dataset, the OLS procedure tends to yield inefficient and distorted
estimates of the values of and Baltagi. The estimation procedure
best suited to situations in which ui varies from country to country
is the fixed effects (FE) model in which ui is not treated as a
random variable. This means that the estimation with the fixed
effects model includes, for each of the 18 countries considered, an
intercept which captures specific country effects due to geography,
institutions and unobservables. Besides fixed-country effects, the
estimation procedure has included also year fixed-effects so as to
capture the impact of yearly shocks. The F test of joint significance
confirm at zero probability level that both country and year fixed-
effects are different from zero. This indicates that their exclusion
from the regression would bias the estimates of the other
parameters.
The results of the regression analysis are presented in Table 13 in a
basic model (column 1) and in two subsequent models where
portfolio inflows/GDP (column 2) and the latter plus the ratio of
the coverage of pensions in the top to the bottom quintile (column
- were added. In the basic model, practically all variables have the
sign expected ex-ante on the basis of the received theory reviewed in
section 3. The addition of portfolio flows and pension has a
minimal effect on the value of the parameters of model 1.
We turn now to the impact of the five sets of variables discussed
before the regression. (i) Initial conditions: in the fixed effects
approach, the time-invariant Gini income 1990 is absorbed in the
country-specific constant term (but its effect is strong, in contrast,
when using the OLS or random effect estimators, not shown here
for reasons of space). (ii) The growth rate of GDP/c (which
measures the impact of the business cycle) has a strong effect on
inequality, falling by a quarter of a Gini point for every one percent
in GDP/c growth. The recovery of 2003-07 appears therefore to
have had an important equalizing effect. (iii) The Gini of the
distribution of the years of education among members of the labor
force (delayed one year) is, as expected, strongly significant,
suggesting that improved access to secondary education had an
important, if slow-moving, effect on the decline of income
inequality.