Changing spectrum estimation

An autoregressive (AR) model with time varying coefficients is used for the modeling of non-stationary covariance time series. A stochastically perturbed linear constraint model is presented as a model of time varying AR coefficients. The overall model for the time series is fitted by using the Kalman filter and Akaike's AIC criterion. An estimate of a changing spectrum is obtained by the fitted model and the fixed interval smoother Examples are given to illustrate the procedure.