Tourism Employment and Economic Growth: Dynamic Panel Threshold Analysis

Abstract

The manuscript reports on findings on the interconnection between tourism employment and economic growth for the selected OECD member states. The dynamic panel threshold regression method was used to analyze the data, where the threshold variable was tourism employment, and the growth of gross national income and value added by activity services were dependent variables in the corresponding models. The dataset covered the period between 2008 and 2020. Both marginal effects indicated positive implications of tourism employment on economic growth. A percent rise in tourism employment leads to an increase in gross national income by 0.15% (in the low regime) and 0.61% (in the high regime). Yet, the results revealed a negative marginal effect of tourism employment on value added by activity services. The outcomes explain that a percent rise in tourism employment in the average country will lead to a decrease in the value added by activity services, as a percentage of value added, by 0.07% (low regime) and 0.09% (high regime). Therefore, the applications of this study are twofold—the first one is its contribution to existing theoretical knowledge through the filling of the literature gaps, and the second one is related to advances in the standing policies. The main limitations and the proposal for future research are the application of random effects and smooth transition threshold models as an alternative to the indicator functions.

1. Introduction

In the last twenty years, the literature has supported that threshold models have a wide application [1,2,3,4] in econometric modeling for nondynamic panels with individual fixed effects [5]. González et al. [6] specify this method and propose a panel smooth transition regression model allowing the variation of constants gradually from one regime to another. However, Kremer et al. [7] stressed the importance of modeling by introducing endogenous regressors with a dynamic panel regression (to overcome the endogeneity problem). In their study, they propose a type of hybrid dynamic model by joining the forward orthogonal deviations transformation and the instrumental variable estimation of the cross-section (motivated by the previous work of [1,8]). Seo and Shin [9] propose how to simultaneously model nonlinear asymmetric dynamics and cross-sectional heterogeneity. These authors suggest the first-difference transformation and estimate their properties by the waning threshold effect asymptotic of the Hansen model.

A few years after, the implication of a dynamic panel threshold model is very useful to test the existence of a specific kind of nonlinear asymmetric dynamics (Zeitun and Goaied [10]), where both threshold variables and regressors are permissible to be endogenous [9]. These characteristics have provided this model with excellent performance in making inferences about underlying data that exceed limitations when it is not possible to use a linear model. In mathematical terms, the conditional expectation of the endogenous inconstant represents a nonlinear function of the exogenous inconstant and the lag of the endogenous inconstant [11]. In other words, this is a nonlinear mechanism process of data. The dynamic panel threshold model proved to be adequate not only for testing nonlinear dynamic transformations but also for being used with a smaller dataset. This is one of the advantages of the dynamic panel threshold model compared with different threshold and switching regression models.

Although the travel industry has been recognized as a strong mechanism in increasing employment in both developed and developing regions, enhancing infrastructural projects, gross national income, GDP, and the growth of value added and creating additional export income [12,13,14,15,16], there are still only a few tourism-related studies applying dynamic panel models [17,18,19,20]. The travel industry impacts employment in many aspects; the scope of job varieties and skills is wide, and it is located across diverse sectors at multiple levels, from municipal to global [21]. Tourism workers are highly flexible [22], frequently in the form of exploitative employment of migrants [23]. Even more, according to WTTC [24], the travel industry contributed to the employment of 289 million persons globally in 2021, which represents 6.1% of the global GDP (approx. USD 5.8 billion). Despite these facts, the literature has essentially ignored the sophistication of the issue [21,25] and the importance of the population size working in the industry. In this respect, the avoidance of dynamic panel models in the field of tourism is not clearly justified. Moreover, Sequeira and Maçãs Nunes [26] shed more light on the fact that the tourism and hospitality industry may be correlated with diverse geographic or cultural features or human capital and may not be a self-determining element of progression. Quantitative forecasting dominates the literature and focuses on causal and dynamic econometric determinants using time series analysis [27]. The recognition of dynamic causal connections between tourism demand drivers and underlying (and more static) nonmarketed assets is central to regionally specific approaches in tourism [28]. Therefore, having data for diverse regions, dynamic panel threshold models represent a more appropriate approach to deal with the potential endogeneity of hospitality and tourism.

Having this in mind, these articulations inspire the primary objective of the research—the application of the dynamic panel regression model in the examination of the role of tourism in the economy sector. In this respect, the significance of the tested variables can shed more light on understanding the theory of tourism effects. Second, a large number of studies in this field use nondynamic models (including threshold models), where it is necessary to improve or question the endogeneity bias (about this issue, see for example the discussion of [29,30]). One of the best features of the dynamic panel threshold model is that in threshold regression, endogenous variables are allowed when assessing the threshold effect [15]. According to this, applying this model solves the endogeneity problem that has been challenged in many previous studies and improves the performance of the classical threshold model for nondynamic panels with individual fixed effects [5]. Third, this is the first study that investigates the nexus between tourism employment and economic growth by complex, dynamic, and threshold effects. A few published studies with a similar approach address the broader connection between economic growth and tourism development [31] or globalization [15], which is certainly of value. However, this study focuses on tourism employment as a significant factor, considering that 10% of laborers are employed in the tourism sector worldwide [24]. The contributions are even greater given the sporadic research in the field of dynamic panel regression testing in the travel industry and its upcoming popularization in international tourism studies.

Given that most studies in this area use different regression models, there are some common features: (a) they mostly include a group of countries or regions and (b) they typically study the effect of tourism indicator(s) on the economic growth of regions, wellbeing, and other macro factors. For example, Brida et al. [32] use a dynamic panel model to examine the long-term elasticities between economic growth (real GDP per capita) and tourism receipts for 27 Brazilian states. Lanza et al. [33] test an almost ideal demand econometric system model for analysis of the long-term impact of specialization in tourism in 13 OECD economies. Rosentraub and Joo [34] find that the tourism industry contributes to regional economic development through sports and amusements, according to the test of a log-linear model in 300 metropolitan areas globally. Çiftçioğlu and Sokhanvar [35] use a trivariate VAR model to detect the nexus between specialization in the travel industry and sustainable development for 30 East Asian-Pacific tourism destinations. Demir et al. [36] report that economic insecurity has a significant undesirable influence on tourism investments, by using a panel econometric model consisting of 101 OECD and non-OECD member countries.

The study aims to test and report findings on the performance of a dynamic panel regression model in the examination of a tourism indicator affecting the economy in the selected OECD member states. The study chose tourism employment as the threshold variable to analyze its impact on economic development through gross national income and the value added by activity services. Although numerous outcomes have proven that tourism employment contributes to economic development (like in the transition period reported in [37]; as a positive force for a country’s development as articulated in [38,39,40]; or through the interaction of both income and environmental issues as claimed in [41]), this study makes a further mathematical (and/or econometrical) contribution to the existing literature by testing dynamic effects of nonlinear threshold regression. Accordingly, the research intends to find the answers to the existing literature gaps and thus contribute to the theoretical knowledge of the travel industry’s role in the OECD members’ economies. The main contribution of the study is shaping our understanding of how tourism impacts employment in the selected countries along with how the implications affect gross national income and the value added by activity services.

2. Theoretical Background

The nexus between tourism and economic growth has been studied over the last five decades. The central place in this theory represents the tourism-led growth hypothesis (TLGH) proposed by [42]. According to the TLGH, tourism is an engine of economic growth in both the short and long run. Numerous studies have confirmed this theory by employing various factors of tourism and economic growth. For instance, there is a direct nexus between tourism and economic growth [32,43,44], with a positive relationship such that more tourism means more growth [44]. Chattopadhyay et al. [45] state that each country needs to develop tourism on their current threshold values to contribute to development. Pulido-Fernández and Cárdenas-García [46] find a positive relationship between tourism growth and economic development. The same authors [46] test the TLGH and state that the nexus between tourism growth and economic development is bidirectional: tourism is a tool of economic development, but a higher level of economic development also influences tourism growth. Zuo and Huang [47] emphasize the importance of tourism specialization as an important factor in economic growth. By applying the system generalized method of moments, ref. [47] states that a meaningful inverted-U and N-shaped nexus exists in testing these factors.

Employment in tourism is a starting point and an important indicator of tourism growth. Yet, there is controversy regarding the inverse relationship between its impact on economic growth and the earnings of tourism employees. Firstly, the human capital development in tourism and its employment contribute to the growth [14,15,38,48]. Moreover, tourism is the only or main industry that contributes to the income of the population in some local communities [49,50]. The most important fact is that tourism is the main industry for employment in many countries [46,51,52]. Secondly, although numerous studies confirm the positive impact of tourism employment on growth, wages for tourism employees are among the lowest in the industry. Szivas and Riley [37] point out that tourism workers come from various industries and consider tourism as a refuge industry. Riley [53] indicates even stronger effects of lower labor performance in the tourism sector, connecting this situation with a high proportion of unskilled workers due to the seasonality of the industry and high intraindustry mobility. However, with the emergence of specialization and higher tertiary education in tourism, this phenomenon changes. This statement was confirmed recently in the study of [54]. Yet, ref. [54] points out that tourism wages are still lower because of technological swapping and high labor-intensive input (previously also studied in [25]). The second reason is that the tourism sector shares labor pools with different sectors [54] because they have similar skill needs [25].

Another important issue relates to the applied methodology in the analysis of the impact of tourism indicators on growth. According to Adamou and Clerides [55], the impact of tourism on economic growth mostly depends on the level of specialization, and such a relationship is nonlinear. Chang et al. [56] confirm this statement by analyzing threshold effects of tourism on economic growth in a cross-country sample. After the introduction of Kremer et al.’s [7] dynamic function of panel threshold and endogenous regressors in the last two years, several studies have appeared that study the relationship between tourism and economic growth. This shows an important improvement in the model performance compared to numerous previous studies, considering the problem of endogeneity and nondynamic functions. What gave a new dimension to the research of this relationship is the possibility of analysis of a dynamic function of panel data and threshold effects. For example, [31] used a dynamic panel threshold model to analyze the nexus between tourism and the informal economy in 117 countries. Chiu et al. [56] confirmed that a dynamic panel threshold model is appropriate for testing the nonlinear effect of globalization on inbound tourism, in the case of 53 countries. These studies showed excellent performance of the dynamic panel regression model in their analyses. Chattopadhyay et al. [45] found a strong nonlinear relationship between variables concerning their threshold effects. This study uses nonlinear panel threshold and U-shape functions. However, most of the studies do not analyze the dynamic panel function, and very few concern a nonlinear relationship (for example, nonlinearity and complexity have been previously demonstrated in [12,57]).

Considering the previously discussed studies, this research states two research questions: 1. Is a dynamic panel threshold regression model (with nonlinear features) adequate for analysis of the nexus between tourism and economic growth? 2. To what extent does tourism employment contribute to economic growth? For the purpose of this study, to explain the importance of a dynamic panel threshold regression model, the next sections first explain the development of Hansen’s threshold autoregression model [5], the involvement of Kremer et al.’s [7] endogenous regressors with dynamic panel regression and its adaptation to the variables of this study as well as a model linear transformation suggested by [11]. In the section after, the variables are tested, and the results are discussed.

3. Materials and Methods

3.1. The Sample

The sample data were extracted on 11 February 2022, from OECD Statistics in the period 2008–2020, for thirty OECD countries [58], representing a population of 1.01 billion inhabitants (Figure 1). The present study analyzes the influence of tourism employment (threshold variable) on (1) the growth of gross national income (Model 1) and (2) the value added by activity services as a percentage of value added (Model 2). Model 1 treats tourism GDP and the growth of value added by activity services as a percentage of value added as independent variables. On the other hand, Model 2 treats the growth of gross national income and tourism GDP as independent variables.

The threshold variable for both models is tourism employment. The variables are the following: tourism employment (toe); the gross national income (gni); the value added by activity services as a percentage of value added (vas); and tourism GDP (togdp) [58]. More details and explanations are presented in

Table 1. Explanation of variables.

	Description	Supporting References
toe	This indicator is calculated by the sum of personnel in tourism activities and the annual average of employees stated in national accounts.	Janta and Ladkin [23], Li et al. [48]
gni	The indicator reflects the growth of the gross domestic product, increased by net receipts from abroad from employee reimbursements (residents who basically live inside the economic area but labor abroad or for residents who live and work abroad for short periods), property revenues (such as dividends or interest), and net taxes.	Ganeshamoorthy [59]
vas	This indicator represents the growth of value added, created by the services industries.	Biagi et al. [60]
togdp	It represents the contribution of all industries directly in contact with visitors to the total GDP of the country. This variable is calculated as a percentage of the total GDP.	Canale and De Siano [61]

.

As for the program language and environment, the models were built in the R program.

Table 2. Descriptive statistics.

	ln(gni)	ln(toe)	ln(togdp)	ln(vas)
Mean	10.41	1.796	0.68	4.248
Std. Dev	0.421	0.363	0.017	0.088
Skewness	−0.564	−0.147	0.747	−0.5
Kurtosis	2.864	2.246	2.519	2.467

Note: calculated by authors. Both skewness and kurtosis values are lying within the acceptable limit. Higher standard deviation is observed for ln(gni) series.

represents descriptive statistics of all the variables used in the study.

3.2. Models

The research considered endogenous regressors including lags of the dependent variable. Compared to the classic Hansen’s threshold autoregression model [5] (including the advantage of drawing interpretations about the primary data-generating procedure), in the dynamic panel threshold regression, the coefficients can take both small and robust numbers of different values. These coefficients depend on the rate of the exogenous stationary variable [9]. Dynamic panel threshold regression has relatively recently started to be used in the field of tourism, showing significant potential in researching the nexus of tourism and factors of the informal economy [31], globalization [15], economic growth [44], and more. Having in mind previous research, the dynamic panel threshold regression model might be considered suitable for analyzing the impact of tourism employment on macroeconomic indicators (such as the growth of gross national income and the growth of value added) and to express the proportionate improvement in tourism employment on the economic growth of aggregate countries under consideration. In our study, we express two regime cases, following [11], as:

$$E(y_t⋮x_t)=\{\begin{array}{c}{\mathit{\chi }}_{it}{\beta }_1^{\text{'}}-first regime\\ {\mathit{\chi }}_{it}{\beta }_2^{\text{'}}-second regime\end{array}$$

(1)

Including the nonlinear transformation of elements $y_t⋮x_t$, between the choice of two regime functions $F(\widehat{z_t}, \widehat{\Theta )}$ where values are between 0 and 1, for the regressor ${\mathit{\chi }}_{it}$ and threshold regimes ${\beta }_1^{\text{'}}$ and ${\beta }_2^{\text{'}}$, the nonlinear transformation of the function is expressed as:

$$y_t=\left[1-F\left(\widehat{z_t}, \widehat{\Theta }\right)\right]{\mathit{\chi }}_{it}{\beta }_1^{\text{'}}+F\left(\widehat{z_t}, \widehat{\Theta }\right){\mathit{\chi }}_{it}{\beta }_2^{\text{'}}+{\epsilon }_{i,t}, \mathit{for} Var\left({\epsilon }_{i,t}\right)={\sigma }^2$$

(2)

where ${\epsilon }_{i,t}$ is normally distributed with zero mean and variance ${\sigma }^2$, with $Var\left({\epsilon }_{i,t}\right)$ being constant. To express these dependencies, we assume Hansen’s threshold autoregression model [5] for selected variables with log-transformed data for series harmonization as:

$$ln{\left(gni\right)}_{i,t}=\{\begin{array}{l}{\alpha }_i+{\beta }_1^{\text{'}}{\mathit{\chi }}_{it}+{\epsilon }_{i,t}\left|threshold variable=ln\left(toe\right)\right|, for {\mathit{\chi }}_{it}=ln{(togdp)}_{i,t}\\ {\alpha }_i+{\beta }_2^{\text{'}}{\mathit{\chi }}_{it}+{\epsilon }_{i,t}\left|threshold variable=ln\left(toe\right)\right|, for {\mathit{\chi }}_{it}=ln{\left(vas\right)}_{i,t} (Model 1)\end{array}$$

(3)

$$ln{\left(vas\right)}_{i,t}=\{\begin{array}{l}{\alpha }_i+{\beta }_1^{\text{'}}{\mathit{\chi }}_{it}+{\epsilon }_{i,t}\left|threshold variable=ln\left(toe\right)\right|, for {\mathit{\chi }}_{it}=ln{\left(gni\right)}_{i,t}\\ {\alpha }_i+{\beta }_2^{\text{'}}{\mathit{\chi }}_{it}+{\epsilon }_{i,t}\left|threshold variable=ln\left(toe\right)\right|, for {\mathit{\chi }}_{it}=ln{\left(togdp\right)}_{i,t} (Model 2)\end{array}$$

(4)

with $ln\left(toe\right)$ as the threshold variable applied to divide the estimated sample into regimes: $\mathrm{for} ln{\left(gni\right)}_{i,t}, ln\left(toe\right)\le \gamma$ in the 1st regime; $\mathrm{for} ln{\left(gni\right)}_{i,t}, ln\left(toe\right)>\gamma$ in the 2nd regime; $ln{\left(vas\right)}_{i,t}, ln\left(toe\right)\le \gamma$ in the 1st regime; and $ln{\left(vas\right)}_{i,t}, ln\left(toe\right)>\gamma$ in the 2nd regime, where ${\mathit{\chi }}_{it}$ is the k × 1 vector of time-varying regressors, including the lagged dependent variable, for i = 1, …n and t = 1, …T.

By applying Kremer et al.’s dynamic function of panel threshold regression [7], the models for estimated variables are expressed as:

$$ln{\left(gni\right)}_{i,t}=\alpha +{\beta }_1^{\text{'}}ln{(togdp)}_{i,t}+{\beta }_2^{\text{'}}\ln {\left(vas\right)}_{i,t}+{\epsilon }_{i,t}\left|threshold variable=\ln \left(toe\right)\right| \left(Model 1\right)$$

(5)

$$ln{\left(vas\right)}_{i,t}=\alpha +{\beta }_1^{\text{'}}ln{\left(gni\right)}_{i,t}+{\beta }_2^{\text{'}}ln{\left(togdp\right)}_{i,t}+{\epsilon }_{i,t}\left|threshold variable=ln\left(toe\right)\right| \left(Model 2\right)$$

(6)

where indicators of function are $ln\left(toe\right)\le \gamma$ and $ln\left(toe\right)>\gamma$, $ln{\left(gni\right)}_{i,t}$ and $ln{\left(vas\right)}_{i,t}$ are scalar stochastic dependent variables of corresponding Models 1 and 2, $ln{(togdp)}_{i,t}$ and $ln{\left(vas\right)}_{i,t}$ are log-transformed scalars of threshold variables in Model 1, $ln{\left(gni\right)}_{i,t}$ and $ln{\left(togdp\right)}_{i,t}$ are log-transformed scalars of threshold variables in Model 2, the regressor ${\mathit{\chi }}_{it}$ is a vector of time-varying regressors, and ${\epsilon }_{i,t}$ is regression error for $i=1,\dots n;t=1,\dots ,T$. The regression error ${\epsilon }_{i,t}$ consists of two error components (according to Seo and Shin, 2016 [9]), an unobserved individual fixed effect ${\omega }_i$, and a zero-mean idiosyncratic random disturbance ${\phi }_i$, where ${\epsilon }_{i,t}={\omega }_i+{\phi }_i$, for the assumption that ${\phi }_i$ is a martingale difference sequence $E\left({\phi }_{i}I{F}_{t-1}\right)=0$, where $F_t$ is a natural filtration at (t) time. Following [8], the forward orthogonal deviations transformation of the error term and the corresponding variances are calculated as:

$${\epsilon }_{i,t}{}^{\diamond }=\sqrt{\frac{T-t}{T-t+1}} · \left[{\epsilon }_{i,t}-\frac{1}{T-t }\left({\epsilon }_{i,t+1}+\dots +{\epsilon }_{i,T}\right)\right], \mathit{for} {\sigma }^2{}_{{\epsilon }_{i,t}{}^{\diamond }}$$

(7)

where ${\epsilon }_{i,t}{}^{\diamond }$ is nonheteroscedastic and serially uncorrelated.

Models include two dynamic threshold regimes: ${\beta }_1^{\text{'}}$ and ${\beta }_2^{\text{'}}$. These regimes (coefficients ${\beta }_1^{\text{'}}$ and ${\beta }_2^{\text{'}}$) indicate the marginal effect threshold variables on the growth of gross national income (in Model 1) and the growth of value added by activity services as a percentage of value-added growth (in Model 2). For both models, the low regime is represented by ${\beta }_1^{\text{'}}$—below the projected threshold value—and the high regime is represented by ${\beta }_2^{\text{'}}$—above the estimated threshold value. We assumed that, according to [5], the regressor ${\mathit{\chi }}_{it}$, $ln{(togdp)}_{i,t}$, $ln{\left(vas\right)}_{i,t}$ (in Model 1), $ln{\left(gni\right)}_{i,t}$, and $ln{\left(togdp\right)}_{i,t}$ (in Model 2) are not time invariant, based on our identification of regimes ${\beta }_1^{\text{'}}$ and ${\beta }_2^{\text{'}}$. The statistical significance of producing regimes of $threshold variable=ln\left(toe\right)$ implies testing of the null hypothesis for ${\beta }_1^{\text{'}}$ = ${\beta }_2^{\text{'}}$. However, $ln\left(toe\right)$ is not identified under the null hypothesis, and distributional properties are nonstandard (by following [45]). In this case, refs. [5,45] suggest a rejection of the null hypothesis of nonthreshold effect when the p-value is lower compared with the critical value:

$$F_1=\frac{{\tau }_0-{\tau }_1\left({\rm Y}\right)}{{\sigma }^2}$$

(8)

where ${\tau }_0$ is the squared standard error of the nonthreshold effect, ${\tau }_1\left({\rm Y}\right)$ is threshold model squared standard error, and the variance of the error term of the threshold regression models is presented as ${\sigma }^2$. In our case, there two threshold models, so according to [5], the F-test is expressed as:

$$F_{tm}=\frac{{\tau }_1\left({\rm Y}\right)-{\tau }_1\left(\widetilde{{\rm Y}}\right)}{{\sigma }^2}$$

(9)

where $({\rm Y})$ is the threshold’s true value and $(\widetilde{{\rm Y}})$ is the threshold’s estimated value. The nonlinear transformation of Kremer et al.’s dynamic function of panel threshold two-regime regression models [7] is expressed as:

$$ln{\left(gni\right)}_{i,t}=\alpha +F\left(\widehat{z_t}, \widehat{\Theta }\right){\beta }_1^{\text{'}}ln{(togdp)}_{i,t}+F\left(\widehat{z_t}, \widehat{\Theta }\right){\beta }_2^{\text{'}}\ln {\left(vas\right)}_{i,t}+{\epsilon }_{i,t}\left|threshold variable=\ln \left(toe\right)\right| \left(Model 1\right)$$

(10)

$$ln{\left(vas\right)}_{i,t}=\alpha +F\left(\widehat{z_t}, \widehat{\Theta }\right){\beta }_1^{\text{'}}ln{\left(gni\right)}_{i,t}+F\left(\widehat{z_t}, \widehat{\Theta }\right){\beta }_2^{\text{'}}ln{\left(togdp\right)}_{i,t}+{\epsilon }_{i,t}\left|threshold variable=ln\left(toe\right)\right| \left(Model 2\right)$$

(11)

for variance ${\sigma }^2{}_{{\epsilon }_{i,t}{}^{\diamond }}$ and the transition of linear of elements $\widehat{z_t}$. According to [11], such transformation asylums an extensive choice of nonlinearities but does not drain all likely cases of nonlinearities.

4. Results

According to the study outcomes, the projected threshold value ln(toe) is 7.51 at a 95% confidence interval using the dynamic panel threshold in Model 1. ${\beta }_1^{\text{'}}$ denotes the marginal effect of ln(toe) on ln(gni) in the low-tourism-employment regime. ${\beta }_2^{\text{'}}$ indicates the marginal effect of ln(toe) on ln(gni) in the high-tourism-employment regime.

Table 3. Dynamic panel threshold estimates for Model 1.

Estimated Threshold Value	7.51
confidence interval (95%)	[3.5 (lower limit), 7.691 (upper limit)]
gamma	1.0675 (0.3316)
Effect of tourism employment
${\beta }_1^{\text{'}}$	0.1461 * (0.0718)
${\beta }_2^{\text{'}}$	0.6098 *** (0.1447)
Effect of Control Variables
initial	0.0125 *** (0.0015)
ln(togdp)	3.5942 *** (0.2292)
ln(vas)	−0.3513 (0.2809)
Number of Observations
Regime 1: Threshold variable less than 7.51	219
Regime 2: Threshold variable greater than 7.51	95
Number of Countries	30

Notes: The calculation is made by authors. Standard errors are shown in parentheses. Similarly, * and *** show significance at 10%, and 1% levels, respectively.

shows dynamic panel threshold estimates for Model 1.

Both ${\beta }_1^{\text{'}}$ (0.1461) and ${\beta }_2^{\text{'}}$ (0.6098) regimes’ dependent coefficients are statistically significant and have positive marginal effects in corresponding regimes. This means that a percent rise in tourism employment of the average country will lead to an increase in gross national income by 0.15% (in the low regime) and 0.61% (in the high regime). The outcomes are compatible with those of [15,48], which also revealed the strong impact of the tourism employment on an increase in gross national income. In addition, ref. [62] proves that there is an unambiguous connection between progress in tourism employment and the growth in gross national income of the region, which could potentially contribute to a possible enlargement in income inequality within the region. Estimates showed that the effect of development in tourism employment is larger under the high regime compared to the low regime. The threshold estimator diagram for Model 1 is shown in Figure 2.

Such results are in line with the findings of [63], which previously confirmed the positive correlation between economic trade and tourism by applying dynamic heterogeneous panel data analysis for the member countries of the OECD. This situation seems logical; the growth of employment in tourism is caused by higher tourist visits and demand increasing in related industries (hospitality, retailers, constructions, etc.), which directly increase gross national income (economic development). Additionally, the growing and developed economies (OECD members) are investing more in tourism infrastructure (including the facilities for luxury tourism), which leads to a higher gross economic contribution. Other studies confirm similar findings, such as [47], in which tourism specialization (as a dynamical process) affected economic growth, and [46], which underlined the fact that tourism development impacts the overall progress of the nations that strategically invest in tourism (such as OECD countries). Even more, ref. [64] stresses that tourism gradually even became one of the main elements of the service economy in most OECD countries as these societies become more developed, transferable, and service-oriented.

The results from Model 2 estimated the threshold value ln(toe) to be 6.56 at a 95% confidence interval (

Table 4. Dynamic panel threshold estimates for Model 2.

Estimated Threshold Value	6.56
confidence interval (95%)	[6.5 (lower limit), 6.784 (upper limit)]
gamma	−0.0762 (0.0734)
Effect of tourism employment
${\beta }_1^{\text{'}}$	−0.0738 ** (0.0244)
${\beta }_2^{\text{'}}$	−0.0911 *** (0.0265)
Effect of Control Variables
initial	−0.0010 (0.0014)
ln(gni)	0.01858 (0.0261)
ln(togdp)	0.2271 * (0.1138)
Number of Observations
Regime 1: Threshold variable less than 59.591	175
Regime 2: Threshold variable greater than 59.591	139
Number of Countries	30

Notes: The calculation is made by authors. Standard errors are shown in parentheses (). Similarly, *, **, and *** indicate significance at 10%, 5% and 1% levels, respectively.

). ${\beta }_1^{\text{'}}$ indicates the marginal effect of ln(toe) on ln(vas) in the low-tourism-employment regime. ${\beta }_2^{\text{'}}$ denotes the marginal effect of ln(toe) on ln(vas) in the high-tourism-employment regime. Both ${\beta }_1^{\text{'}}$ (−0.0738) and ${\beta }_2^{\text{'}}$ (−0.0911) regimes’ dependent coefficients were statistically significant and had negative marginal effects in their corresponding regimes. The outcomes explain that a percent rise in tourism employment in the average country will lead to a decrease in the value added by activity services, as a percentage of value added, by 0.07% (low regime) and 0.09% (high regime). This reinforces the findings of [60], which stress the importance of advancements in tourism employment, which will further guide a reduction in the value added by activity services. Consequently, if the employment rate increases in tourism-related services, this will form one of the fundamental economic facets that directly affects the value-added composition and employment reallocation along with the outcome of foreign exchange inflows of the region (also proved by [65]). Further analysis showed that the effect of development on tourism employment is larger under the high regime compared to the low regime. Both models confirm a direct relationship between tourism employment and tourism GDP. The threshold estimator diagram for Model 2 is shown in Figure 3.

The results from this research help to explain that tourism, as an economic branch, remains one of the lowest-paid sectors (in terms of salaries) in the international service sector. Similar results were obtained in [37], the authors of which claim that tourism employment is the destination for a large portion of “refugee” employees (more precisely, tourism is a second-choice profession) compared to professionals in tourism who choose this industry as their first choice. Second, the study results could also be explained by the previous findings of [38], where there was a high probability that many of the sample countries have a case where tourism employment affects the economy at the macro level but employees in tourism affect quality at the micro level. This inconsistency in policy and strategy can affect the obtained study results. The study outcomes, concerning the nexus of tourism employment and value added by activity services, support the theory of [49]. The author state that the tourism industry had restricted potential for economic productivity increases due to the absence of easily implementable technological development in service-related fields, such as the tourism industry. This results in high demand for tourism employment and a negative impact on relative price increases. Similar findings were confirmed by this study, where a direct relationship between tourism employment and tourism GDP has been revealed. This could help future related studies in understanding the relations, effects, and dynamics of tourism development.

5. Discussion

Despite the connection of study findings with those of numerous studies that confirm the TLGH and the impact of tourism employment on economic growth, there are certain specificities of this study reflected in both the methodology and results. Most studies up to 2010 use surveys [37] or less rigorous econometrics in their analyses (like in [34,49,54], etc.). This does not diminish the importance of their results, but on the contrary, the results of their studies have a significant contribution to the literature and policy makers. There was simply such a methodological trend in tourism research that did not require rigorous econometric and complex mathematical models. This study confirms their results but also reveals more precise values that must be reached to contribute to the growth with the application of a dynamic panel threshold model. In the last decade, some studies in tourism have used more rigorous mathematical and econometric models that are most often used in econometrics and financial economics due to the popularity of quantitative data and the availability of sophisticated software.

In relation to recent studies, which analyze tourism growth and/or the TLGH (like in [35,44,45,46], etc.), this study uses a dynamic model, which allows long-term coefficients for the explanatory inconstant along with the contemporary, short-term ones. This is a potential advantage compared with standard fixed-effects methodology. The study also differs from [15,31] in testing different factors by the application of the dynamic panel threshold model. Chiu et al. [15] analyze the impact of globalization on inbound tourism, while [31] tests the nexus between tourism growth and the informal economy. To the best of or knowledge, there are only a few studies that use dynamic panel threshold analysis and, therefore, this is the first study that empirically analyzes the impact of tourism employment on economic growth with the dynamic panel threshold model.

However, much more important is the fact that all three studies (including this one), posit that a positive impact is achieved only in the case of reaching certain threshold values (in our case, 7.51 (gni) for tourism employment). Our findings, as well as the findings of [15], confirm even a negative impact for specific variables (for a threshold value of 6.56 (vas), both marginal effects are negative in the impact on the value added by activity services; ref. [15] found an even stronger negative impact of competition in the tourism market on inbound tourism). With reference to the studies to use an OECD sample, our study analyzes tourism employment as a significant factor of growth, compared to other studied variables such as tourism revenue and tourism investments (for house prices’ impact in [66]), the impact of inbound tourism on international trade (in [63]), the nexus of tourism investment and economic policy uncertainty [36], the influence of international tourism arrivals on environmental sustainability effects [41], and the impact of tourism spatialization of economic growth in the long run [33].

Apart from the various tested variables in these studies (to the best of our knowledge, our study is the first one that analyzes the impact of tourism employment on economic growth in an OECD case), we highlight the two most important studies linked to our findings that either (a) use a dynamic panel model [63] or (b) analyze the impact on growth [33]. The authors of [63] apply an error correction model to analyze the nexus between international tourism and trade. This cointegration technique assumes that all variables are endogenous. Yet, due to the cointegration nexus, some variables could not adjust to equilibrium errors in the long run, which is weakly exogeneous. Introducing endogenous regressors with a dynamic panel threshold regression will overcome the possible endogeneity issue (as one of the possible solutions). The authors of [33] estimate the impact of tourism specialization on long-term growth. Their study also applies the cointegration technique and ideal demand system model to a relatively smaller OECD dataset but only for the long-term equilibrium among nonstationary variables. Both studies are of great value, with important findings, considering software availability and programs that were used in the earlier period.

6. Conclusions

The study reports the results of threshold regression, with both marginal effects indicating logical implications of tourism employment on gross national income and the value added by activity services. We find that greater investment in tourism employment contributes to economic growth (through higher gross national income and employment growth), while tourism remains one of the lowest-paid industries (in salaries) in the global services sector. In addition, the study contributes to the tourism-led growth hypothesis considering the positive impact of tourism employment on economic growth (gross national income); however, this is achieved only in the case when certain threshold effects are reached. In this case, the tourism-led growth hypothesis is confirmed for the high-threshold regime (0.61) of the impact of tourism employment on economic growth. On the other hand, in the case of the impact of tourism employment on the value added by activity services, certain threshold effects are not reached (threshold regimes are also slightly negative). The most logical conclusion for this is in the fact that the tourism sector is labor-intensive and cannot competitively contribute to value-added services like the technological sectors. In the case of the OECD nonmembers (such as many developing countries), this occurrence would be even stronger, bearing in mind the lower investments of such countries in tourism infrastructure and technologies.

This study has two important implications. The first implication is related to its contribution to the existing literature. While there are numerous studies that analyze the impact of tourism on economic growth, there are far fewer studies that specifically analyze the impact of tourism employment on economic growth. The novelty is greater considering that previous studies do not use dynamic models with threshold effects and particularly that most of them do not consider nonlinear influence (many economic phenomena are characterized by nonlinear behavior and/or have variables characterized by chaotic movements). The second concerns policy implication. During the COVID-19 pandemic, income from tourism and contribution to growth was the most affected by lockdowns and different measures against the pandemic. After the relaxation of these measures, there is a strong trend of growth for tourist services and employment. The study findings are of prospective essence to policy makers to make an effective strategy of employment in tourism that will take advantage of this growing trend. In addition, policy makers should also apply more technology in the tourism/travel industry to have a greater impact on the value-added sectors (such as virtual queues, modern infrastructure, the internet of things, augmented and/or virtual reality, etc.) In this respect, the current research contributes to the existing theory in economy, management, and tourism. More precisely, it provides fresh knowledge with the understanding of the relationship between tourism employment and gross national income on the one hand and the value added by activity services on the other hand. The findings demonstrated empirical evidence that tourism can strongly affect the global economy and international economic growth in the observed regions, which should be the focus of future related research. Lastly, the dynamic panel threshold model turned out to be a good econometric tool for testing the influence of tourism activities on economic development. Misclassifying a steady nonlinear procedure as linear can be equally deceptive in time-series and dynamic panel data analysis.

Nevertheless, the model has potential limitations. In the case of large panel datasets, more rigorous econometrics should concern dynamic heterogeneous panels. Second, there is an issue of complete endogeneity in the dynamic function of panel threshold regression due to the endogeneity bias for factors ln(gni) and ln(vas). Third, the model estimations are computationally intensive, requiring the use of a large amount of memory and making challenges for a large dataset. Future studies could also apply random effects and smooth transition threshold models as an alternative to the indicator functions. In the case of tourism employment, this paper uses aggregated data for the whole tourism sector. It could be indicated as a certain limitation due to the reason that tourism sectors have various types of subsectors with different contributions to growth among countries.