Rob J Hyndman

Inter­na­tional Journal of Fore­cast­ing (2002), 18(3), 439–454.

Rob J. Hyndman¹,Anne B. Koehler²,Ralph D. Snyder¹ and Simone Grose¹

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.

Abstract: We provide a new approach to auto­matic busi­ness fore­cast­ing based on an exten­ded range of expo­nen­tial smooth­ing meth­ods. Each method in our tax­onomy of expo­nen­tial smooth­ing meth­ods can be shown to be equi­val­ent to the fore­casts obtained from a state space model. This allows (1) the easy cal­cu­la­tion of the like­li­hood, the AIC and other model selec­tion cri­teria; (2) the com­pu­ta­tion of pre­dic­tion inter­vals for each method; and (3) ran­dom sim­u­la­tion from the under­ly­ing state space model. We demon­strate the meth­ods by apply­ing them to the data from the M-​​competition and the M3-​​competition.

Keywords: auto­matic fore­cast­ing, expo­nen­tial smooth­ing, pre­dic­tion inter­vals, state space models.

R code

Online art­icle