model for bilateral FDI in our sample of country-pairs and years.
Next, we estimate a dynamic model including lagged outward FDI stocks
on the right-hand-side of the regression (Model E; see Devereux and Freeman,
1995, for such a specification). To avoid an endogeneity bias inherent in dynamic
panel data models with fixed effects (see Baltagi, 2005, p. 135), we use a GMM
estimator based on first differences as proposed by Arellano and Bond (1991).
However, the estimated parameter for the lagged dependent variable is rather
low at 0.22 and not significant at 10 percent (see the notes of Table 3). The
remaining coefficients are very close to the original ones (with the exceptions of
ti,t-1×∆SKij,t-1×I(∆SKij,t-1 >0)andδj,t-1×∆SKij,t-1×I(∆SKij,t-1 >0),
where the signs change).
Furthermore, we exclude the transition economies (Model F) and, alterna-
tively, the transition countries as well as the non-EU members from the sample
(Model G). Especially the transition economies have reduced their statutory
corporate tax rates significantly within the sample period and faced a strong
increase in inward FDI at the same time, which might exert a crucial impact
on the estimation results. However, with the exception of the interaction terms
ti,t-1×∆SKij,t-1×I(∆SKij,t-1 > 0) and δj,t-1 × ∆SKij,t-1 × I(∆SKij,t-1 >0)
we obtain the same signs for the coefficients of interest in Model F as in the
original model. In Model G, several of the originally significant parameter esti-
mates are now insignificant, especially the one of the host country statutory tax
rate. This comes at no surprise, since the sample size is dramatically reduced
to only 846 observations in this experiment.
In the last two exercises, we control for weighted third-country explanatory
variables to make sure that the parameters of the bilateral variables are not
affected qualitatively by the exclusion of third-country effects in the regres-
sions. An inclusion of third-country explanatory variables can be justified from
the perspective of recent theoretical work on complex multinational firms (see
Yeaple, 2003; Ekholm, Forslid, and Markusen, 2005; and Grossman, Helpman,
and Szeidl, 2006). In particular, Model H includes simple third-country aver-
ages of all explanatory variables.25 Model I includes inverse-distance-weighted
25For instance, the third-market parent country corporate tax rate is the simple average of
the lagged corporate tax rate of all other parent countries in a particular year, and similarly
for all other variables that vary across parent countries and years only. In contrast, the third-
market host country corporate tax rate is the simple average of the lagged corporate tax
rate of all other host countries in a particular year, and similarly for all other variables that
vary across host countries and years only. Finally, all variables that vary across country-pairs
are simple averages across the parent and host country dimensions in each year. Not only
the tax parameters but also all other explanatory variables enter in their bilateral and their
third-market form, simultaneously.
27