MEASURING THE INTERDEPENDENCE OF MULTINATIONAL FIRMS' FOREIGN INVESTMENTS.
Bosenberg, Simon ; Egger, Peter H. ; Merlo, Valeria 等
MEASURING THE INTERDEPENDENCE OF MULTINATIONAL FIRMS' FOREIGN INVESTMENTS.
I. INTRODUCTION
In general equilibrium and under resource constraints at the level
of countries, aggregate bilateral foreign investments and foreign
affiliate sales are known to be interdependent across host countries for
a given parent economy (Baltagi et al. 2007, 2008) as well as across
parent countries for investments in a given host economy (Blonigen et
al. 2008). Such interdependence leads to the transmission of
country-specific shocks in the world investment system, whereby
(geographically) more adjacent countries are stronger recipients as well
as transmitters of shocks. All existing evidence on interdependent
foreign investments, however, seems to pertain to aggregate investment
flows or stocks. While such aggregate analysis is useful to understand
the relevance of interdependence at the level of countries, theoretical
models used to motivate empirical studies of interdependence mainly rely
on effects between firms.
This paper contributes to the empirical literature measuring
interdependence of foreign direct investment (FDI) with a particular
focus on whether such interdependence arises across affiliates within
MNEs. To measure whether investment of one particular entity of an MNE
depends on investments carried out by other entities of the same MNE, we
discern different channels of interdependence such as input-output
relationships or mere geographic proximity. This approach relates our
paper to a recent literature on the organization of production along the
value chain (e.g., Antras and Chor 2013; Costinot et al. 2013), because
using weights on input-output dependence allows us to draw conclusions
about the relative position (upstream or downstream) of entities (and
countries) in the global value chain. Moreover, the approach is related
to the literature unveiling vertical international linkages in the
productivity (see Bernstein and Mohnen 1997; Keller 2002; Morrison Paul
and Siegel 1999; Smarzynska Javorcik 2004), growth and volatility
(Burstein et al. 2008; Kleinert et al. 2012; Oberhofer and Pfaffermayr
2013), and, hence, the profitability across units (firms, sectors, and
even countries). Comparing input-output-related interdependencies of
affiliates' investments to geographic-distance-related
interdependencies also permits drawing conclusions about the relative
importance of different channels of interdependence. Finally, while
there is a large literature on productivity spillovers from FDI on
domestic firms or generally across firms, we add to this literature by
providing evidence on interdependencies that occur within a firm, but
across affiliates and countries.
Our analysis is probably most closely related to the study by Chen
(2011). The paper by Chen (2011) is based on subsidiary-level data and
suggests horizontal as well as vertical interdependence in the location
of subsidiaries of MNEs. While the focus of our study is on the relative
importance of different channels (horizontal or vertical linkages)
through which investment at a given location is affected by marginal
investment decisions at other locations, the paper by Chen (2011)
examines the effect of existing production networks on the location of
foreign production. That is, Chen (2011) analyzes how horizontal or
vertical linkages determine the extensive margin of foreign investment.
In contrast, our analysis of interdependencies in the intensive margin
of investment is conditional on network location. While our focus is on
the intensive margin of investment, we explore the relative importance
of adjustments at the intensive margin (changes in investment by old
affiliates) versus adjustments at the extensive margin (investment by
new affiliates).
Our empirical analysis utilizes census-type panel data of all
German MNE parents and their foreign affiliates provided by Deutsche
Bundesbank. One obvious advantage of this dataset is that it allows us
to control for affiliate- and firm-specific characteristics, whose
omission in country-level studies might lead to aggregation bias.
Moreover, the availability of a large number of
firm-affiliate-host-country-year data points permits identification at
relatively great precision in comparison to country-pair-time aggregated
data. Also, the census-type data at the firm and affiliate level help
avoiding a bias associated with missing data points, which affects
virtually all attempts to analyze interdependence at the aggregate
foreign investment level from incomplete (e.g., survey-based or
otherwise selected) datasets. (1) The same, of course, is true for most
firm-level datasets, which cannot provide a similarly complete picture
as our dataset can. (2) Given that group effects are at the heart of our
analysis, the latter points are particularly important.
We model the stock of FDI invested in a foreign entity as to depend
on a weighted function of FDI stock elsewhere within the firm. (3) Our
estimation strategy does not condition on the total amount of FDI
invested in all units of a firm--by focusing on its allocation only--but
it allows aggregate investments to vary. Thus, we do not view
investments to be necessarily substitutive across locations but allow
them to be substitutive or complementary. For estimation and
identification of interdependence effects we specify a spatial
autoregressive (SAR) model which allows for fixed effects at the level
of foreign affiliates. A novel feature of our analysis is that the
structure of interdependence is affiliate-firm-specific to account for
differences in MNEs' vertical and geographical (horizontal)
organization of production. Moreover, using different weights to
distinguish between investments that are relatively more downstream
versus upstream versus geographically proximate enables us to provide
the following novel insights about the interdependence of investment
within firms: (i) what is the relative importance of geographical
proximity versus input-output linkages as channels through which
interdependencies in investment occur and (ii) what is the relative
importance of the upstream versus downstream channel.
The main findings of our analysis can be summarized as follows.
First, vertical (input-output) relations between affiliates seem to play
a greater role in explaining interdependence than horizontal ones (mere
geography). However, conditional on input-output relationships,
investment of a given entity declines with bigger proximate investments
at other entities of the same MNE. Second, while investments in vertical
input-related affiliates exert a complementary impact on investment in a
given affiliate (positive interdependence), the opposite is true for
investments at vertical output-related affiliates. In other words,
input-linked upstream investments of affiliates within an MNE's
affiliate network stimulate investments in more downstream affiliates,
whereas output-linked downstream investments of affiliates reduce
investments in more upstream affiliates. Hence, vertical interdependence
is asymmetric between upstream and downstream relationships. Third,
while vertical interdependence seems to be driven by adjustments at the
extensive margin (investment at other locations by new affiliates),
horizontal interdependence seems to be driven by the intensive margin
adjustments (changes in investment by old affiliates). Fourth, the
relative sensitivity to shocks varies to a relatively large degree
across firms (depending on the location and size of their affiliate
network) and across host countries (depending on the vertical and
horizontal size and structure of the hosted affiliate network).
The finding that more horizontal investments in proximate countries
lead to less investments at a given affiliate, conditional on
input-output relations, is consistent with the use of foreign affiliates
as export platforms (see Ekholm et al. 2007; Tintelnot 2017). Our
results regarding vertical linkages are consistent with findings of the
literature on spillover effects. Smarzynska Javorcik (2004) finds
evidence that positive (productivity) spillovers from foreign affiliates
to domestic firms mainly occur through backward linkages (through the
intermediate input channel), but not through forward linkages. We find a
similar pattern for a given foreign entity which is linked through the
input channel to upstream entities anywhere within the same MNE. Our
results also suggest a U-shaped pattern of investment along the value
chain, similar to the one found in the context of the evolution of
countries and industries (see Shin et al. 2012, for an empirical
application in the global electronics industry). This suggests that the
profitability is particularly high at the root or origin of the
production process or product cycle (where research intensity is high)
and also at the end of it (where input costs are low and the service
intensity is high). In our setting, with Germany as the parent country,
we would expect that processes with high research intensity in the first
part of the curve are mainly located in the parent country (from where
positive technology spillovers are transmitted elsewhere). As Germany is
a mature economy, we expect investments to be gradually shifted to other
countries with greater growth potential and profits. According to our
empirical estimates, for a given foreign affiliate, investments of input
suppliers are positively related to a recipient affiliate's own
investments, while investments of entities to which output is forwarded
are negatively related to an affiliate's own investments. This
suggests that investments are shifted downstream towards the end of the
production line along the right part of the U-shaped curve.
We may also interpret our results in light of the recent literature
on the organization of global value chains. In particular, Alfaro et al.
(2015) suggest that upstream or downstream integration choices depend on
the relative size of the elasticity of demand for a firm's final
product and the elasticity of substitution across sequential inputs. If
inputs are not particularly easy to substitute, investments across
vertically linked units are found to be positively correlated, in which
it is optimal for the firm to integrate only the most downstream stages.
This is consistent with our finding of a positive relationship between
investment of a given affiliate and investments of all other affiliates
within the MNE if the affiliate is linked through the input channel.
Moreover, while we cannot model the outsourcing decision, the findings
in Alfaro et al. (2015) would suggest that the integrated part of the
total value chain for the input-linked affiliates in our data tends to
be more downstream. If instead inputs are easy to substitute,
investments at different stages are negatively correlated and Alfaro et
al. (2015) suggest that a firm finds it optimal to integrate relatively
upstream stages. The negative interdependence found for affiliates
linked through the output channel is in line with the assumption that
the inputs supplied by these affiliates are characterized by a high
substitutability, suggesting that the integrated part of the value chain
we observe in our data for the output-linked affiliates is more
upstream. All this suggests that both goods produced and
relationship-specific investments made along the value chain become more
specialized and less substitutable, which is again in line with the
aforementioned U-shaped curve pattern.
The proposed empirical approach allows us to carry out a number of
interesting experiments. For example, we may gauge the relevance of an
asset reallocation across existing affiliates in response to
location-specific shocks. We are also able to assess the importance of
different channels of interdependence depending on location. This
includes an identification of particularly shock-prone locations from
the viewpoint of German investors (which account for a significant share
of world FDI). On average and conditional on a shock of the same size
across all existing affiliates there, the United States, China, and
Brazil are the most important sources of investment shocks (spillovers)
to German affiliates elsewhere. The German affiliates in Botswana,
Madagascar, Iceland, and Lebanon are the most important recipients of
investment shocks from German affiliates elsewhere. This shows that
intra-firm effects on investment are asymmetric with regard to their
impact on units upstream versus downstream. The general strength of
interdependence among the affiliates in an MNE and its qualitative
impact depend on the (vertical and horizontal) structure of an
MNE's affiliate network.
Our findings also have policy implications. Most obviously, the
degree to which a country is exposed to shocks transmitted from other
countries depends on the structure of FDI in such a country. Moreover,
the results imply that the interdependence of affiliate-level investment
decisions does not permit treating affiliates as independent within the
firm in empirical work without encountering biased and inconsistent
estimates of parameters and comparative static effects. This is relevant
for the analysis of policy effects such as the ones of national or
international tax policy, as responses to national and international tax
incentives have consequences on investments not only within but also
across national borders. Our analysis at the affiliate level suggests
that cross-border effects go beyond the (mechanical) interdependence
emerging and considered in structural general equilibrium models of
multinational firms.
The remainder of the paper is organized as follows. Section II
outlines the main theoretical reasons for investment interrelatedness
and how our paper relates to this literature. Section III describes the
empirical approach as well as the interdependence or linkage measures
used. Section IV describes the dataset used in the present analysis,
before Section V presents the findings. Section VI states a brief
conclusion.
II. REASONS FOR CONTAGIOUS INVESTMENTS AND RELATED LITERATURE
We may distinguish between three different explanations for
interdependence of foreign investments within firms but across
affiliates and countries: (i) vertical input-output linkages within
MNEs; (ii) internal capital markets of MNEs; (iii) correlated learning
over sequential investments of MNEs. Point (i) suggests that single
entities of the MNE are part of a global value chain and intermediate
goods are used by different affiliates at different production stages
within and across different countries. Bernard etal. (2010) show that
such production leads to tangible assets trade within an MNE. More
recently, Atalay et al. (2014) have argued that intra-firm transfers
along vertical production lines are often associated with intangible
assets. Intra-firm trade of intangible inputs may include transfers
related to the simultaneous use of technology and knowledge (McGrattan
and Prescott 2009) across locations and entities, which does not
necessarily require vertical goods linkages but is also relevant in the
context of horizontal FDI (see Markusen's 2002 knowledge capital
model of the multinational firm for theoretical arguments along those
lines; see Carr et al. 2001 and Markusen and Maskus 2002, for aggregate
evidence on vertical versus horizontal MNE activity decisions). Alfaro
et al. (2015) analyze outsourcing decisions along the value chain. They
argue that organizational decisions at a given stage of the value chain
affect all stages of the value chain, because the incentives to make
relationship-specific investments depend on investments made by upstream
suppliers. Depending on the relative size of the elasticity of demand
for the final good and the elasticity of substitution across sequential
inputs, investments at different production stages are found to be
positively or negatively correlated to each other.
Point (ii) argues that entities of MNEs are linked through an
internal capital market. Egger et al. (2014) demonstrate that MNEs may
and do use this market to allocate scarce funds. In particular, funds
are channeled to those affiliates with the highest excess return on
investment. Differences in this excess return are driven by capital
market frictions, differences in productivity, taxes, and local
institutions. Because the allocation of funds involves lending and
borrowing relationships between affiliates, the existence of internal
capital markets provides for a natural reason of why there is
interdependence across all--vertical and horizontal--entities of an MNE.
Point (iii) recognizes that investments might be connected through
complementarities at the extensive margin. Egger et al. (2013) show that
correlated learning causes interdependence since information gathered at
one location by one affiliate can be used to learn about conditions in
other, particularly similar countries. (4)
All of this suggests that an empirical approach modeling investment
at a given affiliate needs to consider the interdependence of such
investments across all affiliates within a firm. (5) Moreover, the
extent of interdependencies may vary between firms as well as
affiliates. A natural approach to model the extent of interdependence is
geographical distance. For instance, Keller and Yeaple (2013) argue that
tangible and intangible transfers from headquarters to affiliates
decline with distance from their headquarters. Another measure to
capture the degree of interdependence--or the closeness of
affiliates--is to model input-output relations between entities.
Depending on a firm's specialization, input-output relations
determine whether a firm is closer to the core of activities or not, and
whether it is further upstream or downstream or not, with potential
consequences for the prioritization of its investment plans. The
relative importance of horizontal linkages through geography among units
in the same sector and of vertical linkages through input-output (or
upstream and downstream) relationships among units in different sectors
at the level of affiliates is not known.
More formally, we could think of profitability of a foreign
investment in affiliate i at some time t, [[PI].sub.it], to be a
function of stocks of assets at all affiliates of the same firm,
[mathematical expression not reproducible] where [N.sub.f] is the number
of all affiliates of firm f to which i belongs. Assume that new
investments at location i and time t are proportional to their
profitability, whereby [mathematical expression not reproducible]
Moreover, let us parameterize the latter as [partial
derivative][FDI.sub.it]/[partial derivative][FDI.sub.jt] =
[w.sub.ijt][delta], where [w.sub.ijt] is a row vector reflecting
channels of influence of investment at j on i (such as input-output
channels, horizontal competition and cross-effects on sales, etc.) and
[delta] is a conformable column vector of importance weights. The
overall level of foreign assets at j and t then induces an overall
partial impact on [FDI.sub.it] of [w.sub.ijt] [delta][FDI.sub.jt]. All
other (non-i) affiliates together would have a joint partial impact on
[FDI.sub.it] of [[bar.FDI].sub.it] [equivalent to] [summation over
(j[not equal to]i[member of][N.sub.ft])] [w.sub.ijt][delta][FDI.sub.jt],
where the summation is over affiliates in the same firm f, [N.sub.ft].
III. EMPIRICAL APPROACH
This paper considers several channels of interdependence
determining individual firms' (intensive) marginal FDIs: one
related to mere geography (horizontal proximity) and other ones related
to input-output relationships (vertical proximity) between foreign
affiliates.
A. Econometric Model
Let us use indices f, i, and j to refer to the ith or jth foreign
affiliate of firm f. Altogether, there are F firms, and the fth firm has
[N.sub.f] foreign affiliates during the sample period. Because firms
enter and exit (and so do affiliates), the number of firms in year t
[member of](1 ..., T) is [F.sub.t] [less than or equal to] F and the
number of firm f's affiliates in year t is [N.sub.ft] [less than or
equal to] [N.sub.f]. Let us denote the set (as opposed to the number) of
foreign affiliates in year t and all years as [N.sub.ft] and [N.sub.f] =
[N.sub.f1] [union] ... [union] [N.sub.fT]. In year t, the stock of FDI
of firm f's affiliate i is determined as
(1) [FDI.sub.it] = [Z.sub.it] [delta] + [u.sub.it], [Z.sub.it] =
[[[bar.FDI].sup.D.sub.it], ([[bar.FDI].sup.S.sub.it]), [X.sub.it]],
[delta] = [[[lambda].sup.D],([[lambda].sup.s]),[beta]],
where [X.sub.it] captures exogenous determinants of FDI. Notice
that we put parentheses around [[bar.FDI].sup.s.sub.it] and
[[lambda].sup.s] in Equation (1). This is to indicate that more than one
concept of s may be considered at a time (e.g., input and output
relationships may enter separately). We usually refer to
[[bar.FDI].sup.D.sub.it] and [[bar.FDI].sup.s.sub.it] as interdependence
or linkage terms of [FDI.sub.it], and they are defined as
(2) [[bar.FDI].sup.l.sub.it] = [summation over (i,j [member
of][N.sub.ft])] [w.sup.l.sub.ijt] [FDI.sub.jt], l [member of] {D,s}.
The parameters [w.sub.ijt.sup.l] in Equation (2) are referred to as
linkage weights in the literature, and they aggregate other
affiliates' investments within a firm and year according to the
distance metric indexed by D and the input-output metric indexed by s.
They are normalized by a scalar as suggested by Kelejian and Prucha
(2010) as
(3) [w.sup.l.sub.ijt] = [w.sup.10.sub.ijt]/max [summation over
(i,j[member of][N.sub.ft])][w.sup.10.sub.ijt], [w.sup.10.sub.ijt]
[greater than or equal to] 0,
where [w.sup.10.sub.ijt] is the unnormalized counterpart to
[w.sup.l.sub.ijt], (6) having the property of [w.sup.10.sub.ijt] = 0 for
all units j = i.
B. Channels of Interdependence (Weights [w.sup.10.sub.ijt])
The channels of interdependence, or weighting schemes, considered
in this paper are based on geographical distance (l = D) and sectoral,
vertical proximity (l = s). While the former is measured by the great
circle distance between the countries two affiliates are based in, the
latter is based on measures of the intensity of input-output relations
consistent with the German input-output table.
Inverse Geographical Distance Weights ([w.sup.D0.sub.ijt]). We
define inverse geographical distance weights as
(4) [w.sup.D0.sub.ijt] = [d.sup.-1.sub.ij] [for all]i [member of]
[N.sub.ft],j [member of] [N.sub.ft],
where [d.sup.-1.sub.ij] denotes either the great circle distance
between the main cities of the (different) countries affiliates i and j
locate in, or the average internal distance of the (same) country i and
j locate in. (7) If the condition in (4) is not met, for example, if we
cross the boundaries of an MNE, [w.sup.D0.sub.ijt] = 0.
In robustness tests, we assume alternative decay functions for the
inverse distance. For this, we specify [d.sup.-1.sub.ij] as
[([d.sup.-1.sub.ij]).sup.2] and [([d.sup.-1.sub.ij]).sup.0.5]. The
reasoning behind these additional tests is That
[([d.sup.-1.sub.ij]).sup.2] puts relatively less weight on distant (more
weight on close-by) affiliates than in the benchmark with
[([d.sup.-1.sub.ij]).sup.1]. And [([d.sup.-1.sub.ij]).sup.0.5] puts
relatively more weight on distant (less weight on close-by) affiliates
compared with the benchmark (see Baltagi et al. 2007; Bode et al. 2012).
In an additional robustness test of the measure of distance within
countries, we give all affiliates within a country the same weight
across countries. Specifically, we set [w.sup.D0.sub.ijt] =
[d.sup.-1.sub.ij] = 1 whenever affiliates i and j are located in the
same country (implicitly assuming that intra-national distance plays no
role).
Input-Output Weights ([w.sup.s0.sub.ijt]). The second type of
interdependence measure, [w.sup.s0.sub.ijt], reflects the amount of
intermediate inputs and/or outputs which the sectors of affiliates i and
j typically exchange with each other by German standards. (8) The
elements [w.sup.I0.sub.ijt],[w.sup.O0.sub.ijt], [w.sup.IO0.sub.ijt]
measure the amount of inputs, outputs and inputs plus outputs,
respectively, the sector of i typically uses from/provides to the sector
of j. Notice that [w.sup.s0.sub.ijt] is time-variant for two reasons:
first, the intensity of input-output relationships changes over time;
second, the sectors in which i and j mainly operate in might change over
time. (9) More precisely, denote the German input-output matrix spanned
by the units of firm f in year t as [[OMEGA].sub.ft], and its (ij)th
element as [[omega].sub.ijt]. In this case, [[omega].sub.ijt] measures
the input of the sector of unit j from the one of unit i at time t. The
three measures of sectoral interdependence are defined as
(5) [w.sup.I0.sub.ijt] = [[omega].sub.ijt] [for all]i [member of]
[N.sub.ft],j [member of] [N.sub.ft],
(6) [w.sup.O0.sub.ijt] = [[omega].sub.ijt] [for all]i [member of]
[N.sub.ft],j [member of] [N.sub.ft]
(7) [w.sup.IO.sub.ijt] = [[omega].sub.ijt] + [[omega].sub.jit] [for
all]i [member of] [N.sub.ft], j [member of] [N.sub.f't], [N.sub.ft]
= [N.sub.f't].
If the conditions in (5) are not met, that is, if we cross the
boundaries of an MNE, [w.sup.I0.sub.ijt], [w.sup.O0.sub.ijt], and
[w.sup.IO.sub.ijt] are zero.
The Model Specification in Matrix Notation. For our approach
towards intra-firm investment interdependencies, it is important to note
that the weighting schemes focus on interdependencies of affiliates i
and j within parent firm f in a given year t. Therefore, the typical
weights matrix for firm f at time t has size [N.sub.ft] x [N.sub.ft] and
is defined as
(8) [w.sup.l.sub.ft] = [[w.sup.l.sub.ijt]] [for all]i,j [member of]
[N.sub.ft],
which has zero diagonal elements for all l = {D, s}. Stacking the
data for all firms / in year t, we obtain a block-diagonal [N.sub.t] x
[N.sub.t] matrix of the form
(9) [W.sup.l.sub.t] = diag ([w.sup.l.sub.ft]).
Thus, using [FDI.sub.t] = ([FDI.sub.fit]), [[bar.FDI].sup.l.sub.t]
= [W.sup.l.sub.t] [FDI.sub.t] = ([[bar.FDI].sup.l.sub.t]), and [X.sub.t]
= ([X.sub.it]) to denote the corresponding stacked vector of elements
across all firms, we may write the model for year t as
(10) [FDI.sub.t] = [[lambda].sup.D] [[bar.FDI].sup.D.sub.t] +
([[lambda].sup.s] [[bar.FDI].sup.s.sub.t]) + [X.sub.t][beta] +
[u.sub.t],
where [FDI.sub.t] , and u, are [N.sub.t] x I vectors, [X.sub.t] is
a matrix of dimension [N.sub.t] xk and [beta] is a k x 1 vector. (10)
Again, we indicate by parentheses in (10) that more than one .v-related
interdependence term may be present at a time. In general, the
interdependence terms [[bar.FDI].sup.l.sub.t] are endogenous. However,
the structure of interdependence of the model delivers valid
instruments. This becomes clear by writing the reduced form of the
deterministic part of the model,
(11) E ([FDI.sub.t]) = (I - [[lambda].sup.D][W.sup.D.sub.t] -
[([[lambda].sup.s][W.sup.s.sub.t])).sup.-1] [X.sub.t][beta],
where several matrices [[lambda].sup.s] [W.sup.s.sub.t] may enter
additively the parentheses of the inverse in (11). A Taylor-series
expansion together with the properties of [W.sup.l] suggests that [(I -
[[lambda].sup.D] - [W.sup.D.sub.t]
[[lambda].sup.s][W.sup.s.sub.t]).sup.-1] [X.sub.t] can be approximated
well by a polynomial function so that [[bar.FDI].sup.D] and
[[bar.FDI].sup.s] can be insgumented well by [bar.[X.sup.D]] =
[W.sup.D]X, [bar.[X.sup.s]] = [W.sup.S]X, [[bar.X].sup.Ds] =
[W.sup.D][W.sup.S]X, [[bar.X].sup.D] = [W.sup.D] [[bar.X].sup.D],
[[bar.X].sup.s] = [W.sup.S][bar.[X.sup.S]], and so on, where it is
sufficient in practice to use up to four powers (see Kelejian et al.
2004). We estimate a two-stage least-squares model with affiliate fixed
effects (FE2SLS). (11)
Some General Remarks on Interdependence. The parameters on the
variables [[bar.FDI].sup.l.sub.it]. with l [member of]{D, I, O, IO)
should be interpreted in the following way. A positive effect of
[[bar.FDI].sup.D.sub.it] means that, conditional on other determinants
of FDI of firm f in affiliate i at time t, an increase in investments in
closer affiliates within the same firm (i.e., ones with a bigger inverse
distance) stimulates investment at the margin in affiliate i at time t.
The latter we dub horizontal complementarity at the intensive foreign
investment margin within the firm. A negative effect of
[[bar.FDI].sup.D.sub.it] means the opposite, pointing to a substitutive
relationship among investments at the intensive margin.
A positive effect of [[bar.FDI].sup.s.sub.it] implies that positive
interdependencies are associated with the interdependence in terms of s
within a parent's affiliates network. For instance, we might
interpret a positive (negative) effect of [[bar.FDI].sup.I.sub.it], on
[FDI.sub.it] as evidence of an upstream vertical complementarity
(substitution) in investments. Similarly, we might interpret a positive
(negative) effect of [[bar.FDI].sup.O.sub.it] on [FDI.sub.it] as
evidence of a downstream it 'i vertical complementarity
(substitution) in investments. A positive (negative) effect of
[[bar.FDI].sup.IO.sub.it] on [FDI.sub.it] could then be dubbed evidence
of a general vertical complementarity (substitution) in investments. One
consequence of a greater such interdependence is the greater
vulnerability of affiliate networks in terms of shocks within the
network. Whether shocks travel at all, primarily, or more strongly
through mere geographic or input-output linkages is a question that only
the data can answer.
IV. DATA AND DESCRIPTIVE STATISTICS
A. Data on the Dependent Variable
The main source underlying our data is the Microdatabase Direct
Investment (MiDi) collected and provided by the German Central Bank
(Deutsche Bundesbank). The database represents an annual unbalanced
panel with German parent firms' individual affiliates as the unit
of observation. The data capture the universe of German MNEs as it is a
legal requirement for firms (and even private households) to report FDI
above a threshold of 3 million [euro] in their balance sheet and if the
participation is at least 10%. Indirect participating interests have to
be reported whenever foreign affiliates hold 10% (50% as of 2007) or
more of the shares or voting rights in other foreign enterprises with a
balance sheet total in excess of 3 million [euro]. (12) For our approach
we use the entire panel for the years 1997 to 2009. The dependent
variable in our approach are the fixed (and intangible) assets of
affiliate i attributable to parent f in year t in logs, [FDI.sub.it], as
available from MiDi and reported in million Euros.
B. Data Underlying the Channels of Interdependence
Data on latitudes (lat,), longitudes ([lon.sub.i]), and
geographical area (area,) underlying the geographic weighting scheme,
[w.sup.D.sub.ijt], are taken from CEPII's GeoDist database. (13)
The data underlying the input-output weighting schemes are taken from
annual input-output tables for the German economy over the period
1997-2007, which are publicly available from the German Federal
Statistical Office. Because input-output tables as of 2008 are not
comparable to the previous ones, (14) we use the one for 2007 for those
2 years. The time variance in input-output shares is minor so that this
procedure seems justifiable. The calculation of inputs and outputs by
the German Federal Statistical Office is based on the concept of a
homogeneous production unit, which is closer to an affiliate (or a
production plant) than a firm. The input, output, and input-output
matrices are of size sector X sector with altogether 71 sectors of
primary (raw material), secondary (manufacturing), and tertiary
(services) type of the German economy, based on the so-called CAP
classification. We aggregated this format to a 60 X 60 table to match it
onto the foreign affiliate statistics as provided by Deutsche
Bundesbank, based on the NACE industry classification (see Table A2 for
details). The columns of the input-output matrix represent the value of
inputs used in a production sector in million Euros, and its rows
represent the value of output of intermediate goods (or services)
produced in million Euros. (15) Hence, both a bigger number of
[w.sup.D.sub.ijt] and of [w.sup.s.sub.ijt] indicates greater proximity
between two affiliates i and j at time t.
C. Data on Explanatory Variables
The vector [X.sub.it] in Equation (1) contains firmtime-specific as
well as country-time-specific determinants of investment of MNE f at
affiliate i and time t. We employ the following affiliate-time-specific
variables from the MiDi database. First, [Sales.sub.it-1] and
[Employees.sub.it-1] capture general lagged characteristics of foreign
entities affecting investments in t. The former reflects
affiliate-specific market size (demand) in logs and the latter an
affiliate's supply capacity in terms of employment, also in logs.
Second, we include [Competition.sub.it-1], the number of German
competitors in the same sector and country as of the previous year. We
calculate this variable by counting all affiliates j[not equal to]i in a
country and year t - 1 by sector. Among the country-time-specific
explanatory variables, we include the following. First, we account for
the Corporate Income [Tax.sub.it]. Higher corporate taxes require a
higher rate of return on investment and, hence, we expect this variable
to be negatively related to affiliates' investments. Information on
the statutory corporate tax rates is gathered from databases provided by
the International Bureau of Fiscal Documentation and annual tax guides
issued by Ernst & Young, PwC, and KPMG. Moreover, we include
Financial [Freedom.sub.it], as published in the Heritage Indicators
Database, which measures the banking efficiency as well as the
independence of the financial sector from government control. At the
extremes, a value of 100 indicates negligible government interference,
whereas a value of 0 indicates repressive government interference. A
greater financial freedom is associated with better access to the local
capital market and lower costs of external financing. We expect this
variable to be positively related to affiliate Vs investments. Also, we
employ the local inflation rate Inflation,, from the International
Monetary Fund's World Economic Outlook, which reflects aspects of
the macro environment affiliate i is operating in. Finally, we include
Capital - Labor [Ratio.sub.it], reflecting relative factor endowments in
affiliate i's market in year t in logs, and [GDP.sub.it], the log
of real gross domestic product (GDP) at constant U.S. dollars of the
year 2000, as a measure for the size of a market at time t. The latter
two explanatory variables are taken from the World Bank's World
Development Indicator Database, where capital-labor ratios are
calculated using the perpetual inventory method to estimate capital
stocks. (16)
D. Descriptive Statistics
Our analysis is based on a sample of 21,598 foreign affiliates of
6,059 German MNEs over the period from 1997 to 2009, resulting in an
unbalanced panel with 134,781 observations. (17) The German affiliates
in our sample are present in altogether 112 countries (for an overview
see Table A1). (18) Using darker color to indicate bigger numbers of
affiliates, Figure 1 shows that affiliates are highly concentrated in
member countries of the European Union, Russia, China, Brazil, Canada,
and the United States, with a maximum of 2,443 affiliates for the
average year located in the United States. At the other extreme are
mainly African and some Asian countries with three or less affiliates in
the average year. Using darker color to indicate higher values of
average fixed and intangible assets per affiliate, Figure 2 suggests
that countries such as China, Brazil, and the United States which host
many affiliates also host larger affiliates, on average. However, also
countries such as Algeria and Cameroon with on average only 14 and 4
affiliates per annum, respectively, receive similar amounts of fixed and
intangible assets per affiliate.
While Figures 1 and 2 consider the number of affiliates and fixed
and intangible assets per affiliate in the average year covered by host
country, Figures 3 and 4 illustrate the geographic distribution of the
number of German parent companies and the fixed and intangible assets
per German parent company in the average year by host country. Overall,
the number of parent companies at a location tends to be large where the
number of affiliates is large, and assets per parent tend to be large
where the number of affiliates per parent and/or the assets per
affiliate are large in the average year. In countries such as
Russia--colored dark-blue in Figure 1 but lighter-blue in Figure 3--a
relatively large number of affiliates is held by a relatively small
number of parent firms. FDI (fixed and intangible assets) per parent
company is on average highest in Cyprus, followed by the United States
and China. Overall, most of and the biggest German MNEs mainly invest in
the European Union, North America, Brazil, Russia, and China.
In a next step, we describe the closeness or proximity of German
foreign affiliates and German FDI per firm and country in terms of three
channels of interdependence: input (vertical upstream) proximity, output
(vertical downstream) proximity, and geographic (horizontal) proximity.
The values in Figures 5, 6, and 7 are calculated by multiplying each
parent firm's weights matrix [W.sup.l.sub.ft] by a vector of ones
to obtain a measure of pure input, output, or geographic distance within
its network of foreign affiliates. This yields a firm-f-specific measure
of proximity in year t and dimension (or proximity channel) l. In each
country, we then calculate the average of this measure of proximity
across all parents weighted by the number of affiliates they hold and
the years they are present. This obtains an average measure of proximity
of affiliates per country within the German affiliate network in
dimension l. In Figure 5 the average affiliate in darker-colored
countries is closer in terms of inputs received from other members of
its network than the average affiliate in lighter-colored countries. In
Figure 6 the average affiliate in darker-colored countries is relatively
closer to other affiliates of its network in terms of output delivered
than in lighter-colored countries. Figures 5 and 6 suggest that input
and output proximity are, in general, relatively similar across
countries. Hence, well-connected affiliates through the downstream
channel tend to be also well connected through the upstream channel.
Nevertheless, there are some interesting differences. For instance, the
average German affiliate in the United States is relatively more related
to other entities within the average MNE in terms of inputs received
than in terms of output delivered. This is consistent, for example, with
the global allocation of production of large German car manufacturers
present in the United States. It generally makes sense to think about
the United States as being a large final market, and not being an
intermediate country in the global value chain. A comparison of Figures
5 and 6 with Figure 7 suggests that there is an obvious difference
between vertical (input-output) and horizontal (merely geographic)
proximity. For instance, while affiliates in Factory Asia (Baldwin 2007)
and South America are well connected vertically, their horizontal
proximity is relatively low. The opposite seems to be true for European
countries, on average. Other interesting examples are countries such as
Uruguay or Namibia, which are both more integrated through the output
channel. For Uruguay, this is consistent with the fact that many German
multinationals provide financial services from Montevideo to affiliated
entities located in South and Latin America. For Namibia, this is
consistent with the fact that some German firms produce raw materials
(e.g., cement) to be exported to other countries in Southern Africa.
Figures 8, 9, and 10 suggest a similar pattern for FDI (fixed and
intangible assets).
V. RESULTS
This section reports the parameter estimates and the consequences
of counterfactual shocks in the foreign affiliate system based on the
estimated model when relying on the specification outlined in Section 3.
Table 2 summarizes parameter estimates on the different (endogenous)
interdependency terms of FDI stock elsewhere in parent company f's
network on the FDI stock at affiliate i. All regressions include
affiliate-level (and, implicitly, parent-level) fixed effects as well as
a full set of aggregate year effects.
In columns 1 to 5 in Table 2 we use four different variables to
capture the channels of interdependence in foreign assets. First,
[[bar.FDI].sup.I.sub.it] is the input-weighted FDI stock of other
affiliates than i as defined above. Second, [[bar.FDI].sup.O.sub.it] is
the output-weighted FDI stock of other affiliates than i. Third,
[[bar.FDI].sup.IO.sub.it] is the input-plus-output-weighted FDI stock of
other affiliates than i. Fourth, [[bar.FDI].sup.D.sub.it] is the
inverse-(geographic-)distance-weighted FDI stock of other affiliates
than i. The weighting matrices that apply to the interdependence terms
are all maximum row-sum normalized to make obvious that the estimated
coefficients on interdependence terms are in the admissible parameter
space. Notice that this scalar-type normalization preserves the notion
of absolute proximity in the affiliate network of any MNE. As suggested
in Section 3.1, we use weighted exogenous regressors (applying the
respective linkage-channel-specific weight) to instrument the
interdependence terms (see Table 1 for those variables and the
respective summary statistics). In what follows, the instruments consist
of a full set of four instruments per linkage term. That is,
input-interdependence, output-interdependence, and
geography-interdependence are modeled separately, so that there are 12
instruments based on [[bar.Sales].sup.l.sub.it-1],
[[bar.Employees].sup.l.sub.it-1], Corporate Income Tax;(,
[[bar.Competition].sup.l.sub.it-1] for l [member of] I,O, D. (19) While
it has been shown above that the weighted exogenous variables can be
used as optimal instruments, there might be a concern about the
exclusion restrictions in the context of weighted variables measured at
the level of the firm. We will address this by using lagged weighted
affiliate characteristics as instruments for affiliate i but only
affiliate characteristics of other multinational firms in the same
country and year that are not related to i and belong to another
multinational firm. (20) Beside linkage terms and the mentioned
affiliate- as well as time-specific effects, we condition on a number of
control variables shown and summarized in Table 1.
The control variables affect affiliates' FDI as expected.
First, larger sales and a bigger number of employees per affiliate have
a positive effect on German MNEs' investment abroad. Second, a
higher level of local (corporate) profit tax rates reduces investment.
More precisely, a one-percentage-point (1 ppt) increase in the tax rate
ceteris paribus reduces local investment by -1.16% for the average
German MNE. This magnitude is broadly in line with previous findings
(for a meta-analysis see De Mooij and Ederveen 2006). Third, a sound
functioning of the financial sector in the host country of the
investment, measured by the financial freedom variable, raises
investment per affiliate there. Fourth, a higher inflation reduces local
foreign investment per affiliate negatively, as does having a German
competitor in the same sector and host country. The former reflects
adverse temporal (or cyclical) macroeconomic conditions, the latter
captures adverse structural (competitive) conditions. While the adverse
competitive effect is quantitatively relatively small, it is
statistically highly significant. Finally, an increase in a host
country's capital-labor ratio (which measures both a relative
capital abundance and the relative development of a host country) exerts
a positive effect on investment while GDP, as a measure of the size of
the host economy is positively related to investments.
A glance at the coefficients on the linkage terms in Table 2
suggests the following conclusions. First, horizontal (geographic)
linkages matter to a smaller extent, whereas vertical linkages via
input-output relations seem to matter more. There is clear evidence of
positive interdependencies to affiliates which are downstream from their
upstream network members. For affiliates which are upstream and close to
their downstream members the opposite seems to be true. This is
consistent with a U-shaped curve pattern in the context of the evolution
of countries and industries (see Shin et al. 2012, for an empirical
application in the global electronics industry). This relationship
suggests that the profitability is particularly high at the root or
origin of the production process or product cycle (where the research
intensity is high) and also at the end of it (where input costs are low
and the service intensity is high). In a developed country such as
Germany which is increasingly specializing in services, it is consistent
with this pattern that firms seek to shift their activity towards and
maximize their profit margins at the end of the production line.
Our findings seem to be also in line with Alfaro et al. (2015). The
latter paper suggests that the elasticity of demand for a firm's
final product, as well as the relative contractibility vis-a-vis stages
located upstream or downstream from a given production stage, determine
upstream or downstream integration choices. If inputs are not
particularly easy to substitute, Alfaro et al. (2015) suggest that the
incentive of a supplier to invest in a relationship-specific input is
higher, the larger the investments by upstream suppliers. This is
consistent with our finding of a positive relationship between
investment of a given affiliate and investments of all other affiliates
within the MNE if the affiliate is linked through the input channel.
Moreover, while we cannot model the outsourcing decision, the findings
in Alfaro et al. (2015) suggest that the integrated part of the total
value chain for input-linked affiliates in our data tends to be more
downstream. The negative interdependence found for affiliates linked
through the output channel is in line with the assumption that the
inputs supplied by these affiliates are characterized by a high
substitutability, suggesting that the integrated part of the value chain
we observe in our data for the output-linked affiliates is more
upstream. All this suggests that both goods produced and
relationship-specific investments made along the value chain become more
specialized and less substitutable.
Our findings finally confirm the results in Smarzynska Javorcik
(2004), who shows that (positive) productivity spillovers from FDI to
domestic firms mainly take place through backward linkages (through the
intermediate input channel), rather than forward linkages. While our
results support the view that such interdependencies also exist within
MNEs and productive assets, there seem to be even negative effects on
affiliates linked through the output or forward channel. (21)
Columns 3 and 4 present specifications with a different decay
function about the spatial process and the impact of inverse distance.
In particular, in column 3 we use the squared inverse distance, in
column 4 the square root of the inverse distance. Once we specify
alternative decay functions, the negative effect of
[[bar.FDI].sup.D.sub.it] becomes insignificant. This confirms that
vertical linkages seem to be more important for interdependence than
geography. In column 5 we test the robustness of the internal-distance
measure. Figure 7 suggests that average geographic proximity is lowest
for affiliates in large countries such as the United States, Canada,
Brazil, or China. This may have to do with the fact that the average
internal distance is used for affiliates within the same country. For
this reason it is of interest to test whether our main results are
affected by another weighting of affiliates within the same country. In
the specification in column 5, we give all affiliates within a country
the same weight, across all countries. Specifically, we set
[w.sup.D0.sub.ijt] = [d.sup.-1.sub.ij] - 1 whenever affiliates i and j
are located in the same country (implicitly assuming that intra-national
distance plays no role). This alternative treatment of affiliates within
the same country has no effect on our findings.
Column 6 shows results where we control for country-time effects.
These results confirm the impact of the three main variables of
interest, [[bar.FDI].sup.I.sub.it], [[bar.FDI].sup.O.sub.it], and
[[bar.FDI].sup.D.sub.it]. This shows that the estimated coefficients are
not biased through unobserved country-specific variables. Of course, all
variables measured at the level of countries and years are not
identified in this specification.
While our focus is on the intensive margin of investment (the
effect of marginal investment decisions at other locations on the level
of investment at a given location), we explore the relative importance
of adjustments at the intensive margin (changes in investment by old
affiliates) versus adjustments at the extensive margin (investment by
new affiliates). Column 6 of Table 2 distinguishes between an extensive
and an intensive margin effect. For this, we assume different slope
parameters on [[bar.FDI].sup.I.sub.it], [[bar.FDI].sup.O.sub.it], and
[[bar.FDI].sup.D.sub.it], respectively, depending on whether an
affiliate that contributes to the respective weighted variable is new
(an extensive margin adjustment) or not (an intensive margin
adjustment). The results suggest that (i) the input-output
interdependence is driven by the extensive margin, while the
coefficients on the intensive margin estimates show the same signs but
are no longer statistically significant; (ii) the geographical
interdependence is driven by the intensive margin. This result suggest
that the impact of vertical integration decisions on investment across
all affiliates in the network is more pronounced than that of horizontal
integration decisions.
We finally run regressions (the basic specification shown in column
1) at the level of industries. This might indicate whether there exist
heterogeneous spatial effects of horizontal foreign investment across
industries. We plot the estimated coefficients on
[[bar.FDI].sup.D.sub.it] by way of a kernel density plot (see Figure
11). The figure suggests that there is significant heterogeneity across
sectors. Based on those sectors with sufficient observations, the
average coefficient estimated [[bar.FDI].sup.D.sub.it] is clearly
negative.2- The results also reveal, however, that the effects may be
positive, depending on the industry affiliate i is operating in. This is
true for about one-third of the sectors we analyze. On average, the
negative substitution effect (as in export platform FDI models)
dominates the positive effects of information spillovers.
VI. ANALYZING THE CONSEQUENCES OF TAX SHOCKS
Based on Specification I in Table 2, we may quantify the effect of
a 1 ppt decrease, for instance, in the corporate profit tax rate of
country r, r = 1, ... R ([DELTA][[tau].sup.r] = -0.01) on the foreign
affiliates' FDI stock as follows:
(12) [mathematical expression not reproducible],
where [[??].sub.[tau]] is the estimated coefficient on the
corporate tax rate (1,16 in our preferred specification) and
[DELTA][[tau].sup.r] has entry (-0.01) only in rows corresponding to
affiliates located in r and zeros elsewhere. Notice that the total
effect in (12) takes into account that such a shock on i through
[DELTA][[tau].sup.r] will not only induce direct (or local) effects on
affiliate i, but it will induce indirect effects on other affiliates
which will themselves induce indirect effects back on i. This is
captured by the inverse [mathematical expression not reproducible] in
(12), which accounts for an infinite series of indirect effects within
each parent's affiliate network.
The effects consistent with (12) may be visualized as follows.
Define [N.sub.rt] and [N.sub.mt] as the sets of affiliates located in
countries r and m, respectively, in year t. Let [N.sub.rt] and
[N.sub.mt] be the respective numbers of affiliates in those countries in
year t. And denote the total number of countries by R and the total
number of affiliates in year t across all countries by [N.sub.t]. Then,
the average total effect of [DELTA][[tau].sup.r] on affiliates located
in r is
(13) [N.sup.-1.sub.rt] [summation over (i[member of][N.sub.rt])]
[DELTA][FDI.sup.r.sub.it].
The average indirect effect of [DELTA][[tau].sup.r] on affiliates
located outside of country r is
(14) [([[N.sub.t] - [N.sup.-1.sub.rt]).sup.-1] [summation over
(m[not equal to]r)] [summation over (i[member of][N.sub.rt])]
[DELTA][FDI.sup.m.sub.it].
The average indirect effect over all [DELTA][[tau].sup.m], m [not
equal to] r on affiliates located in r is
(15) [(R-1).sup.-1][N.sup.-1.sub.rt] [summation over (m[not equal
to]r)] [summation over (i[member of][N.sub.rt])]
[DELTA][FDI.sup.m.sub.it].
For instance, Figure 12 illustrates the geographic pattern of the
total effect as in (13) on the average affiliate in a country. Notice
that we consider a shock in corporate profit tax rates in one country at
a time. Clearly, the direct effect of that shock is always (-1.16) x
(-0.01) so that the geographic pattern is entirely due to the
heterogeneity in all three dimensions of proximity and the respective
parameter weights on the channels in Specification I of Table 2. Figure
13 visualizes the total indirect effect from a shock in tax rates in the
country of affiliate i on affiliates of the same parent company that are
situated in other countries. The results in this figure are obtained as
described in (14), averaging the outcome across all other countries (and
firms as well as affiliates). Hence, this figure illustrates which host
countries tend to be sources of larger or smaller spillovers on FDI in
other countries. Finally, Figure 14 illustrates to which extent
country-specific (one-at-a-time) shocks on corporate profit taxes in
other countries spill over to affiliates in a given host country, on
average. Formally, the results in this figure are obtained as described
in (15). This figure illustrates which host countries tend to be
recipients of larger or smaller spillovers from other countries on FDI.
Due to the asymmetry in input-output tables and the heterogeneity in the
sector membership across affiliates within a parent company, the host
countries which tend to receive high average spillovers from shocks in
foreign taxes abroad are not necessarily also strong sources of
spillovers to affiliates in other countries (to see this, compare the
shading of Figure 14 with that of Figure 13).
The strongest overall positive impacts of an independent negative
shock to local profit tax rates in Figure 12 are found for China, the
United States, and Brazil. The smallest impacts are found for Cyprus,
Malta, and Luxembourg. Comparing the results of Figures 12 and 7 shows
that spillovers are positive and strong mainly to neighboring countries,
on average. However, an inspection of Table 2 suggests that the
researcher would be misguided to conclude that the source of this
pattern is mere geography, since we have seen that the actual channels
are downward closeness and upward distance in input-output space.
Conditioning on input-output relationships, mere geography has little to
contribute to the geographic pattern of spillovers. Figure 13 suggests
that shocks are particularly strongly (positively) transmitted by
affiliates in the United States, China, and Japan to foreign affiliates
and, to a somewhat lesser extent, by affiliates in Italy and Spain. On
the other extreme, a reduction in profit tax rates in adjacent countries
to Germany--such as Austria, Belgium, The Netherlands, and
Switzerland--tends to induce negative effects on the rest of the
affiliate network of German MNEs, on average. The main reason for this
finding is that affiliates in these countries tend to be upstream and
technologically closer to other affiliates downstream rather than to
ones further upstream. Also, they tend to be geographically close to
other affiliates in the network which, on average, leads to negative
spillovers from those countries. Figure 14 suggests that shocks on
corporate profit taxes in other countries which spill over to affiliates
in a given host country positively affect FDIs in several emerging and
South American economies. On the contrary, German MNEs' affiliates
in several European countries tend to receive non-positive spillovers
from a reduction of profit tax rates abroad.
Note that, when looking at Figures 7 and 10 alone, we could
conclude that FDI, from a German perspective, to overseas countries is
mainly related to horizontal FDI (because, e.g., we observe a lot of
German investments in the United States, but these investments are
primarily stand-alone ones). The other figures, however, suggest that
there are linkages between affiliates within an MNE's network
beyond pure geography (using the example from above, this implies that
there is intrafirm shipments to affiliates in the United States from
affiliates that are close in terms of industry closeness rather than
geographic closeness).
VII. CONCLUSIONS
Using a census-type panel dataset of German parent firms and their
affiliates from 1997 to 2009, this paper formulated a model to identify
several channels of spillovers within the German parent firms'
affiliates networks on the affiliate-specific FDIs. Allowing for three
channels of interdependence or spillovers--horizontal linkages (mere
geography), vertical input linkages, and vertical output linkages--we
find that horizontal linkages only matter to a limited extent, whereas
vertical linkages are the main source of spillovers. Moreover, we find
that spillovers from other affiliates are larger if an affiliate is
technologically situated downstream and strongly connected to affiliates
further upstream and if it is less strongly connected to affiliates
further downstream.
We use the regression results to quantify the magnitude of total
effects of shocks, of spillover effects from and to affiliates across
countries within the German multinational firm network. For illustrative
purposes, we use a reduction in corporate profit tax rates by
one-percentage point in one country at a time and calculate its
predicted effect on FDIs across affiliates. Overall, the findings are
illustrative of non-trivial effects of policy shocks on the investments
in a foreign affiliate network. Identical shocks on profit tax rates do
not only lead to quantitatively but even to qualitatively different
effects, depending on where they occur. The findings in this paper
suggest that primarily the technological proximity in input-output space
and less so the geography of an average MNE's affiliate network
matters for the geographic heterogeneity of spillover and total effects
of tax policy on FDI.
APPENDIX
TABLE A1 Countries in the Sample by Continent
Continent Country Code a
Africa Algeria DZA 0
Africa Botswana BWA 1
Africa Cameroon CMR 0
Africa Democ. Rep. of the Congo COD 1
Africa Egypt EGY 0
Africa Ethiopia ETH 1
Africa Gabon GAB 1
Africa Ghana GHA 1
Africa Ivory Coast CIV 0
Africa Kenya KEN 0
Africa Madagascar MDG 1
Africa Malawi MWI 1
Africa Mauritius MUS 0
Africa Morocco MAR 0
Africa Mozambique MOZ 1
Africa Namibia NAM 0
Africa Senegal SEN 0
Africa South Africa ZAF 0
Africa Swaziland SWZ 1
Africa Tunisia TUN 0
Africa Uganda UGA 0
Africa Un. Rep. of Tanzania TZA 0
Africa Zambia ZMB 1
Americas Argentina ARG 0
Americas Barbados BRB 1
Americas Bolivia BOL 0
Americas Brazil BRA 0
Americas Canada CAN 0
Americas Chile CHL 0
Americas Colombia COL 0
Americas Costa Rica CRI 0
Americas Dominican Rep. DOM 0
Americas Ecuador ECU 0
Americas El Salvador SLV 0
Americas Guatemala GTM 0
Americas Honduras HND 0
Americas Mexico MEX 0
Americas Nicaragua NIC 0
Americas Panama PAN 0
Americas Paraguay PRY 0
Americas Peru PER 0
Americas The Bahamas BHS 1
Americas Trinidad & Tobago TTO 1
Americas Un. States of America USA 0
Americas Uruguay URY 0
Americas Venezuela VEN 0
Asia Armenia ARM 1
Asia Azerbaijan AZE 0
Asia Bahrain BHR 1
Asia Bangladesh BGD 0
Asia China CHN 0
Asia Cyprus CYP 0
Asia Hong Kong S.A.R. HKG 0
Asia India IND 0
Asia Indonesia IDN 0
Asia Iran IRN 0
Asia Israel ISR 0
Asia Japan JPN 0
Asia Jordan JOR 0
Asia Kazakhstan KAZ 0
Asia Kuwait KWT 1
Asia Kyrgyzstan KGZ 1
Asia Lebanon LBN 0
Asia Malaysia MYS 0
Asia Oman OMN 1
Asia Pakistan PAK 0
Asia Philippines PHL 0
Asia Saudi Arabia SAU 0
Asia Singapore SGP 0
Asia South Korea KOR 0
Asia Sri Lanka LKA 0
Asia Thailand THA 0
Asia Turkey TUR 0
Asia Un. Arab Emirates ARE 0
Asia Uzbekistan UZB 1
Asia Vietnam VNM 0
Europe Austria AUT 0
Europe Belarus BLR 0
Europe Belgium BEL 0
Europe Bulgaria BGR 0
Europe Croatia HRV 0
Europe Czech Republic CZE 0
Europe Denmark DNK 0
Europe Estonia EST 0
Europe Finland FIN 0
Europe France FRA 0
Europe Greece GRC 0
Europe Hungary HUN 0
Europe Iceland ISL 1
Europe Ireland IRL 0
Europe Italy ITA 0
Europe Latvia LVA 0
Europe Lithuania LTU 0
Europe Luxembourg LUX 0
Europe Macedonia MKD 0
Europe Malta MLT 0
Europe Moldova MDA 0
Europe Netherlands NLD 0
Europe Norway NOR 0
Europe Poland POL 0
Europe Portugal PRT 0
Europe Romania ROU 0
Europe Russia RUS 0
Europe Slovakia SVK 0
Europe Slovenia SVN 0
Europe Spain ESP 0
Europe Sweden SWE 0
Europe Switzerland CHE 0
Europe Ukraine UKR 0
Europe United Kingdom GBR 0
Oceania Australia AUS 0
Oceania New Zealand NZL 0
Note: There is a total of 112 countries in the sample. The column
Code refers to ISO-3 codes
(a) Countries indicated by 1 do not appear in Figures 1-11 of the
paper, as they are based on less than three observations and
therefore not released for display according to the confidentiality
regulations of the Deutsche Bundesbank.
TABLE A2
Merged Industry Classifications (NACE/CPA)
No. Economic Sectors NACE CPA
(a) (Bundesbank)
1 Agriculture, hunting and related 100 1
service activities
2 Forestry, logging and related 200 2
service activities
3 Fishing, operation of fish 500 5
hatcheries and fish farms;
service activities incidental to
fishing
4 Mining of coal and lignite, 1,000 10
extraction of peat
5 Extraction of crude petroleum 1,100 11
and natural gas, service
activities incidental to oil and
gas extraction
6 Mining of uranium and thorium ores 1,200 12
7 Mining 1,300 13
8 Mining and quarrying, other 1,400 14
mining
9 Manufacturing of food and 1,500 15.1-
beverages 15.8
15.9
10 Manufacture of tobacco products 1,600 16
11 Manufacture of textiles 1,700 17
12 Manufacture of textile products 1,800 18
13 Manufacture of leather and leather 1900 19
products
14 Manufacture of wood and wood products 2000 20
15 Manufacture of pulp paper and paper 2,100 21.1
products
21.2
16 Publishing, printing and reproduction 2,200 22.1
of recorded media
22.2-
22.3
17 Manufacture of coke, refined petroleum 2,300 23
products and nuclear fuel
18 Manufacture of pharmaceutical products 2,440 24.4
19 Manufacture of chemicals and 2,400 24(ohne
chemical products 24.4)
20 Manufacture of rubber and plastic 2,500 25.1
products
25.2
21 Manufacture of non metallic mineral 2,600 26.1
products
26.2-
26.8
22 Manufacture of basic metals 2,700 27.1-
27.3
27.4
27.5
23 Manufacture of metal products 2,800 28
24 Manufacture of machinery and equipment 2,900 29
n.e.c.
25 Manufacture of office machinery and 3,000 30
computers
26 Manufacture of electrical machinery and 3,100 31
apparatus n.e.c.
27 Manufacture of radio, television and 3,200 32
communication equipment and apparatus
28 Manufacture of medical, precision and 3,300 33
optical instruments, watches and clocks
29 Manufacture of motor vehicles, trailers 3,400 34
and semi-trailers
30 Manufacture of other transport 3,500 35
equipment (only until 2004), from 2005
onwards 3,510,3,520, 3,530,
3,540, 3,550
31 Manufacure of furniture, manufacturing 3,600 36
n.e.c.
32 Recycling 3,700 37
33 Electricity, gas, steam and hot water 4,000 40.1,
supply 40.3
40.2
34 Collection, purification and 4,100 41
distribution of water
35 Construction sector 4,500 45.1-
45.2
45.3-
45.5
36 Sale, repair of motor vehicles; retail 5,000 50
sale of automotive fuel
37 Wholesale trade and commission trade 5,100 51
(except of motor vehicles and
motorcycles)
38 Retail trade, except of motor vehicles 5,200 52
and motorcycles; repair of personal
and household goods
39 Hotels and restaurants 5,500 55
60.1
40 Land transport; transport via 6,000 60.2-
pipelines 60.3
41 Water transport 6,100 61
42 Air transport 6,200 62
43 Supporting and auxiliary transport 6.300 63
activities; activities of travel
agencies
44 Post and telecommunications (only until 6,400 64
2004), from 2005 onwards 6,410, 6,420
45 Other credit institutions 6,560 65
46 Insurance and pension funding, except 6,600 66
compulsory social security.
47 Activities auxiliary to financial 6,700 67
intermediation
48 Housing enterprises, other real 7,050, 70
estate activities 7,060
49 Renting of machinery and equipment 7,100 71
without operator and of personal and
household goods
50 Computer and related activities 7,200 72
51 Research and development 7,300 73
52 Accounting, book-keeping and auditing 7,412 74
activities; tax consultancy (2005 on)
53 Federal government, Federal states, 7,560, 75.1-
local government and local authority 7,570, 75.2
associations 7,580
Social security and employment 7,590 75.3
promotion
54 Education 8,000 80
55 Health and social work, excluding 8,500 85
non-profit organizations serving
households
56 Sewage and refuse disposal, sanitation 9,000 90
and similar activities
57 Activities of other membership 9,100 91
organizations, excl. Non-profit
organizations serving households
58 Recreational, cultural and sporting 9,200 92
activities, excl. non-profit org.
serving households (only until 2004),
from 2005 onwards 9,210, 9,220, 9,230,
9,240, 9,250, 9,260, 9,270
59 Other service activities n.e.c., 9,300 93
excluding non-profit organizations
serving households
60 Private households with employed 9,550, 95
persons, Other households 9,560
No. Individual Goods
(a) (Statistisches
Bundesamt)
1 Erzeugnisse der Landwirtschaft
und Jagd
2 Forstwirtschaftliche Erzeugnisse
und DL
3 Fische und Fischereierzeugnisse
4 Kohle und Torf
5 Erdol, Erdgas, DL fur Erdol-,
Erdgasgewinnung
6 Uran- und Thoriumerze
7 Erze
8 Steine und Erden, sonstige
Bergbauerzeugnisse
9 Nahrungs- und Futtermittel
10 Getranke
11 Tabakerzeugnisse
12 Textilien
13 Bekleidung
14 Leder und Lederwaren
15 Holz; Holz-, Kork-, Flechtwaren (ohne Mobel)
16 Holzstoff, Zellstoff,
17 Papier, Karton und Pappe Papier-, Karton-
und Pappewaren
18 Verlagserzeugnisse
19 Druckerzeugnisse, bespielte Ton-, Bild- und
Datentrager
20 Kokereierzeugnisse, Mineralolerzeugnisse,
Spalt- und Brutstoffe
21 Pharmazeutische Erzeugnisse
22 Chemische Erzeugnisse (ohne
pharmazeutische Erzeugnisse)
23 Gummiwaren
24 Kunststoffwaren
25 Glas und Glaswaren
26 Keramik, bearbeitete Steine und Erden
27 Roheisen, Stahl, Rohre und Halbzeug daraus
28 NE-Metalle und Halbzeug daraus
29 Giessereierzeugnisse
30 Metallerzeugnisse
31 Maschinen
32 Buromaschinen, Datenverarbeitungsgerate und
-einrichtungen
33 Gerate der Elektrizitatserzeugung,
-verteilung u.a.
34 Nachrtechn., Rundf.- und Fernsehgerate,
elektron. Bauelemente
35 Medizin-, mess-, regelungstechn., optische
Erzeugnisse; Uhren
36 Kraftwagen und Kraftwagenteile
37 Sonstige Fahrzeuge (Wasser-, Schienen-,
Luftfahrzeuge u.a.)
38 Mobel, Schmuck, Musikinstrumente,
Sportgerate, Spielwaren u.a.
39 Sekundarrohstoffe
40 Elektrizitat, Fernarme, DL der
Elektrizitats- u.
41 Fernwarmeversorgung
Gase, DL der Gasversorgung
42 Wasser und DL der Wasserversorgung
43 Vorb. Baustellenarbeiten, Hoch- und
Tiefbauarbeiten
44 Bauinstallations- und sonstige Bauarbeiten
45 Handelsleist. mit Kfz; Rep. an Kfz;
Tankleistungen
46 Handelsvermittlungs- und
Groahandelsleistungen
47 Einzelhandelsleistungen; Reparatur an
Gebrauchsgiitern
48 Beherbergungs- und Gaststatten-DL
49 Eisenbahn-DL
50 Sonst. Landv.leistungen,
Transportleistungen in
Rohrfernleitungen
51 Schifffahrtsleistungen
52 Luftfahrtleistungen
53 DL beztiglich Hilfs- und Nebentiitigkeiten
fur den Verkehr
54 Nachrichtenubermittlungs-DL
55 DL der Kreditinstitute
56 DL der Versicherungen (ohne Sozialversicherung)
57 DL des Kredit- und Versicherungshilfsgewerbes
58 DL des Grundstiicks- und Wohnungswesens
59 DL der Vermietung beweglicher Sachen
(ohne Personal)
60 DL der Datenverarbeitung und von
Datenbanken
61 Forschungs- und Entwicklungsleistungen
62 Unternehmensbezogene DL
63 DL der offentlichen Verwaltung, Verteidigung
64 DL der Sozialversicherung
65 Erziehungs- und Unterrichts-DL
66 DL des Gesundheits-, Veterinar- und
Sozialwesens
67 Abwasser-, Abfallbeseitigungs- u. sonst.
Entsorgungsleistungen
68 DL von Interessenvertretungen, Kirchen u.?.
69 Kultur-, Sport- und Unterhaltungs-DL
70 Sonstige DL
71 DL privater Haushalte
(a) Consecutive sector number introduced for the present analysis.
(b) Consecutive sector number used in the input-output tables
(Statistisches Bundesamt).
ABBREVIATIONS
FDI: Foreign Direct Investment
FE2SLS: Fixed-Effects Two-Stage Least Squares
GDP: Gross Domestic Product
MiDi: Microdatabase Direct Investment
MNE: Multinational Enterprise
SAR: Spatial Autoregressive
doi: 10.1111/ecin. 12517
Online Early publication November 2, 2017
Bosenberg: Researcher, Swiss National Bank, Zurich 8022,
Switzerland. Phone 0041-58631-2168, Fax 0041-586315000, E-mail
Simon.Boesenberg@snb.ch
Egger: Professor, ETH Zurich, Economics, Zurich 8092, Switzerland.
Phone 0041-44632-4108, Fax 0041-446324108, E-mail egger@kof.ethz.ch
Merlo: Professor, ETH Zurich, Economics. Zurich 8092, Switzerland.
Phone 0049-7071 -2975421, Fax 0049-7071 295590, E-mail
valeria.merlo@uni-tuebingen.de
Wamser: Professor, School of Business and Economics, University of
Tubingen, Tubingen 72074, Germany. Phone 0049-178-4303811, Fax
0049-7071295590, E-mail georg.wamser@uni-tuebingen.de
REFERENCES
Alfaro, L., P. Antras, D. Chor, and P. Conconi. "Internalizing
Global Value Chains: A Firm-Level Analysis." NBER Working Paper No.
21582, 2015.
Antras, P., and D. Chor. "Organizing the Global Value
Chain." Econometrica, 81(6), 2013, 2127-204.
Atalay, E., A. Hortagsu, and C. Syverson. "Vertical
Integration and Input Flows." American Economic Review, 2014,
104(4), 2014, 1120-48.
Badinger, H., and P. H. Egger. "Fixed Effects and Random
Effects Estimation of Higher-Order Spatial Autoregressive Models with
Spatial Autoregressive and Heteroskedastic Disturbances." Spatial
Economic Analysis, 10(1), 2015, 11-36.
Baldwin, R. E. "Managing the Noodle Bowl: The Fragility of
East Asian Regionalism." Working Paper Series on Regional Economic
Integration No. 7, Asian Development Bank, 2007.
Baltagi, H. B., P. H. Egger, and M. Pfaffermayr. "Estimating
Regional Trade Agreement Effects on FDI in an Interdependent
World." Journal of Econometrics, 145(1-2), 2008, 195-208.
--. "Estimating Models of Complex FDI: Are There Third-Country
Effects?" Journal of Econometrics, 140(1), 2007, 260-81.
Bergstrand, J. H., and P. H. Egger. "A
Knowledge-and-Physical-Capital Model of International Trade Flows,
Foreign Direct Investment, and Multinational Enterprises." Journal
of International Economics, 73(2), 2007, 278-308.
Bernard, A. B., J. B. Jensen, S. J. Redding, and P. K. Schott.
"Intrafirm Trade and Product Contractibility." American
Economic Review, 100(2), 2010,444-48.
Bernstein, J. I., and P. Mohnen. "International R&D
Spillovers between U.S. and Japanese R&D Intensive Sectors."
Journal of International Economics, 44(2), 1997,315-38.
Blonigen, B. A., R. B. Davies, H. T. Naughton, and G. R. Waddel.
"Spacey Parents: Autoregressive Patterns in Inbound FDI," in
Foreign Direct Investment and the Multinational Enterprise, edited by S.
Brakman and H. Garretsen. Cambridge, MA: MIT Press, 2008, 173-97.
Bode, E., P. Nunnenkamp, and A. Waldkirch. "Spatial Effects of
Foreign Direct Investment in US States." Canadian Journal of
Economics, 45(1), 2012, 16-40.
Burstein, A., C. Kurz, and L. Tesar. "Trade, Production
Sharing, and the International Transmission of Business Cycles."
Journal of Monetary Economics, 55(4), 2008, 775-95.
Carr, D. L., J. R. Markusen, and K. E. Maskus. "Estimating the
Knowledge-Capital Model of the Multinational Enterprise." American
Economic Review, 91, 2001, 693-708.
Chen, M. X. "Interdependence in Multinational Production
Networks." Canadian Journal of Economics, 44(3), 2011,930-56.
Costinot, A., J. Vogel, and S. Wang. "An Elementary Theory of
Global Supply Chains." Review of Economic Studies, 80(1), 2013,
109-44.
De Mooij, R. A., and S. Ederveen. "What a Difference Does It
Make? Understanding the Empirical Literature on Taxation and
International Capital Flows." European Commission Economic Paper
No. 261, 2006.
Egger, P. H., M. Fahn, V. Merlo, and G. Wamser. "On the
Genesis of Multinational Foreign Affiliate Networks." European
Economic Review, 65, 2013, 136-63.
Egger, P. H., C. Keuschnigg, V. Merlo, and G. Wamser.
"Corporate Taxes and Internal Borrowing within Multinational
Firms." American Economic Journal: Economic Policy, 6(2), 2014,
54-93.
Ekholm, K., R. Forslid. and J. R. Markusen. "Export-Platform
Foreign Direct Investment." Journal of the European Economic
Association, 5(4), 2007, 776-95.
Hall, R. E., and C. I. Jones. "Why Do Some Countries Produce
So Much More Output Per Worker Than Others?" Quarterly Journal of
Economics, 114(1), 1999,83-116.
Kelejian, H. H. "Omitted Factors and Spatial Lags in the
Dependent Variable." Letters in Spatial and Resource Sciences,
7(1), 2013,23-33.
Kelejian, H. H., I. R. Prucha, and Y. Yuzefovich "Instrumental
Variable Estimation of a Spatial Autoregressive Model with
Autoregressive Disturbances: Large and Small Sample Results," in
Advances in Econometrics, edited by J. LeSage and R. K. Pace. New York:
Elsevier, 2004, 163-98.
Kelejian, H. H., and I. R. Prucha. "Specification and
Estimation of Spatial Autoregressive Models with Autoregressive and
Heteroskedastic Disturbances." Journal of Econometrics, 157(1),
2010, 53-67.
Keller, W. "Trade and Transmission of Technology."
Journal of Economic Growth, 7(1), 2002, 5-24.
Keller, W., and S. R. Yeaple. "The Gravity of Knowledge."
American Economic Review, 103(4), 2013, 1414-44.
Kleinert, J., J. Martin, and F. Toubal. "The Few Leading the
Many: Foreign Affiliates and Business Cycle Comovements." CEPR
Discussion Paper No. 9129, 2012.
Kuhn, A. "Input-Output-Rechnung im Uberblick, Statistisches
Bundesamt, Fachserie 18." Reihe, 2, 2010, 1-46.
Learner, E. E. Sources of International Comparative Advantage:
Theory and Evidence. Cambridge, MA: MIT Press, 1984."
Lipponer, A. "Microdatabase Direct Investment--MiDi. A Brief
Guide." Bundesbank Technical Documentation, Frankfurt, 2011.
Markusen, J. R. Multinational Firms and the Theory of International
Trade. Cambridge, MA: MIT Press, 2002.
Markusen, J. R., and K. E. Maskus. "Discriminating among
Alternative Theories of the Multinational Enterprise." Review of
International Economics, 10(4), 2002, 694-707.
Mayer, T., and S. Zignago. "Notes on CEPII's Distances
Measures (GeoDist)." CEPII Working Paper 2011-25, 2011.
McGrattan, E. R., and E. C. Prescott. "Openness, Technology
Capital, and Development." Journal of Economic Theory, 144(6),
2009, 2454-76.
Morrison Paul, C. J., and D. Siegel. "Scale Economies and
Industry Agglomeration Externalities: A Dynamic Cost Function
Approach." American Economic Review, 89(1), 1999, 272-90.
Oberhofer, H., and M. Pfaffermayr. "Firm Growth in
Multinational Corporate Groups." Empirical Economics, 44(3), 2013,
1435-53.
Shin, N., K. Kraemer, and J. Dedrick. "Value Capture in the
Global Electronics Industry: Empirical Evidence for the 'Smiling
Curve' Concept." Industry and Innovation, 19(2), 2012, 89-107.
Smarzynska Javorcik, B. "Does Foreign Direct Investment
Increase the Productivity of Domestic Firms? In Search of Spillovers
through Backward Linkages." American Economic Review, 94(3), 2004,
605-27.
Statistisches Bundesamt. "Volkswirtschaftliche
Gesamtrechnungen, Input-Output-Rechnung." 2014. Fachserie 18 Reihe
2, Wiesbaden. Accessed September 23, 2014.
https://www.destatis.de/DE/Publikationen/Thematisch/VolkswirtschaftlicheGesamtrechnungen/InputOutputRechnung/VGRInputOut
putRechnung2180200107004.pdf?
Tintelnot, F. "Global Production with Export Platforms."
Quarterly Journal of Economics, 132(1), 2017, 157-209.
SIMON BOSENBERG, PETER H. EGGER, VALERIA MERLO and GEORG WAMSER*
* We are grateful to Deutsche Bundesbank for granting access to the
MiDi database. The views expressed in this paper are those of the
authors and do not reflect those of the Swiss National Bank. We thank
Mohammed Mardan as well as seminar participants at the NHH in Bergen,
the University of Munich, and the University of St. Gallen for helpful
comments
(1.) Notice that, unlike with trade, only few countries provide
exhaustive datasets on their FDI. With interdependent observations,
omitting some countries or country-pairs from the data leads to
measurement error of the regressors and an associated inconsistency of
the parameters determining aggregate country-level or country-pair-level
FDI.
(2.) Again, much more so than with firm-level exports, only few
countries provide exhaustive datasets on their MNEs and the associated
affiliates and FDIs. With interdependent observations, omitting some
affiliates from the data leads to measurement error of the regressors
and an associated inconsistency of the parameters determining investment
at the level of affiliates and firms.
(3.) We use the foreign affiliate's stock of fixed and
intangible assets attributable to the parent company as a measure of FDI
related to production activities.
(4.) While the sequencing of foreign investments determines the
direction of learning, this is out of the scope of the present study, as
it focuses on the interdependence of investments at the intensive
affiliate margin, that is, at a given size and location of the affiliate
network, at each point in time.
(5.) On a broader scale, the focus on horizontal versus vertical
interdependencies of the present paper relates it to the literature on
affiliate networks (see Egger et al. 2013; Oberhofer and Pfaffermayr
2003) and to the recent literature on the value chains and the
organization of firms (e.g., Antras and Chor 2013).
(6.) Unlike row-normalized weights, the suggested ones bear the
advantage of preserving the notion of absolute proximity in the lth
metric.
(7.) The great circle distance is calculated using the haversine
formula. The internal distance measure and the coordinates of the main
cities are taken from CEPII's GeoDist database (see Mayer and
Zignago 2011, for a description).
(8.) It appears plausible to assume that the covered MNEs use
German technology standards in their affiliate network. Using a
best-practice technology throughout the affiliate network entails one of
the major advantages of MNEs relative to stand-alone firms.
(9.) Moreover, [w.sup.s.sub.ijt] and [w.sup.D.sub.ijt] vary only
across ijt.
(10.) Note that our estimation approach will also allow for
unobserved affiliate heterogeneity (see below).
(11.) Badinger and Egger (2015) and Kelejian (2013) show that
FE2SLS provides consistent estimates of [[lambda].sup.l] and [beta]
under a broad spectrum of assumptions. As shown above, the estimator we
employ suggests using the weighted exogenous variables to instrument for
the endogenous spatial lag. There might be a concern about the exclusion
restrictions in the context of weighted variables measured at the level
of the firm. We will address this by using as instruments lagged
weighted characteristics of other affiliates in the same country and
year that are not related to i and belong to another multinational firm.
We are not concerned that these affiliates may be affected by the same
(exogenous) shocks as i. What is important is that neither affiliate i
nor the firm controlling i may directly affect the characteristics of
other firms in period t - 1. In particular, the latter should be the
case as outcome of affiliate i is measured in period t, and the
instruments are measured in period t - 1.
(12.) The collection of annual statistics is stipulated by law
through the AuBenwirtschaftsgesetz (AWG) (Foreign Trade and Payments
Act). The reporting requirements refer to Sections 56a and 58a of the
AWG. They were enacted in 2002, but are applied consistently for all
years of the panel. For a detailed description of the MiDi database, see
2011.
(13.) See Mayer and Zignago (2011) for a description.
(14.) See Statistisches Bundesamt (2014), p. 4.
(15.) The tables include domestic and imported intermediates used
in production. Intermediates are valued at prices which exclude any
taxes but include subsidies (see Kuhn 2010, p. 15).
(16.) Following Hall and Jones (1999) the capital stock at time t
is generically defined as [K.sub.t] = (1 - [delta])[K.sub.t-1] +
[GFC.sub.t]. Here, [GFC.sub.t] is the gross fixed capital formation at
constant U.S. dollars of 2000 as reported in the World Bank's World
Development Indices Database, and [delta] is the rate of depreciation,
set at 0.133 (see, e.g., Learner 1984). Furthermore, we calculate the
initial capital stock by [K.sub.0] = [GFC.sub.0]/[delta]+[g.sub.K],
where [g.sub.GFK] is the rate of growth of the capital stock, being set
at 0.025, as in Bergstrand and Egger (2007).
(17.) On average an affiliate is about 6.2 years in the sample.
(18.) Figures 1-11 include only 92 of the 112 countries in the
sample due to the confidentiality regulations of the Deutsche
Bundesbank. This results in the deletion of 20 countries with less then
3 affiliates in an average year per country from our graphical
representation. Nevertheless, all affiliates and 112 host countries are
included in the estimation below.
(19.) In an earlier version of the paper we have experimented more
with the set of instruments and have shown that results are robust to
alternative specifications and combinations of instruments. Notice that
one could increase the instrument set by using higher-order powers of
the linkage weights. However, the instrument quality deteriorates with
the order of linkage weights and all that is needed here are only at
least as many instruments as there are linkages in the model.
(20.) We are not concerned that these affiliates may be affected by
the same (exogenous) shocks as i. What is important is that neither
affiliate i nor the firm controlling i may directly affect the
characteristics of other firms in period t - 1. In particular, the
latter should be the case as outcome of affiliate i is measured in
period t, and the instruments are measured in period t - 1.
(21.) It is less straightforward to interpret our findings in the
light of the learning model as suggested in Egger et al. (2013) to
explain international expansion patterns of affiliate networks. On the
one hand, Egger et al. (2013) model the extensive instead of, as in this
paper, the intensive investment margin. On the other hand, Egger et al.
(2013) distinguish between sequential and simultaneous investments and
find qualitative differences in their determinants. Since we focus on
investments within given affiliates it is not possible to explicitly
distinguish between the latter modes. However, as our results suggest
interdependencies predominantly through vertical production linkages,
the results suggest that proximity on a general level (here, mainly in
terms of vertical linkages) matters for marginal investment decisions or
for discrete investment project decisions even in a given affiliate
network.
(22.) Note that we cannot report more details of the sectorlevel
regressions, since the Deutsche Bundesbank has specific requirements
regarding the number of observations per firm, per market, and per
sector.
Caption: FIGURE 1 Average Number of Affiliates for the Average Year
Per Country
Caption: FIGURE 2 Average (Log) Fixed Assets Per Affiliate for the
Average Year Per Country
Caption: FIGURE 3 Average Number of Parent-Firms for the Average
Year Per Country
Caption: FIGURE 4 Average (Log) Fixed Assets Per Parent-Firms for
the Average Year Per Country
Caption: FIGURE 5 Average Input Proximity Across All Affiliates for
the Average Firm Per Country
Caption: FIGURE 6 Average Output Proximity Across All Affiliates
for the Average Firm Per Country
Caption: FIGURE 7 Average Geographic Proximity Across All
Affiliates for the Average Firm Per Country
Caption: FIGURE 8 Average Input-Weighted (Log) Fixed Assets
Proximity Across All Affiliates for the Average Firm Per Country
Caption: FIGURE 9 Average Output-Weighted (Log) Fixed Assets
Proximity Across All Affiliates for the Average Firm Per Country
Caption: FIGURE 10 Average Geography-Weighted (Log) Fixed Assets
Proximity Across All Affiliates for the Average Firm Per Country
Caption: FIGURE 11 Density Plot
Caption: FIGURE 12 Total (Direct Plus Indirect) Effect of a
One-Percentage-Point Tax Reduction Per Country on the Average Affiliate
There
Caption: FIGURE 13 Total (Indirect) Effect of a
One-Percentage-Point Tax Reduction Per Country on Other Affiliates
Outside of the Country Reducing the Tax Rate
Caption: FIGURE 14 Total (Indirect) Effect of a
One-Percentage-Point Tax Reduction from Other Countries (One at Time) on
the Average Affiliate Per Country
TABLE 1 Summary Statistics
Main Variables Mean SD
[FDI.sub.it] 7.725 2.051
[Sales.sub.it--1] 3.109 1.353
[Employees.sub.it-1] 4.475 1.417
Corporate Income [Tax.sub.it] 0.314 0.072
Financial [Freedom.sub.it] 67.703 18.123
[Inflation.sub.it] 3.343 6.436
Capital - [Labor Ratio.sub.it] 10.553 0.974
[GDP.sub.it] 27.417 1.375
[Competition.sub.fit-1] 110.72 156.21
Interdependence Terms Mean SD
[[bar.FDI].sup.I.sub.it] 0.390 1.134
[[bar.FDI].sup.O.sub.it] 0.401 1.134
[[bar.FDI].sup.D.sub.it] 0.241 0.483
[[bar.FDI].sup.Iint.sub.it] 0.348 1.071
[[bar.FDI].sup.Iext.sub.it] 0.041 0.412
[[bar.FDI].sup.Oinr.sub.it] 0.358 1.070
[[bar.FDI].sup.Oext.sub.it] 0.042 0.414
[[bar.FDI].sup.Dint.sub.it] 0.210 0.454
[[bar.FDI].sup.Dext.sub.it] 0.030 0.201
Instruments Mean SD
[[bar.Sales].sup.I.sub.it-1] 0.435 1.149
[[bar.Employees].sup.I.sub.it-1] 0.497 1.330
[[bar.Corporate Income Tax].sup.I.sub.it] 0.014 0.038
[[bar.Competetion].sup.I.sub.it-1] 2.676 5.659
[[bar.Sales].sup.O.sub.it-1] 0.443 1.141
[[bar.Employees].sup.O.sub.it-1] 0.508 1.322
[[bar.Corporate Income Tax].sup.O.sub.it] 0.014 0.038
[[bar.Competetion].sup.O.sub.it-1] 2.536 5.335
[[bar.Sales].sup.IO.sub.it-1] 0.439 1.144
[[bar.Employees].sup.IO.sub.it-1] 0.503 1.325
[[bar.Corporate Income Tax].sup.IO.sub.it] 0.014 0.038
[[bar.Competetion].sup.IO.sub.it-1] 2.607 5.461
[[bar.Sales].sup.D.sub.it-1] 0.311 0.653
[[bar.Employees].sup.D.sub.it-1] 0.346 0.718
[[bar.Corporate Income Tax].sup.D.sub.it] 0.009 0.017
[[bar.Competetion].sup.D.sub.it-1] 2.65 5.479
N = 139,696
Note: Minimum and maximum values are not displayed
due to the confidentiality rules of the Deutsche Bundesbank.
TABLE 2
Analysis of German MNEs' Foreign Investments
Basic Basic Inv. Dist.
Spec. I Spec. II [Decay.sup.2]
[[bar.FDI].sup.I.sub.it] 0.195 *** 0.198 ***
(0.068) (0.068)
[[bar.FDI].sup.Int.sub.it]
[[bar.FDI].sup.Iext.sub.it]
[[bar.FDI].sup.O.sub.it] -0.161 ** -0.167 **
(0.068) (0.068)
[[bar.FDI].sup.Oint.sub.it]
[[bar.FDI].sup.Oext.sub.it]
[[bar.FDI].sup.D.sub.it] -0.038 * -0.042 ** -0.057
(0.020) (0.019) (0.049)
[[bar.FDI].sup.IO.sub.it] 0.033 ***
(0.008)
[[bar.FDI].sup.Dint.sub.it]
[[bar.FDI].sup.Dext.sub.it]
[Sales.sub.it-1] 0.180 *** 0.180 *** 0.180 ***
(0.009) (0.009) (0.009)
[Employees.sub.it-1] 0.335 *** 0.336 *** 0.335 ***
(0.012) (0.012) (0.012)
Corp. Inc. -1.160 *** -1.164 *** -1.152 ***
[Tax.sub.it] (0.155) (0.155) (0.155)
Financial 0.002 *** 0.002 *** 0.002 ***
[Freed..sub.it]] (0.001) (0.001) (0.001)
[Inflation.sub.it] -0.005 ** -0.005 ** -0.005 **
(0.002) (0.002) (0.002)
Cap. - Lab. 0.181 *** 0.177 *** 0.183 ***
[Ratio.sub.it] (0.067) (0.067) (0.067)
[GDP.sub.it] 0.119 * 0.120 * 0.120 *
(0.066) (0.066) (0.066)
[Competition.sub.it-1] -0.0004 *** -0.0004 *** -0.0004 ***
(0.000) (0.000) (0.000)
Year dummies yes yes yes
[R.sup.2] 0.103 0.103 0.103
N 134,702 134,702 134,702
Inv. Dist. No Int.
[square root Dist.
of (Decay)]
[[bar.FDI].sup.I.sub.it] 0.195 *** 0.188 ***
(0.068) (0.068)
[[bar.FDI].sup.Int.sub.it]
[[bar.FDI].sup.Iext.sub.it]
[[bar.FDI].sup.O.sub.it] -0.162 ** -0.154 **
(0.068) (0.068)
[[bar.FDI].sup.Oint.sub.it]
[[bar.FDI].sup.Oext.sub.it]
[[bar.FDI].sup.D.sub.it] -0.013 -0.035 **
(0.014) (0.014)
[[bar.FDI].sup.IO.sub.it]
[[bar.FDI].sup.Dint.sub.it]
[[bar.FDI].sup.Dext.sub.it]
[Sales.sub.it-1] 0.180 *** 0.180 ***
(0.009) (0.009)
[Employees.sub.it-1] 0.335 *** 0.335 ***
(0.012) (0.012)
Corp. Inc. -1.153 *** -1.153 ***
[Tax.sub.it] (0.155) (0.155)
Financial 0.002 *** 0.002 ***
[Freed..sub.it]] (0.001) (0.001)
[Inflation.sub.it] -0.005 ** -0.005 **
(0.002) (0.002)
Cap. - Lab. 0.181 *** 0.186 ***
[Ratio.sub.it] (0.067) (0.067)
[GDP.sub.it] 0.120 * 0.130 **
(0.066) (0.066)
[Competition.sub.it-1] -0.0004 *** 0.0004 ***
(0.000) (0.000)
Year dummies yes yes
[R.sup.2] 0.103 0.103
N 134,702 134,702
Country- Ext. & Int.
Time Eff. Margin
[[bar.FDI].sup.I.sub.it] 0.557 ***
(0.073)
[[bar.FDI].sup.Int.sub.it] 0.105
(0.082)
[[bar.FDI].sup.Iext.sub.it] 2.564 ***
(0.882)
[[bar.FDI].sup.O.sub.it] -0.382 **
(0.072)
[[bar.FDI].sup.Oint.sub.it] -0.072
(0.082)
[[bar.FDI].sup.Oext.sub.it] -2.574 ***
(0.885)
[[bar.FDI].sup.D.sub.it] -0.089 ***
(0.020)
[[bar.FDI].sup.IO.sub.it]
[[bar.FDI].sup.Dint.sub.it] -0.041 **
(0.020)
[[bar.FDI].sup.Dext.sub.it] -0.116
(0.109)
[Sales.sub.it-1] 0.244 *** 0.174 ***
(0.007) (0.010)
[Employees.sub.it-1] 0.680 *** 0.337 ***
(0.009) (0.012)
Corp. Inc. -1.148 ***
[Tax.sub.it] (0.155)
Financial 0.002 ***
[Freed..sub.it]] (0.001)
[Inflation.sub.it] -0.005 **
(0.002)
Cap. - Lab. 0.179 ***
[Ratio.sub.it] (0.067)
[GDP.sub.it] 0.117 *
(.067)
[Competition.sub.it-1] -0.003 *** 0.000 ***
(0.000) (0.000)
Year dummies no yes
[R.sup.2] 0.458 0.093
N 139,561 134,702
Notes: FE2SLS estimations (see Section 3.1); t-statistics in
parentheses. Our estimation sample (unbalanced panel) includes
21,598 foreign affiliates of German MNEs in 112 different countries
over the time period 1997 to 2009. The dependent variable is
[FDI.sub.fit], reported in logs of million Euros.
* (p<0.05), ** (p<0.01), *** (p<0.001).
COPYRIGHT 2018 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2018 Gale, Cengage Learning. All rights reserved.
