The Impact of Green Finance on Urban Carbon Emission Efficiency: Threshold Effects Based on the Stages of the Digital Economy in China

Abstract

As one of the effective ways to achieve “carbon neutrality”, examining the impact of green finance (GF) on carbon emission efficiency (CE) is of great significance for promoting low-carbon development in China. Moreover, the digital economy is a key catalyst in achieving China’s “dual-carbon” targets, as its “greening” characteristic is considered instrumental in promoting urban low-carbon development. However, the effects of the digital economy (Dig) stage on GF on urban CE have not been sufficiently studied. Using panel data from 276 Chinese cities from 2011 to 2021 and constructing a theoretical model based on the Cobb–Douglas production function, this paper analyzes the impact of GF on urban CE. The empirical results indicate that (1) GF can improve CE, and the two have a positive U-shaped relationship, which is still valid after robustness tests. (2) The heterogeneity results indicate that the impact of GF on CE is more significant in non-resource-based cities, low-carbon pilot cities, and cities with higher financial development levels. (3) GF significantly improves urban CE by driving green technology innovation (Gti) and energy efficiency improvement (Eei). (4) The effects of GF on CE have a dual-threshold effect based on the Dig. When the Dig level is excessively high, the positive effect of GF on urban CE will be weakened.

1. Introduction

Since President Xi Jinping proposed the goals of “peak carbon dioxide emissions” and “carbon neutrality” in 2020, several financial policies have been implemented by the Chinese government to encourage low-carbon growth, of which green finance (GF) has been regarded as one of the instruments for realizing the strategic goals of “peak carbon dioxide emissions” and “carbon neutrality” [1]. With a focus on sustainable development, GF utilizes financial instruments and policies such as green credit and green bonds to guide the flow of capital from high-polluting industries to low-carbon industries and projects, and it is a powerful impetus for improving energy efficiency and reducing pollutant emissions [2]. In this transition, cities play a crucial role that cannot be overlooked. Compared with rural areas, cities have intensive and constant economic activity, so the development process of cities largely determines the total amount of carbon emissions. As the world’s largest carbon emitter, China’s urban carbon emissions currently account for 70% of the country’s total carbon emissions. Therefore, the urban carbon emission rate is an essential indicator of a city’s low-carbon economic development and is closely related to technological innovation, industrial upgrading, and resource allocation [3]. Thus, enhancing the urban CE is undoubtedly a crucial strategy for fostering green and low-carbon development in cities, ultimately reducing urban carbon emissions. At the same time, many Chinese scholars believe that there is an urgent need to create a road with Chinese characteristics in GF, inject new financial impetus for green projects, play the role of GF, and promote carbon reduction [4,5]. To ensure that GF effectively promotes urban carbon emission reduction and thus promotes green and low-carbon development, it is essential to investigate not only the effect of urban carbon emission reduction under the incentive of GF and clarify the micro-mechanisms of the role of GF but also to scientifically measure the effectiveness and efficiency of the implementation of relevant specific policies. Exploring the intrinsic mechanisms linking GF and carbon emission reduction holds significant theoretical and practical implications. This understanding could inform adjustments to the industrial energy structure in cities, enhance the alignment of regional green financial policies at a more granular level, and promote cities’ active response to the call for the “dual-carbon” strategic goals.

Although GF is regarded as a key driver for improving CE, the mechanisms and evaluations of its impact remain underexplored. Existing studies predominantly emphasize the roles of green technology innovation (Gti) and industrial upgrading in enhancing CE while neglecting the critical role of resource allocation. This study fills the research gap by examining how GF optimizes resource allocation to improve energy efficiency, fostering sustained improvements in CE and enriching the understanding of GF’s micro-mechanisms. Additionally, while the “green character” of the digital economy (Dig) is regarded as vital for urban low-carbon development, it is seldom incorporated into analytical frameworks. By integrating the Dig as a threshold variable, this study reveals its dual-threshold effect on the relationship between GF and CE, advancing the literature on the interplay between digital transformation and GF. Moreover, existing research has limited focus on how GF can be effectively leveraged to achieve China’s urban “dual-carbon” strategic goals. This study provides actionable insights into urban industrial energy structure adjustments, regional policy alignment, and proactive urban responses to carbon neutrality objectives, offering theoretical and practical significance.

The primary objective of this study is to explore the mechanisms through which GF influences CE, focusing on three key goals: first, examining the role of GF in optimizing resource allocation to enhance energy efficiency and steadily improve CE; second, evaluating the threshold effects of the Dig on the relationship between GF and CE, highlighting its role in promoting low-carbon urban development; and third, scientifically assessing the effectiveness and efficiency of GF policies to provide evidence-based insights for policy design and implementation. To tackle these issues, this study utilizes panel data from 276 Chinese cities spanning 11 years and integrates the Dig as a threshold variable within the research framework. Doing so enriches the analysis of GF’s mechanisms in promoting CE. The findings provide practical insights into how GF influences CE, the intermediary role of the Dig, and the implications for China’s green transformation and low-carbon development strategies.

2. Literature Review and Theoretical Framework

As a financial branch with economic and environmental benefits, GF has become a strong starting point for supporting China in reducing carbon dioxide emissions [6]. Early scholars have studied the concept [7,8], connotation [9], and content of GF [10]. GF is a new financial model that comprehensively considers environmental and economic factors, aiming to balance economic progress and environmentally sustainable development.

With the deepening of research content, more and more attention has been paid to the economic effects of GF policies and instruments [11]. At the level of financial policies, scholars mainly examine the implementation effects of GF policies on regional carbon emission reduction, technological progress, and enterprise green transformation [12] from the pilot zones for GF reform and innovations [13,14] and the green credit policy [15]. At the level of financial instruments, scholars often analyze the impacts of green credit, green bonds [16], and green investments [17] on urban pollution reduction and carbon emission mitigation [18], as well as their effects on green innovation [19], environmental performance [20], investment efficiency [21] of heavily polluting enterprises, and the operational performance of commercial banks [22,23]. To sum up, numerous studies have examined the impact of GF on carbon emission reduction, whether at the level of financial policies or financial instruments. Moreover, most scholars believe that GF plays a crucial role in promoting carbon emission reduction [24] and improving CE [25], and the effect in developed regions is more significant [26], with a noticeable carbon emission reduction effect [27]. This achievement stems from the effect of GF on promoting green innovation [28], upgrading industrial structure [29], and alleviating financing constraints [30,31]. On the one hand, GF can effectively promote enterprises’ Gti, stimulate technology upgrading by using financial subsidies and tax exemptions, and release potential CE [32]. On the other hand, it effectively contributes to the upgrading of industrial structure, gives preferential financing policies to environmental protection enterprises, restricts the financing of heavily polluting enterprises to a certain extent, and ultimately significantly reduces carbon emissions [33]. Moreover, green financial policies [34] and green financial instruments, including green credit and green bonds, can substantially lower financial costs for enterprises, alleviate their financial constraints, and improve their investment level in environmental protection technology [35]. In addition, Zhao et al. (2023) believe that the combined effects of GF and government intervention can significantly promote carbon emission reduction [36]. However, some studies also suggest that green credit, as the core of GF, has a significant inhibitory effect [37] and a spatial spillover effect on per-capita carbon emissions. At the same time, fiscal decentralization and environmental regulation weaken the carbon emission reduction effect of green [38,39]. Zhang et al. (2024) pointed out that, in order to achieve a win–win situation between “development” and “emission reduction”, it is essential not only to strengthen the supportive role of GF but also to leverage the guiding and driving effects of GF agglomeration [40].

However, current studies have limitations in evaluating the effect of GF on carbon emission reduction. Firstly, a substantial number of studies in the existing literature indicate that GF contributes to reducing carbon emission intensity, yet studies are still needed to examine its impact on CE. Secondly, most research has been conducted on a provincial level or the Yangtze River Economic Belt, with limited studies at the city level. Guo et al. (2022) used China’s Yangtze River Economic Belt data. They argued that GF has a significantly negative impact on carbon emissions but has little influence on spatial spillover effects in neighboring provinces [41]. Huang et al. (2024) employed the generalized method of moments (GMM) and the Green Solow model to confirm the significant impact of provincial GF on reducing carbon emissions. They found that the effect of green financial efficiency on carbon emission reductions was more pronounced than GF scale’s [42]. However, provincial data often mask intra-provincial differences. Different cities may vary significantly in terms of economic development, industrial structure, and environmental policies. Many GF policies and low-carbon strategies, such as low-carbon pilot cities and GF reform pilot zones, are implemented at the city level. Gao et al. (2024) evaluated the impact of low-carbon city pilot policies on CE and spatial spillover effects using panel data from Chinese cities. Their findings revealed that these effects were more pronounced in non-resource-based towns and cities with advanced GF systems and collaborative industrial agglomeration [43]. Similarly, Zhang et al. (2023) analyzed the effect of GF reform and innovation pilot zones on urban energy conservation using panel data from pilot provinces and 45 prefecture-level cities in China. Their results indicated that while the pilot zones significantly reduced industrial energy consumption, they had no measurable impact on residential energy consumption. This reduction in industrial energy consumption was primarily achieved by promoting Gti [44]. Therefore, using city-level data allows for a more direct analysis of the actual impact of these policies, while provincial data may underestimate the specific effects of policies at the municipal level. Moreover, what is the internal mechanism of GF in promoting CE? Most of the literature emphasizes the role of GF in driving technological innovation and upgrading industrial structures. Chen and Chen (2021) empirically tested the spatial spillover effect of GF on carbon emissions by using panel data from Chinese provinces, concluding that Gti and the reduction in financing constraints indirectly reduce carbon emissions [45]. Wang et al. (2023) argued that the optimization of energy consumption structure and the upgrading of industrial structure significantly mitigate the impact of GF on carbon emissions. They also discussed the heterogeneity of green credit types and regional variations [46]. Zhao et al. (2024) examined the carbon emission reduction effects of GF from both macro and micro perspectives. They concluded that GF reduces carbon emissions at the macro level through industrial structure upgrades and technological progress [47]. However, few studies have explored whether GF can enhance CE by optimizing resource allocation. Optimizing resource allocation can drive capital to green and low-carbon areas and reallocate resources from high-polluting enterprises to more efficient and cleaner enterprises through emission trading and carbon taxes. It can also support Gti by improving energy efficiency to enhance the overall low-carbon operation of the economy. Resource allocation is key in GF, whether guiding capital flows, supporting technological innovation, optimizing industrial structures, or improving environmental performance. This resource optimization not only directly enhances CE but also promotes the low-carbon transformation of the economy and society, providing solid support for achieving the goals of “carbon peak” and “carbon neutrality”.

Based on the practical and theoretical background outlined above, this paper explores the impact of GF on CE at the city level in China. It proposes a theoretical mechanism, suggesting that GF improves CE by supporting Gti and optimizing resource allocation. It further analyzes the nonlinear effect of the Dig in enhancing CE through GF. Finally, the theoretical inferences are validated using an econometric model. The research framework is presented in Figure 1. The marginal contributions of this paper are as follows. First, we used panel data from 276 Chinese cities from 2011 to 2021. This paper analyzes the relationship, mechanism of action, and heterogeneous impacts of GF on urban CE in China through theoretical and empirical analyses, providing robust evidence for scientifically evaluating the emission reduction effects of GF. Second, based on the Cobb–Douglas production function, this paper constructs a theoretical model to explore the relationship between GF and carbon emissions, examining the profit incentive effect and cost inhibition effect of GF on carbon emissions under the assumption of perfect market competition. The results enrich the theoretical foundations of GF and carbon emissions, providing novel theoretical support for green and low-carbon development. Third, through theoretical analysis, this paper identifies the resource allocation effect of GF in promoting energy efficiency and the green innovation effect in advancing technological progress, supplementing the intermediary mechanism of GF’s impact on CE with empirical data analysis. Fourth, this paper incorporates the nonlinear effect of the Dig into the analytical framework of GF’s impact on CE, thereby enriching existing research findings.

3. Theoretical Hypotheses

3.1. Analysis of the Impact of GF on CE

GF is an innovative financial instrument primarily promoted by financial institutions for encouraging industrial operations toward greening and low-carbon development [48]. It aims to provide essential financial support for environmental protection efforts, support environmentally focused enterprises in implementing green projects, and foster sustainable development.

Through differentiated financial resource allocation strategies, GF guides capital flow to foster the growth of environmentally friendly enterprises while limiting financial support to highly polluting and energy-intensive enterprises, thereby enhancing CE. Firstly, GF provides financial support to environmentally focused enterprises to implement environmental projects through credit support, subsidy guarantees, and preferential interest rates [49]. It facilitates the upgrading of production equipment and technology, which in turn contributes to lower carbon emissions. Secondly, GF can enhance the return on investment of corporate environmental protection projects. Enterprises can leverage green financial instruments and develop innovative products to bridge capital gaps in green projects and enhance capital alignment. This includes issuing environmental protection bonds, establishing green funds, and attracting greater capital inflows, thereby encouraging green enterprises to expand production and achieve economies of scale [50]. Moreover, through punitive and investment-inhibiting effects, GF imposes strict guarantee requirements and interest rates on heavily polluting industries, raising the financing threshold for these industries and reducing their investment returns [51], promoting capital flow to resource saving and ecological protection industries [52,53]. This approach not only alleviates financing difficulties for environmental protection enterprises but also restricts the expansion of high-polluting enterprises, forcing them to adopt environmentally friendly practices, encouraging Gti [54], and ultimately optimizing energy structures and enhancing CE.

However, a review of the existing literature reveals that GF may have a “green blind spot” in improving CE. In the initial stage of GF development, establishing green infrastructure and producing raw materials for downstream enterprises consume substantial resources and may initially increase carbon emissions. However, as GF matures, it encourages the use of cleaner energy and promotes a more efficient allocation of production resources, effectively reducing pollution and enhancing CE.

Hypothesis 1.

GF can significantly improve CE, and the two have a positive U-shaped relationship.

3.2. The Mechanism of GF on CE

(1) Green innovation mechanism. The academic community generally defines Gti as technological innovation that enhances environmental performance, and green technology specifically refers to technologies that reduce pollution or have a positive environmental impact [55]. Previous studies indicate that achieving a “win–win” between economic growth and carbon reduction is challenging. However, a balance between economic growth and carbon emissions may be achievable through Gti [56]. Gti can significantly improve the total-factor carbon productivity of economies with income levels above the international threshold [57], and various forms of green innovation can markedly enhance the carbon emission performance of cities [58]. Although enterprises are committed to developing green technologies, the protracted nature and inherent uncertainty of the process often lead to financing barriers, and traditional financial institutions often lack the ability to provide targeted financial concessions [59]. GF can leverage green financial products to fund research and development in relevant technologies, helping companies overcome financing constraints [60]. It also enables enterprises to employ various credit tools to mitigate overall innovation risks, enhance risk management capabilities [61], and foster a conducive innovation environment. In conclusion, GF promotes Gti by providing funding and establishing a risk-sharing mechanism [62], thereby enhancing CE through technological advancement.

Hypothesis 2.

GF promotes improvements in CE by promoting Gti.

(2) Resource allocation mechanism. To achieve the goals of “carbon peaking” and “carbon neutrality” as well as national high-quality development, energy transition serves as a crucial strategy that requires sustained financial support to enhance energy efficiency (Eei) [63]. Financial institutions leveraging differentiated credit policies can support clean energy, green buildings, and environmental protection enterprises while imposing stricter credit thresholds and limits for high energy-consuming enterprises, thereby facilitating the transformation of the energy structure of “three high” enterprises. This approach promotes traditional enterprises with high carbon emissions to adopt cleaner and more efficient energy [64], thus reducing pollution and carbon emissions.

Driven by the GF pilot policy, the pilot areas have improved the transparency of project information, reinforced the industry’s commitment to ecological protection, and promoted the upgrading of the energy structure within the “three high” industries to enhance CE [65]. In addition, GF facilitates the transition of energy structure from high carbon to low carbon by fostering energy technology innovation, leveraging the incentive mechanisms of the carbon trading market, promoting low-carbon infrastructure construction, and ultimately enhancing CE. In these ways, GF not only accelerates the adoption of renewable energy but also encourages the green transformation of traditional high-carbon industries, providing robust financial support to achieve carbon neutrality.

Hypothesis 3.

GF promotes improvements in CE by promoting Eei.

3.3. The Nonlinear Impact of GF on Urban CE

According to the Environmental Kuznets Curve (EKC) theory, different stages of economic development have varying impacts on environmental quality. While GF affects CE through green innovation and energy efficiency mechanisms, the external economic environment will also influence its effectiveness. As a key engine of China’s economic development, the rapid advancement and application of the Dig have profoundly transformed the operational modes and industrial patterns across various sectors [66]. Furthermore, the Dig has played a crucial role in driving employment, attracting investment, and stimulating consumption [67]. Existing studies indicate that the Dig positively contributes to enhancing CE. On the one hand, digital technologies directly enhance urban CE by improving resource utilization and facilitating the application of clean energy technologies. On the other hand, the development of the Dig has significantly reduced carbon emission intensity [68] and promoted the synergistic effects of pollution reduction and carbon mitigation [69]. However, with the continuous advancement in the Dig, the positive impact of GF on CE may gradually weaken. Firstly, the emission reduction effects of the Dig may gradually undermine the advantages of GF in enhancing CE. Compared to less developed cities in central and western China, cities with advanced Dig exhibit higher levels of industrial maturity, technological advancement, and talent concentration [70]. The expansion of the Dig has significantly increased the energy consumption of data centers and telecommunications infrastructure. While these facilities can be optimized for energy efficiency, overall energy demand and carbon emissions are likely to remain elevated. Particularly in regions reliant on traditional fossil fuels for power generation, the growth of the Dig may further elevate power demand. Consequently, a high level of the Dig could lead to excessive resource investment, thereby offsetting some of the emission reduction effects facilitated by GF. Therefore, when formulating policies and implementing measures, the developmental characteristics of the Dig should be thoroughly considered to ensure the sustainable and effective promotion of GF on urban CE.

Hypothesis 4.

GF has a positive promoting effect on urban CE, but this effect will be weakened when the level of the Dig is too high.

4. Research Design

4.1. Theoretical Model

Referring to Ma et al. (2020) [71] and Bai et al. (2025) [72], this paper constructs a theoretical model to analyze the relationship between GF and carbon emissions.

$$\mathrm{O}=\mathrm{A}\left(\mathrm{M}\right){\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }$$

(1)

where $\mathrm{O}$, $\mathrm{A}$, $\mathrm{M}$, $\mathrm{E}$, $\mathrm{W}$, and $\mathrm{K}$, respectively, represent economic output, comprehensive technology level, economic growth driven by GF, energy, labor, and capital, with $\mathrm{A}>0$, $0<\alpha <1$, and $0<\beta <1$.

$$\mathrm{M}=\mathrm{M}\left(\mathrm{G}\right)={\sigma }_{\mathrm{M}}\mathrm{G}$$

(2)

where ${\sigma }_{\mathrm{M}}>0$, $\mathrm{G}$ represents GF, and ${\sigma }_{\mathrm{M}}$ is the extra impact of GF. This paper derives the carbon emission function based on the pollution function proposed by Li et al. (2022) [73], as shown in Equation (3).

$$\omega ={\mathrm{P}}_0$$

(3)

$$\mathrm{P}={\mathrm{P}}_0+{\sigma }_{\mathrm{P}}\mathrm{M}$$

(4)

$${\mathrm{P}}_0=\sum {\mathrm{e}}_{\mathrm{i}}{\mathrm{c}}_{\mathrm{i}}$$

(5)

where ω represents the total carbon emissions, $\mathrm{P}$ stands for the marginal carbon emissions of the output affected by GF, ${\mathrm{P}}_0$ refers to the initial extra level of carbon emissions, ${\sigma }_{\mathrm{P}}$ captures the change in extra carbon emissions brought about by greenization, ${\mathrm{e}}_{\mathrm{i}}$ denotes the consumed energy type, and ${\mathrm{c}}_{\mathrm{i}}$ represents the carbon emission factor for energy type $\mathrm{i}$, ${\sigma }_{\mathrm{P}}>0$, ${\mathrm{c}}_{\mathrm{i}}>0$.

Greenization reduces carbon emissions and enhances CE by optimizing resource allocation. If GF primarily affects green technology progress, energy use, and carbon emissions, $\mathrm{A}\left(\mathrm{M}\right)$, as shown in Equation (6):

$$\mathrm{A}\left(\mathrm{M}\right)={\mathrm{A}}_0+{\sigma }_{\mathrm{A}}\mathrm{M}$$

(6)

In Equation (6), ${\mathrm{A}}_0$ represents the initial technology level, and ${\sigma }_{\mathrm{A}}$ indicates the progress in green technology driven by GF, where ${\sigma }_{\mathrm{A}}>0$. Then, Equation (7) is as follows:

$$\mathrm{O}=({\mathrm{A}}_0+{\sigma }_{\mathrm{A}}\mathrm{M}){\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }$$

(7)

This paper assumes a perfectly competitive market in which the manufacturer is a price taker. The product price is $\mathrm{r}$, and the unit variable cost is represented by $\mathrm{v}$, while the fixed cost is denoted by ${\mathrm{C}}_0$, and the cost of reducing carbon emission consumption is ${\mathrm{C}}_{\omega 0}+{\sigma }_{\omega }\omega$, where ${\mathrm{C}}_{\omega 0}$ represents the initial cost of carbon emission reduction, and ${\sigma }_{\omega }$ stands for the change in the marginal cost of carbon emission reduction resulting from advancements in green technology. Therefore, the profit function is expressed in Equation (8):

$$\mathsf{\pi }=(\mathrm{r}-\mathrm{v})\mathrm{O}-{\mathrm{C}}_0-{\mathrm{C}}_{\omega 0}-{\sigma }_{\omega }\omega$$

(8)

The representative firms will optimize their level of greenization, selecting the optimal value of $\mathrm{M}$ to maximize profits. Under the constraint of profit maximization objective, the following optimal output level and carbon emissions can be derived by taking the first derivative of the parameter $\mathrm{M}$:

$$\mathit{\partial }\mathsf{\pi }/\mathit{\partial }\mathrm{M}=\left[\right(\mathrm{r}-\mathrm{v}){\sigma }_{\mathrm{A}}-{\sigma }_{\omega }{\sigma }_{\mathrm{A}}{\mathrm{P}}_0-{\sigma }_{\omega }{\sigma }_{\mathrm{P}}-2{\sigma }_{\omega }{\sigma }_{\mathrm{P}}{\sigma }_{\mathrm{A}}\mathrm{M}]{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }$$

(9)

If Equation (9) equals 0, economic growth $\mathrm{M}$ driven by GF and the output $\mathrm{O}$ of the representative enterprises that select the optimal greening level can be summarized as Equations (10) and (11), respectively:

$$\mathrm{M}=\left[\right(\mathrm{r}-\mathrm{v}){\sigma }_{\mathrm{A}}-{\sigma }_{\omega }{\sigma }_{\mathrm{A}}{\mathrm{P}}_0-{\sigma }_{\omega }{\sigma }_{\mathrm{P}}{\mathrm{A}}_0]/2{\sigma }_{\omega }{\sigma }_{\mathrm{P}}{\sigma }_{\mathrm{A}}$$

(10)

$$\mathrm{O}=(\mathrm{r}-\mathrm{v}-{\sigma }_{\omega }\mathrm{P}){\sigma }_{\mathrm{A}}{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }/{\sigma }_{\mathrm{P}}{\sigma }_{\omega }$$

(11)

If the goal of profit maximization is followed, the carbon emissions can be calculated as shown in Equation (12):

$$\omega =\mathrm{P}\mathrm{O}=({\mathrm{P}}_0+{\sigma }_{\mathrm{P}}\mathrm{M})(\mathrm{r}-\mathrm{v}-{\sigma }_{\omega }\mathrm{P}){\sigma }_{\mathrm{A}}{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }/{\sigma }_{\mathrm{P}}{\sigma }_{\omega }$$

(12)

According to Equations (1)–(12), carbon emissions $\omega$ are related to $\mathrm{M}$ and $\mathrm{G}$. Therefore, in this paper, to determine the optimal solutions for GF and carbon emissions, G is derived in both its first and second orders, as illustrated in Equations (13) and (14):

$$\mathit{\partial }\omega /\mathit{\partial }\mathrm{G}=\left[\right(\mathrm{r}-\mathrm{v}){\sigma }_{\mathrm{A}}-2{\sigma }_{\omega }{\sigma }_{\mathrm{A}}{\mathrm{P}}_0-2{\sigma }_{\omega }{\sigma }_{\mathrm{P}}{\sigma }_{\mathrm{A}}{\sigma }_{\mathrm{M}}\mathrm{G}]{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }/{\sigma }_{\mathrm{P}}{\sigma }_{\omega }$$

(13)

$${\mathit{\partial }\omega }^2/{\mathit{\partial }}^2\mathrm{G}=-2{\sigma }_{\mathrm{A}}{\sigma }_{\mathrm{M}}{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }$$

(14)

Equations (13) and (14) describe the variation trend of $\omega$ with $\mathrm{G}$ derived from the first and second derivatives. If Equation (13) equals 0, the optimal carbon emission level G₀ can be determined based on the objective constraints, as shown in Equation (15):

$${\mathrm{G}}_0=(\mathrm{r}-\mathrm{v}-2{\sigma }_{\omega }{\mathrm{P}}_0)/2{\sigma }_{\mathrm{M}}{\sigma }_{\mathrm{P}}{\sigma }_{\omega }$$

(15)

Equation (15) further calculates the turning point of the positive U-shaped curve (${\mathrm{G}}_0$). Since ${\sigma }_{\mathrm{M}}>0$ and ${\sigma }_{\mathrm{A}}>0$, $({\mathit{\partial }\omega }^2/{\mathit{\partial }}^2\mathrm{G})<0$. ${\mathrm{G}}_0$ is the maximum point of carbon emissions ω; when $\omega <{\omega }_0$, carbon emissions gradually increase with the level of GF. When $\omega >{\omega }_0$, it indicates that GF exerts an inhibitory effect on carbon emissions. Therefore, the above derivation results demonstrate that there is a positive U-shaped relationship between CE and GF, validating Hypothesis 1.

For analytical convenience, this paper transforms Equation (13) into differential form, as shown in Equation (16):

$$\mathrm{d}\omega =\left[\right(\mathrm{r}-\mathrm{v}){\sigma }_{\mathrm{A}}{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }/{\sigma }_{\mathrm{P}}{\sigma }_{\omega }]\mathrm{d}\mathrm{G}-[2{\sigma }_{\omega }{\sigma }_{\mathrm{A}}\mathrm{P}{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }/{\sigma }_{\mathrm{P}}{\sigma }_{\omega }]\mathrm{d}\mathrm{G}$$

(16)

In Equation (16), ${\sigma }_{\mathrm{A}}\mathrm{d}\mathrm{f}$ represents the progress in green technology resulting from improvements in GF, $(\mathrm{r}-\mathrm{v})$ denotes the difference between price and variable cost, while $(\mathrm{r}-\mathrm{v}){\sigma }_{\mathrm{A}}{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }\mathrm{d}\mathrm{G}$ indicates the profit increase resulting from green technology progress. $\left[\right(\mathrm{r}-\mathrm{v}){\sigma }_{\mathrm{A}}{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }/{\sigma }_{\mathrm{P}}{\sigma }_{\omega }]\mathrm{d}\mathrm{G}$ represents the increase in carbon emissions attributable to the rise in profits from green technology progress. Thus, it can be interpreted as the profit incentive effect. In concrete terms, ${\sigma }_{\mathrm{A}}\mathrm{P}{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }\mathrm{d}\mathrm{G}$ accounts for the increase in carbon emissions due to higher output, while ${\sigma }_{\omega }{\sigma }_{\mathrm{A}}\mathrm{P}{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }\mathrm{d}\mathrm{G}$ indicates the disposal costs associated with these incremental emissions, so $2{\sigma }_{\omega }{\sigma }_{\mathrm{A}}\mathrm{P}{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }\mathrm{d}\mathrm{G}$ represents the reduction in carbon emissions associated with the incremental disposal cost. Consequently, $[2{\sigma }_{\omega }{\sigma }_{\mathrm{A}}\mathrm{P}{\mathrm{E}}^{\alpha }{\mathrm{W}}^{\beta }{\mathrm{K}}^{1-\alpha -\beta }/{\sigma }_{\mathrm{P}}{\sigma }_{\omega }]\mathrm{d}\mathrm{G}$ in Equation (16) can be viewed as the cost suppression effect. To sum up, GF exerts a dual effect on carbon emissions, both through profit incentives and cost inhibition, and GF influences carbon emissions by driving the progress of green technology, including Gti and Eei. Therefore, Hypothesis 2 and Hypothesis 3 are validated.

Specifically, in the initial stage of GF, the profit incentive effect dominates. Supported by the dual incentives of fiscal policy and tax incentives, enterprises can freely enter and exit the market, leading to a significant influx of firms into the green market competing to enhance green technologies, reduce costs, and achieve higher marginal returns. While most green technologies do not produce direct carbon emissions during use, the construction of related facilities (such as solar panels and wind turbines) requires raw materials and energy. These essentials can generate carbon emissions during production, transportation, installation, and maintenance. Driven by profit incentives, enterprises will allocate substantial funds toward constructing green infrastructure and transforming industrial processes. This results in a temporary increase in carbon emissions and a decline in CE. However, as GF matures, the cost suppression effect becomes predominant. At this stage, significant funding is directed toward the development and deployment of green technologies. Moreover, the scale of green technology applications will continue to expand. As economies of scale are realized, the marginal cost of green technology will gradually decline, facilitating affordability for enterprises and society, further curbing carbon emissions and enhancing CE.

4.2. Data

4.2.1. Dependent Variable

Carbon emission efficiency (CE): The current assessment of CE usually involves two perspectives: single-factor and total-factor CE. The single-factor approach emphasizes ratio analysis based on individual factors, such as carbon emission intensity and carbon productivity. However, the economic production system is complex, and the carbon emissions of a single factor cannot comprehensively reflect the relationships among the economy, environment, and energy. In recent years, many scholars have measured CE from the perspective of total-factor analysis, with the primary method employed being Data Envelopment Analysis (DEA) within the input–output framework. Although the traditional DEA model is widely used in efficiency evaluation, it has limitations, such as failing to measure the impact of undesirable output on efficiency. Therefore, this paper builds on the research of Zhao et al. (2023) [36] and employs the DEA model optimized by Tone, specifically the Super-SBM (Slacks-Based Measure) model based on undesirable output. Moreover, the carbon emission index system for input and output is constructed, with specific indices presented in

Table 1. Urban carbon emission efficiency index.

First-Order Index	Second-Order Index	Third-Order Index
Input	Capital	The capital stock (hundreds of millions of CNY) is calculated using the perpetual inventory method based on the estimation provided by Zhang et al. (2004) [74], with 2011 as the base period.
	Labor	The number of employees in units at the end of each prefecture-level city (ten thousand).
	Energy	Urban direct energy consumption includes natural gas and liquefied petroleum gas, while indirect energy consumption refers primarily to electricity consumption.
Desirable output	GDP	GDP at constant 2011 prices (CNY 10,000).
Undesirable output	Carbon emissions	Referring to the methods of Cong et al. (2014) [75], urban carbon emissions are calculated according to the total amount of range 1, range 2, and range 3.

. Compared to the conventional DEA method, the efficiency values obtained through the Super-SBM method are more objective and accurate.

To gain a deeper understanding of the urban differences and dynamic changes in CE derived from GF, Figure 2 visually presents the results as a topographic map, where darker blue indicates the highest CE level among the 276 cities in China. It can be seen that the CE from 2011 to 2021 is constantly changing, with an overall upward trend. Notably, there is a significant evolutionary trend in 2015 and 2021, attributable to the introduction of the “Energy Efficiency Credit Guidelines” in 2015 and the proposal of the “carbon peaking” and “carbon neutrality” goals in 2020. Furthermore, higher CE levels are concentrated in the eastern region, with CE increasing more rapidly in coastal areas and central cities.

4.2.2. Independent Variable

Green finance (GF): Through various financial instruments and services, GF encourages financial institutions and enterprises to invest capital in sustainable development areas such as environmental protection, energy conservation, and clean energy, thereby promoting long-term sustainable economic and social development [76]. Based on the principles of integrity, scientific rigor, and data availability, this paper draws on the research of Wang et al. (2024) to establish a comprehensive evaluation index system of GF covering seven aspects [77]. Green credit reflects the role of financial institutions in supporting environmental protection projects and indicates the priority that financial institutions place on green development relative to their overall credit activities. Green investment highlights the government’s or enterprises’ commitment to pollution reduction and reflects the integration of green principles into economic growth strategies. Green insurance is an essential tool for risk management in GF. Green bonds are newly developed financial instruments designed to improve the environment by raising funds for environmentally friendly and sustainable projects. The issuance of green bonds can help reduce a company’s carbon dioxide emissions and improve its ESG environmental score. Green support represents the government’s fiscal commitment to environmental sustainability, reflecting public sector support for green development through direct expenditure. Green funds demonstrate the integration of green principles into asset management and investment portfolios. Green equity represents the incorporation of green concepts into financial markets. This variable reflects the maturity of market-based mechanisms in promoting low-carbon and sustainable development. The specific indicator system is presented in

Table 2. The evaluation index system of GF.

Primary Index	Indicator Description	Obs	Mean	Std. Dev.	Min	Index Attribute
Green credit	Credit volume for environmental protection projects/total credit volume	3036	0.049	0.018	0.007	+
Green investment	Investment in environmental pollution control/GDP	3036	0.012	0.005	0.002	+
Green insurance	Environmental pollution liability insurance income/total premium income	3036	0.022	0.008	0.003	+
Green bonds	Total green bond issuance/total bond issuance	3036	0.007	0.003	0.001	+
Green support	Fiscal environmental protection expenditure/general budget expenditure	3036	0.007	0.004	0.001	+
Green funds	Total market value of green funds/total market value of all funds	3036	0.050	0.019	0.007	+
Green equity	Carbon trading, energy use right trading, emission right trading/total equity market transactions	3036	0.025	0.011	0.003	+

. Additionally, based on the multi-dimensional attributes of GF, we selected the seven representative indicators above. The GF index is ultimately calculated using the entropy weight method, and the calculation steps are as follows (performed in Stata 17). The entropy method is widely recognized in comprehensive evaluations for effectively reducing human error [78].

• Standardize the data: To ensure comparability across indicators with different units, the seven types of raw data must be standardized. The following formula is used:

$${\mathrm{X}}_{\mathrm{i}\mathrm{j}}^{\prime }={\displaystyle \frac{{\mathrm{X}}_{\mathrm{i}\mathrm{j}}-\mathrm{m}\mathrm{i}\mathrm{n}\left({\mathrm{X}}_{\mathrm{j}}\right)}{\mathrm{m}\mathrm{a}\mathrm{x}\left({\mathrm{X}}_{\mathrm{j}}\right)-\mathrm{m}\mathrm{i}\mathrm{n}\left({\mathrm{X}}_{\mathrm{j}}\right)}}$$

where ${\mathrm{X}}_{\mathrm{i}\mathrm{j}}$ is the original value of the $\mathrm{j}$ indicator for the $\mathrm{i}$ sample, and ${\mathrm{X}}_{\mathrm{i}\mathrm{j}}^{\prime }$ is the standardized value.

2.

Calculate the information entropy of each indicator: The entropy method determines the weight of each indicator based on its contribution to the dataset. The steps are as follows:

(1)

Compute the proportion for each indicator:

$${\mathrm{P}}_{\mathrm{i}\mathrm{j}}={\displaystyle \frac{{\mathrm{X}}_{\mathrm{i}\mathrm{j}}^{\prime }}{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{X}}_{\mathrm{i}\mathrm{j}}^{\prime }}}$$

(2)

Calculate the entropy value for each indicator:

$${\mathrm{e}}_{\mathrm{j}}=-\mathrm{k}\sum _{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{P}}_{\mathrm{i}\mathrm{j}}\ln \left({\mathrm{P}}_{\mathrm{i}\mathrm{j}}\right),\hspace{1em}\mathrm{k}={\displaystyle \frac{1}{\ln \left(\mathrm{n}\right)}}$$

Here, $\mathrm{k}={\displaystyle \frac{1}{\ln \left(\mathrm{n}\right)}}$ is a constant to ensure that the entropy value lies between 0 and 1, and $\mathrm{n}$ is the number of samples.

(3)

Determine the entropy weight:

$${\mathrm{w}}_{\mathrm{j}}=1-{\mathrm{e}}_{\mathrm{j}}$$

A smaller ${\mathrm{e}}_{\mathrm{j}}$ indicates greater variation among samples, resulting in a higher weight for the corresponding indicator.

3.

Compute the comprehensive score: The comprehensive score for each sample is calculated using the entropy weights:

$${\mathrm{S}}_{\mathrm{i}}=\sum _{\mathrm{j}=1}^{\mathrm{m}}{\mathrm{w}}_{\mathrm{j}}{\mathrm{X}}_{\mathrm{i}\mathrm{j}}^{\prime }$$

where ${\mathrm{S}}_{\mathrm{i}}$ is the GF index for the sample $\mathrm{i}$, and ${\mathrm{w}}_{\mathrm{j}}$ is the weight of the indicator $\mathrm{j}$.

4.2.3. Mechanism Variables

Green Technology Innovation (Gti). This paper draws on the research of Zhou et al. (2022) [61] and utilizes the number of green invention patent applications as a proxy variable for the level of Gti. Based on the IPC Green List code issued by the World Intellectual Property Organization (WIPO), this paper obtains the number of green patent applications in 276 cities in China.

Energy Efficiency Improvement (Eei). Drawing on Shi et al. (2020), this paper adopts green total-factor energy efficiency as a measure of Eei [79]. Specifically, this index considers labor, capital, and energy as input factors, with regional GDP as the desirable output index, and industrial sulfur dioxide, particulate matter, and wastewater discharge as the undesirable output index factors. This is calculated using the Super-CCR (Charnes–Cooper–Rhodes) index method, which not only evaluates improvements in production efficiency but also reflects the roles of technological progress and environmental governance in promoting pollution reduction and mitigating carbon emissions.

4.2.4. Threshold Variable

Digital Economy (Dig). With the rapid advancement of the Dig, various methods for its measurement have diversified. Currently, three primary approaches are commonly used: value-added measurement, satellite accounting, and index construction. Since relying on a single indicator lacks comprehensiveness and scientific rigor, the index construction method is preferred for its advantages in data availability, broader content coverage, and efficient indicator calculation and processing. Taking into account the availability of urban data, based on the research of Zhao et al. (2020) [80], this paper measures the development level of the Dig, focusing on digital inclusive finance and internet development. In terms of internet development, this paper selects two indicators from the aspect of digital infrastructure, the number of internet broadband access users, and the number of mobile phone users. Additionally, two indicators are selected from digital industrialization: the number of employees in the information transmission, computer service, and software industry and telecommunications business income. For digital financial inclusion, three indicators are selected: the coverage breadth, the depth of use of digital finance, and the degree of digitization in inclusive finance, based on the method proposed by Guo et al. (2020) [81].

The specific indicator system is shown in

Table 3. The evaluation index system of the Dig.

Primary Index	Secondary Index	Indicator Description	Obs	Mean	Std. Dev.	Min	Max	Index Attribute
Internet development	Internet penetration rate	Number of internet broadband access users (per 100 people)	3036	25.02	18.80	0.35	189.02	+
	The proportion of internet-related employees	Information transmission computer services and software industry	3036	0.01	0.14	0	0.15	+
	Internet-related output	Telecom business income (ten thousand)	3036	0.10	0.14	0.003	2.17	+
	Number of mobile internet users	Number of mobile phone users (per 100 people)	3036	107.31	76.34	13.89	1016.6	+
Digital inclusive finance	China digital inclusive finance index	Coverage breadth	3036	176.65	74.11	1.88	371.79	+
		Usage depth	3036	180.27	72.21	12.49	354.3	+
		Digitization level	3036	219.04	83.04	2.7	581.23	+

. It employs principal component analysis (PCA) to reduce the dimensionality of key indicators, thereby constructing a comprehensive Dig development index. The main steps of the PCA are as follows:

• Standardize the data: The data for the positive indicators are standardized.

$${\mathrm{Z}}_{\mathrm{i}\mathrm{j}}={\displaystyle \frac{{\mathrm{X}}_{\mathrm{i}\mathrm{j}}-{\mu }_{\mathrm{j}}}{{\sigma }_{\mathrm{j}}}}$$

where ${\mu }_{\mathrm{j}}$ is the mean of the $\mathrm{j}$ indicator, and ${\sigma }_{\mathrm{j}}$ is the standard deviation of the $\mathrm{j}$ indicator.

2.

Construct the correlation matrix: The correlation matrix $\mathrm{R}$ is calculated from the standardized data:

$$\mathrm{R}={\displaystyle \frac{1}{\mathrm{n}-1}}\sum _{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{Z}}_{\mathrm{i}}{\mathrm{Z}}_{\mathrm{i}}^{\mathrm{T}}$$

where $\mathrm{n}$ is the number of samples, and ${\mathrm{Z}}_{\mathrm{i}}$ represents the standardized data matrix.

3.

Calculate the eigenvalues and eigenvectors. The ${\lambda }_{\mathrm{j}}$ eigenvalues represent the contribution of each principal component to the total variance, and the eigenvectors determine the weight coefficients for constructing the principal components as linear combinations of the original variables.

4.

Determine the number of principal components:

(1)

Cumulative contribution rate: Select principal components that explain at least 85% of the total variance:

$$\mathrm{Cumulative} \mathrm{Contribution} \mathrm{Rate}={\displaystyle \frac{{\sum }_{\mathrm{j}=1}^{\mathrm{k}}{\lambda }_{\mathrm{j}}}{{\sum }_{\mathrm{j}=1}^{\mathrm{m}}{\lambda }_{\mathrm{j}}}}$$

where $\mathrm{k}$ is the number of selected components, and m is the total number of eigenvalues.

(2)

Only components with eigenvalues greater than 1 are typically selected.

5.

Construct principal component expressions: Using the eigenvectors, principal components are constructed as follows:

$${\mathrm{F}}_{\mathrm{i}1}={\mathrm{a}}_{11}{\mathrm{Z}}_{\mathrm{i}1}+{\mathrm{a}}_{12}{\mathrm{Z}}_{\mathrm{i}2}+\dots +{\mathrm{a}}_{1\mathrm{m}}{\mathrm{Z}}_{\mathrm{i}\mathrm{m}}$$

$${\mathrm{F}}_{\mathrm{i}2}={\mathrm{a}}_{21}{\mathrm{Z}}_{\mathrm{i}1}+{\mathrm{a}}_{22}{\mathrm{Z}}_{\mathrm{i}2}+\dots +{\mathrm{a}}_{2\mathrm{m}}{\mathrm{Z}}_{\mathrm{i}\mathrm{m}}$$

Here, ${\mathrm{F}}_{\mathrm{i}1}$ and ${\mathrm{F}}_{\mathrm{i}2}$ are the first and second principal components. ${\mathrm{a}}_{\mathrm{i}\mathrm{m}}$ is the coefficient of the eigenvectors corresponding to each variable.

6.

Calculate the comprehensive score: Using the variance contribution rates of the principal components as weights, the comprehensive score (Dig index) is calculated as follows:

$${\mathrm{S}}_{\mathrm{i}}={\displaystyle \frac{{\lambda }_1}{\sum \lambda }}{\mathrm{F}}_{\mathrm{i}1}+{\displaystyle \frac{{\lambda }_2}{\sum \lambda }}{\mathrm{F}}_{\mathrm{i}2}+\dots +{\displaystyle \frac{{\lambda }_{\mathrm{k}}}{\sum \lambda }}{\mathrm{F}}_{\mathrm{i}\mathrm{k}}$$

where ${\mathrm{S}}_{\mathrm{i}}$ represents the comprehensive score for the sample i. ${\lambda }_{\mathrm{k}}$ is the eigenvalue of the principal component $\mathrm{k}$.

7.

Transform the comprehensive score into an index: The comprehensive score is normalized to an index for better comparison across samples:

$$\mathrm{I}\mathrm{n}\mathrm{d}\mathrm{e}\mathrm{x}={\displaystyle \frac{{\mathrm{S}}_{\mathrm{i}}-\min \left(\mathrm{S}\right)}{\max \left(\mathrm{S}\right)-\min \left(\mathrm{S}\right)}}\times 100$$

4.2.5. Control Variables

This paper selects six control variables, with their specific measurement methods detailed in

Table 4. Variable definition.

Category	Symbol	Variable Definition
Independent variable	Gf	Green finance index calculated using the entropy weight method
Dependent variable	Ce	Carbon emission efficiency calculated using the Super-SBM model
Mechanism variables	Gti	The number of green invention patent applications
	Eei	Total-factor energy efficiency calculated using the Super-CCR model
Threshold variable	Dig	Digital economy index calculated using PCA
Control variables	GDP	The per-capita GDP of prefecture-level cities
	Open	The ratio of total import and export volume to regional GDP
	Ts	The ratio of science and technology expenditure to fiscal expenditure
	Pop	The logarithm of the number of people per square kilometer
	Gov	The ratio of government expenditure to regional GDP
	Lfi	The ratio of industrial added value to regional GDP

. The primary purpose of including these control variables is to mitigate potential biases and address omitted variable issues in the model. This ensures a more accurate and reliable assessment of the impact of GF on CE, thereby providing a clearer understanding of its actual effects.

(1) The level of technology and science (Ts): Ts significantly influences a city’s capacity for technological innovation. This capacity is critical in advancing green technology development and indirectly impacts the city’s carbon emissions. Consequently, investments in scientific research and technological development may shape the effectiveness of GF policies. Therefore, controlling for Ts is essential to ensure robust and reliable empirical results.

(2) Economic development (GDP): According to the EKC theory, economically developed regions possess greater capacity and resources to invest in low-carbon technologies and clean energy. Controlling for variations in urban GDP helps mitigate its influence on the effectiveness of GF policies, enabling a more accurate evaluation of their impact on CE.

(3) The level of opening-up (Open): Opening to the outside world can introduce advanced low-carbon technologies and management practices through foreign investment, technology transfer, and international cooperation. The spillover effects of these technologies can enhance the city’s Gti capacity, thereby improving CE. Controlling this variable eliminates its potential interference with CE, allowing the model to identify the independent effects of GF policies accurately.

(4) Industrialization level (Lfi): The higher the Lfi in cities, the greater the pressure on energy consumption and carbon emissions. Controlling for Lfi helps mitigate the bias in CE caused by variations in industrial structure, thereby enabling a more precise observation of the actual impact of GF.

(5) Population density (Pop): Areas with high Pop often exhibit more intensive economic activities and infrastructure development, which may result in higher energy demand. Controlling for Pop helps eliminate the variations in CE caused by differences in population distribution.

(6) Government intervention (Gov): Gov primarily influences low-carbon project investments, the promotion of green technologies, and the implementation of energy conservation and emission reduction policies through fiscal expenditures. Controlling for the effects of Gov ensures a more accurate evaluation of the independent contribution of GF to CE.

4.3. Models

4.3.1. Benchmark Regression Model

To test Hypothesis 1, we construct the panel fixed-effects model. The benchmark regression model is shown in Equations (17) and (18):

$${\mathrm{C}\mathrm{e}}_{\mathrm{i}\mathrm{t}}={\alpha }_0+{\beta }_1{\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}_{\mathrm{i}\mathrm{t}}+\gamma {\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}_{\mathrm{i}\mathrm{t}}+{\mu }_{\mathrm{i}}+{\phi }_{\mathrm{t}}+{\epsilon }_{\mathrm{i}\mathrm{t}}$$

(17)

$${\mathrm{C}\mathrm{e}}_{\mathrm{i}\mathrm{t}}={\alpha }_0+{\beta }_1{\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}_{\mathrm{i}\mathrm{t}}+{\beta }_2{\mathrm{S}\mathrm{G}}_{\mathrm{i}\mathrm{t}}+\gamma {\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}_{\mathrm{i}\mathrm{t}}+{\mu }_{\mathrm{i}}+{\phi }_{\mathrm{t}}+{\epsilon }_{\mathrm{i}\mathrm{t}}$$

(18)

In Equations (17) and (18), ${\mathrm{C}\mathrm{e}}_{\mathrm{i}\mathrm{t}}$ is the CE of city $\mathrm{i}$ in year $\mathrm{t}$; ${\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}_{\mathrm{i}\mathrm{t}}$ is the GF of city $\mathrm{i}$ in year $\mathrm{t}$, and ${\mathrm{S}\mathrm{G}}_{\mathrm{i}\mathrm{t}}$ is the square term of ${\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}_{\mathrm{i}\mathrm{t}}$; ${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}_{\mathrm{i}\mathrm{t}}$ represents the control variable of city $\mathrm{i}$ affecting urban CE in year $\mathrm{t}$; ${\mu }_{\mathrm{i}}$ is the city fixed effect, ${\phi }_{\mathrm{t}}$ is the year fixed effect; ${\epsilon }_{\mathrm{i}\mathrm{t}}$ is the random error term.

4.3.2. Mechanism Regression Model

To further test the mediating effect of GF in enhancing CE through Gti and Eei, as proposed in Hypothesis 2 and Hypothesis 3, this paper draws on the work of Baron et al. (1999) [82] to construct a mechanism regression model based on Equations (19) and (20):

$${\mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r}}_{\mathrm{i}\mathrm{t}}={\alpha }_0+\beta {\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}_{\mathrm{i}\mathrm{t}}+\gamma {\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mu }_{\mathrm{i}}+{\phi }_{\mathrm{t}}+{\epsilon }_{\mathrm{i}\mathrm{t}}$$

(19)

$${\mathrm{C}\mathrm{e}}_{\mathrm{i}\mathrm{t}}={\alpha }_0+\beta {\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}_{\mathrm{i}\mathrm{t}}+\phi {\mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r}}_{\mathrm{i}\mathrm{t}}+\gamma {\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mu }_{\mathrm{i}}+{\phi }_{\mathrm{t}}+{\epsilon }_{\mathrm{i}\mathrm{t}}$$

(20)

In Equations (19) and (20), ${\mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r}}_{\mathrm{i}\mathrm{t}}$ is the intermediary variable, specifically including Git and Eei, and the remaining variables have the same meaning as Equation (17).

4.3.3. Threshold Regression Model

In order to further test the nonlinear effect of the Dig in Hypothesis 4, a threshold effect model, based on Equation (21), is constructed by referring to the work of Hansen (1999) [83].

$${\mathrm{C}\mathrm{e}}_{\mathrm{i}\mathrm{t}}={\chi }_0+{\chi }_1{\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}_{\mathrm{i}\mathrm{t}}({\mathrm{d}\mathrm{i}\mathrm{g}}_{\mathrm{i}\mathrm{t}}\le {\theta }_1)+{\chi }_2({\theta }_1<{\mathrm{d}\mathrm{i}\mathrm{g}}_{\mathrm{i}\mathrm{t}}\le {\theta }_2)+{\chi }_3{\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}_{\mathrm{i}\mathrm{t}}({\mathrm{d}\mathrm{i}\mathrm{g}}_{\mathrm{i}\mathrm{t}}>{\theta }_2)+{\mu }_{\mathrm{i}}+{\phi }_{\mathrm{t}}+{\epsilon }_{\mathrm{i}\mathrm{t}}$$

(21)

where $\theta$ represents the threshold value, and ${\mathrm{d}\mathrm{i}\mathrm{g}}_{\mathrm{i}\mathrm{t}}$ is the Dig of city $\mathrm{i}$ in year $\mathrm{t}$.

4.4. Data Source

According to the principles of scientific rigor and data availability, this paper selects 276 cities in China from 2011 to 2021, excluding missing data and the four municipalities directly under the central government, resulting in a total of 3036 city-year observation samples from 276 prefecture-level cities and above. The data are sourced from the China City Statistical Yearbook, China Finance Yearbook, China Industry Statistical Yearbook, China Tertiary Industry Statistical Yearbook, China Energy Statistical Yearbook, China Agricultural Statistical Yearbook, China Science and Technology Statistical Yearbook, and the official websites of relevant local governments. To facilitate the subsequent empirical research, the data for the sample period are processed as follows: first, linear interpolation is used to address missing data for individual years; second, to mitigate the effects of heteroscedasticity and dimensionality on the empirical results, some statistical data are logarithmically transformed. Furthermore, we winsorize variables at 1% and 99% to remove the impact of certain values on this study. The descriptive statistics are shown in

Table 5. Descriptive statistics.

Category	Symbol	Obs	Mean	Std. Dev.	Min	Max	Cov
Independent variables	Gf	3036	0.332	0.104	0.064	0.650	0.312
Dependent variables	Ce	3036	0.471	0.079	0.275	1.098	0.166
Mechanism variables	Gti	3036	0.043	0.119	0	1.759	2.762
	Eei	3036	0.574	0.165	0.235	2.510	0.288
Threshold variable	Dig	3036	0.104	0.088	0.006	0.891	0.843
Control variables	GDP	3036	10.745	0.560	8.773	13.056	0.052
	Open	3036	0.019	0.042	0	1.211	2.21
	Ts	3036	0.016	0.017	0.0006	0.207	1.028
	Pop	3036	5.727	0.935	0.683	7.882	0.163
	Gov	3036	0.200	0.097	0.044	0.741	0.482
	Lfi	3036	0.353	0.172	0	2.848	0.486

.

As shown in Table 5, the COV of Gti is 2.762, indicating significant variation in Gti among the sample cities and highly dispersed data distribution. This may be due to the close relationship between Gti and economic development, with developed regions typically exhibiting significantly stronger innovation capabilities than less developed ones. The COV of Open is 2.21, suggesting substantial differences in openness to the outside world across cities. This variation may primarily stem from geographical disparities, such as coastal areas being generally much more open than inland regions. The COV of Ts is 1.028, reflecting considerable variation in science and technology levels among Chinese cities, likely attributable to differences in the concentration of scientific and technological resources, research investments, and economic foundations across regions. The Dig variable has a COV of 0.843, indicating that the Dig also varies significantly among the sample cities. This discrepancy may be closely linked to regional economic structure, infrastructure, and population quality differences. The COV of Gov is 0.482, implying relatively small differences in government intervention levels across cities. These differences may result from local governments’ fiscal capacity or intervention intensity variations, constrained by regional economic conditions and policy objectives. Similarly, the COV of Lfi is 0.486, reflecting some variation in industrialization levels across the sample cities, likely influenced by differences in development stages, resource endowments, and historical trajectories. For the remaining variables, the COV values are moderately low, indicating that their distribution is relatively concentrated across Chinese cities.

5. Empirical Results

5.1. Benchmark Regression

This paper tests Hypothesis 1 according to Equation (17), and the results of benchmark regression are shown in

Table 6. Benchmark regression results.

Variables	(1) CE	(2) CE	(3) CE	(4) CE
GF	0.089 *** (0.029)	0.109 *** (0.029)	−0.465 *** (0.084)	−0.520 *** (0.083)
SG			0.738 *** (0.105)	0.843 *** (0.104)
POP		−0.020 *** (0.004)		−0.023 *** (0.023)
GOV		−0.095 *** (0.023)		−0.104 *** (0.023)
LFI		0.005 (0.006)		−0.001 (0.006)
TS		0.600 *** (0.075)		0.606 *** (0.075)
GDP		−0.008 * (0.005)		−0.006 (0.005)
OPEN		0.024 (0.031)		0.032 (0.030)
_cons	0.441 *** (0.010)	0.645 *** (0.060)	0.536 *** (0.016)	0.741 *** (12.200)
N	3036	3036	3036	3036
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
R2	0.819	0.826	0.822	0.830
F	9.309	17.533	29.540	23.896

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

. Column (1) reports that the GF is 0.089, which passes the significance test at the 1% confidence level without considering the control variables. Columns (2) reports the estimation results after incorporating the control variables. GF on CE remains positively significant at the 1% level, further supporting the conclusion that the development of GF enhances CE.

Specifically, in Column (1), an increase of one standard deviation in GF increases CE by approximately 0.0093 (0.089 * 0.104) units. Similarly, in Column (2), an increase of one standard deviation in GF increases CE by about 0.0113 (0.109 * 0.104) units. This result indicates that developing GF can help optimize resource allocation, support Gti, and enhance energy efficiency, thereby reducing carbon emissions. Columns (3) and (4) report the results after adding the quadratic term (SG) of GF according to Equation (18). Regardless of whether the control variables are included, the coefficients of GF become negative, −0.465 * and −0.520 *, respectively, indicating that the effect of GF exhibits a nonlinear pattern. In Columns (3) and (4), the SG of GF is introduced, and its positive coefficient suggests that the impact of GF on CE follows a U-shaped trajectory (negative in the initial stages and positive at later stages). During the early stages of development, GF may face challenges such as low resource allocation efficiency or limited implementation effects. However, as its scale expands and policies are optimized, its role in enhancing CE becomes increasingly evident.

To further validate these findings, this paper employs the U-test to rigorously assess the positive U-shaped relationship between GF and CE, as shown in

Table 7. U-test results.

	Lower Bound	Upper Bound
Interval	0.064046	0.650096
Slope	−0.4115866	0.5766036
t-Value	−3.041951	3.411291
P > |t|	0.0012888	0.0003718
Extreme point	0.308139
99% Fieller interval for extreme point	[0.20542544, 0.36237224]
Overall test of presence of a U shape	t-Value = 3.04   P > |t| = 0.00129

. The test results indicate that the curve first decreases and then increases, with the extreme value falling within the confidence interval and being significant at the 1% confidence level. This further supports the above conclusion, thereby confirming Hypothesis 1.

5.2. Robustness Analysis

Tobit model (Column 1): This model, developed and popularized by Nobel laureate James Tobin in 1958, is effective in addressing issues of limited dependent variables or conducting truncated data regression analysis. Given that the super-efficiency SBM model generates non-negative truncated discrete data, applying ordinary least squares (OLS) may result in biased estimates. To address this, following the methodology outlined by Wu et al. (2021) [84], this paper employs the Tobit model, accounting for individual heterogeneity, thereby enhancing the precision and stability of parameter estimates.

Additional control variables (Column 2): Many factors objectively influence CE. To reduce estimation bias stemming from omitted variables, this paper further controls for human capital levels and financial constraints, drawing on Zhan et al. (2020) [85].

Explanatory variables lagging by one period (Column 3): GF, as a form of government market intervention, may have a delayed effect on CE. Therefore, this paper includes the one-period lag value of GF in the benchmark regression to assess its impact on CE in the subsequent period.

The results are shown in

Table 8. Robustness test results.

	(1)	(2)	(3)
	CE	CE	CE
GF	0.122 ***	0.108 ***	0.074 **
	(0.021)	(0.029)	(0.031)
Control	YES	YES	YES
Constant	0.278 ***	0.736 ***	0.617 ***
	(0.039)	(0.065)	(0.065)
N	3036	3036	2760
City FE	YES	YES	YES
Year FE	YES	YES	YES
R2		0.827	0.838
F		15.438	12.410

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

. Column (1) displays the Tobit model results, which are positively significant; Column (2) shows the results after controlling for human capital levels and financial constraints, which have the same results as above; Column (3) presents the findings when the explanatory variable lags by one period, which is positively significant at the 5% level. After applying these robustness checks to the benchmark regression, the results remain aligned with the initial findings, reinforcing the robustness of the conclusion that GF enhances CE.

5.3. Endogeneity Analysis

Benchmark regression revealed that GF significantly promoted an improvement in CE. However, cities with high CE may also have high levels of GF development, which could lead to the issue of reverse causality. Additionally, while this paper introduces several control variables that influence CE to minimize the effects of omitted variables’ bias on the empirical results, it remains challenging to account for all relevant factors, and potential measurement errors may still exist. Consequently, this paper develops suitable instrumental variables and performs endogeneity analysis utilizing the two-stage least squares (2SLS) method. Specifically, this paper uses the product of the number of commercial bank branches in all prefecture-level cities and the one-period lag of GF as the instrumental variable, drawing on Xiao et al. (2023) [86]. Regarding the relevance assumption, the number of commercial bank branches in prefecture-level cities significantly influences the local level of GF, thus satisfying the correlation assumption. Additionally, CE is unlikely to influence the number of local commercial bank branches, and the lagged CF is not influenced by current shocks. Therefore, the instrumental variable also satisfies the exogeneity assumption.

Table 9. 2SLS test results.

Variables	GF	CE
	The First Stage	The Second Stage
GF		4.967 *** (3.16)
IV	0.587 *** (4.00)
Control	YES	YES
City FE	YES	YES
Year FE	YES	YES
Kleibergen–Paap rk LM		11.202 [0.0008]
Cragg–Donald Wald F		27.98 {16.38}

Note: The value in [] is the p-value, and the value in {} is the critical value of the weak recognition test at the 10% level. Standard errors in parentheses, *** p < 0.01.

presents the results. The first-stage results indicate that the instrumental variable’s impact on GF is significantly positive, demonstrating a strong correlation between the two. In the second stage, GF’s impact on CE passes the 1% significance test. This suggests that after addressing endogeneity, the results of the baseline regression demonstrate robustness.

5.4. Heterogeneity Analysis

5.4.1. Urban Resource Endowments

Based on the list of resource-based cities in the 2013 State Council document, National Plan for the Sustainable Development of Resource-Based Cities (2013–2020), this paper categorizes the 276 sample cities into resource-based and non-resource-based groups and analyzes the impact of GF on CE across different types of cities. As shown in columns (1) and (2) of

Table 10. Heterogeneity analysis results.

Variables	Resource Endowment		Low-Carbon Pilot Cities		Financial Development Levels
	Resource	Non-Resource	Yes	No	High	Low
	(1)	(2)	(3)	(4)	(5)	(6)
GF	0.025	0.141 ***	0.235 **	0.079 ***	0.147 ***	0.022
	(0.032)	(0.040)	(0.092)	(0.027)	(0.047)	(0.029)
Constant	0.096	1.086 ***	0.865 ***	0.537 ***	0.953 ***	0.601 ***
	(0.060)	(0.093)	(0.276)	(0.056)	(0.114)	(0.089)
N	1144	1892	550	2486	1518	1518
Control	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
R2	0.796	0.834	0.861	0.798	0.864	0.829
F	7.568	20.615	6.580	14.070	13.194	4.376

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

, the effect of GF on CE in resource-based cities is insignificant (0.025), while its effect on non-resource-based cities is significantly positive (0.141 ***). This indicates that GF has yet to significantly improve CE in resource-based cities, whereas it has a notably positive effect on non-resource-based cities.

First, resource-based cities heavily depend on the extraction and processing of natural resources. These cities’ economic structures are often dominated by industries such as mining, oil, gas, coal, and forestry, which lead to significant environmental pollution and excessive resource extraction. Second, resource-based cities often experience carbon lock-in, where their economic development is deeply tied to high-carbon industries, making transition efforts particularly challenging. Although GF can promote investment in low-carbon projects, the long-standing reliance on traditional energy sources in these cities hampers the expansion of new green technologies and industries, limiting the impact of GF.

In contrast, non-resource-based cities rely more on tertiary industries and light industries, which can more easily achieve low-carbon transitions with the support of GF. Moreover, the development of emerging industries enhances these cities’ capacity to absorb capital and technology, thereby amplifying the positive impact of GF on CE.

5.4.2. Low-Carbon Pilot Cities

This paper categorizes the 276 sample cities into two groups: low-carbon and non-low-carbon pilot cities, based on the list of low-carbon pilot cities published since 2010. As shown in Columns (4) and (5) of Table 10, the effect of GF on CE is significant for both types of cities. However, the regression coefficient for the low-carbon pilot cities (0.235 **) is significantly greater than that for non-low-carbon pilot cities (0.079 ***), indicating that the positive effect of GF on CE is more pronounced in low-carbon pilot cities. This is probably because low-carbon pilot cities actively pursue carbon emission reductions, improve energy efficiency, and promote sustainable development through the formulation and implementation of low-carbon development strategies supported and guided by government policies. Within the national or local policy framework, these cities take the lead in practicing the model of low-carbon development to address climate change and advance low-carbon economic transformation, which positively impacts CE.

5.4.3. The Level of Financial Development

Due to the varying levels of financial development across cities in China, the effects of GF in different cities on CE are also different. This paper divides cities into those with low financial development and those with high financial development according to the annual median, as shown in Table 10. Columns (5) and (6) show that the cities with a high level of financial development are significantly positive (0.147 **), whereas their impact in cities with lower financial development is not statistically significant (0.022). This suggests that the effect of GF on enhancing CE is more significant in cities with higher financial development.

The reason is that, firstly, cities with higher financial development levels have more comprehensive financial systems, offering diverse financial instruments and sufficient financial support. Secondly, these cities possess more mature financial markets, where participants have strong risk assessment and management capabilities for GF, enabling better support for green projects and promoting carbon emission reduction. Finally, enterprises in these cities have greater financing capacity, allowing them to more easily access the support of GF to upgrade and implement low-carbon technologies, thereby improving CE.

5.5. Mechanism Analysis

Table 11. Mechanism analysis results.

	(1) Gti	(2) CE	(3) CE	(4) Eei	(5) CE	(6) CE
GF	0.221 *** (0.050)	0.052 ** (0.026)	0.047 *** (0.010)	0.114 (0.082)	0.089 *** (0.025)	0.0247 ** (0.010)
Gti		0.259 *** (0.010)	0.373 *** (0.011)
Eei					0.169 *** (0.006)	0.234 *** (0.006)
Sobel			0.036 *** (0.007)			0.058 *** (0.007)
bootstrap	YES			YES
Control	YES	YES	YES	YES	YES	YES
Con_	0.533 *** (0.105)	0.508 *** (0.054)	0.116 *** (0.035)	1.027 *** (0.173)	0.472 *** (0.053)	−0.195 *** (0.033)
City FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
N	3036	3036	3036	3036	3036	3036
R2	0.772	0.861	0.5248	0.676	0.867	0.5479

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

reports the results of the mechanism analysis, focusing on the mediating effects of Gti and Eei. The findings provide critical insights into how GF influences CE.

From the mediating effect results of Gti, in Column (1), the regression coefficient of GF on Gti is 0.221, which is statistically significant at the 1% level. This result indicates that GF effectively promotes Gti, validating its role as a driver of green technology advancement. Column (2) shows that Gti improves CE, and the coefficient of Gti on CE is 0.259, which is also significant at the 1% level. Additionally, the coefficient of GF on CE remains positive and statistically significant, although it is reduced compared to its direct effect. This suggests that Gti serves as a partial mediator in the relationship between GF and CE. In other words, part of GF’s positive impact on CE is transmitted through its influence on Gti. Moreover, the Sobel and bootstrap tests in Column (3) further demonstrate that GF improves CE by promoting Gti, thus verifying Hypothesis 2.

From the mediating effect results of Eei, the regression coefficients in Column (4) are insignificant. In Column (5), however, the coefficient of Eei on CE is 0.169 and significant at the 1% level, indicating that Eei positively contributes to CE improvement. Additionally, the coefficient of GF on CE remains positive and significant, further suggesting the need to examine the mediating role of Eei. The Sobel and bootstrap tests were performed, as shown in Column (6). The results confirm that Eei exhibits a partial mediating effect in the relationship between GF and CE. Specifically, GF indirectly improves CE by enhancing Eei. This result provides robust empirical evidence for Hypothesis 3, underscoring the importance of improving energy efficiency through which GF drives low-carbon development.

5.6. Threshold Effect Analysis

This paper utilizes the Dig as the threshold to investigate the nonlinear characteristics of GF impacts on CE at different stages of the Dig. The test results, presented in

Table 12. Threshold regression results.

Test Parameter	Threshold	F-Value	p-Value	10%	5%	1%
Dig	Threshold 1	85.33	0.0000 ***	30.7426	38.6666	47.3155
	Threshold 2	44.17	0.0033 ***	17.2904	20.0270	29.9197
	Threshold 3	27.64	0.4200	45.3233	53.3662	71.6256

Note: Standard errors, *** p < 0.01.

, are derived from 1000-bootstrap repeated sampling.

The results indicate that the double-threshold effect test is significant at the 1% level, while the triple-threshold test fails to produce statistically significant findings. The double-threshold values are 0.0194 and 0.2039, respectively. Furthermore, Figure 3 displays an LR diagram with the Dig as the threshold variable, with LR values observed in both the upper and lower intervals of the critical curve, confirming the validity of the threshold values. These results collectively confirm the presence of a double-threshold effect based on the Dig in the impact of GF on CE.

Table 13. Digital economy threshold regression results.

	GF (DIG ≤ 0.0194)	GF (0.0194 < DIG ≤ 0.2039)	GF (DIG > 0.2039)
GF	0.093 ** (0.043)	0.135 *** (0.046)	0.043 (0.049)
Control	YES
Con_	0.381 *** (0.152)
N	3036
R2	0.1143

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

shows the results of the Dig. GF can significantly improve CE at both the low Dig stage (DIG ≤ 0.0194) and medium Dig stage (0.0194 < DIG ≤ 0.2039). However, at the high Dig stage, the impact of GF on CE is not significant, thus verifying Hypothesis 4.

6. Conclusions and Recommendations

6.1. Conclusions

Understanding the impact of GF on CE and proposing green finance policies with Chinese characteristics have important theoretical and practical significance for achieving the “dual-carbon” goals, optimizing urban energy structures, addressing regional resource imbalances, strengthening policy synergies, and advancing the global green development agenda. Scholars face limitations in studying how GF influences urban low-carbon development. Moreover, there is a lack of comprehensive research on the internal mechanisms through which GF affects CE and the synergies between the Dig and GF. Therefore, to deepen the theoretical understanding of the relationship between GF and CE, this study provides empirical evidence on how GF enhances CE through Gti and Eei, particularly revealing the U-shaped relationship between the two. This finding enriches the theoretical framework of GF in low-carbon development and offers practical guidance for policymakers to leverage GF better. Additionally, this study highlights the heterogeneity of GF’s impact across cities, providing a basis for regional policy design. Furthermore, it uncovers the nonlinear effect of the Dig on the relationship between GF and CE, attributing this effect to the dual role of the Dig in CE. On the one hand, the Dig positively influences CE by improving resource utilization, promoting green technology innovation, and facilitating the adoption of clean energy technologies, thereby reducing the marginal effect of GF on carbon emission reduction. On the other hand, as the Dig develops, the energy demand for digital infrastructure, such as data centers and communication facilities, increases significantly. Their overall energy consumption remains high, particularly in regions reliant on traditional fossil fuels for electricity, offsetting some of GF’s emission reduction effects. This finding provides a fresh perspective on the relationship between the Dig and sustainable development, encouraging scholars to consider the synergies among financial policy, technological progress, and digital transformation when exploring low-carbon development pathways. Moreover, the results offer actionable insights for policy design, emphasizing the importance of tailoring policies to regional characteristics. For instance, by identifying the more decisive influence of GF in low-carbon pilot cities and cities with higher levels of financial development, this study provides a foundation for formulating differentiated policies. It also underscores the marginal adverse effects of rapid Dig development, advocating for integrating the Dig into GF-related policymaking to achieve a more effective policy mix.

This study further examined the conclusions using data analysis to validate the findings from both theoretical and empirical perspectives. Specifically, utilizing balanced panel data from 276 prefecture-level cities spanning 2011 to 2021, this study employs a panel fixed-effects model, mediation effect model, and threshold effect model to empirically test the direct, indirect, and nonlinear impacts of GF on urban CE. Firstly, GF can significantly enhance CE, and the two have a U-shaped relationship. This conclusion remains robust after various tests, including the U-test, the Tobit model, the addition of control variables, and the instrumental variable method. Secondly, the impact of GF on CE exhibits heterogeneity, with more pronounced effects in non-resource-based cities, low-carbon pilot cities, and cities with higher levels of financial development. Thirdly, the mechanism analysis reveals that GF significantly improves urban CE by promoting Gti and Eei. Fourthly, GF has a nonlinear effect on CE, whereby the positive impact of GF on urban CE is weakened when the Dig becomes excessively high.

6.2. Recommendations for Policy Makers

Firstly, the government should start developing a unified GF evaluation system. Currently, the standards for GF in China are not yet fully developed, and there is a lack of clear frameworks or support in areas such as carbon accounting, product innovation, and environmental oversight. Given the conclusion that GF can facilitate improvements in urban CE, the government can strengthen efforts in the following areas: First, improve the green finance policy framework, formulate and optimize the green finance policy system, strengthen the diversity of green financial products, such as green bonds, green funds, and carbon trading financial instruments, and ensure that funds flow efficiently to energy conservation and environmental protection. In addition, a green finance performance evaluation mechanism should be established to ensure transparency and efficiency in the use of funds. Second, financial institutions should be encouraged to use fintech, big data, and other technical means to establish carbon accounting databases. This would enable the centralized management of carbon emission data from different enterprises and projects. It would also facilitate financial institutions’ better identification of enterprises’ needs, design customized green financial products, and foster innovation in GF. Third, the government can strengthen the supervision of environmental information disclosure by incorporating environmental transparency into the criteria for evaluating enterprise green performance. This approach would help financial institutions direct green investments based on enterprises’ carbon reduction efforts and allow them to develop appropriate risk control strategies promptly. Fourth, we need to strengthen multi-dimensional policy coordination. The positive U-shaped relationship of green finance indicates that its effect on carbon emission efficiency requires long-term policy support and optimization. Therefore, it is necessary to coordinate green finance policies with digital economy development, industrial transformation, and regional development policies to form multi-dimensional policy synergies and further improve carbon emission efficiency.

Secondly, the government can implement targeted GF policies based on each city’s resource endowment, policy coordination, and financial development level. Resource endowment, financial maturity, and spatial differences limit the effectiveness of a single policy. First, for resource-based cities like Zhangjiakou, Chengde, and Tangshan, local governments can guide the transformation of resource-dependent industries toward low-carbon and environmentally sustainable practices, supporting the development of clean energy, renewable energy, and high-tech industries to reduce reliance on traditional resources. Simultaneously, enterprises should be encouraged to take emission reduction measures through financial subsidies and tax incentives. At the same time, strict environmental protection laws and regulations should be formulated and enforced to impose emission limits and monitoring requirements on high-emission industries. Second, cities with high financial maturity can focus on local synergistic development, leveraging economies of scale and financial radiation to drive the common development of surrounding areas. Financially advanced cities can invest in the construction of regional financial infrastructure, such as exchanges and settlement centers, to enhance the financial service capacity of surrounding cities. Additionally, a regional financial information-sharing platforms can be established to promote data interconnectivity, improve surrounding cities’ financial transparency and efficiency, and thereby bridge the financial development gap between them.

Thirdly, the government should emphasize the intermediary role of technological innovation and energy efficiency in GF to reduce pollution and carbon emissions. The government can facilitate the transformation of scientific and technological achievements by establishing a dedicated fund for green technology research and development to support relevant projects and enterprises. Simultaneously, green technology transfer centers should be set up to break down the barriers of green core technologies among enterprises, facilitating improvements in the efficiency of technological output. In addition, the government should support adopting clean and renewable energy and guide the transformation of traditional industries. Specifically, support should be given to energy-intensive industries to undertake technological upgrades, reduce their dependence on fossil fuels, and implement advanced energy-saving technologies and management practices to enhance overall energy efficiency. At the same time, a comprehensive energy use monitoring system should be established to track and evaluate the energy efficiency of various industries and sectors in real time, and energy efficiency assessment reports should be regularly released to publicize the energy efficiency levels of various industries and regions, enhancing transparency and fostering competition.

Fourthly, the government should improve the threshold conditions for GF to boost regional CE. First, to address the excessive growth of the Dig, which weakens the promotion effect of GF on urban CE, the pace of development of the Dig should be reasonably controlled. This will help avoid the blind pursuit of rapid economic growth while strengthening policy synergy between GF and the Dig, enabling the two sectors to complement each other. Second, ensuring the balanced development between GF and the Dig is essential. On the one hand, the government can formulate a comprehensive policy framework that clarifies the goals and priorities of both GF and the Dig, ensuring that they support one another in policy design and implementation, creating a synergy for their mutual development. On the other hand, a corresponding regulatory framework should be developed, considering the particularities of GF and the Dig to ensure that economic growth remains aligned with green development goals.

6.3. Recommendations for Future Research

This paper explores the relationship between GF and carbon efficiency, introduces an intermediary mechanism for optimizing resource allocation, and examines the threshold effect of the Dig, making it a significant contribution to the existing literature. However, several limitations remain that should be addressed in future research. First, due to data availability, this study focuses on 276 cities in China, leaving some cities unexamined. Based on the principle of local adaptation, county-level data should be considered for future analysis. Exploring the impact of GF on CE at the county level remains significant for further research. Second, the mechanism analysis in this paper focuses on Gti and resource allocation optimization. Future research could explore additional mechanisms that might influence the relationship between GF and CE. Finally, this paper only examines the threshold effect of the Dig in the promotion of CE by GF. Future research could investigate other potential influencing factors, which is crucial for elucidating the interplay among the Dig, GF, and urban low-carbon development.