Cambridge Journal of Economics, 1982, 6, 353-374

The neo-Ricardians
Frank Hahn*

I
I want to consider the claim that neoclassical economics is logically faulty, which has been
advanced by the followers of Sraffa.1 I shall first have to give an exposition of SrafFa's
own construction and I shall then discuss what his followers have made of it. I shall then
wish to show that there is no correct neo-Ricardian proposition which is not contained in
the set of propositions which can be generated by orthodoxy. I shall therefore conclude
that the neo-Ricardian attack via logic is easily beaten off.
Before I start, it is important to emphasise the distinction between Sraffa and his
followers. Sraflfa's book contains no formal propositions which I consider to be wrong
although here and there it contains remarks which I think to be false. The book was called
Production of Commodities by Means of Commodities and had the important subtitle Prelude
to a Critique of Economic Theory. With one exception, Sraffa does not claim to have got
beyond the prelude. The exception arises from his view that marginal productivity theory
requires a well-behaved aggregate measure of capital. Since he shows such a measure in
general not to be available (Sraffa, 1960, p. 38) he concludes that the marginal productivity
theory is logically false. It will be rather easy for orthodoxy to stand up to this criticism.
But his followers have gone very much further. They claim that Sraffa's work shows that
orthodoxy cannot logically provide a closed model which treats relative prices and the rate
of profit as endogenous variables determined within the system (see, for example, Gareg-
nani, 1970; Pasinetti, 1966 and 1969). They claim that Sraflfa has established the irrelevance
of the psychological theories of demand and indeed of demand to the determination of
relative prices. They assert that Sraffa shows that distribution has to be determined by
considerations outside the economic model, that it is 'logically prior' to the determination
of relative prices and above all that it is independent of the circumstances of production.
I have of course here lumped several followers together.2 But the preliminary point I wish
to make is simply this: these claims are not found in Sraffa's book.

•University of Cambridge. In 1975, while Taussig Professor at Harvard, I delivered six lectures entitled 'Neo-
classical Economics and its Critics'. For a number of years I toyed with the idea of making these into a slim
volume, but eventually got bored and taken-up with more interesting issues. This paper is a survival from this
venture. I have had comments at various times from K. Arrow, R. E. Rowthorn and I. Steedman. The latter
persuaded me to drop a rather tiresome section in the Socratic spirit.
'All references to Sraffa are to his 1960 book.
Representative statements of some of these views can be found in the writings of Eatwell (1974, 1975, 1977),
Garegnani (1966, 1970), Robinson (1960-80) and other followers (see, for example, Dobb 1973, chs 6, 7, 9:
Harcourt, 1975A, 1975B; Roncaglia, 1975). For an earlier critique by the present author see Hahn (1975).

0309-166X/82/040353 + 22 803.00/0 © 1982 Academic Press Inc (London) Limited

354 F. Hahn

One funher matter merits consideration before we get down to business. I often refer
to neoclassical theory and I had better make clear what I do and do not mean by this
designation. For present purposes I shall call a theory neoclassical if (a) an economy is
fully described by the preferences and endowments of agents and by the production sets
of firms; (b) all agents treat prices parametrically (perfect competition); and (c) all agents
are rational and given prices will take that action (or set of actions) from amongst those
available to them which is best for them given their preferences. (Firms prefer more profit
to less.)
Now there are many writers whom we regard as neoclassical who have either made
mistakes of reasoning or based themselves on special assumptions which have themselves
nothing to do with neoclassical theory. For instance Levhari (1965) was simply wrong in
the reswitching debate. Equally it seems to me impossible (as a matter of intellectual
history) to maintain that the possibility of perfect capital (or labour) aggregation is a
neoclassical doctrine. For instance it was soon realised that Bohm-Bowerk's attempt at
capital aggregation was logically flawed. In any event I shall stick to my definition and will
not care at all about authors regarded as neoclassical, who have proposed conclusions which
cannot be deduced from the basic axioms.
Of course 'neo-Ricardians' and 'Sraffa-followers' pose the same sort of problem. I base
myself here on Sraffa's book and on over twenty years of argument in Cambridge. After
all they call themselves 'The Cambridge School'. But should it be the case that some of
those who regard themselves as carrying on 'the Sraffa revolution' repudiate the views
which I report then I shall be delighted.

n
I shall now give an account of one part of what Sraffa has to say. I will choose the simplest
case where there is no joint production and all goods enter directly or indirectly into the
production of every other (all goods are 'basics'). In particular, I shall concentrate on what
are essential features rather than technical detail (e.g. the possibilities of non-basics). To
make life as simple as possible we will consider a world of two goods and labour.
Suppose that in this world we can observe all inputs and all outputs. We find that good
j uses a,j units of good t and ao; units of labour per unit of output. We do not ask and
we are not encouraged to ask why dus is the case. If we observe gross output Xj of good
; then

laijXj, i = 1,2, ay>0, ij = 1,2

is the amount of good i used to produce goods and

I<*ojXj > Go; >0>> = 1,2

i
is the amount of labour used.
So far all of this is description and the hypothesis is that the economy described has
no joint production and no non-basics. Sraffa now proposes to process this description by
defining a composite good to be called the Standard commodity. This is done as follows.
Let us ask what outputs of the two goods we would have to observe if the ratio of the
gross output of any good to the amount of it used in production were to be the same for
 The neo-Ricardians 355
all goods. If the hypothetical gross outputs are distinguished by an asterisk we want to
solve

_ ! i _ = _ ! ? _ = G*say (1)

where we would like x*> 0 i = 1,2. One can alternatively say that we are looking for
(x*, xf) and G* such that

or in matrix notation
x* = G*Ax* (2)
where x* is the vector (x* x£) and A is the 2 x 2 matrix of elements a,-, (i,j — 1,2). We
now have a purely mathematical problem for which there is a standard mathematical result.1

Theorem 1. If all elements of A are positive and

1-Zfltf X ) , ; = 1,2

then there is (a) a positive vector (A*,A^ unique except for scale and (b) a unique real
number G* > 1 which solves (2).
Since we are taking all elements of A to be positive and since we shall be concerned
with the case of 'Production with a Surplus', 2 the conditions of the theorem are satisfied.
We may therefore write
G* - 1 + g* where g* > 0
and we may write
x*=ktf ,« = 1,2, * > 0
3
where we fix k by:
*2ao; tf — available labour = 1, say.
The composite good consisting of x* units of good one and x* units of good two is then
defined as one unit of standard commodity.
It is to be emphasised that SraflFa does not claim that the world yields the observation
x*. The latter vector is a pure construct as of course is (2) used in its derivation. The
system (2) is called the Standard system.
So far then, nothing has occurred other than the manipulation of data given by A and
(a01, a02). To get any further and in particular to get somewhere where these manipulations
are fruitful the following assumptions are made about the actual world:
(a) The price of every good as output is the same as its price as input.
(b) The rate of profit in the production of every good is the same.

'See e.g. Arrow and Hahn (1971, appendix 2).

That is, A satisfies the so-called Hawkins-Simon conditions.
'We here depart a little from Sraffa in an inessential way. He normalises on the labour used rather than the
labour available.
356 F. Hahn

I shall discuss these assumptions later but accept them now in order to continue with the
story.
Before writing the price equations, we note that Sraffa thinks of wages as paid at the
end of the production process so that no rate of profit is charged on the outlays on
labour.1 We also think of wages as payment over and above subsistence. If the latter is
thought of as a given basket of the two goods per unit of labour we may take subsistence
requirement as already embodied in the coefficients a,y. So now let Pj (/' = 1,2) be the price
of a unit of good j . The assumption that the profit rate, r, is the same in all lines and
assumption (A) can be formally stated as

Pj = RIaijPi+aojzv, ; = 1,2 (3)

where R — 1 + r and w is the wage of one unit of labour. To see that (3) is just a restate-
ment of the assumption notice the identity: 'Price per unit of output = cost of goods per
unit of output plus wages cost per unit of output plus profits per unit of output'.
Express profit per unit of output as a fraction of the cost of goods used per unit of
output. Add the assumption that this fraction is the same whichever good is produced and
you get (3). It will be convenient to define the two row vectors: P = (P,, P2), ao = (a01,
aO2), and to write (3) as
P = RPA + aow (4)
Suppose R is given. What would it mean to have a solution P,w to (4)? It would tell
us what the prices of outputs and the wage would have to be in order that the assumption
of a particular uniform profit rate be correct. The uniform profit rate 'determines' P and
w only in this sense. We also notice that if for given r (i.e. R) P°, zw° solve (4) then so
will kP°, kw° for any k > 0. So there can be no harm in restricting solutions to those values
of P and w for which
P, + P2 + w = 1 (5)
So (4) and (5) constitutes three equations, and if R is given, we also have three unknowns.
But we want a solution for which P,^0 i = 1,2 and w^O.
For this we have

Theorem 2. Under the hypothesis of Theorem 1, the system (4), (5) has a solution (unique
up to scalar multiplication) P , > 0 , i = 1,2 and w > 0 for all R such that
1<R < G * ( o r O < r < g* ) .
If we look at (4) and (5) again we notice that we have one more unknown than equations.
It is for this reason that we have prefaced the remarks about solutions with 'given R'. That
is, we have arbitrarily fixed one of the unknowns. But as far as the formal story goes we
could have arbitrarily fixed any one of the four unknowns and looked for a solution of
the remaining three.2 There is no sense in the sentence: '(4) and (5) still leave R to be
determined'. However, the neo-Ricardians like to think of R (or w) as the 'free' variable
and that is where we return to the Standard commodity.

'Throughout I use rate of profit to stand for the interest rate which producers use to discount future receipts.
2
For instance, the government could control the price of some good in terms of numeraire or in an open economy
this price could be given from abroad.
 The neo-Ricardians 357
The first thing to do is to replace the normalisation (5) by
rPAx* + w = \ (5')
This makes the length of the vector (P,w) depend on r. But that does not matter since
the length of that vector is of no significance. What is the interpretation of this normalis-
ation? It is that we have arbitrarily decided to consider only those prices and wage rates
which make the values added in the production of one unit of standard commodity equal
to one. I re-emphasise that there is absolutely nothing wrong or limiting in this since the
length of price vectors cannot matter to anything. I call the value added [rPAx* + w],
just defined, the Standard net product.l
Now with this new normalisation let P(r), w{r) solve (4), (5') for given r. From (2)
P{r)x* = Gk P(r)Ax* = P(r)Ax*+g*P(r)Ax*
or
P(r)x* - P(r)Ax*
= g* (6)
P{r)Ax*
The left-hand side of (6) SraflFa calls the Standard ratio. Since g* depends only on A, the
Standard ratio is independent of all market happenings such as P,w or r. Moreover, the
reader will quickly verify from (4) that the numerator of the Standard ratio is just another
way of writing the standard net product. Since the solution satisfies (5') it is equal to one.
Hence we can write (6) as:

P{r)Ax* = \ (6')
g*
Substitute (6') into (5') to obtain
r=g*(\-w). (7)
Equation (7) is simply the consequence of the normalisation (5'). Since net product = 1
we can interpret w as the wage in terms of net product or simply the share of one unit
of labour in Standard net product. Then (7) tells us that this share is inversely and linearly
related to whatever the rate of profit happens to be. It is an example of a factor price
frontier. It has caused a good deal of excitement amongst some economists.
Before I consider this, let us make clear what has been achieved by the Standard
commodity. To see this, let us return to the system (4), (5) and for any given r let P(r),
w(r) be the solution. Then from (4)
P(r) = a0 [I - RA]-><b{r) (8)

By Theorem 2 for r < g* the inverse matrix has only positive elements. Let e = (1,1) a
column vector. Then since the solution satisfies (5):
P(r)e + ib{r) = 1
or using (8)

[ao[, -RA]e* + l] ib{r) = 1 W

'Professor Steedman has pointed out to me that Sraffa uses the term Standard national income.
358 F. Hahn
This is the factor price frontier we get from the normalisation (5). Since it is easy to vesify
that the square bracket term is increasing in r we also have that #(r) and r are inversely
related. But (9) is not linear. The merit of using the Standard commodity is then to enable
us to express our solution in normalised form which yields a linear factor price frontier.
Although, as we shall see, the Standard commodity, if it describes an actual and not a
hypothetical basket of goods, has further interest, this is not the case in Sraffa's work.
Its task is to rearrange data so as to yield linearity and that exhausts its merit.
Readers of Sraffa may be surprised at this since they know that he is much concerned
with an 'invariant standard of value'. This has something to do with Ricardian theory but
to a modern theorist it is almost incomprehensible. The normalisation (5) also yields an
invariant standard of value as does any such normalisation. In (5) our 'standard' is the value
obtained from one unit of each good plus one unit of labour. Here w is then the rate at
which one unit of labour exchanges against that basket. A numeraire is a numeraire. The
price of the numeraire can be set equal to one. Sraffa has chosen Standard net product
as numeraire and there's an end to it.
Lastly to complete this part, notice that for R < G*, (8) may be written
P{r) = ao[I + RA + R>A* + . . . . ] * ( r ) (8')
This is the well known reduction of prices to dated labour. Sraffa correctly notes that there
is nothing in the mathematics which tells us how P,(r) will be affected relatively to P 2 (r)
by stipulating a different value of r. But

, 2j
is the value (in numeraire) of the means of production (capital) used in the production
of one unit of Standard commodity. This value depends, in general in a complicated way,
on r. Therefore it cannot be used to determine r without circular reasoning. A generation
of Cambridge students have been taught that this argument is logically fatal to neoclassical
economics. This matter will be discussed later.

m
Sraffa writes:

It is, however, a peculiar feature of the set of propositions now published that, although they do
not enter into any discussion of the marginal theory of value and distribution, they have nevertheless
been designed to serve as the basis for a critique of that theory. If the foundation holds, the critique
may be attempted later (Sraffa, I960, p. vi).

I shall now give a preliminary consideration to that claim. It must be preliminary since
the summary so far has not discussed Sraffa's chapter XII: 'Switch in Methods of
Production'. On the other hand, the summary will do for the rest which elaborates the
story to decomposable matrices (basics and non-basics) and constructs the Standard
commodity for a von Neumann model with fixed coefficients. These further complications
need not concern us since Sraffa's 'foundations' are all there in the simplest case.
Imagine that Mr Sraffa and I, a representative of 'the marginal theory of value and
distribution', arrive on an island. Mr Sraffa suggests that we should look at the very
elaborate statistics of inputs and outputs for a given year which the islanders have collected
and that we should each construct the input-output matrix (A,a0). Our arithmetical skills
 The neo-Ricardians 359
are the same and we produce the same answer. Next, Mr Sraffa suggests that we find what
the composition of output would have had to be in order that the ratio of output of every
good to the amount of it used in production be the same for all goods. No sooner asked
than the computer churns out the answer. Mr Sraffa now performs some more calculations
and announces: 'if in this island the rate of profit in every line is 596 then these are the
relative prices that would rule'. I go into my corner, calculate, and come out with the same
answer. Next we calculate the wage in Standard net value and then we calculate the linear
factor price frontier. Every answer, since we make no mistake, is necessarily true. All we
are doing is to provide descriptions of the island and descriptions of hypothetical states.
I conclude that the only falsifiable entailment of the Sraffa equations is the postulate
of a uniform rate of profit. His followers often interpret this as the state to which the system
tends and they think of Sraffa prices as 'normal' or long-run prices. As will become clear
in the sequel this entailment of Sraffa's theory is also an entailment of the neoclassical
theory of balanced growth. Therefore the two theories cannot be distinguished by that
particular entailment.
We have already seen that Sraffa does not offer a theory of the rate of profit which he
takes as given (but see Sraffa, 1960, p. 33). His followers have seen great merit in this.
For they argue (a) that technology cannot determine the rate of profit since many different
rates are consistent with Sraffian prices, and (b) that Sraffa's system shows quite clearly
that, in the factor price frontier, the working of class struggle for the profit rate and the
real wage are inversely related.
These arguments are false on several grounds. Firstly, if on our island the input output
table had been (B,bo) rather than (A,ao) then a given r would in general have yielded a
different real wage. Thus, it cannot be true that technology has no role in determining
equilibrium shares. Indeed, the Standard commodity for (B,b0) is not the same as for (A,a0)
and if we used one of these to compare shares we would no longer have the linearity of
the factor price relations for each of the technologies.
So what Sraffa followers can reasonably be taken to mean is this: if we know technology
and nothing else we do not have enough information to say what price and what rate of
profit will prevail. We are one equation short. As I argue below exactly the same is true
of neo-classical theory so that on this score we cannot decide between them.
The same applies to the factor price frontier. This frontier can be derived from neoclassi-
cal theory with a choice of techniques. If the inverse relationship between the rate of profit
and the real wage suggests conflict to Sraffa followers then such conflict will also be present
in neoclassical theory. This will become clearer in the sequel.
Exactly the same is true of the fact that Sraffa prices can be found once the rate of
profit is known without any appeal to the preferences of households between goods. This
of course is also true for a special neoclassical model which is discussed below. This special
model ensures constant returns to scale which Sraffa claims not to posit. I have been at
a loss to understand him here. For the claim reduces everything we have discussed so far
to just a fancy way of presenting accounts ex post. If there is enough time for equal rates
of profit to be established then there is also enough time for producers to decide which
technique to use.
Let me break off here to give the neoclassical theory which justifies these claims.

IV
When the neoclassical economist arrives on the island, he is told all the alternative ways
360 F. Hahn

known to islanders in which each good can be produced by means of goods and labour.
This information is summed up by
1
= / / ( < V av> fl
o;)> J = 1»2-
So there are constant returns to scale and no joint production. Each function/ (.) shows
all the combination of inputs capable of producing one unit of output good;. Besides (10)
the neoclassical visitor is given the following information (i) relative input and relative
output prices are the same and the common rate of profit is r and (ii) both goods are
produced. He is told nothing about tastes. On the basis of this information and of a theory
of action by economic agents, the visitor now undertakes to calculate (a) the actual input
output table, and (b) all relative prices.
The well known theory to be used is that island producers of good j maximise their pure
profits.1 Since there are constant returns to scale this implies that pure profits per unit
of output are maximised. The fundamental hypothesis is that each agent treats (R,P,w)
parametrically.
Let ay = (aij,a2j,ao;') and Aj the set of all a, which satisfy (10).2 Then the theory predicts
the choice a;(/?,P,a>) at the market signals (R,P,w) where
aj(R,P,w) solves: max [P; — KL a{j P, — aojw]
A; '
' (11)
Define: Kj(R,P,w) = max [P, -RIa{j P, - a^w]
4
One calls 7ty(.) the unit profit function. It gives the maximum pure profit per unit of output
given the technological possibilities Aj and the market signals (R,P,w). If 7t,(.) < 0,
producer of good ; will not produce, so 7i;(.) ^ 0. If 7t;(.) > 0, the rate of profit is not
r. Hence
nj(R,P,w)=0, ; = 1,2. (12)
But as one easily verifies
knj (R,P,w) = 7ij(R,kP,kw), any k > 0,
so once again the length of the vector does not matter. So we arbitrarily add

P, + P 2 + w = 1. (13)
The system (12), (13) gives three equations in three unknowns (P,w) since R is fixed.
This is not quite satisfactory yet because one would like the solution to be non-negative
and unique. This is a purely technical matter. If there are no non-basics so that every pair
(a,,a 2 ) with a, e Ax and a7 t A2 forms an indecomposable input output system one has
available a theorem very much like Theorem 2. Only now G* is defined as the largest of
the numbers (1 + Standard ratio) which we can find by going through all the input output
systems which can be constructed from Ax and A2 One wants G* > 1. But under
assumptions no more stringent than Sraffa's, we can solve (12), (13) satisfactorily and
uniquely when R is in the appropriate range.

• By pure profits I understand the difference between receipts and costs where the latter include the opportunity
cost of investment exemplified by r times the value of investment.
'Aj is taken to be convex.
 The neo-Ricardians 361
It will be clear that we have solved for our unknowns {P, W) and at(R,P, W), a2(R,P, W)
jointly. It will also be clear that (11) and (12) could be written differently. Let/)•(.) be
differentiable (i.e. a continuum of techniques) and write

&=—, i=l,2,0.

Then^, depends on a, and we could have written down necessary conditions for maximum
profit when all inputs are used

Pjfjia-) = RPh i = l,2, ; = 1,2, ^

Y (14)
w l 2 J
Wh> = > J= >-
Together with (13) this gives seven equations. Because of constant returns to scale1
l/j.iaj) a,y = 1, ; = 1,2, (15)

which adds two equations so we have nine equations for the nine unknowns (ax,avPtvJ).
I emphasise that (13), (14) and (15) are only a rewriting of what we had already.
But we have now reached the first of our conclusions. It will be agreed that you cannot
get more neoclassical than differentiable /J(.) and it will have been recognised that /-, is
the marginal product of i in the production of;'. In (14) and (15) every possible marginal
product has been used. It follows that if you ask the neoclassical economist: but what
'determines' r? he cannot find yet another marginal productivity to do it for him. He is
in exactly the same boat as the Sraffians, i.e. he needs one more equation which cannot
be derived from the production relations which he has given. Moreover, notice the
meaninglessness of a sentence like: 'the marginal product of labour determines the real
wage'. You need to solve all nine equations to find either labour's marginal product or
the real wage. This was patiently explained by Robertson (1931) and it is rather odd that
neo-Ricardians continue to assert that in neoclassical theory the rate of profit or anything
else is 'determined' by marginal productivity.
Before we go on the hunt for the missing equation, I shall get another stumbling block
out of the way and sum up so far.
If/)•(.) is not differentiable (a spectrum of techniques) we must distinguish between a
'right-hand' and a 'left-hand' marginal product. For instance, /^(a,) is the change in output
of; for a small increase in the labour employed and fjo(cij) the change in output of; for
a small decrease in the labour employed. The reader will check that when profits are
maximised

and analogously for other inputs. The inequalities (16) replace (15) and

(RP,w) aj = Pj, ; = 1,2

replace (14). Mathematical economists know how to solve this system too and in it also

'This [Equation (15)] is the famous adding-up equation. It says that under constant returns to scale when each
input receives its marginal product, input payments exhaust the total product.
362 F. Hahn
all 'marginal productivity relations' available have been used up. I know of no neoclassical
economist who insists on the equality of some price ratio to the first differential of a
non-differentiable function.
Sraffa can be interpreted—he does not do so—as dealing with an economy in which (16)
simply says that the wage must be non-negative and not infinitely large. For the neoclassical
economist that is the singular case of fixed coefficients, i.e. where Aj has only a single
member. But this is not what Sraffa asserts. Rather he takes the actual techniques of
production as unexplained. That being so, they may well be the outcome of the kind of
choice which we have described. Mr Sraffa landing on the island will never discover
whether this is so or not. For he does not ask and of course he can calculate Standard
ratios and commodities on the basis of technological observation which are what they are
because of a rational choice of technique. Moreover, for the inquisitive neoclassical
economist who wants to explain the choice of techniques as well as for the easily satisfied
Sraffa, it will be the case that equilibrium prices are independent of preferences.
This argument then shows that nothing, with one exception, that Sraffa says or does
as far as we have discussed it, can conflict with the deepest neoclassical theory. Indeed,
it is irrelevant to it. The exception is the linear factor price frontier. In neoclassical theory
different specified values of R will give rise to different input/output tables and so to
different Standard commodities. However, it is well known and easily verified that the
neoclassical factor price frontier, on any normalisation, is downward sloping so that we
can still have class conflict. But again, since Sraffa simply takes the input/output table
as given and unexplained, this can hardly be a point at issue. The neoclassical economist
if asked to draw the factor price frontier on Sraffa's normalisation for a given technique
will draw the same one. He will note that he would not expect the technique to be the
same at different R but until Chapter XII he can only expect a shrug of the shoulder in
reply. For until that chapter, we are not told whether there is or is not any choice.
Lastly, note the following obvious point. The neoclassical theory may be wrong. Agents
may not treat prices parametrically nor need they maximise profits. That is a virtue not
shared by an accounting system.

So far we have been in familiar waters. But now we set out in search of the 'missing
equation'. Sraffa himself never set out on this voyage and confined himself to the remark
that the equation cannot be one which demands the equality of the marginal product of
'capital' and the rate of profit. We have already seen that the neoclassical economist has
the same view but his reasons are not those given by Sraffa. This matter will be taken
up again later. At the moment I am only concerned to show how neoclassical theory deals
with the missing equation. It is the Sraffa followers who are most at risk here, although
Sraffa himself will be found to have been too reticent on certain crucial matters.
Sraffa writes: 'We retain however the supposition of an annual cycle of production with
an annual market' (page 3). So let us consider an economy which, at the beginning of 1976
has available to it W76 units of wheat and B76 units of barley. They are simply the
outcome of past production decisions. The economy has one unit of labour at the start
of 1976. The agents in the economy can consume some wheat and barley in 1976 and sow
some to produce wheat and barley for 1977. I assume that the world ends in 1977 not
because the theory requires such a disaster but because it makes for simple exposition.
 The neo-Ricardians 363
Now wheat and barley available in 1977 are not the same goods as wheat and barley
available in 1976. So in obvious notation, we have four prices seen in 1976
p76 p76 p77 p77
W B* W* B'
In addition, we write w as the wage of one unit of 1976 labour which gets paid in 1977.
As usual the length of the price vector is without significance and we normalise by:
/>£ + P7B6 + P 7 ; + f>y + w = i. en)
Producers of wheat and barley 1977 face the technological possibilities discussed in the
previous section which include fixed coefficients as a special case. But explicit attention
is now given to dates: inputs are dated 1976 and outputs are dated 1977. If we put
Q = (p£ p;6, p £ py, w)
a five-vector, then the unit profit function for wheat and barley is

*, (fi), « = ^77> B1V

The choice of technique, if there is one, will depend on Q. The equilibrium condition
for producers is:
P/ 7 = la{J P,76 + aojw, ij = W,B, (18)

where the a,ys are the most profitable input coefficients from amongst those available. This
equation does not look like a Sraffa equation and before we proceed this rather crucial
matter must be investigated.
Consider the two numbers R,• = 1 + r,, defined by

Then Rw, for instance, is the number of units of 1977 wheat which can be obtained for
one unit of 1976 wheat. Keynes (1936, Ch. 17, p. 223; but he obtained the idea from SraflFa,
1932, p. 50) called rw ( = RW — 1) the wheat own rate of return on wheat since a man
will not exchange 1976 wheat for 1977 wheat indirectly through storage or production
unless he does as well as the direct exchange. Arithmetic gives

(PyP^iPyPtyR^Rw (19)
where
(py/p 7 ;)-i (20)
is the proportional difference in the relative prices of wheat and barley at the two dates.
The relation (19) can be interpreted as follows. Suppose you have one unit of 1976 wheat.
You can exchange it indirectly for 1977 wheat as follows. Buy P$/P™ units of 1976 barley
and exchange it for Rg (P™/Py>) units of 1977 barley. This can be exchanged for (Pg/Pg)
/? B (P«/Pj 6 ) units of 1977 wheat. So (19) says that the indirect exchange of 1976 wheat
for 1977 wheat yields the same return as does the direct exchange. The left-hand side of
(19) minus one is called wheat own rate of return on barley.
It will now be clear that Sraffa is considering a very special state of the economy where
(20) = 0, that is, where the relative prices of 1976 wheat and barley are the same as those
364 F. Hahn

of 1977 wheat and barley. The neoclassical economist is quite happy with more general
situation.
Sraffa then postulates that on his island
RW=RB = R, say, (21)
so that the wheat own rate of return of wheat is equal to the barley own rate of return
of barley. From the definitions, we now can write (17) as
RPW + RPB + Pw + PB + w = 1 (17')
where the superscript '77' has been omitted since it is now otiose. One sees at once that
the decision to consider only states for which (21) holds reduces the number of unknowns
from (5) [Pg, Pg, P™, P™, w] to four [P^, PB, w, R}. This, as we shall see in a moment,
is of some importance. The reader can check that if (21) then (18) has the usual Sraffa
form.1
Before we consider the very special Sraffa island, let us look at the neoclassical_general
case. We arrive on the island and find that history has given it an endowment W7b, Blft.
We do not worry what history the island has had. On the island all agents treat prices
parametrically and all agents are greedy. We are asked to calculate that value of Q at which
agents who have done as well for themselves as they could take actions which are mutually
compatible, i.e. an equilibrium Q.
I now list the familiar steps:
(a) We only consider values of Q which satisfy (17).
(b) If Q° is an equilibrium where both goods are produced, then it must satisfy.
Ui ( 2 ° ) = 0, i = W77, B77. (22)
For if 7r,(Q°) < 0 g°°d i will not be produced and if ft,(2°) > 0> producers who treat prices
parametrically will wish to produce an unbounded amount. Since available inputs are finite
this could not be consistent with the actions of agents being compatible.
(c) Let X]6 (t = W,B,L) be the difference between the demand for 1976 wheat, barley
and labour and the amount of these three goods available. From elementary theory we know
that this difference depends on prices and on endowments, i.e.

X]6 = X]6 (2, W76, Bn), i = W,B,L.

Let 2° be the set of vectors Q for which (17) and (22) hold. We already know that any
candidate for equilibrium must belong to that set. But if Q° is a member of Q° then
whatever are the amounts of 1977 wheat and barley demand, producers are willing to
supply them, i.e. Q° is compatible with equilibrium on these markets. This of course is
a consequence of our postulate that there are constant returns to scale. Hence

XV ( & W76i B76) = 0 all 2 in 2 ° « = W,B, (23)

1
where X ^ for instance is the excess demand for 1977 wheat.
By accounting, no one can spend more than they get from sales. So one has in view
of (23)
P £ X™ + P7B6 X1' + W X1* = 0 for all Q° in Q° (24)

'Except that I here take wages to be paid in advance.

 The neo-Ricardians 365
If Q> is an equilibrium we need XJ 6 =0, i = W,B,L. In view of (24) it suffices to search
for Q° in Q° such that
Xi6(Q,W76,B76)=0, i=W,B. (25)
(d) Now (17), (22) and (25) give five equations in the five unknowns in Q. If we assume
that at all Q there is a positive demand for 1977 wheat and 1977 barley, then one can prove
under the usual neoclassical conditions on preferences that there exists Q° with strictly
positive components which satisfies all the equations. This solution need not be unique.
(e) When we have calculated an equilibrium Q° we know everything we want to know.
We know the technique of production which will be used, we know the composition
of output of 1977 goods and if we choose wheat as numeraire, we can calculate the
common wheat own rates of return of wheat and barley. Mr Sraffa can calculate Standard
commodities. However, our solution, and so the rate of return in terms of wheat, will not
in general be independent of the amount of 1976 wheat or barley history has provided.
The last observation serves as a useful transition to the special Sraffa case. We now
stipulate (17). As I have already noticed, this means that we lose one unknown and so
it looks as if we have too many (not too few) equations. The model is overdetermined. This
is exactly what commonsense tells us. For suppose that a single element of Q° (i.e. Q°)
solves the equations which we have just discussed for given W76, B76. Suppose Q° does
not satisfy (21). Then imposing (21) makes it impossible to find an equilibrium. So we
must turn the question upside down and ask: is there a set of histories (W76, B7b) such
that if Q° is calculated for any one of them (21) also holds? I want to emphasise that this
question must be faced whether you are a neo-Ricardian or not. For it cannot be part of
the doctrine that you are uninterested whether there is enough 1976 wheat and barley to
meet demand. Yet Mr Sraffa does not consider this matter.
So the neoclassical economist who is always happy to consider interesting special cases
sets to work to find a proper equilibrium for Mr Sraffa:
(a) If (21) then we now confine attention to (17'). Let
Q(R) = (RPW, RPB, / V , PB, W)
so that conditions (22) now become (12) of the previous section which we write as
0, i=W77,B77. (22')
For given R we have (17') and 22') i.e. three equations. If R is properly chosen they
suffice to give us the equilibrium values: (Pw (/?), PB(R), W(R)).
(b) From what has just been said we can write
X]6 = X}6 (R, W76, B76) i = W,B

whenever Q(R) satisfies (17 ') and (22'). For then, given R, all relative prices are known.
So now condition (25) becomes
Xj6 (R, W76, B76) = 0, i=W,B (25')
and as we expected, this is two equations in one unknown when the availability of 1976
wheat and barley are arbitrarily given by the past.
_ (c) In general, then, we must make one of Wlb and B76 into j n unknown. Say we fix
B76 and now search by means of (25J) for the two unknowns W6 and R. If a solution
can be found write it W76 (£ 76 ), R(B76) to indicate that it depends on Blf>. Notice that
366 F. Hahn

R(B76) will_have to fall into the range for which (17') and (22') give positive prices.
By varying B16 the neoclassical economist can generate all histories which are compatible
with Mr Sraffa's demands. If Mr Sraffa lands on an island whose history does not belong
to this set, he will be out of luck.
But the neo-Ricardian followers will not yet be satisfied. For the equilibrium R is still
not independent of endowments. So if we are to help them, we will have to be more special
still.
Let
Z = P^R) X^ + PB(R) X7B6
The reader will verify that Z is the difference (in value) between investment and savings
when we regard the expenditure of producers on 1976 wheat and barley as investment.
Next let all households be alike in preferences and let their total endowment of leisure
be equal to one. Then
V = P^R) W76 + PB(R) Blb + w(R)
is the total 'wealth of households. Text-book economics tells us that given Sraffa's (17 ')
and (22') we can write1
Z = Z(R,V) (26)
Now let us replace (25) by the two equations (26) = 0 and X^ = 0. Clearly if they are
satisfied, then X™ = 0 and X™ = 0 also.
Now to the specialisation. If householders have preferences such that the proportion
of their wealth spent on any good is independent of wealth (their preferences are homo-
thetic) then (26) may be written as
Z = g(R)V (26')
For, recall that there are constant returns to scale. But now (26) = 0 and V ± 0 implies
g(R) = 0 (27)
so that we have at last found one equation in one unknown. History does not enter into
the determination of equilibrium R. But history still matters since we must still satisfy
X™ — 0 and we have already 'determined' R°.
This new specialisation of assumptions has been fully incorporated into the neoclassical
framework. Notice, in particular, that at the R° which solves (27) all the deep neoclassical
marginal productivity relations of (14) hold. Not only do they hold but we need them since
we know that they are only a rewriting of the Sraffa price equations when there is a choice
of technique. This is hardly surprising. It was a neoclassical simplification (which Keynes
disliked) to propose that R was 'determined by savings and investment'.
But even now the followers are restless and unhappy. For the function g(.) depends on
technology since the producer's demand for input is not independent of technology. So
let us see whether the neoclassical economist can find an island where that is not true.
Begin by looking for islands in which the past is like the present. That is if R etc
are today's (1976) equilibrium solutions then they were also yesterday's equilibrium
solutions. Next, we distinguish between households that in 1976 are repaid the wheat and

'Excess demand depends on prices, R and on wealth. But prices are known once R is, i.e. they are determined
by R so that Z depends only on R or V, if producers are always in equilibrium.
 The neo-Ricardians 367

barley 1975 which they lent to producers and households who are paid the wages for their
1975 labour. The former group is called the capitalist group and the latter the workers.
Capitalists receive
R [PW(R) {awwWlb +aWB BJ + PB(R) (aBWW16 + aBB B")] (28)
of which they spend nothing on 1976 wheat or barley, i.e. they re-invest the lot. Workers
receive their wages and they spend all of them on 1976 wheat and barley. So Z may be
written as
Pw(R) (aww Win + aWB B77) + PB(R) (aBW Win + aBB B77) - (28) (29)
where IP77, Bin are the outputs of wheat and barley for 1977.
Now we have the next restriction on the kind of island which we are prepared to visit.
Not only should the past be like the present but the future is also restricted. In particular,
we shall only admit solutions to be equilibrium solutions if for a predetermined number
G we have

— ~ - ~G (30)
W B
Notice that (30) together with Z — 0, X7£ = 0 are now three equations in three
unknowns (R, P776, B76) so that we may be able to tie down a unique history which
the island must have had in order to have Sraffa's (21) and the stipulated future. Moreover,
in view of what has been said we now have
Z={G- R) K
where K is the term in square brackets in (28). Hence for Z = 0 and K > 0, one needs
G = R for an equilibrium. Hence in islands of the kind we have described, R in equilibrium
must have the value G whatever the technological possibilities are.
Now there is an interpretation of G connected with 'animal spirits' which would cause
difficulties to the neoclassical theory. But interpretation here is of no importance. What
has been shown is this. The fundamental neoclassical approach of rational choice and
market clearing prices is quite unaffected by the successive specialisations which we have
discussed. At no stage do the variations impinge on the marginal productivity conditions.
Moreover, the last variation which we have discussed does not lead to the conclusion:
the equilibrium R is independent of endowments or technology. On the contrary, the
dependence on endowments (history) is admitted and one has to search for just that
combination of endowments for which a consistent story can be told. This 'right' combi-
nation of course depends on technology. I find it hard to believe that there is anyone who
has followed this section who can claim that the last story is not a very special case of
the first story or who can continue to believe that there are unknowns for which the
neoclassical model cannot provide an equilibrium account.
Before leaving this part of the argument, let me briefly dispose of some wrong-headed
objections:
(a) I have admitted that the special Sraflfa island must have a special history. But then
is it not true that we cannot talk of the marginal product of some input? For if there
were a little more of one input, history would be different and equilibrium price would
be different and so there would be no neoclassical marginal product. This argument (see,
for example, Robinson, 1974) is false. In the neoclassical model producers treat prices
368 F. Hahn
parametrically and the consequences of their actions are calculated by them at prevailing
prices. If that is so, then for their actions to be equilibrium actions requires that there
are no other actions available which on the hypothesis of given prices are more profitable.
This calculation does not require an actual increase in factors but simply a hypothetical
one. Of course producers may be wrong when they conjecture that prices are independent
of their own actions. This would not invalidate the importance of marginal products in
their calculations and in our theory of their actions. If producers are not to be wrong in
their conjecture, they must be small relatively to the economy (see Hahn, 1978). In both
cases, producers calculate the marginal worth of inputs at given prices.
(b) Sraffa does not require a marginal productivity theory as far as Chapter XII. This
is correct but only because he treats technology as given. It should be noted that there
is a paradox here. For the equalisation of the profit rate, stipulated by Sraffa, can only
be sensible for a world of rational and greedy producers. But rational and greedy producers
will compare gains and losses and will not be satisfied with a given technique if a better
one is available.
(c) In the very special case we looked at the equilibrium R was independent of technol-
ogy. Is it not true, therefore, of this very special case at best that the distribution of income
has nothing to do with technology? Answer, no. For even here to calculate relative shares
we need to know prices and the composition of output and both of these, as we know,
depend on technology. Nor does the fact that in the special case R is known once G is
known remove the importance of the marginal productivity conditions. For now we must
find the technique which is compatible with the given R and in doing so we shall find
equilibrium relative prices.
(d) It is true that in the very special case the equilibrium R is known once G is. But
in less specialised stories, e.g. that leading to (27), that is not the case. There relative
prices, the composition of output and R must be found simultaneously. But in any case
as I have shown, there is no difficulty for neoclassical economies with the special case.
(e) Sraffa does not discuss how the composition of output is to be explained. Neoclassical
theory does. But in my account I have treated endowments as unknown. This is required
for the Sraffa price equations to make sense in a world where inputs precede output. But
neoclassical economics can study the economy for an arbitrary history; Sraffa cannot.
(f) The neoRicardians attach significance to the fact that for any arbitrary if in a given
range one can find Sraffa prices and so R cannot have anything to do with technology or
marginal productivity. This is perfectly correct for a fixed coefficient technology where
marginal products do not exist in any case. But with a variable technology, the given R
will affect the technique of production precisely in such a way as to lead to the marginal
productivity conditions (14) to be satisfied.
(g) let me now return to Equation (8 ') which I reproduce below

P(r) = ao[I + RA + R?A* + ...] &(R) (8')

from which we deduce that no price is simply related to R. Now let us quote Sraffa more
fully. 'The reversals in the direction of the movements of relative prices (when r changes)
in the face of unchanged methods of production cannot be reconciled with any notion of
capital as a measurable quantity independent of distribution and prices' hence there is no
'independent measure of the quantity of capital which could be used, without arguing in
a circle, for the determination of prices and of the shares in distribution' (Sraffa, 1960,
p. 38). Let me couple with this a quotation from Frank Knight (1921): 'If we speak of
 The neo-Ricardians 369
"factors" at all there will thus not be three but a quite indefinitely large number of them.'
Now, in a properly formulated neoclassical theory the vector (W1b, B7b) helps to
determine equilibrium prices and distribution. Suppose that instead of the vector we were
given a number K where1

K = K(W7b, Blb)

Then there may be all sorts of yalues_of Wlb, B76 which give the same K. But we already
know that dif|erent_values of (Wlb, B76) in general generate different equilibria. So even
if for each {Wlb, B76) there is a unique equilibrium, knowing K would not allow us
to 'determine' equilibrium. This argument, the reader will notice, is quite damaging enough
even when K, as here, is independent of R. Moreover, notice that the argument is entirely
neoclassical.
On the other hand, suppose that in fact there is a number C (a measure of capital)
defined by
C = C(R, Wlb, Blb)
For instance, C may be the value of the endowment calculated at the Sraffa prices for
a given R. If now we treat as given C and Wlb> B76 then we shall have one equation
too many, since the information is equivalent to being told R. But that of course is not
'arguing in a circle'.
Lastly, suppose that we are given C and nothing else. Then the investment savings
equation (26) can be written as
Z(R, C) = 0
which now, since C is fixed, is one equation in one unknown. The condition that the wheat
market also clears and the definition
C = P^R) Wlb + PB(R) Blb
then give us what the amount of 1976 wheat and barley would have to be in order to have
a consistent story. Once again 'arguing in a circle' is not the problem. The problem is the
sense to be made of C being given from outside.
All of this has nothing to do with Sraffa's rather special (8') or indeed with the circum-
stance that the relative prices are complicated functions of R. The point is much simpler.
In general, there does not exist a function from the vector of endowments to the scalars
such that knowledge of the scalar (and of preferences and of technology) is sufficient to
allow one to determine a neoclassical equilibrium. If you put it the other way round, it
is even more obvious. In general, the neoclassical equilibrium can be found given the vector
of endowment which may have, say, 108 components. It would be surprising if there were
a single number which gives the same information as the 108 dimensional vector. In fact,
sometimes and in very special cases, this surprising property holds. But neoclassical
economists have shown these special cases to be without interest.
So why the fuss? The answer is partly that Mr Sraffa and his followers are much
concerned with the history of thought and with economists long since deceased. After all,
there have been many attempts to find a scalar representation of the endowment vector,
e.g. the period of production. These earlier economists had no rigorously formulated model

•So K may be thought of as a measure of capital stock.

370 F. Hahn
to help them and they got into trouble. But neoclassical economics is alive and kicking
and has not got stuck in Bohm-Bowerkian arithmetic or even in Wicksellian puzzles.
Another part of the answer is to be found in the simple neoclassical model, in particular
the postwar growth models. Even if some authors regarded these as parables, text books
testify that they can be misused. Sraffa and his followers deserve the credit for making
such misuse less likely in the future. On the other hand, I doubt that they are correct
in their view that the simple (essentially one capital good) models are useless. It is perfectly
true, as we have seen, that one can always construct examples where the answer of the
simple model is false. But we use simple models (e.g. macroeconomics) to gain insights
of a certain kind. Simplification is never without cost and the cost is sometimes loss of
rigour. It remains to be shown that the cost is too high in this instance, i.e. that in actual
problem application the chance of large mistakes is great. I know no Sraffian who has shown
this. Lastly, the fuss is to be understood by a persistent misunderstanding by Sraffians
of what a 'marginal productivity theory of distribution' is. In particular, they have not been
content with enunciating paradoxes of aggregation but have proceeded to use essentially
neoclassical procedures to tilt at some fictitious neoclassical windmills.
The Sraffian picture of neoclassical theory is this. At any moment of time we can observe
something physical called the stock of capital (K) as well as the amount of labour (L).
There is a concave production function

Y = F(K, L)

where Y is output. In a neoclassical equilibrium all inputs are used and must be paid their
marginal products. The latter are known once (K, L) are known. Hence the rate of
profit of capital, the real wage and the distribution of income are all known once F( ),
K and L are known. The concavity of F further implies that the rate of return on capital
is non-increasing (generally decreasing) in K. This construction, to be called the parable,
Sraffians claim to be not logically watertight except in the single-good economy. In this
they are generally correct.
But first let us notice a difficulty with the parable which has nothing to do with aggrega-
tion. So consider a genuine single-good economy—say a wheat economy. Knowing the
endowment of wheat (here K) is not enough for the claimed results of the parable. For
we must know how much of it will be used in production and how much of it will be
consumed. Without a saving-investment equation, the equilibrium rate of profit cannot be
found. Hence, as already noted, even the most primitive neoclassical model claims to be
short of equations when only the endowment and the production functions are known.
There is no way that one could here claim that an independently given K determines the
equilibrium rate of return, etc.
This is a point on which Sraffians seem sometimes to be in a muddle so I make the point
again. Even when an aggregate like K logically exists, one would have to know how much
of it is used in production before being able to claim that one can determine the equilibrium
payments to inputs. If K is a substance which can only be used in production (and not
consumed) then K is not the same 'good' as is output Y and we gain a new unknown price,
namely that of K, which can now differ from the price of Y. Once again we need one
more equation. Not many people have read Solow's famous paper (1956) as far as its dual
formulation to see this. One concludes that there exists no neoclassical formulation whether
in parable or single-good form which claims to determine equilibrium distribution from
a knowledge of endowments and production conditions alone.
 The neo-Ricardians 371
If that is understood, then the question naturally arises as to what neoclassical propo-
sitions are at risk when it is granted that Professor Gorman's necessary conditions for
capital aggregation are extremely unlikely (Gorman, 1959, 1968). For instance, from what
we already know every neoclassical economist can agree that the 'equilibrium rate of profit'
could be determined outside the sphere of production and that whether he uses a parable
or not. For there is nothing in neoclassical theory which excludes the possibility that
workers consume and do not save which together with the postulate of steady state equilib-
rium is enough to give an example of such a theory. Indeed, at first sight, no neoclassical
proposition seems at risk.
But something called 'the marginal productivity theory of distribution' is said by neo-
Ricardians to be at risk. As far as I have been able to gather this is claimed to be so because
one cannot arrive at a measure of 'capital' which is uncontaminated by market happenings.
A fair amount has already been said in this paper on this matter and Professor Bliss (1975,
Chs 5, 8) has lucidly made the case again. So I can be brief and fairly dogmatic.
A crucial and beautiful theorem in neoclassical economics goes as follows.1 Consider
the set of all goods which an economy could feasibly consume. In the definition of goods,
proper attention is paid to the date of their availability. In the definition of feasibility,
proper attention is paid to technological knowhow and the initial endowment of goods. If
this set is a convex and bounded and closed then every feasible efficient consumption plan
can be 'supported' by a set of prices. A consumption plan is efficient when no other feasible
plan gives more of one good without giving less of another. A plan is 'supported' by prices
if at these prices profit-maximising producers treating prices parametrically would choose
productions which result in the goods of the given consumption plan. It is a mathematical
truth that if the feasible set has a differentiable boundary, then every efficient consumption
plan is supported by prices unique up to scalar multiplication. It is now a fact that when
we introduce the given preferences of agents and take them to be convex and stipulate
non-satiation then every Pareto-efficient consumption plan, i.e. a plan such that no other
feasible plan is higher in one agent's preferences without being lower in some agent's
preferences, is also an efficient consumption plan. Moreover, there are again prices such
that if they rule there would exist a distribution of wealth between agents which would
ensure that those prices were in fact the neoclassical equilibrium prices. Lastly, under the
given assumptions every competitive equilibrium is Pareto-efficient.
These results are theorems and they are not at risk. They are not based on any aggrega-
tion hypothesis. They contain all valid neoclassical sentences using the words marginal
productivity. Under the differentiability assumption the support prices are the gradient
vector of the feasible set at the given point. Elementary calculus will explain why it makes
no sense to argue that undertaking the (infinitesimal) changes which give us the gradient
vector will change the support prices.
Suppose, for the sake of exposition, that there is a single consumption good at each
date. Choose the consumption good at a date 0 as numeraire. Then the price of consumption
good date one in terms of numeraire is the (marginal) rate of transformation of consumption
date 0 into consumption date one at the given point. Subtracting unity gives us also the
rate of return in terms of numeraire. At the support prices profits are maximised at the
given production plans and the economy is in equilibrium. The rate of return in terms
of numeraire on any asset which is used or held must be the same. Then the same must
be true of any composite bundle of assets bought at the going prices. Hence, the defined

'In what follows I am considering a world with a long but finite future.
372 F. Hahn
rate of return (minus one) measures the marginal rate at which consumption at date 0
can be transformed into consumption at date one. Putting it this way does not add anything
to what we already know from the theorem. Professor Pasinetti (1969 pp. 525-526)
however, is wrong to think it a tautology, because (a) there would be no sense in the claim
that the rates of return in terms of numeraire are determined by marginal productivity
(marginal transformation rates). To know the latter, we must know where on the feasible
set we are; (b) the fundamental theorem is not in conflict with special classical saving
hypothesis; (c) strictly the latter is embodied in preferences and we need to know these
to get rates of return.
If the frontier is not diflferentiable, then it is not differentiable and support prices are
not unique after normalisation. But left and right hand gradient vectors have already been
discussed and the reader can check that the neoclassical can continue happily and safely.
There is thus no joy for the neo-Ricardians along this route. But what of reswitching?
Towards the end of his book Sraffa writes:

We have been assuming that in a system of single-product industries only one way of producing
each commodity is available with the result that changes in distribution can have no effect on the
method of production employed (Sraffa, 1960, p. 81).

What could be clearer than that? Sraffa says that he has been postulating a world of fixed
coefficients and he is about to drop the postulate. That is of course why I have repeatedly
referred to this chapter. It is the only one which can conceivably be relevant to a critique
of neoclassical economics.
Mr Sraffa proceeds in an impeccable neoclassical manner. That is, he studies an economy
where there is a choice of technique and where the choice is made by greedy (i.e. profit
maximising) agents. This suggests at the outset that switching paradoxes are unlikely
to invalidate any properly formulated neoclassical propositions which invoke marginal
productivity.
If at R the Sraffa prices for technique (A,a0) are the same as those for technique (B,b0)
and if all other techniques make it impossible to earn R times the value of non-labour
inputs, then that R and the associated Sraffa price vector in terms of labour is called a
switch point. That is, every agent is there indifferent between the two techniques. From
this definition one sees that

p = ao[I +RA + R?Ai + . . . . ]

p = bo[I + RB + R*B* + ....)

and subtracting one of these equations from the other we have the polynomial

0 = (ao - b0) + R{a0A - b0B)

which plainly can have a number of real roots R. So there may be many values of R at
which rational producers are indifferent between the two techniques. Suppose they are
R' and R" with R' < R". Then when R' < R", suppose technique (A,a0) is more
profitable than (B,b0). Then at R'=R" there is a switch to technique (B,b0) and for R
such that R' < R < R", (B,bo) is more profitable. But when R=R" the two techniques
are again equally profitable. Then plainly we cannot say that one technique is more capital
intensive than another and that capital intensity is inversely related to R. So the parable
can get into difficulties since it stipulates a concave production function in (K,L).
 The neo-Ricardians 373
It is worth noting that reswitching cannot arise in the case where the production frontier
is differentiable (see Garegnani, 1970, pp. 412-414). On the other hand, differentiability
does not ensure a well-behaved parable production function. In particular, it does not
ensure that we can order techniques by their capital intensity and that the latter is inversely
related to R.
What could be simpler or more neoclassical? What is at risk is a simplified neoclassical
comparative equilibrium analysis and a simplified neoclassical dynamics. Sraffa's point was
a fine technical insight into neoclassical economics but the Sraffians and possibly Sraffa
himself have not exploited it. Matters were further confused by a straightforward mistake
by an eminent neoclassical economist even though he quickly made amends (see Samuelson,
1962; Levhari, 1965; Levhari and Samuelson, 1966; Samuelson, 1966).
There is no doubt that in neoclassical economics as in macroeconomics simple models
are used in order to obtain definite answers and that these simple models will not survive
logical scrutiny. Whether they can be useful nonetheless is a hard question which I cannot
answer. But just as the multiplier collapses when fixed prices are not assumed so does the
neoclassical parable claim that of two economies in Sraffian equilibrium, the one with the
higher rate of interest (profit) will have the lower 'capital' labour ratio. Nor can one argue
that the economy with the higher interest rate will have lower consumption per head. All
of this is simply a reiteration of the proposition that there is no valid aggregation of wheat
and barley into something called capital. But unless one wishes to claim that aggregation
is essential if a theory is to be called neoclassical, so that Arrow-Debreu for instance are
not neoclassical, none of this has any bearing on the main issue of this lecture. Sraffa per-
formed a service in showing how neoclassical arguments can be used to show neoclassical
aggregation parables to be in logical difficulties. But that cannot help with a critique of
marginal theory.
I have said that neither Sraffa nor his followers have made anything of reswitching. By
this I mean that they have continued to believe that it is damaging to neoclassical equili-
brium theory which it is not and have neglected various neoclassical adjustment theories
which are certainly at risk. Certainly the famous Solow parable (Solow, 1956) in which
all equilibrium paths seek the steady state depends on just those possibilities of aggregation
which reswitching examples show not to be available. Indeed, the circumstance that
'capital' consists of a number of heterogeneous objects and that only in very special cases
can their relative prices to be taken as constant cause considerable difficulties to the
'invisible hand'. Professor Robinson was right in arguing that capital aggregation in the
parable where there is only the asset (money is not modelled) has had the consequence
that no agent needs to have expectations concerning the future, especially future relative
prices, when he takes decisions today. This is a very limiting consequence of the simplified
model and suggests that something essential has been missed out. But it was left to neo-
classical economists to attempt to study the precise pathology of the price mechanism which
may result when heterogeneous inputs are modelled explicitly.
The relevant conclusion is straightforward. Reswitching and the general impossibility
of capital aggregation have no bearing on anything which can be called marginal pro-
ductivity theory. Such a theory concerns an economy in full neoclassical equilibrium which,
I have repeatedly argued, has nothing to fear from anything in Sraffa's or in his followers'
work. But on the manner in which such an equilibrium is supposed to come about, neo-
classical theory is highly unsatisfactory. Sraffa's work shows that certain simplified routes
are very risky and not free from logical difficulties. The remarkable fact is that neither
he nor the Sraffians have made anything of this.
374 F. Hahn
Bibliography
Arrow, K. J. and Hahn, F. H. 1971. General Competitive Analysis, San Francisco, Holden-Day;
Edinburgh, Oliver and Boyd
Bliss, C. J. 1975. Capital Theory and the Distribution of Income, Amsterdam, Oxford, North-Holland;
New York, American Elsevier
Dobb, M. H. 1973. Theories of Value and Distribution since Adam Smith: Ideology and Economic
Theory, Cambridge, Cambridge University Press
Eatwell, J. L. 1974. Controversies in the theory of surplus value: old and new, Science and Society,
Autumn
Eatwell, J. L. 1975. Mr Sraffa's standard commodity and the rate of exploitation, Quarterly Journal
of Economics, November
Eatwell, J. L. 1977. The irrelevance of returns to scale in Sraffa's analysis, Journal of Economic
Literature, March
Garegnani, P. 1960. II Capitale Nelle Teorie della Distribuzione, Milan, Guiffre
Gargenani, P. 1966. Switching of techniques, Quarterly Journal of Economics, November.
Garegnani, P. 1970. Heterogeneous capital, the production function and the theory of distribution,
Review of Economic Studies, October
Gorman, W. M. 1959. Separable utility and aggregation, Econometrica, July
Gorman, W. M. 1968. Measuring the quantities of fixed factors, in Wolfe, J. N. (ed.), Value, Capital
and Growth, Edinburgh, Edinburgh University Press
Hahn, F. H. 1975. Revival of political economy: the wrong issues and the wrong argument, Economic
Record, September
Hahn, F. H. 1978. On non-Walrasian equilibria, Review of Economic Studies, January
Harcourt, G. C. 1975A. Decline and rise: The revival of (classical) political economy, Economic
Record, September
Harcourt, G. C. 1975B. Revival of political economy: a further comment, Economic Record,
September
Keynes, J. M. 1936. The General Theory of Employment, Interest and Money, London, Macmillan
Knight, F. H. 1921. Risk, Uncertainty and Profit, Boston and New York, Houghton Mifflin
Levhari, D. 1965. A nonsubstitution theorem and switching of techniques, Quarterly Journal of
Economics, February
Levhari, D and Samuelson, P. A. 1966. The nonswitching theorem is false, Quarterly Journal of
Economics, November
Pasinetti, L. L. 1966. Changes in the rate of profit and switches of techniques, Quarterly Journal
of Economics, November
Pasinetti, L. L. 1969. Switches of technique and the 'rate of return' in capital theory, Economic
Journal, September
Robertson, D. H. 1931. Wage-grumbles, in Economic Fragments, London, P. H. King
Robinson, J. V. 1960-80. Collected Economic Papers, Vols II-V, Oxford, Basil Blackwell
Robinson, J. V. 1974. History versus Equilibrium, London, Thames Polytechnic
Roncaglia, A. 1978. Sraffa and the Theory of Prices, Chichester, New York, Brisbane, Toronto, Wiley
Samuelson, P. A. 1962. Parable and realism in capital theory: the surrogate production function,
Review and Economic Studies, June
Samuelson, P. A. 1966. A summing up, Quarterly Journal of Economics, November
Solow, R. M. 1956. A contribution to the theory of economic growth, Quarterly Journal of Economics,
February
Sraffa, P. 1932. Dr. Hayek on money and capital, Economic Journal, March
Sraffa, P. 1960. Production of Commodities by means of Commodities: Prelude to a Critique of Economic
Theory, Cambridge, CUP