Improving Forecasting Accruracy through Application of Temporal Aggregation

Abstract

Hierarchical forecasting (HF) has traditionally been applied to decrease the time and financial demands of the demand planning process in the cases where the company forecasts demand for a large number of items with a large number of customers. The current surveys show that application of a suitable HF method can result in improved accuracy of demand forecasts on different levels of its cross-sectional aggregation (based on products or territory). However, the area of temporal aggregation does not enjoy sufficient attention in the literature. This paper aims to analyze the influence of the choice of an HF method on the accuracy of corporate forecasts created on different levels of temporal aggregation of the demand. A case study conducted in a manufacturing company of the food industry included time series forecasting in 23 key products of the company on 3 levels of temporal aggregation of sales (yearly, quarterly and monthly sales) using 4 fundamentally different approaches to hierarchical forecasting (bottom-up, middle-out, top-down and optimal combination methods). The forecast accuracy was evaluated through MdAPE indicator. Testing of statistical hypotheses helped to confirm whether choice of an HF method has a significant effect on a change in the monitored forecast error. The study outcomes showed that choice of an HF method affects the accuracy of corporate forecasts. However, the forecasting error was significantly decreased on all the levels of temporal aggregation only when the bottom-up method was applied.