Combining simple exponential smoothing models for time series forecasting

Simple exponentional smoothing is a well-known technique for forecasting univariate time series without trend and seasonality. Forecast combinations such as medians or means are known to improve the accuracy of point forecasts. In this paper we have experimented with combining the forecasts of several simple exponential smoothing models with different smoothing factors. Experimental results, using the M3-competition time series, show that the combined forecasts outperform the forecasts of the model that best fits the series.