![Sample our Global Development journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5929/TOC-Subject-Sample-2021-Global-Development.jpg"})
Abstract
This article examines the impacts of industrial agglomeration and outward foreign direct investment (OFDI) on the total factor productivity (TFP) of Taiwanese firms. A vertical FDI-based model of heterogeneous firms is proposed to analyze how agglomeration economies and technology incompatibilities between parent firms and their affiliates can affect firm productivity. This model suggests that firms located in areas with more concentrated industrial agglomerations are more productive, while those engaging in OFDI may not perform better in terms of TFP. Using plant-level data, this article constructs an indicator of industrial agglomeration to appraise agglomeration economies on firm productivity. Based on the data for 578 manufacturing firms and the agglomeration indicator, we estimate a cross-sectional econometric model to empirically assess the productivity effects of industrial agglomeration and OFDI. The empirical results show that local industrial agglomerations exert a positive contribution to firm productivity, but that FDI in China has no significant effects on Taiwanese firms' TFP.
Acknowledgements
We would like to thank David Gile and two anonymous referees for helpful comments and suggestions. Participants in conferences and seminars at National Chengchi University and National Chung Cheng University also provided helpful suggestions. We are thankful for Shih-yi Fu's great assistance. Finally, we also thank the National Tsing Hua University for financial support under Boost Program Grant 98N2926E1.
Notes
1. FDI statistics are from the Industrial Development and Investment Center, Taiwan Ministry of Economic Affairs ( http://twbusiness.nat.gov.tw/page.do?id=16).
2. The China Statistical Yearbook reports that over 60% of FDI as a whole was undertaken by Taiwan, Hong Kong, and Macau firms in 1990, while the share dropped to 40% in 2003. Hu and Owen (2006) examine the driving forces of FDI in China from different parent countries.
3. See Molnar, Pain, and Taglioni (2007) for a recent review of the evidence in OECD countries.
4. Starting from the late 1980s, developing economies have been increasingly recognized as essential suppliers of OFDI. Their share of global OFDI has increased from 3% during the period 1978–1980 to 12.3% over the period 2003–2005 (UNCTAD 2007, 7).
5. The government invested over 2.8 billion US dollars in the HSP from its founding in 1980 to the end of 2009. The number of employees in the HSP has risen from 8275 in 1986 to 132,161 in 2009 (Source: 2009 Hsinchu Science Park Annual Report, http://www.sipa.gov.tw/english/file/20100601133153.pdf).
6. For example, exports to affiliates of US multinationals grew 128% from 1989 to 1999, while sales to local customers grew by only 89% (see Yeaple 2008). Kiyota and Urata (2008) show that 54.6% of Japanese multinationals simultaneously export and import, while only 6.5% of domestic firms are both exporters and importers.
7. However, Damijan, Sašo, and Janez (2007) find that, for Slovenian manufacturing firms, there is no statistically significant productivity advantage of firms with foreign affiliates over exporting firms.
8. Comprehensive literature reviews on agglomeration and its micro-foundations have been recently provided by Duranton and Puga (2004) and Ottaviano and Thisse (2004).
9. Arkolakis et al. (2011) also construct a model with location-specific productivity for multinational firms.
10. Here .
11. Keller and Yeaple (2009) construct a model with continuum tasks, in which offshoring each task will incur technology transferring costs. The higher the technology-transferring costs for each task are, the fewer tasks a firm will off-shore. They then use intra-firm trade data to proxy the fraction of offshoring tasks for a multinational firm. Since we only have information on whether a firm undertakes FDI in China which can not tell the number of offshoring tasks, we simply use two-task model in this article.
12. We thank an anonymous referee for directing our attention to the work of Busch and Reinhardt (1999).
13. The details for computing TFP are deferred to Appendix.
14. To avoid the possible reverse causality running from TFP to the right-hand-side variables, we adopt TFP2002 to be the dependent variable, which is unlikely to affect the previous FDI decision, cluster status, and so on. We also implement the Generalized Method of Moments (GMM) method to resolve the potential endogeneity for a sensitivity check (please refer to Section 5). We note that due to the data limitations on cluster indicators (up to the year 2000), there are only cross-sectional data at hand. Therefore, the unobserved firm-specific heterogeneity cannot be modeled directly. Moreover, existing dynamic panel data methods based on various instrument sets cannot be successfully applied in our context. To further test the robustness of our empirical results, we have experimented to remove the lagged dependent variable TFP2001 in the regression model. It reveals that doing so does not change our result significantly. We owe the above discussion to an anonymous referee.
15. Firms fall into the low cluster groups if their agglomeration indices are below the mean. It is also noted that four types of firms classified share roughly the similar numbers in our sample.
16. It would be interesting to learn the impacts of some interaction terms (such as FDI i × Cluster i and R&D i × Cluster i ) on TFP. However, those interaction terms are excluded in our empirical settings due to the serious collinearity problem – the sample correlations between FDI i × Cluster i and FDI i as well as R&D i × Cluster i and R&D i are 0.97 and 0.93, respectively. In order to examine the impacts of the interaction term between FDI and Cluster on firm productivity, we have tried another specification where we replace FDI with the interaction term between FDI and Cluster (FDI i × Cluster i ). We got insignificant coefficients on FDI i × Cluster i , which may suggest that, other things being equal, firms undertaking FDI in China benefit from local industrial agglomeration as much as non-FDI firms.
17. As noted above, the most recent survey that provides the information for constructing the agglomeration index is the 2000 survey.
18. Since only FDI firms have corresponding regional variables in China, we calculate the industrial mean level of these four variables as the instrument for non-FDI firms.
19. Note that we estimate the main empirical equation with potential endogeneity using the more efficient GMM method as opposed to the conventional IV approach. Since our model is over-identified, we are able to conduct Hansen's J-tests, which reject the model mis-specification in all cases.
20. Based on the Manufacturing Census reports, we separate industries into high, medium, and low R&D-intensive groups proportionately based on the ranking of their R&D-sales ratios. Food and beverage, clothing, paper and publishing, non-metals, basic metals, and fabricated metals belong to low R&D-intensive industries; Textiles, chemicals, petroleum, rubber, plastics, machinery, and other manufacturing industries are defined as medium R&D-intensive industries; the leather, chemical product, computer and telecommunications equipment, electronic components, electrical, transportation and precision instrument industries belong to high R&D-intensive industries.
21. Domms and Dunne (1998) and Cooper and Haltiwanger (2007) document that plant-level physical adjustment is infrequent and lumpy.
22. Aw (2002) also finds a positive correlation between firm size and productivity growth for Taiwanese manufacturing firms.
23. Recall that the firm production efficiency includes three components: a firm's true productivity (ρ i ), local agglomeration (ag ), and the productivity of task two (a 2l /δ l )). Given the same ρ i and a 2l /δ l , firms located in areas with large ag are more productive.
24. The mean value of the marginal effect is 0.3801.