-----Original Message-----
From: Alejandro Ramos <aramos@btl.net>
To: ope-l@galaxy.csuchico.edu <ope-l@galaxy.csuchico.edu>
Cc: multiple recipients of list <ope-l@galaxy.csuchico.edu>
Date: Wednesday, February 17, 1999 5:28 PM
Subject: [OPE-L:434] New Evidence on Sectoral Prices and Values
Ale: "Where are those "prices" and "values" come from? How do you
get them? How are measured? Are you talking about "market
prices"? Did you take out rent effects (oil, e.g.)?"
The data come from published National Income and Product Accounts
(NIPA) of the Bureau of Economic Analysis (BEA), including NIPA
series on depreciation and intermediate inputs, by industry. Some
aggregation was necessary due to lack of complete data on
intermediate inputs and, for 4 industry groups (out of 42), I had to
estimate intermediate inputs on the basis of benchmark I-O data,
also published by the BEA.
Here's the relationship between my variables and the BEA categories
(the BEA categories are Capitalized). Everything is measured in
current prices:
market price = Gross Output
nonlabor cost = Depreciation + Intermediate Inputs
labor cost = Compensation of Employees + imputed compensation of
self-employed
total cost = nonlabor cost + labor cost
profit = market price - total cost
value = nonlabor cost + monetary value added
monetary value added = (labor cost)*(1+s'), where s' is the average
rate of surplus-value = (aggregate profit/aggregate labor cost)
(Note: this is what is done in the rest of the literature. Instead
of using the actual living labor, and multiplying by the MELT, it is
assumed that s' is uniform.)
surplus-value = value - total cost = monetary value added - labor
cost.
With respect to rent effects, the answer is that I replicated
exactly the industry coverage of the best known studies of the U.S.
economy, those of Ochoa and Shaikh. Government and private
household sectors are not included, and the real estate sector
is also excluded, which obviously has to do with rent. Agriculture
and oil sectors are included. Also *included* are sectors that
Shaikh and Tonak (_Measuring ...) consider unproductive, namely
wholesale and retail trade, finance and insurance, and business,
legal and misc. professional services.
I show in the paper that the inclusion of the latter has important
effects. When these unproductive industries are excluded, the
elasticity of price with respect to value becomes *negative*
(about -0.25), and the elasticity coefficient is significantly
different from zero at the 1 percent level.
Ale: "Could you please give us an example of the spurious
correlation you are pointing out which doesn't involve "values"
and "prices"?"
Sure. Let's say you have a Dog Ownership Theory of Employment. Dog
ownership regulates employment, has predictive power over
employment. To prove that dog ownership is the dominant determinant
of employment, you get data from a bunch of countries of *different*
sizes (the more different the better), i.e., the number of dogs
owned and the number of people employed. Then you measure the
correlation. Lo and behold, in large countries, a lot of dogs are
owned and a lot of people are employed, and in small countries,
fewer dogs are owned and fewer people are employed. You'll get a
very high correlation. You have just proven the empirical strength
of the Dog Ownership Theory of Employment. Congratulate yourself
for your scientificity, sneer at the obscurantist TSS ideologues,
and
suggest that their research be burned.
Ale: "As far as I understand the issue, the matter is that the
absolute size of a given sector does provoke a "correlation" in
itself. For example, imagine two variables, "sales" and "debt"
which are not "correlated" in the sense that *they are not
equal* (i.e. sales/debt is not equal to 1), as it seems the
theory your are testing does regarding prices and values. But as
"big sectors" have "big sales" and "big debts", and "small
sectors" have "small sales" and "small debts", we can get a fine
correlation if we compare the absolute magnitudes. Is this the
argument?"
Yeah. Although there may be an underlying economic relationship
between sales and debt, whereas there's no relationship between
sectoral values and sectoral prices. For instance, sectors that
are doing well (in terms of sales) can more easily bear the
burden of a larger debt.
In any case, the spurious correlation stuff is well known. I take
no credit for it. What is new in this study is (1) price-value
correlations that are NOT affected by industry size and (2) a
demonstration that prices and values are NOT close in the sense
required by the labor theory of relative prices. The latter point
is very important, because a number of people have been willing to
renounce the aggregate sectoral correlations, ecause they had
"small" price-value deviations as backup evidence. Now they
have neither, it seems.
So I want to make absolutely sure that the content of this paper
doesn't get reduced to "Kliman says there's spurious correlation."
Rather, the one-sentence version is "Kliman shows that values and
prices are not correlated, nor are they close in any meaningful
sense."
Ale: "So, to get ride of that "absolute effect" you are trying with
something like this:
W/K and P/K
where K is cost-price and, then, correlating them. Right?"
Yeah. Exactly.
Ale: "Can you perform some test on something like
P/W
directly?"
I'm not sure. Let me think about it. The obvious answer is no,
because you need two variables to have a correlation, and P/W is
only one. On the other hand, you could check the correlation
between P/W and the value composition of capital, and if you found a
statistically significant correlation, that would be evidence
against the labor theory of relative prices. I'll think about it
some more.
Ale: "Could you please explain why did you run "log-linear"
regressions?"
Oh goody. I get to give an answer that I haven't given since I was
a teenager: because everyone else does it!
Just as I chose the industries I did in order to replicate the Ochoa
and Shaikh studies, I also chose the functional form they chose
because they chose it. (This is also the specification of choice of
Cockshott and Cottrell.) Of course, they regressed aggregates,
while I regressed cost-weighted variables.
A secondary reason is that I wanted to bend over backwards to be
fair to the proponents of the labor theory of relative prices, and
the log-linear specification gave results that were more favorable
to their claims than a linear specification. But, of course, not
favorable enough to make any real difference.
Ale: "Does each regression correspond to one year?"
Yes. 21 regression, 21 years. Except the "pooled model" is
different. "Pooling" means you run *one* regression with all of
the industries in all of the years. E.g., I had 42 observations
(industries) for each year, and 21 observations (years) for each
industry, so I had 42 x 21 = 882 observations in all. BTW, I
think that this may be the largest study thus far, and that
the next largest is Ochoa, who had a total of 71 x 9 = 639
observations.
Ale: "Can't you specify some kind of "TSS model" in which there are
time lags? Values of one year "determining" prices of the next?"
Well, none of this has anything to do with TSS, which is an
interpretation of *Marx's* value theory, not the labor theory of
relative prices. But I could regress prices on last year's values.
I didn't, because I was testing *their* hypotheses, and a lagged
effect isn't one of them.
Ciao
Drewk