Spectral density estimation for stationary stable processes (Q794377)

The paper deals with nonparametric spectral density estimation of strictly stationary complex symmetric \(\alpha\)-stable (S \(\alpha\) S) processes X, \(0<\alpha<2\) (especially, these processes have finite moments of order only less than \(\alpha\) so that the standard spectral analysis is not applicable to them). Given an observation of an S \(\alpha\) S process X with a spectral density \(\phi\) (\(\lambda)\) over a finite interval [-T,T] the modified periodogram \(I_ T(\lambda)=C_{p,\alpha}| d_ T(\lambda)|^ p\) is introduced \((0<p<\alpha /2)\), where \(C_{p,\alpha}\) is a normalization constant and \(d_ T(\lambda)\) is the real part of the finite tapered Fourier transform of X(t). The periodogram \(I_ T(\lambda)\) is an asymptotically unbiased estimate of \(\{\phi(\lambda)\}^{p/\alpha}\) but it is not a consistent estimate of it (the covariance structure of \(I_ T(\lambda)\) is established). By smoothing \(I_ T(\lambda)\) via a spectral window a mean-square consistent estimate \(f_ T(\lambda)\) of \(\{\phi(\lambda)\}^{p/\alpha}\) is obtained (with rates of convergence). Finally, \(\phi_ T(\lambda)=\{f_ T(\lambda)\}^{\alpha /p}\) is an estimate of \(\phi\) (\(\lambda)\) which is consistent in probability. A strongly consistent estimate can be obtained for an appropriate subsequence \(\tilde T\) of T (the rates of convergence are again established). The cases of real S \(\alpha\) S processes and one sided intervals [0,T] of observations are also discussed.