Cooperative Innovation Under the “Belt and Road Initiative” for Reducing Carbon Emissions: An Estimation Based on the Spatial Difference-in-Differences Model

Abstract

The Belt and Road Initiative holds significant importance for achieving the United Nations’ Sustainable Development Goals, particularly Goals 9 and 17. Drawing on data from the Web of Science database, the BRI database, and the World Bank database, this study explores the potential carbon emission reduction effects that cooperative innovations may bring to participating countries under the Belt and Road Initiative. The study constructs variable endogenous spatio-temporal weight matrices based on initial spatial weight matrices and, drawing on trends in co-authored publications, innovatively establishes time dummy variables and event dummy variables in a difference-in-differences (DID) model. This approach fully considers the interconnected, shared model of the Belt and Road Initiative and the spatial effects of policy implementation. A spatial DID model was established for 106 BRI participating countries and regions from 2005 to 2021. The results reveal the following: first, cooperative innovation under the BRI significantly reduces carbon emission intensity in participating countries. Second, the BRI primarily achieves carbon reduction through investment, innovation, and trade mechanisms. Third, the results of the global SDID model indicate that cooperative innovation with China negatively impacts carbon emission intensity. Also, this effect exhibits spatial spillover, suggesting that there is a policy spillover effect. Fourth, the decomposition of local policy shock effects indicates that the influences of cooperative innovation exhibit spatial heterogeneity, with varying degrees of direct and indirect effects on carbon emission intensity across different countries.

1. Introduction

The Belt and Road Initiative (BRI), launched in 2013, is a comprehensive policy that integrates political, economic, and cultural tools. By March 2024, China had signed over 200 cooperative agreements with 152 countries and 32 international organizations to jointly build the BRI. This initiative aims to foster “policy coordination, infrastructure connectivity, unimpeded trade, financial integration, and closer people-to-people ties”. However, most BRI countries are developing nations undergoing economic transformation, characterized by “high growth, high energy consumption, and high carbon emissions”. Establishing a green BRI through investment and trade cooperation presents a significant challenge in advancing the initiative. Therefore, the Chinese government released the white paper “Vision and Actions on Jointly Building the Silk Road Economic Belt and 21st-Century Maritime Silk Road” in 2015, introducing the Green Silk Road initiative. This initiative emphasizes “strengthening green and low-carbon infrastructure construction and operation management, fully considering the impact of climate change in construction”, and “highlighting the concept of ecological civilization in investment and trade, enhancing cooperation in ecological environment, biodiversity, and climate change mitigation”. The Sustainable Development Goals (SDGs) adopted by the United Nations in 2015 emphasize “building resilient infrastructure, promoting inclusive and sustainable industrialization, and fostering innovation” (Goal 9) as well as “strengthening the means of implementation and revitalizing the Global Partnership for Sustainable Development” (Goal 17). The BRI promotes international innovation cooperation, contributing to sustainable industrialization by advancing green technological progress, upgrading industrial structures, and improving infrastructure in participating countries, thereby collectively achieving carbon emission reduction targets. Furthermore, as the majority of BRI participant countries are classified as Less Developed Countries (LDCs), the initiative leverages measures such as innovation cooperation, investment, trade, technology transfer, and capacity-building collaborations to enhance their economic development levels, mitigate potential debt risks, and ultimately achieve sustainable development. What is the effectiveness of the Green Belt and Road Initiative? Has the BRI truly contributed to the green development of participating countries? The sustainability of the Belt and Road Initiative has become a focal point for both academia and policymakers.

The current research presents both positive and negative perspectives regarding the impact of the “Belt and Road Initiative” (BRI) on carbon emissions in participating countries. The majority of scholars argue that the implementation of the BRI helps reduce carbon emissions in these countries. Existing literature primarily follows two approaches: most researchers focus on BRI countries as the study area, analyzing how factors such as innovation [1], outward foreign direct investment (OFDI) [2], and trade openness [3,4] affect carbon emissions. Additionally, some studies employ the difference-in-differences (DID) model to evaluate the policy effects of the BRI on carbon emissions [5]. Despite differences in viewpoints and methodologies, most of the current research centers on assessing the policy impact of the BRI itself, particularly evaluating how policy implementation influences carbon emissions in participating countries. However, few studies have explored the impact of cooperation-driven innovation facilitated by the BRI on carbon reduction. Given that promoting connectivity is a fundamental aspect of the BRI, the relationship between cooperation-driven innovation and carbon emissions warrants further investigation.

The BRI has, to some extent, facilitated innovation collaboration between China and its partner countries involved in the initiative. Studies using the DID method have demonstrated that the BRI significantly increased the proportion of collaborative patents from BRI countries within China’s total patent count, effectively enhancing the innovation capabilities and foundations of these countries [6]. Beyond invention patents, existing research highlights the crucial role of collaboration in driving innovation. There is a significant positive correlation between the level of author collaboration and their academic impact [7], and the diversity and quality of partners in corporate innovation activities are critical to their innovation success [8]. Compared to national-level collaboration, international collaboration has a significantly positive impact on innovation performance [9].

The concept of collaborative innovation has gained considerable consensus within the academic community. Collaborative innovation refers to organizational activities centered on innovation, with research and development as the primary focus [10]. From the perspective of division of labor, universities and research institutions are primarily responsible for knowledge production. This is often manifested in the establishment of fundamental research bases, the training of high-level talent, and the publication of scientific papers. Meanwhile, companies and intermediary organizations focus on technological development, market expansion, and learning or imitation, typically reflected in patents, new product output, and technology diffusion [11]. Defining collaborative innovation merely as technological R&D cooperation is undoubtedly too narrow; the concept should encompass cooperation among different entities across all stages of the innovation process [11]. This means that collaborative innovation includes the interactive relationships among different regions and entities, involving both technological and knowledge-based cooperation.

In the current research on the relationship between collaborative innovation and carbon reduction, common indicators include technology spillovers [12], patent licensing, and technology transfer [13,14]. However, few studies have considered co-authored publications as a key indicator of collaborative innovation. Co-authored research papers are the most direct representation of knowledge cooperation and serve as a primary means of studying global knowledge collaboration. The publication of highly cited papers through collaboration is often used to reflect the evolution of global knowledge cooperation networks and global technological cooperation networks [15]. Trends in co-authored publications can directly indicate the level and intensity of innovation collaboration promoted by the BRI among participating countries. Therefore, this study used Web of Science (WoS) data on co-authored publications as a key indicator of the level of innovation collaboration among countries involved in the BRI.

In terms of methodology, existing research also offers room for further advancement. Most studies using the DID model have typically employed a single dummy variable, which has significant limitations: it is not easy to distinguish between the effects of policy interventions and natural variations, making it difficult to determine whether changes in the dependent variable are primarily influenced by policy or by natural factors. The spatial DID approach used in this study extends the traditional double-difference econometric model by incorporating a more generalized spatial lag operator. This generalization applies both in terms of events, providing a more flexible lag operator, and over time, offering a broader lag structure. It generates DID terms for both global regions and local regions. Within the DID framework, our study generalizes eight forms of spatial panel models, selecting the optimal model for estimating policy effects. This approach decomposes policy effects within the data generation process (DGP). Notably, while some spatial DID models only address the decomposition of global effects, this study further decomposes the impacts of policy shocks on local regions. The decomposition of spatial spillover effects involves analyzing not only the overall impact of policy shocks on the treated group but also the localized effects on individual regions within the treated group.

Based on the quasi-natural experiment method of the BRI, this paper uses a spatial difference-in-differences (SDID) model to examine the impact of cooperative innovation (scientific research cooperation) with China under the BRI on the carbon emission intensity of participating countries and its spatial spillover effects. The marginal contributions of this paper are as follows: Among the small literature assessing the policy effects of the BRI, this study expands the design of policy dummy variables and incorporates a decomposition of spatial effects. In the large literature on the impact of innovation on carbon emissions, we examine the environmental effects of collaborative publications as a representation of global knowledge cooperation.

First, the BRI’s promotion of financial integration, trade connectivity, infrastructure development, policy coordination, and people-to-people bonds facilitates knowledge production along the Belt and Road regions, thereby fostering innovation. This paper innovatively focuses on the carbon reduction effects driven by collaborative innovation. In the analysis of policy effects, we designed a framework to examine the impact and mechanisms of cooperative innovation under the BRI on carbon emissions. Specifically, we introduce an event dummy variable for cooperative innovation within the BRI context. Using an overlapping DID (difference-in-differences) approach, the study extends DID dummy variables into event and period dummy variables to construct a spatial DID model.

Second, the policy’s emphasis on connectivity aimed to advance shared development across BRI regions. This high spatial connectivity is expected to enhance socioeconomic development and, notably, environmental governance through various channels. Therefore, we selected BRI countries as the focus to analyze this spatial effect. The study of the environmental effects of cooperative innovation under the BRI emphasizes the control of spatial effects, examining not only spatial spillover effects but also decomposing local spatial effects.

Third, in determining spatial relationships, we designed spatial weight matrices based on the “Five Connectivity Goals” of the BRI. An endogenous spatiotemporal weight matrix is constructed to incorporate the temporal transmission of spatial spillover effects (policy spillover effects) into the matrix design. This matrix uses block matrices to reflect variable temporal transfer and transmission effects. Following the principles of structural alignment and effective relevance, we selected the optimal endogenous spatiotemporal weight matrix.

The remainder of the paper is organized as follows: Section 2 provides a brief review of the literature on carbon emissions in BRI countries and introduces the main research hypotheses. Section 3 outlines the methodology used in this study, while Section 4 presents the empirical specifications. Section 5 concludes the paper by summarizing the findings.

2. Literature Review and Hypotheses Development

2.1. Policy Measures of the Belt and Road Initiative

The policy measures of the BRI have become a significant research topic. Existing studies mainly focus on the design of the policy system [16], the interpretation of its connotations [17,18], and qualitative analyses of China’s capacity [19] to implement the initiative. Some studies also employ quantitative text analysis methods to statistically analyze the structural characteristics of the BRI’s policy tools [20]. Existing research indicates that the BRI embodies an inclusive globalization model that promotes industrial restructuring, growth in foreign investment, diversification of energy routes, and multilateral financial cooperation [21]. The liberalization of trade within the global division of labor and its economic, social, and environmental effects have become focal points for contemporary energy, environment, and climate change economics. Based on the theoretical framework of environmental economics, using causal identification tools to establish causal relationships between key variables becomes an important research approach [22]. As the joint construction of the BRI enters a stage of high-quality development, guided by goals such as the comprehensive connectivity of the Eurasian continent, the Green Silk Road, and the Digital Silk Road, international cooperation, trade and investment, innovation-driven development, and global governance have become themes for win-win cooperation among BRI participant countries and regions.

The BRI advocates for an open and cooperative approach, which helps enhance the innovation systems of participating countries [23]. The BRI has promoted extensive socio-economic cooperation, including high-standard infrastructure projects. China has signed technological cooperation agreements with multiple participating countries and established an international coalition of scientific organizations. Scholars have also noted that the BRI promotes international collaborative innovation [6] and technology transfer [24]. Wang et al. used a difference-in-difference-in-differences (DDD) model to find that the BRI has enhanced the innovation capacity of participating countries, which is a key factor for achieving sustainable development [25]. The BRI has significantly promoted the green innovation upgrades of Chinese enterprises by increasing green entrepreneurship, cooperative innovation, and environmental investment [26]. A Nature editorial pointed out that the BRI has made substantial investments in scientific research, funding a large number of master’s and doctoral programs, and holds great potential for fostering scientific research and knowledge generation [27]. Deng et al. found that China exerts a dual-channel technology spillover effect on BRI countries through trade and investment [28].

2.2. Mechanisms of Carbon Emission Reduction

In terms of the mechanisms of carbon emission reduction, scholars employing the multi-region structural decomposition analysis (SDA) technique have found that the slight decrease in global CO₂ emission intensity is mainly attributed to improvements in sectoral emission efficiency, while international trade has somewhat impeded the overall improvement in global emission intensity. Furthermore, researchers have compared the performances of emerging economies and developed economies, finding that production structures and final demand structures are crucial in reducing CO₂ emission intensity [29].

Scholars utilizing the LMDI decomposition methodology have analyzed the contributions of various factors to carbon emission reduction. They found that the energy intensity effect is the primary driver of the decline in CO₂ intensity, while the CO₂ coefficient effect contributes less to mitigation, and the structural effect tends to increase emission intensity [30]. Danish et al., using the time series method, confirmed that economic policy uncertainty amplifies the negative impact of energy intensity on CO₂ emissions [31]. Mirziyoyeva and Salahodjaev focused on the impact of renewable energy on carbon emission reduction. Their research found that in countries with the highest carbon emission intensity, increases in renewable energy and per capita GDP can lead to reductions in CO₂ emissions [32].

Specifically for countries along the Belt and Road Initiative, Jafarzadeh and Yang conducted a case study on the sustainability of the BRI in Iran’s Aras region using the DPSIR framework and Genetic Algorithm. Their study found that regional heterogeneity, excessive national pressures, and development imbalances significantly impact the sustainable development of the area [33]. Additionally, scholars have discovered that by applying the Stackelberg master–slave mode, Chinese enterprises can incentivize local enterprises in developing countries along the Belt and Road Initiative to coordinate carbon emission reductions through subsidies.

Scholars have also focused on the environmental effects of collaborative innovation. Ding Lili and colleagues examined the impact of corporate collaborative innovation models on carbon reduction under mixed environmental policies [12]. Song et al. investigated the effect of industry–academia–research collaborative innovation efficiency on carbon emissions [14]. Zhao et al. found that collaborative green innovation is more effective in reducing carbon emissions than independent green innovation [34]. Xu et al.’s research revealed that international cooperation in new energy technologies can significantly reduce carbon emission intensity [35]. Zhang et al. found that cooperation between enterprises helps promote eco-innovation [36]. Pandey et al. explored the role and challenges of international technology transfer (ITT) in promoting sustainable development in developing countries, proposing that “collaborative innovation” offers a better framework for advancing international cooperation in this field [37]. Poirier et al. emphasized that international scientific cooperation, particularly between OECD and non-OECD countries, is critical for innovation in climate change mitigation technologies [38]. Additionally, co-authored research papers not only increase the output of sustainability-related studies but also strengthen interdisciplinary connections and the dissemination of academic knowledge [39].

2.3. Environmental Effects of the Belt and Road Initiative on Partner Countries

Regarding the environmental effects of China’s outward direct investment on host countries within the Belt and Road Initiative (BRI), academia has presented several different viewpoints. Numerous studies suggest that the construction of infrastructure projects can lead to significant environmental impacts, such as deforestation, air and water pollution, soil erosion, and loss of biodiversity [40]. Existing research uses structural decomposition analysis methods to identify the driving factors of total carbon emissions, including potential carbon emissions, potential energy intensity, GDP, and carbon reduction technologies [41]. These studies highlight that the carbon emission efficiency levels in countries along the BRI are generally low, and the conclusions about the impact of the BRI on the carbon emission intensity of these countries are mixed. Some studies suggest that the BRI might increase carbon emissions alongside economic growth [42,43], resulting in a moderate increase in global CO₂ emissions and a range of mixed positive and negative outcomes for other types of emissions [44]. Other research indicates that cooperation with China fosters positive environmental effects through green development [45]. Integration into the global value chain can enhance the carbon emission efficiency of manufacturing in BRI countries through the “catch-up effect”, “innovation effect”, and “leadership effect” [46]. Trade openness negatively impacts transportation carbon emissions [3,4]. Additionally, some studies using the Environmental Kuznets Curve (EKC) demonstrate an inverted U-shaped relationship between BRI trade openness and carbon emissions [47].

Scholars have used the quasi-natural experiment of the BRI to examine its spatial spillover effects through the spatial difference-in-differences (SDID) method. The regression results of the SDID indicate that the BRI has significant positive policy spillover effects, reducing carbon emission intensity not only in the participating regions but also in neighboring and not-participating regions. Regarding influencing factors, the BRI primarily promotes the reduction of carbon emission intensity in these regions by fostering income growth, technological advancements, industrial structure upgrades, and infrastructure improvements [5]. Based on this discussion, studies have pointed out that countries with different economic levels benefit to varying degrees from BRI green energy investments [48]. Additionally, research indicates that the BRI has a significant impact on reducing carbon emission intensity, particularly in transportation, electricity, heating, manufacturing, and construction sectors, for countries with both high and low carbon emission intensities [49]. Studies also highlight that the environmental effects of infrastructure development under the BRI framework interact and accumulate across multiple spatial and temporal scales [43]. Furthermore, the BRI stimulates developed countries to increase competitive investments in green and low-carbon emission industries in developing countries, thereby providing more green development resources to countries along the Belt and Road [50].

Based on this, the following research hypotheses are proposed:

Hypothesis 1.

The BRI is conducive to reducing the carbon emission intensity levels of participating countries.

2.4. Comprehensive Effects of Innovation in the BRI

Regarding the promotion of research and development (R&D) innovation and knowledge transfer in countries along the BRI, scholars have used the DID model to compare emerging market countries along the BRI with non-BRI emerging market countries. They found that the BRI positively facilitates the transfer of technology from China to emerging markets along the route [51]. Studies have also indicated that the implementation of the BRI has led to increased R&D expenditure and improved asset utilization by Chinese enterprises establishing branches in countries along the route, further enhancing research outcomes and the efficiency of knowledge transfer in these countries. Therefore, the BRI, as a policy to promote outward direct investment, will improve the overall environment for such investments [52].

Li et al. using the cross-sectional autoregressive distributed lag (CS-ARDL) model, found that green innovation and exports had significant negative effects on consumption-based carbon emissions in BRI host countries [53]. Other scholars have examined the impact of the interaction between natural resources and innovation on the environment, finding that natural resources exacerbate environmental degradation, while technological innovation has the opposite effect. The interaction between natural resources and technological innovation negatively impacts the ecological footprint [53]. Natural resources and financial inclusion have led to increased regional carbon and ecological footprints. Meanwhile, technological innovation, government efficiency, regulatory quality, and human capital help mitigate environmental degradation and enhance environmental sustainability [54].

The BRI has not only accelerated developments in infrastructure, foreign direct investment (FDI), and trade but also strengthened innovation cooperation, yielding positive impacts on sustainable development. Studies indicate that innovation investment in BRI countries significantly enhances green total factor productivity [55]. The BRI’s open cooperation model has expanded communication channels for researchers, and Xu et al. have confirmed the spatial spillover of knowledge and technology among participating countries, which fosters a collaborative increase in green innovation efficiency [56]. Furthermore, resource complementarity and information sharing have led to the formation of scientific innovation networks between universities, research institutions, and enterprises, creating synergistic effects across knowledge, technology, and industrial chains. Under the BRI, innovation capacity has significantly improved the energy efficiency of participating countries, with global collaboration playing a notable moderating role in this outcome [24].

Overall, existing studies on carbon emissions in BRI countries often focus on one or several countries or regions, or on specific industries such as power generation, with few considering spatial effects. Most studies examine the impact of China’s outward direct investment on carbon emissions in countries along the BRI from the perspective of energy consumption and utilization, with few evaluating the policy effects of innovation on carbon emissions in BRI participating countries, especially the effects of cooperative innovation on carbon emission reduction. Most studies believe that the implementation of the initiative has a positive impact on reducing the carbon emission intensity of countries along the BRI. Besides, existing research has deeply explored this impact by industry and region, but few scholars have examined the environmental effects of cooperative innovation within the BRI framework from a policy effect perspective. They have not compared the carbon emission performance of countries along the BRI before and after the initiative, nor have they interpreted the carbon emission reduction performance brought by the initiative or refined the carbon emission reduction effects of the initiative. Therefore, this paper treats cooperative innovation with China as a policy shock and studies the environmental effects of cooperative innovation on BRI participating countries.

Therefore, in examining the environmental effects of innovation and cooperative innovation within BRI countries, existing research primarily utilizes traditional methods, such as DEA and SBM models, to validate these impacts. Studies affirm the critical role of economic and social cooperation, particularly innovation cooperation. However, these models often incorporate indicators representing economic and social cooperation, such as the share of investment or trade in GDP and the degree of informatization, without specifically addressing the environmental impacts, such as emissions reduction, resulting from these cooperative models. Moreover, while studies on carbon emissions frequently consider “spatial” factors, research on the environmental effects of the BRI lacks adequate control spatial effects. Policy effect evaluations using DID and DDD models rarely explore spatial spillover effects and, in particular, seldom include decompositions of local spatial effects. Based on this, the following research hypotheses are proposed:

Hypothesis 2.

Cooperative innovation under the BRI is conducive to reducing the carbon emission intensity of participating countries.

Hypothesis 2a.

The BRI reduces the carbon emission intensity of participating countries through mechanisms such as investment effects, innovation effects, and trade effects.

Hypothesis 2b.

Spatial spillover effects exist in the carbon emission intensity reduction effects of cooperative innovation.

Hypothesis 2c.

The local carbon emission intensity reduction effects of cooperative innovation exhibit spatial heterogeneity.

3. Method and Research Design

3.1. Research Design

The difference-in-differences (DID) method is widely used in empirical research to evaluate policy effects. Its core lies in constructing interaction terms to identify the average treatment effect of a policy intervention on affected individuals (i.e., the treatment group). This approach is grounded in a counterfactual framework, assessing changes in outcomes under two scenarios: with and without the policy intervention. The method has been extensively applied in studies on pollution emissions [57], disease transmission [58], and housing price fluctuations [59]. Recently, the spatial interdependence among individuals has garnered increasing attention from researchers [60]. Taking air pollution as an example, the mobility of air leads to cross-boundary effects, where heavily polluted areas can affect the air quality of adjacent regions. Conversely, regions that implement effective emission reduction measures may inspire neighboring areas to adopt similar strategies, thereby achieving regional optimization of environmental quality. In light of these considerations, particularly the challenges posed by spatial correlation and endogeneity, the spatial difference-in-differences (SDID) model has emerged.

The SDID model is designed to evaluate policy impacts under spatial dependency among variables. Its core innovation lies in incorporating time-policy interaction terms into spatial econometric models [61]. Compared to traditional DID models and regression discontinuity designs (RDDs), the SDID model addresses limitations in observing the magnitude and direction of policy impacts on control groups by analyzing both global and local spatial spillover effects. From the perspective of spatial connectivity, the SDID model surpasses standard spatial econometric models by integrating spatial effects directly into policy effect evaluation frameworks.

3.1.1. Baseline Model: Traditional DID Model

Previous studies have predominantly used cross-sectional data or linear regression methods to investigate the environmental effects of the BRI, which makes it difficult to avoid endogeneity issues. This paper adopts a quasi-natural experimental research approach. First, it treats the BRI as a quasi-natural experiment, with countries that have signed cooperation agreements and engaged in cooperative innovation with China as the experimental group, and other countries as the control group. To determine the net effect of the initiative, a continuous DID method is used to analyze the data. To reduce sample selection bias, appropriate covariates are selected, and propensity score matching is employed to match the experimental group countries with the control group countries. Thus, the results of the continuous DID are re-verified. The OLS regression model for this paper is as follows:

$$\ln C{O}_2=\alpha +{\beta }_1\ln INNO+{\beta }_2\ln OFDI+{\beta }_3\ln POP+{\beta }_4\ln TRA+{\beta }_{5}IND+{\beta }_{6}URB+{\beta }_{7}CAP+\epsilon$$

(1)

In the above equation, CO₂ represents the per capita CO₂ emissions of the sample countries studied, INNO denotes the number of scientific journal publications in BRI participating countries, and OFDI (outward foreign direct investment) stands for China’s direct investment in BRI participating countries. POP, TRA, IND, URB, and CAP represent the total population, openness level, industrial development level, urbanization level, and fixed capital investment level, respectively.

The baseline DID model for this paper is as follows:

$$\mathrm{y}={\alpha }_0+{\alpha }_{1}du+{\alpha }_{2}dt+{\alpha }_{3}du\cdot dt+\epsilon$$

(2)

In the above equation, du is a grouping dummy variable. If individual i is affected by the policy implementation, individual i belongs to the treatment group and the corresponding value of du is 1. If individual i is not affected by the policy implementation, individual i belongs to the control group and the corresponding value of du is 0. dt is the policy implementation dummy variable, which takes a value of 0 before the policy implementation and 1 after the policy implementation. The interaction term du⋅dt represents the interaction between the grouping dummy variable and the policy implementation dummy variable.

3.1.2. Mechanism Testing: Setting up the SDID Model and Decomposing Policy Effects

The SDID model is built upon the Hedonic Pricing Model (HPM) by incorporating events, periods, and interaction terms for DID analysis, adding individual fixed effects, and progressively integrating spatial econometric model forms such as spatial autocorrelation and spatial lag. This approach mitigates spatial multicollinearity, controls endogeneity issues [62], and variable bias, effectively integrating spatial dependency into the DID framework [63]. Furthermore, the design of spatial lag operators extends from the spatial to the temporal dimension, making it more general. Thus, the SDID model, considering spatial spillover effects, integrates eight spatial econometric models into the DID model, creating various SDID models. Fixed effects of panel data further eliminate spatial heterogeneity. The advantage in policy effect analysis lies in analyzing both the overall shock effect on the treatment group and the local single shock effects on individual regions within the treatment group. Two dummy variables are set up for implementation: D₁ for events and D₂ for periods, containing time information of the event.

The basic setup of the generalized spatial difference-in-differences (GNSM-SDID) model is as follows:

$$y=\rho TWy+u+v+X\beta +TWX\theta +D\left(·\right)+TWD\left(·\right)\pi +\mu$$

(3)

$$\mu =\lambda TW\mu +\epsilon$$

(4)

$$D\left(·\right)=D_1{\eta }_1+D_2{\eta }_2+D_1{D}_2{\eta }_3$$

(5)

$$TWD\left(·\right)\pi =TW{D}_1{\pi }_1+TW{D}_2{\pi }_2+TW{D}_1{D}_2{\pi }_3$$

(6)

The six degenerate forms of the GNSM-SDID model are as follows:

$➀$ SXL-SDID model, formula:

$$y=u+\nu +X\beta +TWX\theta +D\left(·\right)+TWD\left(·\right)\pi +\epsilon$$

(7)

$➁$ SAR-SDID model, formula:

$$y=\rho TWy+u+\nu +X\beta +D\left(·\right)+\epsilon$$

(8)

$➂$ SEM-SDID model, formula:

$$y=u+\nu +X\beta +D\left(·\right)+\mu$$

(9)

$➃$ SDEM-SDID model, formula:

$$y=u+\nu +X\beta +TWX\theta +D\left(·\right)+TWD\left(·\right)\pi +\mu$$

(10)

$➄$ SAC-SDID model, formula:

$$y=\rho TWy+u+\nu +X\beta +D\left(·\right)+\mu$$

(11)

$➅$ SDM-SDID model, formula:

$$y=\rho TWy+u+\nu +X\beta +TWX\theta +D\left(·\right)+TWD\left(·\right)\pi +\mu$$

(12)

In the above equations, y represents the dependent variable, ρ represents the spatial autoregressive coefficient in models like SAR, SEM, and SDM. TW denotes the endogenous spatio-temporal weight matrix. u and v represent individual and period fixed effects, respectively, and X represents the set of relevant explanatory variables. TWy and TWX represent the lagged terms of the dependent and independent variables, respectively, indicating the dependent and independent variables of neighboring countries or regions. The SDID model primarily adds dummy variables, D₁ and D₂, to the original spatial econometric regression model, representing the parameters to be estimated. The process of generating event dummy variables involves selecting countries with close cooperative innovation ties with China from the 106 BRI countries as the treatment group, and countries where cooperative innovation has not yet matured as the control group. As the timing of signing cooperation agreements or memorandums of understanding is not concentrated in the same year, the effectiveness of cooperative innovation under this policy does not appear in the same year, indicating a temporal difference in whether the research subjects received policy intervention. Therefore, the process of generating period dummy variables follows the evolution of cooperative innovation relationships, taking the turning point of a significant increase in the number of joint publications with China as the policy effect year. Event dummy variables are set to 1 for the treatment group and 0 for the control group; period dummy variables are set to 1 for the treatment group in the policy effect year and beyond, and 0 for the rest. Specifically, Web of Science was used to search for joint publications by Chinese scholars and scholars from BRI participating countries, using co-authored papers to represent cooperative innovation in knowledge production between China and BRI participating countries. Through the development and evolution of cooperative innovation, event and period dummy variables were determined.

3.2. Selection of Indicators and Data Sources

3.2.1. Explained Variable

The dependent variable is the carbon emission intensity of BRI participating countries or regions, represented by CO₂ emissions per unit of GDP. Existing studies often use indicators such as total carbon emissions, carbon emission efficiency, and carbon emission intensity to represent the carbon emissions of BRI countries [28,40,44]. Considering data availability and representativeness, this paper selected the ratio of total CO₂ emissions to GDP (tons per 10,000 USD) as the indicator for the dependent variable, carbon emission intensity.

3.2.2. Explanatory Variables

The core explanatory variables include innovation performance, OFDI, and trade openness, represented by the number of scientific journal papers from each country, China’s direct investment in these countries, and proportion of import and export trade to GDP respectively (

Table 1. Explanation of variable selection.

Content of Indicator	Description of Indicator	Symbol	Unit
Carbon emission intensity	Ratio of CO2 emissions to GDP	CO2	tons/10,000 USD
Innovation performance	Number of scientific journal papers	INNO	paper
OFDI	China’s direct investment in BRI countries	OFDI	10,000 USD
Openness	Proportion of import and export trade to GDP	TRA	%
Population size	Total population of each country	POP	person
Industrial development level	Proportion of industrial added value to GDP	IND	%
Urbanization level	Urbanization rate	URB	%
Fixed capital investment level	Proportion of gross fixed capital formation to GDP	CAP	%

). This follows the study of existing research, highlighting that innovation capacity influences green development [24], with scientific publications and patents serving as primary indicators of innovation capacity [64]. In this paper, the number of articles published in scientific journals is selected as a measure of innovation capacity [65,66]. Referring to Liu et al., outward foreign direct investment (OFDI) and trade openness are critical factors impacting carbon emissions [3,67]. Based on current research, this paper controls for variables such as population size [5], industrial development level [68], urbanization level, and fixed capital investment level [69]. These variables are represented by the logarithm of population size, the proportion of industrial added value, the urbanization rate [70], and the percentage of gross fixed capital formation in GDP.

3.2.3. Data Sources

The sample countries and regions for this study are derived from the list of 152 countries provided in the “Country Profiles” section of the Belt and Road Portal “https://www.yidaiyilu.gov.cn (accessed on 12 March 2024)” as of March 2024. These primarily include Belt and Road countries and countries that have signed bilateral cooperation documents and other practical cooperation projects with China under the Belt and Road Initiative. Hereinafter, these countries will be referred to as “participating countries”. The data for this study primarily include carbon emission data and socio-economic development data. In the SDID model, the construction of period dummy variables and event dummy variables (i.e., D1 and D2 as mentioned in Section 3.1.2) is based on joint publication data. Joint publication data, representing collaborative innovation, is sourced from the Web of Science.

For the dependent and independent variables, carbon emission data were sourced from the Global Carbon Atlas database, which provides CO₂ emission data for countries worldwide. Socio-economic development data, such as the number of scientific journal articles, GDP, total population, openness, industrial development level, urbanization level, and fixed capital investment level, were mainly retrieved from the World Bank’s World Development Indicators database “https://databank.worldbank.org (accessed on 16 March 2024)”. Data on OFDI in Belt and Road countries were obtained from the China Commerce Yearbook (2013–2022) and the flow of China’s outward direct investment tables. The study compiles CO₂ emissions and socio-economic development data for 106 participating countries from 2005 to 2021 (excluding Afghanistan, Djibouti, Somalia, and Syria due to data unavailability).

3.2.4. Variable Descriptive Statistics

Table 2. Descriptive statistics of the variables.

Variable	Obs.	Mean	Standard Deviation	Min.	Max.
lnINNO	1802	6.010	2.500	0.315	11.376
lnOFDI	1802	7.571	2.422	0.693	13.951
TRA	1802	67.799	37.299	10.202	343.488
lnPOP	1802	16.138	1.599	11.325	19.412
IND	1802	29.435	13.571	4.429	86.669
URB	1802	53.738	21.856	9.375	100
CAP	1802	24.287	7.713	2.000	81.021

provides the descriptive statistics for the variables.

3.3. Design and Optimization of the Spatio-Temporal Weight Matrix

Spatial weight matrices describe the spatial relationships between pairs of sample countries in the study, developed from binary adjacency relationships to characterize spatial proximity, primarily distinguishing whether regions are adjacent and the distances between them. The BRI has entered a new phase of promoting the construction of an interconnected network, positively influencing trade exchanges and technological interactions among participating countries, fostering cooperative innovation among countries. As technological advancements overcome geographical distance, trade exchanges, especially cooperative innovation activities, are no longer constrained by geographical distance. Therefore, drawing from the principle of multidimensional proximities in innovation geography, a spatial weight matrix based on institutional distance is constructed, building on the traditional spatial weight matrix. The spatio-temporal weight matrix (TW) is typically created by combining a temporal weight matrix (T) with a spatial weight matrix (W). The construction approach of the spatio-temporal weight matrix is illustrated in Figure 1. The endogenous spatio-temporal weight matrix was formed by the Kronecker product combination of the standardized spatial weight matrix and the temporal weight matrix based on the annual global Moran’s I. In this combination process, the temporal weight matrix captures how spatial relationships change and propagate over time. The primary advantage of an endogenous spatio-temporal weight matrix is that its elements are generated endogenously, reducing the subjectivity involved in designing weight matrices. Moreover, the elements of the spatial weight matrix include variable temporal effects, allowing for an accurate simulation of the transfer and transmission effects of spatial spillovers across different periods.

3.3.1. Adjacency Spatial Weight Matrix

Based on whether the geographical locations are adjacent, a first-order adjacency (0,1) matrix W₁ is established, with diagonal elements set to 0 and other elements satisfying the following:

$$W_{ij}=\left\{\begin{array}{l}1,\mathrm{Country}\mathrm{i}\mathrm{is}\mathrm{adjacent}\mathrm{to}\mathrm{country}\mathrm{j}\\ 0,\mathrm{Country}\mathrm{i}\mathrm{is}\mathrm{not}\mathrm{adjacent}\mathrm{to}\mathrm{country}\mathrm{j}\end{array}\right\}$$

(13)

3.3.2. Geographical Distance Spatial Weight Matrix

According to the first law of geography, spatial correlation weakens with increasing geographical distance. The geographical distance spatial weight matrix W₂ is set as follows:

$$W_{ij}=\left\{\begin{array}{l}{\displaystyle \frac{1}{d_{ij}}},i\ne j\\ 0,i=j\end{array}\right\}$$

(14)

where d_ij represents the geographical distance between countries (regions) i and j. The geographical distance is sourced from the CEPII database of the World Economic Research Institute. It is calculated based on the bilateral distance between the largest cities of each country. These city-to-city distances are weighted according to the proportion of the city’s population within the total population of the country [71].

3.3.3. Economic Distance Spatial Weight Matrix

The spillover effects vary between countries with different economic levels. The economic distance spatial weight matrix W₃ is constructed using the reciprocal of the absolute annual average differences in real per capita GDP (deflated to the 2000 base year). The spillover effects vary among countries with different economic levels. The economic distance spatial weight matrix W₃ is constructed by using the reciprocal of the absolute difference in the annual average of real GDP per capita (deflated to the base year 2000) between different countries:

$$W_{ij}=\left\{\begin{array}{l}{\displaystyle \frac{1}{\left|\overline{Y_i}-\overline{Y_j}\right|}},i\ne j\\ 0,i=j\end{array}\right\}$$

(15)

where ${\overline{Y}}_i={\displaystyle \frac{1}{t-t_0}}{\displaystyle \sum _{t=t_0}^t{Y}_{it}}$, ${\overline{Y}}_j={\displaystyle \frac{1}{t-t_0}}{\displaystyle \sum _{t=t_0}^t{Y}_{jt}}$.

3.3.4. Language Distance Spatial Weight Matrix

The language distance data also come from the CEPII database. If two countries share a common official language, it is recorded as 1; otherwise, it is 0. Thus, a language distance spatial weight matrix W₄ that includes language difference matrix information is constructed:

$$W_{ij}=\left\{\begin{array}{l}1,\mathrm{The}\mathrm{official}\mathrm{language}\mathrm{of}\mathrm{country}\mathrm{i}\mathrm{is}\mathrm{the}\mathrm{same}\mathrm{as}\mathrm{that}\mathrm{of}\mathrm{country}\mathrm{j}\\ 0,\mathrm{The}\mathrm{official}\mathrm{language}\mathrm{of}\mathrm{country}\mathrm{i}\mathrm{is}\mathrm{not}\mathrm{the}\mathrm{same}\mathrm{as}\mathrm{that}\mathrm{of}\mathrm{country}\mathrm{j}\end{array}\right\}$$

(16)

3.3.5. Institutional Distance Spatial Weight Matrix

Few studies have comprehensively measured the institutional distance between the BRI countries. Institutional distance reflects differences in laws, policies, and market rules between countries. The greater the institutional distance, the higher the communication, learning, and adaptation costs for cooperative innovation. Drawing on the approach of Li et al. [72], this paper used World Governance Indicator (WGI) scores as a basis. The average difference in institutional quality scores between pairs of countries is used as the institutional distance to construct the spatial weight matrix W₅. Institutional quality is measured across six dimensions: voice and accountability, political stability and absence of violence/terrorism, government effectiveness, regulatory quality, rule of law, and control of corruption. The arithmetic mean of a country’s WGI data across these six dimensions was used as the comprehensive institutional quality score. Data were sourced from the World Bank database.

$$W_{ij}=\left\{\begin{array}{l}{\displaystyle \frac{\frac{1}{I{D}_{ij}}}{{\displaystyle \sum _{j=1}^n\frac{1}{I{D}_{ij}}}}},i\ne j\\ 0,i=j\end{array}\right\}$$

(17)

Here, ID represents the institutional distance between country i and country j, calculated as follows:

$$I{D}_{ij}=\left|{\displaystyle \frac{1}{6}}{\displaystyle \sum _{k=1}^6{W}_{ki}-\frac{1}{6}{\displaystyle \sum _{k=1}^6{W}_{kj}}}\right|$$

(18)

In the empirical analysis, we design an endogenous spatio-temporal weight matrix based on the above five spatial weight matrices, incorporating variable time effects to represent spatial spillover effects and their temporal transfer and propagation. Specifically, we generated candidate endogenous spatio-temporal weight matrices from these five spatial weight matrices and calculated the global Moran’s I based on these matrices. Drawing on the studies by Fan and Darren [73], we used the ratio of Moran’s I at different periods to eliminate the influence of the initial spatial weight matrix settings and simulated the temporal transfer and propagation of spatial spillover effects to form the temporal weight matrix. Finally, the Kronecker product of the spatial and temporal weight matrices forms the spatio-temporal weight matrix. We selected the optimal spatio-temporal weight matrix based on the total average coefficient of variation of the model estimation results, following the principles of structural matching, effective correlation, and strongly effective correlation.

4. Empirical Result Analysis

4.1. Baseline Regression and Traditional DID Model Results

The baseline regression utilizes the OLS method to preliminarily examine the coefficients of the explanatory variables, which forms the basis for constructing the traditional DID model. An important prerequisite for using the SDID model is the consistency of trend changes between the treatment group and the control group before the policy implementation, meaning that the trends in carbon emission intensity changes were consistent across countries before the implementation of the “Belt and Road” initiative. Hence, this paper first conducts a parallel trend test to verify the applicability of the DID method. The results show no significant differences in the development trends of carbon emission intensity, including both the experimental and control groups, thus satisfying the parallel trend assumption. Upon verification, the data in this paper can be used for further estimation using the DID model.

After conducting the parallel trend test, the baseline model estimates the policy effects of the BRI from 2005 to 2021 for 106 countries and regions using the DID method, incorporating models established with Stata 17 and MATLAB R2022b software. The baseline model only regresses the dummy variable SDID, representing the implementation of the BRI (i.e., whether a country is a BRI participating country), against carbon emission intensity (see

Table 3. OLS baseline regression and traditional DID model.

Variable	Baseline Model	DID Model
lnINNO	−0.227 ** (−2.26)	—
lnOFDI	−0.246 *** (−3.69)
lnPOP	0.882 *** (5.79)
TRA	−0.215 *** (−4.84)
IND	0.069 *** (5.39)
URB	0.033 *** (3.28)
CAP	0.031 * (1.91)
SDID		−0.048 *** (−3.75)
Spatial rho		0.0875 *** (147.72)
Spatial lambda		0.067 *** (142.59)
Cons.	−11.12 *** (−4.83)	—

Notes: ***, **, and * denote 1%, 5%, and 10% significance levels, respectively.

). The BRI policy dummy variable is significant at the 1% level with a negative coefficient. The policy effect of the BRI is −0.048, indicating that the implementation of the BRI helps reduce carbon emission intensity in participating countries. This confirms Hypothesis 1.

4.2. Empirical Results of the SDID Model

4.2.1. Optimization of the Endogenous Spatio-Temporal Weight Matrix and Setting of Dummy Variables

Figure 1 reports the endogenous spatio-temporal weight matrix established based on five types of spatial weight matrices, while Figure 2 presents the dynamic Moran’s I under different matrices.

Following the selection principles mentioned earlier, the spatial weight matrix was chosen based on the effective correlation coefficient and Fisher’s T-statistic. At a significance level of 1%, the optimal spatial weight matrix is the endogenous spatio-temporal weight matrix TW₅, generated from the institutional distance spatial weight matrix W₅ (see Figure 3). Additionally, a matrix similar in structure to the endogenous spatio-temporal weight matrix, representing carbon emission intensity, is shown (see Figure 4).

Following the construction approach for event dummy variables and period dummy variables mentioned earlier, the dummy variables in the SDID model were generated in MATLAB R2022b. First, generate a research object matrix with a structure similar to the endogenous spatiotemporal weight matrix (Figure 5). Second, generate the dummy variables for the global SDID model as shown in Figure 6. Additionally, generate the DID items in the SDID modeling and the local DID items in the local area policy impact modeling (Appendix A).

4.2.2. Estimation and Decomposition of the Global Policy Impact

We conducted a global SDID model estimation, obtaining parameter estimation results for eight linear SDID models, including the Non-Spatial Model (NSM), Spatial X Lag Model (SXL), Spatial Autoregressive Model (SAR), Spatial Error Model (SEM), Spatial Durbin Model (SDM), Spatial Durbin Error Model (SDEM), Spatial Autocorrelation Model (SAC), and Generalized Nested Spatial Model (GNSM). Both the LM and Robust LM test results are significant at the 1% level, and the optimal model is the Spatial Durbin SDID Model. The decomposition of the global policy impact was performed based on the optimal model in the SDID estimation.

Table 4. Regression results of the DID model.

Variable	SDM-SDID	NSM-SDID	SXL-SDID	SAR-SDID	SEM-SDID	SDEM-SDID	SAC-SDID	GNSM-SDID
lnINNO	−0.171 *** (−4.376)	−0.083 ** (−2.347)	−0.173 *** (−4.378)	−0.089 ** (−2.525)	−0.019 (−0.625)	−0.175 *** (4.450)	−0.095 *** (2.652)	−0.174 *** (4.966)
lnOFDI	−0.182 * (−2.093)	−0.142 * (−1.798)	−0.175 ** (−1.995)	−0.236 *** (−2.872)	−0.224 *** (−2.793)	−0.184 ** (−2.097)	−0.233 *** (−2.820)	−0.172 ** (−1.989)
lnPOP	0.203 ** (1.548)	−0.030 (−0.266)	−0.216 (−1.635)	0.118 (0.985)	0.119 (1.037)	−0.202 (−1.533)	0.110 (0.919)	−0.225 * (−1.711)
TRA	−0.002 ** (−1.875)	−0.001 ** (−1.877)	−0.002 * (−1.832)	−0.002 (−0.937)	−0.001 (−0.305)	−0.002 (−0.978)	−0.002 (−0.976)	−0.005 ** (−2.125)
IND	2.864 *** (15.213)	3.148 *** (18.155)	2.858 *** (15.069)	2.991 *** (16.711)	3.175 *** (18.116)	2.877 *** (15.243)	2.984 *** (16.636)	2.874 *** (15.513)
URB	0.061 *** (7.674)	0.057 *** (7.842)	0.061 *** (7.591)	0.062 *** (8.412)	0.059 *** (7.945)	0.061 *** (7.692)	0.062 *** (8.414)	0.060 *** (7.839)
CAP	−0.005 *** (−15.218)	0.004 ** (2.093)	−0.006 *** (−3.976)	0.004 (0.742)	0.002 (0.368)	−0.006 *** (−3.988)	0.004 (0.770)	−0.006 ** (−2.154)
SDID	−0.248 *** (−3.969)	−0.870 ** (−2.140)	0.242 (0.530)	−0.749 * (−1.832)	−1.069 *** (−2.717)	0.255 (0.562)	−0.720 * (−1.755)	0.286 (0.766)
ρ	0.999 * (2.178)	—	—	−0.392 *** (−3.412)	—	—	−0.393 *** (−3.350)	−7.346 *** (−9203.090)
λ	—	—	—	—	−0.999 ** (−2.150)	−0.999 ** (−2.150)	0.070 (0.184)	−27.307 *** (−67,269.083)
TW∗lnINNO	−0.069 * (−1.653)	—	−0.076 ** (−2.196)	—	—	0.062 (0.208)	—	−1.558 *** (−6.822)
TW∗lnOFDI	−0.118 *** (−7.718)	—	0.564 *** (2.118)	—	—	−1.134 (−0.230)	—	4.262 *** (3.152)
TW∗lnPOP	−10.369 ** (−1.851)	—	−11.121 (−1.576)	—	—	−8.932 *** (−2.247)	—	−16.520 *** (−7.224)
TW∗TRA	−0.148 ** (−2.514)	—	−0.145 (−0.998)	—	—	−0.155 (−1.070)	—	−0.255 *** (−8.543)
TW∗IND	17.464 ** (2.315)	—	15.311 (1.393)	—	—	19.686 (1.590)	—	15.810 *** (5.573)
TW∗URB	0.008 *** (8.019)	—	−0.061 (−0.141)	—	—	−0.156 (−0.390)	—	1.114 *** (10.742)
TW∗CAP	−1.396 *** (−3.982)	—	−1.567 *** (−4.568)	—	—	−1.601 *** (−4.824)	—	−0.352 *** (−2.801)
TW∗SDID	−0.455 ** (−2.140)	—	1.863 (0.295)	—	—	1.439 (0.259)	—	−4.722 *** (−2.816)
Cons.	79.001 (0.468)	−25.55 *** (−10.010)	112.597 (0.662)	−24.241 *** (−9.419)	−27.074 *** (−10.688)	52.758 (0.293)	−24.147 *** (−9.379)	123.541 *** (2.560)

Notes: ***, **, and * denote 1%, 5%, and 10% significance levels, respectively.

presents the parameter estimation results of the optimal SDM-SDID and other models. The estimated coefficient for the SDID term is significantly negative, indicating that cooperative innovation under the BRI significantly reduces the carbon intensity of participating regions or countries, thus confirming Hypothesis 2. The Spatial Durbin Model results show that innovation performance and OFDI have significant negative effects on the carbon emission intensity of BRI countries, with the spatial spillover effects of the lnINNO and OFDI variables being evident. The spatial lag coefficients for these variables indicate that OFDI and cooperative innovation also suppress CO₂ emissions in neighboring regions. Additionally, the level of openness shows a significant carbon reduction effect and spatial spillover effects. Therefore, the BRI effectively reduces the carbon emission intensity of participating countries through investment, innovation, and trade effects, which confirms Hypothesis 2a. The spatial rho is 0.052, indicating a positive spatial autocorrelation, i.e., CO₂ emissions exhibit a positive spatial clustering trend with high-high and low-low clusters. The estimated coefficient of the TW*SDID term is also significantly negative, suggesting that compared to countries without close cooperative innovation relationships, those integrating with China’s cooperative innovation under the BRI have more pronounced CO₂ reduction effects. Cooperative innovation with China not only promotes carbon reduction in the home country or region, but also significantly negatively impacts the carbon emissions in neighboring regions, thus confirming Hypothesis 2b.

From the regression results of other control variables, the regression coefficient of the population size is significantly positive in the model, indicating that population growth correspondingly leads to an increase in carbon emission intensity. The regression coefficients and spatial lag terms of the industrial development level are both significantly positive in the model, suggesting that the energy consumption level of economic development in BRI countries is still relatively high, necessitating continuous green innovation to achieve industrial structure transformation and upgrading. The regression coefficient of the urbanization rate is significantly positive, showing that countries with higher urbanization levels have higher carbon emission intensity, indicating that BRI countries are still in the stage of immature urbanization development. Integrating into the BRI’s interconnected network will facilitate green development, innovative development, and empower high-quality urbanization. The regression coefficient of the percentage of gross fixed capital formation in GDP is significantly negative in the model. Therefore, BRI countries need to continue transforming their development approach to save energy and reduce carbon emissions.

Furthermore, the policy impact effect is decomposed into direct and indirect effects, with the total effect being the sum of the direct and indirect effects. The results of the global policy impact effect decomposition indicate that both the direct and indirect effects are significantly negative, suggesting that international collaborative innovation under the BRI not only reduces carbon emission intensity in the host country or region but also significantly decreases carbon emission intensity in neighboring countries or regions.

4.2.3. Estimation and Decomposition of Local Policy Impact Effects

The original data sequence includes the occurrence of collaborative innovation between individual countries or regions and China. Based on this, we can extracted corresponding elements from the control group for individual countries or regions to form new collaborative innovation dummy variables. These were used to evaluate the individual impact effects of the control group countries or regions. Let r = 1, 2, …, 43 represent the 43 treated countries, with specific numbers listed in the Appendix B. Within the framework of the SDID model, based on the 43 sets of dummy variables formed after decomposition, we obtain the parameter estimation models for all local regions according to the optimal model (Appendix C), as well as the decomposition results of the local policy impact effects (Appendix D). Detailed results are presented in the appendix. We estimate the parameters for the policy impact effect models of all local regions and save the parameter estimation results for each local region individually. Here, r represents the r-th country in the treatment group described previously, with “SDID_r” denoting the policy effect for the r-th country—namely, the impact of BRI cooperative innovation on that country’s carbon emissions. Correspondingly, “SDID_Non_r” represents the policy effect for countries other than the r-th country. “SDID_r” From the local regions, the “SDID_r” terms are mostly negative, indicating that for most treated countries, collaborative innovation under the BRI with China can reduce carbon emission intensity. However, the extent of this effect shows spatial differences. The distribution map of direct effects shows the direct impact of collaborative innovation under the BRI on carbon emissions in various regions. It can be observed that the direct effects in different regions are positive in some regions and negative in others, indicating that the policy has had an emission reduction effect in certain areas while potentially increasing carbon emissions in others.

The analysis of local policy impact decomposition (Figure 7) shows that the regression coefficients for the direct effects of collaborative innovation range from −9.244 to 10.832. In 25 countries, including Turkmenistan, Bahrain, Sri Lanka, Singapore, and Myanmar, the direct effects are all below −3.5, indicating significant emission intensity reduction effects from collaborative innovation with China under the BRI. In 21 countries, including Egypt, Ukraine, and Vietnam, the coefficients for indirect effects are negative, indicating that collaborative innovation exerts spatial spillover effects. The distribution map of total effects, which combines direct and indirect effects, indicates the overall impact of collaborative innovation with China under the BRI on regional carbon emission intensity. Notably, in countries like Kyrgyzstan, Uzbekistan, and Turkey, where the direct effects are positive, the average indirect effects are negative, suggesting substantial externalities in emission intensity reduction through collaborative innovation under the BRI framework. The distribution map of average indirect effects illustrates the average impact of collaborative innovation on the carbon emission intensity of neighboring regions. The varied distribution of total effects demonstrates significant differences in the emission reduction impact of the policy in local regions, supporting Hypothesis 2c.

4.3. Discussion

This paper focuses on the policy effects of knowledge cooperation under the Belt and Road Initiative (BRI), exploring the environmental impact of knowledge collaboration among participating countries and its spatial spillover effects. The aim was to investigate the role of the interconnected cooperation model in reshaping the landscape of knowledge innovation under the BRI, which represents one aspect of scientific and technological cooperation among BRI countries. In the future, further exploration of the evolution of the knowledge cooperation network among BRI countries could be conducted, with a more detailed focus on key areas of knowledge collaboration. By considering the technological development levels of participating countries, future research could delve into the collaborative models that deepen innovation through knowledge cooperation between nations or key cities. Based on these insights, recommendations could be made to improve mechanisms for international scientific and technological cooperation.

Promoting green and low-carbon development, and establishing an international cooperative ecosystem to achieve carbon peak and carbon neutrality goals, has become a common objective among the countries involved in the BRI. Research on the policy effects of the BRI on the carbon emissions of participating countries remains limited, as existing studies primarily analyze the policy effects based on the implementation timeline. This paper innovatively considers the evolving trend of cooperative relations among BRI participating countries, treating cooperative innovation with China as a policy shock, and examines the environmental impacts of such cooperative innovation on BRI participating countries. Using data from the Web of Science, we constructed a cooperative innovation index between China and the BRI countries. Combined with data from the World Bank and the China Commerce Yearbook, we compiled data on the carbon emissions and economic development of 106 BRI countries and regions from 2005 to 2021. We then establish a SDID model. Following a quasi-natural experiment research approach, we consider cooperative innovation between BRI countries and China as an event dummy variable representing policy shock, and the inflection point of cooperative innovation evolution as the time dummy variable representing policy shock. This model helps measure the environmental effects, particularly carbon emission intensity reduction, of the BRI in participating countries. Additionally, during the estimation process of policy shock effects, we incorporate combinations of event and time dummy variables to comprehensively cover the policy shock effects. We reflect the “Five Connections” development in the initial spatial weight matrix and account for time-varying effects in the endogenous spatio-temporal weight matrix, ensuring that the empirical results capture the temporal transfer and transmission effects of spatial spillovers. Furthermore, after obtaining the global model estimation results, we decompose and analyze the local policy shock effects, summarizing the spatial heterogeneity of these effects.

By constructing and optimizing spatial weight matrices, building eight spatial DID (SDID) models that include double-difference terms, and selecting the best model, we empirically analyzed the CO₂ emission effects of collaborative innovation under the Belt and Road Initiative (BRI). First, we performed global SDID model estimations and obtained parameter estimates for the eight linear SDID models. The selected optimal model was the Spatial Durbin Model-SDID (SDM-SDID), based on which we conducted estimations and decompositions of global and local policy shock effects.

The results show the following: first, the traditional DID model only estimates the policy effect of the BRI implementation. The BRI policy dummy variable is significant at the 1% level, with a negative coefficient, indicating that the policy effect of the BRI is −0.048. This suggests that the implementation of the BRI is conducive to reducing CO₂ emission intensity in participating countries. Second, in the estimation of the global policy shock effect, the SDID coefficient is significantly negative, indicating that collaborative innovation under the BRI significantly reduces the carbon emission intensity in participating regions or countries. The spatial lag term of the SDID is also significantly negative, showing that there is a spatial spillover effect in carbon reduction. This implies that, compared to countries that have not established close collaborative innovation relationships, countries involved in collaborative innovation with China and other BRI countries see a more pronounced reduction in CO₂ emissions due to policy effects. Specifically, the BRI reduces carbon emission intensity through the investment effect, innovation effect, and trade effect. Third, at the local level, most “SDID_r” policy shock terms are negative, meaning that for most countries in the treatment group, collaborative innovation with China under the BRI plays a role in reducing CO₂ emissions, although the degree of impact varies across regions. The decomposition of local policy shock effects shows that the direct and indirect regression coefficients of collaborative innovation in most countries are negative, indicating a significant carbon reduction effect through collaborative innovation with China under the BRI.

Based on these findings, we recommend that countries strengthen their collaborative innovation mechanisms with other BRI countries. This can be achieved through policy coordination, technology exchange, personnel exchange, and the construction of innovation platforms, forming tighter cooperation networks to achieve broader regional carbon reduction effects. Given the variation in policy effects across different regions, countries should develop more targeted regional policies based on their development levels and carbon emission characteristics. Countries are encouraged to establish customized cooperation models based on their specific conditions, especially providing more policy support and technical assistance to countries or regions with higher carbon emissions to maximize the benefits of collaborative innovation across a wider area. Additionally, China should play a proactive role as a driving force in fostering innovation cooperation among BRI countries.

Although this study attempts to explore the policy effects of collaborative innovation under the BRI on carbon emission reduction, thereby evaluating its contribution to achieving the SDGs, certain limitations remain. First, due to data availability, this study includes 106 countries in the empirical model, which does not cover all participating countries under the BRI framework. This may compromise the representativeness of the sample. From a technical perspective, the spatial effects of the policy are closely tied to the choice of the spatial weight matrix, which may influence the results. Second, this study treats collaborative innovation under the BRI as a form of connectivity to assess its policy effects. However, discussions across different knowledge domains remain insufficient, leaving room for further exploration. Third, the parameter estimates for the localized policy impacts reveal that the direct effects of collaborative innovation in a few countries show positive regression coefficients. This indicates that collaborative innovation under the BRI has not yet contributed to carbon emission reduction in these countries. This can be attributed to their technological development levels. Moreover, the mechanisms through which collaborative innovation influences carbon emission reduction warrant deeper investigation. While empirical research often focuses on the overall effects of policies, it is crucial to pay closer attention to the spatial heterogeneity of localized effects and to refine case-specific investigations.

5. Conclusions

Promoting green and low-carbon development, and building a collaborative ecosystem for achieving carbon peaking and carbon neutrality goals, has become a shared consensus among countries along the Belt and Road Initiative (BRI). From the perspective of policy effects, research on the impact of the BRI on carbon emissions in participating countries remains relatively scarce. Most existing studies analyze the policy effects of the BRI by using its implementation as a time marker. This paper innovatively considers the evolving trend of cooperative relationships among BRI countries, treating collaborative innovation with China as a policy shock to study the environmental effects of cooperation under the BRI.

Unlike existing policy effect analyses that simply compare carbon emission levels in relevant regions before and after the BRI’s implementation, this study introduces lag operators in both time and event dimensions to minimize the influence of natural variations, thus highlighting the impact of policy shocks. Compared to traditional DID model analyses, we more precisely identify the role of BRI-driven collaborative innovation in carbon reduction while also focusing on the spatial effects of this impact. Building on the existing estimates of global policy shock effects, we further decompose the policy shock effects in local regions, demonstrating not only the overall spatial correlations of policy effects but also the spatial heterogeneity in their impacts.

To achieve the SDGs, BRI member countries should emphasize the critical role of innovation capacity in sustainable development. Nations should strengthen scientific research cooperation, enhance environmental protection standards in infrastructure development, and create a favorable environment for green technological innovation. The global cooperative effects of the BRI should also be prioritized. Participating countries should foster deeper economic collaboration and social exchanges, advancing the goals of “policy coordination, infrastructure connectivity, unimpeded trade, financial integration, and people-to-people bonds” to achieve sustainable development. In particular, LDCs should focus on enhancing their ability to attract investment, strengthening innovation cooperation with more advanced regions, and improving their economic resilience, thereby contributing to their green development goals.

In terms of further research, future studies should focus on a more detailed examination of local effects. First, it is important to study the evolution of the collaborative innovation networks among BRI countries, summarizing the differences in the knowledge fields where various countries engage in collaborative innovation. Based on the current state of knowledge cooperation, the key focus should be on which areas of innovation collaboration among BRI countries have the most significant carbon reduction effects, as well as the possible country-specific effects, providing more targeted references for policymakers. Additionally, based on a summary of the mechanisms, future studies should analyze how different countries can integrate more deeply into the BRI collaborative innovation network to achieve more notable economic and environmental benefits.