4 A General-to-Specific Analysis of FDI Flows
One of the main drawbacks in the FDI literature has been the lack of a coherent and
generally accepted theoretical framework to think about FDI and to form the basis for
empirical analysis. The theoretical vacuum has resulted in an ad hoc selection of FDI
determinants, which complicates direct comparisons across studies. To take an example,
all empirical papers have included some measure of market potential where GDP, GDP
per capita, population or GDP growth are the most commonly used proxies, and valid
theoretical arguments can be put forward for each of them. Which one should we pick?
To what extent is it appropriate to pick the same proxy irrespective of regional belonging?
And when can we expect one variable to have the same impact on FDI irrespective of
regional belonging?
Since potential explanatory variables are highly correlated, it is a challenge to select
several or all of them while avoiding multicollinearity in the model. We therefore use
a general-to-specific model selection approach, which enables us to "test down" among
the large set of explanatory variables. We use the PcGets software, which automatically
selects an undominated, congruent model where statistically insignificant variables are
eliminated and where diagnostic tests check the validity of reductions to ensure a con-
gruent final selection. Equation mis-specification tests include residual autocorrelation,
ARCH, heteroscedasticity, functional form mis-specification, and non-normality. The path
is terminated when all the variables that remain are significant, or a diagnostic test fails.
In some cases insignificant variables are therefore retained. We refer to Hendry (1995,
Chapter 9) for further details on this data-based model selection methodology.
Based on the empirical papers reviewed in Tables 3 and 4, we have collected data on
19 return proxies and 14 risk measures to enter the general-to-specific analysis along with
regional dummy variables for Africa, Asia and Latin America. Data is calculated as an
average over the time period 1980-2004 for a cross-section of 100 developing countries
(43 belonging to Africa, 35 located in Asia and 22 Latin American countries).7 A list of
countries can be found in Appendix. Details on the data are given in Table 5.
4.1 Empirical Findings
Table 6 reports the main results. One of the most important findings is that none of
the regional variables turn out significant, which suggests that regional differences in FDI
inflows can be fully explained by structural characteristics. This means that there is no
African bias (see Asiedu, 2002, among others). Also, we see that growth and inflation are
the only two variables that turn out significant in all specifications although their marginal
7Using averages over 25 years and thus eliminating the time dimension, the cross-sectional approach
allows us to look for deep structural determinants of FDI. The disadvantage is that in some circumstances
our results will not be directly comparable to the panel studies reviewed in the previous section. For
example, it will not be possible to test for agglomeration effects by including a lagged dependent variable.