Simulation Based Forecast Data Generation and Evaluation of Forecast Error Measures

Production planning is usually performed based on customer orders or demand forecasts. The demand forecasts in production systems can either be generated by manufacturing companies themselves, i.e. forecast prediction, or they can be provided by customers. For both alternatives, forecast prediction, as well as the customer-provided forecasts, the quality of those forecasts is critical for success. In this paper, a simulation model to generate forecast data that mimic different forecast behaviors is presented. In detail, an independent forecast distribution and a forecast evolution model are investigated to discuss the value of customer-provided forecasts in comparison to the simple moving average forecast prediction method. Main findings of the paper are that Root-Mean-Square-Error and Mean-Absolute-Percentage-Error describe the forecast error well if no systematic effects are present and Mean-Percentage-Error provides a good measure for systematic effects. Furthermore, systematic effects like overbooking are significantly reducing the value of customer-provided forecast information.