According to a sizable body of work in theoretical international eco-
nomics, knowledge-capital embodied in skilled workers is one of the key
determinants of MNE activity (see Markusen, 2002). In accordance with
that line of reasoning, empirical research identified a key role of the local
supply of skilled labor to play for the set-up of MNE headquarters (see Carr,
Markusen, and Maskus, 2001; Markusen and Maskus, 2002; Blonigen, Davies,
and Head, 2003). While previous evidence is available for investment (and
foreign affiliate sales) at the national level, similar arguments ought to hold
for headquarters location within a country. Our skill measure is the share of
workers with tertiary school education.
A second key factor determining MNE activity according to previous re-
search is host country location size (see Markusen, 2002; Barba Navaretti and
Venables, 2004). Using country (or country-pair) data in previous empirical
work, host country location size is typically controlled for by variables based
on gross domestic product (GDP). However, GDP is not available (as well as
endogenous) at regionally very disaggregated levels. For this reason, in our
preferred specification, we include population density, the independency ratio
of the population (i.e., the number of people aged between 15 and 65 years as
a fraction of total population in the region) and geographical area.5 Notice
that - once including log geographical area as well - we may interpret the
coefficient of log population density as reflecting the elasticity with respect
to population size. The independency ratio is the best measure available to
capture the relative size of the working-age population in a region. Including
log area along with population density also approximates land prices and
provides a measure of the relative abundance of land as such, which may be
used relatively intensively in some of the sectors MNEs operate in.6,7
5 One might think of distance to a metropolitan area as a further possible determinant.
However, this is highly correlated with geographical area and population in the cross-
section, and it is wiped out by the method applied with panel data. Therefore, we do not
include this variable.
6There are various ways of specifying these influences. For instance, it turns out that
including log population and log area obtains similar results to the ones with the preferred
specification, where we use population density and log area. Similarly, using the area
covered with buildings and streets (instead of log total geographical area) along with the
log share of area reserved for building obtains similar results. However, in our preferred
specification, multi-collinearity between the regressors is reduced to the largest possible
extent.
7Another strand of research includes market potential - i.e., some inverse-trade-cost-
weighted average of market size of other regions as a determinant of firm location (see