Tariffs and EU countries foreign direct investment: Evidence from a dynamic panel model

Abstract

According to the models of the multinational enterprise tariffs play a fundamental role in determining the pattern of foreign direct investment (FDI). The aim of this paper is to assess the impact of tariffs on the outward stocks of FDI of the European Union (EU). We estimate a model based on the knowledge–capital theory of the multinational enterprise over the period 1995–2008 by using a sample of five EU countries and 24 partner countries. We consider, first, manufacturing sector as a whole and, then, six manufacturing industries defined at the two-digit level of the Nomenclature statistique des activités économiques dans la Communauté européenne (NACE) classification. Explanatory variables include an index of applied bilateral tariffs and a dummy to capture the presence of bilateral investment treaties (BITs). A dynamic panel model is estimated through the generalized method of moments estimator, taking also into account the endogeneity of regressors. The results show that the pattern of EU outward FDI is a mix of vertical and horizontal FDI. BITs in force have a significant and positive impact on the outward FDI. The impact of tariffs varies across industries and countries, suggesting the predominance of horizontal FDI in some industries, and the existence of vertical FDI in others.

Keywords:

• trade policy
• foreign direct investment
• panel data

JEL Classifications:

• C33
• F23
• F13

Acknowledgements

The authors wish to thank Giovanni Anania, Giuseppe De Arcangelis, Luca De Benedictis, Alessandro Olper and participants at the XIII European Trade Study Group (ETSG) Annual Conference held in Copenaghen, 8–10 September 2011, 52nd Annual Meeting of the Italian Economic Society (SIE) held in Rome, 14–15 October 2011, and at the First Conference of the Italian Association of Agricultural and Applied Economics (AIEAA), held in Trento, 4–5 June 2012, for their useful comments.

Notes

1. An extensive review of this literature can be found in Blonigen (2005) and Blonigen and Piger (2011).

2. With the most notable exception of the USA, country-level statistics provide, by and large, only stock and flows of FDI, while other information, such as the local sales of subsidiaries or the sales back to the parent country, which could be useful to distinguish vertical from horizontal FDI, are not available.

3. Many papers use the outward stocks of FDI (e.g. Baltagi, Egger, and Pfaffermayr 2007, 2008; Stein and Daude 2007), while others pool inward and outward stocks of FDI (e.g. Head and Ries 2008; Blonigen and Piger 2011).

4. The specification proposed by Carr, Markusen, and Maskus (2001) considers the difference between skills or GDP of host and partner countries, rather than the ratio. However, Carr, Markusen, and Maskus (2001) adopt a specification in levels. Studies considering a specification expressed in logarithm generally use the logarithm of the ratio, that is, the difference between the logarithms (Egger and Winner 2006; Baltagi, Egger, and Pfaffermayr 2007; Egger and Merlo 2007; Egger 2001, 2008).

5. Investment costs have been excluded from our specification, because their impact is likely to be captured by the FE and by the lagged dependent variable in the dynamic model. As explained in Section 5, for robustness check, we have included in GMM estimations a variable representing overall political risks of each host country; this may be considered as a reasonable proxy for the overall investment climate. However, we found that this variable does not significantly affect EU FDI.

6. For example, if assets increase because of an acquisition with local funds, this is registered in the M&A data, but not in the FDI data, which include only cross-border investments.

7. More specifically, our sample includes the following countries: Australia, Argentina, Brazil, Bulgaria, Canada, Chile, the Czech Republic, Egypt, Estonia, Hungary, Israel, Japan, Latvia, Lithuania, Mexico, Morocco, Norway, Poland, Romania, Slovakia, Slovenia, Switzerland, the United States and Uruguay.

8. One consequence of this relatively high level of industry aggregation is that an (two-digit) industry is likely to include both inputs and outputs of a certain sector. For example, the vehicle industry includes not only parts and accessories, but also the final products.

9. As suggested by Wooldridge (2002), we have performed the Verbeek and Nijman (1992) test for attrition bias. Test results reject the hypothesis of attrition bias for the manufacturing sector as a whole, but not when the six industries are considered. However, if we run estimations for the six industries dropping year 1995 observations, the hypothesis of attrition bias is rejected also when considering disaggregated data and the results are substantially similar to those obtained when considering the entire period.

10. In our GMM estimations, no observations are lost because of zero FDI for the manufacturing industry as a whole, while in the six manufacturing industries, if we run GMM estimations with the dependent variable in level rather than in logarithm, only about 7.6% observations are zero values.

11. In GMM and FE estimations, because of the inclusion of the FE, the distance, common language and colony variables have been dropped. GMM estimations are carried out by using the Stata command xtabond2 (Roodman 2009).

12. For GMM estimations the tables report a goodness of fit statistic computed as the squared correlation coefficient between actual and predicted levels of the dependent variable, which is equivalent to the standard R² in an OLS regression (Bloom, Bond, and Van Reenen 2007).

13. The coefficient of log (GDPsum) ranges between 1.7 and 2.5 in Baltagi, Egger, and Pfaffermayr (2008) and between 1.7 and 2.6 in Egger and Merlo (2007).

14. It is worth noting that in our OLS estimation, as in other studies (e.g. Tekin-Koru and Waldkirch 2010), distance has a negative and significant effect on FDI, while common language exerts a positive effect on FDI.

15. As a further robustness check, we have run Table 3 estimations by considering EU preferential tariffs instead of EU applied tariffs and the simple average of the EU tariffs provided by WITS instead of the weighted average of EU tariffs computed as described in Section 4. These results, which are available upon request, do not substantially change with respect to those reported in Table 3.

16. To check robustness of our results, we have included in GMM estimations a variable representing overall political risks of each host country, provided by the PRS Group (http://www.prsgroup.com/FAQ.aspx). The overall political index takes into account the following indicators: government stability, military in politics, socioeconomic conditions, religious tensions, investment profile, law and order, internal conflict, ethnic tensions, external conflict, democratic accountability, corruption, bureaucracy quality. The relative coefficient, however, is not significant.

17. The estimates obtained including the tariff in level are not substantially different from those obtained when including tariffs in logarithms; hence, hereafter we report only results with the tariff in level.

18. In the case of developed countries, the BITs dummy variable is equal to one for 17 observations out of 825 and for only two host countries, which are the Czech Republic and Israel. Hence, the effect of the BIT dummy is likely to be captured by the FE.