Nelson winter1982

American Economic Association
The Schumpeterian Tradeoff Revisited
Author(s): Richard R. Nelson and Sidney G. Winter
Source: The American Economic Review, Vol. 72, No. 1 (Mar., 1982), pp. 114-132
Published by: American Economic Association
Stable URL: http://www.jstor.org/stable/1808579
Accessed: 26/10/2010 17:06
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=aea.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
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.
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The
American Economic Review.
http://www.jstor.org

The Schumpeterian Tradeoff Revisited
By RICHARD R. NELSON AND SIDNEY G. WINTER*
The image of the competitive process pre-
sented by Joseph Schumpeter in his Capi-
talism, Socialism, and Democracy has direct
and obvious relevance to the economy of
today, a straightforwardplausibility that the
textbook account of competition cannot
match. In electronics, pharmaceuticals, and
many other industries it is plain that compe-
tition among firms centrally involves their
R&D policies, successes, and failures. And,
as Schumpeterstressed, over the long run the
gains to society from continuing innovation
are vastly greater than those associated with
competitive pricing.
In articulating this view of competition,
Schumpeter also put forth what has been
called "The Schumpeterian Hypothesis": A
market structureinvolving large firms with a
considerable degree of market power is the
price that society must pay for rapid techno-
logical advance. Thus there is a tradeoff be-
tween static efficiency, in the sense of prices
close to marginal production cost, and dy-
namic progressiveness. It is not clear how
much choice Schumpeter thought society ac-
tually had regarding the mix between static
efficiency and dynamic progressiveness, but
many contemporary economists clearly write
as if they think that market structure is a
variable potentially under tight public con-
trol.
It is important here to distinguish between
Schumpeter's general propositions about the
nature and social value of competition in
technologically progressiveindustries and the
specific viewpoint on the role of market
structure represented by the Schumpeterian
hypothesis. One can accept the value of the
formerwhile remainingopen-minded, or even
skeptical about the latter. And one can re-
gard the analysis of Schumpeterian competi-
tion as constituting a promising research
agenda on which a start has barely been
made, while considering that sharply di-
minishing returns may have set in some time
ago for certain lines of effort directed to the
narrowerquestion.'
This is the third of a series of papers
in which we have set forth a model of
Schumpeterian competition, explored some
of the links between technological advance
and marketstructurecontained in that model,
and examined the influence of background
conditions upon those links and on industry
performance more generally. In our 1977
article, we studied how initial market struc-
ture influenced industry technological pro-
gress and price behavior, under a particular
set of assumptions about the cost of innova-
tion and appropriability conditions. In that
paper we noted a tendency for concentration
to grow in industries that start out initially
unconcentrated. The evolution of industry
structure over time, and in particular the
conditions stimulating and retarding growth
of concentration, was the topic of our 1978
article. In the present paper the inquiry is
extended in several ways, most notably in
the treatment of different regimes of exoge-
nous change in technological opportunity.
The focus is on the competitive contest
among innovators and imitators, on how
various technological and institutional condi-
tions, some of which may be subject to in-
fluence by government policy, determine the
nature of that contest, and on the innovation
and price performance of the industry.
*Professors of economics, Yale University. We are
indebted to Peter Reiss for research assistance and to
members of the Yale microeconomics workshop for
useful comments on an earlier draft. Our colleague,
Richard Levin, was especially helpful. We also benefited
from the comments of an anonymous reference. The
work has been supported by grants from the Sloan
Foundation and the National Science Foundation.
IFor an early review of the discussion, see Nelson,
Merton Peck, and Edward Kalachek. For more recent
reviews, see Morton Kamien and Nancy Schwartz, and
F. M. Scherer. Theoretical models of Schumpeterian
competition have recently been put forward by Partha
Dasgupta and Joseph Stiglitz (1980a, b), Glenn Loury,
Carl Futia, and M. Th&ese Flaherty.
114

VOL. 72 NO. I NELSON AND WINTER: SCHUMPETERIAN TRADEOFFS 115
I. TheComplexStructureof the
SchumpeterianArguments
Before presenting a specific model, it is
helpful to consider the issues more generally.
It is plain that the particular view of tech-
nological change taken in most analyses of
Schumpeterian competition involves a con-
siderable abstraction. Many variables in-
fluence the pace and pattern of technological
progress in an industry. There are many dif-
ferent sources of innovation, and many kinds
of policies impinge upon them. However, it
is characteristic of discussion relating to the
Schumpeterian arguments to focus on one
class of innovators-firms in the industry
and on policy influencing the structure (in
some sense) of the industry. We will stick
with these conventional ground rules.
Even within these constraints, a wide range
of possible relationships need to be articu-
lated and examined. We propose there are
three connected, but distinguishable, classes.
One of these concerns the relations between
market structure and the innovation that is
generated. A second involves the nature of
possible tradeoffs between static efficiency
and technological progressiveness implicit in
these relationships. Third, there are ques-
tions regardingthe policy tools available and
their influence over time on structure and
innovativeness.
A. TheRelationshipBetweenMarket
Structureand Innovation
Much of Schumpeter's discussion in Cap-
italism, Socialism, and Democracy stressed
the advantages for innovation of a firm being
large, and was not focused on market struc-
ture per se. He argued that there were
innovation "capability advantages" of large
firm size stemming from economies of scale
in R&D and management, greater capabili-
ties for risk spreading, finance, etc. Almost
certainly, he also had in mind "appropriabil-
ity advantages" of large firms over small
ones. In the economic world of Capitalism,
Socialism, and Democracy, as in the eco-
nomic world of his earlier work, The Theory
of Economic Development, the returns to in-
novation stem from the transient monopoly
of a new product or process provided by
imitator lag. Where patent protection is
spotty and imitation may occur rapidly, the
payoff to an innovator may depend largely
on his ability to exploit that innovation over
a relatively short period of time. Large firms
have a level of production, productive capac-
ity, marketingarrangements,and finance that
enables them quickly to exploit a new tech-
nology at relatively large scale.
The argument that large firms can be more
efficient in R&D, and can quickly reap the
advantages of large scale use of an innova-
tion, has been countered by arguments that
the bureaucratic control structure of large
firms may partially or even fully offset these
latent advantages. While there are extant
theoretical models that have tried to capture
the roles of scale economies in R&D and
appropriability advantages of large size, to
our knowledge there has been no explicit
modeling that tries to come to grips with the
internal control issues.2 This remark applies
to the model we shall present here, as well as
to other models of Schumpeterian competi-
tion.
In the argument above what is favorable
to innovation is firm size, not market power
per se. To the extent that some minimal scale
is necessary for innovation it is of course
possible that the necessary scale may be
achievable only by a monopolist in a product
field, or by a structure involving just a few
firms. But arguments that market power per
se is important to induce innovation must be
of a different stripe.
One such argument is that absence of sig-
nificant competitors is a variable which in its
own right can increase appropriability. Put
another way, market structureinfluences the
speed with which transient quasi rents are
eroded away by imitators. A related but dis-
tinguishable argument is this: absence of
competition, or restrained oligopolistic com-
petition, by leading to high rates of return in
the industry generally, can serve to shelter
firms that do innovative R&D in circum-
stances where, if competition were more ag-
gressive, firms that aim for a "fast second"
2
David Teece has at least begun such an inquiry.

116 THE AMERICAN ECONOMIC REVIEW MARCH 1982
would drive the real innovators out of busi-
ness.
Of course, as a number of commentators
remarked,weak competition may reduce the
spur to innovation. This argument, like the
one about bureaucratic obstacles to innova-
tion in large firms, has not yet been ade-
quately modeled.3 Nor do we treat it in our
model.
Finally, we note that while most analyses
of the connections between market structure
and innovation have viewed the causation as
flowing from the former to the latter, it is
clear that under Schumpeterian competition
there is a reverse flow as well. Successful
innovators who are not imitated quickly may
invest their profits and grow relative to their
competitors. Similarly, a firm that plays an
effective "fast second" strategy may come
ultimately to dominate the industry. As we
stressed in our 1978 paper, market structure
should be viewed as endogenous to an analy-
sis of Schumpeterian competition, with the
connections between innovation and market
structure going both ways. It is surprising
that studies concerned with the Schum-
peterian hypothesis have largely neglected
the reverse causal linkage. An important ex-
ception is Almarin Phillips' study of the
aircraft industry. Recently, Richard Levin
has obtained promising results in exploratory
empirical work on Schumpeterian competi-
tion that takes the endogeneity of structure
into account.
B. TradeoffIssues: Costsand Benefits
Much of the economist's interest in the
Schumpeterian controversy stems from the
observation that aspects of structurethat are
conducive to innovation may be detrimental
to the achievement of Pareto optimality in
the short run. However, like the analysis of
the market structure-innovativenesslinks, the
discussion of the "tradeoff" has tended to be
oversimplified.
With the exception of some recent theoret-
ical work, economists have tended to identify
the deadweight triangle losses associated with
product market power as the sole static cost
of a progressive (concentrated) industry;
there are, however, several kinds of social
costs involved in Schumpeterian competi-
tion, not just one. One of these is, indeed, the
social costs associated with less than com-
petitive output levels. The presence of such
costs is signaled by a gap between marginal
production costs in the most progressivefirms
and price. However, static inefficiencies also
reside in the extent to which the "best tech-
nology" is monopolized by one firm in the
industry, independently of whether that firm
is large enough to have market power. In-
deed, there are two different kinds of costs
associated with limits on the use of technical
information imposed through the patent sys-
tem, or simply by industrial secrecy. One is a
higher average production cost than given
technological knowledge would permit, a cost
associated with a gap between best practice
and averagepractice. Another is the presence
of duplicative or near-duplicative R&D ef-
forts, resulting in a lower "best practice" for
a given amount of cumulative industry R&D
(or more R&D needed to achieve a given best
practice). In addition, of course, there is a
possible distortion of the level of total R&D
effort, which may be greater or less than it
would be in a hypothetical "second best"
optimum in which the other costs are
accepted.
Setting aside for the moment the last of
these considerations, the other costs can be
depicted as in Figure 1.4 Let C equal unit
production costs if industry R&D were spent
perfectly efficiently and all firms had access
to the best known technology. With a de-
mand curve A-B, potential consumer's plus
producer's surplus is triangle ABC. Let c
equal unit production costs with the actual
best practice technology (given some nearly
duplicative R&D efforts) and the jagged line
c-c' represent an industry production cost
schedule arraying costs by firms from the
most efficient (cost c) to the least efficient3More precisely, this issue of the "spur" to innova-
tion provided by more competitive market structure has
been formally treated in the context of models that
assume profit maximization. We do not think that this is
what the spur discussion is all about.
4This graphic representation was suggested to us by
Levin.

VOL. 72 NO. ] NELSON AND WINTER: SCHUMPETERIAN TRADEOFFS 117
P
A
pS
C - d
I d
4
FIGURE 1
firm actually producing (cost c'.) Let actual
output be Q. Actual surplus then is AP'c'c.
The difference between actual and potential
surplus is accounted for by three areas: 1)
P'de, or a conventional deadweight loss as-
sociated with the output shortfall relative to
competitive equilibriumat best practice cost;
2) cc'e, or excess production costs due to a
gap between best and average practice; 3)
CcdB associated with lower best practice due
to inefficient industry R&D.
If there are to be private incentives for
innovative effort, there must be some degree
of private property (at least de facto) in the
fruits of such effort; practically speaking,
this means that some social costs of the sorts
identified above must be accepted. How
should one then approach the problem of
assessing the social gains from more industry
R&D, so as to reach some conclusion regard-
ing the suitability of the total R&D effort
that private incentives would produce? In the
first place, the analysis above points out that
some key determinants of the social gains are
themselves endogenously determined by, or
codetermined with, market structure. For
example, it matters whether R&D expendi-
ture is efficiently allocated from a social
point of view, and this is partly a function of
market structure. But there are also im-
portant exogenous considerations. The gains
from a higher level of industry R&D spend-
ing depend partly on the technological re-
gime that governs R&D outcomes. For exam-
ple, it matters whether diminishing returnsto
industry R&D effort set in early and sharply,
or whether marginal returns continue to be
high at high levels of expenditure. In the
former case, but not the latter, low levels of
industry R&D expenditure may buy society
much of what a significantly higher level
would.
C. WhatAre thePolicy Issues?
While the discussion above identifies
causal links between innovative opportuni-
ties and the market structure that evolves, it
offers no reason to expect that the social
tradeoff might be optimized by the auto-
matic functioning of those links. It is quite
reasonable, therefore, for economists to be
interested in policies that can influence
structure. However, there seems to be a no-
tion implicit in many studies that industry
structureitself is the policy variable, and that
structure should be chosen so as to optimize
the tradeoff. Realistically, it is not clear how
far the policy tools presently existing or pro-
posed can go toward affecting structure in a
durable way. Even if it could be argued that
a lesser degree of concentration than prevails
in an industry would promote innovation,
and hence be desirable on both static and
dynamic grounds, a deconcentration achieved
by policy may get undone over time as the
competitive struggle continues in the now
more innovative environment. And repeated
structural interventions would certainly lead
to behavioral changes that might be costly in
themselves.
On occasion policy can influence structure
directly if not durably, as by requiring dives-
titures, or by forbidding or encouraging
mergers, or entry. But much of policy aimed
to influence industry performance operates
through constraining or requiring certain
behavior, and affects structure indirectly.
Government policies, regarding for example,
the rights and obligations of patent holders,
influence whether imitation is hard or easy.
Or, antitrust policy may or may not permit
firms with a large market share and a strong
technological position to exploit these
advantages and take a larger market share.
Such policies may influence performance di-

rectly, and also indirectly by influencing the
structure that evolves, which in turn de-
termines performance. A further complica-
tion of the policy task arises from the fact
that structure is likely to respond rather
slowly to feasible policy adjustments. It is at
least reasonable to propose that the time
patterns of social costs and benefits ought to
receive some consideration when policy mea-
sures affecting structure are analyzed.
In sum, much of the discussion relating
market structure to technological advance
and static efficiency does not really connect
with the policy instruments that are avail-
able. A serious analysis of policy must start
with recognition of the endogeneity of struc-
ture. The modeling effort that follows is a
step in this direction.
II. A Model
A. GeneralFormulation
While the discussion above provides, we
believe, a useful overview of the issues in-
volved in analyzing the Schumpeterian argu-
ments, it was quite general. In this section,
the framework is greatly simplified and a
model is presented that formalizes a number
of the key relations.5
The model is of an industry in which a
number of firms produce a single homoge-
neous product for sale in a market char-
acterized by a fixed downward-sloping de-
mand curve. All techniques capable of
producing this product are characterized by
constant returns to scale and fixed input
coefficients. Further, techniques differ by
equiproportional amounts in all input coeffi-
cients; productivity differences thus are
"Hicks neutral."
At any time each firm operates a single
technique, the best it knows, to the maximal
level permittedby its existing stock of capital,
purchasing needed complementary inputs on
factor markets. Input prices facing the in-
dustry are constant, independent of total em-
ployment or time. Thus, since all techniques
have the same ratio of other inputs to capital,
cost per unit of capital is constant across
firms and over time. But the cost of a unit of
output is a variable in the model, since out-
put per unit of capital (and all other inputs)
at any time will in general vary across firms,
and will increase over time as better tech-
niques are discovered and implemented.
Analysis of the Schumpeterian tradeoffs
would appear to require explicit recognition
of two different kinds of R&D activity-
innovative R&D which draws on a general
fund of relevant technological knowledge,
and imitative R&D which involves learning
about and copying the techniques of other
firms. We model firms as having R&D poli-
cies that are defined in terms of innovative
and imitative R&D spending per unit of
capital. Thus, as firms grow or decline, so do
their R&D expenditures on innovation and
imitation. Firms may, and likely will, differ
in terms of how much they spend on these
two different kinds of R&D (per unit of
capital). A central question to be explored is
the conditions under which firms that em-
phasize one or the other of these types of
R&D expenditure prosper or survive.
We model both kinds of R&D as a two-
stage sampling process. Within a given period
the probability that a firm may take a "draw"
on the set of innovation possibilites, or the
set of imitation possibilities, is proportional
to the firm's spending on these activities.
Hence, over a run of many periods, the real-
ized average number of innovation and imi-
tation draws per period is proportional to
the firm's average expenditures per period on
these kinds of R&D. In this model, there are
no economies of scale of doing R&D. Big
firms spend more on R&D than do small
firms and thus have a greater chance each
period of an R&D draw. But that increased
chance is only just proportional to their
greater spending. However, there are "ap-
propriability advantages" of large firm size.
Once a firm has acquired access to a new
technique through either innovative or imita-
tive R&D, it can apply that technique to its
entire capacity without further costs. We set
aside issues relating to the embodiment of
technical advance, and assume away any
possibilities that a large firm may be slower
to adopt a new technique found through
5This is a more analytic formulation than that con-
tained in our 1977 and 1978 articles. There also are
certain small changes in terminology and notation be-
tween this version and the earlier ones.

VOL. 72 NO. 1 NELSONAND WINTER: SCHUMPETERIAN TRADEOFFS 119
R&D than a small firm. Similarly, require-
ments to work with a new technique for a
time to get it to operate effectively are
ignored, as are the idiosyncratic characteris-
tics of techniques honed through learning by
doing. An imitation draw will, with certainty,
enable a firm to copy prevailing best prac-
tice. We exaggerate the distinction between
innovation and imitation by disregardingthe
costs, imperfections, and uncertainties that
afflict the imitation process after a suitable
target has been located. These are obviously
quite restrictive assumptions about the im-
plementation of innovation and imitation,
and our analysis can only be suggestive of
what might happen in other cases.
We consider two different specifications of
the distribution from which a firm samples if
it has an innovation draw. These different
regimes of technological change imply quite
different relationships between industry pro-
ductivity growth and industry R&D spend-
ing.
In one of these regimes, which we shall
call the "science-based" case, we view the
distribution sampled by an innovative R&D
draw as improving over time as a result of
events going on outside the industry, for
example, advances in fundamental science
occurring in universities. At any time, firms
sample from a log-normal distribution of
values of the average productivity of capital.
(Recall that all other inputs are proportional
to capital in all feasible techniques.) The
mean (log) of this distribution increases over
time at a rate we call the rate of growth of
"latent productivity." Under this specifica-
tion, what a firm finds today as a result of an
innovation draw is independent of what it
might have found last year or the year be-
fore. And the population being sampled is
better than the one sampled earlier. Innova-
tive R&D by a firm can be interpreted as its
efforts to keep up with a moving set of new
technological possibilities created outside the
industry. Less R&D by a firm or by the
industry as a whole means that the time path
of actual productivity tends to be lower in
the range of exogenously determined possi-
bilities.6
In the other regime, which we call the
"cumulative technology" case, a firm in-
novates by making incremental improve-
ments in its own current technique-not by
drawing on new knowledge created external
to the industry. An innovation draw is, in
effect, a draw on a constant distribution of
proportional increments to its prevailing pro-
ductivity level. Small increments are more
likely than large ones. An innovative R&D
success buys a firm not only a better tech-
nique for the current period, but a higher
platform for the next period's search.
In the context of our model, "market
structure" simply refers to (some statistical
measure of) the degree of concentration of
output or capital. There are no product dif-
ferentiation issues to qualify that interpreta-
tion, and we assume no new entry into the
industry. Market structure and behavior af-
fect the survivability of innovative R&D in
two different ways. First, the sizes of the
other firms in the industry, and their policies
regarding spending on imitative R&D, in-
fluence the amount of time a firm has after
making an innovation before a large share of
the rest of the industry is able to copy it.
Also, market structure and behavior in-
fluence the extent to which firms that have
policies that are not profitable can survive.
Market structure evolves endogenously.
Given the capital stocks and techniques of
the firms in a particular period, output for
that period is determined. The demand curve
then determinesprice, and productivity levels
(given input prices) determine production
cost. For each firm the ratio of price to unit
production cost-which we call the price-cost
margin-then is determined. (Given the as-
sumptions that all inputs are proportioned to
capital and all input prices constant, the rate
of return on capital in production, i.e., ab-
stracting R&D costs, is monotonically related
to the price-cost margin.) We assume that a
firm's desire to expand or contract is
governed by its price-cost margin and its
6Stephen Homer's dissertation examines analytically
some stochastic process models of technological change
akin to that incorporated in our own model. Recent
unpublished work by Katsuhito Iwai obtains interesting
results regarding roughly the same range of problems
with the aid of an assumption that the number of firms
is very large. Both of these works are concerned in
particular with examining the lag (or average ratio)
between latent productivity and average realized pro-
ductivity.

prevailing market share, within constraints
set by the assumed physical depreciation rate
of capital and the firm's ability to finance
investment.
For firms of a given size, the greater the
margin of price over production cost, the
greater the desired proportional expansion.
And, the greater the price-cost margin, the
greater both retained earnings of the firm
and its ability to persuade the capital market
to provide finance. However, R&D expendi-
ture, like production costs, acts as a drag on
retained earnings.
Firms with large market shares recognize
that their expansion can spoil their own
market. The larger a firm's current market
share, the greater must be the price-cost
margin needed to induce a given desired
proportional expansion. By varying the shape
of this relationship, a spectrum of possible
patterns of investment behavior may be re-
presented. In one extreme the expansion of
the firm is not checked by market share at
all; the firm responds as price taker regard-
less of its share. Another investment strategy
we call the "Cournot"strategy- a firm picks
a target capital stock to maximize its profits
under the assumption that all other firms in
the industry keep their output constant.
More formally, the model has the follow-
ing structure:
(1) Qit = AitKit-
The output of firm i at time t equals its
capital stock times the productivity of the
technique it is employing.
(2a) Qt = zQit = -AitKit;
(2b) Pt= D(Qt).
Industry output is the sum of individual firm
outputs. Price is determined by industry out-
put, given the product demand-price func-
tion D(Qt).
(3) 7rit= ( PtAit- c- rim -rin )
The profit on capital of that firm equals
product price times output per unit of capital,
minus production costs (including capital
rental) per unit of capital, minus imitative
and innovative R&D costs per unit of capital.
A firm gets an innovation or imitation
draw in period t with respective probabili-
ties:
(4) Pr(dint= 1)=- anrinKit;
(5) Pr(dimt = I)=- amrimKit.
(Parameters are chosen so that the upper-
bound probability of one is not encountered.)
If a firm gets an innovation draw, it samples
from a distribution of technological oppor-
tunities F(A; t, Ait). This distribution is a
function of time and independent of a firm's
prevailing technique in the science-based
case. It is independent of time per se but
dependent on the firm's prevailing technique
in the cumulative technology case. If a firm
does get an imitation draw, it then has the
option of observing and copying industry
best practice.
For a firm that obtains both an imitation
and innovation draw in the particularperiod,
its following period's productivity level is
given by
(6) Ai + Max (Ait A't,~4ij
Here At is the highest best practice produc-
tivity level in the industry in period t, and Ait
is a random variable that is the result of the
innovation draw. Of course, the firm may
fail to obtain an imitation draw, an innova-
tion draw, or both, in which cases the menu
from which next period productivity is drawn
is shorter.
A firm's desired expansion or contraction
is determined by the ratio of price to produc-
tion cost, P/(c/A), and its market share. But
a firm's ability to finance its investment is
constrained by its profitability, which is af-
fected by its R&D outlays, as well as its
production outlays.
(7 i,t+ l = I(c(O,7it ) it

Here, 8 is the physical depreciation rate and
the gross investment function I(-) is con-
strained to be nonnegative; it is nonde-
creasing in its first argumentand nonincreas-
ing in the other two. Also, we assume
1imS01(l, s,0) 6, that is, a firm that has
price equal to unit cost, negligible market
share and zero R&D expense will have zero
net investment.
There are two key differences between our
model and the other recent models of
Schumpeterian competition cited in foot-
note 1. The strategies or policies assumed of
our firms are not derived from any maximi-
zation calculations. And the industry is not
assumed to be equilibrium.
An essential aspect of real Schumpeterian
competition is that firms do not know ex
ante whetherit pays to try to be an innovator
or an imitator, or what levels of R&D ex-
penditures might be appropriate. Indeed, the
answer to this question for any single firm
depends on the choices made by other firms,
and reality does not contain any provisions
for firms to test out their policies before they
actually commit to them. Thus there is little
reason to expect equilibrium policy config-
urations to arise. Only the course of events
over time will determine and reveal what
strategies are the better ones. And even the
verdict of hindsight may be less than clear,
for differences in luck will make the same
policies brilliantly successful for some firms
and dismal failures for others.7
To understand the process of industry
evolution, we have chosen to focus initially
on cases in which firm R&D policies are
strictly constant over time. This might be
defended as an approximation to empirical
reality by some combination of arguments
involving high set-up and adjustmentcosts in
real R&D programs, bureaucratic sluggish-
ness, and difficulties in distinguishing signal
from noise in the feedback on a prevailing
policy. But, in our view, a more fundamental
justification for this approach is methodo-
logical. If competition is aggressive enough
and the profitability differences among poli-
cies arelarge enough, differential firm growth
will soon make the better policies dominate
the scene, regardless of whether individual
firms adjust or not. If, however, the model
sets the stage for an evolutionary struggle
that is quite protracted (as those in reality
often are), then admitting policy change at
the individual firm level is unlikely to change
the generalindustryenvironment much-and
it certainly complicates the task of under-
standing the dynamic process. To forego the
attempt to understand the process, on the
other hand, is to leave open the question of
the promptness and efficacy of the forces
pressing the system toward equilibrium. It is
also to overlook the shaping role of differen-
tial firm growth as a determinant of the sort
of equilibrium toward which the industry
may be moving. While it would not be dif-
ficult to augment the model by admitting
adaptive R&D policies, in order to focus on
the selection dimension of competition we
have chosen not to do this.
The model defines a stochastic dynamic
system in which productivity levels tend to
rise and unit production costs to fall over
time as better technologies are found. As a
result of these dynamic forces, price tends to
fall and industry output tends to rise over
time. Relatively profitable firms expand and
unprofitable ones contract, and those that do
innovative R&D may thrive or decline. In
turn, their fate influences the flow of innova-
tions.
A number of conclusions about the behav-
ior of the model sketched above can be de-
rived by analytical methods. These relate,
however, to special cases or to individual
components of the model considered in iso-
lation. We cannot, at this point, obtain by
analytical methods conclusions about inter-
esting features of the population of industry
histories that the model implies. And, of
course, analytic methods cannot be used to
explore the implications of the assumption
that all model parameters are in "plausible
ranges." For these purposes we have turned
to simulation. We exploit, however, the fact
that the model is sufficiently transparent in
7The underlying theoretical viewpoint of the present
paper is set forth in detail, and its application amply
illustrated, in our forthcoming book on evolutionary
economic theory.

its logic so that it is easy to develop reason-
able hypotheses about its behavior, hypothe-
ses which can then be tested by reference to
simulation results.
B. SimulationExperiments
We undertook three sets of simulation ex-
periments. They have a common focus on
the conditions for the survival of innovative
R&D in an industry and on the interplay
between innovation and industrial con-
centration. These focal concerns are reflected
in two aspects of the stylized evolutionary
processes we have selected for study. First, in
each experiment half of the firms do innova-
tive R&D as well as imitative R&D, while the
other half do imitative R&D only. (In this
respect the design here is similar to that in
our 1977 article. In our 1978 paper, all firms
engaged in both innovative and imitative
R&D.) Imitative R&D spending per unit
capital is constant across all firms; similarly,
spending on innovative R&D per unit capital
is the same in all the firms that spend at a
postive level. Secondly, we have in each case
established initial conditions in which all
firms are the same size, and the industry is
approximately in equilibrium at the prevail-
ing productivity level. We have also ruled
out entry. These assumptions give us a clear
innovator-imitator struggle and a definite
reference point for the degree of concentra-
tion that exists in the industry at any stage of
the process.
The numbers used to calibrate the model
were chosen so that four simulation "periods"
correspond to one calendar year. Under this
interpretation, the average annual sales to
capital ratio in the industry is in rough accord
with that in technologically progressive
industries. The rate of growth of latent pro-
ductivity is roughly 2 percent per year in our
slow growth condition; 6 percent per year in
the fast growth condition. The ratio of
innovative R&D spending to sales is 6 per-
cent. In the "hard"imitation setting, a given
probability of imitating best practice required
twice as much imitative R&D as in the "easy"
imitation setting. We ran the simulation for
101 periods, or twenty-five years after initial
conditions.
Within this framework, the three sets of
experiments described below correspond to
more specific contexts. In each set we
explored the behavior of the system under
different settings of particular parameter
values.
1. Behavior and Performance in a Science-
Based Oligopoly
Our first set of experiments was concerned
with the behavior of an industry that starts
out initially highly concentrated, and in
which the target markup formulas imputed
to firms produce a regime of relatively
restrainedcompetition. The industry involved
is science based in that latent productivity
evolves over time at a rate determined by
outside forces. All firms start out with pro-
ductivity levels close to then-prevailing latent
productivity.
In the context of this science-based oligop-
oly, we wanted to explore the effect on
behavior and performance of the following
two variables: rate of growth of latent pro-
ductivity, and ease of imitation. To what
extent does the evolution of the industry, in
particular the relative success of the firms
that do innovative R&D and of those that
just lay back, depend on the pace at which
new R&D targets evolve over time? How
does the ease with which one firm can im-
itate the technology of another influence the
evolution? The runs here are, we think, inter-
esting in their own right. They also are inter-
esting as a base of comparison for simulation
experiments in which initially the industry is
less concentrated. Understanding the results
of the four-firm runs greatly facilitates one's
ability to see through the workings of the
model more generally, and thus to interpret
what is going on in other contexts.
Our past experience with running our
model industry, starting it out with a quite
concentrated structure, led us to predict the
following.8 First, latent productivity will be
trackedrelativelywell by the innovative firms
because their level of innovative R&D spend-
8Of course, the relevance of our experience is partly
attributable to the fact that we are working here with
roughly the same values for the model's parameters that
we have explored in the past.

ing, given the other parametersof the model,
is sufficiently high so that each samples the
evolving distribution of new technological
possibilities relatively frequently. Second, the
imitating firms will trackthe innovating firms
relatively closely because their level of imita-
tive R&D expenditure is sufficiently high
that, even under the setting that we define as
difficult imitation, it is unlikely that many
periods go by without their being able to
imitate the best technology in practice. Third,
under some circumstances firms that do not
bear the costs of innovative R&D may tend
to be slightly more profitable than the firms
that do, but the innovative firms will never-
theless survive, make money, and even grow
relative to the imitators. The investment re-
straint shown by the farge firms in this con-
text limits the extent to which a firm more
profitable than its competitors exploits that
advantage by trying significantly to increase
its market share. In this atmosphere of
restrained competition, firms that are not
maximallyprofitable are sheltered.And, even
if they have a run of bad luck, their capaci-
ties for recuperationare not quickly eroded.
Regarding the social merit of the industry
behavior considered in these runs, certain
predictions are relatively obvious. One is that
the industry will be characterized by high
markups over production cost; thus, there
will be significant "trianglelosses." A second
is that, compared with a more competitive
industry, there will be less waste due to the
use of inferior technologies (there will be less
of a gap between average and best practice).
Third, R&D spending will be more efficient
in the sense that less total R&D outlay is
required to keep average practice moving at
a given distance from potential best practice.
Whether price will be higher or lower seems
an open question.
2. A More Competitive Science-Based In-
dustry
In the second simulation experiment we
started the industry initially with sixteen
equal-sized firms, eight of these doing both
innovative and imitative R&D, and eight
doing only imitative R&D. As in the four-firm
case, in some of our runs latent productivity
advanced at a rapid rate, in others latent
productivity grew more slowly, and we set
imitation as "easy"in half of our runs, "hard"
in the other half. One important difference
between the range of experimental variation
in the sixteen-firm runs and the four-firm
runs is this: in some of the sixteen-firm runs
we preserved the assumption, contained in
the four-firmruns, that a firm's target margin
over cost increases sharply with its market
share; in others, we explored what would
happen if a firm with a substantial market
share continued to expand rapidly even if it
only had a moderate gap between price and
cost.
There are several things that can be pre-
dicted with some confidence about industry
performance in the sixteen-firm case as con-
trasted with the four-firm case. Above we
discussed the expected differences in average
price-cost margins, the gap between average
practice and best practice, and R&D ef-
ficiency. Another expected difference be-
tween the four-firm and sixteen-firm runs is
that, in the latter, the initial industry struc-
ture will not prove to be stable. In general,
concentration can be expected to increase
under the force of Schumpeterian competi-
tion. There is one particular competitive
struggle we will want to watch attentively
and to consider as a function of the various
parameter settings-the performance and
survivalof the firms that do innovative R&D,
compared with those that do imitative R&D.
A straightforwardconjecture is that the abil-
ity of innovative firms to prosper and survive
depends positively on the rate of latent pro-
ductivity growth (which determines the aver-
age advance achieved through innovative
R&D) and negatively on the ease of imita-
tion.
3. A CompetitiveIndustrywith a Cumulative
Technology
In this third set of simulation experiments
we preserved the basic assumptions of the
second set, except for the following. Instead
of assuming that innovative R&D yielded
random draws on a moving distribution of
technological possibilities, we assumed that
innovative R&D involved the incremental
improvement of prevailing techniques. The
probability distribution of innovative R&D

successesfor a firm was clusteredon tech-
niquescloseto its existingproductivitylevel.
(Thistechnologicalregimewasnot studiedin
our earlierpapers on Schumpeteriancom-
petition. It is reminiscent, however, of
our earlierevolutionarymodel of econom-
ic growth, see our article with Herbert
Schuette.)
Thetwoassumptionsaboutrateof growth
of latent productivitybuilt into the earlier
twoexperimentswerereplacedin thisexperi-
ment by the followingtwo. Under one as-
sumption,thedistributionof innovativeR&D
funds is packedclose aroundthe prevailing
techniqueanda majoradvancein productiv-
ity is unlikely.Under the otherassumption,
the distributionis morespreadout, and the
firmhasa largerchanceof comingup witha
significantincreasein productivityfrom a
singleinnovation.We attemptedin choosing
the parametersettingsto bring about real-
izedratesof productivitygrowthcomparable
to thosein the science-basedcases.
We variedthe parametersettingsregard-
ing "easeof imitation"in the same way in
theserunsas in the earliersixteen-firmruns.
Wealsoexploredthedifferencethatit would
makeif firmsas they grewlargedid, or did
not, exertinvestmentrestraint.
The most importantcontrastbetweenthis
set of runsandthe earlierset of sixteen-firm
runs, of course,is this: in the formercases
thereis an exogenousforcepushingforward
the frontiersof knowledge.If only a little
innovativeR&Dis done in the industry,the
averagesuccess from that effort will be a
relativelyspectacularadvance,reflectingthe
forward movement of latent productivity
since the priorR&D success.In the present
experimentalcontext, thereis no such out-
side force. If industryinnovativeR&D is
squeezeddown by the dynamicsof Schum-
peteriancompetition,the rateof progressof
best practice and averagepractice can be
predictedto declinealso.
III. TheResultsof theSimulationExperiments
A. TightOligopoly
We report,first,on our simulationexperi-
mentwithanindustrystructureconsistingof
four firms, initially of equal size, two of
whichspendonly an imitativeR&D.Table 1
displaysrelevantdata from these runs.The
firstrowdisplaysforperiod101the shareof
industry capital held by firms that do
innovativeR&D.In general,shareless than
.50 indicatesthat firms that did innovative
R&Dwerelessprofitablethanthosethatdid
not; a shareslightlyover.50is, however,also
compatiblewith inferior profitability.The
secondrowshowsthe averagerateof excess
return(i.e.,overandabovethecapitalrental
rate) in the industryas a whole, as a per-
centage per quarter. Rows 3 and 4 show,
respectively,the percentagemarginbetween
priceandaverageproductioncosts,P - C/C
in the industry as a whole, and the per-
centagemarginfor the firmwith best prac-
ticetechnology.
The next five rows display statistics of
industryconcentrationin period 101: rows
5-8, the shareof industryoutputaccounted
for by the firstand secondlargestinnovator
and imitators;and row9, the inverseof the
Hirschman/Herfindahlindex of (capital)
concentration,that is, the "numbersequiva-
lent"of thevalueof theconcentrationindex.
The next blockof rowsoutlineproductiv-
ity statistics:row 10, averageindustrypro-
ductivity;rows 11, 12, averageproductivity
of innovatorsand imitators,separately;row
13, best practice,all for the final periodof
the run;and row 14, a measureof the aver-
age relativegap over 101 periods between
averageproductivityandlatentproductivity.
Rows 15 and 16 show total industryex-
penditureon innovativeR&D over the 101
periods,andthenumberof innovationdraws.
The bottomrow, 17,is pricein period101.
The symbolsS andF, andE andH, atop
thecolumns,referto slowandfastgrowthof
latent productivity,and easy and hardimi-
tation.The two columnsundereachheading
show the resultsof the differentruns with
the sameparametersettings.
The strikingcharacteristicof all the runs
shown in Table 1 was that for the reasons
explained above, the four firms tended to
stay close togetherboth in productivityand
in size. Under these circumstancesit is not
surprisingthat the rate of growth of best
practiceproductivity,and the rateof growth

TABLEI-FOUR-FIRM RUNS
SE FE SH FH
(1) (2) (1) (2) (1) (2) (1) (2)
1. Innovators'Capital Share .51 .49 .50 .50 .52 .50 .50 .51
2. Excess Return (%Qtr) 4.8 4.8 4.7 4.7 4.8 4.8 4.8 4.8
3. MarginoverAverageCost(%) 33.7 33.3 33.0 31.3 33.2 33.2 32.4 31.4
4. Marginover Best Practice (%) 33.7 33.3 33.0 31.3 33.2 33.2 47.8 38.0
5. Output Share: Largest Innovator .26 .25 .25 .26 .27 .25 .27 .28
6. Output Share: Second Innovator .26 .24 .25 .24 .24 .25 .25 .22
7. Output Share: Largest Imitator .26 .25 .25 .26 .25 .25 .25 .27
8. Output Share: Second Imitator .23 .25 .25 .25 .23 .25 .23 .22
9. Herfindahl Numbers Equivalent 4.0 4.0 4.0 4.0 4.0 4.0 4.0 4.0
10. Average Productivity .28 .28 .82 .94 .28 .29 .86 .75
11. Innovators'Average Productivity .28 .28 .82 .94 .28 .29 .89 .75
12. Imitators'Average Productivity .28 .28 .82 .94 .28 .29 .83 .75
13. Best Practice Productivity .28 .28 .82 .94 .28 .29 .96 .79
14. Average Productivity Gap .05 .04 .22 .18 .03 .08 .15 .17
15. Total Expenditureon Innov. R&D 156. 153. 156. 157. 156. 155. 153. 154.
16. Innovation Draws 77. 76. 79. 78. 72. 88. 62. 81.
17. Price .75 .76 .26 .22 .76 .73 .25 .28
Note: SE denotes the combination of slow latent productivity growth and easy imitation; FE denotes the combination
of fast latent productivity growth and easy imitation; SH denotes the combination of slow latent productivity growth
and hard imitation; FH denotes the combination of fast latent productivity growth and hard imitation.
TABLE2-SIXTEEN-FIRM RUNS; SCIENCE-BASED,INVESTMENTRESTRAINT
SE FE SH FH
(1) (2) (1) (2) (1) (2) (1) (2)
2. Excess Return (%Qtr) .77 .77 .66 .77 .87 .92 .69 .81
3. Marginover Average Cost(%) 7.9 9.8 7.5 .83 .96 10.8 11.9 12.5
4. Marginover Best Practice(%) 12.1 13.7 27.4 12.3 13.6 15.1 37.9 35.5
7. Output Share:Largest Imitator .08 .12 .15 .12 .11 .12 .09 .20
13. Best Practice Productivity .27 .29 .96 .85 .28 .28 1.06 1.00
14. Average Productivity Gap .06 .06 .02 .12 .04 .03 .09 .05
15. Total Expenditure on Innov. R&D 196. 206. 197. 187. 200. 186. 242. 210.
17. Price .66 .63 .21 .21 .65 .65 .21 .22
Note: See Table 1.
of averageproductivity(or theirlevelsat the
end of the run)tendedto be functionsonly
of the rateof growthof latentproductivity.
These variableswereinsensitiveto the ease
or difficultyof imitation.
Table2 presentscomparabledataforruns
that startedout with sixteenfirmsof equal
size. Whatarethe interestingdifferencesbe-
tween the sixteen- and four-firmruns, for
comparableotherparametervalues?
First,for a givenrateof growthof latent
productivity,thereare not many noticeable
differencesbetween the rate of growth of
best practicein the four-firmruns as com-

126 THEAMERICAN ECONOMIC REVIEW MARCH 1982
paredwith the sixteen-firmruns. However,
averagepracticetendedto be higherin the
four-firmrunsthanthe sixteen-firmruns.In
thesenseof makingthemostwidespreaduse
of the best availabletechnologies,as pre-
dictedabove,themoreconcentratedindustry
structurescored better than the less con-
centratedone. Also, the more concentrated
industrygenerallyachievedthis betterpro-
ductivitygrowthperformancewith a smaller
aggregatevolume of innovative R&D ex-
penditure.The concentratedindustrystruc-
turewasmoreefficientin its use of R&D.
On the otherhand,averagemarkupsover
variablecostsweresignificantlyhigherin the
four-firmrunsthanin the sixteen-firmruns.
The static triangleefficientlylosses, there-
fore,weregreaterthere.And in the sixteen-
firm/four-firmrun comparison,the higher
markupsmorethanoffset thehigheraverage
productivityin the former,so thatpricewas
higherin theconcentratedindustrycasethan
in thesixteen-firmcase.9
B. A More CompetitiveScience-Based
Industry
We havediscussedmanyof the important
differencesbetween the four-firmand six-
teen-firmrunsabove.A furthercontrastbe-
tweenthefour-firmrunsandthesixteen-firm
runswasa tendencyforindustrystructureto
changesignificantlyin the latter,but not in
the former.The initial distributionof firm
sizes tended to be moderatelystable where
latentproductivitygrowthwas slow (regard-
less of theeaseof imitation),or wherelatent
productivitygrowthwas rapidbut imitation
easy, so long as profitablefirmsshowedre-
straintregardingtheir output expansionas
they grewlarge.In runswith thesesettings,
the firmsthat did not engagein innovative
R&Dtendedto be moreprofitablethanthose
that did. In other words, innovativeR&D
was not profitableto undertake.But given
the output restraintshowedby the slightly
more profitableimitators,competitionwas
orderlyenoughso that innovatorswerenot
drivenout of business.
Where latent productivity growth was
rapid,andimitationhard,the firmsthatdid
innovativeR&Dfaredmuchbetter,and the
firms that only imitated did much worse,
even thoughprofitablefirmsshowedoutput
growth restraint.In these runs it was ap-
parentthattheimitatorsweregraduallybeing
drivenout of businesseven though the in-
novatorswereshowingconsiderablerestraint
in pushingtheiradvantage.The asymmetry
withthecaseabovewillbe explainedshortly.
Table3 displaysstatisticsforrunsin which
profitablefirmscontinuedto expandaggres-
sivelyeven when they grewlargerelativeto
the market.The comparisonbetweenTables
2 and 3 is quiteinteresting.Undercompara-
ble conditions,in eachandeverycasewhere
profitable firms did not show output re-
straint,the fate of the innovatorswas less
fortunaterelativeto the imitatorsthanwhen
large profitablefirms did show output re-
straint.In everyone of the runs of the ag-
gressivecompetitioncase,by theendinnova-
tors countedfor significantlyless than half
of theindustrycapitalstock.
Thecomparisonis particularlystrikingfor
the caseof rapidgrowthof latentproductiv-
ity and hard imitation. Under restrained
competition,theinnovatorsclearlydominate
and have 70 percent or more of industry
capital by the end of the run. When firms
have aggressiveinvestmentpolicies,the imi-
tatorsprevail,and by much the same deci-
sive margin.Althoughit would be easy to
understandwhy the intensityof the struggle
might affect the marginof "victory,"it is
somethingof a surpriseto find that it also
affects, quite systematically,the identityof
the victor.
Theexplanationforthisphenomenon,and
for the asymmetryof resultsin the invest-
ment restraintcase, residesin the following.
In thismodelanimitatornevercanachievea
higherproductivitylevelthanthebest of the
9In our 1977 article, we probed a wider and denser
range of initial distribution of firm sizes. In the earlier
experiment, as the current one, end-period price tended
to be lower in the sixteen-firm case than in the four-firm
case. However, the runs that initially started out with
eight equal-size firms tended to have roughly the same
end-period price as those that started out with sixteen
equal-size firms. And the runs with thirty-two initially
equal-size firms tended to have higher end-period prices
than the runs that started out with eight and sixteen
equal-size firms.

VOL. 72 NO. ] NELSON AND WINTER: SCHUMPETERIAN TRADEOFFS 127
TABLE3-SIXTEEN-FIRMRUNS;SCIENCE-BASED,AGGRESSIVEINVESTMENT
SE FE SH FH
() (2) () (2) () (2) () (2)
2. Excess Return (%Qtr) .19 .19 .03 .09 .28 .21 -.01 -.15
3. Marginover Average cost(%) 2.3 2.9 1.1 1.8 4.7 4.9 .1 -1.1
4. Marginover Best Practice (%) 2.3 9.6 13.0 4.7 12.7 8.9 6.0 4.1
13. Best Practice Productivity .26 .29 .95 .72 .28 .27 .76 1.02
14. Average Productivity Gap .02 .04 .11 .10 .00 -.05 .05 .33
15. TotalExpenditureonInnov.R&D 162. 165. 174. 176. 174. 164. 175. 188.
17. Price .63 .60 .19 .23 .63 .65 .23 .16
Note: See Table 1.
innovators.If he matchesaninnovator'spro-
ductivityhe willhavehigherprofits,because
he doesnot incurthe innovativeR&Dcosts.
Butif he stopshis outputgrowthat reason-
ablesize,pressureson theinnovatingfirmto
contractarerelaxed,the R&Dbudgetis not
eroded,andthereis a chanceof recoveryfor
the innovatingfirm. (On the other hand, a
largeinnovatorwill stochasticallyextendhis
advantageovera smallimitator.)But,if the
large imitatingfirm continues to grow, it
forces the innovatingfirm to continue to
contract. As the innovatorsR&D budget
contracts, the chances of an innovative
success that will spark recoverydiminish,
and the expectedlead time before the big
imitatorimitatesdiminishesas well.
We think there is a phenomenonhere,
albeit in stylized model form, well worth
pondering.In our modelworld,an imitative
strategymay, if supportedby luck earlyin
the industry's evolution, be a runaway
winner. And certainly imitators will have
good luck at least some of the time. Is it
reallysociallydesirablethattheyshouldpress
theiradvantage?Earlierwe arguedthat the
answermightdependon what a lowerlevel
of innovativeR&Dcostssociety.
In these simulationruns there was little
tendencyforbestpracticeoraveragepractice
in the last periodto be lowerin the aggres-
sive investmentcase than in the restrained
investmentcase.Inpartthisis because,while
in thelattercasetheinnovatorsshareof total
capitaltendedto be less, total industryout-
put and capital tended to be greater,and
total innovatiive R&D spending over the
simulationrun not radicallydifferentthere-
fore in the two cases. In part, the result
pointsto therealsocialadvantageof a struc-
ture dominatedby large imitators:once it
exists, an innovationis rapidlyappliedto a
largefractionof industrycapacity.(Note the
high averageproductivityof imitatorsin the
FH casesof Table3.)
But the central reason is that relatively
sharply diminishingreturns to innovative
R&D are built into this model. A smaller
R&D expendituremeans that the path of
latent productivityis trackedless well, and
that on averagethe differencebetweenin-
dustrybest practiceand latent productivity
is greater.Butevenan occasionalinnovative
R&D hit suffices to keep the averagedis-
tance from being very great. However,the
social costs that occurif imitatorscome to
dominatean industrymightbe significantly
greaterin a regimewherethe opportunities
for today's technicalchange are more in-
fluencedby the industry'sown prior R&D

efforts and less influenced by developments
outside the industry than has been the case
in the runs considered thus far. We now
consider the simulation runs relating to this
variant of our model.
C. CumulativeTechnologicalAdvance
In the simulation runs reported above, the
firm's efforts at innovation were modeled as
taking draws from a population of tech-
niques, the average quality of which was
expanding over time as a result of forces
exogenous to the industry. A firm's prevail-
ing technique had no influence on the distri-
bution of possible outcomes of an innovative
draw. In the simulation runs reported here,
there was no outside augmentation of the set
of possible innovations. And the outcome of
any innovation draw was very much in-
fluenced by the prevailing technique of the
firm making that draw. In particular,techno-
logical advance was cumulative in the sense
defined above.
Tables 4 and 5 display relevant industry
statistics for a set of runs with a similar
structureof parametersettings as in Tables 2
and 3. Many of the same relationships that
held in the earliercases hold as well in these.
The old results about the effect of the char-
acter of innovation and imitation on in-
dustry concentration obtain. The fast in-
novation, hard imitation condition tends to
lead to concentration. Where innovation is
slow and imitation easy, such tendencies were
far less marked. Regarding the competitive
contest between innovators and imitators,
innovators do well when the conditions per-
mit fast technical advance and where imita-
tion is difficult, provided large firms show
restraint in further expanding output. The
imitators do well, and the innovators poorly,
in the opposite parameter settings. When
firms continued to be aggressive in their
output decisions even as they grew large,
profits of both innovators and imitators were
less than were more restrained behavior ob-
tained. But it was especially the innovators
whose fortunes were hurt. Thus the innova-
tor/imitator asymmetry discussed earlier
continues to have force under these different
assumptions regarding the nature of innova-
tion.
What is different about these simulation
runs is that aggressive competitive behavior
has a clear negative effect on both best prac-
tice productivity and average productivity.
Under comparable conditions, each of the
pair of best practice statistics (recall that we
ran each parameter's setting twice) almost
invariably was larger in the run where the
firms restrained their output growth than in
the runs where they did not. As in the earlier
runs, aggressive competitive behavior tends
to generate a structure in which there is at
least one large imitator who is capable of
rather quickly mimicking any new innova-
tion, and who operates with lower costs than
the innovators. As the profitless innovating
firms shrink, so does total industry innova-
tive R&D. In all but two of the comparisons,
industry innovative R&D was less in the
aggressive competition case than in the re-
strained competition case, and, in all but
one, the number of innovation draws smaller.
And, in contrast with the science-based cases,
where such a cutback in industry innovative
R&D and innovation draws had little effect
on the time path of industry best practice
(save to make it more jagged and somewhat
lower), in these runs reduced industry in-
novation shows up in a slower growth of best
practice.
As one would expect, there is less of an
effect on average productivity than on best
practice. Where there is one huge imitator
comprising a large share of industry capital,
average productivity is largely determined by
his productivity, and his productivity stays
close to best practice. Nonetheless, in most
of the cases average productivity was lower
in the aggressivecompetition case than in the
comparable case of more restrained competi-
tion.
The effect of aggressivecompetition on the
price of the industry product was more un-
certain. To some extent lower markups over
costs in the strong competition case offset
higher costs in the case. Nonetheless, and in
striking contradiction to textbook wisdom,

VOL. 72 NO. 1 NELSON AND WINTER: SCHUMPETERIA N TRADEOFFS 129
TABLE4-SIXTEEN-FIRM RUNS; CUMULATIVETECHNICALCHANGE,INVESTMENTRESTRAINT
SE FE SH FH
(1) (2) (1) (2) (1) (2) (1) (2)
3. Marginover Average Cost (%) 8.6 8.6 10.0 8.3 9.0 8.5 10.7 12.6
9. Herfindahl Numbers Equivalent 14.3- 12.6 10.7 9.4 13.1 13.4 7.2 8.1
14. Average Productivity Gap n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a.
15. Total Expenditureon Innov. R&D t97. 190. 210. 182. 206. 188. 235. 208.
17. Price .72 .69 .30 .28 .74 .75 .21 .30
Note: See Table 1.
TABLE5-SIXTEEN-FIRM RUNS; CUMULATIVETECHNICALCHANGE,AGGRESSIVEINVESTMENT
SE FE SH FH
(1) (2) (1) (2) (1) (2) (1) (2)
1. Innovators' Capital Share .23 .29 .26 .40 .32 .35 .38 .48
3. Marginover Average Cost (%) 2.3 2.4 3.0 2.5 4.3 3.3 1.3 2.1
5. Output Share: LargestInnovator .05 .06 .06 .10 .05 .07 .12 .14
14. Average Productivity Gap n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a.
15. TotalExpenditureonInnov.R&D 148. 158. 146. 189. 156. 161. 192. 209.
16. Innovator Draws 80. 84. 83. 86. 64. 88. 99. 115.
17. Price .75 .69 .32 .22 .83 .73 .26 .29
Note: See Table 1.
by and largeend-period industryprice tended
to be higher in the aggressive competition
case than in the restrained competition case.
Furthermore,examination of price trends in
the two contrasting settings suggest that if
the simulation run were longer, price under
aggressive competition would grow progres-
sively higher than what price would be under
more gentlemanly competition. Our 101
period simulation runs end with some in-
novators continuing to exist with nontrivial
capital stocks, but in retreat. As the retreat
continues, industry innovative R&D expendi-
tures should dry up. And ultimately the in-

TABLE6-201 PERIODRUNS; SIXTEENFIRMS,FH
Science Based Cumulative
Aggressive Restrained Aggressive Restrained
I. Innovators' Capital Share .23 .87 .30 .49
2. Excess Return (%Qtr) -.13 2.27 .03 3.68
3. Marginover Average Cost (%) .75 16.24 .13 28.36
4. Marginover Best Practice (%) 12.5 29.1 7.8 31.5
5. Output Share: Largest Innovator .20 .22 .22 .26
6. Output Share: Second Innovator .01 .20 .02 .22
7. Output Share: LargestImitator .73 .08 .55 .24
8. Output Share: Second Imitator .02 .01 .14 .23
9. Herfindahl Numbers Equivalent 1.71 6.25 2.81 4.41
10. Average Productivity 3.48 3.66 2.62 4.77
11. Innovators' Average Productivity 3.54 3.75 2.48 4.80
12. Imitators'Average Productivity 3.47 3.00 2.68 4.74
13. Best Practice Productivity 3.81 4.05 2.81 4.90
14. Average Productivity Gap .16 .15 n.a. n.a.
15. Total Expenditureon Innovative R&D 339.39 484.42 383.00 388.60
16. Innovation Draws 183. 254. 175. 185.
17. Price .04 .05 .06 .04
dustry should settle into something quite
close to a competitive equilibrium with zero
profits and static technology.
Table 6 presents data for simulations we
ran for 201 periods. In the science-based
technology regime the innovators have in-
deed shrunk further in the aggressive invest-
ment behavior case, but best practice pro-
ductivity is not badly affected. On the other
hand, in the cumulative technology regime
period 201 best practice is much lower in the
aggressive competition case than in the case
where competition is more restrained. The
hidden hand has throttled the goose that lays
the golden egg.
IV. SomePolicyandEmpiricalImplications
of theModel
We referred in the introduction to the
unsatisfactory character of the literature on
the Schumpeterian arguments. Undoubtedly,
the difficulties that beset the literature are in
large part a reflection of the complexity and
difficulty of the subject matter. But part of
the problem has been that economists have
not tried to confront those complexities with
a model designed for that task.
The model illustrates some policy conun-
drums that may well be presented by eco-
nomicreality.It leadsus to contemplatethe
possibility that not only may a relatively
concentratedindustryprovidea bettershelter
for R&D than a more fragmentedindustry
structure:productionand technicaladvance
may alsobe moreefficientin sucha setting.
Thus there are "tradeoffs"revealedby the
model, but not necessarilythe same ones
thathavebeenidentifiedpreviously.Further,
the natureof the tradeoffsmaydifferacross
industries.In an industrywheretechnology
is "science based," greater concentration
means highermarkups;it brings the com-
pensatingadvantagesof a smallergap be-
tween averageand best practiceand more
efficient R&D (in the sense that a given
productivitylevel is trackedmore cheaply),
but does not buy a fasterrate of growthof
productivity.However,in an industrywhose
technologyis "cumulative,"a moresheltered
competitiveenvironment,and the associated
higher markups,does lead to more rapid
productivitygrowth.
Theresultsthatshowa tendencyfor firms
that do innovativeR&D to lose out in a
competitive struggle with skillful and ag-
gressiveimitatorsare particularlyprovoca-
tive, and illustratea possibility not much
discussedin theeconomicliterature.Nor has
there been much discussiondifferentiating

VOL. 72 NO. 1 NELSON AND WINTER: SCHUMPETERIAN TRADEOFFS 131
the kinds of regimes for technical progress
under which the social costs are slight (sci-
ence-based industries) or heavy (cumulative
technology industries) of having firms that
invest in innovative R&D driven to the wall
or out of business.
There are some fascinating predicted em-
pirical relationships that flow from our
model, but, as with the policy conundrums,
they are somewhat more subtle than those
many economists seem to think reside in
Schumpeterian competition. Most of the re-
ported empirical tests have rested on the
supposition that the Schumpeterian argu-
ments imply that large firms tend to spend
relatively more on R&D than small firms.
Under the stylized conditions of our experi-
ments, such a correlation can arise only as a
result of selection, that is, of differential
growth of innovators relative to imitators. In
experimental settings where innovative R&D
is profitable, the firms that spend on innova-
tive R&D (and hence have a higher ratio of
total R&D to capital) do tend to grow rela-
tive to the imitators, but in such a setting the
small firms tend to be eliminated. Where
innovative R&D is not profitable, but where
market structure permits it to survive, the
R&D intensive firms tend to be small.
Our model does point to a prediction that
industrieswith rapid technical progress ought
to be markedby high average R&D intensity
and, as the industry matures, a more con-
centrated industry structure than industries
where technical progress is slower. And, in-
terestingly enough, various studies attempt-
ing to explain cross-industry differences in
productivity growth rates have identified
roughly the above relationship. The model
also suggests that concentration is likely to
increase over time in a technologically pro-
gressive industry, another relationship that
seems to hold empirically."1
It should be clear that Schumpeter's ap-
praisal of progressive capitalism continues to
present an imposing challenge to theorists,
econometricians, policy analysts, and schol-
ars of technological change. We hope that
this paper will prompt others to view the
Schumpeterian arguments in a new light.
REFERENCES
Dasgupta,Parthaand Stiglitz, Joseph, (1980a)
"Uncertainty, Industrial Structure,and the
Speed of R&D," Bell Journal of Econom-
ics, Spring 1980, 11, 1-28.
and _ , (1980b) "Industrial
Structure and the Nature of Innovative
Activity," Economic Journal, June 1980,
90, 266-93.
Flaherty,M. Therese,"Industry Structure and
Cost Reducing Innovation," Economet-
rica, July 1980, 48, 1187-211.
Futia, Carl',"Schumpeterian Competition,"
QuarterlyJournal of Economics,June 1980,
94, 675-95.
Horner,Stephen,"Stochastic Models of Tech-
nology Diffusion," unpublished doctoral
dissertation, Ann Arbor: The University
of Michigan, 1977.
Iwai,Katsuhito,"Towards Schumpeterian Dy-
namics," mimeo., Yale University, 1980.
Kamien,MortonandSchwartz,Nancy,"Market
Structureand Innovation," Journal of Eco-
nomic Literature, March 1975, 13, 1-37.
Levin,Richard,"Toward an Empirical Model
of SchumpeterianCompetition," Bell Jour-
nal of Economics, forthcoming.
Loury,Glenn,"Market Structure and Innova-
tion," Quarterly Journal of Economics,
August 1979, 93, 395-410.
Nelson, Richard,Peck, MertonJ. andKalachek,
Edward,Technology,Economic Growth,and
Public Policy, Washington: Brookings In-
stitution, 1967.
______ and Winter,Sidney,"DynamicCom-
petition and Technical Progress," in Bella
Balassa and Richard Nelson, eds., Eco-
nomic Progress, Private Values and Public
Policy: Essays in Honor of WilliamFellner,
Amsterdam: North-Holland, 1977.
______ and _ , "Forces Generatingand
Liniiting Concentration Under Schum-
peterian Competition,"Bell Journal of Eco-
nomics, Autumn 1978, 9, 524-48.
______ and _ , An Evolutionary Theory
of Economic Change, Cambridge: Harvard
University Press, forthcoming 1982.
_ ,____ _ , and Schuette, Herbert,
"Technical Change in an Evolutionary
Model," QuarterlyJournal of Economics,
February 1976, 90, 90-118.'0See, for example, Phillips and our 1978 article.

Phillips, Almarin, Technology and Market
Structure:A Study of theAircraftIndustry,
Lexington, Mass.: D. C. Heath, 1971.
Scherer, F. M., Industrial Market Structure
and Economic Performance, 2d ed.,
Chicago: Rand McNally, 1980.
Schumpeter,Joseph,"The Theoryof Economic
Development, Cambridge: Harvard Uni-
versitypress,1934.
, Capitalism, Socialism, and Democ-
racy,3d ed., New York: Harper& Row,
1950.
Teece,David,"Economiesof Scope and the
Scope of the Enterprise," Journal of Eco-
nomic Behavior and Organization, Septem-
ber 1980, 1, 223-47.

Nelson winter1982

Recomendados

Recomendados

Mais conteúdo relacionado

Destaque

Destaque (20)

Semelhante a Nelson winter1982

Semelhante a Nelson winter1982 (20)

Mais de economicgrowthcucea

Mais de economicgrowthcucea (18)

Nelson winter1982