Y = f(X, P, D)
where Y is the tourism receipts of the province; X is the disposable incomeof the tourists visiting the province; P is the price of tourism goods and servicesin the province; and D is a dummy variable indicating the high-speed railway.We use only domestic tourism receipts as the measure of the high-speedrailway’s tourism impacts on the three provinces. The exclusion of internationaltourism receipts is based on the rationale that, compared with the volume ofnational domestic tourism receipts, which totalled around US$30.60 billion,US$27.27 billion and US$62.41 billion in 2011 in Hubei, Hunan andGuangdong, respectively, that of international revenue is quite small (at US$0.94billion, US$1.04 billion and US$13.91 billion in 2011, respectively), thereforeexerting limited influence on the overall picture.Meanwhile, the disposable income of the domestic tourists to the province(X) is measured as the average of those of the top three origins. In accordancewith criteria set by Dwyer et al (2000), the consumer price index (CPI) in theprovince is used as a proxy for the price of tourism goods and services in theprovince (P). The data above can be obtained from the Yearbook of China TourismStatistics, as well as the Annual Report of Economic and Social Developments of therespective provinces. Finally, a value of 0 is assigned to D for the time seriesperiod before the operation of the high-speed railway, and a value of 1 isassigned to D for the period after the start of operation.Specifically, the regression model of this study is constructed as follows:log(Yt)= β0 + β1log(Xt) + β2CPIt + β3log(Xt)*D
Taking cognizance of its peculiarities, this study employs the auto regressiveand moving average (ARMA) model to refine the time series relationshipsbetween the dependent and independent variables. ARMA is based on theassumption of a stationery time series with ‘smooth’ fluctuations along a time-invariant mean (Chatfield, 2004). It consists of three components of auto
Impacts of the high-speed train on China’s tourism demand163
regression, individual dependence and moving average (Neuman, 2000). ARMAis considered suitable for prediction uses with desired accuracy and scaleparsimony, particularly for estimating the significance of certain variables to thesystem (Monserud, 1986). Furthermore, ARMA contributes to model perfectionby removing the effects of potential persistence factors irrelevant to the variablesunder investigation (Chatfield, 2004). Since the introduction of high-speedrailway can be regarded as a ‘shock’ event to the tourism development of theinvestigated localities, AMRA is deemed as an appropriate statistic platform tocompare the domestic tourism receipts of the three provinces before theintroductionof the high-speed railway with those after the railway introduction,from which the effects of the high-speed railway can be delineated. For themodel testing, the 8.0 version of the software of Statistic Analysis System (SAS)is utilized.
Results
First, a linear test is conducted to explore the regressive nature of the proposedrelationships as follows:
yt = a + bxt + xt2 + ... + xtn + et.
Results of the linear testing are presented in Table 1. According to correspond-ing criteria concerning the value of Dw, the relevance of time series data forHubei province is rejected, indicating little difference between the period beforethe introduction of the high-speed railway and the period after. In other words,the impact of the high-speed railway on domestic tourism receipts of Hubeiare indistinct. The linear regression model for Hubei province is confirmed asfollows:
log(Yt) = –1.0978 + 1.3526log(Xt) – 0.0421 CPIt + 0.071log(Xt)*D(–0.47) (3.41) (–3.43) (5.94).
In contrast, for the provinces of Hunan and Guangdong, respective Dw valueshave exhibited satisfactory time series relevance (sig < 0.05). Thus, the high-speed railway has exerted a distinct influence on the domestic tourism receiptsof these two provinces. To further delineate the influence, an ARMA model isapplied to the series of the linear regression model as follows:
Table 1.Results of linear testing.
Dw
Hunan
Hubei
Guangdong1.62172.27921.6434Significance0.04630.70140.0457R20.98310.78110.9999F247.5252.3511747.2
164TOURISM ECONOMICS
xt = φ0 + φ1xt–1 + φ2xt–2 + Λ + φpxt–p + εt φp ≠ 0 E(εt) = 0, Var(εt) = σ2ε, E(εtεs) = 0, s ≠ t Exsεt = 0, s < t.
After the integrated calculation, the respective final linear regression modelsfor Guangdong and Hunan provinces are presented as follows:
Guangdong:
log(Yt)=0.70817log(Xt)–0.00084CPIt+0.04937log(Xt)*D+0.13833AR(4)(7.69) (–0.12) (7.34) (3.84).
Specifically, from January 2008 to December 2009:
log(Yt) = 0.70817log(Xt) – 0.00084CPIt + 0.13833AR(4).
From January 2010 to December 2011:
log(Yt) = 0.75754log(Xt) – 0.00084CPIt + 0.13833AR(4).
Hunan:
log(Yt) = 2.6431 + 0.64601log(Xt) – 0.00154CPIt + 0.0608log(Xt)*D+ 0.12643AR(4)