Tourism has become one of the biggest economic sectors in the world and it already supports one in every ten jobs globally. The need for accurate forecasts of tourism demand has long been recognised by policy makers and practitioners. This thesis intends to identify several opportunities in tourism demand forecasting and new approaches are proposed.
Search query data have recently been used to forecast tourism demand. Linear models, particularly autoregressive integrated moving average (ARIMA) with exogenous variable models, are often used to assess the predictive power of search query data. However, they are limited by their inability to model nonlinearity due to their pre-assumed linear forms. Artificial neural network models could be used to model nonlinearity, but mixed results indicate that their application is not appropriate in all situations. Therefore, this study proposes a new hybrid model that combines the linear and nonlinear features of component models. The model outperforms other models when forecasting tourist arrivals in Hong Kong from mainland China, thus demonstrating the advantage of adopting hybrid models in forecasting tourism demand with search query data.
Furthermore, the high-frequency nature of search query presents the opportunity of the use of mixed-frequency methods. Generalized dynamic factor model (GDFM) and mixed-data sampling (MIDAS) model are employed to investigate the benefits of utilization of high-frequency information. In addition, to overcome the limitations of traditional MIDAS models, a new model is proposed to integrate MIDAS and seasonal ARIMA (SARIMA) process. The forecasting results suggest that this new model significantly outperforms the benchmark model confirming its superiority. The nowcasting results based on daily data are compared with the forecasts. The results show that nowcasts generally outperform the forecasts.
The third empirical study investigates whether pooling can improve the forecasting performance of tourism demand models. The short-term domestic tourism demand forecasts for 341 cities in China using panel data (pooled) models are compared with individual ordinary least squares (OLS) and naïve benchmark models. The pooled OLS model demonstrates much worse forecasting performance than the other models. This indicates the huge heterogeneity of tourism across cities in China. A marked improvement with the inclusion of fixed effects suggests that destination features that stay the same or vary very little over time can explain most of the heterogeneity. Adding spatial effects to the panel data models also increases forecasting accuracy, although the improvement is small. The spatial distribution of spillover effects is drawn on a map and a spatial pattern is recognised. Finally, when both spatial and temporal effects are taken into account, pooling improves forecasting performance.
Past tourism demand forecasting studies mainly focus on point forecasts, with a few exceptions on interval forecasts. Density forecasts can generate predictive distributions which give a complete probabilistic description of the possible future realizations. Therefore, density forecasts are more informative than point and interval forecasts. However, no study has evaluated and compared density forecasts from alternative models in the tourism context. This study uses proper scoring rules to evaluate the performance of density forecasts generated by several popular time series models. The Continuous Rank Probability Score (CRPS) also enables a direct comparison between point and density forecasts. Using the quarterly tourist arrivals data to Hong Kong from ten major inbound markets, the empirical results suggest that density forecasts perform better than point forecasts evaluated in terms of the CRPS. In terms of the rankings of density forecasts generated by popular time series models, the SARIMA model performs best overall. The innovations state space models for exponential smoothing and structural time series model perform poorly and are significantly outperformed by SARIMA over all forecast horizons. Bootstrap is found to improve the performance of density forecasts for short-term forecasts.
|Date of Award||8 Mar 2019|
- Univerisity of Nottingham
|Supervisor||Chang LIU (Supervisor) & Song Haiyan (Supervisor)|
- Tourism demand
- Search query
- Hybrid model
- Mixed frequency
- Panel data
- Density forecasts