reading discussion

American Economic Association

The Structure of Foreign Trade Author(s): Elhanan Helpman Source: The Journal of Economic Perspectives, Vol. 13, No. 2 (Spring, 1999), pp. 121-144 Published by: American Economic Association Stable URL: https://www.jstor.org/stable/2647122 Accessed: 19-08-2019 19:15 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide

range of content in a trusted digital archive. We use information technology and tools to increase productivity and

facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at

https://about.jstor.org/terms

American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The Journal of Economic Perspectives

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

Journal of Economic Perspectives-Volume 13, Number 2-Spring 1999-Pages 121-144

The Structure of Foreign Trade

Elhanan Helpman

W v r orld merchandise exports amounted to $5.3 trillion in 1997 and exports

of commercial services amounted to $1.3 trillion. These are unprece-

dented volumes that have expanded much faster than income in the postwar period. Figure 1 presents long-term trends in the real volumes of merchan-

dise trade and output. While trade grew at an annual rate of 2.6 percent during this

period, output grew at only 1.5 percent. More than half of the volume of merchan-

dise trade flows amongst developed countries and less than 15 percent flows

amongst developing countries. The rest, about one third, represents North-South

trade between developed and developing countries.

What explains these large volumes of trade? Why do some countries export

computers while others export footwear? Can exports of airplanes be explained in

the same way as exports of paper products? Questions of this type have been

examined for many years. In attempting to answer them, economists have devel-

oped an elaborate analytical apparatus that has been greatly enriched in the last two

decades. They have used the insights from trade theory to examine ever richer data

sets in order to discover systematic patterns of trade flows and to evaluate how well

available theories match these data. Nevertheless, although we do have today better

answers to some of these questions than our predecessors had 40 years ago, the

evolving structure of world trade defies simple explanations. I describe in this paper

what we know about foreign trade and in what ways our understanding has

improved as a result of the last 20 years of research.

The paper is in five parts. In the next section I briefly review early insights:

David Ricardo's theory of comparative advantage and a simplified version of the

m Elhanan Helpman is Professor of Economics, Harvard University, Cambridge, Massachu-

setts, and Tel Aviv University, Tel Aviv, Israel, and Fellow of the Canadian Institute for

Advanced Research, Toronto, Ontario.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

122 Journal of Economic Perspectives

Figure 1

Long-term Trends of World Merchandise Trade and Output, 1950-1994

2,000

Volume Indices, 1950 = 100

1,500

Merchandise trade

1,000

500 , 0 ,-_ , ~~~~ tUc~rchandise output

1950 1960 1970 1980 1990

Source: WTO, Trends and Statistics 1995.

Heckscher-Ohlin theory. In the following section I discuss and evaluate further

developments of the Heckscher-Ohlin approach during the 1960s, 1970s and

1980s. While the first two decades provided mostly theoretical insights, major

empirical innovations appeared in the 1980s. The third part of the paper is devoted

to a discussion of the most recent developments on this front. Next, in part four,

comes a discussion and evaluation of research based on economies of scale and

product differentiation. This work was done mostly in the 1980s and 1990s. An

emphasis on the interplay between theoretical and empirical research characterizes

the entire presentation. Conclusions are provided in the closing section.

Early Insights

David Ricardo's theory of comparative advantage, developed at the beginning

of the 19th century, has played a major role in modern thinking about trade.

Ricardian trade models assume that only labor is used to produce goods and

services, with a given fixed coefficient between labor and output of a particular

product in each country. The theory predicts that a country will export products in

which it has a comparative advantage; that is, products where its labor productivity

is high relative to its labor productivity in other products. The simple Ricardo

model remains useful for thinking about a host of issues, such as the effects of

technological progress on patterns of specialization and the distribution of gains

from trade.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

Elhanan Helpman 123

However, there are hardly any empirical studies of trade flows that build strictly

on Ricardo's insights. A few appeared in the 1950s and '60s (McDougall, 1951,

1952; Stern, 1962), but it quickly became apparent that such models are of limited

use in employing data to analyze the overall structure of foreign trade. The main

difficulty is that Ricardo's theory is mute on a key ingredient: what causes labor

productivity to differ across countries?' Differences in the use of capital comprise

an important source of variation in labor productivity. Capital-rich countries are

able to allocate more capital per worker to all economic activities than capital-poor

countries, but they may do so to a different degree in various lines of business. This

raises the question: what determines the allocation of capital to industries and

thereby labor productivity? But once the role of capital is taken seriously, it may be

best to abandon the exclusive focus on labor productivity and think about what

determines trade flows amongst countries that have additional inputs besides labor.

Eli Heckscher (1919) and Bertil Ohlin (1924) provided a framework for

thinking about trade in this situation. They emphasized the roles of labor, capital

and land in agriculture and industry, trying to explain how their availability shapes

a country's pattern of specialization and trade. Paul Samuelson and his followers

elaborated a two-factor two-sector version that became the cornerstone of modern

trade theory (Samuelson, 1948; Jones, 1956-57, 1965). Samuelson's version is crisp

and elegant. By focusing on labor and capital as inputs and on export- and

import-competing sectors, it cuts to the heart of the matter. The model was widely

adopted as the workhorse of the profession. According to the two-factor, two-sector

version of the Heckscher-Ohlin theory, a country should export the product that is

relatively intensive in using the factor with which the country is relatively well-

endowed. Thinking about labor and capital as the two inputs, it means that a

capital-rich country-a country that has more capital per worker than its trade

partners-should export the capital-intensive product.

The argument can be made in two parts. First, if factor prices are not equal-

ized, then the rental rate on capital relative to the wage rate is lower in the

capital-rich country.2 As a result, it uses in all product lines more capital per worker

than the capital-poor country. But, as shown by Lerner (1952), under these

circumstances the capital-rich country has a cost advantage in the capital-intensive

products, which it ends up exporting. This implies that its exports are more

capital-intensive than its imports. That is, if we were to calculate how much capital

and labor are embodied in the country's exports and how much capital and labor

are embodied in its imports, we should find that for the capital-rich country the

1 Eaton and Kortum (1997) provide the first study of a Ricardian model that contains an explanation of

labor productivity based on a country's technology level. Using extreme value distributions for labor

productivity they derive an equation for bilateral trade flows. They estimate this equiation for a sample of OECD countries, using cumulative investment in R&D and the number of scientists and engineers as

proxies for technology levels. The fit appears to be good.

2 Capital accumulation in conjunction with international (financial) capital mobility tend to reduce divergent rental rates on capital. Nevertheless, differences can persist for long periods of time.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

124 Journal of Economic Perspectives

ratio of capital to labor is larger in exports than in imports. Second, if factor prices

are equalized, then the capital-rich and capital-poor countries use the same ratios

of capital to labor to produce identical products. But because the capital-rich

country has a disproportionately large amount of capital relative to labor, it ends up

producing a disproportionately large amount of capital-intensive products. Other-

wise it cannot maintain full employment of labor and capital. It then follows that

with a similar composition of demand (that results, for example, from identical

homothetic preferences in all countries), the capital-rich country exports capital-

intensive products in this case too. So we find again that in the capital-rich country

the ratio of capital to labor embodied in exports is larger than the ratio of capital

to labor embodied in imports.

There are two concepts of trade in this theory: trade in goods and trade in

factor content. Trade in goods is standard; wheat and airplanes are goods, and they

can be imported or exported. Trade in factor content is different. It refers to the

inputs that are embodied in exports or imports. For example, if a unit of imported

wheat is produced with half a unit of land, three units of labor and five units of

water, then the factor content of the imported wheat is half a unit of land, three

units of labor and five units of water. Using this approach one can calculate the

factor content of exports, the factor content of imports, and the factor content of

net exports, which is the difference between the two.

Leontief (1954) put the prediction that a capital-rich country should export

capital-intensive products to a test. He calculated labor-output and capital-output

ratios for various sectors of the U.S. economy, and then-with the aid of these

coefficients- calculated how much labor and capital are embodied in exports and

how much in imports. Surprisingly, Leontief found that in 1947 the capital-labor

ratio embodied in imports exceeded the ratio embodied in exports by 60 percent!

The surprise emanated from the fact that after the war, the United States was

considered to be the most capital-rich country in the world and the Heckscher-

Ohlin theory predicts for such a country a higher capital intensity of exports than

imports. His finding became known as the "Leontief paradox."

There were attempts to examine additional data sets, but while the paradox is

less pronounced in later data sets, it does not disappear. Leontief proposed his own

resolution for the paradox. He had used the assumption that the U.S. input-output

coefficients also apply to imports, which is the case when foreign suppliers use the

same techniques of production as domestic producers. When countries have access

to the same technologies and factor prices are equalized, this assumption is

justified. But, pointed out Leontief, if a U.S. worker is much more productive than

a foreign worker, then the U.S. should export relatively labor-intensive products.

Why would, however, U.S. workers be so much more productive, especially after

controlling for capital availability? This explanation seems to resolve one paradox

by introducing another.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

The Structure of Foreign Trade 125

Further Developments: 1960s, 1970s and 1980s

The two-factor two-sector version of the Heckscher-Ohlin theory was extended

in the 1960s and '70s; Ethier (1984) offers a review of the literature at this time.

These studies had a variety of purposes, but what is central for this paper is the fact

that from these studies grew a surprisingly simple theoretical specification, in which

two types of relationships could provide the underpinning for an analysis of the

generalized Heckscher-Ohlin theory.

The first set of relationships involves production. Let ai, represent the quantity of input i used in the manufacturing of one unit of output j. I ignore intermediate

inputs. Therefore these coefficients describe primary inputs only, which would

include various types of capital, such as machines and structures; various types of

labor, such as high school dropouts and college graduates; and various types of

land, such as pasture and arable land. Cost-minimizing manufacturers choose these

coefficients from the available technology, taking factor prices as given. As a result,

these coefficients depend in a particular country on its technology and factor

rewards. In the simplest version of this model, the technology is taken to be the

same everywhere and factor prices are assumed to be equalized across countries.3

In this case, the same coefficients are used in all countries. Then, if an economy

fully employs its resources, we can derive a factor-market clearing condition. (If a

resource is not fully employed, replace its quantity with the actually employed

value.) Let Vk represent the quantity of input i in country k. Similarly, let Xk

represent the output of good j in country k. Then full employment of resources

implies

E ai,X>= Vt

for all inputs i and all countries k.

The second set of relationships comes from consumption. Preferences are

assumed to be the same in all countries and homothetic; that is, a ray drawn from

the origin of an indifference map will intersect all indifference surfaces at points

with the same slopes. These assumptions are strong ones. They imply that the

composition of consumption (the share of spending going to a product) is the

same everywhere. Namely, if we denote by sk the share of country k in consumption,

then consumption of good j in country k is given by

C = skXj

3Factor prices are not the same in all countries, not even in the OECD countries. But as O'Rourke and Williamson (1999) demonstrate, trade and migration are powerful forces that lead to convergence of

factor prices. See Williamson (1998) for a review.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

126 Journal of Economic Perspectives

for all goods j and all countries k, where X1 = k Xj is aggregate world output of

good j. When trade is balanced, the share Sk equals country k's share in world

income.

These two sets of fundamental relationships imply empirical specifications.

Each country's production of any good is determined by its primary factor endow-

ments, while each country's consumption of goods is determined by overall spend-

ing.4 Exports will be those products where the country produces more than it

consumes, and imports will be those products where the country produces less than

it consumes. The model allows countries to run trade deficits or surpluses. The

model implies a linear set of relationships between net exports and factor endow-

ments. Each relationship, for petroleum products, forest products, machinery or

chemical products, can be estimated from cross-country data. We do not need data

on production coefficients or technology, because these are assumed to be the

same in all countries. We do need data on net exports, which are readily available,

and data on factor endowments, which are not readily available-at least not in

comparable form.

Ed Leamer has done more than anybody else to promote a research agenda

centered on the construction of such data sets. For example, Leamer (1984) started

a new line of empirical research by estimating this linear relationship using a newly

constructed data set. The simple linear specification performs very well in explain-

ing actual patterns of trade on a cross section of 60 countries in his data set, for

both 1958 and 1975. Examples of the type of effects that were estimated include the

availability of oil raising net exports of petroleum products, but also reducing net

exports of machinery in 1975; and the abundance of literate, non-professional

workers raising exports of labor-intensive manufactures, such as apparel and foot-

wear. Leamer's estimation maintains the assumption of linearity, which works well

for most sectors. In some sectors, however, such as chemicals, the data favor a

non-linear specification.

While estimates of this type are interesting, they do not provide a test of the

generalized Heckscher-Ohlin theory. The reason is that the theory predicts a

relationship between endowments and trade mediated by technology. To test it

therefore requires independent information about all three objects: technology,

endowments, trade.

Two concepts of trade have surfaced so far: trade in goods and trade in factor

content. The former is a natural focus of trade theory, and was examined by

Leamer. The latter was examined by Leontief. Leontief's procedure can be gener-

alized by constructing measures of the factor content of net exports, as suggested

by Vanek (1968). With identical technology coefficients in all countries, this

procedure is straightforward. Just convert net exports of goods into the factor

content of net exports, by multiplying the quantity of net exports of each good with

4Uniqueness of the production structure requires additional assumptions, such as the equality of the number of inputs and outputs.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

Elhanan Helpman 127

the input coefficient aij and summing over all goods, using the common technology matrix. What Vanek has shown, is that under the model's assumptions this measure

of factor content should equal the economy's measure of factor abundance. The latter

is constructed as follows. Begin with the total factor content of production, which

equals the economy's factor endowment, and then subtract out the factor content

that goes into domestic consumption, where consumption of each input is propor-

tional to the country's share of world spending. The result will be a measure of

factor abundance. On net, inputs that are relatively abundant are exported while

those that are relatively scarce are imported. Vanek's key equation is

, = V- S Vi

for all inputs i and all countries k, where FP is the factor content of net exports of input i in country k and Vi = SkVi is the aggregate endowment of input i in the

world economy.

Vanek's (1968) equation was used by Leamer (1980) to point out a shortcom-

ing of Leontief's procedure: whenever there are more than two inputs, that is, more

than just labor and capital, a comparison of the ratio of the embodied quantities of

just two of them in imports and exports does not provide the relevant metric for

rejecting the theory. This may sound like a possible neat resolution of the Leontief

paradox. It is not. As pointed out by Brecher and Chaudhri (1982), the fact that the

United States was a net exporter of labor services has an implication for consump-

tion per worker; that is, U.S. consumption per worker should fall short of world

consumption per worker, which is not born out by the data.

Bowen, Leamer and Sveikauskas (1987), who were the first to use independent

information about endowments, technology and trade, performed two types of tests

on Vanek's (1968) key equation. They used the U.S. technology matrix to calculate

the factor content of net exports F k for all countries. Their tests involved compar-

ing these measures of factor content with the measures of factor abundance

i- s'Vi. They were done with data for 12 inputs and 27 countries (for 1967). One test compared the signs of the two calculations-that is, whether a factor that

was predicted to be exported by the factor abundance measure was also exported

by the factor content measure (and similarly for imports)-and found disagree-

ment one-third of the time. Another test compared the rank order of the inputs in

the calculated measures of the factor content of net exports with the measures of

factor abundance and found disagreement about half the time. This appears to be

bad news for the expanded Heckscher-Ohlin theory. But how bad the news is is

hard to gauge, because in this exercise the theory is not tested against a well-

specified alternative. For this reason it is also possible to take a more positive

attitude and to argue that the theory explains a reasonably large fraction of the

variation-across factors and countries- of the factor content of net trade flows.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

128 Journal of Economic Perspectives

Recent Advances

Although Bowen, Leamer and Sveikauskas (1987) pointed out difficulties with

the Vanek equation, they did not investigate whether the data deviate systematically

from the theoretical predictions. This important task was undertaken by Trefler

(1995). He compiled a new data set, for 33 countries, that disaggregates endow-

ments into nine inputs. The countries in the sample accounted for three-quarters

of world exports and nearly 80 percent of world income in 1983. Again, Trefler first

calculated the factor content of net exports, using the U.S. technology matrix for

all countries, and then compared them to the factor abundance measures. Vanek's

(1968) theoretical prediction is that the two measures should have a correlation

of 1. Trefler found instead a correlation of .28.5 A sign test of the Bowen, Leamer

and Sveikauskas (1987) type was successful in only about one-half of the cases. It

therefore appears that Trefler's data fits the expanded Heckscher-Ohlin theory as

well or as badly as Bowen, Leamer and Sveikauskas's data does.

Trefler (1995) showed clearly in what ways these data deviate from the theo-

retical predictions. First, the measures of the factor content of net exports are

compressed towards zero relative to the factor abundance measures. That is, even

in cases in which both variables have the same sign, the former is much smaller in

absolute value than the latter. This compression is striking. It is not unusual in these

data for the absolute value of the factor abundance measure to be 10 to 50 times

higher than the absolute value of the factor content of trade, as one can see in

Figure 2, which presents these measures for capital (every point represents a

country). Second, whenever a poor country exports an input on net, it exports less

than predicted by its factor abundance measure. And whenever it imports an input

on net, it imports more than predicted by its factor abundance measure. For rich

countries the opposite is true. If one takes the factor content measure of net

exports derived from trade flows and then subtracts from it the factor abundance

measure, derived by subtracting the factor content of domestic consumption from

the domestic factor endowment, that difference has a correlation of .87 with per

capita GDP. Third, poor countries tend to be abundant in more factors than rich

countries, in the sense that they have more inputs for which the domestic endow-

ment exceeds the quantity of the factor embodied in domestic consumption.

Indeed, the correlation between the factor abundance measure and per capita GDP

is -.89.6

5 All the reported calculations use normalized data, which is obtained from the original data by dividing

every input by the standard deviation across countries of the difference

PI _ ( Vk _ S k V)

and by the measure of country size (Sk) 1/2. Here relative country size is measured by its share in aggregate GDP.

b3 These results cannot be explained by trade imbalances, which are much too small for this purpose.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

The Structure of Foreign Trade 129

Figure 2

Factor Abundance and Factor Content Measures: Capital data

0.05

0 0

-0.6 -0.4 . *-0.2 0 * .0.2 . 0.4 0.6 * 0.8

-0.05

u -0.1

-0.15

-0.2J

Factor abundance

Source: Trefler (1995). Every observation has been divided by the quantity of the input in the

country.

This characterization is of lasting value. It gives us a better understanding of

the ways in which the data do not match the theory. It provides a clear theoretical

challenge: How should the model be modified to better fit the data?

Trefler also followed up on the earlier suggestion of Leontief that perhaps

these differences could be explained by differences in productivity across inputs

and countries. Suppose, as suggested by Leontief, that inputs are not equally

productive in all countries. If U.S. labor is, for example, two times as productive as

labor in Italy, then 1000 hours of U.S. labor services are equivalent to 2000 hours

of labor services in Italy. Taking one country as the benchmark, we can then

convert the endowment of all other countries into equivalent units of the bench-

mark country. It is then possible to use Trefler's data to calculate what the

productivity differences would have to be between countries so that a modified

Vanek equation would hold exactly.7 Using this approach, Trefler (1993) first

calculated labor and capital productivity coefficients for a set of countries relative

to the United States, and he then compared them with the wage rate and the return

to capital in these countries relative to the United States. According to the theory,

7Taking one country as the benchmark, we can convert the endowment Vk of country k into T,V equivalent units of the benchmark country, where irk is the productivity of input i in country k relative

to the same input in the benchmark country. Under these circumstances Vanek's equation in the text

becomes

F=v kVj - Sk E TX

for all k and i, where the factor content of net exports F is calculated using the benchmark country's

technology.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

130 Journal of Economic Perspectives

if there is factor price equalization, then in a cross-country comparison the relative

rewards should equal the relative productivity parameters. Indeed, Trefler found

the relative factor rewards to be highly correlated with the calculated relative

productivity parameters.

In a later paper, Trefler (1995) took several alternative approaches to estimat-

ing productivity parameters. In one approach, he assumed that the productivity

parameters are country-specific but not factor-specific; that is, if U.S. labor is two

times as productive as Italian labor, so is U.S. capital and land. Thus, each country's

productivity is represented by a common productivity advantage in all lines of

business, and this advantage can be thought of as a Hicks-neutral technical differ-

ence. Trefler's second observation-namely, that poor countries appear to export

too little of their abundant factors while rich countries export too much-suggests

that Hicks-neutral productivity differences can help explain these data. Trefler's

second approach was to divide countries into two sets, developed and developing,

so that the factor-specific productivity parameters are the same in each group of

countries, but differ across the two groups. Of these two choices, Trefler concluded

that the Hicks-neutral specification performs better.8

To better account for the "missing trade"-that is, the finding that the factor

content of net exports as measured from trade statistics is so much smaller than the

difference between total factor endowments and factors that are embodied in

domestic consumption-Trefler also introduced a home bias in demand. He found

that this bias, together with the Hicks-neutral productivity differences, provides the

best explanation of the data. I find the use of a home bias in demand unappealing.

There is plenty of independent evidence that technologies differ across countries

(for example, Harrigan, 1997). There is no such evidence for demand patterns,

except for biases that are related to income levels.

Trefler uncovered patterns in the gaps between the Heckscher-Ohlin theory

and the actual data on factor content and factor abundance. Apparently, techno-

logical differences between countries help to explain these gaps. The work by Davis,

Weinstein, Bradford and Shimpo (1997) supports this emphasis. These authors set

out to evaluate the two fundamental relationships discussed earlier in this paper:

the production relationship in which sectoral output levels are used to calculate

aggregate demands for inputs which are then required to equal factor endowments;

and the consumption relationship that consumption vectors in different countries

are proportional to each other.

They evaluated the production relationship in a cross-section of countries and

a cross-section of Japanese regions, using Japan's technology matrix in all cases.

They used output data to calculate the demand for inputs. According to the theory,

this vector of input demands should match the vector of factor endowments. To

examine how close these two vectors are to each other they performed on them

rank order tests of the type introduced by Bowen, Leamer and Sveikauskas (1987).

8 Bowen, Leamer and Sveikauskas (1987) also examined productivity differences. Unfortunately, due to a programming error their results are not correct.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

Elhanan Helpman 131

They found that the match is not good in the international data, but is remarkably

accurate in theJapanese regions. One plausible interpretation of this finding is that

techniques of production are very similar across Japanese regions but differ signif-

icantly across countries. The failure of the common technology model to work in

the international data can emanate either from lack of factor price equalization

(which we know to be true from data on wages) or from differences in technolog-

ical opportunities (which we also know to be the case). At the moment there are no

estimates of how much of the failure is attributable to each of these potential

causes.9

In looking at the consumption relationship, Davis, Weinstein, Bradford and

Shimpo (1997) considered how accurately the structure of consumption can be

captured by a model which posits that the share of consumption going to different

goods is the same in all areas. In one exercise, they tested whether regional

"absorption" (consumption plus investment) vectors in Japan are proportional to

the aggregate Japanese absorption vector, concluding that they are. Next they

tested whether the Japanese absorption vectors are proportional to world produc-

tion, concluding that they are.

Since both the fundamental production and consumption relationships seem

to hold well for the regions of Japan, Davis, Weinstein, Bradford and Shimpo

(1997) went on to test a Vanek-style (1968) relationship with Japanese data,

comparing a measure of the factor content of net exports with a measure of factor

abundance. What they found was that even though the fundamental underlying

production and consumption relationships hold across regions ofJapan, the Vanek

relationship does not fit theJapanese data well. As in Trefler's data set, this case too

exhibits "missing trade;" namely, the computed factor content of net exports is

much smaller in absolute value than what is predicted by factor abundance (that is,

by the total factors subtracting out factors used for domestic consumption). For

example, while net exports of noncollege labor services amount to 3.6 percent of

this factor endowment, the factor abundance calculation predicts imports of 31 per-

cent of this factor endowment-a large deviation indeed.

How can these different findings be reconciled? Davis, Weinstein, Bradford

and Shimpo (1997) propose a partial resolution. The test just described used

Japanese production coefficients, assumed to be the same for all countries. If one

instead recognizes that other countries may use different production coefficients,

it is possible to obtain a modified Vanek equation in which the world's aggregate

factor endowment is replaced with the inputs that would have been used if every

country were to use Japan's technology matrix. With this modification in place the

puzzle of the "missing trade" is reduced, although not eliminated.

It is clearly possible to use a Heckscher-Ohlin modeling approach to explain

the trade data if one allows arbitrary differences in techniques of production across

countries. But this answer is not useful; after all, any and every pattern of factor

9 Impediments to trade can cause differences in factor prices. But such impediments have declined

systematically over time. And regional wage differentials persist for long periods of time within countries.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

132 Journal of Economic Perspectives

content can be explained with arbitrary differences in techniques of production.

The challenge is to use key features of countries to explain the reason for differ-

ences in techniques of production, and in turn, to have these differences better

explain the trade data.

One possible approach is to look at models where the differences in tech-

niques of production come about because factor prices are not equalized across

countries. To give an example, consider a world of two countries, two inputs and

two outputs, in which differences in factor composition are wide enough that there

is no factor price equalization. Let the factors be capital and labor, let the products

be food and clothing, and let clothing be capital-intensive. The technologies are

the same in both countries. However, since country A has a high capital-labor ratio

and therefore a relatively low rental rate on capital, it employs more capital and less

labor than country B would employ in the same line of business. Suppose also that

due to the extreme differences in factor endowments, country A manufactures only

clothing and B produces only food. It can be shown that under these circumstances

the capital-rich country A exports on net less capital services than predicted by

Vanek's factor abundance measure and imports on net less labor services than

predicted by this measure (Helprnan, 1998b). A similar conclusion is obtained from

an analysis of the labor-rich country. To see why, note that using the factor

coefficients from country A introduces no bias in the calculation of the factor

content of A's exports. But it biases the calculation of the factor content of A's

imports, because foreign manufacturers use different input-output coefficients-

they use less capital and more labor per unit output. Therefore the use of A's

coefficients overstates the amount of capital and understates the amount of labor

embodied in its imports. As a result net exports of capital services are understated

and so are net imports of labor services. This is why differences in techniques of

production that result from factor price differences help to explain the phenom-

enon of "missing trade."

But how much "missing trade" can be explained by cross-country variations in

factor rewards? At the moment there is no clear answer to this question. Davis and

Weinstein (1998a) have recently shown that, indeed, countries with more capital

per worker use more capital-intensive techniques than countries with less capital

per worker. And moreover, adjustments of technology matrixes for such differences

in factor endowments help to explain the factor content of net exports. However,

Hicks-neutral differences in technology are also required for a good fit, and such

productivity differences remain extremely important even after accounting for

differences in factor composition.10

10 The problem of missing trade appears to be much too severe to be explained with moderate differences between countries. For example, using a regression approach to look at Trefler's (1993) data

set of 33 countries, Gabaix (1997) estimated the affects of factor endowments on the factor content of

net exports. He allows factor-specific productivity levels to differ across countries, but assumes that they

are proportional to factor rewards. Under these circumstances the model predicts a coefficient of 1 on

the factor-augmented endowment measures. Gabaix finds instead coefficients close to zero. In addition

he shows that Trefler's (1993) calculation of cross-country differences in productivity of inputs is

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

The Structure of Foreign Trade 133

Things do not look so bleak when attention is focused on the pattern of

specialization in domestic production rather than trade. Reeve (1998) estimated a

relationship between outputs and inputs for a sample of 20 OECD countries, using

data on 15 sectors and five types of inputs: capital, three grades of labor and arable

land. He found that cross-country variations in factor endowments explain over 40

percent of the variation in output levels. If one allows for cross-country differences

in Hicks-neutral productivity levels-that is, the productivity of a factor is country-

specific but does not vary across inputs in a country-the fit improves significantly

and an additional 7 percent of the variation in output levels is explained. In

decomposing output changes from 1970 to 1985, Reeve finds that shifts in factor

endowments and the techniques of production explain over 80 percent of the

changes in the sectoral output levels. Interestingly, changing factor endowments

contributes about twice as much as changing techniques of production.

This suggests that as imperfect as the theory is, some of its components fare

well empirically. My conclusion from the evidence is that to close the gap between

the theory and the data, we need to model more carefully the cross-country

differences in techniques of production that are driven by both technological

differences and differences in factor rewards. This view is strengthened by Hakura's

(1997) findings, as well as by Davis and Weinstein (1998a). Using data for five of the

original European Union countries, she first confirms the Bowen, Leamer and

Sveikauskas (1987) finding that a Vanek-style (1968) comparison of the factor

content of net exports with a factor abundance calculation has the correct sign

pattern in only about one-half of the observations. But this fraction rises to 70-80

percent when each country's techniques of production are used instead of a

common technology matrix. This fraction rises even further when allowance is

made for non-traded intermediate inputs. Evidently, allowing for differences in

techniques of production can dramatically improve the fit of factor content equa-

tions. Now economists need to identify the forces that induce countries to choose

different techniques of production.

Economies of Scale and Product Differentiation

Concurrent with the refinement of the Heckscher-Ohlin trade theory and the

development of its empirical implications for the factor content of net trade flows,

a new trade theory that emphasizes economies of scale and product differentiation

emerged in the 1980s. Helpman (1984) offers a review of the literature on trade

with economies of scale and Krugman (1995) reviews the literature on trade with

differentiated products and monopolistic competition. At the early stages of its

development, the new theory seemed to threaten the Heckscher-Ohlin orthodoxy.

insensitive to the factor content of net exports, in the sense that virtually identical numbers are obtained

when the factor content of net exports is assumed to be zero.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

134 Journal of Economic Perspectives

But it became clear that these new explanations of trade were actually complemen-

tary to the explanations provided by factor endowments. This called for an inte-

grative view of foreign trade that would allow for an interplay between economies

of scale, product differentiation, and factor proportions. Helpman and Krugman

(1985) developed such an approach, making allowance for sectors that differ in

their sources of scale economies, in conduct, and in market structure.

Much of this research was originally motivated by the observation that large

volumes of trade flow between countries with similar factor proportions and that

significant trade overlap exists within industries, which seemed to imply that a

considerable portion of trade was not being driven by factor endowment differ-

ences. These facts have not changed. In 1996, the 15 countries of the European

Union exported (and imported) a little over $2 trillion worth of merchandise, and

about 65 percent of this trade was within the EU. Europe's imports from Japan

alone exceeded its imports from all of Africa and were more than twice as high as

its imports from all of the Middle East (WTO, 1997, Table Al0). Evidenfly, the EU

countries trade with countries that are similar to themselves more than they trade

with the very different economies of Africa and the Middle East. More broadly, the

industrial countries trade with each other much more than they trade with less

developed countries.

Measures of trade overlap within industries have remained high. These mea-

sures consider imports and exports within each industry, evaluate how much

overlap there is in every sector, and provide a summary statistic of the share of total

trade that consists of such overlaps.'1 To construct such an index for bilateral trade between two countries k and 1, take the minimum of the bilateral import volumes;

for example, if in one industry country k imports $20 billion from I and country I

imports $30 billion from k, choose the $20 billion. Add these volumes up. Divide by

the sum over all industries of all bilateral imports. Multiply by two, to adjust for the

fact that the numerator refers to imports of only one country, while the denomi-

nator refers to imports of both countries. By construction, this index is between

zero and one with higher values representing more trade overlap. To take a couple

of examples, the share of intra-industry trade in the United Kingdom was 53.2

percent in 1970, and then increased to 74.4 percent in 1980 and 84.6 percent in

1990. In Germany, it increased from 55.8 percent in 1970 to 56.6 percent in 1980

and 72.2 percent in 1990 (OECD, 1996, Table 1.8). These two countries represent

a general trend of rising shares of intra-industry trade. Declines in this index are

rare, but it happened in Norway as a result of the oil boom. Shares of intra-industry

trade are especially high in some sectors. In pharmaceuticals, for example, intra-

industry trade accounted for 98 percent of trade in finished products in the United

States in 1991, 70 percent in France and 32 percent in Japan. In the same year,

intra-industry trade accounted for 70 percent of trade in intermediate pharmaceu-

" According to the Heckscher-Ohlin theory such overlaps should not exist. Davis (1995) has proposed, however, that small differences in individual product-related technologies can produce such overlaps

even in a Heckscher-Ohlin framework.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

Elhanan Helpman 135

tical products in the United States, 82 percent in France and 64 percent in Japan

(OECD, 1996, Table 2.12).

Helpman and Krugman (1995) have shown that economies of scale, product

differentiation, and various forms of conduct are compatible with factor price

equalization and, as a consequence, with Vanek-type equations for the factor

content of trade. For this reason the empirical evidence that I reviewed so far is also

relevant for the richer theory developed in the 1980s. Grossman and Helpman

(1991) show conditions under which the Vanek-type equations also remain valid in

economies that invest in research and development.

Although factor price equalization and the employment of identical tech-

niques of production in all countries can take place with economies of scale, their

presence makes this less likely. Scale economies drive countries to specialize in

different products, which enhances the incentives for foreign trade. For this reason,

they help to explain large trade volumes between similar countries. At the same

time, economies of scale make it more likely that countries will employ different

techniques of production. This is especially so when there are dynamic economies

of scale, be they driven by learning-by-doing or investment in research and devel-

opment, in which case some companies have access to technologies that are not

available to rivals.12 Since the empirical evidence does not appear to be consistent with the use of identical techniques of production worldwide, or even within groups

of relatively homogeneous countries, the study of country-specific technological

developments becomes important for understanding international trade.

Intra-Industry Trade

The first comprehensive study of the extent of trade overlap was done by

Grubel and Lloyd (1975). They devised the index of intra-industry trade described

a moment ago, and showed that the index was high in many countries. Even before

the theory of intra-industry trade was properly developed, Loertscher and Wolter

(1980) had established some patterns about trade overlap. In particular, they found

that the share of intra-industry trade is high when the trading partners are highly

developed and at a similar level of development, and when the trading partners are

large and do not differ too much in size.

To explain extensive intra-industry trade, sectors with product differentiation

were introduced into the theory. Many products are differentiated by brand, from

simple goods such as breakfast cereal, toothpaste or clothing, to sophisticated ones

such as cars, computers or magnetic resonance imaging machines. Many producer

goods are differentiated in some way as well, including capital goods such as drilling

machines and intermediate inputs such as microprocessors. The theory applies to

all such goods. It starts by noting that product differentiation typically involves

economies of scale. A brand has to be developed, such as a lightweight laptop, or

12 Such advantages exist at least temporarily, as is evident from a variety of technological races in the electronics and pharmaceutical industries. See Grossman and Helpman (1995) for a review of the

literature on the links between trade and technology.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

136 Journal of Economic Perspectives

it has to be designed, such as a dress fashion. In either case, the company that

develops the product gains some monopoly power, because the market does not

provide a perfect substitute for its unique brand. Moreover, companies are driven

to differentiate their creations from the brands of rivals.

Economies of scale limit the range of products that are profitably supported by

the market; the smaller the economies of scale, the more brands become available.

With international trade, countries specialize in different brands. When every

country demands a wide spectrum of varieties, international trade leads naturally to

trade overlap: brand-specific economies of scale lead to intra-industry trade. Al-

though empirical researchers have sought an association between the degree of

economies of scale in a sector and the extent of its intra-industry trade, the theory

does not appear to imply such a relationship. What matters is that economies of

scale exist, not their size.

According to the theory, the share of intra-industry trade is larger between

countries that are similar both in composition of factor endowments and in size.

Helpman (1987) examined the empirical validity of these results. He used the

absolute difference in GDP per capita to capture the composition of factor endow-

ments and the size of each country's GDP to capture the size variables. For a sample

of 14 industrial countries he estimated such a relationship to assess how well these

variables predicted the share of intra-industry trade in bilateral trade flows between

nations. He found that the partial correlations had the predicted signs for most of

the years between 1970 and 1981; that is, the extent of trade overlap was larger the

more similar were the countries' income per capita, the smaller was the larger

country and the larger was the smaller country.

However, this relationship may not be robust. In Helpman's (1987) results, the

magnitude of the findings appears to weaken over time. Moreover, Hummels and

Levinsohn (1995) both confirm Helpman's findings and call them into question.

They find that Helpman's results continue to hold, for example, if instead of using

income per capita to capture factor differences, one uses income per worker or

absolute differences in capital-labor and land-labor ratios. But when they inserted

country-pair dummy variables into the specification, it appeared that most of the

variation in the share of intra-industry trade could be explained by country-pair

dummies. That is, unspecified characteristics of country pairs explain more than

the variables emphasized by the theory! This finding raises an obvious need to

broaden the theory to arrive at a better empirical specification.

Volume of Trade

Economies of scale and product differentiation strengthen the tendency to

specialize, and in turn, specialization encourages a greater volume of international

trade. To see how trade volumes are determined by specialization, consider an

extreme case in which every country is completely specialized in a subset of

products, each country has the same homothetic preferences for consuming goods,

and trade is balanced, so that spending levels on each good are proportional to

GDP levels (as in the fundamental consumption relationship given earlier). This

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

The Structure of Foreign Trade 137

relationship implies that a given percentage increase in the GDP of a country will

raise its volume of trade (exports plus imports) with any other country by the same

percentage amount; it is called a "gravity equation." 13 Gravity equations go back to Tinbergen (1962); Linnemann (1966) provided a comprehensive analysis. Typical

specifications estimate the effects of income on trade, controlling for the distance

between the trade partners. Distance has a negative effect on the volume of trade.

Equations of this form-aggregated in various ways-have been estimated

time and again, providing a good fit to many data sets. Helpman (1987) used it to

calculate the volume of trade amongst 14 industrial countries as a fraction of their

aggregate income. According to the theory, this fraction is larger the more similar

are the countries' income levels as measured by a similarity index 1 - lk(sk) 2.14 To see how this works as an income similarity index, take the case of two countries.

In this case the index is very close to zero whenever one country is much bigger

than the other, since sk is close to zero for the small country and close to one for

the large country. When the two countries are equal in size sk = 1/ for each, and the similarity index obtains the value 1/?, which is the largest possible value in the

presence of two countries. Helpman found that between 1956 and 1981, this index

increased and the fraction of income traded within the group increased as well.

This co-movement is consistent with models of product differentiation in which

specialization in production is driven by brand proliferation.

A reexamination of Helpman's (1987) evidence with a focus on bilateral trade

flows was performed by Hummels and Levinsohn (1995). They confirmed Help- man's finding for a group of industrial countries. But they also applied the same

approach to a group of mixed countries-some developed, others less developed-

arguing that if product differentiation is the main reason for the good fit, the

equation should not perform well in the mixed sample, since trade between

different economies should presumably be based more on factor differences and

less on trade in similar differentiated products. Although their main equation did

not fit the mixed sample as well as it did the homogeneous sample, it was remark-

ably good nevertheless. They concluded that the evidence does not lend support to

13Assuming that all countries have the same homothetic preferences and that there is complete specialization implies that country k imports from I the fraction sk of I's output. The volume of trade

(imports plus exports) between these two countries therefore equals QklI = skYl + slYk, where yk is the GDP level of country k. If, in addition, spending levels are proportional to GDP levels, Sk = yk/ y

and Qkl = 2YkYl/Y, where Yis the world's GDP. 14 To derive this similarity index recall from the previous footnote that country k's imports from country

I are Sk 1y, where Sk is the share of country k in world spending and Y' represents income of country 1. It follows that k's imports from all other countries equals sk( y - yk), where Ystands for world income.

Next note that assuming that spending is proportional to income implies sl' = Y/ Y. Therefore, the total volume of world trade is given by

E Sk(Y y*) = E (y y=) = 11 - E ( k)2] k k k

This implies that the volume of trade equals the fraction 1 - Sk(s5)2 of world income.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

138 Journal of Economic Perspectives

the view that product differentiation is the key to why the volume of trade seems to

increase with similarity in size of the trading economies.

Evenett and Keller (1998) developed an alternate estimation procedure that

sheds more direct light on whether product differentiation or differences in factor

endowments might explain the empirical relationships between the volume of

trade and the size distribution of countries. Using a mixed sample of countries

(similar to Hummels and Levinsohn), Evenett and Keller divided the observations

of country pairs into those that have less than 5 percent of intra-industry trade and

those that have more, and they assumed that country pairs in the first sample trade

homogeneous products. They then investigated whether the data favors a specifi-

cation in which differences in factor proportions between the countries in the first

sample are large enough to lead to specialization in this group, which would justify

a gravity equation (that is, a link between the volume of trade and the size of

countries) despite the lack of product differentiation, but failed to find evidence in

favor of this hypothesis.15 They also divided the remaining countries into five

classes that differed in the degree of intra-industry trade. They tested whether the

gravity equation provides a better approximation to the data for countries with a

higher share of intra-industry trade, and found that it does. This finding favors the

view that product differentiation is at the core of the relationship between the

volume of trade and the size distribution of countries.

As mentioned earlier, developed economies trade mostly with each other

rather than with less developed countries, and trade within the group of less

developed countries is only a small fraction of total world trade (about 15 percent).

To see how product differentiation helps to account for these facts, consider a

world that consists of developed countries (the North) and less developed countries

(the South). All countries in the North are the same size and have the same factor

endowments; the same is true of countries in the South. However, the composition

of factor endowments differs between the North and the South. In some sectors,

there is product differentiation, in others products are homogeneous. Now sup-

pose that due to differences in factor composition the North specializes in differ-

entiated products while the South specializes in homogeneous products. Since all

the developing countries are the same size, and producing the same homogenous

products, the countries of the South do not trade with each other. However, each

South country imports a fixed fraction of output from each North country. The

North countries all import some fixed fraction of output from every country in the

North and the same fraction from the South region.

Now, for concreteness, suppose that there are 20 countries in the North, each

15 Deardorff (1998) made the argument that a gravity equation can be derived from a generalized Heckscher-Ohlin model with large differences in factor composition that lead to specialization in

production. Although this is a theoretical possibility, the Evenett-Keller evidence suggests that it has little

empirical content. Deardorff's alternative derivation of the gravity equation for economies with homog-

enous products, which is based on random divisions of imports across the exporting countries, is not

even appealing on theoretical grounds. I agree with Grossman's (1998) position on this point.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

Elhanan Helpman 139

one accounting for 4 percent of world output. The North accounts therefore for 80

percent of the world's output and it is four times as large as the South. In this

setting, trade among the North economies will be a large proportion of world trade,

because the North economies are larger to begin with, and they specialize. With

these reasonable sizes of economies the volume of trade among the developed

countries in the North is about twice as large as their trade with the South.16 This is a simple calibration, close to the mark. In the data, this ratio is about 1.6.

Of course, assuming no trade at all between economies in the South is an

extreme condition. Surely, some internal trade does exist in the South, since factor

proportions are not the same in every country. But this extreme formulation helps

to make the main point. Generally there will be less intra-group trade in the South

the more the South specializes in homogeneous products and the less its countries

differ in factor compositions.

A Heckscher-Ohlin approach assuming homogenous products can also be

used to explain such volumes of North-North and North-South trade, without any

need for product differentiation, by incorporating any necessary differences in

techniques of production across goods and sectors as well as suitable cross-country

differences in factor endowments. Davis (1997) gives an example. However, while

I find such arguments to be an interesting intellectual exercise, they are of little

practical value. Explaining the observed trade patterns with only homogeneous

products requires too much fine tuning of the technology to be convincing. The

broad structure of world trade is more naturally explained with the aid of product

differentiation than without it.

To conclude, the available evidence supports the view that economies of scale

with product differentiation are valuable components in the explanation of trade

flows. However, it still leaves room for factor endowments to play a role.

Economies of Scale

Modern trade theory places significant weight on economies of scale and

product differentiation in explaining the structure of foreign trade. Its usefulness

has been gauged by a variety of implications that help to interpret various facts, and

by the fit of some of its implications in various data sets. However, this research

provides no direct evidence on the extent of economies of scale. In fact, most of the

implications that have been examined do not depend on the degree of economies of

scale, just on their existence. Although this is good enough for many purposes, for

a variety of welfare-related questions it is important to have a sense of the size of

economies of scale. Some micro production studies find economies of scale in

single-country data sets. But such studies cannot provide comparable estimates of

economies of scale at the sectoral level and may, therefore, underestimate the size

16 It is shown in Helpman (1998b) that under these assumptions the ratio of North-North to North- South trade equals SN( nN - 1) / 2(1 - sNnN), where SN is the relative size of a country in the North and qN is the number of countries in the North. Using SN = .04 and qN = 20, therefore implies a

ratio of 1.9.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

140 Journal of Economic Perspectives

of scale effects because, as suggested by Marshall, economies of scale at the sector

level can emanate from the interaction of firms and factor markets within an

industry.

Antweiler and Trefler (1997) have developed a methodology for estimating

returns to scale at the sectoral level from international data. They constructed a

data set for 71 countries, with 37 industries and 11 factors, spanning a period from

1972 to 1992. The key theoretical observation that enables them to estimate the

degree of economies of scale is that a variant of the Vanek (1968) equation, which

has been mentioned a number of times already, also holds in the presence of

economies of scale. The Vanek relationship, of course, is to look first at the factor

content of net exports, calculated by applying factor production coefficients to

import and export data, and then to compare that result to the difference between

total factors in an economy and factors used in domestic consumption. In the

Antweiler and Trefler variant of the Vanek approach, the factor content of exports

and imports is calculated separately for each and every trade partner, which allows

all countries to use different techniques of production.

One way to specify the difference in techniques of production is to allow

differences in factor productivity levels across countries, as in Trefler (1993). An

alternative approach, however, is to attribute cross-country variations in productiv-

ity levels to differences in scale. As a result of scale economies, the use of inputs per

unit output declines (on average) as output expands. Assuming that the produc-

tivity parameters are proportional to factor rewards, then scale economies are the

only remaining source of cross-country variations in productivity levels. By param-

eterizing the effects of scale economies, one can therefore estimate the suitable

coefficients from such modified Vanek equations. The results show that productiv-

ity is indeed higher the higher the level of output; costs fall by about 0.15 to 0.20

percent when output rises by 1 percent. These estimates are larger than those often

obtained from micro studies on economies of scale. Moreover, these estimates have

low standard errors. The estimates imply further that output expansion leads to

shifts in the techniques of production that raise the demand for skilled relative to

unskilled workers.

Some other estimates of economies of scale across national economies are

becoming available, as well. Harrigan (1999) uses estimates of production functions

for a sample of OECD countries to assess whether economies of scale at the sectoral

level or Hicks-neutral differences in the technological efficiency of countries better

explain cross-country variations in sectoral levels of total factor productivity. The

results are mixed. Differences in total factor productivity are large, and economies

of scale do not explain them adequately. Neutral technological differences provide

a somewhat better fit. Since one does not exclude the other, it would be interesting

to allow for both scale economies and technological differences, but Harrigan's

data does not allow estimating this decomposition accurately.

Yet another inquiry into economies of scale based on international data is

provided by Davis and Weinstein (1998b). Their point of departure is Krugman's

(1980) home-market effect. Krugman introduced transport costs for varieties of a

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

The Structure of Foreign Trade 141

differentiated product that are produced with economies of scale in a two-country

model with one input (labor) and two sectors, one differentiated and the other

homogeneous. In this sort of environment, where production functions are the

same in both countries, the pattern of specialization is undetermined in the

absence of transport costs. But with transport costs, the size of local demand

determines the profitability of manufacturing differentiated products. The larger

country, with its greater demand, supports the development of a disproportionate

number of brands and therefore exports them on net. The important role of the

size effect is here in demand, not supply.

More generally, in economies where transport costs are important and the size

of the home market matters, shifts in the aggregate demand for a country's

products will have a disproportionately large effect on its output of products with

economies of scale. In pure Heckscher-Ohlin-type economies (no economies of

scale and no product differentiation) with no transport costs, such effects are nil,

while in Heckscher-Ohlin-type economies with transport costs they are less than

proportional. It is therefore possible to discriminate between these alternative

models by estimating the effects of aggregate demand on the output of various

products.

To estimate the demand for goods in a particular sector of a country, Davis and

Weinstein (1998b) first estimate how trade flows are affected by the size of the

economy as compared to its trading partners, accounting for distances between

countries. Their sample consists of 22 OECD countries and 25 sectors. As is typical

for such equations, the trade volume declines with distance. Using these estimates

they then construct relative demand levels for every industry and estimate the effect

of this demand for the country's products, which emanates from all countries

including the home country, on the local supply. Pooling over industries, they find

that a 1 percent increase in demand raises local output by 1.6 percent, which

provides evidence in favor of scale effects. Performing the same estimation for every

industry separately they obtain less accurate estimates. Nevertheless, in 11 out of the

25 sectors the coefficients are significantly larger than one, supporting the presence

of economies of scale. Davis and Weinstein (1997) estimated a similar model on

regional data from Japan, and also found scale effects in a number of sectors.

Conclusions

Some researchers have argued that the modern theory of international trade

with economies of scale and product differentiation is not needed to explain key

features of world trade-such as the existence of intra-industry trade, the connec-

tion between the growth in economies and the growth of trade volumes, or the

difference between North-North and North-South trade volumes. Researchers in

this tradition strive to convince us that there is no need to incorporate economies

of scale or product differentiation into trade theory, since for any given pattern we

can find a structure of a traditional Heckscher-Ohlin model with homogenous

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

142 Journal of Economic Perspectives

goods that also produces it. This is, of course, a legitimate and intellectually

interesting debate. But in my view, product differentiation is so prevalent that it is

hard to see why it is even necessary to justify its presence in economic models. If

anything, it should be necessary to justify the use of models with homogeneous

products only. Just think about the products we consume: food, clothing, furniture,

home appliances, sports equipment, cars-almost anything-they all are differen-

tiated products. The same can be said about intermediate inputs and capital goods,

as well as about a host of services, such as banking, insurance, or travel. Whenever

product differentiation helps to explain certain phenomena, it is only natural to

embrace this assumption, rather than to insist on the construction of models with

homogenous products that also explain them.

All this being said, the ultimate test is in the direct and indirect evidence.

There is plenty of direct evidence, from observation and otherwise, that product

differentiation is prevalent. But this does not imply necessarily that the available

models of international trade with product differentiation provide a good expla-

nation of the data. It is fair to say that none of the existing models does a great job

explaining the data. But adding product differentiation does improve the fit

between theory and data. Moreover, since the inherent richness of the models with

product differentiation has not yet been much explored, such models carry the

potential of providing even better explanations.

We now have a rich theory of international trade that emphasizes economies

of scale, product differentiation and differences in factor composition as key

determinants of the structure of world trade. In combination, these factors explain

significant parts of specialization patterns, volumes of trade, factor content of trade,

and the broad patterns of trade across regions.

But despite the research effort in the last 20 years, these explanations are still

incomplete. This is partly the result of the fact that the nature of world trade is

changing rapidly. Technological change has modified the patterns of specializa-

tion, reduced trading costs and encouraged larger trade volumes; new countries

have joined the trading system; and multinational corporations have spread their

nets more widely than ever before. In the new economy there is plenty of man-

made comparative advantage and product cycles, which are occasionally short-lived.

We need a more technologically-oriented trade theory and more emphasis on

dynamics to understand these developments. Such theories have been elaborated

in the 1990s and they will undoubtedly affect future empirical research.

m This is an extended and elaborated version of the Bernhard-Harms Prize lecture, delivered

at the Kiel Institute for World Economics in June 1998. The original lecture was published in

Helpman (1998a). Most technical details are missingfrom this presentation. Many of them

can be found in my working paper, Helpman (1998b), which also contains more information

about recent studies. Many thanks to Don Davis, Bradford De Long, Gene Grossman, Ed

Leamer, Jim Levinsohn, Alan Krueger, Timothy Taylor, Manuel Trajtenberg and Dan Trefler

for comments on an earlier draft, toJaiho Chung for research assistance, to Jane Trahan for

editorial assistance, and to the National Science Foundation for financial support.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

Elhanan Helpman 143

References

Antweiler, W. and D. Trefler. 1997. "Increasing

Returns and All That: AView from Trade," mimeo.

Bowen, H. P., E. E. Leamer, and L. Sveikaus-

kas. 1987. "Multicountry, Multi-factor Tests of

the Factor Abundance Theory." American Eco-

nomic Review. December, 77:5, pp. 791-809.

Brecher, R. A. and E. V. Chaudhri. 1995. "The

Leontief Paradox Continued." Journal of Political

Economy. August, 90:4, pp. 820-23.

Davis, D. R. 1995. "Intra-Industry Trade: A

Heckscher-Ohlin-Ricardo Approach." Journal of

International Economics. November 39:34, pp.

201-26.

Davis, D. R. 1997. "Critical Evidence on Com-

parative Advantage? North-North Trade in a

Multilateral World." Journal of Political Economy.

Davis, D. R. and D. E. Weinstein. 1997. "Eco-

nomic Geography and Regional Production

Structure: An Empirical Investigation," mimeo.

Forthcoming in the European Economic Review.

Davis, D. R. and D. E. Weinstein. 1998a. "An

Account of Global Factor Trade." NBER Work-

ing Paper 6785.

Davis, D. R and D. E. Weinstein. 1998b. "Mar-

ket Access, Economic Geography and Compara-

tive Advantage: An Empirical Assessment," mimeo.

Davis, D. R., D. E. Weinstein, S. C. Bradford,

and K. Shimpo. 1997. "Using International and

Japanese Regional Data to Determine When the

Factor Abundance Theory of Trade Works."

American Economic Review. June, 87:3, pp. 421-46.

Deardorff, A. V. 1998. "Determinants of Bilat-

eral Trade: Does Gravity Work in a Neoclassical

World?" In The Regionalization of the World Econ-

omy. Frankel,J. A. ed. Chicago: The University of

Chicago Press.

Eaton, J. and S. Kortum. 1997. "Technology

and Bilateral Trade." NBER Working Paper No.

6253.

Ethier, W. J. 1984. "Higher Dimensional Is-

sues in Trade Theory," in Handbook of Interna-

tional Economics. Jones, R. W. and P. Kenen, eds.

Amsterdam: North Holland.

Evenett, S. J. and Keller, W. 1998. "On Theo-

ries Explaining the Success of the Gravity Equa-

tion," mimeo.

Gabaix, X. 1997. "The Factor Content of Trade: A Rejection of the Heckscher-Ohlin-Le-

ontief Hypothesis," mimeo.

Grossman, G. M. 1998. "Comment," in The

Regionalization of the World Economy. Frankel, J. A.,

ed. Chicago: The University of Chicago Press.

Grossman, G. M. and E. Helpman. 1991. In-

novation and Growth in the Global Economy. Cam-

bridge, MA: The MIT Press.

Grossman, G. M. and E. Helpman. 1995.

"Technology and Trade," in Handbook of Interna-

tional Economics, Vol. 3. Grossman, G. M. and K.

Rogoff, eds. Amsterdam: North Holland.

Grubel, H. and P. Lloyd. 1975. Intra-Industry

Trade: The Theory and Measurement of International

Trade in Differentiated Products. London: Macmillan.

Hakura, D. 1997. "A Test of the General Va-

lidity of the Heckscher-Ohlin Theorem for

Trade in the European Community," mimeo.

Harrigan, J. 1997. "Technology, Factor Sup- plies, and International Specialization: Estimat-

ing the Neoclassical Model." American Economic

Review. September, 87:4, pp. 475-94.

Harrigan, J. 1999 (forthcoming). "Estimation

of Cross-Country Differences in Industry Pro-

duction Functions." Journal of International Eco-

nomzcs.

Heckscher, E. F. 1919. "The Effect of Foreign

Trade on the Distribution of Income." Economisk

Tidskrzft. In Heckscher-Ohlin Trade Theory. Heck- scher, E. F. and B. Ohlin. Cambridge, MA: The

MIT Press, 1991 (translated, edited and intro-

duced by Harry Flam and M. June Flanders).

Helpman, E. 1984. "Increasing Returns, Im-

perfect Markets, and Trade Theory," in Hand-

book of International Economics, Vol. I. Jones, R. W.

and P. Kenen, eds. Amsterdam: North Holland.

Helpman, E. 1987. "Imperfect Competition and International Trade: Evidence from Four-

teen Industrial Countries." Journal of theJapanese

and International Economies. 1: 1, pp. 62-81.

Helpman, E. 1998a. "Explaining the Structure

of Foreign Trade: Where Do We Stand?"

Weltwirtschaftliches Archiv. 143:4, pp. 576-89. Helpman, E. 1998b. "The Structure of Foreign

Trade." NBER Working Paper No. 6752.

Helpman, E. and P. R. Krugman. 1985. Market

Structure and Foreign Trade. Cambridge, MA: The

MIT Press.

Hummels, D. and J. Levinsohn. 1995. "Mo-

nopolistic Competition and International

Trade: Reconsidering the Evidence." Quarterly

Journal of Economics. August, 110:3, pp. 799-836. Jones, R. W. 1956-57. "Factor Proportions and

the Heckscher-Ohlin Theorem." Review of Eco-

nomic Studies. 24:1, pp. 1-10.

Jones, R. W. 1965. "The Structure of Simple

General Equilibrium Models." Journal of Political

Economy. 73:6, pp. 557-72.

Krugman, P. R. 1980. "Scale Economies, Prod-

uct Differentiation, and the Pattern of Trade."

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms

144 Journal of Economic Perspectives

American Economic Review. December, 70:5, pp.

950-59.

Krugman, P. R. 1995. "Increasing Returns, Im-

perfect Competition and the Positive Theory of

International Trade," in Handbook of International

Economics, Vol. 3. Grossman, G. and K. Rogoff,

eds. Amsterdam: North Holland.

Leamer, E. E. 1980. "The Leontief Paradox,

Reconsidered." Journal of Political Economy. June,

88:3, pp. 495-503.

Leamer, E. E. 1984. Sources of International

Comparative Advantage. Cambridge, MA, and

London: The MIT Press.

Leontief, W. W. 1954. "Domestic Production

and Foreign Trade: The American Capital Posi-

tion Re-examined." Economica Internazionale. Feb

ruary, 7, pp. 9-36.

Lerner, A. 1952. "Factor Prices and Interna-

tional Trade." Economica. February, 19, pp. 1-15.

Linnemann, H. 1966. An Econometric Study of International Trade Flows. Amsterdam: North Hol-

land.

Loertscher, R. and F. Wolter. 1980. "Determi-

nants of Intra-Industry Trade: Among Countries

and Across Countries." Weltwirtschaftliches Archiv.

116:2, pp. 280-93.

McDougall, G. D. A. 1951. "British and Amer-

ican Exports: A Study Suggested by the Theory

of Comparative Costs, Part I." Economic Journal.

December, 61, pp. 697-724.

McDougall, G. D. A. 1952. "British and Amer-

ican Exports: A Study Suggested by the Theory

of Comparative Costs, Part II." Economic Journal.

September, 62, pp. 487-521.

OECD. 1996. Globalization of Industry. Paris.

Ohlin, B. 1924. "The Theory of Trade," in

Heckscher-Ohlin Trade Theory. Heckscher, E.F. and

B. Ohlin. Cambridge, MA: The MIT Press, 1991

(translated, edited and introduced by Harry

Flam and M. June Flanders).

O'Rourke, K. and J. G. Wflliamson. 1999

(forthcoming). Globalization and History: The Evo-

lution of a 19th Century Atlantic Economy. Cam-

bridge, MA: The MIT Press.

Reeve, T. A. 1998. Essays in International Trade,

Ph.D. dissertation, Harvard University.

Samuelson, P. A. 1948. "International Trade

and the Equalization of Factor Prices." Economic

Journal. June, 58, pp. 163-84.

Stem, R. M. 1962. "British and American Pro-

ductivity and Comparative Costs in International

Trade." Oxford Economic Papers.

Tinbergen, J. 1962. Shaping the World Economy:

Suggestions for an International Economic Policy.

New York: Twentieth Century Fund.

Trefler, D. 1993. "International Factor Price

Differences: Leontief was Right!" Journal of

Political Economy. December, 101:6, pp. 961-87.

Trefler, D. 1995. "The Case of the Missing

Trade and Other Mysteries." American Economic

Review. December, 85:5, pp. 1029-46.

Vanek, J. 1968. "Factor Proportions Theory:

The N-factor Case." Kyklos. 21, 749-54.

Williamson, J. G. 1998. "Globalization, Labor

Markets and Policy Backlash in the Past." Journal

of Economic Perspectives. Fall, 12:4, pp. 51-72.

WTO. 1997. Annual Report 1997, VoL II, France.

This content downloaded from 128.174.198.122 on Mon, 19 Aug 2019 19:15:17 UTC All use subject to https://about.jstor.org/terms