, the correct
answer is "false", given that it apparently doesn't address the problems
raised by Shaikh and Franklin Fisher for interpreting empirical data in
the Solow framework.
> The amazing thing about the Solow model is not that it doesn't
> perfectly describe reality; the amazing thing is that a model of such
> incredible parsimony and abstraction even begins to describe some aspects
> of
> reality.
No, it's a matter of algebra. I am willing to explain the algebra, again.
> > 2. What answer would Anwar Shaikh give to the first question?
> I don't know.
I believe Shaikh's answer would be "false." Given relatively stable
income shares, the ability of data to fit a Cobb-Douglas production
function with the exponent equal to the profit share is a law of
algebra. The supposed empirical tests, I guess, are not potentially
falsifying.
Given Don's answer to this question, I'm not confident of his
assessment of the literature given above.
It might encourage others to read some of this material to note that
Solow's response, previously cited by Chris, is one page and Shaikh's
New Palgrave article on the Humbug Production Function is only four
pages long. I assume this article is in the New Palgrave dictionary
itself, not merely the volume on capital theory that I own.
> > 3. What answer would Robert Solow give to the first question?
> I think he'd probably echo my thoughts above, though no doubt more
> eloquently and with a few more insights.
I enjoy reading Solow. However, I find Don's comment "The amazing
thing..." typically eloquent, whether misguided or not.
Let's see what Solow has to say about his own work:
"Mr. Shaikh's article is based on misconception pure and simple. The
factor-share device of my 1957 article is in no sense a *test* of
aggregate production functions or marginal productivity or of anything
else. It merely shows how one goes about interpreting given time
series if one starts by *assuming* that they were generated from a
production function and that the competitive marginal-product relations
apply. Therefore, it is not only not surprising but it is exactly the
point that if the observed factor shares were exactly constant the
method would yield an exact Cobb-Douglas and tuck everything else
into the shift factor. That is what one would *want* such a method to
do."
I suggest Solow's answer would be "false." He agrees on substance with
Shaikh. Shaikh's supposed misconception, according to my reading of
Solow, is that Shaikh was attacking a position Solow would want to
defend.
(Solow does go on to argue that Shaikh's Humbug example is not nearly
as overwhelming as it first appears; he makes the point neatly with
some statistics. I notice Shaikh gives a different example in his New
Palgrave article and only refers to the Humbug example in passing.)
> > 4. Are your answers to questions 2 and 3 different?
> Yes.
Gee, Don, how should this be scored? Given your answer to (2) and (3), it
is technically correct. But it seems you missed the point. Perhaps Scott
Moss would say it's an example of intellectual dishonesty.
Don Dale
> "Don Dale" <da...@nospam.princeton.edu> wrote:
>
> > "Robert Vienneau" <rv...@see.sig.com> wrote in message
>
> > > Solow's growth accounting framework has been widely applied. Here are
> > > some simple questions:
>
> > > 1. The bulk of this work can be cited as an instance in which (a
> > > version
> > > of) neoclassical theory has passed potentially falsifying empirical
> > > tests.
> > >
> > > True or False?
>
> > Evidence in support of the Solow model per se is mixed; see Baumol AER
> > '86
> > and DeLong AER '88 for a discussion of unconditional convergence, and
> > Young
> > QJE '95 for a sort of case-study approach. More complex models using
the
> > Solow framework as a starting point have fared quite well; see
> > Mankiw/Romer/Weil QJE '92, Barro/Sala-i-Martin JPE '92, and Hall/Jones
> > QJE
> > '99.
>
> I'm going to read Don's answer as "True, more or less." I am only, at
> best, generally aware of this literature. I suggest though, the correct
> answer is "false", given that it apparently doesn't address the problems
> raised by Shaikh and Franklin Fisher for interpreting empirical data in
> the Solow framework.
Robert, I'm surprised at you. Given your preference for decades-old theory,
I would have thought that you, of all people, would have known that it was
the relative stability of factor shares which inspired Cobb and Douglas to
adopt their functional form in the first place. Does this mean that a
Cobb-Douglas aggregate production function is God's truth? Of course not.
Does it mean that a Cobb-Douglas aggregate production function is a damn
good and useful simplification? In my opinion, it does. This is one of the
reasons why Solow's growth accounting method, as extended by Young,
Mankiw/Romer/Weil, etc., has yielded substantial insight into differences in
living standards across both space and time.
> > The amazing thing about the Solow model is not that it doesn't
> > perfectly describe reality; the amazing thing is that a model of such
> > incredible parsimony and abstraction even begins to describe some
aspects
> > of
> > reality.
>
> No, it's a matter of algebra. I am willing to explain the algebra, again.
>
> > > 2. What answer would Anwar Shaikh give to the first question?
>
> > I don't know.
>
> I believe Shaikh's answer would be "false." Given relatively stable
> income shares, the ability of data to fit a Cobb-Douglas production
> function with the exponent equal to the profit share is a law of
> algebra. The supposed empirical tests, I guess, are not potentially
> falsifying.
>
> Given Don's answer to this question, I'm not confident of his
> assessment of the literature given above.
>
> It might encourage others to read some of this material to note that
> Solow's response, previously cited by Chris, is one page and Shaikh's
> New Palgrave article on the Humbug Production Function is only four
> pages long. I assume this article is in the New Palgrave dictionary
> itself, not merely the volume on capital theory that I own.
It's originally from the RES, 1973.
> > > 3. What answer would Robert Solow give to the first question?
>
> > I think he'd probably echo my thoughts above, though no doubt more
> > eloquently and with a few more insights.
>
> I enjoy reading Solow. However, I find Don's comment "The amazing
> thing..." typically eloquent, whether misguided or not.
>
> Let's see what Solow has to say about his own work:
>
> "Mr. Shaikh's article is based on misconception pure and simple. The
> factor-share device of my 1957 article is in no sense a *test* of
> aggregate production functions or marginal productivity or of anything
> else. It merely shows how one goes about interpreting given time
> series if one starts by *assuming* that they were generated from a
> production function and that the competitive marginal-product relations
> apply. Therefore, it is not only not surprising but it is exactly the
> point that if the observed factor shares were exactly constant the
> method would yield an exact Cobb-Douglas and tuck everything else
> into the shift factor. That is what one would *want* such a method to
> do."
>
> I suggest Solow's answer would be "false." He agrees on substance with
> Shaikh. Shaikh's supposed misconception, according to my reading of
> Solow, is that Shaikh was attacking a position Solow would want to
> defend.
Did you miss the bit where Solow writes "All this has literally nothing to
do with the question whether the empirical basis of aggregate production
functions is strong or weak"? This is a direct contradiction of Shaikh's
thesis.
> (Solow does go on to argue that Shaikh's Humbug example is not nearly
> as overwhelming as it first appears; he makes the point neatly with
> some statistics. I notice Shaikh gives a different example in his New
> Palgrave article and only refers to the Humbug example in passing.)
>
> > > 4. Are your answers to questions 2 and 3 different?
>
> > Yes.
>
> Gee, Don, how should this be scored? Given your answer to (2) and (3), it
> is technically correct. But it seems you missed the point. Perhaps Scott
> Moss would say it's an example of intellectual dishonesty.
Perhaps Scott Moss would say any number of things. Why should I care?
Don
Chris Auld
>"Don Dale" <da...@nospam.princeton.edu> wrote:
>I believe Shaikh's answer would be "false." Given relatively stable
>income shares, the ability of data to fit a Cobb-Douglas production
>function with the exponent equal to the profit share is a law of
>algebra. The supposed empirical tests, I guess, are not potentially
>falsifying.
"Supposed empirical tests" implies Solow's decomposition is "supposed"
to be test. It isn't a test. It wasn't proposed as a test. Shaikh's
critique is a non sequiter.
>I suggest Solow's answer would be "false." He agrees on substance with
>Shaikh. Shaikh's supposed misconception, according to my reading of
>Solow, is that Shaikh was attacking a position Solow would want to
>defend.
Solow DOES NOT "agree in substance" with Shaikh. He disagrees in no
uncertain terms. Shaikh's article shows that Solow's decomposition
is not a test of Cobb-Douglas aggregate production functions. But
Solow never proposed it is a test, so it's just "nonsense" and
"misconception" to attack it on those grounds. Do you understand,
Rob?