Neglected dynamics and spatial dependence on panel data: consequences for convergence of the usual static model estimators

Abstract

ABSTRACTNeglected dynamics and spatial dependence on panel data: consequences for convergence of the usual static model estimators. Spatial Economic Analysis. This article assumes that the estimated model is static, whereas the true model is an autoregressive distributed lag error components model including spatial dependence on the disturbances. We derive the probability limits of the ordinary least squares (OLS), Between and Within estimators of a static model on panel data. The results show that asymptotic biases of these estimators vary mainly according to the degree of spatial autocorrelation, the unobservable individual heterogeneity and the data generating process (DGP) of the explanatory variable. The probability limit of the Between estimator converges to the long-run effects, particularly when the individual unobserved heterogeneity is large, when compared with the idiosyncratic error term, and/or there is a strong time dynamic in the regressors. The interpretation is more contrasted for the OLS and Within estimators. The former can be associated under very restrictive conditions with the long-run effects; the latter with the short-run effects.