The Convergence Dynamics of Per Capita International Tourist Arrivals

ABSTRACT

The study highlights the importance of comovements in tests of stochastic convergence in per capita international tourist arrivals for a large multi-country panel. LM unit root tests with allowance for comovements and structural breaks reveals a significant reduction in the number of rejections of the unit root null hypothesis when compared to LM unit root tests without comovements. Additional analysis allowing for Fourier gradual breaks confirms our findings of limited evidence of stochastic convergence. Upon further examination of convergence clubs, we identify four convergence clusters for per capita international tourist arrivals distinguished by whether a country is considered a small island economy.

KEYWORDS:

• Panel unit root tests
• PANIC
• Phillips-Sul club convergence
• stochastic convergence
• structural breaks
• tourist arrivals

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ The share decreased to 5.3% due to restrictions on mobility in 2020 with the emergence of the COVID-19 pandemic. However, with the roll-out of vaccines and the removal of travel restrictions, the tourism industry has begun to recover and forge a new path toward growth. A UNWTO report from March 2022 indicates that the tourism industry is rebounding, with international tourist arrivals increasing by 130% in January 2022 compared to the increase in the entirety of 2021. Arrivals are still lagging behind pre-pandemic levels by around 67%, but the industry is continuing its gradual journey toward recovery.

² Song, Li, and Cao (2018) note that developed countries are more likely to record trade deficits with respect to travel while developing countries tend to observe a trade surplus on their travel account. This highlights the importance of tourism in the economic development and growth of developing countries.

³ Standard panel estimators may be biased when not appropriately accounting for cross-correlations and structural breaks; see Bai (2010), Baltagi, Feng, and Kao (2016), and Baltagi, Kao, and Liu (2017).

⁴ Similarly in the case of the COVID-19 pandemic.

⁵ Stochastic convergence not only assumes strict convergence or divergence, but also exhibits a loss of power in detecting asymptotic comovement due to the presence of a unit root in the variables being examined (Du 2017).

⁶ Methodological approach and presentation parallels Payne et al. (2022), Payne and Lee (2022), and Nazlioglu et al. (2023).

⁷ As discussed by Enders and Lee (2012a, 2021b), the advantages of using the Fourier function rests with the estimation of the number of breaks, and their locations are given by estimating a parsimonious frequency parameter.

⁸ See Payne et al. (2022), Payne and Lee (2022), and Nazlioglu et al. (2023) for details of the testing procedure. The notational presentation parallels Payne et al. (2022) and Payne and Lee (2022).

⁹ Note the Phillips and Sul (2007) procedure evaluates per capita international tourist arrivals ($T{A}_{it}$), not relative per capita international tourist arrivals ($RT{A}_{it}$). Before beginning the convergence analysis, the data is filtered using the Hodrick and Prescott (1997) filter.

¹⁰ The Convergence Clubs package in R is used to implement the clustering algorithm in four steps. First, order the units in the panel in descending order according to the value in the last time period. Next, run the log-t regression (12) to form core groups, maximizing the number of units under the condition that the t-value remains above −1.65. If this condition does not hold for the first two units, drop the first unit and repeat the procedure. If this condition never holds, the entire panel diverges. Once the core group is formed, sieve the data for club membership. Add units one-by-one while the t-statistic remains above the critical value through the estimation of the log-t test each time. Stop once every unit has been checked. Then, for all units that fail the previous condition, repeat the log t-test on those countries to see if $t_{\hat{\beta }}$> −1.65 holds. If this fails, repeat the previous steps until there are no countries left for which $t_{\hat{\beta }}$ > −1.65, at which point all of these remaining countries diverge.

¹¹ Our data do not include the pandemic period, but it could be another major factor that affects all countries globally.

¹² Merged clubs are denoted with the notation “*”.