Title: Size and power properties of autocorrelation and heteroskedasticity robust tests in spatial error models
Authors: Christian Zwatz - University of Vienna (Austria) [presenting]
Abstract: A typical approach for testing linear hypotheses on the regression parameters in regression models with autocorrelated and/or heteroskedastic disturbances is to modify the conventional F-test statistic by using a heteroskedasticity and autocorrelation consistent (HAC) estimator for the covariance matrix. These are nonparametric estimators designed to take the heteroskedasticity and autocorrelation in the data into account. We consider heteroskedasticity and autocorrelation robust testing in spatial error models, i.e. models where the disturbances follow a spatial autoregressive or spatial moving average process. It is well known that tests based on HAC estimators in the case of time series regression models suffer from substantial size and power problems. Based on a general theory about size and power properties of tests in regression models with autocorrelated and/or heteroskedastic disturbances, we show that similar problems also occur in the spatial error model. In particular, we give conditions under which the size of the resulting test is in fact one. We also give conditions under which the size of the test can be controlled by an appropriate choice of critical value.