Linear unbiased approximators of the disturbances in the standard linear model (Q1059962)

The least squares residuals from the standard linear model have a variance matrix which is a function of the \(n\times q\) matrix of observations on the regressors. We examine two classes of residuals which do not suffer from this defect. Our first class of residuals (LUZ residuals) has a variance matrix which is a scalar multiple of an \(n\times n\) idempotent matrix of rank n-q specified by the user, and our second class of residuals (LUS residuals) has a variance matrix which is a scalar multiple of the (n-q)\(\times (n-q)\) identity matrix.