On the benefits of temporal aggregation on forecasting and stock control: empirical evidence from statistical and machine learning methods

Abstract

The ADIDA forecasting approach broke into the operational research literature in 2011, originally as a method for intermittent demand forecasting; 15 years and 226 citations later, with applications spanning from normal and intermittent to hierarchical demand, and inspiring the creation of many other forecasting methods and frameworks – most notably MAPA. This study approaches ADIDA as a flexible aggregate-forecast-disaggregate framework that can be used with various forecasting methods. We provide in this research a dated account of the work citing the original research, and provide empirical evidence that the framework can enhance the performance of advanced machine learning methods as well. The findings indicate that temporal aggregation improves forecasting accuracy, although its effectiveness depends on various conditions. Accurate demand forecasts are a prerequisite for sound inventory and production decisions, since order quantities, safety stock and production schedules are all set with reference to a forecast of future demand. By improving forecast accuracy, ADIDA therefore has direct implications for inventory planning and stock control, beyond its purely statistical performance.