‘Estimating tomorrow’s tourist arrivals’: forecasting the demand for China’s tourism using the general-to-specific approach

Abstract

Accurate forecast of inbound tourism demand is vital for the tourism industry as well as government economic policy and decision making. This article sought to identify the factors which influence the demand for China’s tourism with the aid of econometric models and to generate forecasts of international tourist arrivals to China from five major long-haul source markets. Using the general-to-specific modelling approach, the demand for tourism in China by the residents of Australia, Canada, Germany, the United Kingdom and the United States of America is modelled and forecasted. The empirical results indicate that the ‘word of mouth effect’, income levels in the origin country, the costs of tourism in both China and competing destinations are the crucial factors that determine the demand for China’s tourism by residents of the five origin countries. The forecasts show sluggish growth in tourist arrivals for most of the Western source markets. Findings hold implications for policy formulations.

Keywords

econometric models income modelling price elasticity tourism demand

Introduction

AS one of the fastest growing inbound markets in the world, China’s tourism economy has progressed remarkably over the past two decades. The country ranks fourth in the world in terms of international tourist arrivals and fifth in receipts (World Tourism Organization [UNWTO], 2009a). The potential of growth is still high; as indicated by forecasts from the UNWTO (2001), China will become the largest tourist country and the fourth largest for overseas travel by 2020.

The ever-increasing role of tourism in the generation of wealth and employment cannot be over emphasised. In 2009, the central government of China listed tourism on the top of its agenda of the national economic and social developments. By emphasising tourism as a strategic pillar industry in the national economy, the overall plan calls for further expansion of the tourism market and seeks to make China the world’s flagship Tourism destination by 2020. With an increase of 8 per cent per year, the inbound market is projected to rise to 90 million tourist arrivals. The expected results include 12 per cent annual growth rate in tourism revenues, adding 4.5 per cent gross domestic product (GDP) to the national economy by 2015 as well and creating half a million jobs every year (UNWTO, 2009b).

To be able to realise these monumental goals, accurate forecasts of inbound tourism demand for China are critical for government economic policy and decision making as well as the tourism industry. In this study, forecasting models are developed for tourist arrivals in China from five ofits major long-haul tourism-generating countries: Australia, Canada, Germany, the United Kingdom (UK), and the United States of America (USA). The estimated models are then used to generate forecasts of this demand for the period beginning in the first quarter of 2010 to the last quarter of 2015.

The model

Tourism demand is a complex phenomenon which can be influenced bynumerous exogenous factors – economy, natural disasters, destination image, political situation, etc. Consequently, it is one of the most difficult variables to be predicted. Nevertheless, it is becoming more and more critical for both destinations and private sectors to anticipate demand trends and use such information as a basis of management decisions and planning. Song et al. (2003b) note that published studies on tourism demand analysis can be put into two major categories: those that employ non-causal modelling techniques and those using causal (econometric) approaches.

The non-causal forecasting approaches (most especially, time series models) have been used extensively, over the past four decades, for tourism demand forecasting (Song and Li, 2008). Smeral and Wüger (2005), Du Preez and Witt (2003), Goh and Law (2002) and Gustavsson and Nordstrő(2001) are researchers who employed time series approaches such as the integrated autoregressive moving-average model (ARIMA) or seasonal ARIMA. Though this technique is less costly in terms of data collection and model estimation, it has some major limitations. The most important being that the models are not constructed on the basis of any theory underlining tourists’ decision-making process and so cannot be utilised for the purpose of policy evaluation (Song et al., 2003b).

In contrast, econometric models (which are mainly causal models) play many other useful roles in addition to forecasting (Clements and Hendry, 1998). The ability to examine the causal relationships between the tourism demand variable and its determining factors is considered to be one of its major advantages over the time series approach. Song et al. (2003b) emphasise two reasons why econometric approaches are superior to time series models: they offer researchers with valuable insights into the tourists’ decision-making process and second, its model specification permits forecasters to assess the direction and magnitude of tourists’ response to changes in the determining factors (through an examination of the estimated demand elasticities). Recent studies using econometric approaches include Song and Witt (2006), Li et al. (2006), Witt et al. (2004), Song and Witt (2003), Song et al. (2003a, b), Kulendran and Witt (2003), Tan et al. (2002) and Song and Witt (2000), among others.

Determinants of tourism demand

The demand for tourism is modelled to follow the standard economic theory; own price of a good, the price of a substitute good and consumers’ income constitute the key factors determining the demand of a consumer good. The following equation is thus proposed to represent the demand for China’s tourism by residents from each origin country i:

(1)

where Q_it is the tourism demand variable, measured by tourist arrivals from origin country i to China (where i = 1, 2, 3, 4, 5 represent Australia, Canada, Germany, UK, and USA, respectively) at time t; P_it is the relative consumer price level for tourists from origin country i to China at time t; P_st is the price of tourism in the substitute destination at time t and Y_it is the income of the origin country at time t; and e_it is the residual term used to capture the influence of all other factors that are not included in the model. Hence, it is used to account for the other economic and non-economic factors which influence tourism demand but were not included due to data unavailability.

The data for the dependent variable Q_it (measured by tourist arrivals from country i) were retrieved from the Yearbook of China Tourism Statistics (CNTA, 2011). The income variable, Y_it (representing income in the origin country i), measured by the real GDP (2005 = 100), was inferred from the International Financial Statistics Yearbook published by the DSI Data Service & Information (2010) database. GDP, instead of personal disposable income, is used because the tourist arrival figures include a substantial proportion of business travellers. Song et al. (2003b) and Song and Witt (2003) followed similar approaches and established the significance of GDP in a number of origin countries. It is assumed that this variable will positively influence the demand for China’s inbound tourism.

The own price variable P_it is measured by the consumer price index in China relative to the consumer price index in the origin country, adjusted by the exchange rates between the currencies of China and its origin country, and is thus given as:

(2)

where CPI_cn and CPI_i are the consumer price indices for China and origin country i, respectively; EX_cn and EX_i the exchange rate indices (2005 = 100) for China and the origin country i, respectively. The own price variable include only the living cost components as a result of unavailability of travel cost data. Previous studies revealed the insignificance of travel costs in several tourism demand models (Witt and Witt, 1992; Kim and Song, 1998).

The substitute price variable P_st refers to the weighted average (based on market shares) of the consumer price levels for tourists (measured by the exchange rate-adjusted consumer price index) in five major competing destinations in Asia. It is represented mathematically as:

(3)

where j = 1, 2, 3, 4 and 5 representing Hong Kong, Japan, South Korea, Singapore and Taiwan, respectively; w_j is the share of the international tourist arrivals for country/region j, which is calculated from

w_{j} = T T A_{j} / (\sum_{j = 1}^{5} T T A_{j})

and TTA_j is the total international tourist arrivals in country/region j.

The exchange rates were obtained from the International Financial Statistics Yearbook. The same applies to the consumer price indices except those of Germany which were retrieved from the database of the Organisation for Economic Cooperation and Development (2010). For the substitute prices, the data of tourist arrivals from specific origin country were adopted to calculate the share of the tourism market.

Song and Li (2008) and Witt and Witt (1992) provide further justification for the explanatory variables included and the forms of variables used. The data for other variables that may influence tourism demand (such as the change of consumer taste towards China tourism) were excluded either because of their unavailability or measurement difficulties. As shown in the following section, this however did not affect the goodness-of-fit of the estimated models (the R² indicates very high significance).

Specification of econometric model

The power function in Equation (1) is used in model estimation for the following reasons: first, most previous empirical studies suggest that tourism demand can be modelled better by the power function than the simple linear demand function in terms of model’s statistical significance and forecasting ability (Witt and Witt, 1992; Song et al., 2003b); second, the power function can be transformed into a log-linear specification, which can easily be estimated with ordinary least squares (OLS). The estimated coefficients of the explanatory variables in the log-linear model can be interpreted directly as demand elasticities. From Equation (1), the general tourism demand model is now specified as the following (after taking the logarithm):

(4)

where β₀ = ln A, u_it = ln e_it and β₁, β₂ and β₃ are price, income and substitute price elasticities, respectively; it is expected that β₁ < 0 (the price of tourism would have a negative influence on tourism demand), while β₂ and β₃ > 0 (the income level of the origin country and the price of tourism in the substitute destinations would have positive impacts on tourism demand). In the latter case, complimentary effect is assumed to be the situation where β₃ < 0 (the price of tourism in the substitute destinations has negative impact on tourism demand).

Since the model is to be used for analysing and forecasting tourism demand, it must reflect the dynamic feature of tourists’ decision process. The autoregressive-distributed lag model (ADLM) (Hendry, 1995; Pesaran and Shin, 1995), which was introduced to tourism forecasting by Song and Witt (2000), is employed in this study to capture the dynamic process of tourism demand. The ADLM for Equation (4) is specified as follows:

(5)

As shown in Equation (5), the current tourism demand is influenced by current values of the explanatory variables as well the lagged-dependent and explanatory variables. The specification takes into account the time path of tourists’ decision-making process. Since quarterly data are used in the model estimation, four time lags were introduced for each variable. Song et al. (2003b) cite two reasons why it is necessary to include lagged dependent variables on the right-hand side of the demand model: the effect of ‘word of mouth’ recommendations and the tendency for tourists to return to previously visited destination because they are less uncertain about that destination.

Since the coefficients in Equation (5) are not demand elasticities likethe coefficients of β₁, β₂ and β₃ in Equation (4), some algebraic manipulations are necessary. If the long-run equilibrium is assumed, Equation (5) can be rewritten as:

(6)

where

are price, income and substitute price elasticities, respectively.

A testing down procedure, termed as the general-to-specific procedure (Hendry, 1995), is adopted to eliminate variables which are either insignificant or economically (or theoretically) unacceptable. The test procedure begins with an estimation of Equation (5) using OLS to check the statistical significance of all the variables. The statistically insignificant variables are then eliminated one-by-one from the specification in accordance with the t-statistics of the estimated coefficients starting with the least significant ones. After getting rid of all the insignificant variables and the variables with incorrect signs from the specification, the model is tested with a number of diagnostic statistics to ensure that there is no misspecification in the model. Song and Witt (2000, pp. 34–40) discusses the required diagnostic statistics, which include the tests for heteroscedasticity, autocorrelation, normality and forecasting ability. Based on this methodology, empirical results are presented in the following sections.

Estimates of the demand models

Australia, Canada, Germany, the UK and the USA are important long-haul source markets for China tourism. Effectively modelling and accurately forecasting the demand for China tourism by residents of these countries can provide some useful information for strategy formulation for both the public and private sectors.

To account for seasonal variations in the quarterly data used, three seasonal dummies were included in estimating Equation (5). In addition, a number of dummy variables were incorporated to capture the influence of one-off events on the demand for China tourism. These were the 4 June, 1989 Tiananmen Square Protest (D89) (negative effect expected in 1989), the Gulf War in 1990–1991 (D91) (negative effect expected in 1991), the financial crises in 1996 (D96) and 2008 (D08) (negative effect expected in the respective years), the ‘September 11’ terrorist attack on the USA (D01) (negative effect expected in 2001), SARS in 2003 (D03) (negative effect expected in 2003) and the initial recovery from the global financial crisis (D09) (positive effect expected in 2009). Each of these dummy variables takes the value of 1 in the respective quarter(s) where the event was expected to have effect and 0 otherwise. As a result, the initial ADLM now becomes:

(7)

This model is first estimated for all the five countries, and the general-to-specific procedure is followed to achieve the final specific models. Table1 presents the estimated final models for the five origin countries.

Table 1.

Estimates of demand models (1988Q1–2009Q4)

	Australia	Canada	Germany	UK	USA
Intercept	−1.7675	−2.8697	−7.2529	−1.9566	−6.1188
	(−3.6843)**	(−5.6563)**	(−5.7274)**	(−4.9828)**	(−18.3514)**
ln Q_it₋₁	0.8292	0.7161	0.757	0.7732	0.3026
	(22.1728)**	(15.8056)**	(12.1755)**	(20.1323)**	(9.1723)**
ln Q_it₋₂			−0.1789
			(−2.6443)*
ln Q_it₋₃			0.1511
			(2.7207)**
ln Y_it			1.8538
			(5.8846)**
ln Y_it₋₁	3.1085	5.1475
	(2.9995)**	(3.5788)**
ln Y_it−₂	−2.5381	−6.5093		0.6743
	(−2.5275)*	(−3.3894)**		(5.4987)**
ln Y_it₋₄		2.2763			2.2419
		(3.1163)**			(20.5069)**
ln P_it				−0.1015
				(−2.4396)*
ln P_it₋₂			−0.2778
			(−3.4765)**
ln P_it₋₃		−0.2712			−0.4648
		(−3.9039)**			(−6.8222)**
ln P_it₋₄	−0.0967
	(−1.9904)*
ln P_st₋₁			−0.5316		0.2477
			(−3.5400)**		(3.0839)**
ln P_st₋₂				0.1194
				(2.2783)*
ln P_st₋₃		−0.9392
		(−2.0147)*
ln P_st₋₄		1.2097
		(2.5609)*
D1	−0.1723	−0.0923	−0.0832	−0.1326	−0.1514
	(−9.7278)**	(−1.8979)	(−3.1477)**	(−8.7849)**	(−7.6921)**
D2	0.1212	0.3102	0.0584	0.0719	0.0581
	(2.3178)*	(2.8608)**	(2.0894)*	(4.4107)**	(2.7741)**
D3	−0.0916	−0.0494	−0.0836	−0.0139	−0.0346
	(−5.0825)**	(−1.8478)	(−3.1034)**	(−0.8918)	(−1.7767)
D89	−0.4887	−0.3933	−0.3003	−0.4283	−0.4884
	(−11.4469)**	(−7.6648)**	(−3.8788)**	(−14.4624)**	(−15.0783)**
D91		−0.1833	−0.4067
		(−2.7959)**	(−4.5568)**
D96	−0.2186
	(−3.7563)**
D01					−0.1714
					(−4.2358)**
D03	−1.0033	−0.8884	−1.0526	−0.8531	0.8612
	(−20.6204)**	(−15.9699)**	(−12.7948)**	(−50.8735)**	(65.0786)**
D08	−0.1619	−0.1669		−0.0779	−0.1501
	(−3.0163)**	(−3.6279)**		(−2.2588)*	(−2.6024)*
D09			0.1224
			(2.4171)*
R ²	0.9958	0.9953	0.9876	0.9932	0.9945
Adjusted R²	0.9952	0.9943	0.9854	0.9923	0.9937
F-statistic	1569.0891	1036.8416	435.8441	1099.2050	1185.5800
Jarque–Bera test	4.3070	0.7881	1.7798	2.7611	0.0969
LM test	12.3401	12.3027*	1.5209	4.5179	6.3803
White test	11.4960	27.0096*	31.6534**	12.8191	12.8508
Ramsey regression equation specification error test (RESET)	1.6016	5.0425	4.7121	3.1455	0.2500
Chow forecast	5.8882	3.2591	NA	5.0790	3.6040

Note: The figures in parentheses are t-statistics.

The Breusch–Godfrey LM test is used for testing serial correlation at lag 4. The Jarque–Bera test is used for testing normality. The White test is used for testing heteroscedasticity. The Ramsey’s regression equation error test (or Ramsey RESET test) is used to test the functional form. The Chow forecast is used to test forecasting failure. However, since the dummy variable representing the recovery from the global financial crisis from the year 2009 is included in the demand model, this statistic cannot be calculated.

** and * represent 1 per cent and 5 per cent significant levels, respectively.

Noteworthy from the results is the very high significance of the lagged dependent variables (for ln Q_it) in all the five models. This is a clear indication that the demand for China tourism is dependent upon previous visits. In other words, China’s international tourism demand from each of the five source markets is highly influenced by the ‘word of mouth’ effect and/or consumer persistence (or repeat visits). It is, therefore, critical for China to provide high-quality services in order to attract new and repeat tourists. Another important determinant is the price of China’s tourism product/service. It appears to be significant in each of the five models, though not to the same degree. The income variable is highly significant in all the models, suggesting that the income level of the origin country is a key factor that influences the demand for international tourism to China. In comparison to the income and price variables, the substitute price appears to be less influential in the demand for China tourism. Nonetheless, its value in all but one model (Australia) cannot be ignored. It is important to note that most of theseasonal dummies were very significant, indicating the importance of seasonal variations in the demand for China’s tourism. The important implication of this finding is for China’s tourism industry to innovatively diversify its tourism resources to attract more tourists in the low seasons in order to avoid redundancies.

The Tiananmen Square Protest in 1989, the Gulf War in 1990–1991, SARS in 2003 and the financial crises in 1996 and 2008 all appear to have had negative impacts on China tourism demand. The SARS and the Tiananmen Square Protests, in particular, significantly reduced tourist arrivals from all five countries. The impact of the financial crises in 1996 on tourist arrivals was significant for only the Australian market, whereas those of 2003 were influential in the demand for China tourism from four of the source markets. The 11 September, 2001 terrorist attack on the USA was significant in reducing tourism demand from the USA in 2001.

Validity of the models

Looking at the adjusted R² and the F-statistic, all the five demand models passed the goodness-of-fit test. The diagnostic statistics indicate that the model for the demand from Australia passed all the tests, while those of Canada and Germany failed one or two tests. The Durbin–Watson statistics and Lagrange multiplier (LM) test are used to check the autocorrelation problem but the former is biased towards 2 when a lagged dependent variable is included as an explanatory variable in the model. The White test is used to test heteroscedasticity and the model for the demand from Germany failed this test. This is probably due to the inconsistency of the data set resulting from the reunification of West Germany and East Germany in 1990. None of the models failed the normality tests. The model regarding the demand from Canada failed both the LM and White tests.

Elasticity of demand

On the basis of the demand models in Table 1, the demand elasticities were computed and tabulated in Table 2. The importance of demand elasticities for policymakers and business planners cannot be over emphasised.

Table 2.

Estimated demand elasticities

Model	Income elasticity	Own price elasticity	Cross-elasticity
Australia	3.5170	−0.5148	–
Canada	3.2211	−0.9551	0.9527
Germany	6.8461	−1.0260	−1.9632
UK	2.9729	−0.4475	0.5263
USA	3.2146	−0.3112	−0.1829

Price elasticity has a direct impact on tourist revenues. Knowledge of price elasticity can, therefore, aid suppliers of tourism products/services in China to make appropriate adjustment to their prices so as to maximise the total tourism revenue. As shown in Table 2, tourists from Germany and Canada appear to be more sensitive to the price changes of the tourism product/services in China, compared to the other three source markets (i.e. Australia, the UK and the USA).

The value of the income elasticity is an indication of the responsiveness of the tourism demand to the change in income level in the origin country. Thus, if the tourism product in China is income elastic (>1) for tourists from a particular region, an increase in income level of that country would result in more than proportionate increase in the demand for China tourism by residents from that country. To deal with the fluctuations in the demand for China tourism, it is critical for China to accurately predict the business cycles related to its key tourism origin countries. All the five models generate very high-income elasticities, suggesting that the demand for China’s tourism by tourists from these five countries is highly dependent upon the income levels of their respective countries of origin. The tourism demand from Germany seems to be most remarkably influenced by the economic conditions in that country.

The importance of the price elasticities for substitute destinations is seen in its ability to indicate how much can be generated when there is a price change in competing destinations. Tourists originating from Canada appear to be very much aware of the costs of tourism in the alternative destinations; consequently, a change in the cost of holidaying in the substitute destinations will have a major impact on the demand for China tourism. The important implication of this finding is for China to maintain the cost advantage of its tourism over competing destinations in order to keep attracting tourists from Canada.

It must, however, be noted that although the cross-elasticity for the Germany module indicates high elasticity, it is negative. This means tourists from Germany perceive China’s competing destinations in the sub-region as complimentary. In other words, an increase in the price of tourism in the competing destinations will result in a decrease in the demand for China’s tourism by German tourists. This calls for collaboration with competing destinations in the sub-region in an effort to attract more tourists from Germany.

Forecasts

Tourist arrivals for the period beginning in the first quarter of 2010 to the last quarter of 2015 are forecasted using the estimated demand models presented in the previous section. In doing this, the explanatory variables for the forecast period were first predicted using the Holt–Winter exponential smoothing approach. This approach is easier and also capable of producing reliable forecasts for the explanatory variables in tourism demand models (Song and Witt, 2000). After generating the explanatory variables, they were put into the forecasting models given in Table 1. The forecasts of the dependent variables were then calculated on the basis of the models’ estimated parameters. The actual value of the lagged Q_it was used for the initial observation in the forecast sample. The forecasts for subsequent periods then used the previously forecasted values of ln Q_it. The anti-log of the forecasted values of tourist arrivals is then taken to obtain the actual tourist arrivals (this is because the variables in the demand models are all in algorithm).

The projected quarterly and annual tourist arrivals from the five long-haul tourist-generating countries are presented in Table 3. The forecasts indicate that, among the five source markets for China’s tourism, Australia will be the fastest growing with an average annual growth rate of 11.65 per cent, followed by Canada (10.70 per cent). The second quarter will experience the highest growth rate (12.14 per cent) for the Australian market, while the highest quarterly growth rate will appear in the first quarter (11.17 per cent) for the Canadian market. The number of Australian travellers to China is expected to increase from 1,488,000 in the first quarter of 2010 to 3,128,000 in the last quarter of 2015. Likewise, arrivals from Canada will reach 3,265,000 in the fourth quarter of 2015, from 1,496,000 in the first quarter of2010. Tourist arrivals from the three remaining markets are likely to haveslower growth rates ranging from 3.37 per cent to 5.42 per cent. TheGerman market will experience the slowest growth (3.37 per cent) with the fourth quarter recording the least increase in the number of arrivals. The number of tourist arrivals from Germany is therefore expected to register a lethargic growth from 1.39 million in the first quarter of 2010 to 1.84 million in the last quarter of 2015. Since the tourism product in China is income elastic, the expected sluggish growth in demand from the three Western countries is attributable to the slow recovery from the global financial crises, which have negatively affected the projected values of the GDP.

Table 3.

Quarterly forecasts (‘000) of tourist arrivals in China from five long-haul source markets (2011Q1–2015Q4)

Quarters	Australia	Canada	Germany	UK	USA
2010Q1	148.8	149.6	138.7	132.9	430.4
2010Q2	161.4	164.4	154.3	144	550.9
2010Q3	171.6	175	155.6	149.3	518
2010Q4	179.7	196.2	160.1	148.9	510.2
2011Q1	173.9	145.9	127.6	120.2	379.7
2011Q2	190.6	158.2	142.6	136.5	499.5
2011Q3	191.5	159.7	136.9	137.9	473.3
2011Q4	211.1	185.4	140.2	144.9	492.4
2012Q1	195.9	166.1	135.2	130.3	419.1
2012Q2	213.7	181.3	150.2	149.5	546.8
2012Q3	213.8	183.4	144.6	150.6	515.8
2012Q4	234.9	213.1	150.5	157.8	535.8
2013Q1	217.4	191.1	146.8	141.7	455.9
2013Q2	236.7	208.8	164.5	162.3	594.7
2013Q3	236.4	211.2	157.8	163.3	560.9
2013Q4	259.3	245.6	162.6	171.1	582.7
2014Q1	239.6	220.3	157.8	153.5	495.8
2014Q2	260.6	240.7	176.3	175.9	646.8
2014Q3	260.0	243.5	168.7	176.9	610.1
2014Q4	285.0	283.1	173.4	185.3	633.8
2015Q1	263.3	254.0	168.0	166.2	539.3
2015Q2	286.2	277.5	187.3	190.4	703.5
2015Q3	285.4	280.8	179.1	191.5	663.5
2015Q4	312.8	326.5	184.0	200.6	689.3
2010	661.5	685.2	608.7	575.1	2009.5
2011	767.0	649.2	547.3	539.5	1844.9
2012	858.3	743.9	580.5	588.2	2017.5
2013	949.7	856.7	631.7	638.4	2194.3
2014	1045.3	987.6	676.1	691.5	2386.5
2015	1147.6	1138.9	718.3	748.7	2595.5
Q1 (%)	12.09	11.17	3.90	4.58	4.61
Q2 (%)	12.14	11.04	3.96	5.75	5.01
Q3 (%)	10.71	9.92	2.85	5.10	5.08
Q4 (%)	11.72	10.72	2.82	6.14	6.20
AAGR (%)	11.65	10.70	3.37	5.42	5.25

Note: (1) AAGR denotes the average annual growth rate over 2010–2015 calculated by $(\sqrt[s]{Q_{2015} / Q_{2010}} - 1)$ . (2) Figures in italics are forecasts.

The models are estimated using data from 1988Q1 to 2009Q4. The period from 2010Q1 to 2010Q4 is selected as the ‘holdout’ sample. Smaller mean absolute percentage error (MAPE) and root mean square percentage error (RMSPE) values indicate a better forecasting model, which produces more accurate values of predictions. As presented in Table 4, the five models developed in this study seem to be very competitive in terms of forecasting ability as the MAPE and RMSPE values for the model of Australia are less than 5 per cent (indicative of a highly accurate forecasting ability), while the MAPE and RMSPE values for the models of Canada, UK and USA are between 10 and 20 percentage points, signifying that the models have good forecasting abilities. Though the performance of the model for Germany was relatively worse than the four (MAPE and RMSPE values were around 23 per cent), it is still a reasonable forecasting performance.

Table 4.

Forecast evaluation

Measure	Australia	Canada	Germany	UK	USA
MAPE (%)	1.94	14.45	22.69	11.56	16.27
RMSPE (%)	2.35	15.60	23.74	11.70	16.38

Conclusion

As a complex phenomenon which can be influenced by various factors, the modelling and forecasting of tourism demand is becoming more crucial for both destinations and private sectors. Using the general-to-specific approach, this study modelled and forecasted the demand for China’s tourism by residents from Australia, Canada, Germany, the UK and the USA. Through a careful statistical testing procedure, the best models that passed both statistical and economic tests were chosen for forecasting. The MAPE and RMSPE values demonstrate that generally the models possess good forecasting abilities.

From the estimates of the demand models, the ‘word of mouth’ effect or the behavioural persistence of tourists is the most influential factor in the demand for tourism in China from the five long-haul source markets. The suppliers of tourism products/services therefore need to enhance service quality and promote their brand images in order to attract more tourists to China.

Although the price of tourism in China is significant in all the five models, its elasticity is most crucial with respect to the demand for China’s tourism by residents from Germany. Thus, it is still important for China to pursue cost advantages over competing destinations.

The income level of the origin country is another important determinant of China’s inbound tourism. The income elasticities ranged from 2.973 to 6.846 suggesting wide differences among origin countries with respect to the responsiveness of tourism demand to changes in the income conditions in those countries. Demand for China tourism from residents of Germany, in particular, is greatly influenced by the income levels in that country. It behoves on policy makers in China to closely monitor the economic cycles in the source markets.

Though not as influential as the other factors, the price of tourism in competing destinations cannot be ignored. It is particularly important for tourists originating for Canada and Germany. Whereas an increase in the cost of tourism in substitute destinations will result in more tourists from Canada visiting China, it will also lead to a decline in tourist arrivals from Germany. In other words, tourists from Germany perceive substitute destinations as complimentary. The policy implication of this is that, in order to attract more tourists from Germany to China, players in the industry should collaborate with their counterparts in competing destinations within the sub-region when it comes to pricing and promotion of destinations.

In addition, the significance of seasonal variations in the estimated models calls for diversification and other innovative strategies by both service providers and destination marketers in China in an effort to minimise redundancy in low seasons.

Among the five long-haul source markets, tourist arrivals from Australia are expected to experience the highest growth rates over the forecasting period. Most of the source markets in the Western Hemisphere are predicted to record slow growth rates, with the German market experiencing the slowest growth. Considering the fact that the demand for China’s tourism is highly income elastic, this trend could be attributed to the global financial crisis which is projected to affect GDP in those countries. Policy makers and tourism service providers in China need to consider creative ways to make up for the expected sluggish growth from these source markets. The generic strategies and diversification approaches employed would be very crucial in the quest for achieving the UNWTO target of China becoming the largest tourist country and the fourth largest for overseas travel by 2020.

The forecasts have a few limitations; however, because of the high income elasticity of China’s tourism, the accuracy of the forecasts depends very much on the accuracy of projected GDP. Considering the uncertainties about the recovery process of the global financial crises, the real GDPs can vary. This may subsequently affect the accuracy of the forecasts though the models may be valid. Also, any unforeseen event in the near future could have an effect on the performance of the models since they were not considered in the forecasting process. The above forecasts were generated using econometric models based on the historic trends of tourism demand in China over a period of 22 years (1988–2009). The results are, thus, vulnerable to the potential problem of small sample bias.

Footnotes

Notes

Julian K. Ayeh is a PhD student at the School of Hotel and Tourism Management of the Hong Kong Polytechnic University, Hong Kong.

Shanshan Lin is a PhD student at the School of Hotel and Tourism Management of the Hong Kong Polytechnic University, Hong Kong.

References

Clements

Hendry

(1998) Forecasting Economic Time Series. Cambridge: Cambridge University Press.

CNTA (2011) China Tourism Administration The Yearbook of China Tourism Statistics (1988–2010). Beijing: China Tourism Publishing House.

DSI Data Service & Information Statistical Database (2010) International Financial Statistics Yearbook (Latest edition), http://195.145.59.167/ISAPI/LogIn.dll/login?lg=e, accessed 20 March 2010.

Du Preez

Witt

(2003) Univariate versus multivariate time series forecasting: An application to international tourism demand. International Journal of Forecasting 19(3): 435–451.

Goh

Law

(2002) Modelling and forecasting tourism demand for arrivals with stochastic non stationary seasonality and intervention. Tourism Management 23(5): 499–510.

Gustavsson

Nordström

(2001) The impact of seasonal unit roots and vector ARMA modelling on forecasting monthly tourism flows. Tourism Economics 7(2): 117–133.

Hendry

(1995) Dynamic Econometrics: An Advanced Text in Econometrics. Oxford: Oxford University Press.

Kim

Song

(1998) Empirical analysis of demand for Korean tourism: A cointegration and error correction approach. Tourism Analysis 3(1): 25–41.

Kulendran

Witt

(2003) Leading indicator tourism forecasts. Tourism Management 24(5): 503–510.

10.

Wong

Song

Witt

(2006) Tourism demand forecasting: A time varying parameter error correction model. Journal of Travel Research 45(2): 175–185.

11.

Organisation for Economic Cooperation and Development (OECD) (2010) Statistics from A to Z – Beta version, http://www.oecd.org/document/0,3746,en_2649_201185_46462759_1_1_1_1,00.html, accessed 1 June 2010.

12.

Pesaran, MH and Shin Y (1995) An autoregressive distributed lag modelling approach to cointegration Analysis, Cambridge Working Papers in Economics, Faculty of Economics, University of Cambridge, http://econpapers.repec.org/RePEc:cam:camdae:9514, accessed 20 March 2010.

13.

Smeral

Wüger

(2005) Does complexity matter? Methods for improving forecasting accuracy in tourism: The case of Australia. Journal of Travel Research 44(1): 100–110.

14.

Song

(2008) Tourism demand modelling and forecasting—A review of recent research. Tourism Management 29(2): 203–220.

15.

Song

Witt

(2000) Tourism Demand Modelling and Forecasting: Modern Econometric Approaches. Cambridge: Pergamon.

16.

Song

Witt

(2003) Tourism forecasting: The general-to-specific approach. Journal of Travel Research 42(1): 65–74.

17.

Song

Witt

(2006) Forecasting international tourist flows to Macau. Tourism Management 27(2): 214–224.

18.

Song

Witt

Jensen

(2003a) Tourism forecasting: Accuracy of alternative econometric models. International Journal of Forecasting 19(1): 123–141.

19.

Song

Wong

KKF

Chon

(2003b) Modelling and forecasting the demand for Hong Kong tourism. International Journal of Hospitality Management 22(4): 435–451.

20.

Tan

AYF

McCahon

Miller

(2002) Modelling tourist flows to Indonesia and Malaysia. Journal of Travel and Tourism Marketing 13(1/2): 61–82.

21.

Witt

(1992) Modeling and Forecasting Demand in Tourism. London: Academic Press.

22.

Witt

Song

Wanhill

(2004) Forecasting tourism generated employment: The case of Denmark. Tourism Economics 10(2): 167–176.

23.

UNWTO (2001) World Tourism Organization Tourism 2020 Vision Vol. 3 East Asia & Pacific (English Version), http://www.wtoelibrary.org/, accessed 21 June 2010.

24.

UNWTO (2009a) World Tourism Organization UNWTO World Tourism Barometer 7(3): October, http://www.unwto.org/facts/eng/pdf/barometer/UNWTO_Barom09_3_en.pdf, accessed 20 March 2010.

25.

UNWTO (2009b) World Tourism Organization Milestone in China’s Tourism Policy, Media, http://www.unwto.org/media/news/en/features_det.php?id=5271&idioma=E, accessed 20 March 2010.