Enhancing the accuracy of revenue management system forecasts: The impact of machine and human learning on the effectiveness of hotel occupancy forecast combinations across multiple forecasting horizons

Abstract

Reporting on three separate studies in the context of hotel revenue management systems, this article explores the interaction between two established methods of accuracy enhancement—forecast combinations and learning. In line with theoretical considerations, our empirical investigation suggests that as learning occurs, the capacity of combinations to improve forecast accuracy diminishes in scenarios where the combined elements are independent of each other. Conversely, in the more realistic typical scenario of user overrides of system forecasts, where the elements of the combinations are dependent, the learning-driven efficacy of forecast combinations appears to vary across forecasting horizons. We find no impact of learning on combination effectiveness in the shorter forecasting horizons of 21 days or less and a surprisingly positive impact in the longer horizons. This counterintuitive finding has important practical implications for hotel revenue management practices.