Location choices of multinational firms in Europe: the role of national
boundaries and EU policy
Roberto Basile
ISAE, Rome and University of Macerata
§
Davide Castellani
University of Urbino “Carlo Bo”
Antonello Zanfei
University of Urbino “Carlo Bo”
12/08/2005
Abstract
We examine the determinants of multinational firms’ location choices in Europe by
estimating discrete choice models on a data-set of 5,509 foreign subsidiaries established in 50
regions in 8 EU countries over the period 1991-1999. We find that firms consider regions
across different countries as closer substitutes than regions within national borders. This
reveals that European regions compete to attract FDIs more across than within countries. We
also find that EU regional policies (Cohesion and Structural Funds) play a significant role in
attracting multinationals, while national tax competition attracts FDI only when
agglomeration economies are not controlled for. Differences emerge in determinants of EU
and US multinationals’ location choices.
JEL Classification: F23, O52, R30
Key words: Europe; Foreign Direct Investments; Location; Discrete Choice Models.
§
Corresponding author: Istituto di Scienze Economiche, Università di Urbino, Via Saffi, 2, 61029 Urbino, Italy.
Tel. +39 (0)722 305562, fax. +39 (0)722 305550, e-mail: [email protected]
1
1. Introduction
Accelerating economic integration in Europe over the past decade has favored, inter alia, a
significant flow of international investments from both within and outside the European Union (EU)
borders. As a matter of fact, the EU has attracted over 40% of total world flows of foreign direct
investments (FDIs) in the 1990’s, becoming the largest recipient of multinational activity.
Multinationals account for a growing share of gross fixed capital formation in Europe (from 6% in
1990 to over 50% in 2000) and about one quarter of large firm R&D carried out in Europe has been
conducted under foreign ownership, while the world average is just over one tenth.
However, this increasing inflow of FDI in Europe has not been equally distributed across countries
and regions. From this perspective, this paper analyses patterns and determinants of the location of
multinational activities in Europe over the nineties. In particular, the aim of the paper is twofold.
First, we assess whether national boundaries affect location choices of multinational firms. In fact,
European integration, with the dismantling of trade barriers, free movement of people, goods and
capital and the strong reduction of state aids, has contributed in making country boundaries more
blurred1. Within an integrated economic space, such as the European Union, one could expect a
variety of possible outcomes, ranging from persisting national patterns of localization of foreign
activities, to the emergence of sub-continental regions competing with each other across and within
states for attracting foreign economic activities. A few studies have addressed the issue of national
boundaries in localization decisions, but results are largely inconclusive as they have focused on
specific categories of investors, such as Japanese or French multinationals (Head and Mayer 2004,
Disdier and Mayer 2004, Mucchielli and Puech 2002). In this study, we adopt a wider focus and tackle
the national boundary issue with no restriction on the country of origin of investors.
Second, we focus on the role of EU policies (Structural and Cohesion Funds) as tools to attract
foreign investors in backward regions. As noted in many recent theoretical and empirical works, in the
presence of increasing returns and local externalities, a greater economic integration leads to the
spatial concentration of productive activities (Fujiita, Krugman and Venables, 1999). The uneven
spatial impact of economic integration motivates EU public support in favor of backward regions.
1
This is consistent with the argument put forth by Krugman (1991, p.8): “as Europe becomes a unified market, with free
movement of capital and labour, it will make less and less sense to think of the relations between its component nations in
terms of the standard paradigm of international trade. Instead the issues will be those of regional economics”.
2
Structural and Cohesion Funds aim to contrast this trend towards productive localization in core
regions by creating favorable environmental conditions in the peripheral areas “through investment to
strengthen the economic base in recipient regions” (European Commission, 1996). Using aggregate
data on regional gross value added, Midelfart-Knarvik and Overman (2002) show that European
Structural Funds influenced the location of industry in Europe, thus mitigating the economic forces at
work. Here, using micro data on multinational firms, we analyze whether and to what extent EU
policies have affected the localisation of multinational activities within the continent2.
The analysis makes use of the Elios dataset (European Linkages and Ownership Structure), built at
the University of Urbino and based on Dun & Bradstreet’s Who Owns Whom, which provides
information on location choices of 5,509 affiliates of multinational firms between 1991 and 1999 over
a set of 50 NUTS-1 regions in 8 EU countries (France, Germany, Ireland, Italy, Spain, Portugal,
Sweden and United Kingdom). Parent companies are of different nationalities: the single largest home
country are the US (25%), but the majority are from EU countries (60%). Additional data on regional
and country characteristics are mainly drawn from Eurostat’s Regio and Cambridge Econometrics.
We first test whether national boundaries affect location decisions, by estimating, in a nested logit
framework, to what extent multinational firms consider regions belonging to the same country as
closer substitutes than locations across borders. Second, a mixed logit model is used to identify more
complex substitution patterns between regions and to evaluate the effect of a number of location
determinants. These include market size, agglomeration economies, experience of a multinational firm
on each region, local labour market characteristics and policy measures both at the EU level (namely
Structural and Cohesion Funds) and at the national level (such as corporate tax rates and public
infrastructures).
This paper improves on the existing empirical literature from at least three points of view. First,
this we extend the geographic span of host economies, covering a larger number of EU recipient
countries than most previous contributions. Second, we are able to investigate how the nationality of
the parent firm will determine a different sensitiveness to some location characteristics. Finally, we
estimate, for the first time, a mixed logit model on firms’ location choices, which allows to capture
2
It is worth pointing out that, by assessing the impact of regional policies on the location of foreign investment, we are not
trying to assess whether the geographic distribution of multinational activity eventually contributed to Europe’s economic
growth or regional cohesion.
3
more complex substitution patterns between choices, than standard conditional or nested logit models
used in this context so far.
The main results of our analysis are: (i) country borders do not matter, with the exception of Italy:
foreign investors do not consider regions within countries as closer substitutes than regions belonging
to different countries, suggesting that competition to attract FDI spans across national boundaries; (ii)
EU policy contributed to mitigating agglomeration forces and attracted considerable investments in
peripheral regions; (iii) national tax competition affects location decisions only when agglomeration
economies are not controlled for; (iv) differences emerges in location determinants of EU and US
investments.
The paper is organised as follows. Section 2 describes the dataset and illustrates the regional
distribution of new foreign establishments in Europe over the nineties. Section 3 briefly reviews the
literature concerning the location determinants of foreign firms. Section 4 illustrates the discrete
choice models used for estimation. Section 5 presents the variables introduced in the econometric
model and the empirical findings. Section 6 concludes the paper.
2. Regional distribution of FDI in Europe
Our analysis exploits a novel dataset, built at the University of Urbino, which collects information
from Dun & Bradstreet’s Who Owns Whom on a large sample of firms active in Europe. In particular,
we have data on firms active in 8 countries (France, Germany, Ireland, Italy, Portugal, Spain, Sweden
and the United Kingdom), which inter alia account for over 60% of total inward FDI flows in the EU.
For each firm we have information on name and country of the ultimate owner, sector of activity (2digit SIC), location, year of establishment. Exploiting the information on the country of the ultimate
owner we identified foreign-owned firms and we restricted our analysis to those which were
established over the 1991 to 1999 period. We ended up with a sample of 5,509 foreign-owned firms
locating in the 8 countries considered over 1991-1999. Consistently with Eurostat’s Foreign Direct
Investment Statistics (Eurostat 2002), which reports that 72% of total inward FDIs over the nineties
have been Intra-EU flows, 3,395 (out of 5,509) sample firms are subsidiaries of EU MNEs. Further
supporting the idea that our large sample is a good representation of inward FDIs in the EU, the
4
percentage distribution of foreign-owned firms in our sample across countries is remarkably similar to
the actual distribution of cumulated FDI flows over the same period as registered by Unctad (2002).
Our analysis of the location choice of foreign-owned firms in Europe exploits the information on
the region where each firm in our sample has established itself3. In many cases such pieces of
information were available at a rather fine level of disaggregation (such as NUTS-3 or even cities), but
we had to confine our focus on NUTS-1 regions4, since in some cases (such as for German firms) this
was the only available piece of information and also because this allows to keep computational
complexity tractable in the subsequent econometric analysis.
The distribution of foreign investments (as proxied by the number of foreign-owned firms
established over 1991-1999) in EU regions, reported in Figure 1, suggests that the largest regions of
Germany, France, together with Lombardia in Italy, Cataluna in Spain and South East in the UK
attract more FDIs5. To a closer look, one might notice that the case of Italy is characterized by very
low numbers of newly established subsidiaries in any region but Lombardia, while regional foreign
presence is generally more patchy in other EU countries. One may venture saying that in the case of
Italy a country effect is at play, decreasing the attractiveness of (almost) all regions within the national
boundaries. Finally, interesting differences emerge in the location of EU MNEs relative to firms
whose parents are from countries outside the EU (of which more than 50% are US MNEs). In
particular, Figure 2 suggests that EU multinationals have a higher propensity to establish new
subsidiaries in Southern Europe and in some French regions; while non-EU multinationals tend to
place their activities in Germany and in Anglo-Saxon regions more than their EU counterparts.
Econometric results in Section 5 will shed further light on these aspects.
- Figure 1 and 2 about here 3. Location determinants of FDI
3
Unfortunately, Who Owns Whom does not provide any information on the share of ownership, nor on the type of the
investment, so that we cannot identify different locational patterns for wholly owned vs. shared ownership ventures nor for
greenfield vs. acquisitions.
4
See the Appendix for details on the NUTS classification and the list of regions considered in the analysis.
5
Illustrations in figure 1 are somewhat misleading as regards Portugal and Sweden because we have only one NUTS-1
region for these countries (see Table A.1 in the Appendix), and data are not normalised by population and income.
Regressions will allow more accurate use of the data concerning regions belonging to these as well as to the other countries
considered in this paper.
5
Location choices can be modeled as the outcome of a process where firms compare alternative sites
and choose the profit maximizing one. However, it is worth mentioning that this choice is framed
within the more general decision of whether to serve a foreign market and, eventually, how to do so.
In particular, in the context of our empirical analysis, this means that a firm faces the problem of
whether to serve the EU market, and thereafter it will decide whether to do so with exports, licensing,
collaborative ventures, or some combination. Then, if FDI or a combination involving FDI is the best
choice, the firm decides where to set up their activities. This work focuses on the last step of the
decision process, that is selecting the region where to establish a manufacturing activity among
alternative sites. Therefore, our analysis is conditional on multinational firms having decided to set up
production in Europe6, assuming that the (simplified) model of firm behavior we have illustrated is
appropriate7.
Theoretical literature has identified a number of variables affecting firms’ profits from alternative
locations. In the ‘traditional’ literature (Beckman and Thisse, 1986), determinants of firms’ location
choice comprise measures of costs and accessibility to production factors (labor and raw materials),
transportation costs, size and characteristics of the markets. If the investor produces easily
transportable goods, local demand has little influence on location decisions. By considering the entire
spatial area (Europe in our case) as its outlet market, the firm thus chooses its location on the basis of
cost considerations and, then, exports to nearby locations. On the other hand, when transport costs are
important, the local market size plays a major attraction role.
6
This has implications for the interpretation of our empirical results since the probability of locating anywhere in Europe,
should be conditioned to all prior decisions taken by the firm. One should first take into account the probability that Europe is
chosen instead of other continents. Secondly, location decisions should incorporate the probability of choosing to invest
rather than any other alternative entry strategy (exports, licensing, joint ventures, strategic alliances); and the probability that
FDIs are combined with any of such strategies. This becomes particularly relevant to the extent that some variables affecting
location choice also affect the probability of selecting other strategies. The impact of some variables may also be different if
investment decisions follow a previous entry by different means (e.g. exports followed by FDI) or if multinationals penetrate
a market via a bundle of complementary strategies. Empirical work has hardly been able to model location decision processes
thoroughly. Devereux and Griffith (1998) account exclusively for the export/FDI decision and for the location of US firms in
the EU, but they are constrained by data availability to consider rather aggregated choice sets (countries). On the contrary,
most studies addressing the determinants of location choices of foreign firms at a rather disaggregated level are constrained to
condition their analysis to the export/FDI decision and to focus on individual countries (see, for example, Basile, 2004, and
Mariotti and Piscitello, 1995, for the case of Italy; Crozet, Mayer and Mucchielli, 2003, for the case of France; Barrios, Gorg
and Strobl, 2002, for the case of Ireland; Guimaraes, Figueiredo and Woodward, 2000, for Portugal; Devereux, Griffith and
Simpson, 2003, for the United Kingdom). The present work takes the broader perspective of the location within Europe of
multinational firms originating from any country of origin . Therefore, while it does not take into account alternative entry
modes, it allows to model the firm’s location choice within an integrated economic space more accurately.
7
In fact if serving the EU market is not the primary objective of a multinational firm, the behavioural model may not be the
one illustrated here. For example, when firms go abroad to gain access to specific knowledge sources, the export/FDI
decision looses significance, and the choice of where to locate in the EU (and even the choice of how to operate in that
location) might be defined prior to the decision of selling in Europe.. In these cases we are inclined to believe that our
interpretation of empirical results holds a fortiori. In fact, in these cases location decisions are largely independent of export
6
This literature has also emphasized the role of public policy in affecting the firm’s cost function
and thus its location decision. In particular, the effect of tax regimes (in particular on corporate income
and labour costs), public incentives (at regional, national and supranational level) and infrastructure
(e.g. roads, railways and telecommunications) have been considered.
The literature on foreign firms’ site selection has recently grown alongside with the advances in the
‘new economic geography’ (Fujiita, Krugman and Venables 1999). Following a typical cumulative
causation approach, it is suggested that industrial firms tend to localize where other firms of the same
industry are present. The benefits of this form of externality – connected with the number of
manufacturing plants clustered in a specific area (agglomeration economies) – are well known:
namely, access to a more stable labor market, availability of intermediate goods, production services
and skilled manpower, and knowledge spillover between adjacent firms.
Admittedly, agglomeration economies tend to reach limit values, and agglomeration diseconomies
eventually emerge. Firms operating in markets with relatively large numbers of firms face stronger
competition in product and labor markets. This acts as a centrifugal force which tends to disperse
activities in space. Once the centrifugal forces exceed the effects of the agglomeration economies in a
region, firms will look for locations in contiguous regions where production costs are lower, while at
the same time taking advantage of external economies to some degree, given the short distances
involved. In this case, agglomeration economies would operate at a supra-regional level, giving rise to
an external regional effect. This hypothesis is in line with the process of progressive industrialization
in the periphery proposed in Puga and Venables (1996), where the distance between economies plays a
role in selecting location.
However, in the case of foreign-owned firms, agglomeration economies derive not only from the
generic number of local incumbents, but also from the number of other foreign firms operating in the
same geographical area. As suggested by Head et al. (1999), “if foreign investors - who have less
initial knowledge about regional locations than their domestic counterparts - only receive signals on
costs and benefits of location decision, but face strong difficulties in observing them directly, they
might mimic each others’ location decision”. Finally, agglomeration economies may be generated
among firms belonging to the same business group. The idea is that to the extent that firms gain
and other entry modes and there would be no real need to scale investment decisions by the probability that other decisions
7
experience and get acquainted with a given context, uncertainty is likely to decrease and MNEs will
perceive lower risks from further investments (Castellani and Zanfei, 2004). As a result, MNE
experience will determine persistence in firms’ location choices.
4. Random Utility Discrete Choice Models: CL, NL and MXL
To the extent that location choices can be modelled as the outcome of profit maximization, the
empirical analysis can borrow standard econometric tools developed for the estimation of random
utility maximization problems, which allow to identify the determinants of consumer choice among a
finite number of alternatives (discrete choice models). The profit firm i realizes from location site
(region) j (πιj) can be decomposed into a deterministic, linear-in-parameters, part that depends on
observable attributes of the region (X) and a stochastic part εij:
π ij = β ′X ij + ε ij
(1)
The firm chooses the location that yields the highest profit, that is, π ij > π il ∀l ≠ j
( l = 1,..., L ) .
Under the assumption of independently and identically distributed (iid) error terms, with type I
extreme-value distribution, the probability of choosing location j is (McFadden, 1974):
PijCL = exp( β ′X ij )
∑
L
l =1
exp( β ′X il )
∀l ≠ j ( l = 1,..., L ) (2)
This is known as the conditional logit (CL) model. A major drawback of this model is the
assumption of Independence of Irrelevant Alternatives (IIA), according to which alternatives are
symmetric substitutes after controlling for observable characteristics. This assumption would be
violated if different groups of regions have similar unobservable characteristics, so that errors would
be positively correlated across choices, and CL parameters would be biased (Herriges and Kling,
1997). For example, if some (unobserved) country effect occurs, the choice would not be made among
symmetric substitutes, as the degree of substitution between regions within national boundaries may
be higher than across countries.
are taken (see caveats in footnote 7 above).
8
The Nested Logit (NL) model partially solves this problems, by allowing some correlation between
errors within mutually exclusive groups (nests), while maintaining the hypothesis of no correlation
across nests. The IIA assumption, thus, holds across nests but not within them.
Let us assume that the L alternatives are grouped into K nests, that is each alternative belongs to a
nest Bk. Thus, the probability that a firm chooses region j is
PijNL = Pij | k × Pik =
exp( β ′X i , jk )
exp(δ ′Z ik + λk IVk )
×
∑ exp(β ′X i, jk ) ∑ exp(δ ′Zik + λk IVk )
j∈ B k
(3)
k
where X and Z are the vectors of characteristics specific to the j location in nest k, and to the k
upper nest, respectively. IVk is called inclusive value and measures the average profit that a firm can
expect to obtain from locating in any region within nest k. Its parameter, λk, reflects the degree of
independence among unobserved portions of utility, with lower λk indicating less independence. In
order to be consistent with the random utility (profit) maximization (RUM) behaviour, all the inclusive
value parameters should lie inside the unit interval, that is: 0<λk<1 ∀k = 1, 2,..., K . If this condition is
met, alternatives within the same nest are more similar (or, more precisely, they are closer substitutes)
than alternatives outside the nest. For λk > 1 the model is consistent with RUM only locally, i.e. for
some range of the explanatory variables but not for all values. This can be interpreted as evidence that
the nesting structure is not appropriate and, as shown by Herriges and Kling (1997), this biases the
estimated effect of the various choice determinants. In these cases, estimates can be improved by
trying a different nesting structure. However, in some cases, there is no obvious nesting structure and
the choice has to rely basically on a trial and error procedure, which will eventually lead to single out a
nesting structure which satisfies the 0<λk<1 criterion. Bayesian NL models (Poirier, 1996) provide a
systematic procedure to choose, among all possible combinations of alternatives within a given choice
set, the nesting structure which is more supported by the data. This solution has two major drawbacks.
First, in case of large number of choices, as in our analysis of location choices across up to 50 regions
in Europe, the number of possible combinations (i.e. alternative nesting structures) increases
9
dramatically and computation can be very burdensome.8 Second, these models still rely on relatively
rigid substitution patterns, which, for example, do not allow a region belong to more than one nest. A
more flexible (and computationally more efficient) way to capture correlation among alternatives is
the mixed logit (MXL) model.
Like CL and NL models, MXL models can easily be derived from a random utility model (Train,
2003). Following the error component specification of the MXL model, the profit from location j is
denoted
π ij = β ′X ij + vij + uij = β ′X ij + μ ′Yij + uij
(4)
where X ij and Yij are vectors of variables observed for the firm i and the alternative j (the region
where to invest); β is a vector of parameters to be estimated which are fixed over firms and
alternatives; μ is a random vector with a density g ( μ | Ω ) over all firms; and uij is an iid error term
(with type I extreme value distribution). The term μ ′Yij is interpreted as an error component which
induces heteroskedasticity and correlation over alternatives in the unobserved portion of utility. The
MXL probabilities are the integrals of standard logit probabilities over a density of parameters
PijMXL = ∫
exp(β ′X ij + μ ′Yij )
∑l =1 exp(β ′X il + μ ′Yi )
L
g ( μ | Ω)dμ
(5).
These choice probabilities cannot be calculated exactly because the integral in (5) usually does not
have a closed form solution. Therefore, the choice probabilities are simulated by drawing values of μ
from its distribution. The simulated probabilities are then included in the likelihood function to obtain
the simulated likelihood. Thus, β and μ parameters are estimated through maximization of the
simulated likelihood function.
MXL choice probabilities do not exhibit IIA and any substitution (correlation) pattern can be
obtained by the appropriate specification of Yij and g(.). For example, an analogue to the NL model
can be obtained through a MXL model by grouping alternatives into K nests (k = 1,..,K) and defining
8
For example, Verlinda (2005) considers a model with only 4 choices which yields up to 26 possible a priori tree structures.
In order to determine, a posteriori, which model is most supported by the data, he had to apply the reversible jump procedure
as well as the Laplace approximation to all 26 models.
10
Yij as a vector of dummy variables, d kj , that equal 1 if alternative j is in nest k, and zero otherwise (see
Brownstone and Train, 1999).
5. Econometric results
5.1 Data
We utilized the Elios dataset described in section 2 to identify the location choice of the 5,509
foreign firms during the period 1991-1999. Each firm faces 50 possible choices, then for each
individual, the dependent variable is equal to 1 if firm i is set in region j and zero for all regions
different from j. Independent variables have been selected according to the existing literature on
location choices of multinational firms summarised in section 3 (see Table 2, for a list of variables and
data sources). In particular, we control for: (a) regional market size and potential (higher for regions
which are closer to large markets), (b) agglomeration economies (overall agglomeration, foreign firms
agglomeration, their spatial lags9 and MNE experience), (c) characteristics of the local labour market
(unit labour costs, schooling and unemployment rates, as well as the tax wedge on labour, measured at
the national level, since in Europe there is no room for diversified fiscal treatments within countries),
(d) national policy (national effective average corporate tax rate10 and an index for the regional stock
of infrastructure ), (e) European policy (Structural and Cohesion Funds).
- Table 1 about here With regards to point (e), it is important to remark that while most individual countries have
introduced specific incentives targeted to multinational firms, the EU has no specific policy instrument
‘dedicated’ to the attraction of foreign investments, and foreign firms benefit from ‘generic’ public
incentives, such as those co-financed by the European Union through the Structural and the Cohesion
Funds. Structural Funds have different Objectives: Objective 1 is aimed at boosting the development
of laggard regions (that is regions with a per-capita GDP lower than the 75% of the EU average) and
accounted for about two-thirds of total Structural Funds allocated over the 1989-99 period. Cohesion
Funds are instead distributed to those countries (Ireland, Portugal, Spain and Greece) with a per capita
9
Spatial lags in overall and foreign firm agglomeration variables are expected to capture any congestion effect, which will
discourage location in highly agglomerated regions and favour establishment in regions nearby.
10
See Devereux, Griffith and Klemm, 2002.
11
GDP lower than the 90% of the European average. In our analysis, the effect of European policy is
captured by two variables: a continuous variable measuring the total amount of Structural Funds
allocated over the 1989-93 period to each region and a dummy variable set to 1 if the country receives
Cohesion Funds. Unfortunately, we could not gather any specific variable controlling for incentives or
attraction policies targeted to multinational firms by the individual regions and/or countries. This
might induce some caution in the interpretation of the results, as the effect of such policies may be
picked up by other covariates (such as agglomeration economies and fiscal policy).
5.2 Do national boundaries matter? The NL model results
As we discussed above, the NL model improves on the standard CL by allowing different degrees
of substitutability among regions. In particular, regions whose unobserved portion of profits are
correlated can be grouped into common nests, improving the quality of estimation. In this perspective,
the choice of the nesting structure is crucial. An appropriate nesting structure requires that 0<λk<1 for
all the K nests11, suggesting that errors (i.e. the stochastic component of profits) for the various regions
within a nest are positively correlated or, in other words, that regions within a nest are perceived as
closer substitutes by investing firms. Countries are the natural nests. Cultural specificities, barriers to
trade and to the movement of people should make regions belonging to the same country more similar
than regions from different nation states.
In table 2 we report the results of NL estimations. First notice that the hypothesis of equal
substitution between all regions within Europe, which is implied by the CL model, is rejected from a
Likelihood Ratio (LR) test in all specifications. Therefore, some nesting is required. In column (1)
and (2) we test the conjecture that regions are closer substitutes within countries and we soundly reject
it. Following other works in this literature (see for example Head and Mayer, 2004), we first constrain
the degree of substitution within the various countries to be equal, i.e. λk=λ for all k, and find that the
IV parameter is significantly greater than one. However, more interesting insights emerge from the
analysis of the unconstrained estimation. In fact, we reject the hypothesis of a common IV parameter
for all countries and find that λk is significantly larger than 1 for Germany, Spain and the UK, while in
11
Strictly speaking, λk>1 can be consistent with RUM for some range of the explanatory variables. However, Herriges and
Kling (1996) show that if the number of alternatives is large (thus each choice has a relatively low probability), as it is in our
case, the upper bound of λk, for consistency with RUM rapidly approaches 1 for any range of the explanatory variables.
12
the case of Italy we find evidence that regions within national boundaries are closer substitutes than
regions outside borders12. In other words, a country effect characterizes Italian regions. To illustrate,
one may venture saying that, taking into account differences in observable characteristics, a relatively
advanced region like Emilia Romagna is perceived by MNEs as more similar to, for example, Italy’s
Mezzogiorno than to Baden-Wurttenberg, while the latter is considered more similar to, for example,
Ile de France than to the Berlin region. Indeed, this result provides some more robust explanation to
the fact that almost all Italian regions attract a remarkably lower number of investors than other EU
regions (Basile, Benfratello and Castellani, 2005). When, in column (3) and (4), we include nationwide variables (namely the tax wedge on labor, the effective average corporate tax rate and the
Cohesion funds dummy variable) the IV for Italy becomes not statistically different from 1, but still
three out of five IV parameters (France, Germany and the UK) are outside the unit interval. The fact
that multinational firms do not consider regions within national boundaries as closer substitutes than
regions across countries, reveals that locations compete to attract foreign plant location more across
than within borders.
In sum, we have a rather robust evidence that nesting choices according to national boundaries,
would not be consistent with firms’ profit maximization. This calls for the use of alternative nesting
structures, that is other substitution patterns among regions which would make economic sense. There
are at least two ways to work this problem out. First, we could follow the Bayesian strategy proposed
by Poirier (1996) and search, within all possible combination of choices, for the nesting structure most
supported by the data (and consistent with the RUM). However, as suggested in the previous section,
with the high number of choices in out problem (50 regions) the number of possible nesting structures
becomes intractable and identifying the “best” one (in a statistical sense) would be nearly impossible.
Furthermore, relying on mutually exclusive nests, we would maintain a rather simplified correlation
structure in the unobserved portion of profits. Therefore, we prefer leaving the NL framework and
applying the MXL model (which allows more general correlation patterns and includes the NL as a
special case). As a starting point for our MXL, we use one of the possible NL specification which we
found to be consistent with the RUM. This will allow us to show, inter alia, that MXL indeed
improves the quality of NL estimates. In column (5) and (6) of Table 2 we report the results from one
12
Notice that IV parameters are fixed to 1 in the cases of Ireland, Sweden and Portugal since these nests contain only one
13
nesting structure where regions are grouped according to their eligibility to Objective 1 Structural
Funds and we find that this is consistent with RUM (although this is not necessarily the one which is
best supported by the data) since the IV parameters of both the Objective 1 and non-Objective 1
regions are well below 1. In other words, this suggests that firms consider regions eligible to Objective
1 Funds as substituting more closely with other Objective 1, than with non-eligible regions and vice
versa. In the next section we will build on this result and allow for more flexible pattern of substitution
between regions.
5.3 Location determinants of foreign firms in Europe: the MXL model results
In the previous section, the hypothesis that the country borders matter (or, in other terms, that
“country/region” tree structure is consistent with the RUM) has been rejected and another aggregation
of regions (Objective 1/Non-Objective 1) has been found to be consistent with the RUM. As we have
discussed earlier, this cannot be considered the “true” nesting structure, nor even the more appropriate
pattern of correlation between the error terms in our location choice problem, and can give us
unreliable estimates of the location determinants. In order to mitigate this problem, we have estimated
MXL (error component) models, which generalizes NL and allows to control for more flexible
patterns of correlation in the error terms. Results are reported in Table 3. In column (1) we replicate
the error structure of the NL model.13. In this way, the pattern of correlation of the MXL model is
similar to that of the NL model in columns 3 of Table 214. The two standard deviations enter
significantly in the error component, thus confirming the NL model results.15
In column (2) we exploit the flexibility of MXL and capture more complex substitution patterns
between regions, by introducing three additional error components related to (a) the regional market
size, (b) the MNE experience and (c) the tax wedge on employment.16 The standard deviations of these
region. They are the so-called degenerate nests.
13
In fact, the two error components are normal deviates (which are denoted as ‘Std(Objective 1)’ and ‘Std(Non-Objective
1)’) multiplied by two dummies for each Objective 1 and Non-Objective 1 region. In order to allow these dummies to enter
only the error term, we followed Brownstone and Train (1999) and constrained the mean of their estimated parameter to zero.
14
However, the two models are not exactly equivalent, since the MXL is heteroschedastic, while the NL is homoschedastic.
15
MXL models have been estimated through the GAUSS routine available on Kenneth Train’s website
(http://elsa.berkeley.edu/Software/abstracts/train0196.html), using 25 Halton draws.
16
Unlike the Objective 1 eligibility dummies, which enter only the error component, the three additional elements enter also
the deterministic portion of profits. Each random component is a normal deviate multiplied by the respective variable (a), (b),
and (c). The selection of variables to enter the stochastic portion of profits was determined following a stepwise procedure.
We first estimated a model in which all parameters were assumed to have a normally distributed coefficient in the population
with mean and standard deviation being estimated. Then, those coefficients that did not obtain significant standard deviation
were considered as fixed parameters. Thus, the model reported in Table 3 is that one that seems to better explain the choices
in terms of likelihood.
14
new error components enter significantly, while the standard deviation of Objective 1 regions turns
non-significant. This finding suggests that firms tend to show a higher degree of substitution between:
Non-Objective 1 regions, regions with similar market size, locations where their parent have similar
experience and countries with a similar tax wedge on employment. In terms of policy, one could use
these results to identify, among locations with high substitutability, each regions’ major competitors
for attracting FDI. For instance, our results inform us that multinationals perceive Non-Objective 1
regions like Baden Wurttemberg and Ile de France as close substitutes, hence directly competing for
foreign investment, while the same does not apply for Objective 1 regions like Abruzzo-Molise and
Scotland. From a statistical point of view, our results suggest that MXL models allow to capture more
complex substitution patterns between locations than classical NL models. This is further supported by
the Bayesian information criterion (BIC) which indicates that the specification of the MXL model
reported in column (2) fits the data better than the MXL reported in column 1: the increase in the loglikelihood obtained by including the three additional error components is less than compensated by the
increase in the number of estimated parameters. Thus, for the discussion of the estimated parameters
that enter the non-stochastic portion of utility we rely only on MXL results reported in column 2 of
Table 3.17
Firstly, our tests confirm that demand and agglomeration economies increase the attractiveness of
regions. The spatial lag of foreign firms agglomeration is positive and significant, thus indicating
agglomeration externalities generated by the stock of foreign plants operating in the same industry
across regional boundaries.
Secondly, the characteristics of the local labour market seem to play an important role for the
regional attractinevess of FDIs. In particular, unit labor costs (average wages on labour productivity
ratio) and unemployment rates have a negative effect, while the level of educational attainment has a
positive impact on the regional attractiveness of FDI. Overall, these findings suggest that, holding
labour productivity and the level of human capital constant, foreign investors are very responsive to
differences in labor costs across regions. The reverse also applies. For any given level of labour cost, it
is human capital that makes the difference in attracting foreign investors. One might also observe that
17
It is worth mentioning that, since the continuous variables are in logs, their estimated coefficients are approximations of the
elasticity of the probability of choosing a particular region with respect to the explanatory variable for the average investor.
15
the probability of choosing a particular region is much more responsive to human capital (as proxied
by secondary schooling) than to the labour cost.
Thirdly, we support that the total amount of Structural Funds allocated over the 1989-1993 period
and the membership to a Cohesion country (Ireland, Spain and Portugal in our sample) have positive
and significant effects on the probability of a region attracting the location of a foreign firm. We
interpret these results as evidence that EU policies have contributed to mitigating centripetal
agglomeration forces and have attracted multinationals towards peripheral regions and countries.
Finally, Table 3 shows some unexpected results for other policy variables. In particular, one may
notice that the stock of public infrastructures and the tax wedge on employment turn out nonsignificant, while the effect of corporate taxes seems to be positive and significant. These unexpected
results might be due to a strong correlation of these variables with agglomeration variables. In
particular, as also suggested by Bénassy-Quéré et al. (2003), in the presence of agglomeration forces
only very large tax differentials provide the right incentive to delocalise economic activities. This
finding is also in line with the theoretical predictions of some recent new economic geography models
which cast some doubts on the traditional wisdom that producers should move to whichever country
(region) has the lowest tax rates, and suggest that agglomeration forces create quasi-rents that can be
taxed without inducing delocation (Baldwin and Krugman, 2004). We test for this hypothesis, in
column (3) of Table 3, where we drop our measures of agglomeration from the final specification.
Interestingly, both the corporate tax rate and the tax wedge on labour enter with a negative and very
significant sign in this specification, suggesting that only when agglomeration economies do not play a
role in affecting firms’ location decisions, tax competition is an effective policy measure for attracting
multinational firms. Similarly, when the effect of agglomeration variables is not controlled for, the
coefficient on the regional public infrastructure turns out to be significantly positive, suggesting that in
presence of agglomeration economies only very large differentials in infrastructures would affect
firms’ location.
5.4 US versus EU MNEs
In fact, it can be shown that the elasticity for the average investor equals the coefficient times (L − 1)/L, which in our case
(L=50) is approximately equal to 1.
16
In Section 2, we noticed that EU and non-EU MNEs seem to follow different location patterns. In
Table 4 we investigate this issue further, estimating our richest specification on US and EU firms
separately. Most variables keep their sign and magnitude but two significant differences emerge. First,
corporate tax rate is positive and significant for EU MNEs, while it is negative (although nonsignificant) for US MNEs. This is consistent with the view that US firms are more responsive to
corporate tax rates, probably because the actual tax differential is on average higher than for EU
MNEs, which tend to pay relatively higher taxes in their home countries. Second, labor market
conditions have a remarkably different impact on the attractiveness of regions in the case of US MNEs
as opposed to EU MNEs. In fact, from Table 4 one notices that a high tax wedge on employment
discourages investment from US MNEs significantly but have no effect on EU MNEs, while lower
unit costs attract EU investors, but not US MNEs, which in turn seem to place a greater importance to
the quality of human capital. US and EU multinationals thus appear to pursue different objectives. The
behavior of US based firms is consistent both with a market seeking/asset exploiting strategy and with
an asset seeking strategy (Narula and Zanfei 2005). On the one hand, access to the EU market is an
important motive for US firms to locate within Europe. On the other hand, given their specialization in
skill intensive activities, US multinationals are particularly interested in gaining access to local
applications abilities and knowledge sources, as their limited emphasis on labor cost reduction and
high reactivity to the quality of human capital seem to reveal. Conversely, the greater emphasis of EU
MNEs on labor costs would be consistent with the idea that intra-EU investments are part of a strategy
of reorganization of Continental activities where the local availability of cheap labor plays a greater
role. The different role played by the error components in the case of US and EU firms would further
reinforce these interpretations. In fact, while the former tend to substitute between regions non eligible
for Objective 1 funds, competition to attract EU MNEs seems to be mainly between Objective 1
regions.18
6. Concluding remarks
This paper analyzed the determinants of location choices of multinational firms in Europe. Most of
previous studies focused on location decisions within single countries, often analyzing location at a
18
It is worth mentioning that the stepwise procedure that we used to find the most appropriate error components prompted us
to use the stock of all foreign firms as an error component for US firms, since MNE experience turned out non-significant.
17
rather geographically disaggregated level, but making the hypothesis that firms choose regions within
and not across countries. In other words, firms are usually assumed to choose countries first and then
decide in which region within that country they locate their activities. The process of European
integration is making this perspective rather narrow, since regions can be expected to compete to
attract FDIs with other regions both within and across national boundaries. This study provides
empirical support to the view that country boundaries do not matter. In fact, we find that multinational
firms consider regions across countries as closer substitutes than regions within national boundaries.
This suggests that, when taking location decisions, multinational firms perceive the EU as a relatively
(albeit not completely) integrated area, rather than a collection of independent countries and EU
regions compete with other locations outside their national boundaries (within the EU) to attract
foreign plants. However, Italy turns out as a special case. In fact, MNEs perceive a strong country
effect when locating in Italy, suggesting that they take their location decision on a presumption that
investments in Italian regions would yield systematically lower profits than investments in regions
from other countries sharing similar observable characteristics19.
We also find that regions which received a larger amount of Structural Funds and those belonging
to Cohesion countries are particularly attractive for foreign multinationals. This supports the view that
EU regional policy, creating more favourable conditions for investments in Peripheral regions through
funding (among others) training, infrastructure and R&D activities, have succeeded in counteracting
agglomerative forces which tend to concentrate activities in Core regions. However, further work is
required along these lines. First, one would like to control for more direct measures of EU policies,
such as the actual amount of funds transferred to the various regions for different activities, e.g.
training, infrastructures and R&D. Second, careful measurement of national and regional policies
specifically targeted to foreign investments is required, in order to assess the differential impact of EU
versus national and regional policies correctly. Third, further investigation should be devoted to assess
whether the EU Structural and Cohesion Funds have eventually distorted the efficient allocation of
multinational activity in Europe.
We do not believe this makes much of a difference in the interpretation of the results.
19
This could summarized by a quote from a recent article appeared in a US newspaper (“Italian Puzzle: The Land That
Doesn’t Seem To Fit”, The New York Times, August 20, 2003): “Italy has occupied an odd place in Europe, to potent to be
ignored, but too peculiar to be embraced”
18
Finally, we find important differences among EU and US investors. In particular, they seem to
place a different emphasis on the characteristics of the labour markets: while US firms are not
significantly discouraged by higher wages and place higher importance to human capital, EU MNEs
are attracted towards regions with relatively lower unit labor costs. This is consistent with the latter
being more interested in the reorganization of production within their ‘home market’, and the former
being motivated mainly by market-access and asset-seeking considerations.
19
Appendix: The NUTS classification
The Nomenclature of Units for Territorial Statistics (NUTS) is a hierarchical classification of
administrative areas, used across the European Union for statistical purposes, i.e. for collection,
development and harmonization of Community regional statistics. At the top of the hierarchy (NUTS0) are the individual member states of the EU, below that are levels 1 to 320. Generally speaking,
territorial units are defined in terms of the existing administrative units in the Member States. The
NUTS level to which an administrative unit belongs is determined on the basis of population
thresholds. Where the population of a Member State as a whole is below the minimum threshold for a
NUTS level, that Member State itself constitutes a NUTS territorial unit of that level (thus, Ireland
consists of only one NUTS-2 region, while Sweden consists of only one NUTS-1 region).
NUTS 2 (Basic regions) is the framework generally used by Member States for the application of
their regional policies and is therefore the appropriate level for analysing regional-national problems,
whereas NUTS 1 (major socio-economic regions grouping together basic regions) should be used for
analysing regional Community problems, such as the effect of customs union and economic
integration on areas at the next level down from national areas. NUTS 3, which broadly comprises
regions which are too small for complex economic analyses, may be used to establish specific
diagnoses or to pinpoint where regional measures need to be taken. The NUTS nomenclature serves as
a reference for the framing of Community regional policies: for the purposes of appraisal of eligibility
for aid from the Structural Funds, regions whose development is lagging behind (regions concerned by
Objective 1) have been classified at the NUTS 2 level.
For the present work, we utilized NUTS-1 as the elemental location choice, as it represents the best
solution to the trade off between complexity and exhaustiveness. In fact, NUTS-0 (countries) represent
too large geographical units to study MNEs location behaviour, as countries encompass a lot of
heterogeneity among them and do not account completely for the location factor which MNEs rely
upon. The use of NUTS-2 or NUTS-3 levels would imply the inclusion of too a large number of
alternatives, which would make estimation unfeasible. In Table A.1. we summarize the NUTS
classification for the 8 countries and indicate the list of regions used in the analysis.
20
Here we refer to the nomenclature operating during the period of our analysis. Classification criteria changed in July 2003.
20
Table A.1 – The NUTS classification for France, Germany, Italy, Ireland, Spain, Portugal,
Sweden and the UK
Country
NUTS 1
Countries non-eligible for the Cohesion Fund
Länder:
DE1 BADEN-WUERTTEMBERG; DE2 BAYERN; DE3 BERLIN; DE5
BREMEN; DE6 HAMBURG; DE7 HESSEN; DE9 NIEDERSACHSEN;
DEA NORDRHEIN-WESTFALEN; DEB RHEINLAND-PFALZ; DEC
DE
SAARLAND; DEF SCHLESWIG-HOLSTEIN;
(Germany)
The following regions have been excluded due to the lack of data on
Structural Funds
DE4 BRANDENBURG; DE8 MECKLENBURG-VORPOMMERN; DED
SACHSEN; DEE SACHSEN-ANHALT; DEG THUERINGEN
Z.E.A.T + DOM:
FR1 ILE-DE-FRANCE; FR2 BASSIN PARISIEN; FR3 NORD-PAS-DECALAIS; FR4 EST; FR5 OUEST; FR6 SUD-OUEST; FR7 CENTRE-EST;
FR
FR8 MEDITERRANEE;
(France)
The following region has been excluded due to the lack of data on foreign
plant location
FR9 DEPARTEMENTS D'OUTRE-MER
Gruppi di regioni:
IT1 NORD OVEST; IT2 LOMBARDIA; IT3 NORD EST; IT4 EMILIAIT (Italy)
ROMAGNA; IT5 CENTRO; IT6 LAZIO; IT7 ABRUZZO-MOLISE; IT8
CAMPANIA; IT9 SUD; ITA SICILIA; ITB SARDEGNA
SE
SE Sverige (NUTS 1)
(Sweden)
Standard regions
UK1 NORTH; UK2 YORKSHIRE AND HUMBERSIDE; UK3 EAST
UK
(United MIDLANDS; UK4 EAST ANGLIA; UK5 SOUTH EAST (UK); UK6
Kingdom) SOUTH WEST (UK); UK7 WEST MIDLANDS; UK8 NORTH WEST
(UK); UK9 WALES; UKA SCOTLAND; UKB NORTHERN IRELAND
Countries eligible for the Cohesion Fund
Agrupacion de comunidades autonomas :
ES1 NOROESTE; ES2 NORESTE; ES3 COMUNIDAD DE MADRID; ES4
CENTRO; ES5 ESTE; ES6 SUR;
ES (Spain)
The following region has been excluded due to the lack of data on foreign
plant location
ES7 CANARIAS
NUTS 2
NUTS 3
Regierungsbezirke
Kreise
Régions + DOM
Départements +
DOM
Regioni
Provincie
Riksområden
Län
Group of counties
Counties/Local
authority
regions
Comunidades
Provincias +
autonomas + Ceuta
Ceuta y Melilla
y Melilla
Regional
Authority,
Regions
IE
IE IRELAND (NUTS 1, NUTS 2)
(Ireland)
Continente + Regiones autonomas :
PT1 CONTINENTE;
PT
The following regions have been excluded due to the lack of data on foreign
(Portugal)
plant location
PT2 ACORES; PT3 MADEIRA
21
Comissaoes de
coordenaçao
regional + Regioes
autonomas
Grupos de
Concelhos
Acknowledgments
This paper is part of a research project on “The impact of technological innovation and globalization
on the performances of the Italian and European economy” (Contract n. 2001133591_001) funded by
the Italian Ministry of University and Scientific Research (MIUR). The authors are grateful to Claudio
Cozza and Elvio Ciccardini for excellent research assistantship. A previous version of this paper was
presented at the Western Economic Association International (WEAI) Conference in Denver,
European Trade Study Group (ETSG) Meeting in Madrid, European Network on Industrial Policy
(EUNIP) Conference in Porto, L’Industria Annual conference in Parma and at seminars in Paris, Pisa,
Rome and Urbino. We wish to thank all the participants to those meetings, as well as Nigel Diffield,
Holger Gorg and Andres Rodriguez-Pose, two anonymous referees and the co-editor Jim Tybout, for
helpful comments. Usual disclaimers apply.
References
Baldwin, R. E., Krugman, P., 2004. Agglomeration, integration and tax harmonization. European
Economic Review 48, 1--23.
Barrios S., Gorg, H., Strobl, E., 2002. Multinationals’ location choice, agglomeration economies and
public incentives. Mimeo.
Basile, R., 2004. Acquisition versus greenfield investment: the location of foreign manufacturers in
Italy. Regional Science and Urban Economics 34, 3--25.
Basile, R., Benfratello, L., Castellani, D., 2005. Attracting foreign investments in Europe: are Italian
regions doomed? Rivista di Politica Economica XCV, n. I-II.
Beckman, M., Thisse J., 1986. The location of production activities, in P. Nijkamp (ed.), Handbook of
Regional and Urban Economics, Elsevier, Amsterdam.
Benassy-Quèrè, A., Fontagnè, L., Lahrèche-Rèvil, A., 2000. Foreign direct investment and the
prospects for tax co-ordination in Europe. CEPII Document de travail n. 2000-06.
Brownstone, D., Train, K., 1999. Forecasting new product penetration with flexible substitution
patterns. Journal of Econometrics 89, 109--129.
Castellani, D., Zanfei, A., 2004. Choosing international linkages strategies in the electronics industry.
The role of multinational experience. Journal of Economic Behavior and Organization, 53, 447-475
Crozet, M., Mayer, T., Mucchielli, J.L., 2003. How do firms agglomerate? A study of FDI in France.
Regional Science and Urban Economics.
Devereux, M., Griffith, R., 1998. Taxes and location of production: evidence from a panel of US
multinationals. Journal of Public Economics 68, 335--367.
Devereux, M., Griffith, R., Simpson, H., 2002. Regional selective assistance, agglomeration and firm
location. Presented at the CEPR/IMOP Conference on “FDI and the Multinational Corporation”,
Hydra, Greece, 20-21/9/2002.
Devereux, M.P., Griffith, R., Klemm, A., 2002. Corporate income tax reforms and international tax
competition. Economic Policy 35, 451--495.
Disdier, A.C., Mayer, T., 2004. How different is eastern Europe? Structure and determinants of
location choices by French firms in eastern and western Europe, Journal of Comparative
Economics 32, 280–296
European Commission, 1996. First Report on Economic and Social Cohesion. Luxemburg.
Eurostat, 2002. European Union Foreign Direct Investment Yearbook 2001.
Fujita, M., Krugman, P., Venables, A.J., 1999. The Spatial Economy: Cities, Regions, and
International Trade. Cambridge: MIT Press.
Guimaraes, P., Figueiredo, O., Woodward, D., 2000. Agglomeration and the location of foreign direct
investment in Portugal. Journal of Urban Economics 47, 115--135.
22
Harris, C., 1954. The market as a factor in the localization of industry in the United States. Annals of
the Association of American Geographers 64, 315--348.
Head, C.K., Mayer, T., 2004. Market potential and the location of Japanese investment in the European
Union. Review of Economics and Statistics 86, 959--972.
Head, C.K., Ries, J.C., Swenson, D.L., 1999. Attracting foreign manufacturing: investment promotion
and agglomeration. Regional Science and Urban Economics 29, 197-218.
Herriges, J.A., Kling, C.L., 1996. Testing the consistency of nested logit with utility maximization.
Economic Letters 50, 33--39.
Herriges, J.A., Kling, C.L., 1997. The performance of nested logit models when welfare estimation is
the goal. American Journal of Agricultural Economics 79, 79--802.
Krugman, P., 1991. Geography and Trade. The MIT Press, Cambridge, MA.
Krugman, P., 1992. A Dynamic Spatial Model. National Bureau of Economic Research. Working
Paper, n. 4219.
Mariotti, S., Piscitello, L., 1995. Information costs and location of FDIs within the host country:
empirical evidence from Italy. Journal of International Business Studies 26, 815--841.
Martin, R., 1999. Critical survey: the new “geographical turn” in economics – some critical reflections.
Cambridge Journal of Economics 23, 65--91.
Martinez-Mongay, C., 2000. ECFIN’s Effective Tax Rates. Properties and Comparison with Other Tax
Indicators. Economic Papers European Commission 146, October.
Mayer, T., Mucchielli, J.L., 1999. La localisation à l’étranger des entreprises multinationales. Une
approche d’économie hiérarchisée appliqué aux entreprises japonaises en Europe. Economie et
Statistique 326-327, 159--176.
McFadden, D., 1974. Conditional logit analysis of qualitative choice behaviour, in Zarembka P. (ed.)
Frontiers in econometrics, Academic Press, New York.
Midelfart-Knarvik, K.H., Overman, H.G., 2002. Delocation and European integration. Is structural
spending justified? Economic Policy 17/35, 323--359.
Mucchielli, J.L., Puech, F., 2002. Location and agglomeration of French FDI in European countries
and regions: a probabilistic approach. Mimeo.
Narula R., Zanfei A., 2005. “Globalisation of Innovation” in J.Fagerberg, D.Mowery, R.R. Nelson
(eds.) Handbook of Innovation, Oxford University Press, 2005 (318-345)
Poirier, D., 1996. A Bayesian analysis of nested logit models. Journal of Econometrics 75, 163--181.
Puga, D., Venables, A., 1996. The spread of industry: spatial agglomeration in economic development.
Journal of the Japanese and International Economies 10, 440-464.
Train, K.E., 2003. Discrete Choice Methods with Simulation. Cambridge University Press,
Cambdridge.
UNCTAD, 2002. World Investment Report 2002. Transnational Corporations and Export
Competitiveness. United Nations, New York and Geneva.
Verlinda, J., 2005. A Bayesian analysis of tree specification in nested logit models. Economics Letters
87, 67--73.
23
Table 1 - Variable List and Description
Variables
Market
Size
Description
Source
Type
Log of Value Added in region j
Eurostat
Region
Market
Potential
Log of the sum of value added in all regions r≠j
weighted by the inverse euclidean distance between
the major cities in r and j
Eurostat
Region
Overall
agglomeration
Log of the number of establishments in region j (and
sector s). Also spatial lags are considered
Elios
RegionSector
Elios
RegionSector
Foreign-firms
agglomeration
MNE Experience
Unit labour costs
Unemployment
Rate
Log of the cumulative number of foreign-owned
firms within region j (and sector s). Also spatial lags
are considered
Log of the number of firms in region j controlled by
the same parent of firm n
Log of (average labor costs/labor productivity)
Eurostat
FirmRegion
Region
Log of unemployment rate
Eurostat
Region
Tax wedge on
employment
Log of (sum of social contributions, income taxes
and consumption duties over total employment)
Structural Funds
Log of European Structural Funds expenditure
Elios
MartinezMongay C.
(2000)
European
Commision
Objective 1 region
1 if the region is within Obj.1, 0 otherwise
Cohesion country 1 if the country receives Cohesion Fund, 0 otherwise
Public
Index of infrastructure stock in region j at 1985
Confindustria
Infrastructure
Institute for
Corporate tax rate Log of national effective average corporate tax rate
Fiscal Studies
24
Country
Region
Region
Country
Region
Country
Table 2 – Location determinants of FDI in Europe. NL Regressions. Choosing the nesting structure
Variable
Market Size
Market Potential
EU Structural Funds 1989-1993
1
0,156**
(0,037)
-0,222*
(0,121)
0,052
(0,121)
2
0.174**
(0.039)
0.148
(0.147)
0.059**
(0.012)
3
0.189**
(0.044)
-0.013
(0.140)
0.043**
(0.013)
0.748**
(0.097)
0.116
(0.208)
0.085
(0.089)
-0.068
(0.090)
1.508**
(0.038)
0.472**
(0.054)
0.450**
(0.052)
-0.377
(0.281)
0.886**
(0.299)
-0.231**
(0.063)
0.548**
(0.158)
-0.307**
(0.061)
4
0.226**
(0.046)
0.278*
(0.164)
0.037**
(0.014)
0.724**
(0.134)
-0.360
(0.279)
0.147
(0.095)
-0.045
(0.089)
1.506**
(0.042)
0.455**
(0.054)
0.399**
(0.054)
-0.522*
(0.283)
0.707**
(0.312)
-0.222**
(0.073)
0.238
(0.201)
-0.181**
(0.067)
-0,275**
(0,077)
1,414**
(0,036)
0,276**
(0,044)
0,590**
(0,044)
-0,805**
(0,219)
1,320**
(0,262)
-0,190**
(0,061)
0,765**
(0,128)
-0,114*
(0,058)
-0.251**
(0.079)
1.426**
(0.038)
0.333**
(0.047)
0.498**
(0.050)
-0.399*
(0.247)
0.682**
(0.287)
-0.266**
(0.063)
0.310*
(0.172)
-0.120**
(0.059)
1.040°°
(.004)
1.073°°
(0.035)
0.995
(0.039)
1.080°°
(0.039)
0.823°°
(0.038)
1.098°
(0.056)
1.146°°
(0.004)
1.179°°
(0.048)
1.140°°
(0.049)
1.159°°
(0.046)
0.942
(0.050)
0.980
(0.051)
1.000
(Fixed)
1.000
(Fixed)
1.000
(Fixed)
1.000
(Fixed)
Cohesion Funds Countries
Tax Wedge on Employment
Effective Average Corporate Tax Rate
Public Infrastructure
MNEs Experience
Overall Aggl.
Foreign Firm Aggl.
Spatial Lag of Overall Aggl.
Spatial Lag of Foreign Firms Aggl.
Unit Labour Cost
Secondary School Enrolment Ratio
Unemployment Rate
5
0.122**
(0.033)
-0.055
(0.109)
-0.036**
(0.010)
0.358**
(0.083)
0.116
(0.157)
0.159**
(0.076)
-0.038
(0.067)
1.197**
(0.048)
0.386**
(0.043)
0.335**
(0.045)
-0.290
(0.221)
0.835**
(0.242)
-0.228**
(0.049)
0.612**
(0.128)
-0.234**
(0.049)
6
0.112**
(0.030)
-0.071
(0.098)
0.033**
(0.009)
0.256**
(0.077)
0.064
(0.142)
0.143**
(0.067)
-0.056
(0.061)
1.079**
(0.055)
0.338**
(0.041)
0.308**
(0.041)
-0.287
(0.200)
0.771**
(0.220)
-0.217**
(0.045)
0.572**
(0.115)
-0.205**
(0.044)
IV parameters
UK
France
Germany
Italy
Spain
Ireland
Portugal
Sweden
Objective 1
0.682°°
(0.053)
Non-Objective 1
0.769°°
(0.035)
No. of Firms
5,509
5,509
5,509
5,509
5,509
5,509
Log-likelihood
-16957
-16927
-16924
-16900
-16924
-16918
LR test: CL vs. NL
65,68**
52,20**
14,74**
62,20**
14,74** 27,04**
LR test: Fixed vs Country-specific IV par.
60,94**
47,46**
12,30**
Notes: The dependent variable is equal to 1 if firm i is set in region j and zero for all regions different from j. Standard
errors in brackets. Asterisks denote confidence levels: * p<.10 and ** p<.05. The symbol (°) denotes confidence levels
for the hypothesis that IV parameters are different from 1: ° p<.10 and °° p<.05.
0.848°°
(0.031)
25
Table 3 – Location determinants of FDI in Europe. MXL Regressions. All foreign investors
Variable
Market Size
Market Potential
EU Structural Funds 1989-1993
Cohesion Funds Countries
Tax Wedge on Employment
Effective Average Corporate Tax Rate
Public Infrastructure
MNE Experience
Overall Agglomeration
Foreign Firms Agglomeration
Spatial Lag of Overall Agglomeration
Spatial Lag of Foreign Firms Aggl.
Unit Labor Cost
Secondary School Enrolment Ratio
Unemployment Rate
Error components
Std (Objective 1 Regions)
Std (Non-Objective 1 Regions)
1
0.133**
(0.040)
0.031
(0.134)
0.038**
(0.012)
0.596**
(0.085)
0.044
(0.173)
0.215**
(0.082)
0.024
(0.083)
1.400**
(0.043)
0.458**
(0.050)
0.397**
(0.047)
-0.321
(0.249)
0.867**
(0.268)
-0.208**
(0.061)
0.641**
(0.129)
-0.223**
(0.057)
2
0.161**
(0.043)
0.270*
(0.164)
0.056**
(0.014)
0.691**
(0.102)
-0.220
(0.213)
0.281**
(0.094)
0.054
(0.086)
1.734**
(0.072)
0.470**
(0.053)
0.404**
(0.049)
-0.305
(0.265)
0.755**
(0.286)
-0.279**
(0.065)
0.548**
(0.151)
-0.249**
(0.061)
0.275**
(0.059)
1.114**
(0.155)
-0.289**
(0.061)
0.531**
(0.241)
0.573**
(0.246)
0.356
(0.362)
0.761**
(0.249)
0.297**
(0.124)
1.853**
(0.133)
-2.665**
(1.073)
-16794
5,509
33,761
0.383**
(0.166)
0.380**
(0.190)
0.252*
(0.131)
4.511**
(0.192)
7.231**
(0.658)
-17493
5,509
35,115
Std (Market Size)
Std (MNE Experience)
Std (Tax Wedge on Employment)
Log-likelihood
Number of firms
Bayesian Information Criterion (BIC)
-16927
5,509
34,001
3
1.170**
(0.026)
0.702**
(0.108)
0.030**
(0.014)
0.792**
(0.089)
-2.618**
(0.207)
-0.741**
(0.075)
0.156*
(0.085)
Notes: The dependent variable is equal to 1 if firm i is set in region j and zero for all regions different from j.
Standard Errors in parenthesis. Asterisks denote confidence levels: * p < .10, ** p < .05. Bayesian
Information Criterion = - 2 log-likelihhod + p log(N), where p is the number of parameters and N is the
number of firms. Estimation have been obtained through the Gauss routine provided by Kenneth Train
(http://elsa.berkeley.edu/Software/abstracts/train0196.html), using 25 Halton draws.
26
Table 4 – Location determinants of FDI in Europe. MXL Regressions by country of origin
Variable
Market Size
Market Potential
EU Structural Funds 1989-1993
Cohesion Funds Countries
Tax Wedge on Employment
Effective Average Corporate Tax Rate
Public Infrastructure
MNE Experience
Overall Agglomeration
Foreign Firms Agglomeration
Spatial Lag of Overall Agglomeration
Spatial Lag of Foreign Firms Aggl.
Unit Labor Cost
Secondary School Enrolment Ratio
Unemployment Rate
Error component
Std (Objective 1 Regions)
Std (Non-Objective 1 Regions)
Std (Market Size)
US MNEs
0.280**
(0.095)
0.329
(0.314)
0.084**
(0.028)
0.328*
(0.204)
-1.743**
(0.582)
-0.206
(0.194)
0.190
(0.173)
1.812**
(0.140)
0.445**
(0.120)
0.353**
(0.115)
-0.467
(0.571)
0.864
(0.616)
-0.045
(0.130)
1.264**
(0.325)
-0.341**
(0.126)
EU MNEs
0.112**
(0.055)
0.035
(0.222)
0.104**
(0.019)
1.052**
(0.133)
0.101
(0.287)
1.199**
(0.150)
0.140
(0.110)
1.460**
(0.065)
0.358**
(0.067)
0.479**
(0.064)
-0.385
(0.355)
1.012**
(0.386)
-0.348**
(0.090)
0.437**
(0.198)
-0.789**
(0.079)
0.327
(0.285)
1.199**
(0.280)
0.408**
(0.196)
0.889**
(0.213)
0.056
(0.394)
0.358**
(0.162)
-0.208**
(0.080)
Std (MNE Experience)
Std (Foreign Firms Agglomeration)
Std (Tax Wedge on Employment)
Log-likelihood
Number of firms
Bayesian Information Criterion (BIC)
Notes: as in Table 2
27
1.970**
(0.268)
5.075**
(1.149)
-4097.9
1,405
8,340
4.288**
(1.029)
-9873.0
3,222
19,903
Figure 1 - Number of subsidiaries established in 1991-1999 in EU Nuts 1 regions
Source: Elaborations on Elios (University of Urbino)
Figure 2 - Share of subsidiaries established by EU MNEs in EU Nuts 1 regions (19911999)
Source: Elaborations on Elios (University of Urbino)
28