Instant trend-seasonal decomposition of time series with splines

Luis Francisco Rosales; Tatyana Krivobokova

We present a nonparametric method to decompose a times series into trend, seasonal and remainder components. This fully data-driven technique is based on penalized splines and makes an explicit characterization of the varying seasonality and the correlation in the remainder. The procedure takes advantage of the mixed model representation of penalized splines that allows for the simultaneous estimation of all model parameters from the corresponding likelihood. Simulation studies and three data examples illustrate the effectiveness of the approach.

MoreLess

Year of publication:	2012
Authors:	Rosales, Luis Francisco ; Krivobokova, Tatyana
Publisher:	Göttingen : Courant Research Centre
Subject:	Penalized splines \| Mixed model \| Varying coecient \| Correlated remainder \| Zeitreihenanalyse \| Time series analysis \| Nichtparametrisches Verfahren \| Nonparametric statistics \| Schätztheorie \| Estimation theory