A new approach to statistical forecasting
Author: Makridakis, Spyros INSEAD Area: Technology and Operations Management In: Economic structural change:analysis and forecasting - Hackl, Peter;Westlund, Anders H - 1991 - Book Language: EnglishDescription: p. 233-253.Type of document: INSEAD ChapterNote: Please ask us for this itemAbstract: Available approaches to statistical forecasting suffer from several deficiences that can render their predictions for real-world economic/business series inappropriate. In this paper such deficiencies are illustrated with real-life data and an approach is proposed that corrects their negative impact based on three premises: model selection is not based on historical information but rather on accuracy measures computed from out-of-sample data; two types of model selection are done on out-of-sample the first chooses the best model from those available within a single method, while the second selects the best among four methods run in parallel; the within method or among methods model selection is done for each forecasting horizon separately, making it possible to have different models and/or methods to predict each of the m horizons. In addition to being theoretically appealing, this new approach outperforms the best method of the M-Competition by a large margin when tested empirically.| Item type | Current location | Collection | Call number | Status | Date due | Barcode | Item holds |
|---|---|---|---|---|---|---|---|
|
Digital Library | Available | BC000479 |
Ask Qualtrics
Available approaches to statistical forecasting suffer from several deficiences that can render their predictions for real-world economic/business series inappropriate. In this paper such deficiencies are illustrated with real-life data and an approach is proposed that corrects their negative impact based on three premises: model selection is not based on historical information but rather on accuracy measures computed from out-of-sample data; two types of model selection are done on out-of-sample the first chooses the best model from those available within a single method, while the second selects the best among four methods run in parallel; the within method or among methods model selection is done for each forecasting horizon separately, making it possible to have different models and/or methods to predict each of the m horizons. In addition to being theoretically appealing, this new approach outperforms the best method of the M-Competition by a large margin when tested empirically.
Digitized
There are no comments for this item.