Integration of tourism markets in Australia

Introduction

The convergence of tourism markets by conveying revenue from one country to another benefits local economies (Tugcu, 2014).Consequently, the convergence patterns of tourism demand can undergo scrutiny by the policy makers in order to capture a profound understanding of the implications and the effects on the economy (Abbott et al., 2012; Faulkner, 1988). For instance, the existence of divergent behavior implies that policy makers have to develop more effective (customized) strategies in order to stimulate inbound tourism. Given the fact that the development of the tourism industry is in conjunction with countries’ different economic growth stages (De Vita and Kyaw, 2016), the identification of the different convergence patterns of international arrivals can act as a powerful policy tool evaluating the effect of countries’ tourism development strategies (Mérida et al., 2016; Narayan, 2006).

It is worth mentioning that a small number of surveys have examined the phenomenon of convergence in the tourism literature. In addition, those few studies in order to reveal the convergence patterns of tourism industries have applied the Lagrange multiplier (LM) unit root tests (Lean and Smyth, 2008; Lorde and Moore, 2008; Narayan, 2006; Ozcan and Erdogan, 2017; Tiwari, 2016). In contrast to the few prementioned studies, Lee (2009) has applied a Dickey–Fuller framework, whereas Tan and Tan (2013) have applied a panel setting with multiple structural breaks to reveal the convergence patterns of Singapore’s tourism markets.

Given the aforementioned research gap, this is the first study to examine the convergence across international visitor arrivals in Australia from 28 tourism source markets via the two methodological frameworks introduced by Phillips and Sul (2007, 2009). To the best of our knowledge, only the study by Mérida et al. (2016) has applied only the first convergence test introduced by Phillips and Sul (2007) evaluating the existence of convergence clubs for 12 tourism source markets of Spain. The methodological approaches applied have some unique advantages in comparison with the traditional methods applied in the tourism literature. Apart for its ability to test econometrically for the existence of “convergence clubs,” the applied methodological framework also estimates the convergence paths relative to some identified common trends. Moreover, it accounts directly for transitional heterogeneity and transitional divergence and it does not rely on strong assumptions on trend or stochastic stationarity that are common in visitor arrivals data. Finally, another contribution of our article is the application of the adopted methodology to the Australian case. According to Kulendran (1996), since Australia is geographically isolated, there is not any complementarity or cross-price competition as in other tourism markets (i.e. European). As a result, the inbound tourism has been the core of government and commercial interests in order to promote and develop the tourism industry which is a basic pillar of Australia’s economy (Morris et al., 1995). Given Australia’s isolated geographical position, the understanding and identification of convergence patterns among international visitor arrivals will enable the Australian Tourist Commission to distribute tourism expenditure efficiently based on the different segments of international tourism demand (Tsui and Balli, 2017).

Data and methodological framework

For the purpose of our analysis, we apply monthly international visitor arrivals data from 28 source markets. The data have been extracted from the Australian Bureau of Statistics, have been seasonally adjusted (Valadkhani and O’Mahony, 2015), and are referring to the period from January 1991 to September 2017.

Concerning the methodological framework, Phillips and Sul (2007, 2009) developed the log t test in order to capture the heterogeneity, which is a vital feature in the panel data setting. Within a panel data context, a factor X_i,t (i.e. the observed arrivals) can be expressed in the following form

X i , t = θ i μ t + ε i t

1

In expression (1), the first component θ i μ t represents the distance between X_i,t and the common factor μ_t which is measured from the idiosyncratic element θ_i,t(systematic term) and the second component ε_it representing the error term. The principal contribution of the Phillips and Sul (2007, 2009) convergence test is the reformulation of equation (1). The reformulation contains the measurement of the time varying of systemic (idiosyncratic) factor loading θ_i,t and the incorporation of the error term ε_it alongside a common factor μ_t. The common element θ_i,t measures the deviation among states which is defined by μ_t. Also it must be noted that all groups formed within the clusters or from the observed sample will converge to a steady state when lim m → ∞ θ i t + m = θ . As a result, equation (1) can be reformulated as

X i , t = θ i , t μ t

2

By eliminating the μ_t component, Phillips and Sul (2007, 2009) defined the relative transition parameter h i , t as

h i , t = X i , t N − 1 ∑ i = 1 N X i , t = θ i , t N − 1 ∑ i = 1 N θ i , t

3

In expression (3), h i , t represents the transition path, whereas the μ t component is eliminated. Finally, the time varying coefficient θ i , t is assumed to have the following form

θ i , t = θ i + σ i , t + ϕ i , t , with σ i , t = σ i [ L ( t ) t α ] , σ i > 0 , τ ≥ 0 and ϕ i , t i . i . d . ( 0 , 1 )

4

Then the null hypothesis of convergence for i is expressed as

H 0 : θ i , t = θ , α ≥ 0

5

whereas the alternative hypothesis (non-convergence) as

H 1 : θ i , t ≠ θ , α < 0

6

According to Phillips and Sul (2007, 2009), the creation of cross-sectional ratio H 1 / H t is used to construct the following log t regression, which is employed to test the null hypothesis of convergence

log ( H 1 H t ) − 2 Log L ( t ) = c ^ + γ ^ log t + u t for t = [ r T ] , [ r T ] + 1 , … , T for some r > 0

7

It must be noted that the cross-sectional variance ( H t ) converges to 0 if θ i , t converges to θ and can be defined as

H t = N − 1 ∑ i = 1 N ( h i , t − 1 ) 2 → 0

8

In expression (7), L ( t ) = log ( t + 1 ) , log t = γ ^ and γ ^ = 2 a ^ . Specifically, a ^ is the estimated a in H₀. Moreover, it must be noted that according to Phillips and Sul (2007, 2009), the value of r = 0.3, which is adapted from Monte Carlo simulations. Considering the presumptions summarized by Phillips and Sul (2007, 2009), the null hypothesis of convergence is rejected at the 5% level if t b ^ < − 1.65. Finally, we follow the specific four-step group-clustering algorithm as explained by Phillips and Sul (2009) implementing further the log t regression test, in order to see if the original estimated clubs merge further among themselves forming therefore new (merged) clubs.

Empirical findings

The results obtained when applying Phillips and Sul’s (2007) approach are displayed in Table 1. Initially, we have to test the null hypothesis of convergence for the entire sample, which cannot be rejected since the estimated t-statistics ( t b ^ ) value is larger than −1.65 ( − 0.309 > − 1.65 ) and despite the fact that the estimated value of log t (speed of adjustment) is negative (−0.233). Hence, we continue with the delineation of convergence clubs. From our analysis, we have identified five convergence clubs. Specifically, club 1 consists of Austria, Greece, Taiwan, and Japan, whereas club 2 includes the United Kingdom, Switzerland, Denmark, Norway, Sweden, Italy, Thailand, and Indonesia. In addition, club 3 encompasses Germany, Netherlands, Singapore, China, Hong Kong, and Canada, whereas club 4 contains New Zealand, Ireland, Belgium, France, Spain, Malaysia, Philippines, South Korea, and the United States. Finally, the last club (club 5) has solely one member: India. Interestingly enough, the speed of adjustment ( log t ) is negative for the majority of clubs. This insinuates the subsistence of divergence. Yet, Phillips and Sul (2007, 2009) underlined that if the null hypothesis γ ^ = 2 a ^ is statistically disparate from zero, the null hypothesis cannot be rejected ( − 1.65 < t b ^ < 0 ) . Figure 1 presents the transition curves of the identified clubs under the assumption of the sample’s overall convergence to unity. The transition curves displayed reveal whether the identified clubs converge or diverge from below or above unity over the examined period. We can see that club 5 (India) has an upward trend approaching and eventually reaching cross-sectional average in the last months of 2006. After this point onward, its transition path appears to be divergent. Similarly, the transition path of club 1 (Austria, Greece, Taiwan, and Japan) shows a convergent behavior approaching cross-sectional average (unity) from above. It is evident that from the end of 2002 up to the last months of 2006, club 1 converges toward the sample’s average level of international visitor arrivals (i.e. towards unity). Furthermore, Figure 1 presents a similar transition path among clubs 2, 3, and 4. This phenomenon is more pronounced for clubs 3 and 4 which follow almost an identical transition path, whereas for the case of club 2 it is evident that it had a similar trend with clubs 3 and 4 only up to the third half of 2012. However, after this point onward, club 2 exhibits a transitional divergence. Indeed, when further employing the Phillips and Sul’s (2009) test, which further merges the convergence clubs identified previously, the results suggest (Table 1) that we can further accept the null hypothesis of convergence, since the initial five clubs have been converged into three clubs. Specifically, the first new club (I) contains the originally identified clubs 1 and 2 with a t-statistic value of −0.283. The second club (II) now contains the previously identified clubs 3 and 4 with a t-statistic value of −0.295. Finally, as also estimated previously, India forms a separate third club (III) with a t-statistic value of 0.350.

OPEN IN VIEWER

Table 1.

Convergence club classification.

	Phillips and Sul (2007)				Phillips and Sul (2009)
Category	log t	t-stat	New club	Final classification	log t	t-stat
Full sample [28]	−0.233	−0.309
Club 1 [4]	0.182	0.204	1 + 2	Club I	−0.106	−0.283
Club 2 [8]	0.135	0.232
Club 3 [6]	−0.027	−0.039	3 + 4	Club II	−0.17	−0.295
Club 4 [9]	−0.191	−0.35
Club 5 [1]	−0.483	−0.645	5	Club III	−0.191	−0.35

Note: The numbers in brackets indicate the number of countries within a group.

OPEN IN VIEWER

Figure 1.

Transition path for the evaluated clusters based on Phillips and Sul’s (2007) approach.

The picture of the transition curves of the newly formed clubs is displayed in Figure 2. Club I appears to have a convergent path up to the end of 2006. However, the negative trend after this point onward suggests a divergent transition path. Moreover, club II has a slightly upward trend but as it is observed it converges toward the sample’s average level of international visitor arrivals. Finally, club III (which contains only India) transition path displays and reaches convergence at the third quarter of 2007. However, after this point onward, the upward trends continue suggesting divergence. This finding aligns with those findings by Valadkhani and O’Mahony (2018) suggesting that one of Australia’s largest inbound markets is India and as a result diverges with the other clubs. In addition, it is evident that European inbound markets are shared among clubs I and II suggesting that a different marketing strategy to stimulate further tourism demand is needed. In fact, club II consists (among other inbound markets) of China and New Zealand which is traditionally the largest leading source markets for Australia. In fact, even though the United Kingdom (club I) is also a traditional source of tourism demand for Australia, according to Tsui and Balli (2017) it has recently been overtaken by China (club II). This is also evident of why the two clubs show a divergence path especially after the end of 2006.

OPEN IN VIEWER

Figure 2.

Transition path for the evaluated clusters based on Phillips and Sul’s (2009) approach.

Conclusions

The article by applying the methodological frameworks introduced by the studies of Phillips and Sul (2007, 2009) examines for the first time the convergence patterns of international tourism arrivals in Australia. By using monthly data over the period January 1991 to September 2017, the empirical evidence suggests that the hypothesis of full convergence was attested. Moreover, we have identified five distinct convergence clubs alongside their transition paths. The benefit of revealing and understanding demand patterns has been well highlighted throughout the tourism literature (Faulkner, 1988). To this end, our study presents how different methodological developments can be applied by policy makers in order to “unlock” the different tourism demand patterns. It is evident that the identification of convergence paths and the integration of international visitor arrivals, as presented in this article, can be the first vital step for policy makers to better direct and customize their target policies. Finally, the identification of tourism convergence patterns provides policy makers with the ability to better respond to different pressures in relation to the adjustments of services and facilities provided. Future research can be directed toward the development and presentation of a unified analytical tool/framework, which will be able to identify the convergence patterns that are based on the common ethnic, geographic, and sociocultural characteristics of tourism demand. This in turn will provide the policy makers with the ability to target and respond more efficiently to the different demand changes within dynamic environments.