The integrated periodogram for stable processes (Q1354493)

We consider the linear process \[ X_t= \sum_{j=-\infty}^\infty \psi_j Z_{t-j}, \qquad t\in\mathbb{Z}, \] where \((Z_t)_{t\in\mathbb{Z}}\) is a noise sequence of i.i.d. symmetric \(\alpha\)-stable r.v.'s for \(\alpha\in (0,2)\). This implies in particular that \(Z_t\) and \(X_t\) have infinite variance. In two preceding papers [Stochastic Processes Appl. 47, No. 2, 323-344 (1993; Zbl 0779.60023); Scand. J. Stat. 21, No. 4, 485-491 (1994; Zbl 0809.62081)] we studied the asymptotic behavior of the periodogram \[ I_{n,X}(\lambda)= n^{-2/\alpha} \Biggl| \sum_{t=1}^n e^{-i\lambda t}X_t\Biggr|^2, \qquad \lambda\in [-\pi,\pi]. \] The results obtained there indicate that the self-normalized periodogram \[ \widetilde{I}_{n,X}(\lambda)= I_{n,X}(\lambda)/ \gamma_{n,X}^2, \quad \lambda\in [-\pi,\pi], \qquad\text{with}\qquad \gamma_{n,X}^2= n^{-2/\alpha} \sum_{t=1}^n X_t^2, \] behaves very much like the periodogram for finite variance linear processes. In the present paper we continue our investigation of the spectral analysis of \(\alpha\)-stable processes. Our results can be understood as a study of classical (i.e., finite variance) quantities in spectral analysis when some of the innovations \(Z_t\) assume very large values. In this sense, the theory given below provides some recommendations on how classical estimators and test statistics have to be modified when large \(Z_t\) occur. As appropriate techniques we propose random normalization (we call it self-normalization) and random centering.