affected by the cyclical behavior of the economy in the aftermath of the reform. Note, further, that
in some episodes of trade liberalization, a deep economic downturn is often the trigger of these
reforms, and hence productivity gains from liberalization can be underestimated if a prolonged
recession leads to reduced capacity utilization.
2) Simultaneity bias.Assumethatplanti’s technology is described by the following Cobb-
Douglas production function:
J
yit = β0 + βjxjt + eit; eit = ωit + eit (12)
j=1
where all variables are in logarithms, yit is output and xijt is the jth input. The error term eit
is composed of a stochastic disturbance eit plus an unobserved plant-specific efficiency term ωit.
Note that, since more productive plants are willing to hire more inputs, the error term eit is
positively correlated with factor inputs. This implies that OLS estimates (or between estimates,
as in Tybout and Westbrook, 1995) of the production function coefficients are biased upward, thus
involving biased estimates of ωit.7
In some cases (e.g., in Harrison, 1994) this problem has been tackled by assuming that the
plant-specific efficiency term is time-invariant, which allows to estimate equation (12) using a
fixed effects estimator. This approach only removes the bias originating from the time-invariant
component of plant-specific efficiency, so it does not solve the problem completely. What is more
worrisome, however, is that this approach, by treating plant-specific efficiency as time-invariant,
also removes the possibility to measure how it evolves after trade reform.
Hence, in general, the simultaneity problem is either neglected or tackled improperly in the
literature.
3) Self-selection bias. The literature generally neglects the self-selection bias induced by plant
closing. Pavcnik (2002) shows that, under certain conditions, a negative correlation is to be
7 Estimates of ωit are in fact based on the difference between actual output and output predicted from estimates
of the production function coefficients.
16