Does the Development of Digital Finance Enhance Urban Energy Resilience? Evidence from Machine Learning

Abstract

Amid the escalating global climate crisis, the transition to sustainable energy systems has become imperative. As the world’s largest energy producer and consumer, China has established ambitious dual carbon targets, which present formidable challenges to urban energy systems that remain heavily reliant on conventional energy sources and exhibit inadequate renewable energy development. Drawing on complex adaptive systems theory, this study investigates the extent to which digital finance enhances urban energy resilience, examining both the underlying mechanisms and heterogeneous effects. Employing a multi-period difference-in-differences model with digital finance policies as a quasi-natural experiment, our analysis of panel data from 31 Chinese provinces (2016–2023) demonstrates that digital finance significantly enhances the resilience of urban energy systems and their three constituent subsystems. A mediation analysis reveals the pivotal role of innovative organizations, while machine learning techniques uncover nonlinear relationships moderated by marketization levels, fiscal energy allocations, and initial digital finance development. These findings provide critical insights for policymakers, financial institutions, and energy enterprises seeking to advance sustainable energy governance and foster financial innovation in the energy transition.

1. Introduction

Under the mounting pressure of global climate change, energy transition has become a critical imperative in the development of energy systems worldwide. As the world’s largest energy producer and consumer, China has established ambitious targets to achieve carbon peaking by 2030 and carbon neutrality by 2060. However, the implementation of these dual carbon policies has introduced unprecedented challenges in the realm of energy security [1,2].

Amid the immense pressure from the energy transition, urban energy systems have become increasingly vulnerable. Enhancing the resilience of urban energy systems has emerged as a crucial imperative for ensuring energy security and sustainable development [3]. However, building such complex energy resilience is not merely a technical issue but also a resource allocation problem, which inevitably requires robust and efficient financial support [4]. In this context, the limitations of the traditional financial model have become increasingly apparent, creating an urgent need and vast potential for financial innovation, especially the intervention of digital finance [5].

Although the potential of digital finance has received wide attention, its specific mechanisms, transmission paths, and impact boundaries in enhancing urban energy resilience have not yet been systematically clarified. To address this gap, this study aims to fill this research void and attempts to answer the following core questions: Does the development of digital finance significantly enhance urban energy resilience? If so, through what channels does it exert its influence? Moreover, do its effects vary across different regional characteristics?

To answer these questions, this study first employs the implementation of digital finance policy as a quasi-natural experiment and constructs a multi-period difference-in-differences (DID) model to identify the causal effect of digital finance development on urban energy resilience. Subsequently, this study explores the mediating role of innovative organizations. Finally, we utilize machine learning methods to uncover potential nonlinear relationships. Heterogeneity is analyzed across multiple dimensions, including initial digital finance levels, regional scale, marketization degree, and fiscal energy allocation proportions.

The potential marginal contributions of this study can be delineated across three dimensions.

First, at the theoretical level, this study integrates complex adaptive systems theory into the research on urban energy resilience. It examines the catalytic effects of digital finance as an external shock, as well as its coupling and interaction mechanisms with various subsystems. Furthermore, this study identifies the critical role of innovative organizations as a mediating factor in the formation and diffusion of resilience. This approach not only broadens the scope of energy resilience research but also provides novel insights into multi-agent coordination and system evolution, constructing a more comprehensive theoretical perspective on how financial innovation interfaces with the development of energy resilience.

Second, at the empirical methodology level, in contrast to previous studies that predominantly employ linear regression [6,7,8], this research constructs a two-way fixed effects multi-period DID model to rigorously identify the causal relationship between digital finance development and urban energy resilience. Additionally, this study employs a mediation effect model to reveal the crucial role of innovative organizations in the transmission mechanism. By incorporating machine learning methods, this study further explores potential nonlinear relationships and heterogeneous characteristics, providing a more nuanced and methodologically robust empirical toolkit for analyzing multidimensional interactions in complex economic and social systems.

Finally, at the practical policy level, the findings of this study offer actionable insights for governmental bodies, financial institutions, and energy enterprises in coordinating energy governance and financial innovation under the dual carbon framework. By formulating forward-looking and adaptive strategic combinations and enhancing energy financial service systems, these findings can contribute to the continuous strengthening of urban energy resilience. This, in turn, provides a solid empirical foundation for sustainable urban development and energy security transitions.

The remainder of this study is structured as follows: Section 2 reviews the relevant literature and constructs the theoretical framework and research hypotheses. Section 3 details the research design. Section 4 presents the empirical results. Section 5 explores the underlying mechanisms and analyzes heterogeneity. Finally, Section 6 provides conclusions and the discussion, followed by corresponding policy recommendations.

2. Theoretical Mechanism and Research Hypothesis

2.1. Literature Review

Amid the energy transition under the dual carbon policy, urban energy security faces multifaceted challenges. On the one hand, traditional energy still dominates China’s energy consumption structure [9]. According to data from the National Bureau of Statistics, coal consumption accounted for 55.3% of China’s total energy consumption in 2023. If traditional energy sources are phased out too rapidly while new energy sources have not yet bridged the gap, urban energy supply and demand will become imbalanced [10], causing energy prices to fluctuate frequently due to constrained energy supplies. On the other hand, as current energy storage technology remains immature, the intermittency of new energy sources cannot be effectively resolved as their share grows [11], leading to frequent intermittent power outages in urban energy systems and a subsequent escalation in operational risks [12]. These challenges underscore the complexity of building energy resilience, while the traditional financial system exhibits significant limitations in meeting diverse financing needs, optimizing resource allocation, and managing risks [13].

With the rapid advancement of digital technology, a digital transformation is injecting new momentum into the reform of the financial system. Digital finance, as a core component of “Digital China”, is rapidly emerging and gradually reshaping the traditional financial ecosystem, exhibiting high adaptability in building urban energy resilience [14]. Compared to FinTech, which emphasizes technological attributes, and internet finance, which focuses on financial attributes, the concept of digital finance is more comprehensive and encompasses a broader scope [15,16]. Its core lies in leveraging big data and other technologies to enable flexible and efficient resource allocation. From improving human capital structure [17] and optimizing financial resource allocation [18] to enhancing policy effectiveness [19], the development of digital finance has demonstrated multidimensional potential to advance resilience. Although existing research has examined energy resilience enhancement [20], energy security transitions [21], and green finance [22], there remains a research gap concerning the mechanisms, pathways, and heterogeneous effects of digital finance in fostering urban energy resilience. Therefore, systematically clarifying the mechanisms, transmission paths, and heterogeneous effects of digital finance in enhancing urban energy resilience is of paramount theoretical and practical significance, which constitutes the core focus of this study.

2.2. Complex Systems Theory and Urban Energy Systems

To systematically investigate the core issues mentioned above, this study draws on complex systems theory as its core analytical framework. The theory, first proposed by Holland [23], posits that a complex system comprises numerous independent yet heterogeneous entities. These entities establish dynamic coupling through multidimensional, nonlinear feedback mechanisms and achieve self-organization and self-adaptive evolution in response to changes in the external environment [24].

As China advances its dual carbon goals, the traditional single-model approach to urban energy governance has encountered significant challenges, such as supply–demand imbalances and technological barriers in new energy development. Consequently, there is a pressing need to understand the structural characteristics and dynamic evolution of urban energy systems from a more comprehensive perspective. Complex systems theory provides a robust analytical framework for this purpose [25]. Specifically, as illustrated in Figure 1, the urban energy system can be decomposed into multiple interconnected subsystems, including the energy subsystem, context subsystem, and government subsystem [26].

The energy subsystem consists of providers, users, institutions, and agents, offering key driving forces for the operation of the entire system. The context subsystem provides multidimensional support for stable system operations and primarily includes physical landscape, social heterogeneity, historical evolution, land use, economic diversity, and political narratives. The government subsystem comprises the energy sector, other sectors, and local government, which regulate and optimize the urban energy system. As crucial intermediaries, innovative organizations such as universities and research centers convert external knowledge and resources into the system’s internal drivers [27].

Building on this foundation, the subsystems and innovative organizations continually adjust their behaviors and strategies to adapt to changes in the external environment, allowing the structure and functions of the urban energy system to be dynamically optimized and reconstructed over time. These characteristics align with the core features of complex systems theory—namely, the nonlinear interactions among diverse agents, cross-subsystem feedback mechanisms, and long-term adaptive evolution trajectories—suggesting that urban energy systems are progressively evolving into complex systems involving multi-level and multi-agent coupling. Based on this analysis, examining urban energy systems within the framework of complex systems theory helps deepen our understanding of the diverse drivers and processes that contribute to resilience formation. It also provides a theoretical basis for subsequent research on how the development of digital finance directly and indirectly promotes resilience through innovative organizations [28,29].

2.3. System Resilience Formation

Holling et al. [30] first defined system resilience as a system’s ability to maintain its core functions and achieve structural upgrades when facing external disturbances. In the context of global climate change and energy transition, system resilience has become increasingly significant. For urban energy systems, enhancing resilience is critical for ensuring energy supply security, promoting technological iteration, achieving energy security transitions, and supporting urban sustainable development [31]. When urban energy systems face external shocks, their response process can be divided into four phases: original stable state, disruption state, recovery state, and new stable state [32].

Based on these phases, scholars have typically classified system resilience into four capabilities [33]: (1) bearing capacity, which reflects the system’s ability to maintain core functions in a steady state during short-term disturbances; (2) recovering capacity, emphasizing rapid repair and reconstruction after damage; (3) learning capacity, focusing on strategy and resource allocation optimization through trial and error and information feedback; and (4) conversion capacity, which highlights the system’s ability to achieve higher-level adaptability through institutional innovation, technological upgrading, and industrial restructuring, as shown in Figure 2.

From the perspective of complex system theory, the formation of urban energy resilience exhibits emergent properties driven from the bottom up and from the inside out. In other words, the system’s overall resilience arises from nonlinear interactions among multiple elements at different levels and dimensions. At the micro-level, individual elements within each subsystem engage in continuous local adaptation and functional optimization through feedback mechanisms such as information and resource flows, laying the groundwork for the gradual development of bearing capacity, recovering capacity, learning capacity, and conversion capacity.

At the macro-level, coupling among subsystems further promotes the emergence of overall system resilience. The energy subsystem, context subsystem, and government subsystem influence and reinforce one another through cross-subsystem information exchange, resource allocation, and policy coordination. Through a dynamic equilibrium involving both positive and negative feedback, resilience capabilities dispersed across subsystems become progressively integrated and magnified, ultimately aggregating into system-wide resilience [34,35].

This layered, cross-level mechanism reflects the core tenets of complex systems theory and provides a comprehensive framework for analyzing external factors such as digital finance.

In summary, the existing research has illuminated the challenges facing urban energy systems amidst the energy transition and has acknowledged the potential of digital finance; however, the specific mechanisms and pathways through which it enhances energy resilience remain to be clarified. To address this gap, this study adopts complex systems theory as its core analytical framework, viewing the urban energy system as a dynamic, multi-agent, complex adaptive system. Within this framework, resilience is deconstructed into four core capabilities: bearing, recovering, learning, and conversion. This theoretical perspective provides a foundation for systematically analyzing how external shocks, such as digital finance, affect urban energy resilience.

2.4. Mechanism Analysis of Digital Finance Development

Alongside the intensification of global climate change, economic digital transformation is also accelerating, making digital finance a pivotal means of financial innovation and functional expansion [36]. Traditional financial systems often confront multiple structural bottlenecks—such as outdated human capital structures, inefficient resource allocation, and limited regulatory instruments—when operating under complex, volatile, and highly uncertain market conditions. However, with the rapid advancement of digital technologies (e.g., big data, artificial intelligence, blockchain, and cloud computing), digital finance offers systematic solutions to alleviate these constraints [37].

Digital finance addresses these constraints through three primary mechanisms. First, through online platforms, data analytics, and knowledge sharing, it enhances human capital quality by raising demand for high-end talent and enabling practitioners to process market information more swiftly [38]. Second, by leveraging big data risk control and intelligent matching mechanisms, digital finance improves resource allocation, transcending temporal and spatial barriers to achieve flexible and efficient fund distribution [39]. Finally, real-time regulatory technology and smart contracts strengthen policy alignment and mitigate implementation lags, allowing policymakers to dynamically monitor market feedback and adjust policy tools [40].

On this basis, digital finance—compared with traditional finance—fosters improvements in human capital structure, resource allocation, and policy alignment as an external system shock, facilitating the accumulation and integration of resilience elements at the subsystem level. In the energy subsystem, data-driven intelligent risk control and precise financing provide efficient capital support for new energy and energy storage technology projects, thereby reinforcing bearing capacity and recovering capacity. Simultaneously, continuous information feedback and trial-and-error processes enable learning and strategy updates, enhancing learning capacity and conversion capacity [41,42]. In the context subsystem, information transparency and orderly market participation deepen the alignment between diverse social and economic actors and energy transition objectives, not only sustaining bearing capacity and recovering capacity under shocks but also establishing institutional and structural foundations for long-term learning capacity and conversion capacity [43]. For the government subsystem, regulatory technology and data analytics help policymakers promptly capture market signals and optimize policies, thereby improving bearing capacity and recovering capacity, as well as reinforcing learning capacity and conversion capacity through accumulated experience and institutional innovation [44].

Overall, within a complex systems framework, digital finance promotes information exchange among the three primary subsystems via data-driven processes and creates flexible, efficient resource allocation networks. This process establishes and consolidates nonlinear feedback loops across multiple agents and levels. Consequently, the four resilience capabilities—bearing capacity, recovering capacity, learning capacity, and conversion capacity—disseminate, integrate, and co-evolve from the bottom up and from the inside out, ultimately enhancing the resilience of the entire urban energy system.

Accordingly, this study proposes the following hypotheses:

H1. 

Digital finance development enhances the overall resilience of urban energy systems. 

H1a. 

Digital finance development enhances the resilience of the energy subsystem. 

H1b. 

Digital finance development enhances the resilience of the context subsystem. 

H1c. 

Digital finance development enhances the resilience of the government subsystem. 

2.5. Analysis of Innovative Organizations’ Mediating Effect

Amid the accelerating global energy transition, innovative organizations serve not only as “distribution hubs” for knowledge and resources but also as “accelerators” that drive iterative upgrades of urban energy systems under external shocks. Within the complex systems framework, innovative organizations act as key mediating carriers that absorb external knowledge and resources, integrate them with internal elements, and rapidly translate them into ongoing momentum for dynamic optimization and iteration of urban energy systems [45,46]. As an external shock, digital finance provides significant positive feedback for the assimilation of external resources and advanced concepts by innovative organizations through its advantages in improving the human capital structure, optimizing resource allocation, and enhancing policy alignment. This enriches the technical systems and institutional innovation vitality of urban energy systems.

First, digital finance can enhance the human capital quality of innovative organizations via talent recruitment platforms and online educational resources, thereby strengthening interdisciplinary R&D capabilities and technological reserves [47]. Second, the application of intelligent risk control and big data risk control mechanisms streamlines the cross-regional allocation of financial resources and industrial elements, enabling innovative organizations to acquire project funding and technical support more readily [48]. Finally, real-time regulatory technologies and online policy feedback systems promote efficient information flow and synergy between government agencies and innovative organizations, allowing the latter to promptly adjust their strategies and accelerate the transformation of outcomes within the institutional environment [49].

Building upon these developments, the technical systems and institutional innovation capacities within urban energy systems grow more robust, thereby fueling ongoing gains in system resilience. In the energy subsystem, innovative organizations tackle key technological areas such as distributed generation, energy storage optimization, and load management. This not only bolsters the bearing capacity and recovering capacity in emergencies but also enhances the learning capacity and conversion capacity through repeated experimentation and iterative improvements [50]. In the context subsystem, innovative organizations leverage cross-disciplinary collaboration and resource integration to coordinate social, economic, and environmental elements. While preserving the subsystem’s bearing capacity and recovering capacity during disruptions, this also lays the groundwork for sustained learning capacity and conversion capacity via the continuous generation of new ideas and collaborative models [51,52]. In the government subsystem, innovative organizations provide technical guidance and policy recommendations to decision-making bodies through professional evaluation and forward-looking research. This increases the government’s bearing capacity and recovering capacity for emergencies and, through accumulated experience and process optimization, strengthens ongoing learning capacity and conversion capacity [53].

From a complex systems perspective, once local resilience within each subsystem is enhanced, innovative organizations act as bridges to extend these improvements system-wide. During this process, innovative organizations function both as “hubs” for information and resources and as “accelerators” that transform localized achievements into system-level improvements in resilience. Through iterative collaboration and cross-level linkages, the three subsystems gradually form a nonlinear coupling process that advances from the bottom up and from the inside out. Ultimately, under the combined influence of external shocks and internal evolution, this process leads to a comprehensive enhancement of urban energy resilience.

Accordingly, this study proposes the following hypotheses:

H2. 

Digital finance development indirectly enhances the overall resilience of urban energy systems by promoting the development of innovative organizations. 

H2a. 

Digital finance development indirectly enhances the resilience of the energy subsystem by promoting the development of innovative organizations. 

H2b. 

Digital finance development indirectly enhances the resilience of the context subsystem by promoting the development of innovative organizations. 

H2c. 

Digital finance development indirectly enhances the resilience of the government subsystem by promoting the development of innovative organizations. 

Based on the above, this study posits H1, H1a, H1b, and H1c as direct effects, while H2, H2a, H2b, and H2c represent indirect effects mediated through innovative organizations, as illustrated in Figure 3.

3. Research Design and Status Analysis

3.1. Model Construction

3.1.1. Basic Regression Model

Considering the multifaceted determinants affecting urban energy resilience and aiming to mitigate biases introduced by endogenous variables while accounting for the phased implementation of policies, this study employs a two-way fixed effects, multi-period difference-in-differences (DID) model as the foundational analytical framework [54,55]. The specific formulation of the model is presented in Equation (1):

$${UER}_{it}=a_0+a_1{DID}_{it}+a{Z}_{it}+{\mu }_i+{\gamma }_t+{\epsilon }_{it}$$

(1)

Here, ${UER}_{it}$ represents the dependent variable, denoting the urban energy resilience index for province i in year t. The core explanatory variable, ${DID}_{it}$, is a dummy variable that equals 1 if province i has implemented the policy in year t or later and 0 otherwise. The coefficient $a_1$ captures the net effect of the policy on urban energy resilience. $Z_{it}$ encompasses a set of control variables, while ${\mu }_i$ and ${\gamma }_t$ signify province-specific and year-specific fixed effects, respectively. Lastly, ${\epsilon }_{it}$ is the error term.

3.1.2. Mechanism Effects Model

To examine the mediating role of innovative organizations in the relationship between digital finance policy and urban energy resilience, this study extends the baseline model with two sequential equations as shown in Equations (2) and (3) [56]:

$$M_{it}=b_0+b_1{DID}_{it}+b{Z}_{it}+{\mu }_i+{\gamma }_t+{\epsilon }_{it}$$

(2)

$${UER}_{it}=c_0+c_1{DID}_{it}+c_2{M}_{it}+c{Z}_{it}+{\mu }_i+{\gamma }_t+{\epsilon }_{it}$$

(3)

In these equations, $M_{it}$ denotes the mediating variable, quantified by the development status of innovative organizations, specifically their research and development (R&D) capabilities. A significant $b_1$ in Equation (2) indicates that ${DID}_{it}$ significantly influences the mediating variable. Concurrently, a significantly positive $c_2$ in Equation (3), coupled with a reduction in $c_1$ compared to $a_1$ in the baseline model, substantiates the mediating effect of innovative organizations.

3.1.3. Causal Forest Model

Contrary to traditional DID models that presuppose linear relationships, the causal forest algorithm, an advanced machine learning technique, adeptly captures nonlinear interactions among variables [57,58,59]. This study formulates the treatment effect using the causal forest approach as illustrated in Equation (4):

$$\widehat{\tau }={\displaystyle \frac{{\sum }_{i=1}^n{\alpha }_i(Z_i)({UER}_i-{\widehat{m}}^{(-i)}(Z_i))({DID}_i-{\widehat{exp}}^{(-i)}(Z_i))}{{\sum }_{i=1}^n{\alpha }_i(Z_i){({DID}_i-{\widehat{exp}}^{(-i)}(Z_i))}^2}}$$

(4)

In this context, ${\alpha }_i\left(Z_i\right)$ denotes the adaptive weighting function; ${\widehat{m}}^{(-i)}\left(Z_i\right)$ is the predicted value of $\mathrm{UER}$ accounting for control variables; and ${\widehat{exp}}^{(-i)}\left(Z_i\right)$ represents the estimated propensity score.

3.2. Data and Variables

3.2.1. Explained Variable

The explained variable in this study is urban energy resilience (UER). Building upon prior analyses [60,61,62], an evaluation framework comprising 36 indicators is established, encompassing various subsystems and resilience capacities as detailed in

Table 1. Indicator system for evaluating urban energy systems’ resilience.

Subsystem	Resilience Capacity	Measurement	Direction
Energy subsystem	Bearing capacity	Consumption volatility	−
		Supply stability	+
		Structural diversity	+
	Recovering capacity	Supply chain resilience	+
		Infrastructure support capability	+
		Institutional adjustment capacity	+
	Learning capacity	Transformation capabilities	+
		Talent development and skill enhancement	+
		Technological innovation	+
	Conversion capacity	Structural optimization	+
		Informatization and digitization	+
		Institutional guidance and infrastructure upgrade	+
Context subsystem	Bearing capacity	Social and ecological affordability	+
		Economic diversity and resource allocation	+
		Cultural identity and community resilience	+
	Recovering capacity	Index of economic recovery capacity	+
		Resource allocation efficiency	+
		Policy and infrastructure support	+
	Learning capacity	Space utilization and infrastructure upgrades	+
		Innovation and technological capabilities	+
		Policy innovation and governance learning capacity	+
	Conversion capacity	Resource efficiency conversion capabilities	+
		Fiscal flexibility and resource allocation capacity	+
		International cooperation and openness	+
Government subsystem	Bearing capacity	Ability to identify and analyze risks	+
		Digital science decision-making capabilities	+
		Resource allocation capabilities	+
	Recovering capacity	Emergency response capability	+
		Restore the ability to develop a strategy	+
		Restore efficiency	+
	Learning capacity	Ability to respond and adjust policies	+
		Information transparency and communication skills	+
		Community engagement and feedback capabilities	+
	Conversion capacity	Transformative capacity for sustainable development	+
		Process optimization capabilities	+
		Structural adjustment capacity	+

Note: The “+” sign represents a positive indicator, where a higher value corresponds to higher resilience. The “−” sign represents a negative indicator, where a higher value corresponds to lower resilience.

. Adhering to principles of objectivity and comprehensiveness, the entropy method is employed to calculate resilience scores. Initially, this method computes four capacity values for each subsystem within a province’s energy system. These capacities are then aggregated to derive subsystem resilience scores, which are subsequently combined to obtain the overall UER score for each province. The dataset encompasses panel data from 31 provinces spanning 2016 to 2023.

3.2.2. Core Explanatory Variable

The core explanatory variable is a dummy variable indicating policy implementation ($DID$). Specifically, national-level comprehensive reform pilot zones for sci-tech finance are utilized as proxies for digital finance policies. This substitution is justified by the analogous roles both sci-tech finance and digital finance play in fostering innovation and optimizing financial resource allocation. Although there are subtle differences in their conceptual focus and implementation scope, the policy objectives and developmental aspirations of sci-tech finance pilot zones closely mirror those of digital finance reforms, thereby significantly promoting the advancement of digital finance. The variable $DID$ is operationalized as the interaction term between the policy dummy ($treat$) and the post-implementation dummy ($post$).

Following methodologies from previous research [63,64], and considering potential spatial spillover effects of digital finance policies, the policy influence is extended to the provincial level. Within a province, if any city implements the digital finance policy, the treatment indicator for the entire province is set to 1. Moreover, acknowledging the preparatory phase required for national-level comprehensive financial reforms, the policy implementation year is advanced by one year, consistent with prior studies [65]. For instance, if the policy is enacted in Shandong Province in 2021, the $post$ variable is set to 1 starting from 2020 onward and 0 for preceding years.

3.2.3. Control Variables

In alignment with previous studies [66,67], the following control variables were incorporated: (1) Fiscal decentralization (Fisdec): calculated as the ratio of local government general budget revenue to local government general budget expenditure. (2) Education level (Edu): measured by the natural logarithm of education expenditure. (3) Marketization level (Market): represented by the ratio of total retail sales of consumer goods to regional GDP. (4) Social welfare level (Pub): defined as the ratio of the number of legal entities in public administration, social security, and social organizations to the total number of legal entities within the region. (5) Fiscal energy allocation (Eco): measured by the ratio of local government expenditures on resource exploration, electricity, and information to the local government general budget expenditure.

3.2.4. Mechanism Variables

The mediating variable is the development of innovative organizations, quantified by the regional innovation composite utility value from the China Regional Innovation Capability Evaluation Report 2023, covering 31 provinces from 2001 to 2023 [68]. This indicator effectively captures the innovation capabilities and development status of organizations within each region (Rd).

3.2.5. Data Sources

This study utilizes panel data from 31 provinces in China spanning from 2016 to 2023. Data sources include the China Statistical Yearbook, the China Energy Statistical Yearbook, and the China Environmental Statistical Yearbook, among others. Missing values were addressed using linear interpolation. Descriptive statistics for all variables are presented in

Table 2. Descriptive statistics.

Variable Type	Variables	N	Mean	SD	Max	Min
Explained variable	UER	248	0.218	0.193	0.889	0.0480
Core explanatory variable	DID	248	0.0650	0.246	1	0
Control variables	Fisdec	248	0.452	0.188	0.926	0.0690
	Edu	248	16.31	0.721	18.17	14.44
	Pub	248	0.0820	0.0640	0.533	0.0120
	Eco	248	0.0280	0.0180	0.119	0.00700
	Market	248	0.132	0.0460	0.292	0.0530
Mechanism variables	Rd	248	0.217	0.209	1	0

.

4. Results

4.1. Parallel Trend Test

A fundamental assumption when applying the multi-period difference-in-differences (DID) model is that the trajectories of urban energy resilience for the experimental and control groups are parallel prior to the policy intervention. Consistent with previous studies [69], this research employs an event study methodology to test the parallel trends assumption:

$${\displaystyle {UER}_{it}={\beta }_0+\sum _{k⩾-5,k\ne -1}^3{\alpha }_k{DID}_{it}^k+{\beta }_1{Z}_{it}+{\mu }_i+{\gamma }_t+{\epsilon }_{it}}$$

(5)

In Equation (5), ${DID}_{it}^k$ denotes dummy variables for the experimental group in the k-th year before and after the policy intervention, with all dummy variables for the control group set to 0. This study primarily examines trend changes before and after the policy implementation, with statistical significance assessed at the 95% confidence level. The vertical axis denotes the estimated coefficients, ${\alpha }_k$, while the horizontal axis corresponds to the timeline relative to the policy intervention.

Visually, as shown in Figure 4a, the estimated coefficients for the pre-intervention periods fluctuate around zero. Statistically, the coefficients ${\alpha }_k$ for all pre-intervention periods (k = −5 to k = −2) are not statistically significant at the 95% confidence level, as their confidence intervals clearly contain zero. This lack of a significant pre-existing trend differential between the two groups provides strong evidence that the parallel trends assumption holds for our model, thus validating the reliability of our DID estimation.

4.2. Benchmark Result

Table 3. Benchmark regression results.

	(1)	(2)	(3)	(4)	(5)
	UER	UER	UER_E	UER_B	UER_G
DID	0.0215 *	0.0332 ***	0.0280 ***	0.0358 **	0.0440 **
	(0.0123)	(0.0124)	(0.0105)	(0.0142)	(0.0200)
Control variables	N	Y	Y	Y	Y
City FEs	Y	Y	Y	Y	Y
Year FEs	Y	Y	Y	Y	Y
_cons	0.2674 ***	2.4325 ***	2.1739 ***	2.5640 ***	2.8917 ***
	(0.0063)	(0.6556)	(0.5549)	(0.7495)	(1.0523)
N	248	248	248	248	248

Note: The robustness standard errors, *, **, ***, in brackets represent the significance levels of 10%, 5% and 1%, respectively.

reports the estimated effects of digital finance policies on overall urban energy resilience, as well as the resilience of the energy, context, and government subsystems. Columns (1) and (2) present the regression results without and with control variables, respectively. Column (2) reveals a DID coefficient of 0.0332, indicating that digital finance policies improved overall system resilience by 3.32% at the 1% significance level, thus supporting H1. Columns (3), (4), and (5) report the effects on the energy, context, and government subsystems, with average resilience improvements of 2.80%, 3.58%, and 4.40%, respectively. These results were statistically significant at the 1%, 5%, and 5% levels, respectively, thereby validating H1a, H1b, and H1c.

4.3. Robustness Check

4.3.1. Placebo Test

The comparability between the experimental and control groups represents a crucial assumption in the difference-in-differences (DID) model. This study conducts a placebo test following the method outlined in [70,71]. Figure 4b displays the results of the placebo test using a randomly assigned experimental group, whereas Figure 4c presents the two-way random placebo test. In the two-way random placebo test, this study begins by randomly selecting an equal number of provinces to form the experimental group. Next, the timing of policy implementation is randomly assigned to each randomly selected experimental group. A dummy variable representing the policy is then constructed and used as the independent variable in the regression analysis. This process is repeated 1000 times to yield estimated coefficients for the false policy shocks and their corresponding p-values, which are subsequently compared with the baseline regression coefficients. As shown in Figure 4b,c, the results of the placebo test provide strong evidence for the robustness of our baseline findings. The distribution of the coefficients’ randomly assigned policy shocks is tightly centered around zero, and the majority of their corresponding p-values exceed the 0.1 significance level. Crucially, the estimated coefficient from our baseline regression ($a_1$ = 0.0332) lies far in the tail of this simulated distribution, clearly distinguishing itself from the randomly generated outcomes. This combination of results effectively rules out the possibility that our main findings are driven by unobserved confounding factors, which lends strong support to the credibility of our conclusion that digital finance policies exert a significant positive impact on urban energy resilience.

To account for potential random factors, this study performs a parallel trend placebo test. The results are presented in Figure 4d, where the gray curve represents the estimated results from the placebo test, evenly distributed around the y = 0 line. The black curve represents the baseline regression results, positioned above the majority of the placebo estimates and not overlapping with the gray curve, indicating that the observed positive effect of the policy on overall system resilience is not attributable to random factors.

4.3.2. Bacon Decomposition

This study addresses the issue of negative weights and bias in two-way fixed effects (TWFE) estimators, which arise from inter-group and time-dimensional heterogeneity due to staggered policy implementation in multi-period DID models. Following previous research, we assess the degree of bias in multi-period DID estimation within the TWFE framework [72,73]. The results are presented in

Table 4. Bacon decomposition results.

	(1)	(2)
	Beta	Total Weight
Early_v_Late	0.0031781953	0.0079840324
Late_v_Early	−0.0249783173	0.0059880239
Never_v_timing	0.0219318685	0.9860279437

. Among the groups influencing the TWFE estimated coefficients, the “good control group” accounts for 98%, whereas the “bad control group” contributes only 0.5%. These findings suggest that the bias introduced by the multi-period DID model in estimating the effect of digital finance policies is relatively small, thus confirming the robustness of the baseline regression results [74].

4.3.3. Double Machine Learning Testing

Double machine learning techniques are employed to mitigate estimation bias resulting from limited control variables [75]. To enhance the robustness of the results, this study utilizes double machine learning for testing, applying three configurations: Sample 1:4 (LassoIC regression), Sample 1:9 (LassoIC regression), and Sample 1:9 (RidgeCV regression). The results, displayed in

Table 5. Robustness test results.

	(1)	(2)	(3)	(4)	(5)	(6)
	1:4 LassoIC	1:9 LassoIC	1:9 RidgeCV	Green Finance Pilot	Public Data Policy	Inclusive Finance Reform
DID	0.0251 **	0.0264 **	0.0367 ***	0.0347 ***	0.0323 **	0.0286 **
	(0.0114)	(0.0108)	(0.0109)	(0.0125)	(0.0125)	(0.0130)
gfr				0.0259
				(0.0174)
opd					−0.0091
					(0.0077)
fir						0.0206
						(0.0177)
Control variables	Y	Y	Y	Y	Y	Y
City FEs	Y	Y	Y	Y	Y	Y
Year FEs	Y	Y	Y	Y	Y	Y
_cons	−0.0008	−0.0008	−0.0009	2.6502 ***	2.3224 ***	2.4452 ***
	(0.0025)	(0.0023)	(0.0023)	(0.6698)	(0.6615)	(0.6551)
N	248	248	248	248	248	248

Note: The robustness standard errors, **, ***, in brackets represent the significance levels of 5%, and 1%, respectively.

columns (1), (2), and (3), demonstrate that irrespective of the machine learning algorithm or sample ratio employed, the policy’s effect on urban energy resilience remains significantly positive.

4.3.4. Excluding Other Policies Interference Test

During the sample period, this study also accounts for the potential concurrent effects of other policies on urban energy resilience, alongside the net effect of digital finance policies. To address potential interference, three additional policy dummy variables are included in the model: “national green finance reform pilot zone” to account for green incentives [76], “public data open policy” to control for information enhancement [77], and ”inclusive finance comprehensive reform” to control for traditional financial development [78]. As shown in Table 5, columns (4), (5), and (6) of the regression results table, even after accounting for these policy factors, the positive effect of digital finance policies on urban energy resilience remains significant, thus confirming the robustness of the baseline regression results.

4.3.5. PSM-DID

To mitigate endogeneity resulting from self-selection bias and to address sample differences prior to policy implementation, this study employs the propensity score matching–difference-in-differences (PSM-DID) methodology for re-estimation [79]. Three matching techniques were applied: 1:1 nearest neighbor matching, kernel matching, and radius matching. The results, presented in columns (1), (2), and (3) of

Table 6. PSM-DID results.

	(1)	(2)	(3)
	UER	UER	UER
DID	0.0392 *	0.0498 **	0.0393 **
	(0.0204)	(0.0242)	(0.0199)
Control variables	Y	Y	Y
City FEs	Y	Y	Y
Year FEs	Y	Y	Y
_cons	3.0821 ***	3.1739 ***	3.0815 ***
	(0.6810)	(0.6730)	(0.6793)
N	222	222	224

Note: The robustness standard errors, *, **, ***, in brackets represent the significance levels of 10%, 5% and 1%, respectively.

, demonstrate that, irrespective of the matching method employed, the policy has a significantly positive effect on urban energy resilience.

4.3.6. Other Robustness Tests

To further validate the robustness of the results, a series of additional tests were performed, with the corresponding outcomes presented in

Table 7. Other robustness test results.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Eliminate Special Samples	Adjust Sample Time	Exclude Extreme Outliers	Add Control Variables	Lagged Explanatory Variable	Lagged Dependent Variable	Instrumental	Variable Method
DID	0.0414 ***	0.0396 ***	0.0455 ***	0.0278 **		0.0319 **		0.0359 ***
	(0.0120)	(0.0135)	(0.0130)	(0.0126)		(0.0136)		(0.0132)
L.DID					0.0293 **
					(0.0117)
Trend × id				−0.0002 **
				(0.0001)
Ivpost							0.0175 ***
							(0.0005)
Control variables	Y	Y	Y	Y	Y	Y	Y	Y
City FEs	Y	Y	Y	Y	Y	Y	Y	Y
Year FEs	Y	Y	Y	Y	Y	Y	Y	Y
_cons	2.4826 ***	2.4589 ***	3.0053 ***	9.2556 ***	2.8098 ***	2.4217 ***
	(0.5841)	(0.6790)	(0.7190)	(3.4815)	(0.6480)	(0.6858)
F-value							1052.98
N	216	217	248	248	217	217	248	248

Note: The robustness standard errors, **, ***, in brackets represent the significance levels of 5%, and 1%, respectively.

. First, to account for the potential impact of special political status, the four municipalities (Beijing, Tianjin, Shanghai, and Chongqing) were excluded from the sample, and the model was re-estimated. The results are reported in column (1). Second, to mitigate the potential influence of the COVID-19 pandemic, the year 2020 was excluded from the analysis, and the model was re-estimated, as shown in column (2). Third, all continuous variables were winsorized at the 1% level on both ends to address extreme outliers, with the results presented in column (3). Finally, to control for changes in provincial characteristics over time, a province-specific time trend term was incorporated into the model, with the outcomes displayed in column (4). These additional robustness checks further substantiate the reliability of the baseline regression results.

4.3.7. Other Endogeneity Tests

To address potential endogeneity concerns, this study performs the following tests, with results presented in Table 7. First, to mitigate potential selection bias from pilot provinces, the explanatory variable was lagged by one period, and the results are presented in column (5). Second, to account for potential endogeneity arising from the delayed effects of policy, the dependent variable was lagged by one period, with results presented in column (6). Third, to address the possibility that the experimental group may not be entirely exogenous, this study employs the number of post offices per million people in 1984, interacted with year dummies, as an instrumental variable in a two-stage least squares (2SLS) regression analysis [80]. Post offices, as integral components of the postal system, facilitate communication and financial services, which enhance financial activity and are directly relevant to the development of digital finance policies, thereby fulfilling the relevance requirement for instrumental variables. At the same time, the number of post offices in 1984 does not directly affect current urban energy resilience, thereby satisfying the exogeneity condition. The results are presented in columns (7) and (8). In column (7), the ivpost coefficient is positive and statistically significant at the 1% level, indicating a robust correlation between the number of post offices and digital finance policy. The F-statistic, which far exceeds 10, eliminates concerns of weak instruments. In column (8), the DID coefficient remains significantly positive at the 1% level. These findings confirm that the baseline estimation results remain robust and reliable after addressing potential endogeneity.

5. Further Analysis

5.1. Mechanism Analysis

This study empirically examines whether the policy indirectly enhances the resilience of the urban energy system and its three subsystems by promoting the development of innovation organizations (rd) as a mediator, as theorized in the previous section [81]. The results, obtained through recursive equations, are presented in

Table 8. Mechanism analysis results.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Rd	UER	Rd	UER_E	Rd	UER_B	Rd	UER_G
DID	0.0346 ***	0.0279 **	0.0346 ***	0.0233 **	0.0346 ***	0.0301 **	0.0346 ***	0.0371 *
	(0.0119)	(0.0126)	(0.0119)	(0.0107)	(0.0119)	(0.0144)	(0.0119)	(0.0203)
Rd		0.1539 **		0.1358 **		0.1625 *		0.1983 *
		(0.0725)		(0.0613)		(0.0830)		(0.1168)
Control variables	Y	Y	Y	Y	Y	Y	Y	Y
City FEs	Y	Y	Y	Y	Y	Y	Y	Y
Year FEs	Y	Y	Y	Y	Y	Y	Y	Y
N	248	248	248	248	248	248	248	248

Note: The robustness standard errors, *, **, ***, in brackets represent the significance levels of 10%, 5% and 1%, respectively.

.

As shown in columns (1), (3), (5), and (7) of Table 8, the DID coefficients are significantly positive, indicating that the policy significantly promotes the development of innovation organizations. Furthermore, in columns (2), (4), (6), and (8) of Table 8, the rd coefficients are significantly positive, although the DID coefficients are smaller compared to those in the baseline regression. These findings suggest that digital finance policies indirectly enhance the overall resilience of the system, as well as the resilience of its three subsystems, through the promotion of innovation organizations. This supports H2, 2a, 2b, and 2c.

5.2. Machine Learning Analytics

5.2.1. Counterfactual Forecasting

As shown in Figure 5a, under the baseline scenario, the values of control variables prior to the policy implementation remain unchanged. For the post-policy period, the control variables are forecasted using their growth trends from 2016 to the pre-policy years, alongside the control variables specific to each province. With the predicted control variables, random forest algorithms are employed to estimate the trend values of urban energy resilience for each province. The policy effect of digital finance on urban energy resilience is quantified by subtracting the baseline scenario values from the resilience values of pilot provinces under the policy shock scenario [82].

Taking Shandong Province, the first to implement the policy, as an example, Figure 5b shows that the policy’s effect is not significant in its first year of implementation (2020). A significant positive effect is observed in 2021, indicating a time lag in the policy’s impact. This lag can be attributed to the complexity of systemic adjustments and the involvement of multiple stakeholders, which cause fluctuations during the transition period. Furthermore, the recurrence of the COVID-19 pandemic in Shandong province hinders both the policy’s implementation and digital finance development, resulting in an insignificant effect in 2022. By 2023, with the subsiding of the pandemic and the resumption of economic vitality, the policy yields significant positive outcomes.

5.2.2. Causal Forest

This study utilizes causal forest estimation to capture the nonlinear relationships between variables and estimate the average treatment effect of the policy [83]. As the number of trees in the causal forest increases, the accuracy and stability of the estimates improve accordingly. Columns (1), (2), and (3) in

Table 9. Causal forest results.

	(1)	(2)	(3)
	UER	UER	UER
DID	0.007 **	0.007 **	0.007 **
	(2.017)	(1.983)	(1.992)
Trees	5000	10,000	15,000
N	248	248	248

Note: The robustness standard errors, ** in brackets represent the significance levels of 5%.

 present the regression results obtained using 5000, 10,000, and 20,000 trees, respectively. The results consistently reveal a significant positive impact of digital finance policies on the resilience of urban energy systems.

Furthermore, causal forests can display the conditional average treatment effect (CATE) for each individual causal tree [84]. Figure 6a illustrates the distribution of treatment effects on overall system resilience. The distribution shows that individual treatment effects fluctuate between 0.003 and 0.010, with the majority concentrated in the 0.006 to 0.008 range. This distribution suggests that digital finance policies contribute to enhancing the resilience of urban energy systems. However, the effects also exhibit considerable variability across provinces, indicating significant heterogeneity in the policy impacts.

Expanding on this analysis, the study identifies the relative importance of control variables in the causal analysis [85]. As shown in Figure 6b, the variables ‘market’ and ‘eco’ emerge as key factors influencing the effectiveness of digital finance policies. Heterogeneity analyses are specifically conducted based on these two variables.

As shown in Figure 6c, the relationship between marketization (market) and the effects of digital finance policies follows a U-shaped trend [86,87,88]. This finding indicates that, when marketization is not yet fully mature, excessive administrative control and market barriers limit the technological advantages and financial flexibility of digital finance, thereby partially offsetting the policy benefits. Conversely, when the level of marketization is either very low or very high, the input and resource integration capabilities of digital finance are more effectively leveraged. In regions with low levels of marketization, digital finance compensates for the insufficient coverage of traditional financial systems, offering energy enterprises essential support in risk pricing and data analysis. In regions with high levels of marketization, a robust credit system and a competitive market environment optimize the efficiency of digital finance, enabling it to play a more substantial role in areas such as energy investment and innovation research and development.

Additionally, as shown in Figure 6c, the interplay between eco and digital finance policies manifests a pronounced nonlinear relationship, characterized predominantly by a U-shaped trajectory, albeit with a less conspicuous ascending phase [89,90]. Specifically, under conditions of low fiscal energy allocation ratios, digital finance policies exert a markedly significant impact on augmenting the resilience of urban energy systems. This suggests that in regions experiencing insufficient fiscal investment, digital finance effectively mitigates the limitations inherent in traditional financial mechanisms by delivering adaptable funding support and advanced technical services, thereby bolstering system resilience. As the fiscal energy allocation ratio escalates, the efficacy of digital finance policies progressively diminishes. Intermediate levels of fiscal investment fail to fully harness the technological advantages offered by digital finance, while administrative regulations and market barriers impede the seamless integration of resources. Upon further increases in the fiscal energy allocation ratio, the impact of digital finance policies experiences a partial resurgence. However, this rebound is potentially constrained by factors such as market saturation and the efficiency of policy implementation, rendering the upward trend less pronounced compared to the initial decline. In summary, the fiscal energy allocation ratio plays a pivotal role in mediating the effectiveness of digital finance policies in enhancing urban energy resilience. Consequently, policy formulation must holistically consider the extent of fiscal investment to optimally leverage the potential of digital finance [91].

To further probe the policy’s heterogeneous effects, we extend our causal forest analysis to examine whether the treatment effect varies across regions with different initial digital finance levels and city sizes.

First, we explore the heterogeneity based on the initial digital finance level (IDFL), using the median of the 2016 Peking University Digital Financial Inclusion Index to divide the sample into high- and low-level groups. The results, visualized in Figure 6d, reveal a significant difference. The CATE is stronger in the low-IDFL group compared to the high-IDFL group. This statistically significant difference suggests a “catch-up” effect, where the policy provides greater marginal benefits to regions with less developed financial infrastructure.

Furthermore, we investigate the role of city size, classifying provinces based on the median of their 2016 resident population. As shown in Figure 6d, the analysis yields no significant difference in policy effects between small-size and large-size regions. This finding indicates that the positive impact of digital finance policies on urban energy resilience is quite universal and does not depend on the region’s population or economic scale.

6. Conclusions, Discussion, and Policy Implications

6.1. Research Conclusions

In the context of dual carbon goals and the potential energy security challenges associated with the energy transition, the development of digital finance is expected to play a crucial role in supporting sustainable urban development. From a theoretical standpoint, this study integrates complex systems theory to examine how digital finance, as an external shock, directly and indirectly impacts urban energy resilience through the mediating role of innovative organizations. Empirically, based on panel data from 31 Chinese provinces (2016–2023), this study employs a multi-period difference-in-differences (DiD) model to test the causal relationship, with findings validated through robustness checks, mediation effect models, and machine learning techniques.

This study successfully provides definitive answers to its core questions, with all proposed hypotheses being empirically supported. First, digital finance development significantly enhances urban energy resilience. The overall system resilience improves by 3.32% (significant at the 1% level), with corresponding enhancements in the energy (2.80%), context (3.58%), and government (4.40%) subsystems, a result that confirms the direct effect hypotheses (H1, H1a, H1b, and H1c) and remains robust across multiple robustness checks. Second, this positive impact is partially mediated by the development of innovative organizations, thereby validating the indirect effect hypotheses (H2, H2a, H2b, and H2c). This confirms that digital finance strengthens resilience by fostering regional innovation capacity. Third, a machine learning analysis reveals that the policy’s impact exhibits nonlinear relationships and significant heterogeneity. The policy’s positive effects are notably stronger in regions with lower initial levels of digital finance. In contrast, the regional scale does not significantly alter the policy’s impact, suggesting its broad applicability. Furthermore, the policy’s effectiveness exhibits a U-shaped relationship with both marketization and fiscal energy allocation.

6.2. Discussion

To highlight the uniqueness and generalizability of our findings, this study discusses the findings in relation to the existing academic literature, examines the applicability of the methodological framework in other contexts, and explores the broader international implications of the conclusions.

First, our primary finding that digital finance significantly enhances urban energy resilience represents a substantial theoretical contribution to the emerging literature on financial innovation and energy system transformation. While existing studies have examined green finance [92] and supply chain digitization [93] to explore their importance, our research pioneers an integrated framework that captures the synergistic effects of financial digitalization. By grounding our analysis in complex adaptive systems theory, we reveal the multi-layered, nonlinear dynamics arising from the coupling and interaction mechanisms between digital finance—conceptualized as an external shock—and urban energy systems. Furthermore, while traditional financial intermediation theory primarily focuses on capital allocation and risk management [94,95,96], this study identifies “innovative organizations” as a critical mediating mechanism. In the digital finance context, these organizations serve as transformation agents that actively convert financial resources into energy resilience. This theoretical innovation not only bridges the gap between the financial economics and innovation systems literature but also provides fresh insights into how multi-agent coordination and adaptive evolution processes enhance resilience in complex urban systems.

Second, this study introduces a significant methodological innovation by integrating the multi-period DID model with machine learning techniques to untangle complex policy effects. Diverging from the traditional econometric methods commonly employed in the existing literature [97,98], our integrated framework synthesizes causal inference models, mediation models, and machine learning, thereby ensuring the rigor of causal analysis while enhancing the explanatory power for complex economic phenomena. The adaptability and utility of this framework have been validated in studies across diverse national contexts [59,99,100], equipping researchers and policymakers with a robust toolkit to assess policy impacts, conditional on the availability of suitable panel data and quasi-natural experimental settings. Ultimately, this contribution not only advances the scholarship on digital finance and energy resilience but also offers a replicable analytical paradigm for evaluating policy effects within complex economic systems.

Finally, China’s experience in leveraging digital finance to enhance energy resilience offers valuable insights for global energy transitions. As the world’s largest digital finance market and energy consumer, China’s policy experiments at the intersection of financial digitalization and energy systems demonstrate both frontier innovation and replicability. Our findings are not confined to China’s specific context—research from other transition economies with similar energy structures [101,102,103,104,105] likewise indicates that the positive nexus between financial digitalization and energy resilience transcends specific institutional settings, representing a potentially universal principle. Developing countries commonly face the “environment-energy-economy” trilemma, and digital finance can serve as a critical enabler for addressing this complex challenge, providing a pathway that balances energy security, environmental sustainability, and economic development. These insights hold significant relevance not only for emerging economies and transition countries—including post-communist states in Central and Eastern Europe with comparable coal dependency ratios—but also offer valuable lessons for developed economies seeking to optimize their financial architectures for net-zero transitions. Cultivating a robust digital finance ecosystem may constitute a key strategic lever for accelerating global energy transitions, warranting serious consideration from policymakers across diverse economic contexts.

6.3. Policy Recommendations

Based on the conclusions presented above, this study offers the following policy recommendations.

First, it is crucial to strengthen the development of digital finance infrastructure and enhance policy coordination. The Chinese government should continue to invest in and enhance digital finance infrastructure, including big data, blockchain, and cloud computing, to promote efficient information sharing among various stakeholders in the energy system. Simultaneously, it is essential to establish robust regulatory frameworks and risk prevention mechanisms within digital finance to ensure financial stability, ensure compliance with capital flow regulations, and expand financial services while fostering further financial innovation. Strengthening the linkage between digital finance policies, fiscal subsidies, and digital regulation will provide substantial support for the low-carbon and intelligent upgrade of energy systems.

Second, accelerate the development of innovation organizations and promote the deep integration of digital finance with urban energy systems. Given the crucial mediating role of innovation organizations, the government should provide targeted policy support, such as tax incentives and talent development subsidies, to technology SMEs and startups and promote collaboration between industry and research institutions, facilitating technology incubation and implementation via digital finance platforms. Concurrently, it is important to encourage collaboration between financial institutions and universities or research institutes to conduct interdisciplinary research in areas such as smart energy management, energy storage, and distributed power. Digital finance resources should be strategically directed into key green and low-carbon sectors, generating endogenous momentum to enhance energy system resilience.

Third, we suggest implementing differentiated policies tailored to local conditions. Our finding that the policy yields greater benefits in regions with lower initial digital finance levels suggests that strategic priority should be given to these less-developed areas to maximize the policy’s ‘catch-up’ potential and advance inclusive development. Conversely, the policy’s consistent effect across different regional scales indicates its broad applicability, requiring less tailoring based on population or economic size. For factors like marketization and fiscal allocation, where effectiveness follows a U-shaped pattern, policymakers should be particularly attentive to regions in the intermediate range, potentially implementing complementary measures to overcome market barriers or administrative inefficiencies that might otherwise dampen the policy’s impact.

6.4. Limitations and Future Research Directions

While this study makes significant contributions to theoretical construction, empirical analysis, and policy recommendations, several limitations remain that simultaneously indicate important avenues for future research.

First, this study is constrained by data granularity and an insufficient exploration of micro-level mechanisms. Due to data availability limitations, this research primarily relies on provincial-level panel data. Although this approach effectively reveals macro-level policy effects, it may obscure significant heterogeneity among cities and enterprises within provinces, limiting our ability to capture micro-level behavioral response mechanisms. Future research could leverage city-level or firm-level data to employ multi-level analytical frameworks, thereby exploring the micro-foundations of how digital finance influences energy resilience and revealing heterogeneous response patterns across cities of different scales and enterprises of various types.

Second, the identification of transmission mechanisms requires further refinement. While this study identifies the mediating role of innovative organizations, the pathways through which digital finance affects urban energy resilience are likely more diverse and complex. Future research could construct more comprehensive theoretical frameworks and employ structural equation modeling or machine learning methods to systematically examine the relative importance of multiple transmission mechanisms and their interaction effects.

Third, the external validity of research findings warrants further verification. This study focuses on the Chinese context, and while it provides valuable insights for emerging economies, its applicability in developed countries or under different institutional settings remains to be validated. Future research could conduct cross-country comparative analyses, selecting countries at different developmental stages with diverse institutional environments and energy structures as research subjects. This would test the universality of mechanisms through which digital finance promotes urban energy resilience and identify key institutional factors influencing its effectiveness, thereby providing more robust theoretical foundations for constructing global financial support systems for energy transition.