Escolar Documentos
Profissional Documentos
Cultura Documentos
Multinational sales have grown at high rates off.3 We introduce heterogeneous rms into a
over the last two decades, outpacing the remark- simple multicountry, multisector model, in
able expansion of trade in manufactures. Con- which rms face a proximity-concentration
sequently, the trade literature has sought to trade-off. Every rm decides whether to serve a
incorporate the mode of foreign market access foreign market, and whether to do so through
into the “new” trade theory. This literature rec- exports or local subsidiary sales. These modes
ognizes that rms can serve foreign buyers of market access have different relative costs:
through a variety of channels: they can export exporting involves lower xed costs while FDI
their products to foreign customers, serve them involves lower variable costs.
through foreign subsidiaries, or license foreign Our model highlights the important role of
rms to produce their products. within-sector rm productivity differences in
Our work focuses on the rm’s choice be- explaining the structure of international trade
tween exports and “horizontal” foreign direct and investment. First, only the most productive
investment (FDI). Horizontal FDI refers to an rms engage in foreign activities. This result
investment in a foreign production facility that mirrors other ndings on rm heterogeneity and
is designed to serve customers in the foreign trade; in particular, the results reported in
market.1,2 Firms invest abroad when the gains Melitz (2003).4 Second, of those rms that
from avoiding trade costs outweigh the costs of serve foreign markets, only the most productive
maintaining capacity in multiple markets. This engage in FDI. 5 Third, FDI sales relative to
is known as the proximity-concentration trade- exports are larger in sectors with more rm
heterogeneity.
Using U.S. exports and af liate sales data
* Helpman: Department of Economics, Harvard Univer- that cover 52 manufacturing sectors and 38
sity, Cambridge, MA 02138, Tel Aviv University, and countries, we show that cross-sectoral differ-
CIAR (e-mail: ehelpman@harvard.edu); Melitz: Depart- ences in rm heterogeneity predict the compo-
ment of Economics, Harvard University, Cambridge, MA sition of trade and investment in the manner
02138, National Bureau of Economic Research, and Centre
for Economic Policy Research (e-mail: mmelitz@ suggested by our model. We construct several
harvard.edu); Yeaple: Department of Economics, Univer- measures of rm heterogeneity, using different
sity of Pennsylvania, 3718 Locust Walk, Philadelphia, PA data sources, and show that our results are ro-
19104, and National Bureau of Economic Research (e-mail: bust across all these measures. In addition, we
snyeapl2@ssc.upenn.edu). The statistical analysis of rm-
level data on U.S. Multinational Corporations reported in con rm the predictions of the proximity-
this study was conducted at the International Investment concentration trade-off. That is, rms tend to
Division, U.S. Bureau of Economic Analysis, under an substitute FDI sales for exports when transport
arrangement that maintained legal con dentiality require-
ments. Views expressed are those of the authors and do not
necessarily re ect those of the Bureau of Economic Anal-
3
ysis. Elhanan Helpman thanks the NSF for nancial sup- See, for example, Horstmann and Markusen (1992), S.
port. We also thank Daron Acemoglu, Roberto Rigobon, Lael Brainard (1993), and Markusen and Anthony J. Ven-
Yona Rubinstein, and Dani Tsiddon for comments on an ables (2000).
4
earlier draft, and Man-Keung Tang for excellent research See also Andrew B. Bernard et al. (2003) for an alter-
assistance. native theoretical model and Yeaple (2003a) for a model
1
See Wilfred J. Ethier (1986), Ignatius Horstmann and based on worker-skill heterogeneity. James R. Tybout
James R. Markusen (1987), and Ethier and Markusen (2003) surveys the recent micro-level evidence on trade that
(1996) for models that incorporate the licensing alternative. has motivated these theoretical models.
2 5
We therefore exclude “vertical” motives for FDI that This result is loosely connected to the documented
involve fragmentation of production across countries. See empirical pattern that foreign-owned af liates are more
Helpman (1984, 1985), Markusen (2002, Ch. 9), and Gor- productive than domestically owned producers. See Mark E.
don H. Hanson et al. (2002) for treatments of this form of Doms and J. Bradford Jensen (1998) for the United States
FDI. and Sourafel Girma et al. (2002) for the United Kingdom.
300
VOL. 94 NO. 1 HELPMAN ET AL.: EXPORT VERSUS FDI 301
6
Our controls include the log of capital (book value net
7
of depreciation) per worker, this variable squared, and Part of the cost difference fI 2 fX may also re ect some
4-digit industry xed effects. Controlling for material usage of the entry costs represented by fE, such as the initial cost
intensity does not change the results. of building another production facility.
302 THE AMERICAN ECONOMIC REVIEW MARCH 2004
X D
«21
wj higher than the xed costs of domestic produc-
(1) f I . ~ t ij ! « 2 1 f X . f D . tion, fD. The pro t function pijX is steeper than
wi
both piD and p jI due to the trade costs tij. To-
These conditions will be clari ed in the follow- gether with the rst inequality in (1), these
ing analysis. relationships imply that exports are more prof-
8 11
Asi is well known, our utility function implies that Ai 5 This will be the case so long as the numeraire good is
bEi/[ n0 pi (v)12 « dv], where Ei is the aggregate level of produced in every country and freely traded.
12
spending in country i, ni is the number (measure) of vari- Note that the demand function Aip2 « implies output
eties available in country i and pi (v) is the consumer price i
A (a/a)2« when the price is a/a. Under these circumstances,
of variety v. costs are aAi (a/a)12 « , while revenue is Ai (a/a)12 « . There-
9
We exclude the possibility of exports by foreign af l- fore, operating pro ts are piD 5 (1 2 a) Ai (a/a) 12 « 2 fD.
13
iates. See, however, the Appendix of our working paper, We thank Dani Tsiddon for proposing this gure. In
Helpman et al. (2003), for a discussion of this possibility. the gure fX . fD, which is a suf cient condition for the
10
In a dynamic model with uncertainty, an individual second inequality in (1). Evidently, this inequality can also
rm may choose to serve a foreign market through both be satis ed when fX , fD, and we need only the inequality
exports and FDI. Rafael Rob and Nikolaos Vettas (2003) in (1) in order to ensure that some rms serve only the
provide a rigorous treatment of this case. domestic market.
VOL. 94 NO. 1 HELPMAN ET AL.: EXPORT VERSUS FDI 303
itable than FDI for low-productivity rms and equalized (the numeraire outside good is pro-
less pro table for high-productivity rms. duced everywhere and freely traded). More-
Moreover, there exist productivity levels at over, we can also allow cross-country variations
which exporters have positive operating pro ts in the xed-cost coef cients, as long as these
that exceed the operating pro ts from FDI variations do not lead some countries to stop
[since (aijI )12 « . (aijX)12 «], which ensures that producing the outside good. These generaliza-
some rms export to country j. In addition, the tions are useful for empirical applications.
second inequality in (1) implies (aijX)12 « . We report in our working paper general-
(aiD)12 «, which ensures that some rms serve equilibrium results for a special case in which
only the domestic market. countries only differ in size and trade costs per
The least productive rms expect negative product are symmetric (implying tij 5 t for j Þ
operating pro ts and therefore exit the industry. i). These restrictions apply within each sector,
This fate befalls all rms with productivity lev- so there can be arbitrary variations across sec-
els below (aiD)12 «, which is the cutoff at which tors. Under these circumstances, (2)–(4) and
operating pro ts from domestic sales equal free entry imply that, as long as countries do not
zero. Firms with productivity levels between differ too much in size, wages are the same
(aiD)12 « and (aijX)12 « have positive operating everywhere, all countries share the same cutoffs
pro ts from sales in the domestic market, but aiD 5 aD, aijX 5 aX, aijI 5 aI, and the same
expect to lose money from exports and FDI. demand levels Bi 5 B. Larger countries attract a
They choose to serve the domestic market but disproportionately larger number of entrants
not to serve country j. The cutoff (aijX)12 « is the (relative to country size) and a larger number of
productivity level at which exporters just break sellers (hence, more product variety). We also
even. Higher-productivity rms can export show that larger markets are disproportionately
pro tably. But those with productivity above served by domestically owned rms, i.e., the
(aijI )12 « gain more from FDI. For this reason, market share of domestically owned rms is
rms with productivity levels between (aijX)12 « larger in the home market of a larger country.
and (aijI )12 « export while those with higher pro-
ductivity levels build subsidiaries in country j, A. Exports Versus FDI Sales
which they use as platforms for servicing coun-
try j’s market. It is evident from the gure that We now consider the relative magnitude of
the cutoff coef cients (aiD)12 « , (aijX)12 «, and exports and local FDI sales for a pair of coun-
(aijI )12 « are determined by tries i and j. Let sijX be the market share in
country j of country i’s exporters and let sijI be
(2) ~a iD! 1 2 « B i 5 f D , ; i, the market share in country j of af liates of
country i’s multinationals. The relative size of
(3) ~ t ij a ijX ! 1 2 « B j 5 f X , ; j i, these market shares is then
(4) s ijX V ~a X !
12«
(5) ij 5 t 21
@1 2 ~ t ij! 1 2 « # ~a ijI ! 1 2 « B j 5 f I 2 f X , ; j i. sI V ~a I !
with the exporting cutoff coef cient aX and G(a). We use the Pareto distribution as a
declines with the FDI cutoff coef cient aI. The benchmark. When labor productivity 1/a is
cutoffs, in turn, are determined by the system of drawn from a Pareto distribution with the
equilibrium conditions. shape parameter k, the distribution of rm
A rise in the export costs fX or t, or a decrease domestic sales, indexed by V(a), is also Pa-
in the FDI costs fI, all have similar impacts on reto, with the shape parameter k 2 (« 2 1). 14
the aX and aI cutoffs: they induce an increase in The shape parameter of the Pareto distribu-
aI and a decrease in aX. The relative sales of tion offers a natural and convenient index of
exporters thus decline in all these cases. Recall dispersion, which characterizes heterogene-
that fI encompasses both the country-level xed ity. Given our assumptions, the domestic
costs embodied in fX and the duplicate plant sales of all rms with sales above any given
overhead costs represented by fD. It is therefore cutoff are distributed Pareto with the same
natural to consider the effects of equivalent shape parameter k 2 (« 2 1). A higher
increases in fI and fX (representing higher country- dispersion of rm productivity draws (lower
level costs), and the effects of equivalent k) or a higher elasticity of substitution «, raise
decreases in fI and fD (representing lower over- the dispersion of rm domestic sales and vari-
head plant costs, and hence smaller returns to able pro ts. We now investigate the impact of
scale). Again, we show that the aI and aX cutoffs such changes in heterogeneity on the relative
move in the same directions as before, entailing sales of exporters.
a decrease in relative export sales. The Pareto distribution implies that V(a1)/
These are sensible comparative statics pre- V(a2) equals (a1/a2)k2 (« 2 1) for every a1 and a2
dicting the cross-sectoral variation in relative in the support of the distribution of a. Relative
exports sales. We expect the relative sales of export sales in (5) can then be written as15
exporters to be lower in sectors with higher
transport costs or higher xed country-level
X D
k 2 ~« 2 1!
s ijX aX
costs (even when the latter costs are also borne (7) 5 t1 2 « 21
s ijI aI
by multinational af liates). We also expect
them to be lower in sectors where plant-level
X D
k 2 ~« 2 1!
returns to scale are relatively weak. These re- «21
fI 2 fX 1
sults show how the rm-level proximity- 5 t1 2 « 21 .
concentration trade-off results can be extended f X t« 2 1 2 1
to sectors with heterogeneous rms that select
different modes of foreign market access. It follows that relative export sales decrease
We now shift the focus to the role of rm- with decreases in k and increases in «.16 Thus,
level heterogeneity in explaining the cross- we expect sectors with higher levels of disper-
sectoral variation in relative export sales. Note
from (5) that the function V directly impacts
the relative sales (holding the cutoff levels 14
xed). Recall also that rm sales and variable The cumulative distribution function of a Pareto ran-
dom variable X with the shape parameter k is given by
pro ts are proportional to a12 « in every market.
V(a) therefore captures (up to a multiplicative
XD
k
b
constant) the distribution of sales and variable F~x! 5 1 2 , for x $ b . 0,
x
pro ts among rms that make the same export
or FDI decisions. It also captures the distribu- where b is a scale parameter that bounds the support [b, 1 `)
tion of domestic sales and variable pro ts from below. Log x is then distributed exponentially with a
standard deviation equal to 1/k. Any truncation from below
among all surviving rms. We think of V(a) as of X is also distributed Pareto with the same shape param-
summarizing rm-level heterogeneity in a sec- eter k. X has a nite variance if and only if k . 2. We
tor. It is exogenously determined by the distri- therefore assume that k . « 1 1, which ensures that both the
bution of unit costs G(a) and the elasticity of distribution of productivity draws and the distribution of
rm sales have nite variances.
substitution «. 15
Equations (3) and (4) are used in this derivation.
In order to index differences in rm-level 16
Recall that (fI 2 fX)/fX(t« 2 1 2 1) is greater than 1, by
heterogeneity across sectors, we parametrize assumption; see (1).
VOL. 94 NO. 1 HELPMAN ET AL.: EXPORT VERSUS FDI 305
a Uj
hI
« h 21
5
f hP
f jX
@~ v jt Uj
h !
« h 21
2 1# 21 ,
j Uj 1 2 « h
f hP 1
5 ~v t !h 21 .
~ t Uj U Uj 1 2 « h
B jh 5 w jf jX , f jX ~v j t Uj
h !
«h 2 1
21
(8) h w a hX !
model, we cannot use a measure for a rm that equivalent ways to measure dispersion. The rst
is somehow “representative” of the sector. is to regress the logarithm of an individual
Thus, standard measures of plant-level xed rm’s rank within the distribution on the loga-
costs, such as the number of production workers rithm of the rm’s size. It can be shown that the
at a plant of median size, are not appropriate. estimated coef cient of such a regression is
We seek a measure of such costs that is inde- k 2 (« 2 1), which is exactly the measure of
pendent of rm size or productivity. We follow dispersion that appears in the reduced form of
the model in choosing the number of nonpro- the model.22 The second method is to compute
duction workers per establishment as reported the standard deviation of the logarithm of rm
in the 1997 Census of Manufacturing.20 We sales, which—given our distributional assump-
calculate the average number of nonproduction tion—is computationally equivalent to the slope
workers at the North American Industry Clas- of the conditional expectation of log rank on log
si cation System (NAICS) level.21 Then, we size.23
compute the measure of plant-level xed cost, While our distributional assumption yields a
FP, for every BEA sector as the average of these precise methodology for computing dispersion,
numbers within the BEA sector, weighted by the choice of data is more problematic. We
the NAICS-level sales in the sector. require disaggregated data on the distribution of
sales across rms. Unfortunately, we do not
C. Measures of Dispersion have access to such data on U.S. rms. As a
result, we rely on two alternative sources.
The most novel feature of our model is the First, we use the publicly available data from
relationship between the degree of intra-industry the 1997 U.S. Census of Manufacturing. How-
rm heterogeneity and the prevalence of sub- ever, these data are aggregated into ten different
sidiary sales relative to export sales. To test this size categories, precluding the estimation of
hypothesis, we require data that quanti es the size dispersion measures using regression tech-
extent of this heterogeneity across industries. niques. Nevertheless, we can compute the stan-
As we cannot directly measure the dispersion of dard deviation of log sales by making an
intra-industry productivity levels, we rely on additional assumption: we assume that all es-
guidance from the model to construct alterna- tablishments falling within the same size cate-
tive measures of within-industry heterogeneity. gory have log sales equal to the mean for this
According to our model, the dispersion of rm category. We then treat each of the size catego-
size within a sector captures the joint effect of ries in the many subindustries of the BEA in-
the dispersion of rm productivity and the elas- dustry classi cation as separate observations.
ticity of substitution, which magni es the effect Adopting this method, we calculate the standard
of productivity differences across rms. Since deviation of log sales using the number of rms
the size distribution of rms is observable, we in each size category as weights.
use its dispersion as a measure of rm-level Second, Bureau van Dijck Electronic Pub-
heterogeneity. lishing has recently made available a large data
To quantify this dispersion measure across set of European rms. 24 This database, named
industries, we assume that the stochastic pro- Amadeus, includes information on the consoli-
cess that determines rm productivity levels is dated sales, the national identity, and the main
Pareto, with the shape of the distribution vary- line of business of a large number of European
ing across industries. This assumption is conve-
nient, because it suggests two conceptually 22
It is comforting that the distribution of rm size
closely follows a Pareto distribution; see Robert L. Axtell
(2001).
20 23
This measure does not strictly conform to our model- While the two methods of calculation should be equiv-
ing assumptions, because the number of nonproduction alent, in practice there are moderate to small differences in
workers is not independent of establishment size. their values. We therefore calculate them both ways.
21 24
The new 6-digit North American Industrial Classi - This data set has been used by John W. Budd et al.
cation System replaces the 4-digit Standard Industrial Clas- (2002), who investigate international rent-sharing within
si cation, but provides roughly the same level of industry multinational rms. We thank Matthew Slaughter for bring-
aggregation. ing this data set to our attention.
308 THE AMERICAN ECONOMIC REVIEW MARCH 2004
rms. There are roughly 260,000 rms in this The regression-based measures of dispersion
sample. provide a natural way of evaluating the cross-
We compute each of our two measures of sectional variation in this variable relative to the
dispersion for every industry in two subsets of measurement errors induced by the tting of the
these data: all Western European rms and Pareto distributions. Figure 2, which has been
French rms only. We compute our rm disper- constructed from the sample of Western Euro-
sion measures using French rms only for two pean rms, plots rm rank against rm sales in
reasons. First, using data for multiple countries four sectors on the same log-log scale. In every
raises the issue of industrial composition. plot the dispersion measure is represented by
Within every BEA industry there are many sub- the slope of the regression line while its good-
industries for which countries might produce ness of t is represented by the deviation from
different mixes. France’s industrial structure is linearity. Figure 3 quanti es this comparison by
very similar to the United States, however, and showing the 95-percent con dence intervals
so might share most of the same distributional around the coef cients of dispersion, estimated
aspects of rm characteristics. Second, French as the slopes of the regression lines in these
rms are highly overrepresented in the sample sectors. Evidently, these slopes are precisely
relative to all other Western European coun- estimated in all the sectors, with the exception
tries.25 Our dispersion measures are based on a of ve outliers that we discuss below.27
sample of 55,339 large Western European There are four measures of dispersion calcu-
rms, and a subset of 15,148 French rms.26 lated from the Amadeus data set and one mea-
sure calculated from the U.S. data. The
25
Due to national differences in reporting requirements,
no information on U.K. rms is available, and only an ent cutoffs. Any cutoff above U.S. $2.5 million yields a size
extremely limited number of German rms appear in the distribution that is closely approximated by a Pareto distri-
sample. bution, and a dispersion measure that varies very little with
26
Because small rms are underrepresented throughout the cutoff.
27
the Amadeus database, we rst drop rms with sales below As all 52 manufacturing sectors could not t on one
a cutoff of U.S. $2.5 million per year. Note that, under the graph, only one of the seven food processing sectors (201—
assumption of a Pareto size distribution, our measures of meat products) is represented. The coef cients and con -
dispersion are invariant to the choice of lower bound cutoff. dence intervals for the other six food processing sectors are
We computed the dispersion measures using several differ- very similar to the one represented.
VOL. 94 NO. 1 HELPMAN ET AL.: EXPORT VERSUS FDI 309
correlations between these measures are shown ent: the European measures are computed
in Table 2 (along with our measure of plant- from actual rm-level data while the Ameri-
level xed costs, FP, and the industries’ capital- can measure is calculated from semiaggre-
labor ratio, KL, and R&D intensity, RD). The gated establishment-level data. Given the
table shows that all four measures from Ama- differences in methods of calculation, one
deus are highly correlated with one another, as might argue that the correlations are surpris-
one might expect. The table also shows that the ingly high. Second, there exists an aggrega-
U.S.-based measure of dispersion is positively tion problem. If the composition of output
correlated with the measures of dispersion cal- varies across countries according to compar-
culated from the European data, except that this ative advantage, then within each BEA indus-
correlation is not as high as the correlations try the product mix of goods produced in the
among the four measures of dispersion that United States may differ from the mix pro-
were calculated from the European data. There duced in Europe. For this reason the European
are at least two reasons why this might be so. and American dispersion measures cannot be
First, the method of calculation is very differ- perfectly aligned.
310 THE AMERICAN ECONOMIC REVIEW MARCH 2004
III. Speci cations and Results ef cient of FP is positive and signi cant. We
therefore con rm the predictions of the proximity-
Our aim is to estimate a linearized version of concentration trade-off: rms substitute FDI
(11) that relates the logarithm of relative sales to sales for exports when the costs of international
our measure of rm-size dispersion, the loga- trade are relatively high and the returns to scale
rithm of our proxy for plant xed costs, the are relatively small.
logarithms of transport and tariff costs, and a set Next consider the effects of dispersion. The
of country dummies that we use to control for estimated coef cients on the various dispersion
the differences in fX and v across countries. Of measures are all negative and statistically sig-
course, this linearization precludes any struc- ni cant. Industries in which rm size is highly
tural interpretation of the estimated parameters. dispersed are associated with relatively more
Our goal is limited to testing whether the central FDI sales relative to exports, precisely as the
tendencies in the data are consistent with the model predicts. None of these results changes
partial derivatives implied by (11), and to signi cantly when the set of countries is ex-
assessing the economic signi cance of the panded to include the 11 smaller countries (the
magnitudes associated with the estimated wide country sample).29
coef cients. Although all measures of dispersion yield
We consider several variants of the basic coef cients that are statistically signi cant, the
speci cation in order to raise the level of con- choice of dispersion measure has a noticeable
dence in the results. Given the critical impor- impact on the results. The measures that were
tance of the size distribution of rms, we report derived by tting a Pareto distribution to the
results corresponding to each one of the ve distribution of rm size, yield substantially
measures of dispersion in rm size. We also lower coef cients and higher standard errors
report results for both samples of countries: than the nonparametric dispersion measures,
narrow and wide. Finally, we explore the sen- i.e., the standard deviations of log sales. This
sitivity of the results to alternative assumptions pattern is driven, in large part, by ve sectors
that incorporate other determinants of relative that exhibit the largest differences between the
sales not captured by equation (11). measurement of dispersion by means of the
We begin the analysis by considering a spec- shape of a Pareto distribution and by means of
i cation that controls for sectoral capital and the standard deviation, for both Western Euro-
R&D intensities.28 The results across speci ca- pean and French rms.30 These sectors have the
tions for our two samples and ve measures of lowest number of rms in the data, and they
dispersion are shown in Table 3. The columns yield—without exception—the poorest ts to
correspond to different measures of dispersion, the Pareto distribution, as measured by
beginning with the U.S. standard deviation of R-squares. We believe that in these cases the
log sales, proceeding to the European and nonparametric measures (the standard devia-
French-only standard deviation measures, and tions) better describe the levels of dispersion
ending with the estimated distribution parame- within the sectors. Dropping these ve outliers
ters for the European and the French-only sam- from the sample and reestimating the equations,
ple, respectively. Country xed effects are not we nd that the two different ways of measuring
reported. dispersion yield much more similar results. Af-
First consider the narrow sample of relatively
large countries, studied by Brainard (1997). The
29
coef cients on FREIGHT and TARIFF are neg- The magnitude of the coef cients on virtually all
ative and statistically signi cant in each one of dispersion measures are lower in the wider sample. One
possible explanation is that attenuation bias has affected the
the ve speci cations. These results are consis- magnitudes of the coef cients. Another explanation is that
tent with Brainard (1997). In addition, the co- the process generating FDI in the smaller developing coun-
tries is somewhat different from the process generating FDI
in the larger developed countries.
28 30
In Helpman et al. (2003), we also report estimates The ve outliers consist of the following sectors:
without these controls. The two sets of estimates do not 210—tobacco, 369 — other electronics, 379 — other trans-
differ by much, and the measures of dispersion are highly port equipment, 381—scienti c and measuring equipment,
signi cant in both cases. and 386 — optical and photographic equipment.
VOL. 94 NO. 1 HELPMAN ET AL.: EXPORT VERSUS FDI 311
Notes: T-statistics are in parentheses (calculated on the basis of White standard errors). Constant and country dummies are
suppressed.
independent variable, divided by the standard sales, while capital intensity is; more capital-
deviation of the dependent variable. It converts intensive sectors export less relative to FDI
the regression coef cients into units of sample sales. These results are interesting, but our the-
standard deviations.31 These beta coef cients oretical model offers no guidance concerning
suggest that each one of the ve measures of their interpretation.35
dispersion has a comparable impact to each one Of course, differences in capital intensity
of the standard proximity-concentration vari- may not be the only other source of variation
ables.32 For instance, a one standard deviation across sectors that affects exports relative to
decline in an industry’s freight costs raises the FDI sales. In order to address the possibility that
logarithm of the ratio of exports to FDI sales by some other unmeasured characteristics of sec-
27 percent of a standard deviation; and a one tors fall into this category, we estimate the
standard deviation decline in the dispersion previous speci cation (with the capital and
measures induce comparable changes in the de- R&D intensity controls) adding random indus-
pendent variable, with an average impact of 26 try effects. A bene t of this estimation strategy
percent across the dispersion measures. The im- is that it allows for ef cient estimation in the
pact of tariffs is lower while the impact of presence of common components in the re-
returns to scale is higher. Taken as a whole, siduals that might be induced by unmeasured
these results suggest that rm-level heterogene- industry characteristics. To validate this speci-
ity adds an important dimension to the observed cation, we need to assume that these unmea-
trade-off between exports and FDI sales. sured industry characteristics are uncorrelated
These results strongly support the theoretical with our right-hand-side variables. This is a
model’s predicted link between rm-level het- strong assumption. We feel, however, that it is
erogeneity and the ratio of exports to FDI sales. most likely to hold for our dispersion measures,
Nevertheless, these results have to be inter- which are the focus of the empirical analysis.36
preted with caution, because they may also The results are reported in Table 5. As could
re ect—at least to some degree—variations in be predicted, the standard errors have increased.
industry characteristics that are not captured by But so have the point estimates of the impact of
our parsimonious model. This problem is partly dispersion on exports relative to FDI sales. Im-
taken care of by our control of cross-industry portantly, however, the coef cients for all the
variations in capital and R&D intensities. Note dispersion measures remain highly signi cant.
that both these variables represent characteris- On the other hand, the magnitude of the coef-
tics of an industry’s technology that are not cients on FREIGHT and TARIFF are greatly
captured by our model.33 Furthermore, as reduced, and the coef cients on TARIFF are no
shown in Table 2, these measures of technology longer signi cant. These results support our ear-
are correlated with all the different dispersion lier conclusion that the economic signi cance
measures, although the correlations with the of rm heterogeneity compares favorably with
U.S.-data-based dispersion measure are rather the signi cance of the standard proximity-
weak.34 Table 3 suggests that R&D intensity is concentration trade-off variables in explaining
not a useful predictor of exports relative to FDI the export to FDI sales ratio.
Another robustness check addresses the po-
tential interdependence of the residuals across
31
See Jeffrey M. Wooldridge (2003, Sec. 6.1) for a countries, which may exist even after we con-
further description of this transformation.
32
In the case of FREIGHT, TARIFF, and FP, the coef-
35
cients are averaged across the ve speci cations. Pehr-Johan Norback (2001) nds that R&D-intensive
33
We have restricted our choice of controls to the mea- rms tend to export rather than engage in FDI when tech-
surable characteristics of sectors that are outside the scope nology transfer costs are high, but not when these costs are
of the model, and we have excluded attributes that are low. This suggests that we need a more detailed model in
predicted to endogenously respond to changes in the mod- order to study the role of R&D in the proximity-concentration
el’s exogenous variables. trade-off.
34 36
Capital intensity is measured as the industry’s aggre- The inclusion of industry xed effects would elimi-
gate capital to labor ratio (from the NBER productivity nate the need for this assumption, but would also preclude
database) and R&D intensity is measured as the ratio of any identi cation of sector-level characteristics, such as our
R&D expenditures to sales (from a 1978 FTC survey). dispersion measures.
VOL. 94 NO. 1 HELPMAN ET AL.: EXPORT VERSUS FDI 313
Notes: T-statistics are in parentheses. Constant and country dummies are suppressed.
trol for country xed effects. This type of inter- sented and highly integrated economies of
dependence pattern could be created by the Western Europe.
ability of af liates to re-export a portion of their To address this concern, we treated all the
production to a third country. In this case, a Western European countries as a single aggre-
rm’s decision to operate an af liate in one gate unit and reestimated our speci cation with
country, say Belgium, would not be indepen- the industry controls (capital and R&D intensi-
dent from its decision to locate af liates in other ties) and industry random effects. We found that
neighboring European countries. In the Appen- all the dispersion measures remain highly sig-
dix to our working paper, Helpman et al. ni cant. As could be predicted, the point esti-
(2003), we show that the predicted link between mates on the dispersion measures were slightly
rm-level heterogeneity within sectors and ex- lower, which re ects the fact that the smaller
ports relative to FDI sales is theoretically con- developing countries now receive a greater
sistent with an extended version of the model weight in the sample.37
that explicitly allows for re-exports by af liates.
However, the pattern of interdependence may
be particularly strong among the overrepre- 37
See Table 8 of our working paper for these estimates.
314 THE AMERICAN ECONOMIC REVIEW MARCH 2004
OLS IV
Narrow Wide Aggregated Narrow Wide Aggregated
sample sample Europe sample sample Europe
FREIGHT 21.040 21.011 21.001 21.218 21.118 21.053
(23.997) (24.437) (24.464) (23.862) (24.366) (24.545)
TARIFF 20.365 20.241 20.077 20.188 20.124 20.016
(21.611) (21.081) (20.304) (20.706) (20.478) (20.056)
FP 1.177 1.133 1.086 1.609 1.457 1.344
(4.876) (4.472) (4.166) (3.605) (3.311) (3.182)
DISPERSE (U.S.) 22.343 22.248 22.150 24.321 23.681 23.198
(23.689) (23.655) (23.349) (22.606) (22.248) (22.217)
KL 20.868 20.793 20.848 20.938 20.848 20.890
(23.032) (22.513) (22.593) (22.855) (22.496) (22.599)
RD 20.104 20.086 20.087 20.158 20.127 20.121
(20.851) (20.687) (20.691) (21.233) (20.985) (20.968)
R2 0.373 0.338 0.314 0.328 0.315 0.315
N 961 1,175 678 961 1,175 678
Notes: In the IV speci cations, the U.S. dispersion measure is instrumented using all four European dispersion measures. All
T-statistics (in parentheses) are computed from standard errors that are heteroskedasticity consistent and adjusted for
clustering by industry. Constant and country dummies are suppressed.
Our nal robustness check addresses sources We found that netting out the value of these
of endogeneity bias in the dispersion measures, imports from our FDI sales data had no appre-
including measurement error. To address these ciable impact on the dispersion coef cients, al-
concerns, we instrument the U.S. dispersion though it had a small impact on the size of the
measure using all four European dispersion FREIGHT and TARIFF coef cients. In other
measures. We also use a different method to speci cations, we included the four- rm con-
control for the potential correlation of the resid- centration ratio as a control, in order to assess
uals within sectors by adjusting the standard whether our measures of rm heterogeneity of-
errors for clustering (within sectors).38 These fer information in excess of this crude measure
speci cations are reported in Table 6 for all of concentration. We found that controlling for
previously discussed country samples (narrow, concentration reduces the point estimates of the
wide, and aggregated Europe). Instrumenting coef cients on the dispersion measures, but that
the U.S. dispersion measure signi cantly in- this decline is rather small.
creases the magnitude of both the estimated
coef cient and its standard error. However, as IV. Conclusion
in all the previous speci cations, the effect of
dispersion on relative exports and FDI sales We have developed in this paper a model of
remains statistically signi cant. international trade and investment in which
Finally, we brie y report a number of other rms can choose to serve their domestic market,
robustness checks. One potential complication to export, or to engage in FDI in order to serve
arises from the fact that rms engage in intra- foreign markets. Every industry is populated by
rm trade in intermediate inputs. This trade heterogeneous rms, which differ in productiv-
does not appear in our model, but is of suf cient ity levels. As a result, rms sort according to
size in a number of industries to be of concern. productivity into different organizational forms.
The least productive rms leave the industry,
38 because, if they stay, their operating pro ts will
Under our assumptions on the source of this potential
correlation in the residuals— unmeasured sectoral charac- be negative no matter how they organize. Other
teristics—the previously reported random-effects coef - low-productivity rms choose to serve only the
cients are the ef cient estimators. domestic market. The remaining rms serve the
VOL. 94 NO. 1 HELPMAN ET AL.: EXPORT VERSUS FDI 315
domestic market as well as foreign markets. Multinational Sales and Trade.” American
Their mode of operation in foreign markets Economic Review, September 1997, 87(4),
differs, however. The most productive rms in pp. 520 – 44.
this group choose to invest in foreign markets Budd, John W.; Konings, Jozef and Slaughter,
while the less productive rms choose to export. Matthew J.“International Rent Sharing in
This sorting pattern is con rmed by previous Multinational Firms.” National Bureau of
empirical work and by our own estimates. Economic Research (Cambridge, MA) Work-
Our model embodies standard elements of ing Paper No. 8809, February 2002.
the proximity-concentration trade-off in the the- Doms, Mark E. and Jensen, J. Bradford.“Com-
ory of horizontal foreign direct investment. As a paring Wages, Skills, and Productivity be-
result, it predicts that foreign markets are served tween Domestically and Foreign-Owned
more by exports relative to FDI sales when Manufacturing Establishments in the United
trade frictions are lower or economies of scale States,” in Robert E. Baldwin, Robert E. Lip-
are higher. To these factors, our model adds a sey, and J. David Richardson, eds., Geogra-
role for the within-sector heterogeneity of pro- phy and ownership as bases for economic
ductivity levels. This heterogeneity induces a accounting. NBER Studies in Income and
size distribution of rms, which affects the ratio Wealth, Vol. 59. Chicago: University of Chi-
of exports to FDI sales. cago Press, 1998, p. 235–55.
Using data on exports and FDI sales of U.S. Ethier, Wilfred J.“The Multinational Firm.”
rms in 38 countries and 52 industries, we Quarterly Journal of Economics, November
estimated the effects of trade frictions, econo- 1986, 101(4), pp. 805–33.
mies of scale, and within-industry dispersion of Ethier, Wilfred J. and Markusen, James R.“Mul-
rm size, on exports versus FDI sales. The tinational Firms, Technology Diffusion and
results support the theoretical predictions. In Trade.” Journal of International Economics,
particular, they show a robust cross-sectoral re- August 1996, 41(1–2), pp. 1–28.
lationship between the degree of dispersion in Feenstra, Robert C.“U.S. Exports, 1972–1994:
rm size and the tendency of rms to substitute With State Exports and Other U.S. Data.”
FDI sales for exports. The size of this effect is National Bureau of Economic Research
of the same order of magnitude as trade fric- (Cambridge, MA) Working Paper No. 5990,
tions. We therefore conclude that we have iden- April 1997.
ti ed a new element—namely, within-sectoral Girma, Sourafel; Thompson, Steve and Wright,
heterogeneity—that plays an important role in Peter W.“Why Are Productivity and Wages
the structure of foreign trade and investment. Higher in Foreign Firms?” Economic and
Social Review, Spring 2002, 33(1), pp. 93–
REFERENCES 100.
Hanson, Gordon H.; Mataloni, Raymond J. and
Axtell, Robert L.“Zipf Distribution of U.S. Firm Slaughter, Matthew J.“Vertical Specialization
Sizes.” Science, September 2001, 293, pp. in Multinational Firms.” Mimeo, University
1818–20. of California, San Diego, 2002.
Bernard, Andrew B.; Eaton, Jonathan; Jenson, J. Helpman, Elhanan.“A Simple Theory of Inter-
Bradford and Kortum, Samuel.“Plants and national Trade with Multinational Corpora-
Productivity in International Trade.” Ameri- tions.” Journal of Political Economy, June
can Economic Review, September 2003, 1984, 92(3), pp. 451–71.
93(4), pp. 1268–90. . “Multinational Corporations and Trade
Brainard, S. Lael.“A Simple Theory of Multina- Structure.” Review of Economic Studies, July
tional Corporations and Trade with a Trade- 1985, 52(3), pp. 443–57.
Off Between Proximity and Concentration.” Helpman, Elhanan; Melitz, Marc J. and Yeaple,
National Bureau of Economic Research Stephen R.“Export Versus FDI.” National
(Cambridge, MA) Working Paper No. 4269, Bureau of Economic Research (Cambridge,
February 1993. MA) Working Paper No. 9439, January 2003.
. “An Empirical Assessment of the Horstmann, Ignatius and Markusen, James
Proximity-Concentration Trade-off Between R.“Licensing versus Direct Investment: A
316 THE AMERICAN ECONOMIC REVIEW MARCH 2004
Model of Internalization by the Multinational national Economics, August 2001, 54(2), pp.
Enterprise.” Canadian Journal of Economics, 449 – 69.
August 1987, 20(3), pp. 464 – 81. Rob, Rafael and Vettas, Nikolaos.“Foreign Direct
. “Endogenous Market Structures in In- Investment and Exports with Growing De-
ternational Trade.” Journal of International mand.” Review of Economic Studies, July
Economics, February 1992, 32(1–2), pp. 2003, 70(3), pp. 629 – 48.
109 –29. Tybout, James R. “Plant and Firm-Level
Markusen, James R. Multinational rms and the Evidence on ‘New’ Trade Theories,” in E.
theory of international trade. Cambridge, Kwan Choi and James Harrigan, eds.,
MA: MIT Press, 2002. Handbook of international trade. Oxford:
Markusen, James R. and Venables, Anthony Basil Blackwell, 2003, pp. 388 – 415.
J.“The Theory of Endowment, Intra-industry Wooldridge, Jeffrey M. Introductory economet-
and Multi-national Trade.” Journal of Inter- rics, 2nd Ed. Mason, OH: South-Western,
national Economics, December 2000, 52(2), 2003.
pp. 209 –34. Yeaple, Stephen R.“Firm Heterogeneity, Interna-
Melitz, Marc J.“The Impact of Trade on Aggre- tional Trade, and Wages.” Mimeo, University
gate Industry Productivity and Intra-Industry of Pennsylvania, 2003a.
Reallocations.” Econometrica, November . “The Role of Skill Endowments in the
2003, 71(6), pp. 1695–725. Structure of U.S. Outward Foreign Direct
Norback, Pehr-Johan.“Multinational Firms, Investment.” Review of Economics and Sta-
Technology and Location.” Journal of Inter- tistics, August 2003b, 85(3), pp. 726 –34.