Combining techniques, Forecasting methods, SMAPE, Time series, Variability, Weights

This paper deals with the well-known ?no free lunch? phenomenon that, briefly said, claims that there is no time series forecasting method generally better than any other. To solve this problem, one may try to investigate which method is the most promising one for which time series, e.g., based on time series features such as its length, stationary, seasonality, origin etc. An alternative approach consists in combining more time series forecasting methods in order to eliminate the risk of choosing an inappropriate (or even the worst) method. We adopt the second approach and investigate the possible ways of combining distinct methods.