6"
^
M.I.T.
LIBRARIES
-
DEWEY
^)fe4>^7
ALFRED
P.
WORKING PAPER
SLOAN SCHOOL OF MANAGEMENT
AN EMPIRICAL ASSESSMENT OF THE
FACTOR PROPORTIONS EXPLANATION OF
MULTINATIONAL SALES
S. Lael
Brainard
Massachusetts Institute of Technology
WP#3624-93-EFA
November 1993
MASSACHUSETTS
INSTITUTE OF TECHNOLOGY
50 MEMORIAL DRIVE
CAMBRIDGE, MASSACHUSETTS 02139
AN EMPIRICAL ASSESSMENT OF THE
FACTOR PROPORTIONS EXPLANATION OF
MULTINATIONAL SALES
S. Lael
Brainard
Massachusetts Institute of Technology
WP#3624-93-EFA
November 1993
JAN
2 5 1994
AN EMPIRICAL ASSESSMENT OF THE
FACTOR PROPORTIONS EXPLANATION OF
MULTINATIONAL SALES "
*
Abstract
This paper provides empirical evidence that challenges the factor proportions
explanation of multinational activity.
volumes
were used
proportions and income
that
to demonstrate that
affiliate
on intra-industry ratios and total
a substantial part of trade is explained by factor
tests
similarities rather than differences are applied to affiliate sales with
Some
support for the factor proportions hypothesis
is
derived by
production destined for export to the parent's market, which
is
the
surprisingly similar results.
comparing
The same
category of activity most likely to be motivated by factor proportions considerations, with
that destined for sale in the local market.
Affiliate production destined for export
moderately more responsive to factor proportions differences.
activity differ
more
in their
home
is
However, the two types of
responses to transport costs and destination market income.
Overall, the evidence suggests that only a small part of multinational activity into and out of
the
US
in the late
1980s can be explained by factor proportions differences.
JELNos. F12, F21, F23
"The author takes sole responsibility for the use and interpretation of the data, and for any
errors therein.
am
Andrew Bernard, Julio Rotemberg, and Bill 2^ile for helpful comments
and to Lilan Cheng for excellent research assistance. I am also grateful to Steve Landefeld,
Betty Barker, David Belli, Ned Howenstine, Arnold Gilbert, Smitty Allnut, and Viji
Canagasuriam at the BEA, Nick Orsini at the Census Bureau, and Jim Harrigan at Penn State
**I
grateful to
for their generosity in providing data, and to the National Science Foundation for support
under grant SES-91 11649.
Introduction
I.
The imperfect competition models of the
early 1980s provided an explanation for trade based
on
similarities
between countries rather than differences. They also provided a more natural framework for treating multinational
Although these models diminished the importance of factor proportions
enterprises.
in explaining trade flows,
consistent with empirical evidence, they maintained the primacy of factor proportions in explaining multinational
activities.
Yet in recent years,
this explanation
investment (FDI) and multinational
appears increasingly
A
activities.
large and
at
odds with observed patterns of foreign direct
growing share of multinational
industrialized nations as both the source and destination markets, rather than flowing
1961 and 1988, over half of
countries; this share
all
had risen
direct investment outflows generated
to nearly
70 percent by 1988
from North
activity involves
to South.
Between
by G-5 countries was absorbed by other G-5
(Julius, 1990).
This paper provides empirical evidence that challenges the factor proportions explanation of multinational
activity.
The same
part of trade
is
tests
on intra-industry
The
export to the parent's market.
The
total
volumes
that
were used
to demonstrate that a substantial
predictions of the factor proportions hypothesis for multinational sales
by distinguishing between
and out of the
and
explained by factor and income similarities rather than differences are applied to affiliate sales with
surprisingly similar results.
investigated
ratios
US
in the late
affiliate
further
production destined for sale in the local market and that destined for
Together, the evidence suggests that only a small part of multinational activity into
1980s can be explained by factor proportions differences.
factor proportions explanation for the location of multinational activity focuses
characteristic of
is
on
vertical
expansion
North-South flows (Markusen, 1984; Helpman, 1984; Helpman and Krugman, 1985; Ethier and
A
Horn, 1990; Ethier and Horn, 1991).
factor proportions
factor proportions are identical, the differential in
model with differentiated products predicts
GDP shares
that
when
of the source and destination countries and their joint
income should be the only determinants of trade volumes, and there should be no multinational activities. However,
this
paper finds that equations using only the differential in
share of variation in the
signs and significance.
The
volume of multinational
This finding
factor proportions
is
model
GDP shares
sales as in the
and joint income explain roughly the same
volume of
trade,
and the coefficients have similar
inconsistent with a factor prof)ortions explanation of multinationals.
further predicts that the intra-industry share of trade flows should be
decreasing in factor proportions differences, while multinational activities should arise only in a single direction
between countries with large factor proportions differences.
lower than those of trade on average, they are
still
Although intra-industry
significant,
which
ratios
of
affiliate sales are
strongly counter to the factor proportions
is
Further, tests reported below on
explanation, and they are {wsitively correlated with intra-industry ratios of trade.
intra-industry ratios of trade and affiliate sales find that both are inversely related to factor proportions differences,
and
that a greater share
of the variation
similarities in factor pro[>ortions
The
of
than of trade
affiliate sales
is
explained by
and transport costs.
alternative to the factor proportions hypothesis formulates location advantages in terms of proximity
to customers or specialized suppliers,
scale
in intra-industry ratios
which motivate horizontal expansion across borders
at the
expense of reduced
Proximity-concentration models predict
(Knigman, 1983; Horstman, Markusen, 1992; Brainard, 1992).
multinational activities will arise between countries with high transport costs and trade barriers in industries with
low plant
tests
on
scale economies,
the total
and involve increasing two-way
volumes and intra-industry
predictions, both for
income and
ratios
factor differences
more
activity the
of trade and
roughly consistent with these
affiliate sales are
and for transport
costs.
The
tests find that freight factors
a strong negative effect on trade flows and a negligible or weakly positive effect on
The paper
affiliate
The
The
similar are factor proportions.
have
affiliate sales.
further evaluates the factor proportions explanation of multinational activity
by comparing
production destined for sale in the local market with that destined for export back to the parent market.
factor proportions
model predicts
that all varieties
of a
final
good produced by
a foreign affiliate should be
exported back to the headquarters market, while the proximity-concentration hypothesis predicts that multinationals
will substitute overseas production for trade in final goods, so that flows of final
will be zero.
With a multistage production process,
will generate
one-way trade
two-way trade
while the
US
affiliates
compare the share of
that multinationals
model predicts they will generate
home
on
multinationals export 13 percent of their overseas production to the
US,
affiliate sales
exported back
of foreign multinationals export 2 percent of their
total affiliate
market
small in both directions:
The amount of
owned by US
to the parent
model further predicts
in intermediates, while the proximity-concentration
in intermediates.
aggregate, foreign affiliates
the factor proportions
goods back
production destined for export back
US
is
production to their parents. Tests that
home with
the share destined for local sale
confirm that the two types of activity differ moderately in
their relation to factor proportions differences in a
home. However, they
that is roughly consistent with a factor proportions explanation for the share exported
in their responses to transport costs
more
The
of
its
tests
SIC
differ
and destination market income.
use data on bilateral flows of trade and
trading partners at the 3-digit
way
affiliate sales in
level, as well as direct
both directions between the
measures of transportation costs.
US
The
and 27
analysis
focuses on bilateral flows rather than multilateral flows as would be suggested by theory, due to data limitations.
In addition, a cross-section approach
that for
which the data on
is
taken because most of the independent variables vary at intervals longer than
affiliate sales is available.
Two Models
n.
Guided primarily by findings from case study
framework explaining the decision
to
research, the managerial literature has developed a conceptual
produce abroad rather than export as optimal
in
markets characterized by the
conjuncture of internalization, ownership, and location advantages. In recent years, trade theorists have incorporated
these insights into general equilibrium models.
Broadly, two types of models have emerged to explain the location
Both focus on the choice between exporting and investing across borders
decision.
firms benefit from internalization due to multi-plant economies of scale.
I
starting
from the premise
that
briefly contrast their predictions for the
patterns of trade and of multinational sales.
Factor Proportions
i.
Markusen, Helpman (1984), Helpman and Krugman, and Ethier and Horn (1990) explain
expansion across borders in terms of relative factor endowments and technological parameters.
^
'
vertical
The Helpman
and Helpman, Krugman model incorporates multinationals into a differentiated products, factor proportions model
'
factor
somewhat more concerned with organizational questions against the background of a
proportions motivation for overseas production; they model firm configurations as trading off economies of
Ethier and
Horn
are
scope against increased costs of managerial control and coordination.
^
In the
Markusen model,
factor
explaining multinational production.
endowments
are symmetric, but sector-specific capital plays a key role in
of trade by assuming that the production process in the differentiated products sector can be separated into multiple
stages,
and there are multi-plant economies of scale associated with a firm-specific input which has a public goods
character.
I
briefly review the
and then go on
main predictions of the pure trade model
to discuss the extension to multi-plant firms.'
here,
which
will be familiar to
most readers,
Consider a two-sector, two factor, two-country
world, where countries are distinguished only by their relative factor proportions differences and their relative
incomes. Tastes are homothetic and identical, and there
is
a constant-retums-to-scale,
homogeneous goods
sector,
and a differentiated products sector characterized by Chamberlinian monopolistic competition and increasing returns.
by a
In such a world, there is intra-industry trade in differentiated products motivated
trade in the
homogeneous product, and
countries.
Then,
trading partners.
the net factor content of trade reflects the relative factor
if either there is specialization in the
differentiated products, the total
More
taste for variety
volume of
trade
is
and one-way
endowments of the
production of homogeneous products, or
if all trade is in
a decreasing function of the relative size differential of the
generally, the share of intra-industry trade in total trade
is
a decreasing function of factor
proportions differentials.
When
the production process in the differentiated products sector
is
separable, so that firms can place
headquarters activities in one market and manufacture in either market, the pattern of production will depend on
the factor intensities of each stage of the production process and the relative factor
economies.
If headquarters
activities
and production
have different factor
activities
endowments of
intensities,
the
two
single-plant
multinationals will arise to exploit potential factor cost differentials, for large factor proportions differences.
When
factor
equilibrium, there
is
endowments are
no incentive
sufficiently similar that factor price equalization obtains in the trade
for cross-border investment,
headquarters activities are relatively more capital-intensive.
and the pure trade equilibrium
When
results.
Assume
factor prices are not equalized under trade,
some
of the firms in the differentiated sector locate their headquarters in the relatively capital-abundant economy and
production in the relatively labor-abundant economy, and export back to the headquarters market.
In general
equilibrium, the ability to geographically separate firms' activities in the differentiated sector leads to an enlargement
'
These predictions are derived
in
Helpman, Krugman, 1985.
4
of the factor price equalization
set.
Because cross-border investment
is
motivated solely by factor price differentials,
multinational activities arise only in a single direction within an industry, in single-plant firms, between economies
with strong factor proportions differences.
With a two-stage production process, the existence of multinationals
generates interindustry trade of final goods, and one
would expect
direction between countries with different factor proportions.
may
at
to see multinational activities arising in a single
With
additional stages of production, multinationals
generate interindustry trade of both final goods and intermediates, but again in one direction within an industry
''
each vertical stage.
The
empirical implications of this framework are quite strong.
arise only in
one direction between countries
in an industry
different factor proportions that factor price equalization
predicts that multinationals will
it
and only between economies
would not
are sufficiently different that firms have an incentive to
First,
move
result
under trade. Second,
have sufficiently
proportions
if factor
their plants abroad, the creation
With an additional stage of production,
diminishes the share of intra-industry trade.
that
of multinationals
MNEs may
interindustry trade of intermediate goods in one direction and of final goods in the reverse direction.
increasing role for
Similarly,
it
MNEs
may weaken
weakens the
link
between country
size dispersion
the link between factor proportions differences
and the
total
generate
Third, an
volume of
and the share of intra-industry
trade.
trade.
With
a two-stage production process, the intra-industry share of trade will be positively related to factor proportions
differentials if the headquarters
otherwise.
With an
market
is
a net importer of the final differentiated goods, and negatively related
additional production stage, the intra-industry share tends to decrease with increasing factor
proportions differentials because the additional trade in intermediates
ii.
is inter- industry.
Proximity-concentration Tradeoff
Another group of
paf>ers uses the
same
differentiated products
framework
to explain horizontal expansion
across borders motivated by considerations of access to the destination market at the expense of production scale
economies (Krugman; Brainard, 1992). Consider the same two-sector, two-country world, but assume
that factor
proportions are identical, and firms in the differentiated products sector choose between exporting and cross-border
expansion as alternative modes of foreign market penetration.
The
differentiated sector
is
again characterized by
increasing returns at the firm level due to
production
facilities
some
input, such as
with undiminished value, but
now
R&D,
can be spread
that
there are scale
economies
among any number of
at the plant level,
concentrating production lowers unit costs, and a variable transport cost that rises with distance.
whether
to
such that
Here, the decision
expand abroad via trade or via investment hinges on a trade-off between proximity advantages and scale
advantages from concentrating production in a single location.
In the absence of factor proportions differences
between the two economies, the magnitude of variable
transport costs and the size of scale economies at the plant level relative to the firm level will determine the location
and configuration of production chosen by firms.
Suppose there
is
a simple fixed cost associated with each plant
of F; exporting costs of e^*" per imit, which incorporates both freight factors associated with distance, D, and
barriers, T; per unit variable costs, V(w,r), that are declining in the
amount of
the firm-specific input,
tariff
r,
and
increasing in local wages, w; and costs associated with production of the input C(w,r). Then, with n„ firms in each
market, firms will adopt multinational configurations with plants in both markets
if the increase in variable profits
associated with producing close to consumers in both markets outweighs the additional plant fixed cost.
Combining
the no-defection condition with free entry yields the equilibrium condition:
F(w)
(1)
which simplifies by assuming
price index (which
is
do not take
increasingly accurate as the
a multinational equilibrium,
is
that firms
where
all
^
(i-^n->pxi..))
into account the effect
number of firms
of their potential deviation on the market
increases)'.
Comparative
firms have production facilities in both markets,
is
statics establish that
more
likely the smaller
the fixed production cost relative to the corporate fixed cost, and the higher are transport costs and trade barriers.
Here, as the corporate cost goes to 0, an equilibrium with multi-plant firms
a fixed
number of
firms, the dual-plant equilibrium
is
more
An
*
is
two-way multinational
never sustainable.
In addition, for
likely to hold the larger is the foreign market.
an equilibrium, multinational production completely supplants trade in
corporate services, and there
is
activity in the
final
same
goods, there
is
In such
trade only in 'invisible'
industry.
equilibrium characterized by single-plant, national firms arises under reverse conditions.
Firms
Brainard (1992) gives equilibrium conditions taking into account the effect on the market price index.
6
maintain a single-plant production configuration as long as the increase in fixed costs to open a second plant in the
foreign market outweighs the associated increase in variable profits.
are characterized by the exact reverse conditions as in equation (1).
prevail the higher
is
With
The
single-plant equilibrium
the fixed plant cost relative to the corporate fixed cost
the distance between the markets.
Here, there
factor proportions are equal, all trade
is
two-way trade
intra-industry, as
is
in final
free entry, the equilibrium conditions
is
more
likely to
and the lower are the transport cost and
goods
in the differentiated sector, and, if
would be predicted by a
differentiated products
model
of trade.
In the intermediate range of parameter values, there
coexist with national firms.
In the
is
mixed equilibrium, some
a third equilibrium in which multinational firms
fraction, a,
and exports, and the remaining fraction has production
of firms in each market has a single
both markets.
For a given
and trade
barriers, the
production
facility
number of
firms, the prop)ortion with a single plant rises the smaller are transport costs
facilities in
greater is the fixed plant cost, the smaller are the savings in variable costs from additional
larger
is
is
investment, the
the corporate cost for the two-plant configuration relative to the single-plant configuration, and the smaller
the size of each market.
multinational production.
Thu£,
this
mixed equilibrium, there
In the
The
is
both two-way trade in
final
goods, and two-way
share accounted for by multinational production rises with distance and trade barriers,
and declines with production fixed
activity
R&D
costs.
model explains horizontal expansion across borders motivated by market
access: multinational
can arise in the absence of factor proportion differences and in two directions in the same industry, and
is
undertaken by multi-plant firms.
Of course,
the
two models are not mutually exclusive.
concentration tradeoff are combined,
considerations.
will
emerge
firms
As Brainard (1992) shows
make
in a
also
and a proximity-
on both
hybrid UKxlel, this implies that vertical single-plant multinationals
relative to proximity advantages that national firms
may
factor proportion differences
the decision whether to produce abroad based
for sufficient factor proportions differences,
Single-plant vertical multinationals
When
would
emerge
when
concentration advantages are sufficiently strong
prevail in the absence of factor proportions differences.
for strong factor proportions differences
when proximity
advantages are sufficiently strong that horizontal multinationals would form in the absence of factor price
differentials.
Thus, the addition of factor proportions differences increases the likelihood of concentrating
production in a single location, leading to the coexistence of single-plant multinationals with either multi-plant
multinationals or single-plant national firms.
With a
three-stage production process,
when
factor proportions
considerations dominate, multinationals arise in a single direction between countries and generate interindustry trade
in intermediate
way
and
trade in final
final
goods, while when considerations of proximity dominate, multinational sales supplant two-
goods of unequal magnitudes and may generate intra-industry trade
to a multi-country context,
one would expect
to see a
countries that are dissimilar and a pref)onderance of
in intermediates. Generalizing
preponderance of one-way multinational sales between
two-way
activity
between countries with similar endowment
ratios.
Related Literature
III.
This paper
is
the first to investigate the empirical validity of differentiated products, factor proportions
models of multinationals. The empirical research on multinationals has focused mainly on foreign direct investment
flows, and
much of
the older research
it
has been preoccupied with responses to differentials in corporate income taxation.
on non-tax determinants focused on hypotheses based on
foreign direct investment
is
conceived in terms of physical capital.
Much
factor proportion differences,
These empirical
efforts
of
where
were not successful
at
explaining the location of multinational activity.^
This literature missed two
critical characteristics
foreign direct investment and trade flows
is
of multinational activity.* First, the comparison between
a conceptual mismatch.
When
multinational activities are considered
as a substitute for trade in the presence of trade barriers, for instance, the relevant
trade flows and multinational sales rather than investment.
the current account,
and
this is
Although trade and investment are connected through
an interesting connection from a macroeconomic perspective,
measure of the extent and location of overseas production.
'
comparison should be between
FDI
is
not a good
Second, by their very nature, multinationals cannot
See Caves (1982) for a survey.
With the notoble exceptions of Swedenborg (1979) Blomstrom
and Grubert and Mutti (1991).
'
8
et. al.
(1985) and Lipsey and Weiss (1981)
adequately be incorporated into perfect competition trade nnxiels.
The
literature has so far not directly addressed
the predictions of imperfect com[>etition, general equilibrium models for multinationals.
There
are,
however, several empirical findings
US
per worker ratios of
that are relevant.
parents and their affiliates are inversely correlated, which
proportions account, but capital-labor ratios are uncorrelated.
for
Kravis and Lipsey (1982) find that payroll
Swedish multinationals, Swedenborg (1984) finds
is
consistent with a factor
Using cross-section and time
that affiliate sales are increasing in
series firm-level data
wage
She also finds
exports are uncorrelated, and both increase in the natural resource intensity of the industry.
neither
related to per capita income,
is
which has been used
to
proxy capital-labor differentials
differentials while
that
in the empirical trade
literature.
There
is
explanations of trade.
exceeded
its
body of empirical research
a large
investigating factor proportions
Several researchers have found that the strength of the
empirical predictive power.'
differentiated products have
The
HOS
and differentiated products
theoretical
framework has
predictions of a factor proportions framework elaborated to include
met with more success (Helpman, 1987), but more recently Hummels and Levinsohn
(1993) have questioned the robustness of the results and their interpretation.
None of this work
distinguishes trade
mediated by multinationals from arms-length trade.
Data
IV.
As
is
well
known
countries are daunting.
Analysis
activity
US
of empirical trade, limitations due to incomparability of trade data across
Limitations on multinational sales data are even worse.
(BEA) compiles
between the
in the field
The US Bureau of Economic
the most complete set of data disaggregated by industry, but
and
its
trading partners.
it
covers only bilateral
In order to fully exploit this data set, the analysis below
confined to two-way bilateral relationships. Ideally, of course, a single equation would be used to
multilateral relationships.
'
the
The
analysis focuses
on a cross-section of industry-country
test a full set
hypothesis.
of
pairs for 1989.
See Deardorff (1984) for a survey, and Bowen, Learner, Sveikauskas (1987) for the most complete
HOS
is
test
of
Trade flows and
i.
The
Kong,
They
Argentina, Australia, Austria, Belgium, Brazil, Canada, Columbia, Denmark, France, Germany,
Hong
Ireland, Italy, Japan,
at the 3-digit
Mexico, Netherlands,
SIC
level
was obtained from
level of aggregation available,
which
is
and production structure, and minimize missing
New
Zealand, Norway, Philippines, Singapore, South Korea,
Kingdom, and Venezuela. Data on both
Census Bureau.
the
Data on
wholesale and
sets
retail trade,
between the 3-digit and 2-digit SIC
data includes
utilities,
lowest
at the
which over SO
were excluded, along with
because the 'nontradeable' nature of these activities requires a local presence. Matching
292 were dropped because
all
US
affiliates
Because the multinational data
I
is
is less
familiar than the trade data,
in the analysis,
which
will
I
sales
allocated to wholesale trade.
The
Tlie data
will describe
affiliates
in
it
the 27 countries
and 64 industries included
The reverse
is
sales
in the
true of trade flows:
and imports (20 percent). The
sample
imports
The share of both imports and exports accounted
for
is
total
some
on
the outward side.
detail here.
be termed 'tradeables", gross trade flows are
positively correlated with gross affiliate sales in both directions, but the correlation
and exports (60 percent) than for inward
(BEA
on multinational
did not have matching trade data).
on the inward side but only majority-owned foreign
For the industries included
billion.
was compiled
Industries for
levels.
manufacturing probably understates the true values because some proportion
sales
imports and exports
of classifications yielded data on 64 industries, covering manufacturing and primary industries
categories 150 and
in
bilateral
affiliate sales
percent of revenues are accounted for by services, including finance and
two
countries were selected to
data.
in geographical coverage, income,
Spain, Sweden, Switzerland, Taiwan, United
the
The
data includes trade and multinational sales for 27 countries in total.
maximize diversity
include:
affiliate sales
is
much
stronger for outward
total level
of outward sales for
$ 542 billion, well above that for inward sales of $ 393
$ 380 billion, exceeding exports totalling $269 billion.
by intra-firm
transfers
is
roughly equal
at
one quarter.'
When
imports and exports to and from unrelated parties are included as well as those to and from affiliated parties, the
'
US
There may be some double counting
in the
two-way estimates, since some
parents and foreign-owned affiliates, but for intra-firm trade
10
it
US
affiliates are classified
should be very small.
both as
of trade mediated by
ratio
affiliates is
32 percent on the import
and 37 percent on the export
side,
comparisons must be taken as rough approximations, however, as the
BEA
data on firm trade
On
industry of affiliate sales, while the trade data are classified by the actual product.
foreign affiliates
owned by US
by foreign-owned
parents generates $0. 13 of
US
affiliates in the
down by
Table lA breaks the data
ratios
of sales to trade
in
to
country.
It
US
exports, and $0. 15 of
shows the
statistics for
an industry breakdown.
by
In both cases, trade mediated
countries,
22 percent of outward
sales are to
Netherlands accounting for an additional 43 percent of the
exports, with Japan a close second at 16 percent, and Mexico, the
The
additional 22 percent.
the
of
UK, Norway, Germany, and
Canada accounts
percent.
for
ratio
for 14 percent,
The order
is
somewhat
affiliate sales to
Brazil.
On
exports
is
different for imports
lower than that on the outward
The
last
side, but is
two columns of Table
affiliates
industry codes are listed
affiliate sales variables
UK, Germany,
Canada similarly accounts
UK, and Germany
UK
is
France, and the
for 21 percent of
together accounting for an
particularly high for Ireland,
accounts for 25 percent of
-
affiliate sales,
with Japan accounting for 24 percent, Canada accounting
for 19 percent.
The
ratio
of affiliate sales
to
imports
extremely high for particular countries, such as the Netherlands,
The
1
A report Grubel-Lloyd indices of intra-industry
ratios are aggregated across industries
US
ratios
of trade and similar
by country as follows:
Estimates of the share of trade mediated by multinationals seem to vary widely.
include trade between
parties in the
and trade separately.
UK.
indices for affiliate sales.
'
BEA
affiliate sales, the
and Germany, Japan, and the Netherlands together account for an additional 33
22 percent, and Mexico, Taiwan, and Germany accounting
Switzerland, and the
imports.
imperfect).
is
2.5 on average, and
the inward side, the
imports, while $1 of sales
from the trade and
Canada, with the
total.
by the
average, $1 of sales by
affiliate sales
Descriptions of the
affiliates is subtracted
is classified
of inward and outward
minimize the overlap between the two (although, as noted above, the match
Comparing
is
level
US
US
exports and $0. 11 of
each direction, and aggregate intra-industry ratios for
Table IB gives the same
in the appendix.
generates $0.08 of
US
These
side.'
The
estimates here do not
parents and foreigners other than affiliates, or between foreign parents and unaffiliated
US. For 1989, the
ratio
of merchandise exports shipped by
(excluded) to exports shipped to foreign affiliates (included)
11
is 1.3;
US
parents to foreigners other than
the analogous ratio for imports is 1.1.
f^Min(X°Jc!j)
nx.
(2)
where X^' and X^° are inward
=
^
affiliate sales in industry j
from country
c
and outward
affiliate sales to
country c
respectively for the affiliate sales ratio, and total imports and exports respectively for the trade ratio (net of trade
mediated by
affiliates).
Table IB reports ratios averaged by industry, which are calculated analogously by
aggregating over countries within industries.
The
affiliate sales.
In both cases, the average ratio
is
higher for trade than
intra-industry ratio of affiliate sales exceeds 30 percent for 5 countries
UK, and
is
a 17 percent correlation between the two indices.
Transport Costs
ii.
Two
SO percent. There
is for
(UK, France, Germany,
Japan, and Canada) while that for trade exceeds 30 percent for 18 countries, with Mexico, France, the
Ireland exceeding
it
Most of
types of variables have been used to proxy for freight factors in past research.
work on
the
gravity equations uses measurements of physical land and sea distances between national 'economic centers" to
proxy for transport costs (Bergstrand, 1986, 1989), following the procedure outlined
more
industry level, a
OECD
on a fob
basis
import values reported on a cif basis by the importing country, and the associated value repmrted
by the exporting country.
I
The
results are disappointing,
which Harrigan
however, with freight factors for some
attributes in part to inconsistent reporting procedures.
use an alternative formulation of freight factors, which uses data on freight and insurance charges
reported by importers to the
values.
the
Harrigan (1993) approximates variations in freight factors by product, by using the ratio
industries exceeding 5(X) percent,
Below
Linneman (1966). At
accurate measure of transport costs should reflect specific product characteristics, as well
as geographical factors.
between
in
US
Census Bureau. The
Since no comparable data
is
available from exporters, and there
differences in charges for transporting the
transport, these values are used in the
industry trade (95 ftercent of
all
freight factors are calculated as the ratio
same goods between
outward equations for
all
the
is
no reason
of charges
import
to expect systematic
same locations based on
the direction of
industry /country pairs for which there
industry-country pairs), as well as in the inward equations for
12
to
all
is intra-
industries.
The
resulting series appears
in
more accurate than
either of
its
two predecessors, yielding values between
99.8 percent of the industry-country pairs for which freight factors are reported.
compared with a mean ranging between 140 and 270 percent
series.
The
series also
for different
It
has a
and 100 percent
mean of
8 percent, as
methods of correcting the OECD-based
seems reasonable, with high average values for countries such as Philippines and Singapore,
and for industries such as iron ore and concrete, asbestos, and cut stone, and low average values for countries such
as
Canada and Mexico, and
for industries such as electronic
components and
Country Variables
iii.
Data on national income and per capita income are taken from the
Detailed factor endowments data
capital-labor ratios
following
V.
scientific instruments.
and per worker
GDP
is
IMF
Financial Statistics.
constructed using the methodology outlined by
ratios are constructed using data
Bowen
from the Penn Mark
V World
Empirical Estimates
embedded
in differentiated products models.
However, the
first
same industry and increasing
in the similarity
between the markets
-
and income
distinguish them,
I first
multinational sales.
I
-
-
arising in
two directions
that multinational sales look
arising in a single direction and increasing in factor proportion differences.
To
apply the tests that have been used to assess the differentiated products model of trade to
then exploit the distinction between multinational sales destined for the local market and those
destined for export back to the
home market
in
an attempt to evaluate a factor proportions explanation of
multinational activity.
i.
are
and differently with respect to proximity
and concentration advantages. In contrast, the factor proportions explanation predicts
like inter-industry trade
activities
predicts that multinational sales behave like
differentiated products trade with respect to factor proportions differences
more
Tables
Hummels and Levinsohn.
Both the proximity-concentration and factor proportions explanations of multinational
in the
(1982), and
Comparing
the Total
Volume of Trade with
13
the Total
Volume of
Affiliate Sales (Table 3)
Helpman (1987) develops two
of 14
OECD countries. More recently,
by Hummels and Levinsohn. The
first
tests
of differentiated products models of trade and applies them to a sample
these have been replicated and extended to a sample of non-OECD countries
assumes
that all trade is in differentiated products.
proportions differences, the theory would predict that the volume of trade
This yields an equation explaining the
trading partners.
(3)
where
V^^ is the bilateral
volume of
total
^^
Ju^yJ 1
Y,*y,
Y^
trade, Y; is
income
world income. This equation
pairs over time.
A
is
volume of trade between countries u and
in country
i
and Y,
is
world income.
of the two countries and the greater
among
similar equation should apply to the total
it
i:
'
tested either for trade
concentration, differentiated products mode, but
determined by the relative size of
^*^'^
bilateral trade is greater the smaller is the difference in size
relation to
is
In the absence of factor
The volume of
is their size in
a group of countries or for a set of country
volume of multinational
would not explain multinational
sales in a proximity-
by
sales motivated
factor
proportions differences.
Figure
1
shows the
distribution of total
volumes of
affiliate sales
and trade and compares them
to
income.
Figure la shows the distribution by income. Both multinational activity and trade are skewed towards the countries
whose incomes are most
distribution
similar to the
by per capita income.
US, but
Both
the bias
affiliate sales
incomes and disproportionately so relative to income.
complicated here, since there
is
is
more extreme
for affiliate sales.
and trade are greater
The
Figure lb shows the
for countries with similar per capita
difference between affiliate sales and trade
is
more
disproportionate trade with the 5 countries with the highest per capita incomes but
also with the half with the lowest incomes.
Below
relying
is
on
I
modify equation
it
explains the bilateral volume of trade in particular products, by
the assumption that preferences are identical
a uniform elasticity of
demand
(3) so that
demand among
and homothetic. In the differentiated products model, there
different varieties of a product
for imported varieties within an industry should
and constant expenditure shares, so
depend similarly on
differences in industry elasticities by including dummies.
In addition,
I
relative country sizes.
allow for
include freight factors, which can easily
be incorporated into a pure differentiated products model of trade (Helpman, Knigman).
14
I
that the
The equation
to
be
estimated for affiliate sales
MTOTJ
(4)
where MTOTi^
is
is:
=
o^TGDP,*o^SHDIF,*o^FF|*Y^
the log of total affiliate sales (net of imports) between country
is
the log of the total
shares of country
income of country
i
and the US, and EV
i
and the US, SHDIFj
is
sectors, a proximity
by
model would predict a
to
be equal.
to differ.
affiliates),
rising
I
would
Past research
The same equation
GDP
TGDPj
minus the sum of the squared income
1
on
is
all activity is in
rises
trade
volumes
where the
then used to explain the
differentiated products
trade and multinational sales as the
as their joint
transport costs rise, while a factor proportions model
Below
TTOT/. When
volume of
becomes increasingly similar and
trading partners
of product j,
report results for the case
where they are allowed
trade (net of trade mediated
US
SHDIF
coefficients are constrained and
volume of
and the
j.'"
TGDP
and
the log of
is
i
a variable specific to industry
constrained the coefficients on
total
'^z.P'^^i,
income of the
and a declining volume of trade as
volume of trade
predict only a rising
as relative
income
shares converge.
The
affiliate sales
data suggests a strong similarity between the volumes of trade and affiliate sales: the volumes of gross
and gross trade flows have a correlation of 56 percent, although netting out trade mediated by
Columns
reduces the correlation to 33 percent.
affiliate sales net
ff3=ff4=0.
The
of trade using equation
coefficient
(4).
on RELSH, which
is
1
on
TGDP and SHDIF to differ.
through 3 of Table 3 report estimates of the volunw of
Column
1
reports an estimate of the equation with 0^=02 and
the product of
be predicted by a differentiated products model, and the
TGDP
and SHDIF,
The explanatory power of the equation remains
This result probably reflects collinearity between the
adds freight factors and industry dummies.
share differential rises to unity.
The
coefficient
Columns 4 through 6 of Table
'"
Trade flows mediated by
As explained above, however,
The
fit
on
is
highly significant, as would
Column 2
elasticity is close to 1.
permits the coefficients
constant, but all of the significance
resides in the share differential variable, with the elasticity slightly larger than
the first equation.
affiliates
on
the scaled share differential in
SHDIF and TGDP
variables.
Column
3
of the equation improves markedly, and the elasticity on the
the freight factor
is
significant at -0.35.
3 report analogous estimates of trade volumes net of internal transactions.
affiliates are netted
from both the
affiliate sales
the trade mediated by affiliates includes products
are not sensitive to the use of net versus gross flows.
15
and trade flows to avoid overlap.
from other industries. The results
In
the
column
same
4, the coefficient
on
the product of the share differential and joint
size as in the affiliate sales equation,
coefficients to differ in
and the
column S does not improve
the
The
addition of freight factors and industry
of the trade equation even more than the
the freight factor
is
TGDP is over twice
affiliate sales equation.
nearly twice that of affiliate sales and
In sum, the relative share variable performs as
is
The
is
significant
nearly identical.
is
of the equation. In contrast to the
fit
both variables retain significance, and the elasticity on the
probably reflect collinearity.
of the equation
fit
GDP, RELSH,
that
elasticity
Allowing the
affiliate sales equation,
on SHDIF, but again
dummies
in
and of
the results
column 6 improves
the
fit
of trade volumes with respect to
highly significant.
would be predicted by a
trade in explaining both the volumes of trade and of affiliate sales.
The
differentiated products
model of
finding that the responses of trade and
affiliate sales to the relative share variable are nearly identical suggests either that the predictions
of a differentiated
products model absent factor proportions differences or with specialization in the homogeneous product sector for
trade apply equally well to affiliate sales, or that the [wsitive relationship between relative
border transactions
is better
relative share variable
explained by a different model.
Similar to the findings in
income shares and cross-
Hummels, Levinsohn,
performs almost too well, considering that the group of countries includes those for
factor proportions are sufficiently different
multinational activity to prevail.
from the
US
that
In addition, although the
is
whom
one would expect inter-industry trade and one-way
volume of both
increasing transport costs, the elasticity of the trade response
the
affiliate sales
and trade contracts with
nearly twice that of affiliate sales, weakly consistent
with the proximity hypothesis.
ii.
I
Comparing
Intra-Industrv Ratios for Trade and Affiliate Sales (Table 4)
next turn to the second
test
developed by Helpman, which allows for the co-existence of differentiated
products trade and factor proportions trade.
predicts that the share of
two-way
The
differentiated products
model
in a factor proportions
framework
trade in total bilateral trade should increase with factor proportions similarities,
holding relative country size constant, and
and decrease with factor proportions
affiliate activity
should arise only in a single direction within an industry
similarities.
Figure 2a plots the Grubel-Lloyd index of intra-industry trade against a similar index of intra-industry
16
averaged across industries by country, which
affiliate sales
correlation between the
two
indices.
defined in equation
is
it is
somewhat stronger
is
a clear positive
Figures 2b and 2c plot each of the indices against income-per-worker ratios
by country, as a rough proxy of factor proportions differences. In both
but
There
(2).
for affiliate sales.
Although
cases, there
is
a clear negative relationship,
what would be predicted by a differentiated
this is precisely
products-factor proportions model for trade, the existence of intra-industry affiliate sales and their negative
relationship with per
worker income are strongly counter
the analogous intra-industry indices averaged across countries
The
figures suggest that freight factors have
little
to
this
Figures 3a and 3b plot
model.
by industry against average industry
no influ^ice on
negative relationship on intra-industry ratios of trade.
of
to the predictions
This
freight factors.
intra-industry ratios of affiliate sales
and a
roughly consistent with the predictions of the
is
proximity-concentration hypothesis.
Helpman and Hummels, Levinsohn
test the relationship
between factor proportions differences and
industry ratios of trade for a multilateral sample of country pairs over time.
fit
I
stick
with cross-section analysis and expand the equations to include freight factors.
equation for multinational sales
MJNTRaI
,j.
MINTRAjJ
is
US
their equations to
two-way
i:
in country
affiliate sales to total affiliate sales (net
differential in per capita
differentials.
i
and
MAXGOP; is the reverse.
the trading partners'
endowments of
estimating
as well as freight factors.
I
in
both directions in industry
US
is
is
the
minimum
the country with the highest
GDP
MAXGDP takes the place of the constant. CAP
capital ratios
high-skill workers,
illiterate
Since the
GDP, and
between per capita
measures per capita
In addition,
of exports)
MINTRA^J = 2*min(N0UTiJ,NINi>)/(N0UTj'-l-NINi*). MINGDPj
MINGDP is simply
LAB3
first
is:
the absolute value of the differential
differentials,
The
In both
niCAPi*iiJABli*njIAB2,+n^IAB3t+njLAND,*n^MINGDP,*it.,MAXGDP,
and country
US GDP and GDP
in the sample,
is
=
the ratio of
between the
of
modify
I first
the bilateral and industry-specific nature of the data and then present results for the country aggregates.
cases,
j
Here
intra-
LAB2
between country
i
and the US,
measures per capita unskilled,
labor differentials, and
LAND
LABI
is
the
literate labor
measures per capita arable land
allow for differences in industry responses by including industry-specific variables, D',
Analogous equations are used
to estimate
TINTRAjJ, the
to total trade flows (net of internal transactions) in both directions in industry j
17
ratio
of two-way trade flows
between the
US
and country
i.
The
first
three
columns of the upper panel of Table 4 report estimates of intra-industry shares of
MINTRA. Column
sales,
1
reports an estimate of equation (5). Intra-industry ratios of affiliate sales are decreasing
in differences in the relative
in
income
The
abundance of
skilled labor
and of
levels, but the coefficient
coefficient
on the
on
MAXGDP is
freight factor is negative but insignificant.
3 uses the capital-labor ratio differential,
the coefficients
trade
on
DKL.
significant, as predicted for trade, but those
is
are also decreasing in differences
endowments
worker income
contrary to prediction.
The
MAXGDP
on
coefficient
on
are insignificant.
differential,
of the equation
fit
the factor proportion differential proxies are negative and significant, as
The
in differences in
Following Hummels, Levinsohn, column 2
In both cases, the
by a factor proportions, differentiated products model.
which
They
Relative capital
insignificant.
replaces the factor proportions variables with a single measure of the per
column
weakly decreasing
illiterate labor,
unskilled labor, and increasing in arable 'and ratio differentials.
literate,
affiliate
on
coefficients
falls
DGPL, and
moderately, but
would be predicted
MINGDP
for
are both positive and
are positive and significant in the equation with
DKL,
freight factors remains negative, but is significant only in the
equation with the per worker income differential.
Columns 4 through 6 of
trade.
The
the upper panel of Table 4 report analogous estimates of intra-industry ratios of
signs and magnitudes of the coefficients on the factor pro()ortions differentials are similar to the affiliate
sales equations, but the significance
labor
is
lower.
Compared
ratios is explained.
DGPL, and column
Column
of the coefficient on
literate,
to the affiliate sales results, a smaller fraction
on
DKL.
three equations, the coefficient
wrong
on high
skilled
Similar to the affiliate sales
the factor proportions differential variables are negative and significant, consistent with
on
MINGDP
is
is
half that of affiliate sales in both cases.
sign in the equations with
DKL
equations, the coefficient on freight factors
and
is
In
positive and significant, as predicted, and the magnitude of the
coefficient is less than half that in the corresponding equation for affiliate sales.
the
greater and that
of the variation in intra-industry trade
ratio differential,
a differentiated products model, but the magnitude of the response
all
is
5 replaces the factor proportions differentials with the per worker income differential,
6 replaces them with the capital-labor
equation, the coefficients
unskilled labor
DGPL
and
is
18
coefficient
significant in the equation with
negative and significant and
affiliate sales equations.
The
much
on
MAXGDP has
DKL.
In all three
larger than that in the parallel
Roughly similar
is
used instead, so
results obtain for the signs
do not report them here.
I
and significance of the coefficients when a
tobit specification
Overall, the results of both equations are consistent with the
predictions of a differentiated products, factor proportions model for trade, modified to incorporate transport costs,
but the affiliate sales equation
Moreover, there
that
is
squarely
at
odds with the predictions of the same model for
affiliate sales.
a surprising similarity between the responses of the shares of intra-industry affiliate sales and
is
of trade to differences in particular factor proportions. Indeed, intra-industry shares of affiliate sales and trade
appear to differ most in their responses to freight factors and income differences.
The lower panel of Table 4
reports estimates comparable to those in
Helpman and Hununels, Levinsohn,
which use as the dependent variable the Grubel-Lloyd coimtry aggregate intra-industry
find that the intra-industry index of trade
ratio.
Both
sets
had the predicted negative relationship to variables measuring factor
proportions differences two decades ago, but conclude that the relationship has since deteriorated.
Hummels, Levinsohn
find that
The
contrary to theory.
Columns
and 4 use the
fiill
pair effects are controlled for, the relationship
and scales by the
total
volume of
CMINTRA,
which
is
becomes
positive,
intra-industry trade across
trade, as in equation (2).
set
and columns 4 through 6 repeat the same
tests for trade,
CTINTRA. Columns
of factor differentials, while columns 2 and S include only the per worker income
and columns 3 and 6 include only the capital -labor
factor,
volume of
In addition,
through 3 of the lower panel of Table 4 report results for the aggregate intra-industry ratio of
1
multinational sales,
when country
intra-industry index for each country aggregates the
industries within a country
of authors
ratio differential.
1
differential,
Each equation includes a coimtry
freight
averaged across industries."
Similar to the industry-level results, the coefficients on both types of literate labor are significantly negative
in the affiliate sales equation in
be attributable
MINGDP
is
I
1.
The
coefficient
on the
capital differential is insignificant, but this
to collinearity with other factor variables, similar to the industry equations.
as predicted and significant, while those
of the equation
"
column
is
very good
-
on
around 78. Replacing the
MAXGDP and
list
FF
In addition, the sign of
are positive but insignificant.
19
The
fit
of factor differentials with a single income per worker
experimented with both weighted and simple averaging; the results are not sensitive so
weighted average.
may
I
report only the
DGPL,
differential,
results,
in
column
with some sacrifice in
and a
2,
capital labor differential,
DKL,
in
column
3 yields surprisingly similar
Similar results are also obtained using per capita income differentials, so
fit.
I
do
not report them here.
The
estimates of the intra-industry shares of trade in columns 7 through 9 are very similar to the
Levinsohn findings (with the exception of the freight
factor).
differentials are largely insignificant, with the exception
and arable land (which
However, the
is positive).
coefficient
on
Surprisingly, the
the freight factor
is
In
all
three equations, the coefficients
of unskilled,
MINGDP
and
literate labor
(which
on
Hummels,
the factor
marginally negative)
is
MAXGDP variables do not fare
negative and significant, as would be expected.
much
These
better.
results are
robust to the use of per capita income differentials.
The
of the intra-industry share equations together with the
results
of multinational activity
that a significant part
of factors, rather than differences, which
is
of country-wide intra-industry ratios suggest
than
and trade flows are
fairly similar,
even
complementarity. The similarity breaks
both
total
inconsistent with a factor proportions account.
that the predictions
Indeed, the estimates
of the differentiated products model for trade
down
mediated by
affiliates,
which may
in the responses to increases in freight factors,
much
Comparing
The two
tests
Affiliate Production for
however. While
affiliate sales
Export and Local Sale (Tables 5 and 6)
above do not rule out a factor proportions explanation of multinationals; they simply establish
does not account for a significant amount of multinational
would be
careful test of the
affiliate sales
smaller magnitude, and intra-industry ratios are essentially unaffected.
iii.
differences
fits
in part explain
flows and intra-industry ratios of trade diminish significantly with increasing transport costs,
diminish by a
it
motivated by similarities in income and proportions of certain types
after controlling for trade
their
that
volume equations make a strong case
does trade. The roles of relative incomes and income shares in determining
affiliate sales better
it
is
total
tested using data
HOS
activity.
of factor proportions
from input-output matrices and factor endowments, along the
hypothesis for trade in Bowen, Learner, and Sveikauskas (1987).
industry categories do not permit the degree of vertical separation that
which would require
Ideally, the role
would be necessary
lines
of the
Unfortunately, the
for such a rigorous test,
factor intensities data not only for individual products, but also for separate activities within
20
each product's business system.
activities
However, some
light
based on whether their primary market
distinction
between horizontal
cost differences.
Below
I
can be shed on this issue by distinguishing between
by proximity advantages and
activity motivated
which corresponds roughly
local or export-oriented,
is
vertical activity motivated
produced abroad are exported back
to the headquarters market,
A
significantly different factor proportions.
by factor
be consistent with overseas production destined for export
as the category
most
all varieties
of a
final
good
and such activity arises between countries with
proximity-concentration explanation predicts that overseas production
substitutes for trade in final goods, so that exports back to the
home
to a
exploit this distinction in an attempt to evaluate the factor proportions hypothesis.
Recall a factor profwrtions explanation of multinational activity predicts that
export
affiliate
home market should be
to third markets,
so
I
zero.
Both explanations could
focus on affiliate sales destined for
likely to reflect factor proportions considerations
and contrast
it
with local sales
as the least likely category.
Table 5 reports the shares of
affiliate
production destined for the local market and for export back to the
parent market for inward and outward sales, averaged by country and by industry.
ratios suggests that for the
directions, but particularly
affiliate sales
US, multinational
on the inward
sales account for
92 percent of total
sales are destined primarily for the destination
On
side.
on aggregate, while exports back
ratio
with the exception of Canada.
On
US
of sales destined for sale back
side,
the inward side, only
percent of their production to their parents.
The
side,
and iron ore on the inward
side.
Figure 4a plots the share of inward
of
illiterate
inward
which
fits
Many
home
is
the inward side, local
highest for Singapore,
Hong
a factor proportions explanation fairly well,
Taiwanese and Brazilian
wood
affiliates
export more than 5
home market
products, and other transportation equipment on the
of these industries are resource intensive.
affiliate sales
same
On
to the foreign parent account for only
exported to the parent against the difference in the ratio
labor to unskilled literate labor in the source market relative to the
affiliate sales sold locally against the
market in both
industries with the greatest ratios of exports to the
are iron ores, nonmetallic nonfuel minerals, lumber and
outward
account for only 13 percent.
on aggregate, while exports
Kong, Canada, Taiwan, and Mexico on the outward
the aggregate
the outward side, local sales account for 64 percent of total
to the
affiliate sales
2 percent. Broken down by country, the
A simple glance at
US
factor proportions differential.
21
ratio,
and 4b plots the share of
The share exported
to the parent
market exhibits a clear positive relationship to the factor proportion
relationship.
differential,
while the local share has a negative
Similarly, figure 4c plots the share of outward affiliate sales exported to the
in the ratio of unskilled literate labor to skilled labor in the destination
the share sold locally.
Again, the share exported
home
market relative
US
to the
against the difference
US
and 4d plots
ratio,
increasing in factor proportions differences, while here
is
the share sold locally appears uncorrelated.
I
next present equations that examine these relationships more rigorously.
importance of factor proportions differentials in determining
with their importance in determining
(6)
OXShI
where
all
=
A
destined for export to the
the
home market
affiliate sales sold locally:
X^CAP|*XJLABl^*k^B2^*X^LAB3^*kJHND|*)i^GDP^*k^USGDP*X^F|*ef,
the independent variables are in logs, and
back to the US.
affiliate sales
The equations compare
OXSH
is
the share of total outward sales comprised of exports
similar equation explains the share of outward sales destined for the local market,
OLSH.'^
Table 6 reports estimates for equation (6) for the share of outward sales destined for the local market
column
1
and the share of
sales destined for export
back
to the
US
in
column
share accounted for by local sales decreases in differences in per capita
arable land, while
it
increases in differences in
the share accounted for by exports back to the
illiterate labor,
and decreases
of activity the greater
is
endowments of
US
endowments of
literate unskilled
and
capital, skilled labor,
illiterate labor.
increases in differences in per capita
in differences in unskilled, literate labor.
destination market
2, using a tobit specification.
in
The
and
In contrast,
endowments of
capital
and
In addition, local sales are a larger share
income and the greater are transport
costs, consistent with a proximity-
concentration hypothesis, while sales for export are unaffected by income levels and
fall
with increases in transport
costs.
Estimates of the levels of each type of sales suggest that the difference in responses to factor proportions
differences
is
one of degree rather than kind. Column 3 reports an estimate of the log of the
destined for the local market,
level of
becomes
Gross flows are used
of outward sales
using equation (6), and column 4 reports the same equation for the log of the
outward sales destined for export
capital differentials
'^
OUTL,
level
to the
US, OUTX. Compared
insignificant in the local sales equation,
in this section.
22
to the share estimates, the coefficient
and the coefficient on tmskilled
on
literate labor
differentials
becomes
The two flows
insignificant in the export sales equation.
differ
most
magnitude of
in the
their
responses to destination market income and to freight factors, with local sales three times more responsive to
income, and export sales three times more responsive to freight factors.
Comparable estimates
for
inward sales are reported
columns 5 through 8 of Table
in
a tobit estimate of the share of inward sales destined for sale in the
US, INLSH, based on equation
6 reports an estimate of the share destined for export back to the parent,
response to increases in unskilled,
the
US
on
moderately abundant in unskilled
is
likely to
literate labor differentials,
be destined for export back to the parent the poorer
sales:
falls.
on outward
INXSH.
The
(6),
reports
and column
shares differ in their
with local sales falling and export sales rising. Since
literate labor, this
the freight factor are consistent with the results
Column 5
6.
implies that affiliate production in the
is
US
the parent's market in this resource.
sales
is
The
above and with a proximity hypothesis
results
for local
the share destined for local sale increases with the freight factor, while the share destined for export
The
the size of the
equation
on income are inconsistent, however: the share exported
coefficients
is
home
inward export
sales,
decreasing in the
INX,
illiterate
in
The
level equations for
colunm 8 provide
little
inward local
additional insight.
sales,
The
The
fit
INL,
in
of the export share
column
7,
and for
levels of both types of sales are
labor differential; given the relative scarcity of this resource in the
US,
this implies that
countries which are relatively abundant in illiterate labor are not fertile ground for corporate headquarters.
coefficient
on
GDP
is
significantly positive in both equations,
and
that
on
home
to the source country increases with
country market, while the share sold locally actually decreases.
particularly poor, however.
more
freight factors
is
The
significantly negative,
although smaller in the local sales equation.
A few inferences can be drawn from the two sets of estimates.
in high-skill labor differentials (to countries scarce in high-skill labor)
(to countries
abundant in
illiterate labor).
Outward
Outward
sales
and increasing
sales destined for export
of both types are decreasing
in illiterate labor differentials
back to the
US
differ
from those
destined for local sale in that they are increasing in capital and arable land differentials (to countries with either
extreme scarcity or abundance of capital and scarce
in arable land)
differentials (to countries scarce in unskilled, literate labor).
illiterate
and decreasing
in unskilled, literate labor
Inward sales of both types are weakly decreasing in
labor differentials (from countries abundant in illiterate labor) and weakly increasing in arable land
23
differentials
(from countries scarce in arable land). Inward sales destined for export back to the parent differ from
those destined for local sale in that they are increasing in unskilled, literate labor differentials (from countries scarce
in this resource).
relatively
poor
This implies
in capital,
multinationals that use the
US
multinationals use countries that are relatively abundant in illiterate labor and
arable land, and unskilled, literate labor as offshore production bases.
US
as a base for exports back
home
tend to be relatively poor in unskilled, literate labor
(perhaps suggesting they are locating a different part of the production process in the
abroad).
deficit
Further, the
US
results provide
destined for local sales.
US compared
has a net surplus of sales with countries relatively abundant in
with countries relatively poor in arable land.
The
some evidence
illiterate
production destined for export back
home
in factor proportions differences.
The
The share of both inward and outward
while the export shares decreases.
outward equations support the hypothesis
that the
income
And
tests also
that
is
market
support the
home
are
more
sales destined for the
the coefficients
level of the destination
determinant of local sales than of export sales, but this result
VI.
from
for the factor proportions hypothesis in explaining the share exported
similar to trade in their response to freight factors.
in freight factors,
affiliates
labor and a net
differs
predictions of the proximity hypothesis for local sales, while affiliate sales destined for export
market increases
US
-
back home, while local sales appear to be decreasing
local
to
,
that affiliate
They provide support
Foreign
is
a
on income
in the
more important
not robust in the inward equations.
Conclusion
The
data clearly rejects a pure factor proportions explanation of multinational activities, although factor
proportions appear to explain
some portion of these
activities.
Total volumes of affiliate sales (net of internal trade)
are strongly increasing in similarities in relative income shares, as
identical factor proportions,
in affiliate sales.
trade, they are
and
this variable together
would be predicted
for trade in a
model with
with transport costs explains over 40 percent of the variation
Further, although intra-industry ratios of multinational sales are lower on average than those for
still
by factor proportion
significant,
and the variation in intra-industry
similarities
and relative incomes than
factor proportions explanation of multinationals,
is that
and suggest
ratios
of multinational sales
better explained
of trade. Both findings are inconsistent with a pure
that a substantial part
motivated by similarities rather than differences in factor proportions and incomes.
24
is
of multinational activities
Some evidence
is
for the factor
proportions account
is
derived by distinguishing affiliate production destined for sale in the parent's
US
foreign affiliate production destined for export back to the
and unskilled,
literate labor,
and abundant
in illiterate labor,
is
home
market:
highest in markets scarce in capital, arable land,
while
US
affiliate
production destined for export back
to the foreign parent is highest for countries that are scarce in unskilled, literate labor.
Interestingly, both the tests
on
intra-industry ratios and total
strong dampening effect on trade flows, and a
much
volumes confirm
that freight factors
smaller or insignificant effect on affiliate sales.
have a
In addition,
distinguishing affiliate production destined for export from that destined for local sale reveals an important difference
in their responses to transport costs:
while export sales diminish as freight factors
rise, local sales are either
unaffected or increase.
Taken
activity.
together, these results provide only
weak evidence of factor proportions motivations for
Factor proportion differences are strongest in explaining the f)ortion of affiliate sales that
multinational
is
destined for
export back home, which accounts for 13 percent of foreign affiliate production and between 2 and 8 percent of
affiliate
US
production, while local affiliate sales are decreasing in differences in capital and skilled labor endowments.
In addition, the results
on transport
costs shed
concentration hypothesis (Brainard, 1993).
some
light
on findings
in a
companion paper
That paper finds that the share of
total
testing the proximity-
flows accounted for by
affiliate
sales is increasing in freight factors
and both the share and level of trade are decreasing, consistent with the
proximity-concentration hypothesis.
However, the
robust.
This finding
may be
partially explained
effect
of freight factors on the level of
by the presence of some
affiliate sales is
vertical affiliate activity,
which would be
consistent with the negative relationship between transport costs and affiliate sales destined for export to the
market documented above.
25
not
home
Table lA! Affiliate Sales and Trade Flows, by Country
Net of Trade Mediated by Affiliates'
64 Tradeables Industries
1989, $ Millions
Table IB;
Affiliate Sales and Trade Flows, by Industry
Variable Definitiona
MTOT
TTOT
Sum of affiliate sales (net of trade mediated by affiliates)
Sum of trade flows (net of mediated by affiliates)
MINTRA
TINTRA
Intra-industry ratio of affiliate sales (net of mediated trade)
Intra-industry ratio of trade (net of mediated trade)
FF
Freight factor: transport cost as percent of value
TGDP
SHDIF
RELSH
Total GDP of trade partners
Income share differential
TGDP*SHDIF
DPCY
DKL
CAP
LABI
LAB2
LAB3
LAND
Differential
Differential
Differential
Differential
Differential
Differential
Differential
OUTX
OUTL
OUTXSH
OUTLSH
Outward affiliate sales destined for export back to US
Outward affiliate sales destined for local sale
OUTX /OUT
OUTL /OUT
INX
INL
INXSH
INLSH
Inward affiliate sales destined for export back to parent
Inward affiliate sales destined for local sale
INX/IN
INL/IN
in
in
in
in
in
in
in
per capita income
capital/labour ratio
per capita capital endowment
per capita endowment of skilled labour
per capita endowment of literate, unskilled labour
per capita endowment of illiterate labour
per capita endowment of arable land
28
Table 2A;
Inward
(obs=1731)
1
Correlations
Table
3;
Total Volume of Trade and Affiliate Sales
Affiliate Sales
MTOt/
=
a^TGDPi + a23HDIFi*o^FF/*Y^*
a^,jD^->-\iij
Trade
TTOr/
DEP VAR
relsh
=
mvol
e^TGDPi*djSHDIFi*djFFi^*Y^* Qj^jD^t-e^j
mvol
mvol
tvol
tvol
tvol
—
Table
A.
Intra-induatrv Ratios of Trade and Affiliate Sales
Industry Ratios
Affiliate Sales
MINTRa/
B.
4;
=
nJ^CAP^+K2LABli+n^LAB2+n^LAB3^^n^LAND^+n^MINGDP^+n^MAXGDPJ^
Trade
TINTRA/
= Pj^CAP^ + p^LABli + PjLAB2^*p^LAB3i+py^LAND^*piMINGDP^ + p^MAXGDP^
————————— T— —————————
DEP VAR
mintra
_________J
1
Table
4;
Intra-industrv Ratios of Trade and Affiliate Sales
Country Aggregates
cmintra
DEP VAR
^^__^^^^^^
1
Table
5;
OUTWARD
LOCAL
COUNTRY RATIO
TOTAL
Share of Affiliate Sales Sold Locally and Exported Home
64 Tradeables Industries, 27 Countries, 1989
OUTWARD
EXPORT
HOME
RATIO
INWARD
LOCAL
RATIO
INWARD
EXPORT
HOME
RATIO
BEA
OUTWARD
LOCAL
RATIO
OUTWARD
EXPORT
HOME
RATIO
INWARD
INWARD EXPORT
LOCAL
HOME
RATIO RATIO
Table
Factor Proportions and Local and Export Affiliate Sales
6;
Outward, Levels and Shares
A.
OLSHi
= ^Ji^CAP^*u>^LABl^^lJi^LAB2^*^li^LAB3^*utc,LAND^*lj>^GDP^*<Ji^USGDP^(,i^FFl*\i{
OXShI
=
A,iCAPj+X2LASlj+X3LAB2j + A,^LAS3j+A,5LAWDj+A,5GZ?Pj+A.,£/SGi3P+XgFF/+e^
Inward, Levels and Shares
B.
ILSHi
=
K^CAP^+KjLABlj+KjLAB^^+K^LABJj+KsLAiVDj+KjGriPj+ICjaSGDP+KgFF/'+lA^
IXSh/
=
x^CAP^*X2LABl^+t^LAB2^*x^LAB3^+x^LANDitZ^GDP^+:^USGDP+x^FF/+e{
TOBIT
DEP VAR
_________^
1
outlsh
TOBIT
OLS
OLS
TOBIT
TOBIT
OLS
OLS
Appendix;
010
020
080
090
101
102
107
120
133
140
201
202
203
204
205
208
209
210
220
230
240
250
262
265
271
272
275
281
283
284
287
289
291
299
305
308
310
321
329
331
335
341
342
343
349
351
352
353
354
355
356
357
358
359
363
366
367
369
371
379
381
384
386
390
BEA Industry Definitions
Crops
Livestock, animal specialties
Forestry
Fishing, hunting, trapping
Iron ore
Copper, lead, zinc, gold, silver
Other metallic ores
Coal
Crude petrol extraction, natural gas
Nonmetallic minerals, except fuels
Meat products
Dairy products
Preserved fruits and vegetables
Grain mill products
Bakery products
Beverages
Other food and kindred
Tobacco products
Textile mill products
Apparel and other textile products
Lumber and wood products
Furniture and fixtures
Pulp, paper, board mill products
Other paper and allied products
Newspapers
Miscellaneous publishing
Commercial printing and services
Industrial chemicals and synthetics
Drugs
Soap, cleaners, toilet goods
Agricultural chemicals
Chemical products, nee
Integrated petroleum refining and extraction
Petroleum and coal products, nee
Rubber products
Miscellaneous plastics products
Leather and leather products
Glass products
Stone, clay, concrete, gypsum, other nonmetallic mineral
products
Primary metal products, ferrous
Primary metal products, nonferrous
Metal cans, forgings, stampings
Cutlery, hardware, screw products
Heating equipment, plumbing fixtures, structural
metal products
Metal services; ordnance; fabricated metal products, nee
Engines, turbines
Farm and garden machinery
Construction, mining, and materials handling machinery
Metalworking machinery
Special industrial machinery
General industrial machinery
Computer and office equipment
Refrigeration and service industry machinery
Industrial and commercial machinery, nee
Household appliances
Household audio, video, communications equipment
Electronic components and accessories
Electrical machinery, nee
Motor vehicles and equipment
Aircraft, motorcycles, bikes, spacecraft, railroad
Measuring, scientific, optical instruments
Medical and ophthalmic instruments and supplies
Photographic equipment and supplies
Miscellaneous manufacturing
35
DISTRIBUTION OF US TRADE & AFFILIATE SALES
BY PARTNER'S GDP RANK
1.2
TRADE
GDP
^5 Highest
1
2nd 5
^ 3rd 5
AFFIL.
SALES
EZLower 1
Figure la
DISTRIBUTION OF US TRADE & AFFILM TE SALES
BY PARTNER'S PER CAPTTA INCOME RANK
1.2
TRADE
GDP
^5 Highest
\2nd5
^3rd5
Figure lb
36
AFFIL SALES
^Lower 12
INTRA -INDUSTRY RATIOS
OF TRADE AND AFFILIA TE SALES
Slope 0.63
ven
0.1
Figure 2a
0.3
0.2
0.4
0.5
Intra - Industry Affiliate Sales
INTRA -INDUSTRYAFFILIATE SALES RATIOS
By Country, 1989
Slope
15
Figure 2b
20
-&9e-6
30
Thousands
Per Worker Income Differential
INTRA -INDUSTRY TRADE RA TIOS
By Country, 1989
0.7
fra
0.6
ft;
J"
J
-
0.5
0.4
0.3
I
a
0.2
^
0.1
b
Figure 2c
35
TWO-WAYTRADE RATIOS BY INDUSTRY
AVERAGED OVER COUNTRIES
0.8
101
^
0.4
^
0.3
386
331
30tn
o a2
^
102
0.1
202
D3
a2
0.1
0.3
0.4
0.5
Frei^t Factor
Figure 3a
2-WAYAFFILIATE SALES RATIOS BY INDUSTRY
AVERAGED OVER COUNTRIES
0.9
101
^
0.8
-
a?
S
0.7
<^ 0.6
Si
^
36721QQg
as
^
5S;0.4
^0.3
^
102
3^
a2
140 341
139-
Slope -0.04
54
I
I
f-
0.1
m
20299^,
0.1
Figure 3b
120
133
0.2
0.3
Freight Factor
58
0.4
0.5
3JVl{S
Sn
OJ
3WOff jjodxj
6C
References
Blomstrom, Magnus, Robert Lipsey, and Ksenia Kulchycky (1988): "U.S. and Swedish Direct Investment and
Exports," in Robert Baldwin (ed.), Trade Policy Issues and Empirical Analysis (Chicago: University of
Chicago Press).
Bowen, Harry (1982):
"Statistical
Advantage (Cambridge:
Appendix B,"
MIT
in
Edward
Learner,
Sources of Intemtional Comparative
Press).
Bowen, Harry, Edward Leamer, and Leo Sveikauskas (1987), "Multicountry, Multifactor Tests of
Abundance Theory," American Economic Review pp. 791-809.
the Factor
,
"A Simple Theory of Multinational Corporations and Trade with a Trade-off between
MIT Sloan Working Paper No. 3492-92-EFA.
Brainard, S. Lael (1992):
Proximity and Concentration,"
Brainard,
S.
Lael
(1993):
"An Empirical Assessment of the Proximity/Concentration Tradeoff between
MIT Sloan Working Paper No. 3625-93-EFA.
Multinational Sales and Trade,"
Bureau of Economic Analysis (1990), Foreign Direct Investment
1989 Estimates US Department of Commerce.
in the
US: Operations of
US
Affiliates.
Revised
.
Bureau of Economic Analysis (1990),
US
Direct Investment Abroad. Revised 1989 Estimates .
US
Department of
Commerce.
Multinational Enterprise and Economic Analysis (Cambridge:
Caves, Richard (1982):
Cambridge University
Press).
Deardorff,
Alan (1984): "Testing Trade Theories" in Ronald Jones and Peter Kenen
Economics (Amsterdam: North-Holland).
(eds.).
Handbook of
International
Dunning, John (1988):
Explaining International Production (London:
Ethier, Wilfred (1986):
"The Multinational Firm, " Quarterly Journal of Economics November, pp. 806-33.
.
James Markusen (1991):
Working Paper No. 3825, August.
Ethier, Wilfred, and
Ethier, Wilfred,
Harper Collins).
"Multinational Firms, Technology Diffusion, and Trade,"
NBER
and Henrik Horn (1990), "Managerial Control of International Firms and Patterns of Direct
Investment," Journal of International Economics pp. 25-45.
,
Grubert, Harry, and John Mutti (1991), "Taxes, Tariffs, and Transfer Pricing in Multinational Corporate Decision
Making," The Review of Economics and
Harrigan, James (1993),
Economics
"OECD
Statistics .
1991, p. 285.
Imports and Trade Barriers in 1983," forthcoming, Journal of International
.
Helpman, Elhanan (1984): "A Simple Theory of International Trade with Multinational Corporations," Journal of
Political
Economy
,
p. 451.
Helpman, Elhanan, and Paul Knigman (1985): Market Structure and Foreign Trade (Cambridge:
40
MIT
Press).
Helpman, Elhanan (1987):
Imperfect Competition and Interaational Trade:
Countries," Journal of the Japanese and International Economies .
Horst,
1,
Evidence from
14 Industrial
pp. 62-81.
(1972): "The Industrial Composition of US Exports and Subsidiary Sales
American Economic Review
Thomas
to the
Canadian Market,"
.
A
Horstmann, Ignatius, and James Markusen (1987): "Licensing vs. Direct Investment:
by the MNE," Canadian Journal of Economics p. 464.
Model of Intemalisation
,
Horstmann, Ignatius, and James Markusen (1992): "Endogenous Market Structures
of International Economics 32, pp. 109-129.
in International Trade," Journal
.
Hummels, David, and James Levinsohn( 1993): "Monopolistic Competition and
the Evidence," NBER Working Paper No. 4389.
International Trade: Reconsidering
Kravis, Irving, and Robert Lipsey (1982), "The Location of Overseas Production and Production for Export for
U.S. Multinational Firms, " Journal of International Economics
Julius,
DeAnne
(1990):
Krugman, Paul (1983):
Global Companies and Public Policy (London:
.
12, pp. 201-223.
Royal
Institute
of International Affairs).
"The "New Theories" of International Trade and the Multinational Enterprise,"
in
D.B.
Audretsch and Charles Kindleberger (eds.), The Multinational Corporation in the 1980s (Cambridge:
MIT
Press).
Lipsey, Robert, and Merle Yahr Weiss (1981), "Foreign Production and Exports in Manufacturing Industries,"
Review of Economics and
Statistics . 63, pp.
The
488-94.
"Multinationals, Multi-plant Economies, and the Gains from Trade," Journal of
Markusen, James (1984):
International Economics 16, pp. 205-26.
.
Raff, Horst (1992):
"Foreign Direct Investment versus Exporting
when Product
Quality
is
Unknown," mimeo,
Universite Laval.
Rugman, Alan
(1988):
International
"The Multinational Enterprise,"
in
Ingo Walter and Tracy Murray (eds.), Handbook of
Management (New York: Wiley).
Swedenborg, Birgitta (1979): The Multinational
Operations of Swedish Firms
.
(Stockholm:
Utredningsinstitut).
41
;
2237 037
-
Industriens
MIT LIBRARIES
3
TOaO OOflMBlOM
fl
Date Due
SEP.
2
JUL
^3^1
I2«»
Lib-26-67