Forecasting Exceptional Demand: Not the Same as Forecasting Ordinary Demand

Many forecasting models designed to track and forecast trend and seasonality do well enough for ordinary demand conditions but are challenged by exceptional patterns such as sudden large jumps and turning points. Wil Gorr proposes a framework for exceptional-demand situations that is based on Receiver Operating Characteristics (ROC) analysis, in which models for exceptional demand forecasts work as a team with models to forecast ordinary demands. The ROC framework gives less emphasis to traditional forecast accuracy metrics such as the MAPE in favor of metrics to distinguish between real and false signals of exceptional demand. A similar warning about the use of conventional accuracy metrics is presented in Roy Batchelor’s article, “Accuracy vs. Profitablity”, in this issue of Foresight. Copyright International Institute of Forecasters, 2011