The Impact of Digital Finance on Urban and Rural Household Carbon Emissions: Evidence from China

Abstract

The complex interplay between digital finance (DF) and household carbon emissions (HCEs) represents a critical subsystem within the broader socioeconomic–ecological system driving climate change. This paper presents estimates of HCEs based on panel data for 30 Chinese provinces from 2011 to 2021 and examines the effects and mechanisms of DF on HCEs in urban and rural regions. The results indicate that (1) DF has a negative impact on urban HCEs, while, conversely, it exacerbates HCEs in rural regions; (2) based on the heterogeneity analysis, the impact of DF is primarily driven by its coverage, with the most significant effects seen in eastern China; and (3) two transmission channels are operative: an energy consumption scale effect and an energy consumption composition effect. Further analysis suggests that government expenditure on energy conservation and environmental protection, as well as financial regulation, play moderating roles in these channels. These findings provide new insights into efforts to achieve carbon neutrality in China and offer new perspectives on the role of financial technologies in shaping environmental outcomes within complex socio-technical systems.

1. Introduction

Against the backdrop of geopolitical tensions, frequent public health emergencies, the intensifying greenhouse effect, and extreme weather events, the world faces a severe energy crisis [1]. Carbon emission performance has emerged as a central issue in global governance. In recent years, according to the International Energy Agency (IEA), global CO₂ emissions have continued to rise, hitting 37.4 billion tons in 2023, an increase of approximately 12.8% compared with 2018. China, as the most populous nation with the second-largest economy, has undergone rapid industrialization and urbanization over the past few decades, resulting in a substantial increase in energy consumption. As a result, China has been the largest global carbon emitter since 2006 [2]. In response, President Xi Jinping declared at the 75th Session of the UN General Assembly that China is committed to hitting peak CO₂ emissions before 2030 and achieving carbon neutrality before 2060.

As significant contributors to a nation’s total consumption, households play a vital role in boosting domestic demand and driving economic growth [3]. Thus, rising living standards, increasing household carbon consumption, and the upgraded consumption structure—along with China’s large population—have caused household carbon emissions (HCEs) to surpass those of the industrial sector. This makes households a key source of carbon emissions [4]. Therefore, effectively managing HCEs is critical in enabling not only China but all countries to achieve their low-carbon goals as well as in establishing a sustainable, green, and circular global economic system. Besides government intervention, the structural optimization of energy consumption, socioeconomic transformation, and innovations in the financial industry are also necessary in this regard. In the era of the digital economy, this industry has witnessed profound changes in information and communication technologies, resulting in the emergence of digital finance (DF) [5].

DF is an innovative form of financial advancement combining financial services with digital technologies to overcome the bias toward larger institutions and enhance the accessibility, inclusiveness, and convenience of financing. Facilitated by emerging technologies such as big data, blockchain, cloud computing, and artificial intelligence, DF has been reshaping the finance industry and providing a new approach to financial inclusion. As it becomes the general direction of the industry, DF is transcending the geographical limitations of traditional financial services, lowering the financing costs for small- and medium-sized enterprises (SMEs) and individuals [6,7] and increasing the efficiency of financial resource allocation [8]. Thus, DF provides new tools for and insights into reductions in carbon footprints. Existing research related to this subject falls into three categories. The first category investigates the determinants of HCEs. Scholars have identified a set of key factors that can be categorized as demographic characteristics [9]: household income level and consumption patterns [10], energy intensity and structure [11], and environmental consciousness [12]. Meanwhile, as the lifeblood of economic activities, the financial sector also has significant impacts on carbon emissions [13,14,15,16].

The second category examines the impacts of DF on socioeconomic development. Prior research has highlighted the beneficial impact of DF on regional economic growth [17], a phenomenon that Liu et al. [18] attribute to the promotion of entrepreneurship among SMEs and the stimulation of consumption expenditure. Second, DF can broaden financing channels and the financing scale, thereby easing the financing constraints on SMEs [19], alleviating information asymmetry between banks and enterprises [20], and driving digital transformation [21]. Li et al. [20] utilized panel data of A-share listed enterprises to perform an empirical examination and found that DF significantly enhances the efficiency of capital allocation and the total factor productivity within these companies while reducing credit discrimination against private SMEs. Third, DF enhances access to affordable financial products and services for individuals and populations excluded from traditional financial institutions, thereby positively impacting employment and entrepreneurship [22,23], household income and consumption levels [24,25,26,27], and poverty reduction efforts [28].

The third category establishes a strong link between DF and carbon emissions [29,30]. However, scholars have primarily focused on the production sector and identifying the effects of DF on carbon emissions at the national or regional level. Among the studies comparing the impacts of DF on carbon emissions across countries was that of Khan et al. [29], who utilized the Global Findex database to construct the comprehensive Digital Financial Inclusion Index. They found a strong positive relationship between DF and carbon emissions in 76 emerging economies. Le et al. [31] reached a similar conclusion in their study of 31 Asian countries. Numerous studies have analyzed the impacts of DF on carbon emissions at the provincial and municipal levels in China. Some scholars find that DF increases carbon emissions through mechanisms such as accelerating economic growth [32] or exhibiting an inverted U-shaped relationship [33]. However, most scholars conclude that DF suppresses regional carbon emissions [34,35].

The digitization of finance not only empowers production, but also significantly impacts household lifestyles and consumption patterns. In terms of its digital aspect, mobile payment systems help DF overcome the barriers to transactions imposed by time and space, thereby reducing payment costs and thresholds and, in turn, generate greater consumption demand and enhanced service experiences. In addition, the inclusive nature of DF can alleviate consumption constraints so as to increase the consumption capacity of households. Thus, the impact of DF on HCEs has recently attracted considerable attention from scholars. From the perspective of the household consumption scale and consumption patterns, Qin et al. [36] found that DF can stimulate consumption-based HCEs, since the overall scale effect prevails over the composition effect. Similarly, Pu and Fei [37] found that DF exacerbates city-level residential carbon emissions mainly through changing electricity consumption and transportation in China.

Given the urban–rural inequality and other regional disparities in China [3], consumption patterns and, therefore, carbon emissions may differ between urban and rural households. However, this issue has received limited attention in the existing literature. Zhang et al. [4] identified income growth as the driving factor of urban–rural differences in HCEs but failed to establish the link between DF and HCEs in urban and rural areas. Although Zhou et al. [38] found that the mitigating effect of DF on HCEs is only strongly observed in urban areas, their study did not thoroughly examine the impacts. To address these gaps in the literature, this paper empirically examines the impact of DF on HCEs in urban and rural China, and the mechanisms at work, using panel data for 30 Chinese provinces from 2011 to 2021.

The contributions of this paper are threefold. First, most studies address the impact of DF on carbon emissions at the national or industry level, while this paper focuses on the micro level—specifically, households—and the role of financial innovation in this sector. The integration of DF and HCEs within a single framework can not only enhance policymakers’ understanding of the potential benefits of DF but also highlight the limitations of relying exclusively on production-side strategies, which may hinder the synergistic realization of decarbonization and economic growth. Second, our systematic exploration of the heterogeneity of DF’s impact on HCEs across regions and urban and rural contexts provides crucial insights into the design of targeted and effective emission reduction measures and can enhance the ability of the government to assess the environmental impacts of financial development accurately. Third, the two mechanisms identified here through which DF influences HCEs—the energy consumption scale effect and the composition effect—inform our understanding of the differences between urban and rural households. This understanding has significant policy implications.

The remainder of this study is structured as follows. Section 2 introduces the theoretical analysis, and describes the empirical model, the data, and the methodology in detail. Section 3 presents the empirical results and analysis. Section 4 explains the transmission mechanisms. Section 5 presents the conclusions and the policy implications.

2. Research Design and Methodology

2.1. Theoretical Framework and Research Hypothesis

The disparity between urban and rural development remains one of China’s greatest modernization challenges. Under the household registration system, urban and rural residents have unequal access to social welfare in areas such as education, healthcare, and housing. The most significant factor in this disparity is the income gap [39], which limits rural residents’ lifestyle choices and results in distinct energy consumption patterns. Because urban households have a higher per capita income, they are more inclined to use clean and high-quality energy sources, whereas rural households tend to rely heavily on cheap coal or free biomass fuels [40]. On the other hand, urban residents tend to engage in more conspicuous consumption, resulting in greater energy use for home appliances [41]. By contrast, rural areas, where survival-oriented consumption dominates, use more energy for cooking because they have lower income elasticity for food demand [42]. The urban–rural divide is also evident in financial literacy and financial development [43]. Given these urban–rural disparities, the impacts of DF on HCEs may differ in urban and rural regions [3]. This analysis is the basis for the following hypothesis:

Hypothesis 1.

The development of DF exerts an asymmetrical impact on urban and rural HCEs.

The inclusive nature of DF contributes to income growth in both urban and rural areas [26,44]. Specifically, DF can encourage agricultural investment and support rural entrepreneurship by providing a range of affordable, secure, and convenient financial products, thereby increasing both wage income and income from self-employment [45]. According to the energy consumption framework introduced by Barnes et al. [46] and Khandker et al. [47], when their basic energy needs are fulfilled, households tend to increase their energy consumption scale and diversify their energy sources as their income increases. Thus, household income affects energy use in terms of both overall energy consumption and the structure of energy sources, with potentially opposite effects on HCEs.

Scale effect. An increase in income often leads to higher energy demand [48], especially in rural households. Because overall rural household income remains relatively low on a national scale, most rural residents still prefer low-cost energy sources, such as coal and firewood [49], which result in higher overall carbon emissions [4]. Second, DF overcomes transaction barriers and influences users’ choice of payment method, thereby motivating the purchase of energy-intensive commodities such as housing and cars [37]. Even in rural areas, rising income levels increase the likelihood of owning power-intensive appliances, such as televisions, air conditioners, and refrigerators. Lastly, by increasing income and alleviating liquidity constraints, DF facilitates changes in household consumption structures by making offline payment options available [24]. Nonetheless, since the per-unit carbon intensity of conspicuous consumption is much higher than that of survival-oriented consumption, a shift to the former type of consumption from the latter ultimately leads to an increase in total carbon emissions.

Composition effect. On the one hand, DF enables more households to afford clean energy through increased access to financial and credit services [24]. Fan et al. [50] concluded that the development of new digital infrastructure promotes renewable energy consumption. Shahbaz et al. [51] also confirmed the crowding-out effect of DF on traditional energy sources, finding that DF helps shift coal consumption toward natural gas. As the share of clean energy sources contributing to the total amount of energy consumed increases [18], HCEs can be effectively curbed. On the other hand, DF can raise public environmental awareness and can offer a variety of environmentally friendly products and services for residents to choose from. The growing internet penetration rates in various regions are increasing residents’ access to policies and information related to eco-friendly practices, and the interplay of environmental news and policies can subtly steer residents toward more sustainable and low-carbon energy consumption habits [52]. Based on these considerations, a second hypothesis was formulated:

Hypothesis 2.

The development of DF can affect HCEs through the energy consumption scale effect and the composition effect.

2.2. Model Design

2.2.1. Baseline Model Design

We used the two-way fixed-effects model in the baseline regression represented by Equation (1) to investigate the impacts of DF on HCEs:

$${\mathit{HCE}}_{\mathit{it}}={\alpha }_0+{\alpha }_1{\mathit{lnDFI}}_{\mathit{it}}+\gamma {\mathit{Controls}}_{\mathit{it}}+{\mu }_i+{\delta }_t+{\epsilon }_{\mathit{it}}$$

(1)

where the dependent variable ${\mathit{HCE}}_{\mathit{it}}$ is the per capita HCEs of province i in year t, which can be subdivided into ${\mathit{HCE}\_\mathit{city}}_{\mathit{it}}$ and ${\mathit{HCE}\_\mathit{rural}}_{\mathit{it}}$ depending on the location. The key explanatory variable, ${\mathit{lnDFI}}_{\mathit{it}}$, is the logarithm of the Digital Financial Inclusion Index of province i in year t, which can be subdivided into coverage breadth, ${\mathit{lnCOV}}_{\mathit{it}}$, depth of use, ${\mathit{lnUSE}}_{\mathit{it}}$, and degree of digitization, ${\mathit{lnDIG}}_{\mathit{it}}$. ${\mathit{Controls}}_{\mathit{it}}$ is the set of control variables, ${\mu }_i$ is the province-level individual fixed effects, ${\delta }_t$ is the time fixed effects, and ${\epsilon }_{\mathit{it}}$ is the random disturbance term.

2.2.2. Model Design for Mechanism Analysis

To examine the mechanisms through which DF impacts urban and rural HCEs, three econometric models were constructed:

$$M_{\mathit{it}}={\beta }_0+{\beta }_1{\mathit{lnDFI}}_{\mathit{it}}+\gamma {\mathit{Controls}}_{\mathit{it}}+{\mu }_i+{\delta }_t+{\epsilon }_{\mathit{it}}$$

(2)

$${\mathit{HCE}}_{\mathit{it}}={\theta }_0+{\theta }_1{\mathit{lnDFI}}_{\mathit{it}}+{\theta }_2{M}_{\mathit{it}}+\gamma {\mathit{Controls}}_{\mathit{it}}+{\mu }_i+{\delta }_t+{\epsilon }_{\mathit{it}}$$

(3)

$${\mathit{HCE}}_{\mathit{it}}={\rho }_0+{\rho }_1{\mathit{lnDFI}}_{\mathit{it}}+{\rho }_2{\mathit{lnDFI}}_{\mathit{it}}\times X_{\mathit{it}}+{\rho }_3{X}_{\mathit{it}}+\gamma {\mathit{Controls}}_{\mathit{it}}+{\mu }_i+{\delta }_t+{\epsilon }_{\mathit{it}}$$

(4)

In Equations (2)–(4), $M_{\mathit{it}}$ indicates the mediators of DF on HCEs, including the scale effect and the composition effect. $X_{\mathit{it}}$ denotes the moderating variables, including the ratio of energy conservation and environmental protection expenditure to government general expenditure, ${\mathit{GER}}_{\mathit{it}}$, and the ratio of financial regulation expenditure to government general expenditure, ${\mathit{FRER}}_{\mathit{it}}$. Equation (2) tests the causal effect of DF on the scale and composition of household energy consumption. Equation (3) tests the mediating role of the energy consumption scale and composition. Equation (4) tests the moderating effect of the share of energy conservation and environmental expenditure and financial regulation expenditure on the causal relationship between DF and HCEs.

2.3. Variable Measurement and Data Sources

In this study, we utilized balanced panel data spanning the period 2011–2021 for 30 provinces in China. Owing to data limitations, Hong Kong, Macau, Taiwan, and Tibet were excluded from the dataset. The data come from the EPS database, the “China Energy Statistical Yearbook (2012–2022)”, “China Statistical Yearbook (2012–2022)”, and the “Peking University Digital Financial Inclusion Index of China (2011–2021)”. To mitigate heteroskedasticity, some of the variables were log-transformed.

2.3.1. Dependent Variable: Per Capita HCEs

In the literature, HCEs are typically categorized as either direct or indirect. Direct emissions refer to the amount of CO₂ released from the use of coal, oil, natural gas, electricity, and biofuel for energy generation. Indirect emissions are generated during the manufacturing and use of the goods and services consumed by households. Although some studies in the literature use the sum of direct and indirect emissions for HCEs as their dependent variable [36], the goals of this study were similar to those of an earlier study by Wu et al. [53]. Thus, we focused mainly on HCEs generated from urban and rural residential daily activities, especially from energy consumption. Following the approach of Xu et al. [54] and Zhang et al. [4], we applied the carbon emission coefficient method to measure the total emissions from the primary energy and electricity consumed by urban and rural households, denoted as ${\mathit{HCE}\_\mathit{direct}}_{\mathit{it}}$. The sum of direct HCEs was then divided by the urban and rural populations to obtain the per capita HCEs, denoted as ${\mathit{HCE}}_{\mathit{it}}$. The calculation was carried out as follows:

$${\mathit{HCE}\_\mathit{direct}}_{\mathit{it}}={\mathit{HCE}}_{F,\mathit{it}}+{\mathit{HCE}}_{E,\mathit{it}}$$

(5)

where ${\mathit{HCE}\_\mathit{direct}}_{\mathit{it}}$ represents the direct HCEs produced in urban and rural households, and ${\mathit{HCE}}_F$ and ${\mathit{HCE}}_E$ represent the carbon emissions from fossil fuels and electricity, respectively.

$${\mathit{HCE}}_F=\sum _j{FC}_j\times {\mathit{NCV}}_j\times {\mathit{CC}}_j\times {\mathit{OF}}_j\times 44/12$$

(6)

$${\mathit{HCE}}_E=\mathit{EC}\times {\mathit{CEC}}_E$$

(7)

In this paper, fossil energy is defined as energy produced from coal, oil, and natural gas. ${FC}_j$ and $\mathit{EC}$ represent the household consumption of the $j$th type of fossil energy and electricity, respectively. ${\mathit{NCV}}_j$, ${\mathit{CC}}_j$, and ${\mathit{OF}}_j$ are the net calorific value, carbon content per unit calorific value, and carbon oxidation rate of the $j$th type of fossil energy, respectively. Factor 44/12 is the ratio of the molecular weight of carbon dioxide to the atomic weight of carbon, and ${\mathit{CEC}}_E$ is the carbon emission coefficient of electricity.

$${\mathit{HCE}}_{\mathit{it}}={\mathit{HCE}\_\mathit{direct}}_{\mathit{it}}/{\mathit{POP}}_{\mathit{it}}$$

(8)

Next, the direct HCEs were divided by the urban and rural populations, ${\mathit{POP}}_{\mathit{it}}$, to obtain the per capita carbon emissions, ${\mathit{HCE}}_{\mathit{it}}$.

2.3.2. Core Explanatory Variable: DF

The Digital Financial Inclusion Index (2011–2021) developed by the Institute of Digital Finance at Peking University [21] serves as a proxy for the development status of DF. The index is composed of three key dimensions, coverage breadth (${\mathit{COV}}_{\mathit{it}}$), depth of use (${\mathit{USE}}_{\mathit{it}})$, and degree of digitization (${\mathit{DIG}}_{\mathit{it}}$), and is built from a comprehensive set of 33 sub-indicators. The coverage breadth reflects the extent of financial accessibility, focusing on the coverage rate of online payment accounts; the depth of use captures the variety of financial products, including payments, money market funds, credit, insurance, investments, and loans; and the degree of digitization captures the cost and reliability of these financial services.

2.3.3. Impact Mechanisms and Instrumental Variables

Mediating variables. This research aimed to assess the impact, if any, of DF on HCEs through two core channels: the scale effect and the composition effect. The former effect refers to the total energy consumed by households, which is measured by converting the quantity of total fossil fuels and electricity used in households into tons of standard coal equivalent (TCE). The composition effect refers to the share of clean energy used by households. The method adopted by Zhou et al. [38], where the percentage of natural gas in the total household energy consumption is used as a proxy, is employed here.

Moderating variables. Previous research has demonstrated that government expenditure on energy conservation and environmental protection [55], as well as on financial regulations [56], can influence the effectiveness of carbon reductions. Therefore, the ratios of energy conservation and environmental protection expenditure, and the financial regulation expenditure, to the general government expenditure serve as the moderating variables for further exploring the transmission channels.

Instrumental variables. To address the potential endogeneity issues, a lagged explanatory variable and the internet broadband penetration (IBP) rate are incorporated into the model as instrumental variables. IBP is commonly defined as the number of internet access ports per capita and, since it comprehensively reflects the regional development of digital public infrastructure while having a negligible influence on HCEs, it is a suitable instrumental variable.

2.3.4. Control Variables

Since many other factors affect carbon emissions apart from DF, to eliminate the risk of the omitted variable bias, the control variables for this study encompass individual characteristics, government influences, and regional macroeconomic factors. These variables include: ① Government Intervention ($\mathit{GOV}$). The central government holds the authority and primary responsibility for fiscal spending related to low-carbon governance, while local environmental fiscal expenditure can, in turn, influence the outcomes of carbon emission reductions [57]. ② Environmental Regulation ($\mathit{ENV}$). To advance environmental sustainability, fiscal and industrial policies could be implemented to incentivize producers to adopt eco-friendly technologies while steering consumer spending toward energy-efficient and low-carbon choices, both of which contribute to reducing carbon emissions [58]. ③ Population Aging ($\mathit{AGE})$. Population aging affects carbon emissions mainly through income level, energy demands, and consumption behaviors [9]. ④ Regional Economic Development ($\mathit{RGDP})$. Increasing economic scale necessitates the input of energy and other resources, which often results in a rise in carbon emissions [59]. ⑤ Industrial Structure ($\mathit{SI})$. Compared with traditional energy- and labor-intensive industries, which generate high levels of pollution, knowledge- and technology-driven industries, with their high-value-added potential, have substantially contributed to energy conservation and emission reduction [60].

The definitions, measurements, and descriptive statistics for the variables are specified in

Table 1. Statistical description of key variables.

Variable	Definition	Obs	Mean	SD	Min	Max
HCE_city	Per capita urban HCEs	330	0.499	0.194	0.230	1.892
HCE_rural	Per capita rural HCEs	330	0.603	0.313	0.119	2.164
lnDFI	Logarithm of Digital Financial Inclusion Index	330	5.283	0.669	2.909	6.129
lnCOV	Logarithm of the coverage of DF	330	5.149	0.817	0.673	6.072
lnUSE	Logarithm of the usage penetration of DF	330	5.266	0.652	1.911	6.236
lnDIG	Logarithm of the digitization level of DF	330	5.556	0.681	2.026	6.136
lnCEC	Logarithm of the total energy consumed by urban households	330	6.155	0.792	3.757	7.733
lnREC	Logarithm of the total energy consumed by rural households	330	5.754	0.858	3.163	7.565
CG	Natural gas consumption by urban households as a share of total energy consumption	330	0.224	0.143	0.000	0.661
RG	Natural gas consumption by rural households as a share of total energy consumption	330	0.036	0.065	0.000	0.388
GER	Ratio of energy conservation and environmental protection expenditure to government general expenditure	330	0.030	0.009	0.012	0.068
FRER	Ratio of financial regulation expenditure to government general expenditure	330	0.003	0.003	0.000	0.019
IBP	The number of internet access ports per capita	330	0.475	0.230	0.096	1.074
GOV	Ratio of government expenditure to GDP	330	0.249	0.103	0.107	0.643
ENV	Ratio of investment in industrial pollution to industrial value added	330	0.298	0.295	0.009	2.804
AGE	Population aged 65 and above as a percentage of the total population	330	0.231	0.438	0.046	2.029
lnRGDP	Logarithm of regional GDP per capita	330	10.88	0.444	9.706	12.12
SI	Value added of secondary industry as a share of domestic GDP	330	0.427	0.088	0.158	0.590

.

2.4. Overview of Urban and Rural HCEs and the Development of DF in China

During the period studied, the total HCEs in both urban and rural areas steadily increased, except for a brief decline in 2013. Although rural HCEs remained lower in absolute terms than urban HCEs, their growth rate was slightly higher (Figure 1). By 2021, the national, urban per capita, and rural per capita HCEs were 0.53 tons, 0.49 tons, and 0.60 tons, respectively. In 2013, the national and urban per capita HCEs dropped notably, after which they resumed a gradual upward trend. In contrast, rural per capita HCEs maintained a higher growth rate than those in urban areas, with a sharp increase observed in 2019, outpacing the increase in total emissions. Some researchers attribute this trend to the different income growth rates observed between urban and rural populations over time [4], as well as the significant population increase [53].

Figure 2 documents the difference in descriptive statistics of urban and rural per capita HCEs. Since 2012, the average rural per capita HCEs have remained consistently higher than urban per capita HCEs, and this disparity has widened progressively. The standard deviation (SD) of rural per capita HCEs has also remained consistently higher than that of urban households, though fluctuations in both have been modest. On the other hand, the coefficient of variation (CV) of urban per capita HCEs has stabilized after a sharp drop in 2013, whereas the CV for rural household per capita emissions has steadily declined over the study period.

The secular trend of HCEs, as depicted in Figure 1, and their fluctuations, as illustrated in Figure 2, exhibit a consistent and seemingly synchronized movement throughout the sampling period. Nonetheless, Figure 3 further illustrates the correlation between provincial-level DF and weighted per capita HCEs. It reveals that DF is negatively correlated with per capita HCEs in urban areas but positively correlated with those in rural areas, underscoring the heterogeneity in the cross-sectional data.

Our analysis revealed significant cross-sectional differences among our sample, which corroborates our earlier hypothesis regarding urban–rural disparities. Consequently, in the subsequent sections we conduct further empirical analysis on the potential urban–rural differences.

3. Empirical Results and Discussion

3.1. Baseline Estimation Analysis

Table 2. Baseline regressions.

	(1)	(2)	(3)	(4)
	HCE_City	HCE_Rural	HCE_City	HCE_Rural
lnDFI	−0.187 *** (0.068)	0.216 *** (0.078)	−0.234 *** (0.068)	0.275 *** (0.079)
GOV			−0.292 (0.457)	0.231 (0.525)
ENV			−0.030 (0.037)	−0.011 (0.043)
AGE			−0.045 (0.090)	0.065 (0.103)
lnRGDP			0.298 ** (0.119)	−0.387 *** (0.137)
SI			−0.494 (0.371)	1.073 ** (0.427)
_cons			−1.449 (1.245)	3.003 ** (1.430)
Province FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
N	330	330	330	330
R2	0.065	0.252	0.123	0.299

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

reports the baseline estimation results of Equation (1) after incorporating all the control variables. As shown in column (1) in Table 2, the regression coefficient of DF on per capita urban HCEs was −0.234, which was significantly negative at the 1% level. The coefficient for lnDFI implies that a 1% increase in DF is associated with a decrease of 0.234 units in per capita urban HCEs, with all other factors remaining constant. The arithmetic mean of HCE_city was 0.5, while that of DFI was 231.47. This suggests that a 1% increase in DFI would result in a decrease of approximately 108.5% in HCE_city compared to its mean, indicating a fairly strong negative relationship. Thus, the development of DF may lead to substantially reduced carbon emissions in urban households. Turning to column (2), the regression coefficient of DF on per capita rural HCEs was significantly positive, with an effect size of 0.275 at the 1% level. From this, we can infer that a 1% increase in DF would result in 0.275 incremental units in per capita rural HCEs. The arithmetic mean of HCE_rural was 0.6; thus, a 1% increase in DFI would result in an increase of approximately 106.1% in HCE_rural compared to its mean. This result indicates that the development of DF has exacerbated HCEs in rural areas. These results support Hypothesis 1.

Closely related to this study, Zhou et al. [38] confirmed the mitigation effect of DF on urban HCEs, but noted that they did not impose significant effects on rural HCEs. Nonetheless, their study did not provide an examination of the potential impacting mechanism that causes this disparity. The environmental Kuznets curve suggests that environmental degradation initially worsens with rising income but that this trend reverses when a certain income level is reached. Accordingly, DF may encourage urban residents to shift toward using cleaner energy, thereby reducing per capita carbon emissions through increased energy efficiency. However, in rural areas, DF has primarily expanded access to financial services and products, significantly increasing rural residents’ income and consumption levels, energy use, and, in turn, carbon emissions. These transmission channels will be tested in Section 4.

According to the regression results of control variables, the regression coefficient of $ln\mathit{RGDP}$ on per capita urban HCEs was significantly positive, indicating that economic growth spurs urban production activities, thereby increasing carbon emissions [61]. Conversely, regional economic development appears to help mitigate per capita rural HCEs. Improving the rural economy supports the transformation of rural production, raising production efficiency and reducing excess carbon emissions. Meanwhile, under President Xi Jinping’s rural vitalization strategy and eco-friendly developmental policy, high-pollution industries, such as mining, have been progressively terminated in rural areas. The regression coefficient of $SI$ on per capita rural HCEs suggests a significantly positive effect, in line with our expectations. Traditional secondary industries, including manufacturing and construction, heavily rely on fossil fuels, which in turn generates extensive environmental pollutants [62]. The regression coefficients of $GOV$, $ENV$, and $AGE$ were not significant, indicating that government intervention, environmental regulation, and population aging were not the main drivers of urban and rural HCEs during the period studied.

3.2. Heterogeneity Analysis

3.2.1. Estimation by Sub-Indicators of DF

As discussed in Section 2.3.2, the Digital Financial Inclusion Index consists of three sub-dimensional indicators, each reflecting distinct features of DF. To more comprehensively understand the effect of these second-level indices on per capita HCEs, each of the sub-indicators, ${\mathit{COV}}_{\mathit{it}}$, ${\mathit{USE}}_{\mathit{it}}$, and ${\mathit{DIG}}_{\mathit{it}}$, was substituted for the total index ${\mathit{DFI}}_{\mathit{it}}$ in Equation (1). The regression results are presented in

Table 3. Results of the estimations of the sub-indicators.

	(1)	(2)	(3)	(4)	(5)	(6)
	HCE_City	HCE_Rural	HCE_City	HCE_Rural	HCE_City	HCE_Rural
lnCOV	−0.047 * (0.028)	0.101 *** (0.032)
lnUSE			−0.084 (0.052)	−0.002 (0.060)
lnDIG					0.073 (0.045)	−0.081 (0.052)
_cons	−1.889 (1.273)	3.771 *** (1.446)	−1.334 (1.277)	3.225 ** (1.475)	−1.959 (1.279)	3.581 ** (1.472)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
N	330	330	330	330	330	330
R2	0.096	0.294	0.096	0.269	0.096	0.275

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

. The regression coefficients in columns (1) and (2) are −0.047 and 0.101, respectively, indicating that while the breadth of DF coverage has a significantly negative impact on urban HCEs, it is positively associated with rural HCEs. These results are consistent with the baseline regression results. However, the depth and digitization level of DF use show no statistically significant impact.

These results indicate that the coverage rate of online payment accounts has the most pronounced impact on HCEs, which is consistent with the findings of Pu and Fei [37]. However, the diversified use of DF and the level of digitization did not affect HCEs during the study period. Unlike traditional financial services, DF relies heavily on internet-based operations, so the breadth of the coverage of digital accounts correlates with the effectiveness with which services can be delivered. In other words, coverage breadth is the cornerstone of the depth and digitization level of DF use.

3.2.2. Estimation by Region

China’s vast territory and diverse resource distribution have resulted in significant regional disparities in economic development, income levels, and cultural customs. These disparities are also evident in the uneven development of DF. To explore the regional heterogeneity in the impact of DF on HCEs, the main sample was divided into three sub-samples representing the country’s eastern, central, and western regions in accordance with the classification standard of the National Bureau of Statistics of China.

Table 4. Results of the regional heterogeneity analysis.

	Eastern		Central		Western
	(1)	(2)	(3)	(4)	(5)	(6)
	HCE_City	HCE_Rural	HCE_City	HCE_Rural	HCE_City	HCE_Rural
lnDFI	−0.454 *** (0.071)	1.054 *** (0.251)	−0.001 (0.215)	−0.016 (0.277)	0.076 (0.195)	0.132 (0.119)
_cons	5.514 *** (0.978)	1.185 (3.446)	−5.531 ** (2.497)	−5.308 (3.214)	−11.789 *** (3.419)	1.021 (2.083)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
N	121	121	88	88	121	121
R2	0.595	0.379	0.475	0.510	0.226	0.551

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

presents the estimated results for these three regions.

As columns (1)–(2) in Table 4 show, DF significantly affected per capita HCEs in the eastern region but not in the central and western regions. This finding is consistent with that of Huang et al. [59]. A possible explanation for this finding is that the eastern region is more urbanized and economically developed and, as a result, has more advanced digital infrastructure and higher-quality financial services. The residents of the eastern region, therefore, are likely to use DF instruments more frequently and consistently, with a more noticeable impact on HCEs. By contrast, in the central and western regions, where DF is less developed, DF instruments have not yet exhibited a significant impact on HCEs.

3.3. Robustness Tests

We conducted three robustness checks to validate the reliability of the estimated results. First, to address the potential endogeneity issue in the baseline model, the first-order lags of DF and IBP were used as instrumental variables in the IV-2SLS regression model. The results, shown in columns (1) and (2) in

Table 5. Estimated results of the robustness tests.

	IV-2SLS Regression		Replacing Variables		Winsorize 1%
	(1)	(2)	(3)	(4)	(5)	(6)
	HCE_City 2nd-Stage	HCE_Rural 2nd-Stage	HCE_City	HCE_Rural	HCE_City	HCE_Rural
lnDFI	−0.541 *** (0.154)	0.671 *** (0.195)			−0.252 *** (0.071)	0.305 *** (0.081)
L.lnDFI			−0.174 *** (0.057)	0.240 *** (0.073)
_cons	2.111 * (1.276)	2.059 (1.616)	−0.223 (1.139)	4.285 *** (1.442)	−1.391 (1.243)	2.927 ** (1.427)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
F-statistic value	397.999	397.999
N	330	330	300	300	330	330
R2			0.118	0.336	0.127	0.304

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

, remained unchanged, thereby confirming the reliability of the findings. The second approach involved replacing the core explanatory variable with the first lag term of DF, and the results, shown in columns (3) and (4), align with the baseline results, further confirming the robustness of the baseline model. Lastly, to minimize the impact of extreme values on the accuracy of the sample, the data were trimmed and winsorized at the 1% level, and the results, presented in columns (5) and (6), again coincided with previous results, indicating that the findings of this study are robust and credible.

4. Mechanism Analysis

4.1. Mediating Effect of Energy Consumption Scale and Composition

This section describes the theoretical analysis of Hypothesis 2, which guides the investigation of the mechanisms through which DF affects HCEs in urban and rural regions. The sequential mediation model specified in Equations (1)–(3) allowed us to explore the mediating role of the total energy consumed by households and the share of natural gas consumed by households.

Scale effect. The logarithms of total energy consumption in urban households, ${\mathit{ln}\mathit{CEC}}_{\mathit{it}}$, and rural households, ${\mathit{ln}\mathit{REC}}_{\mathit{it}}$, were used as the mediating variables in testing the potential transmission channel of the scale effect.

Table 6. Estimated results of the impact mechanism: the scale effect.

	City Consumption Scale Effect			Rural Consumption Scale Effect
	(1) Baseline	(2)	(3)	(4) Baseline	(5)	(6)
	HCE_City	lnCEC	HCE_City	HCE_Rural	lnREC	HCE_Rural
lnDFI	−0.234 *** (0.068)	−0.093 (0.088)	−0.171 *** (0.035)	0.275 *** (0.079)	0.176 * (0.096)	0.156 *** (0.044)
lnCEC			0.671 *** (0.234)
lnREC						0.676 *** (0.044)
_cons	−1.449 (1.245)	1.436 (1.601)	−2.412 *** (0.632)	3.003 ** (1.430)	5.200 *** (1.751)	−0.512 (0.816)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
N	330	330	330	330	330	330
R2	0.123	0.511	0.776	0.299	0.325	0.779

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

reports the mechanism test results. Columns (1) and (4) display the baseline regression results of Equation (1); columns (2) and (5) display the regression results of Equation (2); and columns (3) and (6) display the regression results of Equation (3). As columns (2) and (5) show, the estimated coefficient of the mediator was not statistically significant for urban households but was significantly positive at the 10% level for rural households. This result suggests that, for urban households, total energy consumption is not a mediating variable, but for rural households, DF promotes total energy consumption. Furthermore, column (6) in Table 6 shows that total energy consumption in rural households had a significant positive impact on rural HCEs at the 1% level. Even after including the mediating variable, the coefficient of the impact of DF on per capita rural household emissions remained significantly positive at the 1% level, with an effect size of 0.156. The finding that this coefficient is less than the regression coefficient in column (4) suggests a partial mediating effect. Thus, DF increases per capita rural HCEs by increasing the overall energy consumption.

According to Le et al. [31] and Zhao and Yang [63], the development of DF can accelerate the use of fossil fuels, thus leading to an increase in carbon emissions. Furthermore, earlier studies by Wu et al. [53] and Balezentis [64] found that substantial gaps in energy consumption between urban and rural regions have long existed, largely due to differences in economic development stages, cultural practices, and household behaviors. We further conclude based on the results of this study that the development of DF stimulates energy consumption in rural areas, potentially widening this disparity.

Composition effect. Employing the same analytical framework used for the scale effect, the shares of natural gas consumption in urban households, ${\mathit{CG}}_{\mathit{it}}$, and in rural households, ${\mathit{RG}}_{\mathit{it}}$, were used to test the potential impact mechanism of the composition effect. The empirical results are displayed in

Table 7. Estimated results of the impact mechanism: the composition effect.

	City Consumption Composition Effect			Rural Consumption Composition Effect
	(1) Baseline	(2)	(3)	(4) Baseline	(5)	(6)
	HCE_City	CG	HCE_City	HCE_Rural	RG	HCE_Rural
lnDFI	−0.234 *** (0.068)	0.088 *** (0.030)	−0.186 *** (0.068)	0.275 *** (0.079)	−0.012 (0.024)	0.266 *** (0.077)
CG			−0.544 *** (0.132)
RG						−0.720 *** (0.188)
_cons	−1.449 (1.245)	0.068 (0.543)	−1.412 (1.212)	3.003 ** (1.430)	0.596 (0.441)	3.433 (1.402)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
N	330	330	330	330	330	330
R2	0.123	0.312	0.173	0.299	0.271	0.334

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

. The regression results shown in columns (2) and (5) indicate that the share of natural gas consumption in the urban households was significantly and positively associated with DF, while the estimated coefficient was not statistically significant in rural households. Column (3) in Table 7 presents the regression results for the impact of DF on urban HCEs with the mediation of ${\mathit{CG}}_{\mathit{it}}$. As column (3) shows, a significantly negative correlation could be observed between the share of natural gas consumption in urban households and their per capita carbon emissions. In addition, the absolute value of the estimation coefficient of DF was 0.186, which is smaller than the value of 0.234 in column (1) when ${\mathit{CG}}_{\mathit{it}}$ was excluded. These results, taken together, indicate that DF can increase the share of clean energy use in urban households, and in turn reduce urban HCEs, thus supporting Hypothesis 2.

While Shahbaz et al. [51] and Zhou et al. [38] have shown that DF supports the green transition, including the adoption of renewable energy, these results indicate that the composition effect mainly affects urban households, with limited impacts in rural areas. This disparity may reflect the fact that energy consumption in most rural households primarily involves meeting basic subsistence requirements. Thus, while clean energy sources such as liquefied petroleum gas and natural gas have long been available in urban areas, these sources remain scarce in rural regions where the energy infrastructure is underdeveloped.

4.2. Moderating Effects of Government Expenditure on Energy Conservation and Environmental Protection and Financial Regulation

The interaction terms between DF and the relevant moderating variables ${\mathit{GER}}_{\mathit{it}}$ and ${\mathit{FRER}}_{\mathit{it}}$ were used to examine the potential moderating effects through which DF influences urban and rural HCEs.

Table 8. Estimated results of moderating effects.

	Environmental Conservation Expenditure Ratio		Financial Regulation Expenditure Ratio
	(1)	(2)	(3)	(4)
	HCE_city	HCE_rural	HCE_city	HCE_rural
lnDFI	−0.072 (0.091)	0.570 *** (0.102)	−0.307 *** (0.070)	0.328 *** (0.082)
GER	17.562 *** (6.601)	32.133 *** (7.399)
$\mathit{lnDFI}\times$GER	−3.163 *** (1.204)	−6.187 *** (1.349)
FRER			−82.381 *** (25.364)	67.809 ** (29.416)
$\mathit{lnDFI}\times$FRER			13.840 *** (4.578)	−13.184 ** (5.310)
_cons	−1.995 (1.254)	1.851 (1.405)	−1.305 (1.227)	3.314 ** (1.423)
Controls	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
N	330	330	330	330
R2	0.145	0.350	0.164	0.319

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

shows the estimation results of Equation (4). As columns (1) and (2) show, the estimation coefficients of the interaction term ${{\mathit{lnDFI}}_{\mathit{it}}\times \mathit{GER}}_{\mathit{it}}$ on urban and rural HCEs were significantly negative. This result indicates that government spending on energy conservation and environmental protection enhances the effect of DF in reducing urban HCEs while also mitigating the positive effect of DF in promoting rural HCEs. The important role observed for government fiscal expenditure in promoting a green economy and improving carbon reduction efficiency is consistent with the conclusions of Wang and Li [60].

As column (3) shows, the estimation coefficient of the interaction term ${{\mathit{lnDFI}}_{\mathit{it}}\times \mathit{FRER}}_{\mathit{it}}$ on urban HCEs was significantly positive. By contrast, as column (4) shows, the coefficient of the interaction term on rural HCEs was significantly negative. These results imply that government spending on financial regulation can mitigate the increase in rural HCEs resulting from the use of DF but may weaken the effect of DF in reducing the HCEs of urban households, possibly because financial regulation is generally more rigorous and intensive in urban areas than in rural areas. Over-regulation of the financial sector could affect the functioning of the capital market in terms of providing credit support to the energy market, which may lead to resource misallocation and could impair the effectiveness of emission reduction efforts.

5. Conclusions and Implications

Against the backdrop of China’s “dual carbon” goals, the data from 30 Chinese provinces from 2011 to 2021 were used in this study to clarify the relationship between DF and HCEs in urban and rural areas. The main findings suggest that DF reduces per capita HCEs in urban areas but increases per capita HCEs in rural areas. This conclusion remained robust after addressing potential endogeneity issues and conducting a series of robustness tests. Among the three sub-dimensions of DF, coverage breadth was found to be the dominant factor. Regional heterogeneity was observable in the impact of DF on HCEs, being most pronounced in the eastern provinces. The impact mechanism behind the different effects of DF on urban and rural households involves two primary mediating channels: the increase in the proportion of clean energy consumption and the expansion of total energy consumption. The findings also indicate that government fiscal expenditure on energy conservation and environmental protection, as well as financial regulation, can moderate the impact of DF on HCEs.

These empirical results lead us to suggest several policy recommendations. First, policymakers could tailor DF strategies to address the distinct needs of urban and rural regions. Simply put, urban regions have more established and integrated DF ecosystems, while the development of DF in rural regions is still focused on addressing fundamental needs such as broadening financial access, reducing poverty, and unlocking economic potential. Therefore, for rural households, financial services should prioritize essential needs and minimize unnecessary carbon emissions. By contrast, urban households should be guided toward green consumption preferences. Additionally, considering the uneven development of DF across eastern, central, and western China, efforts to promote balanced growth in DF could help to prevent the worsening of the “Matthew effect” caused by the existing digital divide.

Second, the Chinese government could collaborate with large technology firms to leverage the outreach capabilities of digital platforms, raise awareness of the need for carbon neutrality, and promote green and low-carbon consumption in cyberspace. In recent years, personal carbon accounts have emerged as a highly promising global market. For example, Alipay Ant Forest, launched by the Ant Group in 2016, is currently the world’s largest personal carbon account platform. Thus, on the one hand, DF can support the tracking of carbon footprints through online digital carbon accounts that collect data on residents’ energy consumption habits and needs. On the other hand, digital payment platforms such as Alipay and WeChat could encourage residents to live greener lifestyles by issuing energy consumption coupons or introducing rewards for low carbon use when they pay for their household energy online.

Third, in synergistically developing DF and carbon reduction, the government, at both the national and local levels, needs to provide strong policy support for both the digital and green transition at the macro level, while also enhancing household digital and financial literacy at the micro level. These goals can be achieved through subsidies for low-carbon households and those that adopt clean energy, as well as carbon taxes on carbon-intensive products. Additionally, through fiscal means, such as increasing investments in energy conservation, environmental protection, and financial regulation, governments can upgrade and modernize energy and financial infrastructure to promote inclusive and green growth while encouraging households to save energy and cut down on pollution.