Spatial complementarity of FDI: the example of transition countries

Abstract

This article investigates spatial determinants of FDI location. In particular, it focuses on FDI in neighbouring countries and foreign market potential for a panel of 25 transition countries in 1993–2010. The spatial FDI spillovers are found to be positive and economically large. Moreover, omitting spatial FDI leads to a serious misspecification of the model explaining FDI location and biases estimation of the coefficient of the foreign market potential variable, which is found to be a non-robust determinant of FDI location.

The spatial complementarity is stronger for disaggregated data such as bilateral FDI and sector level FDI. There is substantial heterogeneity of spatial FDI spillovers across sectors. Spillovers are large and positive for services sectors and non-significant or even negative for manufacturing sectors.

Notes

 1. The determinants of FDI activities in the region are discussed, among others, by Bevan and Estrin (2004), Carstensen and Toubal (2004) and Lane and Milesi-Ferretti (2007).

 2. Alternatively, positive spillovers that increase firm level productivity can come through technology diffusion (see Keller 2004) on the role of FDI in spreading of technology and the spatial nature of diffusion).

 3. The model with a spatial lag of the dependent variable as one of the regressors is called a spatial autoregressive model (SAR) by analogy with the time-series literature, where AR(1) is a model with a one-period lagged value of the dependent variable included as one of the regressors.

 4. Each MNC is associated with one product variety l. All firms located in the same country i produce according to the same technology. For ease of presentation, I drop the product index.

 5. The earlier version of this article had a spatial autoregressive error structure similar to Kapoor et al. (2007). Accounting for the spatial autocorrelation in the error does not significantly influence the estimation of the coefficients in the model. More importantly, as shown by Badinger and Egger (2009), to estimate the parameters of such a model with good precision, one would need around 200 spatial units.

 6. The sample includes Albania, Armenia, Azerbaijan, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, the Czech Republic, Estonia, Georgia, Hungary, Kazakhstan, Kyrgyz Republic, Latvia, Lithuania, Macedonia, Moldova, Poland, Romania, Russia, Serbia, Slovakia, Slovenia, Tajikistan and Ukraine. Turkmenistan and Uzbekistan have been excluded because of lack of data on human capital. Adding them to the sample does not change the conclusions but requires dropping human capital from the list of explanatory variables.

 7. Eleven observations on FDI inflows are less than or equal to zero. I define the dependent variable as to make the distinction between zero and missing or negative values. The choice of the additive constant does not change the main conclusions of the study. There are also five observations with substantially negative FDI inflows that are dropped from the analysis.

 8. In rare cases, where the data are not available, I use the International Labour Organisation (ILO) data on monthly wages. The ILO data have more gaps than the UNECE data and are available only until 2008.

 9. The Barro–Lee Educational Attainment dataset (2011) is available for 1995, 2000, 2005 and 2010. I use linear interpolation to fill the gaps in the data. For Bosnia and Herzegovina and Georgia the variable is constructed based on the ILO data on the share of labour with tertiary education. For Belarus, the variable is constructed based on 1999 and 2009 census data.

10. Separate data for oil and gas are aggregated into one variable measured in millions of barrels using a conversion factor provided in the BP report – 1bcm of gas is equal to 6.6 millions barrels equivalent. The report includes only countries with substantial oil and gas resources; it also gives information on the total amount of oil and gas in all other countries in the region. To compute values for countries not included in the report, I assume that the remaining totals are distributed among all those countries proportionally to their geographical size.

11. The inverse squared distance weights give higher weights to closer countries. I also estimated and presented the results of the model with weights inversely related to distance, which assign higher importance to the influence of remote countries. The choice of weighting matrix did not have any impact on the conclusions of the study.

12. See for example Campos and Kinoshita (2003), Bevan and Estrin (2004) and Carstensen and Toubal (2004).

13. The data coverage for bilateral FDI inflows is small. In addition, there is a large number of zero and negative FDI inflows. Therefore, I report only results with the bilateral FDI stock from country j to country i as the explanatory variable.