[OPE-L:6967] [OPE-L:459] New evidence on sectoral prices and values

Alan Freeman (a.freeman@greenwich.ac.uk)
Sun, 21 Feb 1999 05:14:04 +0000

It's worth noting that the beginning of this discussion is now in print
with the publication of Riccardo Bellofiore's compilation of papers to the
1994 Bergamo conference, ("Marxian Economics: a re-appraisal", McMillan
1998). I commented Anwar's paper and raised the issue there. I subsequently
developed this in "Time, Money and the Quantification of Value" to the 1997
EEA conference which is on the IWGVT website at

I'd like to pick the thread up with the following remark of Allin's:

"The rest of Andrew's argument boils down to saying, "if you know
the money cost of production of a commodity already, then
knowing its value is not likely to help in predicting its
price". But the notion behind the labour theory of value is
that you can predict prices quite well on the basis of _just_
labour values, without being "given" the money cost of
production, which is a market phenonemon at the same sort of
level as prices."

Sure. But:

(a) No-one disputes that labour values tell you something about prices.
However this is not the same as saying that price is identical to value.
Age is likely to help in predicting height. This neither means that
everyone with the same birthday has the same height, nor that anyone grows
an inch each year of their lives.

The argument which Andrew and I are criticising doesn't merely say that
value is a good guide to price. It says that the difference is so small
that things like the transformation problem don't matter. This is
statistically equivalent to saying that since age predicts height, growth
is theoretically unimportant.

(b) The *statistical* claim made about value and prices in the literature
that we criticise is fallacious, and I continue to regard this as very
dangerous, particularly if it is used as the basis of qualitative
statements, like "the transformation problem is empirically insignificant".
It will rightly be seen by Marx's opponents, once they have cottoned on, as
a trick to avoid confronting the underlying theoretical questions. The
statistical fallacy consists in reporting a variation caused by one factor
as if it were caused by another.

We might try, for example, to investigate whether age is a good predictor
of height by measuring the total height of an aggregate of humans, perhaps
by laying them end to end. Inevitably, we would find that over a large
sample, two humans tended to be twice as long as one human. Likewise two
pizzas are twice as expensive as one pizza, two computers are twice as
expensive as one computer, etc. Unfortunately, this conveys no information
that wasn't already present in the measurement of one human, one pizza, and
one computer. It happens because we have built the same additive relation
into the calculation of aggregate values or ages as into the calculation of
aggregate price or height.

Allin says that the amount of steel in an object isn't as good a predictor
as the amount of labour in it. Sure, because Pizzas don't have any steel in
them. Testicle size is a bad predictor of human height. But this problem
arises just as much with a single Pizza, and a single human, as with many
pizzas or many humans.

What happens when we look at sectoral differences? Even presuming that the
measure of 'quantity' that we obtain in an I/O table really does measure
some homogenous unit of quantity (which it doesn't), the differences
between the values and the price output of the sectors arise for two quite
distinct reasons, which are simply mixed up in the method of correlation
which has become the general practice. On the one hand the prices and
values of the goods produced by each sector bear a different relation to
each other. In some sectors, an hour of labour will sell for say $20 and in
other sectors, for $10. But on the other hand, the sectors are just
different sizes. The variation in aggregate price, and the variation in
aggregate value, will occure because of both these causes. The problem of
spurious correlation is quite straightforward: the second variation swamps
the first.

We would get the same result if we correlated the total heights of
institutions with each other. But the fact that the total height of a
kindergarten is a lot less than the total height of a University tells us
two things that mustn't be mixed up together. On the one hand it tells us
that toddlers are smaller. But on the other, it tells us that Universities
simply have more people in them. You can't report the one source of
variation as if it was a proof of the other.

If we really want to investigate the relation between height and age, the
only reason for considering an aggregate instead of a single individual, is
that we might get a sampling error with an individual. The appropriate
comparison is to compare the *average* height of a kindergarten child with
the *average* height of a University student.

Therefore if we want to study the relation between value and price in its
*pure* form, we should completely eliminate the variation due to quantity.
We should compare the price of an *hour's* labour from one sector with the
price of an *hour's* labour from another. When we do this, we find that
they are smeared out around the MEL with a standard deviation of around
10%. We find in some sectors that prices are as much as thirty percent
greater than in others, and in other sectors that they are as much as
thirty percent less.

This confirms that labour time is a good predictor of price. If, for
example, we looked at price compared with 'quantity of steel' I guess that
we would find the price of an embodied ton of steel varied a lot more than
the price of an embodied hour of labour. We should report this fact as it
stands, not by comparing the price of a ten million embodied tons of steel
in one sector with a thousand embodied tons of steel in another.

However, it does not confirm that labour time is identical to price. A
variation of 30% is NOT insignificant and does NOT justify ignoring the
theoretical problem of transformation.

Finally, my greatest objection to the whole procedure is that it
masquerades as a test of "Marx" by testing a claim that Marx vigorously
rejected and for excellent reasons. For Marx, the core of all his polemics
against Proudhon and the time-chitters, and the centre of his analysis of
the functioning of the market, is that the *divergence* of price from value
is the motor force of the market. It is only *because* you can get more
labour for your pound in one place than another, and moreover more profit
in one place than another, that capital moves from one place to another and
technical progress takes place. The continuous divergence of price from
value is the underlying mechanism by which the market even exists. It is
not just an accident or something discountable, any more than the flow of a
river is an accidental disturbance of the oceans, or living beings can be
conceived of as disequilibriated corpses.

Marx himself never spoke of the labour theory of value. The term was, as
far as I can acertain, invented by Kautsky. Marx speaks of the 'law of
value' and I think marxist discourse should revert to this term not just
because Marx used it, but because it is better. The law of value is a law
of *motion*; when Marx speaks of the empirical relation between value and
price, he always refers to a relation in time. What he actually says is
that prices conform to values on average over a prolonged period, usually
that of the cycle. What we should be investigating (Ochoa did this but his
successors have never followed up on it) is investigating the extent to
which *changes* in price are predicted by *changes* in value. If we did
this, the correlations we would be studying would be genuine ones, because
we would concern ourselves with a genuine independent source of variation,
namely, technical change. And, I suspect, we would find a very close
correlation between the two.