arXiv:physics/0511191v1 [physics.

soc-ph] 22 Nov 2005

The Production Function

Guido Fioretti ∗
University of Bologna
Department of Business Economics
Department of Computer Science
February 2, 2008

Abstract
Productions functions map the inputs of a firm or a productive system
onto its outputs. This article expounds generalizations of the production
function that include state variables, organizational structures and increas-
ing returns to scale. These extensions are needed in order to explain the
regularities of the empirical distributions of some economic variables.

Keywords: Production, Organization, Firms, Growth.

1 Introduction
In the course of its development, the economic theory of production explored
paths that are at odds with general equilibrium theory. Some of these research
directions may be crucial in order to explain the regularities that have been dis-
covered regarding several economic variables, notably the distribution of the size
of firms. In particular, this article expounds some extensions of the function that
maps the production factors onto the products of a firm, the production function.
Most of the issues expounded in the rest of this paper are little known among
economists themselves. If they are, they are known in a lengthy, often verbal form.
∗
Contact address: Guido Fioretti, via di Corticella 23, I - 40128 Bologna. E-mail: fioret-
[email protected].

1
By compacting them in mathematical language, this article hopes to be helpful to
those physicists wishing to apply their methods to economic problems.
The rest of this article is organized as follows. Section 2 adds state variables
to the conventional form of the production function. Section 3 explores the pos-
sibility of production functions that take account of the structural relations within
a firm or an economic system. Section 4 explores the possibility of generalized
increasing returns to scale with reference to the empirical regularities that have
been observed in the distribution of the size of firms. Finally, section 5 frames the
previous issues with respect to the development of economics and its relationships
with other disciplines.

2 State variables in the production function

Any science is tempted by the naive attitude of describing its object of enquiry by
means of input-output representations, regardless of state. For instance, in the first
half of the XIX century it was fashionable for psychologists to study the behavior
of animals in terms of stimuli such as hunger or pain (the input) and the response
in terms of unrest or aggressiveness (the output). No attention was paied to what
was going on inside the brain (the state).
Tipically, microeconomics describes the behavior of firms by means of a pro-
duction function:

y = f (x) (1)
where x ∈ R is a p × 1 vector of production factors (the input) and y ∈ R is a
q × 1 vector of products (the output).
Both y and x are flows expressed in terms of physical magnitudes per unit
time. Thus, they may refer to both goods and services.
Clearly, eq. 1 is independent of state. Economics knows state variables as
capital, which may take the form of financial capital (the financial assets owned
by a firm), physical capital (the machinery owned by a firm) and human capital
(the skills of its employees). These variables should appear as arguments in eq. 1.
This is done in the Georgescu-Roegen production function [14], which may be
expressed as follows:

y = f (k, x) (2)

2
where k ∈ R is a m × 1 vector of capital endowments, measured in physical mag-
nitudes. Without loss of generality we may assume that the first m p elements
represent physical capital, the subsequent mh elements represent human capital
and the last m f elements represent financial capital, wuth m p + mh + m f = m.
Contrary to input and output flows, capital is a stock. Physical capital is mea-
sured by physical magnitudes such as the number of machines of a given type.
Human capital is generally proxied by educational degrees. Financial capital is
measured in monetary terms.
Georgescu-Roegen called the stocks of capital funds, to be contrasted to the
flows of products and production factors. Thus, Georgescu-Roegen’s production
function is also known as the flows-funds model.
Georgescu-Roegen’s production function is little known and seldom used, but
macroeconomics often employs aggregate production functions of the following
form:

Y = f (K, L) (3)
where Y ∈ R is aggregate income, K ∈ R is aggregate capital and L ∈ R is ag-
gregate labor. Though this connection is never made, eq. 3 is a special case of
eq. 2.
The examination of eq. 3 highlighted a fundamental difficulty. In fact, general
equilibrium theory requires that the remunerations of production factors are pro-
portional to the corresponding partial derivatives of the production function. In
particular, the wage must be proportional to ∂ f /∂L and the interest rate must be
proportional to ∂ f /∂K. These partial derivatives are uniquely determined if d f is
an exact differential.
If the production function is eq. 1, this translates into requiring that:

∂2 f ∂2 f
= ∀i, j (4)
∂xi ∂x j ∂xi ∂x j
which are surely satisfied because all xi are flows so they can be easily reverted.
If the production function is expressed by eq. 2 but m = 1 the following con-
ditions must be added to conditions 4:

∂2 f ∂2 f
= ∀i (5)
∂k∂xi ∂xi ∂k
Conditions 5 are still surely satisfied because there is only one capital good.
However, if m > 1 the following conditions must be added to conditions 4:

3
 ∂2 f ∂2 f
= ∀i, j (6)
∂ki ∂x j ∂x j ∂ki

∂2 f ∂2 f
= ∀i, j (7)
∂ki ∂k j ∂k j ∂ki
Conditions 6 and 7 are not necessarily satisfied because each derivative de-
pends on all stocks of capital ki . In particular, conditions 6 and 7 do not hold if,
after capital ki has been accumulated in order to use the technique i, capital k j is
accumulated in order to use the technique j but, subsequently, production reverts
to technique i. This possibility, known as reswitching of techniques, undermines
the validity of general equilibrium theory [31] [38] [13].
For many years, the reswitching of techniques has been regarded as a theo-
retical curiosum. However, the recent upsurge of biodynamic agriculture or the
resurgence of coal as a source of energy may be regarded as instances of reswitch-
ing.
Finally, it should be noted that as any input-state-output representation, eq. 2
must be complemented by the dynamics of the state variables:

k̇ = g(k, x, y) (8)
which updates the vector k in eq. 2 making it dependent on time.
In the case of the aggregate production function 3, eq. 8 combines with 3 to
constitute a growth model. In the case of the microeconomic production func-
tion 2, explicitating eq. 8 requires investigations of the strategies pursued by a
firm.

3 The production function of an organization

A production function in the form of eq. 2 does not distinguish elementary units
within an organization. This may be a serious shortcoming if the structure of in-
teractions between elementary units affects the final outcome. Unstructured pro-
duction functions have the same limitations as using compact chemical formulae
to represent allotropes.
Let us consider an organization composed by n units. Let K denote a m × n
matrix of capital endowments. The i-th column of K, denoted by ki , represents
the m capital endowments of unit i.

4
 Let y = F(K, x) denote the production function of the organization, where x
is a p × 1 vector and y is a q × 1 vector. The organizational production function F
arises from the interaction of n functions yi = f i (ki , xi , y j . . .) with j ∈ {1, . . .n}.
The pi × 1 vector xi entails the inputs that reach unit i. The qi × 1 vector yi entails
the outputs of unit i.
Vector x can be seen as arising from concatenation of the n vectors xi . Like-
wise, x can be seen as arising from concatenation of the n vectors xi . With this
convention, ∑ni=1 pi = p and ∑ni=1 qi = q. However, in general the inputs to the
organization do not reach all units. Likewise, not all the outputs of all units are
necessarily represented in the organizational output. Thus, some components of x
and some components of y may be zero.
Since organizational units may feed their outputs into other units, all outputs
y1 . . . yn , including yl , can be arguments of f l . Which particular y j s appear in each
f i depends on the structure of the organization.
Organizational structures are often quite intricated. Thus, only in a few cases
it is possible to derive a closed form of F from knowledge of the f i s and their
connections. In general, local linearized descriptions are a more affordable goal.
Let us define a q × p matrix Γ(x) entailing the derivatives of the outputs of all
units with respect to the inputs evaluated at x. If input i does not reach unit j, let
us stipulate that γi j = 0. Furthermore, let us define a q × q matrix Ω(y) entailing
the derivatives of the outputs of all units with respect to the outputs of units that
feed into them evaluated at y. If the output i does not feed back into unit j, let us
stipulate that ωi j = 0.
With these matrices, the differentials of inputs and outputs are linked by dy =
Γ(x) dx + Ω(y) dy. Thus, the organizational production function linearized in a
neighborhood of (x, y) takes the form:

Γ(x)
F(x) = x (9)
I − Ω(y)
Matrix Ω entails the structure of connections between organizational units. If
these connections are such that they produce cycles, particular sequences of opera-
tions can be repeated indefinetly giving rise to a routine [27]. Note that according
to this definition routines belong to the organization but not to their members,
who may even be unaware of them. Thus, this formalization may capture fuzzy
but fundamental concepts such as “organizational culture” or, at a more aggregate
level, “social capital”.
In the parlance of artificial intelligence, the circuits embedded in Ω imple-
ment a distributed memory that does not reside in any single unit but belongs to

5
the organization as a whole. No single unit is able to retrieve this knowledge, but
the organization as a whole does if certain environmental stimuli trigger partic-
ular routines. On the contrary, capital endowments K can be seen as localized
memories owned by the single units. Social science prefers the terms distributed
knowledge and localized knowledge, respectively.
Thus, the organizational production function has a lot in common with con-
nectionist models. In particular, let us make the following assumptions:

1. Each component of the input vector x is distributed to one and only one unit.
Thus, pi = 1 ∀i and p = n.
2. Each unit produces only one good yi , which is a component of the output
vector y. Thus, qi = 1 ∀i and q = n.
3. The production functions of organizational units take the form yi = xi +
∑nj=1 ki jy j , where the weights ki j are the elements of the n × n matrix of
capital endowments K.

With assumptions 1, 2 and 3, Γ ≡ I and Ω ≡ K. Thus, eq. 9 takes the form

I/I − K. This equation characterizes the short-term response of associative mem-
ories implemented on neural networks [22].
Similarly to eq. 8, there must be a mechanism for updating the capital endow-
ments of organizational units:

k̇i = gi (K, x, y) ∀i (10)

Eq. 10 is quite general. In fact, it says that the capital endowments of unit
i may change depending on its own capital endowments as well as the capital
endowments of other units, depending on all inputs to the organization and de-
pending on all outputs of the organization.
If assumptions 1, 2 and 3 hold, the analogy between organizational production
functions and neural networks can be further pursued. In fact, eq. 10 corresponds
to the rule for updating the weights of the neurons. In particular, both the back-
propagation algorithm employed in supervised networks and the rules employed
in unsupervised networks are such that k̇i = α x yi , with α ∈ ℜ.
Indeed, hierarchies are often managed like supervised neural networks. In
fact, resources are assigned to subordinated depending on the distance between
current output and a target fixed by the supervisor. On the contrary, regional or
national economies are managed by decentralized decisions depending on local
performance, similarly to unsupervised neural networks.

6
 Fioretti exploited the above similarities in order to model the recognition of the
potentialities of novel technologies by firms and productive systems [9] [10], but
this approach is new to economics. In general, economics ignores organizational
production functions except in the limited case of positive externalities between
firms. In fact, economics resorts to positive externalities in order to explain ge-
ographical agglomeration of industries [23] [11] or, more in general, economic
growth in spite of non-increasing returns to scale of single firms [32] [33].
In these applications, the units are firms and the organizational production
function is the macroeconomic production function. Since firms are independent
from one another, the organizational production function is such that Γ ≡ I.
Positive externalities are supposed to emanate from firms because firms pro-
vide workers with human capital that they diffuse when they move to other firms.
Through a mean field, firms feed back into other firms. In regional models, each
firm feeds back into firms that are geographically close. In growth models, each
firm feeds back into all other firms. Thus, in regional models the elements of Ω
depend both on the relevance of human capital and the geographical distance be-
tween any two firms. On the contrary, in growth models the elements of Ω depend
only on the relevance of human capital.
A further simplification is generally made by assuming that there exists only
one kind of human capital. With this assumption, in regional models the rows of
Ω are proportional to one another by a factor depending on physical distance. On
the contrary, in growth models the rows of Ω are all equal to one another.
Thus, growth models are generally aggregate models. The positive feedbacks
are captured by a term increasing with time that is multiplied to a standard Cobb-
Douglas production function:

Y = A(t) K α L1−α (11)

where α ∈ ℜ. As in eq. 3, Y is aggregate output, K is aggregate capital and L is
aggregate labor. The term A(t) is such that Ȧ > 0.
This function is compatible with general equilibrium theory because it has
constant returns to scale. However, the term A(t) is able to account for economic
growth, allegedly arising out positive externalities between firms operating with
decreasing returns to scale.
Macroeconomic data apparently fit into eq. 11. However, it can be shown
that eq. 11 may fit reality because of algebraic reasons, not because of economic
reasons [19] [34] [26].
By definition, at any time aggregate income Y is equal to the sum of aggregate

7
wages W and aggregate profits Π:

Y (t) ≡ W (t) + Π(t) (12)

This identity can be divided by the amount of labor L(t) in order to be ex-
pressed as y(t) ≡ w(t) + r(t)k(t), where y(t) = Y (t)/L(t) is the income/labor ra-
tio, w(t) = W (t)/L(t) is the wage rate, r(t) = Π(t)/K(t) is the profit rate and
k(t) = K(t)/L(t) is the capital/labor ratio. By differentiating one obtains ẏ/y ≡
w/y(ẇ/w) + rk/y(ṙ/r) + rk/y(k̇/k). By denoting the share of profits α = rk/y
and the share of wages 1 − α = w/y, eq. 12 can be expressed as:

ẏ Ȧ k̇
≡ +α (13)
y A k
where ȦA = [(1 − α)ẇ/w + αṙ/r].
Suppose, as it is roughly the case, that the shares of profits and wages are
roughly constant with time. 1 Then, A is only a function of time and eq. 13 can be
integrated to y(t) ≈ A(t)kα which, by multiplying by L(t), can be written as:

Y ≈ A(t) K α L1−α (14)

which is not a production function [34].
Equation 14 is not a production function in the sense that it does not express
any technological capability. It simply arises out of definitional identities and an
empirical regularity.
Thus, the Cobb-Douglas production function actually has no empirical sup-
port. Macroeconomic production functions of the form 3 may not exist at all. In
spite of widely accepted simplifications, the intricacies of organizational relations
may be unavoidable.

4 Increasing Returns to Scale

Economic theory assumes equilibrium, which is realized under the hypothesis
of decreasing returns to scale. In the basic version the production function 1 is
assumed, with ∂ f /∂xi > 0 ∀i and ∂2 f /∂xi ∂x j > 0 ∀i, j : i 6= j but ∂2 f /∂x2i < 0 ∀i.
1 If wages increase, profits decrease. Thus, the economy slows down, unemployment increases
and, consequently, wages decrease. Thus, the economic system regulates itself around shares of
profits and wages that oscillate within bounds.

8
 However, firms grow at differential rates and some of them become quite large.
In order to explain this fact it is generally assumed that returns to scale may in-
crease until size remains below a certain threshold, i.e. that ∀i ∃δi ≥ 0 : ∀xi > δi
it is ∂2 f /∂x2i < 0 but for at least one i, ∀xi < δi it is ∂2 f /∂x2i > 0.
The distribution of the size of firms has be subjected to a number of empirical
investigations. These have highlighted the following empirical regularities:

• The distribution of the size of firms is skewed to the right [21], deviates
from a log-normal [39] and, according to the most recent studies, follows a
power law [3] [12].

• The empirical investigations carried out since the 1980s have ascertained
that growth rates decrease with firm size [18] [8]. In particular, the most
recent studies fit the data with a Laplace distribution [6] [2] [5].

• The standard deviations of growth rates are known since a long time to
decrease with size [20] [36] [18] [8]. By making use of more recent data
it has been shown that the distribution of the standard deviations of growth
rates follows a power law [6] [2].

In principle, an appropriate choice of the distribution of production functions

across industries and countries can explain any empirically observed distribution
of the size of firms as well as the distributions of their growth rates and standard
deviations. By means of an appropriate combination of increasing returns below
a certain threshold and decreasing returns above it, any empirically observed dis-
tribution can be justified.
However, the observed empirical regularities span a period of several decades,
in which fundamental technological changes have taken place. Across decades,
production functions have certainly changed. Thus, these regularities question the
validity of received economic theory.
A right-skewed distribution can be produced by assuming that firms grow at
a rate independent of their size (Gibrat’s law) [15]. However, the distribution of
the size of firms would be log-normal, whereas some empirical analyses claim
that it is not. Furthermore, growth rates and their standard deviations would be
uniformly distributed, and they are not.
Dynamics of entry and exit from the population of firms can be superimposed
to Gibrat’s law in order to account for the empirically observed regularities. Entry
and exit dynamics may result from internal restructurings of large organizations
[7] [1] [28] [2], or from births and deaths of firms [35] [25] [24] [37] [4] [30].

9
 Conventionally, size is either measured in terms of output or number of em-
ployees. Let us assume that the number of employees (the amount of human cap-
ital) is proportional to output so production functions can take the form of eq. 1.
Under this condition, size is measured by output.
Gibrat’s law translates into a stochastic multiplicative process [15]. The output
at time t obtains from the output at time t − 1 via multiplication by a n × n matrix
of stochastic processes Λt−1 :

yt = Λt−1 yt−1 (15)

In order to avoid that the economy implodes, there must exist at least one
eigenvalue λi of E{Λt−1 } such that |λi | ≥ 1. This implies that for at least one firms
returns to scale are non-decreasing, a circumstance which is incompatible with
general equilibrium theory. The idea is that the economy is inherently unstable
and unpredictable, but that specific entry and exit dynamics cause the empirical
regularities that have been observed in the distribution of the size of firms.
Economics does know a mechanism that may provide regularizing entry and
exit dynamics [29]. When demand is increasing, firms increase their produc-
tive capacity at the fastest rate allowed by their profits and debts. In fact, non-
decreasing returns to scale imply that firms are playing a winner-takes-all game.
They know that at a certain point demand will saturate so most of them will have
excess productive capacity and high debts, but if they choose not to grow, they will
be doomed. By playing the game of growth, they have a chance to become the
monopolist (most often, one of the few oligopolists that share a saturated market).
This is the exit dynamics. The entry dynamics becomes relevant after the loosers
are out of the market and takes place on novel technologies.
An aggregate formalization of this mechanism was provided by Goodwin with
his business cycle model where downswings begin when aggregate capital be-
comes larger than “desired capital”, calculated as a fraction of income [16] [17].
The idea is that available demand calculated as a fraction of income constitutes a
threshold for supply. If supply overcomes this threshold, some producers will go
bankrupt.
Let yti denote the output of firm i at time t, with i = 1, 2, . . .n. Let us suppose
that there are m markets, denoted by an index j = 1, 2, . . .m, with m ≤ n. Let n( j)
denote the number of firms that operate in market j.
Aggregate income at time t is Yt = ∑ni=1 yt−1i . This what consumers can pur-

chase in all markets. Firms correctly predict that at time t the size of market j
j j
will be ct Yt−1 , with 0 ≤ ct ≤ 1 for ∀t, j. Firms know that the market constraint is

10
 n( j) j
∑i=1 yti ≤ ct Yt−1 . However, each firm i operating in market j makes its choices as
j
if yti ≤ ct Yt−1 .
To this exit dynamics, a random entry dynamics should be added. Thus, the
rows of eq. 15 would change as follows:
n( j) j
1. For firms such that ∑i=1 yti ≤ ct Yt−1 , eq. 15 would remain unchanged.
n( j) j
2. For firms such that ∑i=1 yti > ct Yt−1 , an exit dynamics should be devised
that makes some of them disappear.

3. An entry dynamics should be added. Since this may create new firms as
well as new sectors, in general nt 6= nt−1 and mt 6= mt−1 .

Thus, in place of eq. 15 one would have:


n( j) j
 Λt−1
 i yt−1 if ∑i=1 yti ≤ ct Yt−1
yti = At yt−1
n( j) j
if ∑i=1 yti > ct Yt−1 (16)
εt−1
 i

for i > nt
j
where Λt−1i , the i−th row of Λ
t−1 , is such that yt ≤ ct Yt−1 and εt−1 is another
i i

stochastic process.
Matrix At is an algorithm applied at time t to the outcome of eq. 15 in order to
n( j) j
cancel some firms until the requirement ∑i=1 yti ≤ ct Yt−1 is satisfied. It may work
by eliminating the smallest units first, or by reducing the output of all firms by a
fixed proportion and eliminating those falling below zero output, or else.
Simulations of eq. 16 with thresholds provided by supply and demand have
generated distributions of firm sizes very well, besides reproducing the observed
stability of the size of industries [30]. The idea of economies where increasing
returns to scale are ubiquitous systemic constraints produce regular patterns is
intellectally intriguing, technically challenging and empirically sensible.

5 The authors put in context

The previous sections expounded several critical topics surrounding production
functions. Their place within economics, as well as extent to which this paper
provided original connections, can only be appreciated if the authors involved are
framed within the history and the conventional content of this discipline.

11
 Nicholas Georgescu-Roegen graduated in mathematics and physics before be-
coming an economist. Regrettably, his work on production functions is little
known among economists. Since curricula in economics generally do not en-
tail the concept of state variable, even those economists who are acquainted with
the flow-funds model are unable to view its dynamics as a general property of
dynamical systems. Nobody seems to be aware that macroeconomic production
functions are an instance of Georgescu-Roegen’s flow-funds model.
Piero Sraffa, Joan Robinson and Pierangelo Garegnani were marxist 2 economists
who elaborated the theme of reswitching as a criticism to the neo-classical (bour-
geois) theory of value. They had some resonance in the 1960s but were forgotten
thereafter. Since they failed to recognize that the point they were making derives
from a basic property of any function of two or more variables, the debate on
reswitching occupied tons of paper on the journals of that time.
The concept of an organizational production function is foreign to economics,
as well as any connection between production functions and neural networks.
Positive externalities are eventually included in production functions without any
mention of structural issues.
Warren Hogan and Anwar Sheikh were applied economists. Their criticism of
the Cobb-Douglas production function has been totally ignored by the profession,
possibly because it is so destructive. Recently, Scott Moss made it known in the
community of social scientists who make use of agent-based simulations.
The invariant properties of firms size distributions have been discovered by
econophysicists and most often published on this journal. These discoveries are
challenging because they question current economic theory. On the contrary, some
paths that have been abandoned should be possibly resumed.
In particular, George Richardson is fundamental to understand economic dy-
namics without assuming that an equilibrium is necessarily there. His book was
ignored when it was first published but it has been reprinted after more than twenty
years.
Likewise, Richard Goodwin with his non-linear business cycle model is a
prominent reference for non-equilibrium economic studies. Interestingly, Good-
win is also an early example of a physicist who chose economics as his research
field.
2 More precisely, “neo-Ricardians”.

12
References
[1] Luis A. Nunes Amaral, Sergey V. Buldyrev, Shlomo Havlin, Michael A.
Salinger, and H. Eugene Stanley. Power law scaling for a system of in-
teracting units with complex internal structure. Physical Review Letters,
80(7):1385–1388, 1998.

[2] Luis A. Nunes Amaral, Parameswaran Gopikrishnan, Vasiliki Plerou, and

H. Eugene Stanley. A model for the growth dynamics of economic organi-
zations. Physica A, 299(1–2):127–136, 2001.

[3] Robert L. Axtell. Zipf distribution of u.s. firm sizes. Science,

293(5536):1818–1820, 2001.

[4] Aharon Blank and Sorin Solomon. Power laws in cities population, finan-
cial markets and internet sites (scaling in systems with a variable number of
components). Physica A, 287(1–2):279–288, 2000.

[5] Giulio Bottazzi and Angelo Secchi. A stochastic model of firm growth.
Physica A, 324(1–2):213–219, 2003.

[6] Sergey V. Buldyrev, Luis A. Nunes Amaral, Shlomo Havlin, Heiko

Leschhorn, Philipp Maass, Michael A. Salinger, H. Eugene Stanley, and
Michael H.R. Stanley. Scaling behavior in economics: I. empirical results
for company growth. Journal de Physique I, 7(4):621–633, 1997.

[7] Sergey V. Buldyrev, Luis A. Nunes Amaral, Shlomo Havlin, Heiko

Leschhorn, Philipp Maass, Michael A. Salinger, H. Eugene Stanley, and
Michael H.R. Stanley. Scaling behavior in economics: Ii. modeling of com-
pany growth. Journal de Physique I, 7(4):635–650, 1997.

[8] Paul Dunne and Alan Hughes. Age, size, growth and survival: Uk companies
in the 1980s. The Journal of Industrial Economics, 42(2):115–140, 1994.

[9] Guido Fioretti. The investments acceleration principle revisited by means of

a neural network. Neural Computing and Applications, 13(1):16–23, 2004.

[10] Guido Fioretti. Recognizing investment opportunities at the onset of recov-

eries. working paper, submitted to Research in Economics, 2006.

13
[11] Masahisa Fujita, Paul Krugman, and Anthony J. Venables. The Spatial Econ-
omy: Cities, Regions and International Trade, chapter XVI. The MIT Press,
Cambridge (MA), 1999.

[12] Edoardo Gaffeo, Mauro Gallegati, and Antonio Palestrini. On the size dis-
tribution of firms: Additional evidence from the g7 countries. Physica A,
324(1–2):117–123, 2003.

[13] Pierangelo Garegnani. Heterogeneous capital, the production function and

the theory of distribution. The Review of Economic Studies, 37(3):407–436,
1970.

[14] Nicholas Georgescu-Roegen. The economics of production. The American

Economic Review, 60(2):1–9, 1970. Papers and proceedings.

[15] Robert Gibrat. Les Inegalitées Économiques. Librairie du Recueil Sirey,

Paris, 1931.

[16] Richard M. Goodwin. The nonlinear accelerator and the persistence of busi-
ness cycles. Econometrica, 19(1):1–17, 1951.

[17] Richard M. Goodwin and Lionello F. Punzo. The Dynamics of a Capitalist

Economy. Westview Press, Boulder, 1987.

[18] Bronwyn H. Hall. The relationship between firm size and firm growth in the
us manufacturing sector. The Journal of Industrial Economics, 35(4):583–
606, 1987.

[19] Warren P. Hogan. Technical progress and production functions. The Review
of Economics and Statistics, 40(4):407–411, 1958.

[20] Stephen Hymer and Peter Pashigian. Firm size and rate of growth. The
Journal of Political Economy, 70(6):556–569, 1962.

[21] Yuji Ijiri and Herbert A. Simon. Skew Distributions and the Sizes of Business
Firms. North Holland, Amsterdam, 1977.

[22] Teuvo Kohonen. Self-Organization and Associative Memory. Springer,

Berlin, 1989.

[23] Paul Krugman. Geography and Trade, chapter II. The MIT Press, Cam-
bridge (MA), 1991.

14
[24] Moshe Levy and Sorin Solomon. Power laws are logarithmic boltzmann
laws. International Journal of Modern Physics C, 7(4):595–601, 1996.

[25] Patrick McCloughan. Simulation of concentration development from mod-

ified gibrat growth-entry-exit processes. The Journal of Industrial Eco-
nomics, 43(4):405–433, 1995.

[26] Scott Moss. Relevance, realism and rigour: A third way for social and eco-
nomic research. working paper.

[27] Richard R. Nelson and Sidney G. Winter. An Evolutionary Theory of Eco-

nomic Change. Harvard University Press, Cambridge (MA), 1982.

[28] Vasiliki Plerou, Luis A. Nunes Amaral, Parameswaran Gopikrishnan, Mar-

tin Meyer, and H. Eugene Stanley. Similarities between the growth dynam-
ics of university research and of competitive economic activities. Nature,
400(6743):433–437, 1999.

[29] George B. Richardson. Information and Investment. Oxford University

Press, Oxford, 1960.

[30] Matteo Richiardi. Generalizing gibrat: Reasonable multiplicative models of

firm dynamics. Journal of Artificial Societies and Social Simulation, 7(1),
2004. Available at <http:// jass.soc.surrey.ac.uk/ 7/1/2.html>.

[31] Joan Robinson. The production function and the theory of capital. The
Review of Economic Studies, 21(2):81–106, 1953–54.

[32] Paul M. Romer. Growth based on increasing returns due to specialization.

The American Economic Review, 77(2):56–62, 1970. Papers and proceed-
ings.

[33] Paul M. Romer. Endogenous technological change. The Journal of Political

Economy, 98(5):S71–S102, 1990.

[34] Anwar Shaikh. Laws of production and laws of algebra: The humbug pro-
duction function. The Review of Economics and Statistics, 56(1):115–120,
1974.

[35] Herbert A. Simon and Charles P. Bonini. The size distribution of business
firms. The American Economic Review, 48(4):607–617, 1958.

15
[36] Ajit Singh and Geoffrey Whittington. The size and growth of firms. The
Review of Economic Studies, 42(1):15–26, 1975.

[37] Sorin Solomon and Moshe Levin. Spontaneous scaling emergence in generic
stochastic systems. International Journal of Modern Physics C, 7(5):745–
751, 1996.

[38] Piero Sraffa. Production of Commodities by means of Commodities. Cam-

bridge University Press, Cambridge, 1960.

[39] Michael H.R. Stanley, Sergey V. Buldyrev, Shlomo havlin, Rosario N. Man-
tegna, Michael A. Salinger, and H. Eugene Stanley. Zipf plots and the size
distribution of firms. Economics Letters, 49(4):453–457, 1995.

16