Title: Dimension reduction for time series in a blind source separation context
Authors: Klaus Nordhausen - Vienna University of Technology (Austria)
Markus Matilainen - University of Turku/Turku PET Centre (Finland)
Jari Miettinen - Aalto University (Finland)
Joni Virta - Aalto University (Finland)
Sara Taskinen - University of Jyvaskyla (Finland) [presenting]
Abstract: Multivariate time series observations are increasingly common in many fields of science but the complex dependencies between multiple time series often yield to intractable models with large number of parameters. An alternative approach is to first reduce the dimension of the series and then model the resulting uncorrelated univariate time series. Blind source separation (BSS) offers a popular and effective framework for this. We review some dimension reduction tools for time series in a BSS context. In specific, we propose an estimator which is developed for identifying components which exhibit volatility clustering. The theoretical properties of the estimator are discussed and an example is provided to illustrate the method.