【作者】 刘玮；

【导师】 刘汉；

【作者基本信息】 吉林大学 ， 数量经济学， 2020， 硕士

【摘要】 旅游需求预测对旅游业的从业者在投资决策以及产业政策制定方面有着至关重要的作用。在当前网络信息飞速发展的时代,潜在游客在出游前通常会通过网络来获取当地相关信息,并籍此合理安排行程。这些搜索行为信息为研究人员了解游客们的潜在需求提供便利,能够进一步改善旅游需求预测。此外,有研究表明网络搜索数据(如百度指数)作为一种新的数据来源,相较于传统的数据而言有着相当大的改进,能够很好地预测旅游需求。但由于网络搜索数据变量普遍较多,如何选取合适的模型以利用此类数据预测旅游需求值得探讨。因此,本文介绍了两类可以利用百度指数来预测旅游需求的方法——机器学习技术及传统因子模型,并将其应用于大陆对香港旅游需求的预测,进行对比研究。首先,本文探讨并确定了旅游六要素为预测指标,随后提取大陆到访香港人数和旅游六要素相关的百度指数;其次,将预测模型分为随机森林和因子増广回归模型两类,其中因子増广回归模型又包含着两种方法,即广义动态因子模型和主成分分析法。利用这两类模型三种方法,来预测平稳增长阶段和快速增长阶段的大陆对香港的旅游需求;最后基于方向性预测精度检验、DM检验以及模型置信集检验法对不同模型的不同阶段预测效果进行评估。本文实证结果表明:(1)相对于因子分析模型,随机森林在平稳的预测区间对于季节性趋势转换的拐点预测有着较强的优势,可以更快的把握住趋势的切换。对于旅游需求预测的研究,展现出良好的方向性预测功能。(2)当旅游需求处于快速增长阶段时,广义动态因子模型在信息提取方面更加有优势。(3)在相同的样本规模及旅游需求阶段下,主成分分析得到的静态因子没有显示出较大的优势,在预测趋势方面逊色于随机森林算法,预测精度层面稍劣于广义动态因子増广回归模型。(4)由于旅游需求的影响因素关系错综复杂,若在行业情况不明,未来趋势不清晰的情况下,随机森林模型的操作简单方便更加适用于预测。总之,精确地预测旅游需求一方面可以保证资源的高效配置和优质服务,有助于及时调整相关产品或服务的供给,进而避免供需失衡。另一方面,旅游需求预测中得到的诸如旅游需求弹性的信息,有助于旅游业的管理者以及投资人员迅速做出反应,及时安排日程和人员配置、编写旅游宣传册等,使得旅游产品多样化。此外,旅游需求影响因素众多且繁杂,故准确预测旅游需求是一项有意义且具有挑战的研究。更多还原

【Abstract】 Tourism demand forecasting plays a crucial role for tourism practitioners in investment decisions and industrial policymaking.In the current era of the rapid development of network information,potential tourists usually use the network to obtain local relevant information before traveling,and use this to reasonably arrange itineraries.This search behavior information facilitates researchers ’understanding of tourists’ potential needs and can further improve tourism demand forecasts.In addition,some studies have shown that search query data(such as the Baidu Index),as a new data source,has a considerable improvement over traditional data and can predict travel demand well.However,due to many variables in the Internet search data,how to choose a suitable model to use this data to predict tourism demand is worth exploring.Therefore,this paper introduces two types of methods that can use Baidu index to predict tourism demand: machine learning technology and traditional factor models,and applies them to the forecast of Hong Kong’s tourism demand in the mainland,and conducts comparative studies.First,this article explores and determines the six factors of tourism as predictive indicators,and then extracts the Baidu Index related to the number of mainland visitors to Hong Kong and the six factors of tourism.Second,the prediction model is divided into two categories: random forest and factor auto regression model.The factor auto regression model also includes two methods,namely the generalized dynamic factor model and the principal component analysis method,which are used to predict the slow-growth and fast-growth mainland’s tourism demand for Hong Kong.Finally,the prediction effects of different models at different stages are evaluated based on the directional accuracy test,DM test,and model confidence set test.The empirical results in this paper show that:(1)Compared with the factor analysis model,the random forest has a significant advantage in the inflection point prediction of seasonal trend transition in a stable prediction interval,and can grasp the trend switching faster.The research on tourism demand forecasting shows a good directional forecasting results.(2)When tourism demand is at a rapid growth stage,the generalized dynamic factor model is more advantageous in terms of information extraction.(3)Under the same sample size and tourism demand stage,the static factors obtained by principal component analysis did not show significant advantages.They were inferior to the random forest algorithm in predicting trends and slightly worse than the generalized dynamic factors in the prediction accuracy level model.(4)Due to the intricate relationship between the factors affecting tourism demand,the random forest model is more suitable for prediction if the industry situation is unknown and the future trend is unclear.On the one hand,accurately forecasting tourism demand can ensure the efficient allocation of resources and the safety of high-quality services,and help adjust the supply of related products or services on time to avoid imbalances between supply and demand.On the other hand,the information such as the elasticity of tourism demand obtained in the forecast of tourism demand helps the managers and investors of the tourism industry to respond quickly,arrange schedules and staffing in time,and write tourism brochures,which diversify the tourism products.In a word,due to the numerous and complicated factors affecting tourism demand,accurately predicting tourism demand is a meaningful and challenging study.更多还原