© 2018

Multiscale Forecasting Models


  • The book is unique because it contains two new and competitive methods for time series decomposition to improve the accuracy of auto-regressive models

  • The methods are presented in detail through relevant applications

  • Additionally, the methods are compared with other techniques which are conventionally used in forecasting


Table of contents

  1. Front Matter
    Pages i-xxiv
  2. Lida Mercedes Barba Maggi
    Pages 1-29
  3. Lida Mercedes Barba Maggi
    Pages 31-47
  4. Lida Mercedes Barba Maggi
    Pages 49-88
  5. Lida Mercedes Barba Maggi
    Pages 89-118
  6. Back Matter
    Pages 119-124

About this book


This book presents two new decomposition methods to decompose a time series in intrinsic components of low and high frequencies. The methods are based on Singular Value Decomposition (SVD) of a Hankel matrix (HSVD). The proposed decomposition is used to improve the accuracy of linear and nonlinear auto-regressive models.

Linear Auto-regressive models (AR, ARMA and ARIMA) and Auto-regressive Neural Networks (ANNs) have been found insufficient because of the highly complicated nature of some time series. Hybrid models are a recent solution to deal with non-stationary processes which combine pre-processing techniques with conventional forecasters, some pre-processing techniques broadly implemented are Singular Spectrum Analysis (SSA) and Stationary Wavelet Transform (SWT). Although the flexibility of SSA and SWT allows their usage in a wide range of forecast problems, there is a lack of standard methods to select their parameters.

The proposed decomposition HSVD and Multilevel SVD are described in detail through time series coming from the transport and fishery sectors. Further, for comparison purposes, it is evaluated the forecast accuracy reached by SSA and SWT, both jointly with AR-based models and ANNs.


Forecasting Time Series Singular Value Decomposition Hankel Matrix Artificial Neural Networks Singular Spectrum Analysis Stationary Wavelet Decomposition

Authors and affiliations

  1. 1.Universidad Nacional de ChimborazoRiobambaEcuador

About the authors

Lida Mercedes Barba Maggi earned a PhD degree in Informatics Engineering from the Pontificia Universidad Católica de Valparaíso, Chile, in 2017. She is currently affiliated with the Universidad Nacional de Chimborazo in Ecuador. Her research interests include Analysis of time series, Forecast and estimate based on mathematical and statistical models, Forecast and estimate based on artificial intelligence, and Optimization Algorithms.

Bibliographic information

Industry Sectors
Chemical Manufacturing
IT & Software
Consumer Packaged Goods
Materials & Steel
Finance, Business & Banking
Energy, Utilities & Environment
Oil, Gas & Geosciences