Trade, foreign direct investments (FDI) and income inequality: Empirical evidence from transition countries

Abstract

The aim of the article is to verify whether trade and inward foreign direct investments (FDI) may affect income distribution in a sample of 17 Transition Countries (TCs) over the period 1990–2006. In line with most of the previous literature, FDI do not have significant effects on income inequalities, whereas trade, especially when occurs with developed countries, seems to be more relevant. Different results are found when we take into consideration the educational system which represents an important channel through which FDI and trade may affect inequality.

Keywords:

• FDI
• income inequality
• trade
• transition countries

JEL Classifications:

• F10; F23; O15; P20

Acknowledgments

The authors gratefully acknowledge useful comments and suggestions from the participants to the ETSG 2010 Lausanne 12th Annual Conference (9–11 September 2010), David Aristei, Giovanni Cerulli, Francesco Venturini and three anonymous referees on a previous version of this article. Chiara Franco gratefully acknowledges financial support by the Autonomous Province of Trento, as the sponsor of the OPENLOC research project.

Notes

1. The null hypothesis is always rejected at 1% level of significance.

2. This test is run after having estimated the model with fixed effects.

3. We always reject the null hypothesis of unit root at 1% level of significance.

4. The signs and significance of the other openness variables remain the same except for the case of IMPED and IMPEDSQ that are not significant anymore. The table summarizing these results is available upon request.

5. The reliability of the SYS-GMM estimator depends on the validity of the instruments used in the regression. In order to check it, we consider two specification tests. The first, the Sargan test on over-identifying restrictions, is based on the analysis of the moment conditions in the estimation process. Under the null hypothesis instruments are uncorrelated with the error term. Due to the fact that the Sargan test may not be reliable when the number of instruments exceeds the number of regressors, instruments are collapsed and we limit the use of lags (from the second to the third) for those variables which are used as instruments (Roodman 2009).

 In some specifications the Sargan test rejects the validity of instruments at 1% level of significance. In these cases, we lag the instruments one year limiting their use to (3 4).

 In the second test under the null hypothesis the error term e_it is not serially correlated. We do not reject the absence of correlation at first and second order. Taken together, these two specification tests support the use of the GMM estimator.

6. From pairwise correlations we can see that values are never higher than 0.7%, indicating no serious problems of collinearity. However, as the interacted variables may present such problems, we estimate our specifications by orthogonalizing interacted variables using the Gram-Schmidt procedure .We find that the results, both in terms of size and significance of coefficients, are similar to the original ones, thus confirming the reliability of the estimates presented.

7. TIER has a mean of 38.87% and standard deviation of 17.05.