scientific article; zbMATH DE number 1069588
The model studied is \[ y(t)= \mu+ \sum^r_1 (\alpha_j \cos \lambda_jt +\beta_j \sin \lambda_jt) +u(t),\;0<\lambda_j <\pi,\;t=0,\;\pm 1,\;\pm 2, \dots. \] We assume that \(u(t)\) is stationary with zero mean and absolutely continuous spectrum and spectral density \(f(\omega)\) that is continuous with \(f(\omega) >0\), \(\omega\in [-\pi,\pi]\). Other conditions will be imposed later. True values will be indicated as \(\lambda_{0_j}\), \(r_0\) etc. when the emphasis is needed. We are mainly concerned with estimating \(r_0\). The first term on the right could be identified with jumps in a spectral distribution for \(y(t)\), hence the title of the paper. If \(u(t)\) is white Gaussian noise then the maximum likelihood estimates of the \(\lambda_{0_j}\), for \(r\) known, are obtained by forming the regression sum of squares, \(Q(\lambda_1, \dots, \lambda_r)\), from the regression on 1, \(\cos \lambda_jt\), \(\sin \lambda_jt\), \(j=1, \dots,r\). The \(\widehat \lambda_j\) are chosen to maximize this. When the white Gaussian noise assumption is not satisfied there might be better methods but we continue to consider the ones just defined. For the entire collection see [Zbl 0846.00010].
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4356549)
