Iterative estimation correcting for error autocorrelation in short panels

This paper presents an iterative estimation procedure that estimates and corrects for serial correlation of the disturbances in short panels. Controlling for error autocorrelation is a prerequisite for consistent estimation in mod- els with lagged dependent variables. In addition it allows to discern between dierent behavioural mechanisms underlying state persistence. The basic phi- losophy of iterative estimation is to assume some information on the basis of which the parameters of the postulated structural model are easily estimated. These estimates subsequently allow to update the assumed information and the complete cycle is repeated until convergence. The unobserved or latent variables considered here are the residuals from previous periods. While the main result is valid for models that allow for the explicit cal- culation of Cox and Snell's (1968) generally dened residuals, which in turn are allowed to exhibit a very general temporal dependence structure, attention is subsequently restricted to AR correlated disturbances, since an MA error process would require strong assumptions on the initial values for consistency when N ! 1, with T xed. The method is nally applied to short panel data models with xed eects and lagged dependent variables as well.