Persistence modeling for sales prediction: A simple, self-contained approach

Abstract

The accurate prediction of future retail sales is crucial in business. Marketing persistence modeling is a well-established framework for the analysis of retail sales dynamics using a time series approach. However, dominant academic approaches to persistence modeling are rather complex, requiring sophisticated analytic and computing skills. We introduce a decision making approach that allows jointly testing and predicting the evolution of retail sales using a state-space model. We first propose a model, which incorporates the foundations of persistence modeling, and provide a simple estimation method. Secondly, we show a formal procedure for detecting persistence and, hence, selecting parameters to predict future data. An empirical investigation carried out over nearly 30000 Walmart retail sales, shows that our model provides significantly better prediction performance when compared to standard benchmarks. The combination of simplicity and efficiency makes our approach appealing not only for scholars but also for practitioners.