The Factor Content of Bilateral Trade: An

Empirical Test

Yong-Seok Choi
Korea Development Institute

Pravin Krishna
Brown University

The factor proportions model of international trade is one of the

most influential theories in international economics. Its central stand-
ing in this field has appropriately prompted, particularly recently,
intense empirical scrutiny. A substantial and growing body of empirical
work has tested the predictions of the theory on the net factor content
of a countrys trade with the rest of the world, usually under the
maintained assumptions of factor price equalization and identical
homothetic preferences across trading countries (or under quite spe-
cific relaxations of these assumptions). In contrast, this paper uses
OECD production and trade data to test the restrictions (derived by
Helpman) on the factor content of trade flows that hold even under
nonequalization of factor prices and in the absence of any assumptions
regarding consumer preferences. In a further contrast with most of
the existing literature, which has focused on the factor content of a
countrys multilateral trade, our tests concern bilateral trade flows,
thereby enabling the examination of trade flows between only a subset
of countries for which quality data (relatively speaking) are available.

We are most grateful to Steve Levitt and two anonymous referees for many detailed
comments on this paper. We are also grateful to Don Davis, Rob Feenstra, Ron Findlay,
Gordon Hanson, James Harrigan, Scott Taylor, Daniel Trefler, David Weinstein, and sem-
inar participants at the University of Chicago, Columbia University, and the National
Bureau of Economic Research for a number of helpful discussions and comments and to
Elena Urgelles for excellent research assistance. Choi thanks Brown University for pro-
viding its Stephen Robert Fellowship and to the Korea Development Institute, where
revisions to this paper were conducted. The views expressed here do not necessarily
represent those of the Korea Development Institute.

[Journal of Political Economy, 2004, vol. 112, no. 4]

2004 by The University of Chicago. All rights reserved. 0022-3808/2004/11204-0002$10.00

887
888 journal of political economy
We find that restrictions implied by the theory cannot be rejected for
the vast majority of country pairs considered in our analysis.

I. Introduction

The factor proportions model, which predicts that international trade

is driven by differences in factor endowments between countries, is one
of the most influential theories in international economics. In addition
to being used in the study of trade flows between countries, this model
has also served as a platform for innumerable academic and policy
analyses in international trade. They range from the study of the impact
of trade on income inequality within and between countries to the
analysis of the implications of foreign direct investment on welfare and
the impact of immigration on production patterns.
This central standing of the factor proportions model in international
economics has appropriately prompted, particularly recently, intense
empirical scrutiny.1 Researchers testing this framework have largely fo-
cused on an elegant prediction of the model relating to net factor
content of trade that obtains in even its multicountry, multifactor,
and multicommodity version: the well-known Heckscher-Ohlin-Vanek
(HOV) prediction. This holds that under the assumptions that tech-
nologies everywhere are identical, that trade equalizes factor prices
worldwide, and that consumer preferences everywhere are identical and
homothetic, the net exports of factors by a country will equal the abun-
dance of its endowment of these factors relative to the countrys world
income share. Early tests of the HOV prediction in its strict form, how-
ever, proved very disappointing for the theory: In a widely cited and
pioneering study, Bowen, Leamer, and Sveikauskas (1987) reported that
the magnitude and direction of net factor content flows among 27
countries were predicted extremely poorly by the theorya finding that
established a mood of deep pessimism with regard to the empirical
validity of the model.2
This apparent failure of the theory (in its strict form) to match the
data led researchers to amend the theory and to improve on the data

1
See Leamer and Levinsohn (1995), Trefler (1995), Helpman (1998), and Davis and
Weinstein (2003) for comprehensive discussions.
2
Other trade-related predictions of the factor proportions theory did not fare much
better: In a very well-known contribution, Leontief (1953) used data on the factor content
of U.S. exportables and importables to find paradoxically that the former used more
labor relative to capital than the latter in its production, thus rejecting the central pre-
diction of the factor proportions modelthat countries export goods that use their abun-
dant factors more intensively.
bilateral trade 889
3
used in the empirical exercises. In a series of remarkable contributions,
Trefler (1993, 1995) and Davis and Weinstein (2001) variously attempted
particular modifications (some systematic and some ad hoc) of the basic
HOV assumptions and tested the resulting predictions to find much
stronger support for the theory. Thus Trefler (1995) reported that a
variation of the model that postulated Hicks-neutral factor efficiency
differences across country groups performed very well against the stan-
dard HOV prediction. And Davis and Weinstein (2001) articulated a
series of additional departures from the basic HOV framework, includ-
ing the use of bilateral trade estimates from the so-called gravity equa-
tions (themselves valid under the further assumptions of perfect spe-
cialization in tradables and specific assumptions on preferences) to
account for the role of trade costs in restricting trade, to also report
much stronger support for the theory.
Our paper contributes to this literature on empirical testing of the
factor proportions theory. Our methodology contrasts strongly with most
earlier work, however. Nearly all the tests of the factor content predic-
tions of the model (including the ones we have discussed above) have
assumed full factor price equalization (FPE) across countries and iden-
tical homothetic preferences across countries (i.e., they have tested the
HOV prediction) or have attempted very specific relaxations of these
joint assumptionsfor instance, by allowing for factor price differences
to result from Hicks-neutral factor efficiency differences across coun-
tries, as in Trefler (1995). In contrast, this paper implements a test of
restrictions implied by the theory (derived originally by Helpman
[1984]) on the factor content of trade that relies neither on FPE nor
on any restrictions on preferences. We consider this to be a significant
step because, as Helpman (1998) has noted, even casual evidence sug-
gests that full FPE does not hold (as we know from data on wages) and
that preferences are nonhomothetic and vary substantially with income
level. A further and equally important contrast with the existing liter-
ature derives from the fact that while most empirical tests of the theory
(and tests of HOV in particular) have focused on the net factor content
of a countrys multilateral trade, our tests concern bilateral trade flows,
thereby enabling the examination of trade flows between only a subset
of countries for which quality data (relatively speaking) are available.4
Helpmans (1984) result, itself an intuitive (and general) formaliza-
3
Also, a growing literature has examined other aspects and predictions of the neoclas-
sical trade model: Prominent recent contributions include Harrigan (1995, 1997), Bern-
stein and Weinstein (2002), Hanson and Slaughter (2002), Debaere (2003), and Schott
(2003), among others.
4
See, however, the earlier work of Brecher and Choudhri (1993), which does undertake
a bilateral analysis somewhat similar to the one conducted herealthough just between
the United States and Canada (and on an industry by industry basis).
890 journal of political economy
tion of important earlier work by Brecher and Choudhri (1982), is both
straightforward and powerful: even in the absence of FPE, with identical
technologies across countries, it is a simple matter to observe that the
more capital-rich a country is, the more capital and less labor it uses in
all lines of production, while correspondingly achieving a higher wage-
rental ratio. Hence, whatever trade exists between two countries, exports
of the capital-rich country will embody a higher capital-labor ratio than
the exports of the relatively labor-rich country. This, in turn, describes
a clear bilateral factor content of trade. Specifically, the theory implies
that, on average, a country imports those factors that are cheaper in
the partner country and is a net exporter of those factors that are more
expensive there. It is this description that we test using data on OECD
production and trade flows.5
Our results are as follows: The restrictions implied by the theory for
overall (i.e., bidirectional) bilateral trade flows are satisfied for the vast
majority of country pairs in our sample. Having said this, we must note
that in many cases, the theory is only just satisfied. Nevertheless, this
finding stands in strong contrast with many previous tests of the theory
that were conducted under the restrictive assumptions of identical factor
prices and identical homothetic preferences across countries in which
the theory fared very poorlyas was reported in the early tests of Bowen
et al. (1987) and confirmed in more recent implementation of these
tests using the country-specific data (which are also used in the present
study) by Davis and Weinstein (2001), among others. Thus our finding
that the theory, when tested without the imposition of these restrictions,
is not rejected by the data is a significant one. Our results are robust
to a wide variety of methods used to measure factor prices. Allowing
for even a small degree of measurement error in factor prices brings
greater success to the theory. In some configurations, the data are
unable to reject the null that the theory is right in 100 percent of the
cases (i.e., country pairs).
The rest of the paper is structured as follows: Section II presents
Helpmans (1984) basic result regarding restrictions on bilateral trade
flows, incorporating additionally into the analysis the use of interme-
diates in production. We discuss the advantages and disadvantages of
testing these restrictions over standard HOV tests. Section III describes
the data. Section IV describes our empirical analysis and the results.
Section V presents concluding remarks. Appendix A provides a detailed
description of data sources and construction. Appendix B discusses ex-
5
It is worth noting that the theoretical restrictions that we test here are easily extended
to accommodate the possibility of technological differences (aggregate Hicks-neutral dif-
ferences and industry-specific, i.e., Ricardian, differences) across countries. We discuss this
extension in App. B.
bilateral trade 891

Fig. 1.Lerner diagram

tensions to take account of Hicks-neutral and Ricardian technological

differences across countries.

II. Theory
Our analysis considers a freely trading world with many goods and coun-
tries in which production technology is convex, the technology for pro-
ducing any good is assumed (for now) identical across countries, and
perfect competition characterizes both goods and factor markets.
In this framework, as we have noted before, Helpman (1984), building
on the work of Brecher and Choudhri (1982), derived intuitive restric-
tions on the factor content of bilateral trade between countriesrelating
factor content of trade to relative factor scarcities in the trading coun-
tries. The basic insight behind Helpmans result can be easily explained
using a Lerner diagram. Figure 1 depicts a Lerner diagram for the two-
factor/six-good/three-country case.
The isoquants in figure 1, numbered from 1 to 6, describe output
levels of goods 16, respectively, each worth a dollar at free-trade prices.
The factors used in the production of these goods are capital and labor.
The capital-labor ratios of the three countries are represented by the
rays (K/L)c, and their free-trade wage-rental ratios are represented by
the slopes qc, c p 1, 2, 3. In the equilibrium described above, country
1, which has the highest capital-labor ratio, produces goods 1 and 2;
country 2, with an intermediate capital-labor ratio, produces goods 3
892 journal of political economy
and 4; and country 3, with the lowest capital-labor ratio, produces goods
5 and 6. It is a simple matter then to observe that the more capital-rich
a country is, the more capital and less labor it uses per dollar of output
in all lines of production. Hence, whatever trade takes place between
any two countries, the exports of the relatively capital-rich country will
embody a higher capital-labor ratio than the exports of the relatively
labor-rich country. This in turn describes a clear bilateral factor content
pattern of trade even in the absence of factor price equalization and
any assumption regarding preferences.
In what follows, we present Helpmans (1984) result allowing addi-
tionally for the presence of intermediate goods in production. It is worth
noting that, even under the maintained assumption of identical tech-
nologies across countries, nonequalization of factor prices will still result
in the use of different techniques of production across countries. We
denote the direct input matrix, which indicates how much direct input
of each factor is required to produce one dollar of gross output within
each industry, for any country c, by Bc. The input-output matrix for
country c, indicating the amount of output each industry must buy from
other industries to produce one dollar of its gross output, Y c, is denoted
by Ac. For any country c, the trade vector (T c) is the difference between
net production (Qc) and consumption (Cc):

T c p Qc Cc. (1)

In the presence of intermediates in production, we have net and gross

output related as

Qc p (I Ac)Y c. (2)

Then Bc(I Ac)1 is the matrix of total (direct and indirect) factor inputs
required to produce one dollar of net output in each industry (i.e., it
is the overall technology matrix in the presence of intermediate goods)
in country c. This matrix can also be used to determine the factor
content of exports by c. Thus consider bilateral trade between two coun-


tries, with T c c denoting the gross import vector by country c
from


country c and TVc c denoting the corresponding gross import vector of
factor content by country c
from country c. Given the gross import


vector, T c c, and the technology matrix employed in c, Bc(I Ac)1, the


corresponding factor content, TVc c, can be written as



TVc c p Bc(I Ac)1 T c c. (3)

Now, by the definition of the revenue function, we know that, for

country c
, P(p, V c ) p pQc , where P is a revenue function representing


(convex) production technology, V c is the endowment vector in country
bilateral trade 893


c , and p is the commodity price vector in the free-trade equilibrium.6
Given the assumption of identical technologies across countries, it
should be clear that if country c
is given its gross import of factor content


(TVc c) as an extra amount of factor endowment, it could produce with


it at least the value of gross imports (pT c c). This and the concavity of
P in V (used to arrive at the second inequality in what follows) give us
that





p(Qc T c c) P(p, V c TVc c)




P(p, V c ) PV (p, V c )TVc c




p pQc w c TVc c, (4)
where PV is the vector of partial derivatives of P with respect to V, and


w c is the factor price vector in country c
.


Eliminating pQc from both sides of (4) in turn gives us




pT c c w c TVc c. (5)
Further, in country c, since perfect competition implies that every line
of production must break even in equilibrium, we have



pT c c p w c TVc c. (6)
Combining (5) and (6) yields the following inequality:



(w c w c)TVc c 0. (7)
Similarly, for cs imports, we have



(w c w c)TVcc 0. (8)
Equations (7) and (8) together yield




(w c w c)(TVc c TVcc ) 0. (9)
As Helpman (1984) has pointed out, (9) may be interpreted as im-
plying that, on average, country c
is a net importer from country c of
the content of those factors of production that are cheaper in c than
in c
and vice versa.7 It should be readily evident that all the variables
in (9) relate to the equilibrium with trade. Inequality (9) may therefore
6
Note that our assumption of identical production functions across countries implies
that the revenue function is also common across countries (and we therefore have no
country superscript for the revenue function).
7
It is tempting to interpret (9) as a measure of the savings in production costs in country


c
because the gross import vector T c c is imported rather than domestically produced
(measured at equilibrium factor prices in the domestic country). This is incorrect, however.
The cost savings from importing rather than producing domestically, crudely speaking,
require a comparison of autarky equilibria with equilibria with trade. This is not what is
being compared in (9).
894 journal of political economy
be tested using data from the trade equilibria that we observe. This
is precisely what the rest of our analysis attempts to do. In implementing
tests of (9), one needs to take into account the important observation
of Staiger (1986) that when intermediates are freely traded, Helpmans
measure of the bilateral factor content of trade needs to be modified
to exclude the factor content of traded intermediate goods. Therefore,
we perform the tests described above using input-output matrices that
include only domestically produced intermediates.8
Our discussion so far has assumed identical technologies across coun-
tries. It is worth noting here that a relationship quite close to (9) may
be easily derived even if technologies are not identical across countries.
Consider the case in which technologies are instead characterized by
Hicks-neutral differences across countries, where country c is uniformly
more productive than country c
in the production of every good by a
(potentially measurable) factor l. The logic underlying the derivation
of (4) still holdswith the difference that if country c
is given its gross


import of factor content (TVc c) as an extra amount of factor endowment,


it could produce with it at least the value of gross imports (pT c c) times
the ratio 1/l. Equations (5)(9), mutatis mutandis, are then easily de-
rived. Alternatively, we may allow for Ricardian differences in technology
across countries, where technology in industry i in c is more productive
than the same industry i in c
by a factor of l i. Expressions analogous
to (9) (now involving the full set of l is for every industry) may be easily
derived. While we do not implement here tests of the theory with tech-
nological differences of the sort discussed above, we develop these ex-
pressions in detail in Appendix B in order to demonstrate the flexibility
of the framework being used here.
We have derived here theoretical restrictions on the factor content
of bilateral trade flows that may be tested using observable data. These
tests offer some significant advantages over the HOV-based tests that
currently dominate the literature but also suffer from some disadvan-
tages. The primary advantages are that the restrictions that we have
derived do not require that factor prices be equalized across countries
and do not require any assumptions on consumer preferences. Both of
these are significant relaxations of the theoretical assumptions under
which most HOV-based testing of the factor proportions model has been
conducted (from both a theoretical and an empirical perspective, as we
have previously discussed). The focus here on bilateral trade flows also
enables the examination of trade flows between only a subset of coun-
tries for which quality data (relatively speaking) are available. The dis-
8
See App. A for a detailed discussion and a simple example illustrating the need for
the modification of Helpmans measure as suggested by Staiger (1986). On the issue of
the proper measurement of the factor content of trade in the presence of intermediates,
see also Antweiler and Trefler (2002).
bilateral trade 895
advantages of the tests proposed here, on the other hand, are as follows:
While HOV-based tests provide exact predictions regarding the factor
content of trade in each factor, our tests provide only a statement re-
garding the direction and magnitude of the flow of factors, on average.
Further, while HOV tests require information on trade and technology
from the entire trading world, they permit us to focus on only those
factors in which we are interested or on which we have data. In contrast,
the tests proposed here require information on all factors of production.
Thus the tests conducted here offer some significant theoretical and
implementation advantages over HOV tests but are also inferior to HOV
tests in some respects. The two approaches should largely be seen as
complements.

III. Data
The countries we consider in this study are Canada, Denmark, France,
Germany, Korea, Netherlands, the United Kingdom, and the United
States. In order to test the restrictions (7)(9) for any pair of these
countries, we need data on the factor price vector (w), the direct input
matrix (B), and the input-output matrix (A) for each country in the
pair, as well as the gross bilateral import vectors (T) that describe trade
flows between them.

A. Technology
Most previous work that implemented tests of the factor proportions
theory has generally assumed (and used) the same technology matrices
(A and B) across countries (usually U.S. technology matrices) in order
to calculate the factor content of trade of any countrymostly because
of the general difficulty of obtaining the relevant data for a cross section
of countries at any given time.9 Under the maintained assumptions of
FPE as well as identical technologies across countries, the use of the
same technology matrices to represent production in different countries
does not create any problems at the theoretical level. In contrast, be-
cause we choose to abandon the assumption of FPE, we are forced to
confront the fact that, at the theoretical level itself, different technology
matrices across countries are implied even under the maintained as-
sumption of identical technologies across countries. To this end, this
study has required the collection of technology data on both the direct
input matrices and the input-output matrices for each country. As noted
9
Some exceptions may be noted: Trefler (1993), while assuming that the U.S. technology
matrix was basically valid for all countries, rescaled each by a country-specific productivity
parameter. Hakura (2001) used country-specific direct input matrices as well as input-
output matrices, as Davis and Weinstein (2001) did.
896 journal of political economy
earlier, taking trade in intermediates into account implies that we need
to use input-output matrices that include only the usage of domestically
produced intermediates, since Helpmans measure of the bilateral factor
content of trade needs to be modified to exclude the factor content of
traded intermediate goods (as Staiger [1986] has pointed out). Details
on the relevant technology matrices that we used are provided in Ap-
pendix A.

B. Factor Prices

For the purposes of empirical implementation, production technology

was assumed to admit two types of primary input factors: capital and
(disaggregated) labor. The factor price data that we used in this paper
were put together from a variety of sources. Details on the original data
sources and our processing of these data in order to arrive at interna-
tionally comparable factor price vectors and variation in this method-
ology to ensure the robustness of our results are described below (with
some additional details provided in App. A).
When we compiled the data for our analysis, one issue that arose was
the lack of availability of internationally comparable data on factor
prices. A second and equally compelling problem was that the factor
price data that were reported were sometimes inconsistent with gross
domestic product data (i.e., the inner product of factor prices and the
factor endowment vector do not sum to GDP).
Our strategy in dealing with these problems was to collect factor price
data from various sources that were perhaps not directly comparable in
the first instance and then to process them so as to get comparability
across nations and a match with GDP data. This was achieved as follows:
the Annual National Account database of the OECD provides data on
cost components of GDP, with GDP decomposed into the following
terms: compensation of employees, operating surplus, and an aggregate
of other components such as indirect taxes and subsidies. To achieve
consistency of the factor price data with national income accounts, we
started first with returns to aggregates (of labor and capital) and then
moved on to disaggregated returns. Thus, to begin with, we require that
the total return to labor in any country be equal to its compensation
of employees; that is, we set compensation equal to i wi L i, where the
summation is across disaggregated labor categories (described in greater
detail below).
To determine the total return to capital, we have two options: the
first (henceforth referred to as the Capital I method) is to let the op-
erating surplus equal the ex post return to capital in the economy (i.e.,
bilateral trade 897
10
to set operating surplus equal to rK). A second option (henceforth
referred to as the Capital II method) is to let

GDP n

ip1
wi L i p rK,

that is, to let the return to capital equal the residual when employee
compensation is taken out of GDP. To ensure robustness, we perform
our tests using both methods for calculation of the total return to capital.
Given the overall compensation to labor (i wi L i) and the overall
return to capital, we need next the returns to disaggregated labor. This
was accomplished in the following manner. Endowments of labor in
various occupations (L i) and the occupational wage rates (wi) were di-
rectly obtained from various national statistical publications for three
non-European countries and from Eurostats Structure of Earnings
(SOE) for the five European countries in our data set. There are two
problems with using these data directly. First, there is the issue of overall
consistency with the national income accounts because the value of
i wi L i rarely equals the employee compensation data reported in the
national income accounts. In order to achieve this consistency, we con-
struct a modified series of wage rate data as follows. Given the observed
data on occupational wage rate (wi), occupational employment levels
(L i), and compensation of employees, we calculated the modified wage
rate (w i) for each occupation by solving

w i L i p employee compensation,
ip1

wi w i
p , Gi, j n.
wj w j
That is, we took the information about the wage ratios between occu-
pations from the reported wage series wi and made the sum of con-
structed wage rates multiplied by occupational employment levels con-
sistent with the measure of compensation of employees in the national
accounts database.
A second issue had to do with comparability of labor classes across
countries. Publications for different countries use different occupational
classification systems.11 Thus some recategorization of occupational clas-
sifications was inevitable. Data for each of the three non-European coun-
tries (Korea, Canada, and the United States) were reported in a manner
conforming closely to what is referred to as the Industrial Standard
10
To set operating surplus equal to rK requires a strong zero-profit assumption because,
in general, the operating surplus contains other components, such as profit, as well.
11
For details on publication sources, see App. A.
898 journal of political economy
Classification of Occupations (ISCO) 1968 system. However, the occu-
pational classifications of European countries in their structure of earn-
ings data (as reported in Eurostat) were quite different from those of
the non-European countries and could not have been recategorized
easily into the ISCO 1968 system. Also, these data were at a substantially
higher level of aggregation than the data for the non-European coun-
tries. We considered two types of recategorization. The first was simply
to divide the labor force for all countries into production workers and
nonproduction workers (henceforth Euro I categorization). The other
one was to disaggregate the nonproduction workers into three cate-
gories: managerial, clerical, and others (henceforth Euro II catego-
rization).12
Overall then, we have two measures of returns to capital and two
classifications of labor. The Capital I measure is the reported operating
surplus of the economy per unit of capital (and, as measured, is net of
taxes), and the Capital II measure (perhaps theoretically more appro-
priate) is the residual when employee compensation is taken out of
GDP per unit of capital (and is gross of taxes). Labor is disaggregated
at two levels: Euro I separates labor into production and nonproduction
workers. Euro II separates workers into production workers, managerial
workers, clerical workers, and other workers. These factor prices, for
the countries in our sample, are reported in table 1. Wages for both
labor classificationsthe Euro I and Euro II classifications described
aboveare presented in panel A. As can be seen from a comparison,
say, of U.S. and German wages, there is a reasonable degree of diver-
gence between even the OECD countries used in our analysis. Indeed,
the wage gap between Korea and the rest of the OECD is extremely
large, as the figures presented in table 1 indicate. As we have discussed
before, we have used primarily two measures of return to capital. Our
first measure of the rental price of capital (Capital I method), as we
previously discussed, was obtained by dividing the operating surplus by
net capital stock. Panel B in table 1 reports the rental price of capital
calculated in this way for each country. Denmark has the lowest rental
price of capital (5.3 percent), whereas that for the United States is a
bit higher (8 percent) and that for Korea is the highest (15.5 percent).
Our second measure of the return to capital (Capital II method) was
obtained by taking the net return to capital to be the difference between
GDP and employee compensation and dividing this number by the net

12
Appendix table A2 describes the labor categories in greater detail. As in the ISCO
1968 (for non-European countries) and the SOE for European countries, nonproduction
workers in both the manufacturing and nonmanufacturing sectors include managerial,
clerical, and other workers (where other includes assistants, supervisors, accountants,
managers, and salespeople, e.g.). Production workers include all manual workers in the
SOE and all workers classified as production, service, and agricultural workers in the ISCO.
bilateral trade 899

TABLE 1
Factor Prices
United United
Category States Canada Denmark France Germany Kingdom Netherlands Korea
A. Labor (in U.S. Dollars)
Euro I:
Production 13,059 12,592 13,137 14,141 17,151 12,327 17,423 1,638
Nonproduction 20,375 15,657 16,878 23,290 23,496 13,510 23,886 2,822
Euro II:
Production 13,059 12,592 13,333 14,715 18,789 12,595 18,177 1,638
Managerial 26,589 21,165 24,985 40,855 34,011 21,011 36,670 7,189
Clerical 14,869 11,460 17,313 16,221 16,389 9,323 18,363 2,910
Others 21,578 16,960 15,788 22,859 24,544 14,529 25,083 2,495
B. Capital
Capital I .080 .103 .053 .078 .091 .075 .097 .155
Capital II .165 .190 .174 .180 .203 .203 .185 .234
Note.For labor, the factor price figures presented denote average annual compensation in U.S. dollars to an
employee of the designated type. See App. table A2 for a detailed breakdown of labor categories. For capital, the factor
price denotes the rate of return. Rates of return were calculated as follows. Capital I method: operating surplus/K;
Capital II method: (GDP minus compensation to employees)/K, where K denotes net capital stock.

capital stock. This measure of return to capital, consistent with an overall

division of GDP into rewards to labor and capital, is also reported in
panel B of table 1. By the Capital II method, the return to U.S. capital,
for instance, is 16.5 percent and the return to capital in Korea is 23.37
percent. Since the Capital I measure is net of taxes on production (from
the definition of operating surplus) and the Capital II measure is gross
of indirect taxes, the Capital I measure can be expected to be lower
than the Capital II measure of return to capital. This can be seen from
our calculations as well.13

IV. Results

Tests of our basic restriction on the factor content of bilateral trade

flows, equation (9), can be conducted using the factor price data and
the country-specific technology matrices whose construction we have
described in the previous section. Since entering technology and factor
price data into the left-hand side of (9) would simply give us an un-
normalized numerical sum, whose extent of conformance or departure

13
Note that, as we may expect in a world with some degree of capital mobility, the net
of taxes measure of return to capital, the Capital I measure, is closer across countries than
the gross of taxes measure, the Capital II measure.
900 journal of political economy
from the theory cannot be easily ascertained,14 we first rewrite (9) in
the following manner:




w c TVc c w c TVcc



{ v 1. (10)
w c TVc c w c TVcc

Equation (10) has a convenient interpretation. For any country pair,

c and c
, with gross bilateral import flows, T c c and T cc , the ratio in (10)
is the ratio of the sum of the importers (hypothetical) cost of produc-
tion (using the importers factor prices and exporters factor usage) to
the total (actual) cost of production in the exporting countries (i.e.,
using the actual producers factor prices and factor usage). Thus the



first term in the numerator of the ratio in (10), w c TVc c, is the hypothetical


cost of production of the gross import vector of c
from c, T c c, using

c

the factor prices in c , w , and the factor content actually employed in


production of this import vector in the exporting country c, TVc c. The
cost of producing these goods in the exporting country, c, is given by


the first term in the denominator of the ratio in (10), w c TVc c. The second
terms in the numerator and the denominator relate to the trade flow


T cc , the gross import vector of c from c
, and are equal to the hypo-
thetical cost of production in the importer of that flow c and the actual
cost of production in the exporter c
, respectively. We denote this ratio
of costs as v. Clearly, from (9) and (10) above, the theory predicts that
v 1. Importantly (and this is what has motivated our transition from
[9] to [10]), given the relative cost interpretation for v that we have
provided above, actual measures of v for any country pair will give us
an intuitive sense of the extent of the datas conformance to or depar-
ture from the theory for those countries.15
We describe first the values of v obtained using the raw factor price
measures reported in table 1. Results from additional simulation-based
analyses that were conducted to take into account the fact that our
factor price measures may be subject to measurement error are de-
scribed subsequently. The values of v calculated using the Euro I and
Euro II labor classifications and the Capital I measure of return to capital
are presented in tables 2 and 3, respectively. Values calculated using the
Capital II measure of return to capital instead are presented in tables
4 and 5.

14
For instance, if for a given country pair we were to obtain that the left-hand side of
(9) added up to 90,000, we would be able to conclude that the theoretical restriction
that the left-hand side be greater than zero had not been met, but would be unable to
tell how significant a departure this is from the theory.
15
Thus, e.g., if the calculated value of v were to work out to be 0.5 in the case of a
given country pair, this would be a strong violation of the theory, since this would imply
that, on average, costs could be 50 percent lower if domestic production were substituted
for bilateral imports.
bilateral trade 901

TABLE 2
Values of v with Euro I and Capital I Measures

United
Canada Denmark France Germany Kingdom Netherlands Korea
United States .99 1.00 1.03 1.01 .98 1.16 1.95
Canada 1.06 1.01 .99 .97 1.12 1.83
Denmark 1.07 .99 1.04 1.03 2.76
France .99 1.04 1.03 3.00
Germany .97 1.01 2.70
United Kingdom 1.10 2.11
Netherlands 4.04

TABLE 3
Values of v with Euro II and Capital I Measures

United
Canada Denmark France Germany Kingdom Netherlands Korea
United States .99 1.02 1.05 1.01 .98 1.18 1.92
Canada 1.05 1.02 .99 .97 1.14 1.81
Denmark 1.07 .99 1.03 1.04 2.72
France .99 1.04 1.03 2.98
Germany .97 1.00 2.76
United Kingdom 1.11 2.10
Netherlands 4.08

Consider the results presented in table 2 with the Euro I and Capital
I factor price measures. Keeping in mind the theoretical prediction that
v 1, we can see that the theory is satisfied directly for 21 of the 28
country pairs in our sample. Note that even for the seven pairs for which
the theory is not satisfied, v falls below 0.99 in only three cases. Table
3 presents values of v calculated using Euro II and Capital I factor prices.
The move from the Euro I classification to the more disaggregated Euro
II classification does not seem to affect the results by much. The success
rate for the theory stays about the same. Twenty-one of the 28 country
pairs satisfy the theory directly. Of the seven remaining pairs, only three
fall below 0.99. Values of v calculated using Capital II factor prices and
the Euro I labor classification are presented in table 4. As the numbers
presented there indicate, there is now a slight improvement in the extent
to which the data are consistent with the theory. Specifically, 22 of the
28 country pairs in our sample now satisfy the theory. Of the six re-
maining pairs, none fall below 0.99. Values with the Capital II and Euro
II measures in table 5 are even more supportive of the theory. Twenty-
902 journal of political economy

TABLE 4
Values of v with Euro I and Capital II Measures

United
Canada Denmark France Germany Kingdom Netherlands Korea
United States .99 .99 1.05 1.02 1.02 1.12 1.69
Canada 1.01 1.02 1.00 1.01 1.09 1.60
Denmark 1.02 .99 1.04 1.01 2.23
France .99 1.04 1.02 2.52
Germany .99 .99 2.30
United Kingdom 1.07 1.86
Netherlands 3.39

TABLE 5
Values of v with Euro II and Capital II Measures

United
Canada Denmark France Germany Kingdom Netherlands Korea
United States 1.00 1.00 1.06 1.02 1.03 1.14 1.67
Canada 1.01 1.03 1.00 1.01 1.10 1.59
Denmark 1.02 .99 1.03 1.02 2.22
France .99 1.03 1.02 2.51
Germany .99 .99 2.36
United Kingdom 1.08 1.85
Netherlands 3.43

four of the 28 country pairs directly satisfy the theory. Of the rest, none
fall below 0.99.16
The raw values of v and the number of cases for which these values
16
A previous version of this paper also reported measures of the left-hand sides of (7)
and (8), i.e., unnormalized measures of cost differences for imports in each direction for
each bilateral pair of countries. We found that when we considered import flows in each
direction separately, the fraction of cases for which the theoretical restriction is met is
smaller (although still greater than 50 percent). However, for tests of the theoretical idea
that cheaper factors are exported by countries, on average, testing (9) (and [10]) is more
appropriate than testing (7) and (8) individually. The reason is that considering (7) and
(8) separately does not allow for cost differences across countries to be weighted by the
volume of trade (as can be seen from the fact that in testing [7], dividing the left-hand


side of [7] by the trade flow, T c c, or indeed any other scalar does not change the test).
To see why this matters more clearly, consider the following example. Consider a country
that has a cost disadvantage in nearly all industries because of high factor prices relative
to those in a particular partner. Consider further that it exports an infinitesimal amount
to its partner but that bilateral trade between the partners consists nearly entirely of its
imports from the low-cost partner. Here, trade patterns reflect our theoretical intuition:
goods flow from low-cost suppliers to high-cost ones, and factors are exported from the
country in which they are cheaper. However, considering (7) and (8) separately may give
us only a 50 percent success rate for the theory (since imports by the low-cost supplier
[even if infinitesimally small in volume] will violate the theory and exports by the low-
cost supplier will be consistent with the theory). On the other hand, (9) and (10) weight
cost differences by trade flows and would find the theory to be validated. We are therefore
now convinced that (9) provides a more appropriate test of the theory. We are grateful
for discussions with Don Davis and Rob Feenstra on this point.
bilateral trade 903
exceed one are indicative of the degree of success of the theory. A
more formal analysis requires us to take into account the fact that our
calculations of v are subject to stochastic errors and that some assump-
tions about these errors are needed to interpret the results above. One
possibility is to assume a stochastic model in which the estimated value
of the statistic v equals its true value plus an error term that is sym-
metrically and independently distributed with zero mean. The proba-
bility that the value of the statistic exceeds one is .5 under the null
hypothesis and greater than .5 under the alternative (that v 1 1). When
the normal approximation to the binomial distribution (with 28 obser-
vations and with probability of success in any given trial of .5) is used,
the probability of finding 21 or more cases with the value of the statistic
above unity (as we find in our analysis of the data using the Capital I
measure) is .007 and that of finding 22 or more cases to be greater
than one (as we at least do with our use of the Capital II measure) is
.002. Thus the results reported in tables 2, 3, 4, and 5 reject the null
hypothesis that the true value of the statistic is equal to one (against
the alternative that it is greater than one) at even the 1 percent level.17
It should be fully recognized that the findings reported here stand
in strong contrast to previous tests of the theory that were conducted
under the restrictive assumptions of identical factor prices and identical
homothetic preferences across countries. That the theory fared very
poorly when tested under these maintained assumptions is widely
known, having been confirmed in the early tests of Bowen et al. (1987)
and also in more recent implementation of these tests using the country-
specific data (which are also used in the present study) by Davis and
Weinstein (2001), among others. Thus our finding that tests of the
theory that do not impose these restrictions are not rejected by the data
is a significant one.
A point regarding the magnitude of the calculated vs is worth noting:
While they are greater than one in most cases and while the theory is
therefore not met with rejection in the data, it can easily be seen from
even a cursory examination of the results that, in our data, exporters
costs do not seem much lower than importers costs of production (as
reflected in values of v not much greater than one in most cases). How
is this to be understood? What magnitude of v should we expect to see
in the first place? And should one infer that values of v close to one
reflect near equalization of factor prices among the countries in our
sample?

17
See also Brecher and Choudhri (1993) for a similar analysis. As with their tests, this
procedure is subject to the criticism that it assumes v to be drawn from the same distri-
bution for all country pairs. Thus our argument here should be thought of as being only
a tentative one, but one that is nevertheless suggestive of the degree of success of the
theory.
904 journal of political economy
It is worth recalling that the present theoretical framework offers no
further insight into what the value of v ought to be other than to require
it to be greater than one. Nevertheless, given that v simply measures
ratios of production costs, one may imagine that the literature offers
priors on what values of v one should expect to see. This is, however,
not immediately the case. First, it should be recognized that the ratio
v is not a measure of autarky production cost differences between coun-
tries. Rather, it reflects differences in production costs in a trading
equilibrium, which, given the tendency toward equalization of factor
prices through trade, and even in the absence of full equalization of
factor prices, can be reasonably expected to be smaller than any measure
of differences in production costs in autarky. The academic literature,
to date, provides us with few priors on the extent of (economywide)
cost differences across countries in trade equilibria and certainly none,
to our knowledge, that are based in analysis of the data. Thus it is hard
to arrive at judgments about where the production cost ratio should
stand and how our measured values of v compare.
Nevertheless, with values of v so close to one, one may still suspect
that the measured values of v simply reflect nearly full equalization of
factor prices rather than trading patterns (i.e., export of cheaper factors,
on average, as implied by the theory). Could the dot product on the
left-hand side of (9) be close to zero, that is, v be close to one, simply
because each term in the product is zero owing to identical factor prices
across countries? We examined this possibility in two ways. A casual
examination of the wage rates and returns to capital indicates that wage
factor price differences (in particular for labor) on the order of 2030
percent are prevalent among the countries in our sample (even after
we exclude Korea). Thus, for instance, production workers are reported
to earn a 30 percent higher measured wage in the United States than
in Germany, with nonproduction workers higher by about 15 percent.
Wage gaps between the United States and the United Kingdom are even
wider. So, as such, measured factor price differences are quite large in
our sample.18 Second, we conducted an industry by industry analysis in
which values of v were determined for each industry for each country
pair. If our reported findings on v were driven simply by nearly equalized
factor prices across countries, it must be the case that v take values very
close to unity for each industry as well. This is, however, most definitely
not the case. Indeed, combining measures of v across industries for all

18
If anything, the measured values of v are too small given the extent of the difference
in factor prices. To see this, note that, if all else were equal, a 30 percent wage difference
would be reflected in a value of v that is about 1.3 (or a bit less since the share of labor
in production is less than 100 percent). That this is not the case is indicative, among other
things, of the fact that production methods (i.e., technology matrices) are correspondingly
different as well (themselves reflecting the factor price differences).
bilateral trade 905
country pairs and analyzing them gives us the following breakdown: Of
the 476 industrycountry pair combinations, over 250 take values greater
than 1.05 and over 150 take values greater than 1.1.19 This implies im-
mediately that it is not full FPE that is driving our findings of values of
v close to one.
While it should be clear from the preceding discussion that FPE does
not drive our findings of values of v close to unity, one final observation
regarding factor price differences and the success of the theory (frac-
tion of country pairs for which the value of v is greater than one) is
nevertheless worth making. Correlating the success rate of the theory
with measures of factor price differences (aggregating across factors)
gives us a positive correlation.20 Consider the following measure of dif-
ferences in factor price vectors between c and c
:21
2 2


(w c w c)2 (LY) (r r ) (KY) .
c
c 2

If c
has wages that are 10 percent higher and a rental rate of capital
that is 10 percent lower than in c, the expression above takes the value
0.02 (for average values of the ratio L/Y and K/Y values in the sample).
For a 20 percent difference, it takes the value 0.08. For a 30 percent
difference, it takes the value 0.18. In our data, for country pairs for
which the expression above took values greater than 0.02, the theoretical
restriction was satisfied in over 90 percent of the cases. For higher values
of the expression, the success rate increased. For country pairs for which
the expression above took values greater than 0.03, the theoretical re-
striction was satisfied in 100 percent of the casesindicating an in-
creasing success rate with increased differences in factor prices. While
it is unwise to speculate out of sample, this raises the expectation that
the theory would hold with even greater success outside of the OECD
countries we are working with, where factor price differences may be
expected to be even larger.

A. Measurement Error in Factor Prices

A final set of results we present here concerns measurement error in
factor prices. Our data are inconsistent with the theory for roughly one-
quarter of the country pairs (i.e., as discussed above, for six or seven

19
Interestingly, even at the industry level, v takes values greater than one in the over-
whelming majority of cases. In contrast, fewer than 35 observations take values below 0.9,
and fewer than 70 take values below 0.95.
20
We are most grateful to David Weinstein for suggesting this analysis.
21
Note that this difference measure takes value zero if factor prices are equal across
the two countries and uses weights (K/Y ) 2 and (L/Y ) 2 to get away from the problem of
units of measurement of w and r.
906 journal of political economy
country pairs out of a total of 28, v takes values less than unity). However,
a number of these failures are minor in magnitude, with the ratio v
being greater than 0.99 but less than one in a great proportion of these
cases. To what extent could these failures be driven by simply measure-
ment error in factor prices? To examine this, measurement error in
factor prices can be modeled in the following fashion (an alternative
methodology that gives equivalent results is described in n. 22 below):
wobs p wtrue ew, ew N(0, jw2 ). (11)
That is, the observed value of any given factor price, wobs, can be assumed
different from the true value of the factor price, wtrue, by an amount
equal to the measurement error ew, where ew itself is assumed to be
distributed normally with zero mean and variance jw2. Consider a single
factor price at a time. When the values of all other observed factor
prices used in calculating the left-hand side of (9) are taken as true,
for the particular factor price being considered, wtrue can be set equal
to a value w so that the theory is just right (i.e., so that [9] is just satisfied).
Then when a large number of draws of wobs (10,000 draws in our ex-
ercises) under particular assumptions on the magnitude of jw (that, e.g.,
it is 5 percent of the value of wobs) are taken, the left-hand side of (9)
can be computed in each case and its distribution thus obtained. Given
the calculation of (9) using observed factor prices, we can then ask
whether we can reject the null that the theory is right (i.e., that the
left-hand side of [9] is greater than or equal to zero). This can then be
done for all factor prices and the exercise repeated for every country
pair, so we can finally ask how often we are unable to reject that the
theory is right.22
The results of these exercises are presented in table 6, where the
headings of the three columns indicate the extent of measurement error
assumed in drawing wobswith jw equal to 2.5 percent, 5 percent, and
10 percent of the mean of wobs, respectively. For a given combination
of factor price measures chosen (say, Euro I and Capital II), the rows
correspond to the significance level for the test. The entries in the table

22
An alternative exercise (in Bayesian spirit) would model the measurement error in
factor prices in the following fashion:
wtrue p wobs vw, ew N(0, jw2).
Now, under assumptions regarding the magnitude of jw for each factor price, say that it
equals 5 percent of wobs, we can take 10,000 draws on wtrue for each of the factor prices.
The left-hand side of (9) can be computed in each of the 10,000 cases, and the distribution
of the true value of the left-hand side of (9) can be obtained. We can then examine where
along this distribution the number zero lies (the minimum acceptable value of [9] for
the theory to be right). This tells us again, given our observations on factor prices, whether
we are able to reject the null that the theory is right. Given the linearity of (9) and the
normality assumptions, this exercise gives us results that are identical to those obtained
from the analysis described in the text.
bilateral trade 907

TABLE 6
Sign Test Results with Measurement Error
Simulation (%)

Degree of Measurement Error

Significance
Level 2.5% 5.0% 10.0%
Euro I and Capital I
1% 75.0 82.1 96.4
5% 75.0 78.6 92.9
Euro II and Capital I
1% 75.0 85.7 96.4
5% 75.0 78.6 89.3
Euro I and Capital II
1% 89.3 96.4 100.0
5% 82.1 89.3 100.0
Euro II and Capital II
1% 89.3 96.4 100.0
5% 85.7 92.9 100.0

corresponding to a given level of significance and a given level of mea-

surement error indicate the fraction of cases in which we were unable
to reject that the theory is right.23 As the figures in table 6 indicate,
allowing for measurement error in factor prices, we are unable to reject
the null that the theory is right in a very large fraction of cases. With
the standard deviation of measurement error assumed to be even just
10 percent of wobs, the success rate for the theory (i.e., the fraction of
cases consistent with the theory being true) is about 90 percent with
the Capital I rental measure and a full 100 percent with the Capital II
measure.

B. Robustness
In our analysis of the data so far, we have confirmed the robustness of
our findings to particular ways of measuring factor prices. Any of the
combinations of factor measures chosen (gross or net measures of rate
of return to capital and aggregated and less aggregated measures of
wages) gave us results of strong similarity.
In addition, the robustness of our results was checked by performing
the tests of (9) under various other configurations and data construction
methods. These alternative configurations include (i) using different

23
It should be easily recognized that tests of this nature do not necessarily have large
power against alternatives. Our results should then be viewed as only confirming the extent
of consistency of the data with the theory.
908 journal of political economy
depreciation rates (3 percent and 10 percent) in calculating net capital
stocks, (ii) using gross capital stock (readily available from the Inter-
national Sectoral Database [ISDB]) instead of net capital stocks, (iii)
using a variety of other accounting definitions for measuring return to
capital as prescribed by Gollin (2002), and (iv) using the total (domestic
plus foreign) input-output matrix rather than domestic inputs matrix
prescribed by Staiger (1986). None of these variations changed the test
results greatly. The success rate for the theory was about the same as
the results under the configuration we described earlier in the text (i.e.,
using net capital stock calculated using a 5 percent depreciation rate
and with the input-output matrix simply reflecting the usage of domestic
inputs as prescribed by Staiger [1986]).

V. Concluding Remarks
This paper has used OECD production and trade data to test the re-
strictions (derived by Helpman [1984]) on the factor content of trade
flows that hold even under nonequalization of factor prices and in the
absence of any assumptions regarding consumer preferences. We are
unable to reject the restrictions implied by the theory for the vast ma-
jority of country pairs. Our results are quite robust to the factor price
measures used and to a variety of assumptions made in the construction
of necessary variables from observed data.

Appendix A
Data
Countries were selected on the basis of the availability of the relevant data sets.
Of the eight countries chosen, five were European (Denmark, France, Germany,
the Netherlands, and the United Kingdom) and three were non-European (the
United States, Canada, and Korea). All data pertain to 1980, and all relevant
data were converted into 1980 U.S. dollars, unless otherwise stated.

A. Industry Coverage and Labor Disaggregation

Industrial activities are disaggregated according to the International Standard
Industrial Classification (ISIC) system (revision 3, 1968), at the one-digit level
for nonmanufacturing (eight sectors) and at the two-digit level for manufac-
turing (nine sectors), for a total of 17 industrial sectors. European data cate-
gorized according to the Nomenclature Generale des Activites Economiques
dans les Communautes Europeenes (NACE) (taken from Eurostats SOE) were
converted into ISIC code according to table A1.
Labor input factors for non-European countries were disaggregated into seven
categories according to ISCO 1968: professional/technical workers (code 0/1),
administrative/managerial workers (2), clerical workers (3), sales workers (4),
service workers (5), agricultural workers (6), and production workers (7/8/9).
bilateral trade 909

TABLE A1
Seventeen Industries and Their Concordance with ISIC and NACE

Description ISIC Code NACE R6/R25

Agriculture, hunting, forestry, and fishing 1 01
Mining and quarrying 2 12, 14
Food, beverages, and tobacco 31 36
Textiles, apparel, and leather 32 42
Wood products 33 48
Paper, paper products, and printing 34 47
Chemical products 35 17, 49
Nonmetallic mineral products 36 15
Basic metal industries 37 13
Fabricated metal products and machinery 38 19, 21, 23, 25, 28
Other manufacturing 39 48
Electricity, gas, and water 4 06
Construction 5 53
Wholesale and retail trade, restaurants,
and hotels 6 56, 59
Transport, storage, and communication 7 61, 63, 65, 67
Finance, insurance, real estate, and busi-
ness services 8 69A
Community, social, and personal services 9 74

Labor input factors for European countries were disaggregated into production
workers and nonproduction workers. Nonproduction workers comprised top
management executives, other senior executives, clerical workers, assistants, and
supervisors. Table A2 presents more detail on the Euro I and II labor catego-
rizations used in this paper and their concordance with ISCO and SOE
classifications.

B. Technology
The technology matrices comprise two parts: a direct input matrix (B: factor by
industry) and an input-output matrix (A: industry by industry).

1. Direct Input Matrix

The direct input matrix measures how much labor and capital each industry
employs at a given point in time. The disaggregated occupational distribution
of labor was taken from the Economically Active Population by Occupation table
in the International Labour Offices Yearbook of Labor Statistics (194589). Be-
cause, however, the table did not disaggregate the manufacturing sector at the
sectoral level, we relied on each countrys census of population data for non-
European countries24 and on the SOE for European countries in order to obtain
the occupational distribution of labor in each sector of manufacturing. We then
scaled the number of workers in each occupation in each manufacturing sector
24
For the United States and South Korea, the data were available from the Census of
Population conducted in each country in 1980. In the case of Canada, the occupational
distribution in disaggregated manufacturing industries was available only from its 1996
census. Thus we had to assume that the ratio of occupational distribution to total man-
ufacturing workers did not change from 1980 to 1996.
910 journal of political economy

TABLE A2
Concordance of Labor Categories

ISCO 1968 (Non- Structure of Earnings

Euro I Euro II European Countries) (European Countries)
Production Production Service (5) Manual workers
Agricultural (6)
Production (7/8/9)
Nonproduction Managerial Administrative/mana- Top management
gerial (2) executives
Other senior executives
Clerical Clerical (3) Clerical
Others Professional/technical Assistants
(0/1)
Sales (4) Supervisors

so that the total number of occupational workers in manufacturing equaled that

in the International Labour Offices yearbook.
Data on net capital stocks at the sectoral level were not directly available for
the countries in our sample. Measures of net capital stock therefore had to be
constructed.25 We constructed them in this manner: First we calculated the initial
net capital stock in each industry in 1970 by taking the aggregate net capital
stock in 1970 from the Penn World Table and the gross capital stock of each
industry in 1970 from the ISDB of the OECD. Then we calculated the net capital
stock in each industry by distributing the total net capital stock into each industry
using the industrial ratio in the ISDB. Next, we took the data on annual gross
fixed capital formation in each industry during 197180 from the ISDB.26 Taking
the initial disaggregated net capital stock and each years disaggregated capital
formation data, we used the perpetual inventory method to calculate net capital
stock in each industry in 1980. The test results reported in Section IV employed
a depreciation rate of 5 percent.

2. Input-Output Matrix (Indirect Input Matrix)

The entries in the input-output matrix, or indirect input matrix, represent the
quantity of intermediate inputs that a sector purchases from other sectors to
produce one unit of output. The OECDs Input-Output database provides three
sets of input-output matrices for each country. The first is the domestic input-
output matrix, which shows the usage of domestically produced intermediate
goods in each sector. The second is the imported input-output matrix, which
measures how many intermediate goods are imported in each sector. The third
is the total input-output matrix, a simple summation of domestic and imported
input-output matrices.
Given Staigers (1986) proposed modification to the factor content calcula-
tions suggested in Helpman (1984), the domestic input-output matrix, which
includes the factor content of traded intermediate goods, is a more appropriate
choice than the total input-output matrix. To see the argument behind Staigers
25
Measuring capital stock in each industry is important not only because it is a com-
ponent of the direct input matrix but also because it affects the rate of return to capital.
26
Since the Korean data are not included in the ISDB, we obtained the data from the
National Account Department of the Bank of Korea.
bilateral trade 911
modification, consider the following simple three-country, four-commodity case.
Goods 1 and 2 are final goods that use goods 3 and 4 as intermediate goods.
To produce one unit of good 1, we need a units of good 3 and b units of good
4. Also assume that one unit of labor is required for both good 3 and good 4.
Countries A, B, and C produce goods 1, 2, and 3, respectively, and good 4 is
produced by both countries A and B. Now, suppose that country A exports one
unit of good 1 to country B. Country As production cost will be
w Ab w Ca.
If this good were produced in country B, the production cost would be
w Bb w Ca.
For country A to be an exporter of good 1,
w Ab w Ca w Bb w Ca
or
(w B w A)b 0.
The last expression is exactly analogous to what we derived in Section II. Note
that the relevant input-output coefficient is not that of the imported interme-
diate good (a), but that of the domestically produced intermediate good (b).

C. Bilateral Trade
The manufacturing sectors bilateral trade data were obtained directly from the
OECDs Bilateral Trade database for each pair of countries in our sample. These
data provide the bilateral trade flows according to ISIC categorization and are
thus readily conformable with the technology matrix constructed above. The
bilateral trade data for nonmanufacturing sectors were not available. So, follow-
ing Davis and Weinstein (2001), we set bilateral imports of nonmanufacturing
sectors equal to the share of manufacturing imports times total nonmanufac-
turing imports in that sector, where total nonmanufacturing imports were taken
from the OECD Input-Output database.

D. Factor Prices
Section III describes the construction of factor price data in detail, so only a
brief recapitulation is provided here. We calculated the ex post rental rate of
capital by dividing the operational surplus from the OECDs Annual National
Account database by the total capital stock from the OECDs ISDB. The occu-
pational wage rate was taken from the Census of the Population for each non-
European country and from the SOE for the European countries in our sample.
For the purpose of international compatibility, we modified the data as described
in Section III.

Appendix B
A. Hicks-Neutral Technology Differences
An attractive feature of the framework described above is that it relaxes a number
of unrealistic assumptions regarding factor prices and consumer preferences
912 journal of political economy
that have traditionally been made in the empirical literature in this area. How-
ever, as we have already noted, one rather restrictive assumption remains: iden-
tical constant returns to scale technologies across countries. To relax this some-
what, we allow for Hicks-neutral factor efficiency differences across countries
(just as in Trefler [1993, 1995]). The derivation of the restrictions analogous
to (7)(9) is straightforward.
Suppose that all input factors in country c
are more productive than those
in country c by the factor of l (l 1 0). Then, equation (4) becomes



1

p(Qc T c c) P p, V c TVc c
l ( )


1

P(p, V c ) PV (p, V c ) TVc c
l



w c c
c
p pQc T (B1)
l V
because now country c
could do better than country c (in terms of output)


even with only (1/l)TVc c. Applying the zero profit condition in country c,
c
c c c
c
pT p (w )TV , we have the following equation (corresponding to eq. [7] in
Sec. II):
c

(wl w ) T c
V
c
c
0. (B2)

In general, if li is the Hicks-neutral technology parameter describing factor

efficiency levels in country i relative to some benchmark country, (7)(9) can
be rewritten as


w c w c c
c
( lc

T 0,
lc V ) (B3)

c c

(wl wl ) T
c c
V
cc

0, (B4)

and


wc wc
( lc


lc


)

(TVc c TVcc ) 0. (B5)

B. Ricardian Technology Differences

Let the extent by which industry i in country c
is more productive than industry
i in country c be denoted by gi . Using expressions analogous to (B1) above, we
now have the following final expression relating factor prices and factor flows
(and technology):

(


wc
i gi


w c TVic c 0. )
Now, if information on the Ricardian technology parameters, the gs, is available,
the expression above can be tested. Thus the tests proposed by Helpman (1984)
bilateral trade 913
that we have implemented in this paper are easily extended to account for
Ricardian technology differences between countries as well.

References

Antweiler, Werner, and Daniel Trefler. 2002. Increasing Returns and All That:
A View from Trade. A.E.R. 92 (March): 93119.
Bernstein, Jeffrey R., and David E. Weinstein. 2002. Do Endowments Predict
the Location of Production? Evidence from National and International Data.
J. Internat. Econ. 56 (January): 5576.
Bowen, Harry P., Edward E. Leamer, and Leo Sveikauskas. 1987. Multicountry,
Multifactor Tests of the Factor Abundance Theory. A.E.R. 77 (December):
791809.
Brecher, Richard A., and Ehsan U. Choudhri. 1982. The Factor Content of
International Trade without Factor Price Equalization. J. Internat. Econ. 12
(May): 27783.
. 1993. Some Empirical Support for the Heckscher-Ohlin Model of Pro-
duction. Canadian J. Econ. 26 (May): 27285.
Davis, Donald R., and David E. Weinstein. 2001. An Account of Global Factor
Trade. A.E.R. 91 (December): 142353.
. 2003. The Factor Content of Trade. In Handbook of International Trade,
edited by E. Kwan Choi and James Harrigan. Malden, Mass.: Blackwell.
Debaere, Peter. 2003. Relative Factor Abundance and Trade. J.P.E. 111 (June):
589610.
Gollin, Douglas. 2002. Getting Income Shares Right. J.P.E. 110 (April): 458
74.
Hakura, Dalia S. 2001. Why Does HOV Fail? The Role of Technological Dif-
ferences within the EC. J. Internat. Econ. 54 (August): 36182.
Hanson, Gordon H., and Matthew J. Slaughter. 2002. Labor-Market Adjustment
in Open Economies: Evidence from US States. J. Internat. Econ. 57 (June):
329.
Harrigan, James. 1995. Factor Endowments and the International Location of
Production: Econometric Evidence for the OECD, 19701985. J. Internat.
Econ. 39 (August): 12341.
. 1997. Technology, Factor Supplies, and International Specialization:
Estimating the Neoclassical Model. A.E.R. 87 (September): 47594.
Helpman, Elhanan. 1984. The Factor Content of Foreign Trade. Econ. J. 94
(March): 8494.
. 1998. The Structure of Foreign Trade. Working Paper no. 6752 (Oc-
tober). Cambridge, Mass.: NBER.
Leamer, Edward E., and James A. Levinsohn. 1995. International Trade Theory:
The Evidence. In Handbook of International Economics, vol. 3, edited by Gene
M. Grossman and Kenneth Rogoff. Amsterdam: Elsevier.
Leontief, Wassily. 1953. Domestic Production and Foreign Trade: The American
Capital Position Re-examined. Proc. American Philosophical Soc. 97 (Septem-
ber): 33249.
Schott, Peter K. 2003. One Size Fits All? Heckscher-Ohlin Specialization in
Global Production. A.E.R. 93 (June): 686708.
Staiger, Robert W. 1986. Measurement of the Factor Content of Foreign Trade
with Traded Intermediate Goods. J. Internat. Econ. 21 (November): 36168.
914 journal of political economy
Trefler, Daniel. 1993. International Factor Price Differences: Leontief Was
Right! J.P.E. 101 (December): 96187.
. 1995. The Case of the Missing Trade and Other Mysteries. A.E.R. 85
(December): 102946.