RT Generic
T1 Exploring ICA for time series decomposition
A1 García-Ferrer, Antonio
A1 González-Prieto, Ester
A1 Peña, Daniel
A2 Universidad Carlos III de Madrid. Departamento de Estadística, 
AB In this paper, we apply independent component analysis (ICA) for prediction and signalextraction in multivariate time series data. We compare the performance of threedifferent ICA procedures, JADE, SOBI, and FOTBI that estimate the componentsexploiting either the non-Gaussianity, or the temporal structure of the data, orcombining both, non-Gaussianity as well as temporal dependence. Some Monte Carlosimulation experiments are carried out to investigate the performance of thesealgorithms in order to extract components such as trend, cycle, and seasonalcomponents. Moreover, we empirically test the performance of those three ICAprocedures on capturing the dynamic relationships among the industrial productionindex (IPI) time series of four European countries. We also compare the accuracy of theIPI time series forecasts using a few JADE, SOBI, and FOTBI components, at differenttime horizons. According to the results, FOTBI seems to be a good starting point forautomatic time series signal extraction procedures, and it also provides quite accurateforecasts for the IPIs.
YR 2011
FD 2011-05
LK https://hdl.handle.net/10016/11285
UL https://hdl.handle.net/10016/11285
LA eng
DS e-Archivo
RD 20 may. 2024