![Sample our Economics, Finance,Business & Industry journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5931/TOC-Subject-Sample-2021-Economics-Finance-Business-Industry.jpg"})
Abstract
The objective of this paper is to test for the importance of local agglomeration externalities in determining inward foreign direct investment (FDI) intensity, viewed as a measure of regional attractiveness to FDI. The links between the degree of FDI inflow penetration into Italy and its determinants at the regional level are examined using alternative fixed effects panel data model specifications, also extended to include spatial effects. It is found that sectoral and regional specificities are relevant in attracting inward FDI into Italy and the different effects of intra‐sector and inter‐sector spillovers can be distinguished. There is also evidence that the importance of agglomeration spillovers connected to the geographical distance varies across regions, indicating both substitution and complementarity effects from FDI inflows into neighboring regions.
Notes
1. Most recent reviews question the extent of foreign firm/domestic firm performance gaps once industry and scale are controlled for. Bellak (Citation2004) provides a recent review of this issue.
2. For a review of the literature see Görg and Greenaway (Citation2001).
3. The significance of agglomeration may capture the correlation between the location of domestic firms and FDI due to the endowment effect, instead of verifying the agglomeration externalities. In order to disentangle the effect of agglomeration from the effect of the geographical distribution of productive factor endowment (Head et al., Citation1995), a set of control variables for factor endowment has to be introduced.
4. For an application to the empirical analysis of FDI see Campa (Citation1993), Blonigen (Citation1997) and Carr et al. (Citation2001).
5. See Lim (Citation2001) for a survey of the literature.
6. Horizontal FDI flows have market seeking objectives and aim at replicating abroad the parent company activities. Vertical FDI flows are instead mainly resource‐seeking.
7. By using the value added as the scale factor a restriction is introduced in the model, that is the coefficient of the value added is restricted to one in a log‐linear regression model of FDI on value added. To avoid model misspecification, this restriction asks for appropriate diagnostic tests.
8. Districts are geographic areas where the economic activity is highly specialized and the average firm dimension is small.
9. The former have market seeking objectives and aim at replicating abroad the parent company activities, thus possibly displacing exportations. The latter are instead mainly resource‐seeking and often feed intra‐ and inter‐industry flows, thus contributing to make the relationship with trade fairly complex.
10. IV Tobit ML estimates have been obtained using exogenous regressors as instruments for the FDI spatial lag variable.
11. Other FDI determinants were also tested, as regional price index by sector. However their exclusion is not rejected by the standard variable omission tests.
12. The trade openness coefficient is negative, while the district one is not significant.
13. The addition of a price index variable does not modify all results. The trade openness coefficient is negative, while the district one is not significant
14. Another approach to test for cross sectional dependence is to directly test if the cross‐correlations of the errors are zero. In our estimates, cross section dependence is confirmed by Breusch and Pagan’s Lagrangian multiplier (LM) test, while it is not by Pesaran (Citation2004) CD test. See Hsiao et al. (Citation2007) for a description of diagnostic tests of cross section independence for Tobit models.
15. The GME estimator can be viewed as a shrinkage estimator that shrinks the data to the priors (uniform distributions) and toward the centre of their supports. It should be pointed out that unlike ML estimators, the GME approach does not require any explicit error‐distribution assumptions: in fact, the GME method selects the most uniform distribution consistent with the information provided by the constraints. In this respect, we do not need to specify a parametric family for the likelihood function, and the estimation rule is flexible with respect to: (i) the dynamic, stochastic nature of economic data; (ii) a non‐random survey design, as well as to the model selection problems. Within the GME framework, all coefficients and errors are expressed in terms of proper probabilities. The basic idea is that rather than search for the point parameter estimates, each parameter is viewed as the mean value of some well‐defined random variable. The unobserved error vector is also viewed as another set of unknowns, and as in the case of the signal vector, each error is constructed as the mean value of a random variable. Under the GME framework, the full distribution of each parameter and of each error (within their support spaces) is simultaneously estimated under minimal distributional assumptions.
