Orthogonal to backward mean transformation for dynamic panel data models

The within-groups estimator is inconsistent in dynamic panels with fixed T as the individual sample mean of the lagged dependent variable used in the within transformation is contemporaneously correlated with the idiosyncratic error term. This paper suggests transforming the lagged dependent variable into orthogonal deviations from its individual backward mean, which is contemporaneously uncorrelated with the idiosyncratic error term. As this transformation eliminates the individual effects as but not for T fixed, this alternative estimator is consistent for but inconsistent for and T fixed. The inconsistency for fixed T is shown to be negligibly small, though. Moreover, a Monte Carlo simulation shows that overall, it has superior small sample properties compared to other dynamic panel data estimators