Robustness properties of some forecasting methods for seasonal time series: A Monte Carlo study

Most statistical time series forecasting methods are based upon some assumptions on the data generating processes. These assumptions, however, may not be satisfied in practical situations. In this research, we investigate robustness properties of four major forecasting methods for seasonal time series, using Monte Carlo simulations. We ask the question as to whether the various methods have reasonably good forecasting performances for a wide class of time series for which the methods are likely to be used. We discuss some reasons why a forecasting method is (or is not) robust.