Title: On estimating the number of signals in multivariate time series
Authors: Klaus Nordhausen - Vienna University of Technology (Austria) [presenting]
Joni Virta - Aalto University (Finland)
Abstract: A $p$-variate second order source separation (SOS) model is considered under the assumption that only $q<p$ source components exhibit serial autocorrelations and the remaining $p-q$ components can be considered noise. The goal is then to estimate the number of signals $q$ and the signals itself. To solve the problem bootstrap and asymptotic hypothesis tests for the signal dimension as well as the ladle estimator are introduced. The tests and estimators are all based on the two SOS methods AMUSE and SOBI.