Bias Reduction in Estimating Long-run Relationships from Dynamic Heterogeneous Panels

This paper considers the small sample properties of the mean group estimator of the long-run coefficients in dynamic heterogeneous panels, and using Monte Carlo techniques examines the effectiveness of a number of alternative bias-correction procedures in reducing the small sample bias of these estimators. Four different biascorrected estimators of the long-run coefficients are considered. A “naïve” procedure which attempts to bias-correct the estimator of the long-run coefficients by using the bias-corrected estimators of the short-run coefficients proposed by Kiviet and Phillips (1993, Econometric Theory). Two variations of a direct approach which derives biascorrections of the estimators of the long-run coefficients allowing for the variance and covariances of the short-run coefficients, referred to as DBC1 and DBC2, and a bootstrap bias-correction procedure. The “naïve” bias-corrected estimator fails in all cases, and the bootstrap method performs poorly in cases where the true coefficient of the lagged dependent variable is relatively large. Both of the direct biased corrected estimators perform reasonably well, although only the DBC1 estimator which allows for some higher order bias correction terms out-performs the bootstrap method. None of the estimators seem to be effective when the coefficient of the lagged dependent variable is around 0.8. JEL Classifications: C13, C15, C23