Spatial Heterogeneity in Carbon Footprint from Belt and Road Investments Assessed Through a Cross-Sectional Autoregressive Distributed Lag Model

The Belt and Road Initiative (BRI), a monumental global development strategy launched by China¹, has profoundly reshaped the economic and infrastructural landscape across Asia, Africa, and Europe². By channeling unprecedented levels of infrastructure-led foreign direct investment (FDI) into transportation, energy, and port facilities, the BRI aims to enhance regional connectivity and stimulate economic growth. A critical, yet complex, outcome of this enhanced connectivity has been the significant boost to tourism development in participating countries³, facilitated by improved accessibility and infrastructure⁴. However, this triad of infrastructure investment, tourism growth, and economic expansion carries substantial environmental implications, particularly concerning carbon emissions⁵. Understanding the interplay between these factors is paramount for aligning the BRI's economic objectives with global sustainability goals, such as the Paris Agreement and the UN Sustainable Development Goals⁶. This research specifically investigates the consequences of China's infrastructure-led FDI, tourism development, and technological innovation on carbon emissions within the context of BRI participant nations.

There is a significant literature that has entrenched the importance of tourism as an economic development driver, especially in the emerging markets^7,8. The sector is actively encouraged by governments because it brings revenues, contributes to the growth of entrepreneurship, and minimizes unemployment⁹. The industry has shown strong growth globally as international tourist arrivals have been exceeding the projections and bringing in revenues in billions annually, a trend that has been encouraged by the improvement of communication technology, changing visa procedures, and better tourist destinations¹⁰. At the same time, there has been questioning regarding the environmental imprint of this growth¹¹. One line of research points out the adverse effect of tourism, attributing it to the direct correlation of tourism with energy consumption and carbon emissions related to transport¹². Moreover, despite the opportunities that big infrastructure projects under the BRI bring to the tourism sector, they are also known to cause immense environmental stress due to the energy demand and ecological disturbances¹³.

However, existing literature exhibits critical gaps that this study seeks to address. First, while the general link between tourism and emissions is acknowledged, the empirical evidence remains ambiguous and often fails to account for the unique, multi-faceted context of the BRI. The relationship is not monolithic; it is mediated by a host of local factors, including logistical infrastructure, exposure to political instability, environmental standards, and climate conditions¹⁴. Second, a competing narrative in the literature posits that tourism, especially when aligned with eco-tourism principles and green policies, can be a vehicle for environmental sustainability rather than degradation¹⁵. This duality underscores a significant knowledge gap: under what conditions does tourism within the BRI framework exacerbate or mitigate emissions? Third, the role of infrastructure-led FDI is often studied in isolation¹⁶. The synergistic effect of FDI and tourism development, where new airports, roads, and hotels funded by FDI directly enable and stimulate tourist inflows, and its collective impact on the environment is not well understood. This research directly addresses these gaps by conducting a nuanced and comprehensive analysis of the environmental consequences of the BRI's economic drivers. Researchers move beyond broad generalizations to investigate the specific, combined impact of infrastructure-led FDI, tourism development, and technological innovation on carbon emissions. The contributions of this study are threefold. First, it provides novel empirical evidence on the interconnectedness of infrastructure investment, tourism, and carbon emissions within the specific and highly relevant context of the Belt and Road Initiative. By explicitly modeling the interaction between tourism development and FDI, researchers uncover whether their combination creates a synergistic effect that amplifies or mitigates environmental pressure, a dimension largely overlooked in previous studies¹⁷. Second, this research makes a significant methodological contribution by applying a CS-ARDL model and conducting a detailed regional analysis. This enables us to move past one-size-fits-all conclusions and reveal the considerable regional disparities in how these economic activities impact the environment.

Literature review

The tourism industry is widely recognized as a significant catalyst for economic development, particularly in emerging markets¹⁸. Empirical studies consistently demonstrate its capacity to generate income, stimulate entrepreneurial activity, and reduce unemployment, leading governments worldwide to adopt supportive policies for the sector¹⁹. Historical data underscores this dynamic growth, with the industry maintaining an average annual growth rate of about 5%, and international tourist arrivals projected to surpass 1.8 billion²⁰, a projection often exceeded by actual figures fueled by advancements in communication technology, improved tourist infrastructure, and streamlined visa regulations²¹. However, the sector's contribution to economic growth is not uniform, as its success is mediated by a complex array of local factors including logistical infrastructure, political stability, environmental standards, and climate conditions, suggesting that generalized models of tourism-led growth are often inadequate²².

Parallel to the expansion of tourism, the environmental repercussions of economic development have come into sharp focus, revealing a dualistic narrative within the academic discourse²³. One strand of literature firmly associates tourism development with increased energy consumption and transport-related carbon emissions, positioning the industry as a significant contributor to environmental degradation²⁴. This is particularly relevant in the context of large-scale initiatives like China's Belt and Road Initiative (BRI), which, through infrastructure-led foreign direct investment (FDI), enhances connectivity and stimulates tourism and economic activity, but concurrently creates substantial environmental pressures through increased energy demand and ecological disruption²⁵. In contrast, a competing scholarly perspective emphasizes the potential for sustainable tourism, arguing that eco-tourism and green policies can mitigate pollution and ensure long-term ecological sustainability, thus aligning economic growth with environmental stewardship²⁶. This duality extends to the role of FDI, which can be a source of both polluting technologies and advanced, environmentally friendly innovations, creating an empirical ambiguity that is further complicated when FDI is specifically directed towards infrastructure that enables tourism growth²⁷.

The complex interlinkages between tourism, investment, and the environment necessitate a deeper investigation into moderating factors and theoretical frameworks²⁸. Technology innovation is often cited as a critical mechanism for decoupling economic growth from emissions, through the adoption of renewable energy and green technologies²⁹. Furthermore, the theoretical foundation of the Environmental Kuznets Curve (EKC) hypothesis provides a framework for testing the relationship between economic development and environmental degradation, suggesting that emissions may initially rise with income before eventually declining³⁰. The transformative potential of the BRI, with projections suggesting its economies may account for half of global GDP by 2030, makes it a critical case study for examining these dynamics³¹. Yet, a significant gap remains in the literature: a lack of nuanced, empirical studies that simultaneously model the combined impact of infrastructure-led FDI, tourism development, and technological progress on carbon emissions across the heterogeneous regions of the BRI (Figure 1)³². Existing research often treats these drivers in isolation, failing to capture their synergistic effects and regional variations, a gap this research aims to fill by employing a methodological approach that accounts for both common correlated effects and cross-sectional disparities³³.

This study utilized exclusively publicly available, secondary macroeconomic data from official sources, including the World Bank's World Development Indicators (WDI) and the Ministry of Commerce of the People's Republic of China (MOFCOM). The underlying raw data are publicly available from the World Bank (https://databank.worldbank.org/source/world-development-indicators) and the Chinese Ministry of Commerce (MOFCOM). No human or animal subjects were involved in the research. As such, this study did not require review or approval by an institutional review board (IRB) or ethics committee.

Data and variable construction

This research analyzes EKC in the nations along with BRI and examines the correlation between Chinese outward FDI stocks and carbon emissions. Generally, the effects of economic activity on the environment may be broken down into three categories: encompasses what authors labeled as the Technology Innovation (TI) effect, the industrial composition effect, and the economic scale effect. Therefore, apart from the main features of the study, authors are depicting these critical features. The following is a perception of this correlation:

   (1)

In this model, several variables are used to represent various economic indicators. These include:

CO2: carbon emissions, TD: Chinese outward foreign direct investment stocks in Belt and Road Initiative countries, TI: TD indices, GDP: gross domestic product, SQGDP: squares of GDP (used to denote the Environmental Kuznets Curve), and IVA: industrial value-added.

The model uses a subscript to indicate the number of countries and a subscript ε_{it} to indicate time. The parameters in the model are represented by β_{1} + β_{2}  + β_{3}  + β_{4} + β_{5} + β_{6} and β_{7} . The term μ_{i} refers to a cross-sectional specific factor, ε_{it} and represents the mistaken expression.

The BRI has considerably enhanced communication between China and other contributing nations, importantly impacting power usage, fiscal growth, and environmental conditions. Steps taken by the Chinese government to develop the green BRI in line with the Paris Agreement (2015) and the SDGs 2030 include the spread of technological advantages and the growth of efficient and environmentally friendly energy infrastructure³⁴. Gu and Zhou³⁵ examined recently built clean power projects in BRI respective nations with 16 GW of installed capacity and helped reduce greenhouse gas emissions by 49 metric tons, reaffirming these approaches. Additionally, Li et al.³⁶ suggested that the Belt and Road Initiative boosts the hosting nation's power efficacy and economic development by constructing reliable energy and logistical facilities. However, China's FDI worsens ecological problems in hosting countries because of growing demands for power and natural resources. Depending on this rationale, it is projected that China's foreign direct investment abroad may have an impact on carbon emissions in various places that is both beneficial and harmful:

In the latest study, Bekun et al.³⁷ investigated the hypothesis that tourism revenues contribute to lower emissions, but visitor arrivals have the opposite impact. In addition, the synthesis of the scientific studies (see Table 1) emphasized that tourist variables caused specific environmental consequences. To comply, this research conducts an accumulated tourist engagement indicator that aids in removing subjective bias. The overall tourism business indicator might have a beneficial or detrimental effect on CO2.

Comparable to how increased connectivity, fewer barriers to competition (including both people and goods), faster shipments, and infrastructure improvement between BRI member nations encourage the expansion of the tourist industry, which could have a pivotal effect on environmental quality, given the inherent characteristics of the tourism economy, it might be advantageous.

Given the fundamental nature of economic activity, more energy and fuel consumption, which results in increased CO2, are necessary. Nevertheless, technical advancement to create ecologically friendly products and procedures, as well as innovation-led changes in energy supply, can significantly impact environmental sustainability. TI decreases carbon dioxide emissions in this way.

The EKC theory is supported by the economic growth being linked to increasing energy consumption³⁸, which raises emissions levels, but the square of development in the economy lowers carbon dioxide emissions.

One key element influencing the host nations' carbon dioxide emissions is the commercial makeup of those nations (Table 2). IVA may have a beneficial effect on CO2 emissions since increasing industry significance in BRI neighboring countries will result in increased energy consumption and associated emissions³⁹.

Data source and sample selection

The importance of BRI nations is predicated on the assumption that they represent 30% of the world's GDP, 64% of its inhabitants, 39% of its territory, 35% of its commerce, 50% of its power generation use, and 54% of its emissions of CO₂. In addition to these data, the BRI area has experienced enormous growth in Chinese FDI, making it the most effective collaboration platform in the globe⁴⁰. Consequently, authors investigate the ecological effects of China's outbound foreign direct investment, TD, and TI in the chosen 2 (54) Belting and Roads node nations from 2000 to 2020. To do this, the article makes use of yearly data on CO₂ emissions (metric tons), Chinese outward foreign direct investment stocks (constant 2010 US$), the TD indicator (for more information, see Table 3), and the total index (TI), which is calculated as the total amount of patents (for inhabitants and immigrants), the GDP (continual 2011 US$), and the manufacturing values (% of GDP). It is crucial to provide a consistent test statistic when resolving the problem of differential attributes since these factors display diverse measurement units. Considering previous research, the authors transformed all the aspects into the regression model, which produces results in the form of elasticity that facilitate data interpretation. Apart from Chinese external foreign direct investment capital, which is taken from China's Department of Economy and Trade, information for these factors is taken from the World Development Index (WDI).

TD index

Numerous metrics are employed in tourism studies to describe tourist activities. Tourism revenues⁴¹, tourist spending Pata and Balsalobre-Lorente⁴², and admissions⁴³ data are used to quantify comprehensive tourism demand. These tourist models have several drawbacks since they only account for a portion of the link between pollutants. These three tourist variables are very collinear, and if they are utilized concurrently in a unified framework, it might result in the multicollinearity issue, according to Meng et al.⁴⁴, PCA combines common fluctuations from all tourism variables into a unique accumulated score. These tourist metrics are drawn from the WDI dataset, which aids in creating a distinctive composite indicator by fusing the very critical data.

TD index through PCA

The Tourism Development (TD) index was constructed using Principal Component Analysis (PCA) to consolidate the three highly correlated tourism variables (international tourism receipts, expenditures, and arrivals) into a single composite score. The suitability of the data for PCA was first validated using two standard tests. The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy yielded a value of 0.723, exceeding the recommended threshold of 0.6, indicating that the patterns of correlation were compact and thus suitable for factor analysis. Furthermore, Bartlett's Test of Sphericity was statistically significant (p < 0.001), confirming that the correlation matrix was not an identity matrix and that the variables were sufficiently correlated for PCA. Based on the Kaiser criterion (eigenvalue > 1), only the first principal component was retained. This component satisfactorily explained more than 81% of the total variance across all three tourism variables. The factor loadings for all three variables were positive and high, supporting a balanced weighting in the computation of the final index. The resulting TD index scores ranged from -0.681 to 5.157 for the sample, where a lower value indicates a lower level of tourism development, and a higher value indicates a greater level of tourism development among the BRI nations.

Analytical framework

Researchers began the analysis by testing two essential panel data properties: cross-sectional dependence (CSD), which indicates interconnectedness between countries, and slope heterogeneity (SH), which signifies that the variable relationships differ across countries. Confirming these properties was a critical first step, as it necessitated the use of specialized second-generation econometric tests for accurate results.

Cross-sectional dependence (CSD) and slope heterogeneity (SH)

The premise of sloping variability and cross-sectional dependence maintained by second-generation (SGN) unit root (UR; including CS supplement IPS UR testing) and cointegration tests makes it necessary to verify before advancing to the factor's integration order. Because of the rapid fluctuations in financial and monetary projects succeeded by changes in the economy, the banking collapse, globalization, the interconnectedness of royalties, and missing visual and undetected common elements, data series inevitably pose the CSD problem. Prior to using UR and cointegrating tests, CSD enables us to select the most effective method to solve the problems that have been revealed. The research utilized a more sophisticated approach based on Tang et al.⁴⁵, CD testing to validate CSD. Additionally, the article adopts the Zhuang et al.⁴⁶, SH testing, which is better than conventional heterogeneity analyses like SURE that do not permit CSD in panels⁴⁷. The slope heterogeneity analysis formulas are as follows:

   (2)

   (3)

Unit roots test (URT)

A further phase is to use the proper unit root experiments to ensure the integration order once CSD and slope heterogeneity have been verified. The order of integration for all factors is determined by the cross-Sectionally augmented Dickey-Fuller (CADF) and cross-Sectionally enhanced IPS (CIPS) models. First-generation testing is the initial two URT, while SGN analyses are the third URT Since it permits CSD while computing variable unit roots, CIPS is better. The conventional panel unit root tests give false findings when CSD is present. Despite this, the current research employs basic and sophisticated unit root tests for objective and consistent estimations. The following is the CIPS equation:

   (4)

The Equation below uses the symbol to indicate the average across different sections.

   (5)

The CIPS statistical tests are as follows:

   (6)

Panel cointegration tests

Authors then estimate the cointegration connection between carbon dioxide emissions and their causes, including Chinese outbound FDI, TD, TI, and economic expansion, following confirming CDS, slope heterogeneity, and variation stationarity level. The research does this by utilizing the panel cointegration approach, which yields realistic predictions when erroneous factors are cross-sectionally dependent, and panels are diverse. Since this test does not impose typical element constraints while considering CSD and structural breakdowns, it is preferable to first-generation cointegration tests. Additionally, traditional panel estimation methods that use unobserved heterogeneity, fixed effect models, and instrumental variables do not consider the CSD of constant variance, resulting in inaccurate estimations. In contrast to the second-generation cointegration test, the panel cointegration technique addresses structural breaks at various locations for each cross-section in addition to CSD, SH, and serial correlation of mistaken expressions. The paper includes the cointegration test. The structural breakdown and paradigm changes affect genuine values. This analysis provides cointegration predictions without pollution, level break, or paradigm change (Table 4).

Cross-sectionally augmented autoregressive distributed lags

Authors may now consider experimental assessments of the long-term connection between greenhouse gases and their drivers. The traditional prediction model presupposes cross-sectional stability, which is not the case in the initial investigations. Given that the amount of carbon pollution is controlled by the unseen relevant variables that induce reliance on cross-sectional error terms, this is an implausible hypothesis. Several factors, including the unequal and sudden rise of Chinese external FDI and the change in financial and political frameworks, lend conceptual credibility to cross-section reliance on the Belt and Road program. This is because the natural characteristics may be distorted by the presence of unseen relevant variables that are linked with the regression analysis. The CS-ARDL estimation provides the best solution to the issue of cross-section dependence and slope heterogeneity. Propose a dynamic common correlated effects strategy that the CS-ARDL method uses to address these concerns. This leads us to the following derivation of the CS-ARDL formula:

   (7)

Simple autoregressive distributed lags (ARDL) models have the linear function shown above in Equation (7); if the authors continue with this formula, they will get mixed results depending on whether cross-sectional dependencies occur. Equation (7) is generalized to provide Equation (8) by averaging over cross-sections. The cutoff point impact is only present if cross-sectional dependency is considered, and this procedure will help eliminate that potentially misleading assumption.

   (8)

Where X_t-1 = (CO_2i,t–₁, Z_i,t–₁) may be thought of as the mean of the dependent and independent variables, P_w,P_z,P_x Identify all delays within every parameter. CO_2i,t, Carbon dioxide emissions (tCO2) is an example of a dependent variable, and Z_i,t is a vector that includes all the different factors. Cross-sectional averages (including time dummy or pattern variables) are shown by X demonstrating how they may be used to avoid CSD caused by ripple effects. The CS-ARDL uses short-run coefficients as inputs to calculate long-run coefficients. For the average number estimator and the long-run coefficient.

   (9)

The mean group is given as:

   (10)

Short-run coefficients are estimated as:

   (11)

Where,

    (12)

   (13)

    (14)

    (15)

In the event of a disturbance or the timeframe needed to attain constant balance, the mistake correcting procedures for CS-ARDLs are comparable to an organized average panel, which sets the pace of modification towards prolonged balance.

Robustness estimators

Traditional estimation procedures are incredibly opposed to this committee's acknowledged existence of slope heterogeneity and variance decomposition. Moreover, the research uses the Ordinary Connected Effect Average Panel measure established by Pesaran to confirm similar things. Even with CSD, heterogeneity, endogeneity, and non-stationarity, CCEMG can be used effectively (Table 5). It also has correlation problems, especially with the cross-sectional divisions. Following is the CCEMG formula:

   (16)

Granger causality

Granger causality is used to analyze the article's components and their possible contributing relationship. Owing to its core assumptions, this method works better with diverse displays. Against the hypothesis test of causality between variables, the null hypothesis of the DH multiple regression analysis is that there is no causation. Following is a description of the practical structure of the DH Granger causality test:

   (17)

For an autoregressive model, the estimator is denoted by j and the autoregressive variables are represented by βj(j)

The presented results effectively demonstrate the utility of the Cross-sectionally Augmented Autoregressive Distributed Lag (CS-ARDL) technique in capturing complex, real-world economic relationships (Figure 2). For instance, the divergent long-run impact of Foreign Direct Investment (FDI) on carbon emissions is negative in some regions, like Southeast Asia, but positive in others, like South Asia, exemplifying the slope heterogeneity that the CS-ARDL model is specifically designed to handle. This regional variation would likely be obscured by traditional estimators that assume uniform coefficients across all countries. Furthermore, the highly significant and negative Error Correction Term (ECM) across all regions, indicating a 31%-63% speed of adjustment back to long-run equilibrium after a shock, validates the cointegrating relationship identified earlier and confirms the model's dynamic stability (Figure 3). When analyzing these outcomes, one should focus not only on the significance and sign of a coefficient but also on its comparative magnitude across regions; for example, the strong positive interaction term (FDI x TD) in MENA countries suggests that tourism growth in that region significantly amplifies the environmental impact of foreign investment, a nuanced insight crucial for targeted policymaking.

SH and CSD

Confirming the problem of SH and demonstrating that conventional URTs and cointegrating processes cannot provide accurate estimations. Both t-tests and chi-squared tests show that the null hypothesis of slope heterogeneity is false.

Unit root tests

In this analysis, the stationarity properties of the variables, a prerequisite for reliable regression analysis, were confirmed using second-generation unit root tests (CADF and CIPS). The order of integration for each variable, which indicates the number of times a series must be differenced to become stationary, was determined. A variable stationary in its original form is termed I(0), while one that becomes stationary after first differencing is I(1). The results revealed a mixed order of integration among the factors. Specifically, variables like FDI and Technology Innovation (TI) were found to be stationary at level I(0), meaning they do not have a unit root and exhibit a stable mean and variance over time. In contrast, Tourism Development (TD), GDP, and Industrial Value-Added (IVA) were found to be integrated of order one I(1), requiring first differencing to achieve stationarity. The presence of both I(0) and I(1) variables validates the use of the CS-ARDL estimation technique, which is robust to such mixed orders of integration. The CIPS test results are considered most reliable in this context, as they explicitly account for cross-sectional dependence across the panel, a feature confirmed in our initial diagnostic tests. Carbon emissions, FDI, and TI are confirmed to be level-stationary by the CIPS unit root test, whereas TD, industrial value added, and economic expansion are first-difference-stationary. Since the factors are being integrated non-sequentially, authors must use the CS-ARDL and Westerlund and Edgerton estimate methods.

Panel (PN) cointegration

Analogous estimations are standard cointegration methods, and Westerlund and Edgerton's48, state-of-the-art cointegration methods. PN-ADF, PN-PP, board-ADF, and board-PP testing data from Table 6 corroborate the long-run cointegration connection, although Panel v, Panel rho, and group rho indicate no cointegration. The Pedroni cointegration method is susceptible to misrepresentation due to the heterogeneous panel's poor handling of cross-sectional dependence and inability to account for fundamental fractures within panel nations when calculating statistical tests49. Consequently, it is crucial to use an alternate method to verify a long-run cointegration connection.

Brief and prolonged elasticities from CS-ARDL

After establishing the variables' cointegration, this research investigates the prolonged and brief connections between CO2 releases and their causes (Table 7). The CS-ARDL estimator's short-run and long-run estimations for the BRI dataset are done properly in the analysis (Table 8). To begin with, the result exhibits considerable regional heterogeneity in the coefficient's size, location, and importance. Short-run estimations based on observational data indicate that foreign direct investment (FDI) and total international trade (TI) have a detrimental (favorable) effect on carbon emissions over the entire BRI sample, with coefficient values of -0.077 and 0.123 (0.271), respectively. In the BRI sample, the combined impact of TD and FDI on emissions is 0.068, which is statistically significant. SQGDP's sign is notably hostile at -0.132, and the existence of EKC is also verified. If the IVA coefficient is positive, carbon emissions will rise by 0.240. Except in Southern Asia and MENA nations, where FDI had a beneficial impact on CO2 emissions of 0.167 and 0.219 percentage points, respectively, these findings are comparable to those of regional sampling. In the event of a stun or imbalance in a brief period, the ECM from every framework is substantial and adverse at 1% degree, confirming the rapid shift towards prolonged balance. In the case of all BRI countries, the reliability coefficients reveal a correction factor of 41.6%; for South Asian countries, it's 43.9%; for Southeast Asian countries, it's 31.2%; for Central Asian countries, it's 34.6%; for MENA countries, it's 63.4%; for Central and Eastern European countries, it's 33.5%; for Western European countries, it's 36.3%; and for CIS countries, it's 32.1%. A long-term link between variables was confirmed across all BRI regional samples, as shown by their negative ECM values.

Heterogeneous panel causality

Granger causality analyses confirm that shifts in FDI, TD, TI, IVA, and GDP Granger cause carbon dioxide emissions in BRI nations (Table 9). These results show that substantial changes in CO2 emissions result from any strategy targeting these factors. Yet statistical tests and probability values confirm another causation between FDI and CO2 emissions, tourism and CO2 emissions, creativity and CO2 releases, industrial activity and CO2 releases, and financial development and CO2 releases (Figure 4). Such adverse effects make theoretical sense because of the significant roles that international investment, tourism, technology, and GDP play in generating greenhouse gas emissions. Also, these results align with the most current research on advanced and developing nations50.

Data availability:

The datasets generated and/or analyzed during the current study are publicly available from the World Bank's World Development Indicators (WDI) and the Ministry of Commerce of the People's Republic of China (MOFCOM). The underlying raw data are publicly available from the World Bank (https://databank.worldbank.org/source/world-development-indicators) and the Chinese Ministry of Commerce (MOFCOM). 

Figure 1: Trends and Overall Contribution of Chinese Outward FDI, Tourism Growth, and Technological Advancements (2000-2020). The figure shows trends and overall contributions of Chinese outward foreign direct investment (FDI), tourism growth, and technological advancements over the period from 2000 to 2020. The bar graphs on the left depict the annual values of FDI, tourism growth, and technological advancements, highlighting the significant rise in these areas due to the Belt and Road Initiative (BRI). The pie chart on the right side summarizes the overall contribution of each factor over the two decades, providing a visual representation of their relative impact. Please click here to view a larger version of this figure.

Figure 2: Cross-sectional ARDL model results. The figure presents the coefficients and significance levels of key variables, Foreign Direct Investment (FDI), Tourism Development (TD), Technology Innovation (TI), and Gross Domestic Product (GDP), from the Cross-Sectional Autoregressive Distributed Lag (CS-ARDL) model across different regions. Each panel compares the effects of these variables on carbon emissions, highlighting regional variations in their impact. Please click here to view a larger version of this figure.

Figure 3: Environmental Kuznets curve (EKC) validation. The figure illustrates the Environmental Kuznets Curve (EKC) hypothesis, showing the relationship between GDP per capita and carbon emissions for the BRI countries. The graph depicts an inverted U-shaped relationship, validating the study's findings that economic development initially increases environmental impact, but beyond a certain level of GDP per capita, further economic growth leads to a reduction in carbon emissions. Please click here to view a larger version of this figure.

Figure 4: Impact of FDI and tourism development interaction on carbon emissions. The figure presents the interaction between Foreign Direct Investment (FDI) and tourism development, and their combined effect on carbon emissions across different regions. The bar graphs represent the values of FDI and tourism development for each region, with the numerical labels indicating the level of carbon emissions. Different colors are used to distinguish between FDI and tourism development, highlighting how varying levels of these factors interact to influence environmental outcomes. Please click here to view a larger version of this figure.

Table 1: Summary of literature review. This table provides a comprehensive overview of prior studies examining the relationships between tourism, foreign direct investment (FDI), and environmental impacts, particularly carbon emissions, in the context of Belt and Road Initiative (BRI) countries and other regions. It synthesizes key findings on how tourism development, FDI, and technological innovation influence economic growth and ecological sustainability, highlighting both positive and negative environmental effects. Please click here to download this Table.

Table 2: Slope Heterogeneity (SH) Test Results. Table 2 presents the results of the Pesaran and Yamagata (2008) slope heterogeneity (SH) tests for the BRI countries dataset. It includes statistical measures, such as t-tests and chi-squared tests, to evaluate the null hypothesis of slope homogeneity, demonstrating significant heterogeneity in the relationships between variables across different cross-sectional units. Please click here to download this Table.

Table 3: CSD (Pesaran 2015) testing. The table clearly demonstrates that CS also depend on one another. All model variables are found to violate the null hypothesis of cross-sectional independence. Cross-sections are indeed dependent, as seen in Table 3. Rejecting the cross-sectional independent null hypothesis for all modeled components. Thus, the characteristics of one nation may be affected by a disturbance in another country that occurs in a different cross-sectional unit. Please click here to download this Table.

Table 4: Results of stationary analysis. ***shows 1% significance. The null hypothesis states that there is no unit root. This table summarizes the results of stationarity tests, including cross-sectionally augmented Dickey-Fuller (CADF) and cross-sectionally enhanced Im, Pesaran, and Shin (CIPS) unit root tests. It shows that carbon emissions, FDI, and technology innovation (TI) are level-stationary, while tourism development (TD), industrial value-added (IVA), and economic expansion are first-difference-stationary, highlighting the multi-level interdependence of the factors. Please click here to download this Table.

Table 5: Outcomes of Pedroni (Engle-Granger base) board cointegrations. The Pedroni cointegration tests provide mixed but ultimately convincing evidence for a long-run relationship among the variables. While the Panel v-Statistic, Panel rho-Statistic, and Group rho-Statistic fail to reject the null hypothesis of no cointegration, the more reliable Panel PP-Statistic, Panel ADF-Statistic, Group PP-Statistic, and Group ADF-Statistic are all statistically significant at the 1% level. This consistent significance across four of the seven tests, particularly the group-adf statistic, which is designed for heterogeneous panels, allows us to reject the null hypothesis and conclude that a stable long-run cointegrating relationship exists between carbon emissions and its determinants in the Belt and Road Initiative countries. Please click here to download this Table.

Table 6: Westerlund and Edgerton (2008) PN cointegration tests with CSD and structural breaks. The table provides robust evidence of a long-run cointegrating relationship among the variables, even after accounting for cross-sectional dependence and structural breaks. Both test statistics (ZƮ and ZØ) are statistically significant at the 1% level across all three model specifications with no shift, a level shift, and a regime shift. This consistent rejection of the null hypothesis of no cointegration confirms that the variables share a stable long-run equilibrium relationship. Please click here to download this Table.

Table 7: Brief period outcomes from CS-ARDL estimator. The table details the short-run estimates from the cross-sectionally augmented autoregressive distributed lag (CS-ARDL) model for the BRI dataset and regional subsamples. It shows that FDI and TI have a negative (favorable) effect on carbon emissions, while the combined impact of TD and FDI increases emissions. The table also confirms the Environmental Kuznets Curve (EKC) with a negative SQGDP coefficient and reports error correction mechanism (ECM) values indicating rapid adjustment to the long-run equilibrium. Please click here to download this Table.

Table 8: Extended-term results using CS-ASRDL and CCEMG Long-run estimation methods. The table presents long-run estimates from the cross-sectionally augmented autoregressive distributed lag (CS-ARDL) and common correlated effects mean group (CCEMG) models for the Belt and Road Initiative (BRI) dataset and regional subsamples. It shows that foreign direct investment (FDI) is significantly and negatively associated with carbon emissions in the entire BRI dataset (-0.165), Southeast Asia (-0.239), Central Asia (-0.173), Central and Eastern Europe (-0.229), Western Europe (-0.135), and CIS nations (-0.126), indicating a reduction in emissions. Conversely, FDI increases emissions in South Asia and MENA regions due to investments in transportation, ports, industrial areas, and fossil fuel-based energy infrastructure, driven by lax environmental regulations. Tourism development (TD) and its interaction with FDI positively influence CO2 emissions, exacerbating environmental impact. The GDP coefficient (0.532) suggests that a 1% rise in GDP increases emissions, while technology innovation (TI, -0.227) reduces emissions. The negative SQGDP coefficient validates the Environmental Kuznets Curve (EKC) hypothesis, indicating that higher GDP levels eventually offset emission increases. Significance levels are denoted by ***, **, and * for 1%, 5%, and 10%, respectively. Regional estimates vary in size and significance, with CCEMG confirming the robustness of CS-ARDL results, highlighting stronger long-run environmental impacts compared to short-run effects. Please click here to download this Table.

Table 9: Outcomes of heterogeneousness panel causation testing. The table shows a clear unidirectional causal relationship is established. The findings indicate that Foreign Direct Investment (FDI), Tourism Development (TD), Technology Innovation (TI), Gross Domestic Product (GDP), and Industrial Value Added (IVA) all exhibit a statistically significant Granger-causal influence on carbon emissions (CO2). This is evidenced by the significant test statistics and probabilities below the 0.10, 0.05, and 0.01 thresholds for the directions of FDI→CO2, TD→CO2, TI→CO2, GDP→CO2, and IVA→CO2. Conversely, for all variables, the null hypothesis that CO2 does not Granger-cause them cannot be rejected, as all probabilities in the reverse direction are statistically insignificant. Please click here to download this Table.

The results of this empirical work provide a rich and nuanced story regarding the possible environmental impact of economic integration within the framework of the Belt and Road Initiative and transcend the simplistic dichotomies that tend to define the literature. Testing slope heterogeneity and cross-sectional dependence must not be introduced as a methodological nicety at the very outset, but rather as a basic point; it must be recognized that the relations between FDI, tourism, technology, and carbon emissions are not homogeneous across BRI countries and are strongly impacted by the spatial and economic interrelations. This underlying observation explains why authors have adopted a novel use of the CS-ARDL and CCEMG estimators, which are inherently constructed with such a multifaceted panel data structure in mind. The most impressive and distinctive element of this analysis is its disaggregation by region, which demonstrates that the environmental footprint of infrastructure-based FDI is not single-dimensional but varies depending on the economic structure of the recipient region, regulatory environments, and the sectoral composition of the investments. Although FDI flows to Southeast Asia, Central Asia, and European countries have increased greatly, probably because of techno-spillovers and investment in new, efficient technology⁵¹, it has the opposite effect on South Asian and MENA countries. This is an important diffraction, which is lost in the analysis at the full sample level, providing a revelation: that the same macroeconomic driver can produce different environmental effects of opposite impacts based on the local conditions.

This study offers a new insight into an under-researched dynamic, namely the strong interrelationship between foreign direct investment and tourism development. The results of our study indicate that, although tourism development (TD) would positively affect emissions, the association of the same with FDI (FDI x TD) would yield an extra and substantial positive coefficient in most of the regions. It is an implication that the infrastructure developed by Chinese FDI, new airports, roads, and hotels, is a trigger factor that promotes the further carbon footprint of the tourism industry, per se. This synergistic effect provides an answer to studies that had in the past been unable to fully describe this aspect⁵², which mostly focused on FDI and tourism separately. It is clear now that the infrastructure intended to drive economic growth and integration of tourism may unwillingly trap more emission pathways as it facilitates more energy-intensive forms of travelling and activities. Moreover, the fact that the Environmental Kuznets Curve (EKC) hypothesis was repeatedly confirmed over all the regional samples, although they differ greatly in terms of development, is an interesting fact. It denotes a universal possibility of economic maturation to come to the rescue of environmental stress one day within the BRI framework, but the point of turn and the curve slope are highly diverse. The strong and negative correlation with technological innovation (TI) validates that technological innovation (TI) is a very fundamental mitigating force, which underlines the fact that technology transfer is a fundamental ingredient towards the realization of sustainable development in the initiative.

Some of the findings are especially surprising and have serious policy implications. The most surprising thing about the effects of FDI in the two regions is the dramatic difference, which undermines the common wisdom that FDI is either a pollution halo or a pollution haven blanket-wise. The finding that the South Asia and MENA countries are getting higher emissions due to FDI is concerning and indicates that investments in energy and industrial projects that are carbon-intensive are being concentrated, and the environmental governance may be weaker. The other unexpected but methodologically encouraging finding is that the long-run coefficients are larger than the short-run estimates. This means that the environmental impact of FDI and tourism is not instantaneous but rather builds up and develops over the years, which underscores the necessity of a long-term approach in strategic planning. The high speed of adjustment (ECM), particularly in the MENA countries at 63.4, indicates that the adjustments of any policy-induced deviation of the long-run equilibrium are rapid, which indicates the resilience and path-dependence of the current economic-energy models. Lastly, the heterogeneous causality tests cement the directional relationships, which prove that it is true that policies affecting FDI, tourism, and technology indeed will Granger-cause changes in carbon emissions and not the other way round, pitting the onus of action squarely on strategic economic planning. All these novel findings are a multifaceted evidence-based roadmap to customizing the future path of the BRI towards inclusive and sustainable development, which would recognize and utilize regional specifics instead of using a one-size-fits-all strategy.

Policy implications

The policymakers in the partner countries of the Belt and Road Initiative (BRI) need to take a differentiated and strategic stance to achieve economic gains and reduce environmental expenses. In the case of South Asia and MENA, where infrastructure-based foreign direct investment (FDI) causes carbon emissions⁵³, there is a dire need to enhance environmental standards and screening procedures, clearly encouraging green FDI in renewable energy and sustainable infrastructure rather than carbon-intensive ones. On the other hand, every country must be aggressive in promoting the technological breakthrough that has proven to have an emission-cutting impact by encouraging the transfer of green technology and investments in research and development. Besides, the synergistic, emissions-enhancing nature of FDI and tourism development requires the establishment of compulsory environmental impact assessment and sustainable design concepts⁵⁴, like energy-efficient buildings and low-carbon transportation, as the fundamental elements of all tourism-related infrastructural development. Last, the confirmation of the Environmental Kuznets Curve (EKC) hypothesis provides an important long-term outlook, highlighting that the BRI today must strategically invest in a green economic transition to reach the tipping point where further growth is not dependent on environmental degradation and thus would leave the BRI with the legacy of a sustainable, resilient development.

Conclusion

This paper presents detailed and subtle empirical evidence of dynamic relationships among infrastructure-driven foreign direct investment (FDI), tourism development, technological innovation, and carbon emissions in the framework of the Belt and Road Initiative (BRI). These relationships are also strongly supported by the results as being not homogeneous but are highly cross-sectionally dependent and strongly heterogeneous by geography. Although technological innovation remains a decisive factor in the elimination of carbon emissions in all samples, the environmental nature of FDI and tourism greatly relies on the region. The most important message is the dual nature of infrastructure-led FDI that can become an agent to reduce emissions in particular areas, such as Southeast Asia and Europe, but becomes a serious contributor to higher carbon emissions in others, in particular, South Asia and the MENA countries, more so when interacting with tourism development at synergistic levels. Moreover, the confirmation of the Environmental Kuznets Curve (EKC) hypothesis creates a promising outlook that, under the strategic policy guidance, there is an ultimate decoupling of economic growth within the BRI framework and environmental degradation.

Limitations and future research

This study has specific limitations, even though its methodological approach is substantial, which provides promising prospects for the literature in the future. The main limitation is that it is aggregate, whereas the CS-ARDL model may account for regional heterogeneity; it fails to separate the sectoral composition of foreign direct investment (renewable energy versus fossil-fuel power plants) or distinguish between different types of tourism (mass tourism versus eco-tourism), which may not cause the same environmental effect. Additionally, the use of nationally aggregated data can hide important sub-national differences in economic activity and ecological impacts. Subsequent research would thus be of immense benefit to refine the granularity of the current research by using sector-level or firm-level data about Chinese FDI, as well as considering more specific measures of tourism, such as types of tourist expenditure or length of stay. Also, by incorporating quantitative metrics of environmental policy stringency or the quality of governance in the model, a better understanding of which institutional mechanisms hold the key to FDI becoming a pollution halo or a pollution haven could be able to be gained, and the heterogeneity would no longer be discovered, but its underlying causes could be explained.