A one-stage random effect counterpart of the fixed-effect vector decomposition model with anapplication to UK electricity distribution utilities

Abstract

To deal with the presence of slowly changing variables, Plümper and Troeger (2007) proposed a fixed effects vector decomposition (FEVD) estimator, which is a three-stage procedure based on the fixed effects estimator. We show that this estimator moves between fixed effect and ordinary least squares depending on how its second stage is specified. This provides an alternative interpretation of the FEVD estimator that allows us to introduce a one-stage counterpart of the FEVD model, which can be viewed as a partial Mundlak (1978) transformation of the random effects model. We illustrate our approach with an application to UK electricity distribution utilities using the same data as Yu et al. (2009). Our application suggests that our estimator and the FEVD yield similar results when the pooled OLS estimates for a given panel are similar to the random effects estimates.