Journal of International Economics 93 (2014) 108–122

Contents lists available at ScienceDirect

Journal of International Economics

journal homepage: www.elsevier.com/locate/jie

A quantitative approach to multinational production

Natalia Ramondo
University of California at San Diego, School of International Relations and Pacific Studies, 9500 Gilman Drive, MC 0519 La Jolla, CA 92093-0519, United States

a r t i c l e i n f o a b s t r a c t

Article history: I examine new data on the number and revenues of foreign affiliates of multinational firms across a large number
Received 2 June 2008 of country pairs. The data shed light on the behavior of the intensive and extensive margins of multinational pro-
Received in revised form 31 December 2013 duction (MP). To capture the patterns observed in the data, I build and calibrate a multi-country general-
Accepted 10 January 2014
equilibrium model of MP that combines a Lucas (1978) span-of-control with an Eaton and Kortum (2002)
Available online 17 January 2014
type model, and includes both fixed and variable costs of opening affiliates abroad. I use the calibrated model
Keywords:
to calculate the gains that a country would experience from liberalizing access to foreign firms. Those calculations
Multinational production suggest that the welfare losses of closing up to foreign firms would be around 4%, while the gains of liberalizing
Foreign direct investment access to foreign firms would be large, particularly if the variable – rather than the fixed – component of MP costs
Welfare gains were lowered.
Gravity © 2014 Elsevier B.V. All rights reserved.
Calibration

1. Introduction however, reacts disproportionately more than the intensive margin

to changes in distance and country size.
One of the most notable features of economic globalization has been To capture the patterns observed in the data, I build a multi-country
the increasing importance of multinational production (MP) around the general-equilibrium model of MP that combines a Lucas (1978)
world.1 In fact, multinational firms have become one of the most impor- span-of-control with an Eaton and Kortum (2002) type model. MP
tant channels through which countries exchange goods, capital, ideas, occurs in the model when a technology that originates in a foreign
and technologies. By 2007, world sales of foreign affiliates of multina- country is used to produce a good in the host country. But using a for-
tional firms were almost twice as high as world exports. Furthermore, eign technology for production entails a cost, one that is a combina-
over the past two decades, sales of foreign affiliates increased by a factor tion of a variable and a fixed component.
of seven, while exports increased by a factor of five (United Nations The model is novel in that combines a Lucas (1978) span-of-control
Conference on Trade and Development—UNCTAD). type model—which features decreasing returns to scale, fixed costs, and
A natural relevant question is: How large are the gains from hosting perfect competition—with an Eaton and Kortum (2002) type model,
foreign firms? Much attention has been devoted to quantifying the from which I borrow the probabilistic representation of technologies.
gains from trade, but the gains from the activity of multinational firms While having firms with a limited span of control allows me to distin-
could be as large, or even larger, given that we observe flows twice as guish between the extensive (number of affiliates) and intensive (reve-
large. nues per affiliate) margins of MP and to generate distinct predictions
I start by examining new data on the number and revenues of for- about them, having a probabilistic representation of technologies allows
eign affiliates of multinational firms across a broad set of country me to take the model's general equilibrium to the multi-country data.
pairs. As Eaton et al. (2011) do for French exporters, I document the im- Nevertheless, as a simplification, the modeling strategy eliminates
portance of the extensive versus the intensive margin of multinational trade altogether: Affiliates use inputs and sell their output exclusively
activities across countries, and I present evidence on how they in the host country of production.
react to geographical distance between partners, and to host- and I calibrate the model to match the patterns of both revenues and
source-country characteristics. While larger markets receive (and number of affiliates across different country pairs. Including both fixed
send) more and larger foreign affiliates, on average, distant markets and variable components of the cost of engaging in multinational activ-
receive fewer and smaller foreign affiliates. The extensive margin, ities is crucial to matching the extensive and intensive margins of MP.
Moreover, results indicate that the incentives to engage in foreign pro-
E-mail address: [email protected]. duction vary significantly across countries. In particular, overall, MP
1
I use the term “multinational production”, rather than “foreign direct investment”
costs are lower among rich countries than among poorer countries,
(FDI), as most of the previous literature does, because, here, I refer to the activity of foreign
affiliates, such as sales or revenues. FDI is a financial category of the Balance of Payments
which do very little MP. But the value of the fixed cost that foreign affil-
and, as such, is one possible channel through which affiliates finance their activities iates pay in richer countries is higher, as a share of the host country's
abroad. GDP, than the value they pay in poorer countries. The opposite is true

0022-1996/$ – see front matter © 2014 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jinteco.2014.01.004
 N. Ramondo / Journal of International Economics 93 (2014) 108–122 109

for the variable component of MP costs, which is much higher in poorer Table 1
countries. World multinational production and trade.
Source: World Investment Report, UNCTAD (2009).
Finally, I use the calibrated model to calculate the gains from engag-
ing in multinational activities. My calculations suggest that the gains in 1982 1990 2001 2005 2007
real income per capita of moving from autarky to a situation with the World GDP in current U.S. 11,758 22,610 31,900 40,960 55,114
observed flows of multinational activity would be around 4%, reaching dollars (bn)
5% for richer countries and 3.5% for poorer countries. The gains of liber- As % of world GDP:
World sales of foreign 24 25 58 51 58
alizing access to foreign firms would be large for all countries, but larger
affiliates
among the group of poorer countries than among the group of richer World gross product of 5 7 11 10 11
countries. Most of these gains come from lowering the variable – rather foreign affiliates
than the fixed – component of MP costs. World exports 19 19 23 27 31
Early work by Markusen (1984) qualitatively analyzes the welfare ef- As % of affiliates' sales:
world exports of foreign affiliates 26 26 14 18 19
fects of opening up to multinational firms in a model in which
knowledge-based, firm-specific assets can be supplied costlessly within Notes: Exports include goods and non-factor services.
the corporation. Only recently have there been attempts to quantify the
gains from the activity of multinational firms. Efforts in that direction in- managerial know-how, which shapes the firm's productivity, is difficult
clude Burstein and Monge-Naranjo (2009) and McGrattan and Prescott to reproduce at the affiliate level: A manager with certain abilities can
(2009). Both papers extend the neoclassical growth model to allow for control only a limited amount of inputs to production in a given
foreign production. The main difference from my framework is that location.
theirs are suitable for analyzing the transition dynamics when countries
open up to foreign firms. Even though my model is static, by using an
Eaton and Kortum (2002) type model, I am able to introduce several 2. Multinational production in the data
sources of heterogeneity and, hence, to calibrate the model to the ob-
served (gravity) pattern of MP across multiple country pairs. Another Multinational production (MP) – rather than international trade –
attempt to quantify the gains from multinational activity, using an has become the dominant way through which firms serve foreign con-
Eaton and Kortum (2002) type model, is Garetto (2013). The main dif- sumers. Using data for the period 1982–2007, from the United Nations
ference from my framework is that, while I focus on horizontal MP, Conference on Trade and Development (UNCTAD), Table 1 indicates
she focuses on vertical MP.2 that while world exports went from 19% to 31% of world GDP during
More generally, this paper contributes to a recent but growing liter- this period, total sales of foreign affiliates of multinational firms, as a
ature that attempts to quantify the importance of different sources of share of world GDP, increased from 24% in 1982 to 58% in 2007. Gross
welfare gains for countries other than trade. Even though trade is only product of foreign affiliates more than doubled, as a share of world
one possible channel through which countries interact, the previous lit- GDP, during the same period. These magnitudes suggest that foreign af-
erature has typically equated the gains from trade with the overall gains filiates of multinational firms are more important than exports as the
from openness.3 An exception is Ramondo and Rodríguez-Clare (2013), channel through which firms choose to serve foreign consumers.
which introduces trade and MP into an Eaton and Kortum (2002) Additionally, the fact that, between 1982 and 2007, world exports of
framework to quantify the overall gains from openness.4 Another ex- affiliates, as a share of world sales of affiliates, decreased from 26 to 19%
ception is Irarrazabal et al. (2013), which uses a Melitz-type model to suggests that the majority of affiliates' output is sold in the host country
evaluate the joint gains from trade and MP. In comparison with these of production—and, hence, is not exported—and that share increased
two papers, my model features foreign firms as the only way of serving through time. In other words, “horizontal FDI,” which is associated
a foreign market. The gains calculated from my only-MP model, howev- with affiliates that replicate the parents' activities and whose purpose
er, are a valuable benchmark to be compared with calculations coming is to serve the host country of production, seems more pervasive than
from trade-only quantitative models. “vertical FDI,” which is associated with affiliates that specialize in
A final remark regarding this paper's modeling strategy is in order. some slice of the production chain and whose purpose is to export
Most of the literature on multinational firms has introduced this type their output to another party within the corporation. The importance
of firm into an increasing returns to scale and monopolistic competition of horizontal activities by affiliates of multinational firms is in line
model, rather than a decreasing returns to scale and perfect competition with the evidence in Markusen (1995) and, more recently, in
model, as I do here.5 These choices, however, do not have any conse- Ramondo et al. (2013a).7
quence for aggregate MP flows: A model in which firms compete mo- Next, I explore in more detail the patterns of multinational activity
nopolistically and produce under increasing returns to scale would by examining new data on the activities of foreign affiliates of multina-
have similar predictions in that regard.6 A model with a limited span tional firms, for a sample of 35 countries (of which 18 are richer OECD
of control, however, seems more realistic for MP since it reflects that countries), an average over 1996–2001. The data contain two key vari-
ables that allow me to document salient features of the patterns of the
extensive and intensive margins of MP across several country pairs:
2
Horizontal MP refers to affiliates that replicate the parent's activities, while vertical MP the number and revenues of affiliates of multinational firms. Section 4
refers to affiliates that specialize in some slice of the production chain.
3
See, for example, Eaton and Kortum (2002), Anderson and van Wincoop (2004),
describes the data in detail.
Waugh (2010), Fieler (2011), and Donaldson (forthcoming), among many others. Let Mni denote the number of affiliates from i in n, xni revenues per
4
Rodríguez-Clare (2007) is another exception. He introduces international trade and affiliate from i in n, and Xni total revenues of affiliates from i in n. Natu-
international diffusion of ideas into an Eaton and Kortum (2002) model to evaluate the rally, total revenue flows can be written as X ni ≡ xni Mni , the product of
gains arising from both channels.
5 the intensive and extensive margins of MP, respectively. Let Xn denote
Exceptions are Burstein and Monge-Naranjo (2009) and McGrattan and Prescott
(2009), who, as I do, choose as their modeling strategy decreasing returns to scale-plus- country n's GDP. Thus, revenue shares, or MP shares, from i in n are de-
perfect competition at the affiliate level. The first paper introduces managerial know- fined as xni ≡ Xni/Xn.
how, potentially foreign, into a Lucas's (1978) span-of-control setup in which the scarcity
of managers creates a constraint that makes replication of technologies across multiple
7
countries impossible. By contrast, McGrattan and Prescott (2009) assume that technology Using firm-level data for U.S. multinationals, Ramondo et al. (2013a) show that
capital is fully replicable in each foreign location, similar to the assumption in my frame- while it is true that intra-firm trade flows are large (particularly North–North flows) as
work and in the early work by Markusen (1984). a fraction of total trade, they represent a small fraction of affiliate sales for the median mul-
6
This is similar to the results in Arkolakis et al. (2012) about trade-only models. tinational firm, regardless of the destination country or the industry of operation.
110 N. Ramondo / Journal of International Economics 93 (2014) 108–122

Table 2 1), the average revenues per affiliate from i in n decrease by only 2.5%
Multinational production at a glance. (column 4). This implies a distance elasticity of 9% for aggregate MP
All OECD(18) Non- flows (Xni).12
OECD(18) An alternative way of presenting the relative importance of the ex-
Averages tensive and intensive margins of MP is as Eaton et al. (2011) do for
MP shares by i in n, as share of n′ GDP (xni) 0.009 0.019 0.001 French exporters. MP flows can be written as X ni ≡ xni Mni or, alternately,
Number of affiliates from i in n (Mni) 77 191 8 as Xni ≡ xniXn, which yields the identity xni Mni ≡ xni X n . With this identity
MP per affiliate from i in n (xni ) 34 60 7.7
in mind, the relative importance of the intensive versus the extensive
Totals margin of MP can be read from running OLS on
MP into n, as share of n′ GDP (∑ i ≠ nxni) 0.30 0.32 0.011
MP from n, as share of n′ GDP (∑ i ≠ nXin/Xn) 0.19 0.24 0.009 logMni ¼ ax logxni þ aX logX n : ð2Þ
Numbers
Country pairs 1190 306 884 Table 4 shows that, when source-country fixed effects are consid-
Country pairs with zero MP 93 2 91 ered, given the size of the destination market, Xn, a higher MP share
Notes: MP shares by i in n are total revenues of affiliates of multinational firms from coun- into destination n, xni, reflects 49% more affiliates producing there (i.e.,
try i in n, as a share of country n's GDP. MP per affiliate from i in n refers to the ratio of total the extensive margin) and 51% more revenues per affiliate (i.e., the in-
revenues of affiliates to the number of affiliates from i in n, in millions of current U.S. dol- tensive margin), for the average country in the sample. Similarly,
lars. Total MP into (from) n refers to total revenues of foreign affiliates into (from) country
given MP shares into market n, xni, revenues in a larger market reflect
n, as a share of country n's GDP.
55% more affiliates and 45% more revenues per affiliate. While the mag-
nitude of the coefficient on market size is similar to the one found by
Table 2 shows that among the 35 countries included in the sample, Eaton et al. (2011) for French exporters, the coefficient on MP shares
there are 1406 possible bilateral country pairs of which 93% have an is around half as large as theirs: Conditional on market size, the inten-
MP relationship, with zero MP concentrated in poorer countries. sive margin of MP seems to be stronger than the intensive margin of
Among the sub-sample of 18 OECD countries, bilateral MP flows, as a trade.
share of the host country GDP, are, on average, much higher than
among non-OECD countries due to both a larger number of affiliates 3. Model
(i.e., the extensive margin) and higher revenues per affiliate (i.e., the in-
tensive margin).8 Higher bilateral MP shares naturally result in higher To interpret the relationships found in the data, I introduce multina-
total shares into and from this subset of countries (0.32 and 0.24, tional production (MP) in a multi-country general-equilibrium model
respectively). that combines Lucas's (1978) span-of-control model—which entails de-
What is the effect of geography and market size—of both the country creasing returns to scale and fixed costs at the firm level plus perfect
hosting and sending affiliates abroad—on the extensive and intensive competition at the industry level—with the probabilistic representation
margins of MP? This evidence will be useful for the calibration results of technologies from Eaton and Kortum's (2002) model of trade (hence-
in Section 4. In what follows, variables are in logarithms, so the observa- forth, “EK”). The model explains where firms sell, how many firms
tions with zero MP are ignored. enter, and how much they sell, in each destination.
The response of the extensive and intensive margins of MP to dis- In the model, MP by country i in n occurs when a technology from
tance and host- and source-country size can be visualized, respectively, country i is used in country n to produce a good.13 But using a foreign
in Figs. 1, 2, and 3. It is clear from these figures that the bilateral number technology for production entails a cost. This cost of engaging in MP
of affiliates is more responsive than the average revenues per affiliate to by i in n is a combination of a variable and a fixed component. There is
distance and size. These results are confirmed by the ordinary least no trade: Affiliates from country i hosted by n produce by using inputs
squares (OLS) estimates of the following equation: from n and sell exclusively in n.

logY ni ¼ ad logdni þ ah logX n þ as logX i þ uni ; ð1Þ 3.1. A Lucas-Eaton-Kortum model of multinational production

where Yni denotes Mni and xni , alternately, dni is the geographical dis- Consider a set of countries indexed by i ∈ {1,…,I}. Country i has Li
tance between i and n, Xn denotes country n's GDP, and uni denotes all units of labor. A continuum of goods u ∈ [0,1] are produced in quantities
other country-pair specific factors that affect the extensive margin of q(u). They aggregate into a CES composite good Q, given by
MP and are orthogonal to the regressors.9 The expression in Eq. (1) is es- !
Z σ
σ−1
timated replacing Xn and Xi, respectively, by host- and source-country 1 σ −1
Q¼ qðuÞ σ
du ; ð3Þ
fixed effects, as well.10 Table 3 shows the results.11 0
As the figures suggested, the magnitude of the OLS coefficients con-
firms that the extensive margin of MP is much more responsive to dis- with σ N 1.
tance and market size than the intensive margin is. The significant The production function of a firm from i that produces good u in n is
negative effect of geographical distance on both margins of MP survives
ν
the inclusion of fixed effects: While a ten-percent increase in distance qni ðuÞ ¼ zni ðuÞSni ðuÞ ; ð4Þ
decreases the number of affiliates from i in n by almost 6.5% (column
where 0 b v b 1. This production function exhibits decreasing returns to
8
Considering only country pairs with positive MP, the average MP share goes from scale with the span-of-control parameter given by v. The variable qni(u)
0.009 to 0.0095, while the average number of affiliates goes from 77 to 83. denotes output of an affiliate from i in n, and Sni(u) is the amount of the
9
Distance is in thousands of kilometers, from the Centre d'Etudes Prospectives et Infor- input bundle required for production by an affiliate from i in n, defined
mations Internationales (CEPII), and GDP is in current U.S. dollars, from the World Develop-
below.
ment Indicators, an average over 1996–2001.
10
Regressing the source- and host-country fixed effects on the source- and host-country
12
sizes, respectively, delivers almost identical coefficients for the size variables. Head and Ries (2008) find that a ten-percent increase in distance decreases bilateral
11
The relation between MP volumes and gravity has been largely explored and docu- FDI stocks by 12.5%, for a broad set of countries.
13
mented, among others, by Carr et al. (2001), who uses affiliates' sales, and Razin et al. The ability of a firm to replicate its technology across locations can be interpreted as
(2003), and Head and Ries (2008), who uses FDI stocks. None of these papers distinguish the ability to transfer knowledge-based assets within the corporation, as in Markusen's
between the two margins of MP. (1984) model of horizontal FDI. See Bloom and Van Reenen (2007) for empirical evidence.
 N. Ramondo / Journal of International Economics 93 (2014) 108–122 111

(a) Number of affiliates from in ( ) (b) Average MP from in ( )

8

Log of Number of Affiliates from i in n

7

Log of MP per Affiliate from i in n

7 6

6 5

4
5
3
4
2
3
1
2
0
1 -1
0 -2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3
Log of distance between country i and n Log of distance between country i and n
Note: Distance is in thousands of kilometers.

Fig. 1. Multinational production and geography. (a) Number of affiliates from i in n (Mni). (b) Average MP from i in n (xni ). Note: Distance is in thousands of kilometers.

The variable zni(u) represents the efficiency level of firms from i in n 3.2. Equilibrium analysis
in the production of good u, and is given by zni(u) ≡ zi(u)τni. The compo-
nent zi(u) is country- and good-specific and represents the state of the There is an unbounded pool of potential entrants into the production
technology for all firms in i to produce good u. The parameter τni is of each good z. Given the draw zi(z), potential entrants decide whether
country-pair specific, but common across goods, with τii = 1. For n ≠ i, to pay fni and produce good z in country n by hiring local inputs. A firm
a lower τni reflects a lower efficiency level of affiliates from i in n, for from country i opens an affiliate in country n as long as (net) profits are
all goods. The variable τni can be interpreted as a measure of openness non-negative,
between the two MP partner countries. The degree of openness affects
the productivity of foreign affiliates from i in n: If τni = 0, then country πni ðzÞ ¼ max pni ðzÞqni ðzÞ−cn Sni ðzÞ−cn f ni ≥ 0; ð5Þ
Sni ðzÞ
n in completely closed to MP from i; the higher τni, the higher the degree
of openness of n to affiliates from i. where qni(z) is given by Eq. (4). Solving the firm's problem in Eq. (5)
Additionally, I assume that there is a fixed cost of opening an affiliate yields
in n by a firm from i, denoted by fni. This fixed cost can be thought as the
1 ν
costs of forming a subsidiary and production networks in the foreign πni ðzÞ ¼ mðτni zi ðzÞpni ðzÞÞ1−ν cnν−1 −cn f ni ≥0; ð6Þ
country, as well as an overhead cost of production. Fixed costs are
ν
country-pair specific and paid by each affiliate from country i to operate where m ≡ ð1−νÞν 1−ν N0. In a free-entry equilibrium, (net) profits must
in country n, in units of country n's input bundle. Notice that the model be zero, πni(z) = 0. Equivalently, the maximum price for good z that af-
does not distinguish between affiliates and plants; thus, I assume that filiates from i can charge in n has to be equal to the minimum unit cost of
one affiliate equals one plant. Let cn denote the unit cost of country n's production,
input bundle (if labor were the only input into production, cn would
be the wage wn).14 Then, the value of the fixed cost is cnfni. Local firms cn hni
e N0, a constant pni ðzÞ ¼ ; ð7Þ
also bear a fixed cost denoted by fnn and normalized to N zi ðzÞ
defined below.15
To complete the description of the environment, following EK, I as- where
sume that the efficiency parameter zi(u) is a random variable, indepen-
1−ν
dently drawn across goods from a Fréchet distribution with country- f ni
hni ≡ N ; ð8Þ
specific parameter Ti, and common parameter θ, Fi(zi) = exp(−Tiz−θ i ), τ ni
for zi N 0, all i, and θ N max(1, σ − 1). The parameter Ti summarizes
the state of technology in country i, while the parameter θ governs the and N ≡ (1 − ν)ν − 1ν−ν N 0. The unit cost associated with MP from
heterogeneity in efficiency across the continuum of goods (i.e., lower country i in n collapses to cnhni, where hni aggregates the effects of the
θ, more heterogeneity). Draws are also independent across countries, variable and fixed components of doing MP by i in n. With τnn = 1
and fnn conveniently normalized to N e ¼ N1=ðν−1Þ , hnn = 1. Thus, similar
so that F(z1, …,zI) = ∏ Ik = 1Fk(zk). From now on, since goods are iden-
tical except for their country-specific efficiency level, I drop the index u to the setup in Ramondo and Rodríguez-Clare (2013), the parameter hni
and label each good by the vector z ≡ [z1, …,zI]. can be interpreted as the overall efficiency loss incurred by affiliates of
multinational firms from i producing in n.
In a competitive equilibrium, affiliates from the country with the
14
One can think that labor in the host country constitutes the major part of the input
lowest minimum unit cost supply good z to market n,
bundle cost cn, while headquarter services and managerial ability are embedded in the
productivity parameter zi(u). In this sense, some productive resources come from the
home country; profits would be the “remuneration” to these factors of production.
pn ðzÞ ¼ min pni ðzÞ: ð9Þ
i
15
Eq. (12) below makes clear that fnn regulates the ratio of domestic to foreign firms into
n: The lower fnn is, the higher is the ratio of domestic to foreign firms. The calibration of fnn
would, hence, require data on the number of domestic affiliates, which I do not have. This
Expenditures in country n on good z produced by affiliates from i in n
is one of the reasons to normalize fnn. Additionally, fnn is not needed to solve the equilibri- are given by the expenditure function derived from the CES utility func-
um of the model and for the calibration procedure, as explained below in detail. tion, (pn(z)/Pn)1 − σXn, where Pn is the price index associated with the
112 N. Ramondo / Journal of International Economics 93 (2014) 108–122

(a) Number of affiliates from in ( ) (b) Average MP from in ( )

8 7
Log of Number of Affiliates from i in n

Log of MP per Affiliate from i in n

7 6

5
6
4
5
3
4
2
3
1
2 0

1 -1

0 -2
10 11 12 13 14 15 16 10 11 12 13 14 15 16
Log of country n size Log of country n size
Note: Country size refers to GDP in current U.S. dollars.

Fig. 2. Multinational production and the size of the destination market. (a) Number of affiliates from i in n (Mni). (b) Average MP from i in n (xni). Note: market size refers to GDP in current
U.S. dollars.

composite good Qn, and Xn are total expenditures in country n, Xn = good z to country n at price pni(z) (left panel). At the industry level,
PnQn. Substituting pn(z) in Eq. (9) into the familiar CES formula for the there are constant returns to scale, as indicated by the flat supply
price index Pn and integrating yield curve (in blue). With decreasing returns at the affiliate level, the size
of the market determines not only the total quantity of good z sold by
!−1=θ i in n, but also the number of affiliates from country i producing good
X
I
−θ
Pn ¼ γ ðhni cn Þ Ti ; ð10Þ z in country n (right panel).
i¼1

3.2.1. The extensive and intensive margins of multinational production

with γ ≡ Γ(1 + (1 − σ)/θ)1/(1 − σ) N 0 and Γ(.) representing the gamma Total expenditures in country n devoted to goods produced by affil-
function. iates from country i in n aggregate over all goods z for which country i is
Fig. 4 illustrates the workings of the model's equilibrium. Assume the lowest-cost producer,
that there are three possible source countries of affiliates that can pro-
vide good z to country n: k, i, and n (i.e., the local producers). Decreasing h−θ T
returns to scale and fixed costs at the affiliate level deliver U-shaped X ni ¼ X ni −θi X n : ð11Þ
h T
k nk k
unit cost curves that differ across affiliates of different origins. The min-
imum unit cost of production for good z (i.e., for which net profits are
zero) is given by pnk(z), pni(z), and pnn(z), for affiliates from country k, This is MP done by country i in n or, equivalently, total revenues of
i, and n, respectively. With free entry, the technology with the lowest affiliates from i in n. This expression is analogous to the one for bilateral
minimum unit cost is used. In this example, affiliates from country i trade flows in EK, with the only difference being that the unit cost of
are the ones with the lowest minimum unit cost and, hence, provide country n's input bundle drops from this gravity-like equation since

(a) Number of affiliates from in ( ) (b) Average MP from in ( )

8
Log of Number of Affiliates from i in n

7
Log of MP per Affiliate from i in n

7 6

6 5

4
5
3
4
2
3
1
2
0
1 -1

0 -2
10 11 12 13 14 15 16 10 11 12 13 14 15 16
Log of country i size Log of country i size
Note: Country size refers to GDP in current U.S. dollars.

Fig. 3. Multinational production and the size of the source market. (a) Number of affiliates from i in n (Mni). (b) Average MP from i in n (xni). Note: market size refers to GDP in current U.S.
dollars.
 N. Ramondo / Journal of International Economics 93 (2014) 108–122 113

Table 3
Gravity and the margins of multinational production.

Dep. var. All countries OECD(18)

log Mni logxni log Mni logxni

(1) (2) (3) (4) (5) (6) (7) (8)

log dni −0.63⁎⁎⁎ −0.74⁎⁎⁎ −0.16⁎⁎⁎ −0.25⁎⁎⁎ −0.69⁎⁎⁎ −0.79⁎⁎⁎ −0.05 −0.32⁎⁎⁎
(0.039) (0.027) (0.027) (0.043) (0.059) (0.061) (0.046) (0.11)
log Xn 0.49⁎⁎⁎ 0.39⁎⁎⁎ 0.54⁎⁎⁎ 0.37⁎⁎⁎
(0.031) (0.029) (0.050) (0.047)
log Xi 0.93⁎⁎⁎ 0.43⁎⁎⁎ 0.97⁎⁎⁎ 0.30⁎⁎⁎
(0.029) (0.025) (0.049) (0.041)
Fixed effects No Yes No Yes No Yes No Yes
Observations 1097 1097 1094 1094 304 304 301 301
R-squared 0.55 0.97 0.27 0.93 0.65 0.98 0.27 0.96

Notes: Mni is the number of affiliates from i in n; xni is revenues per affiliate from i in n; dni is geographical distance between i and n; and Xn (Xi) is GDP in n (i). Fixed effects refer to two sets
of destination- and source-country fixed effects. Robust standard errors are in parentheses. Levels of significance are denoted.
⁎⁎⁎ p b 0.01.
⁎⁎ p b 0.05.
⁎ p b 0.1.

MP by i in n entails hiring in the host country n. Eq. (11) establishes that Conversely, revenues per affiliate from i in n – the intensive margin – in-
the larger the destination market (higher Xn), or the more productive creases only with the value of the fixed cost,
the firms from i (higher Ti), the larger MP flows Xni; the higher the MP
cost (hni), the lower MP flows Xni. Additionally, more-productive coun- X ni c f
xni ≡ ¼ n ni : ð14Þ
tries (i.e., higher Ti) should have larger market shares, both abroad and M ni 1−ν
domestically; in other words, they should have higher outward and
lower inward MP, as a share of country i's expenditures. Conversely, The empirical regularities documented in Section 2 establish that
less-productive countries should have a higher share of production in both the number of affiliates and revenues per affiliate from i in n de-
the hands of foreigners and smaller market shares abroad. crease with distance, and change with characteristics of the countries
Note that the model with decreasing returns to scale at the firm level hosting and sending affiliates abroad — in particular, with their size.
and both fixed and variable costs of MP has implications about not only From Eqs. (13) and (14), it is clear that, for the model to match the
the aggregate MP flows by i in n, but also the number of affiliates and the MP facts, it has to be that the fixed cost, fni, decreases with distance
average MP per affiliate from i in n. These are the novel features of the and also changes with characteristics of host and source markets —
model with respect to previous work and the ones that allow the among then, country size; at the same time, the variable component
model to make direct contact with the facts presented in Section 2. τni has to decrease with distance (i.e., firms from i going to a more-
The number of affiliates from i in n – the extensive margin – is given distant market n have higher efficiency losses) and vary with host-
by and source-market characteristics. To calibrate the model, rather than
imposing these relationships, I will use Eqs. (13) and (14), as well as
X ni the data on Mni and xni , to calculate directly the matrices of MP param-
Mni ¼ ð1−ν Þ : ð12Þ
cn f ni eters, fni and τni. Section 4 explains the procedure in detail.

Replacing Xni in Eq. (11) into Eq. (12), after some algebra, yields 3.2.2. Closing the model
  I assume that there is an intermediate and final goods sector, denot-
−ð1−ν Þθ−1 θ θ
Mni ¼ λ f ni τni T i X n P n ; ð13Þ ed by the superscripts “g” and “f,” respectively. The CES aggregate good
Q is used for the production of each intermediate good z and for final
where λ ≡ (1 − ν)(Nγ)−θ N 0. We should expect more affiliates from i in consumption. Following EK and Alvarez and Lucas (2007), the produc-
n when the fixed cost is lower (lower fni), the variable component is tion function for each intermediate good z is Cobb–Douglas, with pa-
higher (higher τni), and the host market is larger (higher Xn). rameter β b 1, and combines the CES aggregate good with labor, so
that the input bundle is given by Sg(z) = lg(z)βQg(z)1 − β.16 The unit
cost of the input bundle for intermediate goods in country n is then
Table 4
given by
The margins of multinational production. Decomposition.
g β  g 1−β
Dep. var. log Mni cn ¼ Bwn P n ; ð15Þ
All OECD(18)

log xni 0.62⁎⁎⁎ 0.49⁎⁎⁎ 0.71⁎⁎⁎ 0.59⁎⁎⁎

where B ≡ β−β(1 − β)β − 1 N 0. The unit cost of production for the final
(0.011) (0.022) (0.019) (0.036) good in country n is given simply by the price index associated with the
log Xn 0.55⁎⁎⁎ 0.55⁎⁎⁎ 0.59⁎⁎⁎ 0.59⁎⁎⁎ composite intermediate good, Pgn, in Eq. (10). In a competitive equilibri-
(0.019) (0.017) (0.033) (0.030) um, the price of the final good is equal to its unit cost of production,
Constant 0.13 0.082
(0.26) (0.466) f g
Source country fixed effect No Yes No Yes Pn ¼ Pn : ð16Þ
Observations 1094 1094 301 301
R-squared 0.78 0.96 0.84 0.97 Replacing cgn in Eq. (15) into Eq. (10) and solving for Pgn yields
Notes: Mni is the number of affiliates from i in n; xni is revenues of affiliates from i in n, as a !−1=ðβθÞ
share of country n's GDP, Xn. Robust standard errors are in parentheses. Levels of signifi- g
X
I
−θ
cance are denoted. Pn e
¼γ hni T i wn ; ð17Þ
⁎⁎⁎ p b 0.01. i¼1
⁎⁎ p b 0.05.
⁎ p b 0.1. e ¼ ðγBÞ−1=β N0 and hni is defined in Eq. (8).
where γ
114 N. Ramondo / Journal of International Economics 93 (2014) 108–122

pn(z) Dn(z) D1
n(z)
pn(z)

pnk(z)

pnn(z)

pni(z)
pni(z)

affiliate output Mni(z) M1

ni (z)

Fig. 4. Equilibrium with multinational production.

Total expenditures in country n on the final good are equal to total hni. The lower the cost of hosting foreign affiliates, the higher the gains
income in country n, Xfn = wnLn, while total expenditures on the for the host country.
intermediate-goods sector are Xgn = Xfn/β.17 In this economy, Xfn is Analogous to the results for the gains from trade in Arkolakis et al.
gross domestic product (GDP) and Xgn is gross output. (2012), the gains from MP for country n can be written as a function
Finally, since countries interact only through MP, not trade, the equi- of MP shares. For i = n, the expression in Eq. (11) collapses to
librium for each country can be solved separately and the wages nor-
malized in each economy. I set wn = 1, for all n. " #−1
X nn XI
−θ T i
¼ hni : ð21Þ
3.3. The gains from multinational production X gn i¼1
Tn

The gains to country n from opening up to multinational firms

are computed as the change in the real wage of moving from autarky Replacing Eq. (21) into Eq. (20) yields
(hni → ∞, for all i ≠ n) to a situation with multinational firms (hni b ∞,
for i ≠ n). The real wage for country n in an equilibrium with MP is
 
given by X nn −1=ðβθÞ
GMP n ¼ : ð22Þ
!1=ðβθÞ X gn
wn  −1 X
I
−θ
e
¼ γ hni T i ; ð18Þ
P nf i¼1
If MP flows were normalized by Xfn, the counterpart of GDP in the
while for hni → ∞, for all i ≠ n, Eq. (18) collapses to the real wage under data, replacing Xgn = Xfn/β in Eq. (22), would deliver
autarky,
!−1=ðβθÞ
wn  −1 1=ðβθÞ X
e
¼ γ ðT n Þ : ð19Þ GMP n ¼ 1−β xni ; ð23Þ
P nf i≠n

Dividing Eq. (18) by Eq. (19), the gains from MP for country n are
obtained: where xni ≡ Xni/Xfn. The expression in Eq. (23) is very convenient since the
!1=ðβθÞ gains of moving from autarky to a situation with the observed MP flows
X
I
−θ Ti can be calculated using the data on revenues of affiliates from i in n, as a
GMP n ¼ hni : ð20Þ
Tn share of country n's GDP, as well as some calibrated values for β and θ.
i¼1
Moreover, the gains of moving from the current situation observed in
Opening up to foreign firms gives country n access to better produc- the data to any counterfactual situation (e.g., lower MP costs) can be
tion technologies from the rest of the world, “discounted” by the cost written as GMPn′ = ((1 − β ∑ i ≠ nxni′)/(1 − β ∑ i ≠ nxni))−1/(βθ),
where xni′ indicates the share of MP done by firms from i in n in the
17
Total expenditures in the composite intermediate good are Xgn = PgnQn = Pgn(Qfn + Qgn) =
counterfactual equilibrium. Of course, calculating each x ni′ entails
Xfn + (1 − β)Xgn, where (1 − β) is the expenditure share on the input Q from the calibrating the model parameters and solving for the model's
intermediate-goods sector. Thus, Xgn = (1/β)Xfn. equilibrium.
 N. Ramondo / Journal of International Economics 93 (2014) 108–122 115

Table 5
Data and model variables.

Country Country MP shares No. of affiliates GDP R&D RGDPL

Name Code Outward Inward Outward Inward Data θ = 8.2 θ = 4.2

Argentina ARG 0.05 0.17 221 1227 0.03 0.19 0.24 0.29 0.09
Australiaa AUS 0.15 0.37 603 2474 0.04 0.79 0.88 0.45 0.21
Austriaa AUT 0.09 0.39 1766 4038 0.02 0.57 0.86 0.36 0.13
Beneluxa BNX 1.05 0.62 8465 5198 0.07 0.69 0.87 0.52 0.28
Brazil BRA 0.02 0.21 260 2466 0.08 0.09 0.18 0.30 0.10
Canadaa CAN 0.31 0.51 1766 3440 0.07 0.74 0.84 0.52 0.27
Switzerland CHE 1.08 0.49 5637 3066 0.03 0.70 0.92 0.41 0.17
Chile CHL 0.03 0.21 108 495 0.01 0.12 0.24 0.19 0.04
China CHN 0.00 0.07 250 3781 0.14 0.10 0.07 0.34 0.12
Czech Republic CZE 0.01 0.71 96 1850 0.01 0.31 0.43 0.24 0.06
Denmarka DNK 0.26 0.19 2017 999 0.02 0.75 0.84 0.35 0.13
Spaina ESP 0.05 0.28 486 3394 0.07 0.43 0.65 0.42 0.19
Finlanda FIN 0.59 0.31 1126 1674 0.01 1.48 0.72 0.39 0.16
Francea FRA 0.18 0.22 4921 6514 0.16 0.73 0.76 0.59 0.36
Great Britaina GBR 0.29 0.48 5823 5254 0.16 0.63 0.76 0.59 0.36
Germanya GER 0.46 0.33 21,553 12,724 0.23 0.72 0.80 0.66 0.45
Greecea GRC 0.01 0.11 116 472 0.01 0.33 0.53 0.27 0.08
Indonesia IDN 0.01 0.19 102 1059 0.02 0.05 0.08 0.18 0.04
India IND 0.01 0.05 145 525 0.05 0.04 0.05 0.21 0.05
Ireland IRL 0.26 0.55 499 1159 0.01 0.55 0.80 0.30 0.09
Israel ISR 0.04 0.10 280 225 0.01 0.59 0.59 0.30 0.10
Italya ITA 0.10 0.18 2204 2798 0.13 0.34 0.74 0.47 0.23
Japana JPN 0.20 0.07 8196 1667 0.48 1.12 0.79 0.85 0.72
Korea KOR 0.11 0.11 709 933 0.05 0.56 0.47 0.42 0.18
Mexico MEX 0.04 0.32 357 2062 0.05 0.06 0.28 0.25 0.07
Norwaya NOR 0.25 0.11 1286 673 0.02 0.92 1.17 0.36 0.14
New Zealanda NZL 0.15 0.69 122 470 0.01 0.61 0.62 0.28 0.08
Poland POL 0.01 0.33 139 3402 0.02 0.37 0.28 0.30 0.10
Portugala PRT 0.05 0.58 100 972 0.01 0.35 0.50 0.29 0.09
Russia RUS 0.04 0.04 197 600 0.03 0.84 0.21 0.41 0.18
Swedena SWE 0.44 0.43 2953 4355 0.03 1.06 0.75 0.44 0.20
Thailand THA 0.01 0.49 89 1585 0.01 0.03 0.15 0.16 0.03
Turkey TUR 0.01 0.10 219 622 0.03 0.10 0.21 0.23 0.06
United Statesa USA 0.22 0.19 18,572 8558 1.00 1.00 1.00 1.00 1.00
South Africa ZAF 0.09 0.20 229 881 0.02 0.08 0.15 0.20 0.04
Average All 0.19 0.30 2206 2490 0.06 0.51 0.55 0.37 0.17
Average OECD(18) 0.27 0.34 4560 3649 0.14 0.74 0.78 0.49 0.28
Average non-OECD(18) 0.11 0.25 561 1526 0.04 0.28 0.32 0.28 0.09

Notes: Inward MP shares: Total revenues of foreign affiliates in n, as a share of n's GDP. Outward MP shares: Total revenues of affiliates from n abroad, as a share of n's GDP. Inward number
of affiliates: total number of foreign affiliates in n. Outward number of affiliates: total number of affiliates abroad from n. R&D refers to employment in the R&D sector, as a share of total
employment. Gross domestic product in current U.S. dollars (GDP), R&D, and real GDP per capita (RGDPL) are relative to the U.S.
a
Countries in the OECD(18).

4. Quantitative analysis and Acquisition from Thomson and Reuters. Each observation is at the
country-pair level, an average over 1996–2001. The data refer to non-
4.1. Data financial affiliates in all sectors; no data are available by sector.
The empirical counterpart for MP by i in n in the model, Xni, is total
In contrast with the bilateral trade data, there is no systematic data revenues of affiliates of multinational firms from country i in n, while
set on the bilateral activity of foreign affiliates of multinational firms. I the empirical counterpart for Mni is the total number of affiliates (not
assemble a data set that includes total revenues of affiliates from coun- plants) of firms from i in n. I normalize total revenues of affiliates from
try i in n and the number of affiliates of firms from country i producing country i in n by GDP in country n, in current U.S. dollars, from the
in n. A foreign affiliate is defined as a firm that has more than 10% of its World Development Indicators, an average over 1996–2001.19 Reve-
shares owned by a foreigner. The data include both OECD and non- nues per affiliate from i in n are calculated by dividing total revenues
OECD countries from which I consider a sample of 35 countries with by the number of affiliates from i in n.
real GDP per capita of more than 2000 dollars (PPP-adjusted), and, in In the calibration, I again use data from the World Development In-
general, better-quality data on multinational firms. Countries are listed dicators on research and development (R&D) employment, as a share of
in column 1 of Table 5. total employment, by country, an average over 1996–2001. Data on real
The main information sources are both published and unpublished GDP per capita, PPP-adjusted, are from the Penn World Tables (7.1), an
data from the FDI country profiles found in the FDI statistics recorded average over 1996–2001. Table 5 presents total revenues of foreign affil-
by the Investment and Enterprise Program at UNCTAD.18 Additionally, iates from and into country n, as a share of country n's GDP, total num-
missing values for both bilateral revenues and number of affiliates are ber of affiliates from and to country n, GDP, R&D employment shares,
imputed following the procedure in Ramondo et al. (2013b), which and real GDP per capita, by country.
complements the UNCTAD data with data on cross-border Mergers

19
Taking averages smoothes out year-to-year fluctuations and trade imbalances about
18
Unpublished data are available upon request at [email protected]. which the theory is silent.
116 N. Ramondo / Journal of International Economics 93 (2014) 108–122

Table 6 is 0.53. According to Alvarez and Lucas (2007), the share of intermediate
Calibrated costs of multinational production. goods in the tradable-goods sector is 0.5.21 Hence, I choose β = 0.5.
Calibration with Finward
n Foutward
n θ = 8.2 θ = 4.2
4.2.2. The parameter v
τinward
n τoutward
n τinward
n τoutward
n
I set the span of control parameter v to 0.7, as estimated by Cooper
Argentina 5.16 1.43 0.47 0.60 0.16 0.26
and Haltinwanger (2006) using a model with capital adjustment costs,
Australiaa 10.98 4.47 0.80 0.70 0.37 0.29
Austriaa 11.70 2.74 0.51 0.47 0.19 0.19 plant-level data, and an indirect inference approach.
Beneluxa 18.55 31.49 1.05 1.12 0.51 0.60
Brazil 6.42 0.49 0.64 0.58 0.22 0.25 4.2.3. The parameter θ
Canadaa 15.43 9.29 0.95 0.88 0.46 0.37 It is well-known from quantitative trade models that the parameter
Switzerland 14.59 32.47 0.59 1.04 0.24 0.56
Chile 6.19 0.85 0.33 0.36 0.08 0.17
θ cannot be separately identified from the cost parameters. I set this pa-
China 2.00 0.10 0.42 0.41 0.13 0.15 rameter to 8.2 and 4.2, alternately. These two values encompass the es-
Czech Republic 21.38 0.48 0.46 0.44 0.15 0.15 timates in Eaton and Kortum (2002), Bernard et al. (2003), Alvarez and
Denmarka 5.61 7.87 0.58 0.71 0.20 0.32 Lucas (2007), Simonovska and Waugh (2014), Donaldson
Spaina 8.30 1.39 0.74 0.64 0.31 0.24
(forthcoming), and Ramondo and Rodríguez-Clare (2013).
Finlanda 9.27 17.56 0.68 1.07 0.29 0.49
Francea 6.63 5.33 1.02 0.81 0.52 0.35
Great Britaina 14.29 8.73 1.39 0.87 0.75 0.40 4.2.4. The vector L
Germanya 10.03 13.91 1.14 0.97 0.69 0.47 Similar to the approach in Alvarez and Lucas (2007), I set the vector
Greecea 3.22 0.36 0.41 0.43 0.12 0.16 L to match the observed GDP data, in current U.S. dollars, in each coun-
Indonesia 5.63 0.17 0.29 0.30 0.07 0.17
India 1.38 0.18 0.08 0.00 0.07 0.09
try n. In the model, GDP is given by wnLn. Since wages are normalized to
Ireland 16.49 7.75 0.51 0.70 0.18 0.33 one, it is straightforward to set Ln equal to GDP as observed in the data.
Israel 2.97 1.16 0.28 0.36 0.07 0.14 Since the model abstracts from physical and human capital, the reason
Italya 5.49 3.02 1.09 0.84 0.52 0.36 to choose GDP as the empirical counterpart of wnLn is to interpret Ln as
Japana 2.08 5.95 0.89 0.96 0.47 0.37
efficiency-equipped units of labor in country n, that are different from
Korea 3.23 3.20 0.51 0.79 0.19 0.33
Mexico 9.67 1.12 0.52 0.50 0.16 0.23 the number of people, or workers, in country n.
Norwaya 3.16 7.49 0.52 0.74 0.18 0.32
New Zealanda 20.73 4.47 0.52 0.54 0.19 0.27 4.2.5. The vector T
Poland 10.02 0.23 0.51 0.40 0.19 0.13 Finally, as in Ramondo and Rodríguez-Clare (2013), I assume that
Portugala 17.46 1.61 0.73 0.63 0.27 0.24
Tn/Ln varies directly with the share of R&D employment observed in
Russia 1.23 1.07 0.44 0.54 0.16 0.20
Swedena 12.76 13.28 0.65 0.78 0.29 0.38 the data. For example, the share of R&D employment for Ireland is
Thailand 14.77 0.19 0.37 0.40 0.09 0.21 0.54 the one in the United States, indicating that in Ireland, TIRL/LIRL is
Turkey 2.88 0.21 0.31 0.34 0.08 0.13 half as high as the one in the United States. This assumption reflects
United Statesa 5.69 6.62 1.97 1.09 1.54 0.45
the common idea from semi-endogenous growth models that the
South Africa 5.99 2.78 0.26 0.61 0.06 0.40
Average All 8.90 5.70 0.65 0.65 0.29 0.29 stock of ideas in a country is proportional to its size, Ln, and that coun-
Average OECD(18) 10.08 8.09 0.87 0.79 0.44 0.35 tries have different productivity in the research sector. Additionally, cal-
Average non-OECD(18) 7.52 3.30 0.40 0.49 0.13 0.23 ibrating T's in this way avoids the appearance that small rich countries,
Notes: Finward
n indicates the total value of the fixed costs for foreign affiliates into country n, such as Denmark, have systematically high T's (see Ramondo et al.,
as percentage of country n's GDP, 100 × ∑ k ≠ ncnfnkMnk/(wnLn). Foutward
n indicates the total 2012). I set TUSA = 1.
value of the fixed costs for affiliates from country n abroad, as percentage of country n's
GDP, 100 × ∑ k ≠ nckfknMkn/(wnLn). τinward is the average country-pair specific productiv-
n
4.2.6. MP parameters
ity for foreign affiliates into country n, ∑ i ≠ nτni/(I − 1). τoutward
n is the average country-
pair specific productivity for foreign affiliates from country n, ∑ i ≠ nτin/(I − 1). To calibrate the matrices f and t, it is convenient to use Eq. (8) and
a
Countries in the OECD(18). calibrate first the matrix h = {hni}i ≠ n.
1. Given the set Δ = {β,ν,θ,L,T}, I pick the matrix h (with hnn = 1), to
match exactly the matrix of bilateral revenues of affiliates observed
in the data. To do so, I solve the I × (I − 1) system of equations,
4.2. Calibration procedure
data model
X ni −X ni ðh; ΔÞ ¼ 0;
The parameters to calibrate in the model are: β, v, θ, the vectors L =
{L1,…,LI} and T = {T1,…,TI}, and the matrices t = {τni}i ≠ n, and f = {fni}i ≠ n.
where Xmodel
ni (h;Δ) is given by Eq. (11).
2. The matrix f is computed directly from the data on the bilateral rev-
enues per affiliate, using Eq. (14),
4.2.1. The parameter β
The labor share β is calibrated to match the ratio of GDP to gross out-
put in the large sample of countries. My own calculations for the United xdata
ni
States, using the Annual Industry Accounts from the Bureau of Economic f ni ¼ ð1−νÞ : ð24Þ
cn ðh; ΔÞ
Analysis (BEA), for the period 1996–2001 and non-financial sectors
only, suggest that this ratio is around half.20 Jones (2011), using the
OECD input–output database, calculates for a sample of 35 OECD coun- The equilibrium variable cn is calculated using Eq. (15), which, in
tries and 9 non-OECD countries, typically for the year 2000, a share of turn, uses the expression for the equilibrium price index in
intermediate goods in production that ranges from around 0.4 for Eq. (17), which requires the matrix h, calculated in step 1.
Greece to 0.68 for China, with 0.47 for the United States; the average
21
They match the share of value added in gross output using input–output data from the
BEA (1996–1999), and UNIDO Industrial Statistics database (1998). They include as trad-
able sectors mining, agriculture, manufacturing, and tradable services. Moreover, labor is
interpreted, as I do here, as labor-plus-capital, or value-added, not just compensation of
20
Ratios are very similar whether the Government sector is included or not. employees. These are the reasons why their share is higher than the 0.21 calculated by EK.
 N. Ramondo / Journal of International Economics 93 (2014) 108–122 117

Table 7
Gravity and the costs of multinational production.

Dep. var. All countries OECD(18)

log fni log τni log fni log τni

(1) (2) (3) (4) (5) (6) (7) (8)

log dni −0.31⁎⁎⁎ −0.25⁎⁎⁎ −0.24⁎⁎⁎ −0.18⁎⁎⁎ −0.09⁎ −0.32⁎⁎ −0.12⁎⁎⁎ −0.24⁎⁎⁎
(0.039) (0.049) (0.021) (0.024) (0.049) (0.13) (0.021) (0.054)
log Xn 0.49⁎⁎⁎ 0.26⁎⁎⁎ 0.50⁎⁎⁎ 0.27⁎⁎⁎
(0.048) (0.024) (0.049) (0.022)
log Xi 0.45⁎⁎⁎ 0.16⁎⁎⁎ 0.34⁎⁎⁎ 0.13⁎⁎⁎
(0.032) (0.019) (0.044) (0.019)
Fixed effects No Yes No Yes No Yes No Yes
Observations 1076 1076 1097 1097 304 304 304 304
R-squared 0.15 0.78 0.13 0.86 0.37 0.81 0.46 0.73

Notes: Calibration with θ = 8.2. dni is geographical distance between i and n; and Xn (Xi) is GDP in n (i). Fixed effects refer to two sets of destination- and source-country fixed effects. Robust
standard errors are in parentheses. Levels of significance are denoted.
⁎⁎⁎ p b 0.01.
⁎⁎ p b 0.05.
⁎ p b 0.1.

3. The matrix t is calculated as a residual using Eq. (8), 4.3. Results

Table 6 reports summary statistics for the matrix of calibrated fixed

ν−1
f costs f and of country-pair specific productivity t, by country. The last
τ ni ¼ N ni ; ð25Þ
hni three rows present averages. Results in columns 1 and 2 do not depend
on the value of the parameter θ and represent the value of the fixed cost
paid by all foreign affiliates in country n, and by all affiliates from coun-
with hni and fni coming from steps 1 and 2, respectively. try n abroad, respectively, as a share of country n's GDP:

Some remarks about the calibration procedure are in order. First, the
X M ni cn f ni outward X M cf
data include both positive and zero observations on the bilateral reve- inward
Fn ¼ 100  ; Fn ¼ 100  in i in
:
nues of affiliates. The model does not generate exact zeros in step 1, i≠n wn Ln i≠ w L
n n
but it generates extremely small MP flows by simply picking a very
high hni for those observations with Xdatani = 0.22 In step 2, observations
with zero MP cannot be included because xni ≡ X ni =M ni is undetermined In the calibrated model, these two variables are pinned down direct-
for Xdata = Mdata = 0. Consequently, these observations also are not in- ly by the data on inward and outward MP shares (columns 1 and 2 in
ni ni
cluded in step 3 to calculate τni. These indeterminacies, however, are in- Table 5) and v = 0.7: Using Eq. (14) and xni ≡ Xni/(wnLn) implies that
nocuous since the only variable that matters for computing the Mnicnfni/(wnLn) = (1 − ν)xni. For the average country in the sample, the
equilibrium is the price index in Eq. (17), which requires only the matrix total value of the fixed cost that foreign multinational firms paid for
h from step 1. The statistics presented in the next subsection do not in- opening affiliates in country n (Finward
n ) represents 8.9% of the host
clude calibrated values corresponding to observations with zero MP in country's GDP, while aggregating the value of the fixed cost that multi-
the data. national firms paid for their affiliates abroad (Foutward
n ) represents less
Second, the calibration procedure cannot assign a value to fnn in step than 6% of the home country's GDP. Among the subset of OECD coun-
2 since data on the number of domestic affiliates and, consequently, tries, both the average inward and outward values of the fixed costs
data on revenues per domestic affiliate are not available. As explained are much higher than for the non-OECD countries. The reason is simple:
in Section 3, fnn is normalized to a positive constant. This normalization Finward
n and Foutward
n are pinned down directly by the data on the share of
is innocuous since it does not affect the equilibrium variables of the revenues of foreign affiliates (in terms of the host country's GDP) and of
model (i.e., Pn and cn), which can be calculated immediately after step 1. affiliates abroad (in terms of the home country's GDP), respectively,
Third, even though the calibrated model exactly matches GDP in cur- which are much higher for the former group of countries than for the
rent U.S. dollars, in each country, it is not designed to match real GDP latter group (see Table 5). Additionally, there is great heterogeneity
per capita, whose model counterpart is wn/Pfn. The last two columns of across countries in the fixed cost of engaging in MP activities that purely
Table 5 present the implications of the model regarding this variable, reflects the heterogeneity in MP flows, from Finward
n = 1.23 percent for
for the calibrations with θ = 8.2 and 4.2, respectively. The average coun- Russia, the country with the lowest inward MP shares, and Foutward
n =
try in the data is richer than in the calibrated model (0.55 versus 0.39 for 0.10 percent for China, the country with the lowest level of firm interna-
the calibration with θ = 8.2), while the correlation between this vari- tionalization, to Finward
n = 21 percent for Czech Republic, a small and
able in the data and in the model is around 0.6. The data also present very open country, and Foutward
n = 32 percent for Switzerland, with out-
more variation than the calibrated model (s.d. of 0.31 versus 0.18), ward MP flows that represent more than 100% of its GDP. For the United
with a minimum of 0.05 (India) and a maximum of 1.17 (Norway); States, the largest source of multinational firms, the total value of the
the model with θ = 8.2 achieves a minimum of 0.16 (Thailand) and a fixed cost faced by U.S. affiliates abroad represents almost 7% of U.S.
maximum of 1 (the United States). GDP, while for affiliates of foreign multinationals in the United States,
Finally, from steps 1 and 2 in the calibration procedure, it is clear that these costs represent almost 6% of U.S. GDP.23
the calibrated model matches the extensive and intensive margins of
23
MP exactly as observed in the data. The question is how the two compo- To calculate the value of the fixed cost paid by affiliates from a given source country i
into a host country n, as a share of country n's GDP, the data in Table 2 suffice: With
nents of the MP cost, fni and τni, do that job; I turn to that question next.
Mnicnfni/(wnLn) = (1 − ν)xni and v = 0.7, the value of the fixed cost for foreign affiliates from i
in n represents 0.26% of the host country's GDP (i.e., 0.3 × 0.0087), for the average country
22
Since zero observations represent only seven percent of the observations, including or pair, 0.56% (i.e., 0.3 × 0.0187) for the average country pair belonging to OECD(18), and
ignoring them in the procedure delivers undistinguishable results. 0.021% (i.e., 0.3 × 0.0007) for the average country pair not belonging to OECD(18).
118 N. Ramondo / Journal of International Economics 93 (2014) 108–122

Table 8 that the fixed cost of opening affiliates of firms from country i in n be-
The role of the fixed and variable costs of multinational production.
haves according to the following log-linear equation:
Model with MP costs
f f f f
Fixed and variable Only variable Only fixed log f ni ¼ imn þ exi þ δ logdni þ εni ; ð27Þ
Average
Number of affiliates from i in n 83 5287 1358 where imf and exf denote, respectively, destination- and source-country
Revenues per affiliate from i in n 34 1.5 374 fixed effects, dni denotes geographical distance between i and n, and εfni
captures all other idiosyncratic country-pair specific component that
Median are orthogonal to the regressors. Replacing Eq. (27) in Eq. (26) yields
Number of affiliates from i in n 9 89 42
Revenues per affiliate from i in n 15 1.5 3 f x x f
logxni ≡ −logð1−ν Þ þ δ logdni þ Si þ Dn þ εni ; ð28Þ
Coefficient of variation (s.d. to mean)
Number of affiliates from i in n 105 182 257 where the variables Sxi and Dxn collect all the terms specific to the source
Revenues per affiliate from i in n 57 7 184 and destination countries of foreign affiliates, respectively, that affect
Notes: The magnitudes for the calibrated model with both variable and fixed costs of the intensive margin of MP.24 Estimates in column 4 of Table 3 indicate
MP correspond to the magnitudes observed in the data (considering only observations that δf = −0.25: MP partners that are twice as far away have 25% lower
with non-zero MP flows). Revenues per affiliate are in millions of current U.S. dollars.
fixed costs. This result is confirmed when I estimate Eq. (27) by OLS
(column 2 of Table 7), reconciling the model with the fact that revenues
per affiliate decrease with distance in the data.
Applying a similar reasoning for the bilateral number of affiliates in
Eq. (13) yields
Turning to the variable component of MP costs – the country-pair
specific productivity parameter τni – its average is 0.65 for the calibra- logMni ¼ logλ−½ð1−νÞθ þ 1log f ni þ θlogτni þ logT i þ logX n
tion with θ = 8.2, meaning that a foreign affiliate of a multinational þ θlogP n ; ð29Þ
firm experiences, on average, an efficiency loss of 35% (i.e., 0.35 =
1–0.65) when moving production abroad, with respect to producing where (1 − ν)θ + 1 N 0. Given the OLS estimates in columns 1 and 2 of
in it in home market. Losses are very different depending on the type Table 3, this expression implies that τni has to depend negatively on geo-
of country: Among richer countries, the loss is much lower than graphical distance: The farther away the origin is from the host country
among poorer non-OECD countries — around 10–20% versus 50–60%, of affiliates, the greater the loss in country-pair specific productivity ex-
both for foreign affiliates entering those countries (1 − τinward
n ) and af- perienced by foreign firms (i.e., lower τni). Assume that the variable cost
filiates from those countries going abroad (1 − τoutward
n ). In particular, of opening affiliates of firms from country i in n behaves according to the
U.S. affiliates abroad are, on average, 24% more productive than firms following log-linear equation:
at home when going to a rich country in the OECD, but 10% less produc-
τ τ τ τ
tive when they are hosted by a poorer country. Conversely, the United logτni ¼ imn þ exi þ δ logdni þ εni ; ð30Þ
States receives much more productive affiliates of foreign firms than
any other country. This outcome is driven by the fact that revenues where imτ and exτ denote, respectively, destination- and source-country
per affiliate and the number of affiliates in the United States are, on av- fixed effects, and ετni is an idiosyncratic country-pair specific component
erage, $133 million and 245, respectively, much higher than the aver- that is orthogonal to the regressors. Replacing Eq. (30) in Eq. (29) yields
ages recorded in Table 2. h i
τ f f M M
The values for τni from the calibration with θ = 4.2 are much lower. logMni ¼ logλ þ θδ −δ −ð1−ν Þθδ logdni þ Si þ Dn þ εni ; ð31Þ
A lower θ means more disperse productivity draws within a country
across goods. That translates into MP shares being less elastic to changes where the variables SM M
i and Dn collect all the terms specific to the source
in MP costs—see Eq. (11). Hence, matching the observed shares requires and destination countries of foreign affiliates, respectively, that affect
higher MP costs hni in step 1 of the calibration procedure, which, in turn, the extensive margin of MP, while εni collects the error terms in
translate into lower τ's in step 3. Eqs. (27) and (30).25 Clearly, with θ = 8.2, v = 0.7, and δf = − 0.25,
the OLS estimate in column 2 of Table 3 implies that δτ = −0.18: Affil-
4.3.1. The role of the fixed and variable costs of multinational production iates originating in countries twice as far from the host country experi-
How do the calibrated MP costs reconcile the model with the facts in ence 18% higher losses in their country-pair specific productivity.
Section 2 that relate the intensive and extensive margins of multina- Column 4 in Table 7, again, confirms the result by estimating Eq. (30)
tional activities with characteristics of the host and the source countries by OLS.
and geographical distance between partners? The elasticities of the var- Further, columns 1 and 3 of Table 7, similarly to columns 1 and 3 of
iable and fixed MP costs with respect to distance and the characteristics Table 3, explore the relation between the variable and fixed cost of MP
of the origin and destination markets are intimately related to the way with host- and source-country size, for the sample both OECD and
the model links the bilateral number of affiliates and revenues per affil- non-OECD countries.26 While both components of the costs of doing
iate, respectively, to the MP costs and characteristics of the partner MP increase less than proportionally with the size of the origin and
countries. the destination countries of affiliates, fixed costs are more responsive:
Regarding revenues per affiliate, taking logs in Eq. (14) yields For instance, a host country that is 10% larger has fixed costs almost
5% larger, while it attracts foreign affiliates that experience, on average,
logxni ≡ −log ð1−ν Þ þ log f ni þ logcn : ð26Þ 2.5% lower efficiency losses.
As a final exercise, I assess the importance of including in the model
both fixed and variable costs of engaging in multinational activities in
This expression, together with the estimates in columns 3 and 4 of
24
Table 2, makes clear that, for the model to capture the data, fni has to Sxi ≡ logexfi and Dxn ≡ logcn + logimfn.
25 τ
SM f M f
i ≡ logTi + [(1 − ν)θ + 1]exi + θexi , and Dn ≡ logXn + θlogPn + [(1 − ν)θ + 1]imn +
vary negatively with distance between the country of origin and the θimτn.
destination of affiliates. Additionally, this cost has to vary with charac- 26
Regressing the host- and source country fixed effects on host- and source-country
teristics of the source and destination countries of affiliates. Assume GDP, respectively, delivers virtually the same coefficients.
 N. Ramondo / Journal of International Economics 93 (2014) 108–122 119

Table 9
The gains from liberalizing multinational production.

Calibration with: Gains from MP-autarky Gains from MP liberalization Ln

θ = 8.2 θ = 4.2 θ = 8.2 θ = 8.2

(1) (2) (3) (4) (5)

Argentina 1.022 1.044 2.21 1.08 0.03

Australiaa 1.051 1.101 2.56 1.16 0.04
Austriaa 1.054 1.109 2.26 1.17 0.02
Beneluxa 1.094 1.193 2.28 1.24 0.07
Brazil 1.028 1.055 2.33 1.10 0.08
Canadaa 1.075 1.152 2.26 1.21 0.07
Switzerland 1.070 1.142 2.64 1.20 0.03
Chile 1.027 1.053 2.31 1.10 0.01
China 1.008 1.016 1.79 1.03 0.14
Czech Republic 1.113 1.234 3.03 1.26 0.01
Denmarka 1.024 1.048 2.23 1.09 0.02
Spaina 1.037 1.074 2.27 1.13 0.07
Finlanda 1.042 1.083 2.17 1.14 0.01
Francea 1.029 1.057 1.91 1.10 0.16
Great Britaina 1.069 1.138 2.02 1.19 0.16
Germanya 1.046 1.091 1.82 1.14 0.23
Greecea 1.014 1.027 1.99 1.05 0.01
Indonesia 1.024 1.048 2.26 1.09 0.02
India 1.004 1.011 1.60 1.02 0.05
Ireland 1.081 1.165 2.92 1.22 0.01
Israel 1.012 1.025 1.95 1.05 0.01
Italya 1.023 1.047 2.13 1.09 0.13
Japana 1.009 1.017 1.37 1.04 0.48
Korea 1.014 1.027 1.99 1.05 0.05
Mexico 1.044 1.087 2.57 1.14 0.05
Norwaya 1.013 1.026 1.87 1.05 0.02
New Zealanda 1.109 1.224 2.64 1.26 0.01
Poland 1.046 1.091 2.49 1.15 0.02
Portugala 1.087 1.178 2.87 1.23 0.01
Russia 1.005 1.010 1.61 1.02 0.03
Swedena 1.060 1.121 2.30 1.18 0.03
Thailand 1.07 1.144 2.85 1.20 0.01
Turkey 1.01 1.024 1.94 1.05 0.03
United Statesa 1.02 1.049 1.20 1.08 1.00
South Africa 1.03 1.051 2.30 1.09 0.02
Average All 1.042 1.085 2.20 1.13 0.09
Average OECD(18) 1.048 1.096 2.12 1.14 0.14
Average non-OECD(18) 1.035 1.071 2.26 1.11 0.04

Notes: Columns 1 and 2: changes in the real wage from autarky (i.e., hni → ∞ for i ≠ n) to the calibrated equilibrium. Column 3: changes in the real wage from the calibrated equilibrium to
an equilibrium with 50% higher τni, for all i ≠ n. Column 4: changes in the real wage from the calibrated equilibrium to an equilibrium with 50% lower fni, for all i ≠ n.
a
Countries in the OECD(18).

order to capture the patterns of the extensive and intensive margins of much more of them than observed in the data. Additionally, as hinted
e for all n, i,
MP observed in the data. To do so, I first assume that f ni ¼ N, by the coefficient of variation, the heterogeneity across country pairs
and recalibrate the matrix of country-specific productivity t to match for both margins of MP would be too high, and the distributions too
exactly the total revenues of affiliates from i in n. Alternately, I assume skewed (medians are extremely low in comparison to the means).
that τni = 1, for all n, i, and recalibrate the matrix of fixed costs f to In conclusion, to capture both the extensive and intensive margins of
match exactly the total revenues of affiliates from i in n. What are the MP observed in the data, both fixed and variable components for MP
implications for revenues per affiliate and the number of affiliates costs are necessary. Not only the average magnitudes of the two mar-
from i in n in the two cases? Table 8 shows the results for the calibrated gins, but also their distribution across country pairs, would be off.
model with θ = 8.2.27
A model with only variable costs of engaging in multinational activ-
ities would overstate the extensive margin of MP: There would be, on 5. The gains from multinational production
average, too many very small affiliates. Moreover, as indicated by the
coefficients of variation, the heterogeneity across country pairs in MP I use the calibrated model to evaluate the gains from engaging in
flows would come from the extensive margin exclusively since affiliates multinational activities, both from autarky and to a situation in which
from different source countries operating in the same host country the activity of multinational firms is liberalized. These gains are mea-
would have the same average size (i.e., the same amount of revenues sured as the change in the real wage in country n, wn/Pfn, of going from
per affiliate).28 The distribution of the extensive margin across the calibrated to the counterfactual equilibrium.
country-pairs would be extremely skewed, much more than in the As shown in Eq. (23), the gains from MP-autarky can be written as a
data, as suggested by the differences between the mean and the median. function of the share of foreign affiliates' revenues in the GDP of the host
Conversely, a model with only fixed costs of engaging in multina- country. Using the revenue shares observed in the data, which the cali-
tional activities would generate very large affiliates and, on average, brated model matches exactly, β = 0.5, and θ = 8.2 and 4.2, alternately,
columns 1and 2 in Table 9 present the implied gains of moving from
27
Results for θ = 4.2 yield a very similar pattern (not shown). MP-autarky to an equilibrium with the observed MP flows, for each
28 e xni ¼ Nc
From Eq. (14), with f ni ¼ N, e n =ð1−ν Þ. country in the sample.
120 N. Ramondo / Journal of International Economics 93 (2014) 108–122

0.12

CZE
NZL
0.1
log GMPn = - 0.0074xlog Ln + 0.016
BENELUX
Log of gains from MP-autarky
PRT
0.08 IRL

CAN
THA CHE
GBR

0.06 SWE
AUT
AUS
POL GER
MEX
0.04 FIN
ESP

BRA FRA
CHL ZAF
IDN
DNK ITA USA
ARG
0.02

ISR GRCNOR KOR

TUR
CHN JPN
RUS IND
0
-5 -4.5 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0
Log of country size (relative to U.S.)

Note: Gains from MP are calculated as changes in the real wage from autarky (i.e.,
for ) to the calibrated equilibrium. Calibration with =8.2. Country size refers to GDP
in current U.S. dollars. The coefficient is significant at ten percent.

Fig. 5. Country size and the gains from MP-autarky. Note: Gains from MP are calculated as changes in the real wage from autarky (i.e., hni → ∞ for i ≠ n) to the calibrated equilibrium.
Calibration with θ = 8.2. Country size refers to GDP in current U.S. dollars. The coefficient is significant at 10%.

The gains from MP-autarky range from 0.4% (India) to 11% (Czech OECD(18) country gaining more than the average non-OECD(18) coun-
Republic). Among the richer OECD countries, the gains are larger than try. For example, China would experience increases in the real wage of
the gains for the remaining countries in the sample, indicating that almost 80% if country-pair specific productivities were 50% higher
the latter group is closer to MP-autarky than the former group. The than the ones in the equilibrium with the observed levels of MP. The
United States has gains from MP-autarky of 2.4%, much higher than same country would experience gains of only 3% if the fixed cost of
the gains from trade-autarky of 0.8% calculated by EK. This is simply be- opening foreign firms were reduced by 50%. Nevertheless, these three-
cause MP flows into the United States are higher than U.S. import flows. percent gains would be sustained by revenues of foreign firms in
These calculations suggest that the activity of multinational firms is an China of more than 30% (as a share of China's GDP), much higher than
important channel for the gains from openness. As expected, for θ = the FDI shares observed in 2010 for this country.29
4.2, the calculated gains are larger. Not surprisingly, small countries gain more from liberalizing access
To complement the results in Table 9 and give a better idea of the to foreign firms than large countries do, but less so in the case of lower-
distribution of gains across countries, Fig. 5 shows the gains from MP- ing the fixed cost: An OLS regression (with robust standard errors and a
autarky as a function of country size (Ln). The relationship between constant) delivers an elasticity of gains from lowering the variable MP
the gains from MP-autarky and country size is negative: A regression cost of −0.11 (s.e. 0.023) with respect to country size, while the gains
line with robust standard errors and a constant delivers an elasticity of from lowering the fixed MP cost delivers an elasticity of − 0.014 (s.e.
−0.0074 (s.e. 0.0039). Smaller countries are further away from autarky 0.008).
than larger countries, but the effect is not very big: Doubling size de- Interestingly, the response of the extensive and intensive margins of
creases the gains from MP-autarky by less than 1%. Calculations in MP is different for the two liberalization exercises. When the variable
Waugh (2010) and Fieler (2011) show that small (and rich) countries component of MP costs is liberalized, the average number of affiliates
have the largest gains from trade-autarky. Even though the correlation of firms from country i in n goes up from 83 to 480, while the average
between country size and the gains from MP-autarky is not very strong, revenue per affiliate increases from $34 to $46 millions of current U.S.
the results in Fig. 5 suggest that openness to trade and openness to MP dollars, across country pairs, implying average bilateral MP flows of 6%
are correlated. of the host country's GDP (versus 0.95% in the baseline). When the
The liberalization exercises presented next are meant to illustrate fixed cost of MP is cut in half, the average number of affiliates, across
the workings of the model; they are not meant to be policy experiments. country pairs, goes up to 291, while the average revenues per affiliate
Columns 3 and 4 in Table 9 present the gains, in terms of the real wage, of firms from i in n goes down to $31 millions of current U.S. dollars.
of moving from the calibrated equilibrium to an equilibrium with 50% The two margins combine into MP shares of 2.8%, on average, almost
higher τ's and 50% lower f's for all country pairs, alternately. The largest three times higher as the share observed in the data.
gains are obtained when country-pair specific productivity is raised by
50% for foreign firms. In this case, the gains for the average country
reach 120%, with a lower value among OECD countries and a higher
one among the non-OECD countries. When fixed costs are reduced by 29
FDI shares refer to FDI stocks into China, as a share of China's GDP, rather than to rev-
50%, the gains are much lower, around 13%, with the average enues of foreign affiliates in China (not available yet).
 N. Ramondo / Journal of International Economics 93 (2014) 108–122 121

5.1. Comparison with the literature extensive margin, however, reacts disproportionately more than the in-
tensive margin to changes in distance and country size.
How do the gains from opening up to multinational firms calculated To capture the patterns observed in the data, I build a quantitative
here compare with other calculations in the literature? The closest cal- multi-country general equilibrium model that captures the cross-
culations are the ones in Ramondo and Rodríguez-Clare (2013). In a country patterns of the activity of multinational firms observed in the
model that features both trade and MP, the gains from MP are defined data. Multinational production (MP) in the model occurs when a tech-
as the change in real wages from a counterfactual equilibrium with nology that originates in a foreign country is used to produce a good
trade but no MP to an equilibrium with both international flows. The in the host country. But using a foreign technology for production en-
calculations in that paper suggest that the average gains from MP- tails a cost that is a combination of a variable and a fixed component.
autarky for an OECD country are around 10%. For θ = 4.2 (which encom- The model combines Lucas's (1978) span-of-control model – which en-
passes Ramondo and Rodríguez-Clare's (2013) estimates of θ), the tails decreasing returns to scale and fixed costs at the firm level plus
model I present here delivers gains from MP-autarky of almost 10% for perfect competition at the industry level – with the probabilistic repre-
the same set of OECD countries. This is not, however, very surprising: sentation of technologies from Eaton and Kortum's (2002) model of
Both models match MP shares as observed in the data — the key variable trade. While having firms with a limited span of control allows me to
for calculating the gains of moving from autarky to an equilibrium with distinguish between the extensive and intensive margins of multina-
the observed MP flows. tional activities, having a probabilistic representation of technologies al-
Calibrating an only-MP model using aggregate FDI stocks into a lows me to take the model to the multi-country data. Nevertheless, as a
country, Burstein and Monge-Naranjo (2009) calculate that removing simplification, the modeling strategy eliminates trade altogether: Affili-
taxes on foreign firms would increase a developing country's welfare ates use inputs and sell their output exclusively in the host country of
by 5%, on average. The gains from lowering the costs of engaging in mul- production.
tinational activities would be much higher according to my calculations, I calibrate the model to match the patterns of both revenues and
especially if the variable component of MP costs were lowered. These number of affiliates across different country pairs. Including both fixed
differences might suggest that corporate tax rates applied to foreign and variable components of the cost of engaging in multinational activ-
firms represent a small part of total MP costs.30 Finally, Garetto's ities is crucial to matching the extensive and intensive margins of MP.
(2013) calibration suggests that the losses from closing up to vertical Results indicate that the incentives to engage in foreign production
MP would be very small for the United States – around 0.7% of con- vary significantly across countries. In particular, overall, MP costs are
sumption per capita – while the gains from liberalizing this type of MP lower among rich countries than among poorer countries, which do
could be around 7%. For the United States, using my calibrated model, very little MP. But the value of the fixed cost that foreign affiliates pay
I find that the losses from closing up to horizontal MP would be much in richer countries is higher, as a share of the host country's GDP, than
higher – around 2% – while the gains from liberalizing horizontal MP the value they pay in poorer countries. The opposite is true for the var-
would be of about the same magnitude as Garetto's (2013) – 8% – if iable component of MP costs, which is much higher in poorer countries.
the fixed cost of MP were cut in half, but almost three times higher – I use the calibrated model to calculate the gains that a country expe-
20% – if the variable component of MP costs were reduced by half. riences from opening up to foreign firms. My calculations suggest that
How do the gains from MP compare with the gains from trade? EK the gains in real income per capita of moving from autarky to a situation
calculate a loss of moving to trade-autarky of 3.5% for an average with the observed MP activity would be around 4%; the gains of liberal-
OECD country (using θ = 8.3); the analogous number for MP is almost izing access to foreign firms would be large, coming mainly from lower-
5%, as shown in Table 9. Since the MP flows observed in the data are ing the variable – rather than the fixed – component of the MP cost.
larger than the trade flows, it is not surprising that the gains from autar- Indeed, an important topic left for future research is to understand bet-
ky to the observed equilibrium are higher for MP than for trade. Using a ter the nature of the restrictions on foreign firms.
large sample of countries, both Waugh (2010) and Fieler (2011) find This paper contributes to a recent, but growing literature that at-
that OECD countries are relatively closer to frictionless trade than non- tempts to quantify the gains from liberalizing international flows
OECD countries are, and that they are further away from autarky; other than trade in goods. The focus here is on the activity of multina-
their result is similar to the one I find for MP. The magnitude of the tional firms as one key channel through which countries interact. Even
gains from MP liberalization, however, suggests that there are much though the analysis is restricted to multinational firms as the only way
larger gains to be realized from liberalizing access to foreign firms of serving a foreign market, the calculated gains from the activity of
than from further liberalizing trade, a result similar to that in these firms are a benchmark to be compared with calculations coming
McGrattan and Prescott (2009). from trade-only quantitative models.

Acknowledgments
6. Conclusions
I would like to thank Fernando Alvarez, Costas Arkolakis, Christian
This paper quantifies the gains that a country experiences from Broda, Thomas Chaney, Russell Cooper, Cecilia Fieler, Stefania Garetto,
opening up to multinational firms. I start by examining new data on Hugo Hopenhayn, David Lagakos, Robert Lucas, Robert Shimer, and
the revenues and the number of foreign affiliates of multinational Nancy Stokey for their comments and discussions. I especially thank
firms across a large number of country pairs. As Eaton et al. (2011) do the editor, Jonathan Eaton, and Andrés Rodríguez-Clare, whose insight-
for French exporters, I document the importance of the extensive versus ful comments greatly improved the paper. I also benefited from com-
the intensive margin of multinational activities across countries, and I ments of participants in several conferences and seminars. I thank the
present evidence on how they react to geographical distance between Peter B. Kenen fellowship from the International Economics Section,
partners, and to host- and source-country characteristics. While larger Princeton University, for support and hospitality. Jeff Thurk provided
markets receive (and send) more and larger foreign affiliates, on aver- excellent research assistance. All remaining errors are mine.
age, distant markets receive fewer and smaller foreign affiliates. The

References
30
According to my calculations, the correlation between bilateral corporate tax rates ap- Alvarez, Fernando, Lucas, Robert E., 2007. General equilibrium analysis of the Eaton–
plied to foreign firms and bilateral revenues of foreign affiliates, as a share of the host Kortum model of international trade. J. Monet. Econ. 54 (6).
country's GDP, is very low in the cross section. Anderson, James, van Wincoop, Eric, 2004. Trade costs. J. Econ. Lit. 42, 691–751.
122 N. Ramondo / Journal of International Economics 93 (2014) 108–122

Arkolakis, Costas, Costinot, Arnaud, Rodríguez-Clare, Andres, 2012. New theories, same Jones, Charles I., 2011. Intermediate goods and weak links in the theory of economic de-
old gains? Am. Econ. Rev. 102 (1), 94–130. velopment. Am. Econ. J. Macroecon. 3 (2), 1–28.
Bernard, Andrew, Eaton, Jonathan, Jensen, J. Bradford, Kortum, Samuel, 2003. Plants and Lucas, Robert, 1978. On the size distribution of business firms. Bell J. Econ. 9 (2), 508–523.
productivity in international trade. Am. Econ. Rev. 93 (4), 1268–1290. Markusen, James, 1984. Multinationals, multi-plant economies, and the gains from trade.
Bloom, Nicholas, Van Reenen, John, 2007. Measuring and explaining management prac- J. Int. Econ. 16, 205–226.
tices across firms and countries. Q. J. Econ. 122 (4), 1351–1408. Markusen, James, 1995. The boundaries of multinational enterprises and the theory of in-
Burstein, Ariel, Monge-Naranjo, Alex, 2009. Foreign know-how, firm control, and the in- ternational trade. J. Econ. Perspect. 9 (2), 169–189.
come of developing countries. Q. J. Econ. 124 (1), 149–195. McGrattan, Ellen, Prescott, E., 2009. Openness, technology capital, and development.
Carr, David, Markusen, James, Maskus, Keith, 2001. Estimating the knowledge-capital J. Econ. Theory 144 (6), 2454–2476.
model of the multinational enterprise. Am. Econ. Rev. 91, 693–708. Ramondo, Natalia, Rodríguez-Clare, Andrés, 2013. Trade, multinational production, and
Cooper, Russell, Haltinwanger, J., 2006. On the nature of capital adjustment costs. Rev. the gains from openness. J. Polit. Econ. 121 (2), 273–322.
Econ. Stud. 73, 611–633. Ramondo, Natalia, Rodríguez-Clare, Andrés, Saborío-Rodríguez, Milagro, 2012. Scale ef-
Donaldson, David, 2014. Railroads of the Raj: estimating the economic impact of trans- fects and productivity across countries: does country size matter? NBER Working
portation infrastructure. Am. Econ. Rev. (forthcoming). Paper No. 18532.
Eaton, Jonathan, Kortum, Samuel, 2002. Technology, geography and trade. Econometrica Ramondo, Natalia, Veronica Rappoport, Kim J. Ruhl, 2013. “Horizontal versus Vertical For-
70 (5), 1741–1779. eign Direct Investment: Evidence from U.S. Multinationals”. Mimeo.
Eaton, Jonathan, Kortum, Samuel, Kramarz, Francis, 2011. An anatomy of international Ramondo, Natalia, Rodríguez-Clare, Andrés, Tintelnot, Felix, 2013b. Multinational Produc-
trade: evidence from French firms. Econometrica 79 (5), 1453–1498. tion Data Set.
Fieler, Ana Cecilia, 2011. Non-homotheticity and bilateral trade: evidence and a quantita- Razin, Assaf, Rubinstein, Yona, Sadka, Efraim, 2003. Which countries export FDI, and how
tive explanation. Econometrica 79 (4), 1069–1101. much? NBER Working Paper No. 10145.
Garetto, Stefania, 2013. Input sourcing and multinational production. Am. Econ. Rodríguez-Clare, Andrés, 2007. “Trade, Diffusion, and the Gains from Openness”. Unpublished.
J. Macroecon. 5 (2), 118–151. Simonovska, Ina, Waugh, Michael, 2014. The elasticity of trade: estimates and evidence.
Head, Keith, Ries, John, 2008. FDI as an outcome of the market for corporate control: the- J. Int. Econ. 92 (1), 34–50.
ory and evidence. J. Int. Econ. 74 (1), 2–20. UNCTAD, 2009. The World Investment Report.
Irarrazabal, Alfonso, Moxnes, Andreas, Opromolla, Luca D., 2013. The margins of multina- Waugh, Michael, 2010. International trade and income differences. Am. Econ. Rev. 100
tional production and the role of intra-firm trade. J. Polit. Econ. 121 (1), 74–126. (5), 2093–2124.