Score-driven models for inventory control of intermittent demand

Inventory control for intermittent data involves overseeing inventory levels of products that experience irregular demand. These items may remain unsold for extended periods and then experience a sudden surge in orders, making it difficult to predict future demand. To effectively manage such inventory, companies need specialised strategies that ensure they have enough stock to meet unpredictable spikes in demand while avoiding excess inventory that could lead to waste or storage problems. This study aims to evaluate the performance of forecasting models based on the score-driven (SD) framework as an alternative strategy for inventory control of intermittent demand. Several models were derived using the SD framework, including Poisson, negative binomial, and their hurdle and zero-inflated variants. The SD models were compared with several traditional forecasting methods for intermittent demand using synthetic data. A simulation model was developed to emulate the dynamics associated with inventory control through an order-up-to-level policy. The models were compared using well-known service and inventory level indicators. The proposed SD models consistently achieved closer service and target levels than traditional forecasting methods. Notably, the SD negative binomial model demonstrated the most robust performance across various simulated scenarios.