NBER WORKING PAPERS SERIES

ECONOMIC INTEGRATION
AND ENDOGENOUS GROWTH

Luis A. Rivera
Paul M. Romer

Working Paper No. 3528

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue
Cambridge, MA 02138
December 1990

Conversations with Robert Barro, Gene Grossman, Elhanan Helpman

and Danyang Xie about their work on related issues have been very
helpful, as were comments on earlier versions of this paper by
Ray Riezman and Robert Staiger. The work of the second author was
supported by NSF grant #SES88-22052, by a Sloan Foundation
Fellowship, and by the Center for Advanced Studies in the
Behavioral Sciences, which received support from NSF grant
#BNS87-00864. This paper is part of NBER's research programs in
Growth, in Economic Fluctuations, and in International Studies.
Any opinions expressed are those of the authors and not those of
the National Bureau of Economic Research.
 NBER Working Paper #3528
December 1990

ECONOMIC INTEGRATION AND ENDOGENOUS GROWTH

ABSTRACT
In a world with two similar, developed economies, economic
integration can cause a permanent increase in the worldwide rate

of growth. Starting from a position of isolation, closer

integration can be achieved by increasing trade in goods or by
increasing flows of ideas. We consider two models with different
specifications of the research and development sector that is the
source of growth. Either form of integration can increase the
long—run rate of growth if it encourages the worldwide
exploitation of increasing returns to scale in the research and

development sector.

Luis A. Rivera-Batiz Paul M. Romer

IRPS NBER
D—0l9 204 Junipero Serra Blvd.
University of California, Stanford, CA 94305—8091
San Diego
La Jolla, CA 92093
 I. Introduction

Many economists believe that increased economic integration between the developed
economies of the world has tended to increase the long-run rate of economic growth. If
they were asked to make an intuitive prediction, they would suggest that prospects for
growth would be permanently diminished if a barrier were erected that impeded the flow of
all goods, ideas, and people between Asia, Europe, and North America. Yet it would be
difficult for any of us to offer a rigorous model that has been (or even could be) calibrated
to data and that could justify this belief.
We know what some of the basic elements of such a growth model would be.
Historical analysis (e.g. Rosenberg [1980]) shows that the creation and transmission of
ideas have been extremely important in the development of modern standards of living.
Theoretical arguments dating from Adam Smith's analysis of the pin factory have
emphasized the potential importance of fixed costs and the extent of the market. There is
a long tradition in trade theory of using models with Marshaflian external effects to
approach questions about increasing returns. More recently, static models with fixed costs
and international specialization have been proposed that come closer to Smith's description
of the sources of the gains from trade. (Dixit and Norman [1980], Ethier [1982], Kruginan
[1979,1981], Lancaster[1980]). There are also dynamic models with fixed costs and
differentiated products in which output grows toward a fixed steady state level (Grossman

and Helpman [1989a]).

Recent models of endogenous growth have used these ideas to study the effects that
trade can have on the long-run rate of growth. (See for example the theoretical papers by

Dinopoulos, Oehmke, and Segerstrom [1990], Feenstra [1990], Grossman and Helpman
[1989b,1989c,1989d,1989e,1990], Krugman [1990, Chapter 11], Lucas [1988], Romer [1990],
 2

Segerstrom, Anat, and Dinopoulos [1990], and Young [1990]. Backus, Kehoe, and Kehoe
[1990] present both theoretical models and cross country empirical evidence that bears on
their models.) These models permit a distinction between a one shot gain (i.e. a level
effect) and a permanent change in the growth rate (i.e. a growth effect) that is extremely
important in making an order of magnitude estimate of the benefits of economic
integration. Conventional attempts to quantify the effects of integration using the
neoclassical growth model often suggest that the gains from integration are small. if these
estimates were calculated in the context of an endogenous growth model, integration might
be found to be much more important.
The papers written so far have already demonstrated, however, that the growth
effects of trade restrictions are very complicated in the most general case. Gene Grossman
and Elhanan Helpman [1989b,1989c,1989e,1990] have been particularly explicit about the
fact that no u.niversally applicable conclusions can be drawn. There are some models in
which trade restrictions can slow down the worldwide rate of growth. There are others in
which they can speed up the worldwide rate of growth.
To provide some intuition for the conjecture described in the beginning, that trade
between the advanced countries does foster growth, we narrow the focus in this paper. We
do not consider the general case of trade between countries with different endowments and
technologies. Instead, we focus on the pure scale effects of integration. To set aside the
other "comparative advantage" effects that trade induces in multisector trade models, we
consider integration only between countries or regions that are similar. Therefore, we do
not address the kinds of questions that are relevant for modeling the effects that trade
between a poor LDC and a developed country can have on the worldwide rate of growth.
In the early stages of our analysis of integration and growth, it became clear that
the theoretical treatment of ideas has a decisive effect on the conclusions one draws. In
many of the existing models, flows of ideas cannot be separated from flows of goods. In
others, flows of ideas are exogenously limited by national boundaries regardless of the trade
 3

regime. In either of these cases, economic integration can only refer to flows of goods along
cargo networks. We consider a broader notion of integration, one that assigns an effect to
flows of ideas along communication networks.
Flows of ideas deserve attention comparable to that devoted to flows of goods, for
public policy can influence international communications and information flows to the
same extent that it influences goods flows. Governments often subsidize language training
and study abroad. Tax policies directly affect the incentive to station company employees
in foreign nations. Immigration and visa policies directly limit the movement of people.
Telecommunications networks are either run by government agencies or controlled by
regulators. Some governments restrict direct foreign investment, which presumably is
important in the international transmission of ideas. Others have made the acquisition of
commercial and technical information a high priority task for their intelligence agencies.
Although these are the only ones we consider, it should be clear that flows of goods
and flows of ideas are not the only elements in economic integration. Under some
assumptions about nominal variables and the operation of financial markets, economic
integration will also depend on monetary and institutional arrangements. The growth
models we consider are too simple to consider these effects. It should also be clear that
economic integration is not synonymous with political integration. Firms in Windsor,
Ontario may be more closely integrated into markets in the United States than they are to
markets in the neighboring province of Quebec. Moreover, the notion of full economic
integration does not entail the abolition of citizenship distinctions that have taken place in
Germany's reunification.
The structure of the paper is as follows. Section II lays out the basic features of the
production structure on which all arguments rely. It describes preferences, endowments,
and the nature of equilibrium under the two specifications of R&D. Section III describes
the equilibrium for both models in the closed economy and complete integration cases, and
illustrates the scale effects that are present. Section IV presents the three main thought
 4

experiments concerning partial integration. Sections V and VI describe the general lessons
that can be learned about the relation between the scale of the market and growth and
discuss limitations of the models, extensions, and the relation to other models of

endogenous growth.

IL Specification of the Models

£ Functional forms decentralization i manufacturing sector

The specification of the production technology for the manufacturing sector is taken
from Romer [1990). Manufacturing output is a function of human capital H, labor L, and
a set r(i) of capital goods indexed by the variable i. To avoid complications arising from
integer constraints, the index I is modeled as a continuous variable. Technological
progress is represented by the invention of new types of capital goods.
There are two types of manufacturing activities: production of consumption goods
and production of the physical units of the types of capital goods that have already been
invented. A third activity, research and development (R&D), creates designs for new types
of capital goods. This activity is discussed in next section.
Both manufacturing activities use the same production function. Let z( i) denote
the stock of capital of type i that is used in production and let A be the index of the
most recently invented good. By the definition of A, x(i) = 0 for afl i.A. Output Y
is assumed to take the form

(1) H,L,z(.)) = HaLfif z(i)' di.

 5

Since the production function for manufacturing consumption goods is the same as
that for manufacturing units of any type of existing capital, the relative prices of
consumption goods and all types of existing capital goods are fixed by the technology. For
simplicity, we choose units so that all of these relative prices are 1. Fixed prices imply

that the aggregate capital stock K = f

0
A
z( i) di is well defined, as is aggregate output Y.

In this specification, one unit of any capital good can be produced if one unit of
consumption goods is foregone. This does not mean that consumption goods are directly
converted into capital goods. Rather, the inputs needed to produce one unit of
consumption are shifted from the production of consumption goods into the production of a
capital good. Since inputs are used in the same proportions, it is easy to infer the
allocation of inputs between the different production activities from the level of output of
those activities. Because all of the outputs here have the same production function, the
consumption sector and all of the sectors producing the different capital goods can be
collapsed into a single sector. We can therefore represent total manufacturing output as a
function of the total stock of inputs used in the combined manufacturing sectors and can

describe the division of inputs between sectors by the constraint Y = C+K. For one of the
models of R&D described in the next section, we can use this same observation to combine
the research sector and the aggregate manufacturing sector into a single sector describing
all output in the economy. In the other model, the R&D and manufacturing sectors must

be kept separate.
There are many equivalent institutional structures that can support a decentralized
equilibrium in manufacturing. For instance, the holder of a patent on good j could
become a manufacturer, producing and selling good j. Alternatively, the patent holder
could license the design to other manufactuier for a fee. Formally, it is useful to separate
the manufacturing decision from the monopoly pricing decision of the patent holder, so we
assume that patent holders contract out manufacturing to separate firms. It is also easier
 6

to assume that the patent holder collects rent on its capital goods rather than selling them.
For analytical convenience, we therefore describe the institutional arrangements in the
following, slightly artificial way. First, there are many firms that rent capital goods z( 1)
from the patent holders, hire unskilled labor L, and employ skilled human capital H to
produce manufactured goods. Each of these firms can produce consumption goods for sale
to consumers. It can also produce one of the capital goods on contract for the holder of the
patent. All of the manufacturing firms have the production function given in equation (1),
which is homogeneous of degree one. They are price takers and earn zero profit.
Manufacturing output is taken as the numeraire.
The firm that holds the patent on good j bids out the production of the actual
capital good to a specific manufacturer. It purchases physical units of the good for the
competitive price, by normalization equal to 1. The patent holder then rents out the units
to all manufacturing firms at the profit maximizing monopoly rental rate. It can do this
because patent law prohibits any firm from manufacturing a capital good without the
consent of the patent holder. The patent is a tradeable asset with a price A that is
equal to the present discounted value of the stream of monopoly rent minus the cost of the
machines. It is easy to verify that this set of institutional arrangements is equivalent to
other arrangements. For example, the equivalent licensing fee for each unit of capital sold
by a licensee is the present value of the stream of monopoly rent on one machine minus the
unit cost of manufacturing it.

B. Functional forms decentralization in R&D

We consider two specifications of the technology for R&D that permit easy analytic
solutions. Each specification captures different features of the world, and neither alone
gives a complete description of R&D. We use both of them because they help us isolate the
 7

exact sense in which economic integration can influence long-run growth. As the examples
in the next section shows, it would be easy to come to misleading conclusions about
integration and growth if one generalized from a single example.
The first specification of the technology for producing designs for new capital goods
assu.mes that human capital and knowledge are the only inputs that influence the output of

designs:

(2) A = 5HA.

Here H denotes the stock of human capital used in research. The stock of existing designs
A is a measu.re of general scientific and engineering knowledge as well as practical know-
how that accumulated as previous design problems were solved. (See Romer [19901 for
additional discussion of this specification.) New designs build on this knowledge, so we
refer to this type of R&D process as the knowledge—driven specification of R&D. This
specification imposes a sharp factor intensity difference between R&D and manufacturing.
Neither unskilled labor nor physical capital have any value in R&D. Because of this
difference, the resulting model must be analyzed using a two sector framework.
A useful polar case is a technology for R&D that uses the same inputs as the
manufacturing technology, in the same proportions. If H, L, and denote inputs used
in R&D and B denotes a constant scale factor, output of designs can be written as

(3) A=
BROLPf Z(I)lafl di.

This specification says that human capital, unskilled labor, and capital goods (such as
personal computers or oscilloscopes) are productive in research. But in contrast to the
previous specification, knowledge per se has no productive value. Access to the designs for
 8

all previous goods, and familiarity with the ideas and know-how that they represent, does
not aid the creation of new designs. We refer to this as the lab equipment specification of
R&D.
As noted above, the growth model with a knowledge—driven specification for R&D
has an unavoidable two sector structure. The production possibility frontier in the space of
designs and manufactured goods takes on the usual curved shape. In the lab equipment
model, the production functions of the goods and R&D sectors are the same, so the
production possibility frontier is a straight line, if the output of goods is reduced by one
n.nit and the inputs released are transferred to the R&D sector, they yield B patents.
Thus the price A of a patent in terms of goods is determined on the technology side,
= . Since capital goods and consumption goods have the same production technology,
we integrated them into a single manufacturing sector in the last section. In the lab
equipment model we can go further, and aggregate manufacturing and research into a
single sector. Let H, L, and z(i) denote the entire stock of inputs available in the

economy at date t. Then we can express the value of total output C + K + A/B in terms
of the total stock of inputs,

(4) C+K+A/B = II°Lflf lafl di.

The model's symmetry implies that z(i) = z(j) for all i and j less than A. We can
therefore substitute K/A = z( i) in equation (4) to obtain a reduced form expression for
total output in terms of H, K, L, and A:

(5) C+K+A/B = Ji°LA(K/A)1"

=
 9

The knowledge-driven and lab equipment specifications of the R&D sector lead to
different assumptions about how equilibrium in the R&D sector is decentralized. In the
knowledge-driven model, output of designs is homogeneous of degree 2. By Euler's
theorem, it is not possible for both of the inputs A and H to be paid their marginal
product. We make the assumption that A receives no compensation. Holders of patents
on previous designs have no technological or legal means of preventing designers of new
goods from using the ideas implicit in the existing designs. The stock of A that can be
put to use, with no compensation, by any individual researcher is therefore the entire stock
of knowledge about previous designs, provided that there exists a communication network
that makes this information available. The equilibrium is one with knowledge spillovers or
external effects in the R&D sector (but not in the manufacturing sector.) In this case, we
can describe research as if it were done by independent researchers who use their human

capital to produce designs, which they subsequently sell.

l.a the lab equipment model, output of designs is the same, homogeneous-of-degree-
one production function as in the manufacturing sector. As is the case for the
manufacturing sector, the equilibrium is one in which patents convey market power but in
which there are no other entry restrictions. There are no external effects and no knowledge
spillovers. There is free entry into both R&D and manufacturing. The only restriction is
that no one can manufacture capital of type i without the consent of the holder of the
patent on good i. In this case, we conceive of R&D as being undertaken by separate firms
that hire inputs, produce patentable designs, and sell them for a price

III. Balanced Growth and Integration

The description of the technology given so far represents output as a function of the
inputs H, L, K, and A, and specifies the evolution equations for K and A. To facilitate
 10

the simple balanced growth analysis that we undertake, the stocks of L and H are each
taken as given. Increases in either L or H could be accommodated if we undertook the
more complicated task of solving a nonlinear system of differential equations with growth
rates that vary over time.
The calculation of a balanced growth equilibrium for each of the two specifications
of the R&D technology can be summarized in terms of two linear relations between the
rate of growth and the interest rate that hold along a balanced growth path. One relation
comes from the conditions of equilibrium in production and the other from preferences.
As shown in the Appendix and as illustrated in Figure I, the interest rate implied by
equilibrium in the production sector is decreasing in the rate of growth of output for the
knowledge-driven model:

'6'J —
rtechnology

The term in the denominator depends only on the production function parameters,
A = a(o+fl)1(1-a-fi)1.

The corresponding expression for the interest rate from the lab equipment model is
shown in the Appendix to be a function of the production parameters and the stock of H
and L. It does not, however, depend on the rate of growth:

(7) rtechnology = ri/ZIP,

where r is defined by r = B0(a+fi)fi(1_afl)2fl.

In the knowledge-driven specification, the negative relation between the interest
rate and the growth rate arises because an increase in the interest rate reduces the demand
for capital goods. The calculations in the Appendix show that an increase in the interest
rate reduces the number of units of each capital good that are rented, and thereby reduces
 1].

interest rate

r preferences

r
p

growth rate

Figure I: Determination of r and g in the knowledge-driven specification

the value of a patent. According to the curved production possibility frontier between
designs and manufactured goods, the reduction in the price of the patented design causes a
shift in human capital out of the production of new designs and into the production of
manufactured goods. This shift slows down the creation of technology and thereby slows
growth. In the lab equipment model, only a single value of the interest rate is consistent
with production of both goods and designs. The relative price of patents and final goods is
fixed, so the interest rate is technologically determined.
 12

interest rate

r preferences

growth rate

Figure U: Determination of r and g in the lab equipment specification

It remains to specify the preferences that provide the other balanced growth relation
between the interest rate and the rate of growth. The simplest formulation to work with is
Ramsey preferences with constant elasticity utility,

=' 0E[O,w).

Under balanced growth, the rate of growth of consumption must be equal to the rate of
 13

growth of output. Thus, for any fixed rate of growth g=C/C, we can calculate the implied
interest rate from the consumer's first order conditions for intertemporal optimization:

(8) Tpreferences = p+g.

These preferences yield a positive relation between the interest rate and the growth rate
because when consumption is growing more rapidly, current consumption is more valuable
compared with future consumption, so the marginal rate of substitution between present
and future consumption is higher. Consumers would therefore be willing to borrow at
higher interest rates.
There is a parameter restriction that is necessary to ensure that the growth rate is
not larger than the interest rate. If it is, present values will not be finite and the integral
that defines utility will diverge. In terms of Figures I and LI, the restriction is that the
intersection of the two curves must lie above the 45 degree line. This will always be true if
i is greater than or equal to 1, since in this case, the rpreferences curve always lies above

the 45 degree line. If is less than 1, the Ttechflologr curve must not lie too far up and
to the right.
Because the rate of growth under each specification is determined by the
intersection of two straight lines, it can be calculated directly from the relation between r
and g determined on the preference side, equation (8), and the relation between r and g
determined by the technology, either equation (6) or (7). The balanced rate of growth for
a closed economy under the knowledge-driven model of the research sector is

(9' — I5H-Ap
¼)
 14

The balanced rate of growth for the lab equipment model is

(10) grI/1L15P.

Both of these models have a dependence on scale that is crucial to the analysis of
the effects of trade. To see this, consider two economies that have identical endowments of
H and L. In the long run, these economies will have the same stocks of accumulated
inputs as well, so that scale effects offer the only lasting source of gains from trade and
economic integration.
Suppose that the two economies are physically contiguous, yet are totally isolated
from each other by an impenetrable barrier that impedes the flow of goods, people, and
ideas. If these economies evolve under isolation, the balanced rate of growth in each is
characterized by Figures I and II and calculated in equations (9) and (10). Now suppose
that the barrier is removed, so that the economies are completely integrated into a single
economy. The change from two economies with endowments H and L to one economy
with stocks 2ff and 2L causes an upward shift in the rtechaolo curve in both figures.
Both the rate of growth and the interest rate increase after complete economic integration
takes place, regardless of the specification of the technology for R&D. In both models
(even the knowledge-driven model with no knowledge spillovers), the rate of growth is too
low compared to the rate that would be selected by a social planner.' As a result, one

'For the knowledge—driven model, this is shown in Romer [1990]. For an early
version of the lab equipment model, this is shown in Romer [1987]. See Barro and Sala i
Martin [1990] for a discussion of the optimality of the no intervention equilibrium and of
tax and subsidy policies that can achieve the socially optimal balanced rate of growth in a

variety of endogenous growth models.

 15

would expect integration to be welfare improving. A full welfare analysis, however, would

require explicit consideration of the dynamics along the transition path.

With this discussion as background, the examples in next section are designed to
address three questions. First, can free trade in goods between countries induce the same
increase in the balanced growth rate as complete integration into a single economy? if not,
can the free movement of goods, combined with the free movement of ideas, reproduce the
rate of growth under full integration? And finally, what is the underlying explanation for
the dependence of the growth rate on the extent of the market?

IV. Trade in goods and flows of ideas

In this section we conduct a series of thought experiments about partialintegration.

In the first two experiments, we focus on the knowledge-driven specification for R&D
because it permits a sharp distinction between flows of goods and flows of ideas. In the
third, we consider the lab equipment specification in which ideas have no direct effect on
production.

In the analysis of the knowledge-driven specification, we start with two identical,

completely isolated economies that are growing at the balanced growth rate. We first
allow for trade in goods, but continue to restrict the flow of ideas. To emphasize the
distinction between goods and ideas, we assume that trade in goods does not induce any
transmission of ideas. For example, we assume that is it impossible to reverse engineer an
imported good to learn the secrets of its design. Under these assumptions, we show that
trade in goods has no effect on the long-run rate of growth. Then in the second
experiment, we calculate the additional effect of opening communications networks and
permitting flows of ideas. We show that allowing flows of ideas results in a permanently
higher growth rate.
 16

lii the third experiment, we consider the effects of opening trade in goods under the
lab equipment specification. In this case, trade in goods alone causes the same permanent
increase in the rate of growth as complete integration. Since ideas per ie have no effect on
production, the creation of communications networks has no additional effect.

£ Flows f goods f j j knowledge-driven model

In all of the experiments considered here, the form of trade between the two
countries is very simple. By symmetry, there are no opportunities for intertemporal trade
along a balanced growth path, hence no international lending. Because there is a only a
single final consumption good, the only trades that take place are exchanges of capital
goods produced in one country for capital goods produced in the other.
With the knowledge-driven model of research, it is straightforward to show that
opening trade in goods has no permanent effect on the rate of growth. In balanced growth,

the rate of growth of output is equal to the rate of growth of A, A/A =

£HA, which is
determined by the split of human capital H=HHA between the manufacturing sector
and the research sector. Opening trade in goods has two offsetting effects on wages for
human capital in these two sectors. Before trade is opened, the number of different types
of machines that are used in the manufacturing sector must equal the number that have
been designed and produced domestically. Along the new balanced growth path after trade
is opened, the number of types of machines used in each country approaches twice the
number that have been produced and designed domestically. In their pursuit of monopoly
rents, researchers in the two countries will specialize in the production of different types of
designs and avoid redundancy, so the worldwide stock of designs will ultimately be twice as
large as the stock that has been produced in either country.
 17

With trade in the specialized capital goods, domestic manufacturers can take
advantage of foreign designs and vice versa. Ultimately, the level at which each durable
is used in each country will return to the level that obtained under isolation. From
equation (1) it follows that the increase in A doubles the marginal product of human
capital in the manufacturing sector, increasing it from = aR°Lz1 °A to
= all° L 2A).
For the research sector, opening of trade implies that the market for any newly
designed good is twice as large as it was in the absence of trade. This doubles the price of
the patents and raises the return to investing human capital in research from PASA to
2PA5A. Since the return to human capital doubles in both of the competing sectors, free
trade in goods does not affect the split of human capital between manufacturing and
research. Hence, it does not change the balanced rate of economic growth or the interest
rate. In terms of Figure I, opening trade in goods does not change the position of either
the rpref&en locus or the locus.
rteco1o
This result does not imply that free trade in goods has no effect on output or
welfare. Consider, for example, the extreme case in which two isolated economies start
*
from completely nonintersecting sets of capital goods A and A that have the same

measure. Before trade in goods, the home country will use capital at the level for A
types of capital goods and the foreign country will use capital at the same level for A
different types of capital goods. If existing capital is freely mobile, each country will
immediately exchange half of its capital stock for half of the capital stock of the other
country when trade in capital goods is allowed. Each will then be using capital at the level
on a set of capital goods of measure A÷A. (Over time, the level of usage will climb
back to as capital accumulation takes place because the level of z is determined by r
and g, and on the new balanced growth path these are the same as before.) From the form

of production in manufacturing given in equation (1), it follows that immediately after

trade is opened, output in each country jumps by a factor of 210. This is analogous
 18

to the kind of level effect one encounters in the neoclassical model and in static models of

trade with differentiated inputs in production, (e.g. Ethier [1982J.) In the specific model
outlined here, free trade in goods can affect the level of output and can therefore affect
welfare, but it does not affect long-run growth rates.
if the two different economies start from a position with exactly overlapping sets of
goods prior to the opening of trade, the timing of the effect on output is different, but the
ultimate effect is the same. The level of out put at future dates will differ from what it
would have been without trade in goods and will generally be higher. But once the
transitory effects have died out, the underlying growth rate will be same as it was prior to
the opening of trade in the capital goods.

Flows f information j knowledge—driven model QI research

This second example shows that greater flows of ideas can permanently increase the
rate of growth in the knowledge—driven model of research. Once we allow for flows of
information, we must make some assumption about international protection of intellectual
property rights. In each country, we have assumed that patents protect any designs
produced domestically. Once ideas and designs created abroad become available, the
government could try to expropriate the monopoly rents that would accrue to the
foreigners by refusing to uphold their patents. To simplify the discussion here, we assume
that neither government engages in this practice. A patent in one country is fully
respected in the other. (For a discussion of incomplete protection of intellectual property
rights, see Rivera—Batiz and Romer [1990].)
Consider the two identical economies with the knowledge—driven specification of the

research sector described in the first experiment. Trade in goods has already been allowed,
and this creates the incentive for researchers to specialize in different designs. Over time
 19

the sets of designs that are in use in the two countries will be almost entirely distinct, so
the worldwide stock of knowledge approaches twice the stock of designs in either country.
In the absence of communications links, this means that researchers in each country will
ultimately be using only one half of the worldwide stock of knowledge. In the domestic

cou.ntry, the rate of growth of A is given by A =

SHAA. In the foreign country, it is

given by = 6HA*.
Now suppose that flows of ideas between the two countries are permitted. Research
in each country now depends on the total worldwide stock of ideas as contained in the
*
union of A and A . If the ideas in each country are completely nonintersecting, the
effective stock of knowledge that could be used in research after communication opens

would be twice as large as it was before: A = A= SHA(A+A*) =

2SHAA. Even if the
allocation of H = between manufacturing and research did not change, the rate of
HrHA
growth of A would double. But the increase in the set of ideas available for use in
research increases the productivity of human capital in research and has no effect on its
productivity in manufacturing. This change in relative productivity induces a shift of
human capital out of manufacturing and into research. For two reasons, communication of
ideas speeds up growth.
Increasing the flow of ideas has the effect of doubling the productivity of research in
each country. Compared to the closed economy model, the formal effect is the same as a
doubling of the research productivity parameter 5. This would shift the
T'technology
curve in Figure I upward, and lead to a higher equilibrium growth rate and interest rate.
The algebraic solution for the balanced growth rate of A (and therefore also of Y, C, and
K) can be determined by replacing 5 with 25 in equation (9) to obtain
g = (25H—Ap)/(cA+1).

Doubling the value of the productivity parameter 5 has exactly the same effect on
the rate of growth of output and designs as a doubling of H. And according to the
 20

discussion in section II, doubling H has the same effect on growth as complete integration
of the two economies into a single economy. Flows of both ideas and goods together have
the same effect on the growth rate as does complete integration. Complete integration
would permit permanent migration as well, but since ideas and goods are already mobile
and because the ratio of H to L was assumed to be the same in the two countries,
migration is not necessary to achieve productive efficiency. For symmetric economies,
allowing both trade in goods and free flows of ideas is enough to reproduce the resource
allocation under complete integration.
So far, we have considered the additional effect that free flows of information would
have if free trade in goods were already permitted. It is useful to consider the alternative
case in which flows of information are permitted but flows of goods are not. In this case,
the results hinge on the degree of overlap between the set of ideas that are produced in
each country.
In the absence of trade in goods, there would be no incentive for researchers in
different countries to specialize in different designs either before or after flows of
information are permitted. Moreover, after flows of information are opened, there would
be a positive incentive for researchers in one country to copy designs from the other, and
little offsetting incentives to enforce property rights. If the firm that owns the patent on
good j is not permitted to sell the good in a foreign country, it has no economic stake in

the decision by a foreign firm to copy good j and sell it in the foreign market. (The
domestic firm would of course have both the incentive and the legal power to stop exports
of the copies from the foreign country.) In the extreme case in which identical knowledge
is created in each country, opening flows of information has no effect at all on production.
Alternatively, one could imagine that discovery is a random process with a high
variance so that truly independent discoveries would take place in the different isolated
countries. In this case, permitting the international transmission of ideas would speedup
worldwide growth rates to some extent, even in the absence of trade in goods. With free
 21

communication, each researcher would be working with a larger stock of ideas than would
otherwise have been the case. For example, when the first overland routes to China were
opened in the Middle Ages, transportation of goods was so expensive that the economic
effects of trade in goods was small. But the economic consequences of the ideas that
travelers brought back (e.g. the principle behind the magnetic compass and the formula for
gunpowder) were large.

C. Flows of goods in the lab equipment model of research

The two previous examples show that there is sometimes a separation between
growth effects and level effects. In the first experiment, opening trade in goods had level
effects but no growth effects. In the second experiment, opening flows of ideas had both a
growth effect and a level effect. (Manufacturing output goes down when H shifts into
research, and research output goes up.)
From the first two examples it is tempting to conclude that flows of goods will
generally have level effects of the type that are familiar from neoclassical analysis and that
it is only flows of ideas that have growth effects. The third example considered shows that
this conclusion is wrong. The lab equipment model is constructed so that ideas per s e
have no effect on production. Hence, permitting international flows of ideas can have no
economic effect. Yet we know from the discussion in section UI that complete integration
causes a permanent increase in the rate of growth. The experiment considered in this
example shows that trade in goods is all that is needed to achieve this result.
Recall that when trade in goods is permitted in the knowledge-driven model, this
increases the profits that the holder of each patent can extract because it increases the
market for the good. By itself, this increase in the return to producing designs would tend
to increase the production of designs, but in the knowledge-driven specification, this effect
 22

is exactly offset by the increase in the marginal productivity of human capital in

manufacturing.
In the lab equipment specification, opening trade in goods would cause the same
kind of increase in the profit earned at each date by the holder of a patent if the interest
rate remained constant. But as was noted in section ILB, the price of the patent
= 1/B is determined by the technology. The only way that the larger market can be
reconciled with a fixed price for the patent is if the interest rate increases. A higher
interest rate reduces the demand for capital goods, thereby lowering the profit earned by
the monopolist at each date. The calculation in the appendix shows that the required
increase in the interest rate is by a factor of 2°. When two identical economies are
integrated and 2H is substituted for H and 2L is substituted for L in equation (7), the
same increase in r obtains. In each case, the higher interest rate leads to higher savings.
From Figure LI or from equation (10) it follows that this increase in the interest rate leads
to the same faster rate of growth as complete integration.

V. Scale Effects and Growth

In the last example we noted one incorrect conjecture about why tighter economic
integration leads to faster growth. From the knowledge-driven model one might conclude
that flows of ideas are crucial to the finding that economic integration can speed up
growth. But the lab equipment model shows that closer integration can speed up growth
even in a model in which flows of ideas have no effect on production. A related conjecture
is that knowledge spillovers are fundamental and that increasing the extent of the
spillovers is how integration speeds up growth. The lab equipment model shows that this
too is incorrect, for it has no knowledge spillovers.
 23

Finally, one might conclude that it is the increasing return.s to scale in the

production fu.nction for designs, A =

5HAA, that causes integration to have a growth
effect in the knowledge-driven model. This conjecture seems to us to come dosest to the
mark, but it needs to be interpreted carefully. To see why, recall that the production

function for designs in the lab equipment model, A =

BBLPf Z()1afi di, exhibits
constant returns to scale as a function of H, L, and the capital goods z(i). There is,
nonetheless, a form of increasing returns that is present in this model. It comes from the
fixed cost that must be incurred to design a new good. With integration, this fixed cost
need only be incurred once. Under isolation, it must be incurred twice, once in each
country.
To bring out the underlying form of increasing returns, recall from equation (5) that

we can substitute = K/A into the expression for A and write it as a function of H, L,
K, and A that is homogeneous of degree l+a+fi: A = B°LK Interpreted as a
statement about this kind of reduced form expression, it is correct to say that both models
exhibit increase returns to scale in the production of new designs as a function the stocks of
basic inputs. Consequently, operating two research sectors in isolation is not as efficient as
operating a single integrated research sector. To operate an integrated research sector in
the knowledge-driven model, two things are required. First, one must avoid redundant
effort, that is, devoting resources in one economy to rediscovering a design that already
exists in the other. Trade in goods provides the incentive to avoid redundancy. Second,
one must make sure that ideas discovered in one country are available for use in research in
both countries. Flows of ideas along communications networks serve this function.
In the lab equipment model, trade in goods once again provides the incentive to
avoid redundant effort. Beyond this, all that is needed to create a single worldwide
 24

research sector is to ensure that all types of capital equipment available worldwide are used
in all research activities undertaken anywhere in the world. Since ideas do not matter in
research, trade in the capital goods is all that is needed.
There is one final point worth emphasizing. Sergio R.ebelo [1990] offers a general
observation about multisector models that is relevant for the experiments considered here.

Consider a single sector model of the form C+K÷A B/0(K, A), where F0(.) is a
homogeneous of degree 1. In this example, K and A can denote any two arbitrary
capital goods. If the productivity parameter B0 increases, the balanced growth rate
increases. Consider next a two sector model in which there is an essential fixed factor £
that enters as an input in the homogeneous of degree 1 production function for

consumption and capital of type K: C+K =B1P1(K, A, L). The capital good A, however,
is produced by a homogeneous of degree one function '2(•) of K and A alone:

A = B2)'2(K, A). In this case, a change in the productivity parameter B1 has no effect on
the balanced rate of growth. It has only level effects. In contrast, an increase in B2
increases the balanced rate of growth.
The connection between Rebelo's observation and our results is as follows. We do
not consider changes in technology parameters like B1 and B2, but we do induce changes
in scale for functions that are homogeneous of some degree greater than 1. Increases in
scale are analogous to increases in the productivity parameters. In the knowledge driven
model, trade in goods exploits increasing returns in the sector that produces C and K,
but not in the sector that produces A. It is like an increase in B1 in Rebelo's two sector
model, and induces only level effects. Flows of ideas increase productivity in the research
sector that produces A, and are analogous to an increase in Rebelo's coecient B2.
Finally, trade in goods in the lab equipment model induces a scale effect that is like an
increase in B0 in Rebelo's one sector model.
 25

VI. Limitations of the Models and Extensions

As noted in the introduction, the analysis carried out in this paper takes the form of
thought experiments for idealized cases. These experiments reveal the following general
insight about the connection between economic integration and the rate of economic
growth. In a model of endogenous growth, if economic integration lets two economies
exploit increasing returns to scale in the equation that represents the engine of growth,
integration will raise the long run rate of growth purely because it increases the extent of
the market. Depending on the form of the model, this integration could take the form of
trade in goods, flows of ideas, or both.
This conclusion must be tempered by a large number of qualifications. First, there
is no consensus yet about whether the equation that is the engine of growth is homogeneous
of some degree that is greater than 1 in the basic inputs (as it is in both of the models
considered here) or instead is homogeneous of degree 1 (as it is, for example, in the papers
by Rebelo [19911 and Lucas [1988].)

Second, as noted in the introduciton, we have focused on trade between economies

with identical endowments and technologies to highlight the scale effects induced by
economic integration. In a general two sector framework, trade between economies that
have different endowments or technologies will induce allocation effects that shift resources
between the two sectors in each country. For example, Grossman and Helpman [1990]
show that trade between countries that have different endowments or technologies will
induce shifts between the manufacturing sector and the R&D sector that can either speed
up or slow down worldwide growth. If one wants to take the optimistic conclusions
reached in this paper literally, they are most likely to apply to integration between similar
developed regions of the world, for example between North America, Europe, and Japan.
 26

There are many details of R&D at the micro level that have been ignored in all of
the analysis. We have assumed that giving participants in the economy an incentive to
avoid redundancy in research is sufficient to ensure that no redundancy takes place. We
have also assumed that patents are infinitely lived and, implicitly, that the institutional
structure avoids patent races. We have not considered the role of secrecy in preserving
economic value for ideas. All of these restrictions are very strong. Grossman and Helpman
[1989dJ show how one element of the microeconomic literature on patents, the destruction
of monopoly profits by new discoveries, can be included in an aggregate growth model.
Other extensions will no doubt follow.
The functional forms used here cannot be literally correct. For example, in both of
our models, the output of patents at any date increases in proportion to the resources
devoted to R&D. This permits the solution for balanced growth paths using linear
equations, but it cannot be a good description of actual research opportunities. We would
expect that a doubling of research effort would lead to a less than two fold increase in R&D
output, in large part because of the coordination and redundancy problems at the micro
level that we have ignored. Addressing these issues would help reconcile a model in which
growth rates increase linearly in H in one case, or as a power of Hand £ in the second,
with a historical record showing that growth rates have indeed increased over time, but not
by nearly as much as the functional forms used here would suggest. More precision in the
definition of the input H that is most important for research would also be helpful in this
regard. In terms of their effect on research output, one presumably does not literally want
to equate two people holding high school degrees with one person holding a Ph.D. degree.
Perhaps the most interesting limitation of the models considered here is one that it
shares with many other models: there is no description of how ideas or information affect
the production of goods. Once one admits that ideas per se can influence research output,
it is apparent that they can influence the output of goods as well. Presumably this is what
learning-by-doing models try to capture with the assumption that some production
 27

parameter increases with cumulative experience: producing goods yields both goods and
ideas, and the ideas raise the productivity of the other inputs. Formal models in the
tradition of Arrow [1962] have not yet addressed the importance of communication
networks and information flows. When the learning-by-doing models are used in
international trade, it is implicitly assumed that there is a communication network that
extends throughout one national economy, yet does not cross national boundaries. Little
theoretical attention has been given to the analysis of policy choices that can affect the
efficiency of international communication networks and to explaining historical episodes
(e.g. the emergence of the textile industry in the United States and of the automobile
industry in Japan) that reflect large flows of information from developed industries in one
country to developing industries in another.
Given these limitations and qualifications, our only claim is to have formalized, and
we hope illuminated, an effect that is potentially important. If the discovery of new ideas
is central to economic growth, one should expect that increasing returns associated with
the opportunity to reuse existing ideas will be present. If the increasing returns extend to
the sector of the economy that generates growth, economic integration will induce scale
effects that will raise the long-run rate of growth. And because of the remarkable growth
of exponential functions, policies that affect long-run rates of growth can have very large
cumulative effects on economic welfare. Many other effects may be present as well, but in
future theoretical and empirical work, we argue that scale effects on growth that are
induced by economic integration are worth watching out for.
 28

Appendix

A. Derivation of equation (7)

In the lab equipment model, the value of total production in manufacturing and
research depends only on the aggregate stocks of inputs, not on their allocation between the
two sectors:

Y+ A/B = HLf ()1-a-fl di.

Taking its supply of H and L as given, each representative firm in the manufacturing
sector chooses levels of z(i) to maximize profits. Consequently, the first order condition

for the problem of maximizing Y+ A/B minus total input cost JP(i)z(i) di with
respect to the use of input i yields the economy wide inverse demand curve for good i.
The rental rate p that results when a total of z units of the capital good are supplied is

(A.1) p = (1.afi)B0Lflz().

Input producers choose z to maximize the present value of monopoly rent minus
the unit cost of each piece of capital, A = max (pz/r)-z. The first order condition that
determines the number of machines that the holder of the patent on good i rents to
manufacturing firms is

(A.2) (l_a_fl)2L_(U)r_1 - =
 29

which implies that p/r = (1-a-fi)1. The present discounted value of profit collected by
the holder of the patent can then be simplified to

(A.3) "A = (p/r)- =

Since = 1/B, this implies that = (1-a-fi)/B(a+fl). Substituting this expression into
equation (A.2) yields equation (7) in the text:

-°
rtecolo = Ba+fi(O+p) a÷fi(1 - afi)2

B. Derivation of equation (6).

The demand for the capital goods in this model has exactly the same form as in the
lab equipment model, with the qualification that since all of the demand comes from the
manufacturing sector, H must be replaced by H If we use equation (A.1) with this
replacement to substitute for p in the expression for we have

=
(a+fl)! =
! (1-a-fi)H,L 1-a+fl

Equating the wages of human capital in manufacturing and research yields

PASA = aH°'LA 1-a-fl Combining these expressions and solving for Hgives

H= .a(a+flY'(l-a-fl)'r = Ar. Hence, g = SEA = 6H-5H= SE-Ar.

 30

C. Trade in goods in lab equipment model is equivalent to complete integration.

If the interest rate remained constant, the value of a patent A = r/r would
double when trade in goods between two identical markets is introduced in this model.
The monopolist that sells in two identical markets and faces constant marginal costs of
production will maximize profits in each market independently and earn twice the flow of
profits that would accrue from one market alone. Since the value of the patent must
remain fixed at 1/B by the specification of the technology for producing patents, the
interest rate must increase to restore equilibrium.
As shown above, maximization of profit by the monopolist implies that p/r is
constant, so profit is proportional to . To offset the doubling of profit that the opening of
trade would otherwise induce, r must increase by enough to make the number of units of
capital supplied by the monopolist in each country fall by one half. From equation (A.2),
this will happen if r increases by a factor of This is same the factor increase in r
that results from doubling H and £ when the two countries are combined.
 31

REFERENCES

Arrow, Kenneth J., "The Economic Implications of Learning By Doing," Review of

Economic Studies XXIX (June 1962), 155—73.

Backus, David, Patrick Kehoe, andTimothy Kehoe. "In Search of Scale Effects in Trade
and Growth," University of Minnesota Working paper (1990).

Barro, Robert and Xavier Sala i Martin, "Public Finance in Models of Economic Growth,"
National Bureau of Economic Research, Working Paper No. 3362 (1990).

Dinopouios, Elias, James Oehmke, and Paul Segerstrom. "High Technology Industry
Trade and Investment: The Role of Factor Endowments," University of Florida
working paper (1990).

Dixit, Avinash and Joseph Stiglitz, "Monopolistic Competition and Optimum Product
Diversity." American Economic Review LXXVI (June 1977), 297—308.

_______________ and V. Norman, The Theory of International Trade, Cambridge

England: Cambridge University Press (1980).

Ethier, Wilfred J. "National and International Returns to Scale in the Modern Theory of
International Trade." American Economic Review LXXII (June 1982), 389—405.

Feenstra, Robert. "Trade and Uneven Growth," National Bureau of Economic Research,
Working Paper No. 3276 (1990).
 32

Grossman, Gene and Elhanan Helprnan, "Product Development and International Trade,"
Journal of Political Economy XCVU (December 1989a), 1261—1283.

_______________ and _______________, "Endogenous Product Cycles" National Bureau

of Economic Research, Working Paper No. 2113 (1989b).

_____________ and _____________, "Growth and Welfare in a Small Open

Economy," National Bureau of Economic Research, Working Paper No. 2809

(1989c)

______________ and _______________ "Quality Ladders in the Theory of Growth,"

National Bureau of Economic Research, Working Paper No. 3099, (1989d).

_______________ and _______________, "Quality Ladders and Product Cycles,"

National Bureau of Economic Research, Working Paper No. 3201 (1989e).

_______________ and _______________, "Comparative Advantage and Long Run

Growth" American Economic Review LXXX (September 1990), 796—815.

Helpman, Elhanan, "International Trade in the Presence of Product Differentiation,

Economies of Scale, and Monopolistic Competition: A Chamberlinian-Heckscher-
Olin Approach," Journal of International Economics IX (1981), 305—340.

Krugman, Paul, "Increasing Returns, Monopolistic Competition, and International Trade,"

Journal of International Economics IX (November 1979), 469—479.
 33

"Intraindustry Specialization and the Gains from Trade," Journal of

Political Economy VXXXIX (October 1981), 959—73.

Rethinking International Trade, Cambridge MA: MIT Press (1990).

Lancaster, Kevin, "Intraindustry Trade under Perfect Monopolistic Competition,"

Journal of International Economics X (1980), 151—75.

Lucas, Robert E. Jr., "On the Mechanics of Economic Development." Journal of

Monetary Economics XXII (1988), 3—42.

R.ebelo, Sergio. "Long Run Policy Analysis and Long Run Growth." Journal of Political

Economy, forthcoming (1991).

Rivera-Batiz, Luis A. and Paul M. Romer. "International Trade with Endogenous

Technological Change." University of Chicago, mimeo (1990).

Rosenberg, Nathan, Inside the Black Box, Cambridge England: Cambridge University
Press (1980).

Romer, Paul M. "Increasing Returns and Long—Run Growth." Journal of Political

Economy XCIII (October 1986), 1002—1037.

_______________ "Growth Based on Increasing Returns Due to Specialization."

American Economic Review LXXVII (May 1987), 56-62.
 34

"Endogenous Technological Change." Journal of Political

Economy, XCXXVUI (October 1990), S71—S102.

Segerstrom, Paul, T. C. A. Aiiat and Elias Dinopoulos, "A Schumpeterian Model of the
Product Life Cycle." American Economic Review, forthcoming (1990).

Young, Alwyn. "Learning by Doing and the Dynamic Effects of International Trade."
Quarterly Journal of Economics, forthcoming (1991).