Stochastic Frontiers using a Fixed-effect Vector Decomposition Approach with an Application to ICT and Regional Productivity in Spain

Fixed effects estimators will generally produce imprecise estimates when the data contains variables with relatively low within variance. In a production setting, this can lead to unreliable estimates of key parameters. These problems carry over to production frontiers where technical inefficiency is estimated on the basis of the unit effects (Schmidt and Sickles, 1984) or through the use of true FE stochastic frontier (Greene, 2004; 2005). Plümper and Troeger (2007) propose what they label a Fixed Effect Vector Decomposition (FEVD) estimator which may permit greater precision in estimating variables with low within variance and they provide conditions under which their estimator performs better than the fixed effects estimator. In this paper we extend the FEVD estimator to a frontier setting. This allows the possible advantages of the FEVD estimator to be incorporated into Greene’s true FE frontier model. In an empirical application of the estimator we analyze the effects of ICT capital on regional productivity in Spain using a balanced panel dataset at provincial level over the period 1986-2006. This data contains several variables with relatively low within variance, and the FEVD frontier provides much more plausible estimates of key parameters than the FE estimator or the true FE frontier estimator.