capital stock (Kit-j) and in past realization of marginal costs, we should observe a negative
correlation between export experience and marginal costs. FIML estimation results show, however,
that the coefficients γjy on lagged export experience are generally insignificant. In some cases they
are significant, but with the wrong (positive) sign. Only in a few cases these coefficients are
negative and significant, e.g., in the Moroccan apparel and leather industries.15
In short, these results do not support the learning-by-exporting hypothesis. There are several
reasons, however, to be cautious in interpreting the results. In particular, since the time span
covered by the data is very short, the econometric analysis can only pick up gains in efficiency
which materialize in the short-run (within three years). Further, even if these gains materialize
immediately (which is quite unlikely, given that learning is a gradual process), in the short-run
they can be offset by the sunk entry costs associated with becoming an exporter. Indeed, sunk
entry costs may contribute to explain the positive and significant correlation between exporting
and marginal costs found by the authors in some cases. Hence, this evidence simply suggests that
becoming an exporter does not generate short-run efficiency gains.16
5 Trade, Technology and Wage Inequality in Developing
Countries
According to the traditional trade theory, trade liberalization should pose no serious distributional
issues in developing countries. The reason is that, since developing countries are high-skilled labor
scarce relative to industrial countries, their trade-induced specialization in low-skilled intensive ac-
tivities should increase the relative demand for low-skilled labor, thereby reducing wage inequality
in these countries. This prediction has often been used to argue in favor of trade liberalization
in developing countries, since it would both increase efficiency and lower wage dispersion in these
15FIML estimation results of equation (20) also show that firms with a larger capital stock have lower marginal
costs, and that marginal costs tend to follow a second order autoregressive process.
16 Also, as correctly noted by the authors, their approach does not allow to detect efficiency gains accruing to
workers in the form of higher wages, but that leave average variable costs unchanged.
29