Spectral analysis of segmented data
S. de Waele, PIET M. T. BROERSEN · 2002
Time series analysis is reformulated to allow processing of segmented data. This involves the reformulation of parameter estimation and order selection. Parameter estimation for autoregressive (AR) models is done by fitting a single model to all segments simultaneously. Parameter estimation for moving average (MA) and the combined ARMA models can be derived entirely from long autoregressive models. The finite sample theory required for order selection of AR models has been generalized to segments of data. The resulting algorithm can also deal effectively with segments of unequal length.