(iv) As for the impact of the external conditions, the results
suggest that their direct effect is significant but not large though, as
noted in section 3, their impact might operate via the growth of
GDP. The international terms of trade reduce inequality in a
significant but moderate way (an increase of 100 points in the
related index would reduce inequality by 1.5 Gini points);
remittances appear to raise inequality (as suggested in section 3) but
at a borderline levels of significance; portfolio flows are not
statistically significant, possibly also because of errors of
measurement of this variable. In turn, the FDI stock/GDP appears
to increase inequality in a statistically significant but limited way.
For instance, a doubling of the FDI/GDP ratio from the current
regional average of 20% to 40% for the region as a whole would
increase Gini by 1.2 points, though the effect would be higher, for
instance, in FDI dependent Andean countries. (v) As for the impact
of macroeconomic policy, the results suggest that, as argued in
section 3, a competitive exchange rate affects inequality in a convex
way. Inequality at first falls, then rises beyond a given threshold
requiring excessive nominal devaluations. As for the income and
redistributive policies, the results suggest that the minimum wage
(interacted with the share of formal sector workers) reduces
inequality, but at a borderline level of significance. More
significantly, the ratio of direct to indirect taxes indicates that the
changes in the structure of revenue collection during 2003-07
(Table 5) generated a favorable distributive effect. In turn, social
security expenditure/GDP has a clear and statistically significant
impact on inequality (doubling such expenditure from 10 to 20% of
GDP would reduce inequality by 3.1 points), and the impact would
likely be larger if social assistance could be factored out. In contrast
to ex-ante expectations, the ratio of pension coverage of the top to
the bottom quintile is not significant. This is possibly because it
correlates closely with the share of social security/GDP (see Table
11), or because this variable exhibited little variation over time in
many countries.