On the asymptotic estimate of the maximum likelihood of parameters of the spectral density having zeros

From MaRDI portal





Let \(\{x_ t\}\) be a stationary Gaussian time series with spectral density f(\(\lambda)\) which is a function of an unknown p-dimensional parameter \(\vartheta\). Let \(L_ n(\vartheta,X)\) be the logarithm of the likelihood of the data \(X=(x_ 1,...,x_ n)'\) and \(\tilde L_ n(\vartheta,X)\) be a simpler main part of \(L_ n(\vartheta,X)\) satisfying the condition \[ n^{-1/2}[L_ n(\vartheta,X)-\tilde L_ n(\vartheta,X)]\to 0 \] as \(n\to \infty\) in probability. Then the estimate \({\hat \vartheta}_ n\) maximizing \(\tilde L_ n(\vartheta,X)\) is called the asymptotic estimate of the maximum likelihood, AEML. In the paper, consistency, asymptotic normality and asymptotic efficiency of \({\hat \vartheta}_ n\) are proved for the general case that f(\(\lambda)\) has both weak and strong zeros of polynomial type, i.e. when it admits the representation \[ f(\lambda;\vartheta)=| Q_ m(e^{i\lambda})|^ 2h(\lambda;\vartheta), \] where \(Q_ m(e^{i\lambda})\) is a polynomial of degree m with roots on the unit circle, which are independent of \(\vartheta\), and h(\(\lambda\) ;\(\vartheta)\) has weak zeros also independent of \(\vartheta\). Confidence regions for \(\vartheta\) are also suggested.











This page was built for publication: On the asymptotic estimate of the maximum likelihood of parameters of the spectral density having zeros

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3750864)