Nonparametric high resolution spectral estimation

Summary The uniform rate of convergence of the integrated relative mean square error over a (with the sample sizeT) increasing classI _T of stationary processes is studied for several estimates of the spectral density. The classI _T is chosen in a way such that estimates with a good uniform rate of convergence overI _T may be termed high resolution spectral estimates. By using this criterion several effects are explained theoretically, for example the leakage effect. The advantages uf using data tapers are proved and the use of local and global bandwiths are studied. Furthermore, the behaviors of segment estimates are studied. Simulations are presented for the illustration of some effects.This work has been supported by the Deutsche Forschungsgemeinschaft