Wielding Occam’s razor: Fast and frugal retail forecasting

Abstract

The algorithms available for retail forecasting have increased in complexity. Newer methods, such as machine learning, are inherently complex. The more traditional families of forecasting models, such as exponential smoothing and autoregressive integrated moving averages, have expanded to contain multiple possible forms and forecasting proles. We question the complexity of forecasting and the need to consider such large families of models. Our argument is that parsimoniously identifying suitable subsets of models will not decrease forecasting accuracy, nor will they reduce the ability to estimate forecast uncertainty. We propose a framework that balances forecasting performance versus computational cost. As a result, we consider a reduced set of models. We empirically demonstrate that such a reduced set performs well. Finally, we translate computational benefits to monetary cost savings and environmental impact and discuss the implications of our results in the context of large retailers.