2 and only report the taxation parameters of interest.
In the first sensitivity check, we replace tertiary school enrollment by al-
ternative skill measures in the construction of INT 1ijt, ..., INT4ijt and the
interaction terms involving ∆SKij,t-1 : gross secondary school enrollment in
Model A; the sum of professional, technical, kindred and administrative work-
ers to total employment in Model B (as in Carr, Markusen, and Maskus 2001;
and Markusen and Maskus, 2002); and physical capital stocks per worker in
Model C. As can be seen, most of the point estimates, especially the ones
of statutory tax rates and depreciation allowances, exhibit the same sign as
in Table 2. Only in the case of host country depreciation allowances (δj,t-1)
we obtain positive but insignificant parameter estimates in Models A and C
which should be negative according to the predictions of the theoretical model.
Similarly, the sign of δi,t-1 × ∆SKij,t-1 × I(∆SKij,t-1 > 0) is now positive
in Models A and C (it was and should be negative), and the coefficient of
δj,t-1 × ∆SKij,t-1 × I(∆SKij,t-1 > 0) becomes positive in Models B and C (it
was negative in Table 2, as expected). ti,t-1 × ∆SKij,t-1 × I(∆SKij,t-1 > 0)
is now positive in Models B and C, as it should be (it was negative before).
Generally, we would prefer the original model in Table 2 and Model A in Table
3 over Models B and C , since the sample size is substantially reduced for the
latter ones (to 1011 and 1410 observations, respectively). Further, as can be
seen from Table A3 in the Appendix, the correlation between the interaction
effects is much lower for tertiary school enrollment than for secondary school
enrollment. Therefore, we consider the original model as preferable over Model
A.
The second sensitivity check (Model D) replaces the knowledge-capital
model variables that are not related to corporate taxation by standard grav-
ity model variables (i.e., parent and host country GDP and GDP per capita).
For instance, such a specification is employed by Mutti and Grubert (2004),
Benassy-Quere, Fontagne, and Lahreche-Revil (2005), and also by Blonigen
and Davies (2004) in a paper on the tax treaty effects on FDI. In Model
D, the parameter signs are very similar to their counterparts in the original
model, except for the interaction effects ti,t-1 × ∆SKij,t-1 × I(∆SKij,t-1 > 0),
δi,t-1×∆SKij,t-1×I(∆SKij,t-1 >0)andδj,t-1×∆SKij,t-1×I(∆SKij,t-1 >0),
which have changed their sign. ti,t-1 × ∆SKij,t-1 × I(∆SKij,t-1 > 0) is now sig-
nificantly positive, as expected. However, it must be said that a test on the joint
significance of the skilled labor endowment variables INT 1ijt, ..., INT4ijt indi-
cates that the gravity specification is less suitable than the knowledge-capital
25