variation in the FDI variable is exploited to estimate b2 , which suggests that increased foreign
presence has a negative short-run impact on local firms’ productivity. One possible explanation
for this result is that foreign firms reduce the market share of local firms, thereby reducing their
capacity utilization. Another possibility is that foreign firms, by paying higher wages, attract the
best workers, thereby reducing the productivity of local firms.
4.3 Learning by exporting?
The micro-level evidence shows a positive robust correlation between exporting and productivity.
There are two plausible non incompatible explanations for this stylized fact. One is that, as shown
by Melitz (2002) and discussed in Section 2, more efficient firms self-select into export markets.
The other is the learning-by-exporting argument, according to which exporting causes efficiency
gains. Two recent papers address the question of the direction of causality. Bernard and Jensen
(1999) use data relative to U.S. manufacturing plants for the period 1984-1992. They find that
size, wages, productivity and capital intensity are all higher for exporters relative to non-exporters.
They also find clear evidence that good firms become exporters, since performance is higher ex-
ante for exporters. However, they do not find evidence that exporting improves performance, since
productivity and wage growth are not higher ex-post for exporters relative non-exporters.
Clerides, Lach and Tybout (1998) address the same question and reach similar results. They use
plant-level manufacturing data for Colombia (1981-1991), Mexico (1986-1990) and Morocco (1984-
1991). Their approach is based on the idea that, if exporting fosters productivity growth, then
the productivity trajectory of exporting firms should change after they break into foreign markets.
To test this hypothesis, they estimate econometrically the reduced form of a theoretical model
derived from the hysteresis literature (Baldwin, 1989; Dixit, 1989) which explicitly considers two
possible explanations for the correlation between exporting and productivity, namely, self-selection
and learning-by-exporting.
In particular, they estimate, industry by industry, by full information maximum likelihood
27