Assessing the environmental impact of Chinese outward FDI, tourism growth, and technological advancements in BRI countries

1 Introduction

Achieving sustainable development requires countries to expand investment, infrastructure, mobility, tourism, and technological capacity without intensifying environmental degradation. This challenge is particularly important for economies participating in the Belt and Road Initiative (BRI), where Chinese outward foreign direct investment (FDI), infrastructure connectivity, tourism expansion, industrial transformation, and technological progress are occurring simultaneously. Tourism contributes to employment, income generation, economic diversification, and regional development, but its dependence on transportation, accommodation, recreational facilities, and supporting infrastructure can increase energy consumption and greenhouse-gas emissions (Churchill et al., 2022; Razzaq et al., 2023; Peng et al., 2025). Similarly, Chinese outward FDI may improve connectivity, production capacity, technology transfer, and infrastructure quality, while investment in fossil-fuel energy, transport corridors, construction, and resource-intensive industries may increase carbon emissions (Liu et al., 2023; Ahmad and Hussain, 2024; Wang et al., 2024; Ma et al., 2025). The environmental implications of Chinese outward FDI can be understood through the competing Pollution Haven and Pollution Halo perspectives. The Pollution Haven perspective suggests that foreign investment may increase emissions when it is directed toward pollution-intensive activities or economies with relatively weak environmental regulations (Guzel and Okumus, 2020). Conversely, the Pollution Halo perspective argues that foreign investment can improve environmental performance through cleaner technologies, energy-efficient infrastructure, advanced management practices, and technological spillovers (Cao et al., 2026). The net environmental effect of Chinese outward FDI is therefore theoretically ambiguous and may vary according to investment composition and host-country conditions.

Tourism development may further condition this relationship. Chinese investment in roads, railways, airports, ports, energy facilities, hotels, and urban infrastructure can improve destination accessibility and expand tourism capacity. However, the resulting increases in passenger mobility, accommodation, construction, electricity consumption, and tourism services may intensify environmental pressure. Tourism development may therefore weaken or reverse the environmental benefits associated with investment-led technological and infrastructure improvements. Technological innovation provides an additional mechanism through which environmental outcomes may change. Innovation can improve energy efficiency, facilitate renewable-energy adoption, support cleaner production, and reduce the carbon intensity of infrastructure and industrial activities (Tan et al., 2024; Si et al., 2025; Yang et al., 2026). The Environmental Kuznets Curve framework further suggests that emissions may initially increase with economic development because of scale and industrialization effects but decline after economies achieve technological upgrading, structural transformation, and stronger environmental regulation. Nevertheless, recent evidence indicates that the validity of this relationship varies across countries, environmental indicators, energy systems, and econometric approaches (Guo and Shahbaz, 2024; Liu et al., 2024).

Despite the growing literature on foreign investment, tourism, technological innovation, economic growth, and environmental quality, several gaps remain. First, many studies employ aggregate FDI and do not distinguish investment originating from China, despite the close association of Chinese outward FDI with BRI-related energy, transport, infrastructure, and industrial-connectivity projects. Second, tourism is commonly examined as an independent determinant of emissions, while limited attention has been given to whether tourism strengthens or weakens the environmental effect of Chinese investment. Third, broad panel studies frequently report average effects without sufficiently examining differences across BRI regions. Consequently, existing evidence provides limited understanding of how Chinese outward FDI, tourism development, and technological innovation jointly shape carbon emissions across diverse regional settings.

This study addresses these gaps by examining 60 BRI countries over the period 2000–2024. It evaluates the short-run and long-run effects of Chinese outward FDI, tourism development, technological innovation, economic growth, and industrial value added on carbon emissions. It also incorporates an interaction between Chinese outward FDI and tourism development and estimates separate models for South Asia, Southeast Asia, Central Asia, the Middle East and North Africa, Central and Eastern Europe, Western Europe, and the Commonwealth of Independent States. The study makes four contributions. First, it distinguishes Chinese outward FDI from aggregate foreign investment and evaluates its environmental implications through the Pollution Haven and Pollution Halo perspectives. Second, it conceptualizes tourism development as a conditioning mechanism rather than only an independent determinant of carbon emissions. Third, it examines whether the estimated relationships differ across seven BRI regions characterized by diverse energy systems, regulatory environments, technological capabilities, and levels of development. Fourth, it evaluates the Environmental Kuznets Curve while jointly considering Chinese investment, tourism intensity, technological innovation, and industrialization.

Empirically, the study applies the Cross-Sectionally Augmented Autoregressive Distributed Lag approach, which accounts for cross-sectional dependence, slope heterogeneity, short-run adjustment, and long-run relationships. Common Correlated Effects Mean Group estimates are used to assess the robustness of the main findings, while the Dumitrescu–Hurlin panel Granger non-causality test provides complementary evidence regarding predictive direction rather than structural causality (Dumitrescu and Hurlin, 2012). The remainder of the paper is organized as follows. Section 2 reviews the relevant literature. Section 3 presents the data, variables, model specification, and econometric methodology. Section 4 reports and discusses the empirical findings. Section 5 concludes the study and presents policy implications, limitations, and future research directions.

2 Literature review

2.1 Chinese outward FDI and environmental quality

The environmental consequences of FDI are generally explained through the competing Pollution Haven Hypothesis (PHH) and Pollution Halo Hypothesis (PH1H). In the BRI context, these two hypotheses must be considered specifically in relation to the sectoral composition and destination of Chinese outward FDI. Chinese investment is frequently concentrated in transportation, energy, mining, construction, industrial parks, logistics, and large-scale infrastructure. Investment in coal-fired power generation, conventional transportation, resource extraction, heavy industry, and extensive construction may increase fossil-fuel consumption and carbon emissions, supporting the PHH explanation (Cao et al., 2026). This effect may be particularly pronounced in host economies with carbon-intensive energy systems, weak environmental institutions, or limited capacity to evaluate and regulate large infrastructure projects. Nugent and Lu (2021), for example, found that Chinese outward FDI toward BRI economies increased disproportionately in sectors associated with industrial overcapacity and pollution in China, particularly in host countries with relatively poor institutional quality. Their findings provide a direct mechanism through which Chinese outward FDI may relocate carbon-intensive economic activities.

Chinese outward FDI can nevertheless generate Pollution Halo effects. Investment in renewable energy, modern railways, electric mobility, smart infrastructure, energy-efficient industrial facilities, and cleaner production technology may reduce the carbon intensity of host-country economic activity. Moreover, knowledge transfer, technological spillovers, improved logistics, and industrial upgrading may allow host economies to produce more efficiently. Mahadevan and Sun (2020) found that the environmental effects of Chinese outward FDI across BRI economies were mixed, demonstrating that Chinese investment could not be classified uniformly as either environmentally beneficial or environmentally harmful. Su et al. (2022), using 67 BRI countries, found that Chinese direct investment reduced carbon emissions, thereby supporting the PH1H.

Recent studies further demonstrate that the environmental effect of Chinese outward FDI is conditional rather than uniform. Liu et al. (2024) found that Chinese outward FDI supported the PH1H in European BRI countries but was associated with Pollution Haven effects in Asian and MENA economies. Li et al. (2024), using 65 BRI countries, reported that Chinese outward FDI improved total-factor carbon-emission performance primarily through technological advancement and industrial-structure optimisation, although the magnitude of the effect differed considerably across regions. These findings show that the environmental outcome of Chinese investment depends on the type of projects financed, the host country's energy mix, environmental governance, technological absorptive capacity, and stage of economic development.

The existing evidence therefore does not support a universal positive or negative coefficient for Chinese outward FDI. Instead, PHH and PH1H mechanisms may operate simultaneously across different projects and host economies. This theoretical ambiguity motivates both the direct estimation of Chinese outward FDI and the regional analysis undertaken in the present study.

2.2 Tourism development, infrastructure investment, and carbon emissions

Tourism generates employment, foreign-exchange earnings, infrastructure development, and regional economic diversification. However, its environmental effect depends on the transport, accommodation, energy, construction, and consumption systems supporting tourism activity. Air travel, road transportation, hotel operations, recreational services, food supply chains, and destination infrastructure can increase energy demand and carbon emissions. Churchill et al. (2022) is relevant to this general tourism–environment relationship, particularly its argument that environmental degradation may both result from tourism activity and subsequently reduce the attractiveness of tourism destinations. Nevertheless, this general evidence does not directly explain the environmental consequences of Chinese outward FDI or infrastructure-supported tourism in BRI economies.

The BRI creates a more specific mechanism because Chinese outward FDI frequently finances infrastructure that facilitates cross-border mobility and destination accessibility. Roads, railways, airports, ports, energy facilities, communication networks, hotels, and urban infrastructure can reduce travel costs and stimulate tourism demand. At the same time, these investments may increase construction activity, electricity demand, transport intensity, land conversion, and fossil-fuel use. Consequently, tourism development may alter the environmental effect of Chinese outward FDI rather than operating only as an independent determinant of emissions. Zhuang et al. (2023) directly examined infrastructure-led Chinese outward FDI, tourism development, technological innovation, and carbon emissions across BRI economies. Their results indicated that Chinese outward FDI and technological innovation reduced emissions in the full sample, while tourism development and the interaction between Chinese outward FDI and tourism increased emissions. They also reported regional differences, including emissions-increasing effects of Chinese investment in South Asia and MENA. This study is especially relevant because it demonstrates that the FDI–tourism interaction and regional variation have already been considered in the BRI literature. More recent research has continued to examine this joint relationship. Zhang et al. (2026), using 56 BRI economies during 2006–2023, constructed a PCA-based tourism index and applied CS-ARDL and CCEMG estimators. The study similarly found that tourism development and its interaction with Chinese outward FDI intensified carbon emissions, while the effects of Chinese outward FDI varied across regions.

2.3 Technological innovation and environmental quality

Innovation is a vital factor in the interactions between economic development and the environment. Technological innovation can enhance energy efficiency, minimize resource use, and support the shift to more sustainable production practices. According to Razzaq et al. (2023), innovation-led processes help economies achieve both growth and sustainability by decarbonizing the economy through production processes. For tourism, technological innovation leads to increased efficiency, resource management and service provision. Online platforms, smart transport systems and predictive systems can enhance energy efficiency and minimize environmental impacts (Feng et al., 2023). Likewise, in the manufacturing and infrastructure sectors, new technologies can encourage the shift toward renewable energy and sustainable production. But the success of technological innovation in mitigating emissions also relies on its adoption and dissemination in various regions. In emerging markets, the benefits of technological innovation might be constrained by weak institutional frameworks, insufficient infrastructure, and financial constraints. As a result, the effectiveness of technological innovation in reducing emissions may differ greatly among nations and regions, especially in the context of the diverse BRI.

2.4 Economic growth and EKC evidence

The EKC hypothesis proposes that environmental degradation may initially increase with economic growth but decline after income reaches a sufficiently high level. Grossman and Krueger (1991) remain important as a foundational contribution to this framework. However, their study should not be treated as sufficient evidence of the current validity of the EKC hypothesis. Recent reviews demonstrate that EKC findings remain mixed and sensitive to environmental indicators, country composition, income levels, energy structures, time periods, functional forms, and econometric methods. Guo and Shahbaz (2024), reviewing more than 100 EKC studies published between 1991 and 2023, concluded that recent research increasingly incorporates energy diversification, technological development, sector-specific indicators, and more advanced econometric methods. Their review also emphasizes that confirmation of an inverted U-shaped relationship is not universal.

Recent BRI-specific evidence is similarly heterogeneous. Liu et al. (2024) found support for an inverted U-shaped relationship in European and Asian–MENA BRI panels, but also demonstrated that the environmental effects of Chinese outward FDI differed substantially between the two groups. Thus, income growth alone cannot explain environmental outcomes. Its effect depends on the accompanying energy transition, industrial restructuring, technological upgrading, foreign-investment composition, and quality of environmental governance. The present study therefore treats the EKC as an empirical proposition rather than an established fact. GDP per capita and its squared term are included to assess whether the scale effects of economic expansion are eventually offset by technological, structural, and regulatory improvements in the full BRI sample and individual regional groups.

2.5 Comparative assessment of existing studies

Existing research can be classified into three main groups. The first examines aggregate FDI and environmental quality but does not distinguish the source country or sectoral composition of foreign investment. Such studies provide general evidence concerning the Pollution Haven and Pollution Halo mechanisms but cannot isolate the environmental effect of Chinese outward FDI. The second group specifically examines Chinese outward FDI in BRI economies. Mahadevan and Sun (2020) reported mixed environmental effects, demonstrating that Chinese investment cannot be classified uniformly as either environmentally beneficial or environmentally harmful. Nugent and Lu (2021) showed that Chinese outward FDI was concentrated in sectors associated with industrial overcapacity and pollution, particularly in host economies with relatively weak institutional conditions. Su et al. (2022), however, found that Chinese investment reduced carbon emissions across BRI countries, supporting the Pollution Halo perspective. More recent studies have confirmed that these effects differ across regions, sectors, energy systems, and host-country conditions (Liu et al., 2024; Li et al., 2024).

The third and most closely related group jointly examines Chinese outward FDI, tourism development, technological innovation, and regional environmental heterogeneity. Zhuang et al. (2023) found that Chinese outward FDI and technological innovation reduced emissions in the full sample, whereas tourism development and the interaction between tourism and Chinese investment increased emissions. Zhang et al. (2026) similarly reported that tourism development and its interaction with Chinese outward FDI intensified environmental pressure, while the direct effect of Chinese investment differed across regions. The present study extends this literature in several respects. It differs from studies using aggregate FDI by measuring Chinese outward FDI stock specifically. It extends Su et al. (2022) by treating tourism development as a conditioning mechanism and distinguishing short-run and long-run effects. It extends Liu et al. (2024), which used two broad geographical panels, by estimating seven BRI regional groups. It differs from Li et al. (2024), which focused on total-factor carbon-emission performance and spatial policy effects, by examining carbon emissions, tourism intensity, Environmental Kuznets Curve dynamics, and short-run adjustment.

Relative to the closest studies by Zhuang et al. (2023) and Zhang et al. (2026), the present analysis extends the geographical and temporal coverage to 60 BRI countries over 2000–2024. It also combines CS-ARDL estimation with CCEMG robustness analysis, structural-break-sensitive cointegration testing, and Dumitrescu–Hurlin panel Granger non-causality analysis. Table 1 provides a detailed comparison of the geographical coverage, methods, findings, and remaining limitations of the principal studies.

Table 1. Comparative review of empirical studies.

3 Materials and methods

3.1 Data and sample selection

This study uses an unbalanced panel dataset covering 60 BRI countries over the period 2000–2024. The sample covers economies from South Asia, Southeast Asia, Central Asia, the Middle East and North Africa, Central and Eastern Europe, Western Europe, and the Commonwealth of Independent States. The study uses an unbalanced panel because data availability differs across countries, years, and variables, particularly for Chinese outward FDI, tourism indicators, and patent applications. No missing observations were replaced through mean substitution, linear interpolation, extrapolation, or model-based imputation. Instead, the original observed values were retained. For each econometric specification, only country-year observations with complete information for all required variables and their relevant lags were included. Consequently, the effective number of observations may differ across preliminary tests, full-sample estimations, regional models, robustness checks, and causality analyses. Retaining an unbalanced panel avoids introducing artificially constructed values and preserves the observed cross-country and temporal variation in the data. However, this approach may reduce statistical precision and create sample-selection concerns when missing observations are systematically concentrated in particular countries, regions, variables, or periods. Therefore, the findings should be interpreted with appropriate caution, particularly when comparing estimates across regional subsamples.

3.2 Variable definition and measurement

Carbon emissions are used as the dependent variable, while the remaining variables are included as explanatory variables. Table 2 presents the definitions, measurements, transformations, and data sources for all variables. The dependent variable is carbon emissions (LCO2), measured as the natural logarithm of carbon dioxide emissions in metric tons. This indicator is widely used as a proxy for environmental degradation and allows the estimated coefficients of logarithmically transformed explanatory variables to be interpreted as elasticities. Chinese outward FDI (LFDI) is measured as the natural logarithm of Chinese outward FDI stock in the host BRI countries, expressed in current US dollars. This measure captures investment originating specifically from China and therefore provides a more focused assessment of BRI-related investment effects than aggregate FDI inflows. Because Chinese outward FDI is often concentrated in infrastructure, energy, transport, and industrial-connectivity projects, its coefficient reflects the net environmental effect of these investment activities. Tourism development (LTD) is measured through a composite index constructed using principal component analysis (PCA). The index combines international tourist arrivals, tourism receipts, and tourism expenditure. PCA reduces the dimensionality of correlated tourism indicators and provides a single measure of tourism intensity. The index is standardized to permit comparisons across countries and years. As (LTD) is a composite index rather than a monetary or physical quantity, its estimated coefficient should be interpreted as the effect of a one-unit increase in tourism development on carbon emissions, holding other factors constant.

Table 2. Variables, definition and source.

WDI denotes world development indicators.

Technological innovation (LTI) is measured as the natural logarithm of the number of patent applications reported in the World Development Indicators. Patent applications provide a relatively consistent annual indicator of formal inventive activity across countries and therefore facilitate cross-country and intertemporal comparison. Nevertheless, they do not provide a comprehensive measure of an economy's overall technological and innovative capacity. Patent-based indicators may underestimate innovation in developing economies because many technological improvements occur through informal experimentation, adaptation of imported technologies, incremental process improvements, organizational changes, user innovation, and unregistered knowledge. Firms may also protect innovations through secrecy, lead-time advantages, informal know-how, trademarks, or utility models rather than formal patents. Moreover, patent counts do not account for differences in patent quality, technological significance, commercial value, successful implementation, or environmental relevance. Patenting behavior may also vary because of differences in registration costs, intellectual-property institutions, administrative capacity, industrial composition, and firm-level capabilities. Accordingly, LTI is interpreted as a proxy for formal patent-based inventive activity rather than as a complete measure of technological innovation.

Economic growth is captured by the natural logarithm of real GDP per capita in constant 2015 US dollars (LGDP). The squared term of logarithmic GDP per capita (LGDP²) is included to test the EKC hypothesis. An inverted U-shaped EKC relationship is supported when the coefficient of LGDP is positive and statistically significant, while the coefficient of LGDP² is negative and statistically significant. Industrial value added (IVA) is used to capture the level of industrialization in each economy. It is measured as industry, including construction, value added as a percentage of GDP, using data from the World Development Indicators. The indicator covers mining, manufacturing, construction, electricity, gas, water, and other industrial activities. IVA is entered into the empirical model in its original percentage form. Since the dependent variable is expressed in natural logarithms, the IVA coefficient represents the approximate percentage change in carbon emissions associated with a one-percentage-point increase in the industrial share of GDP, holding all other factors constant. A positive coefficient indicates that a greater industrial contribution to GDP is associated with higher carbon emissions.

Finally, the interaction term (LFDI × LTD) is constructed by multiplying LFDI and LTD. It captures whether tourism development moderates or amplifies the environmental effect of Chinese outward FDI. A positive and statistically significant interaction coefficient indicates that higher tourism development strengthens the emissions effect of Chinese outward FDI, whereas a negative coefficient indicates that tourism development moderates this effect.

3.3 Construction of tourism development index

Tourism development is measured using a composite index constructed through principal component analysis. The index combines three World Development Indicators: international tourist arrivals, international tourism receipts, and international tourism expenditure. These indicators capture complementary dimensions of tourism development. Tourist arrivals measure the physical scale of international tourism activity, whereas tourism receipts and expenditure capture its economic intensity. Because the three indicators are measured in different units and exhibit differences in scale, each indicator was converted into a standardized z-score before conducting PCA. The analysis was performed on the correlation matrix using pooled country-year observations for which complete data were available for all three tourism indicators. Standardization prevents indicators with larger numerical values from exerting disproportionate influence on the extracted component.

The suitability of the indicators for PCA was assessed using the Kaiser–Meyer–Olkin measure of sampling adequacy and Bartlett's test of sphericity. The KMO value was 0.742, exceeding the commonly accepted minimum value of 0.60. Bartlett's test was statistically significant, χ²(3) = 2,674.38, p < 0.001, rejecting the null hypothesis that the correlation matrix is an identity matrix. These results indicate that the indicators share sufficient common variation for principal component extraction. Appendix Table A1 presents the complete PCA results. The first principal component has an eigenvalue of 2.432 and explains 81.07% of the total variance. The second and third components have eigenvalues of 0.382 and 0.186 and explain 12.73% and 6.20% of the variance, respectively. Only the first component has an eigenvalue greater than one. Accordingly, the first principal component was retained based on the Kaiser criterion, the scree-plot pattern, and the high proportion of total variance explained. All three tourism indicators have strong and positive loadings on the retained component: 0.910 for international tourist arrivals, 0.903 for tourism receipts, and 0.887 for tourism expenditure. These loadings show that the component represents a common underlying dimension of tourism scale and economic activity rather than being dominated by a single indicator.

The unstandardised first-component score was constructed as follows:

PC1=0.584(ZARR)+0.579(ZREC)+ 0.569(ZEXP)

where ZARR denotes the standardized value of international tourist arrivals, ZREC denotes the standardized value of international tourism receipts, and ZEXP denotes the standardized value of international tourism expenditure. The resulting component score was subsequently standardized to have a mean of zero and a standard deviation of one and is denoted by LTD. Higher values of LTD indicate greater tourism development.

3.4 Model specification

To examine the environmental effects of Chinese outward FDI, tourism development, technological innovation, economic growth, and industrialization, the following baseline model is specified:

LCO2,it=αi+β1LFDIit+β2LTDit+β3(LFDIit×LTDit) +β4LTIit+β5LGDPit+β6LGDPit2+β7IVAit+εit(1)

Where (i) denotes country and (t) denotes year. LCO₂ represents carbon emissions; LFDI denotes Chinese outward FDI; LTD is the tourism development index; LTI represents technological innovation; LGDP and LGDP² capture economic growth and its non-linear effect; and IVA denotes industrial value added. The interaction term (LFDI × LTD) captures whether tourism development modifies the environmental effect of Chinese outward FDI. A positive interaction coefficient indicates that tourism development strengthens the emissions effect of Chinese outward FDI, whereas a negative coefficient suggests a mitigating effect. The EKC hypothesis is supported when the coefficient of LGDP is positive and the coefficient of LGDP² is negative.

3.5 Econometric methodology

3.5.1 Cross-sectional dependence and slope heterogeneity

The empirical analysis begins with tests for the existence of CSD and slope heterogeneity. The CSD occurs when there are common unobserved factors that affect several countries, such as unobserved global economic shocks, financial market crises, or oil price shocks. For the BRI economies, these interdependencies are particularly significant given the growing economic integration, connectivity and exposure to global markets. This study uses the Pesaran (2015) CSD test, considered a robust approach for testing for cross-sectional dependence, to detect the presence of CSD. A rejection of the null hypothesis of CSD suggests that standard first-generation panel estimators may not be suitable for use.

In addition to CSD, it is essential to examine slope heterogeneity, which refers to variations in the association between variables across countries. Given the diversity of BRI economies in terms of income levels, energy structures, and institutional quality, it is unlikely that a single set of coefficients can adequately capture these relationships. Therefore, the Pesaran et al. (2008) slope heterogeneity test is employed to assess whether slope coefficients differ significantly across cross-sectional units. Accounting for heterogeneity is crucial to avoid aggregation bias and to ensure that the estimated relationships reflect country-specific dynamics.

3.5.2 Unit root analysis

Once cross-sectional dependence has been detected, it is important to investigate the stationarity of the variables. To be able to use valid regression analysis in panel data, unit root analysis is imperative. Traditional first-generation unit root tests (e.g. Levin-Lin-Chu, Im-Pesaran-Shin) are invalid in the presence of cross-sectional dependence. To solve this issue, the second-generation panel unit root tests, which are the Cross-sectionally Augmented Dickey-Fuller (CADF) and Cross-sectionally Augmented IPS (CIPS) tests proposed by Pesaran, are used in this paper. By taking into consideration cross-sectional dependence, these tests are explicitly done through cross-sectional averages of the variables in the tests.

The general form of the CIPS test is expressed as:

CIPS=1N∑i=1NCADFi(2)

Where CADFi represents the cross-sectionally augmented Dickey-Fuller statistic for country i. These tests allow for heterogeneous autoregressive coefficients and provide a more reliable assessment of the integration order of the variables. Establishing the order of integration is particularly important in determining the appropriate econometric methodology, as the presence of a mixture of I(0) and I(1) variables justifies the use of autoregressive distributed lag (ARDL)-based approaches.

3.5.3 Panel cointegration analysis

The next question to be answered, after establishing the long run integration order of the variables is whether the carbon emission and its factors are cointegrated in the long run. When the variables will be non-stationary, the cointegration analysis will play a significant role as it will avoid spurious regression and determine the long-run relationship. Given the cross-sectional dependence and possible structural changes in the BRI economies, the panel cointegration test by Westerlund and Edgerton (2008), which takes such factors into consideration, is employed in the paper. The Westerlund method, in contrast to standard panel cointegration tests (such as the Pedroni or Kao tests) allows for heterogeneity among cross-sectional units and takes into account structural breaks that can occur as a result of policy and/or economic shocks, and/or large-scale projects such as the BRI. The fact that the null hypothesis of no cointegration is rejected, provides evidence of a long-run equilibrium relationship, and therefore the modeling of both the long-run and the short-run dynamics within the same framework.

3.5.4 Cross-sectionally augmented ARDL (CS-ARDL) model

Following the confirmation of cross-sectional dependence, slope heterogeneity, and mixed orders of integration among the variables, this study employs the Cross-Sectionally Augmented Autoregressive Distributed Lag (CS-ARDL) model. The CS-ARDL framework is appropriate because it allows the simultaneous estimation of short-run dynamics and long-run relationships in heterogeneous panel data while controlling for unobserved common factors through cross-sectional averages. This feature is particularly relevant for BRI economies, which are jointly exposed to common shocks related to global energy prices, trade conditions, investment flows, technological diffusion, and environmental policy changes.

The long-run relationship between carbon emissions and its determinants is specified as follows:

LCO2it=αi+β1iLFDIit+β2iLTDit+β3i(LFDIit×LTDit) +β4iLTIit+β5iLGDPit+β6iLGDP it2+β7iIVAit +μit(3)

where i denotes country and t denotes time. LCO₂ represents the natural logarithm of carbon emissions; LFDI denotes Chinese outward foreign direct investment; LTD represents the tourism development index; LTI denotes technological innovation; LGDP and LGDP² capture economic growth and its non-linear effect; and IVA represents industrial value added. The interaction term (LFDI × LTD) captures whether the environmental effect of Chinese outward FDI changes with the level of tourism development.

The general CS-ARDL specification is expressed as:

LCO2it=αi+Σ=11pϕi1LCO2i,t−1 +Σ=01qβ i′−1 Xi,t−1+Σ=01mγi−1LCO¯2,t−1+Σ=01m δi′X¯t−1+εit(4)

where X_it is the vector of explanatory variables, including LFDI, LTD, LFDI × LTD, LTI, LGDP, LGDP², and IVA. The terms LCO¯2,t− 1, and X¯t− 1 represent the cross-sectional averages of the dependent and explanatory variables, respectively. Their inclusion controls for unobserved common factors and cross-sectional dependence across countries. ε_*it* is the disturbance term.

For interpretation, the model is reparameterised into an error-correction form:

ΔLCO2it=αi+λi[LCO2i,t−1−θ1iLFDIi,t−1−θ2iLTDi,t−1 −θ3i(LFDIi,t−1×LTDi,t−1)−θ4iLTIi,t−1−θ5iLGDPi,t−1 −θ6iLGDP i,t−12−θ7iIVAi,t−1]+Σ1p1p−1 ψ1−1ΔLCO2i,t−1+Σ=01q−1κ1′1ΔXi,t−1+Σ=01mρ1′1ΔZ¯t 1+εit (5)

where λ_*i* is the error-correction coefficient, θ_*i* denotes the long-run parameters, and κil represents the short-run coefficients. A negative and statistically significant error-correction coefficient confirms convergence toward the long-run equilibrium after a short-run disturbance. Its absolute value indicates the speed at which disequilibrium is corrected in each period.

Given the annual frequency of the data and the relatively limited time dimension of the panel, a parsimonious lag-selection procedure was adopted. Alternative CS-ARDL specifications with a maximum of two annual lags were evaluated using both the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The evaluated specifications included CS-ARDL(1,0), CS-ARDL(1,1), CS-ARDL(1,2), CS-ARDL(2,0), CS-ARDL(2,1), and CS-ARDL(2,2). As reported in Appendix Table A2, the CS-ARDL(1,1) specification generated the lowest AIC value of −3.126 and the lowest BIC value of −2.944. It was therefore selected as the preferred specification. The model includes one lag of carbon emissions and one lag of each explanatory variable. This structure captures the persistence of carbon emissions, short-run adjustment, and delayed effects of the explanatory variables while avoiding excessive parameter proliferation and preserving observations in the relatively short annual panel.

Following the estimation of the preferred CS-ARDL(1,1) specification, post-estimation diagnostic tests were conducted to assess the adequacy of the dynamic structure and cross-sectional augmentation. The Pesaran residual cross-sectional-dependence test was used to determine whether the inclusion of cross-sectional averages adequately absorbed unobserved common shocks and interdependencies across the sampled BRI economies. Failure to reject the null hypothesis of residual cross-sectional independence indicates that the cross-sectional augmentation sufficiently controls for the common factors affecting the panel. The Wooldridge test for serial correlation in panel data was applied to examine whether first-order temporal dependence remained in the estimated residuals. Failure to reject the null hypothesis of no first-order residual serial correlation indicates that the selected lag structure adequately captures the dynamic dependence in the model. The sign, magnitude, and statistical significance of the error-correction coefficient were also examined. A negative and statistically significant coefficient confirms convergence toward the long-run equilibrium, while its absolute value indicates the proportion of disequilibrium corrected in each year.

3.5.5 Robustness estimation

In order to enhance the validity and robustness of the empirical results, the paper uses the Common Correlated Effects Mean Group (CCEMG) estimator developed by Pesaran. This approach is well suited to account for unobserved common effects that may affect all cross-sectional units. By including cross-sectional means in the regression, the CCEMG estimator helps to reduce the bias caused by omitted common factors and delivers consistent estimates even when there is heterogeneity and cross-sectional dependence. The robustness check with CCEMG estimation is complementary to the CS-ARDL estimation, as it checks whether the findings are robust to a different estimation approach. The robustness check across estimators adds to the credibility of the results.

3.5.6 Panel granger non-causality analysis

The Dumitrescu–Hurlin panel Granger non-causality test is employed to examine the temporal and predictive relationships between the variables. This procedure is suitable for heterogeneous panel data because it allows the autoregressive parameters and predictive relationships to vary across countries. The null hypothesis states that variable (X) does not Granger-cause variable (Y) in any cross-sectional unit, whereas rejection of the null indicates the presence of Granger causality for at least some, but not necessarily all, countries in the panel (Dumitrescu and Hurlin, 2012). In this study, Granger causality refers to predictive or temporal precedence rather than structural causality. A finding that (X) Granger-causes (Y) indicates that the lagged values of (X) contain statistically significant information for predicting (Y) beyond the information contained in the lagged values of (Y). It does not establish that an exogenous or policy-induced change in (X) directly causes a change in (Y) (Dumitrescu and Hurlin, 2012). The Dumitrescu–Hurlin procedure does not fully resolve endogeneity arising from simultaneity, reverse causality, omitted variables, measurement error, anticipation effects, common shocks, or non-random investment and policy decisions. The results may also be sensitive to the selected lag order, sample period, availability of consecutive observations, country composition, and statistical power. Accordingly, the test is used as complementary evidence of temporal direction and predictive relationships rather than as a substitute for a structural causal-identification strategy.

3.5.7 Scope and limitations of the econometric procedures

Although the selected econometric procedures address major concerns associated with heterogeneous and interdependent panel data, each test has limitations. The Pesaran cross-sectional dependence test determines whether the sampled countries are affected by common shocks and interdependencies, but it does not identify the specific sources or transmission mechanisms of such dependence (Pesaran, 2015). Similarly, the slope-heterogeneity test determines whether the estimated relationships differ across countries but does not explain which institutional, technological, regulatory, or structural characteristics generate those differences (Pesaran and Yamagata, 2008). The CADF and CIPS unit-root tests account for cross-sectional dependence, but their outcomes may remain sensitive to the selected lag structure, deterministic components, structural changes, sample length, and the unbalanced nature of the panel. Likewise, the cointegration procedure identifies whether a stable long-run relationship exists among the variables but does not establish the direction of causality. Although the Westerlund–Edgerton test accommodates cross-sectional dependence, heterogeneity, and possible structural shifts, its findings may depend on the assumed break structure and available time observations (Westerlund and Edgerton, 2008).

The CS-ARDL estimator accounts for short-run adjustment, long-run relationships, slope heterogeneity, and unobserved common factors through the inclusion of cross-sectional averages. Nevertheless, its reliability depends on the appropriateness of the lag structure, correct model specification, and the assumption that the regressors are at least weakly exogenous after controlling for the included dynamics and common factors. It cannot fully eliminate bias arising from omitted country-specific time-varying variables, measurement error, simultaneity, or non-random investment selection (Chudik and Pesaran, 2015). The CCEMG estimator provides a useful robustness check by accounting for heterogeneous coefficients and unobserved common factors. However, consistency between the CS-ARDL and CCEMG estimates strengthens the robustness of the reported relationships rather than establishing structural causality (Chudik and Pesaran, 2015). Accordingly, the CS-ARDL and CCEMG coefficients are interpreted as conditional short-run and long-run associations, while the Dumitrescu–Hurlin findings are interpreted as evidence of predictive direction. Establishing structural causal effects would require additional identification strategies and more detailed sectoral, project-level, institutional, or policy data.

4 Result and discussions

4.1 Descriptive statistics and correlation analysis

Table 3 reports the descriptive statistics for the variables used in the empirical analysis. The results show substantial variation in LCO₂, LFDI, LTI, LGDP, and IVA across BRI countries and over the study period. This variation reflects the diverse economic structures, levels of industrialization, energy systems, and development conditions within the sample. The tourism development index has a mean of zero because it is standardized during its construction, while its minimum and maximum values indicate considerable differences in tourism intensity across countries and years.

Table 3. Descriptive statistics.

*Source:* Authors' calculations.

Table 4 presents the pairwise correlations among the principal study variables. Because the empirical specification also contains an interaction term and a squared GDP term, pairwise correlations alone may not provide a complete assessment of multicollinearity. Variance inflation factors were therefore calculated for all explanatory variables, including LFDI × LTD and LGDP². As reported in Appendix Table A3, the VIF values range from 2.36 to 4.87, with a mean VIF of 3.49. All values remain below the conventional threshold of 10, while the corresponding tolerance values exceed 0.10. The results therefore indicate that multicollinearity is not sufficiently severe to undermine the stability of the estimated coefficients.

Table 4. Pairwise correlation matrix.

Source: Authors' calculations.

Figure 1 presents the aggregate trends in Chinese outward FDI, tourism development, and technological innovation across BRI countries from 2000 to 2024. All three indicators show a clear upward trajectory, with particularly rapid growth after 2015. The figure provides descriptive evidence of expanding investment connectivity, tourism activity, and innovation capacity in BRI economies, motivating the subsequent econometric analysis of their environmental effects.

Figure 2 presents the relative descriptive shares of Chinese outward FDI, tourism development, and technological innovation across the sampled BRI countries during 2000–2024. Chinese outward FDI represents the largest share (46.1%), followed by tourism development (33.8%) and technological innovation (20.1%). The figure provides a simple comparative overview of the three indicators; their effects on carbon emissions are examined through the subsequent econometric analysis.

4.2 Cross-sectional dependence and slope heterogeneity

Cross-sectional dependence (CSD) and slope heterogeneity are first examined to justify the selected econometric approach. Given the high degree of economic integration among BRI countries through trade, investment, and exposure to common global shocks, disturbances in one country may be transmitted to others.

Table 5 presents the results of the Pesaran and Yamagata (2008) slope heterogeneity test. Both the Delta statistic (14.229) and the adjusted Delta statistic (17.691) are statistically significant at the 1% level. Therefore, the null hypothesis of homogeneous slope coefficients is rejected. This finding indicates that the relationship between carbon emissions and its determinants is not uniform across the sampled BRI countries. Such heterogeneity is expected because BRI economies differ substantially in income levels, industrial structures, energy dependence, environmental regulations, technological capacity, and the sectoral composition of Chinese outward FDI. For example, Chinese investment may support carbon-intensive infrastructure and fossil-fuel projects in some countries, while in others it may facilitate cleaner transport, renewable energy, or technological upgrading. Similarly, the environmental effects of tourism, innovation, and industrial activity may vary according to local energy systems and policy conditions. Therefore, estimators that accommodate heterogeneous slope coefficients, such as the CS-ARDL model, are appropriate for the subsequent analysis (Pesaran and Yamagata, 2008; Chudik and Pesaran, 2015).

Table 5. Slope heterogeneity test.

^*** indicate statistical significance at the 1% level. *Source:* Authors' calculations.

Table 6 reports the Pesaran CSD test results. The null hypothesis of cross-sectional independence is rejected for all variables at the 1% significance level. This confirms that carbon emissions, Chinese outward FDI, tourism development, technological innovation, economic growth, and industrial value added are influenced by common shocks and interdependencies across BRI countries. This result is expected because BRI economies are connected through trade, investment, energy markets, tourism flows, regional infrastructure networks, and common exposure to global economic conditions. Therefore, changes in one country may affect other countries through spillover effects. The presence of CSD justifies the use of second-generation panel unit root tests and the CS-ARDL estimator, which explicitly account for unobserved common factors and cross-country dependence (Pesaran, 2015; Chudik and Pesaran, 2015).

Table 6. Cross-sectional dependence test.

^*** indicate statistical significance at the 1% level. *Source:* Authors' calculations.

4.3 Unit root tests

Before estimating the long-run and short-run relationships, the stationarity properties of the variables were examined using the second-generation Cross-sectionally Augmented Dickey–Fuller (CADF) and Cross-sectionally Augmented Im–Pesaran–Shin (CIPS) unit root tests. These tests are appropriate because they account for the cross-sectional dependence identified in the previous subsection. Table 7 reports the CADF and CIPS test results at level and first difference. The results indicate that LFDI and LTI are stationary at level and are therefore treated as I(0) variables. In contrast, LCO₂, LTD, LGDP) and IVA do not exhibit consistent stationarity at level but become stationary after first differencing. These variables are therefore treated as I(1). Overall, the findings confirm that the variables are integrated at a mixture of I(0) and I(1), with no evidence that any variable is integrated of order two. This integration pattern supports the application of the CS-ARDL approach, which is suitable for modeling dynamic relationships among variables with mixed orders of integration, provided that none is I(2).

Table 7. Second-generation panel unit root test results.

C denotes a model with an intercept, while C&T denotes a model with an intercept and trend. ^** and ^*** indicate statistical significance at the 5% and 1% levels, respectively. The symbol “—” indicates that first-difference results are not reported because the variable is stationary at level. *Source:* Authors' calculations.

4.4 Panel (PN) cointegration

After confirming that the variables are integrated at a mixture of [I(0)] and [I(1)], panel cointegration tests were conducted to examine whether carbon emissions, Chinese outward FDI, tourism development, technological innovation, economic growth, and industrial value added share a stable long-run relationship. Table 8 reports the Pedroni panel cointegration test results. Although the panel (v)-statistic and panel rho-statistic do not reject the null hypothesis of no cointegration, the panel PP-statistic, panel ADF-statistic, group PP-statistic, and group ADF-statistic are statistically significant at the 1% level. The majority of the test statistics therefore reject the null hypothesis of no cointegration. These findings provide evidence that the variables move together in the long run across the sampled BRI countries.

Table 8. Pedroni cointegration test results.

Null speculation: No co-integrations occur among factors. BRI area-wise (sub) sample also affirms the prolonged cointegrating relationships. Thus, not stated because of conciseness. *Source:* Authors' calculations.

Table 9 presents the Westerlund and Edgerton (2008) cointegration test results, which are particularly useful because they allow for cross-sectional dependence, heterogeneity, and potential structural shifts. The null hypothesis of no cointegration is rejected under the no-shift, level-shift, and regime-shift specifications. Both ZT(N) and Z/Ø (N) statistics are statistically significant, with significance levels ranging from 1% to 5%. This confirms the existence of a robust long-run equilibrium relationship even after allowing for possible structural changes in the sample period.

Table 9. Westerlund and Edgerton (2008) cointegration tests and structural breaks.

^** and ^*** indicate statistical significance at the 5% and 1% levels, respectively. *Source:* Authors' calculations.

According to the BRI's regional (sub) samples, there is a long-term cointegration relationship. While this was not reported in detail, it was done to maintain brevity. In 2005, the test assumed that there had been no shift in parameters, whereas each cross-sectional unit has its own distinct breakpoint for level shift and regime shift. Overall, the cointegration results from Tables 8, 9 support the estimation of both short-run dynamics and long-run coefficients using the CS-ARDL approach. The findings indicate that deviations from the long-run relationship between carbon emissions and its determinants are temporary and tend to adjust over time (Westerlund and Edgerton, 2008; Chudik and Pesaran, 2015).

4.5 Short-run and long-run CS-ARDL estimates

Table 10 presents the short-run CS-ARDL estimates for the full BRI sample and regional subsamples. The error-correction term [ECM(-1)] is negative and statistically significant in all specifications, ranging from −0.321 in Southeast Asia to −0.543 in MENA countries. This confirms that short-run deviations from the long-run equilibrium are temporary and converge over time. For the full BRI sample, the coefficient of −0.427 indicates that approximately 42.7% of the disequilibrium is corrected within 1 year. The relatively faster adjustment in MENA suggests a stronger tendency to return to the long-run emissions path following short-run shocks, whereas the lower adjustment speed in Southeast Asia implies a more gradual correction process.

Table 10. Short-run CS-ARDL estimation results.

^*, ^**, and ^*** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. *Source:* Authors' calculations.

For the full BRI sample, the short-run coefficient of Chinese outward FDI is −0.087. Because both carbon emissions and Chinese outward FDI are expressed in logarithms, this coefficient indicates that, at the average level of the standardized tourism index, a 1% increase in Chinese outward FDI is associated with an approximately 0.087% reduction in carbon emissions in the short run. This magnitude is modest but suggests that efficiency improvements, technology transfer, or the modernization of productive and infrastructure capital may begin to influence environmental performance shortly after investment occurs. However, this main coefficient must be interpreted together with the positive FDI–tourism interaction coefficient of 0.068. The short-run marginal effect of Chinese outward FDI is:

∂LCO2/∂LFDI=-0.087+0.068LTD(6)

Since the tourism index is standardized with a mean of zero, the coefficient of −0.087 represents the FDI effect at the average level of tourism development. At one unit above the average tourism level, the estimated FDI effect becomes −0.019%, calculated as −0.087 + 0.068. Thus, increasing tourism intensity substantially weakens the short-run emissions-reducing effect of Chinese outward FDI. Based on the point estimates, the effect becomes positive when the tourism index exceeds approximately 1.28 units above its mean. This calculation demonstrates that the environmental effect of Chinese investment is conditional on the tourism intensity of the host economy.

The regional estimates further demonstrate this conditionality. Chinese outward FDI increases short-run emissions in South Asia and MENA, where the coefficients are 0.167 and 0.219, respectively. The corresponding positive interaction coefficients of 0.113 and 0.175 indicate that tourism development strengthens these emissions-increasing effects. In contrast, Chinese outward FDI has negative short-run coefficients in Southeast Asia, Central Asia, and Western Europe. In these regions, the positive interaction term does not mean that tourism directly makes FDI environmentally beneficial. Rather, it indicates that higher tourism development reduces the magnitude of the environmental benefit associated with Chinese outward FDI. Several mechanisms may explain the negative FDI coefficients. Chinese investment may replace older and less efficient capital, facilitate the adoption of cleaner production technologies, improve logistics efficiency, support renewable-energy capacity, or finance relatively efficient transport and urban infrastructure. These effects may be stronger in economies with greater technological absorptive capacity, cleaner energy systems, stricter environmental requirements, and more effective project screening. However, these mechanisms should be interpreted cautiously. The study uses aggregate Chinese outward FDI stock and does not directly observe the sectoral distribution, energy source, technological quality, or environmental standards of individual projects. Therefore, the negative coefficients support a net Pollution Halo interpretation but do not establish which particular investment channel is responsible for the reduction in emissions.

Tourism development has a positive and statistically significant short-run relationship with carbon emissions in the full BRI sample, South Asia, Central and Eastern Europe, MENA, and CIS countries. This finding indicates that tourism expansion increases energy demand through air and road transport, accommodation facilities, restaurants, recreational services, and tourism-related infrastructure. The absence of significance in some regional groups suggests that the emissions consequences of tourism depend on differences in transport systems, energy efficiency, tourism intensity, and environmental policies. The result is consistent with studies showing that tourism can increase emissions when it relies on energy-intensive mobility and conventional infrastructure (Churchill et al., 2022; Shan et al., 2023; Razzaq et al., 2023).

Technological innovation has a negative and significant effect on emissions in the full sample, South Asia, MENA, Western Europe, and CIS countries. This result suggests that innovation contributes to lower carbon intensity through energy-efficient technologies, cleaner production processes, improved resource management, and the adoption of low-carbon infrastructure. The insignificant coefficients in some regions may indicate that innovation capacity alone is insufficient unless it is effectively diffused into production, transport, and energy systems. This finding is consistent with the argument that technological progress can improve environmental performance when supported by institutional capacity and technology adoption mechanisms (Tan et al., 2024; Si et al., 2025). Economic growth has a positive and significant short-run effect on carbon emissions in all regional groups, indicating that higher output and income levels initially increase energy demand, production activity, transport use, and consumption-related emissions. The squared GDP term is negative and significant in the full sample, Southeast Asia, Central and Eastern Europe, and Western Europe, providing preliminary evidence of a non-linear growth–environment relationship. This pattern is consistent with the EKC argument that emissions may rise during early stages of development but decline when economies achieve higher efficiency, cleaner production structures, and stronger environmental regulation (Grossman and Krueger, 1991; Zhou et al., 2022). Industrial value added has a positive and significant effect on carbon emissions across all samples. This confirms that industrial expansion remains an important source of environmental pressure in BRI economies, particularly where manufacturing, construction, heavy industry, and energy generation are dependent on fossil fuels. The result emphasizes the importance of industrial upgrading, energy efficiency, renewable-energy adoption, and cleaner production standards in reducing the environmental cost of industrial development.

Table 11 presents the long-run CS-ARDL estimates for the full BRI sample and regional subsamples. The long-run results provide stronger evidence that the environmental effect of Chinese outward FDI is conditional on tourism development and regional circumstances. For the full BRI sample, the long-run Chinese outward FDI coefficient is −0.245. At the average level of the standardized tourism index, a 1% increase in Chinese outward FDI is therefore associated with an approximately 0.245% reduction in carbon emissions. The magnitude is greater than the corresponding short-run estimate of −0.087%, suggesting that technology transfer, capital modernization, infrastructure efficiency, and production restructuring may require time before their full environmental benefits materialize. The positive interaction coefficient of 0.089 shows that tourism development progressively weakens this emissions-reducing effect. The long-run marginal effect of Chinese outward FDI is expressed as:

∂LCO2/∂LFDI=-0.245+0.089LTD(7)

Table 11. Long-run CS-ARDL estimation results.

^*, ^**, and ^*** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. *Source:* Authors' calculations.

At the mean tourism level, the estimated effect is −0.245%. At one unit above the mean, it decreases in absolute magnitude to −0.156%, while at two units above the mean it falls further to −0.067%. Based on the point estimates, the net effect becomes positive when the tourism index exceeds approximately 2.75 units above its mean. The result therefore does not imply that Chinese outward FDI is environmentally beneficial under all tourism conditions. Rather, the environmental benefit diminishes as tourism intensity increases and may be reversed in highly tourism-intensive economies. The regional coefficients provide additional insight. Chinese outward FDI increases long-run emissions in South Asia and MENA, with coefficients of 0.268 and 0.273, respectively. In these regions, the interaction coefficients are also positive, indicating that tourism expansion reinforces an already emissions-increasing investment effect. These results may reflect the concentration of investment in construction, conventional energy, road transport, airports, urban development, and other infrastructure that operates within carbon-intensive energy systems.

The regional results demonstrate that the environmental effect of Chinese outward FDI is not uniform across BRI economies. Chinese outward FDI has a negative and statistically significant long-run association with carbon emissions in Southeast Asia, Central Asia, Central and Eastern Europe, Western Europe, and the Commonwealth of Independent States, with estimated coefficients of −0.238, −0.183, −0.239, −0.226, and −0.136, respectively. These differences indicate that the environmental consequences of Chinese investment depend on the sectoral destination of investment, the condition of the existing capital stock, the host country's energy structure, environmental regulation, infrastructure quality, and technological absorptive capacity. In Southeast Asia, the negative coefficient may reflect Chinese investment in renewable-energy generation, railway systems, efficient logistics, electric mobility, digital infrastructure, and modern industrial equipment. Such investments may reduce emissions by improving energy efficiency, supporting cleaner electricity generation, lowering the carbon intensity of transportation and production, and replacing relatively inefficient domestic capital.

In Central Asia and CIS economies, existing energy, transport, and industrial infrastructure is often relatively old and inefficient. Chinese investment in modern power systems, railways, logistics networks, transport corridors, and industrial equipment may therefore generate efficiency gains relative to the aging and more carbon-intensive capital it replaces. Even where the regional energy mix remains dependent on fossil fuels, infrastructure modernization may reduce energy consumption and emissions per unit of production or transportation. The negative coefficients in Central and Eastern Europe and Western Europe may be associated with stronger environmental regulations, more rigorous project-screening requirements, energy-efficiency standards, skilled labor, efficient electricity networks, and greater technological absorptive capacity. These conditions may increase the likelihood that Chinese outward FDI transfers cleaner technologies, adopts modern production processes, and complies with relatively stringent environmental requirements. Chinese investment may therefore be more likely to generate Pollution Halo effects in these regions. However, these interpretations represent plausible regional mechanisms rather than directly estimated project-level channels. The Chinese outward FDI variable measures aggregate investment stock and does not distinguish renewable-energy projects from fossil-fuel investments or separately identify transport, construction, manufacturing, mining, tourism, and digital-infrastructure projects. Therefore, the negative regional coefficients represent average conditional associations and should not be interpreted as evidence that every Chinese-funded project is environmentally beneficial.

The regional FDI coefficients must also be interpreted jointly with the interaction between Chinese outward FDI and tourism development. Because the tourism index is standardized with a mean of zero, each standalone FDI coefficient represents the estimated effect of Chinese outward FDI at the average level of tourism development. The positive FDI–tourism interaction coefficients indicate that tourism growth reduces the magnitude of the negative FDI effect. As tourism intensity increases, investment-supported expansion in airports, roads, hotels, urban facilities, passenger mobility, construction, and tourism services raises energy demand and may offset part of the emissions reductions associated with technological upgrading and more efficient infrastructure. Thus, the regional findings support a conditional Pollution Halo interpretation rather than the conclusion that Chinese outward FDI is universally emissions-reducing.

The positive interaction operates through a connected infrastructure–energy mechanism. Chinese outward FDI may finance roads, railways, airports, ports, hotels, electricity networks, communication facilities, and urban services that expand the capacity and accessibility of tourism destinations. Infrastructure is therefore the enabling channel through which investment facilitates tourism growth. Once tourism activity expands, greater passenger mobility, hotel occupancy, construction, food services, recreational activity, and electricity consumption increase energy demand. Energy consumption is thus the more immediate channel through which infrastructure-supported tourism affects carbon emissions. The interaction coefficient captures the combined net effect of these processes; it does not empirically separate emissions arising from infrastructure construction from those generated by the subsequent operation of tourism transport, accommodation, and services.

Tourism development has a positive and statistically significant long-run effect on carbon emissions in the full sample and all regional groups. This finding indicates that sustained tourism expansion increases environmental pressure through higher demand for transport, accommodation, food services, recreational facilities, and tourism-supporting infrastructure. Since the tourism variable is a PCA-based composite index, the coefficients indicate the effect of a one-unit increase in tourism development rather than an elasticity. The strongest tourism-related emission effects are found in Western Europe and MENA countries, whereas Central Asia records the smallest positive coefficient. These differences may reflect variation in tourism intensity, transport modes, energy efficiency, and the carbon content of tourism infrastructure. The result is consistent with evidence that tourism can increase emissions when expansion depends on energy-intensive mobility and conventional accommodation systems (Churchill et al., 2022; Shan et al., 2023; Razzaq et al., 2023).

Technological innovation has a negative and statistically significant coefficient in all specifications, confirming its long-run role in reducing carbon emissions. The results suggest that innovation supports energy efficiency, cleaner production, renewable-energy adoption, improved resource management, and lower carbon intensity. The largest emissions-reducing effect is reported for Central and Eastern Europe, followed by Central Asia and CIS countries. These findings support the view that technological upgrading is an important channel through which BRI economies can pursue growth while reducing environmental pressure (Tan et al., 2024; Si et al., 2025). The coefficients of GDP per capita are positive and statistically significant, whereas the squared GDP per capita coefficients are negative and statistically significant across the full sample and all regional groups. This sign pattern provides long-run support for the EKC hypothesis. It suggests that carbon emissions initially rise with income growth because of scale effects, industrial expansion, and increased energy demand; however, after a certain stage of development, cleaner technologies, structural transformation, and stronger environmental regulation begin to reduce emissions.

The regional differences in the estimated coefficients indicate that the timing and strength of this transition vary across BRI regions. Industrial value added has a positive and statistically significant effect on carbon emissions in the full sample and every regional subsample. This result confirms that industrial expansion remains a major long-run source of environmental pressure in BRI economies. It highlights the importance of cleaner industrial technologies, energy-efficiency standards, renewable-energy integration, and the gradual transition away from carbon-intensive production structures.

In Table 12, the EKC turning point is derived from the estimated long-run relationship between carbon emissions and GDP per capita. The EKC hypothesis is supported when the coefficient of logarithmic GDP per capita is positive and statistically significant, while the coefficient of its squared term is negative and statistically significant. Under these conditions, carbon emissions initially increase with income and decline after the estimated turning point.

Table 12. Environmental kuznets curve (EKC) turning points from long-run CS-ARDL estimates.

^** and ^*** indicate statistical significance at the 5% and 1% levels, respectively. *Source:* Authors' calculations based on the long-run CS-ARDL estimates.

The logarithmic turning point is calculated as:

LGDP*=-βLGDP/(2βLGDP2)(8)

where βLGDP is the estimated long-run coefficient of logarithmic GDP per capita and βLGDP² is the estimated long-run coefficient of squared logarithmic GDP per capita. The turning points reported below are expressed in logarithmic terms because GDP per capita enters the CS-ARDL specification in logarithmic form.

The EKC hypothesis is supported in the full BRI sample and all regional groups because the coefficient of GDP per capita is positive and statistically significant, whereas the coefficient of squared GDP per capita is negative and statistically significant. This indicates an inverted U-shaped relationship between economic growth and carbon emissions. In the initial stage of development, income growth increases carbon emissions through higher production, energy consumption, transport demand, and industrial activity. After the estimated turning point, further economic development is associated with lower emissions because of technological upgrading, energy efficiency, structural transformation, and stronger environmental regulation.

The estimated logarithmic turning points vary across regions, ranging from 0.861 in Western Europe to 1.878 for the full BRI sample. The relatively lower turning point in Western Europe may reflect its earlier transition toward cleaner technologies, energy-efficient production, and stricter environmental standards. In contrast, the comparatively higher turning points for the full BRI sample, MENA, and Central and Eastern Europe indicate that the environmental benefits of economic growth may arise later in these groups. These differences highlight the importance of regional energy structures, industrial composition, technological capacity, and environmental governance in shaping the growth–environment relationship.

4.6 Post-estimation diagnostic results

Table 13 reports the post-estimation diagnostic results for the preferred CS-ARDL(1,1) model. The Pesaran residual cross-sectional-dependence statistic is 1.214, with a p-value of 0.225. Therefore, the null hypothesis of no residual cross-sectional dependence cannot be rejected. This finding indicates that the cross-sectional averages included in the augmented specification adequately capture the common shocks and interdependencies affecting the sampled BRI economies. The Wooldridge test for residual serial correlation produces a statistic of 1.536 and a p-value of 0.219. The statistically insignificant result indicates that the null hypothesis of no first-order residual serial correlation cannot be rejected. Therefore, there is no evidence of remaining temporal dependence in the residuals, and the selected one-year lag structure appears to capture the relevant short-run dynamics adequately.

Table 13. Post-estimation diagnostic results for the CS-ARDL(1,1) model.

^*** indicates statistical significance at the 1% level. Source: Authors' calculations.

The error-correction coefficient is −0.427 and statistically significant at the 1% level. Its negative sign confirms convergence toward the long-run equilibrium following a short-run disturbance. The magnitude indicates that approximately 42.7% of the disequilibrium between observed carbon emissions and their estimated long-run level is corrected within one year. Overall, the diagnostic findings support the adequacy of the cross-sectional augmentation, the CS-ARDL(1,1) lag structure, and the error-correction representation.

4.7 Robustness results

To assess the robustness of the baseline CS-ARDL findings, Table 14 reports estimates obtained using the Common Correlated Effects Mean Group (CCEMG) estimator. The CCEMG approach accounts for unobserved common factors and cross-sectional dependence while allowing slope coefficients to differ across countries. The estimated signs and significance levels are broadly consistent with the long-run CS-ARDL results, supporting the reliability of the main findings. Chinese outward FDI has a negative and significant effect on carbon emissions, indicating that Chinese investment is, on average, associated with lower emissions in the full BRI sample. Tourism development has a positive and highly significant coefficient, confirming that tourism expansion increases environmental pressure through higher transport, accommodation, and service-related energy demand. The positive and significant interaction term between Chinese outward FDI and tourism development further indicates that tourism intensity amplifies the emissions effect associated with FDI-supported development.

Table 14. Robustness results: CCEMG estimation.

^** and ^*** indicate statistical significance at the 5% and 1% levels, respectively. *Source:* Authors' calculations.

Technological innovation has a negative and significant coefficient, reaffirming its role in reducing emissions through energy efficiency, cleaner production, and technological upgrading. GDP per capita is positive and significant, whereas its squared term is negative and significant, providing additional support for the inverted U-shaped EKC relationship. Finally, industrial value added exerts a positive and significant effect on carbon emissions, showing that industrial expansion remains carbon-intensive across BRI economies. Overall, the CCEMG results confirm the main CS-ARDL conclusions. The consistency of findings across estimators strengthens confidence in the reported long-run relationships.

4.8 Panel granger non-causality results

Table 15 presents the Dumitrescu–Hurlin panel Granger non-causality results. Rejection of the null hypothesis indicates that the lagged values of the variable listed before the arrow contain statistically significant information for predicting the variable listed after the arrow for at least some sampled countries (Dumitrescu and Hurlin, 2012). The null hypothesis of no Granger causality from Chinese outward FDI to carbon emissions is rejected. This finding indicates that past changes in Chinese outward FDI help predict subsequent changes in carbon emissions. However, it does not establish that an exogenous increase in Chinese outward FDI structurally causes emissions to rise or decline. Chinese investment decisions may themselves be influenced by infrastructure demand, economic growth, energy availability, environmental regulations, institutional conditions, and other host-country characteristics. Tourism development also Granger-causes carbon emissions, indicating that its lagged values provide useful information for predicting subsequent environmental pressure. This result is consistent with the environmental effects associated with tourism-related transport, accommodation, construction, services, and energy consumption. Nevertheless, the test does not independently identify or isolate these transmission mechanisms.

Table 15. Dumitrescu–Hurlin Panel Granger non-causality test results.

^** and ^*** indicate statistical significance at the 5% and 1% levels, respectively. *Source:* Authors' calculations.

Technological innovation Granger-causes carbon emissions, suggesting that past patent-based innovation activity helps predict subsequent changes in emissions. Although this finding is consistent with the emissions-reducing innovation coefficient reported in the main estimations, it does not demonstrate that patent applications directly cause emission reductions. The relationship may also reflect technological adoption, cleaner production, industrial upgrading, environmental regulation, or other associated developments. GDP per capita and industrial value added also Granger-cause carbon emissions, indicating that their lagged values contain useful information for predicting subsequent emissions. These findings are consistent with the scale, composition, and industrialization effects identified in the CS-ARDL estimates.

The reverse relationships from carbon emissions to the explanatory variables are statistically insignificant. However, failure to reject reverse Granger non-causality should not be interpreted as definitive evidence that feedback effects are absent. The findings may depend on the selected lag structure, sample period, available observations, country composition, and statistical power. Therefore, the results provide evidence of predictive direction rather than structural causal effects (Dumitrescu and Hurlin, 2012).

5 Conclusion and policy implications

5.1 Conclusion

This study examined the short-run and long-run environmental effects of Chinese outward FDI, tourism development, technological innovation, and the interaction between Chinese outward FDI and tourism development in 60 Belt and Road Initiative (BRI) countries over the period 2000–2024. By distinguishing Chinese outward FDI from aggregate FDI, incorporating the FDI–tourism interaction, and conducting regional analysis, the study provides a more focused assessment of how investment-led connectivity and tourism expansion jointly shape carbon-emission outcomes across BRI economies. The preliminary diagnostics confirmed the presence of cross-sectional dependence and slope heterogeneity, indicating that BRI countries are influenced by common shocks while the effects of the explanatory variables differ across countries. The unit-root and cointegration results further supported the use of the CS-ARDL approach by confirming a mixture of [I(0)] and [I(1)] variables and a stable long-run relationship among carbon emissions and their determinants.

The short-run estimates show that Chinese outward FDI reduces carbon emissions for the full BRI sample, although this effect varies substantially across regions. In South Asia and the Middle East and North Africa, Chinese outward FDI increases emissions, whereas its effect is negative in several other regions. Tourism development generally increases short-run emissions, particularly in regions where tourism depends heavily on transport, accommodation, and conventional energy-intensive infrastructure. The positive FDI–tourism interaction indicates that tourism growth amplifies the short-run environmental pressure associated with investment-led development. Technological innovation reduces emissions in the full sample and several regional groups. The negative and significant error-correction terms confirm that short-run deviations converge toward long-run equilibrium.

The long-run findings reinforce these conclusions. Chinese outward FDI and technological innovation reduce carbon emissions in the full BRI sample, supporting the view that investment can generate environmental gains when it is associated with technology transfer, efficiency improvement, and cleaner infrastructure. However, the positive long-run effects of tourism development and the FDI–tourism interaction show that tourism-intensive and infrastructure-led expansion can create persistent environmental pressure when it relies on carbon-intensive transport, construction, and energy systems. The regional results highlight that Chinese outward FDI is not environmentally uniform: it increases emissions in South Asia and the Middle East and North Africa, but reduces emissions in Southeast Asia, Central Asia, Central and Eastern Europe, Western Europe, and the Commonwealth of Independent States countries.

The findings also provide support for the EKC hypothesis in the full sample and regional subsamples. However, this result should be interpreted as one component of the broader findings rather than the main contribution of the study. The principal contribution lies in demonstrating that the environmental implications of Chinese outward FDI depend on tourism intensity, technological capacity, industrial structure, energy systems, and regional conditions. Overall, the study advances the literature by showing that Chinese outward FDI cannot be treated as a uniformly clean or polluting source of capital. Its environmental effect is conditional on the form of investment and the host-country context. Likewise, tourism expansion is not inherently sustainable unless supported by low-carbon mobility, energy-efficient accommodation, and effective environmental governance. The consistency of the CS-ARDL results with the CCEMG robustness estimates and the causality findings strengthens confidence in these conclusions.

5.2 Policy implications

The empirical magnitudes indicate that environmental policies should focus not only on the volume of Chinese outward FDI but also on its sectoral composition and interaction with tourism development. For the full BRI sample, the estimated long-run FDI effect is −0.245 at the average tourism level. However, this effect weakens to −0.156 when tourism development increases by one unit and to only −0.067 when it increases by two units. Therefore, governments should not assume that infrastructure-oriented FDI will remain environmentally beneficial as tourism intensity expands. Tourism-linked investment approvals should be conditional on measurable low-carbon requirements. Airport, road, railway, hotel, urban-development, and recreational projects should incorporate renewable-energy supply, energy-efficient building codes, low-emission public transport, rail-based mobility, waste-management systems, and carbon-reporting obligations. Since infrastructure expansion enables tourism growth while the resulting transport, accommodation, and service activities increase energy demand, environmental assessments should cover both construction-stage and operational emissions. These measures are particularly important where the tourism index approaches the level at which the net environmental effect of FDI becomes positive.

Regional differentiation is also necessary. The positive FDI coefficients of 0.268 in South Asia and 0.273 in MENA indicate that Chinese investment in these regions is associated with higher long-run carbon emissions. Governments in these regions should apply stricter project-level carbon-intensity thresholds, mandatory environmental-impact assessments, green-finance eligibility criteria, energy-efficiency standards, and post-investment emissions monitoring. Projects involving fossil-fuel generation, carbon-intensive construction, road-dominated transport, and heavy industry should face stronger screening and mitigation requirements. In contrast, regions where FDI coefficients are negative should preserve these benefits by expanding renewable energy, clean transport, efficient logistics, and environmental-technology transfer rather than assuming that all future investment will automatically remain emissions-reducing. The technological-innovation coefficient of −0.237 indicates that innovation represents a quantitatively important emissions-reduction channel. BRI governments should therefore complement support for patent creation with policies that facilitate the commercial adoption and diffusion of cleaner technologies. Appropriate measures include green research grants, concessional finance for energy-efficient equipment, tax incentives for low-carbon production, university–industry research partnerships, technology-transfer requirements in foreign-investment agreements, and technical assistance for domestic firms. Particular emphasis should be placed on the conversion of formal innovation into operational reductions in energy use and carbon intensity.

The positive industrial-value-added coefficient of 0.536 identifies industrial structure as a major policy priority. Industrial policy should promote energy audits, mandatory efficiency standards, electrification of production, renewable-energy procurement, cleaner manufacturing, circular production systems, and the gradual retirement of obsolete carbon-intensive capital. Industrial expansion should be evaluated against emissions intensity and energy efficiency rather than only its contribution to GDP. Finally, the estimated adjustment coefficient indicates that approximately 42.7% of disequilibrium is corrected annually. Policy evaluation should therefore use multi-year monitoring frameworks rather than expecting immediate environmental adjustments. Investment and tourism projects should be subject to annual emissions reporting and periodic reviews that assess whether their actual environmental outcomes are consistent with the conditions under which they were approved.

5.3 Limitations and future research directions

This study has several limitations that also provide important directions for future research. First, the analysis uses an unbalanced panel because data availability differs across countries, years, and variables, particularly for Chinese outward FDI, international tourism indicators, and patent applications. No missing observations were replaced through mean substitution, interpolation, extrapolation, or model-based imputation. Although retaining the observed data avoids introducing artificial values, systematic missingness may affect the composition of the estimation sample, reduce statistical precision, and limit the comparability of regional estimates. Future research could conduct sensitivity analyses using balanced subsamples, alternative sample periods, different country groupings, and carefully justified missing-data procedures. Second, Chinese outward FDI is measured using aggregate investment stock. The indicator does not distinguish investment in renewable energy, fossil-fuel energy, transport, construction, manufacturing, mining, tourism, digital infrastructure, or other sectors. It also does not differentiate greenfield projects, mergers and acquisitions, reinvested earnings, and other forms of investment. Consequently, the present study cannot directly identify which sectors or project types explain the positive FDI coefficients in South Asia and MENA or the negative coefficients in Southeast Asia, Central Asia, Central and Eastern Europe, Western Europe, and CIS economies. Future research should use sectoral and project-level Chinese investment data to distinguish environmentally beneficial investment from carbon-intensive projects and to identify the specific mechanisms underlying regional heterogeneity.

Third, technological innovation is measured through patent applications. This indicator captures formal inventive activity but may underestimate informal, incremental, adaptive, organizational, process-based, and unregistered innovation, particularly in developing economies. Patent counts also do not fully reflect differences in technological quality, commercial value, successful implementation, diffusion, or environmental relevance. Future studies could employ green patents, patent citations, research and development expenditure, technology imports, digital adoption, scientific publications, environmental-technology exports, and other indicators of innovation quality and diffusion. Fourth, tourism development is measured through a PCA-based index combining international tourist arrivals, tourism receipts, and tourism expenditure. Although the index captures the common scale and economic intensity of tourism activity, it does not separately identify emissions arising from aviation, road transport, accommodation, construction, food services, recreational activities, and supporting infrastructure. Similarly, the positive interaction between Chinese outward FDI and tourism development represents the combined infrastructure–energy mechanism and does not independently quantify construction-stage emissions and the operational emissions of tourism services. Future research could decompose tourism development into its main components and formally examine whether transport intensity, aviation activity, hotel capacity, energy consumption, or infrastructure construction mediates the FDI–emissions relationship.

Fifth, the regional classifications used in this study may conceal substantial differences among countries within the same BRI region. Countries grouped within a single region may differ considerably in income, energy dependence, environmental regulation, institutional quality, industrial structure, tourism intensity, and technological absorptive capacity. Future research could employ country-specific estimations, panel-quantile regressions, threshold models, spatial econometric approaches, and classifications based on income, energy composition, institutional quality, or environmental performance. Such approaches may reveal nonlinearities, distributional differences, and cross-border environmental spillovers that cannot be fully captured through broad regional averages. Sixth, industrial value added is measured as industry, including construction, value added as a percentage of GDP. This broad indicator does not distinguish manufacturing, mining, construction, electricity, gas, and water-related activities. As a result, it cannot identify which industrial subsectors are primarily responsible for the estimated positive effect on carbon emissions. Future studies should use disaggregated industrial data to separate the environmental effects of manufacturing, extractive industries, construction, utilities, and energy-intensive production.

Finally, the CS-ARDL and CCEMG estimators account for cross-sectional dependence, heterogeneous coefficients, dynamic adjustment, and unobserved common factors. Nevertheless, these procedures cannot fully eliminate endogeneity caused by omitted country-specific time-varying factors, simultaneity, measurement error, anticipation effects, and non-random investment allocation. Their estimates should therefore be interpreted as conditional short-run and long-run associations rather than definitive structural causal effects. Similarly, the Dumitrescu–Hurlin panel Granger non-causality test identifies predictive and temporal precedence rather than structural causality (Dumitrescu and Hurlin, 2012). Future research could strengthen causal identification by applying valid instrumental-variable strategies, natural experiments, difference-in-differences designs based on project implementation, policy discontinuities, synthetic-control methods, and detailed project-level panel data containing credible sources of exogenous variation. Future studies may also extend the environmental scope beyond territorial carbon dioxide emissions by examining consumption-based carbon emissions, methane, nitrous oxide, ecological footprints, local air pollution, land degradation, water stress, and biodiversity outcomes. Such extensions would provide a broader understanding of the environmental consequences of Chinese outward FDI, tourism development, technological change, and industrial expansion across BRI economies.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: David Anderson, Center College, United States

Reviewed by: Ali Aljofan, University of Science Malaysia (USM), Malaysia

Zhaoyang Zhao, Sichuan University, China