Openness, inequality and poverty: Endowments matter

Abstract

Using tariffs as a measure of openness, this paper finds consistent evidence that the conditional effects of trade liberalization on inequality are correlated with relative factor endowments. Trade liberalization, measured by changes in tariff revenues, is associated with increases in inequality in countries well-endowed with highly skilled workers and capital or with workers that have very low education levels. Similar, although less robust, results are also obtained when decile data are used instead of the usual Gini coefficients. Taken together, the results are strongly supportive of the factor-proportions theory of trade and suggest that trade liberalization in poor countries where the share of the labor force with little education is high raises inequality. Simulation results also suggest that relatively small changes in inequality as measured by aggregate measures of inequality, like the Gini coefficient, are magnified when estimates are carried out using decile data.

Keywords:

• international trade
• income distribution
• poverty

JEL Classification:

• F11
• F16
• D3

Acknowledgements

The authors thank Jean-Marie Baland, Olivier Cadot, Marco Fugazza, Sylviane Guillaumont, Sébastien Jean, Jaya Krishnakumar, Alessandro Nicita, Branko Milanovic, Marcelo Olarreaga, Mathias Thoenig, Jeffrey Williamson, Adrian Wood and seminar participants at UNCTAD (Geneva), T2M conference (Paris – PSE), the Universities of Milan and Namur for very helpful comments on an earlier draft. They also thank Branko Milanovic for providing them with his WYD data set and for answers to numerous queries. Annexes with detailed data sources and robustness checks are available from the authors upon request at http://www.unige.ch/ses/ecopo/demelo/Jaime.html

Notes

1. Perhaps this should not come as a surprise and not concern us too much if, via other channels, such as growth, increased openness reduces poverty. After all, HO theory should only be expected to inform us about the relation between endowments and factor rewards in response to a reform-induced change in relative factor demands rather than between endowments and overall income inequality, which is determined by many other factors. And, as pointed out by Baldwin (2004, 517) in his review of the trade liberalization and growth literature, since trade liberalization is rarely applied in isolation, it makes little sense to try to isolate its effects from those of associated policies. Anderson (2005) gives a survey of the conflicting evidence on openness and inequality, and Winters et al. (2004) survey the evidence between trade liberalization and poverty. Spilimbergo et al. (1999), Milanovic (2005a) and Bensidoun et al. (2005) are the studies most closely related to ours.

2. Let superscripts 1 and 0 stand for the change in factor prices associated with a change in trade policy. Then if technological choices are not affected by the change in trade policy: (w ¹ − w⁰) (V^C − s^C V^W ) ≥ 0 where s^C is country c's share in world (i.e. sample) income, and V^w is the sample average endowment. In the more general setting with no restrictions on the number of goods or factors, or on the rest-of-the-world, all that one can establish from profit maximization is a positive link between changes in prices and net exports and if technology choice is insensitive to price changes, then this correlation extends to factor price changes (see Ethier 1980 and Bensidoun et al. 2005).

3. Among the studies that control for heterogeneity, Edwards (1997) finds no evidence that openness or trade liberalization increases inequality. When including fixed effects, Barro (2000) finds no correlation between inequality and openness, echoing Ravallion's (2001) results.

4. To correct for heteroskedasticity, we used two estimators: the standard heteroskedasticity-consistent White (1980) estimator and the panel-corrected standard errors (PCSEs) estimator proposed by Beck and Katz (1995) which is shown to be as good or slightly superior to the robust estimator in Monte Carlo studies for small samples (see Beck and Katz 1996, Table 2). Hence we only report the PCSEs results in subsequent tables. Ethnicity is dropped from the FE estimates because it is time-invariant.

5. As a first exercise, not reported here, we replicated the same specification as Spilimbergo et al. (1999) confirming their results (i.e. a result in conformity with factor-endowment predictions for human capital but in contradiction with predictions for physical capital when using their openness variable (‘adjusted’ trade ratio instead of tariffs). However, when using our preferred measure tariffs, increases in inequality are associated with relatively abundant endowments in capital following a reduction in tariffs (i.e. the coefficient on the interaction between relative endowment in capital per unit of labor, K/L, and the lagged tariff, is negative). To our knowledge, this plausible set of results has not been found in previous studies. However, with tariffs, the significance for the human capital endowment interactive term with tariffs disappears.

6. Bensidoun et al. (2005) argue that the Heckscher–Ohlin model is too restrictive, relying on factor-price-equalization (FPE) and hence identical production techniques in equilibrium. Using a more general approach that relaxes the FPE assumption (but still relies on other restrictive assumptions, like homothetic preferences and unchanged production techniques following trade liberalization), they show that factor price changes are correlated with the factor content of trade, leading them to test their model using constructed estimates of the labor-capital content of net exports instead of factor endowments on a similar D-S inequality data set for 53 countries. However, their results are not strictly comparable with ours (different sample with no SSA countries and a different definition of variables).

7. The index of human capital endowment (average years of schooling) is now replaced by these three different categories of skill levels. We take the NO, BS and SK variables from the Barro and Lee (2000) data set, which is available on a five-year basis that corresponds to the 5-year averages used for all our variables.

8. Owing to the high correlation between SK/L and K/L (ρ = 0.84), the coefficient on K/L changes sign and is almost significant statistically.

9. WYD can be downloaded from http://econ.worldbank.org/projects/inequality (accessed May 30, 2008).

10. We also ran the same regression without taking the logarithm of the LHS-variables, obtaining similar results. Regarding reverse causality, as previously, we ran the same regression using future trade rather than past values and the results become mostly insignificant, suggesting that reverse causality should not be a problem here (results are available upon request).