ation: parent and host country statutory corporate tax rates, their depreciation
allowances, and the host economy’s withholding tax rate.
The figure covers the same sample of host countries and years that is consid-
ered in the empirical analysis. We report the mean as well as the minimum and
maximum values of each tax parameter and year. Obviously, there is enough
time variation in every component, rendering parameter estimation by a fixed
effects model feasible.
Table 2 summarizes the results for (i) the full sample (covering all country-
pairs and years, independent of the implemented method of double taxation
relief), (ii) the sub-sample covering the country-pairs which apply the exemp-
tion method, and (iii) the sub-sample covering country-pairs which apply the
credit method.21
To facilitate the comparison of the point estimates with the theoretical
hypotheses, we indicate in Table 2 whether our findings are in line with the
predictions (labelled ’Y ’) or not (’N’). If the expected effects are ambiguous
according to Table 1, i.e., for the entries ’+/-’ (e.g., the credit method), this
is indicated by a ’?’ in the table. All estimated standard errors and test statis-
tics are corrected for heteroskedasticity and autocorrelation, following Newey
and West (1987). In general, the model fit is quite well. Inspired by Razin,
Rubinstein, and Sadka (2005) we check for a potential bias due to sample se-
lection arising from missing FDI data. We follow Wooldridge (1995, 2002) in
applying testing procedures that are suited for panel data. However, in our
sample of OECD country-pairs, there is no indication for endogenous selec-
tion into the sample, given the adopted specification. This can be seen from
the insignificant parameters of the inverse Mills’ ratio reported in Table 2.
With regard to the usual knowledge-capital variables, we find across the board
that the impact of overall and relative country size (ΣGDPijt and ∆GDPi2jt)
is in line with theory, whereas the skill-endowment interaction term parame-
ters (INT1ijt, ..., INT4ijt) are insignificant in almost all specifications.22 In
21In our sample, the exemption (credit) method applies for around 59 percent (34 percent)
of the observations (either through unilateral tax law or through bilateral tax treaties). The
remaining observations use the deduction method (around 7 percent). Since the latter group
contains only 149 observations, it is not possible to estimate the tax effects precisely for the
deduction sub-sample.
22In this regard, our findings do not match with the empirical results in Carr, Markusen,
and Maskus (2001) and Markusen and Maskus (2002). However, there are four reasons for
this. First, we use stocks of outbound FDI rather than foreign affiliate sales as our dependent
variable (however, Blonigen, Davies, and Head, 2003, indicate that the results tend to be very
similar for FDI and foreign affiliate sales). Second, we apply a log-linear specification (as
suggested by Mutti and Grubert, 2004) rather than one in levels. Third, our sample covers a
panel of OECD parent and host countries that are relatively similarly endowed, which differs
21