5. Estimation methodology
As mentioned above, official data on FDI inflows became only available after August 1994. Consequently the analysis will concentrate on the period 1995–2005. Taking into account the relatively short time period covered by these FDI data, it is not appropriate to use time series analysis for the estimations. Cross-sectional estimations are also known to be inefficient as only seventeen and fifteen home countries for approved FDI and realized FDI, respectively, are available. It was therefore decided to opt for a panel data set in the estimations of Eq. (2).
Panel data sets (Hsiao, 2003, 2005; Plasmans, 2006) allow for several advantages including the use of three estimation procedures, i.e.: pooled OLS, fixed-effects (FE), and random effects (RE) estimations. If the assumption holds that the unobservable individual country-specific effects are not very different, pooled OLS estimations are the most simple and efficient method. The FE estimations allow for the unobservable country heterogeneity. However, the use of a fixed-effects model will ‘‘kill’’ the time-invariant variable DIST which is considered to be an important factor influencing FDI, and will make FE estimations less efficient than the RE estimation counterpart. Like the FE model, RE estimations take into consideration the unobservable country heterogeneity effects, but incorporate these effects into the error terms, which are assumed to be
uncorrelated with the explanatory variables. One model against another will be tested using appropriate testing techniques for a panel data set.