Time series forecasting, model competitions and the theta method

In this talk we will present recent stochastic expansions of the theta method, the winner of the largest time series competitions (M3-Competion), which caught researchers' attention due to its performance and simplicity. The Theta method, as implemented in the monthly subset of the M3-Competiton, decomposes the seasonally adjusted data into two "theta lines''. The first theta line removes the curvature of the data to estimate the long-term trend component. The second theta line doubles the local curvatures of the series to approximate the short-term behavior. The proposed Dynamic Optimised Theta Model is a state space model that optimally selects the best short-term theta line and dynamically revises the long-term theta line. The superior performance of this model is demonstrated through an empirical application in the M3-Competition database.

Referências:
Makridakis, S. & Hibon, M. (2000). The M3-competition: results, conclusions and implications. International Journal of Forecasting , 16, 451{476.

Fiorucci, J. A., Pellegrini, T. R., Louzada, F., Petropoulos, F. & Koehler, A. B. (2016). Modelfor optimising the theta method and their relationship to state space models. Internationa Journal of Forecasting, 32(4), 1151-1161.