Bias-Corrected Estimation for Spatial

Maximum likelihood (ML) or quasi-maximum likelihood (QML) estimator of the spatial parameter in the spatial autoregressive model can be very biased. The biasness depends heavily on the magnitude of the error standard deviation and on the spatial layout. Second-order approximations to the bias and MSE are given by modifying the results of Bao and Ullah (2007a). A bootstrap procedure is introduced for the practical implementation of the bias-correction. Monte Carlo simulation shows that this bootstrap procedure works well and that the bias-corrected ML or QML estimator outperforms the regular ML or QML estimator