Factor Residuals in SUR Regressions: Estimating Panels Allowing for Cross Sectional Correlation

This paper describes a method for estimating panels by imposing a factor structure on the residuals. The method allows SUR estimation of panel models by providing a full-rank estimator of the system covariance matrix when the usual estimate is rank-deficient. We charactersie completely the circumstances when this is possible. When the usual estimator is of full rank, our procedure provides a more parsimonious representation of the covariance matrix, which can lead to efficiency gains in finite samples. Monte Carlo analysis of convergence regressions and PPP regressions in the Heston-Summers data-set indicates that the proposed estimator has better performance in terms of RMSE and bias than standard panel or SUR estimators (where available), as well as offering unbiased inference.