Search

The Univariate Moving Window Spectral method is extended to the Multivariate Moving Window Spectral method (MWS). This further mitigates the bias encountered in time series model parameter estimates, that otherwise results in multiple-period lead time forecast error divergence. Compared to time domain methods, the spectral approach provides for better estimation of cyclical components in time series, When recombined by the MWS paradigm, better long range forecasts are possible. The method is illustrated by a tri-variate sales, price and income case study, The multivariate MWS method requires little user expertise, explains the data better, and forecasts better. It is applicable to a broad range of biomedical, physical, economic, and social science time series. Keywords: Global parametrization Forecasting Moving window Spectral analysis Frequency domain

To link to full-text access for this article, visit this link: http://dx.doi.org/10.1016/j.cie.2008.03.009 Byline: Dennis Ridley (a), Pierre Ngnepieba (b) Keywords: Global parametrization; Forecasting; Moving window; Spectral analysis; Frequency domain Abstract: The Univariate Moving Window Spectral method is extended to the Multivariate Moving Window Spectral method (MWS). This further mitigates the bias encountered in time series model parameter estimates, that otherwise results in multiple-period lead time forecast error divergence. Compared to time domain methods, the spectral approach provides for better estimation of cyclical components in time series. When recombined by the MWS paradigm, better long range forecasts are possible. The method is illustrated by a tri-variate sales, price and income case study. The multivariate MWS method requires little user expertise, explains the data better, and forecasts better. It is applicable to a broad range of biomedical, physical, economic, and social science time series. Author Affiliation: (a) SBI, Florida A & M University and School of Computational Science, FSU, Tallahassee, FL 32306-4120, USA (b) Department of Mathematics, Florida A & M University Tallahassee, FL 32307, USA Article History: Received 2 February 2007; Accepted 19 March 2008 Article Note: (footnote) [star] This manuscript was processed by Area Editor E.A. Elsayed.