Determinants of U.S. Textile and Apparel Import Trade

potentially create correlation problems among individual variables and the covariates. This can
be corrected by specifying the model to capture the differences in behavior over time and space.
Conceptually, the difference in the nature of individual effects can be classified into the fixed
effects which assume each country differs in its intercept term; and the random effects which
assume that the individual effects can be captured by the difference in the error term.

The Hausman test was run to check if the fixed or random effects model is more efficient.
We use the Hausman’s (p. 1261) notation where equation 9 in the time series and cross-section
framework is written as:

Xijt = Zijt β + μij + μijt                                            (10)

where

X_ij_t = trade observation from country i to j at time t (t = 1,...,T);

Zijt = a corresponding trade determinant vector;

μij = the trade effect associated with a country pair; and
μijt = the error term.

Equation 10 has the main advantage of allowing different effects of Zijt on Xijt for each
country pair to be captured. By assuming individual effects, we proceeded to test if μij is fixed or
random. Hausman’s essential result is that “the covariance of an efficient estimator with its
difference from an efficient estimator is zero” (Greene, 1990). Results indicate a Hausman m-
statistic of 22.81 and 15.20 for the specified models for textiles and apparel imports,
respectively, at a χ² value of 15.09 at the 1% level and 5 degrees of freedom. Thus, we reject the
assumptions of orthogonality between μij and right-hand side variables in favor of the existence
of individual country fixed effects, which supports covariance specification of the models. The
covariance matrix is estimated by a two-stage procedure leading to the estimation of model
regression parameters by General Least Square (GLS) approach. The covariance model

12

<<   10   11   12   13   14   15   16   >>

More intriguing information