Title: Zero-crossings of time series: Sign-prediction, mean-square error and a holding-time constraint
Authors: Marc Wildi - University of Applied Sciences Zurich (Switzerland) [presenting]
Abstract: An extension of classic time series approaches is proposed, which addresses zero-crossings of a zero-mean stationary time series. Specifically, we subject the original optimization criterion to a novel holding-time constraint which conditions the expected duration between consecutive crossings. Formally, the solution to this prediction problem is obtained by assigning more weight to components of the classic estimate, which are in better accordance with the constraint. Besides an analysis based on simple forecast and signal extraction exercises, we also provide an application of our novel approach to business-cycle analysis. The latter example illustrates that improved smoothing capability (noise suppression) does not necessarily conflict with timeliness (relative lead) in a real-time nowcasting setting.