Modeling FDI based on a spatially augmented gravity model: Evidence for Central and Eastern European Countries

Abstract

Based on a spatially augmented gravity model, the current paper isolates spatial interrelationships in foreign direct investment (FDI) to Central and Eastern European Countries (CEECs) not only across the destination but also across the origin country dimension of FDI. Results show that (i) spatial interrelationships across destination countries are present and are consistent with the predominance of vertical-complex FDI in total FDI; (ii) spatial correlation across origin countries is given in earlier years of transition, while spillover and competition effects cancel over the whole sample period; and (iii) agglomeration forces gain in importance for FDI to CEECs.

Keywords:

• foreign direct investment
• spatial econometrics
• Central and Eastern Europe
• third country effects

JEL Classifications:

• C33
• F21

Acknowledgements

We gratefully acknowledge the suggestions of two anonymous reviewers, who helped to substantially improve the submitted version of the paper.

Notes

1. Blonigen et al. (2007) additionally find that spatial interrelationships are stable over time. The results of Garretsen and Peeters (2009) indicate the importance of agglomeration forces for outward FDI, and Shepotylo (2012) shows the relevance of spatial interdependencies in FDI to a broad sample of transition economies.

2. Baltagi, Egger, and Pfaffermayr (2007) also consider third-country effects in FDI. Specifically, they include spatially lagged explanatory variables as well as spatial autoregressive errors in their empirical model. Their estimation results ‘illustrate that third-country effects are important and lend support to the presence of complex FDI’ (p. 262).

3. Shepotylo (2012) also focuses on transition countries. However, his analysis abstracts from spatial interrelationships across origin countries.

4. We are grateful to the referee who advised us in this respect.

5. Besides this spillover effect a demonstration or herding effect might lead to a positive spatial correlation across origin countries as well: FDI from one origin country can be seen as a device to reduce uncertainty about the locational quality among investors from other countries leading them to also invest in the CEECs (also see Barba Navaretti and Venables 2004, p. 148; Barry, Goerg, and Strobl 2003).

6. Note that models based on uni- and bilateral factors are in principle capable to show whether FDI in a particular destination country is purely horizontally or purely vertically motivated. Yet, they fail to uncover complex-vertical or export-platform motives, which necessitate a multilateral perspective (see, e.g. Blonigen et al. 2008).

7. Blonigen et al. (2008) also include a variable capturing a ‘parent market proximity effect’ in their model. This variable is calculated similarly to the surrounding market potential of a destination country. Its theoretical motivation is quite different, however. Blonigen et al. (2008) argue that large markets (z) that surround a particular origin country (i) of FDI exert incentives for higher outbound FDI from country i to another country (j). According to Blonigen et al. (2008) the presence of large nearby countries (z) will lead to high exports to these countries (due to large market size and relatively low transport costs). This, in turn, implies that output of origin country i is diverted away from country j. To be able to service country j, FDI activity of MNEs from country i in country j might increase. Note that, in this respect, the model of Blonigen et al. (2008) assumes the presence of a purely horizontal FDI motive. Yet, which motive prevails is an empirical question. The motive for FDI will be indicated by the coefficient of the spatial lag variable capturing the destination dimension of FDI. We therefore do not consider a variable capturing the ‘parent market proximity effect’ in our empirical application.

8. In a robustness check we also show the results when μ_ij are considered as fixed effects.

9. As N ≫ T the asymptotic properties of our estimator can be derived with N → ∞ and T fixed. This implies that we can be ‘agnostic’ about the amount of temporal persistence of our variables (see Wooldridge 2001, p. 175).

10. A detailed description of the wild bootstrap is outlined in Appendix A.1. The Matlab code for the whole routine is available from the authors upon request.

11. As it is not clear if the traditional Hausman test is applicable here (see Mutl and Pfaffermayr 2011) we also estimate under assumption of fixed country-pair effects. Results are displayed in Table A8.

12. Details on the testing down procedure are available upon request.

13. Appendix A.2 explicates how these average direct effects are derived based on equation (2.43) given in LeSage and Pace (2009). Note, that standard errors for average direct effects are derived via a bootstrap approach.

14. Estimations are carried out using the xtivreg command (with the vce(boot) option) of Stata 11.1.