Green Finance, Trade-Embodied Carbon, and the Sustainable Transition of China’s Manufacturing Sector: Evidence from Provincial Panel Data

Abstract

Mitigating trade-embodied carbon is essential for the sustainable, low-carbon transition of China’s manufacturing sector amid increasingly integrated domestic and global production networks. This study measures total trade-embodied carbon, embodied carbon outflows, and embodied carbon exports within a China-embedded global multi-regional input–output framework. Using a panel dataset covering 30 provinces, 15 manufacturing industries, and 7 benchmark years from 2002 to 2020, the study employs high-dimensional fixed-effects models to examine the effect of green finance—defined as finance directed toward environmentally sustainable and low-carbon activities—on trade-embodied carbon. The results show that green finance significantly reduces trade-embodied carbon, with a relatively stronger effect in the domestic trade dimension. Mechanistic analysis indicates that this effect operates through both technological and structural channels. Heterogeneity analysis further suggests that the carbon mitigation effect of green finance is more pronounced in the eastern and central regions and in energy-intensive industries. This study extends the analysis of the environmental effects of green finance from the value-chain trade perspective and provides empirical evidence to advance the low-carbon transition of manufacturing under intertwined domestic and global production networks.

1. Introduction

Climate change associated with rising carbon emissions has become a central issue in global policy debate. As cross-border and cross-regional production fragmentation has deepened, an increasing share of emissions has become embedded in trade flows rather than being linked only to production for local use, a pattern that is especially evident in manufacturing [1,2]. This issue is particularly important in the context of China. In 2024, China’s manufacturing value added reached approximately USD 4.66 trillion, accounting for roughly 28% of the global total, while its share of merchandise exports was 14.2%, ranking first globally [3]. The scale of China’s manufacturing and exports has made trade-related emissions particularly salient. According to a 2022 sectoral assessment by the International Energy Agency (IEA), direct carbon emissions from China’s manufacturing sector were about 3.8 billion tonnes, accounting for 41.3% of global manufacturing emissions [4]. China’s interprovincial production networks are closely intertwined with its external trade linkages. As a result, part of the emissions generated by manufacturing production is transferred across regions through interprovincial trade, whereas another substantial share is embodied in exports and ultimately driven by foreign final demand [5,6,7]. Under these conditions, production-based indicators alone cannot fully capture the environmental burden associated with manufacturing activity, nor can they accurately reflect how emission responsibility is distributed across regions and countries. By contrast, trade-embodied carbon accounting links the physical location of emissions to the final demand that drives them and thus provides a more appropriate framework for identifying carbon transfer, evaluating the environmental effects of trade, and designing differentiated mitigation policies. Against the background of China’s “Dual Carbon” goals, accurately measuring and effectively reducing trade-embodied carbon in manufacturing has therefore become an important and urgent task for advancing the sustainable transition of the sector [8].

In this context, reducing trade-embodied carbon requires not only stricter environmental constraints but also sustained financial support for low-carbon transformation. Low-carbon transformation in manufacturing is costly, time-consuming, and often subject to considerable uncertainty [9]. Cleaner production, technological upgrading, and structural adjustment all require sustained investment over an extended period. According to a UNDP technical report, China will need approximately USD 14 to 24 trillion to meet its dual carbon goals between 2020 and 2060 [10]. Given the scale of this financing demand, public finance and administrative regulation alone are unlikely to be sufficient. As a market-based institutional arrangement, green finance incorporates environmental objectives into financial decision-making and capital allocation [11]. As a financial tool for sustainable development, green finance channels capital toward environmentally sustainable and low-carbon activities. Through instruments including green credit, green bonds, and green funds, it can relax financing constraints on low-carbon activities and support green innovation, cleaner production, and industrial upgrading [12,13]. Thus, green finance plays an important role not only in carbon mitigation, but also in supporting cleaner and more sustainable manufacturing development under increasingly intertwined domestic and global production networks.

Although a substantial body of research has explored the environmental implications of green finance and the factors shaping carbon emissions, its relationship with trade-embodied carbon has received much less direct empirical attention. Existing studies have mainly focused on production-based emissions, carbon intensity, or firm-level environmental performance [14,15]; however, relatively little attention has been paid to whether green finance can also affect the carbon embodied in trade flows. This issue is particularly important in the context of China’s manufacturing sector, where domestic interprovincial production linkages are closely intertwined with international trade linkages. Moreover, although technological progress and structural upgrading are often regarded as key channels in the low-carbon transition [13,16], the specific mechanisms through which green finance affects trade-embodied carbon have not yet been sufficiently tested. The potential heterogeneity of this effect across regions and manufacturing sectors also remains insufficiently explored. As a result, existing research has not yet provided sufficient provincial-level evidence on how green finance contributes to the sustainable transition of manufacturing from the perspective of value-chain trade.

This paper examines whether and how green finance affects trade-embodied carbon in China’s manufacturing sector. Using a China-embedded global multi-regional input–output framework, it measures total trade-embodied carbon and further distinguishes between embodied carbon outflows and embodied carbon exports. On this basis, the study explores the direct effect of green finance on trade-embodied carbon, examines its technological and structural transmission channels, and assesses potential regional and sectoral heterogeneity. In this way, it connects the literature on green finance with the study of trade-embodied carbon and provides a new perspective for understanding the sustainable transition of manufacturing trade within intertwined domestic and global production networks.

The remainder of the paper is structured as follows. The next section reviews the related literature, followed by the theoretical analysis and research hypotheses. Trade-embodied carbon is then measured and its current characteristics are described. Then, the empirical model, variables, and data sources are introduced, followed by the empirical results, including the baseline estimates, endogeneity tests, robustness checks, mechanism analysis, and heterogeneity analysis. The final two sections present the conclusions and discuss the main implications and limitations of the study.

2. Literature Review

With the deepening of global value chains, embodied carbon emissions have increasingly been transferred through domestic and international trade networks [1]. Accordingly, the measurement of trade-embodied carbon and its determinants has become an important topic in environmental and trade research. The environmentally extended multi-regional input–output (EE-MRIO) model has become a widely used tool in this field because it can trace inter-sectoral and cross-border production linkages [7,17]. Based on this framework, existing studies have examined trade-embodied carbon from the perspectives of trade openness, environmental regulation, technological progress, and industrial structure [2,18,19]. Previous research has predominantly relied on either global MRIO tables or interprovincial input–output tables. The former is suitable for tracing embodied carbon transfers across countries but usually treats China as a single economy and therefore cannot fully capture interprovincial production linkages. The latter can reveal domestic embodied carbon transfers across provinces but cannot connect these domestic linkages with China’s position in international trade. To address this limitation, some studies embed China’s interprovincial input–output tables into a global MRIO framework, thereby constructing a China-embedded EE-MRIO table [5,6]. This embedded framework is especially suitable for China, because its manufacturing production is shaped by both strong interprovincial linkages and deep integration into global value chains.

Green finance is increasingly regarded as a key market-oriented instrument for environmental governance. The existing literature has mainly explored its environmental and economic effects from both macro and micro perspectives. At the macro level, Lee and Lee (2022) showed that green finance improves green total factor productivity, indicating its role in enhancing overall environmental and economic performance [20]. Liu and Zhu (2024) further found that green finance helps to reduce carbon emission intensity and improve carbon emission efficiency at the regional level [21]. At the micro level, Yu et al. (2021) showed that green finance helps to ease financing constraints on green innovation [9]. Hu et al. (2021) found that green credit policy promotes green innovation in heavily polluting firms [12]. Flammer (2021) showed that corporate green bonds support environmentally beneficial and long-term investment activities [22]. However, most of this literature focuses on production-based carbon outcomes, emphasizing emissions generated within regions or firms [17,23]. Whether green finance also affects the carbon embodied in trade remains underexplored, even though this issue is particularly relevant for manufacturing sectors embedded in domestic and global production networks.

A growing but limited body of literature has begun to connect green finance with trade-related outcomes. Some studies have suggested that green finance can improve the environmental performance of exporting firms, promote cleaner production, and support the green upgrading of export structure and product quality [24,25,26]. By directing capital toward low-carbon technologies and environmentally friendly activities, green finance may reshape the comparative advantage of regions and sectors in trade, making production and exports cleaner and more value-added [2,16]. These findings imply that green finance may also influence the carbon embodied in trade. However, direct empirical evidence on the relationship between green finance and trade-embodied carbon remains scarce. Existing studies have rarely examined trade-embodied carbon within a unified analytical framework that distinguishes between domestic and international trade, even though these two dimensions occupy different positions and perform different functions in manufacturing production networks. In addition, while technological progress and industrial upgrading are often discussed as important channels, the specific mechanisms through which green finance affects embodied carbon in manufacturing trade have not been sufficiently examined. The potential heterogeneity of this effect across regions and sectors also remains underexplored. To address these gaps, this study investigates the impact of green finance on trade-embodied carbon in China’s manufacturing sector within a unified analytical framework.

3. Theoretical Analysis

3.1. The Direct Impact of Green Finance on Trade-Embodied Carbon

Green finance is an important market-based tool for environmental governance. By incorporating environmental objectives into capital allocation and financial pricing, it helps to internalize environmental externalities and reshape production incentives in manufacturing [25]. In developing economies, manufacturing has long been associated with high energy consumption and carbon emissions. Green finance influences the carbon footprint of traded goods by altering the marginal cost of capital for enterprises [12]. Specifically, green credit, green bonds, and green funds channel more favorable financing terms toward low-carbon technologies, cleaner production, and environmentally friendly projects, while imposing tighter financing constraints on pollution- and energy-intensive activities [27]. This differential allocation of financial resources encourages firms to adopt cleaner energy use and production methods, thereby reducing the carbon intensity of manufactured products. Since trade-embodied carbon is transmitted through domestic and international production linkages, this effect can be reflected in both domestic outflows and international exports [2].

Hypothesis 1 (H1). 

Green finance reduces trade-embodied carbon in China’s manufacturing sector.

3.2. The Mediating Roles of Technological and Structural Effects

Technological progress is an important channel through which green finance may affect trade-embodied carbon. Unlike conventional financial instruments and command-and-control environmental regulation, green finance operates through a market-based incentive structure and the risk-adjusted pricing of low-carbon assets [28]. From the perspective of the green investment process, low-carbon technological upgrading in manufacturing is not a one-step adoption of technology but a capital-intensive and highly uncertain trajectory that often spans laboratory research, prototype development, pilot-scale trials, equipment upgrading, and eventual commercialization [29]. Throughout this process, low-carbon technological upgrading typically requires substantial upfront capital investment and involves long payback periods and considerable uncertainty. Under conventional credit arrangements, these features are likely to translate into higher risk premiums and tighter financing constraints. By internalizing environmental externalities into financial pricing, green finance changes this dynamic. Through targeted financial instruments, it can provide more patient capital for low-carbon projects, fill financing gaps in the pilot and scaling-up stages, ease financing constraints on green investment, and reduce the risk premium attached to environmentally sustainable projects [9]. In this way, green finance is more likely to support continuous R&D expenditure and the commercialization of green innovation.

Through these mechanisms, green finance is more likely to promote green innovation and R&D investment, which further drives technological upgrading, improves energy and production efficiency, and increases the share of cleaner and higher value-added products in manufacturing trade, thereby reducing the carbon intensity embodied in both domestic trade flows and export activities [21]. Therefore, green finance is expected to reduce trade-embodied carbon through the technological effect.

Hypothesis 2 (H2). 

Green finance reduces trade-embodied carbon through the technological effect.

Green finance also affects trade-embodied carbon through the structural effect. This channel is closely related to the information disclosure function and capital-orientation mechanism that are distinctive to green finance. Unlike conventional financial instruments or pure administrative intervention, green finance is typically accompanied by stricter environmental disclosure, ESG reporting, and project identification requirements. These arrangements help to reduce information asymmetry between financial institutions and manufacturing activities, making it easier for low-carbon projects to be identified, evaluated, and financed. At the same time, by incorporating environmental risks into credit allocation and investment decisions, green finance makes financial resources more likely to flow toward cleaner, more efficient, and more technology-intensive sectors, while carbon-intensive and low-efficiency activities face tighter financing conditions [20,30].

Through this information-based and environmentally oriented capital reallocation process, green finance can accelerate structural transformation in manufacturing. On the one hand, it promotes industrial structure advancement by increasing the relative importance of cleaner and more advanced manufacturing sectors and facilitating the adjustment of obsolete and high-emission capacity [13]. On the other hand, it promotes industrial structure rationalization by enhancing the allocation efficiency of capital and labor across sectors, thereby reducing structural distortions and improving overall production efficiency. Together, these changes help lower the carbon intensity embodied in both domestic and cross-border manufacturing trade [31]. Therefore, green finance is expected to reduce trade-embodied carbon through structural transformation.

Hypothesis 3 (H3). 

Green finance reduces trade-embodied carbon through the structural effect.

4. Measurement and Characteristics of Trade-Embodied Carbon

4.1. Measurement of Trade-Embodied Carbon

To measure provincial trade-embodied carbon, this study constructs a China-embedded global multi-regional input–output table (CMRIO) by disaggregating the China account in the OECD Inter-Country Input–Output (ICIO) table into 30 provincial accounts (excluding Tibet, due to missing data) using China’s interprovincial input–output table. The benchmark OECD-ICIO table reports 45 sectors, whereas China’s provincial input–output table contains 42 sectors. To ensure consistency between the global and provincial accounts, the two datasets are harmonized and aggregated into a unified classification of 21 sectors, comprising 18 goods-producing sectors and 3 service sectors. The manufacturing sectors used in the empirical analysis correspond to N3–N17 in Appendix A 

Table A2. Sector concordance between China’s Interprovincial IO table and OECD-ICIO table.

Unified Sector No.	Unified Sector Name	Industry Classification	Corresponding GBT Sector(s)	Corresponding OECD-ICIO Sector(s)
N1	Agriculture, Forestry, Animal Husbandry and Fishery Products & Services	Primary sector	A01–A05	A01_02, A03
N2	Coal Mining, Oil & Gas Extraction, Metal Ore Mining, Non-Metallic Mining & Others	Mining	B06–B12	B05_06, B07_08, B09
N3	Food and Tobacco	Manufacturing	C13–C14	C10T12
N4	Textiles, Apparel, Leather & Related Products	Manufacturing	C15–C19	C13T15
N5	Wood Processing	Manufacturing	C20	C16
N6	Furniture, Paper, Printing, Cultural & Educational Products, Misc. Manufactures	Manufacturing	C21–C24, C43	C17_18
N7	Petroleum, Coke and Nuclear Fuel Processing	Manufacturing	C25	C19
N8	Chemical Products	Manufacturing	C26	C20, C21
N9	Non-Metallic Mineral Products	Manufacturing	C30	C22, C23
N10	Metal Smelting and Rolling	Manufacturing	C31	C24
N11	Metal Products	Manufacturing	C33	C25
N12	General and Special Equipment Manufacturing	Manufacturing	C34–C35	C31T33
N13	Transport Equipment Manufacturing	Manufacturing	C36–C39	C29
N14	Electrical Equipment	Manufacturing	C38	C27
N15	Communication Equipment, Computers & Other Electronics	Manufacturing	C39	C26
N16	Instruments, Machinery & Equipment Manufacturing	Manufacturing	C40	C28
N17	Other Transport and Other Equipment Products	Manufacturing	C41–C42	C30
N18	Electricity, Heat, Gas and Water Production & Supply	Utilities	D44–D46	D, E
N19	Construction	Construction	E47–E50	F
N20	Wholesale & Retail, Transportation, Storage and Postal Services	Wholesale, retail and transportation-related services	F, G51-G60	G, H49, H50, H51, H52, H53
N21	Services	Other services	H-U61-96	I, J58T60, J61, J62_63, K, L, M, N, O, P, Q, R, S, T

. This embedded framework enables a more comprehensive measurement of provincial trade-embodied carbon by jointly capturing domestic interprovincial and international trade linkages. Detailed procedures for data preprocessing, sector concordance, and balancing are reported in Appendix A.

The construction proceeds in four steps. First, the OECD-ICIO benchmark table is preprocessed to improve accounting consistency and comparability with the Chinese provincial input–output table. In particular, household final consumption expenditure, non-profit institutions serving households, and direct purchases abroad by residents are merged into a unified category of private consumption. In addition, for country-sector accounts in which total input and total output do not match exactly, the discrepancy is assigned to changes in inventories of the rest of the world (ROW), using total input as the balancing benchmark. Second, the sector classifications in the two datasets are harmonized. After concordance, both the OECD-ICIO table and the Chinese provincial table are aggregated into the same 21-sector classification so that the intermediate-input, final-demand, value-added, and total-output blocks are comparable across national and subnational accounts. Third, the China account in the OECD-ICIO table is split into provincial accounts. In this step, the totals reported in China’s interprovincial input–output table are used as the main constraints for provincial intermediate input, final demand, value added, total input, and total output, while the structural coefficients of the original China account in the OECD-ICIO table are used to infer the allocation of provincial exports across destination economies and the distribution of imported products across source economies. Fourth, the embedded table is rebalanced. Residual discrepancies generated during the embedding process are absorbed by the ROW account and inventory adjustments to restore accounting balance.

The construction of the embedded MRIO framework relies on two boundary conditions. First, the estimation of interprovincial–foreign linkages is based on a proportionality assumption, whereby a given sector in different provinces is assumed to follow the same export destination and import source structure as the national benchmark China account in the OECD-ICIO table [32]. Second, aggregating the detailed OECD-ICIO and Chinese provincial tables into 21 broad sectors may introduce aggregation bias, as this procedure implicitly assumes a certain degree of within-sector technological homogeneity [33]. Accordingly, the findings are better interpreted as reflecting overall patterns and relative differences in trade-embodied carbon across provinces and sectors, rather than representing highly disaggregated carbon-accounting relationships.

The measurement of value-added trade and its associated embodied carbon is fundamentally based on the accounting framework of the Leontief inverse. The Leontief inverse matrix captures the total direct and indirect intermediate input requirements from all sectors and regions needed to satisfy one unit of final demand. For exposition, this study first illustrates the calculation using a simplified embedded global input–output model with two foreign countries and three domestic subnational regions. Specifically, the domestic economy consists of three subnational regions, denoted by $E$, $M$, and $W$, while the external economy consists of two foreign countries, denoted by $H$ and $L$. Within this simplified framework, the accounting expressions for bilateral value-added exports and their embodied carbon are first derived and then extended to the measurement of total trade-embodied carbon outflows and trade-embodied carbon exports for a general region $r$.

Under the embedded subnational global input–output model, subnational value-added exports are measured using the standard trade in value-added (TiVA) approach. For value-added exports from region $E$ to country $H$, the accounting expression can be written as follows:

$$\begin{array}{c}{TiVA}^{EH}=v^E\left[B^{EW}{B}^{EM}{B}^{EE}{B}^{EH}{B}^{EL}\right]{\left[Y^{WH}{Y}^{MH}{Y}^{EH}{Y}^{HH}{Y}^{LH}\right]}^T\\ =v^E{B}^{EE}{Y}^{EH}+{\displaystyle \sum _{r\in \left\{W,M\right\}}}{v}^E{B}^{Er}{Y}^{rH}+v^E{B}^{EH}{Y}^{HH}+v^E{B}^{EL}{Y}^{LH}\end{array}$$

(1)

where $B$ is the Leontief inverse matrix derived from the embedded subnational global input–output table, $v^E$ denotes the value-added coefficient vector for region $E$, $Y^{rH}$ represents final demand in country $H$ for products originating from region $r$, and $T$ denotes matrix transpose. Equation (1) gives bilateral value-added exports from region $E$ to country $H$. Following Meng et al. (2018) [1], this study replaces the value-added coefficient vector $v$ with the carbon emission coefficient matrix $F$ to obtain the corresponding embodied carbon in bilateral value-added exports:

$$\begin{array}{c}P\left(E^{EH}\right)=F^E\left[B^{EW}{B}^{EM}{B}^{EE}{B}^{EH}{B}^{EL}\right]{\left[Y^{WH}{Y}^{MH}{Y}^{EH}{Y}^{HH}{Y}^{LH}\right]}^T\\ =F^E{B}^{EE}{Y}^{EH}+{\displaystyle \sum _{r\in \left\{W,M\right\}}}{F}^E{B}^{Er}{Y}^{rH}+F^E{B}^{EH}{Y}^{HH}+F^E{B}^{EL}{Y}^{LH}\end{array}$$

(2)

Equation (2) characterizes the embodied carbon in bilateral value-added exports from region $E$ to country $H$. By aggregating the bilateral results across destinations, this framework can be extended to measure the total trade-embodied carbon for a general region $r$. In this study, total trade-embodied carbon is divided into two components. The first is trade-embodied carbon outflows, defined as embodied carbon originating in region $r$ but ultimately absorbed by final demand in other domestic subnational regions. The second is trade-embodied carbon exports, defined as embodied carbon originating in region $r$ but ultimately absorbed by foreign final demand. For terminological consistency, transactions among domestic subnational regions are referred to as inflows and outflows, whereas transactions between subnational regions and foreign economies are referred to as imports and exports.

However, directly applying the standard Leontief inverse to the embedded global MRIO table introduces a measurement challenge. The block matrices $B^{rr}$ and $B^{rs}$ derived from the full embedded global input–output table capture not only domestic interregional production linkages, but also indirect effects transmitted through international production networks. Here, $B^{rr}$ denotes the Leontief inverse block for production linkages within region $r$, while $B^{rs}$ denotes the corresponding block linking production in region $r$ to absorption in another domestic region $s$. Therefore, if these original blocks are used directly, the resulting domestic embodied carbon flows already contain international feedback effects, making it challenging to distinguish purely domestic transmission from internationally mediated transmission.

To address this issue, this study adopts a hypothetical extraction approach. Specifically, under a counterfactual setting in which cross-border intermediate input linkages are assumed to be negligible, the original direct requirement matrix $A$ is adjusted to a domestic-only matrix $A^d$, defined as follows:

$${\mathit{A}}^{\mathit{d}}=\left[\begin{array}{ccc}{\mathit{A}}^{\mathit{R}\mathit{R}}& 0& 0\\ 0& {\mathit{A}}^{HH}& 0\\ 0& 0& {\mathit{A}}^{\mathit{L}\mathit{L}}\end{array}\right],{\mathit{A}}^{\mathit{R}\mathit{R}}=\left[\begin{array}{ccc}{\mathit{A}}^{\mathit{W}\mathit{W}}& {\mathit{A}}^{\mathit{W}\mathit{M}}& {\mathit{A}}^{\mathit{W}\mathit{E}}\\ {\mathit{A}}^{\mathit{M}\mathit{W}}& {\mathit{A}}^{\mathit{M}\mathit{M}}& {\mathit{A}}^{\mathit{M}\mathit{E}}\\ {\mathit{A}}^{\mathit{E}\mathit{W}}& {\mathit{A}}^{\mathit{E}\mathit{M}}& {\mathit{A}}^{\mathit{E}\mathit{E}}\end{array}\right]$$

Accordingly, the corresponding counterfactual Leontief inverse is given by ${\mathit{B}}^{\mathit{d}}={(\mathit{I}-{\mathit{A}}^{\mathit{d}})}^{-1}$, which can be written in block form as follows:

$${\mathit{B}}^{\mathit{d}}=\left[\begin{array}{ccc}{\mathit{B}}^{\mathit{d}\mathit{R}\mathit{R}}& 0& 0\\ 0& {\mathit{B}}^{\mathit{d}HH}& 0\\ 0& 0& {\mathit{B}}^{\mathit{d}\mathit{L}\mathit{L}}\end{array}\right],{\mathit{B}}^{\mathit{d}\mathit{R}\mathit{R}}=\left[\begin{array}{ccc}{\mathit{B}}^{\mathit{d}\mathit{W}\mathit{W}}& {\mathit{B}}^{\mathit{d}\mathit{W}\mathit{M}}& {\mathit{B}}^{\mathit{d}\mathit{W}\mathit{E}}\\ {\mathit{B}}^{\mathit{d}\mathit{M}\mathit{W}}& {\mathit{B}}^{\mathit{d}\mathit{M}\mathit{M}}& {\mathit{B}}^{\mathit{d}\mathit{M}\mathit{E}}\\ {\mathit{B}}^{\mathit{d}\mathit{E}\mathit{W}}& {\mathit{B}}^{\mathit{d}\mathit{E}\mathit{M}}& {\mathit{B}}^{\mathit{d}\mathit{E}\mathit{E}}\end{array}\right]$$

Based on this setting, total trade-embodied carbon for region $\mathit{r}$ can be expressed as follows:

$$\begin{array}{cc}P\left(E^r\right)& =\left[{\displaystyle \sum _{s\ne r}}{F}^r{B}^{drr}{Y}^{rs}+{\displaystyle \sum _{s\ne r}}{F}^r{B}^{drs}{Y}^{ss}+{\displaystyle \sum _{s\ne r}}{\displaystyle \sum _{t\ne s,r}}{F}^r{B}^{drs}{Y}^{st}\right]\hfill \\ & +\left[{\displaystyle \sum _{s\ne r}}{F}^r{(B}^{rr}-B^{drr})Y^{rs}+{\displaystyle \sum _{s\ne r}}{F}^r{(B}^{rs}-B^{drs})Y^{ss}+{\displaystyle \sum _{s\ne r}}{\displaystyle \sum _{t\ne s,r}}{F}^r{(B}^{rs}-B^{drs})Y^{st}\right]\hfill \\ & +\left[{\displaystyle \sum _{G\in \left\{H,L\right\}}}{\displaystyle \sum _{t\ne r}}{F}^r{B}^{rG}{Y}^{Gt}\right]+[{\displaystyle \sum _{G\in \left\{H,L\right\}}}{F}^r{B}^{rG}{Y}^{GG}+{\displaystyle \sum _{G\in \left\{H,L\right\}}}{\displaystyle \sum _{K\in \left\{H,\mathit{L}\right\},K\ne \mathit{G}}}{\mathit{F}}^{\mathit{r}}{\mathit{B}}^{\mathit{r}\mathit{G}}{\mathit{Y}}^{\mathit{G}K}]\hfill \\ & +[{\displaystyle \sum _{\mathit{G}\in \left\{H,\mathit{L}\right\}}}{\mathit{F}}^{\mathit{r}}{(\mathit{B}}^{\mathit{r}\mathit{r}}-{\mathit{B}}^{\mathit{d}\mathit{r}\mathit{r}}){\mathit{Y}}^{\mathit{r}\mathit{G}}+{\displaystyle \sum _{\mathit{s}\in \left\{\mathit{E},\mathit{M},\mathit{W}\right\},\mathit{s}\ne \mathit{r}}}{\displaystyle \sum _{K\in \left\{H,\mathit{L}\right\}}}{\mathit{F}}^{\mathit{r}}{(\mathit{B}}^{\mathit{r}\mathit{s}}-{\mathit{B}}^{\mathit{d}\mathit{r}\mathit{s}}){\mathit{Y}}^{\mathit{s}K}]\hfill \\ & +\left[{\displaystyle \sum _{\mathit{G}\in \left\{H,\mathit{L}\right\}}}{\mathit{F}}^{\mathit{r}}{\mathit{B}}^{\mathit{d}\mathit{r}\mathit{r}}{\mathit{Y}}^{\mathit{r}\mathit{G}}+{\displaystyle \sum _{\mathit{s}\in \left\{\mathit{E},\mathit{M},\mathit{W}\right\},\mathit{s}\ne \mathit{r}}}{\displaystyle \sum _{K\in \left\{H,\mathit{L}\right\}}}{\mathit{F}}^{\mathit{r}}{\mathit{B}}^{\mathit{d}\mathit{r}\mathit{s}}{\mathit{Y}}^{\mathit{s}K}\right]\hfill \end{array}$$

(3)

In Equation (3), $\mathit{Y}$ represents the final demand matrix. The superscripts indicate the source and destination; for example, ${\mathit{Y}}^{\mathit{r}\mathit{s}}$ denotes the final demand in destination region $\mathit{s}$ for products originating from source region $\mathit{r}$. The indices $\mathit{r}$, $\mathit{s}$, and $\mathit{t}$ denote domestic subnational regions, where $\mathit{t}$ represents a third domestic region distinct from $\mathit{r}$ and $\mathit{s}$. By contrast, $\mathit{G}$ and $K$ denote foreign countries (or foreign economies). ${\mathit{F}}^{\mathit{r}}$ denotes the carbon emission coefficient vector for region $\mathit{r}$.

According to Equation (3), total trade-embodied carbon for region $r$ can be decomposed into four components. The first bracket captures the embodied carbon generated through pure domestic input–output linkages and ultimately absorbed by final demand in other domestic subnational regions, denoted by VAODPC. The second and third brackets jointly capture the embodied carbon that is indirectly shaped by international production linkages but ultimately absorbed by final demand in other domestic subnational regions, denoted by VAOIPC. Accordingly, trade-embodied carbon outflows for region $r$ can be written as follows:

$${\mathrm{T}\mathrm{E}\mathrm{C}\mathrm{O}}_r={\mathrm{V}\mathrm{A}\mathrm{O}\mathrm{D}\mathrm{P}\mathrm{C}}_r+{\mathrm{V}\mathrm{A}\mathrm{O}\mathrm{I}\mathrm{P}\mathrm{C}}_r$$

(4)

Furthermore, the fourth and fifth brackets jointly capture the embodied carbon generated through international production linkages and ultimately absorbed by foreign final demand, denoted by VAEIPC. The sixth bracket captures the embodied carbon that can still be generated through domestic input–output linkages under the counterfactual setting and is ultimately absorbed by foreign final demand, denoted by VAEDPC. Accordingly, trade-embodied carbon exports for region $r$ can be written as follows:

$${\mathrm{T}\mathrm{E}\mathrm{C}\mathrm{X}}_r={\mathrm{V}\mathrm{A}\mathrm{E}\mathrm{I}\mathrm{P}\mathrm{C}}_r+{\mathrm{V}\mathrm{A}\mathrm{E}\mathrm{D}\mathrm{P}\mathrm{C}}_r$$

(5)

Therefore, Equation (3) decomposes total trade-embodied carbon for region $r$ into two aggregate indicators, namely, trade-embodied carbon outflows and trade-embodied carbon exports.

$${\mathrm{T}\mathrm{E}\mathrm{C}}_r={\mathrm{T}\mathrm{E}\mathrm{C}\mathrm{O}}_r+{\mathrm{T}\mathrm{E}\mathrm{C}\mathrm{X}}_r$$

(6)

where ${\mathrm{T}\mathrm{E}\mathrm{C}}_r$ denotes total trade-embodied carbon in region $r$.

4.2. Characteristics of Trade-Embodied Carbon

Figure 1 and Figure 2 present the characteristics of provincial trade-embodied carbon in China’s manufacturing sector in 2002 and 2020. A comparison between the two years reveals three main patterns. First, the overall scale of provincial trade-embodied carbon increased markedly over time. Second, the spatial concentration of high-value provinces became more pronounced. Third, the sources of growth differed across regions, with inland provinces showing stronger growth in embodied carbon outflows, whereas coastal provinces exhibited a more pronounced increase in embodied carbon exports.

In terms of spatial distribution, provinces with relatively high trade-embodied carbon in 2002 were mainly concentrated in traditional industrial and manufacturing regions, such as Hebei, Liaoning, Shandong, Jiangsu, and Guangdong, whereas western and remote provinces remained at relatively low levels. By 2020, trade-embodied carbon had increased in almost all provinces, and high-value provinces became increasingly concentrated in North China, parts of Northeast China, and the eastern coastal manufacturing belt.

From a structural perspective, the growth of trade-embodied carbon in most inland and resource-intensive provinces was driven mainly by the expansion of embodied carbon outflows. These provinces are predominantly resource-based regions, where the emphasis on energy supply and primary production tends to generate higher embodied carbon outflows. This pattern was particularly evident in provinces such as Inner Mongolia, Hebei, Shanxi, and Shaanxi. By contrast, in coastal export-oriented provinces, the increase in trade-embodied carbon was more closely associated with embodied carbon exports. In major manufacturing hubs such as Guangdong, Jiangsu, Zhejiang, and Fujian, embodied carbon linked to international trade became increasingly prominent. This pattern highlights their roles as advanced manufacturing centers and reflects their deep integration into both domestic and global production networks.

5. Materials and Methods

5.1. Model Setting

5.1.1. Baseline Regression Model

To evaluate the effect of green finance on trade-embodied carbon, this study specifies the following baseline model:

$${CE}_{ijt}={\beta }_0+{\beta }_1{GF}_{it}+\sum {\beta }_{m}Controls+{\mu }_i+{\lambda }_j+{\delta }_t+{\theta }_{ij}+{\epsilon }_{ijt}$$

(7)

In Equation (7), ${CE}_{ijt}$ represents trade-embodied carbon in region $i$ in sector $j$ at time $t$. It is alternatively measured by total trade-embodied carbon (${TEC}_{ijt}$), trade-embodied carbon outflows (${TECO}_{ijt}$) or trade-embodied carbon exports (${TECX}_{ijt}$). ${GF}_{it}$ denotes the level of green finance in region $i$ at time $t$. $Controls$ denotes a vector of control variables, including both province–sector-level and province-level factors. Here, ${\mu }_i$, ${\lambda }_j$, and ${\delta }_t$ denote province, sector, and year fixed effects, respectively; ${\theta }_{ij}$ represents province–sector interaction fixed effects; and ${\epsilon }_{ijt}$ is the error term.

Given the high dimensionality of the fixed effects, the baseline model is estimated using the high-dimensional fixed-effects (HDFE) estimator proposed by Correia (2016) [34]. This estimator is adopted as the main specification because it can efficiently absorb province, sector, time, and province–sector interaction effects, thereby reducing omitted variable bias arising from unobserved heterogeneity. In addition, two alternative estimators are employed for robustness analysis. First, OLS estimates are reported as a benchmark linear comparison to examine whether the estimated relationship between green finance and trade-embodied carbon is sensitive to the specific implementation of the baseline fixed-effects estimator [35]. Second, the Poisson Pseudo Maximum Likelihood (PPML) estimator is reported as a supplementary robustness check. Unlike the baseline HDFE and OLS specifications, which use logarithmic forms of the dependent variables, the PPML specification is estimated using the level values of trade-embodied carbon. This approach provides an additional check on whether the main findings remain stable under an alternative estimation framework that is less sensitive to heteroskedasticity [36]. In this case, the focus is on the sign and significance of the estimated coefficient, rather than on the direct comparability of coefficient magnitudes across models.

5.1.2. Mechanism Test Model

To identify the channels through which green finance influences trade-embodied carbon, this study augments the baseline model in Equation (7) by introducing mediating variables. A mediation framework is then applied to test whether the effect of green finance operates through technological and structural channels [37,38].

Specifically, the mechanism analysis is conducted using the following set of equations:

$$M_{ijt}={\alpha }_0+{\alpha }_1{GF}_{it}+\sum {\alpha }_{m}Controls+{\mu }_i+{\lambda }_j+{\delta }_t+{\theta }_{ij}+{\epsilon }_{ijt}$$

(8)

$${CE}_{ijt}={\gamma }_0+{\gamma }_1{GF}_{it}+{\gamma }_2{M}_{ijt}+\sum {\gamma }_{m}Controls+{\mu }_i+{\lambda }_j+{\delta }_t+{\theta }_{ij}+{\epsilon }_{ijt}$$

(9)

where $M_{ijt}$ denotes the mediating variables, which capture the potential transmission channels of green finance. In this study, two main mechanisms are considered: the technological effect and the structural effect. In addition to the stepwise mediation regressions, the Sobel test and the bootstrap method with 1000 replications are used to examine the statistical significance and robustness of the indirect effects [39].

5.2. Variable Description

5.2.1. Dependent Variables

The dependent variables—namely, total trade-embodied carbon (${TEC}_{ijt}$), trade-embodied carbon outflows (${TECO}_{ijt}$), and trade-embodied carbon exports (${TECX}_{ijt}$)—were as defined in the previous section. Their original unit is million tonnes of carbon dioxide equivalent (Mt CO₂e). In the empirical analysis, their natural logarithmic forms, ${lnTEC}_{ijt}$, ${lnTECO}_{ijt}$, and ${lnTECX}_{ijt}$, are used.

5.2.2. Core Explanatory Variable: Green Finance Index (GF)

Building on the existing literature on province-level green finance measurement, this study develops a composite index to capture the level of green finance development at the provincial level more comprehensively [40,41]. The composite index is built from six components, namely, green credit, green investment, green insurance, green bonds, green funds, and green equity. Green credit is proxied by the share of environmental protection project loans in total credit. Green investment is measured as environmental pollution control investment relative to GDP, whereas green insurance is represented by the ratio of environmental pollution liability insurance income to total premium income. For capital market instruments, green bonds are captured by the share of green bond issuance in total bond issuance, and green funds by the share of green fund market capitalization in total fund market capitalization. Green equity is measured by the proportion of carbon trading, energy use rights trading, and pollutant discharge rights trading in total equity market transactions. To construct the overall index, this study adopts the entropy weighting method (EWM), which helps to reduce subjectivity in weight assignment [42,43].

As a robustness check, this study further constructs an alternative green finance index using global principal component analysis (GPCA) based on the same six dimensions. Compared with the entropy-weighted index, the GPCA-based measure extracts the common variation across the underlying indicators and thus provides an alternative data-driven proxy for the level of green finance development [44].

5.2.3. Mediating Variables

The technological effect is characterized by green innovation and innovation input. Therefore, this study uses two proxy variables for the technological effect. ${GP}_{it}$ denotes the number of green patents in province $i$ in year $t$, originally measured as the number of granted patents. In the empirical analysis, its natural logarithmic form, ${lnGP}_{it}$, is used. A higher level of green patenting generally indicates stronger green innovation capacity. ${RD}_{it}$ denotes R&D expenditure in province $i$ in year $t$, originally measured in billions of current USD. In the empirical analysis, its natural logarithmic form, ${lnRD}_{it}$, is used. Higher R&D expenditure reflects greater innovation input and stronger support for technological upgrading [29,45].

The structural effect is characterized by industrial structure advancement and industrial structure rationalization. Therefore, this study uses two proxy variables for the structural effect. ${AIS}_{it}$ denotes industrial structure advancement in province $i$ in year t, measured by the ratio of operating revenue of high-technology industries (Note: The identification of high-technology industries is based on the OECD standard [46] and China’s Classification of High-Technology Manufacturing Industries (2017) [47], under which six sectors are included: pharmaceutical manufacturing; aircraft, spacecraft, and related equipment manufacturing; electronic and communication equipment manufacturing; computer and office equipment manufacturing; medical instruments and measuring instruments manufacturing; and information chemicals manufacturing.) to the operating revenue of industrial enterprises above the designated size [19]. A larger value indicates a higher level of industrial upgrading. ${RIS}_{it}$ denotes the industrial structure rationalization index of province $i$ in year t, which is measured by the reciprocal of the Theil index, where ${Theil}_{it}={\sum }_k\left({\displaystyle \frac{Y_{kit}}{Y_{it}}}\right)\ln \left({\displaystyle \frac{Y_{kit}/L_{kit}}{Y_{it}/L_{it}}}\right)$. The Theil index measures the mismatch between output structure and employment structure across industries. By construction, a higher value of ${RIS}_{it}$ indicates a more coordinated and efficient allocation of resources across industries [48].

5.2.4. Control Variables

Following Lee et al. (2022) and Su et al. (2024) [20,49], this study adopts the STIRPAT framework, which is widely used in environmental research, as the basic analytical framework for selecting control variables. Accordingly, a set of control variables related to economic, demographic, and technological factors is incorporated into the model to account for the influence of macro-structural determinants on trade-embodied carbon. In addition, given the important role of trade activities in resource allocation and carbon transfer, trade openness is further introduced into the extended model to better reflect the factors influencing carbon emissions under open-economy conditions.

${GDP}_{ijt}$ denotes the gross domestic product of industry $j$ in province $i$ in year $t$, originally measured in millions of current USD. In the empirical analysis, its natural logarithmic form, ${lnGDP}_{ijt}$, is used. It is included to capture the economic scale of regional industrial activity. According to the EKC hypothesis, economic growth may influence trade-embodied carbon through both expansion effects and efficiency-related adjustments [50,51]. GDP is therefore included to control for the role of economic scale in the formation of trade-embodied carbon.

${POP}_{it}$ denotes the total resident population of province $i$ in year $t$, originally measured in units of 10,000 persons. In the empirical analysis, its natural logarithmic form, ${lnPOP}_{it}$, is used. It captures regional population size and reflects development pressure on resources and the environment, which may in turn affect trade-embodied carbon.

${EI}_{ijt}$ denotes energy consumption per unit of output for industry $j$ in province $i$ in year $t$, originally measured in tons of standard coal equivalent per million of current USD of output. In the empirical analysis, its natural logarithmic form, ${lnEI}_{ijt}$, is used. A higher value indicates lower energy efficiency. This variable is included to capture the energy-use efficiency of industrial production, which may directly affect the level of trade-embodied carbon [52].

${EINV}_{it}$ denotes environmental infrastructure investment in province $i$ in year t, originally measured in millions of current USD. In the empirical analysis, its natural logarithmic form, ${lnEINV}_{it}$, is used. It captures the intensity of regional investment in environmental governance. Higher environmental infrastructure investment may improve pollution-control capacity and facilitate cleaner production, thereby affecting trade-embodied carbon [53,54].

${LQ}_{it}$ denotes labor quality and is measured by the number of students enrolled in higher education institutions in province $i$ (millions of persons). It is included to proxy regional human capital and skill endowment. Higher labor quality may facilitate technological adoption, green innovation, and energy-efficiency improvement, thereby affecting trade-embodied carbon [55,56].

${FDI}_{it}$ denotes foreign direct investment received in province $i$ in year $t$, originally measured in millions of current USD, which captures foreign capital inflows and external openness. In the empirical analysis, its natural logarithmic form, ${lnFDI}_{it}$, is used. On the one hand, FDI may reduce trade-embodied carbon through technology spillovers, managerial improvements, and efficiency gains [57]. On the other hand, it may increase trade-embodied carbon if pollution-intensive activities are relocated to regions with relatively lower environmental standards [58].

${OPEN}_{it}$ denotes the total value of goods imports and exports of province $i$ in year $t$, originally measured in millions of current USD, which captures the degree of trade openness. In the empirical analysis, its natural logarithmic form, $ln{OPEN}_{it}$, is used. Greater openness can increase trade-embodied carbon through scale expansion, but may also reduce trade-embodied carbon through technological spillovers, structural upgrading, and efficiency improvement [2,18].

5.3. Data Source

This study measures trade-embodied carbon within a China-embedded environmentally extended multi-regional input–output (EE-MRIO) framework. The international input–output information is drawn from the OECD Inter-Country Input–Output (ICIO) tables, which report production linkages for 77 economies and 45 sectors from 2000 to 2020. For the China-related component, both the multi-regional input–output tables and provincial carbon emission inventories are obtained from the CEADs China Carbon Accounting Database for the benchmark years 2002, 2007, 2010, 2012, 2015, 2017, and 2020 [15,17,23,59,60]. These benchmark years are used because the China-related input–output tables and provincial carbon-emission inventories required to construct the embedded MRIO framework are available only for discrete benchmark years, with the most recent update extending data coverage to 2020. Although the sample does not extend beyond 2020, it still allows for a preliminary assessment of how green finance affects trade-embodied carbon under the policy context following the introduction of China’s “Dual Carbon” goals. On this basis, the input–output matrices were decomposed, and the indicators of total trade-embodied carbon (TEC), embodied carbon outflows (TECO), and embodied carbon exports (TECX) were computed in MATLAB R2025a following the approach of Meng et al. (2018) [1]. As the analysis centers on China’s manufacturing sector, the final sample includes 15 manufacturing industries in 30 provinces (excluding Tibet due to missing data) over 7 benchmark years.

Green finance (GF), the main explanatory variable in this study, was constructed by the authors from a range of official statistical yearbooks and financial databases. Information on green credit and green bonds was drawn from the China Financial Yearbook; green investment was measured using data from the China Industrial Statistical Yearbook and the China Energy Statistical Yearbook; green insurance was captured using data from the China Insurance Statistical Yearbook; and the indicators for green funds and green equity were obtained from the CSMAR database. These six dimensions were finally aggregated into a provincial composite index using the entropy weighting method.

The mediating variables were obtained from several specialized databases and official yearbooks. Green patents (GP) were obtained from CNIPA patent statistics and the CSMAR database, whereas R&D expenditure (RD) was measured using data from the China Science and Technology Statistical Yearbook. The variables used to construct the industrial structure advancement index (AIS) and the industrial structure rationalization index (RIS) were obtained from the Wind database, and the corresponding indices were calculated by the authors.

The control variables were primarily collected from official statistical yearbooks and administrative databases. Among them, GDP, population (POP), and FDI were primarily obtained from the China Statistical Yearbook, provincial statistical yearbooks, and provincial statistical bulletins. Labor quality (LQ) was obtained from the China Education Statistical Yearbook and provincial education statistical yearbooks. Environmental infrastructure investment (EINV) was obtained from the China Environmental Statistical Yearbook and related provincial environmental statistics. Trade openness (OPEN) was obtained from the database of the General Administration of Customs of China. Energy intensity (EI) was calculated using China’s multi-regional input–output tables and provincial carbon emission inventories from the CEADs China Carbon Accounting Database.

The descriptive statistics of the main variables employed in the empirical analysis are presented in

Table 1. Descriptive statistics.

Type	Variables	Mean	S.D.	Min	Max	Obs.
Dependent variable	lnTEC	−0.014	2.455	−14.359	6.001	3150
	lnTECO	−1.697	2.869	−13.816	3.712	3150
	lnTECX	−1.055	3.588	−14.458	5.946	3150
Independent variable	GF	0.435	0.100	0.229	0.629	3150
Mediating variables	GP	2321.738	4490.374	3.000	35,699	3150
	RD	3.828	5.912	0.013	36.765	3150
	AIS	0.301	0.222	0.002	0.872	3150
	RIS	0.201	0.125	0.008	0.608	3150
Control variables	GDP	1632.265	5554.978	0.002	106,164	3150
	POP	4492.687	2754.083	528.600	12,624	3150
	EI	644.951	407.249	126.975	2502.291	3150
	EINV	248.382	245.662	0.706	1663.235	3150
	LQ	1.700	0.706	0.320	4.125	3150
	FDI	161.083	327.424	0.700	2745.000	3150
	OPEN	9899.750	18,667.250	19.664	102,402	3150

Note: Because the original values of trade-embodied carbon variables differ substantially in magnitude, lnTEC, lnTECO, and lnTECX are reported in natural logarithmic form, whereas all other variables are reported in their original form.

.

6. Empirical Results

6.1. Baseline Regression Results

Before reporting the baseline regression results, this study first conducts multicollinearity diagnostics using both pairwise correlation analysis and the variance inflation factor (VIF). The correlation analysis shows that the absolute values of the pairwise correlation coefficients among the explanatory variables are all below the commonly used warning threshold of 0.8, suggesting that no severe linear association exists among the regressors [61]. The VIF results further indicate that the maximum VIF is 2.53 and the mean VIF is 1.96, both well below the commonly used threshold of 10 and the stricter threshold of 5 [62]. Overall, these results suggest that multicollinearity is unlikely to materially affect the baseline estimation. For brevity, the detailed correlation matrix and VIF statistics are reported in Appendix C.

The baseline estimation results for the effect of green finance on trade-embodied carbon are reported in

Table 2. Results of the baseline estimations.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Variables	lnTEC	lnTEC	lnTECO	lnTECO	lnTECX	lnTECX
GF	−1.414 ***	−1.632 ***	−2.449 ***	−2.486 ***	−1.915 ***	−1.976 ***
	(0.489)	(0.475)	(0.878)	(0.911)	(0.401)	(0.392)
lnGDP		0.268 ***		0.042		0.842 ***
		(0.035)		(0.045)		(0.045)
lnPOP		2.047 ***		2.361 **		1.507 ***
		(0.637)		(1.041)		(0.529)
lnEI		0.486 ***		0.380		0.906 ***
		(0.168)		(0.305)		(0.166)
lnEINV		−0.001		−0.001		−0.007 ***
		(0.002)		(0.004)		(0.002)
LQ		0.020 ***		0.024 ***		0.016 ***
		(0.003)		(0.005)		(0.003)
lnFDI		−0.037		−0.069		−0.216 **
		(0.088)		(0.165)		(0.086)
lnOPEN		−0.061		0.097		−0.467 ***
		(0.101)		(0.153)		(0.114)
Constant	−1.232 ***	−10.43 **	−3.164 ***	−14.14 *	−6.730 ***	−9.475 **
	(0.420)	(4.468)	(0.757)	(8.143)	(1.055)	(4.044)
Province Effect	YES	YES	YES	YES	YES	YES
Industry Effect	YES	YES	YES	YES	YES	YES
Time Effect	YES	YES	YES	YES	YES	YES
Observations	3150	3150	3150	3150	3150	3150
R-squared	0.722	0.754	0.604	0.612	0.703	0.819

Note: Standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

. Total trade-embodied carbon (TEC) is used as the dependent variable in Models 1 and 2, trade-embodied carbon outflows (TECO) in Models 3 and 4, and trade-embodied carbon exports (TECX) in Models 5 and 6. In each case, the odd-numbered models report the benchmark specification, whereas the even-numbered models additionally include the full set of control variables. Province, industry, and time fixed effects are included throughout.

Across all specifications, the coefficient of green finance remains negative and statistically significant, suggesting that green finance helps to reduce trade-embodied carbon. Focusing on the specifications with the full set of controls, a 0.01 increase in the green finance index is associated with an approximately 1.632% decrease in total trade-embodied carbon, a 2.486% decrease in trade-embodied carbon outflows, and a 1.976% decrease in trade-embodied carbon exports. These results indicate that the carbon-mitigation effect of green finance is strongest in the domestic outflow dimension, followed by the export dimension, and then total trade-embodied carbon. This pattern is broadly consistent with existing studies showing that green finance and related instruments can lower carbon emission intensity, enhance carbon emission efficiency, ease financing constraints on green innovation, and support environmentally beneficial long-term investment [12,20,21,22]. It is also consistent with related evidence that green finance can promote cleaner export upgrading and improve the environmental performance of exporting firms [24,25,26]. Relative to this literature, our findings further show that the environmental effects of green finance extend to the value-chain perspective of trade-embodied carbon, with a particularly pronounced effect along domestic production linkages. Overall, the baseline estimates provide consistent evidence that green finance significantly reduces trade-embodied carbon in China’s manufacturing sector. Therefore, Hypothesis 1 is supported. Given that the three dependent variables yield consistent signs and significance levels, the subsequent robustness and mechanism analyses focus primarily on TEC, which captures the aggregate level of trade-embodied carbon, whereas the corresponding results for TECO and TECX are reported as supplementary evidence.

6.2. Endogeneity Test Results

Potential endogeneity in the relationship between green finance and trade-embodied carbon may arise from reverse causality and omitted variables. On the one hand, green finance may affect trade-embodied carbon through resource allocation and low-carbon transformation. On the other hand, changes in regional trade-embodied carbon may also influence local green finance policies and the allocation of financial resources. To mitigate these concerns, this study adopts an instrumental variable strategy.

This study uses two instrumental variables. The first is a Bartik-style (shift-share) instrument [63], which is constructed as the interaction between the number of bank branches per 10,000 people in 2002 and the leave-one-out national green finance level:

$${IV\_Bartik}_{it}={Branch}_i^{2002}\times {\overline{GF}}_{-i,t}$$

(10)

where ${Branch}_i^{2002}$ denotes the number of bank branches per 10,000 people in province $i$ in 2002, and ${\overline{GF}}_{-i,t}$ denotes the leave-one-out national green finance level in year $t$, calculated as follows:

$${\overline{GF}}_{-i,t}={\displaystyle \frac{1}{N-1}}\sum _{k\ne i}{GF}_{kt}$$

(11)

where $N$ is the total number of provinces in the sample, and ${GF}_{kt}$ is the green finance index of province $k$ in year $t$. As a Bartik-style (shift-share) instrument, this interaction term combines a province-specific initial exposure with a common national shock. The initial financial infrastructure of a province, measured by ${Branch}_i^{2002}$, shapes its capacity to absorb and implement subsequent green finance policies. Provinces with denser historical banking networks tend to be more responsive to nationwide green finance expansion, providing a clear first-stage channel linking the instrument to current green finance. At the same time, bank branch density in 2002 reflects a historically determined initial condition of financial development that predates current carbon-reduction policies and is therefore plausibly exogenous to contemporaneous shocks affecting regional trade-embodied carbon [64,65,66]. Moreover, the leave-one-out national trend, ${\overline{GF}}_{-i,t}$, captures nationwide changes in green finance while excluding province $i$’s contribution, thereby reducing the possibility that the instrument is mechanically driven by local contemporaneous conditions. In this Bartik-style design, identification comes from differences in provinces’ exposure to a common national green finance expansion through their initial financial infrastructure [67]. Although historical financial infrastructure may be correlated with general industrial activity, the identified variation in the interaction term reflects heterogeneous provincial responses to a common national shock rather than contemporaneous local industrial shocks. In addition, the model includes province, industry, and year fixed effects, together with a rich set of control variables such as GDP and population, which absorb persistent regional characteristics and common macroeconomic changes. Under this identification strategy, the interaction term influences current trade-embodied carbon only through its effect on provincial green finance.

The second instrument is the one-period lag of the provincial green finance index (IV_L.GF). It is constructed as the lagged value of green finance for each province in the panel. The rationale is that green finance development exhibits strong persistence due to institutional continuity, policy implementation inertia, and the gradual accumulation of green financial resources. Therefore, past green finance is expected to be highly correlated with current green finance. Meanwhile, conditional on province, industry, and year fixed effects, province–industry interaction fixed effects, and other controls, the lagged value is less likely to affect current trade-embodied carbon except through its impact on the current level of green finance [68]. To further strengthen identification, this study uses the lagged instrument jointly with the Bartik-style instrument rather than relying on it alone.

Table 3. Results of the endogeneity test.

	First Stage	Second Stage
Variables	GF	lnTEC
IV_Bartik	0.455 ***
	(0.029)
IV_L.GF	0.904 ***
	(0.044)
GF		−2.193 ***
		(0.404)
Controls	YES	YES
Province effect	YES	YES
Industry effect	YES	YES
Time effect	YES	YES
Observations	2700	2700
Durbin–Wu–Hausman test	18.384 ***
Underidentification test	1070.258 ***
(Kleibergen–Paap rk LM stat)
Weak instrument test	1016.514
(Cragg–Donald Wald F stat)
Overidentification test	0.593
Hansen J statistic (p value)

Note: Standard errors are reported in parentheses. Due to the use of the one-period lag of the provincial green finance index as an instrumental variable, the first benchmark-year observations are excluded from the IV estimation sample; therefore, the number of observations reported in Table 3 is 2700. *** p < 0.01.

reports the instrumental-variable estimation results. The Durbin–Wu–Hausman test is significant at the 1% level, rejecting the null hypothesis that green finance is exogenous and thereby confirming the necessity and appropriateness of adopting the instrumental variable approach. In the first-stage regression, both instruments, IV_Bartik and IV_L.GF, enter positively and significantly, indicating that they are strongly correlated with green finance. The underidentification test is significant, suggesting that the model does not suffer from underidentification. The weak instrument test further indicates that instrument weakness is unlikely to be a concern, while the Hansen J test fails to reject the null hypothesis that the overidentifying restrictions are valid, thereby supporting the overall validity of the instruments. In the second stage, the estimated coefficient on green finance remains significantly negative, which is consistent with the baseline results. This finding suggests that the carbon reduction effect of green finance remains robust after accounting for potential endogeneity.

6.3. Robustness Test Results

To ensure that the baseline results are not driven by specific variable definitions, estimation methods, omitted policy factors, sample composition, or the treatment of extreme observations, this study conducts a series of robustness checks.

As reported in

Table 4. Results of robustness test 1.

	Model 1	Model 2	Model 3	Model 4	Model 5
Variables	NTEC	lnTEC	Add. ER	OLS	PPML
GF	−0.748 **		−1.688 ***	−1.073 ***	−0.551 **
	(0.350)		(0.475)	(0.179)	(0.230)
GF2		−1.581 ***
		(0.398)
ER			−0.006 **
			(0.002)
Constant	6.242 **	−8.243 *	−11.65 **	−4.492 ***	1.393 **
	(3.074)	(4.303)	(4.608)	(0.888)	(0.640)
Controls	YES	YES	YES	YES	YES
Province effect	YES	YES	YES	YES	YES
Industry effect	YES	YES	YES	YES	YES
Time effect	YES	YES	YES	YES	YES
Observations	3150	3150	3150	3150	3150

Note: Standard errors are reported in parentheses. In the PPML specification, the dependent variable is estimated in its original level. *** p < 0.01, ** p < 0.05, * p < 0.1.

, Model 1 replaces the dependent variable with net trade-embodied carbon (NTEC) (Note: In Model 1, the dependent variable is replaced by net trade-embodied carbon (NTEC), which is calculated as NTEC = TEC − TIC, where TEC denotes total trade-embodied carbon and TIC denotes trade-embodied carbon inflows. A higher value of NTEC indicates a greater net outward transfer of embodied carbon.). Model 2 replaces the core explanatory variable with an alternative measure of green finance. Model 3 further adds environmental regulation intensity (ER) as an additional control variable [69]. This indicator is constructed from the total frequency of environment-related keywords in provincial government reports, including terms related to environmental protection, pollution control, energy consumption, emission reduction, ecological governance, green development, and low-carbon transition. As a text-based proxy, it captures the degree of local policy attention and regulatory pressure on environmental issues and can also partly reflect broader industrial or regional policy interventions [70]. Therefore, incorporating this variable helps to alleviate the concern that the estimated effect of green finance may be biased by omitted environmental or policy factors that simultaneously influence both green finance and trade-embodied carbon. Models 4 and 5 then re-estimate the baseline specification following the model setting in Section 5.1.1, using OLS and PPML, respectively, as robustness checks to examine whether the main findings remain stable under alternative estimation methods. It should be noted that, unlike the baseline HDFE and OLS specifications, the PPML model is estimated using the level values of the dependent variable rather than its logarithmic form. Therefore, the PPML results are used mainly to verify the sign and significance of the estimated effect of green finance, rather than to make direct comparisons of coefficient magnitudes across models.

Model 6 excludes the four centrally administered municipalities [71]. Models 7 to 9 apply winsorization to the variables at the 1%, 5%, and 10% levels [72], respectively. These robustness checks are intended to examine whether the estimated effect of green finance on trade-embodied carbon remains stable under alternative variable definitions, estimation strategies, sample restrictions, and outlier treatments. As shown in

Table 5. Results of robustness test 2.

	Model 6	Model 7	Model 8	Model 9
Variables	Exclude Munic.	Winsorize 1%	Winsorize 5%	Winsorize 10%
GF	−1.803 ***	−1.585 ***	−1.305 ***	−1.303 ***
	(0.517)	(0.387)	(0.343)	(0.337)
Constant	−24.86 ***	−7.185 **	−0.511	2.204
	(5.838)	(3.646)	(2.995)	(2.548)
Controls	YES	YES	YES	YES
Province effect	YES	YES	YES	YES
Industry effect	YES	YES	YES	YES
Time effect	YES	YES	YES	YES
Observations	2730	3150	3150	3150

Note: Standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05.

, the coefficient on green finance remains significantly negative in all four models. These results indicate that the baseline conclusion is robust to excluding municipalities and to different winsorization thresholds.

6.4. Mechanism Test Results

6.4.1. Technological Effect

To examine whether green finance mitigates trade-embodied carbon through technological channels, this study uses green patents (GP) and R&D expenditure (RD) as mediating variables. The results are reported in

Table 6. Results of the mechanism test: Technological effect.

	Model 1	Model 2	Model 3	Model 4
Variables	lnGP	lnTEC	lnRD	lnTEC
GF	0.328 ***	−1.545 ***	9.475 ***	−1.515 ***
	(0.096)	(0.469)	(0.759)	(0.489)
lnGP		−0.264 **
		(0.126)
lnRD				−0.012 **
				(0.006)
Constant	−5.791 ***	−11.96 **	−53.21 ***	−11.08 **
	(0.937)	(4.656)	(6.234)	(4.637)
Controls	YES	YES	YES	YES
Province effect	YES	YES	YES	YES
Industry effect	YES	YES	YES	YES
Time effect	YES	YES	YES	YES
Observations	3150	3150	3150	3150
Mediation Effect Tests
Sobel Z-test	−1.790 *		−1.970 **
Bootstrap (1000 reps)
Indirect effect	−0.087 **		−0.114 **
	(0.044)		(0.056)
95% Confidence interval	[−0.441, −0.016]		[−0.291, −0.042]

Note: Standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

. Green finance significantly promotes both green patenting and R&D expenditure, whereas lnGP and lnRD are both significantly negatively associated with trade-embodied carbon. After the mediating variables are included, the coefficient on green finance remains significantly negative but decreases in absolute value, indicating partial mediation through both channels. Overall, these results support Hypothesis 2.

The significance of the indirect effects is further examined using the Sobel test together with a bootstrap procedure based on 1000 replications. As shown in the lower panel of Table 6, the Sobel Z-statistics are significant for both the GP and RD channels. In addition, the estimated indirect effects are significantly negative, and the corresponding 95% bootstrap confidence intervals do not include zero. These findings indicate that green finance reduces trade-embodied carbon partly through technological channels.

6.4.2. Structural Effect

To examine whether green finance reduces trade-embodied carbon through structural channels, this study uses industrial structure advancement (AIS) and industrial structure rationalization (RIS) as mediating variables. The results are reported in

Table 7. Results of the mechanism test: Structural effect.

	Model 1	Model 2	Model 3	Model 4
Variables	AIS	lnTEC	RIS	lnTEC
GF	0.453 ***	−1.133 **	0.197 ***	−1.282 ***
	(0.046)	(0.480)	(0.029)	(0.474)
AIS		−1.311 ***
		(0.191)
RIS				−1.773 ***
				(0.399)
Constant	3.069 ***	−10.81 **	0.432	−9.661 **
	(0.604)	(5.031)	(0.417)	(4.263)
Controls	YES	YES	YES	YES
Province effect	YES	YES	YES	YES
Industry effect	YES	YES	YES	YES
Time effect	YES	YES	YES	YES
Observations	3150	3150	3150	3150
Mediation Effect Tests
Sobel Z-test	−6.860 ***		−4.443 ***
Bootstrap (1000 reps)
Indirect effect	−0.594 ***		−0.349 ***
	(0.113)		(0.098)
95% Confidence interval	[−0.815, −0.372]		[−0.541, −0.158]

Note: Standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05.

. Green finance significantly promotes industrial structure advancement and significantly improves industrial structure rationalization, whereas both AIS and RIS are significantly negatively associated with trade-embodied carbon. After the mediating variables are included, the coefficient on green finance remains significantly negative, indicating partial mediation effects through both structural channels. Overall, these results support Hypothesis 3.

The significance of the indirect effects is further examined using the Sobel test together with a bootstrap procedure based on 1000 replications. As shown in the lower panel of Table 7, the Sobel Z-statistics are significant at the 1% level for both the AIS and RIS channels. In addition, the estimated indirect effects are significantly negative, and the corresponding 95% bootstrap confidence intervals do not include zero. These findings indicate that green finance reduces trade-embodied carbon partly through structural channels.

6.5. Heterogeneity Analysis Results

6.5.1. Regional Heterogeneity

As shown in

Table 8. Results of the heterogeneity analysis: Regional heterogeneity.

	Model 1	Model 2	Model 3	Model 4
Variables	Eastern	Central	Northeastern	Western
GF	−2.144 ***	−2.424 **	−5.327	0.469
	(0.607)	(0.996)	(5.756)	(0.974)
Constant	−26.88 **	108.8 ***	29.19	−67.15 ***
	(10.647)	(22.976)	(56.473)	(12.590)
Controls	YES	YES	YES	YES
Province effect	YES	YES	YES	YES
Industry effect	YES	YES	YES	YES
Time effect	YES	YES	YES	YES
Observations	1050	630	315	1155
R-squared	0.871	0.866	0.797	0.715

Note: Standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05.

, the effect of green finance on trade-embodied carbon varies across regions, indicating substantial regional heterogeneity. The estimated coefficient on green finance is significantly negative in both the eastern and central regions but remains insignificant in the northeastern and western regions. Specifically, a 0.01 increase in the green finance index is associated with an approximately 2.144% decrease in trade-embodied carbon in the eastern region and a 2.424% decrease in the central region. By contrast, the coefficients for the northeastern and western regions do not reach conventional significance levels, suggesting that the carbon mitigation effect of green finance is concentrated mainly in the eastern and central parts of China [73]. One possible explanation is that these regions generally have more developed financial systems, stronger industrial bases, and more favorable conditions for green investment and technology diffusion, which allow green finance to play a more effective role. By comparison, the lack of statistical significance in the northeastern and western regions may be closely related to their stronger reliance on resource-based and heavy industries. The northeastern region has long been an important heavy industry base in China, while the western region plays a major role in energy extraction and primary resource processing. These industrial characteristics are often associated with stronger path dependence and carbon lock-in effects in carbon- and energy-intensive production systems [49,74]. Under such conditions, low-carbon transition usually requires more than marginal technological improvement and often involves longer investment cycles, larger adjustment costs, and deeper structural transformation [75]. Therefore, the current scale and allocation efficiency of green finance in these regions may be insufficient to overcome the inertia of resource-dependent industrial structures in the short-term, thereby weakening its observable effect on trade-embodied carbon.

6.5.2. Sectoral Heterogeneity

Table 9. Results of the heterogeneity analysis: Sectoral heterogeneity.

	Model 1	Model 2
Variables	Energy-Intensive Industries	Non-Energy-Intensive Industries
GF	−2.478 ***	−1.938 ***
	(0.699)	(0.570)
Constant	−9.965 ***	−10.55 ***
	(2.798)	(2.575)
Controls	YES	YES
Province effect	YES	YES
Industry effect	YES	YES
Time effect	YES	YES
Observations	840	2310
R-squared	0.763	0.739

Note: Standard errors are reported in parentheses. *** p < 0.01.

reports the heterogeneity analysis by industry group, distinguishing between energy-intensive and non-energy-intensive sectors. The coefficient on green finance is significantly negative in both subsamples, indicating that the carbon mitigation effect of green finance operates across different types of manufacturing industries. Specifically, a 0.01 increase in the green finance index is associated with an approximately 2.478% decrease in trade-embodied carbon in energy-intensive industries and a 1.938% decrease in non-energy-intensive industries. Comparatively, the mitigating effect of green finance is stronger in energy-intensive industries. One possible explanation is that such industries generally have higher carbon intensity and face greater pressure for low-carbon upgrading and, so, the role of green finance in supporting cleaner production, technological renovation, and emissions reduction can be more clearly reflected [76,77].

7. Conclusions

Based on the embedded environmentally extended multi-regional input–output (EE-MRIO) framework and high-dimensional fixed-effects models, this study investigates the effect of green finance on trade-embodied carbon in China’s manufacturing sector and its implications for the sector’s sustainable transition. The main findings are summarized as follows.

First, trade-embodied carbon in China’s manufacturing sector shows clear spatial expansion and structural differentiation. During the sample period, the overall scale of trade-embodied carbon increased markedly, with high-value provinces becoming increasingly concentrated in North China and the eastern coastal manufacturing belt. From a structural perspective, the growth in inland and resource-intensive provinces was driven mainly by embodied carbon outflows in domestic trade, whereas coastal provinces experienced a more pronounced increase in export-embodied carbon in international trade. This pattern reflects the differentiated roles of inland and coastal provinces in domestic and global production networks.

Second, green finance significantly and robustly reduces trade-embodied carbon. This mitigating effect holds not only for total trade-embodied carbon but also for embodied carbon outflows and embodied carbon exports, with a relatively stronger effect in the domestic trade dimension. The main findings remain stable after addressing endogeneity and conducting a series of robustness checks.

Third, green finance reduces trade-embodied carbon through both technological and structural channels. On the one hand, green finance facilitates green innovation and R&D investment, thus contributing to carbon mitigation through technological advancement. On the other hand, it facilitates industrial structure advancement and rationalization, thereby reducing trade-embodied carbon through structural upgrading. These channels are consistently supported by the mechanism analysis.

Fourth, the carbon mitigation effect of green finance varies across regions and sectors. At the regional level, the effect is more evident in the eastern and central regions, while no significant effect is found in the northeastern and western regions. At the sectoral level, the effect is larger in energy-intensive industries than in non-energy-intensive industries, suggesting that green finance plays a stronger role in supporting low-carbon transition in high-emission sectors.

8. Discussion and Limitations

8.1. Contributions and Implications

This study makes three key contributions to the literature. First, by bringing a value-chain trade perspective into the analysis of the environmental effects of green finance, it extends the existing literature and highlights the role of green finance in reducing trade-embodied carbon and facilitating the sustainable transition of manufacturing within interconnected domestic and global production networks. Second, it develops an integrated analytical framework to examine the mechanisms through which green finance influences trade-embodied carbon, and shows that technological progress and structural upgrading are two key channels linking green finance to low-carbon and sustainable manufacturing development. Third, by accounting for both regional and sectoral heterogeneity, the study provides a more refined understanding of the sustainability effects of green finance and generates policy-relevant evidence for supporting greener and more sustainable manufacturing transformation in China.

Several policy implications emerge from these findings. First, green finance policies should be designed in a more regionally differentiated manner. Provinces with stronger financial foundations and more advanced industrial upgrading are better positioned to expand pilot programs and green financial innovation, whereas central, inland, and resource-dependent provinces require greater support in institutional capacity, project identification, and access to financing for low-carbon investment [78]. Second, policy instruments should better reflect sectoral heterogeneity. For energy-intensive industries, the policy focus should prioritize transition finance, using instruments such as transition loans and transition bonds to support emission-reduction-oriented technological upgrading [79]. For non-energy-intensive industries, green finance should play a greater role in supporting firms’ green transformation and improving the green competitiveness of their products. Third, policy transmission should be calibrated to different trade channels. In domestic supply chains, greater attention should be given to the development of green supply chain finance to strengthen low-carbon coordination among upstream and downstream firms and reduce embodied carbon transfer across interregional production linkages [80]. In export-oriented value chains, policy support should place greater emphasis on green upgrading, green certification, and compliance with climate-related trade standards [81]. Finally, the role of green finance in reducing trade-embodied carbon can be strengthened through closer coordination with industrial upgrading policies, regional development strategies, and the implementation of the “dual carbon” goals. In particular, policy efforts should support the development of green value chains, such that green finance can better promote the joint transformation of production processes, supply chain linkages, and trade structures.

8.2. Limitations and Future Study Directions

This study has several limitations that should be acknowledged, as detailed below.

First, the analysis is constrained by data timeliness. The measurement of trade-embodied carbon relies on China’s input–output tables, and the latest available table currently extends only to 2020. Although the 2020 data can preliminarily reflect the policy context of China’s “Dual Carbon” goals, they cannot fully capture the subsequent evolution of green finance policies or the more recent structural adjustments in China’s manufacturing sector after 2020.

Second, the China-embedded MRIO framework is subject to both approximation and aggregation constraints. Because province-specific foreign destination and source structures are not directly observed, province–foreign linkages are inferred from the structural coefficients of the national China account in the OECD-ICIO table. In addition, harmonization of the OECD-ICIO table and China’s provincial input–output table requires sector aggregation. In the empirical analysis, manufacturing is classified into 15 broad sectors, which may smooth heterogeneity in production technology and carbon intensity within the same sector. These constraints imply that the estimated trade-embodied carbon measures are better suited to identifying overall patterns and relative differences across provinces and sectors than representing highly disaggregated carbon accounting relationships.

Third, the empirical analysis is conducted at the provincial industry level, which also defines the scope of inference of this study. Since trade-embodied carbon is inherently constructed within an input–output framework, the estimated relationships in this paper are better understood as meso-level manifestations of the relevant transmission channels, rather than as direct identification of firm- or plant-level behavioral responses. The proposed mechanisms are consistent with the existing micro-level literature, which shows that green finance can ease financing constraints, promote green innovation, and support low-carbon investment. In this sense, the mechanism analysis in this study is intended to capture the meso-level consequences of these micro foundations for trade-embodied carbon. Moreover, although firm-level validation would be a useful extension, the micro datasets commonly used in related research—such as the Chinese Industrial Enterprise Database and the Chinese Customs Database—mainly cover earlier periods. For a policy field such as green finance, whose institutional development and policy implementation have evolved substantially in recent years, these earlier micro datasets may not adequately capture its more recent characteristics and effects. Therefore, they would provide only limited support as a direct validation of the mechanisms examined in this study.

Fourth, although the green finance index covers six core dimensions, some qualitative aspects of green financial development remain difficult to capture at the provincial level. In particular, differences in policy implementation, screening standards, financial product quality, and risk-sharing arrangements are not fully observable in the current measure. As a result, the index may not completely reflect the institutional depth and operational quality of green finance across provinces.

Fifth, several advanced econometric extensions are not pursued in this study due to the structure of the available data. The trade-embodied carbon measures are constructed from a limited number of discrete benchmark years rather than from continuous annual observations, which does not support formal dynamic panel estimation and also constrains rigorous structural break testing. In addition, although spatial spillovers are theoretically plausible, they are difficult to model rigorously within the province–industry framework adopted here. The dependent variable is measured at the provincial industry level, whereas available spatial weight matrices can be constructed only at the provincial level. It is therefore challenging to build a spatial industry weight matrix that simultaneously reflects interprovincial proximity and inter-industry linkages. Applying a common provincial spatial matrix to all industry observations would risk masking cross-industry heterogeneity, while aggregating trade-embodied carbon to the provincial level would weaken the analytical advantage of this indicator relative to direct carbon emissions by discarding the industry dimension that is central to the present study.

Future research can proceed in several directions. First, as more recent benchmark data become available, future studies could update the China-embedded MRIO framework to examine whether the effect of green finance has changed under the “Dual Carbon” strategy and the evolving policy environment after 2020. Second, future research could refine the identification of province–foreign production linkages, reduce aggregation bias by using more detailed sectoral information or alternative embedding strategies, and further improve the accuracy of trade-embodied carbon accounting. Third, subsequent studies could strengthen the measurement of green finance by incorporating richer information on policy implementation, institutional quality, and financial product design. Finally, with more suitable high-dimensional weight matrices, finer time-series observations, and better integrated multi-source data, future work could further explore the dynamic adjustment, spatial spillover, structural change, and micro-level foundations of the relationship between green finance and trade-embodied carbon.