Laws of production and laws of algebra:

Humbug II
Chapter 5 of: Edward J. Nell (ed.), Growth, profits, and property.
Essays in the Revival of Political Economy,
Cambridge University Press, 1980
Shaikh

The theoretical basis modity other things being equal -the lower its
relative price.
Recent debates on capital theory have focused The next step in the analysis requires its
on the notion of capital as a factor of production, to the case of production. Initial endow-
which along with labor, to ments are now assumed to contain not just con-
the in sumer goods but also means of production, such
Though the intricate point and counterpoint of as land, machines, raw materials, etc.; in addi-
the controversy often obscure this simple fact, it tion, since the game cannot continue unless
has become increasingly clear that what is at every individual has at least some wealth, it is
stake in the current debate is in essence the generally assumed that each and every initial en-
same issue with which the classical economists, dowment includes potentially saleable labor ser-
particularly Ricardo, grappled that of the divi- vices. By assumption, the ultimate objective of
sion of income between wages and profits. The every individual is consumption: means of pro-
argument thus rages around eco- duction and labor services, however, are not
nomic theory, whose aim it is to represent the directly consumable. At this point, therefore,
workings of a competitive capitalist economy. In production is introduced as a roundabout way of
a sense this is a return to relevance, since much consumption, a process in which inputs are
of modern mathematical economics has stu- transformed into outputs. In order to translate
diously concerned itself, not with descriptive. any given initial endowment into the production
but instead with normative theory, such as the possibilities inherent in it. neoclassical econom-
study of optimal and efficient growth paths, etc.. ics commonly relies on the assumption of a well
(Lancaster, 1968, pp. 9-10). behaved neoclassical production function, one
In neoclassical theory, the model of pure ex- for each commodity produced.
change occupies a central position, for it illus- Each individual then faces three basic
trates simply and elegantly the fundamental methods of arriving at some preferred final allo-
truths of the paradigm, truths which any more cation, methods which he or she is free to use in
complex representations may modify but cer- any combination permitted by the initial endow-
tainly cannot undermine. Thus, in the model of ment and consistent with the utility function.
exchange. trading begins with selfish indi- First, he can trade any of the
viduals each having an arbitrarily determined means of production in his possession for other
initial endowment of goods, and proceeds to a goods he desires; second, he may rent out the
final state in which no one individual can im- n f
prove his or her basket of commodities without and/or rent out his labor power: and third, if his
making someone else worse off. Such a situation
is known as a pareto-optimal allocation, and it
implies a set of final exchange ratios between
This chapter is an expanded, revised version of a
commodities that is, a set of
paper entitled Laws of Production and Laws of
tive prices. What is more, given The Humbug which
of well-behaved neoclassical utility functions for appeared as a note in of
each individual, the equilibrium prices of the Vol. LVI, No. 1 (February, pp.
model of exchange will along with comment by Robert The
the higher the relative availability of some postscript to this chapter assesses comment.
 81

initial endowment so permits, he may choose to tures most problematical gifts the natural
become a producer, renting and/or buying selfishness of every individual!
means of production and labor-power and com-
bining these with the elements of his initial en-
dowment to turn out one or more commodities Scarcity pricing parables and the aggregate
via a well-behaved neoclassical production func- production function
tion. Ruled only by his enlightened self-interest,
which dictates that more is better, and con- Traditionally, several models have been used to
strained only by his native abilities and initial en- extend scarcity pricing to the theory of distribu-
dowment, he is assumed to eventually arrive at tion. The simplest, and by far the most widely
some most efficient combination of the used in both the theoretical and empirical litera-
trader-rentier-producer modes, t h e r e b y a t - ture, is the aggregate production function model.
taining his personal optimum in the form of some we is an aggregated ver-
final allocation. sion of the general equilibrium model outlined
Because preferences (utility functions) and above, constructed as an empirically useful
initial endowments are of the analy- approximation, supported the
sis, the whole structure of equilibrium is ruled Even the sophisticates, the so-called
by them, so that once again, the forces of con- high-brows of neoclassical theory, at one time,
sumer sovereignty lead us ineluctably to took this and similar parables seriously:
Pareto-optimality . Equilibrium relative prices . . . In various places I have subjected to de-
are once again prices, a term which now tailed analysis certain simplified models in-
covers the prices of consumption goods, the volving only a few factors of production . . .
wage rate for labor services, and the rental [These] simple models or parables do. I think.
sale prices of means of production (Hershleifer, have considerable heuristic value in giving in-
1970). sights into the fundamentals of interest theory
Under carefully fashioned assumptions in- in all of its complexities. (Samuelson, 1962,
volving well-behaved utility and production
functions, these sorts of models are determinate The originators of the production func-
in the sense that one or more possible equilibria tion theory of distribution (in the static
can be shown to exist. But the model, as out- sense, where I still think it should be taken
lined here, contains no reference to the uniform fairly seriously) were Wicksteed, Edgeworth,
rate of profit which is supposed to characterize and Pigou. (Hicks, 1965, p. 293, footnote 1)
competitive capitalism. The explanation of this Though aggregate or surrogate production
rate of profit is what (descriptive) neoclassical function models occupy the bulk of the theoreti-
capital theory is all about. Moreover, given that cal and empirical literature on the distribution of
the basic parables of the theory have already income in a capitalist society, the essential char-
identified the equilibrium price of every good or acteristic of this and all other parables of neo-
service as a scarcity price, one that reflects its classical theory concerns their attempt to ex-
individual and social scarcity, the task that con- plain the wage rate and the rate of profit as scar-
fronts the theory is clear: somehow, the rate of city prices of labor and capital, respectively,
profit too must be explained as the scarcity price determined in the final analysis by efficiency
of some thing with both the price and quantity of considerations. It was precisely this techno-
this thing to be mutually determined in some cratic apologia for capitalism which became the
market. This market, it turns out, is the capital target of the neo-Keynesian counterattack of
market, in which demand is determined by indi- the during the so-called Cambridge cap-
viduals preferences for present versus future ital controversies.
consumption their taste for investment One of the most striking, and for neoclassical
(Dewey, 1965) and supply is determined by the economics most devastating, results of the
technological structure. The price that suppos- above capital controversies was the proof that
edly emerges from this interaction is the of version of the neoclassical parable, in which
interest, the scarcity index of the quantity of the rate of profit varied inversely with the quan-
and with the addition of a few more con- tity of capital and the wage rate inversely with
venient assumptions, the rate of profit is made the quantity of labor (so that each at least be-
equal to this rate of interest. haved like a scarcity price) was valid in static
cun then, it is argued, the distri- conditions prices in possible
bution in is competitive equilibria were proportional to labor
sequence the values. I apply,
in fact, within this wondrous construct, capital- to that particular version of the parable
ism itself represents the resolution of one of known as the aggregate (or surrogate)
 82 Shaikh

tion function, in which the wage rate and the rate former can be most generally described as fitting
of profit not only move inversely to the quan- a function of the Q(t) = to
tities of labor and capital, respectively, but are observed data while the latter consists of
also equal to and determined by their respective that aggregate marginal products of capital
marginal products. Considering that the neoclas- and labor are equal to their respective unit
sical in a
counterrevolution the classical school, ify structural coefficients. In general, for both
against Ricardo and Marx in particular (Dobb, time series and cross-section data, the
1970, p. and above all, against the labor Douglas function wins out: the sum of coeffi-
theory of value in form, it is gratifying to dis- cients usually approximate closely to unity
cover in the end these parables themselves (thus implying constant returns to scale), with
depend on the simple labor theory of value. The the additional bonus of a close agreement
irony is inescapable. between the labor exponent and the share of
These and other inimical results were not lost wages in the value of output (thus support-
on the faithful. As awareness of the internal in- ing aggregate marginal productivity theory)
consistencies of neoclassical theory began to (Walters, 1963, p. 27).
grow, many were led to abandon it. But for In a recent paper, Franklin Fisher concedes
others, hope died hard; and hope, it seems, lay that the requirements under which the produc-
in the data. As a neoclassical theorist, can tion possibilities of a technically diverse
only that the relevant question is what is economy can be represented by an aggregate
relevant: should we make our predictions on the production function are far too stringent to be
basis of what Mrs. Robinson has called perverse believable (Fisher, 1971, p. 306). He proposes
technical behavior the therefore to investigate the puzzling uniformity
that been repeatedly of the empirical results by means of a simulation
1971, p. 254, emphasis added) experiment: each of industries in this simu-
What has been repeatedly observed, it is lated economy is assumed to be characterized
argued, is the empirical efficacy of aggregate by a microeconomic Cobb-Douglas production
production functions. spite of the very function relating its homogeneous output to its
strongest theoretical requirements for their homogeneous labor input and its own
existence, the use of such functions flourishes machine stock. The conditions for theoretical
the current justification being that their empiri- aggregation are studiously violated, and the
cal basis appears strong. In study after study, question is, how well, and under what circum-
empirically derived functions appear to strongly stances, does an aggregate Cobb-Douglas func-
support both the constancy of returns to scale tion represent the data generated? In such an
and the equality of marginal products with economy, the aggregate wage share is often vari-
rewards: in for hnth able over time, so that in general an aggregate
series and cross-section studies (within any one Cobb-Douglas would not be expected to give a
country), the Cobb-Douglas function appears to good fit. What seems to surprise Fisher, how-
dominate the field, ever, is that when the wage share happens coin-
For the neoclassical faithful, these results rep- cidentally to be roughly constant, a
resent their salvation; no matter what those crit- Douglas production function will not only fit the
ics from Cambridge say, the real world, it data well but provide a good explanation of
would is Or is it? The wages,
answer is simple: no. m aggregate Cobb
strength production is illusion, that the the
due not to some mystical laws of production, but constancy is presence
instead, to some rather prosaic laws of algebra. of production
To see why, however, we must first examine is runs the
how production functions are estimated. the success
is due to
the constancy shut-e.
(Emphasis added.) (Fisher, 1971, p. 306).
The empirical basis of aggregate production It is obvious that so long as aggregate shares
functions are roughly constant, the appropriate economet-
ric test of aggregate neoclassical production and
The most popular methods of estimating ag- distribution theory requires a Cobb-Douglas
gregate production functions have been the function. Such a test would then apparently cast
single equation least squares method and the some light on the degree of returns to scale
factor shares method (Walters, 1963). The (through the sum of the coefficients), and the
applicability of aggregate marginal productivity Dividing through by
theory (through the comparison of the labor and
capital exponents with the wage and profit
shares, respectively). What is not obvious, how-
ever, is that so long as aggregate shares are con- By definition, the profit and wage shares,
stant, an aggregate Cobb-Douglas function respectively, are
having apparently constant returns to scale
will always provide an exact fit, for any data
whatsoever.
so WC may write,
to
= + where (1 + (3)
B
These
propositions, it will be shown, are It is important to note that all relations given
consequences of constant shares, and it will be so far are true for aggregate data at
argued that the puzzling uniformity of the empir- all, irrespective of production or distribution
ical results is due in fact to this law of algebra conditions.
and not to some mysterious law of production. Suppose now we are faced with particular
In fact, in order to emphasize the independence data which for some unspecified reasons exhibit
of any laws of an constant shares, so that Re-
illustration is provided in the form of the rather membering that the dotted variables are time
implausible data of the Humbug economy, for derivatives etc.), we can immedi-
even data such as this is perfectly consistent ately integrate the identity (3):
with a Cobb-Douglas function having constant
returns to scale, neutral technical change,
and satisfying marginal productivity rules, so
long as shares are constant. dt In k +

where for convenience the constant of integra-

tion is as In Rewriting, we have,

Laws of algebra = =

Let us begin by separating the aggregate data in where by definition

any time period into output data the value of
output), distribution data (W, wages and prof- B = exp
its, respectively), and input data (K, L, the index
numbers for capital and labor, respectively). Equation (4) is strikingly reminiscent of a con-
Then we can write the following aggregate iden- stant returns to scale aggregate Cobb-Douglas
tity for production function with shift
But in fact, it is not a function at all,
+ but merely an algebraic relationship which
Given index numbers c a n always holds for output-input data L,
always write: even data which could not conceivably come
q(t) w(t) +
from any economy, so long as the distribution
data exhibits a constant ratio. Furthermore,
and are the and since the term in identity is a weighted
capital -labor ratios, respectively, and w(t) average of the rates of and
are the wage and respectively, it seems empirically reasonable to
profit The is expect of L would give
therefore the fundamental identity relating out- capital-labor ratio which is weakly correlated
put, distribution, and input data. Defining the with With measures for which the above is
share of profits in output as and the share of
wages as 1 s, we can differentiate identity 2 to so that B will also be solely a
arrive at identity 3 (time derivatives are denoted function of time. Then we can write
by dots, and the time index, t, is dropped to sim-
= (5)
plify notation):
and since and we get
Q = B(t)
 The algebraic relationship just given has vintages of machines, etc., but the
propcrtics. First, it is homoge- has remained unchanged.
neous to the first degree in and L. Second, approach is by now a familiar one.
since = the partial derivatives Equation (6) expresses the assumption of a con-
are equal to w, respectively. stant returns to scale aggregate production func-
And third, the effect of time is neutral, as tion, with the parameter A(t) expressing the as-
incorporated in the shift parameter B(t). What sumption of neutral technical change.
we have, actually, is mathematically identical to
a constant returns to scale Cobb-Douglas pro-
duction function having neutral technical change For such a function, the marginal product of
and satisfying marginal productivity rules. capital is = A(t) =
And yet, as we have seen, production since A(t) = By assumption, this marginal
whatsoever presented being g e n - product is equal to the rate of profit
erated by such so long as shares are
constant and the measures of capital and labor
such that k is uncorrelated with Therefore,
precisely because is a mathematical rela- and by rewriting, we can express this in terms of
tionship, holding true for large classes of data as- the profit share s:
sociated with constant shares, it cannot be inter- k rk
share of profit in output
preted as a production function, or any produc-
tion relation at all. If anything, it is a distributive
relation, and sheds little or no light on the under- expressed purpose was to distinguish
lying production In fact, since the between shifts of the assumed production func-
constancy of shares has been taken as an empiri- tion (due to technical change) and move-
cal datum throughout, equation (5a) does not ments along it (due to changes in the
shed much light on any theory of distribution labor ratio,
either. Figure 5.1 illustrates the geometric assump-
I emphasized earlier that the theoretical basis tion implicit in paper. Points A, and
of aggregate production function analysis was are observed points, at times and respec-
extremely weak. It would seem now that its tively, while represents the adjusted point
apparent empirical strength is no strength at all, after neutral technical change has been re-
but merely a statistical reflection of an algebraic moved. Thus points and lie on the under-
relationship. For the neoclassical old guard, the lying production function.
retreat to data is really a rout. Algebraically, in terms of Equation the
aim of his procedure is to partition output per
worker into A, the technical change shift
parameter, the underlying production
Applications function to which 1 just referred. In order to do
this, first differentiates Equation (6):
It is obvious that one can apply Equation (5a) in
many ways. The section that follows will reex-
amine famous paper on measuring tech-
nical change. The humbug production func-
tion section will present a numerical example
to illustrate the generality of Equation The (Value of
section on Fishers simulation experiments will
extend the preceding analysis; and the final sec-
tion will touch briefly on cross-section produc-
tion function studies.

Technical change and the aggregate production

function: In what is considered a semi-
nal paper Robert intro-
duced in 1957 a novel method for measuring the
contribution of technical change to economic
growth. Since that several refinements of
original calculations have been estab-
lished, all aimed at providing better measures of
labor and capital by taking account of education, Figure 5.1
Rearranging, be well represented by the functional form in
= B(t) a form which is mathematically
= + identical to a constant returns to scale
1 Douglas function, with neutral technical change
Since from = and from and marginal products equal to factor re-
k rk wards. In fact, the algebra indicates that
underlying production function should
be of the form:
being the share of profit in gross output, we can
write =
f(k) + k
is of course the (roughly) constant share and
is a constant of integration which depends only
Equation (8) is derived from the assumptions of
on the initial points of the data.
a constant returns to scale aggregate production 1909-1942 in his
function, with distribution determined by mar- for these years the average profit share s
ginal productivity rules. Equation de- Moreover, since in any period =
rived earlier from an identity and therefore from Equation in period we
always true any production and distribution may write = B(0) which gives us
behavior, is mathematically identical to (8). = (0) For
It follows therefore that =
that is, measure
this residual B(t) represents the shift parameter
A(t) (compare Equations (3) derived from an
technical change is merely a weighted average of identity, and Equation (8) derived from
the growth rates of the wage rate, and the rate lows assumptions), so that B(0) = A (0); as
of profit, mentioned earlier, he takes A(0) = From
data provide him with a series for Table 1, p. 315 of his article, we get =
gross output per worker capital per worker = 2.06, which when combined with B(0) =
and profit share for the United States from A(0) = gives 0.725.
1909-1949. From this data, he calculates the Thus, on purely algebraic considerations one
rates of change and and using these would expect underlying production
rates along with the data for the profit share he function to be characterized by
derives a series for k/A =
To the series for represents the = + k
rate of change of technology; since a scatter dia- This, of course, is virtually identical to
gram of on k shows no apparent correla- regression result, equation as it
tion, he concludes that technical change is es- should it is a law of not a law of
sentially neutral. By setting A(0) = 1, he is able
to translate the rate of technical change into
a series for A(t), the shift parameter. Finally,
since by definition = he is able to The humbug production function. The analysis of
combine his derived series for A(t) with his the laws of algebra led to the conclusion that
given series on to derive the underlying pro- production series k whatsoever, be
duction function q/A(t). represented being by
Plotting f(k) versus k, gets a diagram Douglas production neutral
with noticeable curvature, and notes with obvi- technical change and satisfying marginal produc-
ous satisfaction that the data gives a distinct tivity rules, so long as shares are constant
impression of diminishing returns and the measures of capital and labor such that
1957, p. 380). fact, underlying is uncorrelated is possible to illus-
production function to be extremely well repre- trate the generality of the above result by means
sented by a Cobb-Douglas function: of a numerical example. Consider, for example,
= + = (9)
an economy with illus-
trated in Figure 5.2 and having the same profit
Given our preceding analysis in the section share as in data for the United States.
on laws of algebra, it is not difficult to see why The Humbug data set gives us a series for k,
results turn out so nicely. We know for and from which we can calculate rates of
instance that his data exhibit roughly constant change and k/k. From these, in turn, we
shares, and the residual term = is derive = (The calculations
correlated with k. From purely algebraic consid- appear in Figure 5.5
erations, therefore, one would expect the data to Plotting B/B on gives us a scatter diagram
 years, share is
Moreover, since SO, = 2.00, and
for Humbug data, we
the constant term to be In In
B(O) Algebraic consid-
erations therefore tell us that the constant term
will be -0.459 and the
The actual regression of on k, presented
below, gives virtually identical results.
-0.453 =
The function is of course much more trou-
blesome. A simple glance at Figure 5.3 tells
that no linear or log-linear function will suffice
for a numerical approximation.
even in this case a approximation is

corrected of

Figure 5.3
 , I I I I I I
2.0 2.4 2.8 3.2 3.6 4.0
Figure 5.4

Combining these two fitted functions, one ar- by a microeconomic Cobb -Douglas production
rives at a numerical specification for even the function:
Humbug data (Table 5. I)!
where = 1, . . . , N (14)
Fishers simulation experiments. Earlier, I men-
tioned Franklin Fishers extensive (and expen- (The are constant over time, but in general
sive) simulation experiments, in which he finds, A,(t), and K,(t) are not.)
to his surprise, that aggregate Cobb-Douglas At any instant of time, the total stock of labor
functions seem to work for his simulated L(t) in the economy is given. The basic proce-
economy even when the theoretical conditions dure followed in the model is to allocate this
for such an aggregate function are carefully vio- given supply among the existing industries so as
lated, so long as the particular simulation run to equalize the industry marginal products of
happens to have roughly constant wage (and
labor MPL): this of course
hence profit) shares (Fisher, 1971, p. 306). yields the maximum aggregate output Q(t)
It is worth noting at this point that what Fisher
means by aggregate production functions work-
In general, the marginal product of a
ing, is not simply that they give a good fit to Douglas function is =
Since these are all equalized for the various in-
gross output or gross output per worker
but also that the estimated marginal dustries to a single ievel, we can denote this
products of labor, and presumably of capital, common level by and write:
closely approximate the actual wage and profit
rates, respectively (Fisher, 1971). W(t)
1 have already demonstrated in section on the
represents the imputed rental (uniform
laws of algebra why in general an aggregate
wage rate) of a unit of labor, so that the wage bill
Cobb-Douglas may be expected to work, in the
in the industry is:
sense explained earlier, for data which
constant wage shares. In this section, however,
it will be shown that even Fishers massive com-
Thus, the aggregate wage bill is:
puter simulation is in reality only an application
of the laws of algebra. = =

The the Fishers simu- so that the wage share in total output Q(t)
lated economy consists of N industries, each is:
producing the same type of output Q, using
homogeneous labor but its own type wage share =
of machine stock Thus and are both
quantities of the same good, produced by indus-
tries i and j, respectively, whereas and are Finally, since Q,(t) is the gross output of the
stocks of different types of machines. industry, and = Q,(t) its wage bill,
Each industry is assumed to be characterized the difference between the two, the gross
Table 5. I. data

Actual
share of Humbug Humbug
property output per capital per
Year income worker worker

B/B

1909 0.335 0.80 2.00 -0.125 0.000 -0.125 1 .ooo 0.800

1910 0.330 0.70 2.00 -0.143 0.000 -0.143 0.875 0.800
0.335 0.60 2.00 +O. 167 0.000 167 0.750 0.800
1912 0.330 0.70 2.00 0.000 -0.017 0.875 0.800
1913 0.334 0.70 2.10 0.000 0.048 -0.016 0.860 0.814
1914 0.325 0.70 2.20 -0.143 0.000 -0.143 0.846 0.826
1915 0.344 0.60 2.20 0.000 0.725 0.828
1916 0.358 0.80 2.20 0.000 0.045 -0.016 0.965 0.830
1917 0.370 0.80 2.30 -0.250 0.000 -0.250 0.948 0.843
1918 0.342 0.60 2.30 0.000 0.044 -0.015 0.710 0.845
1919 0.354 0.60 2.40 0.000 0.042 -0.015 0.700 0.857
1920 0.319 0.60 2.50 +O. 167 0.000 +O. 167 0.690 0.870
1921 0.369 0.70 2.50 +o. 143 0.000 +o. 143 0.805 0.870
1922 0.339 0.80 2.50 -0.250 0.040 -0.264 0.921 0.869
1923 0.337 0.60 2.60 0.000 0.678 0.885
1924 0.330 0.80 2.60 -0.063 0.019 -0.069 0.902 0.887
1925 0.336 0.75 2.65 0.067 0.019 -0.073 0.810 0.893
1926 0.327 0.70 2.70 0.019 0.780 0.897
1927 0.323 0.75 2.75 0.018 0.830 0.903
1928 0.338 0.80 2.80 -0.250 0.000 -0.250 0.880 0.908
1929 0.332 0.60 2.80 0.000 0.036 -0.012 0.660 0.908
1930 0.347 0.60 2.90 0.000 0.052 -0.018 0.652 0.920
1931 0.325 0.60 3.05 +O. 167 0.000 +O. 167 0.641 0.935
1932 0.397 0.70 3.05 0.000 -0.049 0.019 0.74% 0.335
1933 0.362 0.70 2.90 +o. 143 0.000 +o. 143 0.764 0.916
1934 0.355 0.80 2.90 0.000 0.052 -0.018 0.874 0.916
1935 0.351 0.80 3.05 -0.125 0.000 -0.125 0.860 0.930
1936 0.357 0.70 3.05 0.143 0.033 +O. 132 0.752 0.930
1937 0.340 0.80 3.15 0.250 0.000 -0.250 0.852 0.940
1938 0.331 0.60 3.15 0.000 0.032 -0.011 0.638 0.940
1939 0.347 0.60 3.25 0.000 0.031 -0.011 0.633 0.948
1940 0.357 0.60 3.35 0.000 0.626 0.960
1941 0.377 0.80 3.35 0.000 0.070 -0.026 0.843 0.950
1942 0.356 80 3 000 0.820 0.975
1943 0.342 0.80 3.45 -0.250 0.000 -0.250 0.832 0.964
1944 0.332 0.60 3.45 0.000 0.044 -0.015 0.624 0.964
1945 0.314 0.60 3.60 +O. 0.000 +O. 167 0.614 0.978
0.312 -0.014 0.717 0.975
1947 0.327 0.70 3.55 0.721 0.970

in the industry, is treated as the imputed chine being a different type. An index of
rental of its unique machine stock K,(t). De- capital has therefore to be constructed,
fining this gross profit and it is known that in gcncrnl any such
as (t), we have: will violate the strict conditions under which
the microeconomic Cobb-Douglas production
= functions can bc theoretically aggregated into
= gross profits in industry
a macroeconomic Cobb -Douglas production
Since output and labor L,(t) are homoge- function (Fisher, 1971, pp. 307-08). On the
neous across industries, their respective basis of aggregation theory, therefore, one would
gregates are derived by simple addition. But not expect the macroeconomic variables in this
since each industry has a type of simulated economy to behave as if they were
chine, an capital hy even
derived by adding machines together, each if aggregate shares happen to remain roughly
constant over time. That, of course, is the rea- implies that each firms output and
son for Fishers surprise at his results. employment grow at roughly the same rate. That
Fisher chooses to construct an aggregate is, dropping time subscripts and denoting time
index in two steps. First, he runs the model derivatives by dots:
economy over its 20-year period, from which he
gets the gross profits of any given industry,
for each of 20 years. Similarly, over each of the
20 years he knows the machine stock K,(t) in the It is the central result
same industry: the ratio of the sums of
of this paper that constant shares, any ag-
these two is the average rate of return in the
gregate data Q , L whatsoever can be
industry:
described by a function of the form Q(t) =
providing the residual is
a function of time. What we must there-
= 20 year average rate of return
fore do for Fishers experiments. in order to see
in industry (19) why aggregate Cobb-Douglas functions work
for them, is to examine this residual
The units of each average return are output By definition, from Equation (3)
per machine type Thus Fisher can use these
in any one period t to aggregate the individual in-
dustry machine stocks into an aggregate index of
= = Fishers
index of capital is denoted by J. Thus =
J(t) = = Q/Q and K/K = and:

It is useful to note that in the above expression - L- -

L J L
the are nut functions of time, since they repre-
sent average rates of r e t u r n the whole
20-year period.
Since and are profit
The From Equation and wage shares, respectively, we need only
the wage share is examine the rates of change of

wage share = The first is easy. In all of his simulations,

Fisher specifies that labor grows at an average
rate of 3% trend with small random deviations
Now, as Fisher notes, since the parameters from the trend (Fisher, 1971, p. 309). Ignoring
are independent of time, the wage share will be small random deviations
roughly constant over time only if the relative
outputs are roughly constant over
time (Fisher, 1971, p. 321, footnote 21). Let us L

denote these roughly constant relative outputs The growth rate of the aggregate capital index
by Pi, and the constant wage share by (1 s), is a bit more complicated. In Equation (20)
J(t)
the lack of time subscript denoting their con- we defined
stancy:
J,(t) =

where the are constant over time.

tiating this with respect to time,

In each industry, the wage bill, as derived in

Equation is = From
the aggregate wage bill is = (1
and dividing one by the other, we get:

1 I

Finally, to prepare us for the last step, we Dividing through we get:

need to note that the rough constancy of relative 1
relative employment
During all his simulations, Fisher assumes that
each capital K,(t) grows at an essentially
rate one which in general differs from
industry to industry. Thus,

(27) Given that the constant wage share I =

we can write the profit share = 1
and this in turn implies But by definition SO that

Therefore is a weighted average of the Thus,

with weights which sum to one, since J(t)
(This type of weighted average is = = (1 a,
as a convex combination, and implies
that will always be between the largest and where = (1 From this, we at long last
smallest .)
Finally, we come to the growth rate of ag-
gregate output From Equation + 11

we know so
in important to note that the
Q(t) Q;(t) and when summed over each
sum to I.
From this, we can derive
In the
expression (32) for the basic structural
parameters are and Of these, repre-
Of the terms in expression we sents the rate of growth of the machine stock
know that L/L over any given simulation run, whereas rep-
from and from (27). To resents the rate of technical change in the in-
this, we need only add the fact that in general, dustry. (Since the are constant over any given
ignoring small random Fisher as- run, changes in the shift parameter repre-
sumes that the shift parameter grows at an es- sent the only possible technical change in any in-
sentially constant rate, which differs from dustry.)
to industry. Fisher partitions his simulations into two
basic groups. In the first of these, which he calls
(30) Hicks experiments, he sets all = 0. Thus,
in each of these experiments, there is technical
All of this gives change 0) but no growth in the size of the
machine stock = 0). Under these condi-
tions, reduces to a constant over time.

But = = constant wage share, constant over time)

from (22). So
Thus, for Hicks can expect
(1 from purely algebraic considerations that
In +
Combining the expressions for j/J, and where is a constant of integration.
we return to the all important residual From the laws of algebra (Equation we
of equation (25); know that in general if is solely a function of
time, any data associated with constant shares
can be represented by the functional form
(since Fisher uses J as an index nf capital,
what we previously called = K/L is now
= =
Taking natural logs,
 In = In B(t) + In + j= In j and combining the
= In In terms into a single constant
and combining the constants into. a single con-
stant we get For the capital experiments, therefore, purely
In = + algebraic considerations lead us to expect that
What we have shown therefore is that for Fishers data can be represented by what
be a Cobb-Douglas production func-
purely (as opposed
to econometric) lead us to the tion with a constant level of technology and a
conclusion that whenever shares are (roughly) marginal product of labor equal to the actual
constant Fishers aggregate data can be gen- wage. Once this is precisely the result
erated by what to be a Cobb-Douglas gets his experiments.
production function with a constant rate of It is important to note that Fisher himself
technical change and a marginal product of labor
never presents the exact regression results in-
equal to the actual wage. volved (an understandable omission considering
This is precisely Fisher gets his that there were a total of 1010 runs of this simu-
Hicks experiments: for this set of experiments,
lated economy, each run covering a 20-year
the functional form which repeatedly works the period). Instead, he tells us only that the best fits
best (in the sense that the estimated marginal to the aggregate data were derived from an equa-
product of labor most closely approximates the tion of the form = + for
actual wage) is one which assumes constant re- Hicks experiments, and one of the form =
turns to scale and a constant rate of technical j for experiments. To Fisher
change . this result comes as a surprise. But it should not,
have Fishers complicated
We now turn to the second set of experiments,
what Fisher calls his Capital experiments, in and expensive experiments have merely redis-
which all = 0. In this set of experiments, covered the laws of algebra.
therefore, there is positive or negative growth of Cross-section production
the machine stock 0) but NO technical
The direct analogy to constant shares in time
change = 0). Equation the genera1 series is the case of uniform profit margins
the now becomes: its per dollar sales) in cross-section data. Using
the subscript i for the industry (or firm), and
defining as the uniform profit
margin, we can rewrite Equation (3) as
In Equation (36) each term in the brackets is a
convex combination (a weighted average whose
weights sum to one) of the so that each term
lies between the largest and the smallest One Then, so long as the term in brackets is
would therefore expect the of these related with the above equation is alge-
terms to be close to zero; in addition, since the braically similar to a simple linear regression
constant wage share I = is itself a model = + with the term in brackets
convex combination of the parameters it it- playing the part of the disturbance term
self will be within range of any data the bracketed
since the unweighted average of the is term is small and uncorrelated with the depen-
0.75, the profit share will be roughly around dent variable the best fit will be a
0.25. is cross-section Cobb-Douglas production func-
to be small, multiplying it by will yield tion with constant returns and factors paid their
a number even closer to zero. In capital experi- marginal products.
ments algebraic considerations would therefore There are still other ways in which one may
lead us to expect: explain the apparent success of a Cobb-Douglas
in cross-section studies, the best single refer-
ence being Phelps Browns (1957) critique. In
subsequent note, Simon and Levy (1963) show
so that that any data having uniform wage and profit
B where is a constant rates across the cross section can be closely
approximated by the ubiquitous Cobb-Douglas
In setting this result into the general functional function having correct coefficients, even
form of Equation (5) = and nat- though the data reflect only mobility of labor and
ural logs of both sides, + + capital, not any specific production conditions.
Once again, it would seem that the apparent And what are these facts? Simply, that again
empirical success of the Cobb-Douglas function and again, aggregate Cobb-Douglas production
having correct coefficients is perfectly con- functions work that is, they not only give a
sistent with wide varieties of data, and cannot be good fit to aggregate output, but they also gener-
interpreted as supporting aggregate neoclassical ally yield marginal products which closely
production and distribution theory. approximate factor rewards. Since the aggregate
production function is the simplest form of
the grand neoclassical parable, its apparently
Summary and conclusions strong empirical basis has often been taken as
providing a good measure of support for the old
It is characteristic of theoretical parables that time religion, regardless of what the theory says,
they illustrate truth para- The of this has to
digm, truths which more developed theoretical show that these empirical results do not, in fact,
structures may modify and elaborate, but cannot have much to do with production conditions at
undermine. In the neoclassical progression of all. Instead, it is demonstrated that when the dis-
parables from simple exchange to capitalism as data (wages and profits) exhibit con-
the final solution to Mans natural greed, one stant shares, there exist broad classes
central theme which emerges right in the begin- tion data (output, capital, and labor) that can
ning is the conception of equilibrium prices as always be related to each other through a func-
scarcity prices: relative prices which reflect tional form which is mathematically identical to
the relative scarcity of commodities. Cobb -Douglas production with
In their most developed form, neoclassical constant returns to scale, neutral technical
parables have sought to present the notion of change, and marginal products equal to
scarcity pricing as an explanation of the distribu- tor
tion of income between workers and capitalists. Since this result is a mathematical conse-
Here, the task is to portray a capitalist economy quence of any (unexplained) constancy of
in such a way that the wage and profit rates may shares, it is true even for very implausible data.
be seen to be scarcity prices of labor and For instance, data that spell out word
capital, respectively. But for this to be even a HUMBUG were used as an illustration, and
logical possibility, it is at the very least neces- it was shown that even the humbug economy
sary that the wage and profit rates behave as can bc by Cobb-Douglas
they were scarcity prices i.e., that the profit tion function having all the previously men-
rate fall as the capital-labor ratio rises, and the tioned properties.
wage rate fall as the labor-capital ratio rises. Similarly, we have examined
This correlation is minimally necessary for the paper on measuring technical change; and here
internal consistency of the parable (though of too it is shown that the underlying production
course its existence would hardly justify the im- function which he isolates, by removing the ef-
plied causation). fects of technical change, can be algebraically
Alas, the grand neoclassical parables have anticipated, even down to the fitted coefficients
fallen on hard times, and after repeated demon- of his regression.
strations of their logical inconsistencies, they Next, Franklin Fishers mammoth simulation
have been abandoned by the high-brows of the experiments are examined and once again it be-
theory; not without regret, though, for as comes clear that the laws of algebra can antic-
uelson so insightfully notes, within the parable ipate the laws of simulation from the structure of
the apologist for capital and for thrift has a less the experiments alone.
difficult case to argue (Samuelson, 1966). Lastly, in the final part of this chapter, the
If all this for those nos- analysis is to provide a
talgic for the old time parables of neoclassical tion for cross-section aggregate production func-
writing, we must remind ourselves that scholars tions. The overall impact of these discussions, it
are not born to live an easy existence We is hoped, will be to demonstrate that the
respect, and appraise, the facts of life to which the neoclassical hangers-on clutch so
uelson, 1966). desperately is as empty as their own abstrac-
Not everyone was ready to give up the old tions.
time parables though, and those who chose to ig-
nore the previously mentioned facts of life
sought succor where else? in the facts. Postscript
The real world, whose vulgar intrusions neo-
classical theory had in the past so carefully The point of this chapter is to demonstrate that
avoided, became its last refuge. Facts, after all, as long as distributive shares are constant, it is
are always better than facts-of-life. an algebraic law that the Cobb-Douglas
 tion fits almost any data. Hence, data would be consistent with a neoclassical pro-
paper and the Humbug data stand on the same duction function having neutral technical
footing. change and marginal products equal to factor
has recently claimed that all along the rewards.
intention of his 1957 paper was to yield an Obviously, given that the underlying func-
exact Cobb-Douglas and tuck everything else tion f(k) was numerically specified by the laws of
into the shift factor 1974, p. 121). But algebra (Equation (12) and note 9, in this
his own printed words give quite a different chapter), all that would have been necessary for
impression: in the original paper, after he has a complete numerical specification was a fitted
derived the so-called shift factor A(t), ex- function for B(t). However, since such a fitted
pressly states his intention to discuss the shape function was not necessary to the logic of my
of I) and the (underlying) ag- argument, I was content with
gregate production function 1957, p. B(t) versus time, as in Figure 5.3.
317). To this end, he constructs a graph of f(k) A glance at Figure 5.3 is sufficient to indicate
versus noting with obvious satisfaction that that no simple linear or log-linear function will fit
in spite of the amount of a priori doctoring B(t). And yet this is precisely the that
which the raw figures have undergone, the fit is uses in his regression, He naturally gets
remarkably 1957. p. 317). giving a very poor fit. How clever.
rise to an inescapable impression of curvature, In this version of the paper, for the sake of
of persistent but not violent diminishing re- completeness, I do actually specify a fitted func-
turns 19.57, p. 318). tion for B(t), with an = (Equation 13).
If, as now claims, he knew all along But the logic of the argument does not require
that the underlying production function would this step; it only requires that the so-called shift
be a Cobb-Douglas, then why bother recon- factor be a function solely of time: there is
structing it? Why the surprise at the tightness nothing in neoclassical theory, no law of pro-
of fit and the inescapable impression of curva- duction or of nature, which requires B(t) to be
ture? Why does need regression analy- linear or log-linear. Struggling under the weight
sis to confirm the visual of dimin- Myers seem
ishing returns . . 1957, 319). If have forgotten that linearity is merely a conve-
had indeed understood his own method, nient assumption whose applicability must at all
he should have known that regardless of the times be not merely assumed.
amount of a priori doctoring of the data, the laws
of algebra dictate that the fit of f(k) versus
would be very tight as well as being inescapably Notes
curved. But it is hardly necessary to rediscover
these algebraic artifacts by means of graphs and . the core of the theory of a private owner-
. .

ship economy is provided by the theory of ex-

change (Walsh, 1970, p. 159).
Having just said that his method and his edu- Garegnani in fact does not state it this way. He
cation lead him to conclude that even the shows that the necessary and sufficient condition is
Humbug economy is neoclassical, next that the wage-curves all be straight lines, and
asserts the very opposite. With the help of Sam- shows that this in turn is true when all industries
uel L. Myers, he runs a regression of the form have the same capital-labor ratios, i.e., when
In = a, + + b on the Humbug data, prices are proportional to labor values (Garegnani,
and finds to his obvious that this 1970, 421)
Q(t) value of output; K(t) value of the utilized
not only to a very poor fit but also gives rise to a stock of capital; L(t) employed stock of labor;
negative coefficient for The moral seems t time.
clear: production functions do not work for I thank Professor Luigi Pasinetti for having pointed
the Humbug data, whereas they do for real data this out in his comments on an earlier version of
1974, p. 121). this paper.
But once again, his method and education be- R. R. Nelson a summary of subsequent
tray him, The laws of algebra show that almost refinements (Nelson, 1964).
any production data associated with a constant In order to isolate shifts of the aggregate produc-
profit share could be cast in the form Q = tion function from movements along it
The Humbug data was an illustration of 1957, p. 314).
The discrete equivalent for is AA/A, where
this, and it was sufficient for my purpose in the A A = A(t + 1) A(t). Thus A(t + 1) = A(t) +
original paper to show that even in this case the AA/A]; in 1909, = 0, and by setting A(0) =
underlying function was extremely well derives a series for A(l), A(2) . . . , from
fitted by the Cobb-Douglas = = the data on
and that the so-called shift factor was Since calculations contained an arithmeti-
a function of Hcncc, Humbug cal rcprcscnting ycnrs
 1943-1949 clearly lay outside the range of any hy- so that
pothesized curve. After expressing some
Suluw them out of his
1957, 318).
9 The deviation of the numerical value of the con-
stant term is explained on pp. 20-21 of 1 6 Fisher (1971, p. 309) assumes A, + so
1957 paper. that
I wish to thank Larry Heinruth and especially Peter 17 The function form in Fishers equation, the
Brooks, for the time and effort expended in best form for Hicks experiments, is log =
deriving this function. Two in- + where his corresponds
volved in the fitting. First, a two-year moving to our and his J/L to ourj. Fisher uses log for
average was constructed from the data for natural logarithms (Fisher, 1971, p. 313).
B(t), by means of the formula = [B(r) 18 Fisher has two ranges of and
in which the year 1909 represents = in both the unweighted average = 0.75
1, 1910 by = 2, etc. Second, the function of (Fisher, 1971, p. 309).
Equation (13) was fitted to this moving average 19 The functional form Fisher finds best for Capi-
B(t), a = tal experiments is =
11 Since the fitted function has = 16 parameters to which, allowing for notation differences, is iden-
it, and since there are T = 38 data points in the tical to equation 5.38 (Fisher, 1971, 313).
moving average the corrected for degrees 20 Yet confronted with the humbug data, says:
of freedom is (Goldberger, 1964): If you ask any systematic method or any edu-
cated mind to interpret those data produc-
tion the
the answer will be that they are exactly what

= = would be produced by technical regress with a pro-

duction function that must be very close to
12 By definition = Applying
Cobb-Douglas 1957, p. 121). What kind
this to the expression for in Equation (14) of systematic method or educated
that can interpret almost any data, even the
yields =
humbug data, as arising from a neoclassical pro-
13 From Equation = where is duction function?
constant over time. Thus 21 uses the form = +
since the general form under consideration is
dt = Pi so that A(t) +
has obviously specified A(t) as log-linear: A(t)
and a,

References
Similarly for employment from (23).
In = + + (small ran- Brown, P. 1957. The Meaning of Fitted
dom deviations). Ignoring the small deviations, and Douglas Functions, of Eco-
differentiating gives (Fisher, 1971, p. 309) nomic&,
Dobb, M. H. 1970. Some Reflections on the Sraffa
dt K,(t) System and the Critique of the So-called
Classical Theory of Value and Distribution, un-
= published.
Dewey, D. 1965. York:
Columbia University Press.
Dropping the time subscript, and differentiating,
Ferguson, C. E. 1969. The Theory
Production Distribution. Cambridge: Cam-
+
bridge University Press.
1971. Capital Theory Up to Date: A Comment on
+ (1 Mrs. Robinsons Article, Journal
Economics, Vol. IV, No. 2, 250-4.
Fisher, F. 197 1. Aggregate Production Functions and
the Explanation of Wages: A Simulation Experi-
ment, of Economics Statistics

Garegnani, P. 1970. Heterogeneous Capital, the Pro-

+ (1 Function and the Theory of Distribu-
tion, Review Studies, 37, 407-36.
Goldberger, A. S. 1964. Econometric New
York: John Wiley Sons.
1970. Investment. interest. and
1 Englewood Cliffs, N. J.: Prentice-Hall.
Hicks. J. R. 1965. New York: Simon, H., and Levy, F. 1963. A Note on the
Oxford University Press. Cobb-Douglas Function, Economic
1968. Economics. New Vol.
York: Macmillan. R. M. 1957. Technical Change and the Ag-
Nelson, R. R. 1964. Aggregate Production Functions gregate Production Function,
and Medium-Range Growth Projections,
Economic 1 9 7 4 . Law
of Production and Laws of Algebra:
Samuelson, P. A. 1962. Parable and Realism in The Humbug Production Function: A Comment.
tal Theory: The Surrogate Production Function, Review Economics and Statistics, Vol. LVI,
29: 193-206. No. 1,121.
1966. Summing Up, Quarterly Walsh, V. C. 1970. to Contemporary
LXXX, New York: McGraw-Hill.
Shaikh, A. 1974. Laws of Production and Laws of Walters, A. A. 1963. Production and Cost Functions:
Algebra. The Humbug Production Function: A An Econometric Survey, No.
Comment, Economics and Statistics, l-2, l-46.
Vol. LVI, No. 1, 115-20.