Accurate demand forecasts are imperative for organisations to make strategic long-term decisions. Demand forecasting for the long-term is challenging and further complications can arise when a demand series has periods where no demand is observed. Motivated by operational decisions made at an automotive manufacturer, we contribute novel methodology to determine long-term forecasts for spare parts.
We first introduce a stochastic demand forecasting model for products in the final phase of the life cycle when demand is in decline. Theoretical results on the bias and variance of the parameter estimates motivate an extension which uses the demand history of parts with the same declining pattern. In experiments on real data, we demonstrate that our extension reduces the mean absolute percentage error, achieves a higher fill rate and incurs less leftover inventory.
We then outline an approach for long-term demand forecasting throughout the product life cycle. We extend our model by pooling the incomplete demand histories of products with similar life cycle behaviour to estimate joint model parameters. We validate our approach on 175 automotive spare parts and find that our extension improves forecast accuracy even for cases when the peak of demand is yet to be observed.
As a third contribution, we develop material to communicate forecasting and modelling topics to wider audiences. We design outreach content based on core principles and consider aims to address recruitment shortages and gender disparity in the mathematical sciences. We reflect on our impact using teacher feedback.