and region was collected, we worked out the database to order it in six comparable sectors at the
regional level. We label the sectors as follows:
- Traditional manufacturing (including food & beverage, textile, paper, metal products, wood
furniture),
- Mechanical, machinery and automotive,
- Electrical, Electronic and high-tech,
- Chemical,
- Financial and business services (including holding),
- Other services.
For every year we have the data corresponding to the six selected sectors and we organize them
in such a way to merge sectors and years. For Catalunya we get a final cross-section matrix of 48
observations. In case of Baden Württemberg we get a cross section matrix of 35 observations because
we dropped out the finance sector. Our choice is motivated by the following consideration: this sector
attracts a large amount of FDI and we do not know in which sectors the direct investments from the
banks are realized. Then, we preferred to remove the finance sector from the sample in the case of
Baden-Württemberg to avoid additional difficulties of interpretation. As for Lombardia, we reduced
the sample to 35 observations since data were not available for all the periods. The sample period
starts from 1997 and ends in 2002. Finally, for all the regions, data on gross fix capital formation stops
in 2001.
The general equation for our estimation is the following:
FDIp = α + βx + ε,
where FDIp represents the annual per-capita inflows of FDI in each region and βx is a vector of
variables selected as proxies for FDI determinants. We followed the literature to select the potential
FDI determinants. We isolate variables related to the local business climate (such as openness to trade,
R&D investments, human capital), as well as some macro-indices (local GDP as proxy for local
wealth as well as gross capital formation). Moreover, among those determinants, we tested two
indicators of productivity, real productivity (measured as real value added per employee) and unit
labor costs (compensation of employees per unit of value added).
One should reasonably expect that all these factors should display a positive correlation with
the amount of FDI inflows. They proxy the local factors that investors are likely to look for when they
decide to invest. The only one that is expected to show a negative coefficient is the unit labor cost
since an increase in this indicator means a decrease in productivity and hence a less attractive
determinant for investors.
We applied the cross section technique for each regional matrix at two dimensions (by year and
by sector). We run the regressions by estimating the matrix with the OLS technique and applying the
White correction for controlling for heteroskedasticity problems. In the regressions for Catalunya and
Baden-Württemberg we control for fixed effects by sector, in order to capture the possible
heterogeneity among sector principally due to their own productive structure (LSDV estimators).
Because of the reduced number of available years in the cross section, we just perform POLS
estimations for Lombardia. The variables selected for each region i and sector j annually are
summarized in Boxes 1, 2 and 3.
In addition, to control for size effects we normalize to population all variables we are using.
This means that we analyze the determinants of FDI per-capita inflows in each of our three regions.
15