The BLUE in continuous-time regression models with correlated errors (Q2313274)

The authors deal with the problem of finding the best linear unbiased estimate (BLUE) for continuous-time regression models of the form \[y(t)=\theta^{\top}f(t)+\epsilon(t),\quad t\in\mathcal{T}\subset [A,B],\] where \(\theta\in\mathbb R^m\) is a vector of unknown parameters, \(f(t) = (f_1(t),\dots,f_m(t))^{\top}\) is a vector of linearly independent functions on \(\mathcal{T}\), and \(\epsilon(t),t\in [A,B]\) is a random error process with \({\mathbb E}[\epsilon(t) ]= 0,t\in [A, B]\) and covariances \({\mathbb E}[\epsilon(t)\epsilon(s)]=K(t,s)\). It is assumed that \(\epsilon(t),t\in [A,B]\) has continuous (in the mean-square sense) derivatives \(\epsilon^{(i)}(t),i= 0,1,\dots, q\) and \(\mathcal{T}\) is the set where the observations of \(y (t)\) and derivatives of \(y(t)\) are available. An excellent survey of classical results on the BLUE is given in the book by \textit{W. Näther}, [Effective observation of random fields. Leipzig: BSB B. G. Teubner Verlagsgesellschaft (1985; Zbl 0612.62133)]. For references of the study of relations between solutions of the Wiener-Hopf equations and the BLUE through the RKHS theory see [\textit{K. Ritter}, Average-case analysis of numerical problems. Berlin: Springer (2000; Zbl 0949.65146)]. The authors of this article investigate properties of the general linear estimator of the parameter \(\theta\) in the continuous-time regression model of the form \[ \hat{\theta}_G=\int_\mathcal{T}G(dt)Y(t)= \sum_{i=0}^q\int_\mathcal{T}y^{(i)}G_i(dt), \] where \(Y(t) = \{(y(t),\dots, y^{(q)}(t))^{\top}\}\) is the observation vector containing the process \(y (t)\) and its \(q\) derivatives, \(G(dt)=(G_0(dt),\dots,G_q(dt))\) is a matrix of size \(m\times (q +1)\). The columns of this matrix are signed vector-measures \(G_0(dt),\dots,G_q(dt)\) defined on Borel subsets of \(\mathcal{T}\). Several general statements concerning the explicit form of the best linear unbiased estimator (BLUE) are proved. There are derived the explicit forms of the BLUE for many specific models including the cases of continuous autoregressive errors of order two and integrated error processes (such as integrated Brownian motion). The results are illustrated by examples.