Real-time signal extraction (RTSE) concerns the determination of optimal asymmetric filters towards the end of a time series where ⎯otherwise desirable⎯ symmetric filters cannot be applied anymore. The attractiveness of this particular estimation problem resides in the generality of its scope. For illustrative purposes we here stress realtime monitoring of the US-economy as well as multi-step ahead forecasting. Traditionally, the estimation problem addressed by RTSE is tackled in the methodological framework of the classical maximum likelihood paradigm. We here question the adequacy of this general parametric approach. In particular, we review a statistical apparatus⎯the DFA⎯ consisting of optimization criteria, diagnostics and tests which accounts for alternative user-relevant aspects of the estimation problem. Interestingly, this customization relates to an uncertainty principle which entails a fundamental shift of perspective. As a result, RTSE emerges as an autonomous discipline with proprietary concepts and statistics. With little suggestive power we may interpret the DFA as a generalization of the traditional model-based approach to more general enquiries about the future than the classical one-step ahead inference.
Multivariate Real-Time Signal Extraction by a Robust Adaptive Regression Filter