[OPE-L:6966] [OPE-L:458] Value-price correlations

David Laibman (DLaibman@brooklyn.cuny.edu)
Sun, 21 Feb 99 16:05:00 EST

(Sorry! I have forgotten the thread name.)

Dear OPEople,

I have been reading the discussion between Andrew, Allin, and Alejandro
on Andrew's new work showing absence of correlation between labor values and
observed money prices. I am technologically challenged; the only way I
could study these posts carefully would be to print them all, but then after
a few weeks I would not have a working printer left, and I can't afford to
replace toner/drum, etc., at the moment. So I must work entirely from
memory. I also have not personally done this sort of empirical study, so my
contribution to the discussion (in the spirit of Jerry's encouragement of
participation) is truly more questions than answers.

Surprise! For once, I find myself agreeing with Andrew and Alan! I
have always been suspicious of the labor values derived from I-O data, and of
the high correlation coefficients between them and observed prices. With
Andrew, I see no reason in theory why we should expect to find a high
correlation. Alan's point, further studied by Andrew, that a high
correlation would emerge simply from the fact that absolute (rather than
unit) values and prices are both highly correlated with the size of
industries seems on the mark. Alejandro's point about the general
reliability of the data is also quite important. The Shaikh-Ochoa matrix
inversion procedure produces *not* unit values (direct plus indirect labor
time per unit of output), but rather: direct plus indirect labor time per
dollar (or other money unit) value of sales. Without obtaining a separate
series proxying physical output by sector, the coefficients themselves are
distorted (I suspect), except along the principal diagonal. This alone, and
independently of the scale effect, may explain the high correlations found in
those studies and related ones. There is also the matter of heterogeneous
labor, and I doubt if there is an agreed-upon method of dealing with this
complication.

Even 80-sector I-O data are highly aggregated. Full disaggregation,
were it achievable, would most likely reveal wider dispersion in compositions
of capital. If it were possible to measure unit labor values, unit
production prices and market prices at a moment in time, one would want to
know the extent of deviation among all three measures, and I would expect
each deviation (production price from value; market price from production
price) to be significant. Even here, however, there is the question: how
large is significant? If labor values turned out to "explain" 93% of
observed prices, is that a lot or a little? I think we can have significant
difference among the measures, and still retain Allin's insight that the
price difference between a peanut and a computer (or, as Joan Robinson once
put it, between an automobile and a packet of pins) is most centrally
explained by the different quantities of labor time these commodities
contain.

I don't know enough about the statistical measures to weigh in on
Allin's critique of Andrew's proposal for eliminating the scale effect. I
would only note that *if* Allin's critique is correct, that would only mean
that some other way of removing that spurious source of correlation between
"values" and "prices" would have to be found.

Finally, a question for Andrew. I do not want to go over the whole TSS
terrain again, but the question of empirical measurement of values (even if
only hypothetical) raises one issue I can't resolve. You may identify *value
added* in any period of time with the current labor performed in that period
(leave the heterogeneity problem to one side). But to estimate *indirect
labor time* and gross value, *some* unit value must be applied to consumed
means of production. Even if we are time-sensitive and want to make this
*yesterday's* unit value, the same problem arises when we try to calculate
*that* one, and so on in infinite regress. We can't use observed money
prices of inputs; this would compromise the whole project, since the idea is
to compare values with precisely the observed price magitudes (and, if Andrew
is correct, find them essentially uncorrelated). Since the TSS approach
insists on different market values at different dates, it is not clear how
one could calculate unit values at a moment of time on these assumptions,
even in principle.

Cheers,

david

David Laibman