Overcoming non-admissibility in ARIMA-model-based signal extraction
This article analyses the situation in which the decomposition of a time series into orthogonal balanced components as performed by the ARIMA-model-based (AMB) method is non-admissible. It is shown that considering top-heavy models for the components can solve the problem. The top-heavy decomposition is derived and the improvement achieved is illustrated by an application to a class of models often encountered in practice. Two empirical applications allows comparison with the results yielded by the AMB decomposition of an approximated model by using an ad hoc filter such as X11-ARIMA and by direct specification of the structural time series models.
Bibliographic Reference: An article published in: Journal of Business & Economic Statistics, Vol.19, No.4 (2001), pp.455-464
Record Number: 200114197 / Last updated on: 2002-01-02
Original language: en
Available languages: en