that the instrumental variables model leads to a parameter estimate of TAX
which is closer to the NEWMNE-based estimates in Table 3 than to the
M N E-based ones. The difference between the parameter for TAX in the
ZINB model of Table 3 and the IV-GMM model in Table 4 is not statistically
significant at commonly used levels of significance.
The second column in Table 4 reports estimates based on the quasi-
differenced GMM estimator derived in Windmeijer (2006). We use the in-
struments as described before and in the table footnote. However, we elim-
inate the time-invariant variables EAST and AREA from this regression.
It turns out that the parameter of T AX is significantly different from zero
and exhibits the same sign as in the benchmark regressions of Table 3. The
estimated elasticity of the number of MNEs with respect to T AX is much
bigger than before. However, the confidence interval is larger, too.27 So,
we take the results from the quasi-differences GMM estimator mainly as in
indication that the qualitative findings about a significant negative effect of
higher business tax rates on the location of foreign MNE headquarters in
Germany are robust.
Overall, our findings suggest that a marginal reduction of business tax
rates in the average municipality is not enough to attract foreign MNE head-
quarters. Municipality size and the availability of workers, especially skilled
workers, are important. Only those municipalities which have a sufficiently
attractive environment to be on the map of foreign MNEs when planning to
locate subsidiaries in Germany will be able to attract foreign headquarters
by reasonable reductions in their tax rates. Municipalities with less favor-
able environments would have to trade these off with sizable reductions in
the business tax rate. For instance, the estimates in Table 3 imply that a
one-percent reduction in business tax rates increases the number of foreign
MNE headquarters in a region by about 0.05. Hence, the preferred model
indicates that the average municipality has to reduce its tax rate by about
20% (or 2.8 percentage points) to attract a single foreign headquarters. The
average municipality may not want to do so, since the associated losses in
tax income from national firms may easily outweigh the expected raise of tax
27As mentioned before, we do not see the panel model results as the most reliable ones.
The reason is that there is little variation in tax rates and numbers of headquarters per
municipality on average over the short time span covered by the data. To be more specific,
4,080 municipalities changed their tax rate at least once during that time span but in only
345 of those did the number of MNE headquarters actually change (increase or decline).
Hence, the T AX parameter is identified from a fairly small fraction of observations.
19