Time invariant variables and panel data models: A generalised Frisch-Waugh theorem and its implications

Mundlak ("On the pooling of time series and cross-section data", Econometrica, Vol. 46 (1978), pp. 69-85) showed that when individual effects are correlated with the explanatory variables in an error component (EC) model, the GLS estimator is given by the within. In this paper we bring out some additional interesting properties of the within estimator in Mundlak's model and go on to show that the within estimator remains valid in an extended EC model with time invariant variables and correlated specific effects. Adding an auxiliary regression to take account of possible correlation between the explanatory variables and the individual effects, we find that not only the elegant results obtained by Mundlak but also the above mentioned special features carry over to the extended case with interesting interpretations. We obtain these results using a generalised version of the Frisch-Waugh theorem, stated and proved in the paper Finally, for both the EC models with and without time invariant variables we have shown that the estimates of the coefficients of the auxiliary variables can also be arrived at by following a two-step procedure.