No label defined (Q4940288)

The authors consider the linear functional errors-in-variables model \[ (1)\qquad y_{i}=\xi_{i}\beta_0+\delta_{i},\quad x_{i}=\xi_{i}+\varepsilon_{i}, \] where the vectors \(y_{i}\) and \(x_{i}\) are observed, \(i=1,2,\ldots,n.\) The error terms are \(\delta_{i}\) and \(\varepsilon_{i}\) with the expected values zero. The \(k\)-dimensional row vectors \(\xi_{i}\) are the nonrandom nuisance parameters. The \(k\times r\)-dimensional matrix \(\beta_0\) is the true value of the parameter \(\beta\) to be estimated. The \(r+k\)-dimensional row vectors \(z_{i}\) of observations are of the form \(z_{i}= (y_{i},x_{i})=\xi_{i}(\beta_0,I_{k})+E_{i}\), with \(E_{i}=(\delta_{i},\varepsilon_{i})\), where \(I_{k}\) denotes the \(k\times k\) unit matrix. For each observation the components of \(E_{i}\) are correlated: \[ {\text{cov}}(E_{i})=\Gamma=\begin{pmatrix} \Gamma_{11}&\Gamma_{12}\\ \Gamma_{21}&\Gamma_{22}\end{pmatrix}. \] There exists a weak dependence between \(E_{i}\) and \(E_{i+m}.\) This dependence is described by mixing conditions. The main result of this paper states that the situation in the errors-in-variables model differs from the situation in ordinary regression models. If \(k\geq r\), then consistency of the weighted least squares estimator holds only in the case where the weight matrix is proportional to \(\Gamma^{-1}.\) The other result of the paper is the consistency of the properly weighted least squares estimator for a linear functional model with dependent errors.