New Insights into the Role of Personal Income on International Tourism

Abstract

This study aims to investigate the role of personal income in the income elasticity of tourism demand and, more specifically, the hypothesis that the richest and poorest individuals both tend to react less to changes in income than middle-class individuals, who tend to be more sensitive. To that end, this study applies different strategies within the context of a gravity model, using yearly data from 1995 to 2016 and bilateral tourism flows between 192 countries. Results show that income elasticity is determined to a significant extent by per capita income in the origin country and they confirm the inverted-U relationship between income elasticity and personal income. The study indicates that middle-income countries are more elastic than low- and high-income ones, while high-income countries display an inelastic or nonsignificant relationship.

Keywords

international tourism income elasticity gravity model demand analysis

Introduction

Tourism demand modeling can be tackled from different perspectives. Numerous surveys on the subject have shown that aggregate tourism demand modeling has been the most popular method to do that (Song and Li 2008; Peng et al. 2015; Li et al. 2006; Crouch 1994; Morley 1992). Generally speaking, with this approach, different regression models are built to explore the correlation between bilateral tourism flows (origin-destination) and a set of determining variables in order to accurately quantify the relationship (normally in terms of elasticities) between the change in the aggregate level of tourism flows and the change in any of the explanatory variables. The final goal is often to forecast tourism demand, a key issue for the industry (Song and Li 2008). Whatever the case, from a literature review, the income in source markets has been demonstrated to be a dominant explanatory variable and it is the most widely discussed determinant of international tourism demand (Peng, Song, and Crouch 2014).

Although studies of tourism demand modeling usually assume that income’s effect on tourism demand remains stable, irrespective of changes in the remaining factors, more recently an increasing body of literature has started to draw attention to the variability of the estimated income elasticity, striving to explain why the relationship between tourism and income can vary as a result of different factors. Empirical literature has identified various factors that could explain the variability of the income elasticity, such as financial and economic crises (Smeral 2009), the point in the business cycle (Smeral and Song 2013; Gunter and Smeral 2016), structural changes (Song, Witt, and Li 2009), different time periods (Gunter and Smeral 2016, 2017), different destination–origin pairs (Peng, Song, and Crouch 2014), and the income level at the destination country or continent (Martins, Gan, and Ferreira-Lopes 2017).

Controversy over the income elasticity’s assumed constancy is nothing new. According to the hypothesis put forward by Morley (1998), low-income and high-income countries are expected to have a low elasticity, while middle-income countries are thought to be the most elastic. In other words, people in the wealthiest countries would not relinquish their holiday in the event of an economic recession (and neither would they significantly increase their travel activities in the event of economic growth) because they would understand travel to form part of their regular consumption (a necessary good). That is, they might adjust their travel expenses but they would continue to travel. However, people in middle-income countries would be the most sensitive, reacting significantly to changes in income levels because they understand tourism to be a luxury good. Finally, people in the poorest countries would not have enough money for international travel. So, since their demand for tourism is limited, they would not react to changes in income levels.

In this framework, the objective of our study is to investigate the relationship between personal income and the income elasticity of tourism demand and to evaluate the veracity of the hypothesis about the inverted-U relationship between income elasticity and personal income. Although various empirical applications that have explored variability in income elasticities (Gunter and Smeral 2016, 2017; Rosselló, Aguilo, and Riera 2005; Smeral and Song 2013; Song and Wong 2003; Song and Li 2008; Song, Kim, and Yang 2010) seem to point to the validity of U-relationship hypothesis, none of these papers has directly attempted to confirm the hypothesis that personal income determines the income elasticity of tourism demand, in the sense that the richest and poorest people tend not to react very much to changes in income, while the middle classes are the most sensitive individuals. It is worth noting that an in-depth analysis of the special relationship between income and tourism is an issue of strategic importance in long-term tourism forecasts. Thus, in a world where future predictions of the world GDP point to a positive trend in coming decades (PwC 2017), if the GDP is acknowledged to play a key role in long-run tourism projections (UNWTO 2011), then an accurate insight into how income determines tourism flows is crucial for strategic planning by both public administrations and private-sector tourism stakeholders.

To achieve the above objective, this research study proposes the application of various different strategies, using aggregate tourism demand models for international tourism in the context of a gravity model. The models are estimated for different income and timing subsets, using interactions between the income elasticity and level of income in the source market. Yearly data from 1995 to 2015 and for bilateral flows between 200 countries were used to estimate a gravity model.

The rest of the study is organized as follows. The next section reviews some previous papers that have explored and quantified the relationship between tourism demand and income. The third section presents the methodology and data used to make an in-depth analysis of the special relationship of the national income variable and its effect on tourism demand. The fourth section presents the empirical application and, finally, the fifth section summarizes the results, before going on to present the general conclusions, point out the limitations of the study and propose research ideas to extend the analysis in the future.

Income and Tourism Demand: A Review of Recent Literature

Economic utility theory has led economists to specify demand as a function of determining variables, with a decisive influence for individuals. According to the microeconomic theory of tourism demand, the independent variables in the models should include measures of income, fares, prices, and other variables that tourists encounter at a destination (Morley 1992). The empirical results have shown that the dynamics of tourism demand are mainly determined by income and prices, and more specifically, past research has demonstrated that income in the source market is a dominant explanatory variable and it is the most widely discussed determinant of international tourism demand (Crouch 1992, 1994, 1995; Lim 1999; Peng, Song, and Crouch 2014; Peng et al. 2015).

In empirical exercises using aggregate data, the nominal or real GDP and their per capita forms are the most popular proxies of tourist income (Greenidge 2001; Turner and Witt 2001). Whichever income variable is used, most of the empirical studies demonstrate that in accordance with economic theory, income has a positive effect on tourism demand. Crouch (1996) shows that, in empirical exercises, multiple values are obtained in estimates of the income elasticity of tourism demand modeling. The point is that international tourism should be considered as a luxury product, as indicated by the fact that most studies have estimated an income elasticity of demand of over 1, showing that, as income rises, tourism consumers spend an increasing amount of their income on international travel (Peng, Song, and Crouch 2014). However, in the meta-analysis by Crouch (1996), although the mean income elasticity was 1.86, the standard deviation was 1.78, showing that many empirical applications obtained a value of between 0 and 1 (indicating that tourism is a necessary good), with some of them even obtaining a negative value for the income elasticity (indicating that tourism can be an inferior good). Consequently, although in estimations of tourism demand, the income elasticity is considered to remain constant, it has been agreed that many determinants can cause the value of this elasticity to differ.

Song, Kim, and Yang (2010) proposed the use of confidence intervals in the estimation of tourism demand elasticities. They argued that point estimates give a single value for the parameter of interest, but it does not provide information about its degree of variability. Hence, point estimations could provide biased estimations of true elasticities if it is assumed that this elasticity is a nonlinear function of other parameters in the model. In the context of time series, the assumption of parameter constancy can also be tackled using time-varying parameter models that take account of the possibility of parameter changes over time (Song and Wong 2003; Li, Song, and Witt 2005; Li et al. 2006; Song et al. 2011).

Within the cross section framework, Peng, Song, and Crouch (2014) show that income elasticities can differ considerably across different origin–destination pairs, depending on how deluxe the destinations are. For example, the estimated income elasticity of the tourism demand for Aruba varies from 1.43 for American tourists to 2.52 for Dutch visitors (Croes and Vanegas 2005). Dogru, Sirakaya-Turk, and Crouch (2017) point out that a Giffen good (a good for which the demand grows as its price increases, and vice versa) is always an inferior good, and hence this type of destination is lacking in quality and it will be chosen by tourists when they have no other alternative. Additionally, Papatheodorou (2001) suggests also the Veblen effect of conspicuous consumption, which induces the tourist to pay a higher price for a functionally equivalent good simply for reasons of fashion, image, and prestige.

Economic theory suggests that time can also influence elasticity values. In the short term, tourists’ responses to income and price changes may be constrained by existing travel arrangements. However, in the long run, tourists have enough time to adjust as necessary and they are likely to display more income-elastic behavior (Peng, Song, and Crouch 2014). According to these last authors, the estimated long-run income and (absolute) price elasticities of international tourism demand are expected to be higher than the short-run estimates. In their empirical framework, Song, Witt, and Li (2009) show that tourism demand modeling exercises have often failed to consider long-run co-integration relationships and short-run dynamics. However, in many empirical studies that have calculated the elasticities in both the short and long run, the values of both the long-run income and own-price elasticities are higher than their short-run counterparts, showing that tourists are more sensitive to income/price changes in the long term (Pesaran 1997; Pesaran, Shin, and Smith 2001; Li, Song, and Witt 2005; Mervar and Payne 2007; Seetaram, Forsyth, and Dwyer 2016).

It is also important to add that divergences in income elasticity estimations can also be explained by the way in which tourism demand is measured. Martins, Gan, and Ferreira-Lopes (2017) used the inbound visitor population and on-the-ground expenditures as tourism demand measures, showing that the per capita world GDP is more important in explaining arrivals, although relative prices become more important when expenditures are used as a proxy for tourism demand. They also showed that an increase in the per capita world GDP, the depreciation of the national currency, and a decline in relative domestic prices help to boost tourism demand. Rosselló and He (2019) also show how the use of different measures of tourism demand can explain different income elasticities.

It has also been suggested that the economic environment can also determine income elasticities. Using a global panel data set of 208 destination countries and 7 origin countries between 1985 and 2002, Eugenio-Martin, Martin-Morales, and Sinclair (2008) found that economic development makes a difference to tourist decision making, showing that in countries with high GDPs, differences in economic development are not significant, whereas in developing countries they are. Gunter and Smeral (2016), in reference to the structural changes observed frequently within the leisure sector, showed that structures changed in the long term and its positive impact on certain goods, such as international travel, slows down or disappears, also leading to a decline in income elasticities from period to period. Gunter and Smeral (2017) found that income elasticities were different in distinct growth periods between 2004 and 2014 for both the EU15 and EU28, showing how, on the one hand, in slow growth periods, the income elasticity had a value above 1 while, on the other hand, in the fast growth periods, the elasticity value was below 1. Additionally, the reaction of tourism demand to income changes could be asymmetric during different phases of the business cycle (Gunter and Smeral 2017; Smeral and Song 2013; Smeral 2018). Overall, it could be argued that the economic environment can be captured by income indicators (like the per capita GDP) and, consequently, these studies could be considered to be specific case studies of the more general hypothesis explored in this article, which proposes that the income elasticity will depend on the level of personal income.

Martins, Gan, and Ferreira-Lopes (2017) recently conducted an empirical analysis of world tourism demand of special interest for the purposes of this research study. Using an unbalanced panel of 218 countries over the period 1995–2012, they analyzed the role of the per capita world GDP on tourism demand (among other variables), partitioning the data by the income level (at the destination) and finding robust results of the existence of differences in the estimated value of the income elasticity, depending on the per capita GDP at the destination. In some way, our study expands the study by Martins, Gan, and Ferreira-Lopes (2017), focusing on income’s effect on tourism demand and taking into account not just differences in personal income at the destination but also in the origin country, thus linking current discussion on income’s effect on tourism demand with the hypothesis about the inverted-U relationship between income elasticity and personal income.

Methodology and Data

Methodology

Because personal income at a national level is expected to have a high structural component, a gravity equation was used to explain tourism flows, while also taking into consideration the potential time variability of the determinants. The gravity equation suggests that different bilateral international flows (i.e., trade, tourism, migrations, foreign direct investment, etc.) are expected to increase with the economic size of a country in absolute levels and to decrease as the distance between country pairs grows. Additionally, a set of other determining variables, such as origin and destination characteristics, can be included.

The gravity model has been extensively used in empirical exercises to explain international trade because of its goodness of fit (Deardorff 1998; Anderson and Wincoop 2003). Since tourism is a special type of trade in services, gravity equations have also been used to estimate the magnitude of tourism flows in different contexts (Eilat and Einav 2004; Kimura and Lee 2006; Santana-Gallego, Ledesma-Rodríguez, and Pérez-Rodríguez 2010; Fourie and Santana-Gallego 2011; Falk 2016; Okafor, Khalid, and Then 2018; Park and Pan 2018). Although the use of the gravity equation method has been backed up by international trade theory for many years, only recently has its use been justified in the field of tourism, based on consumer theory (Morley, Rosselló, and Santana-Gallego 2014).

Given the aim of this paper, which focuses on the specific effect of per capita personal income on the income elasticity, the gravity equation that was used considered the fixed effects between country pairs and destination–period pairs and so it centered its attention on structural factors relating to the origin country, which determine the outbound tourism demand. One of the consequences of this choice is that time-invariant country pair characteristics (such as distance or common borders) and time-variant destination country characteristics (such as temperature, prices, or development level in the destination) are not included explicitly in the model since they are absorbed by the fixed effects. Therefore, traditional variables included in gravity models such as distance or characterization at destination level are not required in the specification since they are already controlled by the inclusion of the set of fixed effects. This is a common practice when gravity models are estimated in order to avoid omitted factor bias and centering the attention of the variables of interest for the research (Balli, Ghassan, and Jeefri 2019; Fourie, Rosselló-Nadal, and Santana-Gallego 2019). Analytically, the baseline model can be written as

\ln (T o u_{i j t}) = a + b \ln (G D P p c_{i t}) + c_{k} X_{i t}^{k} + α_{i j} + δ_{j t} + ε_{i j t},

(1)

where the dependent variable $T o u_{i j t}$ reflects tourism demand from origin country i to destination country j during period t; $G D P p c_{i t}$ is the per capita personal income indicator of the origin country; $X_{i t}^{k}$ is a set of determining variables relating to the origin country during period t; and a, b, and c are parameters to be estimated. Moreover, a set of fixed effects that control for dyadic ( $α_{i j}$ ) and destination-year ( $δ_{j t}$ ) characteristics was added to equation (1). Finally, $ε_{i j t}$ is a well-behaved disturbance term. It should be noted that the Tou and GDPpc variables are taken in natural logs, so coefficient b can be interpreted as elasticity.

From the initial equation (1), the dependence of the income elasticity or, in other words, the specific function f between Tou and GDPpc is evaluated following different strategies. First, we consider the nth-order Taylor polynomial of the relationship between tourism demand and the income variables. The introduction of an nth-order Taylor approximation allows for measuring a nonlinear relationship between dependent and independent variables and, consequently, if there is an inverted-U relationship or not. Analytically:

\begin{array}{l} \ln (T o u_{i j t}) = a + b_{1} \ln (G D P p c_{i t}) + b_{2} {[\ln (G D P p c_{i t})]}^{2} + \dots \\ + b_{n} {[\ln (G D P p c_{i t})]}^{n} + c_{k} X_{i t}^{k} + α_{i j} + δ_{i t} + ε_{i j t} . \end{array}

(2)

Thus, assuming that f is derivable and continuous, with the estimation of b₁, b₂, . . . b_n, it is possible to evaluate the function’s behavior within the range of the observed GDPpc variable from a specific data set. The choice of the appropriate n can be done considering a high value and reducing the order of the polynomial following the general to specific strategy (Hoover and Perez 2004). The existence of a maximum can be found through the first derivative of equation (2) in relation to ln(GDPpc), after verifying that the second derivative takes a negative value within the income variable’s range.

The second strategy considers different subsamples of origin countries, selected according to the ordered values of the average GDPpc, and it estimates different (b) parameters for equation (1) for the different subsamples. The effect of per capita personal income on the income elasticity of tourism demand is derived, in this case, from the different estimations of the b parameters. Analytically, we calculate the average GDPpc of the origin countries for the sample period. In our database, we have 192 origin countries with GDPpc data available, so we divide the sample into w groups with the same number of countries (48) in each group, four in our case w={1,2,3,4}, where 1 represents the group of countries with the lowest incomes and 4 the countries with the highest. Although the number of selected groups can be modified without significant differences in the expected results, in our case, when 5 or more groups were considered, the number of insignificant parameters very high. Alternatively, in order to compare our estimates, we also employ a common classification of countries according to the level of income, such as the one provided by the World Bank.

For both strategies, we evaluate general estimations with two different subsamples interacting with the proposed one. On the one hand, we consider the differences in the level of development at the destinations since, according to Martins, Gan, and Ferreira-Lopes (2017), the level of development at a destination can be a significant determinant of the income elasticity. On the other hand, according to previous results on how the economic crisis has affected tourism demand (Gunter and Smeral 2016, 2017; Papatheodorou, Rosselló, and Xiao 2010; Smeral and Song 2013; Smeral 2018), we consider different subsample periods for the years before and after the international economic crisis.

Finally, for estimation purposes, we use a standard OLS estimator with fixed effects as benchmark estimation (OLS-FE). In addition, the Poisson pseudo-maximum-likelihood (PPML) estimator proposed by Silva and Tenreyro (2006, 2011), which correctly accounts for the existence of heteroscedastic residuals, is also used. PPML also allows for zeros in the dependent variable. However, one drawback to tourism data is the fact that it is not possible to discriminate between zero tourism flows and missing values. For this reason, as proposed by Neumayer and Plumper (2016), the PPML estimator is only applied for positive bilateral tourism flows. For both estimation methods, robust standard errors clustered by pairs are used in order to deal with possible heteroscedasticity and autocorrelation.

Data

In accordance with previous literature, the dependent variable $T o u_{i j t}$ used in this study consists of the number of international tourist arrivals from origin country i to destination country j during year t, using data from the United Nations’ World Tourism Organization (UNWTO 2017). As mentioned earlier, one of the main limitations of this database is the fact that it does not report zero tourism flows between country pairs. Missing data can stem from gaps in the data collection procedure or because bilateral tourism flows between some country pairs are recorded only if they are above a certain threshold. Marginal tourism flows from origin countries to destinations are grouped together by the UNWTO under the “other countries” label. Although we acknowledge that the estimates can suffer from a sample selection bias because no zeros are reporte, only positive tourism flows are considered in the analysis. The data set includes 198 countries for the period 1995-2016.

The per capita personal income indicator $G D P p c_{i t}$ is measured through the real per capita GDP of each origin country, and is taken from the World Development Indicators. As mentioned in the methodological section, because the variable of interest is origin specific, we should consider how the control variables are related to the origin country too, while other potential determining variables referring to country pairs (like the distance) or to the destination (the coastline, visa restrictions, etc.) will be captured by the fixed effects.

The control/explanatory variables ( $X_{i t}^{k}$ ) includes Pop, the population of the origin country in order to capture the dimension effect (through the population). Indeed, we want to use population as a control for the size of the country while GDP per capita measures the income level. Here, it is important to mention that GDP per capita and population are not correlated enough (−0.27) to produce statistical problems in our regression. The variable R-Law, the rule of law, is a proxy for the origin country’s law and governance system, which captures perceptions of the extent to which agents have confidence in and abide by the rules of society and, in particular, the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence. This variable ranged from −2.5 (weak) to +2.5 (strong). Population and Rule of law variables are taken from The World Development Indicators. Finally, Terror was used as a proxy for the level of instability and insecurity in the origin country, measured as the number of victims of terrorist attacks per 10,000 inhabitants. The data were obtained from the Global Terrorism Database, which defines terrorism as “the threatened or actual use of illegal force and violence by a non-state actor to attain a political, economic, religious, or social goal through fear, coercion, or intimidation.”

The main descriptive statistics relating to these variables are presented in Table 1.

Table 1.

Descriptive Statistics and Diagnostic Tests.

	Obs	Mean	SD	Minimum	Maximum
Tou_ijt	206,881	75,753	868973	0	7.90E+07
GDPpc_ijt	206,881	20,419	23,006	170.582	192,989
Pop_ijt	206,881	5.75E+07	1.79E+08	9,230	1.40E+09
Rlaw_ijt	206,881	0.36	1.049	−2.13	2.1
Terror_jt	206,881	30.03	158.86	0	3926
Diagnostic tests
Hausman test	χ²(27) =552.76		Prob > χ² = 0.0000
Poolability test	F(14232, 186681) = 162.84		Prob > F = 0.0000
Wald test heteroscedast.	χ² (14233) = 1.0e+08		Prob > χ² = 0.0000
Panel Unit root test
Fisher-type unit-root test	LnTou_ijt	LnGDPpc_ijt	Pop_ijt	LnRlaw_ijt	LnTerror_ijt
Inverse normal (Z)	−101.9***	13.7	−148.1***	−153.35***	−25.2***
Inverse Logit t (L)*	−165.6***	−28.8***	−266.9***	−240.62***	−53.6***

Moreover, we present some diagnostic tests that confirm that panel fixed effect estimation method is adequate and that there is evidence of heteroscedasticity, so we have to compute robust standard errors. Finally, Fisher-type unit-root tests based on Phillips-Perron tests, which work well in unbalanced panels, are applied. These tests strongly reject the null hypothesis that all the panels contain unit roots for all the series considered.

Empirical Application

The estimation results of the baseline model, equation (1), are presented in Table 2 using the whole sample first and then two different periods relating to the pre-crisis period (1995–2007) and the crisis and postcrisis period (2008–2016). In general, the estimated parameters of the control variables yield the expected signs and sizes, suggesting that the model is correctly specified.

Table 2.

Baseline Model.

	All Sample		1995–2007		2008–2016
	OLS-FE	PPML	OLS-FE	PPML	OLS-FE	PPML
Ln(GDPpc_ijt)	0.774***	0.910***	0.623***	0.886***	0.748***	1.057***
Ln(GDPpc_ijt)	(0.0327)	(0.164)	(0.0459)	(0.170)	(0.0457)	(0.252)
Ln(Pop_ijt)	0.449***	0.803***	0.143	1.171***	0.698***	0.395*
Ln(Pop_ijt)	(0.0605)	(0.260)	(0.0988)	(0.229)	(0.0858)	(0.236)
Rlaw_ijt	0.0684***	−0.189**	0.00932	−0.157*	0.0757***	−0.0251
Rlaw_ijt	(0.0198)	(0.0825)	(0.0241)	(0.0941)	(0.0244)	(0.0512)
Terror_ijt	0.0569	0.140***	0.171*	−0.111	0.0403	0.146***
Terror_ijt	(0.0375)	(0.0533)	(0.0964)	(0.142)	(0.0274)	(0.0516)
Observations	205,998	205,998	103,211	103,211	102,296	102,296
R-Square	0.968		0.973		0.981

Note: Robust standard errors clustered by pairs in parentheses. ***p<0.01, **p<0.05, *p<0.1.

Destination–year and country-pair fixed effects are included but not reported.

As expected, countries with bigger populations generate more international tourists. Despite the clear evidence that this result provides, it should be highlighted that, if we focus on PPML estimations, the parameter for the postcrisis period is lower than the pre-crisis one (and significant only at a 10% level). This should be tied in with the higher propensity to travel domestically during and after the crisis period, an option that is more easily available in big countries. The coefficient for terrorism at the origin country is very low or even insignificant in most of the regressions. However, the positive sign, when it is significant, shows that if there are attacks in the origin country, people tend to travel abroad more. Additionally, a good law and governance system in the origin country is also related to a higher propensity to travel internationally.

As for the variable of interest, personal income, all the estimated parameters show a positive significant sign, highlighting the expected positive relationship between income and tourism demand. Moreover, the income elasticities estimated by the PPML model are always greater than the ones for the OLS-FE estimates. Focusing specifically on the results of the PPML model, the values of the coefficients show a mean income elasticity of around one, which is in line with previous estimates found in past literature (Peng et al. 2015; Li et al. 2006; Lim 1997, 1999; Crouch 1994, 1996, among others). The pre-post crisis analysis shows an increase in the income elasticity values in the most recent period. Again, this coincides with the related literature, which propounds relatively higher income elasticities during sluggish growth periods as compared with fast ones (Smeral 2014, 2018). As a possible explanation for this, Smeral (2014) suggests that consumers behave like loss averters. That is, consumers will not reduce their consumption and travel plans during the slowdown period when they expect a recession.

However, as argued in the methodological section, the variability of the income elasticity can be related to the country’s income level, and this issue is analyzed by considering a fourth-order Taylor polynomial, the order that obtained better results in accordance with equation (2), whose estimated parameters are shown in Table 3. Thus, Table 3 shows that the coefficients of the polynomial of the GDPpc variable confirm the existence of a complex (nonlinear) relationship between personal income and the income elasticity at a country level. Additionally, according to the estimated coefficients of the polynomial, an inverted U–shape relationship is found, providing evidence of the hypothesis that low-income and high-income countries tend to be low-elastic while medium-income countries are the most elastic.

Table 3.

Nonlinear Model.

Method PPML	All Sample	1995–2007	2008–2016
Ln(Pop_ijt)	1.160***	1.354***	0.504**
Ln(Pop_ijt)	(0.137)	(0.247)	(0.202)
Rlaw_ijt	0.122**	−0.091	0.148**
Rlaw_ijt	(0.0519)	(−0.135)	(0.0575)
Terror_ijt	−0.138**	−0.142*	−0.012
Terror_ijt	(−0.0643)	(−0.085)	(−0.042)
Ln(GDPpc_ijt)	23.67**	8.545	39.94**
Ln(GDPpc_ijt)	(11.930)	(18.790)	(−19.200)
Ln(GDPpc_ijt)²	−4.667**	−1.714	−8.315**
Ln(GDPpc_ijt)²	(−2.14)	(−3.434)	(−3.436)
Ln(GDPpc_ijt)³	0.417**	0.164	0.756***
Ln(GDPpc_ijt)³	(0.170)	(0.273)	(0.268)
Ln(GDPpc_ijt)⁴	−0.0137***	−0.0056	−0.0247***
Ln(GDPpc_ijt)⁴	(−0.005)	(−0.008)	(−0.008)
Observations	205,998	103,211	102,296

Note: Robust standard errors are clustered by pairs in parentheses. Destination–year and country-pair fixed effects are included but not reported.

***

p < 0.01, **p < 0.05, *p < 0.1.

Further insights into these results can be obtained by calculating the reference elasticity for the mean of the GDPpc income and the turning points. Thus, using the all sample equation, if we take a GDPpc at US$20,419 and we simulate an increase of 1% on this GDPpc we get a mean increase of 1.3% in tourist arrivals or, in other words, an income elasticity of 1.3. The turning point analysis, that is, where the GDPpc is expected that, on average, a country will display a higher income elasticity, gets the maximum at US$40,134. Thus, when a country’s per capita GDP exceeds this threshold, its income elasticity is expected to become less elastic. According to data for 2017, a total of 29 countries exceeded this value. It is important to note that for the postcrisis period, this threshold has increased to US$59,874. Furthermore, again the income elasticity is found to be larger for the crisis and recovery period from 2008 to 2016.

Finally, the income elasticity’s variability and its relationship with a country’s income level are also evaluated by considering different subsamples (according to the income level) and the estimations of the income elasticity parameter. In this case, the results are presented in Table 4.

Table 4.

Income Groups by Origin Countries.

	Low Income		Low-Middle Income		High-Middle Income		High Income
Method PPML	Average Income	World Bank	Average Income	World Bank	Average Income	World Bank	Average Income	World Bank
Ln(GDPpc_ijt)	0.586**	0.447***	0.783***	0.763***	1.099***	0.681***	−0.0507	0.150
Ln(GDPpc_ijt)	(0.233)	(0.126)	(0.135)	(0.149)	(0.217)	(0.240)	(0.171)	(0.235)
Ln(Pop_ijt)	1.339*	−1.022**	0.671	1.409**	0.554	−0.0239	1.322***	1.137***
Ln(Pop_ijt)	(0.732)	(0.444)	(0.524)	(0.551)	(0.425)	(0.394)	(0.179)	(0.175)
Rlaw_ijt	0.166	0.191*	−0.133	−0.0913	0.259***	0.273***	−0.104	−0.146
Rlaw_ijt	(0.114)	(0.109)	(0.250)	(0.202)	(0.0773)	(0.0863)	(0.0783)	(0.0971)
Terror_ijt	−0.0199	−0.0162	0.242***	0.209***	0.493**	0.652**	−0.339**	−0.355**
Terror_ijt	(0.0721)	(0.0709)	(0.0534)	(0.0522)	(0.226)	(0.258)	(0.165)	(0.166)
Countries	48	39	48	58	48	39	48	56
Observations	37,044	30,942	43,521	51,805	49,928	41,072	74,367	81,063

Note: Robust standard errors are clustered by pairs in parentheses. Destination–year and country-pair fixed effects are included but not reported.

***

p < 0.01, **p < 0.05, *p < 0.1.

Table 4 presents the results for the estimate income elasticity for the different income groups according to both classifications, average income levels for the whole sample period and World Bank classification, using 2016 as reference year. Results show that there are different income elasticities according to the income level in the origin country. For the average income classification, it seems clear that low and higher-income countries are less elastic than middle-income countries. First, it is noteworthy that the coefficient of the income variable is highly significant in lower-to-middle-income and middle-to-upper-income countries with striking differences in elasticity values. Low-income countries have a less elastic value, which, additionally, is only significantly different from 0 at a 10% confidence interval. In the case of high-income countries, the elasticity coefficient is statistically equal to zero. Thus, the results indicate that income only has a clear significantly positive impact on tourism in the case of middle-income countries.

Additionally, middle-to-upper-income countries are more income elastic when compared with lower-to-middle-income ones. This definitely shows that the former are more sensitive to an increase in income and ultimately to tourism demand. In other words, it was found that people from high-income origin countries do not relinquish their holiday (or do not significantly increase/decrease their travel behavior in the event of economic growth/an economic crisis) because they understand travel to be part of their regular consumption (a necessary good), whereas people from medium-income origin countries are the most sensitive ones, reacting significantly to changes in income levels because they understand tourism to be a luxury good. In the case of people from the poorest countries, they do not have enough money to make international trips so they do not react to changes in income levels.

Estimates using the World Bank income classification yield relatively similar conclusions. That is, middle-income countries (low and high) present larger income elasticities than the low and high income groups, although the income elasticity of upper-to-middle-income countries is lower than that for the low-middle income group. In any case, the average income classification is our preferred one since the classification of countries in each group of income according to the World Bank could change (depending on their economic growth). Moreover, the number of countries in each group is quite different which might cause that the estimations lose power and representativeness.

Furthermore, it is possible to explore this relationship further, considering the pre-crisis period (1995–2007) and the crisis and postcrisis period (2008–2016). For simplicity, here we focus our attention on the income elasticity estimations and we use the average income classification. Figure 1 shows that there are differences in the behavior of the elasticity, attributable to the different time periods under consideration and the income groups of the origin countries. The estimates for the full sample reproduce the results for the whole period, shown previously in Table 5. In this case, it is important to note that, in keeping with the previous strategy, the inverted-U shape is clearer for the crisis and postcrisis period (2008–2016). During this period, a maximum of 1.28 was estimated for the income elasticity for medium-to-high-income countries.

Figure 1.

Income elasticity: Different income groups and time periods.

Table 5.

Income Elasticity by Income Level of the Destination Country.

		Income Level Origin Countries
	Method PPML	Low Income	Low-Medium Income	High-Medium Income	High Income
Income level of destination countries	Low developed	0.521***	0.600***	0.577	−0.423***
	Low developed	(0.191)	(0.161)	(0.550)	(0.0839)
	Observations	15,635	17,517	19,452	31,959
	High developed	0.617**	0.823***	1.272***	0.933***
	High developed	(0.314)	(0.171)	(0.196)	(0.174)
	Observations	21,409	25,983	30,458	41,849

Note: Robust standard errors are clustered by pairs in parentheses. Destination–year and country-pair fixed effects are included but not reported. PPML = Poisson pseudo-maximum likelihood.

***

p < 0.01, **p < 0.05, *p < 0.1.

The final segmentation considered in this article explores the role of income in the destination country as an additional determinant of the income elasticity. For this purpose, we divided the sample into two main groups. The first one (low-developed destinations) includes low-income and low-middle income destinations, while the second one (high-developed destinations) includes high-middle and high-income countries. The reason for this division is to see whether, according to previous literature (Martins, Gan, and Ferreira-Lopes 2017), differences in income elasticity values exist because of the level of income at the destination.

Table 5 shows the interaction between the income level of the origin country (w) and the destination’s income level. In general, we observe higher income elasticity for tourism demand in the case of destinations with a higher level of development. However, significant differences in the estimated income elasticities cannot be observed when origin countries with low incomes are analyzed. Hence it seems that outbound tourism is inelastic in countries with a low income, whatever the level of development at the destination. In other words, in the case of poor countries, changes in income levels do not lead to big changes in outbound flows, neither to destinations with a high level of development nor to less developed ones. As in previous cases, countries with (low and high) middle incomes are characterized by the highest income elasticities. However, the maximum elasticity corresponded to origin countries with high-middle income visiting destinations with a high level of development and being not significant for the flow toward less developed countries.

The most interesting case was high-income origin countries. Although on average these countries are income inelastic in behavior, when their inhabitants travel to more developed countries, the income elasticity becomes positive and significant. Thus, it seems that in the case of origin countries with high incomes, on the one hand, travel to high-income destination countries is not regarded as a necessary good, since the elasticity is increasingly close to one and, on the other hand, travel to less developed countries is seen as an inferior good. In other words, in situations of economic expansion (crisis), high-income countries would reduce (increase) their trips to less developed countries and increase (reduce) them to more developed ones. Thus there seem to be clear variations in the income elasticity depending on the income in both the origin and destination countries.

Conclusion

Because of the tourism sector’s substantial economic relevance and public administrations’ expectations that this sector will boost the growth and development of economies, extensive literature can be found that deals with the evaluation of the determinants of tourism demand. The identification, analysis, and measurement of the impacts of the determinants of tourism demand are central to any attempt to understand and explain past changes and to anticipate possible trends in future tourism flows. In this context, from a literature review, it can be seen that tourist income, which is often assessed by taking the GDP in the origin country, is a dominant explanatory variable in tourism demand modeling.

This article conducted an in-depth analysis of the relationship between tourism and income by investigating heterogeneity in the nexus between income elasticity and tourism demand, reviving Morley’s hypothesis (1998) that low-income and high-income countries tend to have low income elasticity while medium-income countries are the most income elastic. Using yearly data for 208 origin countries and 196 destination countries for the period 1995–2016, different gravity models were estimated for different income and time subsets, based on the pre– and post–financial crisis and using interactions with the level of income at the destination market.

The results show that the income elasticity is significantly determined by per capita income in the origin country, verifying the inverted-U relationship between income elasticity and per capita personal income. Thus, outbound tourists from middle-income countries are found to be the most elastic, while tourists from high- and low-income countries show a lower (sometimes not significant) relationship. The segmentation analysis further qualifies these results. On the one hand, the inverted-U relationship seems to be more significant, and the peak in the income elasticity is higher during the crisis and postcrisis period (2008–2016). On the other hand, when the segmentation analysis was based on income at the destination countries, the results for the wealthiest origin countries reflect very different behaviors depending on whether tourists from these countries travel to high-income countries or to the poorest ones. Future research should go in depth in these results especially in reference to visiting low-income and low-to-medium-income country groups from less developed countries.

In any case, the impossibility of obtaining data at the individual level means that the results must be interpreted with caution and refer to large average trends. The use of aggregated data limits the faculty to evaluate more complex relationships that may exist between personal income and international tourism consumption. Additionally, the change observed in the two periods analyzed limits the possibility of extending these results to the long run and, therefore, contingent the long-term tourist predictions to the sensitivity that consumers will show in relation to income increases.

If income in source markets is acknowledged to play a key role in generating tourism demand, then long-term forecasts needed for strategic planning by public and private tourism managers must be made with as much precision as possible. The results of this study should be useful in facilitating more accurate research into long-term tourism projections.

Further studies should explore different ways to extend this research. On the one hand, at an aggregate level, different segmentations could be considered that might offer a further insight into the specific relationship between income and tourism demand. On the other hand, the relationship between income and tourism consumption could also be explored, using survey data and micro-econometric techniques to examine income elasticities at an individual level. In this way, our knowledge of this particular relationship could be enhanced, using different case studies of destination and/or origin countries.

Footnotes

Acknowledgements

We wish to thank UNWTO Statistics Department for kindly providing us with the tourist data for this study.

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) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: We acknowledge the Agencia Estatal de Investigación (AEI) and the European Regional Development Funds (ERDF) for its support to the project ECO2016-79124-C2-1-R (AEI/ERDF, EU).

ORCID iDs

Aon Waqas-Awan

Jaume Rosselló-Nadal

Maria Santana-Gallego

Author Biographies

Aon Waqas-Awan is has Master’s in Management Sciences from IMSciences, Peshawar (Pakistan) and a PhD in Applied Economics at the University of the Balearic Islands (Spain). He has worked as Lecturer at Shaheed Benazir Bhutto University, Shaheed Benazirabad (Pakistan). His research interests include tourism economics with special focus on climate change and long-term issues.

Jaume Rosselló-Nadal is a Full Professor of Economics at the University of the Balearic Islands (Spain). His research interests include tourism demand modelling and environmental issues. Main results of his work have been published in academic Journals like Annals of Tourism Research, Tourism Management, Tourism Economics, Journal of Sustainable Tourism and Journal of Travel Research.

María Santana-Gallego is Associate Professor of Economics at the University of the Balearic Islands (Spain). Her research focuses on the international economics and tourism economics. She has widely published in peer-reviewed international journals on these topics, such as Tourism Management and Annals of Tourism Research, among others.

References

Anderson

J. E.

Van Wincoop

2003. “Gravity with Gravitas: A Solution to the Border Puzzle.” American Economic Review 93:170–92.

Balli

Ghassan

H. B.

Jeefri

E. H.

2019. “Towards Understanding GCC Outbound International Tourism.” Journal of Policy Research in Tourism, Leisure and Events. https://doi.org/10.1080/19407963.2018.1556671.

Croes

R. R.

Vanegas

2005. “An Econometric Study of Tourist Arrivals in Aruba and Its Implications.” Tourism Management 26 (6): 879–90.

Crouch

G. I.

1992. “Effect of Income and Price on International Tourism.” Annals of Tourism Research 19 (4): 643–64.

Crouch

G. I.

1994. “The Study of International Tourism Demand: A Review of Findings.” Journal of Travel Research 33 (1): 12–23.

Crouch

G. I.

1995. “A Meta-analysis of Tourism Demand.” Annals of Tourism Research 22 (1): 103–18.

Crouch

G. I.

1996. “Demand Elasticities in International Marketing: A Meta-analytical Application to Tourism.” Journal of Business Research 36 (2): 117–36.

Deardorff

A. V.

1998. “Determinants of Bilateral Trade: Does Gravity Work in a Neoclassical World?” In The Regionalization of the World Economy, edited by Frankel

Jeffrey A.

Chicago: University of Chicago Press.

Dogru

Sirakaya-Turk

Crouch

G. I.

2017. “Remodeling International Tourism Demand: Old Theory and New Evidence.” Tourism Management 60:47–55.

10.

Eilat

Einav

2004. “Determinants of International Tourism: A Three Dimensional Panel Data Analysis.” Applied Economics 36:1315–28.

11.

Eugenio-Martin

J. L.

Martin-Morales

Sinclair

M. T.

2008. “The Role of Economic Development in Tourism Demand.” Tourism Economics 14 (4): 673–90.

12.

Falk

2016. “A Gravity Model of Foreign Direct Investment in the Hospitality Industry.” Tourism Management 55:225–37.

13.

Fourie

Santana-Gallego

2011. “The Impact of Mega-Sport Events on Tourist Arrivals.” Tourism Management 32:1364–70.

14.

Fourie

Rosselló-Nadal

Santana-Gallego

2019. “Fatal Attraction: How Security Threats Hurt Tourism.” Journal of Travel Research 59 (2): 209–19.

15.

Greenidge

2001. “Forecasting Tourism Demand: An STM Approach.” Annals of Tourism Research 28 (1): 98–112.

16.

Gunter

Smeral

2016. “The Decline of Tourism Income Elasticities in a Global Context.” Tourism Economics 22 (3): 466–83.

17.

Gunter

Smeral

2017. “European Outbound Tourism in Times of Economic Stagnation.” International Journal of Tourism Research 19:269–77.

18.

Hoover

K. D.

Perez

S. J.

2004. “Truth and Robustness in Cross-Country Growth Regressions.” Oxford Bulletin of Economics and Statistics 66 (5): 767–98.

19.

Kimura

Lee

H.-H.

2006. “The Gravity Equation in International Trade in Services.” Review of World Economics 142:92–121.

20.

Song

Witt

S. F.

2005. “Recent Developments in Econometric Modeling and Forecasting.” Journal of Travel Research 44:82–99.

21.

Wong

K. K.

Song

Witt

S. F.

2006. “Tourism Demand Forecasting: A Time Varying Parameter Error Correction Model.” Journal of Travel Research 45 (2): 175–85.

22.

Lim

1997. “Review of International Tourism Demand Models.” Annals of Tourism Research 24 (4): 835–49.

23.

Lim

1999. “A Meta-analytic Review of International Tourism Demand.” Journal of Travel Research 37 (3): 273–84.

24.

Martins

L. F.

Gan

Ferreira-Lopes

2017. “An Empirical Analysis of the Influence of Macroeconomic Determinants on World Tourism Demand.” Tourism Management 61:248 260.

25.

Mervar

Payne

J. E.

2007. “Analysis of Foreign Tourism Demand for Croatian Destinations: Long-run Elasticity Estimates.” Tourism Economics 13 (3): 407–20.

26.

Morley

C. L.

1992. “A Microeconomic Theory of International Tourism Demand.” Annals of Tourism Research 19 (2): 250–67.

27.

Morley

C. L.

1998. “A Dynamic International Demand Model.” Annals of Tourism Research 25 (1): 70–84.

28.

Morley

C. L.

Rosselló

Santana-Gallego

2014. “Gravity Models for Tourism Demand: Theory and Use.” Annals of Tourism Research 48:1–10.

29.

Neumayer

Plumper

2016. “Spatial Spill-overs from Terrorism on Tourism: Western Victims in Islamic Destination Countries.” Public Choice 169 (3/4): 195–206.

30.

Okafor

L. E.

Khalid

Then

2018. “Common Unofficial Language, Development and International Tourism.” Tourism Management 67:127–38.

31.

Papatheodorou

. 2001. “Why People Travel to Different Places.” Annals of Tourism Research 28(1): 164–179

32.

Papatheodorou

Rosselló

Xiao

2010. “Global Economic Crisis and Tourism: Consequences and Perspectives.” Journal of Travel Research 49 (1): 39–45.

33.

Park

S. Y.

Pan

2018. “Identifying the Next Non-stop Flying Market with a Big Data Approach.” Tourism Management 66:411–21.

34.

Peng

Song

Crouch

G. I.

2014. “A Meta-analysis of International Tourism Demand Forecasting and Implications for Practice.” Tourism Management 45:181–93.

35.

Peng

Song

Crouch

G. I.

Witt

S. F.

2015. “A Meta-analysis of International Tourism Demand Elasticities.” Journal of Travel Research 54 (5): 611–33.

36.

Pesaran

M. H.

1997. “The Role of Economic Theory in Modeling the Long Run.” Economic Journal 107 (440): 178–91.

37.

Pesaran

M. H.

Shin

Smith

R. J.

2001. “Bounds Testing Approaches to the Analysis of Level Relationships.” Journal of Applied Econometrics 16 (3): 289–326.

38.

PwC (PricewaterhouseCoopers LLP). 2017. “The World in 2050.” PricewaterhouseCoopers LLP:UK. https://www.pwc.com/gx/en/issues/the-economy/assets/world-in-2050-february-2015.pdf (accessed October 24, 2018).

39.

Rosselló

Aguilo

Riera

2005. “Modeling Tourism Demand Dynamics.” Journal of Travel Research 44:111–16.

40.

Rosselló

2019. “Tourist Arrivals versus Tourist Expenditures in Modelling Tourism Demand.” Tourism Economics. doi:10.1177/1354816619867810.

41.

Santana-Gallego

Ledesma-Rodríguez

F. J.

Pérez-Rodríguez

J. V.

2010. “Exchange Rate Regimes and Tourism.” Tourism Economics 16 (1): 25–43.

42.

Seetaram

Forsyth

Dwyer

2016. “Measuring Price Elasticities of Demand for Outbound Tourism Using Competitiveness Indices.” Annals of Tourism Research 56:65–79.

43.

Silva

J. S.

Tenreyro

2006. “The Log of Gravity.” Review of Economics and Statistics 88 (4): 641–58.

44.

Silva

J. S.

Tenreyro

2011. “Further Simulation Evidence on the Performance of the Poisson Pseudo-Maximum Likelihood Estimator.” Economics Letters 112 (2): 220–22.

45.

Smeral

2009. “The Impact of the Financial and Economic Crisis on European Tourism.” Journal of Travel Research 48 (1): 3–13.

46.

Smeral

2014. “Forecasting International Tourism with Due Regard to Asymmetric Income Effects.” Tourism Economics 20 (1): 61–72.

47.

Smeral

2018. “Variations in Seasonal Outbound Travel across the Business Cycles.” Journal of Travel Research 57 (7): 936–46.

48.

Smeral

Song

2013. “Varying Elasticities and Forecasting Performance.” International Journal of Tourism Research 17:140–50.

49.

Song

Kim

J. H.

Yang

2010. “Confidence Intervals for Tourism Demand Elasticity.” Annals of Tourism Research 37 (2): 377–96.

50.

Song

2008. “Tourism Demand Modeling and Forecasting—A Review of Recent Research.” Tourism Management 29 (2): 203–20.

51.

Song

Witt

S. F.

Athanasopoulos

2011. “Forecasting Tourist Arrivals Using Time-Varying Parameter Structural Time Series Models.” International Journal of Forecasting 27 (3): 855–69.

52.

Song

Witt

S. F.

2009. The Advanced Econometrics of Tourism Demand. New York: Routledge.

53.

Song

Wong

K. K. F.

2003. “Tourism Demand Modeling: A Time-Varying Parameter Approach.” Journal of Travel Research 42 (1): 57–64.

54.

Turner

L. W.

Witt

S. F.

2001. “Factors Influencing Demand for International Tourism: Tourism Demand Analysis Using Structural Equation Modeling, Revisited.” Tourism Economics 7 (1): 21–38.

55.

UNWTO (United Nations World Tourism Organization). 2011. Tourism Towards 2030: Global Overview. Madrid: UNWTO.

56.

UNWTO (United Nations World Tourism Organization). 2017. Compendium of Tourism Statistics. Madrid: UNWTO.