We consider estimation of a panel data model where disturbances are spatially correlated in the cross-sectional dimension, based on geographic or economic proximity. When the time dimension of the data is large, spatial correlation parameters may be consistently estimated. When the time dimension is small (the usual panel data case), we develop an estimator that extends the cross-sectional model of Kelejian and Prucha. This approach is applied in a stochastic frontier framework to a panel of Indonesian rice farms where spatial correlations represent productivity shock spillovers, based on geographic proximity and weather. These spillovers affect farm-level efficiency estimation and ranking. Copyright 2004, Oxford University Press.

In transition economies, there may be a significant mismatch between the types of skills that workers possess and the types of skills that the new economy demands. We consider this problem of human capital mismatch along the dimensions of training type (holding the level) and occupation. We document that in the Czech Republic and Poland the wage rate grew faster in business occupations than in technical occupations in the 1990's, and that in response the technical training/occupations contracted while the business training/occupations expanded. We do not find this pattern in Hungary. We construct a neoclassical model with endogenous occupational choice and calibrate it to the Czech and Polish data. We estimate that the discounted sum of output loss due to human capital mismatch amounts to 44% of the aggregate output of the beginning year of transition.

This paper considers estimation of a panel data model with disturbances that are autocorrelated across cross-sectional units. It is assumed that the disturbances are spatially correlated, based on some geographic or economic proximity measure. If the time dimension of the data is large, feasible and efficient estimation proceeds by using the time dimension to estimate spatial dependence parameters. For the case where the time dimension is small (the usual panel data case), we develop a generalized moments estimation approach that is a straightforward generalization of a cross-sectional model due to Kelejian and Prucha. We apply this approach in a stochastic frontier framework to a panel of Indonesian rice farms where spatial correlations are based on geographic proximity, altitude and weather. The correlations represent productivity shock spillovers across the rice farms in different villages on the island of Java. Test statistics indicate that productivity shock spillovers may exist in this (and perhaps other) data sets, and that these spillovers have effects on technical efficiency estimation and ranking.