its in a given host country (tjwt); and the parent and host country-specific net
present values of depreciation allowances (δit , δjt). To avoid a possible endo-
geneity bias, we use the lagged values of all tax variables.20
Regarding the parent country’s statutory tax rate, we have no clear-cut hy-
pothesis on the sign of the parameter estimate. However, the above theoretical
model suggests that its impact depends on the skilled-to-unskilled labor endow-
ment ratio in the parent country relative to the host (see Table 1). We include
the interaction term ti,t-1 × ∆SKij,t-1 × I(∆SKij,t-1 > 0), where we expect a
positive sign for the exemption as well as the credit method. Additionally, we
include the present values of depreciation allowances, δi,t-1 and δj,t-1 , and two
interaction terms between the depreciation allowances and the skill differences,
δi,t-1×∆SKij,t-1×I(∆SKij,t-1 >0)andδj,t-1×∆SKij,t-1×I(∆SKij,t-1 >0).
Similar to the parent country statutory corporate tax rate, the inclusion of these
interaction terms is necessary, since the theoretical model does not predict a
clear-cut relationship between the depreciation allowances and outbound FDI.
The theoretical model predicts a negative parameter for the first and a positive
one for the second interaction term.
Overall, the empirical specification reads as follows
F DIijt = β1ti,t-1 + β2tj,t-1 + β3tjw,t-1 + β4δi,t-1 + β5δj,t-1
+ β6 [ti,t-1 × ∆SKij,t-1 × I(∆SKij,t-1 > 0)]
+ β7[δi,t-1 × ∆SKij,t-1 × I(∆SKij,t-1 > 0)]
+ β8 [δj,t-1 × ∆SKij,t-1 × I(∆SKij,t-1 > 0)]
+ β9ΣGDPijt + β10∆GDPi2jt + β11INT1ijt + β12INT2ijt
+ β 13 INT3 ijt + β 14 INT 4 ijt + μij + λt + εijt (22)
where μij and λt are fixed country pair and time effects, respectively. ε is an
identically and independently distributed stochastic error term.
6 Empirical analysis
Baseline results: We provide details on the data sources and descriptive
statistics for all variables in the Appendix (Tables A1 to A3) and in Figure 4.
The figure illustrates the change in all considered components of corporate tax-
20Given the large number of taxation-related variables (eight) it seems infeasible to employ
instrumental variable methods. However, with panel data one can use lagged values of the
tax variables to get rid of the endogeneity problem as long as the contemporaneous variation
in foreign direct investment does not cause a change in taxation in the past.
20