Modelling tourism competitiveness in small Pacific island countries

Abstract

This study models overall and bilateral tourism competitiveness in small Pacific island countries (PICs), namely, Cook Islands, Fiji, Tonga, Samoa and Vanuatu. The pooled mean group approach, which corrects for cross-sectional dependence and non-stationarity, is used for estimation with quarterly data from 2002 to 2019. The findings indicate that for Fiji and Vanuatu, other PICs are competing destinations and that Fiji and Vanuatu face the strongest bilateral competition amongst the selected PICs. Cross-price elasticities are insignificant for Tonga and are generally negative for the Cook Islands and Samoa. Thus, while for Fiji and Vanuatu, the Cook Islands is a competing destination, Fiji and Vanuatu are complementary destinations for the Cook Islands. Therefore, destinations that more closely resemble each other face stronger competition, and the nature and strength of competitive behaviour between two destinations are different for each concerned destination.

Keywords

tourism competitiveness tourism demand panel data pooled mean group small Pacific island countries

Introduction

Tourism competitiveness is a key element of tourism demand (Pashardes and Sinclair, 2005). The multi-dimensionality of tourism competitiveness has important implications for the development of tourism infrastructures such as airports, telecommunications and transportation networks (Andereck et al., 2005), and the preservation of historical buildings and archaeological sites (O'Connell, 1998). Because competitiveness relative to other destinations depends in part on the natural, built and social environment of the country, these elements become increasingly important for destination competitiveness (Pashardes and Sinclair, 2005). This study assesses overall and bilateral tourism competitiveness in selected PICs namely, Cook Islands, Fiji, Tonga, Samoa and Vanuatu using the pooled mean group (PMG) approach. The microeconomic demand theory serves as the theoretical foundation for this study.

Research on relative price competitiveness is important because it plays a role in determining tourism demand from different countries (Pashardes and Sinclair, 2005). Tourism competitiveness helps policymakers isolate each country’s comparative advantage and identifying the tourism product which contributes most to revenue (Pashardes and Sinclair, 2005). This assists in firms’ pricing and investment decisions concerning the tourism product and governments in planning the location and scale of tourism infrastructure (Pashardes and Sinclair, 2005). Tourism competitiveness helps plan marketing strategies. With non-competitive behaviour, tourism planners can collaborate to improve tourism performance through joint marketing campaigns (Li et al., 2005). Understanding PICs’ tourism competitiveness is important because of their huge dependence on tourism for growth and because they have been severely weakened by the travel restrictions resulting from COVID-19. Their tourism industries are unlikely to revive fully by 2021/22 (c.f. Singh et al., 2021). As of April 2021, Fiji is experiencing a second wave of the more transmissible B1617/delta variant of COVID-19. Any tourism sector recovery plan will therefore require a careful evaluation of tourism competitiveness and demand elasticities.

Source market specific (SMS) models for PICs have been developed by Kumar et al. (2020) using the ARDL approach. However, SMS models have limited use in macroeconomic policy formulation due to the potential of heterogeneity in demand elasticities across source markets (elasticity heterogeneity hereafter) per destination (c.f. Peng et al., 2015). Elasticity heterogeneity implies that changes in macroeconomic conditions can have dissimilar effects on each source market. This makes overall macroeconomic policy formulation, for example, on tourism competitiveness, or monetary policy’s effect on tourism difficult. Although analysing different source markets separately has the advantage of more fine-grained and targeted policies when elasticities differ,¹ overall models are better suited for macroeconomic policy formulation by presenting an overall picture of tourism demand. We develop overall tourism demand models using the PMG technique. Following Dogru et al. (2021) on panel data methods in tourism demand, appropriate attention is paid to issues such as cross-sectional dependence, non-stationarity and dynamic effects for robust outcomes.

Our findings indicate that for Fiji and Vanuatu, the other PICs are competing destinations. Fiji and Vanuatu face the strongest bilateral competition because they are more similar to each other compared to the other destinations in terms of culture and geography, both being Melanesian countries with noticeably warmer climates. Cross-price elasticities are insignificant for Tonga. This suggests that Tonga does not compete with the other PICs, probably because their tourism sector is relatively small, and a sizeable share of tourists travel to Tonga to visit friends and relatives (Table 1). Bilateral cross-price elasticities are negative for the Cook Islands from Fiji, Vanuatu, Tonga and Samoa. Overall cross-price elasticity is negative for Samoa and is negative for Fiji, Cook Islands and Vanuatu indicating complementary behaviour. Thus, while for Fiji and Vanuatu, Cook Islands is a competing destination, the relationship is complementary for the Cook Islands. This divergent competitive behaviour is also observed between Fiji and Samoa and Vanuatu and Samoa.

Table 1.

Selected key indicators of PICs.

Key indicators	Cook Islands	Fiji	Tonga	Samoa	Vanuatu
Land area (thousand km²)	236.7	18,274	748	2,842	12,189
Total population in 2019	17,548	889,953	104,494	197,097	299,882
Visitor arrivals in 2019	187,226	870,000	54,000	164,000	116,000
Average annual GDP growth rate (2002–2019)	3.73	2.44	1.52	2.07	3.05
Average annual inflation rate CPI (2002–2019)	2.82	3.36	5.52	3.93	2.30
Exchange rate 2020 (per USD)	1.38	2.04	2.29	2.50	107.81
Agriculture value added 2019 (percent of GDP)	5.10	11.58	17.07	9.75	25.84
Industry value added 2019 (percent of GDP)	12.7	15.59	15.39	15.59	11.05
Services value added 2019 (percent of GDP)	82.1	53.95	50.49	74.66	62.70
Tourism’s contribution in 2019 (percent of GDP)	67	40	21	20	48
RGDP per capita 2019 (constant USD)	12,759	4738	4355	3869	2887
Percent of tourists from Australia in 2019	44	16	21	19	58
Percent of tourists from New Zealand in 2019	18	61	43	40	13
Purpose of visit in 2019 – Holiday (percent of tourists)	77	75	42	31	76
Purpose of visit in 2019 – VFR (percent of tourists)	7	8	42	31	9

Source: World Banks’ WDI database and World Data Atlas, respective Central Bank’s quarterly reviews, and Ministry of Finance and Bureau of Statistics, and national tourism organization updates, World Travel and Tourism Council.

The key contribution of the study is the application of the pooled mean group approach to address the cross-sectional dependence and non-stationarity problems in panel data and the estimation of overall tourism demand (c.f. Dogru et al., 2021). The findings are important for strategic planning of tourism resources to maximize tourism revenue and the development of tourism sector recovery plans. We distinctly use the PMG technique to develop overall tourism demand models for the PICs, which to our knowledge no other research has done. The theoretical implication of the results is that demand elasticities can vary significantly across source markets which creates the need for panel estimates. Similar destinations face stronger competition, and the competitive relationship between two destinations can be different for each of the concerned destinations. A further implication is that cross-sectional dependence and non-stationarity need to be addressed for reliable elasticities and inference.

Literature review

A detailed exposition of the evolution of tourism demand modelling is in Song et al. (2019). Tourism demand is modelled with either the econometric, time series or artificial intelligence models. This review focuses on econometric models because they provide estimates of demand elasticities, consolidate knowledge and test the predictions of theories (Song and Li, 2008). Econometric models focus on establishing the structure of causality and determining how various explanatory variables affect tourism demand (Song et al., 2019). Econometric models begin by specifying a potential causal relationship as supported by theory and proceed to eliminate insignificant explanatory variables (Song et al., 2019). Song et al. (2019) report that 111/211 (52.61%) of the published research on tourism demand between 1968 and 2018 have used econometric models. Modelling tourism demand is still evolving with recent developments progressing towards mixed frequency (Zhang et al., 2021; Wen et al., 2021; Nguyen and Valadkhani, 2020) and panel data methods (Dogru et al., 2021). Panel data incorporate information on both intertemporal movements and cross-sectional heterogeneity of the tourism demand data (Song et al., 2019). This allows researchers to capture information that cannot be captured with time series and cross-sectional models without risking multicollinearity or reducing degrees of freedom.

Panel data models have been applied in the tourism demand literature. Fixed and random effects was applied by Seetaram and Dwyer (2009) who find that immigration is a major determinant of Australian inbound tourism from its nine key source markets. Kusni et al. (2013) find that tourism prices, the SARS pandemic and the global financial crisis are important determinants of tourism demand for Malaysia using the fixed effects model. Long et al. (2019) also find that the fixed effects model improves the forecasting capacity of tourism demand over the pooled least squares in Chinese provinces. Panel fully modified least squares (FMOLS) was applied by Dogru et al. (2017) and Ulucak et al. (2020) in Turkey. Dogru et al. (2017) find including prices and exchange rates separately results in misleading outcomes and that the industrial production index is not a good proxy for tourist’s income. Ulucak et al. (2020) find that tourism is generally income and price inelastic in Turkey. This highlights that although international tourism is generally considered a luxury product (Inchausti-Sintes et al., 2021), for some destinations, it is a necessity (Waqas-Awan et al., 2021). The system GMM (SGMM) method is applied by Li et al. (2017) who find that tourism in 19 major Chinese cities is impacted by climatic factors, and by Muryani et al. (2021) who find that per-capita income, relative prices and room availability positively affect tourism demand in Indonesia. Kumar and Kumar (2020) estimate the tourism demand model for the top nine tourist destinations using the gravity model and the PMG technique and find that ICT positively impacts tourism demand.

Dogru et al. (2021) suggest that research with panel data should consider the effects of cross-sectional dependence, non-stationarity and slope heterogeneity. Cross-sectional dependence implies that a shock occurring in one unit affects other units in the panel and has important implications for stationarity testing and estimation of long-run parameters. Sarafidis and Robertson (2009) show that the moment conditions, and hence, estimates derived from standard dynamic panel techniques such as GMM are invalid with cross-sectional dependence, and this holds for any lag length of the instruments used. Another shortcoming is the use of static panel regressions. Tourism demand is fundamentally dynamic as a tourist’s former experience can affect current tourism demand through repeat visitation and/or word-of-mouth (Dogru et al., 2021). The SGMM estimator is the most common dynamic panel data method employed in tourism demand. However, the approach may suffer from an over identification bias in panels where the time series component exceeds the number of cross-sections, and non-stationarity may lead to weak instruments (Dogru et al., 2021). Panel research on tourism demand seldom applies stationarity tests, and estimates assuming stationarity may violate the Gauss–Markov assumptions (Dogru et al., 2021). Slope heterogeneity indicates that slope coefficients are different across cross-sectional units. The choice of the appropriate methodology applied thus depends on, among other reasons, the characteristics of the dataset. However, elements such as cross-sectional dependence, non-stationarity and slope heterogeneity are important considerations (Dogru et al., 2021).

Estimating the cross-price elasticity helps ascertain the nature of the competition between destinations. Research provides mixed evidence on tourism competitiveness. Through a meta-analysis of tourism demand elasticities in 195 studies published between 1961 and 2011, Peng et al. (2015) suggest that most destinations tend to be competitors. A significant and positive cross-price elasticity indicates a substitution effect and strong competitors for a destination. Li et al. (2005) however note that different degrees of substitution may emerge by assessing the bilateral relationship between destinations. The implication in this setting is to adopt appropriate strategies based on attributes specific to competing destinations or to focus on differentiated market segments to benefit from their comparative advantages. However, Li et al. (2005) also note that two destinations in geographic proximity can also benefit from each other as tourists are more likely to package such destinations into a single trip. Empirical evidence that Greece and Italy were regarded as complementary destinations by United Kingdom tourists was provided by Lyssiotou (2000) and Li et al. (2004). According to Li et al. (2005), when complementary effects are in place, the destinations involved should consider launching joint marketing programmes to maximize tourism sector performance.

Summary statistics of the Pacific island countries

Table 1 describes selected key indicators of the PICs. Real GDP per capita is the highest (lowest) in the Cook Islands (Vanuatu) amongst the selected PICs. Average inflation rates amongst the PICs have been below 4% from 2002 to 2019 (Table 1). Cook Islands was the fastest growing country relative to other PICs and recorded an average of 3.73% growth in real GDP. Key source markets for the PICs are short-haul markets Australia and New Zealand (Table 1). About 75% of the tourists visit Fiji, Cook Islands and Vanuatu for holiday. For Tonga and Samoa, a relatively even share of tourists are noted for holidays and visiting friends and relatives (Table 1).

Theory, model and methods

Demand elasticities

Theoretical insights are drawn from microeconomic demand theory. The effect of a change in price $p_{i}$ on the change in quantity demanded of $x_{i}$ is shown by the Slutsky (1915) equation(s)

\frac{\partial x_{i}}{\partial p_{i}} = \frac{\partial x_{i}^{c}}{\partial p_{i}} - x_{i} \frac{\partial x_{i}}{\partial y} < 0

(1)

or in elasticity terms

ε_{i}^{p} = ε_{i}^{c} - s_{i} ε_{i}^{y} < 0

(1.1)

where

\partial x_{i} / \partial p_{i}

is the change in quantity demanded of

x_{i}

due to change in

p_{i}

\partial x_{i}^{c} / \partial p_{i}

is the substitution effect holding utility constant,

- x_{i} \partial x_{i} / \partial y

captures the income or purchasing power effect of a change in

p_{i}

on the change in quantity demanded of

x_{i}

ε_{i}^{p}

is the own price elasticity

(ε_{i}^{p} = \partial x_{i} / \partial p_{i} \times p_{i} / x_{i})

ε_{i}^{c}

is the compensated demand price elasticity, or the own price substitution effect

(ε_{i}^{c} = \partial x_{i}^{c} / \partial p_{i} \times p_{i} / x_{i})

ε_{i}^{y}

is the income elasticity

(ε_{i}^{y} = \partial x_{i} / \partial y_{i} \times y_{i} / x_{i})

and

s_{i}

is the share of income devoted to

x_{i}

(s_{i} = p_{i} x_{i} / y)

The substitution effect is negative with convex indifference curves because an increase in $p_{i}$ discourages consumption of $x_{i}$ and encourages the consumption of a relatively cheaper alternative $x_{j}$ , holding the level of utility and $p_{j}$ constant. The substitution and income effects are both negative with normal goods $(ε_{i}^{y} > 0)$ ; the income effect is positive with inferior goods $(ε_{i}^{y} < 0)$ and the magnitude of $ε_{i}^{u}$ , $ε_{i}^{y}$ and $s_{i}$ determines $ε_{i}^{p}$ . A strong positive correlation between $p_{i}$ and quantity demanded of $x_{i}$ $(ε_{i}^{p} > 0)$ implies an inferior good that may follow Giffen’s paradox.

Similarly, the price of $x_{j}$ $(p_{j})$ affects the quantity demanded of $x_{i}$ . A Slutsky decomposition of the cross-price substitution and income effects is given by

\frac{\partial x_{i}}{\partial p_{j}} = \frac{\partial x_{i}^{c}}{\partial p_{j}} - x_{j} \frac{\partial x_{i}}{\partial y} > 0

(2)

and in terms of elasticity

ε_{j}^{p} = ε_{j}^{c} - s_{j} ε_{i}^{y} > 0

(2.1)

where

\partial x_{i} / \partial p_{j}

is the change in quantity demanded of

x_{i}

due to change in

p_{j}

;

\partial x_{i}^{c} / \partial p_{j}

is the substitution effect holding utility constant;

- x_{j} \partial x_{i} / \partial y

is the income effect of a change in

p_{j}

on the change in quantity demanded of

x_{i}

;

ε_{j}^{p}

is the cross-price elasticity

(ε_{j}^{p} = \partial x_{i} / \partial p_{j} \times p_{j} / x_{i})

;

ε_{j}^{c}

is the compensated cross-price elasticity or the cross substitution effect

(ε_{j}^{c} = \partial x_{i}^{c} / \partial p_{j} \times p_{j} / x_{i}); ε_{i}^{y}

is the income elasticity (

(ε_{i}^{y} = \partial x_{i} / \partial y_{i} \times y_{i} / x_{i})

) and

s_{j}

is the share of income devoted to

x_{j}

(s_{j} = p_{j} x_{j} / y)

Goods $x_{i}$ and $x_{j}$ are gross substitutes if $ε_{j}^{p} > 0$ , gross complements if $ε_{j}^{p} < 0$ , net substitutes if $ε_{j}^{c} > 0$ and net complements if $ε_{j}^{c} < 0$ . As gross substitutes (complements), the demand for $x_{i}$ increases when $p_{j}$ increases (decreases). The cross substitution effect (net substitution effect) is positive with convex indifference curves because an increase in $p_{i}$ compels consumers to purchase the relatively cheaper alternative good $x_{j}$ , holding the level of utility and $p_{j}$ constant. The income effect depends on $s_{j} ε_{i}^{y}$ – if $x_{i}$ is a normal (inferior) good, $s_{j} ε_{i}^{y}$ will be negative (positive). If $x_{i}$ is a luxury good, the cross-price elasticity could be negative.

We thus observe the following three theoretical restrictions on demand elasticities

ε_{i}^{y} > 0 \to ε_{i}^{p} < 0 and ε_{j}^{p} \neq 0

ε_{i}^{y} < 0 \to ε_{i}^{p} \neq 0 and ε_{j}^{p} > 0

ε_{i}^{y} = 0 \to ε_{i}^{p} < 0 and ε_{j}^{p} > 0

The Slutsky restrictions indicate that a positive income elasticity must be accompanied by a negative price elasticity although there are no restrictions on price elasticities with negative income elasticities. The income elasticity of zero suggests that the price elasticity must be unconditionally negative. However, external effects on the utility can be derived from the consumption of $x_{i}$ (Leibenstein, 1950). These include conspicuous consumption of status symbols goods and network externalities which include the bandwagon and snob effects. If $x_{i}$ is a Veblen good, its demand will increase as its price increases yielding a positive price elasticity and highly elastic income elasticity. The bandwagon effect refers to the positive (negative) effect on the consumption of $x_{i}$ due to other peoples’ consumption of $x_{i}$ .

Model

The model specification is drawn from Song et al. (2003) which gives us

q_{i, t} = A p_{i, t}^{α} y_{i, t}^{β} s p_{t}^{θ}

(3)

where

q_{i, t}

is visitor arrivals from source country

i

to a PIC destination at time t,

p_{i, t}

is the tourism price faced by tourists from the source country

i

visiting a PIC destination at time t,

y_{i, t}

is the tourist’s income measured by the source country’s real GDP and

s p_{t}

is the neighbouring country substitute tourism price at time t.

The tourism price indicator is computed as

p_{i, t} = \frac{c p i_{t}}{c p i_{i, t}^{*}} e x_{i, t}

(4)

where

c p i_{t}

and

c p i_{i, t}^{*}

are the consumer price indices for the PIC destination and the source country

i

, respectively, and

e x_{i, t}

is the exchange rate of the source country

i

with reference to the destination’s currency.²

To assess overall tourism competitiveness, the substitute price is calculated as follows

s p_{t} = \sum_{l = 1}^{5} ω_{l} \frac{c p i_{l}}{e x_{l}}

(5)

where l represents competing destinations,

c p i_{l}

is the consumer price index of the competing destination,

e x_{l}

is the exchange rate of the competing destination expressed with reference to USD and

ω_{l}

is the share of competing destinations expressed as

ω_{l} = \frac{q_{l}}{\sum_{l = 1}^{5} q_{l}}

(6)

and

q_{l}

represents total visitor arrivals of the competing destinations.

Bilateral competitiveness is assessed using the ratio of the CPI to exchange rates of the selected competing PIC destination with reference to the USD.

Taking the log of equation (6), we get the basic long-run model for estimation

\ln q_{i, t} = π + α \ln p_{i, t} + β \ln y_{i, t} + θ \ln s p_{t} + u_{i, t}

(7)

where

π

is the intercept, and

α

β

and

θ

are price, income and competing destinations’ price elasticities, respectively, and

u_{i t}

is the error term.

We expect that $α < 0$ , which implies that greater tourism prices harm visitor arrivals. However, with conspicuous consumption of luxury goods, the price elasticity may be positive. Elastic price elasticity generally indicates that tourists have choices to engage as tourists and hence are very sensitive to changes in prices (Crouch, 1994). We also expect that $β \geq 1$ because the tourist’s income exerts a positive effect on visitor arrivals and because international travel is a luxury (Crouch, 1994). A negative income elasticity may indicate that income growth would change the preference of tourists to better alternatives (Crouch, 1994). We expect that $θ > 0$ would imply that PICs are competing destinations. On the other hand, $θ \leq 0$ indicates non-competitive behaviour.

Seasonality test

Seasonality reflects the pursuit of climatic and non-climatic seasonal factors such as sunshine and beaches and factors such as school and public holidays (Ridderstaat et al., 2014). Seasonality relates to concurrent seasonal changes in other economic variables with which the dependent variable interacts (Thomas and Wallis, 1971). We use the US Census Bureau’s X-13 ARIMA-SEATS method to test for seasonality in visitor arrivals which tests for stable and moving seasonality. Seasonality testing is important because seasonality dummies are appropriate under stable seasonality only (Ridderstaat et al., 2014).

Structural breaks

Tourism demand is affected by structural events such as financial crises and pandemics (c.f. Smeral, 2017). We include structural breaks to control for the effects of the 2007/08 global financial crisis. To control for the effects of pandemics in our sample, we focus on whether the pandemics may have affected the PICs’ key source markets. As a result, we control for the effect of the 2002/03 SARS-CoV-1 pandemic and the 2009 H1N1 swine flu pandemic. The dummy variables are set to one for the period the hypothesized break occurs.

Panel data methods

To test for the presence of unit roots in the data, we first test for cross-sectional dependence (Pesaran, 2007). Traditional panel unit root tests which do not control for cross-sectional dependence may have severe size distortions (O’Connell, 1998). Pesaran (2007) shows that although orthogonalization procedures to eliminate cross-sectional dependence are an alternative, augmenting the standard augmented Dickey-Fuller test regression with cross-sectional averages of the lagged level and first difference variables yields similar, asymptotic results. To examine unit roots, we use Pesaran (2007) cross-sectional dependence robust CADF unit root test. This test is conducted by estimating with OLS, the following equation

Δ x_{i, t} = μ_{i, t} + η_{i} x_{i, t - 1} + ϵ_{i, t}

(8)

where

Δ

is the first difference operator,

x_{i, t}

is a panel variable,

μ_{i, t}

is the individual deterministic component (constant, trend term, fixed effects),

η_{i}

is the AR coefficient and

ϵ_{i, t}

is the error term. The test assumes the null hypothesis of no unit roots H₀:

η_{i} = 0

. To eliminate the cross dependence, the standard ADF regression in equation (8) is augmented with the cross-section averages of lagged levels and first differences of the individual series.

We invoke the Pedroni (2004) panel-based Engle–Granger cointegration test. This test performs an ADF unit root test on the residuals of the estimated level model in equation (7)

Δ u_{i, t} = ρ_{i} u_{i, t - 1} + ε_{i, t}

(9)

where

Δ

is the first difference operator,

u_{i, t}

is the panel regression’s residual and

ε_{i, t}

is the error term. The Pedroni test we consider assumes individual AR coefficients and assumes the null hypothesis of no unit roots H₀:

ρ_{i} = 0

. The null hypothesis of no cointegration is rejected if the relevant p-values of the test are less than 5%.

We apply the pooled mean group approach of Pesaran et al. (1999) for estimation. The model is specified below

Δ \ln q_{i, t} = ζ_{1} + ζ_{2} \ln q_{i, t - 1} + ζ_{3} \ln p_{i, t - 1} + ζ_{4} \ln y_{i, t - 1} + ζ_{5} \ln s p_{t - 1} + lagged short run terms + ν_{t}

(10)

The PMG is a panel ARDL model which is estimated with the maximum likelihood approach and allows for unrestricted short-run slope heterogeneity like the Mean Group estimator, but long-run parameters remain the same across cross-sections like the fixed effects estimator. Pesaran et al. (1999) show that the poolability of long-run coefficients has valid theoretical foundations. The advantages include PMG’s ability to provide robust estimates with cross-sectional dependence, and it can be applied with a mixture of $I (0)$ and $I (1)$ data. PMG also provides robust estimates when the number of periods exceeds the number of cross-sectional units (Kumar et al., 2020).

Data and results

Data

Visitor arrivals data, price, income and substitute prices for each destination from each key source market are pooled together to derive the panel dataset required for estimation. Australia, New Zealand, the United States, Canada, Europe, United Kingdom and Japan are source markets considered for Fiji. Source markets for the Cook Islands include Australia, New Zealand, the United States, Canada and Europe. For Vanuatu, source markets include Australia, New Zealand, the United States, Europe and Japan. For Tonga and Samoa, source markets include Australia, New Zealand and the United States Source markets are considered to depend on the availability of data. Given data availability, we use a sample from 2002Q1 to 2019Q3 for Fiji (497 observations), the Cook Islands and Vanuatu (355 observations each), 2002Q1 to 2017Q4 for Tonga (192 observations) and 2002Q4 to 2019Q3 for Samoa (204 observations).

Fiji visitor arrivals data are from the Reserve Bank of Fiji’s quarterly reviews from 2002Q1 to 2020Q1. For the Cook Islands, the Ministry of Finance and Economic Management, tourism and migration statistics provide arrivals data from 1993Q1 to 2020Q1. For Vanuatu, the Vanuatu National Statistics Office international arrivals statistics and tourism news archives provide arrivals data from 2002Q1 to 2019Q4. The National Reserve Bank of Tonga’s quarterly bulletin provides arrivals data for each source market from 1995Q1 to 2017Q4 and 2019Q1 to 2020Q1. Source market-level arrival data were not available in 2018 for Tonga. Samoa Bureau of Statistics, migration statistics provides arrivals data from 2002Q3 to 2019Q3. Total arrivals for the Solomon Islands are sourced from the Central Bank of Solomon Islands and are available from 1997Q1 to 2019Q4.

Fiji CPI data are from the Reserve Bank of Fiji’s quarterly reviews from 2002Q1 to 2020Q1. For the Cook Islands, the Ministry of Finance and Economic Management provides CPI data from 1993Q1 to 2020Q1. The Vanuatu National Statistics Office, CPI quarterly movement provides CPI data for Vanuatu from 1998Q1 to 2020Q1. For Tonga, the National Reserve Bank of Tonga’s quarterly bulletin provides CPI data from 1999Q1 to 2020Q1, from the Samoa Bureau of Statistics for Samoa from 2005Q1 to 2020Q1, and from the Central Bank of Solomon Islands, Honiara retail price index from 2002Q1 to 2020Q1. CPI data for the source markets are sourced from the OECD (2021) main economic indicators, available at the Federal Reserve Economic Data (2021) (FRED) database, from 1960Q1 to 2021Q1. The base period for all CPI series was 2015Q1.

Gross domestic product data for Australia are sourced from the Australian Bureau of Statistics (2021), Australian National Accounts from 1959Q3 to 2021Q1. Data for New Zealand are sourced from the Reserve Bank of New Zealand (2021), Gross Domestic Product from 1987Q2 to 2021Q1. Data for the USA are sourced from the US Bureau of Economic Analysis (2021), Gross Domestic Product from 1947Q1 to 2021Q1. Data for Germany, Canada, UK and Japan are sourced from the OECD (2021) main economic indicators, available on the FRED database over the desired period.

Data for bilateral exchange rates (nominal), with reference to USD, are obtained from the OECD (2021) main economic indicators, available on the FRED database for the source markets. Exchange rate data for PICs are obtained from the Exchange Rates United Kingdom (2021) database. Arrival data for Fiji from Europe are not available from 2006Q1 to 2006Q4 and is interpolated for each quarter using the automatic log-linear interpolation method. Samoa’s CPI data from 2002 to 2004 are converted to the quarterly frequency with annual data sourced from the World Development Indicators using the quadratic average conversion approach. Total arrivals data for Samoa are not available from 2002Q1 to 2002Q3. Because these data are needed to compute substitute prices, the quadratic sum approach is used to convert the annual arrivals in 2002 to quarterly to derive the data series from 2002Q1 to 2002Q3. Shahzad et al. (2017) suggest that the quadratic approach of frequency conversion produces reliable decompositions of the data.

Seasonality

The results from the census X-13 ARIMA-SEATS seasonality tests indicate stable seasonality in visitor arrivals (Table 2).

Table 2:

X-13 ARIMA-SEATS seasonality test.

Source market	Stable seasonality		Moving seasonality
Panel A: Fiji
Australia	F (3,67)=301.77^A	$χ^{2}$ (3)=63.15^A	F (16,48)=0.78
New Zealand	F (3,67)=579.46^A	$χ^{2}$ (3)=65.06^A	F (16,48)=1.58
USA	F (3,67)=77.46^A	$χ^{2}$ (3)=53.13^A	F (16,48)=0.46
Canada	F (3,67)=47.95^A	$χ^{2}$ (3)=40.11^A	F (16,48)=1.61
Europe	F (3,67)=16.59^A	$χ^{2}$ (3)=30.52^A	F (16,48)=2.85^A
UK	F (3,67)=15.44^A	$χ^{2}$ (3)=29.96^A	F (16,48)=0.80
Japan	F (3,67)=90.38^A	$χ^{2}$ (3)=55.22^A	F (16,48)=1.93
Panel B: Cook Islands
Australia	F (3,67)=124.27^A	$χ^{2}$ (3)=58.35^A	F (16,48)=2.88^A
New Zealand	F (3,67)=1302.84^A	$χ^{2}$ (3)=65.63^A	F (16,48)=2.01^A
USA	F (3,67)=8.09^A	$χ^{2}$ (3)=18.59^A	F (16,48)=0.74
Canada	F (3,67)=749.61^A	$χ^{2}$ (3)=63.53^A	F (16,48)=5.14^A
Europe	F (3,67)=16.59^A	$χ^{2}$ (3)=30.52^A	F (16,48)=2.85^A
Panel C: Vanuatu
Australia	F (3,54)=36.51^A	$χ^{2}$ (3)=50.73^A	F (13,39)=2.16^A
New Zealand	F (3,54)=294.10^A	$χ^{2}$ (3)=58.89^A	F (13,39)=1.98
USA	F (3,54)=46.31^A	$χ^{2}$ (3)=53.13^A	F (13,39)=2.59^A
Europe	F (3,54)=34.05^A	$χ^{2}$ (3)=44.96^A	F (13,39)=1.76
Panel D: Tonga
Australia	F (3,54)=59.33^A	$χ^{2}$ (3)=49.77^A	F (13,39)=0.55
New Zealand	F (3,54)=97.99^A	$χ^{2}$ (3)=56.07^A	F (13,39)=1.35
USA	F (3,54)=94.49^A	$χ^{2}$ (3)=40.29^A	F (13,39)=0.83
Panel E: Samoa
Australia	F (3,44)=15.96^A	$χ^{2}$ (3)=30.15^A	F (10,30)=0.69
New Zealand	F (3,44)=18.67^A	$χ^{2}$ (3)=31.88^A	F (10,30)=0.54
USA	F (3,44)=12.55^A	$χ^{2}$ (3)=26.24^A	F (10,30)=1.10

Source: Estimated in EViews 12. A-seasonality at 1%.

Cross-sectional dependence, unit roots and cointegration

Cross-sectional dependence is found (Table 3). The CADF unit root test indicates that the data are a mixture of

I (0)

and

I (1)

(Table 4). Cointegration is confirmed using the Pedroni cointegration test (Table 5).

Table 3:

Pesaran’s cross-sectional dependence test.

Variables	Fiji	Cook Islands	Vanuatu	Tonga	Samoa
Null hypothesis: No cross-sectional dependence
$\ln q_{i, t}$	3.16^A [<0.01]	0.10 [0.91]	17.04^A [<0.01]	9.09^A [<0.01]	9.09^A [<0.01]
$\ln p_{i, t}$	7.93^A [<0.01]	15.41^A [<0.01]	1.49 [0.14]	5.61^A [<0.01]	8.58^A [<0.01]
$\ln y_{i, t}$	27.24^A [<0.01]	24.74^A [<0.01]	16.34^A [<0.01]	12.68^A [<0.01]	13.09^A [<0.01]
$\ln s p_{t}$ -overall	38.61^A [<0.01]	26.65^A [<0.01]	26.65^A [<0.01]	13.86^A [<0.01]	14.28^A [<0.01]
$\ln s p_{t}$ -bilateral	38.61^A [<0.01]	38.61^A [<0.01]	38.61^A [<0.01]	38.61^A [<0.01]	38.61^A [<0.01]

Source: Estimated in EViews 12. A-cross-sectional dependence at 1%. p-value in [.].

Table 4:

Pesaran’s CADF unit root test.

Variables	Fiji		Cook Islands		Vanuatu		Tonga		Samoa
—	Level	1st diff.	Level	1st diff.	Level	1st diff.	Level	1st diff.	Level	1st diff.
Null hypothesis: No unit roots
$\ln q_{i, t}$	−1.89 [0.37]	−6.08^A [<0.01]	−2.08 [0.23]	−10.72^A [<0.01]	−3.98^A [<0.01]	−10.71^A [<0.01]	−1.74 [0.53]	−6.19^A [<0.01]	−2.28 [0.17]	−6.19^A [<0.01]
$\ln p_{i, t}$	−1.46 [0.82]	−4.19^A [0.82]	−1.84 [0.44]	−9.71^A [<0.01]	−2.37^A [<0.01]	−8.43^A [<0.01]	−2.02 [0.33]	−4.20^A [<0.01]	−2.04 [0.31]	−4.67^A [<0.01]
$\ln y_{i, t}$	−2.61^A [<0.01]	−5.16^A [<0.01]	−1.89 [0.39]	−10.01^A [<0.01]	−2.89^A [<0.01]	−10.71^A [<0.01]	−0.68 [0.99]	−3.95^A [<0.01]	−1.45 [0.73]	−4.33^A [<0.01]
$\ln s p_{t}$ -overall	−6.19^A [<0.01]	−10.72^A [<0.01]	−6.19^A [<0.01]	−10.72^A [<0.01]	−6.19^A [<0.01]	−10.72^A [<0.01]	−6.19^A [<0.01]	−10.72^A [<0.01]	−6.19^A [<0.01]	−10.72^A [<0.01]
$\ln s p_{t}$ -bilateral	−6.19^A [<0.01]	−6.19^A [<0.01]	−6.19^A [<0.01]	−6.19^A [<0.01]	−6.19^A [<0.01]	−6.19^A [<0.01]	−6.19^A [<0.01]	−6.19^A [<0.01]	−6.19^A [<0.01]	−6.19^A [<0.01]

Source: Estimated in Stata 16. A-indicates stationarity at 1%. p-value in [.].

Table 5:

Pedroni panel cointegration test results.

Competitiveness assumption	Fiji	Cook Islands	Vanuatu	Tonga	Samoa
Null hypothesis: No long-run relationship
Overall	−5.73^A [<0.01]	−12.27^A [<0.01]	−15.34^A [<0.01]	−9.41^A [<0.01]	−6.71^A [<0.01]
Bilateral	—	—	—	—	—
Fiji	—	−19.00^A [<0.01]	−15.47^A [<0.01]	−8.05^A [<0.01]	−8.37^A [<0.01]
Cook Islands	−11.79^A [<0.01]	—	−15.32^A [<0.01]	−8.79^A [<0.01]	−7.62^A [<0.01]
Vanuatu	−12.35^A [<0.01]	−19.27^A [<0.01]	—	−8.28^A [<0.01]	−7.46^A [<0.01]
Tonga	−11.29^A [<0.01]	−19.19^A [<0.01]	−15.96^A [<0.01]	—	−7.71^A [<0.01]
Samoa	−13.65^A [<0.01]	−17.74^A [<0.01]	−15.37^A [<0.01]	−9.01^A [<0.01]	—
Solomon Islands	−15.73^A [<0.01]	−16.22^A [<0.01]	−16.41^A [<0.01]	−8.75^A [<0.01]	−7.74^A [<0.01]

Source: Estimated in EViews 12. A-cointegration at 1%. p values in [.]. The test assumes no deterministic trend. One lag is used in the cointegration test.

Long run and short run

The PMG estimates are presented in Tables 6–10. Fiji, Tonga and Samoa are generally price inelastic destinations, whereas the Cook Islands have a unit price elasticity. Lower price elasticity indicates that PICs are unique destinations (c.f. Peng et al., 2015). Own price effects are insignificant in Vanuatu. The income elasticity exceeds positive one for Fiji, Cook Islands and Tonga in both the overall and bilateral models. Vanuatu can be considered an income inelastic destination, whereas for Samoa, the income elasticity is at minimum, unit elastic. The global financial crisis had a significant negative effect on tourism demand in Fiji, insignificant effects in the Cook Islands and Vanuatu, but positive effects in Tonga and Samoa. One reason for the positive effect could be the purpose of travel (Table 1), as travellers may wish to visit their friends and relatives back home during the crisis. SARS-CoV-1 generally had insignificant effects on arrivals in Fiji, Cook Islands, Vanuatu and Tonga, but had a negative effect in Samoa. The H1N1 pandemic harmed arrivals to Fiji and insignificant effects in the Cook Islands. The effect was positive and significant in Vanuatu, Tonga and Samoa which suggests that tourists from these destinations’ source markets began substituting away from destinations where the swine flu pandemic was prevalent to safer destinations in the PICs. However, H1N1 reduced tourism demand in Fiji possibly because its key source markets were impacted. Seasonality effects indicate that quarter 3 is the best quarter for tourism in Fiji and Vanuatu, and quarter 2 is the second-best quarter for tourism in Vanuatu. One reason for the observed pattern is that winter is the peak period for tourism in the Pacific as Australians and New Zealanders search for the winter sun.³ Quarter 1 is the worst quarter for Cook Islands, Tonga and Samoa due to the Pacific cyclone season.

Table 6:

Fiji.

Variables	Overall	Bilateral
Variables	Overall	Cook Islands	Vanuatu	Tonga	Samoa	Solomon Islands
Panel A: Long run
$\ln P_{i, t}$	−0.76^A [0.1989]	−0.47^A [0.1159]	−0.40^A [0.0942]	−0.37^A [0.0995]	−0.44^A [0.0908]	−0.60^A [0.1291]
$\ln Y_{i, t}$	1.48^A [0.2583]	1.71^A [0.1107]	1.54^A [0.1095]	1.60^A [0.1172]	1.43^A [0.1179]	2.79^A [0.1817]
$\ln P_{s, t}$	0.27^B [0.1383]	0.22^A [0.0687]	0.42^A [0.0659]	0.32^A [0.0437]	0.36^A [0.0553]	−0.51^A [0.0799]
Panel B: Short run
$E C T_{t - 1}$	−0.17^B [0.0700]	−0.27^B [0.1079]	−0.30^B [0.1386]	−0.31^B [0.1575]	−0.30^A [0.1379]	−0.34^A [0.0846]
$Δ I n Q_{i, t - 1}$	−0.31^A [0.0662]	−0.18^A [0.0507]	−0.17^B [0.0681]	−0.15^C [0.0847]	−0.16^B [0.0684]	—
$Δ I n Q_{i, t - 2}$	−0.20^B [0.0972]	—	—	—	—	—
$Δ I n P_{i, t}$	0.27^C [0.1468]	0.26^C [0.1479]	0.24^C [0.1363]	0.25^C [0.1335]	0.23^C [0.1434]	0.21^B [0.0895]
$Δ I n P_{i, t - 1}$	0.01 [0.2361]	—	—	—	—	—
$Δ I n Y_{i, t}$	−1.18 [0.8915]	−1.13 [0.7082]	−1.36^B [0.6561]	−0.97 [0.7468]	−1.24^B [0.7460]	−1.47^C [0.8210]
$Δ I n Y_{i, t - 1}$	0.57 [0.6429]	—	—	—	—	—
$Δ \ln P_{s, t}$	0.03 [0.0211]	0.11 [0.0757]	0.12 [0.1059]	−0.03 [0.0951]	0.06 [0.1141]	0.26 [0.3511]
$Δ \ln P_{s, t - 1}$	−0.10 [0.0843]	—	—	—	—	—
Q1	−0.10 [0.1314]	−0.10 [0.1059]	−0.10 [0.1064]	−0.09 [0.1081]	−0.10 [0.1009]	−0.07 [0.1545]
Q2	0.01 [0.0605]	0.08 [0.0916]	0.08 [0.0931]	0.09 [0.0947]	0.07 [0.0969]	0.09 [0.0706]
Q3	0.25^B [0.1019]	0.27^A [0.1002]	0.28^A [0.1020]	0.28^A [0.1060]	0.27^B [0.1029]	0.29^B [0.1217]
GFC	−0.04^B [0.0199]	−0.03^B [0.0207]	−0.04^B [0.0200]	−0.03^B [0.0205]	−0.03 [0.0211]	−0.01 [0.0259]
H1N1	−0.07^A [0.0120]	−0.07^A [0.0285]	−0.07^A [0.0268]	−0.05^A [0.0365]	−0.07^B [0.0258]	−0.04 [0.0246]
SARS-CoV-1	0.01 [0.0194]<	0.01 [0.0193]	0.01 [0.0144]	0.03^A [0.0116]	0.02 [0.0135]	−0.01 [0.0238]
Constant	−5.31^B [2.1180]	−10.34^A [3.9527]	−9.81^B [4.4913]	−11.06^B [5.6496]	−9.28^B [4.2087]	−22.57^B [5.4485]
Panel C: Statistics
Specification	ARDL (3,2,2,2)	ARDL (2,1,1,1)	ARDL (2,1,1,1)	ARDL (2,1,1,1)	ARDL (2,1,1,1)	ARDL (1,1,1,1)
RMSE	0.1033	0.1130	0.1149	0.1146	0.1145	0.1165
SER	0.1154	0.1227	0.1240	0.1244	0.1243	0.1263
AIC	−1.4539	−1.3708	−1.3853	−1.3694	−1.3798	−1.3668

Source: Estimated in EViews 12. A, B, C-significance at 1, 5, 10%. Standard errors in [.].

Table 7:

Cook Islands.

Variables	Overall	Bilateral
Variables	Overall	Fiji	Vanuatu	Tonga	Samoa	Solomon Islands
Panel A: Long run
$\ln P_{i, t}$	−1.11^A [0.1782]	−0.34^C [0.1866]	−0.37^A [0.1529]	−0.43^A [0.1460]	−0.39^B [0.1911]	−1.01^A [0.2658]
$\ln Y_{i, t}$	2.06^A [0.3074]	3.16^A [0.2472]	3.13^A [0.1822]	3.17^A [0.1796]	3.15^A [0.2738]	2.49^A [0.4156]
$\ln P_{s, t}$	0.28 [0.2466]	−1.03^A [0.2309]	−0.83^A [0.1398]	−0.66^A [0.1064]	−0.62^A [0.1596]	−0.08 [0.2232]
Panel B: Short run
$E C T_{t - 1}$	−0.32^A [0.1198]	−0.33^B [0.1256]	−0.36^B [0.1481]	−0.36^B [0.1483]	−0.33^A [0.1296]	−0.32^A [0.1193]
$Δ I n P_{i, t}$	0.06 [0.4280]	−0.15 [0.4043]	−0.16 [0.4340]	−0.07 [0.4746]	−0.12 [0.4329]	−0.06 [0.5143]
$Δ I n P_{i, t - 1}$	0.48 [0.3117]	—	0.37 [0.3366]	0.24 [0.3752]	—	0.31 [0.3096]
$Δ I n Y_{i, t}$	−0.28 [1.1460]	−1.53 [1.3884]	−0.83 [1.3672]	−1.73 [1.2533]	−1.59 [1.4327]	−1.08 [1.1429]
$Δ I n Y_{i, t - 1}$	1.64 [0.5250]	—	1.05 [0.7216]	1.02 [0.6231]	—	1.56 [0.6119]
$Δ \ln P_{s, t}$	−0.05 [0.1443]	0.22 [0.1806]	0.24 [0.2624]	0.08 [0.1215]	0.05 [0.1529]	−0.37 [0.2347]
$Δ \ln P_{s, t - 1}$	−0.23 [0.1695]	—	−0.05 [0.3298]	0.20 [0.1529]	—	0.08 [0.4223]
Q1	−0.29^C [0.1559]	−0.28^C [0.1673]	−0.25 [0.1600]	−0.29^C [0.1699]	−0.29^C [0.1617]	−0.24 [0.2040]
Q2	−0.18 [0.4076]	−0.21 [0.4049]	−0.18 [0.4022]	−0.19 [0.4020]	−0.22 [0.4119]	−0.16 [0.4030]
Q3	−0.11 [0.2416]	−0.11 [0.2453]	−0.10 [0.2496]	−0.11 [0.2518]	−0.11 [0.2475]	−0.10 [0.2658]
GFC	−0.09 [0.0460]	−0.06 [0.0381]	−0.09 [0.0583]	−0.09^C [0.0536]	−0.10^C [0.0527]	−0.09^C [0.0259]
H1N1	−0.02 [0.0056]	−0.06 [0.0381]	−0.02 [0.0705]	−0.01 [0.0581]	−0.03 [0.0483]	−0.03 [0.0609]
SARS-CoV-1	0.09^A [0.0376]	0.09^A [0.0324]	0.10^A [0.0366]	0.08^A [0.0282]	0.08^A [0.0338]	−0.08^B [0.0342]
Constant	−16.19^B [6.8324]	−25.18^A [10.1569]	−28.93^B [12.3435]	−28.73^B [11.9081]	−26.09^B [10.6362]	−19.59^B [8.2194]
Panel C: Statistics
Specification	ARDL (1,2,2,2)	ARDL (1,1,1,1)	ARDL (1,2,2,2)	ARDL (1,2,2,2)	ARDL (1,1,1,1)	ARDL (1,2,2,2)
RMSE	0.1041	0.1077	0.1025	0.1040	0.1087	0.1059
SER	0.1151	0.1168	0.1133	0.1150	0.1180	0.1171
AIC	−1.3231	−1.3735	−1.3765	−1.3544	−1.3561	−1.3206

Source: Estimated in EViews 12. A, B, C-significance at 1, 5, 10%. Standard errors in [.].

Table 8:

Vanuatu.

Variables	Overall	Bilateral
Variables	Overall	Fiji	Cook Islands	Tonga	Samoa	Solomon Islands
Panel A: Long run
$\ln P_{i, t}$	−0.06 [0.2117]	−0.06 [0.2104]	0.04 [0.2077]	−0.27 [0.1916]	−0.26 [0.2026]	−0.22 [0.2038]
$\ln Y_{i, t}$	0.75^A [0.2405]	0.71^A [0.2461]	1.00^A [0.2107]	0.29 [0.1879]	0.35 [0.2285]	0.24 [0.2599]
$\ln P_{s, t}$	0.68^A [0.1679]	0.79^A [0.1889]	0.39^A [0.1065]	0.71^A [0.1001]	0.68^A [0.1146]	0.65^A [0.1154]
Panel B: Short run
$E C T_{t - 1}$	−0.60^A [0.0506]	−0.60^A [0.0491]	−0.59^B [0.0527]	−0.63^A [0.0902]	−0.63^A [0.0672]	−0.66^A [0.0426]
$Δ I n P_{i, t}$	−0.08 [0.1959]	−0.12 [0.2094]	−0.10 [0.1461]	−0.15 [0.1575]	−0.08 [0.1561]	−0.05 [0.1122]
$Δ I n Y_{i, t}$	−0.69 [1.0519]	−0.60 [1.0119]	−1.19 [1.0944]	−0.66 [0.8118]	−0.69 [1.0489]	−0.91 [1.1116]
$Δ \ln P_{s, t}$	−0.01 [0.1619]	0.19 [0.2743]	−0.12^A [0.0484]	−0.06 [0.1617]	0.09 [0.0742]	−1.03^B [0.4971]
Q1	−0.09 [0.1619]	−0.11 [0.1621]	−0.09 [0.1772]	−0.11 [0.1544]	−0.12 [0.1710]	−0.05 [0.1990]
Q2	0.17^A [0.0508]	0.17^A [0.0519]	0.18^A [0.0495]	0.16^B [0.0694]	0.13^B [0.0498]	0.11^B [0.0482]
Q3	0.43^A [0.0321]	0.43^A [0.0279]	0.44^A [0.0311]	0.42^A [0.0576]	0.40^A [0.0377]	0.41^A [0.0329]
GFC	0.02 [0.0493]	0.01 [0.0478]	0.04 [0.0519]	0.04 [0.0533]	0.07 [0.0581]	0.09^C [0.0568]
H1N1	0.12^A [0.0620]	0.13^A [0.0637]	0.12^B [0.0576]	0.10 [0.0776]	0.10 [0.0705]	0.07 [0.0677]
SARS-CoV-1	−0.01 [0.0312]	−0.01 [0.0303]	−0.01 [0.0348]	0.03 [0.0348]	0.03 [0.0359]	0.01 [0.0385]
Constant	−10.47^A [1.3981]	−10.08^A [1.3447]	−13.71^B [1.7065]	−2.86^A [0.4991]	−3.91^A [0.7059]	−1.40^B [0.6630]
Panel C: Statistics
Specification	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)
RMSE	0.1886	0.1874	0.1899	0.1867	0.1874	0.1835
SER	0.2047	0.2035	0.2061	0.1150	0.2034	0.1992
AIC	−0.1844	−0.1961	−0.1694	−0.2202	−0.2034	−0.2360

Source: Estimated in EViews 12. A, B, C-significance at 1, 5, 10%. Standard errors in [.].

Table 9:

Tonga.

Variables	Overall	Bilateral
Variables	Overall	Fiji	Cook Islands	Vanuatu	Samoa	Solomon Islands
Panel A: Long run
$\ln P_{i, t}$	−0.82^A [0.2001]	−0.71^A [0.2154]	−0.73^A [0.2116]	−0.64^A [0.2204]	−0.49^B [0.2272]	−0.58^A [0.2226]
$\ln Y_{i, t}$	1.38^A [0.3427]	1.62^A [0.2670]	1.60^A [0.2402]	1.73^A [0.2586]	1.93^A [0.3360]	1.97^A [0.4138]
$\ln P_{s, t}$	0.14 [0.2118]	−0.01 [0.2174]	0.04 [0.1065]	−0.09 [0.1718]	−0.25 [0.1771]	−0.25 [0.1968]
Panel B: Short run
$E C T_{t - 1}$	−0.70^A [0.0222]	−0.70^A [0.0312]	−0.71^A [0.0661]	−0.70^A [0.0246	−0.71^A [0.0424]	−0.72^A [0.0664]
$Δ I n P_{i, t}$	0.01 [0.3704]	−0.06 [0.2202]	−0.09 [0.3378]	−0.02 [0.2931]	−0.02 [0.2577]	0.05 [0.2404]
$Δ I n Y_{i, t}$	0.09 [1.8377]	0.29 [1.7706]	0.16 [1.9624]	0.29 [2.0357]	0.10 [1.6231]	0.20 [1.5612]
$Δ \ln P_{s, t}$	−0.27^C [0.1565]	0.12 [0.4332]	−0.09 [0.3744]	−0.03 [0.3062]	0.05 [0.2013]	−0.34 [0.5045]
Q1	−0.67^A [0.1870]	−0.67^A [0.1569]	−0.68^A [0.1858]	−0.67^A [0.2031]	−0.65^A [0.1532]	−0.60^A [0.1319]
Q2	−0.06 [0.1419]	−0.06 [0.1543]	−0.08 [0.1862]	−0.08 [0.1649]	−0.05 [0.1179]	−0.05 [0.1048]
Q3	−0.10 [0.1159]	−0.09 [0.1154]	−0.11 [0.1357]	−0.11 [0.1362]	−0.08 [0.0530]	−0.06 [0.0860]
GFC	0.07^C [0.0460]	0.09^C [0.0471]	0.09^C [0.0452]	0.08^C [0.0459]	0.07 [0.0530]	0.07 [0.0609]
H1N1	0.17^A [0.0417]	0.17^A [0.0247]	0.18^A [0.0618]	0.16^A [0.0411]	0.15^A [0.0282]	0.15^A [0.0157]<
SARS-CoV-1	0.06 [0.0607]	0.06 [0.0591]	0.07 [0.0545]	0.07 [0.0594]	0.05 [0.0659]	0.05 [0.0697]
Constant	−21.41^A [2.4046]	−25.51^A [2.7180]	−25.45^B [2.2146]	−27.48^A [2.6357]	−31.39^A [4.5121]	−33.00^B [5.8746]
Panel C: Statistics
Specification	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)
RMSE	0.1413	0.1415	0.1413	0.1417	0.1418	0.1409
SER	0.1556	0.1557	0.1555	0.1560	0.1561	0.1550
AIC	−0.6999	−0.6942	−0.7062	−0.6927	−0.6822	−0.6937

Source: Estimated in EViews 12. A, B, C-significance at 1, 5, 10%. Standard errors in [.].

Table 10:

Samoa.

Variables	Overall	Bilateral
Variables	Overall	Fiji	Cook Islands	Vanuatu	Tonga	Solomon Islands
Panel A: Long run
$\ln P_{i, t}$	−0.65^A [0.2389]	−0.43^C [0.2551]	−0.37 [0.2638]	−0.33 [0.2717]	−0.40 [0.2814]	−0.42 [0.2781]
$\ln Y_{i, t}$	0.95^A [0.5589]	1.78^A [0.6583]	1.57^B [0.6695]	2.01^A [0.7064]	1.64^B [0.6748]	1.65^B [0.6724]
$\ln P_{s, t}$	−1.02^A [0.1932]	−0.50^B [0.2371]	−0.19^C [0.1142]	−0.35^B [0.1753]	−0.11 [0.1269]	−0.16 [0.1776]
Panel B: Short run
$E C T_{t - 1}$	−0.87^A [0.1458]	−0.83^A [0.1501]	−0.80^A [0.1277]	−0.80^A [0.1643]	−0.80^A [0.1366]	−0.79^A [0.1350]
$Δ I n P_{i, t}$	0.44 [0.3300]	0.24 [0.5784]	0.45 [0.4275]	0.44 [0.5103]	0.36 [0.3888]	0.36 [0.4208]
$Δ I n Y_{i, t}$	0.09 [1.8377]	−1.67 [1.8284]	−1.36 [1.9087]	−1.86 [1.5932]	−1.23 [2.1334]	−0.68 [2.6108]
$Δ \ln P_{s, t}$	−0.27^C [0.1565]	−0.36 [0.3195]	0.14 [0.1173]	−0.17 [0.3649]	−0.35^A [0.0689]	−0.05 [0.7802]
Q1	−0.67^A [0.1870]	−0.13 [0.2702]	−0.18 [0.2435]	−0.16 [0.2434]	−0.15 [0.2717]	−0.14 [0.2199]
Q2	−0.08 [0.1305]	0.05 [0.1207]	0.05 [0.1068]	0.06 [0.1121]	0.03 [0.1251]	0.05 [0.1103]
Q3	0.04 [0.1431]	0.09 [0.1581]	0.09 [0.1421]	0.09 [0.1495]	0.08 [0.1683]	0.10 [0.1481]
GFC	0.13^A [0.0397]	0.06^B [0.0275]	0.03 [0.0181]	0.02 [0.0212]	0.04^C [0.0184]	0.04^B [0.0182]
H1N1	0.12^A [0.0436]	0.07^A [0.0413]	0.09^A [0.0308]	0.10^A [0.0165]	0.13^A [0.0323]	0.12^A [0.0243]
SARS-CoV-1	−0.14^B [0.0699]	−0.11^C [0.0650]	−0.11 [0.0671]	−0.08 [0.0701]	−0.08 [0.0599]	−0.08 [0.0697]
Constant	−13.57^A [2.6116]	−31.90^A [6.4571]	−25.45^B [2.2146]	−37.64^A [8.3235]	−28.86^A [5.1460]	−28.69^B [5.2651]
Trend	0.02^A [0.0032]	0.01^A [0.0011]	0.01^A [0.0011]	0.01^A [0.0007]	0.01^A [0.0011]	0.01^A [0.0008]
Panel C: Statistics
Specification	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)	ARDL (1,1,1,1)
RMSE	0.1450	0.1544	0.1557	0.1558	0.1418	0.1554
SER	0.1601	0.1704	0.1718	0.1719	0.1561	0.1716
AIC	−0.6338	−0.5180	−0.4998	−0.5011	−0.6822	−0.5037

Source: Estimated in EViews 12. A, B, C-significance at 1, 5, 10%. Standard errors in [.].

Overall cross-price elasticity is positive and significant for Fiji and Vanuatu; thus, the other PICs can be considered as competing destinations. The implication is that growth in tourism in competing destinations may come at the detriment of Fiji’s tourism sector. Capturing this market may require tourism policymakers to differentiate the tourism product through value-adding products to entice tourists to spend a greater proportion of their holidays in Fiji and Vanuatu, respectively. Bilateral estimates reveal that Fiji faces the strongest competition with Vanuatu, followed by Samoa, Tonga and the Cook Islands. Vanuatu on the other hand faces the strongest competition from Fiji followed by Tonga, Samoa, Solomon Islands and the Cook Islands. Fiji and Vanuatu face the strongest bilateral competition because they are more similar to each other compared to the other destinations. Song et al. (2003) claim that substitute destinations should be similar in terms of culture and geography. Being both Melanesian countries with a noticeably warmer climate, Fiji and Vanuatu are more similar to each other than the other PICs.

Cross-price elasticities are insignificant for Tonga. This suggests that Tonga is not in active competition with the rest of the PICs. This could be because their tourism sector is still relatively small (Table 1) and developing and a sizeable share of tourists travel to Tonga to visit friends and relatives. Thus, Tonga’s tourism sector evolves independently of tourism in other PICs. Bilateral cross-price elasticities are negative and significant for the Cook Islands from Fiji, Vanuatu, Tonga, and Samoa. Overall cross-price elasticity is negative for Samoa and is negative for Fiji, Cook Islands and Vanuatu indicating complementary behaviour. Thus, while for Fiji and Vanuatu, Cook Islands is a competing destination, the relationship is complementary for the Cook Islands. The same can be argued for Fiji-Samoa and Vanuatu-Samoa. With complementary behaviour, Li et al. (2005) suggest that the destinations involved should launch joint marketing campaigns. However, policymakers in Fiji and Vanuatu may be cautious of such policies as the Cook Islands is their competitor.

The results echo the earlier findings of Pratt (2013) who surveyed the distinguishing attributes of PICs considered by Australian tourists such as repeat visitation, beaches, and beautiful scenery, and weather/climate. Fiji and Vanuatu are perceived as close competitors and closely associated with as a tropical island destination with positive feedback either from word-of-mouth, advertising or previous visits (Pratt, 2013). The Cook Islands is not strongly linked to any particular attribute that distinguishes its destination image (Pratt, 2013) which agrees with the insignificant cross-price elasticity estimate (Table 7). Pratt (2013) further notes that Samoa and Tonga are close bilateral competitors. Yet, our results suggest otherwise because substitute prices between Tonga and Samoa are insignificant (Tables 9 and 10). Overall, however, many attributes are shared by the PICs which can complicate research on destination competitiveness (Pratt, 2013).

Conclusion, policy implications and research outlook

This study tests for tourism competitiveness in small Pacific island countries using quarterly panel data over the period 2002–2019. Estimations were done using the pooled mean group approach. The findings imply that for Fiji and Vanuatu, other PICs are competing destinations. Fiji and Vanuatu face the strongest bilateral competition amongst the PICs. Cross-price elasticities are negative and significant for the Cook Islands and Samoa and insignificant for Tonga. Thus, while for Fiji and Vanuatu, Cook Islands is a competing destination, the relationship is complementary for the Cook Islands. The results suggest that the nature and strength of competition are different for each concerned destination.

For tourism competition policy, targeted marketing campaigns need to be developed to tap into the source market pool of tourism in PICs. For example, the key source market for both Fiji and Vanuatu is Australia, and Fiji and Vanuatu both have strong bilateral competition with each other. Policymakers in Fiji and Vanuatu could then devise marketing campaigns targeted especially for Australian tourists. Policy planners in Tonga should focus on developing their distinct brand of tourism as a niche market. Samoa and Cook Islands on the other hand generally have negative cross-price elasticities. Thus, policymakers there need to collaborate with their counterparts in other PICs to sustainably develop their tourism sector such as through joint marketing campaigns, or developing direct, visa-free flight routes. One example is to develop a direct travel route from Fiji to the Cook Islands, without requiring the passenger to transit in Auckland.⁴ A regional partnership, such as the South Pacific Tourism Organization (SPTO) can also play a pivotal role in regional tourism development.

Demand elasticities are also important for macroeconomic policy formulation. Because tourism demand is determined by financial market indicators such as prices and the exchange rates, monetary policymakers need to be aware of how any policy action may impact prices, and subsequently, tourism. For example, an unwarranted expansionary monetary policy may create inflationary pressure which could reduce arrivals. For competing destinations, this will lead to an increase in arrivals as tourists now choose to travel to relatively cheaper destinations. The implication is that monetary policymakers need to devise policies keeping in mind how their policies may impact tourism, which is a key driver of growth for PICs. Macroeconomic policymakers also need awareness of macroeconomic policy in other PICs as this impacts their tourism demand.

Data limitations were noted. CPI data for Samoa were not available over the period 2002Q1–2004Q4. Annual data from the World Bank were converted to quarterly using the quadratic average approach. Total visitor arrivals data for Samoa were not available from 2002Q1 to 2002Q3. Annual data from the World Bank were converted to quarterly using the quadratic sum approach. Visitor arrivals data from Europe to Fiji were not available from 2006Q1 to 2006Q4 and was interpolated using the log-linear interpolation approach. The discussion of price and income in demand theory assumes a specific product. Derivation of demand elasticities assumes that the nature or quality of the product does not change. This may not be entirely valid for tourism demand. Future research can thus examine time-varying demand elasticities to bypass this shortcoming. As identified by Song et al. (2019), methodologies used to estimate tourism demand models are still evolving. Although this study has demonstrated the usefulness of panel data techniques in modelling tourism demand, future research could predict tourism demand using Bayesian structural time series methods.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Nikeel N Kumar

Notes

Author biographies

Nikeel N Kumar is currently a teaching assistant and doctoral candidate with the School of Economics at the Auckland University of Technology. His current research focuses on tourism demand analysis and economic growth.

Arvind Patel is currently employed as Professor in the School of Accounting, Finance and Economics at the University of the South Pacific (USP). A graduate of the University of South Pacific, University of New South Wales and the University of Queensland in Australia, Professor Patel has a research interest in corporate governance, auditing, financial accounting, financial economics, and tourism and economic growth. Professor Patel has published in top-tier journals in information systems, accounting and economics. His current research interest focuses on how COVID-19 has impacted the Pacific region.

Rup Singh is an academic economist and senior lecturer in Economics at the University of the South Pacific. His research interests are economic analysis, macroeconomic policy and trade. Rup teaches official statistics and quantitative methods at USP.

References

Andereck

Valentine

Knopf

, et al. (2005) Residents’ perceptions of community tourism impacts. Annals of Tourism Research 32(4): 1056–1076.

Crouch

(1994) The study of international tourism demand: A review of findings. Journal of Travel Research 33(1): 12–23.

Dogru

Bulut

Sirakaya-Turk

(2021) Modeling tourism demand: Theoretical and empirical considerations for future research. Tourism Economics 27(4): 874–889.

Dogru

Sirakaya-Turk

Crouch

(2017) Remodeling international tourism demand: Old theory and new evidence. Tourism Management 60: 47–55.

Inchausti-Sintes

Voltes-Dorta

Suau-Sánchez

(2021) The income elasticity gap and its implications for economic growth and tourism development: The Balearic vs. the Canary Islands. Current Issues in Tourism 24(1): 98–116.

Kumar

(2020) Relationship between ICT and international tourism demand: A study of major tourist destinations. Tourism Economics 26(6): 908–925.

Kumar

Patel

, et al. (2020) Modelling inbound international tourism demand in small Pacific Island countries. Applied Economics 52(10): 1031–1047.

Kusni

Kadir

Nayan

(2013) International tourism demand in Malaysia by tourists from OECD countries: A panel data econometric analysis. Procedia Economics and Finance 7: 28–34.

Leibenstein

(1950) Bandwagon, snob, and Veblen effects in the theory of consumers’ demand. The Quarterly Journal of Economics 64(2): 183–207.

10.

Song

Witt

(2005) Recent developments in econometric modeling and forecasting. Journal of Travel Research 44(1): 82–99.

11.

Song

(2017) A dynamic panel data analysis of climate and tourism demand. Journal of Travel Research 56(2): 158–171.

12.

Song

Witt

(2004) Modeling tourism demand: A dynamic linear AIDS approach. Journal of Travel Research 43(2): 141–150.

13.

Long

Liu

Song

(2019) Pooling in tourism demand forecasting. Journal of Travel Research 58(7): 1161–1174.

14.

Lyssiotou

(2000) Dynamic analysis of British demand for tourism abroad. Empirical Economics 25(3): 421–436.

15.

Muryani

Permatasari

Padilla

MAE

(2021) Determinants of tourism demand in Indonesia: A panel data analysis. Tourism Analysis 25: 77–89. DOI: 10.3727/108354220X15984209532647

16.

Nguyen

Valadkhani

(2020) Dynamic responses of tourist arrivals in Australia to currency fluctuations. Journal of Hospitality and Tourism Management 45: 71–78.

17.

Pashardes

Sinclair

(2005) Special issue: Tourism competitiveness introduction. Tourism Economics 11(1): 5–10.

18.

Pedroni

(2004) Panel cointegration: asymptotic and finite sample properties of pooled time series tests with an application to the PPP hypothesis. Econometric Theory 20(3): 597–625.

19.

Peng

Song

Crouch

, et al. (2015) A meta-analysis of international tourism demand elasticities. Journal of Travel Research 54(5): 611–633.

20.

Pesaran

(2007) A simple panel unit root test in the presence of cross-section dependence. Journal of Applied Econometrics 22(2): 265–312.

21.

Pesaran

Shin

Smith

(1999) Pooled mean group estimation of dynamic heterogeneous panels. Journal of the American statistical Association 94(446): 621–634.

22.

Pratt

(2013) Same, same but different: Perceptions of South Pacific destinations among Australian travelers. Journal of Travel & Tourism Marketing 30(6): 595–609.

23.

O'Connell

(1998) The overvaluation of purchasing power parity. Journal of International Economics 44(1): 1–19.

24.

Ridderstaat

Oduber

Croes

, et al. (2014) Impacts of seasonal patterns of climate on recurrent fluctuations in tourism demand: Evidence from Aruba. Tourism Management 41: 245–256.

25.

Sarafidis

Robertson

(2009) On the impact of error cross-sectional dependence in short dynamic panel estimation. The Econometrics Journal 12(1): 62–81.

26.

Seetaram

Dwyer

(2009) Immigration and tourism demand in Australia: A panel data analysis. Anatolia 20(1): 212–222.

27.

Shahzad

SJH

Shahbaz

Ferrer

, et al. (2017) Tourism-led growth hypothesis in the top ten tourist destinations: New evidence using the quantile-on-quantile approach. Tourism Management 60: 223–232.

28.

Singh

Lal

Khan

, et al. (2021) The COVID-19 experience in the Fiji Islands: Some lessons for crisis management for small island developing states of the Pacific region and beyond. New Zealand Economic Papers, 1–6. DOI: 10.1080/00779954.2020.1870534.

29.

Smeral

(2017) Tourism forecasting performance considering the instability of demand elasticities. Journal of Travel Research 56(7): 913–926.

30.

Song

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

31.

Song

Qiu

RTR

Park

(2019) A review of research on tourism demand forecasting: Launching the annals of tourism research curated collection on tourism demand forecasting. Annals of Tourism Research 75: 338–362.

32.

Song

Wong

KKF

Chon

KKS

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

33.

Slutsky

(1915) On the consumer budget theory. Journal of Economists and Statistics Magazine, 1–26.

34.

Thomas

Wallis

(1971) Seasonal variation in regression analysis. Journal of the Royal Statistical Society. Series A (General) 134(1): 57–72.

35.

Ulucak

Yücel

İlkay

SÇ

(2020) Dynamics of tourism demand in Turkey: Panel data analysis using gravity model. Tourism Economics 26(8): 1394–1414.

36.

Waqas-Awan

Rosselló-Nadal

Santana-Gallego

(2021) New insights into the role of personal income on international tourism. Journal of Travel Research 60(4): 799–809.

37.

Wen

Liu

Song

, et al. (2021) Forecasting tourism demand with an improved mixed data sampling model. Journal of Travel Research 60(2): 336–353.

38.

Zhang

Law

, et al. (2021) A hybrid MIDAS approach for forecasting hotel demand using large panels of search data. Tourism Economics. https://doi.org/10.1177%2F13548166211015515