The tourism and hospitality sectors are recognized as significant contributors to the global economy, driving employment and contributing to national GDPs. Given their economic importance, the accuracy of demand forecasting within these sectors is crucial. These industries are subject to various fluctuating factors, including economic trends, environmental changes, and public health crisis events, necessitating forecasting methods that are both precise and flexible. This thesis aims to address this need by focusing on enhancing demand forecasting through applying Bayesian ensemble methods.

Structured with a sequence of empirical studies, the thesis begins with an in-depth investigation of the impact of time series decomposition on the efficacy of Bagging forecasting models. This initial study meticulously assesses how appropriate Bagging techniques can enhance the forecasting performance, addressing a crucial gap in the existing literature and providing a novel perspective on optimizing Bagging forecasting methodologies in the face of fluctuating tourism demand. Taking the benefit of Bagging, the subsequent two studies delve into the synergistic integration of Bagging methods with Bayesian statistical approaches. The second study proposes a Bayesian Bagging (BayesBag) method for time series forecasting, which overcomes instability and overfitting issues in tourism and hospitality demand modeling. This newly developed model is then applied to forecast international tourism demand in 10 European countries during both the pre- and post-COVID-19 pandemic periods. The third study is built upon the second study by further extending the BayesBag method into a more comprehensive Bayesian ensemble forecasting method, which incorporates the Bayesian model combination method in the aggregation step of the whole Bagging process. The novel Bayesian ensemble method is used to evaluate the policy intervention impact on Australian wine export to China, showing superior forecasting performance compared with the benchmark models. Representing the techniques and designs in the literature, these studies contribute not only to the tourism and hospitality literature but also to the general field of forecasting methodologies, providing practical implications for industry stakeholders and policymakers. By demonstrating the potential for improving accuracy through decomposition and Bayesian methods, the study advocates for the adoption of the proposed techniques to bolster strategic planning and enhance operational efficiency within the sectors. Furthermore, the thesis emphasizes the need for forecasting models to account for a wider range of external influences, from economic policies to global health crises. With these built-in external factors, the Bayesian ensemble methods are able to more accurately monitor the empirical impact of such events and provide robust reference to support timely business decisions and policy making. Ultimately, this research significantly contributes to the tourism and hospitality demand analysis both theoretically and practically.