A SUCCESSIVE FILTERING TECHNIQUE FOR IDENTIFYING LONG-TERM TRENDS

ASSIMAKOPOULOS, V

dc.contributor.author	ASSIMAKOPOULOS, V	en
dc.date.accessioned	2014-03-01T01:10:46Z
dc.date.available	2014-03-01T01:10:46Z
dc.date.issued	1995	en
dc.identifier.issn	0277-6693	en
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/11433
dc.subject	CYCLES	en
dc.subject	FILTERS	en
dc.subject	LONG-TERM TRENDS	en
dc.subject.classification	Management	en
dc.subject.classification	Planning & Development	en
dc.title	A SUCCESSIVE FILTERING TECHNIQUE FOR IDENTIFYING LONG-TERM TRENDS	en
heal.type	journalArticle	en
heal.identifier.primary	10.1002/for.3980140104	en
heal.identifier.secondary	http://dx.doi.org/10.1002/for.3980140104	en
heal.language	English	en
heal.publicationDate	1995	en
heal.abstract	The most reliable component of a time series, for forecasting purposes, is the shape of the long-term trend. However, the presence of cycles makes it difficult to identify and predict the changes in such trends. While up to now emphasis has been given in identifying and measuring cyclical behaviour, this paper presents a technique that aims at removing cyclical effects from the long-term trends. This technique is based on a transformation that is successively applied on the original time series. Each time the transformation is applied, an observation is selected and replaced by the average of its adjacent observations. This results in the elimination of the cyclical component. Independently of their depth, the cycles are being removed in ascending order relatively to their length. This leads to 'brushing off' the long-term trends from any cyclical effects. A specialized software has been developed in Pascal. The proposed technique was applied in a set of time series from the M2- and M-competition and the results are presented in this paper.	en
heal.publisher	JOHN WILEY & SONS LTD	en
heal.journalName	JOURNAL OF FORECASTING	en
dc.identifier.doi	10.1002/for.3980140104	en
dc.identifier.isi	ISI:A1995QF66700003	en
dc.identifier.volume	14	en
dc.identifier.issue	1	en
dc.identifier.spage	35	en
dc.identifier.epage	43	en