The Impact of Digital Service Trade on the Carbon Intensity of Well-Being Under Sustainable Development Goals

Abstract

Reducing the carbon intensity of well-being (CIWB) is essential for advancing environmental sustainability and socio-economic development. The expansion of digital service trade has emerged as a novel engine of global economic growth and a promising pathway for pollution reduction and carbon mitigation. This study investigates the nonlinear impact of digital service trade on CIWB, identifying an inverted U-shaped relationship—initially increasing CIWB, then reducing it beyond a certain threshold. In the financial digital service trade sector, this effect is mediated by energy structure transition, whereas in the technology-intensive sector, it is driven by green technological innovation. In contrast, digital service trade in the insurance, pension, and audiovisual sectors directly suppresses CIWB. Moreover, rising public environmental awareness helps leverage and strengthen the inhibitory effect of digital service trade on CIWB. Regionally, except for North America (which displays a consistently inhibitory effect), Asia, Africa, Europe, and Oceania reflect patterns similar to the overall sample. In regions with higher economic and internet development levels, the inverted U-shaped curve is steeper, and its turning point is located further to the left. Temporally, the relationship mirrors the full-sample patterns prior to the enforcement of the Paris Agreement, while an inhibitory effect emerges afterward. These findings offer policy implications for achieving the United Nations’ 2030 Sustainable Development Goals.

1. Introduction

Reducing the carbon intensity of well-being (CIWB) is essential for advancing environmental sustainability and socio-economic development. In recent years, global well-being has steadily improved alongside the ongoing development of the world economy. According to the Human Development Report 2023/2024 released by the United Nations, the global Human Development Index has risen from 0.601 in 1990 to 0.739 in 2022, with a forecasted potential to exceed the threshold for a very high Human Development Index (0.800) by 2030 [1]. However, the continued growth of the world economy has come at a significant environmental cost. As urbanization and industrialization processes increase the demand for fossil fuels, emissions of greenhouse gases and air pollutants have also risen [2], exacerbating global climate change and environmental pollution—issues that have become a major focus of the scientific community. According to the Global Carbon Budget database, global CO₂ emissions from fossil fuels were approximately 37.94 billion tons in 2022, an increase of about 477 million tons from the previous year, reflecting a growth rate of about 1.27% [3]. Global climate change and environmental pollution severely threaten atmospheric systems, ecological balance, and well-being. CO₂-dominated greenhouse gas emissions have led to rising global temperatures, increased frequency of extreme weather (such as heatwaves, heavy rainfall, and droughts), and habitat destruction for flora and fauna (such as polar ice melt and rising sea levels) [4,5]. Additionally, air pollution severely threatens human health, causing respiratory and cardiovascular diseases [6] and jeopardizing global environmental sustainability, biodiversity, and human survival. Against this dire backdrop, the available carbon space for human society and economic development is becoming increasingly limited. Many developed countries have already occupied a significant share of this carbon space, while developing countries must balance carbon reduction efficiency with equity [7], which is a critical factor constraining the sustainable development of global well-being. The ongoing rise in CO₂ emissions underscores a critical shortfall in coordinated energy efficiency and emission mitigation strategies, posing a significant obstacle to realizing the United Nations’ 2030 Sustainable Development Goals (SDGs). CIWB—the level of CO₂ emissions generated per unit of well-being—illustrates a fundamental dilemma: while economic growth remains essential for boosting human welfare, such progress becomes unsustainable if it increases carbon output. Therefore, one of the central challenges of global sustainability is to harmonize economic development with equitable, low-carbon improvements in well-being.

Meanwhile, the expansion of digital service trade has become a novel engine of global economic growth and a promising pathway for pollution reduction and carbon mitigation. Digital service trade refers to all cross-border service trade digitally delivered. Recently, digital service trade has become the most rapidly expanding trade segment, emerging as a dynamic and transformative force in the world economy. The IMF, OECD, UNCTAD, World Bank, and WTO in Digital Trade for Development reported that global digital service exports reached USD 3.82 trillion in 2022, representing 54% of service exports and 12% of total exports. During 2005–2022, digital service trade grew nearly fourfold, expanding at an average annual rate of 8.1%, which exceeded the growth of goods exports (5.6%) and other service exports (4.2%) [8]. While driving global economic growth, digital service trade also plays a growing role in curbing pollution and lowering CO₂ output, emerging as a novel solution for tackling climate change and ecological degradation. Digital service trade can reduce overall carbon emissions by substituting digital alternatives for physical goods and traditional services, as this transition reduces the demand for transportation and manufacturing, which are major sources of greenhouse gases [9]. For example, online education, virtual meetings, and remote medical consultations reduce the carbon footprint of studying abroad, international business travel, and cross-border medical treatments. Additionally, digital service trade can enhance energy efficiency through advanced data analysis and intelligent tools. To illustrate, intelligent power grids, building information modelling, and smart home solutions can optimize energy distribution and reduce energy waste, thereby lowering overall carbon emissions. Furthermore, digital service trade can optimize the energy structure by promoting the demand for clean energy. Digital services—including AI, cloud storage, and cloud computing—are key enablers of green energy investment. Because data centers and computing centers have enormous electricity demands, service providers are turning to renewable energy—including solar, wind, and hydrogen—to meet substantial energy needs and reduce costs, thus lowering overall CO₂ emissions. The continued expansion of digital service trade drives rapid global economic growth and plays a leading role in addressing climate change and mitigating ecological degradation.

As a comprehensive indicator for measuring environment and well-being, studying CIWB and its driving factors is crucial for advancing sustainable development. Nevertheless, a review of the existing research reveals some gaps in the literature. Few studies have directly examined the effect of digital service trade—a new factor—on CIWB. Most of the existing literature focuses on factors such as economic and social development (Jorgenson, 2014; Wang et al., 2022; Givens, 2015) [10,11,12], institutions (Mayer, 2017; Briscoe et al., 2021) [13,14], policies (Herziger et al., 2020; Ergas et al., 2021) [15,16], education (Barak et al., 2024; Kelly, 2020) [17,18], and urbanization (Wang et al., 2022; McGee et al., 2017; Sharmin et al., 2021) [19,20,21]. In light of this, the present study will theoretically and empirically explore the effect of digital service trade on CIWB and its mechanisms. This paper first proposes a theoretical hypothesis that digital service trade may have a nonlinear impact on CIWB, considering the current global context of climate change and environmental pollution, focusing on energy structure transition, green technology innovation, and public environmental awareness. It then uses balanced panel data from 152 countries (regions) for 2005–2022 to construct a nonlinear regression model, combined with U test and plotting predictive margins, to empirically analyze the effect and mechanisms of digital service trade on CIWB. Robustness checks are performed using methods such as two-stage least squares, difference moment estimation, and system moment estimation.

This study seeks to fill the identified gaps in the above research fields, with its main contributions including the following points: (1) Expansion of the study on factors affecting CIWB. While previous research has predominantly examined the impact of factors such as economic and social development, institutions, policies, education, and urbanization on CIWB, this study further explores the impact of digital service trade, a new factor, on CIWB. This provides a valuable addition to the existing research. (2) Theoretical mechanisms linking digital service trade to CIWB. The theoretical pathways through which digital service trade affects CIWB are clarified. The theoretical analysis specifies the overall nonlinear effect, the heterogeneous impacts of different categories of digital service trade, their transmission channels, and the nonlinear moderating role of public environmental awareness. (3) Utilization of recent and comprehensive authoritative statistical data. The current literature in this field often relies on outdated data or lacks comprehensive coverage of countries or regions. Hence, this study compares sample sizes and time spans from various authoritative databases, ultimately matching and consolidating balanced panel data from 152 countries or regions from 2005 to 2022. This methodology improves the timeliness and comprehensiveness of the study, offering a more thorough reflection of trends in CIWB and representing the latest developments in this area. (4) Thorough examination of the nonlinear impact mechanism of digital service trade on CIWB. This study discovers an inverted U-shaped effect of digital service trade on CIWB through theoretical and empirical analysis. It further examines the intrinsic transmission mechanisms and external moderation mechanisms that contribute to this nonlinear relationship. This deepens the understanding of the factors affecting CIWB and enriches the study of nonlinear impact mechanisms in this field. (5) Extensive analysis of heterogeneity in the research topic. This study conducts a detailed heterogeneity analysis based on factors including geographic location, economic and internet development levels, and the effectiveness of the Paris Agreement. This comprehensive examination of the heterogeneous effects of digital service trade on CIWB provides a nuanced understanding of the variations in impact.

2. Literature Review

Recent research on CIWB has revealed numerous and complex influencing factors. These factors involve not only economic and social development, but also institutional, policy, educational, and urbanization aspects. Economic development is generally considered a significant determinant of carbon intensity. Some studies suggest that economic growth is often accompanied by increased carbon emissions, being closely linked to increases in consumption and production activities, although it also typically brings improvements in well-being (Jorgenson, 2014; Wang et al., 2022) [10,11]. Particularly in developing countries, accelerated economic development and urbanization have a marked influence on CIWB (Givens, 2015; Wang et al., 2022) [12,19]. These regions often face issues such as low energy efficiency and inadequate pollution control during economic transformation, leading to increased CIWB (Li et al., 2019; Wang et al., 2022) [22,23]. However, economic development is not always positively correlated with carbon intensity. Structural changes in economic development, especially the shift toward service and high-tech industries, can reduce CIWB (Jorgenson & Givens, 2015; Sweidan, 2018) [24,25]. Additionally, institutional and policy factors also play a crucial role in carbon intensity. Mayer (2017) points out that democratic institutions can promote stricter environmental protection policies, thereby reducing CIWB [13]. Ergas et al. (2021) find that gender climate plays a critical role in implementing climate change policies, with gender-sensitive policies significantly enhancing the effectiveness of environmental protection, thus affecting CIWB [16]. Furthermore, Barak et al. (2024) found through testing the Environmental Kuznets Curve hypothesis that a country’s environmental policies and economic development have a complex, nonlinear relationship with CIWB [17]. The impact of education is also a factor that cannot be ignored. Studies show that higher education levels are associated with lower CIWB, as higher education often correlates with increased environmental awareness and more energy-efficient lifestyles (Kelly, 2020) [18]. Meanwhile, social systems and welfare policies can influence CIWB by improving social equity and resource distribution. For example, Herziger et al. (2020) found that good social welfare policies can mitigate the pressure of economic growth on carbon emissions, achieving a dual benefit of economic growth and ecological conservation [15]. Briscoe et al. (2021) indicate that social intersectional indicators, such as race and gender, can affect the fairness and effectiveness of environmental policies, further supporting the view that improving social systems can reduce CIWB [14]. The effect of urbanization on carbon intensity is even more complex. From one perspective, urbanization may lead to a rise in CIWB, as energy demand and transportation discharges in urban areas are typically higher (McGee et al., 2017) [20]. From another perspective, urbanization may also lower CIWB by enhancing energy efficiency and driving green technology innovations (Sharmin et al., 2021) [21]. In summary, CIWB is influenced by various factors, including economic and social development, education, urbanization, institutions, and policies. A comprehensive exploration of these influencing factors is crucial for formulating effective economic and environmental policies and balancing economic development with environmental protection.

There is scarce literature directly analyzing the effect of digital service trade on CIWB. However, a few relevant studies have explored this issue from the broader perspective of international trade. For instance, Wang et al. (2023) argue that international trade reduces CIWB among high- and upper-middle-income economies, but increases it among lower-middle- and low-income economies [26]. Similarly, Xu et al. (2020) found that international trade facilitates the achievement of SDGs, improving SDG scores for most developed countries but lowering them for many developing countries [27]. Additionally, there is research on the environmental impacts of digital service trade. Some studies indicate that digital service trade can nonlinearly affect CO₂ emissions, with the scale of digital service trade having a bidirectional effect on CO₂ emissions depending on countries’ technological levels and policy environments (Chen & Jiang, 2023) [28]. For example, digital service trade liberalization can enhance carbon efficiency, especially in high-income countries, by reducing their dependence on traditional energy through the introduction digital technologies (Fang et al., 2023) [29]. Ji et al. (2023) and Wang et al. (2023) investigated the impact of digital trade on regional carbon emissions, revealing that this impact varies across regions due to differences in industrial structure, energy efficiency, consumption patterns, and policy orientation [30,31]. Li et al. (2024) further confirmed these findings, indicating significant differences in the effects of digital trade on CO₂ emissions during the urbanization process [32]. Moreover, Liang and Hao (2023) discussed how service trade can effectively promote carbon emission reductions. It is suggested that optimizing the allocation of service trade can significantly lower overall carbon intensity [33]. Ma et al. (2023) focused on participation in digital global value chains, showing that such participation helps improve resource efficiency and thus reduces CO₂ emissions [34]. Conversely, Wang et al. (2023) pointed out that digital trade restrictions negatively impact green transformation in low-income countries, where the anticipated carbon reduction effects are not achieved [35]. He and Xiang (2024) demonstrated that combining trade liberalization and digital trade can promote carbon reduction, although the results vary by country and region [36]. Zhang et al. (2022) explored the interaction between financial development and digital trade in shaping the total volume and the structure of carbon emissions [37]. In summary, although research on the effect of digital service trade on CIWB is still limited, the existing literature suggests that its effects differ across countries and regions, influenced by factors such as trade scale, energy structure, technological innovation, and dependence on energy and technology. Therefore, to comprehensively understand the effect of digital service trade on CIWB, further research is needed to explore its specific manifestations and mechanisms in different contexts, enabling the formulation of effective policies to enhance well-being and reduce global CO₂ emissions.

Compared to the existing literature, this study is the first to examine the impact of digital service trade on CIWB, thereby addressing a gap in the research on the determinants of CIWB. To further enhance analytical precision, a nonlinear regression model has been developed to explore whether the influence of digital service trade on CIWB exhibits nonlinear characteristics, along with the underlying mechanisms behind such patterns. This modelling choice is grounded in the recognition that the effect of digital service trade on CIWB may be inherently complex and multifaceted. Employing a linear regression model alone could overlook important nonlinear relationships, even when statistically significant results are obtained. The nonlinear regression model employed in this study differs fundamentally from traditional linear regression models. While linear models are limited to fitting straight-line relationships, they often produce significant estimation bias when the data exhibit a “rise-then-decline” pattern. In contrast, nonlinear regression models can capture underlying curvilinear trends with greater precision. This improves the accuracy of model fitting and the ability to interpret complex dynamics, thereby offering more robust policy insights to support achieving the SDGs.

3. Theoretical Analysis and Hypothesis Development

During its initial phase, the expansion of digital service trade often leads to an increase in CIWB. This is primarily because the initial development of digital service trade requires substantial support from digital infrastructure, such as data centers, computing centers, and 5G base stations. Establishing this digital infrastructure and ensuring its functioning typically involves significant energy consumption, predominantly from fossil fuels, which inevitably leads to increased carbon emissions. Moreover, the economic benefits during the early stages are often overshadowed by energy consumption [38], potentially resulting in a rise in CIWB. In addition to digital infrastructure, the production, use, and replacement of related digital devices also contribute to higher energy consumption, further driving up carbon emissions [39]. Increased carbon emissions inevitably cause various ecological degradation and climate change issues, negatively impacting human health and well-being and ultimately increasing CIWB. Once digital service trade reaches a relatively mature stage, its mitigating effect on CIWB begins to manifest. Over time, the widespread use of digital technologies can improve production efficiency, drive green innovation, and reduce energy consumption [40]. Technological advancements enhance the efficiency of data processing and storage, reducing energy demand. Many data centers, computing centers, and 5G base stations are also transitioning to renewable energy sources, further reducing carbon emissions. The growth of the digital service sector, alongside an increased share of the service industry in the national economy, supports the green transformation of industries, thereby reducing CIWB overall. Furthermore, the emergence of digital platforms has enabled the development of the sharing economy model [41]. Services such as shared mobility, cloud computing, and cloud storage reduce reliance on physical resources, decreasing carbon emissions while facilitating residents’ daily lives, thus lowering CIWB.

Digital service trade is an important branch of international trade in statistical terms, comprising thirteen detailed categories [42]. Although digital service trade and CIWB may generally exhibit an inverted U-shaped relationship, different categories of digital service trade could have varying impacts on CIWB. Intuitively, insurance and pension services, as well as audiovisual and related services, have a notable inhibitory effect on CIWB. Insurance and pension services can reduce the financial burden of high medical expenses due to illnesses by providing pensions and insurance products related to retirement, such as life insurance, health insurance, and medical insurance [43]. This allows residents to receive timely and adequate treatment, thereby extending average life expectancy. Additionally, by offering green insurance products such as environmental pollution liability insurance, green building insurance, and clean energy insurance [44], these services can reduce the environmental costs for businesses during low-carbon transitions, thereby decreasing carbon emissions. Thus, insurance and pension services clearly reduce CIWB. According to the 2010 Extended Balance of Payments Services Classification (EBOPS-2010), audiovisual and related services fall under personal, cultural, and recreational services. This category of digital service trade itself generates minimal environmental costs and can enhance individual well-being and social welfare, thereby evidently reducing CIWB.

Secondly, digital service trade in the financial sector during the early stages may hinder the transition of energy structure. This is because early digital financial services tend to invest in projects with high short-term returns, with lower risks and initial costs, whereas renewable energy projects typically require long-term investments with higher initial costs and risks [45]. As a result, funds may be directed more towards traditional energy projects rather than renewable ones. Additionally, the early development of digital financial services requires substantial support from digital infrastructure, which often relies on traditional energy sources. This reliance exacerbates fossil fuel consumption, impedes the energy structure transition, generates significant carbon emissions, and increases CIWB. However, as digital financial service trade matures, big data and artificial intelligence may be used to precisely assess energy market trends, as well as the returns, costs, and risks of projects. These analyses help investors make long-term decisions and optimize investment portfolios [46], making them more inclined to invest in renewable energy, which offers higher long-term returns despite higher initial costs and risks. Furthermore, the emergence of cross-border digital financial platforms supports the trading and circulation of green financial products, including green bonds, stocks, and funds, which aim to support sustainable development and promote investment in renewable energy projects [47].

Finally, intellectual property charges not included elsewhere, telecommunications, computer and information services, research and development services, and professional and management consulting services, as well as other technology-intensive digital service trades, may initially suppress green technology innovation. This is because while green technology innovation offers environmental benefits and long-term returns, it involves higher costs and risks. During the initial phase of technology-intensive digital service trade, companies may adopt a more cautious approach, focusing on traditional technology R&D with lower costs, risks and quicker returns, rather than investing in green technology innovation, which has higher costs and risks but offers long-term benefits. Even if companies enter the green technology field at this stage, they may still prefer to directly introduce green technology patents rather than engage in green technology R&D due to the longer development cycles, higher investments, and greater risks associated with green technologies compared to traditional ones [48]. In the short term, this approach may make achieving large-scale green technology innovations difficult, thereby impeding energy saving and emission reduction and increasing CIWB. However, as the scale of technology-intensive digital service trade grows, international collaboration and knowledge exchange in green technology will become more frequent. Such collaboration and exchange will accelerate the updating and iteration of green technologies [49], potentially reducing the development cycle, risks, and costs associated with green technology. Consequently, green technologies’ environmental and economic benefits will rise, fostering green technology innovation and promoting energy conservation and emission reduction, thereby reducing CIWB.

The previous discussion theoretically analyzed why the relationship between digital service trade and CIWB exhibits an inverted U shape. The next crucial step is to explore how to maximize and enhance the mitigating effect of digital service trade on CIWB. In recent years, rising public environmental awareness has driven a surging preference for low-carbon and environmentally friendly lifestyles and production practices [50]. This shift has led governments, businesses, and consumers to emphasize green development, promoting green consumption and production. Increased environmental awareness typically leads consumers to prefer products and services with a lower carbon footprint. As a low-carbon service form, digital service trade is more readily influenced by public environmental awareness. Consumer preferences for environmentally friendly options in digital service trade help reduce carbon emissions and positively impact social welfare, thus reducing CIWB. Enhanced environmental awareness means governments will likely implement stricter environmental policies, regulations, and standards [51], further expanding the scale of environmentally beneficial digital service trade. Increased environmental awareness may also lead companies to emphasize environmental benefits and social responsibility simultaneously [52], rather than solely focusing on economic gains. This shift will drive digital service trade to adopt more eco-friendly energy and technologies and encourage greater investment in energy structure transition and green technology innovation projects. Thus, overall CO₂ emissions will be lowered, and residents’ environments will be more livable, which in turn will help lower CIWB. In summary, heightened environmental awareness can increase public recognition of carbon footprints, make consumers more willing to support clean and digital import–export services, encourage enterprises to engage in digital service trade within the renewable energy and green technology patent sectors, and prompt governments to refine environmental regulatory systems continuously. This collective effort will foster green, low-carbon lifestyles and production methods, continuously scale up digital service trade, and further reduce CIWB.

Drawing on the above analysis, this study proposes the following hypotheses.

H₁: 

Overall, digital service trade exerts an inverted U-shaped impact on CIWB, initially raising and then lowering it.

H₂: 

Digital service trade in the insurance, pension, audiovisual, and related sectors directly inhibits CIWB.

H₃: 

The financial digital service trade exerts an inverted U-shaped effect on CIWB by influencing the energy structure transition.

H₄: 

Technology-intensive digital service trade exerts an inverted U-shaped effect on CIWB by influencing green technology innovation.

H₅: 

Improving public environmental awareness helps leverage and strengthen the inhibitory effect of digital service trade on CIWB.

4. Description of Empirical Models, Variables, and Data

4.1. Empirical Modelling

To test whether Hypothesis 1 holds, this study references Blanchflower (2021) [53], Enders et al. (2012) [54], and Aghion et al. (2005) [55] by introducing a squared term to capture potential nonlinear effects within the estimation framework. By observing the coefficients of the independent variable and its squared term and calculating the turning point of the curve, we can determine if there is an inverted U-shaped relationship. This approach is a classic method for testing nonlinear relationships. As shown in Equation (1), a fixed-effects model with the squared term of digital service trade is introduced as the benchmark regression model, and robustness checks and heterogeneity analyses are conducted based on this model. In the equation, i stands for country or region, t stands for year, CIWB is the carbon intensity of well-being, ${\alpha }_0$ is a constant term, DST is digital service trade, ${\alpha }_1$ is the coefficient of digital service trade, ${DST}^{ 2}$ is the squared term of digital service trade, ${\alpha }_2$ is the squared term coefficient of digital service trade, $X_{it}^{\prime }$ is the control variable in the form of a row vector, $\beta$ is the coefficient of the control variable in the form of a column vector, ${\mu }_i$ is region fixed effects, ${\lambda }_t$ is year fixed effects, and ${\epsilon }_{it}$ is an error term.

The coefficients ${\alpha }_1$ and ${\alpha }_2$ constitute the central parameters under examination. Evidence of a nonlinear, inverted U-shaped association between digital service trade and CIWB emerges when ${\alpha }_1$ is statistically greater than zero, whereas ${\alpha }_2$ holds a significantly negative value. In contrast, a U-shaped relationship is indicated if ${\alpha }_1$ is significantly negative and ${\alpha }_2$ is significantly positive. A significant negative ${\alpha }_1$ with an insignificant ${\alpha }_2$ suggests a negative correlation between digital service trade and CIWB, while a significant positive ${\alpha }_1$ with an insignificant ${\alpha }_2$ points to a positive correlation.

$$CIW{B}_{it}={\alpha }_0+{\alpha }_{1}DS{T}_{it}+{\alpha }_{2}DS{T}_{it}^2+X_{it}^{\prime }\beta +{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(1)

Meanwhile, to evaluate the stability of the baseline estimation, several alternative specifications are employed, including independent variables lagged one period, two-stage least squares, a dynamic panel model, and 1% two-tail winsorizing. The specification incorporating the lagged independent variable is presented in Equation (2). In the equation, ${\mathit{DST}}_{i,t-1}$ and ${\mathit{DST}}_{i,t-1}^{ 2}$ denote the lagged value and its squared term of DST. The corresponding dynamic panel formulation, detailed in Equation (3), incorporates ${\mathit{CIWB}}_{i,t-1}$ and ${\mathit{CIWB}}_{i,t-2}$ as one- and two-period lags of CIWB, respectively. Other symbols are defined above. The model settings for other robustness-checking methods are consistent with the benchmark regression model.

$$CIW{B}_{it}={\alpha }_0+{\alpha }_{1}DS{T}_{i,t-1}+{\alpha }_{2}DS{T}_{i,t-1}^2+X_{it}^{\prime }\beta +{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(2)

$$CIW{B}_{it}={\theta }_0+{\theta }_{1}CIW{B}_{i,t-1}+{\theta }_{1}CIW{B}_{i,t-2}+{\theta }_{3}DS{T}_{it}+{\theta }_{4}DS{T}_{it}^2+X_{it}^{\prime }\beta +{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(3)

In addition, to test whether Hypotheses 3 and 4 hold, a transmission framework is developed based on the methodology of Bartram et al. (2024) [56], aiming to identify the channels through which digital service trade influences CIWB. As shown in Equations (4) and (5), the method first regresses the independent variable on the mechanism variables (Mech) and then regresses the dependent variable on the mechanism variables (Mech), thereby conducting the transmission mechanism analysis. In the equation, Mech is the transmission mechanism variable, including energy structure transition (EST) and green technology innovation (GTI). The other symbols are defined above.

$$Mec{h}_{it}={\gamma }_0+{\gamma }_{1}DS{T}_{it}+{\gamma }_{2}DS{T}_{it}^2+X_{it}^{\prime }\beta +{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(4)

$$CIW{B}_{it}={\delta }_0+{\delta }_{1}Mec{h}_{it}+X_{it}^{\prime }\beta +{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(5)

Finally, to test whether Hypothesis 5 holds, this study constructs a nonlinear moderation mechanism model based on the method of Haans et al. (2016) [57]. As shown in Equation (6), the model incorporates public environmental awareness (PEA), along with its interaction terms with both DST and its squared value, as extensions to the baseline regression framework. In the equation, PEA is public environmental awareness. The other symbols are defined above.

$$\begin{array}{l}CIW{B}_{it}={\eta }_0+{\eta }_{1}DS{T}_{it}+{\eta }_{2}DS{T}_{it}^2+{\eta }_{3}PE{A}_{it}+{\eta }_{4}PE{A}_{it}\times DS{T}_{it}+{\eta }_{5}PE{A}_{it}\times DS{T}_{it}^2\\ +X_{it}^{\prime }\beta +{\mu }_i+{\lambda }_t+{\epsilon }_{it}\end{array}$$

(6)

The subsequent discussion will present detailed descriptions of the measurements for the aforementioned variables.

4.2. Variable Measurements

(1)

Carbon Intensity of Well-Being(CIWB). Based on the research by Barak et al. (2024) [17], Wang et al. (2023) [26], and Li et al. (2019) [22], this study measures CIWB using the ratio of per capita CO₂ emissions to average life expectancy at birth. As shown in Equation (7), Emissions_it denotes the per capita CO₂ emissions of country i in year t, while life_it represents the average life expectancy at birth for residents of country i in year t. CIWB_it refers to the CIWB of country i in year t, which measures the per capita CO₂ emissions generated per unit of well-being. Human activities involving fossil fuel consumption during economic development lead to increased carbon emissions. Concurrently, while increasing carbon emissions, economic development also enhances well-being. Economic development is commonly regarded as a key driver of improvements in societal welfare. Effective sustainable development policies require reducing CIWB [10]. The numerator, per capita CO₂ emissions, is sourced from the Global Carbon Budget (GCB) database [3]. According to GCB estimates, greenhouse gas (GHG) emissions from energy use (fossil fuel combustion) account for over three-quarters of global GHG emissions. Therefore, the GCB database reports per capita CO₂ emissions from fossil fuel use, measured in tons. Emissions from non-fossil fuel sources are not included in the GCB statistics. The denominator, average life expectancy at birth, measured in years, is sourced from the WDI database (https://databank.worldbank.org/reports.aspx?source=World-Development-Indicators, accessed on 15 January 2025). Next, this section presents the spatial distribution of CIWB, measured using the methods above, as shown in Figure 1. Overall, CIWB in most countries in 2022 decreased compared to 2005. This indicates that these countries have made progress in reducing carbon emissions while maintaining a high level of well-being. Regionally, North America and Oceania still exhibit high levels of CIWB, although there was a decline from 2005 to 2022. Most European countries experienced a significant reduction in CIWB in 2022 compared to 2005, while Russia showed a noticeable increase. From 2005 to 2022, the trends in CIWB in South America, Asia, and Africa vary significantly; China and several neighboring countries continue to maintain high CIWB. This suggests that although economic development has improved well-being, it has also led to higher carbon emissions, especially in regions experiencing rapid industrialization. While CIWB decreased in most South American, Asian, and African countries during this period, some countries, such as Suriname and Saudi Arabia, showed a notable increase.

$$CIW{B}_{it}={\displaystyle \frac{Emission{s}_{it}}{Lif{e}_{it}}}$$

(7)

(2)

Digital service trade (DST). This paper measures DST according to the statistical definitions outlined in the Handbook on Measuring Digital Trade (second edition), jointly published by IMF, OECD, UN, and WTO. According to the handbook, DST includes the following 13 service categories from EBOPS-2010: insurance and pension services; financial services; intellectual property charges not included elsewhere; telecommunications, computer, and information services; research and development services; professional and management consulting services; architectural, engineering, scientific, and other technical services; trade-related services; other business services not included elsewhere; audiovisual and related services; health services; educational services; and heritage and recreational services [42]. The data were obtained from the UNCTAD database (https://unctadstat.unctad.org/datacentre/, accessed on 15 January 2025). Due to excessive missing values in architectural, engineering, scientific, and other technical services, trade-related services, other business services not included elsewhere, health services, educational services, and heritage and recreational services, these six categories were excluded. Consequently, the analysis retained seven categories of digital service trade: insurance and pension services (DST_Ins), financial services (DST_Fin), intellectual property charges not included elsewhere (DST_Int), telecommunications, computer, and information services (DST_Tel), research and development services (DST_Res), professional and management consulting services (DST_Pro), and audiovisual and related services (DST_Aud). Summing these categories provides the total digital service trade value, measured in millions of USD.

(3)

Other variables. Besides the aforementioned core variables, transmission mechanism variables, moderating mechanism variables, and control variables are also considered in this study. The transmission mechanism variables include energy structure transition (EST) and green technological innovation (GTI). Energy structure transition (EST) is measured by the proportion of renewable energy consumption in total final energy consumption (unit: %), with data from the WDI database. Green technological innovation (GTI) is assessed using patents of technology development in climate change mitigation (unit: items), with data from the OECD database (https://data-explorer.oecd.org/, accessed on 15 January 2025). The moderating mechanism variable is public environmental awareness (PEA), which is measured by the proportion of the population with primary reliance on clean fuels and technology (unit: %), with data obtained from the UNSDG database (https://unstats.un.org/sdgs/dataportal/database, accessed on 15 January 2025). The control variables include the level of economic development (PGDP), urbanization rate (City), degree of openness (FDI), energy intensity (Prima), and industrial structure (Indus). The level of economic development (PGDP) is measured by per capita GDP (unit: USD); urbanization rate (City) is measured by the proportion of urban population in the total population (unit: %); degree of openness (FDI) is assessed as the share of net inflows of foreign direct investment in GDP (unit: %); energy intensity (Prima) is measured by the energy intensity level of primary energy (unit: megajoules per USD); and industrial structure (Indus) is measured by the ratio of value added in the service sector to that in the industrial sector (unitless). Data for the control variables are drawn from the WDI database.

4.3. Data Sources and Processing

This study draws on data sourced from the GCB, WDI, UNCTAD, OECD, and UNSDG databases. Based on data availability and cross-database matching, this study ultimately retained balanced panel data for 152 economies from 2005 to 2022, employing linear interpolation and regression imputation methods to handle a few missing values. To control for price-level distortions, relevant variables are adjusted to constant 2005 values. To reduce data volatility and address heteroscedasticity, this paper transforms the relevant variables into their natural logarithms. Data processing and empirical analysis are conducted using Stata/MP 18.0 software, with summary metrics of the main variables in

Table 1. Results of descriptive statistics.

Variables	N	Mean	Std. Dev.	Min	Max
CIWB	2224	0.0170	0.0183	0.0002	0.1511
DST	2224	7.2853	2.5687	0.3920	13.4171
DST_Ins	2224	5.0072	2.3339	0.0000	11.1893
DST_Fin	2224	5.0724	2.7896	0.0000	12.0155
DST_Int	2224	4.4370	3.1648	0.0000	11.7949
DST_Tel	2224	5.9851	2.4565	0.0000	11.9795
DST_Res	2224	3.2989	3.2664	0.0000	11.7368
DST_Pro	2224	5.3540	2.9998	0.0000	11.8879
DST_Aud	2224	2.8927	2.7781	0.0000	10.3853
PGDP	2224	8.8033	1.4187	5.5735	11.7657
City	2224	59.5059	22.2989	11.4820	100.0000
FDI	2224	5.9030	22.8275	−394.4716	449.0828
Prima	2224	4.5770	2.6641	0.3500	24.8200
Indus	2224	2.9129	2.3429	0.4896	34.1286
EST	2224	29.5753	26.2176	0.0000	94.7000
GTI	2224	2.1890	2.7653	−18.0218	9.6860
PEA	2224	70.665	36.0362	0.1000	100.0000

. Furthermore, the results of the multicollinearity test indicate that the highest VIF among the variables is 3.07, well below the conventional threshold of 10, suggesting that serious multicollinearity is not present.

5. Empirical Analyses

5.1. Benchmark Regression Analysis

In panel regression analysis, we can use the Chow and Hausman Test to validate the appropriateness of choosing a fixed-effects model. The Chow Test rejects the null hypothesis that “intercepts are equal across all groups”, indicating significant individual differences among samples, which makes a mixed model inappropriate. The results of the Hausman test indicate a rejection of the null hypothesis, which assumes no correlation between the intercept and explanatory variables. This outcome favors the use of a fixed-effects specification over its random-effects counterpart. Additionally, we should pay attention to issues like heteroscedasticity, autocorrelation, and cross-sectional correlation. Due to the robustness of Driscoll–Kraay standard errors in heteroskedasticity, autocorrelation, and cross-sectional dependence, this study primarily employs Driscoll–Kraay standard errors for estimation. Additionally, to enhance the conclusions’ robustness, this study also uses country (or region)-level cluster-robust standard errors and heteroskedasticity-robust standard errors as references.

Table 2. Results of benchmark regression.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	CIWB	CIWB	CIWB	CIWB	CIWB	CIWB
DST	−0.0005 ** (0.0002)	0.0047 *** (0.0006)	−0.0005 (0.0004)	0.0047 *** (0.0013)	−0.0005 ** (0.0002)	0.0047 *** (0.0007)
DST×DST		−0.000434 *** (0.000045)		−0.0004 *** (0.0001)		−0.0004 *** (0.0001)
PGDP	0.0183 *** (0.0011)	0.0192 *** (0.0011)	0.0183 *** (0.0027)	0.0192 *** (0.0027)	0.0183 *** (0.0012)	0.0192 *** (0.0011)
City	0.000342 *** (0.000038)	0.000190 *** (0.000039)	0.0003 *** (0.0001)	0.0002 ** (0.0001)	0.000342 *** (0.000038)	0.000190 *** (0.000043)
FDI	0.000015 ** (0.000007)	0.000014 ** (0.000007)	0.000015 *** (0.000005)	0.000014 *** (0.000005)	0.000015 ** (0.000006)	0.000014 ** (0.000006)
Prima	0.0019 *** (0.0002)	0.0019 *** (0.0002)	0.0019 *** (0.0004)	0.0019 *** (0.0004)	0.0019 *** (0.0003)	0.0019 *** (0.0002)
Indus	−0.000014 (0.000128)	−0.000046 (0.000120)	−0.000014 (0.000168)	−0.000046 (0.000167)	−0.000014 (0.000083)	−0.000046 (0.000070)
RFE	Y	Y	Y	Y	Y	Y
YFE	Y	Y	Y	Y	Y	Y
Constant	−0.1694 *** (0.0109)	−0.1804 *** (0.0110)	−0.1694 *** (0.0258)	−0.1804 *** (0.0259)	−0.1651 *** (0.0099)	−0.1771 *** (0.0096)
Chow Test	142.44 [<0.0001]	165.77 [<0.0001]	142.44 [<0.0001]	165.77 [<0.0001]	142.44 [<0.0001]	165.77 [<0.0001]
Hausman Test	52.24 [<0.0001]	108.70 [<0.0001]	52.24 [<0.0001]	108.70 [<0.0001]	52.24 [<0.0001]	108.70 [<0.0001]
N	2224	2224	2224	2224	2224	2224
Within R2	0.3096	0.3535	0.3096	0.3535	0.3772	0.4168

Note: ** and *** indicate significance at the 5% and 1% levels, respectively. Standard errors are reported in parentheses, and p-values are shown in brackets. RFE = region fixed effects; YFE = year fixed effects; Y = included. The Chow Test reports F-statistics, while the Hausman Test reports Chi-squared statistics.

reports the regression estimates using different types of standard error corrections. Columns (1) and (2) apply heteroskedasticity-robust standard errors, (3) and (4) adopt cluster-robust standard errors at the region level, and (5) and (6) employ Driscoll–Kraay standard errors. Columns (1), (3), and (5) report linear models excluding the squared term of DST. In this setup, the estimated coefficients for digital service trade are significantly negative at the 5% level in columns (1) and (5), but statistically insignificant in column (3), suggesting that digital service trade may reduce CIWB. Columns (2), (4), and (6) display the nonlinear specifications incorporating the squared term of DST. In all three cases, the squared term is negatively signed and highly significant at the 1% threshold, while the linear term shows a strong positive association at the same significance level. These findings suggest a nonlinear association linking digital service trade with CIWB. A comparison of the linear and nonlinear models reveals a substantial improvement in coefficient significance and a noticeably higher R-squared value within the nonlinear models. The link between digital service trade and CIWB comprises linear and nonlinear components, while the predominant relationship is negative over a broad range. However, there is a turning point where, if the value of digital service trade is on the left side of this point, the relationship turns positive. According to Equation (1), this turning point is given by the first-order condition of Equation (1), and the value of DST at this turning point, calculated from column (6), is 5.4211, which lies between the upper and lower limits. Therefore, the relationship between DST and CIWB is indeed an inverted U-shaped nonlinear relationship.

Moreover, the economic significance of the coefficients warrants further discussion. Unlike in linear regression models, where the marginal effect of an independent variable on the dependent variable is constant, the marginal effect in nonlinear regression models varies depending on the value of the explanatory variable itself. Specifically, the marginal effect of DST on CIWB can be obtained by taking the first derivative of Equation (1) with respect to DST.

$${\alpha }_1+2{\alpha }_{2}DST$$

(8)

By substituting the coefficient from column (6) into Equation (8), the marginal effect of DST on CIWB is derived as 0.0047–0.0008 × DST. This indicates that the effect of DST on CIWB is not constant, but varies with the level of DST. When DST is relatively low, the effect is positive; however, at higher levels, the effect becomes negative. To estimate the maximum potential reduction in CIWB due to DST, it is assumed that DST takes the maximum value observed in the sample (13.4171). Under this condition, the marginal effect becomes 0.0047–0.0008 × 13.4171 = –0.0060. This implies that an additional unit increase in digital service trade at this level would, on average, decrease CIWB by 0.0060 units, corresponding to an approximate 35.29% decline relative to its mean value.

To validate the above judgment further, this paper follows the approach of Lind and Mehlum (2010) [58]. It uses Stata’s utest command to test the nonlinear relationship and recalculate its turning point. The utest is employed to formally assess whether a nonlinear relationship exists within a specified interval. The null hypothesis of this test assumes a monotone or U shape, while the alternative hypothesis assumes an inverted U shape. The t statistic and p-value obtained from the U test help determine whether to reject the null hypothesis, thus indicating the curvature pattern. In addition, the test reports the turning point and interval, allowing for precise identification of the curve’s location.

As shown in

Table 3. U test results.

(1) Overall test of inverted U shape and turning point
t statistic	6.23
p > |t|	<0.0001
Turning point	5.4211
95% Fieller interval for extreme point	[4.7965, 5.9227]
(2) Upper and lower limits and their corresponding slopes
	Lower limit	Upper limit
Interval	0.3920	13.4172
Slope	0.0044	−0.0069
t statistic	6.23	−8.32
p > |t|	<0.0001	<0.0001

, the overall test result leads to the rejection of the null hypothesis, suggesting a statistically significant inverted U-shaped link between DST and CIWB. Additionally, the estimated turning point for DST is 5.4211, consistent with the value derived in Table 1. Moreover, the upper and lower bounds of digital service trade are 0.3920 and 13.4172, respectively, with a significantly positive slope at the leftmost end and a significantly negative slope at the rightmost end, and the turning point lies within the range of these bounds. To visually present the curve, this paper calculates and plots the fitted values of CIWB concerning the digital service trade at intervals of 1 while controlling for other variables. As shown in Figure 2, the results further confirm that the digital service trade and CIWB exhibit a significant inverted U-shaped relationship. The range with the positive slope is significantly narrower than the negative one, further corroborating the baseline regression results: digital service trade and the CIWB exhibit a negative correlation over a broad range while displaying an overall nonlinear relationship. These results preliminarily confirm that Hypothesis 1 of this study holds. Notably, the curve’s turning point occurs at DST = 5.4211, where DST represents the natural logarithm of the scale of digital service trade. The calculation shows that the scale of digital service trade at this turning point is approximately USD 225 million. This finding suggests that countries should strive to expand their scale of digital service trade to exceed USD 225 million, thereby contributing to both reducing emissions and the enhancement of well-being.

5.2. Robustness Checks

To validate the reliability of the benchmark results, various robustness checks were carried out across multiple dimensions, as shown in

Table 4. Results of robustness checks.

Variables	(1)	(2)	(3)	(4)	(5)
	DST Lagged One Period	2SLS	Dif-GMM	Sys-GMM	Winsorizing
DST		0.0143 ** (0.0062)	0.0014 *** (0.0003)	−0.0002 ** (0.0001)	0.0051 *** (0.0010)
DST×DST		−0.0012 *** (0.0003)	−0.000181 *** (0.000025)	0.000020 *** (0.000004)	−0.0004 *** (0.0001)
L.DST	0.0046 *** (0.0009)
L.(DST×DST)	−0.0004 *** (0.0001)
L.CIWB			0.6404 *** (0.0070)	0.8954 *** (0.0058)
L2.CIWB			−0.1467 *** (0.0070)	−0.0668 *** (0.0047)
Controls	Y	Y	Y	Y	Y
RFE	Y	Y	Y	Y	Y
YFE	Y	Y	Y	Y	Y
Constant	−0.1746 *** (0.0075)		−0.1820 *** (0.0073)	−0.0085 *** (0.0006)	−0.1501 *** (0.0138)
DWH Test		42.09 [<0.0001]
Weak IV Test		14.79 {7.03}
AR (1) Test			−2.84 [0.0045]	−3.00 [0.0026]
AR (2) Test			−0.71 [0.4798]	−1.24 [0.2168]
Sargan Test			125.02 [0.9999]	128.08 [0.9999]
U Test	5.21 [<0.0001]	2.25 [0.0188]	4.28 [<0.0001]	2.34 [0.0097]	5.08 [<0.0001]
N	2068	2204	1760	1912	2224

Note: ** and *** indicate significance at the 5% and 1% levels, respectively. L. and L2. denote the first- and second-order lagged terms, respectively. The DWH Test reports the chi-squared statistic, while the Weak IV Test reports the Cragg–Donald Wald F statistic. The values in braces represent the Stock–Yogo weak ID test critical values (10% maximal IV size). The AR Test reports the z statistic, the Sargan Test reports the chi-squared statistic, and the U test reports the t statistic.

: (1) Independent variable lagged one period. To address the potential endogeneity arising from reverse causality and lagged effects, the analysis incorporates a one-period lag of DST and re-estimates the model accordingly. As shown in column (1), the results continue to support the benchmark regression conclusions, suggesting a lagged effect of DST on CIWB. (2) Two-stage least squares (2SLS). While the estimates reported in column (1) help address reverse causality, they fall short of resolving endogeneity arising from unobserved confounders. This could lead to a high correlation between the independent variable and the disturbance term, with severe endogeneity potentially causing substantial bias in parameter estimates. Therefore, fixed broadband and mobile cellular subscriptions were selected as instruments, and the model was re-estimated using the 2SLS approach. Column (2) yields consistent findings, reinforcing the conclusions drawn from the baseline model. (3) Dynamic panel regression. Considering that the current CIWB might also be influenced by past factors, this study further constructed a dynamic panel model. It re-ran the regression using the difference Generalized Method of Moments (Dif-GMM) and system Generalized Method of Moments (Sys-GMM). The estimates in columns (3) and (4) are consistent with the baseline results, and they further confirm that CIWB is influenced by past factors. (4) Two-tail winsorizing (1%). To reduce the impact of outliers, this study performed 1% two-tail winsorizing on all variables and re-ran the regression. Column (5) produces results aligned with the baseline findings and further reinforces the main conclusions. The findings provide additional empirical support for Hypothesis 1.

5.3. Mechanism Analysis

Building on the preceding theoretical framework, different types of digital service trade affect CIWB through various mechanisms. Therefore, this section will first analyze the impact of different digital service trades on CIWB, as shown in

Table 5. Impact of different digital service trades on CIWB.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	CIWB	CIWB	CIWB	CIWB	CIWB	CIWB	CIWB	CIWB	CIWB
DST_Ins	−0.0004 (0.0003)	−0.0003 *** (0.0001)
DST_Ins×DST_Ins	0.00001 (0.00003)
DST_Fin			0.0012 *** (0.0003)
DST_Fin×DST_Fin			−0.00017 *** (0.00003)
DST_Int				0.0016 *** (0.0003)
DST_Int×DST_ Int				−0.00027 *** (0.00004)
DST_Res					0.0031 *** (0.0004)
DST_Res×DST_Res					−0.00034 *** (0.00004)
DST_Tel						0.0011 *** (0.0002)
DST_Tel×DST_Tel						−0.00022 *** (0.00004)
DST_Pro							0.0014 *** (0.0002)
DST_Pro×DST_Pro							−0.00023 *** (0.00003)
DST_Aud								0.0004 (0.0003)	−0.0003 * (0.0002)
DST_Aud×DST_Aud								−0.0001 ** (0.0001)
Controls	Y	Y	Y	Y	Y	Y	Y	Y	Y
RFE	Y	Y	Y	Y	Y	Y	Y	Y	Y
YFE	Y	Y	Y	Y	Y	Y	Y	Y	Y
Constant	−0.1639 *** (0.0100)	−0.1640 *** (0.0102)	−0.1658 *** (0.0106)	−0.1676 *** (0.0097)	−0.1672 *** (0.0087)	−0.1605 *** (0.0087)	−0.1692 *** (0.0089)	−0.1654 *** (0.0260)	−0.1696 *** (0.0261)
U Test	Trivial failure to reject H0		4.17 [<0.0001]	4.71 [<0.0001]	7.42 [<0.0001]	5.32 [<0.0001]	6.61 [<0.0001]	Trivial failure to reject H0
N	2224	2224	2224	2224	2224	2224	2224	2224	2224
Within R2	0.3771	0.3771	0.3877	0.4112	0.4155	0.4091	0.4141	0.3168	0.3100

Note: *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

. Columns (3) to (7) indicate that DST_Fin, DST_Int, DST_Res, DST_Tel, and DST_Pro all exhibit an inverted U-shaped relationship with CIWB. Columns (1) and (2) show a negative correlation between DST_Ins and CIWB. Columns (8) and (9) show a negative correlation between DST_Aud and CIWB. This indicates that priority should be given to developing digital service trade in the insurance, pension, and audiovisual sectors. Meanwhile, the scale of digital service trade in other sectors should also be expanded to exceed the turning point, thereby contributing to reduced emissions and improved well-being. The above results provide strong evidence to support Hypothesis 2 proposed in this study.

Regarding the intrinsic transmission mechanisms, the theoretical analysis presented earlier suggests that insurance and pension services can reduce economic pressure due to illness by providing pensions and related insurance products, enabling residents to receive appropriate treatment and thus extending the average lifespan. Additionally, by offering green insurance products, these services can reduce the environmental compliance costs faced by enterprises during their transition to low-carbon operations, thereby lowering carbon emissions. Thus, insurance and pension services can evidently reduce CIWB. Similarly, audiovisual and related services, which fall under personal, cultural, and recreational services, generally have minimal environmental costs and can enhance individual well-being and social welfare. Thus, these services can also reduce CIWB. In summary, insurance and pension services, as well as audiovisual and related services, have an obvious inhibiting effect on CIWB. This also partly explains the coexistence of linear and nonlinear components, as digital service trade tends to show a broadly negative association with CIWB. This phenomenon is related to the existence of these two categories of digital service trade. Since the mechanisms through which insurance and pension services, as well as audiovisual and related services, affect CIWB are relatively direct, and it is difficult to find quantifiable mechanism variables, this paper will not conduct an empirical analysis of these mechanisms. Instead, the focus will shift to the nonlinear transmission mechanisms of other digital service trades.

The theoretical analysis above suggests that financial digital service trade (DST_Fin) exerts a nonlinear impact on CIWB through energy structure transition. Likewise, technology-intensive digital service trade (DST_Tec), including DST_Int, DST_Res, DST_Tel, and DST_Pro, exerts a nonlinear impact on CIWB through green technology innovation. Therefore, according to Equations (4) and (5), this paper first regresses the transmission mechanism variables on the independent variables and then regresses the dependent variables on the transmission mechanism variables to uncover the intrinsic transmission mechanisms.

Table 6. Analysis of intrinsic conduction mechanisms.

Variables	(1)	(2)	(3)	(4)
	EST	CIWB	GTI	CIWB
DST_Fin	−1.0606 *** (0.3180)
DST_Fin×DST_Fin	0.1306 *** (0.0375)
EST		−0.00017 *** (0.00002)
DST_Tec			−0.4484 *** (0.1550)
DST_Tec×DST_Tec			0.0329 *** (0.0123)
GTI				−0.00015 *** (0.00003)
Controls	Y	Y	Y	Y
RFE	Y	Y	Y	Y
YFE	Y	Y	Y	Y
U test	3.34 [0.0005]		2.12 [0.0176]
Turning point	4.0593		6.8155
Constant	196.1550 *** (17.7516)	−0.1307 *** (0.0113)	−4.1271 (3.2305)	−0.1644 *** (0.0104)
N	2224	2224	2224	2224
Within R2	0.2233	0.4044	0.0232	0.3771

Note: *** indicates significance at the 1% levels.

(columns 1 and 2) shows that financial digital service trade exhibits a U-shaped relationship with energy structure transition, while energy structure transition negatively correlates with CIWB. This suggests that at lower levels, financial digital service trade hinders the energy structure transition, thereby contributing to an increase in CIWB. Conversely, a sufficiently large scale of financial digital service trade promotes the energy structure transition, thereby reducing CIWB. This results in an inverted U-shaped relationship between financial digital service trade and CIWB. Columns (3) and (4) show that technology-intensive digital service trade exhibits a U-shaped relationship with green technological innovation, while green technological innovation is negatively correlated with CIWB. This implies that when the scale of technology-intensive digital service trade is small, it hinders green technological innovation, thus increasing CIWB. Conversely, a sufficiently large scale of technology-intensive digital service trade promotes green technological innovation, thereby reducing CIWB. This contributes to the observed nonlinearity in how technology-intensive digital service trade influences CIWB. The above results provide strong evidence supporting Hypotheses 3 and 4 proposed in this study.

The previous discussion thoroughly examined how digital service trade relates nonlinearly to CIWB and the intrinsic transmission mechanisms through which it exerts influence. The next crucial step is to explore how to maximize the suppressive impact of digital service trade on CIWB and avoid its exacerbating impact, which involves analyzing the external moderation mechanisms of the nonlinear link between digital service trade and CIWB.

Section 3 theoretically posits that public environmental awareness (PEA) may be a positive moderator in the nonlinear association between digital service trade and CIWB. To investigate this, this study adopts Haans’ (2016) [57] method to establish a nonlinear moderation effect model to examine the moderating role of public environmental awareness on the nonlinear link between digital service trade and CIWB, as illustrated in Equation (6) in Section 4. To alleviate multicollinearity issues, both public environmental awareness (PEA) and digital service trade (DST) were centered, and the regression results are shown in

Table 7. Analysis of external adjustment mechanisms.

Variables	(1)	(2)	(3)
	CIWB	CIWB	CIWB
DST	−0.0016 *** (0.0002)	−0.0016 *** (0.0003)	−0.0013 *** (0.0003)
DST×DST	−0.0004 *** (0.0001)	−0.0004 *** (0.0001)	−0.0004 *** (0.0001)
PEA		0.000019 * (0.000010)	0.000026 ** (0.000011)
PEA×DST			−0.000012 *** (0.000002)
PEA×DST×DST			−0.000003 *** (0.000001)
Constant	−0.1658 *** (0.0083)	−0.1624 *** (0.0079)	−0.1592 *** (0.0079)
Controls	Y	Y	Y
RFE	Y	Y	Y
YFE	Y	Y	Y
U test	6.23 [<0.0001]	6.31 [<0.0001]	6.19 [<0.0001]
Extreme point	−1.8642	−1.8214	−1.5995
N	2224	2224	2224
Within R2	0.4168	0.4173	0.4220

Note: *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

. For interpreting the results of nonlinear moderation effect models, Haans (2016) primarily focuses on two aspects: the curve’s shape (whether it is flat or steep) and the turning point’s position (whether it shifts to the left or right). Therefore, this paper investigates the impact of PEA on the steepness of the curve and location of the turning point.

Regarding the influence of the moderating variable on the steepness of the curve, Equation (6) can be viewed as a quadratic function of the independent variable DST, with the curvature of the quadratic function determining the curve’s steepness. Clearly, for an inverted U shape to exist, the curvature must be negative, and a smaller curvature value corresponds to a steeper slope. Conversely, a curvature closer to 0 results in a flatter curve. By taking the second derivative of Equation (6) concerning DST, the curvature (l) can be derived as follows:

$$l=2{\eta }_2+2{\eta }_{5}PEA$$

(9)

Taking the first derivative of PEA with respect to Equation (9), the marginal impact of public environmental awareness on curvature is derived as follows:

$${\displaystyle \frac{dl}{dPEA}}=2{\eta }_5$$

(10)

According to Table 7, ${\eta }_5$ is significantly negative at the 1% level, implying that the curve becomes steeper as public environmental awareness increases. Specifically, to the left of the turning point, rising public environmental awareness aggravates the CIWB-increasing effect of digital service trade. Conversely, to the right of the turning point, rising public environmental awareness aggravates the CIWB-decreasing effect of digital service trade.

The turning point’s position, affected by the moderating variable, is identified as the digital service trade value where the first derivative of Equation (6) equals zero.

$$DST=-{\displaystyle \frac{{\eta }_1+{\eta }_{4}PEA}{2{\eta }_2+2{\eta }_{5}PEA}}$$

(11)

The marginal effect of public environmental awareness on the turning point’s location is obtained from the first derivative of Equation (11) with respect to PEA, as shown below:

$${\displaystyle \frac{dDST}{dPEA}}={\displaystyle \frac{{\eta }_1{\eta }_5-{\eta }_2{\eta }_4}{2{({\eta }_2+{\eta }_{5}PEA)}^2}}$$

(12)

It is clear that the denominator $2({\eta }_2+{\eta }_{5}EEA{)}^2$ in Equation (12) is positive. As calculated from Table 7, the numerator ${\eta }_1{\eta }_5 - {\eta }_2{\eta }_4$ is negative. Therefore, the marginal effect of PEA on the turning point’s position is negative, implying that greater PEA causes the turning point to shift leftward.

In summary, increasing public environmental awareness steepens the curve and shifts its turning point leftward. To visualize how public environmental awareness (PEA) influences the shape and turning point of this curve, this section defines samples with below-average PEA as low PEA, and those with above-average PEA as high PEA. The fitted values of CIWB concerning digital service trade were calculated and plotted under different levels of public environmental awareness, with other variables controlled. As shown in Figure 3, in countries or regions with higher public environmental awareness, the turning point is around DST = 5.3, whereas in countries or regions with lower public environmental awareness, the turning point is around DST = 6.3. The turning point in countries or regions with higher environmental awareness is further to the left. Meanwhile, countries or regions with higher public environmental awareness exhibit a steeper curve than those with lower ones. This implies that countries or regions with higher public environmental awareness reach the turning point more easily, and beyond this point, digital service trade has a more pronounced suppressive effect on CIWB. The above results provide strong evidence supporting Hypothesis 5 proposed in this study.

5.4. Heterogeneity Analysis

To explore the regional heterogeneity in the link between digital service trade and CIWB, this study divides the 152 countries or regions in the sample into six regions: Asia, Africa, Europe, North America, South America, and Oceania. Regression analyses are then re-run using the benchmark regression model.

Table 8. Results of regional heterogeneity analysis.

Variables	Asia	Africa	Europe	North America		South America		Oceania
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
DST	0.0045 ** (0.0017)	0.0022 *** (0.0006)	0.0122 *** (0.0023)	0.0026 (0.0025)	−0.0050 *** (0.0014)	0.0001 (0.0015)	−0.0003 (0.0002)	0.0051 *** (0.0008)
DST×DST	−0.0003 *** (0.0001)	−0.0002 *** (0.0001)	−0.0008 *** (0.0001)	−0.0006 *** (0.0002)		−0.00003 (0.00010)		−0.0010 *** (0.0001)
Controls	Y	Y	Y	Y	Y	Y	Y	Y
RFE	Y	Y	Y	Y	Y	Y	Y	Y
YFE	Y	Y	Y	Y	Y	Y	Y	Y
U test	2.67 [0.0059]	3.80 [0.0002]	5.40 [<0.0001]	Trivial failure to reject H0		Trivial failure to reject H0		5.72 [0.0004]
Turning point	6.6489	5.4327	7.6943	None		None		2.4731
Constant	−0.1152 *** (0.0171)	−0.0991 *** (0.0131)	−0.2218 *** (0.0137)	−0.2278 *** (0.0400)	−0.1932 *** (0.0425)	−0.0925 *** (0.0268)	−0.0920 *** (0.0255)	−0.0747 *** (0.0109)
N	455	551	663	287	287	154	154	114
Within R2	0.2672	0.3439	0.6662	0.7066	0.6954	0.5537	0.5531	0.7192

Note: ** and *** indicate significance at the 5% and 1% levels, respectively.

reveals that in Asia, Africa, Europe, and Oceania, digital service trade exhibits an inverted U-shaped association with CIWB, aligning with the baseline regression findings. However, this pattern is not observed in North America and South America. A linear regression model was further constructed to investigate whether it is linear. The analysis reveals that digital service trade is negatively correlated with CIWB in North America, whereas no linear relationship exists in South America. The negative correlation in North America may stem from extensive engagement in digital service trade, which likely surpassed the turning point during the sample period and entered a phase where digital service trade significantly reduces CIWB.

Additionally, comparing the turning points of the curves in Asia, Africa, Europe, and Oceania, Oceania’s turning point is the furthest to the left, followed by Africa and Asia, with Europe’s turning point being the furthest to the right. This indicates that Oceania reaches the turning point with a smaller scale of digital trade, followed by Africa and Asia, while Europe requires a larger scale of digital service trade to cross that threshold. Moreover, differences in the steepness of the curves can be assessed by comparing the coefficients of the squared terms. As mentioned earlier, the smaller the squared term coefficient, the steeper the inverted U-shaped curve. The 95% confidence intervals of the squared term coefficients in digital service trade are plotted in Figure 4. Oceania’s coefficient of the squared term is significantly smaller than Asia and Africa’s, with non-overlapping 95% confidence intervals; Europe’s coefficient is also significantly smaller than Africa’s, with non-overlapping 95% confidence intervals. This indicates that Oceania’s curve is steeper than Asia and Africa’s, and Oceania’s digital service trade has a more pronounced suppressive effect on CIWB than Asia’s and Africa’s after surpassing the turning point. Compared to Africa, Europe exhibits a steeper curve, with its digital service trade exerting a stronger downward influence on CIWB beyond the turning point.

The turning point in Oceania appears furthest to the left with the steepest curve. This phenomenon may be attributed to its relatively small economic scale, advanced digital infrastructure, and early adoption of green energy policies. As a result, digital service trade in Oceania has surpassed the high-carbon stage more rapidly, with subsequent technological innovation and the application of clean energy substantially lowering carbon intensity. Conversely, the turning points in Asia and Africa are situated further to the right, with noticeably flatter curves. Such patterns may stem from disparities in digital infrastructure maturity, continued reliance on high-carbon energy sources in certain countries, and slower rates of technology diffusion and policy implementation, which delay and weaken the emissions reduction impact of digital service trade. Although positioned the furthest to the right, Europe’s turning point is characterized by a steeper curve than that of Africa. This distinction is likely linked to Europe’s highly industrialized economic structure, where digital service trade expansion tends to drive higher energy consumption. Nevertheless, after surpassing the turning point, Europe has experienced a rapid decline in carbon intensity, supported by stringent environmental policies, advanced low-carbon technologies, and a well-established carbon market. In contrast, Africa’s emissions reduction from digital service trade remains limited due to weak infrastructure, low clean energy adoption, and poor policy enforcement.

In addition to analyzing regional heterogeneity, this paper also explores the following types of heterogeneity, as shown in

Table 9. Results of other heterogeneity analyses.

Variables	High GDP per Capita	Low GDP per Capita	High Proportion of Internet Users	Low Proportion of Internet Users	Before Paris Agreement	After Paris Agreement
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
DST	0.0042 *** (0.0008)	0.0021 *** (0.0005)	0.0038 *** (0.0009)	0.0016 *** (0.0004)	0.0027 *** (0.0004)	−0.0002 (0.0002)	−0.0010 *** (0.0003)
DST×DST	−0.0004 *** (0.0001)	−0.00017 *** (0.00004)	−0.0004 *** (0.0001)	−0.00012 *** (0.00003)	−0.00023 *** (0.00003)	−0.00007 * (0.00004)
Controls	Y	Y	Y	Y	Y	Y	Y
RFE	Y	Y	Y	Y	Y	Y	Y
YFE	Y	Y	Y	Y	Y	Y	Y
Constant	−0.2509 *** (0.0142)	−0.0947 *** (0.0118)	−0.2384 *** (0.0168)	−0.0806 *** (0.0129)	−0.1723 *** (0.0135)	−0.1478 *** (0.0121)	−0.1462 *** (0.0122)
U test	4.89 [<0.0001]	3.85 [0.0001]	4.00 [0.0001]	3.63 [0.0002]	6.16 [<0.0001]	Trivial failure to reject H0
Turning point	5.3577	6.1473	5.1597	6.5401	5.8937	None
N	1055	1169	1156	1055	1409	815	815
Within R2	0.6088	0.3038	0.5942	0.2246	0.3708	0.2909	0.2902

Note: * and *** indicate significance at the 10% and 1% levels, respectively.

: (1) Economic-Development-Level Heterogeneity. The impact of digital service trade on CIWB may vary across countries or regions with different levels of economic development. We use GDP per capita to measure economic development level (Econ). The categorical variable econ is set to 1 if GDP per capita exceeds the sample mean, and 0 otherwise. We then perform grouped regressions. As shown in columns (1) and (2), the inverted U-shaped association between digital service trade and CIWB persists across both high and low levels of economic development. In more economically advanced regions, the turning point occurs earlier, suggesting that a lower level of digital service trade is sufficient to trigger the shift. Beyond turning point comparisons, differences in curve steepness can be assessed by examining the coefficients of the squared term. As discussed earlier, a smaller coefficient of the squared term indicates a steeper inverted U-shaped curve. This section plots the 95% confidence intervals of the squared term coefficients for digital service trade. As shown in Figure 4, more developed regions exhibit notably smaller squared term coefficients for digital service trade, with non-overlapping confidence intervals. This suggests that their curve is steeper, and digital service trade has a more pronounced suppressive effect on CIWB after crossing the turning point.

(2) Internet-Development-Level Heterogeneity. Differences in internet development may influence how digital service trade affects CIWB across countries or regions. We measure internet development level (Inter) by the proportion of the population using the internet. The categorical variable inter equals 1 if the proportion of internet users exceeds the sample mean, and 0 otherwise. We then perform grouped regressions. As shown in columns (3) and (4), an inverted U-shaped association between digital service trade and CIWB is evident regardless of internet development level. However, the turning point is positioned further to the left for countries or regions with higher internet development, indicating that these countries or regions reach the turning point with a lower scale of digital service trade. Similarly, as illustrated in Figure 4, countries with higher internet development levels have smaller squared term coefficients for digital service trade, indicating a steeper curvature and a stronger post-turning-point suppressive effect of digital service trade on CIWB.

(3) Heterogeneity Before and After the Paris Agreement. The Agreement seeks to curb climate change by keeping the rise in global average temperature well below 2 °C, with efforts to limit it to 1.5 °C to prevent severe consequences. Therefore, signing the Paris Agreement may alter how digital service trade interacts with CIWB in a nonlinear manner. Considering that the Agreement came into effect on 4 November 2016, this paper divides the sample period into before the Paris Agreement (2005–2016) and after the Paris Agreement (2017–2022) and performs grouped regressions. As shown in columns (5) to (7), before the Paris Agreement, the relationship between digital service trade and CIWB had an inverted U shape; after the Paris Agreement, the relationship became negatively correlated.

6. Conclusions and Policy Implications

6.1. Conclusions

This paper employs balanced panel data from 152 countries (regions) from 2005 to 2022. It constructs a nonlinear regression model, combined with U tests and plotted predictive margins, to empirically analyze the nonlinear impact of digital service trade on CIWB. The robustness of the results is tested using methods such as two-stage least squares, difference moment estimation, system moment estimation, independent variables lagged one period, and 1% two-tail winsorizing. Additionally, this paper examines both the intrinsic transmission mechanisms and external moderating mechanisms of the nonlinear impact of digital service trade on CIWB. Furthermore, the samples are categorized according to factors such as geographical location, economic development level, internet development level, and the status of the Paris Agreement. Heterogeneity analysis is performed by conducting regressions on these sub-samples.

The findings reveal an overall inverted U-shaped link between digital service trade and CIWB, whereby digital service trade initially increases CIWB but subsequently exerts a suppressive effect, with the latter predominating. The estimated turning point occurs at approximately USD 225 million in digital service trade volume, lending empirical support to Hypothesis 1. This suggests that maintaining digital service trade volumes above this threshold is crucial for maximizing the mitigating impact on CIWB. Mechanism analysis further indicates that financial and technology-intensive digital services influence CIWB primarily through energy structure transformation and green technological innovation. Meanwhile, digital services in the insurance, pension, and audiovisual sectors exhibit a direct suppressive effect on CIWB. In addition, enhanced public environmental awareness reinforces the negative relationship between digital service trade and CIWB. These results validate Hypotheses 2 through 5 and highlight the importance of countries prioritizing digital service trade in the insurance, pension, and audiovisual industries, while ensuring that other sectors maintain trade volumes above the identified turning point. Building upon the confirmed core hypotheses, this study extended its analysis to examine regional and temporal heterogeneity. Regionally, the relationship between digital service trade and CIWB exhibits heterogeneity. Except for North America, where the effect is primarily suppressive, the results from Asia, Africa, Europe, and Oceania are consistent with the overall sample. Moreover, in countries or regions with higher economic and internet development levels, the inverted U curve is steeper and the turning point occurs at a lower trade volume. This suggests that expanding digital service trade and improving economic and internet infrastructure are essential for realizing greater CIWB mitigation in these regions. From a temporal perspective, before the implementation of the Paris Agreement, the findings mirror those of the full sample. After its enforcement, however, the suppressive effect becomes more pronounced. This underscores the need for continued advancement of the Paris Agreement’s implementation and the negotiation of future climate accords.

The limitations of this study lie primarily in the incomplete availability of statistical data on digital service trade, CIWB, and other relevant variables across certain countries or regions. As a result, the dataset employed does not encompass all global economies. Moreover, adopting a global perspective to analyze the impact of digital service trade on CIWB provides a broad and generalized view, yet lacks the granularity required for detailed micro-level analysis. Future research could focus on conducting more micro-level empirical analyses using panel data from enterprises or households within a specific country. A digital service trade policy implemented by that country could be regarded as a quasi-natural experiment, and a DID model could be employed to examine how digital service trade policies affect the CIWB of micro-level entities.

6.2. Policy Implications

To advance the Sustainable Development Goals, countries (regions) need to leverage the beneficial effect of DST in reducing CIWB by implementing feasible and effective policy measures that promote its healthy and green development. First, digital service trade should be expanded to USD 225 million, which is set as a policy benchmark. Countries that have not yet reached this threshold are encouraged to support the expansion of digital service exports through fiscal subsidies, tax incentives, and low-interest loans. Simultaneously, digital infrastructure construction should be accelerated, particularly in low- and middle-income countries. Regional cooperation mechanisms, shared platform resources, and interoperable standards should be established to improve the accessibility and efficiency of digital service trade.

Second, priority should be given to the development of digital services in the insurance, pension, and audiovisual sectors. Special development funds should be established, accompanied by green labelling certification and preferential government procurement policies. Financial instruments such as green credit and policy guarantees can be utilized to reduce corporate financing costs. Enterprises should be guided to incorporate environmentally friendly principles into product design, fostering the development of services with green functionalities—such as remote environmental education, online elderly care management, and green insurance evaluation systems—that enhance human well-being while minimizing carbon footprints. Third, green innovation should be strengthened in finance-related and technology-intensive digital service sectors. A dedicated fund for green digital innovation could be established to encourage the application of artificial intelligence, big data, and cloud computing in optimizing energy efficiency. The development of green financial products and the export of green technologies should also be supported. The transition to cleaner energy structures and the dissemination of low-carbon technologies can be accelerated by advancing the integration of digitalization and environmental sustainability. Fourth, public environmental awareness should be enhanced to provide a social foundation for the low-carbon transformation of digital services. National environmental education initiatives could incorporate environmental protection and digital literacy into primary, secondary, and higher education curricula. In addition, digital platforms should be encouraged to display carbon footprint information, thereby guiding consumers toward green services. Social organizations and media should be supported in disseminating environmental values through short videos, social media, and other diverse communication channels, thereby fostering a culture of green consumption. From a regional perspective, Asia, Africa, Europe, and Oceania should adopt differentiated strategies based on their economic and digital development levels. Economically advanced nations should accelerate the intelligent and sustainable upgrading of digital services, fostering integrated local–regional–global ecosystems. Meanwhile, countries with underdeveloped digital foundations may bridge existing gaps through capital and technology inflows, institutional guidance, and international cooperation to ensure the sustainable expansion of digital service trade.

Finally, global climate governance cooperation must be reinforced, with digital service trade integrated into Nationally Determined Contributions (NDCs) and future climate agreements. Establishing international standards for carbon footprint accounting in digital services and a global monitoring platform would provide scientific foundations for policymaking and pathway optimization. A cross-sectoral coordination mechanism should also be developed to align digital trade, environmental, and economic policies, enabling a dynamic cycle of “expansion in scale—structural optimization—low-carbon development”. Ultimately, this will contribute to constructing a green, inclusive, and efficient global digital service trade system. By implementing these targeted policy measures, countries can harness the potential of digital service trade to drive green transitions and high-quality growth and demonstrate greater responsibility in addressing global climate challenges, fostering a harmonious synergy between well-being and sustainability.