On estimating the hidden periodicities in linear time series models (Q1107250)

The author investigates the model \[ X(n)=\sum^{p}_{j=1}\alpha_ je^{in \lambda_ j}+\xi_ n \] where p, \(\lambda_ 1,...,\lambda_ p\) are constants, \(\alpha =(\alpha_ 1,...,\alpha_ p)'\) is a random vector and \(\xi_ n\) is a stationary random sequence with spectral density f(\(\lambda)\). An estimator \(\hat p\) of the parameter p is (roughly speaking) based on the number of peaks of the periodogram which are sufficiently high and sufficiently narrow. It is proved that \(\hat p\) is strongly consistent and that the corresponding estimators for \(\lambda_ i\) and \(\alpha\) are also strongly consistent. The method is demonstrated on two numerical examples.