Multivariate real time signal extraction by a robust adaptive regression filter
Résumé
We propose a new regression-based filter for extracting signals online in moving windows from multivariate high frequency time series. This fast and robust filtering procedure considers the local covariance structure between the single time series components. It tackles the bias variance trade-off problem for the optimal choice of the window width by choosing the size of the window adaptively, depending on the current data situation. Furthermore, the signals are estimated at the recent point in time. Moreover, we present an advanced algorithm of our filter for replacing missing observations in real time. Thus it can be applied in online-monitoring practice.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...