Digital Finance, Industrial Structure, and Total Factor Energy Efficiency: A Study on Moderated Mediation Model with Resource Dependence

Abstract

China’s main energy structure is dominated by coal. The burning of coal is a major source of greenhouse gas emissions, making China the largest carbon emitter. Facing double pressure on ecological protection and economic development, improving energy efficiency is more practical than reducing coal utilization. In this context, digital finance can be a vital engine that supports a transition to a low-carbon economy. Based on panel data for 2011 to 2019 of 30 provinces in China, this study probes the effect of digital finance on the total factor energy efficiency and constructs a moderated mediating effect model to analyze the mechanism of action. The results show that: (1) digital finance is able to improve the total factor energy efficiency, (2) the industrial structure plays a mediating effect, which is regionally heterogeneous, and (3) the above transmission path is affected by the degree of regional resource dependence. With the deepening of resource dependence, the role of digital finance in driving energy efficiency through the industrial structure is enhanced. This research demonstrates the effectiveness of digital finance in energy efficiency improvement and develops ideas for ecological governance and sustainable development.

1. Introduction

While industrialization and urbanization have been driving socio-economic development, the world is facing the threat from the scarcity of energy sources and environmental pollution by fossil fuel consumption. China, as the largest energy consumer in the world, whose Environmental Performance Index (EPI) score has increased 8.4% over a decade, ranking 120 in 180 countries in 2020 [1]. The growth of energy consumption in China has accounted for three-quarter of the global growth [2]. With a high sense of responsibility and crisis, the Chinese government is committed to reaching the target of the carbon emission peak by 2030 and carbon neutrality by 2060, which is called the “dual carbon” target [3]. Moreover, China proposed the “Modern Energy System Plan under the 14th Five-Year Plan (2021–2025)”, which laid great stress on green and low-carbon transformation in energy production and consumption patterns by improving energy efficiency [4]. Under the dual constraints of environment and resources, improving total factor energy efficiency is the key to breaking the energy obstacles in the process of realizing “dual carbon” target and green development.

Financial innovation supports a long-term mechanism for energy conservation and promotes green development [5]. With the application data-driven technologies, digital finance has quietly risen and changed the traditional financial industry at an amazing speed in terms of service mode, operational systems, and other aspects. Research suggests that digital finance can provide better financial inclusion for developing countries by enhancing the four basic functions of finance: resource allocation, savings mobilization, risks diversification, stimulation, restraining, and supervision [6,7]. In China, digital finance is growing rapidly. “The 49th Statistical Report on China’s Internet Development” reports that China had 1032 million netizens as of December 2021, an increase of 42.96 million from December 2020, and that the internet penetration had reached 73.0% [8]. The internet access gap between urban and rural areas continued to narrow. These signs indicate that China has greater advantages in developing digital finance. According to current studies, the role of digital finance in the promotion of economic growth is confirmed [1]. Typical studies such as Romer believed that finance could enhance the productivity of the total factor through the development of technology-based theory of endogenous economic growth [9]. Furthermore, many scholars found that digital finance based on artificial intelligence, big data, and technology platforms can effectively reduce financial friction caused by asymmetric information, thus driving total factor productivity [10,11]. However, the deterioration of the ecological environment restricts the progress of economic development. In response to this issue, scholars further explore the function of digital finance in ecological protection. Chang found that digital finance can effectively promote the capital flow to green and high-tech industries by accurately identifying green industries [12]. Moreover, digital finance can guide enterprises to green production and individuals to green investment and consumption. Combined with fin-tech, financial business is becoming more sustainable by promoting green finance [13]. However, some scholars concluded that digital finance causes a rise in energy consumption via a rebound effect and contributes to environmental degradation [14,15]. Few studies focused on the influence of digital finance on energy efficiency from threshold effects and spatial spillover, and found that digital finance works differently in different regions in China. All the efforts mentioned above have explored how energy efficiency is associated with digital finance. However, there are few studies on the analysis of influence mechanism from the perspective of transmission path, interfering factors, and regional heterogeneity. This research focuses on the following questions. What is the effect of digital finance on total factor energy efficiency and the mechanism of action from the perspective of the industrial structure? Is there a regional heterogeneous mechanism of digital finance on total factor energy efficiency in China? There is also the question of whether this action mechanism is interfered by the degree of resources dependence as the large differences in resource endowments in China’s different regions. These questions will be answered in this article.

The contributions of this research are as follows. First, this research adopts the total factor energy efficiency method and incorporate carbon emissions into non-desired outputs, which enriches the research on energy efficiency measurement. Second, digital technology, industrial structure, and resources endowment are included in the energy efficiency analysis framework, and a moderated mediation effect model is built to explore the complex mechanism with the moderation of resource dependence, which expands the breadth and depth of the study on energy efficiency improvement. Finally, this research develops ideas for coal pollution reduction and provides a decision-making basis for ecological governance, which is of practical significance.

2. Literature Review

The relationship between financial systems and green development is one of the hot spots that has attracted widespread attention. Before that, many scholars have affirmed the contribution of finance to economic growth based on the “theory of endogenous growth” [9].

However, the development of economies is usually accompanied by environmental degradation. Therefore, studies have emerged on the influence mechanism of the level of financial development on environmental pollution and have emerged and come to different conclusions. Shahbaz et al. [16] concluded that the level of carbon emissions can be lowered with financial development in the long-term perspective based on the empirical analysis of financial industry in Malaysia. On the contrary, Pradeepta et al. [17] argued that the increasing levels of financial development are adverse to environmental sustainability with the expansion in energy consumption. The empirical study of Zhao et al. [18] showed that the direct impact of financial efficiency and financial depth on environmental pollution was positive and negative, respectively.

With the deepening of research, the question arises whether it is possible to achieve sustainable economic growth and simultaneously reduce environmental pollution. In response to this question, many scholars carry out studies of green development from the perspective of total factor energy efficiency, which will be abbreviated as T-F-E-E in the following contents. Patterson [19] believed that energy efficiency is to produce the same number of services or effective outputs using less energy input. Some scholars emphasized the reduction of energy consumption caused by technological progress and improved management, which is concluded as technical energy efficiency [20]. Hu and Wang [21] first proposed the concept of T-F-E-E and defined it as the ratio between the target energy input required for a certain output and the actual energy input according to the best production practice on the premise that all factors other than energy factor inputs remain constant. With this definition, there is significant research that suggests that using financial systems as a means of capital allocation has a profound effect on T-F-E-E. Zhu [22] argued that finance can support the technological innovation of enterprises, which improves the clean energy utilization rate and the output per unit of energy consumption, thus promoting T-F-E-E. Some scholars believe that there is regional heterogeneity in the effect of finance on T-F-E-E. In economically highly developed areas, the improvement of financial industry level plays a significant role in promoting T-F-E-E due to the more efficient resource allocation, while in economically less developed areas, the development of financial industry plays a greater marginal role in improving T-F-E-E [23,24].

Stepping into the digital era, digital finance, relying on digital innovation technologies such as cloud computing, blockchain, and big data, has become a crucial apart of modern financial construction and has produced enormous benefits to human society. According to the study by the Development Research Group of the World Bank [25], digital finance can stimulate economic growth through asset accumulation. Most studies on digital finance have focused on its role in optimizing entrepreneurship mechanisms, reducing the urban-rural income gap, promoting inclusive growth, alleviating poverty, and stimulating regional innovation, focusing mainly on the economic benefits of digital finance. Representatively, Liu [26] proposed that the development of digital finance enhances financial equity, which, in turn, increases educational opportunities by financing vulnerable groups. Chen [27] confirmed that the effects of digital finance on poverty alleviation can be explained by easing information and credit restrictions, widening social networks, and spurring entrepreneurial activities. Some scholars have evaluated digital finance’s ecological benefits of digital finance in terms of promoting marine eco-efficiency, improving the productivity of the green total factor and reducing carbon emissions. The empirical research [28] showed that digital finance is able to improve marine ecological efficiency in China. Zhang [29] found that the efficiency of carbon emissions is significantly contributed by the synergy of digital finance and green technology innovation; however, due to the spatial correlation, it inhibits carbon emissions efficiency of the surrounding cities.

Summing the matter up, scholars have quantitatively studied the function of digital finance and factors influencing T-F-E-E, providing enormous valuable conclusions. However, there are still a few gaps that need to be tackled. First, compared to the economic benefits of digital finance, the environmental effects have not been fully discussed. Additionally, in the context of the ‘dual carbon’ target, there are diverse opinions on whether digital finance can improve energy efficiency. That is to say, the way in which digital finance can have an effect on energy efficiency and the mechanism of action deserves further study. Second, when exploring the effect of digital finance on energy efficiency at the macro level, many scholars overlook regional heterogeneity, which is particularly important for China, where regional economic development is uneven. Third, much of the existing literature uses a simple mediating effect model to clarify the impact pathway of digital finance on T-F-E-E. However, this impact pathway is complex and can be influenced by moderating factors. Therefore, this study delves the direct and indirect mechanism of digital finance’s impact on energy efficiency, using an econometric model, considering the moderating variables and taking full account of regional heterogeneity, potential endogeneity and related robustness issues, to provide practical and theoretical support for the development of digital finance and green transformation.

3. Methodology

3.1. Theoretical Framework

Through the review of existing literature, it is believed that the coverage breadth, use of depth, and digitalization of digital finance can effectively mitigate resource misallocation. Influenced by digital finance, high-tech and green industries have a more friendly investment environment than industries with high energy consumption and high pollution, and the investment efficiency is significantly improved. On the other hand, supported by digital technology, digital finance provides a disclosure of information on eco-friendly projects, green products, and environmental protection activities, which helps investors to identify green investment opportunities and leads consumers to green consumption, to improve the participation of the whole society in environmental protection activities, reducing energy consumption, and improving T-F-E-E. Thus, this study builds the following hypothesis:

Hypothesis 1. 

The development of digital finance can significantly improve T-F-E-E.

Based on the ‘structural dividend hypothesis’ from Lewis [30], the results of existing studies have demonstrated that energy efficiency is corelated with industrial structure. Normally, optimization of industrial structure includes two levels of meaning: the advancement and the rationalization of the industrial structure, both of which are related to energy efficiency [31]. In the process of the industrialization of the secondary industry on traditional energy and is an agglomeration of sectors with high energy consumption. Due to the development mode of high pollution, high input, and low output, the secondary industry consumes the most energy and produces the highest proportion of carbon emissions among the three industries. Along with the influence of unreasonable industrial structure and layout, the gas emission is increasing, and the ecological environment is deteriorating. According to existing theories, industrial structure optimization influences energy efficiency through the following ways. On the one hand, the advancement of industrial structure can gradually reduce the secondary industry’s dependence on energy consumption by reducing costs. On the other hand, the rationalization of industrial structure can promote the rapid flow of factors and the optimal integration and efficient distribution of energy resources, accelerating the green transformation of the industry by improving the production efficiency of enterprises. Thus, this study assumes that digital finance can promote T-F-E-E by acting on the industrial structure.

Moreover, China is a country that boasts a vast territory with uneven regional economic development and growth momentum. The financial system in the eastern coastal area is relatively complete and the industrial structure is more advanced and rational, while the availability and quality of financial services in the western and central regions are relatively poor, as is the level of industrial structure. Therefore, regional heterogeneity should be discussed.

In summary, this study proposes the following hypotheses:

Hypothesis 2. 

The improvement of T-F-E-E through digital finance can be realized through the role of digital finance in optimizing industrial structures, including the advancement and rationalization of the industry structure.

Hypothesis 3. 

The mediating effect of the industrial structure has regional heterogeneity in China.

Natural resources, as the premise of production and reproduction, are necessary for the development of human society. The conclusions of the existing literature have offered sufficient evidence that the dependence on natural resources certainly limit economic development and deteriorate ecological environment [32], but the direction of this impact is up for debate. Since energy consumption is higher in resource-dependent areas, the improvement of resource misallocation by digital finance in this region is particularly significant. Based on this theoretical analysis, the hypothesis is proposed as follows:

Hypothesis 4. 

The degree of resource dependence strengthens the role of digital finance in promoting T-F-E-E.

3.2. Model Construction

According to the theoretical framework, the following regression model is first constructed to investigate the impact of digital finance on the T-F-E-E.

$$TFE{E}_{it}={\alpha }_0+{\alpha }_{1}D{F}_{it}+{\alpha }_2{\mathrm{Controls}}_{it}+{\epsilon }_{it}^{}$$

(1)

In Equation (1), subscript i represents the provinces, and the subscript t indicates the year. TFEE denotes T-F-E-E. DF is the core explanatory variable denoting digital finance. Controls represents a series of control variables. ε is the random error term. The Hausman test result suggests that a fixed-effects model should be used for regression.

To test whether digital finance promotes T-F-E-E by enhancing industrial structure optimization, regression models of digital finance with advancement and rationalization of industry structure are set:

$$IS{U}_{it}={\beta }_0+{\beta }_{1}D{F}_{it}+{\beta }_2\mathrm{Controls}{s}_{it}+{\epsilon }_{it}$$

(2)

$$IS{O}_{it}={\gamma }_0+{\gamma }_{1}D{F}_{it}+{\gamma }_2{\mathrm{Controls}}_{it}+{\epsilon }_{it}$$

(3)

$$E{E}_{it}={\eta }_0+{\eta }_{1}D{F}_{it}+{\eta }_{2}IS{U}_{it}+{\eta }_3{\mathrm{Controls}}_{it}+{\epsilon }_{it}$$

(4)

$$E{E}_{it}={\theta }_0+{\theta }_{1}D{F}_{it}+{\theta }_{2}IS{O}_{it}+{\theta }_3{\mathrm{Controls}}_{it}+{\epsilon }_{it}$$

(5)

Here, ISU stands for advancement of industry structure, and ISO denotes the rationalization of industry structure. The definitions of the remaining variables are the same as those in Equation (1).

Equations (2) and (3), respectively, test the impact of digital finance on the advancement and rationalization of industrial structure. Equation (4) investigates the influence of digital finance on T-F-E-E when controlling the variable of the advancement of industrial structure. Coefficient η₁ demonstrates the direct effect of digital finance on T-F-E-E. The product of coefficient β₁ and coefficient η₂ indicates the mediating effect of the advancement of industry structure. Similarly, when the rationalization of industrial structure is controlled in Equation (5), the direct effect of digital finance on T-F-E-E is suggested by the coefficient θ₁, and the mediating effect played by the rationalization of industrial structure is represented by the product of coefficients γ₁ and θ₂.

Taking the mediating variable ISU_it as an example, the test procedure of mediating effect is as follows: First, the significance of the coefficient α₁ in Equation (1) is tested. If it is significant, digital finance plays a significant role in T-F-E-E and proceeds to the next step; otherwise, the test of mediating effect is terminated. Second, Equations (2) and (3) are tested. In the case where both β₁ and η₂ are significant, there is a partial mediation effect if η₁ is also significant; if η₁ fails the significance test, then a full mediation effect exists. If at least one of the coefficients β₁ and η₂ fails to pass the significance test, the bootstrap method should be adopted. The same procedure is conducted to examine the mediating effect of variable ISO_it.

Considering that resources endowment may interfere with the path of action of digital finance, this study constructs econometric models 6 to 10 for extended analysis by using the degree of resource dependence as a moderating variable.

$$E{E}_{it}={\chi }_0+{\chi }_{1}D{F}_{it}+{\chi }_{2}Re{s}_{it}+{\chi }_{3}D{F}_{it}\times Re{s}_{it}+{\chi }_4{\mathrm{Controls}}_{it}+{\epsilon }_{it}$$

(6)

$$IS{U}_{it}={\pi }_0+{\pi }_{1}D{F}_{it}+{\pi }_{2}Re{s}_{it}+{\pi }_{3}D{F}_{it}\times Re{s}_{it}+{\pi }_4{\mathrm{Controls}}_{it}+{\epsilon }_{it}$$

(7)

$$IS{O}_{it}={\rho }_0+{\rho }_{1}D{F}_{it}+{\rho }_{2}Re{s}_{it}+{\rho }_{3}D{F}_{it}\times Re{s}_{it}+{\rho }_4{\mathrm{Controls}}_{it}+{\epsilon }_{it}$$

(8)

$$\begin{array}{l}E{E}_{it}={\mu }_0+{\mu }_{1}D{F}_{it}+{\mu }_{2}Re{s}_{it}+{\mu }_{3}D{F}_{it}\times Re{s}_{it}+{\mu }_{4}IS{U}_{it}+{\mu }_{5}IS{U}_{it}\times Re{s}_{it}+\\ {\mu }_6{\mathrm{Controls}}_{it}+{\epsilon }_{it}\end{array}$$

(9)

$$\begin{array}{l}E{E}_{it}={\lambda }_0+{\lambda }_{1}D{F}_{it}+{\lambda }_{2}Re{s}_{it}+{\lambda }_{3}D{F}_{it}\times Re{s}_{it}+{\lambda }_{4}IS{O}_{it}+{\lambda }_{5}IS{O}_{it}\times Re{s}_{it}+\\ {\lambda }_6{\mathrm{Controls}}_{it}+{\epsilon }_{it}\end{array}$$

(10)

In Equations (6)–(10), Res denotes the degree of dependence on resources, and the moderated mediation pathway is demonstrated in Figure 1.

The moderating effect is proceeded as the following steps. First, Equation (6) tests the moderating effect of the degree of resource dependence on Path 1 in Figure 1. If the coefficient χ₃ is significant, there is a moderating effect. Secondly, the significance of the coefficients π₁, π₃, ρ₁, and ρ₃ in Equations (7) and (8) are examined, respectively. Finally, the significance of the coefficients μ₄, μ₅, λ₄, and λ₅ in Equations (9) and (10) is tested. If π₃ and μ₄ are simultaneously significant, then the degree of dependence has a moderating effect on path 2. If π₁ and μ₅ are simultaneously significant, then the moderating effect of resource dependence in path 3 is significant. The same procedure is conducted to investigate the moderating effect in paths 4 and 5. Furthermore, the magnitude of the mediating effect is related to the value of the moderating variable, and the relationship is as follows: mediating effect = (π₁ + π₃ Res_it) × (μ₄+ μ₅ Res_it) or (ρ₁+ ρ₃ Res_it) × (λ₄+ λ₅ Res_it).

3.3. Variables and Data Description

3.3.1. Explained Variable: T-F-E-E

Based on the research methodology from Tone [33], this study adopts the SBM model and ML index to measure the regional T-F-E-E considering the undesirable output.

Table 1. Input and output for the SBM model.

Indicator Type	Name	Symbol	Measurement Method	Measuring Unit
Input	Input of Labor Force	$x_1$	The Number of Employees in Each Province at The End of The Year	ten thousand people
	Energy Resource Input	$x_2$	Total Energy Consumption	ten thousand tons
	Capital Input	$x_3$	Perpetual Inventory Method	hundred million yuan
Desirable Output	Economic Output	$y_1^g$	GDP	hundred million yuan
Undesirable Output	Atmospheric Pollution	$y_1^b$	Carbon Emissions	million tons
		$y_2^b$	Industrial Sulfur Dioxide Emissions	ten thousand tons
		$y_3^b$	Industrial Dust Emissions	ten thousand tons
	Water Pollution	$y_4^b$	Industrial Wastewater Discharge	ten thousand tons
	Solid Waste Pollution	$y_5^b$	General Industrial Solid Wastes	ten thousand tons

demonstrates the specific input and output indicators. Each province is set as a DMU with the same inputs, desirable outputs, and undesirable outputs, which are, respectively, composed of three vectors $X=\left[x_1,\cdots ,x_n\right]\in R^{m\times n}$, $Y^g=\left[y_1^g,\cdots ,y_n^g\right]\in R^{S_1\times n}$ and $Y^b=\left[y_1^b,\cdots ,y_n^b\right]\in R^{s_2\times n}$. The following SBM model is constructed for a specific DMU:

$$\begin{array}{l}\min \rho =\frac{1-\frac{1}{m}{{\displaystyle \sum }}_{i=1}^m\frac{s_i^{-}}{x_{ik}}}{1+\frac{1}{s_1+s_2}\left({{\displaystyle \sum }}_{r=1}^{s1}\frac{s_r^g}{y_{rk}^g}+{{\displaystyle \sum }}_{r=1}^{s2}\frac{s_r^b}{y_{rk}^b}\right)}\\ \left\{\begin{array}{l}X\lambda +s_i^{\overline{i}}=x_k\hfill \\ Y^g\lambda -s^g=y_k^g\hfill \\ Y^b\lambda -s^b=y_k^b\hfill \\ \lambda ,s_i^{-},s^g,s^b⩾0\hfill \end{array}\right.\end{array}$$

(11)

In model 11, ρ is the efficiency score; λ is the weight vector; $s_i^{-}$ is the input slack variable; $s^g$ and $s^b$ are the desirable output and undesirable output slack variables. Applying the directional distance function to undesirable output to the Malmquist model, the ML index from period t to period t + 1 is:

$$M{L}_t^{t+1}={\left[\begin{array}{c}\frac{1+D_0^t\left(x^t,y^t,b^t;y^t,-b^t\right)}{1+D_0^t\left(x^{t+1},y^{t+1},b^{t+1};y^{t+1},-b^{t+1}\right)}\\ \frac{1+D_0^{t+1}\left(x^t,y^t,b^t;y^t,-b^t\right)}{1+D_0^{t+1}\left(x^{t+1},y^{t+1},b^{t+1};y^{t+1},-b^{t+1}\right)}\end{array}\right]}^{1/2}$$

(12)

3.3.2. Explanatory Variable: Digital Finance

This research employs The Peking University Digital Financial Inclusion Index of China (2011–2020) published by Institute of Digital Finance Peking University [34], which covers three dimensions to portray the digital finance index, including the coverage breadth (Cover), use depth (Deep), and the degree of digitization (Digi).

3.3.3. Mediating Variable: Industrial Structure

According to the above theoretical analysis, the level of regional industrial structure should be measured from the perspective of both the advancement and rationalization of industrial structure. The advancement of the industrial structure (ISU) is measured by the proportion of the tertiary industry’s added value in that of the secondary industry in each province. To measure the rationalization of the industrial structure (ISO), this study adopts Thiel index, which is calculated as follows:

$$ISO={\displaystyle \sum }_{i=1}^n\left(\frac{Y_i}{Y}\right)\ln \left(\frac{Y_i}{L_i}/\frac{Y}{L}\right)$$

(13)

In Equation (13), $Y$ represents the output value of different industries. $L$ denotes the number of employments in each region. The types of different industries and the number of different industrial sectors are denoted by $i$ and $n$.

3.3.4. Moderating Variable: Degree of Resource Dependence

Since the proportion of people employed in the mining industry in a region provides a good measure of the degree of dependence on the primary industrial sectors related to natural resources, this study measures the degree of regional resource dependence by calculating the proportion of mining employees to the total employed population. According to the current industry classification standards, the mining industry mainly includes energy extraction, metal mining, and non-metal mining.

3.3.5. Control Variables

To explain the effect of digital finance on T-F-E-E more accurately, the following four variables are selected based on the existing literature and economic and social realities: The level of government fiscal spending (fis) which is calculated as the share of local fiscal expenditure in local GDP, the degree of openness to the outside world (open) which is the proportion of export and import trade in provincial GDP, the level of financial development (fin) which is demonstrated as the proportion of banking institutions’ the balance of RMB deposits and loans in the regional GDP, and the level of foreign direct investment (fdi) which is expressed by the growth rate of total investment of foreign-invested enterprises in each province.

Given the availability of data, this study selects panel data from Chinese provinces from 2011 to 2019, excluding Tibet, Hong Kong, Macao and Taiwan. The data are obtained from the China Energy Statistical Yearbook, China Statistical Yearbook and Provincial Statistical Yearbooks. The Digital Finance Index comes from the Peking University Digital Financial Inclusion Index of China (2011–2020). Compiled by the Institute of Digital Finance of Peking University, the index is scientific, representative, reliable, and authoritative to describe the development of digital finance in China.

3.3.6. Descriptive Statistics of Variables

The results of variables’ descriptive statistical analysis and their correlation coefficients are reported in

Table 2. Statistical description and correlation coefficient matrix.

Panel A	Statistical Description
	N	Mean	Median	p50	p75	Std	Min	Max
TFEE	270	0.172	0.089	0.018	0.248	0.299	−0.533	1.641
DF	270	203.3	213.4	144.2	275.6	91.55	18.33	410.3
Wide	270	196.7	194.5	110.6	284.4	101.1	1.96	397
Deep	270	198	189.9	134.9	273.4	91.35	6.76	439.9
Digi	270	278.4	313.6	228.3	379.1	118	7.58	462.2
ISU	270	1.193	1.017	0.826	1.286	0.681	0.518	5.169
ISO	270	8.903	5.219	3.318	9.055	10.18	1.286	56.69
Res	270	0.039	0.033	0.009	0.056	0.041	0	0.222
fis	270	0.255	0.226	0.183	0.292	0.129	0.11	1.222
fin	270	3.173	2.968	2.461	3.531	1.139	1.528	8.131
open	270	0.279	0.143	0.09	0.349	0.297	0.013	1.464
fdi	270	0.368	0.204	0.142	0.487	0.363	0.048	1.792
Panel B	Correlation Coefficient
	TFEE	DF	Wide	Deep	Digi	ISU	ISO	Res
TFEE	1.000	0.494	0.507	0.560	0.353	0.173	0.464	−0.329
DF	0.494	1.000	0.979	0.949	0.898	0.440	0.272	−0.256
Wide	0.507	0.979	1.000	0.943	0.850	0.466	0.290	−0.253
Deep	0.560	0.949	0.943	1.000	0.768	0.454	0.367	−0.351
Digi	0.353	0.898	0.850	0.768	1.000	0.302	0.076	−0.131
ISU	0.173	0.440	0.466	0.454	0.302	1.000	0.643	−0.260
ISO	0.464	0.272	0.290	0.367	0.076	0.643	1.000	−0.369
Res	−0.329	−0.256	−0.253	−0.351	−0.131	−0.260	−0.369	1.000

. The differences between the mean (0.172) and median (0.089) of T-F-E-E (TFEE) indicate some variation in T-F-E-E among the study samples, and the same is true for digital finance (DF). According to the statistics of industrial structure advancement and industrial structure rationalization, the degree of industrial structure advancement fluctuates more during the sample period compared to the degree of industrial structure rationalization. In contrast, there are fewer disparities and dispersions in the degree of resource dependence over the years. Moreover, considering that the problem of multicollinearity may arise due to the high correlation among variables, thus reducing the accuracy of the estimation results, this study tests the correlation coefficients among the explanatory variables. Except for Wide, Deep, and Digi which stand for different dimensions of digital finance and therefore have high correlation coefficients with digital finance and with each other, the correlation coefficients among other explanatory variables do not exceed 0.8, so the problem of multicollinearity does not arise in the relative reliability of the model, which ensures the estimation results.

4. Results and Discussion

4.1. The Impact of Digital Finance on T-F-E-E

Table 3. The effect of digital finance on T-F-E-E.

	TFEE
	(1)	(2)	(3)	(4)	(5)
DF	0.0013 ***	0.0011 ***
	(0.0001)	(0.0001)
Wide			0.0010 ***
			(0.0001)
Deep				0.0012 ***
				(0.0001)
Digi					0.0006 ***
					(0.0001)
fis		−0.0033	0.0102	−0.0023	0.0095
		(0.1109)	(0.1104)	(0.1087)	(0.1171)
fin		−0.0654 **	−0.0760 **	−0.0521 *	−0.0203
		(0.0322)	(0.0327)	(0.0298)	(0.0334)
open		−0.7078 ***	−0.7039 ***	−0.7121 ***	−0.6671 ***
		(0.1359)	(0.1352)	(0.1327)	(0.1485)
fdi		0.1856 **	0.1610 **	0.1223	0.2827 ***
		(0.0781)	(0.0780)	(0.0774)	(0.0829)
cons	−0.0983 ***	0.2909 ***	0.3449 ***	0.2593 ***	0.1578
	(0.0218)	(0.0947)	(0.0964)	(0.0913)	(0.0966)
N	270.0000	270.0000	270.0000	270.0000	270.0000
R2	0.4334	0.5520	0.5561	0.5693	0.5009

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

suggests the regression results of digital finance on T-F-E-E. The regression coefficient of the variable DF on the dependent variable TFEE in column (1) is 0.0013, which is significant at the 1% level, which reveals that digital finance has a significantly positive effect on total factor energy efficiency. To avoid omitting significant variables, control variables were added and regressed again. The estimation results in column (2) represent that the fitting degree is significantly improved by adding control variables, which confirms the accuracy and reasonableness of the selection of control variables. Moreover, the coefficient of DF remains significant at the 1% level, supporting Hypothesis 1. Columns (3) to (5) report the impact of digital finance’s three dimensions on T-F-E-E, respectively. From the results, it can be found that the breadth of coverage, depth of use, and digitalization of digital finance all have positive effects on T-F-E-E, and such positive effects are all significant at the 1% level, demonstrating that the improvement of T-F-E-E by digital finance is mainly achieved by expanding its coverage, deepening the use depth and improving the degree of digitalization, among which the use depth has the greatest effect on improving T-F-E-E.

4.2. Digital Finance, Level of Industrial Structure, and T-F-E-E

According to the above mediating effect test procedure,

Table 4. Mediating effect test results.

	The Advancement of Industry Structure			The Rationalization of Industry Structure
	TFEE	ISU	TFEE	TFEE	ISO	TFEE
	(1)	(2)	(3)	(4)	(5)	(6)
DF	0.001 ***	0.002 ***	0.001 ***	0.001 ***	0.009 ***	0.001 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.002)	(0.000)
ISU			0.038 ***
			(0.057)
ISO						0.003 ***
						(0.005)
fis	−0.003	−0.071	−0.001	−0.003	−1.012	0.000
	(0.111)	(0.126)	(0.111)	(0.111)	(1.606)	(0.111)
fin	−0.065 **	0.102 ***	−0.069 **	−0.065 **	−0.034	−0.065 **
	(0.032)	(0.037)	(0.033)	(0.032)	(0.466)	(0.032)
open	−0.708 ***	−0.473 ***	−0.690 ***	−0.708 ***	−1.317	−0.703 ***
	(0.136)	(0.155)	(0.139)	(0.136)	(1.967)	(0.136)
fdi	0.186 **	0.355 ***	0.172 **	0.186 **	0.985	0.182 **
	(0.078)	(0.089)	(0.081)	(0.078)	(1.131)	(0.078)
cons	0.291 ***	0.541 ***	0.270 ***	0.291 ***	7.544 ***	0.266 ***
	(0.095)	(0.108)	(0.100)	(0.095)	(1.371)	(0.101)
N	270.000	270.000	270.000	270.000	270.000	270.000
R2	0.552	0.754	0.553	0.838	0.971	0.839

Standard errors in parentheses; ** p < 0.05, *** p < 0.01.

presents the mediating effect test results. As shown in column (2), the coefficient of variable DF indicates that digital finance has a positive contribution to enhance the industrial structure advancement. The main reason may be that digital finance is able to reach to the bottom stratum of society, provide reeducation opportunities, and improve the quality of employed people, thus promoting the rationalization and efficiency of financial resource allocation and the advancement of industry. The results in columns (1) to (3) show that the estimated coefficients α₁, β₁, and η₂ all pass the 1% significance test, which proves that the advancement of industrial structure is significant as a mediating variable, indicating that the influence of advanced regional industrial structure on the T-F-E-E is significantly positive. Meanwhile, the coefficient η1 is also significant, suggesting a partial mediating effect played by the advancement of industrial structure accounts for 5.87% of the total effect. Similarly, columns (4) to (6) reveal the mediating effect in the rationalization of industrial structure. Column (4) illustrates that digital finance positively contributes to the rationalization of industrial institutions, which may be mainly due to the fact that digital finance relying on big data technology mitigates the degree of information asymmetry between banks and enterprises, corrects the credit distortion system, and thus alleviates the degree of financial mismatch. Moreover, the estimated coefficients α₁, γ₁, and θ₂ pass the 1% significance test, and the direct effect of digital finance on the T-F-E-E is 0.001 after controlling for industrial structure rationalization, and the mediating effect played by the rationalization of industrial structure accounts for 8.34%. In conclusion, the above research suggests that digital finance can promote the regional T-F-E-E by promoting the level of regional industrial structure, which verifies Hypothesis 2.

4.3. The Impact of Different Dimensions of Digital Finance

To further identify the transmission path of digital finance to T-F-E-E, this research probes into the three dimensions of digital finance. The regression results are represented in

Table 5. The relationship among digital finance’s three dimensions, the advancement of industrial structure, and T-F-E-E.

	Cover		Deep		Digital
	ISU	TFEE	ISU	TFEE	ISU	TFEE
	(1)	(2)	(3)	(4)	(5)	(6)
Cover	0.002 ***	0.001 ***
	(0.000)	(0.000)
Deep			0.002 ***	0.001 ***
			(0.000)	(0.000)
Digi					0.001 ***	0.000 ***
					(0.000)	(0.000)
ISU		0.029 **		0.022 **		0.146 ***
		(0.057)		(0.055)		(0.054)
fis	−0.051	0.011	−0.071	0.001	−0.051	0.017
	(0.126)	(0.111)	(0.128)	(0.109)	(0.140)	(0.116)
fin	0.085 **	−0.079 **	0.147 ***	−0.054 *	0.175 ***	−0.046
	(0.037)	(0.033)	(0.035)	(0.031)	(0.040)	(0.034)
open	−0.470 ***	−0.691 ***	−0.512 ***	−0.698 ***	−0.410 **	−0.607 ***
	(0.154)	(0.138)	(0.156)	(0.136)	(0.178)	(0.148)
fdi	0.316 ***	0.152 *	0.266 ***	0.113	0.510 ***	0.208 **
	(0.089)	(0.080)	(0.091)	(0.079)	(0.099)	(0.086)
cons	0.626 ***	0.327 ***	0.453 ***	0.244 **	0.328 ***	0.110
	(0.110)	(0.103)	(0.107)	(0.095)	(0.116)	(0.097)
N	270.000	270.000	270.000	270.000	270.000	270.000
R2	0.757	0.556	0.747	0.571	0.698	0.516

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

and

Table 6. The relationship among digital finance’s three dimensions, the rationalization of industrial structure, and T-F-E-E.

	Cover		Deep		Digital
	ISO	TFEE	ISO	TFEE	ISO	TFEE
	(1)	(2)	(3)	(4)	(5)	(6)
Cover	0.007 ***	0.001 ***
	(0.002)	(0.000)
Deep			0.007 ***	0.001 ***
			(0.002)	(0.000)
Digi					0.005 ***	0.001 ***
					(0.001)	(0.000)
ISU		0.005 **		0.004 **		0.007 **
		(0.004)		(0.004)		(0.005)
fis	−0.949	0.014	−1.026	0.003	−0.875	0.015
	(1.624)	(0.110)	(1.623)	(0.109)	(1.623)	(0.117)
fin	0.102	−0.077 **	0.317	−0.052 *	0.189	−0.022
	(0.481)	(0.033)	(0.444)	(0.030)	(0.463)	(0.033)
open	−1.607	−0.697 ***	−1.715	−0.703 ***	−0.666	−0.663 ***
	(1.988)	(0.135)	(1.981)	(0.133)	(2.058)	(0.148)
fdi	0.875	0.157 **	0.638	0.117	1.824	0.271 ***
	(1.148)	(0.078)	(1.156)	(0.077)	(1.150)	(0.083)
cons	7.601 ***	0.310 ***	6.938 ***	0.226 **	6.596 ***	0.114
	(1.417)	(0.102)	(1.360)	(0.096)	(1.339)	(0.101)
N	270.000	270.000	270.000	270.000	270.000	270.000
R2	0.185	0.558	0.185	0.572	0.186	0.505

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

. In Table 5, the coefficients of the variables Cover, Deep and Digi in columns (1), (3), and (5) are all significant at the 1% level according to the mediation effect test procedure. In addition, the coefficients of the variables ISU in columns (2), (4), and (6) pass the significance test, and the coefficients of Cover, Deep and Digi are all significantly positive. Thus, the advancement of industrial structural plays a mediating role in the process of improving energy efficiency in all three dimensions: coverage breadth, use depth, and digitalization, where the mediating effects account for 5.80%, 4.40%, and 14.6% of the total effect, respectively, indicating that the mediating effect played by industrial structure advancement in the process of promoting energy efficiency in the digitalization dimension is particularly obvious. Similarly, as shown in Table 6, industrial structure rationalization also plays a mediating role in the effect of all three dimensions, with the proportion of the total effect being 3.17%, 2.96%, and 3.48%, respectively, indicating that the rationalization of industrial structure plays a particularly significant mediating role in the improvement of energy efficiency by the digitalization dimension compared to the other dimensions. In summary, all three dimensions of digital finance can promote T-F-E-E by improving the level of industrial structure, which further verifies Hypothesis 2.

4.4. Regional Heterogeneity Analysis

According to the geographical location and economic development level of each region, the sample is divided into the regional subsamples of central, western, and eastern. The eastern region includes Beijing, Guangdong, Fujian, Jiangsu, Hainan, Hebei, Liaoning, Zhejiang, Shanghai, Tianjin and Shandong. The Central region includes Anhui, Chongqing, Henan, Heilongjiang, Hubei, Hunan, Jiangxi, Shanxi and Jilin; The western regions include Gansu, Shaanxi, Guangxi, Guizhou, Ningxia, Inner Mongolia, Qinghai, Sichuan, Xinjiang and Yunnan.

The regression results by region are reported in

Table 7. The regression analysis results of regional heterogeneity (ISU).

	East			Middle			West
	TFEE	ISU	TFEE	TFEE	ISU	TFEE	TFEE	ISU	TFEE
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
DF	0.002 ***	0.002 ***	0.002 ***	0.001 ***	0.001 ***	0.001 ***	0.001 ***	0.002 ***	0.001 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
ISU			−0.092			0.131			0.154 ***
			(0.126)			(0.083)			(0.037)
fis	−0.029	−0.393 **	−0.065	0.604	−0.280	0.640	0.079	0.231	0.043
	(0.207)	(0.181)	(0.213)	(1.044)	(1.523)	(1.032)	(0.062)	(0.175)	(0.057)
fin	−0.074	−0.031	−0.077	−0.052	0.267 *	−0.087	−0.029	0.075	−0.041 **
	(0.066)	(0.058)	(0.066)	(0.108)	(0.157)	(0.109)	(0.022)	(0.062)	(0.020)
open	−0.509 **	−0.504 **	−0.555 **	−0.736 *	−1.227 **	−0.575	0.248	−1.232 *	0.437 **
	(0.240)	(0.210)	(0.249)	(0.371)	(0.541)	(0.381)	(0.221)	(0.625)	(0.206)
fdi	0.141	0.575 ***	0.194	0.796 ***	0.662 *	0.709 ***	0.076	0.236	0.039
	(0.134)	(0.118)	(0.153)	(0.264)	(0.385)	(0.267)	(0.084)	(0.237)	(0.077)
cons	0.394	1.228 ***	0.507	−0.093	0.139	−0.111	−0.017	0.466 **	−0.089
	(0.266)	(0.233)	(0.308)	(0.135)	(0.196)	(0.134)	(0.066)	(0.186)	(0.062)
N	99.000	99.000	99.000	81.000	81.000	81.000	90.000	90.000	90.000
R2	0.602	0.790	0.851	0.679	0.797	0.778	0.431	0.751	0.539

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

and

Table 8. The regression analysis results of regional heterogeneity (ISO).

	East			Middle			West
	TFEE	ISO	TFEE	TFEE	ISO	TFEE	TFEE	ISO	TFEE
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
DF	0.002 ***	0.014 ***	0.002 ***	0.001 ***	0.005	0.001 ***	0.001 ***	0.004 ***	0.001 ***
	(0.000)	(0.004)	(0.000)	(0.000)	(0.006)	(0.000)	(0.000)	(0.001)	(0.000)
ISO			−0.005			0.008			0.022 **
			(0.008)			(0.005)			(0.012)
fis	−0.029	−1.792	−0.038	0.604	−2.964	0.627	0.079	−0.575	0.092
	(0.207)	(2.760)	(0.208)	(1.044)	(24.128)	(1.034)	(0.062)	(0.589)	(0.061)
fin	−0.074	−0.757	−0.078	−0.052	2.173	−0.069	−0.029	−0.036	−0.028
	(0.066)	(0.879)	(0.066)	(0.108)	(2.488)	(0.107)	(0.022)	(0.208)	(0.021)
open	−0.509 ***	1.617	−0.500 ***	−0.736 ***	−4.088	−0.704 ***	0.248	−5.886 ***	0.380
	(0.240)	(3.201)	(0.241)	(0.371)	(8.574)	(0.368)	(0.221)	(2.105)	(0.229)
fdi	0.141	2.063	0.152	0.796 ***	−3.403	0.823 ***	0.076	1.100	0.051
	(0.134)	(1.794)	(0.136)	(0.264)	(6.102)	(0.262)	(0.084)	(0.800)	(0.084)
cons	0.394	14.493 ***	0.472	−0.093	1.086	−0.101	−0.017	3.031 ***	−0.085
	(0.266)	(3.548)	(0.293)	(0.135)	(3.111)	(0.133)	(0.066)	(0.626)	(0.074)
N	99.000	99.000	99.000	81.000	81.000	81.000	90.000	90.000	90.000
R2	0.850	0.972	0.851	0.770	0.661	0.777	0.808	0.879	0.817

Standard errors in parentheses; ** p < 0.05, *** p < 0.01.

. according to which, the total effect of digital inclusion finance on enterprise innovation (α₁) is significantly positive at the 1% level in both central and western regions, indicating that digital inclusion finance promotes T-F-E-E in different regions, but the mechanism of action shows significant differences among regions. The total effect of digital finance on T-F-E-E is significantly positive at the 1% level in all regions, indicating that digital finance is able to promote energy efficiency in different regions. However, the mechanism of effect is clearly different among different regions.

Firstly, focusing on the role path of industrial structure advancement, as shown in columns (3) and (6) of Table 7, the coefficient of variable ISU fails to pass the significance test, suggesting that the advancement of industrial structure does not play a mediating role in the process of digital finance improving T-F-E-E in the eastern and central regions, so digital finance promotes regional T-F-E-E more directly or through other paths. Column 9 shows that the coefficients of variables ISU and DF are significant, indicating that there is a mediating effect of the advancement of industrial structure in the western region, and this mediating effect is 30.80% of the total effect.

Similarly, as shown in columns (3) and (6) of Table 8, the variable ISO’s coefficient is not significant, indicating that the rationalization of industrial structure does not play a mediating effect in the central and eastern regions. Meanwhile, the variable DF’s coefficient in column (5) is not significant, suggesting that the development of digital finance does not significantly promote the rationalization of industrial structure in the central region. A reasonable explanation is that the enterprises in the central region, after financing constraints are relived, give priority to investing funds in skills research and development, or upgrading to high-end manufacturing, but have not yet focused on the coordinated development of individual industrial structures and the improvement of transformation ability, which will be an issue to be focused on in the process of industrial structure upgrading in the central region. Moreover, columns (7) to (9) indicate that there is a mediation effect played by the rationalization of industrial structure in the western region, which accounts for 8.80% of the total effect.

In conclusion, the development of digital finance can significantly contribute to the improvement of T-F-E-E in all regions of China, and in the western region, the advancement and rationalization of industrial structure plays an obvious mediating role in this promotion process. Therefore, Hypothesis 3 is verified.

4.5. Robustness and Endogenous Test Results

To ensure the robustness and credibility of the empirical results, a detailed robustness test is carried out.

To solve the endogeneity problem, this study uses the one-period lagged core explanatory variable for the regressions. The regression results for the lagged period of the digital finance index (DFt-1) are shown in

Table 9. Regression with the one-period lagged core explanatory variables.

	The Advancement of Industry Structure			The Rationalization of Industry Structure
	TFEE	ISU	TFEE	TFEE	ISO	TFEE
	(1)	(2)	(3)	(4)	(5)	(6)
DFt−1	0.001 ***	0.002 ***	0.002 **	0.001 ***	0.009 ***	0.001
	(0.000)	(0.000)	(0.001)	(0.000)	(0.002)	(0.001)
ISU			−0.145 ***
			(0.036)
ISO						0.009 ***
						(0.003)
fis	−0.051	−0.062	−0.015	−0.051	−0.980	0.142
	(0.111)	(0.127)	(0.141)	(0.111)	(1.617)	(0.141)
fin	−0.053 **	0.127 ***	−0.026	−0.053 **	0.185	−0.124 ***
	(0.026)	(0.036)	(0.026)	(0.026)	(0.454)	(0.021)
open	−0.146	−0.502 ***	0.167	−0.146	−1.597	0.188 *
	(0.089)	(0.156)	(0.108)	(0.089)	(1.976)	(0.109)
fdi	0.358 ***	0.297 ***	0.121 *	0.358 ***	0.745	0.051
	(0.062)	(0.090)	(0.065)	(0.062)	(1.146)	(0.068)
cons	−0.056	0.385 ***	−0.597 ***	−0.056	6.685 ***	−0.255
	(0.072)	(0.105)	(0.221)	(0.072)	(1.338)	(0.240)
N	270.000	270.000	270.000	270.000	270.000	270.000
R2		0.750	0.527		0.191	0.512

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

. It can be found that there is still a positive correlation between digital finance and T-F-E-E. After adding the variable ISU and variable ISO, respectively, a partial mediation effect is still found according to the test steps. The parameter estimates and significance of all variables in the robustness test are generally consistent with the original regression, which indicates that the empirical results of this study are robust.

To avoid the bias caused by the contingency of variables selection, the following variable substitution approach is conducted:

First, the core explanatory variables are replaced. Drawing on the practice of Xie [35], the Internet penetration rate is used as the substitution variable of digital finance. Internet penetration and digital finance are closely related, and there is no direct role between Internet penetration and T-F-E-E in terms of industry heterogeneity characteristics. Therefore, it is reasonable to use Internet penetration (INT) as a substitution variable.

Table 10. Replacing the core explanatory variable.

	The Advancement of Industry Structure			The Rationalization of Industry Structure
	TFEE	ISU	TFEE	TFEE	ISO	TFEE
	(1)	(2)	(3)	(4)	(5)	(6)
INT	0.001 ***	0.002 ***	0.002 **	0.001 ***	0.009 ***	0.001
	(0.000)	(0.000)	(0.001)	(0.000)	(0.002)	(0.001)
ISU			−0.145 ***
			(0.036)
ISO						0.009 ***
						(0.003)
fis	−0.051	−0.062	−0.015	−0.051	−0.980	0.142
	(0.111)	(0.127)	(0.141)	(0.111)	(1.617)	(0.141)
fin	−0.053 **	0.127 ***	−0.026	−0.053 **	0.185	−0.124 ***
	(0.026)	(0.036)	(0.026)	(0.026)	(0.454)	(0.021)
open	−0.146	−0.502 ***	0.167	−0.146	−1.597	0.188 *
	(0.089)	(0.156)	(0.108)	(0.089)	(1.976)	(0.109)
fdi	0.358 ***	0.297 ***	0.121 *	0.358 ***	0.745	0.051
	(0.062)	(0.090)	(0.065)	(0.062)	(1.146)	(0.068)
cons	−0.056	0.385 ***	−0.597 ***	−0.056	6.685 ***	−0.255
	(0.072)	(0.105)	(0.221)	(0.072)	(1.338)	(0.240)
N	270.000	270.000	270.000	270.000	270.000	270.000
R2		0.750	0.527		0.191	0.512

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

reports the new regression results. The results of columns (1) and (4) show that the regression coefficient of Internet penetration rate on T-F-E-E is significantly positive, and the mediating effect still exists after adding ISU and ISO, which verifies the robustness of the empirical results.

Second, for the explanatory variables, the study remeasures energy efficiency using the traditional method (that is, the ratio of real GDP to total energy consumption for each region).

Table 11. Replacing the explained variable.

	The Advancement of Industry Structure			The Rationalization of Industry Structure
	EE	ISU	EE	EE	ISO	EE
	(1)	(2)	(3)	(4)	(5)	(6)
DF	0.0001 ***	0.0017 ***	0.0001 ***	0.0001 ***	0.0085 ***	0.0001 ***
	(0.0000)	(0.0002)	(0.0000)	(0.0000)	(0.0020)	(0.0000)
ISU			−0.0038
			(0.0059)
ISO						−0.0007
						(0.0005)
fis	0.0049	−0.0713	0.0046	0.0049	−1.0124	0.0042
	(0.0114)	(0.1264)	(0.0114)	(0.0114)	(1.6058)	(0.0114)
fin	0.0034	0.1017 ***	0.0038	0.0034	−0.0344	0.0034
	(0.0033)	(0.0367)	(0.0034)	(0.0033)	(0.4663)	(0.0033)
open	0.0381 ***	−0.4734 ***	0.0363 **	0.0381 ***	−1.3175	0.0373 ***
	(0.0140)	(0.1548)	(0.0143)	(0.0140)	(1.9675)	(0.0140)
fdi	0.0136 *	0.3554 ***	0.0149 *	0.0136 *	0.9853	0.0142 *
	(0.0080)	(0.0891)	(0.0083)	(0.0080)	(1.1315)	(0.0080)
cons	0.0670 ***	0.5413 ***	0.0691 ***	0.0670 ***	7.5435 ***	0.0719 ***
	(0.0098)	(0.1079)	(0.0103)	(0.0098)	(1.3715)	(0.0103)
N	270.0000	270.0000	270.0000	270.0000	270.0000	270.0000
R2	0.2092	0.7538	0.9528	0.2092	0.2021	0.9531

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

reports the regression results of digital finance on energy efficiency (EE) calculated by the above method. Columns (1) and (4) show the mediating effect still exists, further demonstrating the robustness of the empirical results.

4.6. The Moderating Effect of Resource Dependence Degree

As the results are shown in

Table 12. The moderating effect of the degree of resource dependence.

	The Advancement of Industry Structure			The Rationalization of Industry Structure
	EE	ISU	TFEE	TFEE	ISO	TFEE
	(1)	(2)	(3)	(4)	(5)	(6)
DF	0.001 ***	0.002 ***	0.001 ***	0.001 ***	0.008 ***	0.001 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.003)	(0.000)
ISU			0.063
			(0.056)
ISO						0.004 **
						(0.004)
Res	0.120	−2.472	0.479	0.120	−4.651	0.530
	(1.406)	(1.652)	(1.405)	(1.406)	(21.092)	(1.496)
Res × DF	−0.010 ***	0.000	−0.016 ***	−0.010 ***	−0.009	−0.010 ***
	(0.002)	(0.003)	(0.004)	(0.002)	(0.036)	(0.002)
Res × ISU			2.732 **
			(1.174)
Res × ISO						0.090 ***
						(0.114)
fis	−0.013	−0.068	−0.027	−0.013	−1.015	−0.013
	(0.108)	(0.126)	(0.107)	(0.108)	(1.613)	(0.108)
fin	−0.033	0.097 **	−0.053	−0.033	−0.012	−0.037
	(0.032)	(0.038)	(0.033)	(0.032)	(0.482)	(0.033)
open	−0.561 ***	−0.525 ***	−0.574 ***	−0.561 ***	−1.271	−0.571 ***
	(0.138)	(0.162)	(0.140)	(0.138)	(2.064)	(0.139)
fdi	0.148 *	0.367 ***	0.176 **	0.148 *	0.970	0.147 *
	(0.076)	(0.090)	(0.081)	(0.076)	(1.146)	(0.077)
cons	0.149	0.702 ***	0.155	0.149	7.713 ***	0.130
	(0.127)	(0.149)	(0.133)	(0.127)	(1.905)	(0.132)
N	270.000	270.000	270.000	270.000	270.000	270.000
R2	0.582	0.756	0.593	0.582	0.202	0.584

Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

, the pattern of interaction Res × DF is an antagonistic interaction since both χ₁ and χ₂ work in the positive direction, and the interaction is of an opposite sign, which means that both the variable Res and the variable DF have compensatory effects on the variable TFEE. According to columns (2) and (5), the estimated coefficient of variable DF is significant, but the coefficient of the cross-term Res × DF is not, which indicates that the degree of resource dependence does not play a significant moderating effect on path 2 and path 4 of Figure 1. One explanation for this result is that the degree of regional resource dependence can be reflected by the characteristics of regional industrial structure, while it has no obvious influence on the process of realizing the reasonable financial resources allocation, so it will not significantly intervene the digital finance’s effect on the advancement and rationalization of industrial structure.

According to the regression results in columns (3) and (6), the estimated coefficients of the cross terms Res × DF, Res × ISU, and Res × ISO are significant, indicating that the degree of resource dependence plays a moderating role in both path 3 and path 5 of Figure 1, respectively, which in turn affects the role of digital finance in enhancing energy efficiency. Finally, the mediating effect considering the moderating variables can be calculated as the product of the following coefficients: 0.002 × (0. 063 + 2.732 × Res) and 0.008 × (0.004 + 0. 090 × Res). It can be concluded that the coefficient of resource dependence on the mediating effect is always positive, indicating that with the deepening of resource dependence, the role of digital finance in improving T-F-E-E via upgrading the level of industrial structure is enhanced.

5. Conclusions

This study theoretically and empirically analyses the mechanism of digital finance on T-F-E-E. A moderated mediation model is constructed and strongly revealed the important mediating role of industry structure and the moderating role of resource dependence in this mechanism. The following conclusions were drawn.

First, digital finance has a significant driving effect on the growth of T-F-E-E, in which the use depth of digital finance plays a major role

Second, digital finance can improve T-F-E-E by optimizing the level of industrial structure, and this conclusion still holds after considering endogeneity issues and robustness tests, such as substituting variables. Compared to the central and eastern regions, digital finance has a more significant effect on T-F-E-E by enhancing the level of industrial structure in the western region, indicating that digital finance has, indeed, eased the financing constraints of the real economy in underdeveloped regions to achieve green transformation.

Third, the transmission path of digital finance is influenced by the degree of regional resource dependence, and the higher the degree of regional resource dependence, the stronger the positive enhancement effect of digital finance on T-F-E-E. Meanwhile, as the degree of resource dependence deepens, the effect of energy savings and emission reduction played by industrial structure optimization is strengthened, which in turn promotes the mediation channel for digital finance to exert its financial efficacy.

These conclusions provide inspiration for improving energy efficiency and sustainable development. Based on the research conclusions, the three policy suggestions are presented.

First, the development of digital finance should continue to be promoted. Digital financial facilities must be fully improved so that digital finance can support regional green transformation and sustainable economic development.

Second, digital finance policies are supposed to be formulated according to local conditions. Considering the region’s resource endowment and industrial structure, we should strengthen the coverage, depth and digitization of digital finance for the real economy should be strengthened, give full play to its advantage in breaking the space-time blockage in the flow of financial factors, reduce energy consumption, and enhance the efficiency of energy factor allocation.

Finally, it is necessary to accelerate the popularization and application of digital technology in different field and optimize the industrial structure. Relevant institutions should fully enhance the role of digital technology in driving economic development, provide targeted support to the real economy, lead the capital flow to high-tech and high value-added emerging industries, strengthen the intrinsic dynamics of economic growth, and promote the sustainable development of the green economy.