Вы находитесь на странице: 1из 65

f petri

Fabio Petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 1

Microeconomics for the critical mind

CHAPTER 5.

november 2011

THE THEORY OF THE FIRM, PARTIAL EQUILIBRIA, PERFECT COMPETITION,


AND THE ATEMPORAL (ACAPITALISTIC) GENERAL EQUILIBRIUM WITH
PRODUCTION
5.1. The present chapter extends the marginalist theory of competitive general equilibrium to
include production (without capital goods and without a rate of interest; capital goods and rate of
interest raise special problems in the marginal/neoclassical approach and will be discussed in
Chapters 7, 8 and 9). However, in order to make our treatment of production decisions sufficiently
general and thus also useful for subsequent chapters, we admit the presence of capital goods when
discussing the single firm and the single industry; we only leave them out when we come, in the
third part of the chapter, to discuss the general equilibrium of atemporal production and exchange
as this kind of general equilibrium is called (explanation for this terminology must wait for
Chapters 7 and 8).
We start with the necessary notions about the theory of production and of price-taking firms.
Price-taking behaviour, which we have already assumed in the study of consumers, means the
economic agent treats prices as given parameters in her maximizations; hence a buyer (respectively,
a seller) believes that the price at which she can purchase (respectively sell) additional units of a
good is the same as the price of the previous units; therefore for a firm, revenue from sales or
expenditure on inputs are linear functions of the quantity sold or bought. A discussion of when the
price-taking assumption is legitimate, and of its connection with the notion of competition, is
provided in Part III of the chapter.
The firms we study in this chapter produce undifferentiated goods: each product is produced
by many firms and is so standardized (and unaccompanied by marketing expenses) that consumers
are indifferent as between the several producers. As a result the only problem of the price-taking
firm is how much to produce and with what combination of inputs.
PART I
PRODUCTION POSSIBILITIES SETS AND PRODUCTION FUNCTIONS
5.2.1. Imagine a world where a variety of consumption goods is produced through the use of
unproduced (original) factors, i.e. different types of land and of labour; there are no produced

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 2

means of production i.e. no capital goods, and no need to date goods no role for time, no interest
rate. We want to understand how the marginal approach determines the competitive general
equilibrium of such an economy.
There are two types of agents: consumers, with given endowments of factors and given
preferences; and firms. In the background, ensuring respect of private property and of contracts,
there must be an institutional setup, something like a state with laws, police and courts, and this
requires resources, but for the moment we neglect this aspect.
Firms have production possibility sets. A production possibility set Y is the set of all
combinations of inputs and outputs that are possible for a given firm. These combinations are called
production processes or production plans. A production process is a vector, of inputs and of
outputs, with n elements if the possible outputs and inputs total to n.
In modern general equilibrium theory, the preferred formalization of a production process is
as a vector of netputs, a vector where negative numbers indicate (net) inputs and positive numbers
indicate (net) outputs of goods (or services).
Why the specification '(net) inputs' and '(net) outputs'? Netputs are especially convenient
when one studies intertemporal equilibria with produced intermediate means of production (i.e.
capital goods). Then inputs and outputs are dated. A firm may for example consider a plan including
the production of 100 units of a circulating capital good at time t, and also the utilization of 80 of
those units at time t to obtain other products at time t+1; one says then that the planned netput of
that capital good by that firm at time t is 20; its what the firm can sell of that capital good to outside
agents according to that plan. Then the inner product of a netput vector and of the vector of input
and output (discounted) prices yields the (discounted) profit of adopting that production plan, with
negative netput entries (amounts of inputs) contributing to cost and positive netput entries (amounts
of outputs) contributing to (discounted) revenue.
In the same way as for consumer theory, inputs and outputs can also be distinguished by the
location in which they are available, and by the state of nature with which they are associated[1].
When goods are dated, it will be generally legitimate to assume that if Y includes a productive
process with netputs distinguished by their dates, (yt,..,yT), then it also includes the same sequence
with all netputs dates increased by the same number, indicating that all that matters is the lag
between inputs and outputs, and not the moment when the productive process is started;
furthermore it is natural to assume that outputs cannot precede their inputs in time.
But now we leave aside questions requiring the dating of commodities. The most acceptable
1

Cf. footnote 12 in chapter 4, and chapter 12??.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 3

interpretation of the model we are going to present in Part III is that it depicts the equilibrium of an
economy where flows (per time unit) of services of non-produced factors (types of land and types
of labour) produce flows (per time unit) of consumption goods, either with no lag (continuous-flow
production), or in production cycles of one period length with the product coming out all together at
the end of the period; and that there is no intertemporal transfer of purchasing power (no loans) and
thus no interest rate. This is called the atemporal economy but in fact it may well refer to an
economy in time, with production taking time, but with no interest rate and no loans; the
equilibrium can be conceived as the normal situation which the economy gravitates toward, with
average prices constant through time or changing sufficiently slowly for the change to be negligible.
The endowments of the economy consist therefore only of non-produced factors, and we assume
that the inputs consist only of services of these factors; the products are only consumption goods,
that are sold as they come out.
5.2.2. When the vector of outputs to be produced by a firm is given, the input requirement
set is the set of input vectors that allow the production of that vector of outputs. In the vectors of the
input requirement set the inputs are measured as positive quantities. If one assumes that some of the
quantities of inputs can be left idle, the input requirement set for a certain output vector includes all
vectors x of inputs that allow producing at least that output vector, plus all vectors xx.
Firms will generally only utilize efficient production processes. In terms of netputs we say
that yY is efficient if there is no other y'Y such that y'y and y'y; in other words it is not
possible to produce the same outputs with less of some input, or to produce more of some output
with the same inputs. If the output vector z0 is given, an input vector x (inputs being measured
now as positive quantities) is efficient if no vector xx exists such that the netput vector (z,-x)Y.
When one considers production processes that produce only one output, it is often assumed
that the economically relevant production processes that produce that output can be described by a
production function q=f(x1,,xn)=f(x) where output q is the maximum output obtainable from the
vector of inputs x=(x1,,xn), the latter measured as positive quantities. The set of input vectors that
a production function q=f(x) associates with a given output q is called the isoquant associated with
q; note that its elements need not be all efficient: if x is an efficient input vector associated with
output q, and if the addition to x of some additional amount of an input, e.g. x 1, is unable to
increase production, the new input vector x=(x1+ x1,,xn) is also part of the isoquant associated
with q. In common parlance in economics when one speaks of output obtainable from certain
inputs, one means the maximum output. The reason why the economically relevant production

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 4

processes can include some non-efficient input vectors is that if an input is free (zero price), a firm
need not limit the demand for it, so it might even demand a uselessly large amount of it.
It is sometimes assumed that in the short period the quantity of some inputs cannot be varied;
these are called fixed inputs. The short-period production function has two alternative
representations: it can include among the inputs the given quantities of the fixed factors, but for
brevity it can also describe output as a function of the sole variable inputs.
The equivalent of the production function for production processes that produce several
outputs simultaneously is a transformation function implicitly defined by an equation T(x1,
,xn;q1,...,qm)=0, where, again, inputs are measured as positive quantities[ 2]. For all inputs fixed
and all outputs but one fixed, the equation yields the maximum obtainable output of the last good,
and for all outputs and all inputs but one fixed, the equation yields the minimum necessary amount
of the last input.
5.3. When the production possibilities set Y is a set of netput vectors, some axioms that may
be postulated on it are:
1

0Y, inactivity is one possibility

2 YRn+ = 0, no production of outputs without inputs


3 YY = , production is irreversible
4 Y is convex
5 Y is bounded above
6

for any good i, and any positive scalar q, the vector y=(0,...,0,q i,0,...0) is Y (free

disposal).
Axioms 1, 2 and 3 are unproblematic and are assumed in what follows. Axiom 4 implies
perfect divisibility of all inputs and outputs; its connection with returns to scale will be discussed
presently. Axiom 5 is convenient in a first stage of some mathematical proofs, and it is justified by
referring to limited factor endowments that do not allow producing more than certain maximum
quantities, but this means mixing up endowments with technologies, so it will not be assumed in
what follows. Axiom 6 postulates that for each good there is available a process that uses that good
alone as an input and produces nothing, so one can always get rid of any amount of any good
without any cost; it is called the free disposal assumption. When the free disposal assumption is
made, then any firm can couple any production process with free disposal processes, and the result

Exercise: Obtain the transformation-function representation of a production function.

f petri

Adv Micro

chapter 5 firms and productionGE

p. 5

10/03/2016

is a property or assumption sometimes called monotonicity of the production possibilities set(3):


if yY, then y' such that y'y is also Y,
because y' is a process that employs at least as much of each input as y ( 4), and produces not more
of each output than y, and it is always possible to obtain y' from y by use of free disposal processes.

q=f(x)
q

Fig. 5.1. Production possibilities set resulting from a production function q=f(x) plus free disposal.

Free disposal may appear a questionable assumption in many situations (it is often costly to
get rid of goods, or to prevent some output from coming out), but it can be argued to be
fundamentally harmless. An ability costlessly to get rid of excess inputs is not needed as long as
these inputs have a positive cost, because then firms will not buy them to start with, so a free
disposal assumption of excess costly inputs is superfluous but then also harmless; and if inputs are
costless and with a negative marginal product, they can be left idle and all one needs is to
distinguish the technological from the economic production method (as was done in 3.3.4). As to
undesired outputs, if they must not be produced it can be assumed that, whatever disposal process is
necessary in order not to produce them (or in order to dispose of them), its inputs (and costs) are
included among the inputs (and costs) of the desired output. When there is a choice about how
much to produce of an undesired side product, then a formalization can usually be found in which
the abatement of its production is counted as an output, and then the problem becomes again the
3

Some authors call monotonicity the following slightly different assumption: let q stand for a vector of
outputs and let V(q) stand for the set of input vectors (measured as positive quantities) that allow the
production of at least q; if an input vector x is in V(q), and x'x, then x'V(q).
4
Remember that inputs are negative numbers, so a greater use of an input means a greater absolute value
of a negative number i.e., algebraically, a smaller number indicating input use.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 6

usual problem of cost minimization and profit maximization, this time with joint production. Thus
the free disposal assumption is not really necessary but it is also generally harmless.

5.4. Axiom 4 stipulates that if yY and yY, then for 0<a<1 all netputs y=ay+(1a)y are
also Y; strict convexity of Y means that if y and y are efficient, then y is not efficient. Convexity
requires divisibility of inputs and of outputs.
Exercise: ) Suppose a single output q is produced by two inputs x 1 and x2 according to the
production function q=ax1+bx2, a,b>0. Draw some isoquants and prove that the production
possibility set is convex.
) Suppose two outputs z1 and z2 are jointly produced by a single input x with transformation
function z12+z22x=0. Prove that the production possibility set is convex and that for fixed x the
locus of possible efficient combinations of z1 and z2 (the transformation curve) is concave.
) Suppose a single output q is produced by two inputs x1 and x2 according to the production
function q=x12+x22; draw some isoquants and prove that the production possibility set is not convex.
An assessment of Axiom 4 requires a discussion of returns to scale, a notion given different
meanings in the history of economic thought. The term originally referred to the dependence of
returns, that is of net earnings, on the scale of normal output. The usual meaning in modern
analyses is technological returns (in terms of output) to the scale of input employments, or
degree of homogeneity of the production function when all inputs are variable. A different meaning
that we will discuss later (5.21??) is firm returns in terms of output from the scale of total cost, a
notion especially helpful in the presence of indivisibilities; this definition of returns to scale takes
input prices as fixed. A still different meaning is industry returns in terms of output from total
industry costs, a very complex Marshallian notion, historically important in partial equilibrium
analysis, that takes into account (i) returns to scale at the firm level, (ii) what happens to the prices
of the industry's inputs, and (iii) external effects, when the industry's output changes( 5); this notion
of returns (in this meaning the appendage to scale is usually absent) will be discussed when we
arrive at the industry's supply curve.
Production functions allow a quick definition of technological returns to scale. Let x be the
vector of inputs to a production function f(x). We say that f(x) exhibits
constant returns to scale (CRS) if f(tx)=tf(x) for t>0;
increasing returns to scale (IRS) if f(tx)>tf(x) for t>1;
decreasing returns to scale (DRS) if f(tx)<tf(x) for t>1.
In words, constant returns to scale means that doubling all inputs doubles output; increasing
5

This is the notion of returns in the famous controversy originated by Piero Sraffa with his 1925 and 1926
articles.

f petri

Adv Micro

chapter 5 firms and productionGE

p. 7

10/03/2016

returns, that doubling all inputs more than doubles output; decreasing returns, that doubling all
inputs yields less than double output. Returns to scale can be variable: for example a production
function might exhibit returns to scale that are increasing at first, then constant, then decreasing.
Local returns to scale for a given vector of inputs x are ascertained by checking which one of the
three inequalities holds for small variations of t around 1 (cf. 5.6.1 below).
Technical returns to scale can also be defined in terms of netputs and of frontier of the
production possibility set F(Y). I give a taste. If netput yF(Y) implies tyF(Y), with t any positive
scalar in a neighbourhood of 1, we say that Y exhibits local CRS. There are increasing returns to
scale at least locally, if there is some yF(Y) and some scalar t>1, such that tyY and ty F(Y),
that is, yY and ty such that yty.
Note that if there are increasing returns to scale at least locally, Y is not convex.
Proof: by contradiction. By Axiom 1, y=(0,...,0)Y, so convexity requires that for any yY also t'y
with 0<t'<1 is in Y; then consider netputs yF(Y), ty and y'ty of the above definition of locally increasing
returns in terms of netputs: if Y is convex, y"=t'y' with 0<t'<1 is in Y; choose t=1/t, then y"y and y"y, so y
cannot be in F(Y), contradicting the assumptions; hence Y cannot be convex.

This shows that Axiom 4 excludes increasing returns to scale. Now, increasing returns to scale
are argued by many economists to be often present in reality; therefore Axiom 4 is a restrictive
assumption; in spite of this, we will generally assume it because it is necessary for the standard
theory of general equilibrium, whose presentation is now our main aim.
As an Exercise, you are asked to prove that a convex production possibility set with a single
output implies a quasiconcave production function. (Hint: prove first that the input requirement set
is convex. The definition of quasiconcavity was given in Ch. 4, 4.5)
5.5. Some clarification on the notions of productive process, inputs, indivisibilities can be
useful. A production process usually starts with certain inputs and lasts some time, during which
time the inputs become transformed into a series of other things before reaching the final form of
the saleable product. There is therefore some arbitrariness in the description of a production
process; one might always break it into two successive processes, the first one producing
intermediate products which are the inputs to the second one. For example, the production of a sofa
from wood planks, springs and fabric might be broken down into a series of stages, each one
producing an incomplete sofa closer and closer to completion. The intermediate products might
themselves be priced, as made evident by the fact that sometimes the process originally performed

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 8

in its entirety by a firm is broken down into processes performed by different firms, for example
production of a car has historically been more and more broken down among separate firms which
produce the seats, the windshield, the tires, the brakes, etc., parts which are finally assembled into a
car by a different firm. Which inputs are initial inputs that appear in the production function, and
which are intermediate stages that are only implicitly considered (but might become explicit if the
process were broken down into successive subprocesses performed by different firms), depends on
the organization of production and, from the theorist's point of view, is largely arbitrary. This makes
the notion of intermediate input ambiguous. Flour is an intermediate input in the production of
bread from wheat, but it is an initial input in the production of bread from flour.
The term 'intermediate input' is sometimes used as synonym of circulating capital good,
which means a capital good that is fully destroyed (one might say, that disappears into the product)
in a single utilization. These two notions are best kept distinct; it is best to reserve the term
'intermediate inputs' for the products produced and re-utilized inside a production process, and
therefore not appearing among the inputs of the production function (they must not be paid for).
The inputs appearing in the production function are all the 'initial' inputs that must be paid for( 6);
when they are capital goods, they can be either circulating capital goods, or durable capital goods;
in the latter case they reappear among the outputs, obviously with the alterations caused by
utilization(7).
We will generally assume that all inputs and outputs are perfectly divisible, i.e. can be
represented by continuous variables. This is a good approximation for land and for labour time[ 8],
but it is obviously unrealistic for capital goods and for many products; however, if the analysis deals
with big quantities the assumption may still be acceptable if the indivisibilities are small relative to
total input use or total output.
Much more debatable is the assumption of differentiability of the production function, and
here we come to a very important issue. In most industries a different productive process requires,
not different proportions among the same capital goods, but different capital goods, and for each
ensemble of capital goods a rather rigid labour input. The amount of labour services needed to
assemble a car, for example, is rather strictly determined by how mechanized the production
6

In the case of intertemporal production functions, 'inital' inputs are not necessarily entering the process
at the same date, the term 'initial' must be interpreted as meaning not produced by the production process
itself.
7
It is also possible to imagine cases in which a durable capital good is an intermediate good, because the
production process considered lasts a number of periods, and produces itself a durable capital good which is
then entirely utilized during the remainder of the process.
8
Labour time is, physically speaking, perfectly divisible, but regulations often limit this divisibility; for
example a firm may be obliged to hire full-time labour only. Even then, if the firm employs 1000 labourers a
perfect divisibility assumption may be an acceptable approximation.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 9

process is; and there is no substitutability between the parts to be assembled, a car needs exactly one
engine, five tyres, etc., each one of these parts needing in turn rigid quantities of material inputs and
rigid amounts of labour determined by the machinery used. There generally is no, or nearly no,
variability of proportions among the same inputs. Differentiability is a very unrealistic assumption
when all inputs are physically specified and production uses capital goods. But then how come the
differentiability assumption is so widely accepted? The reason would appear to be a historical one,
namely the Marshallian habit, shared by a majority of economists of his generation and of the next
one, and still widespread to-day, to describe production functions as combinations of labour and
land with capital treated as a single factor K, measured as an amount of value; implicitly, the prices
of the several capital goods are treated as given, and the production function is determined as
follows: the firm is assumed to determine, for each given vector of non-capital inputs and each
given K, the vector of capital goods of value K that maximizes production; thus, given the amounts
of labour and land, small increases of K can well imply a totally different vector of capital goods;
along a total factor productivity curve, capital so conceived changes not only in quantity (an amount
of exchange value) but also in form (physical composition)[ 9]. With such a specification of the
capital input, the assumption of smooth variability of proportions between the inputs acquires much
greater plausibility: increasing only the number of tyres certainly does not increase the output of
cars; on the contrary if what can be increased is the value of the capital goods used, which can be
associated with a change in the capital goods used, then it is likely that a way can be found to use
the capital increase so as to increase the number of cars produced by a given number of workers.
Unfortunately in more recent times this origin of the use of smooth production functions and of
nicely decreasing marginal product curves appears to have been lost sight of, with the result that in
modern microeconomic theory and general equilibrium theory this treatment of capital has gone out
of fashion[10], inputs are all measured in technical units, the several capital goods are each treated as
a separate factor; but continuity and differentiability of production functions are still commonly
assumed. I must follow now this usual practice in order to introduce the readers to this literature;
but it is important to realize that we are encountering here an instance of survival, in a context no
longer justifying them, of assumptions that were originally justified by the treatment of capital as a
single value factor of variable form. (More on this in chapter 7.)
9

Analogously an isoquant in terms of, say, labour and K indicates, for each level of labour, the minimum
value of capital required to produce the given output; again, small movements along the isoquant can well
mean a passage to a production method requiring very different capital goods.
10
It is not difficult to understand why. The treatment of capital as a quantity of value in the firms
production function is only legitimate if the prices of capital goods are given. But these prices cannot be
taken as given when the purpose of the analysis is not a partial-equilibrium one but rather the determination
of income distribution which affects relative prices.

f petri

Adv Micro

chapter 5 firms and productionGE

p. 10

10/03/2016

So unless differently specified, we assume production functions to be continuous and


differentiable functions of perfectly divisible inputs.
5.6. Returns to scale.
5.6.1. In Ch. 4 we met homogeneous functions (cf. Ch. 4 fn. 35??). CRS implies that the
production function is homogeneous of degree 1 [11]. If the production function is homogeneous of
degree k>1, it has increasing returns to scale; if homogeneous of degree k<1 it has decreasing
returns to scale. Even when a function f(x) is not homogeneous, we can determine its local degree
of homogeneity by increasing all independent variables by a common small percentage, say, .1%,
and observing whether f(x) increases by more or less than .1%. This indicates the local
technological returns to scale (of course, under perfect divisibility of all inputs). Their most
widely used measure is the scale elasticity of output (or simply elasticity of scale) with respect to
inputs. Let x be the vector of inputs in an initial situation with q=f(x), and consider f(tx) with t>0;
the scalar t measures scale, and the scale elasticity of output in q=f(x) is defined as
e = [df(tx)/f(tx)]/(dt/t) = (q/t) (t/q) = ln q / ln t evaluated in t=1, with x fixed.
Note that this definition does not require differentiability of f(x), it only requires
differentiability of f(tx) with respect to t, i.e. with respect to proportional variations of all inputs,
therefore it is applicable to production functions where inputs, or some of them, are perfect
complements. But if f(x) is differentiable, then by the derivative rule of a function of function it is e
=

1
n
[xif/xi]. By Euler's theorem on homogeneous functions, if the production function
f ( x ) i 1

has constant returns to scale then i(xif/xi)=f(x) so e=1. According as e is equal to, more than, or
less than, 1, the production function exhibits locally constant, locally increasing, or locally
11

Two properties of homogeneous functions (already briefly indicated in footnote 35 of ch. 4) are of great
relevance for CRS production functions. A continuous function f(x1,...,xn) is homogeneous of degree k if,
with t a positive scalar, f(tx 1,...,txn) = tkf(x1,...,xn). The first property is that if a function homogeneous of
degree k is differentiable, then its partial derivatives are homogeneous of degree k1; the proof is by
differentiating both sides of f(tx 1,...,txn) = tkf(x1,...,xn) with respect to xi, and indicating with f(tx)/(tx i) the
partial derivative of f relative to the i-th independent variable, calculated in tx: one obtains
f (tx )
f ( x)
t
tk
which implies that the partial derivative calculated in tx is the partial derivative
(txi )
xi
calculated in x multiplied by tk1; thus if f is a CRS production function its marginal products are
homogeneous of degree zero i.e. depend only on factor ratios (hence the expansion path, the locus of
tangencies between isoquants and isocosts, is a ray from the origin). For the second property, differentiate
n

both sides of f(tx1,...,txn)=t f(x1,...,xn) with respect to t, obtaining

f (tx)

( (tx ) x ) kt
i 1

k 1

f ( x) , and then

set t=1: one obtains i(xif/xi)=kf(x), a result sometimes called Eulers theorem for homogeneous
functions; for CRS production functions it is k=1, hence if each factor is paid its physical marginal product
the payment to factors exhausts the product, a result also called the product exhaustion theorem.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 11

decreasing returns to scale.(12)


If all relevant inputs are taken into account in the specification of the production function,
then an exact doubling of all inputs allows the exact replication of the same plant twice and
therefore it should permit at least a doubling of production: for integer t>1, f(tx)tf(x), i.e. there
must be at least constant returns to scale. Why do I say 'at least'? Because there may be some
indivisibility of the technological process, which was not fully exploited at the original scale, and
can be better exploited at a bigger scale, allowing for increasing returns to scale. Thus consider the
production function of a firm that extracts and transports oil and also produces all the pipes for the
pipeline. The pipes are intermediate products in the overall production process and appear neither
among the inputs nor among the outputs of the oil production-and-transport function. The inputs
include, for example, the steel needed to make the pipes. Now, up to a point the carrying capacity of
pipes increases more than proportionally with the increase in the steel utilized to make pipes,
because the steel utilized is within certain limits roughly proportional to the diameter of the pipe
but the carrying capacity is proportional to the square of the diameter. Doubling the amount of
produced and transported oil may then require pipes of double diameter that use less than double the
steel, with less than double the cost. A similar issue arises with tanks. This example shows that,
when the production function reflects vertically integrated production processes which include the
production and utilization of intermediate goods, a doubling of all inputs need not correspond to a
replication of the same production method twice, and a doubling of output need not require a
doubling of inputs(13). Still, the replication of the same production method twice (the building of a
second plant identical to the first one) is always possible and therefore returns to scale for integer
t>1 are at least constant. What allows the existence of increasing technical returns to scale is the
existence of indivisibilities (either of goods, or of processes) which are not fully taken advantage of
at small scales of production.
The result f(tx)tf(x) need not hold for fractional increases in scale, again owing to
indivisibilities. It may be impossible to increase all inputs by, say, 30% if some inputs are
indivisible; or the indivisibilities can be in the production process, e.g. the production process might
be vertically integrated and include the internal production and utilization of a large indivisible
12

The extension of these definitions to transformation functions is left to the reader (rays of outputs will
replace the single output; the sole complication is that, since with transformation functions generally a given
vector of inputs does not uniquely determine the vector of outputs, it is now possible to imagine cases where
returns to scale differ according to which output ray one considers).
13
The same can happen if capital goods are aggregated into the single factor (value of) capital; then
double the capital and double the non-capital inputs need not correspond to purchasing twice as many of the
same capital goods, it may for example mean the use of a different fixed plant that costs twice as much but
allows more than double the production. Cf. below, 5.20, the notion of scale returns to cost.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 12

capital good. Below I will assume that this problem is of minor importance, we will see that when
one considers an entire industry with free entry it is generally plausible to treat the industry output
as coming from a CRS production function even when at the firm level there are relevant
indivisibilities
5.6.2. Plausibly, even if indivisibilities cause increasing returns at small scales of production,
at each stage of technical knowledge there is a finite production scale beyond which returns to scale
are no longer increasing: we can call it minimum optimal scale. Beyond a certain dimension larger
pipes and tanks are no longer convenient because they require special reinforcing structures. In a
perfectly competitive industry with free entry firms must produce at minimum average cost and
therefore will tend to adopt plants of at least minimum optimal scale, possibly several of them. As
long as this efficient scale of production is small relative to total industry output, it will be
approximately true that the aggregate of firms composing an industry can be seen as having a
production function exhibiting constant returns to scale where the constancy is generated by
variations in the number of identical efficient plants. For this reason, below we generally assume
constant technical returns to scale for industries. This assumption may appear valid for only a very
restricted set of industries, given the observable tendency of firms in many industries to grow as
large as they can. But the advantages of size can be due to many other reasons besides increasing
technical returns to scale, reasons that do not concern us now( 14). Anyway globalization has
increased competition in many industries where minimum efficient plant size is very large; for
example, one can buy cars produced all over the world, Korean cars compete with German and
USA cars, thus even for industries where minimum plant size is very large there often appears to be
sufficient competition for the assumption of price equal to average cost to be broadly acceptable for
many analyses.
5.6.3. What about decreasing technical returns to scale? They are difficult to defend if the
inputs appearing in the production function are really all relevant inputs, because then, as argued
above, identical duplication of plant and process should yield double output. Decreasing returns to
scale can be admitted only if some relevant inputs are fixed in quantity and do not appear in the
production function. This is the case in short-period analysis when only variable inputs are made to
appear in the production function; but for long-period analysis, the sole way which appears
14

We only mention at this stage the possible advantages in terms of funds for marketing expenditures, or
for research and development (R&D) expenditures, and the possibility to obtain discounts on some input
prices .

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 13

acceptable to arrive at decreasing returns to scale is to postulate a greater difficulty of control and
co-ordination over the performance of subordinates[15]. It might be answered that if one duplicates
not only the plant but also the managers, there is no reason why there should be a greater difficulty
of control. The objection to this is that managers must be controlled too, and the owner of the firm
or top manager may have greater difficulty in avoiding subordinate managers shirking or
embezzling when she must control many managers; furthermore, the information that the top
manager must process increases, and its reliability decreases, as it must pass through a greater
number of bureaucratic layers, making correct decisions more difficult. Whether this objection is
sufficient to justify an optimal size of price-taking firms is still an object of disagreement among
theoreticians[16]. But for the purposes of value theory what is important is the behaviour of
industries, and then the moment one admits free entry the difficulties of control may explain why
individual firms are limited in size, but industry output can be varied by variation in the number of
firms; so at the industry level there will be CRS anyway in long-period analysis as long as one can
assume that minimum efficient size is small relative to total demand.
5.6.4. Finally a word on the difference between production process and production method
when there are CRS. Any netput vector in the production possibility set is a production process.
With CRS, if (x,q) is the netput representation of a single-output production process then (tx,tq)
with t>0 is a production process too, and if the first one is efficient so is the second. One means then
by production method a set of ratios between inputs and outputs: a vector (x,q)Y and a vector
(tx,tq), t1, are considered two production processes representing the same method activated at
two different activity levels.

15

Decreasing economic returns to scale (i.e. profits that increase less than proportionately with output)
can arise because an output increase raises the rentals of some inputs; but this is best kept separate from the
issue of technical returns to scale.
16
For example, Scherer and Ross p. 106 agree with the traditional position well represented by EAG
Robinson ??ref on the greater difficulty of control as a cause of ultimately decreasing returns to scale, and
their arguments on the difficulty of control increasing with size are prima facie convincing. Edith Penrose on
the contrary wrote: We do not know how effective the decentralization of authority can be as a means of
keeping costs per unit of output from rising as a firm expands. Reliable empirical evidence does not exist and
all studies of the matter are inconclusive, but there is no evidence that a large decentralized concern requires
supermen to run it....Neither is there significant evidence that the ability to fill the higher administrative
positions is excessively rare or that the demands on the men occupying these positions exceed their ability to
cope with them effectively. E. Penrose, Limits to the growth and size of firms, AER 1955, vol. 45 (2),
May, 531-43, p. 542. Cases supporting Penrose, for example MacDonalds or CocaCola or United Fruits,
easily come to mind. Perhaps in many cases the advantages of increasing size are so great (especially when
one considers the scale economies in marketing, R&D, transport costs, employee training, etc.), as to more
than counterbalance the increasing difficulties of co-ordination; also, tying decentralized managers pay to
results may be often sufficient to motivate them.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 14

5.7. Activity analysis


The distinction between production process and production method comes useful when, given
the general lack of smooth substitutability between different capital goods, one esteems that it is a
closer approximation to reality to represent the production possibilities of a firm as including only a
finite number of CRS fixed-proportions production methods, which can be activated at any activity
level (we still assume divisibility); then the firm can also activate all the linear combinations of
processes achievable from those methods. In this case the production possibility set is a polyhedral
convex cone, with the efficient methods as edges and the vertex in the origin; and the isoquants are
said to be of the activity-analysis type. These isoquants appeared in Appendix 2 to chapter 3, cf.
Fig. 3.15?? which is reproduced here for ease of reference as Fig. 5.1bis??.
Availability of a single method implies L-shaped isoquants, corresponding to a production
function of type q=min{x1,x2} in the two-inputs case (represented in red in Fig. 5.1bis) ; this is
called a Leontief technology.

f petri

Adv Micro

land

chapter 5 firms and productionGE

10/03/2016

p. 15

method 1

method 2
isocost

A
B'
method 3
B
C

labour

Fig. 5.1bis (Fig. 3.15). Activity-analysis isoquant with three alternative methods. The red
broken line is the isoquant associated with method 2 alone.

f petri

Adv Micro

chapter 5 firms and productionGE

p. 16

10/03/2016

5.8. Marginal product, isoquant, transformation curve


5.8.1. We assume now that the firm produces a single output and its efficient production
possibilities are represented by a production function f(x) which is twice continuously differentiable
(that is, it has partial derivatives that are differentiable, and their partial derivatives are continuous).
Then we can define the marginal product of factor xi as MPif/xi; and the locus of input
combinations which yield an assigned level of output, the isoquant, is a differentiable surface (in Rn
if there are n inputs) by the implicit function theorem.
If some factors are fixed, the locus of combinations of the remaining inputs which yield the
given output is called a restricted isoquant or short-period isoquant (note that it may be empty).

x2

q2
MRT
isoquant
transformation curve

TRS

x1

q1

Fig. 5.2 Standard isoquant and standard transformation curve in R2.


Suppose we consider an isoquant restricted to the two inputs i and j. The condition that
production be equal to an assigned quantity q^ and that all inputs be given except for xi and xj:
f(x1,...,xi,...,xj,...,xn) = q^ , with xh given except for h=i,j
implicitly makes xj a function of xi. The slope of this function xj=xj(xi), i.e. the slope of the
restricted isoquant when xi is treated as the independent and xj as the dependent variable, is the
equivalent in production theory of the consumer's marginal rate of substitution; it is called the
technical rate of substitution of factor j for factor i, to be indicated as TRS ji; by the derivative rule
for differentiable implicit functions the TRS ji is given by TRSji dxj/dxi = (f/xi)/(f/xj) =

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 17

MPi/MPj, the well-known result of first-year textbooks. The TRS is a negative quantity if both
marginal products are positive; economists sometimes speak of the TRS as a positive quantity, then
they are implicitly referring to its absolute value.
In Chapter 3 we saw that marginal products need not be positive, and therefore technological
isoquants need not be downward-sloping. (The reader is invited to re-read in Chapter 3 the
distinction between technological and economic isoquants.) We supply now a numerical example.
Assume
q = (x1x2 .8x12 .2x22)1/2
The reader can check that this production function has CRS. Marginal products are given by
q/x1=

1
1
( x1 .4x2); provided it is q>0, which will be the case as
1 / 2 (x2 1.6x1), q/x2 =
2q
2q1 / 2

long as 1<x2/x1<4, the marginal product of factor 1 is positive as long as x 2>1.6x1, and the marginal
product of factor 2 is positive as long as x 2<2.5x1. Hence both marginal products are positive and
isoquants are negatively sloped only as long as 1.6<x 2/x1<2.5. Outside this fairly restricted range of
factor proportions, one of the two marginal products is zero or negative, implying an upwardsloping isoquant.
For a differentiable transformation function T(x1,,xn;q1,...,qm)=0, since there is more than
one output, an input has several marginal products, one for each output (Exercise: write the
expression for them from the rule of derivation of implicit functions). The determination of the TRS
between two inputs requires that all other inputs and all outputs be fixed. If all inputs, and all
outputs but two, are fixed, the locus of efficient combinations of the two remaining outputs (where
efficiency means that one output is maximized when the other one is given) is called a
transformation curve(17) and its slope is called the marginal rate of transformation, MRT, and is
given by MRTji dqj/dqi = (T/qi)/(T/qj).
With constant returns to scale, marginal products only depend on factor proportions and not
on the scale of production. This is because the partial derivatives of a function homogeneous of
degree one are homogeneous of degree zero, i.e. do not vary for equiproportional variations of all
independent variables. Thus along a ray from the origin all isoquants have the same TRS.
5.8.2. The reader will have noticed the resemblance between isoquants and indifference
curves, and between marginal products and marginal utilities. Thus, for example, convex isoquants
imply that the production function is quasiconcave. However, there are some differences.
17

It is also called a production possibility frontier for the firm (in order to distinguish it from the
production possibility frontier for the entire economy).

f petri

Adv Micro

chapter 5 firms and productionGE

p. 18

10/03/2016

A first difference reflects the cardinal character of production possibilities versus the ordinal
nature of preferences. Any strictly increasing transformation of an increasing utility function
represents the same preferences; on the contrary, any transformation of a production function alters
the production possibilities. Therefore, quasiconcavity is not as important a notion as in utility
theory: in production theory we also want to know returns to scale, which are arbitrary in ordinal
utility theory.
Also, as long as CRS and perfect divisibility are assumed, an isoquant is necessarily convex.
This is because under these assumptions if x and x' are two input vectors belonging to the isoquant
associated with output q, then the firm's production possibilities set Y includes all netput vectors (
tx,tq) and (tx',tq) for any t>0, so the firm can produce q by using any convex combination of the
two processes (ax,aq)+((1a)x',(1a)q), where 0 a 1, and therefore the isoquant cannot go
above the segment connecting any two points of it, cf. Fig. 5.3b. On the contrary, in utility theory it
would be a most special case if all linear combinations of two consumption bundles on the same
indifference curve yielded the same utility (the consumption goods would be locally perfect
substitutes).

x2

x2'

B
C

x1'

Figure 5.3a Short-period restricted isoquant with


incompatibility between the two variable factors.

x1
Fig. 5.3b. With divisibility and CRS, isoquants
cannot be strictly concave, point B of input use
allows the same production as A or C through a
linear combination of the activities (processes)
corresponding to A and C.

For the same reason, under divisibility and CRS, production functions are concave because,
given any two efficient vectors (-x',f(x')) and (-x",f(x")), it is possible to produce at least any linear
combination of these vectors and therefore f(ax'+(1-a)x")af(x')+(1-a)f(x").
This is no longer necessarily the case with restricted, or short-period, isoquants. For example,
with a given machine for chemical reactions, it might be possible to produce a given output by
using a certain chemical process, or another chemical process based on different components, but

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 19

any joint use (or even alternate use) of the two processes on the same machine might be impossible
because the residues of the components of one process would react with the components of the
other process and destroy the machine. In this case, assuming free disposal, the restricted isoquant
would have the shape shown in Figure 5.3a. The meaning of such an isoquant is that, given the
other inputs, the desired output can be produced either with the quantity x 1' of input 1, or with the
quantity x2' of input 2, but not with both. However, cases such as this one can be considered highly
unusual and therefore we leave them aside.
5.9. Profit maximization and WAPM
The most common assumption about the behaviour of firms is that they aim to maximize
profit. The meaning of 'profit' here and in the entire chapter is the marginalist one, i.e. in this
chapter 'profit' stands for what is left to the entrepreneur (the owner of the firm) after paying all
costs including interest on capital advances(18). Even when the apparent aim of the firm is another
one, e.g. sales maximization, a case can usually be made that this does not entail significantly
different choices from the ones aimed at maximizing long-run profit. More relevant is the
possibility of inefficiency, but as long as management strives for profit maximization the fact that
the goal is only imperfectly realized does not alter the broad pattern of industry behaviour. For
example, the tendency to invest more in the industries that offer better profitability prospects will
still exist even if on average management is not very good at minimizing costs. The first-year
textbook short-period supply curve of the firm, coinciding with the marginal cost curve, most
probably remains upward-sloping and therefore suggests an increase in output if the product price
rises, even if marginal cost reflects inefficiencies. And the occasional episodes of managers
pursuing strategies of personal enrichment at the expense of the profitability of their firm usually
end up rather quickly in the disappearance of the firm, whose market shares are absorbed by better
run firms. We accept profit maximization as broadly valid as a survival condition, especially in
competitive environments[19]. A monopolist entrepreneur not threatened by takeovers might indulge
in other aims, e.g. to have political influence, or play golf, or be generous toward employees; firms
in competitive environments and under threat of takeovers end up being taken over or going
18

And including an allowance for risk too; but we are not considering risk for the moment. (The reader
may be surprised by our mentioning interest here; but as we said, the treatment of firms aims to be general,
the assumption that there are no capital goods and no interest will only be made when we come to the
general equilibrium model to be studied at the end of this chapter.)
19
From the voluminous and often inconsistent evidence, it appears that the profit maximization assumption
at least provides a good first approximation in describing business behavior. Deviations, both intended and
inadvertent, undoubtedly exist in abundance, but they are kept within more or less narrow bounds by
competitive pressures, the self-interest of stock-owning managers, and the threat of managerial displacement
by important outside shareholders or takeovers. Scherer and Ross p. 52.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 20

bankrupt if they do not struggle to survive, and this requires profit maximization.
In what follows the prices of outputs are indicated as pj, j=1,...,m; the rentals (prices of the
services[20]) of inputs are vi, i=1,...,n. The vectors p=(p1,...,pm) and v=(v1,...,vn) are to be conceived
as row vectors unless otherwise indicated, and the vectors of inputs x and of outputs q of a
production process as column vectors. At given prices (p,v), for each production process with netput
(x,q) total cost is vx and profit is :=pqvx. With the netput notation y=(x,q), one can put all
prices of inputs as well as of outputs into a single vector; if one uses the symbol P=(v,p) for this
encompassing row vector, then profit can be represented more compactly as :=Py. If inputs and
outputs are paid at different dates, the several prices of the previous expression are to be intended as
discounted or capitalized to a common date: for the moment, the date at which output is sold. If, as I
will mostly assume in what follows, the firm produces a single output, then vector q has only one
nonzero element.
A result, somewhat analogous to the weak axiom of revealed preferences, that derives
immediately from profit maximization is the following: if at prices P=(v,p) the firm chooses a
netput vector y, the profit must be at least as great as with any other netput in Y, i.e. PyPy, y;
This is sometimes called the Weak Axiom of Profit Maximization, WAPM. It has the following
implication: if netput y is chosen at prices P and netput y* is chosen at prices P*, it must be
PyPy* and P*y*P*y>; re-write these inequalities as P(yy*)0, P*(yy*)0 and add
them to get (PP*)(yy*)0, which is often written
Py0.
In words: the dot product of the vector of price changes and the vector of netput changes is
non-negative and generally positive; geometrically, the two vectors form an acute angle. The
interpretation requires to remember that inputs are negative numbers in the netput representation.
Thus if price i is the only one to change and it increases, the WAPM implies that netput i must
increase or at least not decrease: if netput i is an input, an algebraic increase of its negative quantity
means a smaller utilization of that input.
5.10. Optimal employment of a factor
Let us first consider the decision of how much to employ of a single input, when the
quantities of the other inputs are given. The firm is competitive, i.e. price taker, and produces a
single output. The production function is q=f(x), differentiable. Profit is =pqvx=pf(x)vx where q
20

As pointed out in Ch. 3, it is preferable to speak of input rentals to mean price of the services of inputs;
but input price is so often used to mean input rental, that in this chapter we use the two terms
interchangeably.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 21

is output (a scalar), p its price, x the vector of inputs (positive quantities). Interior maximization of
with respect to xi under xi>0 requires the first-order condition pf/xi vi = 0, i.e. the equality
between marginal revenue product of the factor and 'price' (i.e. rental) of the factor, where the
marginal revenue product of a factor is the derivative of revenue pq with respect to the employment
of the factor, i.e. (under price-taking) pf/xi. With the usual symbols:
pMPi=vi.
This can also be expressed as equality between marginal product of the factor, and real rental
of the factor measured in terms of the product, MPi=vi/p.
The second-order condition is that the marginal revenue product must be decreasing in x i. The
increase in profit if the firm employs one more small unit of factor i is pMP ivi, and if it is positive,
or if it becomes positive for further increases of the factor, the firm finds it convenient to expand the
use of the factor, so the optimal level of factor employment must be where one more unit of the
factor no longer increases profit and further units would only make things worse.
Careful: there may be no positive solution to this maximization problem; in other words, it
may happen that no positive value of x i, however small, avoids a marginal revenue product inferior
to the given rental. In this case, since it must be xi 0, the firm reaches a 'corner solution' with xi=0
and pMPi vi. It is possible, although a fluke, that at xi=0 it is pMPi=vi.
5.11. Cost minimization
5.11.1. If two variable factors i and j are both demanded in positive amounts, then pMP i=vi
and pMPj=vj imply the well-known condition MPi/MPj=vi/vj; however, this last equality can be
satisfied when the two other ones are not, and this will mean, as we now show, that a different
problem is being solved: cost minimization.
A necessary condition for profits to be maximized is that the total cost of producing the profitmaximizing output be minimized. Profit maximization can be achieved in two steps: first, for each
level of output, minimize cost, and find how this minimized cost varies with output, i.e. find the
cost function; second, maximize profit by finding the level of output that maximizes the difference
between revenue, and the minimized cost.
The cost function is the minimum-value function
c(v,q) = min vx s.t. f(x)q .
This will be a short-run cost function if some factors cannot be varied; usually the other
ones, called variable factors, are the sole factors that are made to appear in x.
Assume there is a unique cost-minimizing solution x for each given (v,q). Let x=H(v,q) be the

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 22

vector function indicating how the solution changes with v and q; H(v,q) is a vector function called
the conditional input demand function. Then c(v,q)=vH(v,q). The word 'conditional' derives from
the fact that these are the input demands conditional on the level of output.
There is a strict similarity between the cost function in production theory and the expenditure
function in consumer theory[ 21], and between the firm's conditional input demand function and the
consumer's compensated (or Hicksian) demand function: mathematically they are just the same
thing (this is why we use the same symbol H to indicate the conditional factor demand function
too). Therefore we need not prove the properties we now list because the proofs are the same as for
the expenditure function, cf. chapter 4.
As long as f(x) is continuous and x is such that f(x) is strictly increasing in a neighbourhood
of x, the cost function has the following properties:
(1) c(v,q) is nondecreasing in vi
(2) c(v,q) is homogeneous of degree 1 in v
(3) c(v,q) is continuous in vi, for v>>0.
(4) c(v,q) is strictly increasing in q as long as q>>0
(5) c(v,q) is concave in vi.
If the production function is continuous we can replace the constraint f(x)q with the
constraint f(x)=q; if it is also differentiable, for interior solutions (x>>0) we can use the Lagrangian
approach with equality constraint (the Kuhn-Tucker conditions are the more general necessary firstorder conditions). Formulating the problem as one of maximization of c(v,q), the Lagrangian
function is vx(qf(x)) where q and v are given(22). The first-order conditions for an interior
solution yield
vi = f/xi,

i=1,...,n

from which one derives the well-known condition


vi/vj = MPi/MPj .
This is interpretable geometrically. Assume all input levels apart from those of inputs i and j
to be given and to cause a cost B= si,j(vsxs). For each given total cost C, the expression
vixi+vjxj=CB makes xj a linear function of xi. This function is called a restricted (two-dimensional)
21

Indeed a number of economists call cost function the consumers expenditure function.
In the consumer maximization problem we write the constraint in the Lagrangian function as (pxm),
here we write it as (qf(x)): the difference derives from the fact that, in order to obtain a non-negative
value for the Lagrange multipliers in Kuhn-Tucker theory, the constraint (the expression in parenthesis) must
be written in such a way that it is constrained to be non-positive (if there is a minus sign before the ): in the
case of the consumer, the constraint is mpx, in the case of the firm it is f(x)q. Another way of putting the
thing is, that in the consumer problem a relaxation of the constraint requires an increase of m, in the cost
minimization problem a relaxation of the constraint requires a decrease of q.
22

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 23

isocost. Its geometrical representation is a downward-sloping straight line in (x i,xj)-space, with


slope equal to vi/vj and intercepts (CB)/vi on the abscissa and (CB)/vj on the ordinate axis. It is
the locus of quantities employed of the two factors that cause the same total cost C. Increases in C
induce a parallel outward shift of the isocost line. Cost minimization requires that the isocost be as
close as possible to the origin under the condition that it has a point in common with the given
isoquant corresponding to the desired output. If the isoquant is smooth, the condition v i/vj =
MPi/MPj imposes that the isocost be tangent to the i,j-isoquant associated with q. In order for this
tangency actually to indicate a point of minimum cost, any isocost closer to the origin must have no
point in common with the isoquant. This is ensured if the isoquant is convex.
5.11.2. The partial analogy with the utility maximization problem is graphically clear in the
two-factors case: in both cases we have a map of straight lines and a map of curves; the difference
is that in order to maximize utility we look for the point, on a given straight line (the budget line),
that touches the curve (the indifference curve) farthest from the origin; while in order to minimize
cost we look for the point, on a given curve (the isoquant), that touches the straight line (the isocost)
closest to the origin.
x2

isocosts

isoquant
x1

Figure 5.4
With n factors, the isoquant is a surface of dimension n1 in R n, the isocost is a hyperplane;
the first-order conditions for an interior solution imply tangency between isoquant and isocost, and
the second-order sufficient condition is that the isoquant surface be convex. The thing is graphically
evident with two factors. As in the UMP, convexity of the isoquants ensures that, when the firstorder conditions are satisfied, the solution is a global maximum of C, that is to say, a global
minimum of C. Mathematically, this second-order condition can be expressed, as in the UMP

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 24

problem, as the condition that at the solution x* it is d 2(C)>0 for displacements from x* that satisfy
the constraint f(x)=q, and it can be again shown that this corresponds to the condition that the
leading, or naturally ordered, principal minors (starting from the third one) of the bordered
Hessian[23] alternate in sign starting from positive (remember that the problem is formulated as the
maximization of C[24]).
For the reasons explained earlier, convexity of the isoquant surface always obtains when CRS
and divisibility are assumed and all factors are variable.
The condition vi/vj = MPi/MPj can be realized without the conditions MP i=vi/p, MPj=vj/p
being realized; when so, factor employments are not optimal, the output level is not profitmaximizing. Cost minimization is only one part of what is necessary for profit maximization: one
must also choose the optimal output level. This latter choice can also be examined in terms of cost
function and revenue function, see below. But before, we explore the cost function a little more.
5.12. WACm; Kuhn-Tucker conditions and cost minimization; Shephards lemma.
5.12.1. Cost minimization has an implication analogous to the WAPM. If for a given output q
and given input rentals v the firm finds it optimal to utilize an input vector x=H(v,q), it must mean
that any other input vector capable of producing q (or more) must cost at least vx, in other words,
vxvx for all x such that f(x)q, a result sometimes called weak axiom of cost minimization,
WACm for short. If at input prices v the firm chooses input vector x and at input prices v* the
firm chooses input vector x* to produce the same output, proceeding in the same way as for the
WAPM one reaches the conclusion (vv*)(xx*)0, more often expressed as
vx0.
This implies, for example, that if only one input price changes, the demand for that input must
change in the opposite direction. (Note that if inputs were measured as negative quantities the
inequality sign would be reversed, still this is not the same result as was derived from the WAPM,
because here output is kept fixed.)
In general, not all inputs will be used in positive amounts by a firm; the condition v i/vj =
MPi/MPj must hold for inputs both used in positive quantities; the more general necessary firstorder conditions for cost minimization are derivable from the Kuhn-Tucker theorem. In this case,
with q the given output, the function to be maximized is vx and the constraints are qf(x)0,
and xi0, i=1,...,n. The first-order conditions are therefore (multiplying both sides by 1):
23

Exercise: derive the bordered Hessian in the two-factors case and show that its determinant is positive
if and only if the analogous bordered Hessian of the UMP in the two-goods case has a positive determinant.
24
If the problem is formulated as the minimization of C, then the leading principal minors of the bordered
Hessian must be all negative.

f petri

Adv Micro

vi = 0f/xi + i,

chapter 5 firms and productionGE

10/03/2016

p. 25

i=1,...,n, with the complementary slackness condition that if i>0,

that is to say, if vi>0f/xi, then xi=0.


The Lagrange multiplier 0 in this problem can be interpreted through an application of the
Envelope Theorem. We define the marginal cost function MC(v,q) as the derivative of the cost
function with respect to output: MC(v,q)=c(v,q)/q . It tells us by how much total cost increases if
the firm increases output by one (small) unit. Now let M be the value function of the maximization
problem with constraint

m a x ( vx ) s . t . f x q ;
x

hence M = c(v,q); the Lagrangian function of the maximization problem is


vx0(qf(x)),
and the Envelope Theorem implies
M/q= 0, i.e. c/q=0:
the Lagrange multiplier 0 in the cost minimization problem is the marginal cost.
Another way to prove this result is the following, where fif(x)/xi:

f dx ... 0 f n dx1
c( v ,q ) v1 dx1 ... vn dx1

( from the first order conditions ) 0 1 1


0 .
q
f 1 dx1 ... f n dxn
f 1 dx1 ... f n dxn

5.12.2. Let us now consider again the function c(v,q) as the value function of the
maximization problem with constraint, m a x (vx) s. t. f x q . Let us apply the Envelope
x

Theorem to the derivative of this value function with respect to factor rentals; we obtain (now we
indicate the Lagrange multiplier as simply ):

c v , q
f x
f ( x )
0 );
xi
xi (because
vi
vi
vi

the quantity xi that appears in this expression is in fact the conditional demand for input i, because it
is the quantity of this input demanded when output is kept fixed at the level q; so we obtain:
Shephard's Lemma for the firm:

c( v ,q )
xi ( v , q ) .
vi

In words: The conditional factor demands are the partial derivatives of the cost function with
respect to factor rentals.
This is the original Shephard's Lemma, which was later extended to consumer theory.
5.13. The profit function and Hotelling's Lemma

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 26

5.13.1. The profit function (p,v) is defined as the value function of the (unconstrained)
maximization problem
maxx pf(x)vx,
which asks to find the input vector that maximizes profit, as a function of output price and input
rentals. (The function (q)=pqC(q), profit as a function of output under cost minimization, is not
called profit function.) In short-period analysis the inputs are the variable ones.
The profit function is only defined when the condition p=MC yields a definite optimal output.
This is not the case if the production function has constant returns to scale (or if there is perfect
replicability of plants, cf. 5.??): then average and marginal cost coincide and are constant, and for
a price-taking firm, if p>AC, profits grow endlessly with increases in output, so there is no optimal
output, while if p=AC there is an infinity of solutions all yielding zero profit. Thus the profit
function requires sufficiently decreasing returns to the scale of variable inputs[ 25]; this can be
justified in short-period analyses by taking some inputs as fixed; on the contrary, a long-period
profit function requires assumptions on returns to scale (a U-shaped LAC curve) that not all
economists find plausible. But precisely in long-period analyses the profit function is irrelevant
even when it can be defined, because entry will anyway maintain profits equal to zero, so the
determination of equilibrium industry output does not require consideration of the profit function, as
we explain later. As to the short-period profit function, it is based on a rather misleading definition
of profit as revenue minus variable cost, neglecting the need to include among costs the quasirents
of fixed factors as opportunity costs. These quasirents, if included in total cost and if the
entrepreneur is neither better nor worse than other entrepreneurs at maximizing profit, would
always bring profit to zero even in short-period analysis because they are the opportunity cost of
using the fixed plant instead of renting it out to other entrepreneurs (these would be ready to pay for
the use of the fixed plant a maximum amount equal precisely to what would bring down their profit
to zero). Therefore the profit of the short-period profit function is the sum of true profit and of the
quasirent to be attributed to the fixed factors. As a consequence, it may be the case that this profit
is positive but the entrepreneur would do better to sell the firm because other entrepreneurs would
get more out of that set of fixed factors and so they value the fixed factors more than she does.
However, the profit function is widely used in microeconometric practice, so we list its main
properties:

25

Exercise 5.2: Why the stress on sufficiently? Is decreasing returns to scale without qualifications a
sufficient condition for the existence of a profit function? Try exploring the case in which C(q)
asymptotically approaches from below a function a+bq, with a,b>0.

f petri

Adv Micro

chapter 5 firms and productionGE

p. 27

10/03/2016

Properties of the profit function. Suppose that the production function f: Rn+R+ is
continuous, strictly increasing and strictly quasiconcave and that the profit function (p,v), i.e. the
value function of the problem max x pf(x)vx, is well defined for given (p',v') and continuous in (p,v)
in a neighbourhood of (p',v'); then in that neighbourhood (p,v) is
(1)
(2)
(3)
(4)
(5)

increasing in p
decreasing in v
homogeneous of degree one in (p,v)
convex in (p,v)
differentiable in (p,v)>>0.

PROOF: We only prove the least intuitive of these properties, convexity, i.e. that for any scalar a such
that 0a1 it is a(p,v)+(1a)(p',v')[ap+(1a)p',av+(1a)v']. Define paap+(1a)p' and vaav+(1a)v'.
Let x, x', xa and q=f(x), q'=f(x'), qa=f(xa) be the solution inputs and outputs associated respectively with
(p,v), (p',v') and (pa,va). Then it is
(p,v) = pqvx pqavxa
(p',v') = p'q'v'x' p'qav'xa
which imply a(p,v)+(1a)(p',v') a(pqavxa)+(1a)(p'qav'xa) = paqavaxa = (pa,va).

5.13.2. Now let us apply the Envelope Theorem to the profit function. The latter is the value
function of a maximization problem without constraints, so we obtain:
/p = f(x)=q i.e. the partial derivative of the profit function with respect to the output price
yields the optimal output;
/vi = xi i.e. the partial derivative of the profit function with respect to input rental v i
yields the (unconditional) demand for input i measured as a negative number (i.e. in accord with
the netput notation).
These two results constitute Hotelling's Lemma (sometimes called the Derivative Property).
Thus the partial derivatives of the profit function, when the latter exists, give us the output
supply function and (the negative of) the input demand functions[ 26]. Since the profit function is
convex, we reach the result that, when the profit function is well defined, both /p and /vi are
non-decreasing and generally increasing in p, respectively in vi, that is to say:
supply is a non-decreasing, and generally an increasing, function of output price (because
convexity of implies /p is non-decreasing and generally increasing in p, but /p=q);
the unconditional demand for an input, when measured as a positive quantity, is a nonincreasing, and generally a decreasing, function of the input own rental: there are no Giffen inputs.
26

Exercise: Prove that if the firm is a multiproduct one and profit depends on inputs and outputs, then
Hotellings Lemma, again derived from the Envelope Theorem, generalizes to equality between the partial
derivative of the profit function relative to any one output price, and supply function of that output.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 28

5.14. Conditional and unconditional factor demands, inferior inputs, rival inputs,
substitution effect and output effect.
5.14.1. When the profit function is well defined, for each output price p and vector of factor
rentals v there is an optimal output and an associated vector x of optimal factor utilizations (the
latter need not be unique, but I will assume it is). In this case we can define the supply function (of
the individual firm) S(v,p) that indicates how optimal output changes with factor prices v and the
output price p, and the (vectorial) unconditional factor demand function x(v,p) that indicates the
associated optimal factor employments; it is
x(v,p)=H(v,S(v,p)).
If the profit function is defined for the short period, i.e. with some inputs fixed, then only the
variable inputs appear in x and in v. When the profit function does not exist, the output supply
function and the input demand functions do not exist either: no unique profit-maximizing output
exists. When these functions can be defined, what about the sign of their partial derivatives?
If profit is considered a function of q, its maximization requires solving the problem
maxq pqC(v,q),
whose first-order necessary condition is the equality of product price p, and marginal cost
MC(v,q):=C/q; the second-order sufficient condition is that MC must be rising at the optimal q.
Hence
S(v,p)/p>0
the (inverse) supply curve (optimal q as a function of p, with p in ordinate and q in abscissa) is
increasing (as long as to increase output is possible), a result actually already reached via
Hotellings Lemma plus the convexity of the profit function; but it can be useful to see the same
result from different perspectives.
We cannot reach a result on the sign of x i(v,p)/vi directly from the condition vi=pMPi,
because xi is not the only input use that will change when v i changes; but Hotellings Lemma and
the convexity of the profit function imply xi(v,p)/vi0: the own-rental effect is negative.
5.14.2. Does the above result on input use imply S(v,p)/vi 0 ? in other words, is it always
the case that the optimal output, when it exists, decreases if the price of a factor (in positive use)
rises? Perhaps surprisingly, not always. The reason is that the factor might be an inferior input,
defined as an input whose conditional demand falls as output increases, that is, such that
xi(v,q)/q<0 (at least at the given input rentals and in a neighbourhood of the initial q). It is indeed

f petri

Adv Micro

chapter 5 firms and productionGE

p. 29

10/03/2016

possible that an increase of q is best achieved by increasing some input and decreasing some other
input.

pq
xj

total cost,
revenue

expansion path

c
c

xi
Fig. 5.4bis(a)

q
Fig. 5.4bis(b)

This possibility is shown in Fig. 5.4bis(a), where, assuming two variable factors with given
prices, several isoquants are drawn and for each one the tangency with an isocost is shown; the
locus of tangency points is called the expansion path of the firm for given input prices; this
expansion path can exhibit a decrease of the utilization of one factor as higher isoquants are
reached. An example of inferior input can be labour in some agricultural productions on a given
specialized land where, as long as the required output is not large, it can be optimal to achieve it
with abundant use of labour nearly unassisted by machinery, but when output becomes large, it
becomes convenient to mechanize production which reduces the needed amount of labour. Another
example can be the use of fuel in a chemical production process (carried out in a fixed plant) that
requires heat, such that when the process is run on a small scale, heat must be provided by flames
under the cauldron, which means consumption of fuel, but the greater the scale on whichz the
process is run, the more heat is produced by the chemical reaction itself, and the need for fuel
consumption decreases: fuel is an inferior input[27].
The relevance of inferior inputs for the sign of S(v,p)/vi comes from the fact that locally the
increase in the rental of an inferior input shifts the marginal cost curve downwards, so the p=MC
condition is achieved for a greater q although at a smaller profit, as in Fig. 5.4bis(b) where the
increase in the price of an inferior factor shifts the cost curve from c to c. This means that
S(v,p)/vi 0 only if factor i is not inferior.
The proof that the marginal cost curve shifts downwards, that is, MC/v i<0 if input i is
27

Exercise: suppose the chemical process in the text produces output q with chemical input x and gas y,
according to the production function {q=f(x,y) with q=x1/2 if y1/x, q=0 if y<1/x}. Assume px=1/3, py=3, and
fixed cost (due to the presence of fixed machinery) is 2. Confirm that y is an inferior input; derive the cost
function and prove that minimum average cost is reached for q=3.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 30

inferior, is an interesting application of the notions studied thus far. The cost function is assumed
twice continuously differentiable, so second-order cross partial derivatives coincide.
Proof. Let input i be inferior; differentiate both sides of x i(v,p)=Hi(v,S(v,p)) with respect to p, and
apply Shephard's Lemma and the coincidence of cross partials to obtain:

xi ( v, p ) H i (v, q ) S (v, p )
c (v, q ) S (v, p )
c(v, q ) S (v, p ) MC S (v, p )

.
p
q
p
q vi
p
vi
q
p
vi
p
S( v, p)
> 0 and we are assuming that x i is inferior, the conditional demand for factor i decreases
p
MC
x i ( v, p)
when output increases with given input prices, so it is
< 0, hence it must be
< 0. Or
v i
p

Since

also, the above shows that

H i (v, q ) MC

, and the left-hand side is <0 by the definition of inferior


q
vi

input.

I will not go into the intricacies of inferior inputs except to prove that when the production
function is differentiable a necessary (but not sufficient) condition for an input x i to be inferior is
that it be rival of some other input xj, which means that an increase in x j decreases the marginal
product of xi[28]; this above all as a way to introduce rival inputs, that have some right to attention
on their own and are briefly discussed in the next paragraph.
Proof. Let w = 1/p v be the vector of real factor rentals in terms of the firm's output; f(x)=q is the
firm's production function. Assume that marginal cost is locally strictly increasing so the profit function ( w)
is well defined at the optimal q. Maximization of (w) determines the unconditional factor demands x(w).
These satisfy the well-known condition wi = f(x)/xi, all i, that for brevity I indicate as
Dxf(x) = w,
where Dx means the vector of partial derivatives of the subsequent function with respect to the variables in x.
Let us assume that x(w) is invertible, i.e. to each vector x there corresponds a unique vector w. Let w(x(w))
be this inverse function[29]. Since w = Dxf(x), it is Dx(w(x(w)) = Dx2f(x(w)). And by the derivative rule for
inverse vectorial functions, the Jacobian Dx(w(x(w)) is the inverse of the Jacobian Dwx(w):
Dx(w(x(w)) = [Dwx(w)]1 = Dx2f(x(w)).
But then, inverting everything:
Dwx(w) = [Dx(w(x(w))]1 = [Dx2f(x(w))]1.
Now we use a result in the theory of matrices (cf. e.g. Takayama, Mathematical Economics, 1973, p.
393, Theorem 4.D.3) that states that if the off-diagonal elements of a square nonsingular matrix are all non28

Rivalry can arise, for example, if there is some third input x h, whose services co-operate with either x i
or xj, and such that when xj increases it is convenient to allocate x hs services to co-operate mainly with x j; or,
one input can have direct negative side effects on the efficiency of another input, for example owing to
chemical interactions fertilizers might decrease the marginal productivity of pesticides in fruit production. Of
course when the production function is twice continuously differentiable the cross partial derivatives
coincide so rivalry is reciprocal.
29
As I have used x(w) to represent the function that makes x depend on w, its inverse should be actually
represented as w=x1(x).

f petri

Adv Micro

chapter 5 firms and productionGE

p. 31

10/03/2016

negative then its inverse is non-positive. If no inputs are rival, the marginal product of no input decreases
when some other input is increased, so the Hessian of the production function has non-negative off-diagonal
elements, hence its inverse is non-positive, and therefore D wx(w) is non-positive.
Now fix the product price p so we can return from w to v. It is S(v,p)/vi=

(
j

f ( x) x j (v, p )
)
x j
vi

< 0 because we have shown that the second terms in the parentheses are non-positive. Therefore absence of
rival inputs implies that supply cannot increase when an input price increases.

5.14.3. The possibility of rivalry among inputs has some relevance for the marginalist or
neoclassical approach to income distribution. As explained in Chapter 3, according to this approach
each factor tends to earn, in the long period, a rental equal to (the value of) its full-employment
marginal product. In the long period, competitive industries behave like firms with CRS production
functions, and this prevents rivalry if factors are only two, as I prove below; but if factors are more
than two, an increase in the equilibrium use of a factor because of an increase in its supply can
cause a decrease of the full-employment marginal product of a rival factor in fixed supply: thus the
marginal approach does not exclude the possibility that an increase in the supply of a factor causes a
decrease of the equilibrium rental of another factor. However, cases of rivalry appear rare and
specific to certain industries, so marginalist/neoclassical economists unanimously consider the
likelihood of the decrease of equilibrium rental just illustrated to be zero for factors, like most types
of labour, the demand for which comes from very many industries.
Proof that with two factors and CRS there cannot be rivalry. If factors are only two, call them x and y,
since marginal products only depend on the proportion x/y owing to CRS[ 30], if an increase of x with y fixed
causes a decrease of MPy it must mean that an increase of y with x fixed causes an increase of MP y, and this
means a non-convexity of output as a function of y with x fixed; but such a non-convexity is excluded by
profit maximization plus CRS: in Fig. ??, where the curve represents output as a function of y with factor x
fixed, all points on the segment AB can be reached by a linear combination of the factor vectors (y A,x) and
(yB,x); this generalizes what was explained in Chapter 3 on the difference between technological and
economic total productivity functions or isoquants, and we conclude that with CRS and two factors,
increasing marginal products are impossible. This excludes rivalry.

output
B
A
30

Cf. footnote 11.

f petri

Adv Micro

chapter 5 firms and productionGE

yA

10/03/2016

p. 32

yB
Fig. ??

The possibility just mentioned of a decrease of the equilibrium rental of a factor when supply
of another factor increases is due to rivalry, not inferiority: an input is inferior if the demand for it
decreases when output increases at given factor rentals, but in this case with CRS all inputs increase
in the same proportion as output, and I have argued that constant returns to scale industries is the
only legitimate assumption for the long-period behaviour of competitive industries with possibility
of plant replication, while inferior inputs require non-constant returns to scale[ 31]. Now, the
determination of equilibrium factor rentals is a long-period issue, owing to the complex, timeconsuming adjustments (changes in outputs, shifts of labourers across firms, etc.) required for
equilibrium to be approached on factor markets; therefore input inferiority, differently from rivalry,
is irrelevant for the marginalist theory of income distribution[32].
Input rivalry can also cause perverse effects of shifts in the composition of demand for
consumption goods on equilibrium factor rentals. As illustrated in Ch. 3, if one leaves aside possible
perverse income effects then a shift in the composition of demand in favour of goods that use a
factor in a higher-than-average proportion tends to raise that factors equilibrium rental if technical
coefficients are fixed, and if there is technical substitutability the effect on the factor rental is
normally considered to be of the same sign, only weaker. But if a factor is specialized and used only
by one industry and is rival of other inputs in that industry, then when demand for the industrys
product rises the industry increases the use of other inputs in order to satisfy the increased demand
and this can cause a decrease of the marginal product of the specialized factor and hence a decrease
of the demand for it if its rental remains the same: the excess supply of the factor will then cause the
rental of the factor to decrease; so it is not impossible that a rise in the equilibrium output of the
industry, although associated with a higher output price, be associated with a lower equilibrium
rental of the specialized factor. This possibility looks exceptional, but it cannot be excluded.
31

In the Exercise in footnote 27?? the industry would increase output by increasing the number of optimal
plants (owned by the same firms, or by newly formed firms), each one producing 3 units, thus at the industry
level the inferiority of factor y disappears. Indeed constant returns to scale imply that expansion paths are
rays from the origin.
32
Except possibly for specialized inputs to be associated with other specialized inputs, e.g. special fertilizers
to be used on very special lands for the production of specific products.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 33

5.14.4. Finally, we consider xi(v,p)/vj. It should be intuitive that its sign is ambiguous: if
factor j is not inferior, optimal output decreases when vj increases, and this would reduce the
demand for xi if factor proportions were fixed, on the contrary these proportions change (and in
directions that depend on the specific production function) both because optimal factor proportions
change with q even with fixed relative factor prices, cf. Fig. 5.4bis(a), and because relative factor
prices change and this alters optimal factor proportions for each level of q. The possibility that
factor j be inferior adds further uncertainty by rendering the sign of the change in optimal q
uncertain. All this can be analyzed more formally (but without great gains in clarification).
Differentiating x(v,p)=H(v,q(v,p)) with respect to vj, one gets

xi(v,p)/vj = Hi(v,q)/vj +

H i ( v, q ) S( v, p)

= (cross substitution effect) + (output


q
v j

effect).
This shows that the direction of change of x i is the composition of two directions of change:
along the original isoquant as indicated by the first term on the right-hand side, that indicates the
change in xi if output were kept fixed; and along the expansion path owing to the change in q at
given factor rentals, as indicated by the second term. Both directions are of uncertain sign. The sign
of S/vj depends on whether input j is inferior; the sign of

H i ( v, q )
depends on whether input i
q

is inferior. But even assuming that neither input is inferior (so the sign of

H i ( v, q ) S( v, p)

is
q
v j

negative), still the sign of x i(v,p)/vj for ji is uncertain, because the sign of the cross substitution
effect Hi(v,q)/vj is uncertain as long as there are more than two factors. The reason is that when v j
rises and hence xj(v,p) decreases, the changes in other inputs are affected by two forces: first, it is
necessary to restore q to its given level, and this would tend to increase the demand for the inputs
other than xj; but second, the optimal proportions among the inputs other than x j will change in
favour of the factors whose marginal product relative to the other marginal products rises when xj
decreases: now, it is possible that the decrease of xj affects the marginal product of xi strongly and
negatively, so much so that the optimal proportion among the factors other than x j changes against
xi so much that the compensated demand for xi decreases.

5.15. Elasticity of substitution.


Cost minimization requires that firms which intend to produce a given output locate

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 34

themselves on the point of the corresponding (convex) isoquant where the absolute slope equals the
ratio between factor rentals. A change in the relative rental of two factors may induce no
substitution, little substitution, extensive substitution. For example, if factors are perfect
complements then isoquants (with two factors) are L-shaped and cost-minimizing firms will always
locate themselves at the kink: changes in relative factor prices induce no substitution. In order to
measure the effect of changes in relative factor rentals on the proportions in which firms find it
optimal to combine two factors, economists use the elasticity of substitution. We limit ourselves to
the two-factors case. We have already met this notion in chapter 4. When applied to production, this
elasticity is the absolute value of the ratio of the percentage change in the factor proportion x 1/x2
(along a given isoquant) to the percentage change of their TRS, i.e. MP1/MP2:

d ( x1 / x 2 )
d( x1 / x 2 ) v1 / v 2
x1 / x 2

=
.
21
d( v1 / v 2 ) x1 / x 2
d ( MP1/MP2 )
MP1/MP2
The second of the above two expressions for the elasticity of substitution is based on the
assumption that the firm chooses the factor proportion that satisfies TRS 21=v1/v2; thus the elasticity
of substitution measures the sensitivity, of the proportion in which factors are demanded, to relative
factor rentals; its usefulness lies above all in that it gives an indication of what happens to the
relative shares of factors in total cost as relative factor rentals vary. The relative share of factor 1 in
total cost is given by (v1x1)/(v2x2), which can be re-written as (x1/x2)(v1/v2) or (x1/x2)|TRS21|. When
v1/v2 increases, x1/x2 decreases; an elasticity of substitution equal to 1 means that the two changes
neutralize each other and relative factor shares in total cost do not change. An elasticity of
substitution less than 1 means that when factor 1 becomes relatively more expensive, x1/x2
decreases less than in proportion, so the relative share of factor 1 in cost increases.
This result is used in one-good, two-factor general equilibrium neoclassical models to derive
predictions on factor shares from the elasticity of substitution. In these models the economy
produces with a CRS production function, and income distribution is determined by fullemployment marginal products, hence the product exhaustion theorem holds (cf. footnote 11), and
total factor cost is also total revenue. Changes in factor supplies will then alter factor shares in a
direction that depends on the elasticity of substitution. Thus assume the factors are labour and land,
rigidly supplies and fully employed. Suppose labour immigration raises labour supply, causing the
real wage to decrease relative to land rent. Check your understanding of the issues with the
following questions. If the elasticity of substitution is less than one, the percentage decrease of the

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 35

ratio of wage to rent, needed to ensure the full employment of the increased supply of labour, will
have to be greater or smaller than the percentage increase in labour supply? and the share of wages
in national income will decrease or increase? The correct answers are in a footnote on the next page.
(The popularity of the Constant-Elasticity-of-Substitution (CES) production function among
macroeconometricians derived from the claim that the share of labour in national income did not
change much for many decades in the USA. The empirical evidence was however immediately
disputed, and certainly looks much less convincing now, because after the 1980s the share of wages
has decreased considerably; furthermore there are formidable aggregation problems behind any
attempt to study an economy as if it were producing a single output; also, it is totally unclear why
technical progress should not alter the elasticity of substitution over the decades; and last but not
least, the validity of the marginal/neoclassical approach to income distribution can be disputed with
strong arguments, as will be explained in later chapters.)
5.16. Integrability of conditional factor demands
We touch very briefly on the duality between some of the notions explained in this chapter.
We have seen that cost function and conditional factor demands stand to the production function in
exactly the same relationship as expenditure function and Hicksian (or compensated) consumer
demands stand to the utility function. Therefore the result reached in consumer theory, that the
expenditure function or the Hicksian consumer demands allow the reconstruction of the utility
function (more precisely, of its convexification), also holds for production theory: the cost function
contains the same economically relevant information as the production function, and from it one
can recover the (convexified) isoquants. Of course this is only possible if the chosen function really
is a cost function, i.e. if there exists a production function that generates it; the conditions
guaranteeing it are listed in the following proposition (we omit the proof, cf. Varian, 1992, p. 85):
Let c(v,q) be a differentiable function which is
(i) non-negative if (v,q) is non-negative,
(ii) non-decreasing in (v,q),
(iii) concave in v, and
(iv) satisfying homogeneity of degree 1 in v ;
then c(v,q) is the cost function of a production function.
It can be convenient, in applied work, to start directly from a cost function rather than from a

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 36

production function[33].
As to conditional factor demands, they can be 'integrated' to yield the production function that
generated them, with the same procedure that derives the utility function from a system of
compensated demands for consumption goods.
5.17. Functional separability: Leontief separability (and weakly separable utility). Suppose
we can separate inputs into two sub-vectors, x=(x1,...,xm) and y=(y1,...,yn), respectively with prices v
and w, and that the production function satisfies the following condition: f(x,y)=f(z(x),y), where z()
has the characteristics of a production function: it is as if inputs x produced a single intermediate
good which then produces the final output in combination with inputs y. This can reflect a true
production of an intermediate good, or be simply a property of the production function.
If z(x) is differentiable and f(z,y) is also differentiable, then f/xi=(f/z)(z/xi); as a result,
the Leontief weak separability condition holds: the marginal rate of substitution between any two
x-goods is independent of the amounts of y-goods:
MRSxj,xi = (f/xi)/(f/xj) = (z/xi)/(z/xj).
Vice-versa if the Leontief weak separability condition holds, then a differentiable f(x,y) can be
written as f(z(x),y) where z(x) is a scalar function. (We omit the proof.)
Under weak separability, the firm can adopt a two-stage cost-minimization procedure: it can
first determine the cost-minimizing input combination of the x-inputs for each level of z, and the
resulting cost of z; and then it can determine the cost-minimizing input combination of (z,y) for
each level of output. If f() has constant returns to scale, so does z(); then the cost function for the
good z can be written as (v)z, with (v) representing the unit price of z.
The production function must be additively separable if one wants that not only the marginal
rate of substitution between two inputs, but also the marginal product of an input, be independent of
the amounts of other inputs. It is not easy to think of realistic examples to which such an
assumption might apply, except when a firm uses physically separate processes that produce the
same good using inputs of different quality, and one treats the inputs of each process as a single
input because in each process they are combined in fixed proportions: this might perhaps be the
case for some agricultural or mineral product produced on lands, or by mines, of different quality.
5.18. Supply curves: Short-period Marshallian analysis.
33

Answers to the questions posed on the previous page: greater, decrease. Note that the decrease in the share
of wages in national income does not imply a decrease of the total wage bill, because the increased supply of
labour raises total output, it is therefore possible that total wage payments increase, but by a lower
percentage than the increase of total output so that the share of wages decreases.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 37

5.18.1. A traditional approach distinguishes short-period from long-period profit


maximization and short-period from long-period supply curves of firms. The difference consists in
the fact that in the short period a firm is unable to alter the quantities of some of the inputs.
This can derive from contracts that it would be too costly to modify, for example contracts
that oblige a firm to use the services of a certain accounting firm for a number of years; or from
time constraints, for example moving out of certain buildings to a new location may be a complex,
time-consuming affair. In the end time constraints are the essential element, because, given
sufficient time, everything can be modified.
However, the argument most commonly used to distinguish the short-period from the longperiod supply curve of the firm is the least convincing. This is that once a firm has a fixed plant, the
latter is given for the firm until it becomes so old as to justify scrapping and replacement with a new
fixed plant. This argument is not convincing because firms can always sell or rent out their fixed
plants to other firms, or buy or rent the fixed plants of other firms (a firm is a legal entity, not to be
confused with its plants). If one re-reads the inventor of the short-period/long-period distinction,
Alfred Marshall, one finds that what he had in mind was referred not to the individual firm but to
the industry: the short period was the analytical period within which there was not enough time
significantly to change the amounts available to the entire industry of specialized durable inputs
necessary to the industry and requiring considerable time for their production. Thus Marshall's
short-period analysis of the fishing industry took as given the number of fishing ships and of
experienced fishermen, not the number of firms nor how the fishing ships were divided among
different firms (Marshall ?? p.??). Marshalls approach permits the determination of short-period
supply curves of industries, without requiring a fixity of factors for firms that is much more difficult
to justify.
However, for completeness we also explain the usual textbook approach. This can be seen as a
pedagogical first step toward a better appreciation of what long-period analysis means, a first step
that also allows a treatment of indivisibilities. We cover this ground quickly because we do not have
much to add to what is learned on these issues in introductory economics courses.
5.18.2. So let us take the 'fixed plant' of the firm as given in the short period, where the fixed
plant includes all inputs whose (maximum) quantity is treated as given[ 34]; this means that one can
even omit the fixed plant from the inputs appearing in the production function; the other inputs, the
sole ones affecting production in the short period, will be called variable inputs; the corresponding
34

It is possible not to use the entire amount of a fixed factor; this will be just one instance of the difference
between technological and economic production function.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 38

production function is called the short-period, or restricted, production function. Cost


minimization can only operate on the variable inputs, and the short-period cost function results from
the minimization of the cost of variable inputs.
Correspondingly, there will be fixed costs and variable costs. Variable cost is the total cost of
variable inputs. The costs that are fixed (in the sense of not depending on the quantity produced) but
only exist as long as the firm exists, i.e. disappear if the firm is closed down, are called quasi-fixed
costs. Fixed costs proper are those costs independent of the quantity produced, that must be borne
by the owners of the firm even if production is discontinued and the firm is closed down. Fixed
costs are due to irrevocable contracts that oblige the owners of the firm to pay them in all instances,
e.g. the repayment of debts. Fixed costs proper do not include, for example, those overhead labour
costs (manager's secretary and analogous accounting labour etc.) independent of the quantity
produced but which can be eliminated by closing down the firm and firing all workers. Fixed costs
do not necessarily coincide with the cost of fixed factors. A firm might have a debt to be repaid, that
causes a fixed cost but is due to past expenses and has no connection with the firms present fixed
plant.
For the question whether the firm should close down when profit is negative, fixed cost
proper should not make a difference since the owners of the firm must still bear it even if the firm
closes down; quasi-fixed cost on the contrary does make a difference and therefore it must be
included in the variable cost. But for simplicity in what follows there are no quasi-fixed costs.
The (short-period) variable cost function, to be indicated as VC(v,q), results from the
choice of variable inputs that minimizes variable cost for each assigned level of output.
It is plausible that, since there are fixed factors, the short-period production function will
exhibit decreasing returns to scale at least after a certain level of output; as a result, at least beyond
a certain level of output VC(v,q) will increase more than in proportion with output. This is
formalized by assuming that the short-period marginal cost, i.e. the derivative of variable cost
with respect to output,
MC(v,q):=VC/q
is an increasing function of q at least beyond a certain level of output. In what follows I take v as
given and for brevity I often drop the indication of the functional dependence on q. Let us now
define (short-period) average variable cost as
AVC(v,q) VC/q;
it is possible that initially AVC is a decreasing function of output, indicating that the fixed plant was
planned to be optimal for a certain level of production and up to that level variable cost increases

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 39

less than in proportion with output. If all these magnitudes are continuous functions of otuput we
can study the relationship among them. For q=0 it is AVC=MC because for the first small unit of
output average variable cost coincides with the increase in cost. Then if AVC is initially a
decreasing function of q, MC is a decreasing function of q too and is less than AVC (in order for the
average cost to decrease, the additional units of output must cause an additional cost lower than the
average). AVC remains a decreasing function of output as long as MC<AVC. But if MC becomes an
increasing function of q at least from a certain level of output onwards, then sooner or later it
becomes equal to AVC, and from that level of output onwards MC>AVC and AVC becomes an
increasing function of q (because the additional units of output cause an additional cost greater than
the average). It follows that AVC reaches its minimum where its curve crosses the MC curve; if one
knows the two functions, this minimum can be determined simply by solving MC=AVC for q>0.
However, if MC is increasing from the very start, then the minimum AVC is reached for q=0. (The
mathematical proof of these statements is easy and left to the reader as Exercise; be sure to check
the second-order conditions.)
Now define short-period average (total) cost as AC=(FC+VC)/q=AFC+AVC. AFC is
average fixed cost, defined as FC/q. If FC>0 then AC>AVC; the vertical distance between the AC
curve and the AVC curve decreases as q increases, because it measures AFC. For the same reason as
for AVC, AC is a decreasing function of q as long as it is greater than MC, and an increasing
function of q as long as it is smaller than MC, and as a result it too reaches a minimum where it
crosses the MC curve. All these relationships are shown in Fig. 5.5: a particularly important point is
X, where the AC and the MC curve cross each other; this point determines the minimum average
cost, MinAC, associated with the given fixed plant and the given factor rentals, and the
corresponding quantity of output q^. As long as the price at which the firm sells its output is greater
than MinAC, the firm makes a positive profit.
How does the firm maximize profit in the short period? By equalizing marginal cost and
output price. This can be shown as follows. Let R=pq stand for the firm's revenue, and SC(q) for its
short-period total cost function, whose derivative with respect to q is the marginal cost MC(q). Then
=RSC(q)=pqSC(q)
and the first-order condition for a maximum is pMC(q)=0. The second-order condition is
dMC(q)/dq<0,
i.e. MC must be increasing where it equals the given output price.
The condition p=MC(q) implies a supply curve of the firm which coincides with part of the
MC curve (except that now the independent variable is the one on the vertical axis) if on the vertical

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 40

axis we also measure the output price. The part of the MC curve which coincides with the supply
curve is the part above the AVC curve. The reason is that the firm finds it convenient to go on
producing even if p<AC, as long as p>AVC, because in this way it can at least minimize its loss: the
excess of revenue over variable cost compensates at least partially for the fixed cost which must be
borne anyway. But if p<AVC then the firm minimizes its loss by not producing at all(35).

35

In concrete situations, a firm may decide to go on producing even when the price is below minimum
average variable cost, if it esteems that this is a temporary situation and that the interruption would damage
profit more than continuing to produce at a loss for a time.

f petri

Adv Micro

chapter 5 firms and productionGE

p. 41

10/03/2016

p
MC=MVC
AC
AVC

p*=MinAC

AFC
O

q^

Fig. 5.5. Average (short-period) cost AC and average variable cost AVC when marginal cost MC is
initially decreasing; average fixed cost AFC is a rectangular hyperbola and equals ACAVC. Note that if
AVC included some quasi-fixed costs then the AVC curve would not start at the same level as the MC curve,
but would have initially a shape similar to that of the AFC curve.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 42

5.18.3. The idea that a firm will continue to operate even when making a loss, as long as
revenue covers at least the variable costs, can derive from considering fixed cost proper a sunk cost,
to be paid whether one continues to produce or not, for example interest on the debt contracted to
start production; but this kind of explanation is endangered by the variable confines of the firm as a
legal entity, for example the uncertain existence of fixed costs proper in the presence of limited
liability. A clearer theory is obtained if one reformulates the question as regarding, not the survival
of the firm, but rather the survival of a fixed plant. The key is to consider the cost of using a fixed
plant a residually determined quasirent. The analogy with land is clarificatory. Imagine that
someone buys a land with borrowed money in order to rent it to firms, and then discovers that the
rent she can earn is insufficient to repay the interest on the debt; this is no reason not to rent the land
out to firms: as long as rent is positive, it is not convenient to leave the land idle; perhaps our owner
will go bankrupt, but then the land will be bought (for a price appropriate to its rent-earning
capacity), and utilized or rented out to firms, by someone else. Fixed plants are like lands in that,
once created, it is best to utilize them as long as they can earn a positive rental. The rental earned by
the fixed plant is not made explicit in the usual formalization of short-period firm cost and profit,
but it can be derived from it because it is the difference between revenue and cost of variable
factors. Indeed, a firm might lease its fixed plants to other entrepreneurs; what maximum rental will
an entrepreneur be ready to pay for the right to use a fixed plant she does not own? A rental equal to
the maximum residual obtainable after subtracting all other variable costs[ 36] from the revenue one
can earn by operating the fixed plant; such a rental would reduce profit to zero. If the rental is less
than that, the profit of the leasee is positive, and entrepreneurs must be expected to compete for the
right to use the fixed plant, therefore the rental will rise to the zero-profit residual just discussed. If
the entrepreneur is also the owner of the plant, she should include in the costs the opportunity cost
of the use of the fixed plant the revenue the owner gives up when deciding not to lease the fixed
plant to other entrepreneurs , and this opportunity cost is the maximum rental thus determined,
called quasirent by Alfred Marshall because of its analogy with the rent of land (the difference is
that fixed plants deteriorate). The entrepreneur who first purchases the fixed plant is in the same
position as the person who purchases a land; she may go bankrupt if the plant's quasirent falls below
the level expected at the time of purchase rendering it impossible to repay the debt incurred to
purchase the plant, but as long as quasirent is positive the plant will not be shut down, it will be
bought by some other entrepreneur at its new value (the present value of its new expected quasirents
36

Inclusive of quasi-fixed costs if these are positive.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 43

for its remaining economic life(37)) and will be kept in operation.


The considerations just developed have an interesting implication. If one neglects those fixed
costs which are, in the short period, ineliminable and therefore not an opportunity cost (e.g.
payments for debts contracted in the past), then a correct imputation of costs, inclusive of
opportunity costs and hence of quasirents of fixed plants, renders the short-period profit always
equal to zero, even when it appears positive with the usual formalization: the reason why it appears
positive is that short-period total cost as usually defined does not include the quasirent earned by the
fixed factors.
5.18.4. The above considerations also show that in order to determine the short-period supply
curve of an industry what is necessary is the supply curves of the several fixed plants in the
industry; how their property is subdivided among firms is not relevant (as long as efficiency is
independent of ownership). The short-period supply curve of the industry is given by the horizontal
sum of the parts, of the short-period marginal cost curves of the single plants, which lie above the
respective AVC curves.
price
p

q1

q2

q1+q2

Fig. 5.6. Horizontal sum of the short-period supply curves of two price-taking firms having different
minimum AVC. Aggregate supply at price p, the segment ef, is the horizontal sum of the supplies of the two
firms, the two segments ab and cd.

(In order to determine the short-period supply curve of the price-taking firm the consideration
of fixed costs, as well as of the AC curve, is not necessary, but these notions become necessary for
long-period analysis, and this is why they have been introduced.)

37

We are here introducing a rate of interest. This part of the chapter is concerned with the theory of the
firm, and we want to present this theory so that it is applicable also to economies where there is a rate of
interest.

f petri

Adv Micro

chapter 5 firms and productionGE

p. 44

10/03/2016

5.19. From short-period to long-period supply


5.19.1. The short-period AC and MC curves were derived for a given fixed plant and given
fixed and quasi-fixed costs. We can now extend the analysis to the long period i.e. to the analytical
situation where we treat all inputs as variable, by imagining that the firm can choose among a set of
fixed plants, and for each one of them it can derive the AC and MC curves and find the minimum
average cost and the associated output level. We need not assume perfect divisibility of the elements
which go to form a fixed plant: the firm can be confronted with a finite number of alternative fixed
plants, for each one of which it can determine the AC and MC curve. The smallest of the minimum
average costs associated with the several alternative fixed plants is the true minimum average cost,
to be indicated as MinLAC; let us indicate the associated output as q*.
The long-period average cost curve LAC is the inferior envelope of the short-period average
cost curves, cf. Fig. 5.7. Thus each point of the LAC curve is associated with a type of fixed plant,
but not generally with that fixed plants minimum-average-cost output. If the fixed plant consists of
a single factor whose amount can be varied continuously (no indivisibilities), then each short-period
supply curve has a point in common with the long-period supply curve, where the two curves have
the same slope.

LAC

q
Fig. 5.7
Proof. Assume the fixed factor is factor n; consider the unconditional long-period demand x n(v,q) and
assume v is given; for a given level q of output, cost minimization determines a demand x n(v,q) for factor
n; if its fixed amount is just xn*=xn(v,q), then for q different from q long-period cost cannot be greater than
short-period cost SC, because the additional short-period constraint (the fixed amount of x n) cannot possibly
permit a lower cost and will generally imply a higher cost, hence c(q)SC(q,x n*); and for q=q long-period
and short-period cost coincide; hence the long-period average cost curve LAC coincides at q with the shortperiod average cost curve based on x n=xn*, and is not above it (and generally below it) for q different from
q. For each q, there will be a x n*=xn(v,q) for which one can repeat the above reasoning; if both kinds of
average cost curves are smooth, at the q for which the given x n* is optimal the two curves must be tangent
to each other. Repeating the reasoning for all q completes the proof.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 45

Note that some of the short-period AC curves in Fig. ?? have been drawn as having no point
in common with the LAC curve. The reason is that if the fixed plant consists of several factors
combined in fixed amounts, the combination may be a suboptimal one for all output levels; then the
AC curve corresponding to that fixed plant will be everywyere strictly above the LAC curve.
If there is perfect divisibility and constant returns to scale, then MinLAC can be reached for
any q; if there are indivisibilities and replicability of plant, then there is a minimum efficient scale
of output q* that allows the firm to achieve an average cost equal to MinLAC, and the firm can
reach the same minimum average cost by producing 2q* with two fixed plants identical to the one
which produced q*, or by producing 3q* with three fixed plants, etc. The AC and MC curves with
two fixed plants are the curves with one fixed plant, 'stretched' rightwards so as to reach the same
value on the ordinate for a double value on the abscissa. In Fig. 5.7b we see the AC curves and MC
curves with one, two and three fixed plants of the same type. A price-taking firm considers that it
can sell any amount of product at the given price, therefore as long as the output price p is greater
than MinLAC, the firm finds it convenient to grow without limits by replicating infinite times the
plant associated with MinLAC; if p=MinLAC, then the firm's maximum profit is zero and the firm
is indifferent between producing q*, 2q*, 3q* etcetera; if p<MinLAC, in the long period the firm
does not produce.

AC
MC1

MC2

AC1

AC2

MC3

AC3

MinLAC
q*

2q*

3q*

Fig. 5.7b. Average and marginal cost curves with one, two or three identical plants.

5.19.2. Let us see how these considerations connect with profit maximization.
Mathematically, the problem is
maxq (q)=R(q)C(q)=pqC(q)
where C(q) is the long-period cost function, and R=pq is revenue. For levels of production requiring
the use of a high number of fixed plants, if replication of plants does not cause a decrease in

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 46

efficiency it is legitimate to treat C(q) as proportional to q, because average cost remains nearly
constant as q varies owing to the fact that the variability in the number of plants ensures that, even
when q is not an integer multiple of q*, each plant produces a quantity very close to q*; for example
in betweeen q=100q* and q=101q* the production of each plant differs from q* by at most 1%
and therefore remembering the U-shape of the average cost curve of a plant average cost is very
nearly equal to MinLAC; in Fig. 5.7 this is evident already with three plants. Therefore in this case
it is legitimate to treat the long-period average cost as constant, equal to MinLAC. It is then also
equal to the long-period marginal cost LMC, the derivative of the long-period cost function c(q).
Therefore if p>LMC=MinLAC, no q satisfies the condition p=LMC for a price-taking firm; the
profit (q)=(pMinLAC)q increases without limit by increasing q; the problem max q =pqc(q) has
no solution. If p=MinLAC, supply is indeterminate because =0 for any output level; only
p<MinLAC yields a determinate solution, q=0. Therefore in this case there is no supply function of
the firm, the firms supply is either zero, or infinite, or indeterminate.
But this fact does not create problems to the theory. All one needs to assume is that if
p>MinLAC the firm will plan to expand productive capacity (i.e. cost-minimizing output) by
expanding or replicating plant, and if p<MinLAC the firm will plan to reduce productive capacity;
but variations of productive capacity take time, and since the firm will not be the only one to take
such decisions, and since it is unlikely that there be perfect synchronization of the decisions of the
several firms (possibly including new entrants), it is legitimate to assume that generally the
expansion or contraction of industry productive capacity will be gradual; thus the short-period
supply curve shifts gradually, and the short-period equilibrium price, determined by the intersection
of the demand curve with the short-period supply curve, tends toward MinLAC[ 38]. If the minimum
average cost is not the same for different firms, the less efficient firms will be eliminated by
competition, and only the firms with the least MinLAC will survive: in the long period, competition
enforces productive efficiency in the sense of minimization of average cost. The total number of
plants will be such as to bring output price as close as possible to MinLAC without falling below it.
The industry supply curve is derived as follows. Let LMC n(q) stand for the horizontal sum of
the long-period marginal cost curves of n identical efficient plants, with q their total output and q*
the single-plant minimum-average-cost output; the supply curve consists of a discontinuous series
of upward-sloping segments yielding a saw-like shape, the n-th segment being the portion of the
LMCn(q) curve corresponding to the semi-open interval [nq*, (n+1)q*). All segments start at a
38

Synchronized decisions to alter productive capacity might cause phenomena akin to cobweb cycles (cf.
Ch. 6 ??) but for long-period decisions such phenomena are less likely because there is time to revise
decisions in the light of information of what other producers in the industry are doing.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 47

height equal to MinLAC; as n increases the segments become flatter, the supply curve approaches a
continuous horizontal line at a level equal to MinLAC. If the downward-sloping demand function
for this good crosses this supply curve more than once, the equilibrium intersection is the one
corresponding to the greatest output at a price not less than MinLAC, qE in Fig. 5.8.

LMC1

demand curve
LMC2

MinLAC
q*

2q*

3q* etc.

qE

Fig. 5.8. Long-period industry supply curve with plants that reach minimum
average cost at output level q*, and a demand curve crossing the supply curve twice. In
this case in equilibrium there is room for 6 plants. The LMC curves are drawn as
straight-line segments only for simplicity.
Let us for example suppose that, at p=MinLAC, the demand for the industry's output falls
between 100q* and 101q*. There is therefore room in the industry for 100 plants, each one
producing slightly more than q*, and therefore having a long-period marginal cost just slightly
above MinLAC. The equilibrium price will be so close to p=MinLAC, that to assume that the
equilibrium price is equal to MinLAC is an excellent approximation. There is no room in the long
run for 101 plants because LMC101(q) crosses the downward-sloping demand curve at a price below
MinLAC, and firms would make losses. The conclusion is that, with perfect replicability of plants,
as long as the output of a single plant is only a small fraction of total output, the long-period
industry supply curve can be treated, to all relevant purposes, as a horizontal straight line (if input
prices are given) at a level equal to minimum average cost.
However, one may feel uneasy with the fact that the theory does not determine the size of
individual firms. We pass to discuss this issue.
5.20. The size and the number of firms
Economists still discuss on the issue of the long-period equilibrium size of competitive firms.
When there is product differentiation, the size of a firm is limited by the market for its product(s),
the question becomes what determines the size of that market, and answers are not difficult to find:
the quality of the product relative to the tastes of consumers, if it is a consumption good; the

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 48

number of firms that use technologies needing that product, if it is a productive input; the extent of
competition from rival products, and the marketing strategies of the firm and of its rivals. Now, it is
important to understand that perfect product homogeneity is a very rare occurrence; usually
location, or the importance for customers of past tradings that have built confidence in their usual
suppliers, are sufficient to render it costly for a firm to extend its sales by subtracting customers
from other firms, although often the divergence from perfectly competitive behaviour remains small
enough that one may continue to apply the long-period theory of the perfectly competitive industry.
In other cases the size of the firm is determined by the nature of the product: rock bands are firms
too, and their product requires a certain size of the workforce; there is no possibility of replicating
the plant identically within the same firm. Similar considerations apply to all firms based on a
strict, creative interaction among few persons. Apart from these cases, one finds the disagreement
among economists mentioned in 5.6.3, with some (e.g. Edith Penrose) arguing that in many
instances firms are able to grow to enormous sizes without any increase in average cost and
therefore the limits to size must be found either on the demand side or on the need for own capital
or collateral, and others arguing that the general case is U-shaped LAC curves because of coordination difficulties that increase with size.
In the discussion whether LAC curves are U-shaped or not, we find here the second meaning
of returns to scale mentioned in 5.4, returns to the scale of total cost or, briefly, (scale) returns to
cost, obviously a notion that assumes given input prices. These returns are defined by the elasticity
of output to total cost, and need for their definition neither that all inputs be increased in the same
proportion, nor differentiability of the production function, nor divisibility of all inputs; as total cost
increases, there may well be discontinuous changes in the quantities employed of some inputs, e.g.
some capital goods may be replaced by capital goods of a different type, the fixed plant may
change, or fixed plants may be indivisible and be discretely increased from one to two, three etc.;
there may then be some discontinuities in maximum output as total cost increases, and the point
elasticity of maximum output to total cost will not be defined at those points; but everywhere else,
and everywhere for discrete changes in total cost, the elasticity of output to cost will be well
defined, and therefore returns to cost is a more general notion than technical returns to scale. When
returns to cost are constant, output increases in the same proportion as total cost, and therefore
average cost is constant; when returns to cost are increasing, successive increases in total cost yield
increasing returns i.e. bigger and bigger increases in output, so average cost is a decreasing function
of output; when returns to cost are decreasing, average cost is an increasing function of output.
Constant returns is also used for the case when MinLAC is reached only for the optimal outputs

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 49

corresponding to replication of indivisible plants. Scale economies is another term used to indicate
increasing returns to cost.
Assuming now that maximum output is a continuous function of total cost and that the LAC
curve is U-shaped, MinLAC is reached where the returns to costs thus defined, in passing from
locally increasing to locally decreasing, are locally constant. If the production function is
differentiable with respect to all inputs, then the locally constant returns to costs at MinLAC, that is,
the equality of average and marginal cost, imply locally constant technical returns to scale[ 39] and
therefore imply that the payment to each factor of its marginal revenue product exhausts revenue if
p=MinLAC. Thus the fact that the long-period cost curve is U-shaped entails no contradiction
between assuming zero profits of competitive firms in equilibrium, and assuming that each factor is
paid its marginal revenue product.
If the firm's long-period AC curve is U-shaped, the firm's dimension is no longer
indeterminate: profit is maximized when LMC(q)=p. When this is the case, the long-period industry
supply curve is derived in the way already shown, with firms replacing plants in the reasoning:
the number of firms is endogenous, because competition also means free entry, and in the long
period there is time for entry. The conclusion is again that, as long as the minimum optimal
dimension of firms is small relative to total industry output, if factor prices (factor rentals) are given
then to all practical effects the long-period supply curve of the industry is horizontal at a price equal
to minimum average cost.
But even when the U-shaped cost curve is not accepted, the supply curve of the industry
remains horizontal at the MinLAC level; the size of the firms composing the industry is then simply
irrelevant. One can for example take it as determined by historical accidents, or by limits to the
growth of individual firms deriving from limits to possible indebtedness.
In conclusion both when there are, and when there arent, constant returns (to cost) by firms,
the assumption of competition with free entry implies an essentially horizontal long-period industry
supply curve once input prices are given, as long as the minimum quantity that allows a firm to
minimize average cost is small relative to the total demand forthcoming at a price equal to that
average cost. When this is the case, since in the long period competition eliminates inefficient firms,
one can assume a common technology within the industry. We can therefore treat the industry's
39

We prove this for the two-factors case. At the point of minimum average cost it is MC =

v1
==
MP1

v1
v2
v1 x1 v2 x2
AC
v
MP2
f ( x1 , x2 ) ; this can be re-written f(x1,x2) = MP1x1+MP1 2 x2 = MP1x1+MP2x2 , which
implies that the production function is locally homogeneous of degree 1.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 50

long-period aggregate production function as exhibiting constant technical returns to scale even
when we cannot do so for single firms.
This conclusion will allow us, when in Part III of the chapter we formulate the neoclassical
competitive general equilibrium with production, to assume product prices equal to minimum
average costs, and that at those prices supply adapts to the demand forthcoming at those product
prices (and at the factor prices that determine them).
So far in this long-period analysis we have taken all input prices as given. In this case, except
for flukes only one production method will be the cost-minimizing one for the production of each
product. However, when we come to determining factor rentals endogenously, it may well happen
that two (or more) methods will co-exist in the production of the same product: the different factor
employments need not imply differences in average cost if factor rentals, for some of the factors,
adapt so as to ensure the same average cost with both methods. One typical such case is that of
extensive differential land rent: when the same product is produced on lands of different fertility, a
differential rent will arise on more fertile lands, that will make the utilization of different lands
equally convenient. Here we need not add to what was said on this topic in Chapters 1 and 3.
5.21. Aggregation
5.21.1. When the number of firms in an industry is given, and for each firm a profit function,
and therefore a supply function, exists, then the industrys competitive supply can be determined as
if forthcoming from a single multiplant price-taking firm that operates all the individual production
functions[40]. Formally, the aggregability condition is that, given the individual production
possibility sets (whose elements are netput vectors) Y1, ... , Yj, ... ,YJ of the individual firms (where
J is the number of firms), the aggregate production possibility set be
Y = Y1 + ... + YJ = {yRn: y = j yj for some yjYj, j=1,...,J}.
In words, the aggregate firm must have no additional production possibilities at its disposal
beyond a simultaneous activation of the production processes available to the individual firms, at
most one per firm. In this case, the aggregate firm can do no better in terms of profits than the sum
of the individual firms' profits, because it can do no better than copy what the individual firms
40

This is true as long as the possibility is excluded that a single management of all the factors of the
individual firms (including the fixed factors which need not explicitly appear in the short-period production
functions) would achieve cost reductions. For example, the fusion of five small farms into a single big farm
might permit the utilization of big agricultural machinery which was uneconomical for each individual farm;
or there might be unnecessary duplication of some indivisible factors (for example, each separate farm might
need to buy its own tractor if no sharing is allowed, while four shared tractors would suffice for the five
farms). Only when one excludes such phenomena can one conclude that the industry behaves in the same
way as if a single firm were to operate all the individual production functions.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 51

would choose autonomously.


When this is the case then, conversely, the firms in the industry maximize aggregate profits;
since profit maximization requires cost minimization, this also shows that the allocation of the
industry output among the several firms in the industry is cost-minimizing. Indeed, since for each
firm output price equals marginal cost, marginal cost is the same in all active firms, which is an
obvious efficiency condition: if marginal cost were different in two (active) firms, a small transfer
of production to the firm with the lower marginal cost would decrease aggregate cost. Analogously,
for each variable (and hence, transferable) factor the marginal product will be the same in all firms
where the factor is used, because equal to the factor rental divided by the output price( 41), again an
obvious efficiency condition: if the marginal product of a factor were not the same in all firms (and
lower in the firms not using it), transferring a small amount of the factor to the firm where it has the
greater marginal product would increase total production. These considerations show that the
aggregate firm obeys the conditions for profit maximization when each individual firm does.
5.21.2. In long-period analysis with free entry, again the industrys behaviour can be derived
as coming from the decisions of a single firm, a constant-returns-to-scale firm which, for each
vector of factor rentals and each output level, adopts the factor employments corresponding to
minimum average cost[42]. Competition eliminates less efficient firms and thus causes the
production function to become the same for all firms. If the single firms have a CRS production
function, then the industry acts like a giant firm with that same production function. If the
individual firms production function yields U-shaped average cost curves, then (assuming a
sufficiently small minimum efficient size relative to aggregate output) because of the possibility of
replication of plants or firms the industry acts like a single CRS firm, with a production function
which, for each vector of relative factor rentals, yields the same optimal factor employments per
unit of output as the average-cost-minimizing choice of the individual firms.
We illustrate with a numerical example. The firms production function is q = 2(1+x 11x21)1 .
2
If both factors are multiplied by a scalar t, we obtain q(t) = 1 1 ; it is convenient to put
t 2 x1 x2
x1x2=1/A; then q(t) = 2t2/(t2+A); and the scale elasticity of output (remember that it is evaluated at
t=1) is e=2A/(1+A) which is 1 according as A1. Thus for each given factor proportion this
41

Assuming that marginal products can be defined. The reader is reminded that factor rentals must equal
marginal revenue products.
42
Here as elsewhere in this chapter it is assumed that the average-cost-minimizing factor proportions are
uniquely determined.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 52

function produces a U-shaped cost curve, because it exhibits locally increasing returns to scale
when x1x2<1, locally decreasing returns to scale when x1x2>1, and locally CRS (and minimum
average cost) when x1x2=1 i.e. when q=1. Therefore in the long period each firm produces one unit
of output, and the isoquant map of the industrys long-period production function is the radial
expansion of the isoquant x2=1/x1 associated with q=1; in order to know how much an industry
input vector (x1,x2) produces, we must view it as t times the vector (x 1*,x2*) that produces 1 unit of
output with the same factor proportion as (x1,x2), where t is determined by x1=tx1*, x2=tx2* such that
x1*x2*=1; in other words, inputs (x1,x2) produce t units of output where t 2 = x1x2. Hence the
industrys production function is Q =

x1 x2 . The reader can check that it has CRS[43].

We can use this example to clarify a formal problem arising with U-shaped cost curves of
individual firms in long-period analysis with free entry. Suppose that factor rentals are v 1=v2=1;
then in this example MinLAC=1. Suppose that the demand curve is decreasing, and at the product
price p=1 demand is 100.5 units. There is room for 100 firms; if there were 101 firms, price would
go below 1 and all firms would make negative profits. But with 100 firms the equilibrium price is
slightly above 1 and profits are positive; hence, if we assume that firms enter as long as
p>MinLAC, there will be entry; no equilibrium exists if we define it as simultaneously requiring
rigorously demand=supply and profits=0. However, if the aim is to determine a long-period
equilibrium (the average situation around which the economy oscillates), this is not a problem
because even if firms do enter and bring the total number of firms above 100, the number will
subsequently decrease, and we can still assume that the average around which the price oscillates is
1. Long-period equilibrium only aims at determining the average around which actual market
variables oscillate. (Furthermore it is plausible that potential entrants try to make an estimate of the
effect of their entry, and generally they will realize that it is likely that the small profits associated
with 100 firms will disappear if they enter, and hence will not enter.)

43

To render explicit the industrys production function starting from the firms production function is not
always possible, but, as the example indicates, one procedure to find the amount produced from the amounts
of factors employed by the industry is as follows. For simplicity assume only two factors. Let f(x 1,x2) be the
individual firms production function yielding a U-shaped LAC curve; let x 1,x2 be the given industry inputs;
let Bx2/x1; find x1* such that f(x1*,Bx1*) minimizes average cost; let q* f(x 1*,Bx1*); let tx1/x1*; then the
industry output is Q=tq*. Exercise 5.3: assume that the individual firms production function is q = x 1x2
(x13/2x23/2)/3 and B=9; find x1*, q*, and Q if the industrys employment of input 1 is x1=100.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 53

PART II
5.22. PARTIAL EQUILIBRIUM
5.22.1. As a first step toward the study of the marginalist/neoclassical competitive general
equilibrium of production and exchange, we discuss in greater detail the construction that, more or
less explicitly, we have been using in this chapter (for example in Fig. 5.8?? and in the text
discussing it): the determination, via the intersection of a supply curve and a demand curve, of the
so-called particular equilibrium, as it was originally called, or (its current usual denomination)
partial equilibrium, of a single market studied in isolation. The prices and quantities on other
markets are taken as given and are considered essentially unaffected by changes in the market under
study: this is called the assumption of coeteris paribus (latin for 'other things remaining the same').
A famous 1926 article describes this approach as follows:
This point of view assumes that the conditions of production and the demand for a
commodity can be considered, in respect to small variations, as being practically
independent, both in regard to each other and in relation to the supply and demand of all
other commodities. It is well known that such an assumption would not be illegitimate
merely because the independence may not be absolutely perfect, as, in fact, it never can be;
and a slight degree of interdependence may be overlooked without disadvantage if it
applies to quantities of the second order of smalls, as would be the case if the effect (for
example, an increase of cost) of a variation in the industry which we propose to isolate
were to react partially on the price of the products of other industries, and this latter effect
were to influence the demand for the product of the first industry. (Sraffa 1926, p. 538)
The partial equilibrium approach can also be used for the study of imperfectly competitive
markets, for example a monopolistic market.
Sometimes it may be legitimate to isolate not one, but two (or perhaps even more)
interdependent markets: one example is the study of the effects of changes of the supply conditions
and hence of the price of one product on the demand and hence on the equilibrium price of a
complementary or of a substitute product; another example is the determination of the supply curve
of an industry that uses a specialized factor, whose rental rises when, owing to a rise in the demand
for the industrys product, the industrys output rises.
5.22.2. The partial-equilibrium supply curve of a competitive industry can be a long-period or

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 54

a short-period one. The long-period supply curve is horizontal if factor rentals are given[ 44]. The
short-period supply curve is upward-sloping even with given factor rentals, owing to the given
amounts of some factors. Their derivation has been illustrated already; we add some considerations
on their legitimacy and on the analogy between them.
Besides needing the price-taking assumption, the use of a partial-equilibrium long-period
supply curve is legitimate in two cases. The first one is when the industry uses only a small fraction
of the total supply of each one of the factors of production it utilizes; then a variation in the quantity
produced by that industry will not exert an appreciable influence on its factors rentals, because it
will cause a very small percentage variation of the demand for them. As a result, factor rentals can
be taken as given. If the influence on the rental of some factor were significant, this would alter the
cost conditions of other products too and then their prices would change, rendering the coeteris
paribus assumption illegitimate.
The second case is when there is a specialized factor demanded only by the industry one is
isolating. This might be for example a special type of land indispensable to (and only demanded
for) the production of one product, say a famous wine. In this case the specialized factors rental is
determined endogenously; it will rise as demand for the product rises, so as to maintain the
producers profit at zero; the marginal product of the remaining factors (whose rentals are given)
decreases as output increases; the supply curve is upward-sloping. The independence between
supply curve and demand curve, necessary for partial equilibrium analysis, additionally requires
that the changes in the incomes of the owners of the specialized factor do not appreciably influence
the demand for the product of the industry under analysis.
Marshall generalized this second case to include cases where the supply to the industry of
some specialized factors, differently from the supply of specialized land, is variable in the long
period because those factors are produced factors, but it varies sufficiently slowly relative to the
supply of the other factors for it to be treated as given in shorter-period equilibration processes.
Some types of fixed plants may indeed take a long time to be built; this authorizes treating their
supply as fixed for equilibration processes on time horizons of, say, a few months. (As pointed out
earlier, what is needed for the determination of the short-period supply curve and hence for shortperiod partial equilibrium analysis is that the supply of fixed factors to the industry, not to each
firm, be given.)
The supply curve is valid only for comparatively small variations of the quantity produced,
44

This assumes that if there are profits to be made, there will always be someone (existing or new firms)
ready to add further plants in the industry. This assumption appears strongly confirmed by experience when
adequate account is taken of barriers to entry, that we are assuming absent here and will be discussed in
Chapter 11.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 55

because for considerable variations the coeteris paribus condition becomes doubtful[ 45]. Particular
difficulties arise when one attempts comparative statics of partial equilibria of a capital good: a
change in the supply conditions of a capital good (due for example to the discovery of a new,
cheaper production method, or to a tax) alters the costs and hence the prices of all goods for whose
production that capital good is used, and many of these may be in turn inputs to the production of
that capital good, so the supply curve of the capital good shifts for more reasons than the direct
effect of the first change[46]; and the derivation of the demand curve is not easy to conceive.
5.22.3. Alfred Marshall attempted to argue that the long-period partial-equilibrium supply
curve of a product can also be downward-sloping, owing to two effects of increases of the
dimension of an industry: first, the possibility better to exploit scale economies; second, an increase
in external effects or externalities. He was thus trying to make room in the theory of partial
equilibrium for a phenomenon no doubt often observed, an association between increase in
production and decrease in price of a product produced by an industry where it was difficult to deny
the existence of competition among producers. But in two articles, in 1925 and 1926, Piero Sraffa
showed that the decreasing supply curve is incompatible with competitive partial equilibrium. He
remembered that the existence of unexploited scale economies is incompatible with competition
with undifferentiated products, because competition requires firms to be rather small relative to total
industry demand, and the perfect substitutability for the buyer among the products of the different
firms in the industry implies that any small price reduction by a firm will attract to the firm enough
buyers to make it able to sell the increased output that allows the exploitation of scale
economies[47]. As to externalities, he noticed that the positive external effects due to increases of
economic activity, e.g. greater ease in finding repairmen or transportation firms or skilled workers,
45

Marshall Principles p. 384 fn. of the original 8th edition (1920; p. 318 fn. in the after-1949 reset
editions): the ordinary demand and supply curves have no practical value except in the immediate
neighbourhood of the point of equilibrium. The reason is partly different for demand curves, cf. below in
the text and also ??(consumer surplus).
46
Changes in input use can be sometimes very surprising when these interrelations are taken into account;
their exploration has started only recently and is still proceeding, cf. Opocher and Steedman ??
47
Marshall had argued that unexploited scale economies exist, but require time to be exploited because firms
are slowed down in their expansion by the need for collateral and therefore for accumulated profits, and this
prevents firms from becoming indefinitely large because the founders of successful firms pass the firm to
their children who are much less competent and cause the firm to decline and die; he compared the firms in
an industry to trees in a forest, some of which are growing, while others are dying. This picture is
occasionally confirmed by facts, but nowadays more and more it is the case that firms are owned by many
shareholders and run by hired managers, and a decline due to incompetent owner-entrepreneur heirs is rare;
therefore a decline before scale economies can be exploited is generally implausible; the more so, because
there are more and more giant firms, conglomerates which have the financial potential to set up very large
firms from the start.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 56

are very seldom internal to an industry, they generally concern large groups of firms belonging to
different industries but connected by common location or by similar needed skills; these effects
cannot be admitted in the partial equilibrium analysis of one industry because they extend to other
products, altering their prices, and therefore violating the coeteris paribus condition. Sraffa
concluded, and subsequent economists have admired the cogency of his critique, that competitive
long-period partial equilibrium theory can admit only constant-cost industries, or increasing-cost
industries in the sole case of an industry being the sole demander of a specialized factor. (When the
expansion of an industry affects the rental of a factor also used by other industries, then the costs of
these other industries are affected as much as in the first industry, the prices of the products of those
other industries are relevantly affected, and again the coeteris paribus condition does not hold.) This
does not mean that unexploited scale economies do not exist, it only means that their causes and
effects require a different approach: Sraffa suggested to abandon the assumption of perfect
competition and admit that the general case is rather one of differentiated products, which, if
coupled with free entry, ensures nonetheless a broad tendency of prices toward average costs[48].
5.22.4. The demand curve is a function that specifies the quantity demanded of the product
under investigation as a function of its price. The legitimacy of assuming the existence of a partialequilibrium demand curve requires:
1) Given prices of other goods49; this implies a given income distribution. (This in turn
implies that partial equilibrium analyses cannot study changes in the price and quantity of a product
induced by changes in income distribution.)
2) Given incomes of consumers. This requires, in addition to a given income distribution, a
given level of utilization of resources (in particular, a given level of labour employment): this was
traditionally justified by an assumption of full utilization of resources, the tendential result of the
working of market economies according to the marginal approach, as we know from Chapter 3.
3) Given preferences, unaffected by actual non-equilibrium consumptions.
These three sets of conditions too are subsumed under the expression coeteris paribus. These
givens are, and must be, assumed not to change (or more precisely, to change only negligibly)
48

The reading of both Sraffas articles, the 1925 and the 1926 one??refs in English, is strongly recommended
as they are excellent examples of penetrating reasoning attentive to the economic justifications of theoretical
constructs.
49
It is also possible to consider the demand curve as derived under an assumption that the prices of some
strongly interconnected goods change when the quantity demanded of the first good changes; for example if
one considered probable that a big increase of taxes on gasoline would considerably decrease the demand for
cars, any estimate of the demand curve for gasoline not restricted to the very short period should try to take
into account the effect on the car market, including the possible effect on the price of cars. Fortunately, for
most industrially produced goods the price is rather insensitive to demand, cf. Chapter 11.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 57

during the adjustment processes toward the partial equilibrium, otherwise the equilibrium would
lack the persistence necessary to give a good indication of the average behaviour of the market; it
would be impossible to assume, for example, that the demand curve has the persistence that allows
a monopolistic firm to form a reasonably correct idea of its position and slope (a necessary premise
to the derivation of the marginal revenue curve).
It is then clear that the partial equilibrium method requires that income distribution and
aggregate demand can be considered given (although not necessarily determined according to the
marginal/neoclassical approach), and that preferences can be considered sufficiently unaffected by
the disequilibrium adjustments. On this last issue, we have remembered at the beginning of chapter
4 Marshalls own admission that experience irreversibly affects tastes. We may add here that when a
price changes significantly, obliging consumers to a relevant change in consumption habits, it seems
plausible that people will not know in advance how their own behaviour is going to change; they
will experiment and discover and develop new consumption habits that could not be predicted from
past evidence, and that often can only be described as due to a formation, or discovery, of
preferences until then undefined[50]. As a consequence, the assumption of a well-defined demand
curve for prices different from the prevailing one may be questioned, as again admitted by Marshall
himself (cf. above ch. 4 4.22): hence the assumption (that will be met frequently in Chapter 11)
that firms know the demand curve facing them must be treated with suspicion[51].
But we must introduce the reader to the dominant analyses, so we do not further question the
notion of partial-equilibrium demand curve; however, let us not forget that this notion can be
considered reasonably well defined only for consumption goods (and perhaps for some non-basic
capital goods), only for rather small departures from the until then prevailing price, and only as long
as the incomes of consumers (hence income distribution and the aggregate level of activity of the
economy) are given.
5.23. Stability of partial equilibria.
50

The dependence of preferences on experience can be used for a further criticism of the marginal
approach. It is not only that if consumption of a certain good has never been experienced, the preference for
it cannot but be vague, and open to modification by experience. There is also the fact that repeated
experience can permanently alter preferences (e.g. people can develop a taste for listening to certain kinds of
music, for drinking good wine, for practicing certain sport activities). Now, whether and how many times a
good is experienced can depend on prices. This questions the assumption of preferences independent of
prices, on which the marginalist/neoclassical determination of equilibrium is based.
51
Firms must have had the possibility to explore how demand depends on price in an economic situation
undergoing very little change in the variables impounded in the ceteris paribus clause. This may be an
acceptable assumption in some cases, but not in many other ones; for the latter cases, one will need theories
explaining firm behaviour without an assumption that there is a well-defined and known demand curve.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 58

5.23.1. If the good is homogeneous (undifferentiated) then on average all units of the good
must sell at the same price. Of course this can only be approximately true but, as already pointed
out several times, the equilibrium can only aim at describing the average resulting from the trialand-error higgling of the market. We can speak therefore of the price of the good.
Given a supply curve and a demand curve, that represent supply price and demand price as
functions of the quantity of the good[52], equilibrium obtains where aggregate demand for the good
equals aggregate supply, or equivalently where supply price and demand price coincide.
Let us examine the stability of equilibrium. Apart from the implausible case of Giffen goods
the demand curve for a consumption good can be assumed to be downward-sloping. (For capital
goods the issue is more complex, and as argued earlier the partial equilibrium method is seldom
acceptable, but an argument can still be put forward to the effect that an increase in the price of a
capital good, with other factor prices given, will tend to reduce the demand for that capital good,
both because of technical substitution, and because of the rise in cost and hence in relative price of
the consumption goods using that capital good as an input.) The supply curve is either horizontal, or
upward-sloping[53]. In the latter case the stability of equilibrium is clear, under the assumption that
price tends to rise if demand exceeds supply, and tends to decrease if supply exceeds demand. When
the supply curve is horizontal it is a long-period supply curve, and the adjustment goes on in a
succession of short-period situations, in each one of which the number of plants is given and the
supply curve is a short-period, upward-sloping one. The short-period equilibrium price is stable, and
if higher than minLAC it induces in the long period an increase in the number of plants in the
industry, that is, a shift of the short-period supply curve to the right that causes the short-period

52

The Marshallian preference for considering price the dependent variable thus supply price is the price
necessary to induce a given supply to be forthcoming, and demand price is the price that induces a given
demand has the advantage that one can still speak of a supply function even when the supply curve is
horizontal.
53
Actually, Marshall considered at length the possibility that the long-period supply curve of a product be
decreasing, owing to economies of scale achieved by an average increase of firm dimension with the growth
of industry size, or owing to cost reductions due to economies of scale external to the firm but internal to the
industry, a special case of externalities. The first cause was soon judged incompatible with the perfect
competition assumption, that must assume that in the long period firms are at the minLAC size; what
Marshall was implicitly admitting was demand limits to the expansion of individual firms, and this can only
be discussed by abandoning price-taking and turning to theories of imperfect competition. An example of the
second cause might be the increasing average skill of specialized skilled labour when the dimension of an
industry grows and with it grows the number of workers who have acquired high skills owing to work
experience, and the result is a community where expertise is greater, with a reduction in costs. But such
externalities can only act very slowly, on a time scale superior to that of the adjustments contemplated in
short-period or long-period equilibration: the time scale one considers when one discusses economic growth.
Furthermore externalities external to firms but internal to an industry are very rare, generally positive
externalities do not reduce costs only in one industry and are therefore incompatible with partial equilibrium.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 59

equilibrium price to decrease and thus to tend toward minLAC[ 54]; the reverse process will go on if
the short-period equilibrium price is lower than minLAC. Thus we obtain stability of the longperiod equilibrium too.
5.23.2. The analysis suggests that it seems legitimate to assume a tendency of the quantity
produced of the several products to adapt to the demand for them, at prices that tend to equal
minimum average cost. In Chapter 6 it will be seen that the consideration of adjustment lags can
raise doubts on this conclusion, but it will be argued there that the difficulties are not very serious.
So we have some justification for believing that the assumption that will be made in the formulation
of the general equilibrium equations in Part III of this chapter, of equilibrium product prices equal
to minimum average costs and of quantities produced equal to the demand for them at those prices,
reflects actual tendencies. However, the stability of product markets thus assumed rests on given
factor prices and given demand curves; therefore it does not prove the stability of the general
equilibrium of production and exchange, which requires in addition the stability of factor markets
(and, to such an end, cannot assume given demand curves for products because changes in factor
rentals change incomes and demands). This will be discussed in chapter 6.
5.24. Welfare analysis
5.24.1. Let us now prove that in a partial-equilibrium framework the competitive equilibrium
of a single market, determined by the intersection of demand curve and supply curve, is a Paretoefficient allocation. But we must clarify, allocation of what? Of the quantity produced, x, among
consumers, and of income among consumers and producers, under a trade-off (analogous to a
production function) between income (=cost) and x. We consider two groups of maximizers:
consumers maximize utility, producers maximize their income i.e. profit. The given prices of all
other goods allow their treatment as a Hicksian composite commodity, whose price can be made
equal to 1 and whose quantity therefore can be legitimately identified with expenditure on goods
other than x, or residual income y; thus consumer hs utility depends on x h and yh, and when a
consumer pays p for a unit of x-good she is giving up p units of y-good; the producer gives away
amounts of y-good as cost to produce the unit of x-good; thus it is as if x were produced by using
54

Note that for the process to push the price toward minLAC it is not necessary that all firms have optimal
plants; it suffices that there be entry of optimal plants until supply obliges the price to equal minLAC; this
process will generally be faster than the process of closure and replacement of older plants (the economic life
of plants and durable capital goods is generally much longer than the time required by their production), so
one must consider it normal that a price very close to minLAC co-exists with plants that are not optimal and
earn residual quasirents.

f petri

Adv Micro

chapter 5 firms and productionGE

p. 60

10/03/2016

quantities of y as input. In equilibrium, consumers have the same MRS between x-good and y-good;
through an opportune choice of units for utility the equilibrium marginal utility of income can be
rendered equal to 1 for each consumer. This means p*, the equilibrium price of x, equals the
marginal utility of x for all consumers active on the market; producers are interested in maximizing
their income i.e. profit, so they equalize price and marginal cost; hence MC(x*)=MU(x*) where x*
is equilibrium output.
Proof of the Pareto efficiency of the perfectly competitive partial equilibrium. We first prove that
production of a quantity different from the equilibrium quantity x* cannot be Pareto efficient. Pareto
efficiency means that a Pareto improvement (a change that makes somebody better off without making
anybody worse off) is impossible. If x<x*, there will be some consumer ready to pay for one more unit of x
a demand price pd greater than the equilibrium price p*, and there will be some producer who can produce
one extra unit at a marginal cost not greater than p* (we admit the possibility of CRS and constant MC) and
would be therefore ready to sell it at a supply price p sp*, hence these two agents can strike a mutually
advantageous bargain (to produce and exchange one extra unit at any p intermediate between p d and ps)
without making anybody else worse off. If x>x*, there will be some producer with a marginal cost not
smaller than p* who is ready to pay a sum greater than or equal to p* for the right to reduce production by
one unit, and there will be some consumer ready to renounce one unit of x for a recompense less than p*,
hence again these two agents can strike a mutually advantageous bargain. Thus x=x* is a necessary condition
for Pareto efficiency.
Let us now prove that no different allocation of the production of x* among producers can be a Pareto
improvement, and no different allocation of x* among the consumers can be Pareto efficient. A different
allocation of the production of x* among firms either leaves marginal costs unchanged, or causes an increase
in marginal cost in at least one firm, whose profit decreases: in either case there isnt a Pareto improvement;
in the first of these two cases Pareto efficiency does not uniquely determine the allocation of the production
of x* among firms. A different allocation of x* among consumers means that at least one has more x-good
and at least one has less x-good than in equilibrium; this means that their MRSs differ so they can strike a
mutually advantageous exchange.

Consumer and producer surplus and welfare changes.


5.24.2. We have seen in chapter 4 the definition of consumer surplus in an industry.
Producer surplus is analogously defined as the area above the supply curve up to the horizontal
price line, cf. the triangle ADC in Fig. 5.10. It is a more complex notion than consumer surplus. It
intends to measure the maximum amount of money that producers, that is, the ensemble of
entrepreneurs and factor suppliers involved in producing the good, would be ready to pay in the

f petri

Adv Micro

chapter 5 firms and productionGE

p. 61

10/03/2016

aggregate rather than forgo the possibility to produce and sell the good at the given price. It too
assumes a constant marginal utility of income.
price
B

Industry supply curve

competitive price p*

demand curve

quantity

Fig. 5.10. Marshallian total surplus is the area of triangle ABC, the sum of consumer surplus (the area of
triangle DBC) and producer surplus (the area of triangle ADC).

MC

AVC

p, MC, AVC

p
A

D
B

O
C

output

Fig. 5.11. Graphical proof that for a firm the area above the supply curve up to the price line (trapeze
ABDp) equals revenue minus variable cost. When price is p and therefore the supply of the firm is OE,
revenue is rectangle OEDp, and the area under the supply curve OABD of the firm equals total variable cost,
because the area of rectangle OABC is variable cost up to output OC, and, by integration, the area under the
marginal cost curve from B to D is the addition to variable cost caused by increasing output from OC to OE.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 62

In the short period, the producer surplus of a single firm at a given product price and
quantity supplied is defined as total revenue minus total variable cost or equivalently, as pure profit
plus total fixed cost. (Remember that fixed cost is by definition a cost that the producer must bear
whether she produces or not. Thus in the short period the entrepreneur finds it convenient to
produce as long as she is able to more than cover variable cost.) That this is equivalent to the area
above the short-period supply curve up to the horizontal price line is easily proved by remembering
that total variable cost is the integral of marginal cost and therefore it is the area under the supply
curve, cf. Fig. 5.??. Considering fixed cost to be a sunk cost which cannot be avoided, the
entrepreneur, rather than be excluded from the market, will be ready to pay up to the amount that
would leave her with what is just sufficient to cover variable cost, amount which is measured by
that area. The sum of these areas is the area above the industrys supply curve.
In long-period analysis all cost is variable; if the industrys supply curve is horizontal
because all factor rentals are given, producer surplus is zero. But if the industry uses a specialized
input whose rental rises with industry supply, the long-period industry supply curve is upwardsloping, so producer surplus is positive. However, profits are zero all the same, because at each
point of the supply curve the specialized inputs price is given and each firm treats input prices as
given and produces the quantity that minimizes average cost, i.e. that causes average and marginal
cost to coincide[55]. How is the positive producer surplus reconciled with the zero profits? The point
is that the rise in the rental of the specialized factor due to the expansion of production implies an
income gain for the suppliers of that factor, so they would be ready to pay rather than see production
of the good forbidden. The maximum amount they would be ready to pay, again misleadingly called
producer surplus (it is in fact a consumer surplus of the consumers who supply the factor), is
actually measured by the area above the factors supply curve (in the graph with factor supply and
factor rental on the axes) for reasons similar to those defining the consumer surplus on the demand
side[56]; for example if the supply of the factor is rigid (implying a zero reservation price of the
factor owners), the producer surplus is the area of the rectangle formed by the axes, the vertical
supply line, and the factor rental horizontal line. This area coincides with the area above the
industrys long-period supply curve because the area below the latter curve is the payment to the
other factors: this is made clear by noticing that, if total production came from a single firm with the

55

Thus a long-period industry supply curve has average and marginal cost coincide at all points even when it
is upward-sloping.
56
Again, under the assumption of constant marginal utility of money less easily justifiable in this case! It is
very unusual that income from ownership of a specialized factor of production be a very small part of one's
overall income.

f petri

Adv Micro

chapter 5 firms and productionGE

p. 63

10/03/2016

specialized input as fixed factor, the supply curve would be the marginal cost curve of this firm, and
the area under it would measure its variable cost.
Proof. We prove it for a simple case. Assume that the good is produced by unspecialized labour L,
whose wage w is constant and for simplicity equal to 1, over a specialized land T in rigid supply, whose
surface, again for simplicity, is measured in such units as to make it equal to 1; the CRS differentiable
production function is q=f(L,T); each marginal product is a decreasing function of the ratio of the factor to
the other factor, and becomes negative above a sufficiently high ratio, in which case as explained in ch. 3 the
factor will not be fully utilized, only the amount yielding a zero marginal product will be utilized. Land is
fully employed if possible, so we consider q a function of L only, q=f(L) which we assume invertible, L(q)=f 1

(q) whose derivative is 1/MPL(q). Optimal labour employment requires w=MP Lp where p is the product's

price; since w=1 is constant, it must be p=1/MPL, and an output increase requires p to rise if MPL is
decreasing; this causes the land rental to rise too, because land receives its marginal revenue product too,
and MPT rises as labour employment rises, except initially when land is not fully utilized and its marginal
product is zero. Thus is a function of q, =(q), and initially it is zero: the opportunity cost of land T is
zero, the entire land revenue is 'rent' or producer surplus. At each q, supply price is defined by equality with
average cost, p(q)=((q)+L(q))/q, and the function p(q) thus defined is the supply curve. We want to show
that, given the quantity produced q*, total industry revenue p(q*)q* minus the area under the supply curve
equals (q*). So what we need to show is that the area under the supply curve is L(q*). Now, p also satisfies
q*
1
p(q)=1/MPL(q); therefore the area under the supply curve is the definite integral
=L(q*)
MP
(q)
L
0

L(0)=L(q*).

dq

Exercise: Assume wine V is produced by labour L over 1 unit of specialized land in fixed supply
according to the production function V=T1/2L1/2. Labours wage in terms of other goods is fixed and equal to
w. Prove that the industrys long-period inverse supply curve is p=2wV and that the area below it for any
given V equals the wage payments to the labour employed to produce that V.

Actually, short-period producer surplus in a competitive industry is determined in the same


way, because as argued in 5.14, the entrepreneurs' short-period profits should be seen as earnings
accruing to the fixed factors.
If for some reason (e.g. rigid factor contracts stipulated in the past under different
conditions) there are profits, then this is a subtraction of part of the surplus from the factor
suppliers, but the area above the marginal cost curve still measures producer surplus, only now
consisting partly of entrepreneurial profits and partly of factor suppliers' surplus.
Marshallian aggregate producer surplus, or simply producer surplus for brevity, is the sum
of the individual producer surpluses in a market. And the sum of consumer and producer surplus is
called the Marshallian aggregate total surplus. It is the area of the triangle formed by demand
curve, supply curve, and ordinate axis.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 64

It is easy to see graphically that the competitive partial equilibrium maximizes total surplus.
Any price or quantity different from the equilibrium ones would ration either buyers or sellers,
because there would not be enough production or enough demand.
It is important to be aware of the limits of the above conclusion: the argument bringing to it
has neglected externalities, has taken incomes and factor property as given, and concerns the market
for a consumption good.

f petri

Adv Micro

chapter 5 firms and productionGE

10/03/2016

p. 65

c EXERCISES
5.11 The production function is Cobb-Douglas with CRS, q=Axy1-. Let factor prices be vx
and vy. Find the cost function C(vx,vy,q) and confirm Shephards Lemma by showing that indeed
C/vx=x.
5.12 Explain rigorously why constant returns to scale and divisibility imply that isoquants
are convex.

Вам также может понравиться