3.1. Introducing a New Approach for the Income Level of a Country’s Tourist Arrivals
In this section, our paper introduces a new method for measuring the income level of a country’s tourist arrivals in the literature for the first time. However, this approach requires tourism data on a cross-country basis over a period of time. Given the limitation of cross-country tourism data, this study focuses on Mediterranean countries. To calculate the new index, we follow the approach suggested by [
8], which is popularly known as the PRODY measure” for calculating the income level of a country’s exports.
In a year, a country is visited by tourists having different income levels from various countries. Further, every country has a tourism basket consisting of tourists from different countries. In the literature, scholars have generally used tourism expenditure (or tourist arrivals) as a proxy of tourism. However, with these approaches, it is impossible to observe every visitor country’s economic contribution to host country’s tourism basket. In this new approach, we can observe the contribution of every country to host country. In addition, tourist expenditure is dependent on a number of factors, such as average length of stay. Therefore, our paper focuses on the number of tourism arrivals instead of tourism expenditures.
We define the measurement of income level of a country’s tourist arrivals using the following equation:
where
is the measure of income level of a country’s tourist arrivals. More specifically, a country
i attracts tourists from various countries, and thus,
n is the number of countries from which, country
i attracts tourists.
is the number of tourist inflows (arrivals) from country
j to country
i at time
t, and it needs to be divided by the total tourist inflows of country
i at time
t (
.
is the (constant US
$) per capita GDP of country
j at time
t.
To make this process more simplified, let us give a numerical example to explicitly introduce our measure of income level of a country’s tourist arrivals: For instance, Spain (country i) attracts tourists from 24 countries (n) in 2014 (t). According to the statistical database of the World Tourism Organization (UNWTO), the total number of tourist inflows (from 24 countries) to Spain in 2014 is 60,011,074 (, and 10,615,746 ( of them came from France (country j) in 2014. In addition, according to the World Development Indicators (WDI) the per capita GDP (constant US$) of the France (country j) in 2014 is US$ 35,667 ().
Note that similar calculations need to be done for other 23 countries that Spain attracts tourists from to calculate the income level of a country’s tourist arrivals for Spain in 2014. In this example, our income level of a country’s tourist arrivals measure for Spain in 2014 can be calculated as below:
These calculations are done for every year from 1995 to 2014 and for each of the 8 countries in the sample. Thus, our measure of the income levels of a country’s tourist arrivals is simply a weighted average of the per capita GDPs of countries in terms of their share of tourist inflows to the visiting country. In other words, it represents the average income level of tourism attraction associated with the share of tourist inflows. Our measure of ILTA in Equation (1) is not sensitive to the number of partner markets (n). The different number of partner markets is related to the data availability of the Tourism Statistics Database of the UNWTO. A higher level of measure indicates that a country attracts tourists from countries that have higher levels of income. Therefore, the income level of a country’s tourist arrivals (ILTA) is a real measure of attraction of tourist inflows from high-income countries (partners). It is noteworthy to mention that this measurement is calculated over time and it is subject to cross-country comparisons; therefore, a rise in the calculation is an indication of an upgrade to the income level of a host country’s tourist arrivals via a broader tourism partnership (e.g., common market deal and visa freedom), especially with high-income economies.
Furthermore, a lower level of
ILTA can provide risks for the sustainability of tourist inflows and earnings. Further, low
ILTA implies that “the tourism basket” consists of tourist inflows from poor countries, and this sort of tourism partner can increase barriers due to its instability. For example, if a poor tourism partner faces a financial or political crisis, the people live in such a country that can easily postpone or forego their international tourism demand due to the uncertainty or rising of the real value of their exchange rate (i.e., depreciation of local currency). In addition, demand for international tourism is elastic; that is, the income elasticity of international tourism demand is higher than one [
37]. This issue can also be related to the evidence that households in low-income countries have a limited budget for tourism spending, and thus they can easily postpone or forego the demand for international tourism, since their main spending has to be on “compulsory goods and services” (e.g., clothing, food, housing, health, etc.).
However, this can be a less serious issue for the international tourism demand of high-income countries. For instance, the average tourism budget in high-income countries is generally greater than the average tourism budget in low-income countries. The price stability and other kinds of uncertainties can also be successfully tolerated in high-income economies due to their strong institutional set up (e.g., bureaucracy quality, central bank independence, less-corruption, etc.). For all of these reasons, it can be suggested that attracting tourists from high-income economies will be more beneficial for a host country. In other words, a higher level of income level of a country’s tourism basket can be crucial in providing a sustainable tourism market, especially for countries whose economic growth depends on substantial tourism earnings.
3.2. Nature of Data and Measurement
The present study constructs a balanced panel dataset using annual data from 1995 to 2014 in eight Mediterranean countries, namely Egypt, France, Greece, Italy, Morocco, Spain, Tunisia and Turkey. Since our tourism data is only available from 1995, we begin our sample period from there. Our variables are measured as follows: economic output (EO) is measured through GDP in constant 2010 US$; environmental pollution (EP) is the total CO2 emissions in million metric tons; income level of a country’s tourist arrivals (ILTA) (The annual data for tourist inflows (arrivals) were obtained from the Tourism Statistics Database of the UNWTO) is as per the authors’ calculation, which we defined in the previous section; non-renewable energy consumption (NREC) is the sum of coal, gas, and petroleum in Quad Btu; renewable energy consumption (REC) is the sum of all renewable energy sources in kilotons of oil equivalent; capital (CAP) is the gross fixed capital formation in constant 2010 US$; labor (LBR) is the total labor force; per capita (PI) is the GDP per capita in constant 2010 US$; and finally, population (POP) is the total population. The required data on EO, CAP, LBR, PI, and POP are sourced from the World Development Indicators (WDI), while data on the EP and NREC are obtained from the International Energy Statistics (IES), and finally, data on the REC is acquired from the OECD online data source. Thus, our variables are measured in different units, hence we convert the dataset into natural logarithms before our empirical analysis begins to avoid the issues that are associated with the distributional properties of the data series. Precisely, if the variables are measured in different units, as in our case, it would be difficult to interpret the estimated parameters; hence we convert our variables into natural logarithms.
3.3. Estimation Strategy
The main purpose of this research is to investigate the impact of the income level of a country’s tourist arrivals on economic output and environmental pollution in a sample of eight Mediterranean countries. To achieve these objectives, we built our empirical models by making use of theoretical frameworks such as growth model and environmental model i.e., STIRPAT (Stochastic Impacts by Regression on Population, Affluence and Technology). Using these theoretical approaches, our empirical models are defined in the following:
where,
EO,
CAP,
LBR,
NREC,
REC,
ILTA,
EP,
POP,
PI represent the economic output, capital, labor, non-renewable energy consumption, renewable energy consumption, income level of a country’s tourist arrivals, environmental pollution, population and per capita income, respectively. Our main focus in this paper is to examine the impact of tourism quality on economic growth and carbon emissions, so we are not focusing on the “country level innovative policies which aim to promote sustainable tourism” across these economies. However, future studies may consider addressing this issue. The purpose of Equation (3) is to examine the role of income level of a country’s tourist arrivals on economic output by accounting other important factors in the model, such as capital, labor, non-renewable energy and renewable energy consumption. Similarly, Equation (4) explores the effect of income level of a country’s tourist arrivals on environmental pollution along with population, per capita income, non-renewable and renewable energy uses. On the other hand, vi in Equations (3) and (4) represents individual fixed country effects; cross-sections and time periods are denoted by the subscripts
i and
t , respectively. See also the STIRPAT approach and the tourism-led growth hypothesis literature for the motivation of the empirical models in Equations (3) and (4).
We began our analysis by employing two panel unit root tests to explore the order of integration of the considered variables. More specifically, we employed the Levin, Lin and Chu (LLC) [
38] unit root test under the assumption of a common unit root process. On the other hand, Im, Pesaran, and Shin (IPS) [
39] test works under the assumption of an individual unit root process. Both of these tests have the same null hypothesis i.e., a unit root, while they slightly differ in terms of the alternative hypothesis. For instance, the alternative hypothesis for the LLC test is ‘no unit root’, whereas for the IPS test ‘some cross-sections may not have a unit root’. The findings obtained from these panel unit root tests helped us to select the appropriate econometric techniques to achieve the study objectives. Given the nature of our variables, which were confirmed from the panel unit root tests, we then applied the following methods for the empirical investigation.
Most of the regression models use the conditional mean of a dependent variable while estimating the cause and effect relationship among the variables in the models. However, there is an increasing interest among the econometricians and researchers to employ a quantile regression framework to understand the varying association between dependent and independent variables. More specifically, [
40] proposed a quantile regression technique, which provides the estimates of linear association between independent variables and a specified quantile of the dependent variable. This approach helps us to understand how the 10th or 90th percentile of the dependent variable is affected by the right-hand side variables. Despite these advantages, the empirical studies in the tourism literature fail to employ a quantile regression approach. Therefore, our study employs a pooled quantile regression framework to estimate the effect of the income level of a country’s tourist arrivals on economic performance and environmental pollution at a varying percentile, i.e., from 0.1 to 0.9.
Finally, our study explores the direction of (short-run) causalities by employing a method that allows and accounts heterogeneity across the given cross-sections. For this reason, we follow the approach recommended by [
17]. The detailed discussion of this model can be found in [
17]. These authors develop a simple framework to test the null hypothesis of homogeneous non-causality against the alternative hypothesis of heterogeneous non-causality. More specifically, the null hypothesis of no causality in any cross-section is tested against the alternative hypothesis of causality at least for a few cross-sections. Since this test is designed to examine the short-run dynamic causalities between the variables, we apply this test on the first difference data series. The detailed discussion of the empirical models is avoided to conserve the space in the paper.