Exploring the Relationship Between Green Finance and Carbon Productivity: The Mediating Role of Technological Progress Bias

Abstract

In the context of global climate change, achieving a green and low-carbon economic transition is essential for sustainable development. This study constructs a model using data from 30 provinces collected between 2006 and 2020 to investigate how green finance influences China’s carbon productivity and the transmission mechanism mediated by factor-biased technological progress. The findings reveal the following: (1) The Moran’s index test for carbon productivity across Chinese provinces demonstrates significant spatial clustering. (2) Green finance exhibits substantial spillover effects on carbon productivity in surrounding regions. (3) Capital-biased and energy-biased technological progress significantly mediate the relationship between green finance and carbon productivity, indicating that green finance enhances carbon productivity by optimizing the allocation of capital, labor, and energy factors. (4) Regional heterogeneity analysis indicates that capital-technology-biased and energy-factor-technology-biased approaches can significantly enhance carbon productivity in Central and Northeastern China. Notably, energy-factor innovation delivers far greater environmental efficiency gains in these regions than in Eastern and Western China.

1. Introduction

China’s green finance policy has been consistently improved by fostering a low-carbon and green economy and enhancing the market for green finance, which is constantly growing and primarily centered on green credit and augmented by green bonds, carbon financing, and green insurance [1]. Because sustainable development projects are expensive, risky, and have a protracted investment return cycle, they place increased demand on financial intermediaries. In addition to effectively addressing the drawbacks of conventional financial investment, growing the green finance market can help China achieve its carbon emission reduction goals.

Most of the existing research on how green finance affects carbon productivity points to a favorable relationship between the two. One study uses quartile regression to confirm the beneficial impact of green finance on increasing carbon productivity, examining how this relationship varies across different quartiles and in response to the continuously changing environment. Green finance support in ten nations was analyzed [2]. A larger-scale study shows that green credit’s carbon emission reduction effect is superior to that of green venture capital, but the two types of financial products both have significant roles in reducing carbon emissions [3]. Green finance also plays a significant role in enhancing carbon productivity [4].

The pace and quality of economic development are determined by technological advancements, which is why green finance is essential for promoting enterprises that are involved with green technological innovation. Green technological innovation cannot be developed without the support of green finance [5]. Some scholars have researched energy-saving environmental protection enterprises [6]. One study found that an enterprise’s green credit can effectively foster technological innovation in environmental protection enterprises, but this role has a lag effect. The degree of financing constraints will also have an impact on the green credit’s ability to foster technological innovation in enterprises. Building a spatial econometric model based on provincial green patent data, it was discovered that green credit has a considerable promotion effect on raising the level of green technological innovation in a region and that R&D expenditure is the primary means of realizing this effect [7]. Furthermore, it has been confirmed that technological advancements have a moderating influence on the process of green financing, which enhances environmental quality and fosters superior economic development.

A crucial route to reaching carbon neutrality in the framework of dual carbon targets is improving research and investment in cutting-edge energy technology. There is a wealth of current research on the relationship between progress in technology and carbon productivity. The work covers both broad and specialized technological progress, as well as endogenous and external technological progress. Based on current research findings, several academics believe that technological advancements can significantly increase carbon productivity. On the basis of the climate change model RICE, the variable of research and development (R&D) was introduced [8]: as time passes, the stock of knowledge will facilitate a greater level of R&D, which will affect the output rate and lower the carbon emissions per unit of output. In terms of classifying technological progress, it has been maintained that while environmental and capital-biased technologies help to increase carbon production, broad and energy-biased technological advancement decrease it [9].

This raises the following questions: What effect does green finance actually have on carbon productivity? Is there a spillover effect of green finance on carbon productivity? Can green finance affect carbon productivity through different technological progress? What are the processes via which green financing contributes to carbon productivity at different levels of agglomeration, considering the variations in carbon productivity levels between regions? Based on these questions, this study uses panel data from 30 provinces in China to firstly construct the economic geography weight matrix and explore the spatial distribution characteristics of carbon productivity through spatial measurement; secondly, the relationship between green finance and carbon productivity is examined by using the spatial Durbin model; lastly, using the intermediary model, the process underlying how various technology advancements affect carbon productivity is confirmed.

2. Mechanism Analysis and Hypothesis

2.1. Green Finance and Carbon Productivity

Green finance refers to financial services provided to support economic activities that promote environmental improvement, address climate change, and encourage resource conservation and efficient utilization [10,11,12]. Carbon productivity denotes the carbon dioxide intensity per unit of GDP generated [13]. According to current research, we have found that by encouraging the development of green technology in industrial businesses and enhancing the effectiveness of their factor use, green finance growth reduces carbon emissions per unit. There is no obvious economic profit in the pre-innovation period because green innovation has its own characteristics of sunk investment, uncertain results, and high financing costs. In contrast to traditional innovation, green innovation places more emphasis on resource conservation, environmental protection, maintenance of the ecological environment, and the improvement of product yield [14]. In contrast to environmental protection enterprises that are relatively new in China, it is easier to gain the favor of financial institutions in the traditional financial services system due to the pressure of profitability, which is often achieved through the assessment of corporate collateral assets, follow-up business capacity, and the current market size to screen customers [15]. In this context, it is difficult to obtain financing support for those who really have market potential for green technology. In order to increase the relevant scientific research investment in industrial enterprises and to further enhance their level of innovation in relation energy saving and environmental protection, as well as to achieve carbon emission reduction, green finance development can, therefore, greatly lower the investment threshold of environmental protection companies and their affiliates, solve the financing conundrum, and build a green project library or use green special funds [16].

Hypothesis 1.

The development of green finance has a positive contribution to carbon productivity in each region.

2.2. Mediating Role of Technological Progress Bias

Green finance aims to alleviate the financial barriers to green technological advancement and foster the growth of a green economy. Its advancement is also a crucial step towards China’s dual-carbon objectives. Green finance furthers technological advancement in various areas by directing financial resources, boosting enterprise innovation, and altering production structures, all of which can lower carbon emissions.

Capital-biased technical advancement can have an impact on carbon emissions through the use of green finance. By encouraging businesses to invest more in fixed assets, adopt and promote advanced green technology, increase the rate at which capital factors are utilized, encourage technological advancement and industrial upgrading, and support the transformation of the economic growth model, rational allocation of green financial resources can help mitigate the conflict between economic growth and carbon emissions. Capital investment is a requirement for businesses to engage in technological advancement activities. Similarly, investing in fixed assets is a condition for businesses to employ and promote green technology. Businesses will spend more on capital factors and use fewer energy factors in order to maximize profits when the price of production factors changes, which will lower carbon emissions.

Hypothesis 2.

Green finance promotes carbon productivity by influencing the factor bias of capital technology progress.

Energy-biased technology advancement can have an impact on carbon emissions through green finance [13]. Green finance can help reduce carbon emissions by limiting the amount of funding available to high-energy-consuming and high-pollution businesses, compelling them to modernize and alter their processes, increasing the efficiency of their energy use and lowering their emissions of polluting gases. Green finance, on the one hand, encourages the growth of environmentally friendly businesses by providing energy-saving and environmental protection industries with favorable loans and other forms of financial support, which encourages the sector to expand and lowers its energy consumption per GDP unit. On the other hand, green finance can direct economic actors toward ecological environmental protection, boost investments in industries that save energy and protect the environment, highlight the green transformation market, and direct different types of enterprises toward green initiatives like the production of clean energy, energy conservation, and carbon reduction, as well as toward further reducing carbon emissions. Reducing energy consumption at the source lowers carbon emissions and increases carbon productivity. Green initiatives such as reducing carbon emissions have also set higher standards for green innovation in both corporate and societal spheres.

Hypothesis 3.

Green finance contributes to carbon productivity improvement by influencing the factor bias of energy technology progress.

Figure 1 demonstrates the conceptual framework of this paper.

3. Data Sources and Modeling

3.1. Data Sources and Variable Definitions

3.1.1. Data Sources

In this paper, 30 provinces in China (excluding Tibet, Hong Kong, Macao, and Taiwan) are selected as the research object. The data covering the years from 2006 to 2020, were obtained from China Statistical Yearbook, China Population and Employment Statistical Yearbook, China Energy Statistical Yearbook [13] (https://www.stats.gov.cn/sj/ndsj/, accessed on 23 September 2025), and the statistical yearbooks of each of the respective provinces.

3.1.2. Definition of Variables

Explained variable—carbon productivity (CP). Carbon productivity is measured using the ratio of Gross Regional Product/Regional CO₂ used in the wider literature. The specific calculation formula is

$$CP={\displaystyle \frac{GDP}{E}}$$

(1)

$$E={\sum }_{i=1}^{30}{\sum }_{j=1}^8{AC}_{ij}·{NCV}_j·{CC}_{i,j}·O_j·{\displaystyle \frac{44}{12}}$$

(2)

where GDP denotes gross domestic product; E denotes CO₂ emissions by region (industrial end-use consumption of raw coal, coke, crude oil, gasoline, kerosene, diesel oil, fuel oil, and natural gas was selected to calculate CO₂ emissions according to the methodology in the 2006 IPCC Guidelines for National Greenhouse Gas Inventories); ${AC}_{ij}$ denotes the i region’s $\mathrm{j}$ type of energy consumption; ${NCV}_j$ denotes the average low-level heat content of the first energy source; ${CC}_{i,j}$ denotes the carbon content per unit calorific value; $O_j$ denotes the carbon oxidation rate of energy source; and 44/12 denotes the ratio of the relative molecular mass of CO₂ to that of carbon.

Core explanatory variable—green finance (GF). Green finance injects green capital into the economic system. According to the People’s Bank of China’s “Guiding Opinions on Establishing a Green Financial System,” this article draws upon the measurement methods of Zhang [17], selecting two primary indicators: green credit and green investment. Green credit is represented by the proportion of green loans relative to total credit, while green investment is measured by the density of listed environmental protection enterprises. Finally, the entropy method is employed to calculate weighted scores for each indicator, thereby determining the green finance development level of each province.

Control variables—Energy structure (ES), measured by the ratio of coal consumption to total primary energy consumption, and industrial agglomeration (IA). using the location entropy index to calculate the degree of industrial agglomeration [18], the specific calculation formula is

$$\mathrm{I}\mathrm{A}={\displaystyle \frac{{\mathrm{A}}_{\mathrm{i}\mathrm{t}}/{\mathrm{G}}_{\mathrm{i}\mathrm{t}}}{{\mathrm{A}}_{\mathrm{t}}/{\mathrm{G}}_{\mathrm{t}}}}$$

(3)

where ${\mathrm{A}}_{\mathrm{i}\mathrm{t}}$ and ${\mathrm{G}}_{\mathrm{i}\mathrm{t}}$ denote regions $\mathrm{i}$ in the $\mathrm{t}$ year industrial output and gross domestic product (GDP), and $A_t$ and $G_t$ denote the total industrial output and gross domestic product in the $\mathrm{t}$ year.

The degree of environmental regulation (ER), which measures the total investment in environmental pollution control as the share of the output value of the secondary industry in GDP, labor skills (LS), which is measured by the percentage of education above bachelor’s degree in each region, trade level (TL), which is measured by the total amount of trade in each province, and the optimization of the economic structure (ESO), which is measured by the ratio of the value added of the tertiary industry to the value added of the industry in each province, are used.

Mediating variables. We selected labor–capital technological progress bias (Bias_LK) and energy-enhancing technological progress bias (Bias_E) as mediating variables. The specific estimation process is as follows.

Based on Acemoglu’s [19] definition of technological progress bias and incorporating Dai ‘s [20] estimation methodology, this study quantifies the labor–capital bias in interprovincial technological progress across China. The specific estimation process is as follows: The most commonly used production function in the previous literature is the Cobb–Douglas production function. While this function provides a more intuitive measure of resource allocation in economic activities, it struggles to capture technological progress as represented by the capital–labor marginal productivity ratio. Therefore, drawing on David and Klundert’s [21] research on the direction of technological progress, this article assumes a CES production function:

$$Y_t={\left[\gamma {\left(A_{Kt}{K}_t\right)}^{\frac{\sigma -1}{\sigma }}+\left(1-\gamma \right){\left(A_{Lt}{L}_t\right)}^{\frac{\sigma -1}{\sigma }}\right]}^{\frac{\sigma }{\sigma -1}}$$

(4)

where $Y_t$ is the total output over time; $K_t$ and $L_t$ are the inputs of capital and labor in different periods; $A_{Kt}$ is capital-enhanced technological progress; $A_{Lt}$ is labor-enhanced technical progress in different periods; $\gamma$ is the capital income share; $1-\gamma$ is the share of labor income $\left(0<\gamma <1\right)$; and $\mathsf{\sigma }$ is the capital–labor elasticity of substitution $\left(\mathsf{\sigma }>0\right)$.

Again, according to Acemoglu’s definition

$$M_t={\displaystyle \frac{{MP}_K}{{MP}_L}}={\displaystyle \frac{\gamma }{1-\gamma }}{\left({\displaystyle \frac{A_{Kt}}{A_{Lt}}}\right)}^{\frac{\sigma -1}{\sigma }}{\left({\displaystyle \frac{K_t}{L_t}}\right)}^{-\frac{1}{\sigma } }$$

(5)

$${\displaystyle \frac{\partial M_t}{\partial (A_{Kt}/A_{Lt})}}={\displaystyle \frac{\gamma }{1-\gamma }}·{\displaystyle \frac{\sigma -1}{\sigma }}{\left({\displaystyle \frac{A_{Kt}}{A_{Lt}}}\right)}^{-\frac{1}{\sigma }}{\left({\displaystyle \frac{K_t}{L_t}}\right)}^{-\frac{1}{\sigma } }$$

(6)

When $\sigma >1$, ${\displaystyle \frac{\partial M_t}{\partial (A_{Kt}/A_{Lt})}}>$ 0, a rise in $A_{Kt}/A_{Lt}$ will lead to a rise in $M_t$, an increase in capital factor inputs will lead to an increase in marginal output, at which time technological progress is biased in favor of capital; similarly, when $\sigma <1$, technical progress is biased toward labor; when $\sigma =1,{\displaystyle \frac{\partial M_t}{\partial (A_{Kt}/A_{Lt})}}$ $=$ 0, there is neutral technical progress.

We follow the assumption that the price of capital and labor is equal to marginal output; it can obtain the rate of return on capital $r_t$ and wage rate ${\omega }_t$. Combining Equations (4) and (7) gives us the technical efficiency of capital $A_{Kt}$ and the technical efficiency of labor $A_{Lt}$.

$${\displaystyle \frac{r_t}{{\omega }_t}}={\displaystyle \frac{{MP}_K}{{MP}_L}}={\displaystyle \frac{\gamma }{1-\gamma }}{\left({\displaystyle \frac{A_{Kt}}{A_{Lt}}}\right)}^{\frac{\sigma -1}{\sigma }}{\left({\displaystyle \frac{K_t}{L_t}}\right)}^{-\frac{1}{\sigma } }$$

(7)

$$A_{Kt}={\displaystyle \frac{Y_t}{K_t}}{\left[{\displaystyle \frac{r_t{K}_t}{\gamma \left(r_t{K}_t+{\omega }_t{L}_t\right)}}\right]}^{\frac{\sigma }{\sigma -1}}$$

(8)

$$A_{Lt}={\displaystyle \frac{Y_t}{L_t}}{\left[{\displaystyle \frac{{\omega }_t{L}_t}{\left(1-\gamma \right)\left(r_t{K}_t+{\omega }_t{L}_t\right)}}\right]}^{\frac{\sigma }{\sigma -1}}$$

(9)

Finally, the level of technological progress bias is measured [20]:

$${\mathrm{D}}_{\mathrm{t}}={\displaystyle \frac{\mathsf{\gamma }}{1-\mathsf{\gamma }}}·{\displaystyle \frac{{\mathrm{A}}_{\mathrm{L}\mathrm{t}}}{{\mathrm{A}}_{\mathrm{K}\mathrm{t}}}}·{\displaystyle \frac{\mathrm{d}\left({\mathrm{A}}_{\mathrm{K}\mathrm{t}}/{\mathrm{A}}_{\mathrm{L}\mathrm{t}}\right)}{\mathrm{d}\mathrm{t}}}$$

(10)

Regarding the elasticity of substitution $\mathsf{\sigma }$ estimation, the supply-side system model of the standardized CES production function established by Klump and others is primarily used [22]; this method standardizes the first-order conditions and associates the estimated parameters for the three equations of the production function, capital and labor, which are formed by the CES production function. In addition, based on the elasticity of substitution estimation method of León-Ledesma, the capital technical efficiency ${\mathrm{A}}_{\mathrm{K}\mathrm{t}}$ and labor technical efficiency ${\mathrm{A}}_{\mathrm{L}\mathrm{t}}$ are set as “Box–Cox type” [22,23]:

$$\mathrm{l}\mathrm{o}\mathrm{g}\left({\displaystyle \frac{Y_t}{\overline{Y}}}\right)=\mathrm{l}\mathrm{o}\mathrm{g}\left(\xi \right)+{\displaystyle \frac{\sigma }{\sigma -1}}\mathrm{l}\mathrm{o}\mathrm{g}\left[\left(1-\gamma \right){\left[{\displaystyle \frac{L_t}{\overline{L}}}exp(g_L(t,t_0))\right]}^{\frac{\sigma -1}{\sigma }}+\gamma {\left[{\displaystyle \frac{K_t}{\overline{K}}}exp(g_K(t,\overline{t }))\right]}^{\frac{\sigma -1}{\sigma }}\right]$$

(11)

$$\mathrm{l}\mathrm{o}\mathrm{g}\left({\displaystyle \frac{r_t{K}_t}{Y_t}}\right)=\mathrm{l}\mathrm{o}\mathrm{g}\left(\gamma \right)+{\displaystyle \frac{\sigma -1}{\sigma }}\mathrm{l}\mathrm{o}\mathrm{g}\left(\xi \right)-{\displaystyle \frac{\sigma -1}{\sigma }}\mathrm{l}\mathrm{o}\mathrm{g}\left({\displaystyle \frac{Y_t/\overline{Y}}{K_t/\overline{K}}}\right)+{\displaystyle \frac{\sigma -1}{\sigma }}{g}_K(t,\overline{t})$$

(12)

$$\mathrm{l}\mathrm{o}\mathrm{g}\left({\displaystyle \frac{{\omega }_t{L}_t}{Y_t}}\right)=\mathrm{l}\mathrm{o}\mathrm{g}\left(1-\gamma \right)+{\displaystyle \frac{\sigma -1}{\sigma }}\mathrm{l}\mathrm{o}\mathrm{g}\left(\xi \right)-{\displaystyle \frac{\sigma -1}{\sigma }}\mathrm{l}\mathrm{o}\mathrm{g}\left({\displaystyle \frac{Y_t/\overline{Y}}{L_t/\overline{L}}}\right)+{\displaystyle \frac{\sigma -1}{\sigma }}{g}_L(t,\overline{t})$$

(13)

$$g_K(t,\overline{t})=\overline{t}{\displaystyle \frac{{\theta }_K}{{\lambda }_K}}\left[{\left({\displaystyle \frac{t}{\overline{t}}}\right)}^{{\lambda }_K}-1\right]$$

(14)

$$g_L(t,\overline{t})=\overline{t}{\displaystyle \frac{{\theta }_L}{{\lambda }_L}}\left[{\left({\displaystyle \frac{t}{\overline{t}}}\right)}^{{\lambda }_L}-1\right]$$

(15)

where $\overline{Y},\overline{K},\overline{L},\overline{t}$ denote the sample means of total output, capital input, labor input, and time; $\mathsf{\xi }$ is the scale factor;${ \theta }_K$ and ${ \theta }_{L }$ are the technical growth parameters; and ${\lambda }_K$ and ${\lambda }_L$ represent the technology curvature.

By means of total output in each year ($Y_t$), capital inputs ($K_t$), labor inputs ($L_t$), capital income ($r_t{K}_t$), and labor income (${\omega }_t{L}_t$) data can be used to estimate the system, which can obtain the factor elasticity of substitution ($\sigma$) and the share of capital income ($\gamma$).

Energy is an important driving force of social production. This paper introduces the energy factor into the measurement of technological progress bias in order to explore the role of energy in the perspective of technological progress bias path. Existing studies classify the factor-enhanced technological progress bias into three types: capital–energy–labor-enhanced, capital–labor–energy-enhanced and labor–energy–capital-enhanced. Based on the development history of China, it is believed that the capital–energy–labor-enhanced bias is more closely aligned with the factor use situation in China. Therefore, this paper chooses to study its impact on carbon productivity from the perspective of the energy-enhanced type. We refer to León-Ledesma to extend the technical progress measure to the three-stage CES production function, draw on the work of Guo to introduce factor prices in the estimation equation, and set the production function in the following form [24,25]:

$$Y_t={\left[\gamma {\left({KE}_t\right)}^{\frac{\sigma -1}{\sigma }}+\left(1-\gamma \right){\left(A_{Lt}{L}_t\right)}^{\frac{\sigma -1}{\sigma }}\right]}^{\frac{\sigma }{\sigma -1}}$$

(16)

$${KE}_t={\left[\mu {\left(A_{Kt}{K}_t\right)}^{\frac{\tau -1}{\tau }}+\left(1-\mu \right){\left(A_{Et}{E}_t\right)}^{\frac{\tau -1}{\tau }}\right]}^{\frac{\tau }{\tau -1}}$$

(17)

where $A_{Kt}$ denotes energy-enhanced technological progress; ${KE}_t$ is energy and capital factor synthetic goods;${ E}_t$ is the energy input; $\mu$ is the factor share; and $\tau$ is the capital–energy elasticity of substitution $\left(\tau >0\right)$. The specific index measurements are consistent with the previous section.

The descriptive statistics of the samples are shown in

Table 1. Descriptive statistics of variables.

Variables		Mean	SD	Min	Max
Carbon productivity	CP	0.2705	0.1724	0.0249	1.2240
Green finance	GF	−0.5170	0.0935	−0.7137	−0.3294
Capital–labor technological progress bias	Bias_LK	−0.1592	0.8720	−2.0399	13.3741
Energy-enhanced technological progress bias	Bias_E	−0.0064	0.0416	−0.2335	0.1914
Energy structure	ES	8.2589	1.5305	0.8604	9.9979
Industrial agglomeration	IA	0.9286	0.2236	0.3242	1.3573
Environmental regulation	ER	2.0156	1.9072	0.0009	9.6841
Labor skills	LS	6.9831	6.5480	0.7825	43.8000
Trade level	TL	0.1077	0.1849	0.0003	1.0916
Economic structure	ESO	1.3497	0.8762	0.5342	7.1810

. Carbon productivity is measured by the ratio of regional GDP to total CO₂ emissions. The province with the lowest productivity stands at 0.0249, while the highest reaches 1.224, indicating significant disparities in overall carbon productivity levels. Green finance is constructed using two indicators: green credit and listed environmental protection enterprises. The overall difference is relatively small, with a gap of only 0.3843, suggesting a relatively balanced development of green finance across Chinese regions. The overall gap in labor–capital–technology progress bias is substantial, while the gap in energy-enhancing technology progress bias is relatively small. Significant differences exist in the extremes of energy structure, industrial agglomeration, environmental regulations, labor force level, trade level, and economic structure.

3.2. Spatial Measurement Models

3.2.1. Spatial Auto-Correlation Test

Everything is related to everything else, but, according to the first law of geography, things that are in closer proximity are more closely associated. Because carbon emissions only take into account geographical location and ignore spatial distribution factors, the final correlation analysis results may have a significant deviation. To determine if variables are spatially associated within a region, spatial auto-correlation analysis is employed. In this study, the spatial auto-correlation of industrial carbon emissions is tested using Moran’s index to determine whether spatial correlation exists and whether the spatial econometric model should be used for analysis. The specific calculation formula is as follows:

$${Moran}^{\prime }s I={\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\sum }_{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{W}}_{\mathrm{i}\mathrm{j}}({\mathrm{l}\mathrm{n}\mathrm{C}\mathrm{P}}_{\mathrm{i}}-\overline{\mathrm{l}\mathrm{n}\mathrm{C}\mathrm{P}})({\mathrm{l}\mathrm{n}\mathrm{C}\mathrm{P}}_{\mathrm{j}}-\overline{\mathrm{l}\mathrm{n}\mathrm{C}\mathrm{P}})}{{\mathrm{S}}^2{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\sum }_{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{W}}_{\mathrm{i}\mathrm{j}}}}$$

(18)

where ${lnCP}_{i },{lnCP}_j$ are the industrial carbon productivity of the i and j provinces; $W_{ij}$ denotes the economic–geographic nested matrix (W), where the exact calculation of the matrix is shown below; n denotes the number of provinces; $\overline{lnCP}$ is the logarithmic mean of carbon productivity; and $S^2$ is the sample (log carbon productivity) variance. Next, for Moran’s index, the value range is [−1,1], and different values correspond to different correlations. When the value is in the range (0,1], there is a positive spatial correlation of the variable; the closer the value to 1, the stronger the positive correlation. When the value is in the range [−1,0), there is a negative spatial correlation of the variable; the closer the value to −1, the stronger the negative correlation. The economic geography nested matrix W considers the dual influence factors of the level of regional economic development and geographic latitude and longitude location to construct a spatial weight matrix. The specific calculation formula is as follows:

$$\mathrm{W}={\mathrm{W}}_{\mathrm{d}}\times {\mathrm{W}}_{\mathrm{e}}$$

(19)

$${{(\mathrm{W}}_{\mathrm{d}})}_{\mathrm{i}\mathrm{j}}=\left\{\begin{array}{c}{\displaystyle \frac{1}{{\mathrm{d}}_{\mathrm{i},\mathrm{j}}}}, \mathrm{i}\ne \mathrm{j}\\ \\ 0, \mathrm{i}=\mathrm{j}\end{array}\right.$$

(20)

$${{(\mathrm{W}}_{\mathrm{e}})}_{\mathrm{i}\mathrm{j}}=\left\{\begin{array}{c}{\displaystyle \frac{1}{|{\mathrm{G}}_{\mathrm{i}}-{\mathrm{G}}_{\mathrm{j}}|}}, \mathrm{i}\ne \mathrm{j}\\ \\ 0, \mathrm{i}=\mathrm{j}\end{array}\right.$$

(21)

where $d_{i,j}$ denotes the geographic distance between provinces calculated based on geographic latitude and longitude; G is the average value of GDP per capita in each province; $W_d$ is the geographic distance weight matrix, expressed as the ${\mathrm{d}}_{\mathrm{i},\mathrm{j}}$, the reciprocal representation; and $W_e$ is the economic distance weight matrix based on the level of economic development, denoted by the $|G_i-G_j|$ inverse number.

3.2.2. Spatial Durbin Model

Spatial measurement models address the spatial correlations that cannot be observed in linear regression. The three common forms of this are the spatial lag model (SLM), the spatial error model (SEM), and the spatial Durbin model (SDM). Among them, the expression of SDM is as follows:

$$Y=\alpha +\delta \mathit{W}Y+\theta \mathit{W}X +\beta \mathit{X}+\epsilon ,\epsilon ~(0,{\sigma }^2)$$

(22)

where Y denotes the dependent variable; X is the independent variable; α is the constant term; β is the regression coefficient of the explanatory variable; ρ is the spatial regression coefficient, which takes the value in the range of [−1,1] and illustrates the extent to which the explanatory variable interacts with the explanatory variables in the neighboring cells; and ε denotes the random error term, which is usually considered to be independently distributed. The introduction of spatial weight matrices in the SDM overcomes the problem of biased or invalid least squares estimation in traditional econometric models.

3.3. Mediated Effects Model

Drawing on previous research [26], the mediation model method of stepwise regression is used to test the mediation effect of technological progress bias, and the specific formula of the model is constructed as follows:

$${CP}_{it}={\beta }_0+C·{GF}_{it}+\beta ·X_{it}+{\epsilon }_1$$

(23)

$$M_{it}={\alpha }_0+\alpha ·{GF}_{it}+\beta ·X_{it}+{\epsilon }_2$$

(24)

$${CP}_{it}=c_0+C^{\prime }·{GF}_{it}+b·M_{it}+\beta ·X_{it}+{\epsilon }_3$$

(25)

where $M$ is the mediating variable and $X$ is the total effect, c = ab + c’. ab denotes the mediating effect, which is indirect effect, and c’ denotes direct effect. $X$ is a set of control variables.

The mediating effect was tested using the stepwise regression method: in the first step, the total effect of green finance on carbon productivity in Equation (23) was examined, as was whether the coefficient of measurement c was significant; in the second step, the effect of green finance on the degree of the mediating variable in Equations (24) and (25) was examined, as was the effect of the mediating variable on carbon productivity and whether the coefficients of measurement a and b were significant, confirming if there was a mediating effect; and in the third step, the direct effect of green finance on carbon productivity in Equation (25) is examined.

4. Empirical Results and Analysis

4.1. Empirical Analysis and Testing of Spatial Econometric Models

4.1.1. Empirical Analysis of Spatial Econometric Models

This research chooses the 2020 carbon productivity data for spatial visualization analysis to investigate the spatial distribution characteristics of carbon productivity in each province and city (Figure 2). The provinces with missing data are shown in white; the darker the color, the higher the degree of carbon productivity. In general, the northern region has a higher carbon productivity than the southern region, the eastern region has a higher carbon productivity than the western region, and each province and city has a notable high–high and low–low agglomeration in its carbon productivity.

Figure 2 displays the spatial distribution of China’s carbon productivity in 2020. The map reveals that carbon productivity exhibits spatial clustering patterns; darker shades indicate higher productivity levels. Northern provinces demonstrate higher carbon productivity than southern provinces, while eastern coastal regions show higher productivity than western inland areas.

Table 2. Moran index of carbon productivity.

Year	Moran’s Index	Statistical Value	t-Value
2006	0.0487 ***	2.4772	0.0066
2007	0.0498 ***	2.5101	0.0060
2008	0.0896 ***	3.6946	0.0001
2009	0.0940 ***	3.8275	0.0001
2010	0.0652 ***	2.9685	0.0015
2011	0.0654 ***	2.9748	0.0015
2012	0.0691 ***	3.0855	0.0010
2013	0.0804 ***	3.4210	0.0003
2014	0.0818 ***	3.4633	0.0003
2015	0.0903 ***	3.7150	0.0001
2016	0.0855 ***	3.5745	0.0002
2017	0.0757 ***	3.2826	0.0005
2018	0.0701 ***	3.1153	0.0009
2019	0.0719 ***	3.1670	0.0008
2020	0.0662 ***	2.9979	0.0014

Note: *** represent significant at 1% level, respectively.

presents the Moran index of carbon productivity for each province and city between 2006 and 2020. The results demonstrate a positive correlation and clear spatial aggregation of inter-regional carbon productivity. The results also pass the 1% significance test, thereby confirming the validity of the previous model analysis.

Through

Table 3. Spatial model test results.

Methods	Hausman (Fixed Time)	Hausman (Fixed Space)	LM-Lag	Robust LM-Lag	LM-Error	Robust LM-Error
statistical value	449.6372 ***	77.3860 ***	38.8794 ***	9.5382 ***	40.6183 ***	11.2771 ***
P-value	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000

Note: *** represent significant at 1% level, respectively.

, it can be seen that both LM-lag and LM-error pass the 1% significance test, and the SDM can be the first to be considered. Though the Hausman method in time-fixed and space-fixed both pass the 1% significance test, both can reject the original hypothesis that the time or space effect is not significant and so establish the time–space double fixed-effects model.

In summary, the spatial Durbin model with double fixation in time–space is as follows:

$$\begin{gathered} {lnCP}_{it}=\alpha +\delta {\sum }_{i=1}^n{W}_{ij}{lnCP}_{jt}+{\beta }_1{lnGF}_{it} +{\beta }_2{X}_{{control}_{it}}+{\theta }_1{\sum }_{j=1}^n{W}_{ij}{lnGF}_{jt}+ {\theta }_2{\sum }_{j=1}^n{W}_{ij}{X}_{{control}_{jt}}+ \\ {\sigma }_i+{\mu }_t+{\epsilon }_{it} \end{gathered}$$

(26)

where ${lnCP}_{it }$ represents the carbon productivity of province i in year t; ${lnGF}_{it }$ represents green finance in province i in year t; ${lnCP}_{jt}$ and ${lnTECH}_{jt}$ represent the carbon productivity and green finance of a neighboring province j in year t, respectively; $W_{ij}$ is the economic-geographical nested matrix (W); ${\sum }_{i=1}^n{W}_{ij}{lnCP}_{jt}$ and ${\sum }_{j=1}^n{W}_{ij}{lnTECH}_{jt}$ denote the spatial lag terms of carbon productivity and green finance, respectively; $X_{{control}_{it}}$ and ${ X}_{{control}_{jt}}$ are control variables for this province and neighboring provinces, respectively$; \delta$ represents the spatial regression coefficients; $\beta$ represents linear regression coefficients; $\theta$ represents lag term impact coefficients; ${ \sigma }_i$ represents spatial fixed effects; ${\mu }_t$ represents time fixed effects; and ${ \mu }_{it}$ is the error term.

Once the model has been established, we examine the model output using the previously established spatial Durbin model effect decomposition. The findings, as presented in

Table 4. Decomposition of the effect of green finance and carbon productivity.

Variant	Direct Effect	Indirect Effect	Aggregate Effect
GF	0.0748 (0.4689)	1.9204 ** (2.4777)	1.9952 ** (2.3732)
ES	0.0282 *** (7.1979)	−0.0014 (−0.1272)	0.0268 ** (2.4691)
IA	−0.1871 *** (−5.5159)	−0.2175 ** (−2.5081)	−0.4046 *** (−4.7629)
ER	−0.0123 *** (−3.1853)	−0.0044 (−0.3418)	−0.0168 (−1.2367)
LS	0.0997 (0.6816)	−0.4471 (−1.5532)	−0.3474 (−1.0215)
TL	0.0559 (1.6179)	0.1006 (0.8035)	0.1566 (1.1301)
ESO	−0.0196 ** (−2.0969)	−0.0237 (−1.0156)	−0.0434 ** (−1.8237)
R2	0.8273
log-L	551.0004
size	450

Note: ** and *** represent significant 5% and 1% levels, respectively.

, demonstrate that green finance has a positive impact on the direct effect of carbon productivity, suggesting that local green finance enhances local carbon productivity. The enhancing effect of green finance on carbon productivity in nearby areas is significant at the 5% level, suggesting that green finance has a significant spillover effect. Increasing investment in green financial activities spurs economic growth while simultaneously reducing pollution emissions. Furthermore, the implementation of green finance initiatives not only lowers pollution emissions but also stimulates economic growth and advances the green transformation of the economy [27]. This is because green finance improves overall carbon productivity and has a greater impact on neighboring regions than on the local area, meaning that inter-regional financial activities must be coordinated more closely, a finding that supports hypothesis H1. The carbon productivity of the region will increase significantly when the energy structure is optimized; the carbon productivity of the local and neighboring regions will decrease significantly when the degree of industrial agglomeration increases; the carbon productivity of the local region will be inhibited more strongly by environmental regulation to a greater extent than that of the neighboring regions; and the carbon productivity will increase when the labor force’s skill level increases because highly skilled laborers are more resource-efficient and can effectively reduce CO₂ emissions during the production process [28]. Because variations in the export product trade structure can have varying effects on carbon emissions, and because trade structure in the tertiary industry tends to promote a decline in the implicit carbon of the products, an increase in trade level promotes carbon productivity [29]. The local carbon productivity is significantly inhibited by the optimization of the industrial structure, whereas the neighboring areas are not inhibited as much.

4.1.2. Endogeneity Test

Considering that the model may suffer from endogeneity issues due to omitted variables, multicollinearity bias, or measurement errors, this article conducts the following endogeneity tests. Capital input, labor input, and energy input are selected as instrumental variables. These variables are highly correlated with the green finance variable but not strongly correlated with carbon productivity. The two methods employed for endogeneity testing are the DWH test and the Hausman test. Since both methods yield p-values below 0.05, it can be concluded that green finance is an endogenous variable.

Using the instrumental variables method for 2SLS regression, the regression results indicate that the coefficient of green finance on carbon productivity is 0.7317, passing the 1% significance level test and remaining positive. An over-identification test for the instrumental variables method shows an exogeneity p-value greater than 0.05, confirming that the selected instrumental variables are uncorrelated with the disturbance term. Thus, the instrumental variables are appropriately chosen and can eliminate endogeneity.

4.1.3. Robustness Tests

In this study, we test resilience via the replacement space weight matrix approach, using the economic distance matrix (${\mathrm{W}}_{\mathrm{e}}$) and geographic distance weight matrix (${\mathrm{W}}_{\mathrm{d}}$) for the replacement spatial weight matrix’s robustness test. The spatial Durbin model with bilateral fixed effects is used in both tests, and the outcomes demonstrate that each variable’s sign and direction, as well as its significance, are largely consistent with the previously mentioned empirical findings, suggesting that the study’s findings are somewhat robust.

Table 5. Robustness test of the substitution matrix.

Variant	Direct Effect	Indirect Effect	Aggregate Effect	Direct Effect	Indirect Effect	Aggregate Effect
	${\mathbf{W}}_{\mathbf{e}}$			${\mathbf{W}}_{\mathbf{d}}$
GF	0.0158 (0.1023)	92.6472 (0.6705)	92.6630 (0.6706)	0.2732 (0.8817)	6.3678 (1.0076)	6.6411 (1.0061)
ES	0.0282 ** (8.0777)	1.5735 (0.9267)	1.6022 (0.9432)	0.0285 *** (7.3776)	−0.0844 (−0.9821)	−0.0558 (−0.6415)
IA	−0.2288 *** (−7.8504)	−17.1823 (−1.2285)	−17.4112 (−1.2444)	−0.2177 *** (−6.8220)	0.2327 (0.3180)	0.0150 (0.0205)
ER	−0.0135 *** (−3.9072)	−0.2726 (−0.1530)	−0.2861 (−0.1605)	−0.0177 *** (−3.8318)	−0.1878 * (−1.8004)	−0.2056 * (−1.9272)
LS	0.0184 (0.1293)	−1.3248 (−0.0321)	−1.3064 (−0.0316)	0.1070 (0.5696)	−15.1970 *** (−3.3738)	−15.0899 *** (−3.2774)
TL	0.0504 (1.4812)	−10.8253 (−0.6563)	−10.7749 (−0.6529)	0.0946 ** (2.2403)	1.2525 (1.4466)	1.3472 (1.5140)
ESO	−0.0248 *** (−2.8027)	−4.5241 (−0.9617)	−4.5489 (−0.9666)	−0.0240 ** (−2.5656)	0.0269 (0.1292)	0.0028 (0.0138)
R2	0.2906
log-L	542.0702
size	450

Note: *, **, and *** represent significant at 10%, 5%, and 1% levels, respectively.

displays the test results.

4.2. Analysis of Inter-Mediation Effects

Regression tests are carried out using the mediation effect model to investigate the relationship between technical advancement bias and green finance and carbon productivity in more detail. The results are displayed in

Table 6. Results of the mechanism path test of green finance on carbon productivity.

Variant	Bias_LK	Bias_E	CP
GF	−0.0302 * (−1.67)	−0.1156 ** (−1.91)	0.2573 *** (11.43)
Bias_LK			0.1581 *** (2.70)
Bias_E			0.0416 ** (2.38)
con_s	−0.5880 *** (−81.05)	−0.5882 *** (−81.16)
R2	0.2305	0.2320
control variable	Yes
size	450	450	450

Note: *, **, and *** represent significant at 10%, 5%, and 1% levels, respectively.

. The allocation of the labor and capital factors of production is tightly linked to the support of green financial activities, which can optimize the factor structure by reducing energy factor inputs in energy-intensive industries. Green finance has a significant impact on both labor–capital technological progress bias and energy-enhanced technological progress bias. Furthermore, because modifications to the factor input structure can lessen the upgrading of the factor structure of high-emission industries, lower carbon emissions, and raise carbon productivity, the labor–capital technology progress bias and energy-enhanced technology progress bias both play a major role in promoting carbon productivity. In other words, by affecting the allocation of capital and energy factors, green finance can actually boost carbon productivity. This conclusion supports hypotheses H2 and H3.

4.2.1. Analysis of Regional Heterogeneity

To investigate the function of technical progress bias as a mediator in the relationship between carbon productivity and green finance, we split the sample data into four main regions based on China’s economic development level. The results in

Table 7. Mediating effects of regional heterogeneity.

Region	East			Central			West			Northeast
	Bias_LK	Bias_E	CP	Bias_LK	Bias_E	CP	Bias_LK	Bias_E	CP	Bias_LK	Bias_E	CP
GF	−0.1219 *** (−2.74)	−0.2726 *** (−3.85)	0.3577 *** (6.41)	−0.0814 *** (−0.90)	−0.4434 ** (−2.4)	0.3621 *** (7.91)	−0.0193 (−1.22)	−0.1210 * (−1.75)	0.2784 *** (8.67)	0.2041 (1.13)	−0.3458 * (−15.66)	0.5378 *** (5.26)
Bias_LK			0.9269 *** (9.19)			0.0408 (0.76)			0.1980 (1.25)			0.1643 * (1.91)
Bias_E			0.5076 *** (8.20)			−0.0357 (−1.39)			0.0834 ** (2.32)			0.7399 *** (8.84)
con_s	−0.5942 *** (−44.26)	−0.5921 *** (−48.71)	−0.5761 *** (−48.22)	−0.6129 *** (−40.42)	−0.6128 *** (−41.59)	−0.6121 *** (−40.48)	−0.6220 *** (−48.88)	−0.6235 *** (−49.01)	−0.6207 *** (−48.89)	−0.6524 *** (−20.44)	−0.6148 *** (−15.66)	−0.6660 *** (−22.48)
R2	0.2555	0.2890		0.4212	0.4519		0.3221	0.3285		0.4094	0.4403
control variable	Yes
size	150	150	150	90	90	90	165	165	165	45	45	45

Note: *, **, and *** represent significant at 10%, 5%, and 1% levels, respectively.

indicate that technological progress biases related to labor–capital and energy-enhanced labor have a significant mediating effect in the east. This means that in the more-developed eastern region, achieving high carbon productivity requires co-regulation of capital, labor, and energy factors, and reducing the inputs of capital, labor, and energy factors in production activities can lower carbon emissions and raise carbon productivity. The labor–capital technological progress bias and the energy-enhanced technological progress bias do not have a significant mediating effect in the central region; however, the results show that green finance significantly negatively impacts both labor–capital and energy-enhanced technological progress bias. This suggests that the growth of green finance can conserve factor resources and reduce the input of factors of production in the central region. As in the west, the energy-enhanced technological progress bias plays a significant mediating role in the northeast, where green financial development raises carbon productivity throughout the region by lowering the use of energy factors and, consequently, carbon emissions. This suggests that improving energy technology and reducing the use of energy factors is one way to increase carbon production efficiency in the region.

4.2.2. Analysis of the Mediating Effect of Agglomeration Level Heterogeneity

We categorize the carbon productivity distribution in each province and city in Section 4.1.1 into three types of aggregation: high–high agglomeration, middle–middle agglomeration, and low–low agglomeration.

Table 8. Types of aggregation.

		Provincial Classification
High-high agglomeration area	(1)	Inner Mongolia, Ningxia, Shanxi, Liaoning, Hebei
	(2)	Xinjiang, Qinghai
Middle-middle agglomeration area	(3)	Gansu, Shanxi, Hubei, Anhui, Jiangxi, Hunan, Guizhou, Yunnan, Guangxi, Shandong
	(4)	Heilongjiang, Jilin
Low-low agglomeration area	(5)	Sichuan, Chongqing
	(6)	Henan, Jiangsu, Zhejiang, Shanghai, Fujian, Guangdong, Hainan

Note: Beijing and Tianjin do not play a significant role in aggregation and are not considered here.

displays the agglomeration degree classification for each region. This allows us to more thoroughly investigate the mechanism by which green finance can alter carbon productivity.

Table 9. Mediating effects by aggregation type.

			High-High Agglomeration Area
			(1)								(2)
			Bias_LK		Bias_E			CP			Bias_LK			Bias_E			CP
GF			−0.1933 * (−1.94)		0.4012 *** (2.71)			0.3104 *** (5.86)			−0.0004 (−0.02)			0.3949 ** (2.6)			0.8003 *** (10.66)
Bias_LK								−0.0632 (−1.04)									−0.2350 (−0.47)
Bias_E								0.0902 ** (2.25)									0.3589 *** (4.22)
con_s			−0.6644 *** (−26.29)		−0.5921 *** (−48.71)			−0.6659 *** (−25.87)			−0.8277 *** (−28.44)			−0.1566 *** (−4.84)			−0.8277 *** (−28.97)
R2			0.3538		0.3830						0.8023			0.8418
control variable			Yes
size			75		75			75			30			30			30
	Middle-middle agglomeration area									Low-low agglomeration area
	(3)					(4)				(5)					(6)
	Bias_LK	Bias_E		CP		Bias_LK	Bias_E		CP	Bias_LK		Bias_E	CP		Bias_LK	Bias_E		CP
GF	−0.0341 ** (−2.23)	−0.3560 *** (−3.82)		0.3639 *** (10.76)		0.4166 *** (1.93)	0.2934 ** (2.46)		0.5074 *** (8.19)	0.1514 (1.50)		−0.0917 ** (−2.18)	0.8918 *** (12.81)		−0.0234 (−0.47)	0.0421 (−15.66)		0.9337 *** (11.25)
Bias_LK				0.4118 ** (2.30)					0.1414 ** (2.73)				−0.0361 (−0.28)					−0.0564 (−0.34)
Bias_E				0.1298 *** (4.55)					0.2424 ** (2.69)				−0.1646 (−0.56)					−0.0306 (−0.39)
con_s	−0.6192 *** (−55.08)	−0.6301 *** (−55.16)		−0.6160 *** (−54.52)		−0.7293 *** (−20.43)	−0.7205 *** (−20.89)		−0.7670 *** (−24.50)	−0.7034 *** (−44.45)		−0.7025 *** (−46.32)	−0.7065 *** (−44.07)		−0.6490 *** (−47.53)	−0.6486 *** (−47.56)		−0.0146 (−0.54)
R2	0.4575	0.4898				0.7413	0.7593			0.8655		0.8761			0.5525	0.5522
control variable	Yes
size	150	150		150		30	30		30	30		30	30		105	105		105

Note: *, **, and *** represent significant at 10%, 5%, and 1% levels, respectively.

displays the results of the regression analysis. The results indicate a significant mediating effect for energy-enhanced technological progress bias when testing the mediating effect of green finance on carbon productivity in the high-high agglomeration areas (1) and (2). This suggests that maintaining high carbon productivity in this region can be achieved by controlling the inputs of energy factors. The regression coefficients (−0.0341, −0.3560, 0.3639) and (0.4166, 0.2934, 0.5074) indicate the mediating effect of green finance on carbon productivity in the middle-middle agglomeration area, where both the labor–capital technological progress bias and the energy-enhanced technological progress bias have a significant mediating effect; these values suggest that increasing factor inputs in the right way can lead to a higher level of carbon productivity rather than lowering it. The results of testing the mediating effect of green finance on carbon productivity in the low-low agglomeration area, where neither bias has a mediating effect, are shown in (5) and (6). These results indicate that increasing carbon productivity in low-carbon productivity regions does not depend solely on factor inputs.

5. Conclusions

This study utilizes data from 30 provinces over the period 2006–2020 to construct a model examining how green finance influences China’s carbon productivity, with a focus on the transmission mechanism mediated by factor-biased technology progress. Based on the preceding analysis, the following conclusions can be drawn: (1) Carbon productivity in the sample data passed Moran’s I test, exhibiting significant spatial clustering. Eastern regions demonstrate higher carbon productivity than western regions, while northern regions show higher productivity than southern regions. Significant high–high and low–low clustering patterns are observed across provinces and municipalities. (2) Under the economic geography weighting matrix, the spatial Durbin model decomposition reveals that green finance significantly boosts carbon productivity in neighboring regions. After testing for endogeneity issues, the results remain significantly positive using 2SLS regression. (3) The mediating effect of capital- and energy-factor-biased technological progress significantly influences the relationship between green finance and carbon productivity. Green finance enhances carbon productivity by affecting regional capital and energy factor allocation. Energy factor inputs exert a greater impact on carbon emissions than capital and labor factors, making energy factor technological progress a more efficient pathway to reduce emissions and boost carbon productivity. (4) Regional heterogeneity analysis reveals that energy factor technological progress bias exerts a stronger influence on carbon productivity in Central and Northeast China. This stems from these regions being China’s natural resource hubs, where social production relies heavily on energy. Consequently, energy factor innovation delivers substantially greater environmental efficiency gains in these areas compared to Eastern and Western China.

According to our analysis of the article, on the one hand, green finance can encourage the development of low-carbon technologies, direct the distribution of production factors and increase carbon productivity; on the other hand, the effects of green finance and technological advancement are not limited to local communities. Through mechanisms like technology spillovers and competitive impacts, it produces notable spatial spillover effects that propel a coordinated increase in carbon productivity over neighboring regions. The innovative contributions of this article are as follows: Unlike other studies on green finance, this article examines the relationship between green finance and carbon productivity by treating labor-capital technological progress preferences and energy-enhancing technological progress preferences as mediating variables. While prior studies predominantly focus on macro perspectives such as environmental regulation policies and digitalization, this research examines how micro-level factor allocation influences provincial green finance investment, thereby enhancing carbon productivity. Current literature remains relatively scarce in analyzing green finance investment from the perspective of factor inputs.

To better leverage green finance in advancing China’s low-carbon economic transition and accelerating the achievement of dual carbon goals, this article proposes three recommendations based on the findings above: (1) Improve the green finance development system to fully exploit its financing advantages. Empirical research indicates that green finance significantly enhances China’s carbon productivity. All regions should refine their green finance frameworks to provide financial services supporting corporate energy conservation and emission reduction initiatives. As an innovative model for promoting sustainable economic development, green finance has gradually become a vital driving force for China’s green development. Concurrently, the market must refine incentive and constraint policies to provide support and safeguards for standardizing green finance operations and advancing green industry growth. Additionally, green enterprises should be helped to secure investments from financial institutions to obtain adequate capital support and (2) focus on corporate green technology R&D and emphasize green finance’s role in driving technological advancement. Analysis of intermediary effects reveals that green finance enhances carbon productivity through capital-driven technological progress and energy-related technological progress. Governments should increase corporate investment in green technologies and improve factor utilization efficiency. The intermediary role of factor innovation to rationally allocate and utilize factor resources should be fully leveraged. Governments should mobilize and incentivize greater social capital investment in green industries, guiding more social capital away from high-pollution, high-emission sectors toward green industries to effectively reduce financing costs during green industrial development. Concurrently, financial support for clean technology R&D should be strengthened, encouraging financial institutions to engage in green investments and supporting enterprises in green and clean technology innovation activities. Additionally, the risk-dispersing function of green finance should be leveraged to mitigate corporate financing risks, and more social capital should be diverted from high-pollution, high-emission industries toward green industries, effectively lowering financing costs during green industrial development.