Cross-sectional estimation of FDI spillovers when FDI is endogenous: OLS and IV estimates for Mexican manufacturing industries

Abstract

Cross-sectional estimates of Foreign Direct Investment (FDI) spillovers are biased when the variable capturing the cross-industry variation of foreign participation is endogenous to the estimated regression model. In this article I introduce an original instrument for this problematic FDI variable, capturing the general FDI intensity of manufacturing industries. I use this instrument to estimate FDI externalities in a cross-section of Mexican manufacturing industries. The main findings show that, in contrast to the common criticism that Ordinary Least Squares (OLS) estimation produces an upward bias in the estimated FDI externality effect, for the sample of Mexican manufacturing industries OLS estimation causes a downward bias. Controlling for the tendency among foreign manufacturing firms to gravitate towards low productivity industries, the instrumental variable estimations provide robust evidence of significantly larger positive FDI spillover effects.

Acknowledgements

I would like to thank Gilles Duranton, Henry Overman, Vassilis Monastiriotis, Steve Gibbons, Jan Willem Gunning, Chris Elbers and two anonymous referees of this journal for their valuable comments and suggestions. Of course, the usual disclaimers apply and all remaining errors are mine.

Notes

¹ Having said this, these solutions may also face limitations. As Hanson (2005) argues, there are a potentially large number of unobserved shocks to host economy firms’ productivity, and if some of these shocks are also related to industry foreign participation, it may not always be possible to fully control for the endogenous component of the FDI variable. Also of course, there is the problem that differencing that is implied by fixed effects estimation may exacerbate measurement error problems (Griliches and Hausman, 1986).

² Findings from the analysis of Mexican plant level data for the second half of the 1980s are more diverse. Aitken et al. (1996) do not find evidence of positive wage spillovers from FDI. In contrast, Aitken et al. (1997) find evidence of positive FDI externalities in the form of market access spillovers. Grether (1999) reports a negative association between the industry level of foreign participation and relative efficiency of Mexican manufacturing plants.

³ See Appendix for the list of variables, definitions and summary statistics.

⁴ The two-digit industry dummies are for the industries of food, drinks and tobacco, clothing and leather, wood and wood products, paper and paper products, chemicals and plastic products, nonmetallic minerals, basic metals, metal products and equipment and miscellaneous industries.

⁵ Of course, cross-sectional estimates such as those presented in Table 1 can be criticized for several reasons (Hanson, 2001). Important to mention is that I did estimate for the effects of a wider range of control variables. First, following e.g. Driffield and De Propris (2006) and Girma et al. (2008), I estimated for the presence of FDI spillovers through forward and backward linkages. The findings suggest the absence of such FDI spillovers, as both variables carry insignificant coefficients. In related work, I do find evidence of significant positive inter-industry FDI spillovers in Mexican industries, but only at the regional level (Jordaan, 2008a, b). Second, I also estimated for productivity effects of the industry level of trade openness and the industry presence of Maquiladora FDI. Again, the estimated effect of these variables is insignificant (Jordaan, 2004). In addition to this issue of specification, the use of cross-sectional industry data calls for caution in the interpretation of the magnitude of the estimated coefficients, as they are likely to be affected by aggregation bias. Accepting this, the prime focus of the present analysis lies in the identification of the bias that may occur from using OLS estimation techniques to identify intra-industry FDI spillovers.

⁶ For example, in 2004 over 20% of the stock of world wide outward FDI was under US control (UNCTAD, 2005).

⁷ See Appendix for data sources.

⁸ The Moulton problem may arise because of the different level of industry aggregation of the variables FOR and US. In particular, several six-digit industries are grouped in the same four-digit industry. If the errors of these six-digit industries are correlated, the estimated SE of the association between FOR and US will be biased downwards. I control for this problem by allowing for clustered six-digit industry SEs at the four-digit level. Also, I have calculated the numerical approximation as introduced in Moulton (1990). Assuming the worst case scenario of perfect correlation between errors of six-digit industries within four-digit industries gives a Moulton-adjusted SE of 0.18. This leads to a t-statistic of 2.33, indicating that even in the worst case scenario the Moulton problem does not affect the significance of the estimated association between FOR and US.

⁹ In more recent years, there has been a notable increase in FDI flows to Mexico from source countries other than the US. However, in 1993 the vast majority of FDI in Mexico was US-owned.

¹⁰ Based on data from www.bea.gov.

¹¹ Based on data from www.bea.gov.

¹² Estimating the regression model with 3SLS produces qualitatively similar results.

¹³ Keller and Yeaple (2005) also find in their estimations of FDI externalities in the US that their IV estimations produce larger externality effects from FDI compared to their OLS estimations.