T AX is related to an increase in the expected number of headquarters of
between 150% and 450% percent. However, recall that many municipalities
do not attract foreign headquarters at all. So, it is useful to compute the
marginal effect of TAX evaluated at the sample mean.23 We find that, for
the average municipality, a one-percent reduction of the municipal tax rate
- which is equivalent to a decline by about 0.14 percentage points - leads to
an increase in the number of headquarters of foreign MNEs by about 0.05 in
the (preferred) ZINB model for NEWMNE in Table 3.
A higher fraction of skilled workers in a municipality raises the number
of MNE headquarters as predicted, e.g., by the knowledge-capital model of
the multinational enterprise (Carr, Markusen, and Maskus, 2001; Markusen,
2002; Markusen and Maskus, 2002). Larger regions attract more headquar-
ters as predicted by a variety of models of MNEs (see Markusen, 2002; and
Barba Navaretti and Venables, 2004). This is reflected in the positive coeffi-
cients for population density (POPDEN) and geographical area (AREA).
Region size should not only matter in terms of geographical area or the num-
ber of potential consumers but also with regard to the size of the available
work force. Hence, not surprisingly, we find that MNEs tend to locate in those
regions where the fraction of the population of working age (IDEP RAT) is
large.
The coefficient of the relative size of the area covered with buildings
and streets (BUILT) suggests that such space is particularly important for
MNE location. After controlling for the latter, the share of investments in
a municipality’s total expenditures (INV ) seems to be of minor importance
for MNE headquarters location. Finally, as expected, foreign MNEs are less
inclined to locate their headquarters in the New (formerly Eastern German)
Lander.
5.1 Instrumental variables estimation
The results in Table 3 indicate that conditioning on the included covariates
eliminates a large part of the bias of the TAX parameter contaminating an
23In nonlinear models the partial effect of a variable on the conditional mean of the
dependent variable is a function of the other regressors and parameter estimates as well,
and needs to be evaluated at some value of the covariates. In applied research, it is common
to evaluate marginal effects at the first moment of the distribution of the covariates, i.e.,
at the sample mean. See the Appendix for the derivation of the marginal effects and their
standard errors for the models estimated in Table 3.
16