A0260
Title: A new SSA-based procedure for detecting structural changes in a time series
Authors: Adelaide Freitas - University of Aveiro (Portugal) [presenting]
Alberto Silva - University of Aveiro (Portugal)
Abstract: Some procedures adopted to detect eventual structural changes in a time series using Singular Spectral Analysis consist of applying a single decomposition method to two different trajectory matrices (base and test) iteratively throughout the series. Then, distances between some eigenvectors and an appropriate subspace are computed and compared in each iteration. A method is proposed to assess differences when two decomposition methods (robust and ordinary) are applied to the same trajectory matrix. These differences will be more accentuated when there is an eventual change in the direction of some principal components (eigenvectors) in case of interrupting the linear recurrent formula. One advantage of this strategy lies in the possibility of interpretation in terms of the principal components that the visualization of the results provides.