DETERMINING THE NUMBER OF TERMS IN A TRIGONOMETRIC REGRESSION

Abstract. We consider the estimation of the number of sinusoidal terms in a time series contaminated by additive noise with unknown correlation structure. The method fits sinusoidal terms by least squares and models the noise component using a high order autoregression. A criterion based on the minimum description length principle is used to select the number of sinusoidal terms and the order of the noise model. The small sample efficacy of the model selection procedure is examined by simulations and the analysis of some astronomical data. Consistency is proved under quite general conditions on the noise spectrum.