Governance Quality and Outbound Tourism Expenditure: Evidence from Symmetric and Asymmetric Panel ARDL Models

Abstract

This study examines the relationship between governance quality and outbound tourism expenditure. Annual panel data for 54 countries over the period 1996–2023 are analyzed using symmetric and asymmetric panel ARDL models. The results highlight the existence of a long-run relationship between governance quality and outbound tourism expenditure. The symmetric ARDL results indicate that governance quality exhibits a significant negative long-run effect on outbound tourism expenditure, suggesting that improvements in governance quality reduce residents’ spending on international travel. The effect remains significant only in low-governance countries when the sample is separated into low- and high-governance countries. The asymmetric ARDL results further reveal that positive governance shocks significantly reduce outbound tourism expenditure, while negative shocks increase it only in high-governance countries. Overall, the findings highlight the important role of institutional quality in shaping tourism consumption behavior and provide policy insights for managing tourism-related outbound expenditures.

1. Introduction

The outbound tourism expenditure has been considerably rising in the 21st century as the tourism sector grows. Global outbound tourism expenditure has exceeded US$1 trillion annually recently (UNWTO, 2023). This outbound tourism expenditure represents a major channel of cross-border consumption and financial outflow. Although the outbound tourism expenditure indicates the rising level of income and globalization, at the same time it also creates policy concerns about loss of domestic spending, pressure on the balance of payments, and challenges for sustainable tourism development, which can lead to economic imbalances and hinder local community benefits from tourism activities. These concerns highlight the importance of identifying not only economic but also institutional drivers of outbound tourism expenditure. Therefore, understanding the structural drivers of outbound tourism expenditure is crucial for tourism sectors and public policy.

The empirical tourism studies have predominantly investigated the relationship between governance quality and inbound tourism. They consistently find that better governance attracts more tourism. By contrast, the relationship between governance quality and outbound tourism expenditure remains a critical and under-researched gap in the literature. No prior study has simultaneously examined the long-run, heterogeneous, and asymmetric dimensions of this relationship using a multi-country panel framework. The previous researcher has mainly focused on the economic factors of outbound tourism expenditure, particularly income growth, exchange rates, and transportation costs (Akdağ et al., 2022; Gopalan & Khalid, 2023). These determinants of outbound tourism expenditure are undeniably important, but they offer an incomplete explanation of tourism behavior. Tourism decisions are not made independently of institutional conditions. Governance is a key institutional factor that shapes the outbound tourism expenditure behaviors through public trust, service quality, policy credibility, and domestic welfare. Thus, an empirical study is required to cover this gap and find the association between governance quality and outbound tourism expenditure.

Institutional economics suggests that governance quality shapes economic behavior in the long run by influencing transaction costs and uncertainties (W. Li & Abiad, 2009). In this context, governance affects household consumption decisions, including tourism choices, through several mechanisms. For example, there is high domestic consumption and investment when the rule of law is strong, corruption is controlled and bureaucracies are effective (Sarkar & Hasan, 2001; H. Q. Nguyen, 2023). By contrast, the weak quality of governance is associated with capital flow and external consumption. Households tend to seek a high quality of services and goods abroad (Hidri, 2025; Sun et al., 2025). These findings also suggest that governance quality may influence outbound tourism expenditure by shaping the behaviors of residents. This channel has received little direct attention in tourism research and needs to be explored.

Based on the above background, this study addresses four gaps. The first relates to the role of governance in shaping outbound tourism expenditure. Secondly, governance shapes outbound tourism expenditure either in the long run or in the short run. Thirdly, the influence of governance on tourism is likely to be heterogeneous across governance levels. Fourth, the effect of governance on outbound tourism expenditures may be asymmetric over time. The governance booms and busts may have different effects on the outbound tourism expenditure.

Therefore, this study investigates the association between governance quality and outbound tourism expenditure using a symmetric and asymmetric panel framework. The analysis employs the Pooled Mean Group Autoregressive Distributed Lag (PMG-ARDL) model for short-run dynamics and long-run relationships. Furthermore, Pooled Mean Group Nonlinear Autoregressive Distributed Lag (PMG-NARDL) approach is applied to capture asymmetric effects of positive and negative governance quality changes. The study uses panel data for 54 countries over the period 1996–2023. Moreover, Additional robustness checks and panel causality test are applied.

The study makes the following theoretical, methodological, and empirical contributions to the literature. Theoretically, it extends tourism demand research by grounding the governance-outbound tourism nexus in institutional economics. It shows that domestic trust, regulatory quality, and public services influence people’s travel decisions. Methodologically, it applies both PMG-ARDL and PMG-NARDL panel frameworks, capturing long-run equilibrium relationships, short-run dynamics, and the asymmetric impact of governance quality on outbound tourism expenditure. Empirically, it demonstrates the long-run relationship between governance quality and outbound tourism expenditure is considerably stronger than the short-run relationship. In addition, positive governance shocks have a stronger influence on reducing outbound tourism expenditure than negative shocks. Moreover, split the sample into low- and high-governance countries and highlight the important institutional heterogeneity in tourism behavior across different governance environments.

Moreover, the next section presents the literature review, followed by the data and methodology, results and discussion, and finally the conclusion.

2. Literature Review

There is enough literature on the economic determinants of outbound tourism. Income is one of the most dominant determinants of outbound tourism in prior literature. Higher income levels correlate with increased spending on international travel and tourism (Lim & McAleer, 2001; Gedecho et al., 2022). Beyond income, outbound tourism is also affected by macroeconomic variables, including exchange rates, financial development, and economic stability, which influence relative prices and household purchasing capacity (Chi, 2020; Gopalan & Khalid, 2023; Obi et al., 2025). These macroeconomic conditions directly affect the affordability of outbound tourism expenditure. However, while these studies consistently confirm the importance of economic variables, they largely treat tourism demand as a purely price- and income-driven phenomenon, thereby overlooking institutional and behavioral dimensions. Nevertheless, the current body of research offers limited empirical evidence directly investigating the relationship between governance quality and outbound tourism expenditure.

Institutional economics provides strong theoretical and empirical evidence that governance quality shapes economic behavior by influencing transaction costs and uncertainties (W. Li & Abiad, 2009). For instance, economies with a strong rule of law, less corruption and excellence bureaucracies lead to high domestic consumption and investment (Sarkar & Hasan, 2001; H. Q. Nguyen, 2023). By contrast, the weak quality of governance is associated with capital flow and external consumption. Households tend to seek a high quality of services and goods abroad (Hidri, 2025; Sun et al., 2025). While these findings suggest a substitution effect between domestic and foreign consumption, prior studies have not explicitly linked this mechanism to outbound tourism expenditure, leaving an important gap in tourism-specific applications. This behavior may extend to tourism, where residents allocate more spending to international travel. These findings suggest that governance quality may influence outbound tourism expenditure by shaping the behaviors of residents. This channel has received little direct attention in tourism research and needs to be explored.

In the tourism sector, the influence of governance is mostly incorporated in inbound tourism studies. Several studies report that improvements in governance are associated with higher levels of tourism. The rule of law, which is well-established, attracts tourists by ensuring safety and fairness (Brau et al., 2011). Moreover, corruption control and high political stability are positively associated with tourist arrivals and tourism revenues (Poprawe, 2015). These studies consistently highlight the positive role of governance in attracting inbound tourism; however, they implicitly assume that governance primarily influences foreign visitors, without considering how domestic residents respond through outbound tourism decisions. In contrast, the exiting literature is scarce for the association of governance with outbound tourism. The existing studies on governance and outbound tourism either consider governance implicitly or link the outcome of governance with outbound tourism. For instance, Gozgor and Demir (2018) investigate how economic policy uncertainty influences outbound tourism expenditure and show that higher uncertainty significantly reduces residents’ international tourism spending.

From a methodological perspective, the existing limited literature on governance and tourism mostly uses linear and short-run models. These models may yield biased estimates when long-run equilibrium relationships exist between tourism expenditure and its determinants (Yamaka & Ramos, 2026). Recent methodological advances emphasize the importance of modeling approaches in tourism analysis. The panel ARDL estimators have increasingly been adopted to capture both short-run and long-run relationships (M. H. Pesaran et al., 1999). Compared to conventional static or purely short-run models, ARDL-based approaches allow for heterogeneous dynamics across countries and provide more reliable estimates in the presence of mixed integration orders, making them particularly suitable for panel tourism data. Therefore, it is important to examine the relationship between governance and spending on outbound tourism, using a method that accounts for long-term relationships. In addition, a growing literature highlights that tourism demand responses are not always linear and symmetric. It can be nonlinear and asymmetric. The positive and negative change in the determinants of tourism may influence the tourism in different directions and magnitudes (Yamaka & Ramos, 2026). The ARDL /NARDL framework may be the most appropriate model for such a situation. It demonstrates that positive and negative changes in key drivers may exert unequal effects (Shin et al., 2014). This represents a methodological advancement over linear models, which assume symmetric responses. In tourism economics, asymmetric effects have been documented for income, exchange rates, and uncertainty shocks (Zhang et al., 2024). Although asymmetric governance effects remain largely unexplored. Institutional improvements may generate stronger behavioral responses than institutional deterioration. This justifies examining asymmetric governance effects on outbound tourism expenditure.

Human development and digital connectivity have emerged as important complementary factors in shaping tourism behavior. Higher human development improves health, education, and living standards, potentially reducing outbound tourism expenditure by improving domestic quality of life (Chattopadhyay et al., 2022; Rivera, 2017). Similarly, an increase in internet uses facilitates travel planning and information access (Jamal & Habib, 2020). Moreover, higher income levels are consistently linked to increased international travel and greater spending on tourism (Lim & McAleer, 2001; Gedecho et al., 2022).

The above literature shows that governance quality may affect outbound tourism expenditure through several ways. Improvements in governance enhance institutional quality, increase public confidence in domestic services and reduce uncertainty in economic activities. As a result, residents may allocate a larger share of their leisure and tourism expenditure within the domestic economy rather than abroad. In addition, outbound tourism expenditure can be viewed as a substitution decision between domestic and international consumption. Improvements in governance strengthen the domestic institutional environment and make local leisure opportunities more attractive relative to foreign travel. Consequently, residents may substitute international tourism consumption with domestic alternatives. Furthermore, changes in governance can lead to asymmetric behavioral responses. Improvements in institutions might have stronger and longer-lasting effects on tourism behavior. In contrast, a decline in institutional quality could result in weaker or delayed responses. This provides a basis for analyzing both symmetric and asymmetric relationships in the empirical model.

In summary, the literature reveals four clear gaps. Firstly, the governance quality is mostly studied in relation to inbound tourism as compared to outbound tourism expenditure. Secondly, most existing studies have not fully identified short-run and long-run equilibrium relationships and dynamic adjustment. Thirdly, the asymmetric effects of governance changes on outbound tourism have not been examined. Fourthly, the different effect of governance on outbound tourism expenditure across different governance levels is also undiscovered. This study addresses these gaps by applying ARDL and NARDL to a governance and outbound tourism dataset across different governance levels.

3. Empirical Method

3.1. Data and Variables

In this study, we use data from 54 countries as cross-sectional and a 27-year time span from 1996 to 2023 to construct a panel dataset. We selected these countries based on data availability. The panel dataset was further divided into two subsamples representing high-governance and low-governance countries based on the ICRG Quality of Governance index rankings (see Appendix A,

Table A1. Sample Countries Classified by Governance Level.

High-Governance Countries

Low-Governance Countries

1 Australia, 2 Austria, 3 Belgium, 4 Canada, 5 Finland, 6 Germany, 7 Ireland, 8 Cyprus, 9 Italy, 10 Japan, 11 New Zealand, 12 Norway, 13 Portugal, 14 Singapore, 15 Spain, 16 Sweden, 17 Switzerland

18 Albania, 19 Algeria, 20 Bahrain, 21 Bangladesh, 22 Brazil, 23 Bulgaria, 24 China, 25 Colombia, 26 Costa Rica, 27 Croatia, 28 Ecuador, 29 Guatemala, 30 Honduras, 31 Hungary, 32 India, 33 Indonesia, 34 Israel, 35 Jamaica, 36 Kuwait, 37 Kenya, 38 Kazakhstan, 39 Latvia, 40 Malaysia, 41 Mali, 42 Mexico, 43 Morocco, 44 Panama, 45 Paraguay, 46 Poland, 47 Romania, 48 South Africa, 49 Tunisia, 50 Togo, 51 Thailand, 52 Uganda, 53 Ukraine, 54 Zambia

for the country list). The preliminary tests are only conducted for the full sample, while the PMG-ARDL and PMG-NARDL estimations and post-estimation tests are performed separately for the full sample as well as for the high-governance and low-governance subsamples. In this study, the dependent variable is the outbound tourism expenditure per capita (OTEP). This is calculated by dividing the total spending on outbound tourism by the population size. Moreover, the main independent variable of this study is the ICRG Quality of Governance index (Teorell et al., 2024). The other covariates of this study include Gross Domestic Product (GDP), Human Development Index (HDI), and Internet Users (INT). The selection of variables is grounded in tourism and institutional economics literature. Governance quality serves as a proxy for institutional efficiency and public trust. It further represents service quality and shapes consumer behavior. GDP accounts for income levels and purchasing power. HDI indicates overall societal welfare and living conditions. Internet usage measures access to information and the efficiency of travel planning. Together, these variables form a comprehensive framework to explain outbound tourism expenditure. Further details about all these variables are shown in

Table 1. Variable definitions and measurement.

Variables	Symbol	Definition	Source	Unit
ICRG Quality of Governance index	G	Governance quality index (control of corruption, rule of law and bureaucratic effectiveness)	International Country Risk Guide (ICRG)	Index (0–1)
Outbound Tourism Expenditure per capita	lnOTEP	Total outbound travel expenditure of residents divided by total population	UNWTO	In millions of dollars
Gross Domestic Product	lnGDP	Gross domestic product	World Bank	Constant US dollars
Internet User	lnINT	Individuals using the internet	World Bank	% of population
Human Development Index	HDI	Human development index measuring achievements in health, education and standard of living	World Bank/UNDP	Index (0–1)

.

3.2. Preliminary Tests

3.2.1. Cross-Sectional Dependence Tests

Commonly, panel data analysis begins with checking for cross-sectional dependence (CD) across countries. The results could be biased and unrealistic without considering the cross-sectional dependence (M. H. Pesaran, 2007; Westerlund, 2006). This study incorporate three commonly used CD tests: the Pesaran scaled LM (H. Pesaran, 2004), the Pesaran CD (M. H. Pesaran, 2015), and the Breusch–Pagan LM (Breusch & Pagan, 1980). For all three CD tests, the null hypothesis is based on the assumption that cross-sectional countries are independent of each other, while the alternative hypothesis allows for the existence of CD across countries.

The results of the CD tests are presented in

Table 2. Cross-Sectional Dependence Tests.

Variable	Breusch–Pagan LM	Pesaran-Scaled LM	Pesaran CD
lnOTEP	11,815.85 (0.000)	89.66 (0.000)	83.68 (0.000)
G	7485.97 (0.000)	31.28 (0.000)	27.29 (0.000)
G+	20,171.51 (0.000)	148.30 (0.000)	134.32 (0.000)
G−	22,808.85 (0.000)	166.20 (0.000)	146.56 (0.000)
lnGDP	33,109.87 (0.000)	185.95 (0.000)	180.18 (0.000)
HDI	29,864.09 (0.000)	191.03 (0.000)	162.57 (0.000)
lnINT	27,712.93 (0.000)	183.03 (0.000)	163.02 (0.000)

Note: Values in parentheses are p-values. The null hypothesis is cross-sectional independence.

. The results revealed that there is CD across all variables. The results provide evidence against cross-sectional independence, as significance is confirmed at the 1% level for all cases. This outcome suggests that shocks affecting one country may transmit to others, reflecting interconnected economic structures. In tourism studies these types of results are very common because international tourism flows are influenced by global economic conditions and regional spillovers (Chen et al., 2025). Therefore, accounting for CD is essential to ensure the reliability and robustness of subsequent estimations. Given the existence of CD, the analysis proceeds with second-generation unit root and robust cointegration techniques.

3.2.2. Panel Unit Root Analysis

The next important step after analysis CD tests is to examine the presence of unit roots. Panel unit root tests are generally divided into two main categories: first-generation and second-generation tests. In this analysis, both types are applied based on the CD test results. The first-generation tests used in this study are the Breitung (2001), the Levin, Lin and Chu (LLC) (Levin et al., 2002) and the Im, Pesaran and Shin (IPS) (Im et al., 2003). Moreover, Cross-sectionally Augmented Dickey–Fuller (CADF) and Cross-sectionally Augmented IPS (CIPS) statistics are applied as second-generation tests due to the presence of CD. These second-generation tests are particularly suitable when cross-sectional dependencies are present among countries.

The results presented in

Table 3. Panel unit root test.

Variable	LLC	Breitung	IPS	CADF	CIPS
At Level
lnOTEP	−5.745 ***	−3.911 ***	−3.001 ***	−2.338	−2.924
G	−8.182 ***	−1.568 *	−7.281 ***	−2.911 ***	−2.645
G+	−1.897 **	−1.813 **	–	−2.006	−2.391
G−	−10.339 ***	−0.523	−7.756 ***	−3.243 ***	−2.899
lnGDP	0.864	−0.183	1.323	−1.925	−1.562
HDI	1.049	5.824	7.469	−2.226	−2.424
lnINT	−9.776 ***	2.604	−8.816 ***	−2.909 ***	−2.912
At First Difference
lnOTEP	−20.974 ***	−12.009 ***	−24.174 ***	−3.473 ***	−5.342
G	−16.068 ***	−5.816 ***	−17.335 ***	−3.964 ***	−4.231
G+	−9.042 ***	−6.013 ***	–	−3.064 ***	−4.151
G−	−17.411 ***	−5.880 ***	−18.110 ***	−4.035 ***	−4.247
lnGDP	−13.558 ***	−6.770 ***	−15.164 ***	−3.677 ***	−4.302
HDI	−11.897 ***	−7.277 ***	−14.432 ***	−3.129 ***	−4.586
lnINT	−10.070 ***	−4.289 ***	−12.380 ***	−3.854 ***	−4.443

Note: ***, **, * denote p-values at the 1%, 5%, and 10% levels. – indicates not estimated.

suggest the presence of mixed integration orders across the variables, as some variables are found to be stationary in their level form, I(0), whereas others become stationary only after applying first differencing, indicating integration at order one, I(1). The mixed order of integration among the variables is an important justification for using the Panel ARDL framework because ARDL can accommodate a combination of I(0) and I(1) variables without requiring all variables to be integrated at the same order. Furthermore, considering the potential asymmetric effect also leads us to apply the NARDL model (Shin et al., 2014).

3.2.3. Slope Heterogeneity Test

After confirming the existence of CD and stationarity properties, it is necessary to examine slope coefficients of cross-sectional units. Assuming identical slope coefficients for all countries may lead to misleading models and conclusions. Thus, this study uses the slope heterogeneity test developed by Hashem Pesaran and Yamagata (2008) along with two additional versions of it. This approach enables a comprehensive assessment of heterogeneity across countries. The first version, proposed by Blomquist and Westerlund (2013), incorporates heteroskedasticity and autocorrelation. The second version incorporates the CD by taking averages of countries. The null hypothesis posits that the slope coefficients are equivalent for all countries. A rejection of this null hypothesis signifies the existence of heterogeneous slopes among the countries. The results presented in

Table 4. Slope Heterogeneity Test Results.

Test Type	Delta Statistic	Adjusted Delta	p-Value	Decision
Standard Delta Test	15.999	18.930	0.000	Reject H0
Robust Test (HAC)	22.604	26.745	0.000	Reject H0
CSD-Controlled Test (CSA)	−30.696	−16.031	0.000	Reject H0

Note: The null hypothesis assumes slope homogeneity.

, uniformly reject the null hypothesis of slope homogeneity at the 1% significance level. The existence of slope heterogeneity suggests that estimators predicated on the assumption of identical short-run dynamics may produce biased estimates. Therefore, the PMG-ARDL model is the preferred methodology, owing to its ability to account for allowing for variation in short-run dynamics across cross-sectional units while maintaining common long-run coefficients.

3.2.4. Cointegration Tests

After identifying the order of integration and slope heterogeneity, panel cointegration tests are conducted to detect a cointegration between the variables. So, the study applies the Westerlund (2007) and Kao (1999) panel cointegration tests. The Westerlund test adopts an error-correction framework and enables robust inference using bootstrap procedures in the presence of CD. The Kao test, a method based on residuals similar to the Engle–Granger approach, assumes that the cointegrating relationships are homogeneous across all countries. The results in

Table 5. Kao and Westerlund Panel Cointegration Test Results.

Test	Statistic	Symmetric Model	Asymmetric Model
Kao (1999)	Modified DF t	−4.888 ***	−5.761 ***
	DF t	−5.089 ***	−5.792 ***
	ADF t	−5.915 ***	−6.923 ***
	Unadjusted Modified DF t	−8.222 ***	−9.294 ***
	Unadjusted DF t	−6.497 ***	−7.188 ***
Westerlund (2007)	Gt	−1.115	−1.761
	Ga	−1.284	−0.320
	Pt	−6.283 **	−6.510 *
	Pa	−1.247	−0.295

Note: The null hypothesis of both tests is no cointegration. The Westerlund test reports bootstrap p-values to account for cross-sectional dependence. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

, indicate that the Kao (1999) test decisively rejects the null hypothesis significantly at the 1% level. Furthermore, the Pt statistic from the Westerlund (2007) test demonstrates statistical significance in both symmetric and asymmetric models, thereby offering additional support for cointegration. As a result, the preliminary test outcomes indicate a long-run equilibrium relationship among outbound tourism expenditure, governance quality, gross domestic product, human development, and internet usage. Therefore, supports the use of the panel ARDL/NARDL framework in later analyses.

3.3. Model

The model of this study draws on the governance and tourism development literature (Chulaphan & Barahona, 2021; Gozgor & Demir, 2018; Khan et al., 2020; Özbozkurt et al., 2018; Das & Dirienzo, 2010). This study assumes that quality of governance, along with other covariates, has a considerable effect on outbound tourism expenditure per capita. The model is as:

$${\mathrm{O}\mathrm{T}\mathrm{E}\mathrm{P}}_{\mathrm{i}\mathrm{t}}={\mathsf{\alpha }}_{\mathrm{i}}+{\mathsf{\beta }}_1{\mathrm{G}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }}_2^{\prime }{\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(1)

where ${\mathrm{O}\mathrm{T}\mathrm{E}\mathrm{P}}_{\mathrm{i}\mathrm{t}}$ denotes outbound tourism expenditure per capita for country $\mathrm{i}$ at time $\mathrm{t}$. ${\mathrm{G}}_{\mathrm{i}\mathrm{t}}$ represents quality of governance, capturing bureaucracy effectiveness, control of corruption, and rule of law. ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}$ is a vector of control variables including GDP, human development index, and internet uses. The subscript $\mathrm{i}$ refers to the cross-sectional dimension (countries), $\mathrm{t}$ denotes the time dimension, ${\mathsf{\alpha }}_{\mathrm{i}}$ captures unobserved country-specific effects, and ${\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$ is the error term.

3.4. Methodology

This study applies both panel ARDL and NARDL models for exploring the association between governance quality and outbound tourism expenditure. The NARDL captures the asymmetric effect of positive and negative changes in governance quality on outbound tourism expenditure (Shin et al., 2014). Moreover, among the available ARDL estimators, this study adopts the Pooled Mean Group (PMG) estimator proposed by M. H. Pesaran et al. (1999) based on Hausman test. The PMG estimator is particularly appropriate when slope heterogeneity exists in the short run while the long-run relationship among variables is assumed to be homogeneous across cross-sectional units. This characteristic makes PMG especially useful for macro-panel analyses. These analyses are based on the idea that countries might react differently in the short term, but ultimately, they may share a common long-term equilibrium relationship. This specification is consistent with the findings of the slope heterogeneity tests reported earlier, which indicate that slope coefficients vary across countries. Compared to alternative estimators, the PMG approach provides a balanced framework by allowing short-run heterogeneity while maintaining long-run homogeneity across countries.

Alternative estimators such as Common Correlated Effects Mean Group (CCEMG) and Augmented Mean Group (AMG) are not used as the primary estimation techniques because these approaches mainly focus on estimating heterogeneous long-run coefficients and do not explicitly model the dynamic error-correction mechanism that is central to ARDL-type models. Likewise, compared to the Mean Group (MG) estimator, which may be inefficient due to full heterogeneity, and the Dynamic Fixed Effects (DFE) estimator, which imposes restrictive homogeneity assumptions, PMG offers a more appropriate compromise for this study. In a similar manner, estimators such as Fully Modified Ordinary Least Squares (FMOLS) and Dynamic Ordinary Least Squares (DOLS) mainly focus on estimating long-run cointegration relationships and therefore do not adequately capture the short-run adjustment process toward the long-run equilibrium. Thus, the PMG-ARDL framework is more suitable as it captures both short-run dynamics and long-run equilibrium relationships in a unified model. To further ensure the robustness of the empirical findings, additional robustness checks are performed using alternative estimators. The analysis uses Dynamic FE (cluster-robust) estimation, along with fixed-effects regressions that use Driscoll–Kraay standard errors. This estimation methods help to address potential issues related to CD, serial correlation, and heteroskedasticity in the dataset. Moreover, lag lengths for the ARDL/NARDL specifications were determined using the Akaike Information Criterion (AIC). AIC is widely recommended for dynamic panel settings because it balances model fit and parameter parsimony (H. Pesaran et al., 2001; M. H. Pesaran et al., 1999).

3.4.1. The Symmetric Panel ARDL

The ARDL equation is as follows:

$$\begin{array}{cc}\hfill {\Delta \mathrm{l}\mathrm{n}\mathrm{O}\mathrm{T}\mathrm{E}\mathrm{P}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_{0\mathrm{i}}& + {\mathsf{\beta }}_{1\mathrm{i}}{\mathrm{l}\mathrm{n}\mathrm{O}\mathrm{T}\mathrm{E}\mathrm{P}}_{\mathrm{i},\mathrm{t}-1}+{\mathsf{\beta }}_{2\mathrm{i}}{\mathrm{l}\mathrm{n}\mathrm{G}}_{\mathrm{i},\mathrm{t}-1}+{\displaystyle \sum _{\mathrm{k}}^{\mathrm{K}}}{\mathsf{\gamma }}_{\mathrm{k}}{\mathrm{l}\mathrm{n}\mathrm{X}}_{\mathrm{k},\mathrm{t}-1}\hfill \\ & + {\displaystyle \sum _{\mathrm{i}=1}^{{\mathrm{n}}_1}}{\mathsf{\theta }}_{1\mathrm{i}\mathrm{j}}\Delta {\mathrm{l}\mathrm{n}\mathrm{O}\mathrm{T}\mathrm{E}\mathrm{P}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}+{\displaystyle \sum _{\mathrm{i}=1}^{{\mathrm{n}}_2}}{\mathsf{\theta }}_{2\mathrm{i}\mathrm{j}}{\Delta \mathrm{l}\mathrm{n}\mathrm{G}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}+{\displaystyle \sum _{\mathrm{i}=1}^{{\mathrm{n}}_3}}{\mathsf{\delta }}_{\mathrm{k}\mathrm{i}\mathrm{j}}{\Delta \mathrm{l}\mathrm{n}\mathrm{X}}_{\mathrm{k},\mathrm{t}-\mathrm{j}}\hfill \\ & + {\mathsf{\mu }}_{\mathrm{i}\mathrm{t}}\hfill \end{array}$$

(2)

where $\Delta$ is the first differences, $\mathrm{i}=1,2,\dots \mathrm{N},\mathrm{t}=1,2,3,\dots \mathrm{T},$ ${\mathsf{\beta }}_{0\mathrm{i}}$ is the constant coefficient, ${\mathsf{\beta }}_{\mathrm{i},\mathrm{t}-1}$ are the long run coefficients, and ${\mathsf{\theta }}_{1\mathrm{i}\mathrm{j}}^{\mathrm{d}}$ are the short run coefficients for tourism arrivals and governance. Where ${\mathrm{G}}_{\mathrm{i}\mathrm{t}1}$ is the governance indexes, and ${\mathrm{X}}_{\mathrm{k},\mathrm{t}-1}$ is the vectors for explanatory variables.

To include the error-correction term the equation can be written as:

$$\begin{array}{cc}\hfill {\Delta \mathrm{l}\mathrm{n}\mathrm{O}\mathrm{T}\mathrm{E}\mathrm{P}}_{\mathrm{i}\mathrm{t}}=& {\mathsf{\delta }\mathrm{E}\mathrm{C}\mathrm{T}}_{\mathrm{i},\mathrm{t}-1}+{\displaystyle \sum _{\mathrm{i}=1}^{{\mathrm{n}}_1}}{\mathsf{\vartheta }}_{1\mathrm{i}\mathrm{j}}{∆\mathrm{l}\mathrm{n}\mathrm{O}\mathrm{T}\mathrm{E}\mathrm{P}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}+{\displaystyle \sum _{\mathrm{i}=1}^{{\mathrm{n}}_2}}{\mathsf{\vartheta }}_{2\mathrm{i}\mathrm{j}}{\Delta \mathrm{l}\mathrm{n}\mathrm{G}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}\hfill \\ & +{\displaystyle \sum _{\mathrm{i}=1}^{{\mathrm{n}}_3}}{\mathsf{\delta }}_{\mathrm{k}\mathrm{i}\mathrm{j}}{\Delta \mathrm{l}\mathrm{n}\mathrm{X}}_{\mathrm{k},\mathrm{t}-\mathrm{j}}+{\mathsf{\mu }}_{\mathrm{i}\mathrm{t}}\hfill \end{array}$$

(3)

where ${\mathsf{\delta }\mathrm{E}\mathrm{C}\mathrm{T}}_{\mathrm{i},\mathrm{t}-1}={\Delta \mathrm{l}\mathrm{n}\mathrm{G}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}-{\mathsf{\varphi }}_{0\mathrm{i}}-{\mathsf{\varphi }}_{1\mathrm{i}}{\mathrm{l}\mathrm{n}\mathrm{O}\mathrm{T}\mathrm{E}\mathrm{P}}_{\mathrm{i},\mathrm{t}-1}-{\mathsf{\varphi }}_{2\mathrm{i}}{\mathrm{l}\mathrm{n}\mathrm{X}}_{\mathrm{k},\mathrm{t}-1}$. $\mathsf{\delta }$ is the speed of adjustment, indicating how quickly the variable returns to equilibrium after a short-run shock. The parameters of the error correction term (ECT) are derived using the estimated long-run parameters obtained from the ARDL framework, as expressed below:

$${\mathsf{\varphi }}_{0\mathrm{i}}=-{\displaystyle \frac{{\mathsf{\beta }}_{0\mathrm{i}}}{{\mathsf{\beta }}_{1\mathrm{i}}}}, {\mathsf{\varphi }}_{1\mathrm{i}}=-{\displaystyle \frac{{\mathsf{\beta }}_{2\mathrm{i}}}{{\mathsf{\beta }}_{1\mathrm{i}}}}, {\mathsf{\varphi }}_{2\mathrm{i}}=-{\displaystyle \frac{{\mathsf{\beta }}_{3\mathrm{i}}}{{\mathsf{\beta }}_{2\mathrm{i}}}}$$

(4)

3.4.2. The Asymmetric Panel ARDL

In the symmetric panel ARDL framework, it is assumed that positive and negative changes in governance quality exert identical effects on outbound tourism expenditure per capita. In the asymmetric panel ARDL model we relax this assumption and allow positive and negative changes in governance quality have different effects on outbound tourism expenditure. In the asymmetric panel ARDL framework, governance quality is decomposed into positive and negative partial sums as follows:

$${\mathrm{G}}_{\mathrm{i},\mathrm{t}}^{+}={\displaystyle \sum _{\mathrm{q}=0}^{\mathrm{Q}}}\mathrm{m}\mathrm{a}\mathrm{x}\left(\Delta {\mathrm{G}}_{\mathrm{i},\mathrm{t}-\mathrm{q}},0\right)$$

(5)

$${\mathrm{G}}_{\mathrm{i},\mathrm{t}}^{-}={\displaystyle \sum _{\mathrm{q}=0}^{\mathrm{Q}}}\mathrm{m}\mathrm{i}\mathrm{n}\left(\Delta {\mathrm{G}}_{\mathrm{i},\mathrm{t}-\mathrm{q}},0\right)$$

(6)

where ${\mathrm{G}}_{\mathrm{i},\mathrm{t}}^{+}$ represents the cumulative positive changes (improvements) in governance quality for country $\mathrm{i}$ at time $\mathrm{t}$, and ${\mathrm{G}}_{\mathrm{i},\mathrm{t}}^{-}$ represents the cumulative negative changes (deteriorations) in governance quality.

Accordingly, the asymmetric panel ARDL specification is expressed as:

$$\begin{array}{cc}\hfill \Delta \ln \mathrm{O}\mathrm{T}{\mathrm{E}\mathrm{P}}_{\mathrm{i}\mathrm{t}} {=\mathsf{\beta }}_{0\mathrm{i}}& + {\mathsf{\beta }}_{1\mathrm{i}}\ln \mathrm{O}\mathrm{T}{\mathrm{E}\mathrm{P}}_{\mathrm{i},\mathrm{t}-1}+{\mathsf{\beta }}_{2\mathrm{i}}^{+}\ln {\mathrm{G}}_{\mathrm{i},\mathrm{t}-1}^{+}+{\mathsf{\beta }}_{2\mathrm{i}}^{-}\ln {\mathrm{G}}_{\mathrm{i},\mathrm{t}-1}^{-}+{\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{K}}}{\mathsf{\gamma }}_{\mathrm{k}}\ln X_{i,t-1}^k\hfill \\ & + {\displaystyle \sum _{\mathrm{j}=1}^{{\mathrm{n}}_1}}{\mathsf{\theta }}_{1\mathrm{i}\mathrm{j}}\Delta \ln \mathrm{O}\mathrm{T}{\mathrm{E}\mathrm{P}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}+{\displaystyle \sum _{\mathrm{j}=1}^{{\mathrm{n}}_2}}\left({\mathsf{\theta }}_{2\mathrm{i}\mathrm{j}}^{+}\Delta \ln {\mathrm{G}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}^{+}+{\mathsf{\theta }}_{2\mathrm{i}\mathrm{j}}^{-}\Delta \ln {\mathrm{G}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}^{-}\right)\hfill \\ & + {\displaystyle \sum _{\mathrm{j}=1}^{{\mathrm{n}}_3}}{\mathsf{\delta }}_{\mathrm{k}\mathrm{i}\mathrm{j}}\Delta \ln {\mathrm{X}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}^{\mathrm{k}}+{\mathsf{\mu }}_{\mathrm{i}\mathrm{t}}\hfill \end{array}$$

(7)

where all variables are defined as in the symmetric specification, and the superscripts $+$ and $-$ capture positive and negative governance shocks, respectively.

To incorporate the error-correction mechanism, Equation (7) can be re-written as:

$$\begin{array}{cc}\hfill \Delta \mathrm{l}\mathrm{n} \mathrm{O}\mathrm{T}{\mathrm{E}\mathrm{P}}_{\mathrm{i}\mathrm{t}}=& {\mathsf{\delta }}_{\mathrm{i}}\mathrm{E}\mathrm{C}{\mathrm{T}}_{\mathrm{i},\mathrm{t}-1}+{\displaystyle \sum _{\mathrm{j}=1}^{{\mathrm{n}}_1}}{\mathsf{\vartheta }}_{1\mathrm{i}\mathrm{j}}\Delta \mathrm{l}\mathrm{n} \mathrm{O}\mathrm{T}{\mathrm{E}\mathrm{P}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}\hfill \\ & + {\displaystyle \sum _{\mathrm{j}=1}^{{\mathrm{n}}_2}}\left({\mathsf{\vartheta }}_{2\mathrm{i}\mathrm{j}}^{+}\Delta \mathrm{l}\mathrm{n} {\mathrm{G}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}^{+}+{\mathsf{\vartheta }}_{2\mathrm{i}\mathrm{j}}^{-}\Delta \mathrm{l}\mathrm{n} {\mathrm{G}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}^{-}\right)+{\displaystyle \sum _{\mathrm{j}=1}^{{\mathrm{n}}_3}}{\mathsf{\delta }}_{\mathrm{k}\mathrm{i}\mathrm{j}}\Delta \ln {\mathrm{X}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}^{\mathrm{k}}+{\mathsf{\mu }}_{\mathrm{i}\mathrm{t}}\hfill \end{array}$$

(8)

where the asymmetric error-correction term is defined as:

$$\mathrm{E}\mathrm{C}{\mathrm{T}}_{\mathrm{i},\mathrm{t}-1}=\ln \mathrm{O}\mathrm{T}{\mathrm{E}\mathrm{P}}_{\mathrm{i},\mathrm{t}-1}-{\mathsf{\varphi }}_{0\mathrm{i}}-{\mathsf{\varphi }}_{1\mathrm{i}}^{+}\ln {\mathrm{G}}_{\mathrm{i},\mathrm{t}-1}^{+}-{\mathsf{\varphi }}_{1\mathrm{i}}^{-}\ln {\mathrm{G}}_{\mathrm{i},\mathrm{t}-1}^{-}-{\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{K}}}{\mathsf{\varphi }}_{2\mathrm{i}}^{\mathrm{k}}\ln {\mathrm{X}}_{\mathrm{i},\mathrm{t}-1}^{\mathrm{k}}$$

(9)

The parameter ${\mathsf{\delta }}_{\mathrm{i}}$ represents the speed of adjustment. The long-run coefficients in the asymmetric model are recovered from the estimated ARDL parameters as:

$${\mathsf{\varphi }}_{0\mathrm{i}}=-{\displaystyle \frac{{\mathsf{\beta }}_{0\mathrm{i}}}{{\mathsf{\beta }}_{1\mathrm{i}}}}, {\mathsf{\varphi }}_{1\mathrm{i}}^{+}=-{\displaystyle \frac{{\mathsf{\beta }}_{2\mathrm{i}}^{+}}{{\mathsf{\beta }}_{1\mathrm{i}}}}, {\mathsf{\varphi }}_{1\mathrm{i}}^{-}=-{\displaystyle \frac{{\mathsf{\beta }}_{2\mathrm{i}}^{-}}{{\mathsf{\beta }}_{1\mathrm{i}}}}, {\mathsf{\varphi }}_{2\mathrm{i}}^{\mathrm{k}}=-{\displaystyle \frac{{\mathsf{\gamma }}_{\mathrm{k}}}{{\mathsf{\beta }}_{1\mathrm{i}}}}$$

(10)

3.5. Post-Estimation Test

3.5.1. Hausman Test

In this study, the panel ARDL model is estimated using three different estimators: the Pooled Mean Group (PMG), the Dynamic Fixed Effects (DFE), and the Mean Group (MG). Consistent with the study’s econometric strategy, the PMG estimator is treated as the baseline specification because it allows heterogeneous short-run dynamics across countries while constraining long-run coefficients to be homogeneous. This method is particularly useful for macro-panel tourism datasets, where countries might show different short-term reactions but have similar long-term relationships, as suggested by M. H. Pesaran et al. (1999). In the MG estimation approach, coefficients are first obtained independently for each cross-sectional unit, after which these individual estimates are averaged to produce overall parameters, thereby fully accommodating heterogeneity across units (M. H. Pesaran & Smith, 1995). In contrast, the DFE estimator imposes homogeneity on short- and long-run coefficients, except for the intercepts. Hausman test is employed to determine the most appropriate estimator among all three.

3.5.2. Wald Test

To examine whether positive and negative shocks have different long-run effects, a Wald test for long-run asymmetry is applied following the approach of Shin et al. (2014). The PMG–NARDL framework allows the breakdown of governance quality into positive and negative partial sums, which makes it possible to test whether the long-run effects of these shocks differ. Let ${\mathsf{\theta }}^{+}$ and ${\mathsf{\theta }}^{-}$ represent the long-term coefficients associated with positive and negative shocks, respectively. The Wald test is performed to evaluate the following hypotheses:

$${\mathrm{H}}_0:{\mathsf{\theta }}^{+}={\mathsf{\theta }}^{-}$$

$${\mathrm{H}}_1:{\mathsf{\theta }}^{+}\ne {\mathsf{\theta }}^{-}$$

The null hypothesis posits that positive and negative shocks produce identical long-run effects, thereby suggesting a symmetric relationship. Conversely, a rejection of the null hypothesis implies that positive and negative shocks have different long-run effects, which indicates the existence of long-run asymmetry. The Wald test is applied to the estimated long-run coefficients obtained from the PMG–NARDL model.

3.6. Panel Causality Analysis

Lastly, the Dumitrescu–Hurlin panel Granger causality test is applied to explore dynamic lead–lag relationships between governance quality and outbound tourism expenditure per capita. This test allows for heterogeneity across countries and identifies predictive causality rather than structural causality. The null hypothesis of no causality is rejected if the standardized Z-bar statistic is significant.

$${\mathrm{Y}}_{\mathrm{i}\mathrm{t}}={\mathsf{\alpha }}_{\mathrm{i}}+{\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{K}}}{\mathsf{\gamma }}_{\mathrm{i}\mathrm{k}}{\mathrm{Y}}_{\mathrm{i},\mathrm{t}-\mathrm{k}}+{\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{K}}}{\mathsf{\beta }}_{\mathrm{i}\mathrm{k}}{\mathrm{X}}_{\mathrm{i},\mathrm{t}-\mathrm{k}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(11)

3.7. Robustness Checks: Alternative Estimators

While the PMG-ARDL estimator serves as the principal estimation method, supplementary robustness checks are performed to ascertain the stability of the findings across different estimation techniques. Specifically, the analysis incorporates dynamic fixed-effects models and fixed-effects estimations employing Driscoll–Kraay standard errors. These alternative estimators are employed to mitigate potential issues concerning CD and serial correlation. It is crucial to note that these estimators are utilized exclusively as robustness checks, intended to confirm the consistency of the core PMG-ARDL/NARDL results, rather than to supplant the primary estimation approach.

In summary, the empirical strategy of this study proceeds in three sequential stages. In the first stage, preliminary tests are conducted to establish the presence of cross-sectional dependence, unit root properties, slope heterogeneity across countries, and panel cointegration among the variables. In the second stage, both PMG-ARDL and PMG-NARDL models are estimated, with model selection guided by Hausman tests comparing PMG, MG, and DFE estimators, separately for the full sample and for high- and low-governance country subgroups. In the third stage, post-estimation tests are performed, including Wald tests, Dumitrescu-Hurlin panel Granger causality analysis, and robustness checks using alternative estimators.

4. Results and Discussions

4.1. Descriptive Statistics

It is necessary to know the basic characteristics of all variables. So, for this purpose, we run descriptive statistics, which are shown in

Table 6. Descriptive statistics of variables.

Variable	Observation	Mean	Std. Dev.	Min	Max
lnOTEP	1512	4.985	1.911	−0.465	8.584
G	1512	0.611	0.210	0.250	1.000
lnGDP	1512	25.693	1.822	21.451	31.097
HDI	1512	0.758	0.142	0.275	0.967
lnINT	1512	3.153	1.395	0.001	4.615

Note: ln denotes natural logarithm.

. Descriptive statistics are computed for 1512 country–year observations to summarize the key characteristics of the variables. These statistics provide insights into the central tendency, dispersion, and overall distribution of the data used in this study. Outbound tourism expenditure per capita (lnOTEP) displays substantial cross-country variation, indicating heterogeneous outbound tourism behavior across economies. The governance quality index (G) shows meaningful dispersion, reflecting differences in institutional environments, which may influence the level of outbound tourism expenditure across various countries. Similarly, economic development (lnGDP), human development (HDI), and internet usage (lnINT) exhibit wide ranges, confirming that the sample includes both developing and advanced economies.

4.2. Symmetric Panel ARDL Results

The results of symmetric panel ARDL for outbound tourism expenditure per capita are displayed in

Table 7. PMG-ARDL Results.

Variable	Full Sample	High Governance	Low Governance
Long-run association (LR)
G	−1.144 ** (0.569)	−1.358 (1.044)	−2.369 *** (0.553)
lnGDP	0.712 *** (0.089)	1.907 *** (0.328)	0.537 *** (0.082)
HDI	−9.623 *** (1.794)	−30.802 *** (5.752)	−3.454 ** (1.480)
lnINT	0.274 *** (0.056)	0.257 * (0.132)	0.112 ** (0.047)
Short-run association (SR)
G	−0.603 (0.469)	−0.980 (0.892)	−0.366 (0.593)
lnGDP	0.153 ** (0.067)	0.154 (0.143)	0.110 (0.081)
HDI	27.337 *** (3.962)	27.221 *** (10.469)	28.311 *** (2.892)
lnINT	−0.087 (0.064)	−0.171 (0.117)	−0.077 (0.086)
Error-correction term (ECT)	−0.250 *** (0.021)	−0.244 *** (0.039)	−0.280 *** (0.032)

Note: Standard Errors in parentheses. ***, **, * denote p-values at the 1%, 5%, and 10% levels. Dependent Variable: Outbound Tourism Expenditure per capita.

. In the first stage, we estimate symmetric panel ARDL specifications using PMG and DFE estimators and choose between them with a Hausman test, which indicates that PMG is preferred (M. H. Pesaran & Smith, 1995; M. H. Pesaran et al., 1999; Blackburne & Frank, 2007; Hausman, 1978). In the second stage, we estimate the mean group (MG) model and compare MG with PMG using another Hausman test, which again favors PMG. Finally, we split the sample into High Governance and Low Governance countries using the ICRG Quality of Governance index countries’ ranking and repeat the same sequence—PMG versus DFE, then MG versus PMG—with Hausman tests guiding model selection. The result of Hausman test for symmetric ARDL is presented in

Table 8. Hausman test results for panel ARDL models.

Model	Comparison	χ2 (df)	p-Value	Decision
Full Sample	PMG vs. DFE	0.240 (4)	0.993	PMG preferred
Full Sample	MG vs. PMG	1.510 (4)	0.826	PMG preferred
High Governance	PMG vs. DFE	0.120 (4)	0.998	PMG preferred
High Governance	MG vs. PMG	4.590 (4)	0.333	PMG preferred
Low Governance	PMG vs. DFE	0.670 (4)	0.955	PMG preferred
Low Governance	MG vs. PMG	2.990 (4)	0.559	PMG preferred

Note: Hausman test determines the preferred estimator between MG, PMG, and DFE.

. This stepwise estimation strategy helps ensure the selection of the most appropriate estimator across different model specifications.

The panel ARDL results show that governance quality has statistically significant long-run effects on outbound tourism expenditure per capita. The magnitude of the Error Correction Term (ECT) (approximately −0.26 to −0.29) implies that around one-quarter of short-run disequilibrium adjusts toward the long-run equilibrium each year, indicating a moderate speed of adjustment. This may supports the use of a long-run equilibrium framework in modeling outbound tourism behavior (M. H. Pesaran et al., 1999). Such a speed of adjustment reflects a gradual convergence process toward equilibrium across countries.

In the full sample, the quality of governance exhibits a significant negative effect on outbound tourism expenditure per capita in the long run. In high-governance countries, governance quality does not exert a statistically significant influence on outbound tourism expenditure per capita. This result aligned with Putterman (2013) and may indicates institutional saturation, where further improvements in governance might no longer translate into noticeable changes in residents’ international travel spending behavior. In low-governance countries, governance quality has a stronger, negative, and statistically significant effect on outbound tourism expenditure per capita. This finding consistent with Kim et al. (2018) and C. Nguyen (2021) and may suggests that improvements in institutional quality reduce residents’ propensity to spend on international travel. Stronger governance may improve domestic trust, regulatory quality, and public service delivery. These institutional improvements might reduce the need for residents to seek consumption and leisure opportunities abroad (North, 1991). Strong domestic institutions also support the development of domestic tourism markets and may encourage residents to allocate a larger share of their tourism expenditure domestically rather than abroad (Sun et al., 2025; M. Li et al., 2025; Rigelský et al., 2021). These findings underline how governance quality affects outbound tourism expenditure across low- and high-governance countries.

The GDP displays a positive and significant influence on outbound tourism expenditure in all samples. The results indicate that outbound tourism expenditure is strongly income-driven. This may be aligned with Wang (2014), where higher income relaxes the budget constraints and increases the affordability of international travel. The internet users use it as a proxy for digitalization, which also exerts a positive long-run effect in all samples. This reflects the role of digital connectivity in facilitating information access, booking efficiency, and international mobility (Ruan et al., 2025). This indicates that technological development may further enhance international travel accessibility. By contrast, HDI shows a negative long-run association with outbound tourism expenditure, particularly in high-governance economies. This finding is constant with Biagi et al. (2016), Akbar et al. (2025), and may suggests that improvements in domestic living standards, public services, and social welfare reduce incentives for residents to seek welfare abroad, aligning with substitution effects emphasized in quality-of-life and consumption theory.

4.3. Asymmetric Panel ARDL (NARDL) Results

The panel NARDL model results are reported in

Table 9. PMG-NARDL Results.

Variable	Full Sample	High Governance	Low Governance
Long-run association (LR)
G+	−2.397 *** (0.560)	−5.697 * (3.102)	−2.876 *** (0.554)
G−	1.572 ** (0.611)	4.931 ** (2.326)	0.609 (0.657)
lnGDP	0.604 *** (0.075)	1.647 *** (0.299)	0.542 *** (0.078)
HDI	−4.980 *** (1.494)	−26.603 *** (5.322)	−2.066 (1.401)
lnINT	0.292 *** (0.048)	0.676 *** (0.177)	0.176 *** (0.046)
Short-run association (SR)
G+	0.238 (0.666)	1.194 (1.107)	−0.699 (0.834)
G−	−1.191 * (0.660)	−2.443 * (1.351)	−0.624 (0.833)
lnGDP	0.153 ** (0.064)	0.125 (0.141)	0.091 (0.073)
HDI	26.423 *** (3.933)	28.547 *** (9.618)	27.623 *** (2.696)
lnINT	−0.125** (0.063)	−0.211 * (0.124)	−0.128 (0.083)
Error-correction term (ECT)	−0.277 *** (0.025)	−0.254 *** (0.043)	−0.307 *** (0.036)

Note: Standard Errors in parentheses. ***, **, * denote p-values at the 1%, 5%, and 10% levels. Dependent Variable: Outbound Tourism Expenditure per capita.

, where outbound tourism expenditure per capita (lnOTEP) is used as the dependent variable. The result of Hausman test for NARDL is presented in

Table 10. Hausman test results for panel NARDL models.

Model	Comparison	χ2 (df)	p-Value	Decision
Full Sample	PMG vs. DFE	0.250 (5)	0.998	PMG preferred
Full Sample	MG vs. PMG	0.780 (5)	0.979	PMG preferred
High Governance	PMG vs. DFE	0.200 (5)	0.999	PMG preferred
High Governance	MG vs. PMG	2.230 (5)	0.817	PMG preferred
Low Governance	PMG vs. DFE	0.390 (5)	0.996	PMG preferred
Low Governance	MG vs. PMG	3.770 (5)	0.583	PMG preferred

Note: Hausman test determines the preferred estimator between MG, PMG, and DFE.

. The key explanatory variable, quality of governance, is decomposed into positive changes (G⁺) and negative changes (G⁻) to capture potential asymmetric effects (Shin et al., 2014). The results provide clear evidence of long-run asymmetry in the relationship between governance quality and outbound tourism expenditure per capita. This decomposition allows for a more detailed examination of how improvements and deteriorations in governance affect tourism behavior differently.

Positive governance shocks (G⁺) have a statistically significant negative long-run effect on outbound tourism expenditure per capita in the full sample, high-governance countries, and low-governance countries. This indicates that improvements in institutional quality consistently reduce residents’ spending on international travel across all groups. In contrast, negative governance shocks (G⁻) increase outbound tourism expenditure per capita in the full sample and high-governance economies, but they do not exert a statistically significant effect in low-governance countries. This difference highlights the asymmetric sensitivity of outbound tourism expenditure to changes in governance quality. Moreover, this asymmetric pattern suggests that governance improvements have a stronger and more systematic influence on outbound tourism behavior than governance deterioration. Stronger institutions improve confidence in domestic economic conditions, public services, and regulatory environments, which reduces the incentive to seek consumption and leisure opportunities abroad. These results are in line with findings of Rigelský et al. (2021), Sun et al. (2025), and M. Li et al. (2025). Such patterns indicate that positive institutional changes may have more persistent effects than negative shocks. By contrast, short-term declines in governance quality do not immediately translate into higher outbound tourism expenditure because of adjustment costs, habit persistence, and income constraints faced by households (Wang, 2014). These findings are consistent with nonlinear adjustment theories and asymmetric behavioral responses under institutional uncertainty (Shin et al., 2014).

4.4. Wald Test Results

The Wald test results are summarized in

Table 11. Wald Test Results.

Model	Test	χ2 Statistic	p-Value	Inference
Full Sample	Long-run asymmetry	35.49	0.000	Long-run asymmetry
Full Sample	Short-run asymmetry	2.25	0.134	Symmetry not rejected
High Governance	Long-run asymmetry	3.38	0.066	Long-run asymmetry
High Governance	Short-run asymmetry	0.02	0.879	Symmetry not rejected
Low Governance	Long-run asymmetry	0.35	0.553	Symmetry not rejected
Low Governance	Short-run asymmetry	0.07	0.799	Symmetry not rejected

Note: Wald test examines long-run and short-run asymmetry.

. The results indicate the presence of long-run asymmetry in the full sample, as the null hypothesis of symmetric effects is rejected (χ² = 35.49, p = 0.000). This suggests that positive and negative governance shocks have different long-run impacts on outbound tourism expenditure. However, the short-run Wald test is statistically insignificant, indicating no evidence of short-run asymmetry. This may reflect the limited immediate response of tourism expenditure to short-term institutional changes. For the high-governance group, evidence of long-run asymmetry is observed at the 0.06% significance level, while short-run symmetry cannot be rejected. In contrast, for the low-governance group, both long-run and short-run Wald tests are insignificant, suggesting symmetric effects of governance shocks on outbound tourism expenditure. This difference across groups highlights the role of institutional context in shaping asymmetric responses. Overall, the results imply that asymmetric effects mainly emerge in the long run.

4.5. Dynamic Lead–Lag Relationships: Panel Granger Causality Results

Table 12. Panel Granger (Dumitrescu–Hurlin).

Lag	Direction	Z-Bar	p-Value	Causality?
1	ICRG → OTEP	1.375	0.169	No
1	OTEP → ICRG	3.791	0.000	Yes
2	ICRG → OTEP	1.273	0.203	No
2	OTEP → ICRG	1.271	0.204	No
3	ICRG → OTEP	4.577	0.000	Yes
3	OTEP → ICRG	1.270	0.204	No
4	ICRG → OTEP	3.849	0.000	Yes
4	OTEP → ICRG	2.670	0.008	Yes

reports Dumitrescu–Hurlin tests. The results reveal limited short-run causality but significant bidirectional predictive relationships at longer lags. Governance Granger causes outbound tourism expenditure per capita at higher lag orders, suggesting that institutional changes precede long-term adjustments in outbound tourism behavior. Conversely, outbound tourism expenditure also Granger-causes governance at certain lags, indicating feedback effects through economic openness, exposure to international norms, and institutional learning. These findings highlight the dynamic and endogenous interaction between governance and outbound tourism, justifying the use of dynamic panel frameworks rather than static models (Dumitrescu & Hurlin, 2012; Hurlin & Venet, 2005).

4.6. Robustness Checks: Alternative Estimators Results

Table 13. Symmetric Results of Dynamic Fixed Effects (Cluster-Robust) and Fixed Effects with Driscoll–Kraay Standard Errors.

Variable	Full Sample		High Governance		Low Governance
	Dynamic FE (Cluster-Robust)	FE with Driscoll–Kraay SE	Dynamic FE (Cluster-Robust)	FE with Driscoll–Kraay SE	Dynamic FE (Cluster-Robust)	FE with Driscoll–Kraay SE
L.OTEP	0.782 *** (0.028)	0.782 *** (0.053)	0.700 *** (0.073)	0.700 *** (0.096)	0.805 *** (0.028)	0.805 *** (0.046)
G	−0.585 *** (0.213)	−0.585 * (0.331)	−0.315 (0.242)	−0.315 ** (0.140)	−0.689 ** (0.261)	−0.689 (0.412)
lnGDP	0.196 *** (0.027)	0.196 ** (0.078)	0.295 *** (0.080)	0.295 *** (0.105)	0.204 *** (0.030)	0.204 ** (0.090)
HDI	−1.708 *** (0.638)	−1.708 (1.353)	−5.585 *** (1.654)	−5.585 (3.345)	−1.249 (0.758)	−1.249 (1.150)
lnINT	−0.024 (0.020)	−0.024 (0.030)	0.070 * (0.038)	0.070 (0.054)	−0.051 ** (0.023)	-0.051 (0.049)
Constant	−2.196 *** (0.559)	−2.196 ** (0.905)	−0.857 (1.237)	−0.857 (1.939)	−3.020 *** (0.622)	−3.020 ** (1.345)

Note: Standard Errors in (parentheses. ***, **, * denote p-values at the 1%, 5%, and 10% levels.

and

Table 14. Asymmetric Results of Dynamic Fixed Effects (Cluster-Robust) and Fixed Effects with Driscoll–Kraay Standard Errors.

Variable	Full Sample		High Governance		Low Governance
	Dynamic FE (Cluster-Robust)	FE with Driscoll–Kraay SE	Dynamic FE (Cluster-Robust)	FE with Driscoll–Kraay SE	Dynamic FE (Cluster-Robust)	FE with Driscoll–Kraay SE
L.OTEP	0.768 *** (0.031)	0.768 *** (0.056)	0.688 *** (0.063)	0.688 *** (0.098)	0.792 *** (0.034)	0.792 *** (0.051)
G+	−1.111 *** (0.261)	−1.111 * (0.619)	−1.564 ** (0.629)	−1.564 ** (0.602)	−1.048 *** (0.308)	−1.048 (0.683)
G−	−0.134 (0.261)	−0.134 (0.114)	0.446 (0.349)	0.446 (0.414)	−0.342 (0.332)	−0.342 * (0.196)
lnGDP	0.205 *** (0.028)	0.205 ** (0.080)	0.319 *** (0.091)	0.319 *** (0.109)	0.208 *** (0.031)	0.208 ** (0.090)
HDI	−1.543 ** (0.655)	−1.543 (1.268)	−5.244 *** (1.604)	−5.244 (3.182)	−1.109 (0.788)	−1.109 (1.119)
lnINT	0.008 (0.026)	0.008 (0.013)	0.124 *** (0.039)	0.124 * (0.072)	−0.026 (0.029)	−0.026 (0.030)
Constant	−2.881 *** (0.583)	−2.881 ** (1.108)	−2.113 (1.523)	−2.113 (1.857)	−3.519 *** (0.669)	−3.519 ** (1.567)

Note: Standard Errors in parentheses. ***, **, * denote p-values at the 1%, 5%, and 10% levels.

presents robustness checks using alternative estimation strategies. Across all models, governance quality retains a statistically significant negative long-run effect on outbound tourism expenditure per capita. The persistence of results after controlling for cross-sectional dependence and serial correlation confirms that the main findings are not driven by model specification. The results show that the lagged dependent variable has a positive and statistically significant coefficient, indicating strong inertia in outbound tourism behavior. GDP continues to exert a positive effect, while HDI maintains a negative association in most specifications. These results validate the reliability and stability of the PMG-based findings.

4.7. Discussion

The results confirm that governance quality has a long-run association with outbound tourism expenditure. This relationship varies across country groups. In high-governance countries the effect is not statistically significant. This finding may reflect diminishing marginal effects of governance improvements as well as limited within-country variation in governance quality in already high-performing institutional environments, which reduces statistical power in panel estimation. When governance has already reached a high level, further improvements do not produce noticeable changes in residents’ international travel spending (Putterman, 2013). In low-governance countries the results are different. Governance quality has a stronger and statistically significant negative effect on outbound tourism expenditure per capita. This means that institutional improvements reduce residents’ tendency to spend on international travel (Kim et al., 2018). This result suggests that the domestic substitution channel dominates alternative mechanisms such as income growth and increased openness, which are typically associated with higher outbound tourism demand (Rosselló & Santana Gallego, 2022). Better institutions raise public confidence in domestic services and economic stability, which can reduce the motivation to substitute domestic consumption with foreign travel. Strong institutions also support the growth of domestic tourism markets. As a result, residents allocate more of their tourism spending within their home country rather than abroad (Sun et al., 2025).

The asymmetric results may be interpreted using insights from general consumption theory, particularly Reference-dependent preferences (Koszegi & Rabin, 2009), which suggests that individuals evaluate outcomes relative to a reference point rather than in absolute terms. When governance improves, residents update their domestic reference point upward. Public services, institutional reliability, and domestic safety are perceived as increasingly satisfying relative to this new benchmark. The relative attractiveness of international travel therefore declines, and residents redirect leisure and consumption spending toward domestic markets. This explains the significant negative long-run effect of positive governance shocks on outbound tourism expenditure. When governance deteriorates, the behavioral response is different. A decline in institutional quality lowers residents’ confidence in domestic public services and regulatory stability. This widens the perceived welfare gap between domestic and international consumption options, raising the incentive to seek goods and leisure experiences abroad. This is consistent with the positive effect of negative governance shocks, particularly in high-governance economies.

The effects of governance deterioration differ between high- and low-governance contexts, whereas improvements exhibit relatively uniform effects across groups. In high-governance countries, even modest institutional deterioration is perceived as a meaningful and credible departure from an established norm. This triggers a precautionary reallocation of spending toward international markets, consistent with evidence that policy reversals impose larger behavioral costs in high-credibility environments (Masson & Drazen, 1994). The absence of a significant negative shock effect in low-governance countries may reflect a different behavioral mechanism. Acemoglu (2005) argue that residents of institutionally weak states have already internalized high levels of uncertainty into their expectations. Further deterioration that falls within the range of already anticipated instability does not alter travel behavior, because residents have no stable domestic reference point to revise downward. Taken together, these results suggest that the asymmetry is not simply a matter of magnitude but reflects fundamentally different behavioral mechanisms. Governance improvements generate a systematic and sustained reduction in outbound spending through upward reference-point revision. Governance deterioration produces context-dependent increases in outbound spending, amplified in high-credibility environments and absent in low-governance ones. This pattern is consistent with nonlinear adjustment theory and asymmetric behavioral responses under institutional uncertainty (Shin et al., 2014).

The control variables also produce results that are consistent with prior literature. GDP per capita has a positive and significant association with outbound tourism expenditure. Higher income relaxes household budget constraints and makes international travel more affordable (Wang, 2014). Internet penetration is positively associated with outbound tourism in the long run. Greater digital access lowers information costs and improves booking efficiency. This makes international travel more accessible to residents (Ruan et al., 2025). The Human Development Index shows a negative long-run association with outbound tourism expenditure. This effect is particularly evident in high-governance economies. When domestic living standards improve, residents have less incentive to seek welfare and leisure experiences abroad. This finding is consistent with Biagi et al. (2016) and Akbar et al. (2025). It supports the substitution hypothesis in quality-of-life and consumption theory. Improvements in domestic human development redirect leisure spending from international to domestic markets.

5. Conclusions

The study examines the long-run and short-run association between governance quality and outbound tourism expenditure by applying PMG-ARDL and PMG-NARDL. The results of PMG-ARDL and PMG-NARDL, along with robustness checks and panel causality analysis, show a long-run association between governance quality and outbound tourism expenditure. The finding indicates that improvement in governance quality reduces the outbound tourism expenditure. The ARDL shows a significant negative association between governance quality and outbound tourism expenditure in the low governance sample, while for the high governance sample there is no significant effect. The NARDL shows that positive shocks reduce outbound tourism expenditure in both low-governance and high-governance countries, while the negative shocks of governance only significantly enhance the outbound tourism expenditure in high-governance economies. The covariates used in this study are gross domestic product, human development and internet users. These are also important and relevant to policy implications. The gross domestic product is positively associated with outbound tourism expenditure. This suggests that high income is associated with more outbound tourism expenditure. The human development index is significantly negatively associated with outbound tourism expenditure. The higher living standard in home countries reduces the outbound expenditure on goods and quality life abroad. Lastly, the internet user is positively associated with outbound tourism expenditure. This indicates that more digitalization and internet use support information access, booking efficiency, and international mobility. The robustness checks using alternative estimators confirm the stability of these findings. In summary, the study highlights stronger long-run effects, potential asymmetries, and institutional heterogeneity across governance level. The evidence suggests that outbound tourism expenditure is not driven solely by income growth but is also shaped by governance quality.

The results yield several important policy implications for governments and tourism policymakers. First, the results provide additional evidence that may support policy efforts to improve governance quality, including strengthening political will and public awareness of its broader economic implications. Second, the influence of governance quality on outbound tourism expenditure is predominantly in the long run. Temporary declines in governance do not immediately increase outbound tourism, while sustained improvements in institutions lead to clear long-run changes in tourism behavior. This may highlight the importance of stable governance reforms rather than short-lived policy measures. Third, complementary policies in digital infrastructure and human development remain important in shaping the effect of governance quality on outbound tourism expenditure.

Despite its contributions, this study has some limitations. Governance is measured using a composite index, which may hide the effects of individual institutional components. Future studies could examine specific governance dimensions, such as corruption control, rule of law, and bureaucratic quality. Moreover, future research could use bilateral tourism flow data to better capture how governance in both origin and destination countries affects travel decisions. Furthermore, future research may apply alternative econometric approaches, such as quantile panel or time-varying models, to examine whether governance effects differ across countries with different levels of outbound tourism expenditure.