This dissertation presents two studies of the forecast of occupancy in the United States’ hotel industry. The first is a quantitative study of the forecast accuracy performance of moving average, simple exponential smoothing, additive, and multiplicative Holt-Winters method, and Box-Jenkins forecasting procedures on weekly aggregated occupied room data from 10 geographic markets in the United States. In addition, this researcher also examined the performance of combined forecasts. The additive Holt-Winters method was found to be the most accurate in forecasting in seven of the 10 markets, even though it was not the most accurate in the training set. In three of the markets, the seasonal autoregressive integrated moving average method produced the highest level of accuracy.

The second study is a qualitative study designed to understand how the sample of revenue management experts uses their tacit knowledge of future demand in specific markets to modify statistically based forecasts of hotel occupancy. The researcher interviewed revenue managers. Four of these were working on a revenue management team, which supported groups of franchised hotels for a major global brand. These managers worked directly with the multiple hotels they supported in their assigned geographies. The remaining six revenue managers were located on the property they supported. Two of these managers also supported one or more properties in their geographic area in addition to their property. Marriott International, Hilton Worldwide, Starwood Hotels and Resorts Worldwide, Intercontinental Hotels Group, and Wyndham Hotels and Resorts were in the sample. The revenue managers oversaw the revenue management function in the limited and select service, full service, and luxury quality tiers.

Each of the revenue managers did use external sources of information to adjust forecasts based upon their local markets; however, there was little training or consistency in how this process occurred. This results in a sub-optimal situation in which the knowledge, skills, and abilities in the application of expert judgement vary widely. There appears to be no consistent process, training, or knowledge transfer capabilities in place for this human element.

This presents an opportunity for forecast accuracy improvement across each of the major brands represented in the sample. Much of the literature has demonstrated that rule-based forecasting results in more accurate forecasts, particularly when there is good domain knowledge and that knowledge has a significant impact (Armstrong, 2006). Standardizing practices that result in greater accuracy and creating a more robust structure across brands could prove to be quite beneficial.