Title: Robust time series forecasting with exponential smoothing methods
Authors: Ruben Crevits - KU Leuven (Belgium) [presenting]
Christophe Croux - Edhec Business School (France)
Abstract: Simple methods like exponential smoothing are very popular for forecasting univariate time series. The R-package for forecasting with exponential smoothing has been downloaded numerous times. The method chooses whether or not to include a (damped) trend or seasonality effects, both of which often occur in real time series. We provide a robust alternative for the exponential smoothing forecaster. For each variation of exponential smoothing we present a robust alternative. The robust method is developed by robustifying every aspect of the original exponential smoothing variant. We compare the standard non-robust version with our robust proposal in a simulation study. The methodology is applied to data from the M3 competition, which includes time series from microeconometrics, macroeconometrics, demographics, finance and industry.