基于Bernstein Copula函数的中国入境旅游需求预测

Forecasting Chinese Inbound Tourism Demand with Bernstein Copula Function

旅游需求的序列相关结构是旅游学研究中长期被忽略的一个问题。在旅游预测建模中,往往假定线性的或者是某种特定的非线性序列相关结构。这种假定虽然为模型构建带来一定的便捷性,但是很可能会影响预测的精确性。该研究引入Bernstein Copula函数刻画中国入境旅游需求的序列相关结构,以构建预测模型进行实证分析。实证结果表明,Bernstein Copula模型在旅游预测中具有其优越性。研究的结果为旅游需求建模提供了一个新的思考方向。

Tourism demand modeling and forecasting has long been an attractive topic in the tourism demand literature, because of its great impact on decision making of governments and tourism related businesses. Many researchers have highlighted the necessity of tourism demand forecasting. China is one of the most popular destinations in the world, and the rapid development of the tourism sector in China has caused tourism demand forecasting to become increasingly essential. This paper proposes the Bernstein Copula model as an alternative to analyze serial dependence structure of China inbound tourism demand for forecasting. Forecast endeavors should be underpinned by knowledge of serial dependence structure; however discussion of the latter has been insufficient in the tourism forecasting literature. In the traditional tourism demand forecasting model, the serial dependence structure is always been predetermined, either as the linear structure or some certain nonlinear structure. This restriction can reduce the forecasting accuracy of the traditional models. The proposed Bernstein Copula model is thus appealing, as it possesses some advanced properties which make it applicable and appealing for high dimensional associations. First of all, Bernstein polynomials are closed under differentiation, which leads to the computational convenience of the Bernstein copula for high dimensional associations. Second, any copula can be approximately represented by certain Bernstein copula with only simple restriction on the coefficients. Actually, the Bernstein copula allows for arbitrary dependence structure between dependent variables and covariates. Thirdly, different from many traditional models, the Bernstein Copula does not require the tourism demand variable follow any given distribution （usually normal distribution）. Our empirical results indicate that China inbound tourism demand follow normal distribution, but its serial dependence structure is probably nonlinear. To illustrate the benefit of using the Bernstein Copula model for tourism demand forecasting, we compare the forecasting performance of the Bernstein Copula model with those of several benchmarks, including the Seasonal Naive （S-Naive） model, the Autoregressive （AR） model, as well as the Seasonal Autoregressive Integrated Moving Average （SARIMA） model. The compare results show that the Bernstein Copula model produces smaller root mean square error and mean absolute percentage error than the three benchmarks, which indicates that the Bernstein Copula performs better in forecasting China inbound tourism demand. The contribution of this study is not introducing the Bernstein Copula model as the universally best approach for forecasting tourist demand. Instead, it contributes to the existing tourism demand and forecasting research by highlighting the importance of serial dependence structure to tourism demand forecasting. The consideration of the serial dependence structure generalizes the existing time series model into a broaden setting, in which both linear and nonlinear serial associations can be addressed and the restricted distribution assumption of the demand series involved can be released.