Abstract

Tourism can reflect the living standards and social development of a country. Predicting and accurately forecasting tourism demand can not only ensure efficient resource allocation and safe high-quality services, but also help timely adjust the supply of related products or services to avoid imbalance between supply and demand. In order to obtain higher economic benefits, the forecast of tourism demand is particularly important. Forecast models chosen by early scholars often ignore relevant variables which have indicative roles in tourism demand. Moreover, these models do not make full use of relevant information to improve the forecast effect. In recent years, with the popularity of the Internet,search behavior and attention to destination before travel have grown closely related to the tourism demand of a given region. Therefore, the Baidu Index data which reflect the search behavior and attention to destination can be used to forecast regional tourism demand. Since the data frequency of the Baidu Index data often differ from regional tourism demand, traditional models cannot forecast regional tourism demand with mixed- frequency data. Doing so may lead to loss of information or inflated information, and then result in inaccurate analysis results and forecast. Hence, in order to overcome the limitations of the same-frequency model which can only be built with same-frequency data, according to the modeling theory of the mixed-frequency model, eight weekly Baidu Index data reflecting tourism demand for Sanya have been selected to construct univariate MIDAS models and multivariate MIDAS models respectively to forecast the monthly tourism demand for Sanya. The forecast results show that: first, since the Baidu Index has certain indications for tourism demand, the addition of Baidu Index data can improve the forecast effect; second, since the mixed- frequency model can make full use of data information, the forecast effect of the mixed- frequency models are more accurate than that of same-frequency models, and the univariate MIDAS models with different weekly Baidu Index data have different forecast effects. Furthermore, since multivariate models combine more information from Baidu Index data, in general, the forecast results of the multivariate models are better, and the forecast results of the C- MIDAS models with different weighting are different. This also proves that the combination of principal component analysis and mixed- frequency forecast can further improve the forecast effect when the number of forward forecast steps is small. At the same time, the MIDAS model can now forecast regional tourism demand according to the newly released weekly Baidu Index data. Through the combination of the Baidu Index and the mixed- frequency model, it is possible to forecast the number of tourists in Sanya for July and August 2018. The forecast results are in line with reality, showing that the number of tourists in Sanya is still growing at a higher rate than 10%. The MIDAS model can make real- time regional tourism demand forecasts more accurate. Therefore, the combination of the Baidu Index and the mixed- frequency model provides new ideas for the forecast of regional tourism demand, as well as a decision-making basis for tourism departments, which can now forecast regional tourism demand in advance to realize vigorous and healthy development of regional tourism.

Key words

Baidu Index; MIDAS Model; Sanya Tourism Demand; Mixed-frequency Forecast

        DOI: https://doi.org/10.19765/j.cnki.1002-5006.2019.10.014