Mean-based error measures for intermittent demand forecasting

摘要: To compare different forecasting methods on demand series, we require an error measure. Many measures have been proposed, but when is intermittent some become inapplicable because of infinities, give counter-intuitive results, and there no agreement which best. We argue that almost all known rank forecasters incorrectly series. propose several new with correct forecaster ranking patterns. call these ‘mean-based’ they evaluate forecasts against the (possibly time-dependent) mean underlying stochastic process instead point demands.