Some results on consistency of LS estimates (Q1344608)

Consider the linear regression model \[ y_i= x_i' \beta+ e_i, \qquad 1\leq i\leq n, \quad n\geq 1; \] \(\widetilde {x}\equiv (x_1, x_2, \dots)\) and \(e\equiv (e_1, e_2, \dots)\) are sequences of known \(p\)-vectors and random errors, respectively; \(\beta= (\beta_1, \dots, \beta_p)'\) is the unknown vector of regression coefficients. Write \(S_n= x_1 x_1'+ \cdots +x_n x_n'\) and assume that \(S_n^{-1}\) exists when \(n\geq n_0\). Denote by \(u_{nij}\) the \((j,i)\)-element of the \(p\times n\) matrix \(S_n^{-1} (x_1 \vdots \cdots \vdots x_n)\). Then the LS estimate of \(\beta\) is \(\widehat {\beta}_n= (\widehat {\beta}_{n1}, \dots, \widehat {\beta}_{np})\), where \[ \widehat {\beta}_{nj}= \sum_{i=1}^n u_{nji} Y_i= \beta_j+ \sum_{i=1}^n u_{nji} e_i. \] Therefore, consistency of \(\widehat {\beta}_{nj}\) in some sense \(C\) is equivalent to the convergence of \(\sum_{i=1}^n u_{nji} e_i\) to zero in the same sense \(C\); such as weak consistency \(W\), strong consistency \(S\), and consistency in the \(r\)th mean \(M_r\).