Our data concern stocks of FDI and stocks of immigrants and emigrants. A relevant
specification problem in our empirical analysis is endogeneity, which may arise from two main
different sources. The first is the possible presence of omitted variables. Regarding the outward
FDI, it may happen that a third variable, for example the liberalization of restrictions with a partner
country, simultaneously drives investments towards that economy and emigrants inflows from the
country. Regarding the inward FDI, an unobserved positive productivity shock in one of our
countries may simultaneously attract capital and immigrants, which induces a correlation between
the error term and the network variables. The inclusion of the dummies and of the time fixed effects
should, however, obviate for this type of endogeneity.
The second source is reverse causality, which can arise since migration and FDI may
influence each other. The employees of multinationals can be transferred to the foreign firm’s
subsidiaries, or, from the latter to the firm’s headquarters. In this case, causality would run from
FDI to migration, therefore biasing OLS estimates. This reverse causality problem, however, is
more likely to arise in specific cases. Given the bilateral nature of the data considered, it is more
likely to concern the interactions between outward FDI and emigration and between inward FDI
and immigration. Furthermore, it is more likely to involve the migration of skilled individuals,
especially the OECD residents. The latter face low immigration restrictions from foreign countries
and are more likely to be sent abroad by the multinationals of the home country. Hence, we expect
that the reverse causality problem, if present, may introduce an upward bias especially in the
coefficients of the OECD-immigrants in the regressions of the inward equations and of the
emigrants in the outward FDI regressions (regarding Italy).6
Following previous studies using gravity models (Gould, 1994; Girma and Yu, 2002), we
estimate a dynamic version of the above equation using a lagged dependent variable in addition to
the other regressors; this allows to account for the autoregressive component of the FDI stocks, and
to check for the robustness of the results, too.
IV. The Data
Four different databases have been utilized, each corresponding to one of the four countries
considered, France, Germany, Italy and the United Kingdom. The countries of origin included and
the time periods considered vary according to data availability. The partner economies of each of
our four countries are listed in the Appendix.
6 Some recent studies point out that, instead, endogeneity appears to introduce a downward bias, thereby
producing an underestimation of the impact of migration on FDI and bilateral trade (Combes and
Lafourcade, 2005; Javorcik et al., 2006).
11