Good or bad? The influence of FDI on productivity growth. An industry-level analysis

Abstract

This paper attempts to reconcile the often inconclusive evidence on the role of FDI in the process of economic development by taking into account the heterogeneity both among industries and among countries. Using a comparable database at the industry level for 35 countries in the OECD, Asia and Eastern Europe from 1987 to 2002, we test for the influence of both stage of development and sectoral FDI patterns in the relationship between FDI and productivity growth. In certain industries and for the catching-up countries, a significant and positive relationship emerges when FDI coincides with high investment or export orientation.

Keywords:

• FDI
• labour productivity
• heterogeneity
• manufacturing sector
• panel data analysis
• openness

JEL Classifications:

• C33
• F14
• F21
• L60

Acknowledgements

The authors are grateful for the financing of the joint project through HU2005-0018 de Acciones Integradas 18/2005 / Wissenschaftlich-Technisches Abkommen Projekt No. 20/2006. Julia Woerz thankfully acknowledges the support at the Vienna Institute for International Economics Studies (wiiw) while working on this paper and funding by Oesterreichische Nationalbank through Jubiläumsfondsprojekt No. 10214.

Notes

 1. In a few cases, not shown in the figure, the stock of inward FDI valued at the end of the year exceeds the industry's output of the same year, leading to a ratio above 100%. This may happen as a result of heavy foreign investment in a specific year. As a consequence of these investments, one would expect strong increases in output in the following years, for the theoretical reasons given below.

 2. We classified countries into advanced OECD members (Australia, Austria, Denmark, Finland, France, Germany, Iceland, Ireland, Italy, Japan, The Netherlands, Norway, Sweden, UK, USA), and different groups of catching up countries: OECD members (Greece, Mexico, Portugal, Spain, Turkey), the four Asian Tigers (Taiwan, Hong Kong, Korea, Singapore), Emerging Asia (Indonesia, Malaysia, Philippines, Thailand) and CEECs (Croatia, Czech Rep., Hungary, Latvia, Poland, Slovak Rep., Slovenia). See also Appendix Table A1.

 3. The same effect can also be achieved through imports of such goods. In this sense, FDI represents an alternative means to increase the number of available varieties in addition to trade, even if there are qualitative differences between the two.

 4. Spillovers occur when multinationals are unable to capture all the productivity effects that follow in the host country's local firms as a result of the presence of the multinational (Caves 1996).

 5. In the empirical estimation, we additionally face the restriction imposed by the data that domestic and foreign investment cannot clearly be separated. Hence, we proxy for the former by total gross fixed capital formation, while the latter will be given by the growth of FDI inward stocks in a certain industry.

 6. Results for output growth are available from the authors upon request.

 7. The results are extremely robust to alternative specifications, for instance when only industry fixed effects are included or when time invariant country dummies are used as well as when industry and country specific time varying trends are included instead.

 8. We also normalized FDI by employment; however, we believe that this gives a different flavour to the analysis. FDI-to-employment ratios are higher for less labour intensive industries, yielding potentially higher growth rates in the labour intensive industries. This induces a bias towards labour intensive industries. The results were surprisingly not too different again.

 9. In this setting, an endogeneity bias is possible only if the relationship is driven by forward-looking expectations, which are excluded here, especially because growth rates are extremely difficult to predict for catching-up countries. Following Nair-Reichert and Weinhold (2001), we are using one lag in our empirical specification, which is also justified by looking at simple model selection criteria such as adjusted R-squared. The explanatory power of the model is maximized when using one lag as opposed to including 2–4 lags. Degrees of freedom limit the number of further lags to be included.

10. In the System-GMM specification we are using up to three lags as instruments.

11. It would be highly desirable to have detailed information on each individual industry contained in this group, since petroleum extraction is not only very capital intensive, but also very closely tied to endowments and international technology and the distribution networks of big oil multinationals and thus not relevant for every country in the sample. Chemicals on the other hand cover a very wide spectrum of economic activities, ranging from low-skill, resource intensive production to high-skill, technology intensive activities (such as pharmaceuticals). However, for the present sample, covering a wide range of countries, any further disaggregation was not possible.