An Application of First-Order Autoregressive Markov Regime Switching Models for Logarithmic Demand Forecasting in Backtesting "Paghi Poco" Supermarkets in Sicily

Abstract

This work evaluates the forecasting performance of first-order autoregressive Markov Regime Switching (MRS) models to forecast demand in four retail units of the “Paghi Poco” supermarket chain, a national retail chain based in Sicily. In mind of the rapid pace of digitalization in retail and despite the increasing availability of point-of-sale (POS) data, we underline the value of analytical forecasting models that will aid operations management by modelling demand shifts. Specifically, we show that the MRS models will capture the regime shifts between low and high demand regimes that are present in the historical sales data by a back-testing methodology that draw upon an extended history of sales data, which contributed to a better understanding of market dynamics. The results show improvements in forecasting model performance when compared to traditional simpler models, demonstrating the probability of improved decision-making and enhanced agility and resilience to the retail supply chain via the MRS approach.