A bias and variance analysis for multistep-ahead time series forecasting

Souhaib Ben Taieb, Amir F. Atiya

Research output: Contribution to journalArticleResearchpeer-review

Abstract

Multistep-ahead forecasts can either be produced recursively by iterating a one-step-ahead time series model or directly by estimating a separate model for each forecast horizon. In addition, there are other strategies; some of them combine aspects of both aforementioned concepts. In this paper, we present a comprehensive investigation into the bias and variance behavior of multistep-ahead forecasting strategies. We provide a detailed review of the different multistep-ahead strategies. Subsequently, we perform a theoretical study that derives the bias and variance for a number of forecasting strategies. Finally, we conduct a Monte Carlo experimental study that compares and evaluates the bias and variance performance of the different strategies. From the theoretical and the simulation studies, we analyze the effect of different factors, such as the forecast horizon and the time series length, on the bias and variance components, and on the different multistep-ahead strategies. Several lessons are learned, and recommendations are given concerning the advantages, disadvantages, and best conditions of use of each strategy.
Original languageEnglish
Article number7064712
Pages (from-to)62-76
Number of pages15
JournalIEEE Transactions on Neural Networks and Learning Systems
Volume27
Issue number1
DOIs
Publication statusPublished - 1 Jan 2016
Externally publishedYes

Keywords

  • Bias
  • machine learning
  • Monte Carlo simulation
  • multistep-ahead forecasting
  • nearest neighbors
  • neural networks (NNs)
  • time series
  • variance

Cite this

@article{dd3ab5ddefb64d138ef4cbcc643db14a,
title = "A bias and variance analysis for multistep-ahead time series forecasting",
abstract = "Multistep-ahead forecasts can either be produced recursively by iterating a one-step-ahead time series model or directly by estimating a separate model for each forecast horizon. In addition, there are other strategies; some of them combine aspects of both aforementioned concepts. In this paper, we present a comprehensive investigation into the bias and variance behavior of multistep-ahead forecasting strategies. We provide a detailed review of the different multistep-ahead strategies. Subsequently, we perform a theoretical study that derives the bias and variance for a number of forecasting strategies. Finally, we conduct a Monte Carlo experimental study that compares and evaluates the bias and variance performance of the different strategies. From the theoretical and the simulation studies, we analyze the effect of different factors, such as the forecast horizon and the time series length, on the bias and variance components, and on the different multistep-ahead strategies. Several lessons are learned, and recommendations are given concerning the advantages, disadvantages, and best conditions of use of each strategy.",
keywords = "Bias, machine learning, Monte Carlo simulation, multistep-ahead forecasting, nearest neighbors, neural networks (NNs), time series, variance",
author = "{Ben Taieb}, Souhaib and Atiya, {Amir F.}",
year = "2016",
month = "1",
day = "1",
doi = "10.1109/TNNLS.2015.2411629",
language = "English",
volume = "27",
pages = "62--76",
journal = "IEEE Transactions on Neural Networks and Learning Systems",
issn = "2162-237X",
number = "1",

}

A bias and variance analysis for multistep-ahead time series forecasting. / Ben Taieb, Souhaib; Atiya, Amir F.

In: IEEE Transactions on Neural Networks and Learning Systems, Vol. 27, No. 1, 7064712, 01.01.2016, p. 62-76.

Research output: Contribution to journalArticleResearchpeer-review

TY - JOUR

T1 - A bias and variance analysis for multistep-ahead time series forecasting

AU - Ben Taieb, Souhaib

AU - Atiya, Amir F.

PY - 2016/1/1

Y1 - 2016/1/1

N2 - Multistep-ahead forecasts can either be produced recursively by iterating a one-step-ahead time series model or directly by estimating a separate model for each forecast horizon. In addition, there are other strategies; some of them combine aspects of both aforementioned concepts. In this paper, we present a comprehensive investigation into the bias and variance behavior of multistep-ahead forecasting strategies. We provide a detailed review of the different multistep-ahead strategies. Subsequently, we perform a theoretical study that derives the bias and variance for a number of forecasting strategies. Finally, we conduct a Monte Carlo experimental study that compares and evaluates the bias and variance performance of the different strategies. From the theoretical and the simulation studies, we analyze the effect of different factors, such as the forecast horizon and the time series length, on the bias and variance components, and on the different multistep-ahead strategies. Several lessons are learned, and recommendations are given concerning the advantages, disadvantages, and best conditions of use of each strategy.

AB - Multistep-ahead forecasts can either be produced recursively by iterating a one-step-ahead time series model or directly by estimating a separate model for each forecast horizon. In addition, there are other strategies; some of them combine aspects of both aforementioned concepts. In this paper, we present a comprehensive investigation into the bias and variance behavior of multistep-ahead forecasting strategies. We provide a detailed review of the different multistep-ahead strategies. Subsequently, we perform a theoretical study that derives the bias and variance for a number of forecasting strategies. Finally, we conduct a Monte Carlo experimental study that compares and evaluates the bias and variance performance of the different strategies. From the theoretical and the simulation studies, we analyze the effect of different factors, such as the forecast horizon and the time series length, on the bias and variance components, and on the different multistep-ahead strategies. Several lessons are learned, and recommendations are given concerning the advantages, disadvantages, and best conditions of use of each strategy.

KW - Bias

KW - machine learning

KW - Monte Carlo simulation

KW - multistep-ahead forecasting

KW - nearest neighbors

KW - neural networks (NNs)

KW - time series

KW - variance

UR - http://www.scopus.com/inward/record.url?scp=84925431469&partnerID=8YFLogxK

U2 - 10.1109/TNNLS.2015.2411629

DO - 10.1109/TNNLS.2015.2411629

M3 - Article

VL - 27

SP - 62

EP - 76

JO - IEEE Transactions on Neural Networks and Learning Systems

JF - IEEE Transactions on Neural Networks and Learning Systems

SN - 2162-237X

IS - 1

M1 - 7064712

ER -