Estimation procedure

3.1 Estimation equation

To estimate the impact of the presence of foreign firms on local productivity suggested above, we use the following estimation equation based on a Cobb-Douglas production function:

ln Y_it = β_K ln K_it + β_L ln L_it + β_R ln R_it + δ ln FDI_ij,t-1 + β_S SHARE_it + α_i + υ_t + ε_it (1)

where Y_it, K_it, L_it, and R_it are the value added, capital stock, labor, and R&D stock of firm i in industry j at time t, respectively. α_i, υ_t, and ε_it are the firm-specific constant term, the yearspecific constant term, and the error term, respectively.

In the equation, ln FDI_ij,t-1 represents the extent of foreign firms' activities in industry j in year t-1. We take the size of the corresponding coefficient, δ, to represent the extent of knowledge spillovers from MNEs. More specifically, we employ several alternative measures of MNE penetration at the industry level: the amount of total MNE labor force and their educated labor force in the industry, FL_ij,t-1 and FE_ij,t-1, respectively, in logs. We assume that FL_ij,t-1 represents the extent of foreign firm production activities at the industry level,⁷ while by using FE_ij,t-1 we examine whether MNE employment of educated workers is a channel of FDI spillovers.⁸ We use a one-year lag because we assume that there is a time lag from when an MNE engages in production and when its knowledge spills over to domestic firms.

In addition, Haskel, Pereira, and Slaughter (2002) and Keller and Yeaple (2009) argue that the presence of MNEs may have a negative effect on local production, since MNEs grab market share from, and undermine the monopoly power, of domestic firms. To incorporate this effect of MNEs unrelated to spillovers, we include in the equation (1) the market share of firm i in the industry, SHARE_it.⁹ If we had failed to incorporate this variable, then the coefficient on the FDI variable would capture the MNE positive spillover effect as well as the negative effects of their increasing market share, and thus could become negative even in the presence of the positive spillover effect, as Aitken and Harrison (1999) suggest.

3.2 Estimation method

There are two major econometric issues when estimating equations such as production functions: estimation biases, due to the endogeneity of regressors, and unobservable firmspecific effects. In particular, an estimation using OLS may suggest that the extent of FDI has a positive impact on domestic output, when in fact the correlation reflects the fact that industries with a high productivity level attract more FDI.

To correct for these potential problems, we employ the system GMM estimation as detailed in Blundell and Bond (1998) and apply the estimation method to equation (1) and its first difference. Using first differences eliminates firm-specific fixed effects, whereas GMM estimation corrects for endogeneity. In the system GMM, we estimate equation (1) in addition to its first-difference, because instruments are weak if the regressors have very similar unit root properties. More specifically, instruments used for the regressors in the level equation are Δz_i,t-1 and earlier Δz_i where z = lnK, lnL, lnR, SHARE, lnFDI, and Δz_it = z_it - z_i,t-1. Similarly, instruments for the regressors in the first-difference equation are z_i,t-2 where z is defined as above. Since our data set covers a four-year period, we actually use data for the period 2002–2003 as regressors and data for the earlier period as instruments.

We apply two-step estimations of the GMM system to obtain higher levels of efficiency. In addition, we use the methodology of robust standard errors developed by Windmeijer (2005), which are consistent in the presence of any pattern of heteroskedasticity and autocorrelation, and correct for finite sample biases found in the two-step estimations.

Download this Paper [ PDF 194.5KB| 27 pages ].

Comment(s)

There are [0] comment(s) for this entry. Post a comment.