Original language | English |
---|---|
Title of host publication | Wiley StatsRef: Statistics Reference Online |
Publisher | Wiley |
DOIs | |
Publication status | Published - 4 Nov 2020 |
Forecasting with the Theta Method
Research output: Chapter in Book/Report/Conference proceeding › Entry for encyclopedia/dictionary › peer-review
Electronic versions
DOI
Theta method is the most successful univariate time series forecasting method of the past two decades, since its origination in 1999. The method's success has been demonstrated in applications in demand forecasting, marketing, and supply chain forecasting contexts; nevertheless, the success in extensive blind empirical forecasting competitions also involved thousands of series from finance, economics, and a wide range of applications and thus attested for generalizability. The superior performance of Theta method originally was proved in the M3 completion in 2000, where the Theta method was the only method that outperformed Forecast Pro and dominated a pool of 18 academic methods and 5 software packages. The performance in M4 competition was also very commendable where the method outperformed – in both accuracy and computational time – all standard forecasting benchmarks, most notably the well‐celebrated ETS method and the ARIMA models in all their variants. Theta method has also been an essential part of most of the well‐performing combinations that participated in M4. In essence, Theta method and Dampen Trend Exponential Smoothing are the two benchmarks that any new forecasting method should outperform, in order to pass the test of time.