INTERNATIONAL JOURNAL OF TOURISM RESEARCH Int J Tourism Res 13, 1–16 (2011) Published online 26 March 2010 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/jtr.779 Taiwan’s International Tourism: A Time Series Analysis with Calendar Effects and Joint Outlier Adjustments Hui-Lin Lin1,*, Lon-Mu Liu2, Yi-Heng Tseng3 and Yu-Wen Su1 Department of Economics, National Taiwan University, Taipei, Taiwan Public Economics Research Center, College of Social Sciences, National Taiwan University, Taipei, Taiwan Department of International Business, Yuan Ze University, Chung-Li, Taiwan ABSTRACT In this paper, we examine monthly tourist arrivals from Japan, Hong Kong and the USA between January 1971 and December 2008 Our purpose is to find events or variables that affect Taiwan’s international tourism We find that the Chinese New Year has a positive effect on tourist arrivals from Hong Kong, but negative effects for other countries Through outlier detection, we obtain a better understanding of the effects of non-recurring events that have impacted Taiwan’s international tourism Using transfer function model with automatic outlier detection and adjustment, we find that the exchange rate influences tourist arrivals from Japan and Hong Kong Copyright © 2010 John Wiley & Sons, Ltd Received November 2009; Revised 22 February 2010; Accepted 23 February 2010 Keywords: international tourism; Taiwan; time series; calendar effects; outliers INTRODUCTION D uring the past 40 years, the economic structure of Taiwan (the Republic of China) has changed greatly It has evolved from a mainly agriculture-based economy in the 1970s to a technology-based economy in recent years The rapid growth of *Correspondence to: Dr H.-L Lin, Economics Department, National Taiwan University, No.21 Hsu-Chow Road, Taipei 100, Taiwan E-mail: huilin@ntu.edu.tw Taiwan’s economy inevitably gives rise to many new economic and societal issues, such as a high demand for energy, increases in air and water pollution, and an M-shaped income distribution (i.e the rich gets richer and the poor gets poorer) To partially address such societal issues, the Taiwan government has refocused its attention on international tourism, which was overshadowed by manufacturing and technology-based industries in the past More specifically, according to the statistical reports of the World Travel and Tourism Council, the annual output value of the Taiwan tourism sector amounted to US$19.7 billion in 2008 or 4.7% of Taiwan’s gross domestic product (GDP), which totalled US$419 billion According to the national statistics released by the directorate-general of Budget, Accounting and Statistics of Taiwan in 2007, the manufacturing sector accounted for 23.76% of GDP In that same year, the agriculture sector accounted for only 1.45% of GDP Relatively speaking, the tourism sector is about one-fifth of the size of the manufacturing sector and is thus already an important component of Taiwan’s economy Globally, Taiwan has a moderate ranking in tourism According to the World Economic Forum (2009), the Travel and Tourism Competitiveness Index is composed of three major subindices of the travel and tourism sector, namely, the regularity framework sub-index, the business environment and infrastructure sub-index, and the human, cultural and nature sub-index Based on the index, Taiwan was ranked 30th in the Travel and Tourism Competitiveness Index among the 124 countries reported In comparison, China was ranked 71st and Korea 42nd, while Japan was ranked 25th, Hong Kong the sixth and the USA the fifth based on this index Copyright © 2010 John Wiley & Sons, Ltd In recent decades, there has been keen interest in tourism studies in how tourism demand is affected by various cultural, economic and institutional factors, as well as major ‘one-time’ events In such studies, tourist arrivals have been the most frequently used dependent variable in quantitative analyses (e.g Martin and Witt, 1989; Kulendran and King, 1997; Song and Witt, 2006) Lim (1997) reviewed 124 tourismrelated studies and concluded that 67 of these studies used tourist arrivals and 54 used tourism expenditures as the dependent variable Lim (1997) also reviewed several commonly used explanatory variables, such as income, relative tourism prices, transportation costs, exchange rates, the time trend, seasonal factors, economic activity indicators, lagged dependent variables, marketing and promotion, as well as various qualitative factors Among such explanatory variables, dummy variables were typically used to deal with the influence of qualitative factors, including well-known factors such as seasonal variation (e.g Goh and Law, 2002; Hui and Yuen, 2002) and ‘one-time’ events (e.g Ryan, 1993; Chen et al., 1999; Goodrich, 2001; Huang and Min, 2002; Kim et al., 2006; Athanasopoulos and Hyndman, 2008) Such an approach was also used by Wang (2008) to study four major local or international disasters potentially relevant to Taiwan’s international tourism: the Asian financial crisis in 1997, the major earthquake on 21 September 1999 in Taiwan, the terrorist attacks on 11 September 2001 in the USA and the outbreak of severe acute respiratory syndrome (SARS) in 2003 In most studies, traditional regression models with dummy variables (e.g Witt and Witt, 1995; Wang, 2008) or Autoregressive Integrated Moving Average-related models (e.g Goh and Law, 2002; Chu, 2008) were typically employed Recently, a rough sets approach was used to study tourism (Goh et al., 2008) It has the advantage of being straightforward and directly interpretable It considered various economic and non-economic factors as well as month in a year However, it did not consider effects because of one-time events or calendar variation as shown in this paper In this paper, both ARIMA and transfer function time series models will be used Effects because of calendar variation are included in the models, and the onetime events are handled through automatic Copyright © 2010 John Wiley & Sons, Ltd H.-L Lin et al outlier detection and estimation in the context of time series modelling In this research, our primary interest is to study major factors or events that affect international tourism in Taiwan Such factors or events may be classified as recurring or nonrecurring in nature Both will be studied in this paper On recurring factors, besides calendar variables, we focus on investigating the impact of exchange rate as previous researches (see e.g Crouch et al., 1992; Lim, 1997) demonstrate that exchange rate has a significant influence on tourism However, they did not apply time series models using joint estimation of model parameters and outlier effects With rigorous time series analysis, these models will allow Taiwan to obtain information and knowledge to better allocate its resource for promotion and expansion of international tourism as well as providing a better ongoing tourism service Before studying international tourism in relation to Taiwan, we first provide an overview of worldwide international tourism at both the national and regional levels in Section 2: International Tourism Worldwide and Taiwan This is then followed by an introduction to the international tourist arrivals into Taiwan We have an extensive collection of monthly tourist arrivals data into Taiwan from various countries and regions between 1971 and 2008, with each series having 456 observations In Section 3: Time Series Models for The Analysis of Taiwan's Tourism, Box–Jenkins time series models with calendar effects are introduced The parameters of such models are estimated using a joint estimation method of model parameters and outlier effects in Section 4: Analysis of Calendar Effects The effects of recurring and non-recurring events are presented and discussed in that section as well In Section 5: The Effects of Exchange Rates on Taiwan's Tourism, we examine the effects of foreign exchange rates on international tourist arrivals into Taiwan from major countries In Section 6: Discussion and Conclusion, we provide a discussion as well as the conclusion to this paper INTERNATIONAL TOURISM WORLDWIDE AND TAIWAN Even though our primary interest is to study international tourism in Taiwan, it is Int J Tourism Res 13, 1–16 (2011) DOI: 10.1002/jtr Taiwan’s International Tourism Table International tourist arrivals into various regions and countries Country Worldwide Euro area USA ASEAN China Japan Taiwan 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 538 991 962 573 555 988 599 266 326 614 287 034 639 895 637 688 000 108 683 748 532 703 790 318 691 603 848 763 037 433 803 813 233 850 778 039 908 000 000 201 613 206 207 430 975 220 470 384 233 882 000 245 313 000 260 090 915 258 187 000 262 947 000 259 565 000 260 115 000 268 505 000 284 903 000 315 935 000 43 490 000 46 636 000 47 875 000 46 377 000 48 509 000 51 238 000 46 927 000 43 581 000 41 218 000 46 086 000 49 206 000 50 978 000 55 986 000 23 999 000 24 611 000 23 939 000 22 872 000 26 964 000 30 860 000 33 875 000 35 800 000 30 801 000 40 916 000 43 495 000 48 189 000 54 068 000 20 034 000 22 765 000 23 770 000 25 073 000 27 047 000 31 229 000 33 167 000 36 803 000 32 970 000 41 761 000 46 809 000 49 913 000 54 720 000 345 000 837 000 218 000 106 000 438 000 757 000 772 000 239 000 212 000 138 000 728 000 334 000 347 000 331 934 358 221 372 232 298 706 410 248 624 037 617 137 602 037 248 117 950 342 378 118 519 827 716 063 ASEAN, Association of Southeast Asian Nations important to have a good understanding of international tourism worldwide Table lists the annual tourist arrivals worldwide as well as the tourist arrivals in several important regions and countries Except for Taiwan, tourist arrivals for the different regions and counties are obtained from the World Tourism Organization The annual tourist arrival data for Taiwan are provided by the Tourism Bureau, Ministry of Transportation and Communications, China To facilitate a better understanding of tourism growth in each region/country, annual tourist arrivals are indexed to 1995 levels (i.e the numbers in 1995 are set to 100) and displayed in Figure (A,B) The tourist arrival indices for the world as a whole and the USA are displayed in Figure (A,B) to facilitate the visual comparison From Table and Figure 1, we find that the Euro area accounts for more than one-third of worldwide international tourism each year and that the USA accounts for roughly 6–8% of worldwide tourism However, the growth of international tourist arrivals in the Euro area, along with the growth in the USA, has slowed substantially in recent years International tourism has grown at a significantly faster rate in Asia, including China, Japan and the Association of Southeast Asian Nations (ASEAN) area, despite the 11 September terrorist attacks in 2001 and the SARS epidemic in 2003 While the growth of international tourism in Taiwan has been smaller in comCopyright © 2010 John Wiley & Sons, Ltd parison with that in other countries or regions, the pace seems to have picked up following the SARS epidemic in 2003 In this study, our primary interest is to study major factors or events that affect international tourism in Taiwan Based on the total tourist arrivals data for 2008, the international tourists visiting Taiwan came primarily from the following five regions or countries: Japan (28.3%), Hong Kong (16.1%, including Macao), the ASEAN area (16.7%), the USA (10.1%) and Europe (5.2%) In Figure (A), the total tourist arrivals in each month between January 1971 and April 2008 are displayed From this graph, we find that the total tourist arrivals exhibit a general upward trend While this trend was severely affected by the SARS epidemic in 2003, it resumed with higher growth following the SARS outbreak As the total number of tourist arrivals is an aggregate of many time series, its properties are harder to interpret and less meaningful in their application To improve our study, tourist arrivals from major countries are displayed in Figure (B–D) The solid lines in Figure (B–D) represent monthly tourist arrivals from Japan, Hong Kong and the USA, the three principal sources of international tourism for Taiwan, and it is these that are the primary focus of this study The dashed lines in Figure (C,D) represent monthly tourist arrivals from the European and ASEAN areas As these two series are also an aggregation of tourist arrivals Int J Tourism Res 13, 1–16 (2011) DOI: 10.1002/jtr H.-L Lin et al Tourist Arrivals (Indexed to 1995) 280 (A) Worldwide United States China 240 Japan Taiwan 200 160 120 80 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 280 Tourist Arrivals (Indexed to 1995) (B) Worldwide Euro Area 240 ASEAN United States 200 160 120 80 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Figure International tourist arrivals into various regions and countries ASEAN, Association of Southeast Asian Nations from various countries, they are not our focus in this research study and are provided for information only Tourist arrivals from Japan, Hong Kong and the USA were obviously impacted by the SARS epidemic in 2003 However, their historical temporal patterns are quite different The average number of tourist arrivals from Japan is much higher than that from Hong Kong and the USA, but its growth rate has been much smaller than the corresponding growth rates for the other two areas in recent years The numbers of tourist arrivals from the USA are Copyright © 2010 John Wiley & Sons, Ltd much smaller than the corresponding numbers of arrivals from Japan and Hong Kong, but they display a persistent upward trend The numbers of tourist arrivals from Japan and Hong Kong have sometimes declined or have remained the same for extended periods of time As in the cases of many other tourist arrival time series, the numbers of international tourist arrivals in Taiwan seem to fluctuate seasonally We display the average monthly tourist arrivals into Taiwan from Japan, Hong Kong and the USA, as well as the total international Int J Tourism Res 13, 1–16 (2011) DOI: 10.1002/jtr Taiwan’s International Tourism 400000 (A) 350000 Tourist Arrivals Worldwide 300000 250000 200000 150000 100000 50000 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 140000 (B) Tourist Arrivals 120000 Japan 100000 80000 60000 40000 20000 1972 1974 70000 (C) Tourist Arrvials 60000 Hong Kong Euro Area 50000 40000 30000 20000 10000 1972 70000 Tourist Arrivals 60000 1974 1976 1978 1980 (D) United States ASEAN 50000 40000 30000 20000 10000 1972 1974 1976 1978 1980 Figure Monthly international tourist arrivals into Taiwan (1/1971–12/2008) ASEAN, Association of Southeast Asian Nations Copyright © 2010 John Wiley & Sons, Ltd Int J Tourism Res 13, 1–16 (2011) DOI: 10.1002/jtr H.-L Lin et al 200000 100000 80000 150000 60000 100000 40000 50000 20000 Average Monthly Tourist Arrivals into Taiwan Average Monthly Tourist Arrivals into Taiwan (Worldwide) 6 Worldwide Japan 10 Hong Kong United States 11 12 Month Figure Average monthly tourist arrivals into Taiwan (1/1971–12/2008, excluding the year 2003 because of the severe acute respiratory syndrome epidemic) tourist arrivals into Taiwan in Figure Note that the data for 2003 are excluded from the monthly averages because the SARS epidemic affected international tourism severely in that year Except for October and December, the seasonal pattern for total Taiwan international tourist arrivals is very similar to the seasonal pattern for Japan as Japan is the major source of Taiwan’s international tourism In the case of Japan, the numbers of tourist arrivals are higher between January and March, lower between April and October (with July the lowest in a year), become higher in November and decline to a lower level in December This pattern is rather different from the international tourist arrivals in the USA and Europe where the summer months and Christmas period tend to have higher numbers of tourist arrivals The monthly tourist arrival pattern for Japan may, to a large degree, be related to the differences in climate between Taiwan and Japan The climate in Taiwan between January and March is much more temperate than that in Japan and is thus more appealing to Japanese tourists The summer months (particularly between June and October) in Taiwan are much hotter than in Japan and are thus less appealing to Japanese tourists The climate in Taiwan in November and December may be warmer than in Japan, but Taiwan seems to Copyright © 2010 John Wiley & Sons, Ltd lose Japanese tourists to the USA/Europe in December As for Hong Kong and the USA, the tourist arrival patterns are somewhat different from that for Japan In these two areas, the summer months (June to August) and December continue to have relatively high numbers of tourist arrivals into Taiwan, and the tourist arrivals in October are particularly high because of the most important governmentsponsored national holiday celebration that is held on 10 October each year TIME SERIES MODELS FOR THE ANALYSIS OF TAIWAN’S TOURISM International tourist arrivals may be affected by external factors that can be classified as recurring variables and non-recurring events Non-recurring events, such as the 11 September terrorist attacks and the SARS epidemic, can only be represented by discrete indicator variables Recurring variables such as exchange rates and other economic variables are data collected systematically and can be represented by various forms of time series As tourist arrivals in Taiwan are compiled as monthly data, tourist arrivals may be influenced by calendar variation Calendar variation is recurring in nature, and it is very important to account for its effects in the analysis of monthly Int J Tourism Res 13, 1–16 (2011) DOI: 10.1002/jtr Taiwan’s International Tourism time series We shall discuss two important kinds of calendar effects below In Taiwan, all official monthly economic statistics are compiled according to the Gregorian calendar, just as in all other countries in the world However, the dates of all traditional Chinese festivals and holidays are set according to the Chinese lunar calendar The parallel use of these two calendars gives rise to substantial issues in the analysis of monthly time series data for these countries The most problematic issue is that where important traditional festivals or holidays fall during different Gregorian calendar months from year to year For example, even though the Chinese New Year is always on the first of the first month each year according to the Chinese lunar calendar, it may fall in either January or February based on the Gregorian calendar As tourist arrivals may be greatly affected by the Chinese New Year, the observed time series may vary substantially, depending on whether a particular month (January or February) includes the Chinese New Year or not Such effects are referred to as moving-holiday effects (Liu, 1980, 1986, 2006) In addition to moving holidays, the number of tourist arrivals may depend on the days of the week As the composition of days of the week varies from month to month and year to year, the observed series may be affected by such variation as well Such effects, which are by and largely because of the composition of trading days (or work days) in each month, are referred to as trading-day (or working-day) effects (Hillmer et al., 1981; Hillmer, 1982; Bell and Hillmer, 1983) A general time series model for tourism analysis Assuming that Yt is a time series that may be subject to the influences of recurring variables and non-recurring events, a general time series model for Yt can be written as Yt = C + f (ω , X t ) + N t , N t = θ ( B) at , ( D B) φ ( B) at ∼ i.i.d N (0, σ a2 ) , t = 1, 2, , n (1) where B is the backshift operator (i.e BYt = Yt−1), C is a constant term, f (ω , X t ) represents Copyright © 2010 John Wiley & Sons, Ltd the total exogenous effects at time t, D(B) is the differencing operator, φ(B) is the autoregressive operator, θ(B) is the moving average operator, n is the number of observations, and at’s are independently and identically distributed (i.i.d.) in normal distribution with mean and variance σ a2 The operators D(B), φ(B) and θ(B) can be expressed in simple or multiplicative form as shown in Box and Jenkins (1976) The function f (ω , X t ) can be either in linear or nonlinear form In this study, we consider a class of linear and non-linear dynamic relationship functions (often referred to as transfer functions) described in Box and Jenkins (1976) Using the terminology of transfer function modelling (Box and Jenkins, 1976; Liu, 2006), Nt is referred to as the disturbance or noise of the model In the above model, X t contains variables X1t, X2t, , Xmt that are used to characterise the effects because of various recurring variables, and ω is a vector of parameters reflecting the effects of such variables Even though the effects because of non-recurring events (e.g the SARS epidemic, 11 September attacks, etc.) may be included in f (ω , X t ) with the X it ’s being indicator variables, it is more flexible to treat such events as outliers (Fox, 1972; Chang et al., 1988) Using an estimation procedure developed by Chen and Liu (1993), we can automatically detect outliers (nonrecurring events) and perform joint estimations of the outlier effects and model parameters Such an approach allows us to account for the effects of both known and unknown nonrecurring events more effectively Model (1) can also be expressed in the following alternative form D (B ) Yt = C′ + D (B ) f ( ω , X t ) + N ′t , θ (B) N ′t = at φ (B) (2) where the differencing operator is applied to both response and explanatory variables The latter form of model is used in the estimation of the model parameters Time series models with calendar effects We are interested in economic variables or certain tourism-related events that may affect tourist arrivals to Taiwan from a prospective Int J Tourism Res 13, 1–16 (2011) DOI: 10.1002/jtr H.-L Lin et al country/city To study the effects of such variables or events, we must include important calendar effects in the model first Calendar effects are of interest in this study themselves Furthermore, they also may be viewed as important ‘nuisance’ effects that must be taken care of before further econometric modelling is conducted In addition to Chinese New Year (referred to as H1t), there are two other major moving holidays that potentially may affect Taiwan international tourism These are the Dragon Boat Festival (H2t) and the Autumn Moon Festival (H3t) The Dragon Boat Festival is on May of the Chinese lunar calendar and may vary between May and June of the Gregorian calendar The Autumn Moon Festival is on 15 August of the Chinese lunar calendar and may vary between September and October of the Gregorian calendar A model with such moving-holiday effects can be expressed as f (α , β1 , γ , H1t , H 2t , H 3t ) = α 1H1t + β1H 2t + γ 1H 3t (3) if such moving-holiday effects are the same over years (i.e staying constant), where the variable Hit (i = 1,2,3) represents the proportion of a particular holiday in the t-th month Here, we assume that extra tourist arrival changes (either increases or decreases) because of Chinese New Year are distributed uniformly during a 10-day period beginning days prior to the New Year and days during the festival As for the Dragon Boat and Autumn Moon Festivals, we assume that the tourist changes are distributed uniformly during a 5-day period beginning days prior to the festival and days over the duration of the festival The assumptions for the length and effect distribution of the festivals are not crucial as most of the festivals fall in the same months instead of splitting across two adjacent months If these moving-holiday effects increase (or decrease) linearly over the years (i.e having an upward or downward trend), then the following model may represent the effects more adequately: f (α 1, α , β1, β , γ 1, γ , H1t , H 2t , H 3t , K t ) = α 1H1t + α H1t × K t + β1H 2t + β H 2t × K t + γ 1H 3t + γ H 3t × K t Copyright © 2010 John Wiley & Sons, Ltd where Kt is for all Kt in the first year, for all Kt in the second year and so on As for trading-day effects, the following model may be considered f ( ξ1 , , ξ7 , W1t , , W7t ) = ∑ ξ i Wit (5) i =1 where Wit, i = 1, , , represent the number of Mondays, Tuesdays, and Sundays in the t-th month, respectively, and ξi, i = 1, 2, , are the effects because of Monday, Tuesday, and Sunday To avoid multicollinearity, it is desirable to restrict trading-day effects to vary around zero, or equivalently imposing ξ1 + ξ2 + + ξ7 = Thus the model in Equation (5) can be written as f (δ1 , , δ , D1t , , D 6t ) = ∑ δ i D it (6) i =1 where Dit = Wit − W7t and δi = ξi, i = 1, 2, , are the effects because of Monday, Tuesday, , Saturday, and the effect for Sunday is (−ξ1 −ξ2 −ξ6) In addition to D1t, , D6t, Hillmer (1982) and Bell and Hillmer (1983) include an additional term δ7D7t in Equation (6), where D7t = W1t + W2t + + W7t is the length of a month The interpretation of δ7 depends on the form of a model For a stationary time series, the δ7 parameter represents the average of daily effects and is used to adjust for the length of a month A similar interpretation holds if only the first-order differencing operator (1–B) is present in the model However, when the model includes the seasonal differencing operator (1–B12), the parameter δ7 reflects the effect because of leap year that may or may not be important, and may be omitted from the model in some situations The model in Equation (6) implies that the trading-day effects are constant over time If the trading-day effects increase (or decrease) linearly from year to year, then the following model may be more appropriate: f (δ1 , , δ , λ , , λ , D1t , , D7t , K t ) (4) 7 i =1 i =1 = ∑ δ i D it + ∑ λ i ( D it × K t ) (7) Int J Tourism Res 13, 1–16 (2011) DOI: 10.1002/jtr Taiwan’s International Tourism where λi represents the linear trend for the day of the week Combining the model in Equations (4) and (7), a calendar effect model (abbreviated as f(•)) that includes both moving-holiday and trading-day effects can be expressed as f (•) = α 1H1t + α H1t × K t + β1H 2t + β H 2t × K t + γ 1H 3t + γ H 3t × 7 i =1 i =1 (8) K t + ∑ δ i D it + ∑ λ i ( D it × K t ) Using the model identification method described in Liu (1986, 2006), we find that the disturbance term (Nt) in Equation (1) can be represented as 12 ˆ t = (1 − θ1 B ) (1 − θ12 B ) a t N (1 − B ) (1 − B12 ) (9) for the monthly tourist arrivals from Japan, Hong Kong and the USA as well as worldwide monthly totals ANALYSIS OF CALENDAR EFFECTS To conduct a rigorous examination of the potential calendar effects on tourist arrivals into Taiwan from various countries (Japan, Hong Kong and the USA) and worldwide, we employ the calendar effect model with a trend in Equation (8) and the disturbance model in Equation (9) Both the original and logtransformed series are examined The results for these two scales (original and logtransformed) are largely consistent and will be presented using one or the other for the purpose of simplifying the interpretation and in order to increase clarity As the tourist arrival time series in Taiwan are all subject to outliers (e.g the SARS epidemic in 2003), the outliers must be identified, and their effects must be jointly estimated with model parameters By using the joint estimation method described in Chen and Liu (1993) and the Scientific Computing Associates Statistical System (Liu and Hudak, 1992), we find that the trading-day effects are not significant, and only the Chinese New Year has a significant impact on international tourism among the three moving-holidays discussed above Furthermore, the Chinese New Year effects can be simply represented by the trend parameter (α2) as the intercept parameter α1 is insignificant Thus, the above calendar effects model can be simply expressed as Yt = α H1t × K t + N t (10) where Nt is the ARIMA model shown in Equation (9) In the table below, we list the model parameter estimates obtained by the joint estimation method of Chen and Liu (1993), where the critical value 4.0 is used for outlier detection Thus, major outliers such as those because of the SARS epidemic are automatically detected and adjusted during the joint estimation of model parameters and outlier effects Here, a larger critical value for outlier detection is used as the series are long and we are only at the stage of obtaining appropriate model parameter estimates The number of outliers detected using this critical value for each model is reported at the end of each row in Table A smaller critical value for outlier detection will be used later when we try to detect nonrecurring events in the time series More of the details are discussed later in this section The results in the above table show that, except for Hong Kong, international tourist arrivals decrease during the Chinese New Year period, particularly for tourists arriving from Japan Chinese New Year is the most important holiday for families to get together during the year Therefore, hotels are primarily booked Table Parameter estimates of models α2 Japan Hong Kong USA Worldwide −479.52 (t = −12.29) 233.90 (t = 8.93) −6.13 (t = −0.41) −398.09 (t = −5.31) Copyright © 2010 John Wiley & Sons, Ltd θ1 0.48 (t = 10.93) 0.64 (t = 16.49) 0.70 (t = 19.88) 0.49 (t = 11.58) θ12 σa Number ofoutliers 0.56 (t = 13.91) 0.56 (t = 13.04) 0.64 (t = 16.29) 0.67 (t = 17.68) 4834.36 2981.74 1582.69 8686.26 10 Int J Tourism Res 13, 1–16 (2011) DOI: 10.1002/jtr Chinese Tourists’ Satisfaction with Hong Kong 87 HTS1 Satisfaction with Hotels HTS2 HTS3 STS1 STS2 Satisfaction with Retail Shops Overall Satisfaction STS3 TTS1 TTS2 Satisfaction with Tour Operators TTS3 Figure The overall tourist satisfaction index model Note: HTS, STS and TTS refer to the measurement indicators of tourist satisfaction with hotels, retail shops and tour operators, respectively three indicators means that a satisfaction index can be assigned to each corresponding tourist Such an individual satisfaction index variable can further facilitate market segmentation analysis In addition, the sectoral TSI framework is readily applicable to the product level (e.g a hotel) once a sub-sample has been extracted based on the name of the service provider encountered by a respondent Crossproduct comparisons within a service sector are thus possible, although such analysis is beyond the scope of the current study The second step in the proposed satisfaction evaluation system is to estimate the overall TSI for the destination based on the sectoral TSI results obtained in the first step The weights for the overall TSI are retrieved by employing a second-order confirmatory measurement model Second-order confirmatory factor analysis provides a structure by which the firstorder factors become manifest variables of the second-order factors Such analysis explicates Copyright © 2010 John Wiley & Sons, Ltd the pattern of constructs assumed to underlie the variables, in other words, how each firstorder factor (sector) contributes to the higherorder factor (destination) In this model (see Figure 2), the constructs are formed by respondents’ level of satisfaction with individual sectors, each of which is measured by the three satisfaction items In turn, the factor loadings indicate the contributions made by sectoral satisfaction to overall satisfaction; hence, they are adopted as the weights for estimating the overall TSI Given the objective weights obtained from the second-order confirmatory factor analysis of destination satisfaction, this procedure has a strong scientific basis, which in turn guarantees the robustness of the overall TSI estimation The weights are then introduced in the second equation to calculate the overall TSI As explained above, tourist satisfaction within a particular tourism sector equals the weighted average of the mean values of its three Int J Tourism Res 13, 82–96 (2011) DOI: 10.1002/jtr 88 H Song et al satisfaction indicators (y–31, y–32 and y–33) multiplied by a scaling constant of 10 Thus, each TSI is expressed on a 0–100 scale The higher tourists’ average scores on the satisfaction questions, the greater the calculated sectoral TSI The overall TSI, as the weighted average of the sectoral TSIs, is given in Equation (2), where the γi-values are the factor loadings derived from the confirmation measurement model for the overall TSI calculation m Overall TSI = ∑ γ i Sub-TSIi i =1 m ∑γ i (2) i =1 The proposed two-step TSI framework has the following four distinct advantages: (1) Tourist satisfaction within each individual sector is measured by the same set of indicators Thus, the TSIs at the sectoral level are directly comparable (2) The overall TSI is computed on the basis of the sectoral TSIs using a weighting scheme that is determined by the tourists’ own evaluations As a result, free-of-charge and other public services can be included in its compilation (3) Tourist satisfaction assessment at both the sectoral and destination levels can be estimated, thereby providing more comprehensive information for destination management (4) The proposed TSI system can be used repeatedly to capture the dynamics of tourist satisfaction, thus reflecting changes in a destination’s competitiveness over time RESULTS Hong Kong was selected as the destination for this investigation because of its status as a well-developed, mature tourism destination Kozak (2001) has found that the relationship between overall satisfaction and loyalty is stronger for mature destinations than for less familiar destinations Loyalty is an important outcome of the current framework, as well as a desirable strategic result for many tourism operators and regulators The mainland Chinese source market is the target population Copyright © 2010 John Wiley & Sons, Ltd in this pilot study, and any competitive advancement in this segment should be of interest to both the government and service industry in Hong Kong Because the primary purpose of this study was to test the validity and reliability of the proposed two-step framework, only three key tourism-related sectors were included in the survey (hotels, local tour operators and retail shops), and, accordingly, only these sectors are included in the computation of the overall TSI to demonstrate the aggregation method A face-to-face street intercept survey employing a self-administered questionnaire facilitated the data collection, which took place over days in November 2008, and a convenience sampling method was adopted at a variety of locations in Hong Kong The final sample included 279 valid responses Respondents’ details are presented in Table The SmartPLS 2.0 M3 (University of Hamburg, Hamburg, Germany) software program (Ringle et al., 2005) was employed for the first step of the analytical procedure As tourist surveys are usually subject to non-responses, missing values ([...]... Entrepreneurship Theory and Practice 22(4): 5–21 Yoon S 2 002 The antecedents and consequences of trust in online-purchase decisions Journal of Interactive Marketing 16(2): 47–63 Int J Tourism Res 13, 17–31 ( 2011 ) DOI: 10.1 002/ jtr INTERNATIONAL JOURNAL OF TOURISM RESEARCH Int J Tourism Res 13, 32–40 ( 2011 ) Published online 14 July 2010 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1 002/ jtr.795... level of significance within the accepted range (