Utilizing green finance to promote low-carbon transition of Chinese cities: insights from technological innovation and industrial structure adjustment

Abstract

Green finance (GF) has emerged as a promising tool to promote low-carbon development, while knowledge is rather limited regarding the underlying mechanism. This article aims to address this void by constructing a city-level GF index covering seven dimensions and identifying the main pathways through which GF can facilitate the low-carbon development of cities. Using a balanced panel data covering 277 Chinese cities from 2010 to 2020, the results show that: (1) China’s GF development exhibits an overall spatial differentiation of ‘high in the east and low in the west’, while the distribution of carbon intensity (CI) displays an overall spatial differentiation of ‘high in the north and low in the south’; (2) GF significantly decreases CI of cities, which is robust to employing DID strategies and IV estimations; (3) The role of GF on CI varies with the level of CI whereas not with the level of GF. Specifically, the mitigating effect of GF on CI is significant in both high GF and low GF groups, but only in high CI group; and (4) GF promotes low-carbon transition of cities through mainly on adjusting industrial structure rather than stimulating technological innovation. Despite we also demonstrate green finance enhances green innovation, due to multi-factors, such technology progress it brings may not always translate into a tangible improvement in green productivity. For most developing countries including China, the future policy objective of green finance should focus on enhancing sustainable technological progress.

Introduction

Climate change has become a global concern, and the need for sustainable development has become increasingly urgent¹. Cities, responsible for over 75% of global carbon emission despite occupying just 2% of the world’s land area, play a pivotal role in achieving low-carbon and sustainable development^2,3. The United Nations 2030 Agenda adopted at the 2015 Sustainable Development Summit proposed 17 Sustainable Development Goals (SDGs), which can be summarized in three dimensions: social, economic, and ecological, aiming to jointly facilitate sustainable socio-economic advancement at the regional level⁴. Given the multi-objectives of the SDGs, the nature of low-carbon transition of cities lies in achieving socio-economic advancement while maintaining environmental sustainability.

China, as an important economic growth player in the Asia–Pacific region⁴, has experienced a long-term and high-speed growth since its reform and opening up. However, this crude approach with high-energy consumption and high-emissions is no longer sustainable in the context of uncontrolled carbon emissions and ecological degradation. For example, China’s cumulative carbon emissions from energy combustion, industrial process, and land use have exceeded 261.4 billion tons (Gt) since the Industrial Revolution in 1750, ranking third after the United States and Europe⁵. However, in terms of annual emissions, China's energy-related carbon emissions reached 11.4 to 12.1 Gt^6,7, ranking first in the world. As the world’s leading carbon emitter, China has announced two ambitious goals to be achieved by 2030: reaching a peak carbon emission and reducing its carbon intensity by 60% to 65% relative to the 2005 level. Recently, at the 75th General Debate of the United Nations General Assembly, the Chinese government pledged to enhance its Nationally Determined Contributions (NDCs) and aim to achieve carbon neutrality by 2060⁸. Nevertheless, another issue that cannot be ignored is that China is still the world's largest developing country. Unlike industrialized countries, China's industrialization and urbanization processes have not yet been completed, indicating that pursuing economic growth remains China's main goal for a long period of time, both currently and in the future. Obviously, seeking the coordination between socio-economic development and environmental improvement is a systemic transformation that China must undergo to achieve the goal of carbon neutrality before the mid-21th century.

Actually, since the mid-1960s, policymakers worldwide have sought various environmental regulations to balance the conflict between environmental governance and economic development, including control and command (CAC) regulation and market-based instruments (MBIs)⁹. Green finance, as a typical MBI, refers to the integration of environmental considerations into financial decision-making, which has emerged as a promising tool to support and promote sustainable development in both developed and developing economies¹⁰. Compared with traditional finance, green finance not only provides funding for enterprise development, but also aiming at encouraging these enterprises to address environmental issues such as climate change by setting standards for measuring the impact of commercial activities¹¹. It is evident that the introduction of green finance aims to promote coordinated development of economic, environmental and social structures to achieve sustainable development¹¹.

Numerous studies have examined the role of green finance in areas such as green productivity¹², green innovation¹³, environmental equality^4,14, carbon emissions¹⁵, energy efficiency and energy structure optimization^16,17,18. Most of these studies support the positive role of green finance in promoting sustainable and low-carbon development, however, leaving two research directions that warrant further investigation. The first direction involves developing an appropriate indicator to measure the development of green finance at the city level. Existing literature on China’s green finance measurement mostly employs a certain indicator of green credit, green securities, green insurance and green investment as proxies^19,20, or constructs a comprehensive index using these indicators^12,14,21. Due to data limitation, however, existing comprehensive green finance indices rarely consider green funds and carbon finance, which are important components of China's green finance that have experienced rapid development (see “Typical facts” section). Other studies therefore shift the perspective to using the pilot programs related to green finance as policy shocks^22,23, but most of these measurements are conducted at the provincial level. Given that China's green finance policies are piloted at the city level, existing policy evaluation literature based on provinces may overestimates the policy effects.

The second research direction involves the insufficient examination of underlying mechanisms. Technological innovation (TI) and industrial structure (IS) adjustment are often considered as the two key pathways to achieving low-carbon development. On the one hand, as technological innovation requires significant upfront investments, green finance can bridge this financing gap by providing funding and support for technological innovation¹². On the other hand, green finance can also promote the adjustment of industrial structure by channeling capital towards cleaner and sustainable industries while restricting access to credit and investment in energy-intensive sectors²⁴. Although both technological innovation and industrial restructuring enable cities to reduce reliance on fossil fuel consumption and towards low-carbon pathways, their underlying mechanisms are different. Technological innovation aims at improving energy efficiency by financing cleaner technologies, new processes, and energy-saving equipment. In contrast, the adjustment of industrial structure aims at reducing the scale of pollution-related production capacity by compulsorily controlling for the flow of capital to energy-intensive sectors. Comparatively, industrial restructuring is a relatively crude and short-term strategy; In the long run, the low-carbon transition of cities should however rely more on green innovation pathway with sustainable technological progress. Currently, it remains unclear which path green finance mainly relies on to promote the low-carbon transition of cities.

To fill the above research gaps, using a balanced panel data covering 277 Chinese cities from 2010 to 2020, this article aims to investigate how green finance can facilitate low-carbon development in China by focusing on two potential avenues: technological innovation and industrial structure adjustment. Unlike previous studies that mostly used the levels of carbon emissions to measure low-carbon development, we focus on the decrease in carbon intensity, which reflects the transition of urban development models by emphasizing the dependence of economic growth on carbon emissions. Actually, the Chinese central government has been using carbon emissions per unit of GDP as the official indicator when establishing the milestones for the comprehensive green and low-carbon transition of the economy²⁵. Therefore, utilizing carbon intensity to measure the level of low-carbon development enhances the connection between our findings and policy formulation. To accurately capture the level of green finance development, we construct a comprehensive GF index covering seven dimensions of green credit, green bonds, green insurance, green investment, green support, green funds, and green equity, which allows us to employ a fixed-effects model to obtain the marginal effect of green finance on carbon intensity. Also, to overcome potential endogeneity, we consider China's green finance pilot zones as policy shocks and estimate the average policy treatment effect using a DID strategy.

Our study contributes to the existing literature in fourfold. Firstly, we provide a comparative analysis of the potential pathways through which green finance can promote low-carbon development, which will offer valuable insights into the most effective strategies for a better understanding of the role of green finance in promoting sustainable development. Secondly, utilizing the advantages of Python web scraping technology, we develop a city-level green finance index covering seven major dimensions of green financial products and services in China, which accurately captures the level of green finance development in each city by incorporating additional consideration for green funds and carbon finance. Thirdly, unlike industrialized countries with mature financial markets, China's green finance is still in its early stages, thus our findings enrich the knowledge about the role of green finance in developing countries. Fourthly, our results indicate that the current green finance policy in China mainly promotes the low-carbon development of cities through industrial structure adjustment. Although our results indicates that green finance can promote green innovation, such technological advancement has not yet led to a decrease in carbon intensity. These findings hold valuable implications for policymakers, financial institutions, and urban planners seeking to integrate green finance into their low-carbon development strategies.

The remainder of this paper is structured as follows. “Literature review, theoretical hypothesis, policy background and typical facts” section provides the literature review, theoretical hypothesis, and policy background in China. “Method and data” section introduces the method, variables and data. “Results” section presents the empirical results. “Discussion” section discuss the main findings. “Conclusion and policy implications” section details the conclusion and policy implications.

Literature review, theoretical hypothesis, policy background and typical facts

Literature review

Measurement of green finance

Although China’s green finance market was not officially established until 2007, a body of studies have attempted to measure the level of China's green finance development. According to the methods used, these studies can be roughly divided into three categories. The first category, due to the data limitation, mostly uses a certain indicator such as green credit or green bonds as proxy variables^19,20,26. As a single indicator cannot accurately reflect the actual development level of green finance, the second category tends to construct a comprehensive index using multi-indicators. For example, Bai et al.²¹ took green credit, green securities, and green investment as a comprehensive index. While Lee and Lee¹² and Zhou et al.¹⁴ developed a comprehensive index considering green credit, green securities, green insurance, and green investment. Liu et al.⁴ established a comprehensive index including green credit, green securities, green insurance, green investment, and carbon finance. Xiong et al.²⁴ constructed a comprehensive index, covering green credit, green insurance, green investment, and Green fiscal support. Overall, the existing comprehensive indices rarely include green funds and carbon finance. The third category usually adopts a policy evaluation framework, using China's pilot projects related to green finance as policy shocks to measure the level of green finance development^22,23, and most of which are measured at the provincial level. However, as China's green finance policies are piloted based on cities, the existing policy evaluation literature based on provinces may overestimates the policy effects.

Evaluation of benefits of green finance

The emergence of green finance has innovatively connected environmental governance and financial development, attracting an increasing number of research to evaluate the effectiveness of this new financial tool in addressing climate challenges, which mainly focuses on four topics. (1) Green innovation and green productivity. Lee and Lee¹², for example, found that green finance development significantly improves provincial green productivity in China. Zhou et al.²² concluded that green credit promotes green technology innovation by increasing liquidity, debt financing, and corporate profits, but which mainly exist in heavily polluting enterprises. However, the quality improvement of green innovation brought about by green credit is not always significant. Wang et al.¹³ argued that green finance positively affects green innovation in countries with lower level of green finance. Conversely, it negatively affects green innovation in countries with better green innovation or environmental performance. (2) Environmental performance²⁷. Sun et al.²⁸, for example, concluded that green finance could promote firm’s ESG performance, which is moderated by market concentration and social trust²⁹. However, this positive impact on regional ecological efficiency is not obvious in general³⁰, which varies for different levels of economic development¹⁴. (3) Carbon emissions^{15,21,28,31,32,33}, and (4) Energy efficiency, energy intensity and energy structure optimization^{16,17,18,20,26,28}. Most of these studies support the positive role of green finance in reducing carbon emissions and improving energy efficiency, however, the underlying mechanisms have not well-examined.

Hypothesis development

As summarized by Liu et al.³⁰ and Wen et al.³⁴, the development of green finance can lead to corresponding resource allocation effect and innovation-driven effect.

At the macro-level, through green financial product and instrument innovation, financial institutions can effectively utilize the resource allocation effect of finance systems and guide the capital flow to low-energy consumption and low-emission industries³⁰. On the one hand, for example, by increasing the financing cost of energy-intensive industries and limiting their access to credit²⁴, green finance can directly force these industries to reduce the production scale or exit the market²¹. On the other hand, according to the “signalling theory”, green finance can utilize the transaction prices in the securities market to transmit policy signals of green development to the public, further strengthening the guiding role of credit funds in promoting the participation of social capital in green industrial investment²¹. In other words, by channeling the flow of capital resources from high-emission and low-efficiency industries to low-emission and high-efficiency industries, green finance can play an important role in promoting the optimization and upgrading of industrial structure towards a more sustainable path²¹.

At the micro-level, although technological innovation is a high-investment and high-risk activity (Wen et al. 2024), green finance can induce innovation-driven effect by providing capital support and external supervision for green innovation activities of enterprises^12,17,30. On the one hand, by reducing the financing cost of green projects through green credit and green bonds, or by providing a risk-sharing mechanism for innovation activities through green insurance³³, green finance can create incentives for these enterprises to adopt cleaner production technologies, and investment in green innovation^13,22. On the other hand, by incorporating environmental assessment into the decision-making process of investment and financing, finance institutions actually play the role of external regulator, which can force enterprises to enhance efforts in green technological innovation³⁴. According to the “innovation compensation theory”, as a typical MBI, the technological innovation-driven effect triggered by green finance will offset the cost of environmental compliance, achieving a win–win situation in which environmental performance and economic performance are simultaneously improved²¹.

In summary, the resource allocation effect of green finance can promote the optimization and upgrading of industrial structure, while the innovation-driven effect of green finance can provide capital support and external supervision for technological innovation, which can ultimately enable an increase in the efficiency of economic output and/or a decrease in the scale of carbon emissions³⁰. Drawing upon the above mechanism analysis, we propose the following two hypotheses:

Hypothesis 1

The development of green finance is conducive to the reduction of carbon intensity.

Hypothesis 2

Green finance reduces carbon intensity via two channels: industrial structure adjustment and technological innovation.

Policy background in China

To accelerate the transition of economic development towards a sustainable model, in 2007, the Ministry of Ecology and Environmental of China (MEEC), the People's Bank of China (PBC), and the China Banking and Insurance Regulatory Commission (CBIRC) jointly issued the Opinions on Implementing Environmental Protection Policies and Regulations to Prevent Credit Risks, marking that the concept of green credit has officially become a major policy tool in China's environmental regulation system. In 2012, the CBIRC issued the Guidance on Green Credit, which clearly proposed to promote banking and financial institutions to actively adjust their credit structure to support the transition and upgrading of the industrial structure. In 2016, Chinese seven ministries jointly issued the Guidance on Building a Green Finance System, which pointed out China’s definition of green finance and formulate an overall strategy framework for green finance development. In 2017, with the approval of the State Council, the Green Finance Reform and Innovation Zones (GFRIZ) were piloted in eight cities from five provinces, including Zhejiang, Jiangxi, Guangdong, Guizhou, and Xinjiang.

Following the introduction of the GFRIZ, China’s green financial products and services have developed dramatically. By the end of 2021, according the data from Ren et al.²⁵: (1) the volume of green credit (domestic and foreign denominated green loans) exceeded RMB 15.9 trillion, marking a 33% increase compared to the previous year; (2) the stock size of green bonds reached RMB 1.16 trillion, with the scale of new green bonds (excluding green local government bonds) increasing by approximately RMB 607.24 billion; (3) the scale of green funds was close to RMB 800 billion, and the number of green-investment-related theme funds exceeded 50.

Compared to industrialized countries with mature financial markets, the development of China's green finance is still in the early stage. However, under the strong leadership of the central government, China’s green finance system is distinct, functioning neither as an independently developed system nor as a purely market-based system, but a collaboration of financial institutions, enterprises, markets, and the government³⁵. This government-led green finance system has several unique advantages in dealing with climate change issues (see Ren et al.²⁵ for details).

Typical facts

Distribution and evolution of green finance development

To capture the panoramic view and spatial evolution of China's green finance development, we construct a comprehensive GF index by using 277 Chinese cities from 2010 to 2020 (see “Method and data” section for details). As show in Fig. 1, during the study period, GF index experienced a growth by 34.5%, with an average annual growth rate of 3.7%. However, the rapid development of green finance in China is mainly driven by green credit and green funds. Green bonds and green investment, on the other hand, have developed slowly, which are prominent shortcomings that constrain the development of green finance in China.

Figure 1

Mean and annual changes of green finance index in 277 sample cities from 2010 to 2020.

Although the bar graphs in Fig. 1 describes the annual evolution trend of GF index and the relative development level of the seven sub-dimensions, it cannot capture the spatial distribution characteristics between different cities well. To this end, we further plot the spatial distribution map of GF index using Arc GIS software. As shown in Fig. 2, darker colors indicate higher levels of green finance development. Overall, it presents three major characteristics: (1) Cities with high levels of green finance development are mostly distributed in the GFRIZ pilot zones and eastern coastal areas of China. Such as Shanghai (Eastern coastland), Guangdong (GFRIZ pilot & Eastern coastland), Heilongjiang, Hebei (Eastern coastland), Liaoning (Eastern coastland), Guizhou (GFRIZ pilot), Hubei, Shandong (Eastern coastland), Jiangsu (Eastern coastland), and Zhejiang (GFRIZ pilot & Eastern coastland) are the top ten provinces in the rankings. It indicates that a high correlation between green finance development and urban economic vitality and policy guidance; (2) Although the development level of green finance in each city has improved over the past decade, the spatial differentiation pattern between cities has persisted, or which even shows an expanding trend of polarization. Factors such as endowment, geographical location, and infrastructure constraints contribute to this pattern; and (3) In some urban agglomerations, such as Beijing-Tianjin-Hebei, the Yangtze River Delta, and the Pearl River Delta, green finance is showing a trend of spreading from core cities to surrounding cities.

Figure 2

Spatial distribution of green finance in 277 sample cities in 2010 and 2020.

Distribution and evolution of urban carbon intensity

Similarly, as shown in Fig. 3, darker colors represent higher levels of carbon intensity (carbon emissions per RMB 10,000 Yuan of GDP). It can be seen that the carbon intensity of Chinese cities is decreasing over time, but showing a clear spatial pattern characterized by “high in the north and low in the south”. This pattern can be attributed to two reasons. On the one hand, as the cold winter climate, all northern cities need heating, with coal-fired boilers serving as the primary heat source in most of these cities. Despite the widespread development of “coal to gas/electricity” technologies in recent years³⁶, coal-fired power still remains a significant source of power generation. On the other hand, many cities in the north, especially in the northeast region, are traditional old-industrial-bases, with the industrial structure dominated by high-energy consumption and high-emissions industries, such as automobile manufacturing, steel, metallurgy, chemical and building materials.

Figure 3

Spatial distribution of carbon intensity in 277 sample cities in 2010 and 2020.

Method and data

Baseline fixed-effects model

This article uses a two-way fixed-effects model to examine the impact of green finance on urban low-carbon development. Considering the inertia inherent in economic growth and carbon emissions, we add a lagged term of carbon intensity to the model, which would alleviate the high autocorrelation of the dependent variable in the time series. Therefore, the following dynamic panel model is established.

$${CI}_{it}={\alpha }_{0}+{\alpha }_{1}{G{F}_{it}+{\alpha }_{2}CI}_{i,t-1}+{\alpha }_{3}{Z}_{it}+{\gamma }_{i}+{\delta }_{t}+{\varepsilon }_{1,it}$$

(1)

where ${CI}_{it}$ presents the carbon intensity of city $i$ in year $t$; ${CI}_{i,t-1}$ presents the lagged carbon intensity; $GF_{it}$ presents green finance index; ${Z}_{it}$ presents a vector of time-varying city-level controls that may affect carbon intensity; ${\gamma }_{i}$ and ${\delta }_{t}$ present the city fixed-effects and year fixed-effects, respectively; ${\varepsilon }_{1,it}$ presents the random error term.

Mediating effect model

As mentioned before, we assume that green finance can promote the low-carbon development of cities by simulating technological innovation and adjusting industrial structure. The aim of this section is to identify the primary pathway through which green finance promotes the low-carbon transition of cities. Thus, we introduce two mediating variables, namely technological innovation and industrial structure, and establish the following mediating effect model.

$${M}_{it}={\beta }_{0}+{\beta }_{1}G{F}_{it}+{\beta }_{2}{Z}_{it}+{\gamma }_{i}+{\delta }_{t}+{\varepsilon }_{2,it}$$

(2)

$${CI}_{it}={\theta }_{0}+{\theta }_{1}G{F}_{it}+{\theta }_{2}{M}_{it}+{\theta }_{3}{CI}_{i,t-1}+{\theta }_{4}{Z}_{it}+{\gamma }_{i}+{\delta }_{t}+{\varepsilon }_{3,it}$$

(3)

where ${M}_{it}$ presents the mediating variables, namely technological innovation and industrial structure of city $i$ in year $t$, respectively; ${\varepsilon }_{2,it}$ and ${\varepsilon }_{3,it}$ present the random error term.

Specifically, as shown in Fig. 4, we identify the mediating effect in four steps. Step 1, checking the significance of ${\alpha }_{1}$. If ${\alpha }_{1}$ is significant, proceed to the second step of the testing procedure. Step 2, checking the significance of ${\beta }_{1}$ and ${\theta }_{2}$, respectively. If both are significant, proceed to step 4; If at least one is not significant, proceed to step 3. Step 3, using the Bootstrap method to test the hypothesis ${{H}_{0}=\beta }_{1}{\times \theta }_{2}$. If the test results are significant, proceed to step 4³⁷; Otherwise, stop the analysis. Step 4, checking the significance of ${\theta }_{1}$. If ${\theta }_{1}$ is significant, indicating the existence of partial mediating effect³⁸; Otherwise, it indicates the existence of complete mediating effect³⁹.

Figure 4

Flowchart for mediation effect testing.

Endogeneity consideration

In this study, the empirical analysis may be subject to endogeneity interference. On the one hand, as shown in the typical facts analysis, the level of green finance development is highly correlated with the economic vitality of cities, and since economic activities are the main source of carbon emissions⁸. This means that green finance and carbon intensity may be jointly influenced by some confounding factors that we cannot observe. In theory, on the other hand, cities with high carbon intensity may be more inclined to develop green finance. The above two points will both lead to endogeneity in the baseline model. Therefore, we employ exogenous policy shocks and instrumental variables (IVs) to address the potential endogeneity, respectively.

(1) DID strategy. Following Ren et al.²⁵, we take GFRIZ pilot programs as the policy shock, thus employing the DID strategy to overcome the potential endogeneity.

$${CI}_{it}={\rho }_{0}+{\rho }_{1}{G{FRIZ}_{it}+{\rho }_{3}CI}_{i,t-1}+{\rho }_{4}{Z}_{it}+{\gamma }_{i}+{\delta }_{t}+{\varepsilon }_{4,it}$$

(4)

$$G{FRIZ}_{it}=treated_{i}\times time_{t}$$

(5)

where $G{FRIZ}_{it}$, a dummy variable, indicates the interaction of $treated_{i}\times time_{t}$. If city $i$ located in the GFRIZ pilot zones, $treated_{i}$ equals one and zero otherwise. $time_{t}$ equals one if GFRIZE was implemented in and after year $t$ and zero otherwise.

The consistency of the DID estimation requires the validity of the ‘parallel trend assumption’, which means that without the GFRIZ shock, the time trend in carbon intensity in the treatment group should parallel that of the control group prior to the policy shock^9,40. Following Jia⁴¹, we choose the previous year before the GFRIZ shock as the baseline, and thus employ an event study method to conduct a parallel trend test as follows.

$${CI}_{it}={\varphi }_{0}+\sum_{T=2}^{4}{\widetilde{\beta }}_{-T}{GFRIZ}_{i,t-T}+{\sum_{T=0}^{3}{\widetilde{\beta }}_{T}{GFRIZ}_{i,t+T}+\varphi }_{2}{Z}_{it}+{\gamma }_{i}+{\delta }_{t}+{\varepsilon }_{5,it}$$

(6)

where ${\widetilde{\beta }}_{-T}$ represents the influence coefficients of 3 years prior to the treatment, and ${\widetilde{\beta }}_{T}$ represents the influence coefficients of current period and 3 years following the treatment.

It should be noted that, although the parallel trend assumption is valid, endogeneity challenges still cannot be ignored in the standard DID model. . This is because that policy-makers may consider the basic conditions of each city when selecting pilot areas, and prioritize conducting pilot projects in cities with good financial foundations (such as Guangdong) or high levels of green development (such as Guizhou). In other words, whether a city is listed as a GFRIZ pilot area may not be a random. Therefore, we further integrate propensity score matching (PSM) and DID estimation to establish a PSM-DID framework to solve the possible sample selection bias.

(2) IV selection. We know that finding a purely exogenous IV is rather difficult, so we turn our perspective to the endogenous variable itself. Specifically, we use the lagged terms of endogenous variable as IVs to overcome the possible reverse causal relationship between carbon intensity and green finance.

Variable preparation

Dependent variable

We measure the level of low-carbon development of cities by their carbon intensity, which equals the ratio of carbon emissions to the gross domestic product of each city.

Independent variable

Referring to Lee and Lee¹², Zhou et al.¹⁴, and Liu et al.³⁰, the level of green finance development of cities is captured by a comprehensive index, which consists of seven dimensions of green finance products and services, namely green credit, green bonds, green insurance, green investment, green support, green funds and green equity. The detailed indicators are described in Supplementary Table S1. Different from Liu et al.³⁰ and Lee and Lee¹² , we follow Xiong et al.²⁴ to synthesize the above seven indicators into a comprehensive index ($GF$) by using the entropy-weighted method, which ensures that the weight assigned to each indicator is proportionate to its contribution to the overall evaluation.

Mediating variable

(1) Technological innovation. As Zheng et al.⁴² stated that, despite some disadvantages, patents serve as the best indicator for assessing the innovativeness of a country. In innovation research, patents are widely regarded as a statistically sound way to proxy technological innovation^42,43,44, technological innovation is measured by the number of patents applications for green technologies. Considering that the number of patent applications is related to the population size of the city, this variable is standardized by the total population of cities.

(2) Industrial structure. As the industrial sector is responsible for 70% of China's total energy consumption and 80% of its carbon emission⁴⁵, a comprehensive study of energy, environment, and climate change in China should initially focus on the industrial sector^8,46. Therefore, the industrial structure of cities is measured by the proportion of added value of the secondary industry to GDP in each city.

Control variable

To ensure the robustness of the estimation results, the following control variables are considered in the model. Urbanization ($UR$), measured by the ratio of the urban population of each city to the resident population at the end of the year, reflects the impact of population structure of cities. Science and education investment ($LnSE$), measured by the natural logarithmic of fiscal expenditure on the areas of R&D and education, reflects the level of technology and education development in each city. Fixed assets investment ($LnFI$), measured by the natural logarithmic of per capita fixed assets investment of cities, indicates the impact of economic activity and scale expansion. Government expenditure intensity ($GOV$), measured by the ratio of fiscal expenditure to GDP of cities, reflects the breadth and strength of the role played by the government in resource allocation. Furthermore, to control for the interference of unobservable confounding factors at the city level and temporal trends, city dummy variables and year dummy variables are included.

Data and descriptive statistics

The carbon emission data is from the CEADs database. The green patent data is from the China National Intellectual Property Administration (CNIPA). The data of green finance indicators is obtained by using a combination of manual and Python web scraping methods. In which, the original data sources include the official website of the National Bureau of Statistics of China, the official website of the Ministry of Science and Technology of China, the official website of the People's Bank of China, China Statistical Yearbook, China Urban Statistical Yearbook, China Science and Technology Statistical Yearbook, China Energy Statistical Yearbook, China Financial Yearbook, China Agricultural Statistical Yearbook, China Industrial Statistical Yearbook, China Third Industry Statistical Yearbook, China Environmental Bulletin, and the official platforms for government transparency and statistics departments of cities. The remaining data comes from the China Statistical Yearbook and the China Urban Statistical Yearbook.

The descriptive statistics, presented in Table 1, provide the basic information of variables used in the study. As can be seen, the mean value and standard deviation of $GF$ are 0.3259 and 0.0993, respectively, indicating that the overall level of green finance is not high, and their differences between cities are relatively small. Similarly, the standard deviation of $CI$ (0.0215) also indicates that the distribution of carbon intensity between cities is relatively concentrated.

Table 1 Descriptive statistics.

Results

Baseline regression results

Table 2 presents the baseline regression results. As shown in model 1 of Table 2, after controlling for a set of covariates, city fixed-effects, and year fixed-effects, the static panel estimates show a significant negative correlation between green finance and carbon intensity, which is significant at the 5% level. In models 2 to 4, we further introduced the lagged term of the dependent variable as one of the independent variables, thus establishing a dynamic panel model. The FE estimators of model 2 show that there is still a significant negative correlation between GI and CI, significant at the 1% level. As Nickell⁴⁷ pointed out that, for dynamic panels, FE estimations are inconsistent. In line with Che et al.⁴⁸, we replace FE estimation with System-GMM estimation. Specifically, in model 3, we introduce the first-order lagged term of dependent variable with allowing a maximum of 2 lagged-terms. In model 4, both the first-order and second-order lagged terms of the dependent variable are considered, and a maximum of three lagged-terms are allowed. Note that there are two prerequisites that must be met when using GMM estimation. First, the difference of the disturbance term does not have second-order or higher-order autocorrelation⁴⁹, i.e. $\text{Cov}\left({\Delta \varepsilon }_{it},{\Delta \varepsilon }_{i,t-k}\right)=0,k\ge 2,\forall i$. The results of Arellano-Bond test in model 3 and model 4, namely AR(1) and AR(2), both show that there is no second-order autocorrelation in the difference of the perturbation term. Second, all IVs must be valid. The results of Sargan test in model 3 and model 4 both show that the null hypothesis of “all IVs are valid” cannot be rejected at the 1% level. Overall, the baseline regression results of models 1 to 4 are quite robust, indicating that the development of green finance can significantly reduce carbon intensity of cities. Thus, Hypothesis 1 is verified.

Table 2 Baseline regression results.

In addition, we have also made some interesting findings from the estimation results of control variables. First, the carbon emission trajectory of a city shows a clear “ratchet effect”, which may be related to the “path-dependence” formed by China's extensive economic development model of high-growth and high-emissions. Second, in most cases, urbanization, science and education investment do not significantly affect carbon intensity. Third, after controlling the ratchet-effect of carbon emission trajectory, fixed-assets investment and government consumption seem to be the key factors contributing to the rise of carbon intensity, which is mainly because China's fixed-assets investment and fiscal expenditure are mainly invested in energy-intensive and high-emission industries such as infrastructure (about 25%), manufacturing (about 30%), and real estate (20%).

Robustness checks

To avoid the possible interference from extreme values caused by variable measurement or sample selection, as well as from other green development policies, we have made a series of efforts in Table 3 to check the robustness of the estimators in Table 2. Specifically, in model 5 of Table 3, we truncate the 1% and 99% percentiles of continuous variables; In model 6, we exclude the top four cities in China, namely Beijing, Shanghai, Guangzhou, and Shenzhen, as the they are super-mega cities with highly concentrated financial, population, and many other resources; In models 7 to 10, we exclude cities that were listed as the GFRIZ pilot zones, innovative city pilot projects, carbon trading pilot projects, and low-carbon city pilot projects during the study period, respectively. From the results of models 5 to 10, it can be seen that the estimated coefficients of GI have not changed significantly compared to that of Table 1, indicating the main results are not sensitive to possible outliers after eliminating the impact of the above policies.

Table 3 Robustness checks.

Heterogeneity analysis

(1) Heterogeneity in carbon intensity. We divide the full sample into high CI group and low CI group based on the 50th percentile of carbon intensity in each year. The results of model 11 and model 12 in Table 4 show that the estimated coefficient of GI is significantly negative in the high CI group (significant at the 1% level), but not significant in the low CI group, indicating green finance mainly promotes the low-carbon transition of cities with high carbon intensity.

Table 4 Robustness checks.

(2) Heterogeneity in green finance. Similarly, we divide the full sample into high GF group and low GF group based on the 50th percentile of green finance in each year. From the results of models 13 and 14 in Table 4, it can be seen that the estimated coefficient of GI is significantly negative in both the high GF group and the low GF group, indicating that the mitigating effect of green finance on carbon intensity will not vary across cities depending on the level of financial development.

Mechanism analysis

Technological innovation

As shown in Table 5, the estimated coefficients (${\beta }_{1}$) of GF are both significant at the 1% level in model 15 (${\beta }_{1}$) and model 16 (${\theta }_{1}$), respectively. While the estimated coefficient of TI is not significant in model 16 (${\theta }_{2}$). According to the mediating effect identification steps introduced in the method section, if one of ${\beta }_{1}$ and ${\theta }_{2}$ is not significant, we need to further use the Bootstrap method to check the existing of mediating effect. The results of Sobel-Goodman Mediation test using Bootstrap (1000) show that the direct effect of GF on CI is significant (−0.014, $p=0.010$), while the indirect effect of GF on CI through TI is still not significant (−0.0001, $p=0.308$). Thus, we can conclude that the mediating effect of technology innovation in the linkage between green finance and carbon intensity does not exist. This indicates that green finance can stimulate technological innovation, whereas the above technological progress it brings will not significantly translate into the improvement of green productivity. Therefore, the first channel of green finance reducing carbon intensity through technological innovation effect in Hypothesis 2 is not validated.

Table 5 Mechanism analysis.

Industrial structure adjustment

In model 17 of Table 5, the estimated coefficient of GF is significant negative at the 1% level. Meanwhile, in model 18, the estimated coefficients of GF and IS are significant negative and positive at the 5% and 1% level, respectively. Therefore, we can conclude that the mediating effect of industrial structure adjustment exist significantly in the linkage between green finance and carbon intensity. In other words, green finance can promote the reduction of carbon intensity by accelerating industrial structure adjustment. Thus, the second channel of green finance reducing carbon intensity through industrial structure adjustment in Hypothesis 2 is validated.

Endogeneity processing

(1) Standard DID estimation. As described in Eqs. (4) and (5), we take GFRIZ pilot programs as policy shocks to overcome the potential endogeneity. However, before establishing the DID model, we need to conduct a parallel trend test as expressed in Eq. (6). The results of parallel trend test displayed in Fig. 5 show that the estimated coefficients for each period before the GFRIZ policy shock cannot reject the null assumption of 'significantly zero', indicating that there is no statistically difference in carbon intensity between the treatment group and the control group. Therefore, we can conclude that the standard DID model has passed the parallel trend test. In model 19 of Table 6, we further employ the standard DID strategy described in Eqs. (4) and (5) to estimate the average treatment effect of GFRIZ policy on carbon intensity in pilot cities. The estimations show that, after the implementation of GFRIZ policy, the carbon intensity of pilot cities has significantly decreased by 0.001 (at the 5% level).

Figure 5

Parallel trend test.

Table 6 DID and PSM-DID regression.

(2) PSM-DID estimation. To avoid estimation bias caused by matching methods, 1-on-2 nearest neighbor matching, kernel matching, and local linear regression matching are employed in models 20 to 22 in Table 6, respectively. The results of three different matching methods all showed a significant decrease in carbon intensity in the pilot cities (at the 10% level).

(3) IV-GMM. In the baseline GMM estimations of Table 2, we treated green finance as an exogenous variable. To overcome possible reverse causal effect of carbon intensity on green finance development, we use the first-order and second-order lagged terms of GF index as IVs. According to Bellemare et al.⁵⁰, we cannot use the lagged term of endogenous variable as its IV unless both of the following conditions are met simultaneously: (i) The endogenous variable is highly correlated with its lagged term; (ii) The endogenous variable is a stationary auto-regressive process. For the first condition, the regression coefficient of GF index on its lagged term is significant at the 1% level, meeting the correlation requirement between GF index and its lagged term. For the second condition, the p-values of the panel unit-root tests for GF index using HT-test (allow the auto-regressive coefficients to be the same across different cross-sections) and Fisher-test (allow for different auto-regressive coefficients within different cross-sections) are both less than 0.000, indicating GF index is a stationary auto-regressive process. Moreover, the results of Arellano-Bond test show that there is no second-order autocorrelation in the difference of the perturbation term, and the results of Sargan test show that all IVs are valid. Therefore, we can use the first-order and second-order lagged terms of GF index as IVs. The estimated coefficient of GI, as shown in model 23 of Table 6, remains negative and significant at the 1% level.

Discussion

This article provides a city-level evidence on the role of green finance in promoting low-carbon transition in China, as well as reveals the underlying mechanisms. The findings enrich the understanding of environmental and economic performance of green finance development in developing countries.

Firstly, the negative impact of GF on CI indicates that a better development of green finance can make a lower level of carbon intensity, which highlights the significant role of green finance in accelerating cities towards a more sustainable and cleaner pathway in developing countries. This finding aligns with existing literature that has demonstrated the positive impact of green finance development on carbon emissions reduction^{15,21,31,32,33,51} and improvement in energy efficiency or optimization of energy structure^{16,18,20,26,52}. However, by comparison, we shift the research perspective on green finance from environmental indicators such as carbon emissions and energy consumption effects to emphasizing the relative nexus between economic growth and environmental sustainability, which can better reflect the transformation of urban development models.

Secondly, further analysis exploring channels indicates that green finance promotes the low-carbon development of cities through mainly on adjusting industrial structure rather than stimulating technological innovation. Therefore, for developing countries including China, the future policy objective of green finance should focus on enhancing sustainable technological progress. This finding aligns with Xiong et al.²⁴ who stated that the development of green finance significantly promotes the rationalization and upgrading of the industrial structure by suppressing the development of high energy consuming industries. Similarly, the above finding is also supported by Hu and Zhang⁵³ who decomposed the impact of green finance on industrial structure optimization into two effects: advancement of industrial structure and rationalization of industrial structure. They found that the implementation of GFRIZ pilot projects has significantly promoted the rationalization of industrial structure, but also inhibited the advancement of industrial structure. However, their further analysis showed that the positive effect of GFRIZ pilot projects on the industrial structure was greater than the negative effect, thus promoting the optimization of industrial structure on the whole. We suggest that this is mainly related to the policy transition of supply-side reforms in China, which has been vigorously promoted by both the central and local governments in recent years. The eliminating of overcapacity and destocking, identified as the primary tasks of supply-side reforms, is particularly concentrated in high-polluting and high-emission industrial sectors. The resource allocation effect of financial tools is brought into play to direct funds to green projects and industries, promoting industrial structure adjustment⁵⁴. This speculation can be verified from a recent report released by IEA⁶ who claimed that global carbon emission reduction in 2021 is also a result of economic slowdown, with 155 million tons of carbon dioxide stemming from the reduction of energy intensive industrial capacity, which mainly concentrated in China, the European Union, Japan, South Korea, and North America. In China, the proportion of added value of the secondary industry in GDP dropped by 10 percentage points from 47.7% in 1978 to 37.8% before the COVID-19 epidemic. Finance, as a core resource allocation tool in China, has played an important role in it.

The existing research has shown that the enhancement of technological innovation can bring about an increase in green productivity⁵⁵. However, unlike Lee and Lee¹² who stated that green finance development promotes green productivity in China, we find no evidence to support that the technological innovation effect induced by green finance will necessarily translate into a real improvement of green productivity. This finding is supported indirectly by Wang et al.¹³ who concluded that green finance positively affects green innovation in the emerging countries and countries with lower level of green finance, but negatively affects green innovation in countries with better green innovation or environmental performance. Similarly, Bi et al.⁵⁶ found that China’s green fiscal policy facilitates green technological innovation in cities with high levels of green innovation, industrial, and resource endowments, but also inhibits green technology innovation in neighboring cities. Both Wang et al.¹³ and Bi et al.⁵⁶ indicate that green policies do not necessarily bring about technological innovation effect. Li et al.⁵⁴ further pointed out that the effect of green finance on green productivity is not a simple linear relationship. Specifically, they found that only when the level of green finance development is within a certain threshold range can it truly stimulate the improvement of green productivity; Once this threshold is crossed, it will actually hinder the development of green productivity. We suggest that there are two potential reasons for this result. On the one hand, it may be related to the current low quality of patents in China. According to the data from the World Intellectual Property Organization (Available at https://www.wipo.int/portal/en/index.html), China's international patent applications under the Patent Cooperation Treaty (PCT) system surpassed the United States for the first time since 2019, ranking first in the world. In 2020, it continued to rank first with 69,000 applications, accounting for 24.9% of the world's total applications. However, the CNIPA recently claimed that the licensing rate of China’s invention patents is only 10.4%, with an industrialization rate of less than 36% in 2021. Moreover, the original motivation of some enterprises applying for green patents related to environmental innovation, such as renewable energy technologies and electric vehicles, may be simply to “deceive” fiscal green subsidies and tax incentives, or simply to engage in “green-washing” behavior to cater to external investors.

On the other hand, it may also be related to the energy rebound effect brought by technological progress. As of the end of 2022, for example, the breakthrough innovations in electric vehicles and renewable energy related technologies have led to a market penetration rate of over 40% for new energy vehicles in China, with a total ownership of 13.1 million new energy vehicles, increasing by 67.13% compared to the previous year⁵⁷. The rapid growth of electric vehicles inevitably leads to a significant increase in additional electricity demand. In 2021, national electricity consumption increased by 10.3% compared to the previous year, representing a growth rate increase of 7.1 percentage points⁵⁸. However, coal-fired power remains the main source of electricity supply in China (over 60%), thus the replacing traditional fuel vehicles with electric vehicles is actually shifting carbon emissions from the transportation sector to the electricity generation and supply sectors.

Conclusion and policy implications

Faced with the increasingly urgent global climate challenge, considerable efforts have been made to explore the role of green finance in promoting low-carbon development, while knowledge is rather limited regarding the underlying mechanisms, especially in developing countries. This article therefore addresses this void by constructing a city-level comprehensive index of green finance development from seven dimensions, and identifying the main pathways through which green finance can facilitate the low-carbon development of cities in China.

Using a balanced panel data covering 277 Chinese cities from 2010 to 2020, our typical fact analysis first reveals that China’s green finance development exhibits an overall spatial differentiation of ‘high in the east and low in the west’, while the carbon intensity level displays an overall spatial differentiation of ‘high in the north and low in the south’. Second, the baseline FE and GMM estimators indicate that the development of green finance significantly decreases the level of carbon intensity in cities. This finding remains robust when using GFRIZ pilot programs as policy shocks and using lagged terms as IVs. Third, green finance in cities with a higher level of carbon intensity is more likely to have a larger reduction effect on carbon intensity, whereas in cities with higher and lower levels of green finance, the impact remains significant. This finding indicates the positive role of green finance in the low-carbon development of cities varies with the level of carbon intensity, but not with the level of green finance development. Finally, further mechanism analysis suggests that green finance primarily promotes low-carbon development in cities by adjusting industrial structure rather than stimulating technological innovation. Despite we have also proved that green finance development enhances green innovation, the technology progress it brings may not always translate into a real improvement in green productivity due to multifaceted factors.

Our findings offer several implications for policymakers, financial institutions, and other stakeholders seeking to integrate green finance into China’s low-carbon development planning.

• For policymakers: From the perspective of top-level design, utilizing the resource allocation function of finance to forcibly restrict the flow of financial resources to energy intensive industries can indeed achieve the goal of rapid reduction in carbon emissions, which would inevitably come at the expense of economic expansion. In addition, the space for adjusting the industrial structure of a country or city cannot be infinite, so this strategy is destined to be a short-term choice. For many developed countries, where economic development remains their currently top-priority, they should adhere to synchronous promotion of industrial structure adjustment and technological innovation. In the long-run, whether developing or developed countries, the future policy objective of green finance should aim at enhancing a sustainable technological progress.

• For financial institutions: Institutions should carry out continuous green financial products and services innovation over a longer horizon, and allocate greater financial resources towards the fields of green projects. Even though it may not bring about enhancement of green innovation immediately, as Kudratova et al.⁵⁹ argued that, the profit on green projects that benefit sustainable development is generally higher than that of traditional projects.

• For regulators: Patent authorization and trading departments should strengthen the risk assessment of green patent value to prevent the adverse selection of low-quality innovation driving out high-quality innovation, and thus making more financial support for green projects and their innovators who are truly committed to promoting environmental sustainability.

• For corporate sector: First, high-quality innovation corporate should actively strengthens information disclosure related to patent quality to distinguish them from low-quality competitors and attract more external investors to promote the progress of green technologies. Second, high-pollution corporate can also set up special funding or green corporate bonds for green technologies to enhance their sustainable competitiveness.

Although the indications from our study recommend fruitful policies related to improvement of green finance markets in developing countries, there is a clearly need for future research on how green finance promotes the low-carbon transition of cites in OECD and other nations, through stimulating technological innovation or adjusting industrial structure. Comparing the differences in the impact pathways of green finance between developed and developing economies would enrich policy insights. Moreover, consideration of other mediating variables such as the quality of green innovation and patent information disclosure is highly suggested for further studies.