Answers are grounded in this paper and cite their sources.
In the linear model y=Xβ+u, the error vector u is not observable. Following Theil [8] we approximate J'u, a vector of selected components of u, by A'y with A chosen so that ϵ(A'y) = 0 and U(A'y) = σ2I. We study Σa = U(A'y—J'u) and give a new proof that fr Σa is minimized at A = B, where B'y is the vector of BLUS residuals. The claim in [9] that Σa — Σb is positive semi-definite is shown not to hold in general; we prove, however, that the BLUS residuals minimize each individual characteristic root of Σa.
No takes yet. Share an insight, caveat, or question.
Grossman et al. (1972) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: