Forecasting With The Theta Method

Preface

The Theta method is the most successful univariate time series forecasting technique of the past two decades, since its origination in the late 1990s. Jointly with the damped trend exponential smoothing, these are the two benchmarks that any newly proposed forecasting method aspires to outperform, so as to pass the test of time. This performance was originally demonstrated in the M3 competition in 2000 that the Theta method was the only method that outperformed Forecast Pro (www.forecastpro.com) and dominated a pool of 18 other academic methods and five more software packages.

The method originated in 1999 from Kostas and his supervisor Professor Vassilis Assimakopoulos, and was first presented in the International DSI 99¹ while the first full‐fledged academic article and description came in 2000 in the International Journal of Forecasting. In the same journal came the first critique three years later, led by Professor Rob Hyndman and his team in Monash for the similarity of the basic model from the method to simple exponential smoothing (SES) with drift.

The journal made a mistake in that they had neither used our review on the paper (as our objections were left unanswered) nor had they offered us a commentary to be published along that article – that is a standard practice in similar situations. This created more confusion as an immediate clear answer from us and Vassilis could not see the light of day; and it took a few more years until we found the means and media ...