ON SIZE AND POWER OF HETEROSKEDASTICITY AND AUTOCORRELATION ROBUST TESTS

摘要: Testing restrictions on regression coefficients in linear models often requires correcting the conventional F-test for potential heteroskedasticity or autocorrelation amongst disturbances, leading to so-called and robust test procedures. These procedures have been developed with purpose of attenuating size distortions power deficiencies present uncorrected F-test. We develop a general theory establish positive as well negative finite-sample results concerning properties large class tests. Using these we show that nonparametrically parametrically corrected F-type tests time series stationary disturbances either equal one nuisance-infimal zero under very weak assumptions covariance model generic conditions design matrix. In addition suggest an adjustment procedure based artificial regressors. This resolves problem many cases so-adjusted do not suffer from distortions. At same their function is bounded away zero. As second application discuss case heteroskedastic disturbances.