Title: Multivariate automated circulant SSA
Authors: Eva Senra - Universidad de Alcala (Spain) [presenting]
Juan Bogalo - University of Alcala (Spain)
Pilar Poncela - JRC (Italy)
Abstract: Circulant Singular Spectrum Analysis (CSSA) is an automated version of SSA that allows to extract the unobserved components associated to any frequency in a time series in an automated way. We generalize the technique to a multivariate setup and automatize it in the same way by the use of block circulant matrices applied to a new multivariate trajectory matrix. With the multivariate extension (M-CSSA) we can decompose the estimated signal at any frequency into the sum of $M$ orthogonal components that will allow to characterize the main sources of the fluctuations and obtain more robust signals for each individual time series. We also specify a time series factor model for an estimated vector signal that allows to obtain the common factors in the frequency domain from the information obtained with M-CSSA. Finally, we illustrate the application of the technique to a group of series.