Efficient Estimation of Rarely Changing Variables in Fixed Effects Models

The estimation of slowly and rarely changing variables in panel data with unobserved unit effects suffers from inefficiency of the fixed effects estimator. Point estimates lack reliability when the within variance (the variation across time) remains small. The estimate's efficiency can be enhanced if the between variance (the variation across space) increases. However, this comes at the cost of increasing potential omitted variable bias. We describe a three-stage estimator (called xtfevd) that allows to maintain the between variance of cross-sectionally dominant variables while estimating the variables with sufficient within variation by fixed effects. Monte Carlo simulations show that this procedure performs better than the fixed effects model in cases where the between variance exceeds the within variance by at least factor 2.5. Thus, xtfevd outperforms the standard fixed effects model if the within variance is small and the between variance significantly larger