List

Journal of Forecasting (2005), 24(1), 17-37.

Rob J. Hyndman1, Anne B. Koehler2, J. Keith Ord3 and Ralph D. Snyder1

  1. Department of Econometrics and Business Statistics, Monash University, VIC 3800, Australia.
  2. Department of Decision Sciences and Management Information Systems, Miami University, Oxford, OH 45056, USA.
  3. 320 Old North, Georgetown University, Washington, DC 20057, USA.

Abstract: Three general classes of state space models are presented, based upon the single source of error formulation. The first class is the standard linear state space model with homoscedastic errors, the second retains the linear structure but incorporates a dynamic form of heteroscedasticity, and the third allows for non-linear structure in the observation equation as well as heteroscedasticity. These three classes provide stochastic models for a wide variety of exponential smoothing methods. We then use these three classes to provide exact analytic (matrix) expressions for forecast error variances that can be used to construct prediction intervals one or multiple steps ahead. These formulas are then specialized to non-matrix formulas for fifteen state space models that underlie nine exponential smoothing methods, including all the widely used methods. In cases where an ARIMA model also underlies an exponential smoothing method, there is an equivalent state space model with the same variance expression. We also discuss relationships between these new ideas and previous suggestions for finding forecast error variances and prediction intervals for the exponential smoothing methods. Simpler approximations are developed for the more complex schemes and their validity examined. The paper concludes with a numerical example using a non-linear model.

Keywords: forecast distribution, forecast interval, forecast error variance, Holt-Winters method, structural models.

Errata:

p26
Replace first equation with the following:
\(v_h \approx s^2_{n-m+1+(h-1)*}\left[\tilde{\mu}_h^2(1+\alpha^2\sigma^2)^{h-1}(1+\sigma^2)(1+\gamma^2\sigma^2)^k – \ell_n^2\right]\)
p26
Replace \(s_{n-1} = 0.90,~s_{n-1} = 1.10\) with \( s_{n-2} = 0.90,~s_{n-3} = 1.10.\)

R code

Online article

  Posts

1 2 3 5
September 20th, 2016

smoothAPC package for R

September 20th, 2016

stR package for R

September 14th, 2016

Grouped functional time series forecasting: an application to age-specific mortality rates

August 30th, 2016

Forecasting large collections of related time series

August 22nd, 2016

thief package for R

June 21st, 2016

Exploring time series collections used for forecast evaluation

June 9th, 2016

Associations between outdoor fungal spores and childhood and adolescent asthma hospitalisations

May 25th, 2016

ISCRR time series workshop

May 19th, 2016

Visualising Forecasting Algorithm Performance using Time Series Instance Spaces

May 6th, 2016

Automatic foRecasting using R

February 29th, 2016

On sampling methods for costly multi-objective black-box optimization

February 19th, 2016

Dynamic Algorithm Selection for Pareto Optimal Set Approximation

February 4th, 2016

Forecasting uncertainty in electricity smart meter data by boosting additive quantile regression

January 30th, 2016

Bayesian rank selection in multivariate regressions

January 25th, 2016

Probabilistic Energy Forecasting: Global Energy Forecasting Competition 2014 and Beyond

January 24th, 2016

Long-term forecasts of age-specific participation rates with functional data models

January 1st, 2016

Bagging exponential smoothing methods using STL decomposition and Box-Cox transformation

January 1st, 2016

Fast computation of reconciled forecasts for hierarchical and grouped time series

December 31st, 2015

Measuring forecast accuracy

November 26th, 2015

Forecasting hierarchical and grouped time series through trace minimization

November 2nd, 2015

Forecasting big time series data using R

October 7th, 2015

Optimal forecast reconciliation for big time series data

October 5th, 2015

Google workshop: Forecasting and visualizing big time series data

September 16th, 2015

Unbelievable

August 29th, 2015

Forecasting with temporal hierarchies

August 25th, 2015

New IJF editors

August 17th, 2015

Machine learning bootcamp

August 7th, 2015

Statistical issues with using herbarium data for the estimation of invasion lag-phases

June 30th, 2015

Exploring the feature space of large collections of time series

June 26th, 2015

Exploring the boundaries of predictability: what can we forecast, and when should we give up?

June 25th, 2015

Automatic algorithms for time series forecasting

June 23rd, 2015

MEFM: An R package for long-term probabilistic forecasting of electricity demand

June 19th, 2015

Probabilistic forecasting of peak electricity demand

June 10th, 2015

Do human rhinovirus infections and food allergy modify grass pollen–induced asthma hospital admissions in children?

June 8th, 2015

STR: A Seasonal-Trend Decomposition Procedure Based on Regression

June 4th, 2015

Probabilistic time series forecasting with boosted additive models: an application to smart meter data

June 1st, 2015

Large-scale unusual time series detection

May 26th, 2015

Visualization of big time series data

May 22nd, 2015

Probabilistic forecasting of long-term peak electricity demand

April 20th, 2015

A note on the validity of cross-validation for evaluating time series prediction

April 4th, 2015

Discussion of “High-dimensional autocovariance matrices and optimal linear prediction”

April 1st, 2015

Change to the IJF editors

February 23rd, 2015

Visualization and forecasting of big time series data

January 12th, 2015

Visualizing and forecasting big time series data

December 24th, 2014

Bivariate data with ridges: two-dimensional smoothing of mortality rates

December 17th, 2014

MEFM package for R

October 21st, 2014

Optimally reconciling forecasts in a hierarchy

September 23rd, 2014

Forecasting: principles and practice (UWA course)

September 1st, 2014

Outdoor fungal spores are associated with child asthma hospitalisations – a case-crossover study

August 1st, 2014

Efficient identification of the Pareto optimal set