Abstract:
For Gaussian stationary processes, a time series Hellinger distance $T(f,g)$ for spectra $f$ and $g$ is derived. Evaluating $T(f_\theta,f_{\theta+h})$ of the
form $O(h^\alpha)$, we give $1/\alpha$-consistent asymptotics of the maximum
likelihood estimator of $\theta$ for nonregular spectra. For regular spectra,
we introduce the minimum Hellinger distance estimator
$\widehat{\theta}=\operatorname{arg}\min_\theta T(f_\theta,\widehat{g}_n)$,
where $\widehat{g}_n$ is a nonparametric spectral density estimator. We show
that $\widehat\theta$ is asymptotically efficient and more robust than the
Whittle estimator. Brief numerical studies are provided.