You are on page 1of 22

Schumpeterian Competition

Author(s): Carl A. Futia


Source: The Quarterly Journal of Economics , Jun., 1980, Vol. 94, No. 4 (Jun., 1980), pp.
675-695
Published by: Oxford University Press

Stable URL: https://www.jstor.org/stable/1885663

REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/1885663?seq=1&cid=pdf-
reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Oxford University Press is collaborating with JSTOR to digitize, preserve and extend access to
The Quarterly Journal of Economics

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
SCHUMPETERIAN COMPETITION*

CARL A. FUTIA

This paper describes a stochastic model of the process of competition via tech-
nological innovation as it might occur within a single industry. Individual firms un-
dertake R&D projects in the hope of acquiring a decisive competitive advantage over
their rivals. But such advantages and the economic rents arising from this are only
temporary; they eventually disappear in the face of imitation, entry, and innovation
by other firms. At the industry's long-run equilibrium, concentration and the pace of
technological innovation are jointly determined by the conditions of entry and the
extent of innovative opportunity. The model implies relationships among these vari-
ables that have in fact been detected in the empirical R&D literature.

This paper is a theoretical exploration of the relationship between


market structure and the rate of technological innovation within
various industries. In the following pages I shall construct a dynamic
model of the Schumpeterian process of competition via technological
innovation. Although highly stylized, this model rests upon explicit,
choice theoretic foundations. It is capable of explaining a diverse
number of empirical regularities that describe the relationship be-
tween concentration and innovation.
Economists have expended considerable research effort in an
attempt to unravel the connections between market structure and the
rate at which technological innovation takes place. The reader should
consult Kamien and Schwartz [1975] for an excellent survey of the
literature. Although much empirical work has been done, theoretical
progress on this problem has been slow. (See Dasgupta and Stiglitz
[1977], Loury [1976], and Nelson and Winter [1976] for some very
interesting recent contributions.) I believe that the theoretical model
discussed in this paper achieves more success in its confrontation with
the qualitative aspects of the empirical evidence than has been
achieved to date.
Before discussing the structure of the model, I would like to list
some of the empirical facts that are its qualitative implications. I must
first warn the reader that some experts would object to my use of the
term "fact" when referring to these regularities. In addition to the
usual disagreements about data quality and statistical methodology,
there are also serious disputes about the nature of the "true" struc-
tural model generating the data. For example, Schmookler [1966]

* This paper represents the views and assumptions of the author, not necessarily
those of the Bell System.

? 1980 by the President and Fellows of Harvard College. Published by John Wiley & Sons, Inc.
The Quarterly Journal of Economics, June 1980 0033-5533/80/0094-0675$02.10

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
676 QUARTERLY JOURNAL OF ECONOMICS

would dispute the importance of any measure of technological op-


portunity in determining innovation rates.
Still, there is enough agreement to make these regularities worth
noting. They are, in any event, implications of the model developed
in this paper.
First of all, there appears to be a positive, statistical correlation
between measures of aggregate industry R&D input (relative to sales)
and measures of industry concentration. On the average, higher levels
of industry expenditure on R&D inputs are associated with higher
levels of industry concentration. Second, this positive correlation
between concentration and innovative input appears to decline as
measures of the extent of innovative opportunity open to the industry
increase. For example, one can separate industries into a high op-
portunity cohort and a low opportunity cohort according to the facility
with which products can be differentiated via innovation [Comanor,
1967]. One finds that the high opportunity cohort exhibits a smaller
positive correlation between R&D expenditure and concentration
than does the low opportunity cohort. Third, industries with greater
innovative opportunities tend to be more concentrated. And finally,
measures of industry R&D input relative to industry sales appear to
achieve their maximum values at moderate levels of industry con-
centration rather than at very high or very low concentration
levels.
For a summary of these and related results, as well as for refer-
ences to the original papers, the reader should consult Kamien and
Schwartz [1975], Section VI; and Scherer [1970].
In the following pages I shall develop a dynamic, stochastic model
of the process of technological innovation within a single industry.
I shall show that the industry structure converges (in a stochastic
sense) to a long-run steady state. At the long-run equilibrium, both
industry concentration and the pace of technological innovation are
endogenously determined by the exogenous parameters describing
barriers to entry and the extent of innovative opportunity. I shall
analyze the long-run relationship between concentration and inno-
vation determined in this model, and I shall argue that it is consistent
with the four empirical relationships listed above.
This model is constructed from a distinctly Schumpeterian
perspective (see Schumpeter [1934,1942]). The principal incentive
resulting in R&D activity by individual firms is the prospect of al-
tering market structure and acquiring market power via a successful
and decisive product or process innovation. The economic rents
arising from successful innovation are temporary, however. These

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
SCHUMPETERIAN COMPETITION 677

rents are eventually whittled away though an inexorable process of


imitation, entry, and innovation by other firms. I should note that in
this model pure size endows a firm with absolutely no cost advantage
in undertaking R&D activities. It is the prospect of attaining market
power in the future rather than the fact of market power in the present
that drives firms to innovate. Only by successful innovation can a firm
weather Schumpeter's "perennial gale of competition" that this model
depicts.

I. A STOCHASTIC MODEL OF INDUSTRY DYNAMICS

I shall now describe a stochastic model of the Schumpeterian


"process of creative destruction." From the formal point of view, this
model is a Markov process defined on the set of short-run, price-
quantity equilibria for the industry under study.
A short-run industry equilibrium is defined by a pair of positive
integers (n,j). The integer n denotes the number of identical firms
coexisting at the equilibrium determined by the data (nj). The integer
j is an index number that identifies the product produced or the
production technique employed by all firms at the equilibrium (n,j).
If one wishes to model a process of cost reduction, 1/] might be in-
terpreted as an index of minimum average cost associated with
technique j. If one wishes to model a process of product innovation,
j might be interpreted as an index of product quality (in the broad
sense of the word). In either case, one should regard product or
technique j as representing a decisive improvement over product or
technique k whenever k < j.
I shall assume that whenever we observe this industry we find
it in just such a short-run equilibrium state. In particular, we observe
it only after products or techniques inferior to those associated with
the observed short-run equilibrium have been driven from the market.
This is, of course, an extreme simplification that is at odds with the
facts and is certainly not necessitated by economic theory. Products
of different qualities and techniques with different unit costs can and
do coexist in the marketplace. I do not believe that incorporating the
coexistence phenomenon into this model would alter its qualitative
implications, but it would certainly make an analysis of these an order
of magnitude more difficult.
In the next section of this paper, I shall associate with each
short-run equilibrium (n,j) a single period R&D game played by the
n firms coexisting at that equilibrium. The Nash equilibrium for the
R&D game will determine the probability that at least one of the n

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
678 QUARTERLY JOURNAL OF ECONOMICS

firms innovates while the configuration (nj) prevai


that this probability will depend only upon n but
fore I denote the probability that innovation occurs at the short-run
equilibrium (nj) by g(n). I shall furthermore assume that, whenever
innovation occurs, it occurs one unit at a time.
Note that I am assuming that only the firms within the industry
undertake industry relevant R&D projects. In the world, of course,
outside innovators occasionally play crucial roles in the development
of the industry's products or techniques. Unfortunately, there are very
little useful data on the extent of these extra-industry innovative ef-
forts. In any case, the empirical work mentioned earlier did not
measure such inputs. Therefore I have chosen not to incorporate this
difficult-to-observe variable into the model. So long as it does not
distort the qualitative properties of g(n), the implications of the model
will not be altered.
These assumptions are formalized as follows. Let NtJt denote
the random variables recording (respectively) the number of firms
and the product or technique that prevails at date t. Then I assume
that

prob(Jt+l = 1 + Jt IJt =jNt = n) = AWn)


prob(Jt+l = Jt IJt = i, Nt = n) = 1- 8(n).

The fundamental feature of this model is that the introduction


of a new product or technique will generally alter industry structure
(i.e., the number of firms coexisting in short-run equilibrium). Fur-
thermore, even when such innovation does not take place, the eco-
nomic rents earned by existing firms are apt to be whittled away by
the imitation and entry of additional firms into the industry.
The extent to which market structure can be altered by innova-
tion and by entry is governed by what I shall call the condition of
entry. I denote this real valued variable by the symbol v; high values
indicate an easy-to-enter industry and low values a difficult-to-enter
industry. I shall assume that the condition of entry relevant to the
industry is fixed over time and, most importantly, exogenous to the
industry under study. The reader should keep in mind that several
of the commonly noted entry "barriers" are, in principle, under the
control of firms within the industry. Capitalized advertising expen-
ditures and limit pricing policies are examples. In this model, v is best
thought of as measuring the implicit or explicit legal restraints upon
entry (e.g., patent law requirements, antitrust policies). Alternatively,
if innovation is not directed toward changing the relationship of

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
SCHUMPETERIAN COMPETITION 679

minimum efficient scale to market size (


where it is so directed), or to changing min
these variables can be included in the measurement of entry
conditions.
To describe the process of entry and imitation parameterized by
v, I shall adopt the following formalism. Imagine the R&D game de-
fined by the equilibrium (n,j) to be a race among the n firms to be the
first to introduce product or technique j + 1. This race may have no
winner, i.e., no firm innovates. This event has probability 1 -,(n).
In such a case I shall assume that all n firms are sure to remain eco-
nomically viable during the subsequent period, and I shall say that
there are n sure survivors of the R&D race. If this race has a winner,
it is assumed to have exactly one. This event has probability AL(n). In
such a case I shall assume that only the single innovating firm (among
the n firms in the race) is sure to remain economically viable during
the subsequent period. I shall then say that there is one sure survivor
of the R&D race.
To facilitate the exposition, I now introduce the random variable
St denoting the number of sure survivors of the R&D race defined by
the short-run equilibrium (Nt,Jt). Then

prob(St = nlINt = n, Jt = j) = 1 -u(n)


prob(St = 1 I Nt = n, Jt = j) =(n).

Of course, the number of firms in the industry at date t + 1 will


always be at least as large as the number of sure survivors at date t.
But it will often happen that Nt+1 > St. Such an event can result from
successful imitation and entry by additional firms. If St = Nt, then
the new entrants come from outside the industry. If St = 1, then
among the entrants may well be some of the unsuccessful innovators
at date t who nonetheless quickly succeeded in imitating the single
successful firm.
I shall represent this process of innovation and imitation by as-
suming that the conditional distribution of the number of firms at
date t + 1, Nt+l is completely determined, once one specifies the
number of sure survivors at date t and the condition of entry. I es-
tablish the notation

X(l,k,v) prob(Nt+l = k|St = 1).

I shall restrict v to lie in the interval [0,1], and introduce the following
assumptions.

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
680 QUARTERLY JOURNAL OF ECONOMICS

Al.

L X(l,k,v) = 1
ke 1

for all v and all 1. Furthermore, X is a continuous function of


V.

A2. k < 1 implies that X(l,k,v) = 0.

A3. There is an n such that k > hi implies X(l,k,v) = 0.

A4. For any fixed m,


m

E X(lkV)
k=1

is a decreasing function of 1 and v.

A5. If v = 0, then X(l,l,0) = 1 for all 1. If v = 1, then X(l,Th, 1) = 1 for


all 1.

Assumptions Al, A2 require no comment. Assumption A3 places an


upper bound ff upon the number of firms that can coexist in short-run
equilibrium. This assumption will aid the comparative statics analysis
in later sections. Assumption A5 identifies some properties of the
extreme values of the condition of entry. If v = 0, then entry is
blockaded and imitation is impossible. If v = 1, then entry is "free,"
and Nt always assumes the value ff. Thus, ff should be interpreted as
the number of firms that would coexist in a perfectly competitive
market. Assumption A4 implies that the conditional expectation of
the number of firms in the industry at date t + 1, E(Nt+ I St = 1), is
an increasing function of the number of sure survivors at date t and
the condition of entry. This in particular implies that

E(Nt+llSt = 1) < E(Nt+l|St = Nt).


I assume, therefore, that successful innovation will in the short run
tend to increase industry concentration from the level it would have
attained had there been no innovation.
This description of the entry process implies that entry in this
model always occurs via imitation and not via innovation. New firms
always adopt the product or technique of an existing firm. This re-
striction is a consequence of my decision not to incorporate extra-
industry R&D efforts into the model. In particular, it prevents this
theory from describing a world in which there is a perfect patent.
It is now possible to write down the transition probabilities that

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
SCHUMPETERIAN COMPETITION 681

describe the movement of the industry from o


to the next:

prob(Nt+1 = m, Jt+1 = jNt = n, Jt = i)

0 ifj i, i + 1
= (n)X(l,m,v) if] = i + 1

[1 - /l(n)]X(n,m,v) if j

These transition probabilities, together with a spe


distribution of No, J0, completely specify the
Markov process {(Nt, Jt), t = 0,1,2, . . .}. I wish to
properties of this chain. Unfortunately, Jt+1 >
one, so in general there will be no invariant probability distribution
over industry states to which this process converges.
This difficulty can be avoided by considering the Markov process
{Nt, t = 0,1, . . } in isolation. This is made possible by the fact that the
innovation probabilities do not depend upon the index number
identifying the industry's technique or product. Thus,

p(n,m) prob(Nt+l = mINt = n)


= jt(n)X(l,m,v) + [1 - jt(n)]X(n,m,v)
defines the transition probabilities for the process {Nt}. For exactly
the same reason, we lose no essential information about the long-run
rates of innovation by passing from I(Nt,Jt)} to {Nt}.
The process Nt } has very useful asymptotic properties. Let 7ro
denote the probability density of No. Denote by p 1(n,m) the proba-
bility of moving from an n firm equilibrium to an m firm equilibrium
in precisely 1 periods.

THEOREM A. Assume Al, and suppose that inf jt(l) > 0. Then there
1>2
is a unique probability density 7r* over industry sizes with the
following properties:
ao

(a) E 7r*(n)pt(n,m)= =r*(m)


nl

for all m,t.


(b) There are positive constants a,b such that

a
SUP 7rwo(n)pt(n,m) - r* (m) _ 1+ <
m n=1 1+bt

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
682 QUARTERLY JOURNAL OF ECONOMICS

Theorem A tells us two things. First of all, assertion (a) says that
there is a long-run, stochastic equilibrium for this process of
Schumpeterian competition. Assertion (b) tells us that this equilib-
rium is unique and that it will be the limit of the process starting from
any initial state. This allows us to interpret wx*(n) as the fraction of
the time n firms are observed in the industry over any infinitely long
sample path.
The proof of Theorem A is a straightforward exercise in the
theory of Markov chains. For example, one can invoke Theorems 3.5,
4.6, and 4.11 of Futia [1976] to obtain the result. Alternatively, one
can use the assumption that inf 4(1) > 0 to show that the chain is
1>2

positive recurrent with a single recurrent class and is noncyclic. Ap-


pealing to Kemeny, Snell, and Knapp [1976], Theorem 6-38 then
proves all of Theorem A except the rate of convergence result. This
last fact can be obtained as a consequence of Orey [1971, proposition
6.1, p. 26].
The long-ruit equilibrium 7r* defines two important statistics:
00

n*= Z nwr* (n)


n=1

/1* = Z Is (n)7r* (n).


n=1

The number n* is the long-run expected value of the number of firms


in the industry and is an index of long-run concentration. The number
A* is the long-run expected value of the innovation probability. I refer
to A* as the long-run pace of innovation. Intuitively speaking, 1/11*
is the expected value of the waiting time between successive innova-
tions. The equilibrium statistics n*jt* will naturally depend upon
the condition of entry v. These relationships will be explored in sub-
sequent sections.

II. THE DETERMINATION OF INNOVATION PROBABILITIES

To complete the model of Schumpeterian competition described


in the last section, it is necessary to specify how the innovation
probabilities Mt(n) are determined by the industry's short-run equi-
librium configuration. To do this, I shall associate with each equi-
librium (n,j) a single period R&D game played by the n firms at that
equilibrium. The pure strategy, Nash equilibrium for this game will
completely determine the innovation probability 4t(n).
The assumptions I am about to make concerning this R&D game

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
SCHUMPETERIAN COMPETITION 683

are very restrictive. However, it is not my inte


eral framework for analyzing the R&D decision problem. I only wish
to "rationalize" the so far arbitrary innovation probabilities M(n). And
I wish to do this in a way that is simple enough to allow a reasonably
extensive analysis of the model's equilibrium properties. In later
sections of this paper, I shall argue that this particular "rationaliza-
tion" of the innovation probabilities has equilibrium implications that
are consistent with empirical observation.
Imagine that the short-run equilibrium (n,j) prevails at date t.
At that date, each of the n coexisting firms is assumed to choose a level
of input (in physical units) that it will devote to R&D activities at date
t. Denote the input level chosen by firm i by xi and assume that it lies
in the unit interval [0,1]. A firm's pure strategy for the game I shall
define is such an input level.
I assume that at date t each firm wishes to choose a level of R&D
input that will maximize the difference between the expected value
of the economic rents it can appropriate at date t + 1 (before sub-
tracting the R&D costs incurred at t + 1) and the R&D costs associ-
ated with its input choice at date t. This is a sensible comparison, since
R&D efforts at date t cannot influence the firm's market position until
date t + 1. As long as we work within a model where all new entrants
are imitators, the restriction of the optimization horizon to but a single
period is only a (necessary) computational simplification. Longer
horizons will not alter the qualitative behavior of a firm's choice of
R&D input.
Of course, in general the optimal choice of R&D input for firm
i will depend upon the choices made by other firms at that date. I
attempt to recognize this interdependence by assuming that the vector
of R&D input choices observed at date t constitutes a pure strategy,
Nash equilibrium for the one-period game defined by these objective
functions.
I begin to define the objective function for each firm more pre-
cisely by first defining the level of economic rents appropriated by
each firm at the short-run equilibrium (nJ). If (n,j) prevails at date
t, I shall assume that each of the n firms appropriates economic rents
at that date in an amount that can be written in the form 6(n)R(j).
Furthermore, I assume that

A6. (a) 0(1) = 1,0(n) > 0.


(b) n.0 (n) is a nonincreasing function of n.
(c) 0(n) is a decreasing, convex function of n.
A7. R(j) = -yiR for some constant T > 1.

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
684 QUARTERLY JOURNAL OF ECONOMICS

In view of A6 (a), the number R (j) should be interpreted as the


level of monopoly rents appropriated by a single firm industry with
product or technique j. Assumption A6 (b) simply asserts that ag-
gregate industry rents do not increase as the number of competing
firms increases. An interesting example of a function 0 satisfying A6
is

0(n) = no3 for some/3? 1.

The parameter : measures the speed with which aggregate rents de-
cline to zero with the number of competing firms. If 3 = 1, the industry
would always succeed in maximizing aggregate (short-run) rents re-
gardless of the number of coexisting firms. Such a situation could arise
if, for example, price and quantity fixing agreements were both fea-
sible and legal. If such agreements were not possible, then : might be
large (e.g., 2 or 3) and rents would decline rapidly to the competitive
level.
Assumption A7 is principally a device that provides a convenient
index of the extent of innovative opportunity. I wish to argue that the
constant -y is such an index. It is sensible to suppose that the greater
is the extent of innovative opportunity, the bigger is the "bang per
R&D buck" accruing to each firm. Holding market structure and R&D
inputs fixed, one can see that the "bang" is roughly proportional to
the percentage increase in monopoly rents R (j) that can be achieved
by successful innovation. This is just

Rj + 1) - R) = y - 1 byA7.
R (j)
Thus, I shall regard y as a legitimate index of the extent of innovative
opportunity facing the industry. For a discussion of the difficulties
one encounters in attempting to define an empirical counterpart to
this opportunity concept, the reader should consult Wilson [1977].
I continue the construction of a firm's objective function by de-
scribing how the vector of R&D inputs determines the probability that
any given firm innovates. Denote a vector of R&D inputs that could
be chosen at the equilibrium (n,j) by (xl1, . . , xn). I begin by assuming
that there is a function p(x1, . . ., xn) that records the probability that
at least one firm innovates, i.e., introduces product or technique
j + 1 at the following date.
Next, I suppose that the R&D activities carried on by the in-
dustry at date t can be thought of as a race among the n firms coex-
isting at date t to be the first to introduce product or technique j +
1 at date t + 1. This race may have no winner; as in the previous sec-

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
SCHUMPETERIAN COMPETITION 685

tion I then say that there are n sure survivors of the R&D race at date
t. All n firms remain economically viable at date t + 1. This is an event
with probability 1 - p(xi, . . ., Xn). On the other hand, there may be
a winner of the R&D race. If there is one, I assume that there is only
one. The winning firm is the sole sure survivor of the R&D race and
is the only firm certain to remain economically viable at date t + 1.
The probability that firm i wins, conditioned on the event that there
is a winner, I denote by a-(x1,.. , ,Xn). Thus, p(xi,... ., xn)aLi(xi,.
xn) is the unconditional probability that firm i wins the R&D race at
date t.
To complete the specification of the objective functions, it is
necessary to associate with each choice of R&D input a monetary cost.
I assume that if firm i selects input xi when the short-run equilibrium
is (n,j), then it incurs a cost of yiC(xi). The constant y is the same
constant that provides a measure of innovative opportunity. Thus,
the costs of doing R&D are assumed to grow at exactly the same rate
as the private benefits. Such an assumption is necessary to insure that
the optimal choices of R&D inputs are not invariably corner solutions
as t a- oo. Furthermore, I shall assume that the cost function C(x) is
increasing and strictly convex in x. This amounts to supposing that,
in any reasonable sense, the marginal cost of innovative output in-
creases with the quantity of such output.
It is now possible to write down firm i's objective function defined
by the short-run equilibrium (n,j). To do this, first note that the ex-
pected value of the economic rents firm i can appropriate at date t +
1 can be calculated by considering three mutually exclusive and ex-
haustive events.
The first event occurs when firm i wins the R&D race at date t.
This event has probability p(xi, . . ., x n)ai(x, ... , X, ). Firm i is then
certain to remain economically viable at date t + 1. If at date t + 1
there are k firms in the industry, firm i will appropriate rents in the
amount 0(k)R (j + 1). We can therefore denote the expected value of
these rents (weighting by the probabilities of various k) by

,q(1,V),Yi+1R,
where

,q(IV)- E (1,k,v)0(k).
k=1

(Recall that 1 denotes the number of sure survivors of the R&D race
at date t; if innovation occurs, 1 = 1.)
The second event occurs when none of the n firms wins the R&D

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
686 QUARTERLY JOURNAL OF ECONOMICS

race. This event has probability 1 - p(x,. . ., Xn). In this case all n
firms, including, of course, firm i, are sure to remain economically
viable at date t + 1. The expected value of the rents firm i can ap-
propriate in this event can be written as q1(n,v)TiR.
The third event occurs when some firm other than firm i wins
the R&D race. This has probability p(xl,. . ., xn) [1 - ai(x1,.. *, ,n)]
Firm i may or may not remain economically viable at date t + 1 de-
pending upon the ease with which it can quickly imitate the innovator.
It will turn out that I do not need to be precise about this imitation
process. Thus, for the moment, let me denote the expected value of
the rents appropriated by i in this third event by i7',yi+1R.
Taking these three events together, one can now write the ob-
jective function fi(x1, . . ., xn) for firm i at the equilibrium (n,j) as

fi(x1,. . ., xn) = p(X1,... , Xn)ti(xl,.. Xn), (xlv))i+lR


+ [1 - p(x, ... , xn)]ij(n,v)yiR
+ p(xi, ... , Xn)[l - 1(x, ., Xn)]q',yi+lR
- C(Xi)yi.
Notice that the optimal choice of xi for firm i will in no way depend
upon j. Since innovation probabilities depend only upon input
choices, they too will be independent of j.
It is necessary to simplify this choice problem still further in order
to facilitate the study of the industry's long-run equilibrium state.
To this end, I introduce the following assumptions:
n

A8. p(Xi,... ,Xn)=(a Exi.


i=1

A9. ai(xi,... , Xn) = xi/ L 'Xk


k=1

Alo. C(xi) = (b/2)xW.


All. Each firm simplifies the calculation of its optimal input choice
by replacing

*11,v)- E X(1,k,v)0(k)
k=1

by

O(M(I,0 ) = 0 E hXA(l,k,v ))
k=1

where we have introduced the notation


0o

M(l,v) L kX(l,k,v).
k=1

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
SCHUMPETERIAN COMPETITION 687

One must be careful to choose the constant "a" in A8 so that for no


value of n < n- (cf. A3) does p exceed one.
Assumptions A8, A9 have two important consequences. First of
all, firm i's optimal choice of xi will in fact be independent of the
choices made by the other n - 1 firms (i.e., the Nash equilibrium is
a dominant strategy equilibrium). Second, firm i's choice of xi will
in no way depend upon the level of rents it appropriates when it loses
the R&D race to an innovator, but nonetheless imitates quickly
enough to remain viable.
Next, I suppose that the optimal choice x i of xi always satisfies
the relevant first-order condition. Thus,

(1) x = (a/b)R[,yO(M(1,v)) - O(M(n,v))].


Since fi is concave in xi, this x will indeed maximize fi. Thus
n-tuple (x1, ..., x ) constitutes a pure strategy, Nash equilibrium
for this one-period R&D game. Since the R&D cost function is the
same for all firms, x 7 = x. I shall denote this common value of R&D
input by x(n,v,,y).
The reader should note that x (n,v, -y) is roughly proportional to
the difference between the expected value of the rents a firm appro-
priates when it wins the R&D race and the expected value of the rents
it appropriates when no innovation occurs.
Finally, I define the innovation probability ,u(n) associated with
the short-run equilibrium (n,j) to be p(x(n,v,,y), . . ., x(n,v,-y)) =
anx (n,v,'y) and denote this probability by It(n,v, y). This definition
then completes the description of our model of Schumpeterian
competition.

III. CONCENTRATION AND INNOVATION:


A PRELIMINARY STUDY

Are increases in industry concentration associated with increases


or with decreases in the pace of technological innovation?
To answer this question, it seems reasonable first to examine the
properties of the innovation probability A(n,v,-y) defined in the last
section. One should note immediately that this innovation probability
increases as short-run concentration decreases (i.e., as n increases).
For an increase in n will increase M(n,v), decrease 6(M(n,v)), and
therefore increase x(nv,-y) (see equation (1)). Since ,L(n,v,y) =
anx (n,v,y), we conclude that gt increases with n.
This fact suggests that a policy maker might be able to "eat his
cake and have it too," for a shift in the long-run equilibrium distri-
bution -r* to the right (so as to raise n*) would presumably increase

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
688 QUARTERLY JOURNAL OF ECONOMICS

the extent of long-run price competition and also i


pace of innovation A*. For

8*- Lg(n,v,^y)w*(n),
n=1

and A is an increasing function of n.


But is the situation really as simple as this? Clearly it is not, for
the extent of long-run concentration is not an exogenous variable in
this model. Concentration can be altered only by changes in the
condition of entry, and changes in the condition of entry will directly
affect the innovation probabilities g(n,v,,y) by changing the ease with
which firms may alter market structure via innovation.
In fact, it is not hard to see that, with one additional assumption,
an increase in v so as to ease entry and (hopefully) raise n * will result
in a decrease in ti(n,vyy) for each n.
A12. DvM(l,v) is a nonincreasing function of 1.
Thus, I assume that the effect of a given change in the condition of
entry upon concentration does not increase as the number of sure
survivors of the R&D race increases.
Assumption A12 together with the assumed convexity of 0 implies
that ,u(n,v,,y) decreases as v increases (i.e., as entry conditions ease).
Easing entry conditions may well raise n*, but since g(n,v,,y) de-
creases as v increases, the effect upon g* of easier entry is
ambiguous.
The point of this tale should be easy to see. An important in-
centive for undertaking R&D projects is the prospect that innovation
can result in greater economic rent, through increased market power.
Any government policy that changes the possibilities of altering
market structure through innovation will directly affect these private
incentives. But in general, industry concentration can only be changed
by such policies. Therefore, it is of great importance to study the re-
lationships between concentration and innovation in a model where
industry structure is explicitly variable and endogenously determined,
for in a static equilibrium model with industry concentration as an
exogenous variable, the effects upon R&D effort of the policy changes
necessary to alter concentration will be ignored, more likely than
not.

IV. THE DETERMINANTS OF LONG-RUN CONCENTRATION

How do changes in the condition of entry v, as well as in the ex-

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
SCHUMPETERIAN COMPETITION 689

tent of innovative opportunity My, affect long-run industry concen-


tration (a quantity inversely proportional to n*)? The short-run
consequences of such parameter changes are easy to unravel.
Consider, for example, the case of an easing in the entry condi-
tion, i.e., an increase in v. From assumption A4 (see Section II), we
know that such a change in v will, for any fixed number of sure sur-
vivors of the R&D race at date t, increase the expected number of
firms in the industry at date t + 1. Furthermore, assumption A12 (see
Section III) implies that an easing of entry conditions will, for each
n, decrease aggregate R&D input and thus decrease the innovation
probability ,g(n,vyy). This will tend to increase the number of su
survivors of the R&D race at date t, since all firms survive if no in-
novation occurs. Another application of A4 tells us that the probability
distribution of the number of firms at date t + 1 shifts to the right as
the number of sure survivors at date t increases. Combining these
three observations allows us to infer that, starting from any short-run
equilibrium at date t, an increase in v will increase the expected value
of the number of firms in the industry at date t + 1.
A shorter argument shows that, in the short run, a decrease in
My will increase the expected value of the number of firms in the in-
dustry. From equation (1) (see Section II), a decrease in the extent
of innovative opportunity will decrease each firm's R&D input and
therefore decrease the innovation probability Au(n,v,'y). This will shift
the distribution of the number of sure survivors at date t to the right
and will, by A4, increase the expected value of the number of firms
in the industry at date t + 1.
We know now that easing the condition of entry or decreasing
the extent of innovative opportunity will, in the short run, decrease
industry concentration. But will these short-run effects persist and.
result in the corresponding changes in n*? Deducing a long-run effect
from a short-run effect is seldom easy in a stochastic model of this sort.
In fact, it is logically possible (although I have not succeeded in con-
structing an illustrative example) that easing entry conditions could
increase rather than decrease long-run concentration, for we know
(see Section III) that the innovation probability gt(n,v,,y) increases
with n. Therefore, an increase in v will, by increasing the number of
firms in the short run, also increase the probability of innovation. But
innovation tends to reduce the number of firms in the industry. This
indirect "innovation effect" could offset the direct effects of an in-
crease in v and thus lead to an increase, rather than a decrease, in
long-run concentration.
To rule out such "cobweb" phenomena, I shall, from this point

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
690 QUARTERLY JOURNAL OF ECONOMICS

on, require that this model of Schumpeterian competition have a kind


of monotonicity property. Consider the quantity

g(n,l,v,,y) L {g(n,v,y)X(1,k,v) + [1 - g(n,v,,y)]X(n,k,v)}.


k=1

A13. For all values of lv,)y, g(n,l,v,,y) is a nonincreasing function of


n.

If the stochastic model satisfies assumption A13, I shall say that it is


stochastically monotone increasing. The terminology arises from the
observation that A13 implies E(Nt+il INt = n) is an increasing func-
tion of n. Thus, under A13, the greater the number of firms in the
industry at date t, the greater the expected value of the number of
firms in the industry at date t + 1.
Assumption A13 is a nontrivial restriction. But the class of
models that it delineates is by no means vacuous. Consider, for ex-
ample, a two-state model, i.e., one where X(l,k,v) = 0 unless k = 1 or
k = ii. It is easy to see that A13 must be satisfied.
It is also easy to see that, more generally, A13 will be satisfied so
long as g(n,v,,y) does not increase too rapidly with n. For if g(n,v,,y)
were independent of n, then A1-A4 imply A13. Since A3 places an
upper bound upon n, a slight perturbation of g(n,v, y) so that A in-
creases with n will preserve A13.
If one adopts assumptions Al-A13, then one can apply the main
theorem in Daley [1968] to infer that increasing v or decreasing ^y will
increase n*. In this case, as one expects, high entry barriers or ex-
tensive innovative opportunities are associated with high levels of
long-run industry concentration.

V. CONCENTRATION AND INNOVATION


Under what circumstances will an easing of the condition of entry
increase the long-run pace of innovation, ,u*? When will such a change
decrease A*? Since n* is positively related to the condition of entry
it is equivalent to asking for the precise relationship between long-run
concentration and the long-run pace of innovation.
The first empirical fact this model must account for is an ob-
served positive correlation between concentration and various mea-
sures of innovative output or input (see the introduction). I shall do
this by showing that, "on average," A* is a declining function of v.
This, of course, is a difficult property to define precisely. What I shall
show is that A* depends continuously upon v (notation: A*(v)) and
that, in general, A* (0) > A* (1).

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
SCHUMPETERIAN COMPETITION 691

That A* depends continuously upon v i


this stochastic model is a finite Markov chain. Theorem A asserts that
this chain has a unique invariant distribution, 7r*. These facts, to-
gether with Al, imply that ir*, hence A* is a continuous function of
v.

I now argue that, in general, gt* (0) >,g* (1). The key assumptio
required is A5. This describes the properties of the entry process for
the two extreme values of the condition of entry. If entry is free (v =
1), then A5 implies that the density lr* is a spike at the point n = if.
If entry is blockaded (v = 0), then A5, together with the fact that.
A(n,v,,y) > 0 for n > 1, implies that lr* is a spike at the point n = 1.
Thus, g*(O) = g(1,0,,y), and A* (1) = A(n-, 1,y). But since A(n,v,y) =
anx (n, v, y), we conclude from equation (1) that

1*(O) = (a/b)RO(l)(y - 1) > (a,b)Rii6(ii)('y - 1) = g*(1).

This inequality is strict, provided that y > 1 and that aggregate in-
dustry rents strictly decrease as the number of firms in short-run
equilibrium increases.
Thus, we find that blockaded entry results in a higher long-run
pace of innovation than does free entry. For in neither case can an
individual firm alter market structure in the long run via innovation.
In such cases, ,* is strictly proportional to the increase in aggregate
industry rents that arises from successful innovation. This increase
is smaller in a "competitive" industry than in a monopolized one
under assumptions A6, A7. This result should not seem strange, since
by "free entry" in this model I mean a situation in which a private firm
has no property rights in the innovations it makes (aside from those
implicitly associated with the upper bound, n-, on industry size). In
such a situation aggregate competitive rents would be small both
before and after the innovation (thus remaining essentially
constant).
Although the pace of innovation declines, "on average," as the
conditions of entry ease, it is still of interest to know for which con-
dition of entry * attains its greatest value. It is helpful to distinguish
two possibilities. It could happen that ,u* is greatest when entry is
blockaded. On the other hand, it could happen that ,*is greatest for
some intermediate value of v, thus causing ,u*(v) roughly to assume
the shape of an inverted "U". The second possibility finds consider-
able support in the empirical literature.
I now wish to discover conditions under which ,u* (v) attains its
maximum at an intermediate value of v. It would be sufficient to ex-

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
692 QUARTERLY JOURNAL OF ECONOMICS

hibit conditions under which Dv* (v) I v=o > 0. U


generally impossible to evaluate this derivative in closed form. Fur-
thermore, the stochastic monotonicity techniques used in the last
section are of little help here, principally because A(n,v,,y) is not a
state variable for the Markov process.
To circumvent these problems and gain some insight into the
behavior of A*(v), I am forced to resort to a heuristic method of
analysis. I shall associate with this stochastic model a heuristically
derived, deterministic, "certainty equivalent" model. The variable
corresponding to A* is denoted by it. I shall then evaluate the deriv
ative Dv, (v) I v=o. If the deterministic model is a good analogue of the
stochastic model, then Dv () I v0=o should be positive in the same
circumstances that DvA*(v) I v=0 is positive. This is the case in the
numerical example that appears in the Appendix to this paper.
Recall from Section I that the unconditional expectations of the
random variables {Nt I satisfy the equation,
ENt+1 = E{g(Nt,v,'y)M(1,v) + [1 - g(Nt,v,'y)]M(Nt,v)}.
I let nt denote the deterministic analogue of ENt, and I define At re-
cursively by the equation,

A = g(At,v,'y)M(1,v) + [1 -(tV, )]M(AtV).


I assume that this difference equation has a stable, stationary solution,
and I denote this equilibrium value of At by h. This is, of course, the
deterministic analogue of n*. The reader can check that if n is a stable
solution of this equation, then an increase in v or a decrease in oy will
increase h. This evidence supports the hope that our heuristic leap
of faith might be justified, for n* has exactly the same properties. I
now define 4(v) by the equation, 4(v) _(,A(v,,y),v,,y). Note that 4(0)
- ,*(0) and ,(1) = ,*(l)
Using equation (1) together with the definition of A, one finds
that

DV (v) A V= = (positive constant) [DvM(1,0)]


p~~~~~~~~~

X DO(1) + A(O))D1M(1,O)J
1- (1- MLO)
This expression is of considerable interest, for it asserts that the sign
of D V ,(v) I v= o is determined by the sign of the sum of the two effects
(in brackets) that were discussed in section three.
The first term, DO (1), is negative and represents the tendency
for innovation probabilities, A(n,v,,y), to decrease in response to an
easing of entry conditions. This tends to lower A. If n is a stable so-
lution to the difference equation, then the second term is positive. For
increasing v will then increase n; since A(n,v,'y) increases with n, A

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
SCHUMPETERIAN COMPETITION 693

will thus tend to increase. The reader should note that these two terms
in brackets can be varied independently by changing, for example,
the function 0 that records the behavior of a firm's rents as industry
concentration falls. One can conclude that if rents do not fall too
rapidly as concentration decreases, then jt will attain a maximum at
an intermediate value of v.
It might be helpful to illustrate this conclusion with a simple
example. Let M(l,v) (1 - v)l + v-n, and let 0(n) n-0 for some : >
1. Then the term in brackets is just

If, for example, 0 < 2 and the waiting time between innovations ma
by a monopolist (recall that 4(0) = A* (0)) exceeds two periods, then
Dv (0) > 0.
This example shows that the certainty equivalent version of a
particular parameterization of the stochastic model can explain the
apparent midpoint maximum in the innovation versus concentration
relationship. In the Appendix I show that this parameterization of
the stochastic model actually behaves as the analysis of its certainty
equivalent suggests.

APPENDIX

In the previous section I derived, using heuristic methods, a


"sufficient condition" for the pace of innovation, At*(v), to attain
maximum value at an intermediate value of the condition of entry.
To check upon the validity of the analysis, I have computed numeri-
cally the equilibrium density, wr*, for a number of very simple par-
ameterizations of the stochastic model. In all cases the conclusions
of the heuristic analysis were supported. In this Appendix I shall re-
port the numerical results obtained in the simplest example I have
studied.
These numerical analyses of simple versions of the stochastic
model yield a surprising and encouraging further result. One finds
that the positive correlation between the pace of innovation and
concentration declines as the extent of innovative opportunity in-
creases. This phenomenon has been noted in empirical work (Co-
manor [1967], and Scherer [1967,1970]).
I now discuss a simple numerical example to illustrate these
points. In this example there are two possible industry sizes: n = 1,
n = 10 = ii. The entry process implicitly defined by X is given by the
equations

X(1,1,v) = 1 - v
X(1,10,v) = v
X(10,1Ov) = 1.

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
694 QUARTERLY JOURNAL OF ECONOMICS

TABLE I

CORRELATION BETWEEN A* AND (-n*)


v = 0.1,0.2,0.3,0.4,0.5,0.6

ly

A 1.1 1.5 2.0


1 0.668 0.486 0.174
2 0.900 0.886 0.873

From this it follows that


M(l,v) = 1 + 9v and M(10,v) = 10.
The function 0 recording the behavior of a firm's rent sh
by 0(n) = n-0 for some f 1. The constants in the model are assumed
to have the values,
a =0.1, b = 1, R = 5.0.

These data are sufficient to determine the innovation probabili-


ties:

A(l,v,-y) = 0.05 [y (1 +9v) (1

A(10,v,-y) = 0.5 Fy (1 + 9v)3 101


The innovation probabilities, together with X, determine the equi-
librium probability 7r* that there is exactly one firm in the in-
dustry:
* _(10,v,y)(1 - v)
v (1 - v)g(1,v,y),
Then the probability 7r*o that there are ten firms in the ind

Since a two-state model is always stochastically monotone, it


should be true in this example that an increase in v, a decrease in oy,
or an increase in /B should cause n* to increase. Direct computation
shows that this is the case.
This numerical example also satisfies the conditions set down
in Section V (see the example at the end of that section), which should
guarantee that IA* attains its greatest value at intermediate values of
v. Again a simple computation shows that the heuristic analysis is
vindicated.
This numerical example exhibits two properties (shared by the
other examples I have computed), which, it seems to me, are impos-
sible to derive analytically.
First of all, for any fixed value of v, an increase in by will increase
A *. Thus, industries with greater innovative opportunities will, in the
long run, innovate more frequently.
The second property is more surprising. Compute the correlation
coefficient between A* and (-n*) for each pair of parameter values
(-y,fl). The data points are those determined by the conditions of entry
v = 0.1,0.2,0.3,0.4,0.5,0.6. The results are reported in Table I. Note

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms
SCHUMPETERIAN COMPETITION 695

that for fixed /, the correlation between 1&* and (-n*) is positive and
that it declines as innovative opportunity increases.
This phenomenon can best be explained as follows. The rela-
tionship between A* and n* can be pictured as an inverted "U" w
A* first increasing and then decreasing as n* increases. An incre
in by results in an approximately parallel, upward shift in this curve.
Consider two of these curves, the lower one determined by a low value
of y, the higher by a high value of y. A fixed value of v determines a
point on the low y curve and a point on the high oy curve. Increasing
-y lowers n*. Since the two curves are approximately parallel, this
implies that the six points on the low y curve determined by the six
values of v lie farther to the right, relative to the curve's peak, than
the corresponding points on the high y curve. Thus, the correlation
between tt* and n* determined by the points on the low y curve is a
larger negative number than the correlation determined by the points
on the higher by curve, and increasing y lowers the positive correlation
between A* and (-n*).

BELL LABORATORIES

REFERENCES

Comanor, W., "Market Structure, Product Differentiation, and Industrial Research,"


this Journal, LXXXI (1967), 639-57.
Daley, D. J., "Stochastically Monotone Markov Chains," Z. Wahrscheinlichkeitstheorie
und Verw. Gebiete, X (1978), 305-17.
Dasgupta, P., and J. Stiglitz, "Market Structure and the Nature of Innovative Activity,"
paper presented to the World Congress of the International Economic Association
held in Tokyo, Japan, in September 1977.
Futia, C., "A Stochastic Approach to Economic Dynamics," Bell Laboratories, Murray
Hill, NJ, 1976.
Kamien, M., and N. Schwartz, "Market Structure and Innovation: A Survey," Journal
of Economic Literature, XIII (March 1975), 1-37.
Kelly, T. M., "The Influence of Firm Size and Market Structure on the Research Efforts
of Large Firms," Ph.D. thesis, Oklahoma State University, 1970.
Kemeny, J., J. Snell, and A. Knapp, Denumerable Markov Chains (New York:
Springer-Verlag, 1976).
Levin, R., "Technical Change, Economies of Scale, and Market Structure," Ph.D. thesis,
Yale University, 1974.
Loury, G., "Market Structure and Innovation," unpublished paper, Northwestern
University, 1976.
Nelson, R., and S. Winter, "Dynamic Competition and Technical Progress," Institute
of Public Policy Studies Discussion Paper No. 82, Yale University, 1976.
Orey, S., Limit Theorems for Markov Chain Transition Probabilities (New York: Van
Nostrand, 1971).
Scherer, F., "Market Structure and the Employment of Scientists and Engineers,"
American Economic Review, LVII (June 1967), 524-31.
"Market Structure and Technological Innovation," Ch. 15 in F. Scherer, In-
dustrial Market Structure and Economic Performance (Chicago: Rand McNally,
1970).
Schmookler, T., Invention and Economic Growth (Cambridge, MA: Harvard Uni-
versity Press, 1966).
Schumpeter, J., Theory of Economic Development (London: Oxford University Press,
1934).
-, Capitalism, Socialism and Democracy (New York: Harper & Row, 1942).
Wilson, R. W., "The Effect of Technological Environment and Product Rivalry on R&D
Effort and the Licensing of Inventions," Review of Economics and Statistics, LIX
(May 1977), 171-78.

This content downloaded from


217.248.61.125 on Fri, 02 Apr 2021 08:51:51 UTC
All use subject to https://about.jstor.org/terms

You might also like