Adjusted Grid Search to Find Hyper-parameters in SARIMAX Models: Efficiently Filling The Shelves in Kruidvat Stores

Abstract

Replenishment processes, promotions and assortment choices in retail are determined by forecasting sales. Sales are affected by external factors such as seasonality, weather or promotions. This relationship can differ per store or product. Currently, researchers focus on developing a SARIMAX model for one product and store specifically to predict sales, and the novelty of this research is that it develops an algorithm that automates SARIMAX modeling to allow for model generation tailored to a store and product. The algorithm was tested with a case study at Kruidvat. It was found that the algorithm works good enough to potentially save Kruidvat more than 5 million Euros annually.