Research on the Nonlinear and Spatial Effects of Digital Financial Information Flow on Industrial Structure Upgrading

Abstract

In the digital economy era, digital inclusive finance represents a paradigmatic reconstruction of key economic information flows. This study integrates multi-source panel data of 27 cities in the Yangtze River Delta from 2011 to 2023. By constructing an economic geography composite spatial weight matrix and a nonlinear spatial panel model, this study analyzes the impact of the diffusion of digital inclusive financial information on industrial structure upgrading. The results show that: (1) digital financial inclusion exerts a significant direct effect and spatial spillover effect on industrial structure; (2) the local effect exhibits a “U-shaped” curve with an accelerating characteristic on the right side; the spatial spillover effect demonstrates an “inverted U-shaped” curve, revealing the transformation law and threshold effect of the diffusion and aggregation of digital financial information benefits; (3) digital payment and digital credit constitute the core information flows driving the coordinated upgrading of industries; and (4) entrepreneurial activity exerts a partial mediating effects, and exhibits a spatial mediating effect, while the technological innovation only demonstrates a significant local mediating effect. The findings provide quantitative evidence to support the optimization of the digital financial information ecosystem and the realization of coordinated industrial upgrading in the Yangtze River Delta.

1. Introduction

Industrial structure upgrading serves as a critical engine for high-quality economic development and a core indicator of the transformation between old and new growth drivers. The traditional growth model driven by factor input and scale expansion is no longer viable. The coexistence of structural slowdown and overcapacity has compelled regions to expedite the adjustment of industrial structures. Big data, cloud computing, and information and communication technologies are deeply penetrating various sectors. As a result, massive transaction data, credit data, and business operation data have gradually undergone digital precipitation and cross-entity circulation, reshaping the underlying logic of financial resource allocation and industrial evolution. Pioneer regions such as the Yangtze River Delta have taken the lead in entering the stage of industrial evolution marked by “servitization” and “high-tech orientation”. This process is inseparable from the efficient support of modern service sectors including finance and information services.

However, constrained by structural predicaments such as information asymmetry and low efficiency in credit information transmission, the traditional financial system struggles to accurately match the financial service demands arising from industrial structure upgrading. In the era of the deep integration of digital economy and big data application, the digitization and inclusion of financial services essentially represent a paradigm shift in the flow of critical economic information. As a new form of information infrastructure, digital inclusive finance integrates digital finance with traditional finance through the strategic leverage of information and communication technologies (ICT) to enhance financial inclusiveness [1]. Endowed with diverse network connectivity advantages [2], digital financial inclusion offers a more efficient communication space than traditional finance. Capable of deep integration with various industries to advance industrial supply-side reform, digital financial inclusion is recognized as a critical engine for industrial structure optimization [3]. Global research and practice have also corroborated this trend: studies conducted in Africa and Asia have revealed that digital finance exerts a significant and positive impact on inclusive growth or industrial upgrading [4,5]. Financial inclusiveness is critical to sustainable development and economic growth in the Asia-Pacific region, and it exhibits a nonlinear effect [6]. This nonlinear pattern has also been validated in China. Studies have demonstrated that digital inclusive finance significantly facilitates the upgrading of rural industrial structures, with the presence of a single threshold effect [7,8].

However, traditional studies exploring the relationship between digital inclusive finance and industrial structure upgrading typically rely on the implicit assumption of inter-regional independence and homogeneity, employing ordinary least squares for estimation. Such approaches overlook the pervasive spatial correlations among economic units, which may introduce biases into the estimation results and hinder a reasonable explanation of inter-regional interaction mechanisms [9]. In reality, amid the growing convenience of information flow, the connections between economic, technological, and financial information across regions have become unprecedentedly tight. Regions that are geographically proximate or share similar economic characteristics often demonstrate stronger spatial interaction and information dependency. Since the 1970s, the emergence of spatial econometrics has laid a methodological foundation for addressing such issues. Its core proposition holds that an economic–geographical phenomenon or attribute value in a given spatial unit exhibits spatial autocorrelation with the identical phenomenon or attribute value in its adjacent spatial units [10]. This theoretical lens is particularly critical for understanding the interactive dissemination of digital financial information in both cyberspace and geographical space. The spatial spillover effect of such information may be a key mechanism driving the coordinated upgrading of regional industries. Prior studies have employed spatial models to identify the local “U-shaped” and peripheral “inverted U-shaped” nonlinear patterns [11]. While spatial econometric methods have been incorporated into this research domain, three critical gaps remain unaddressed in the existing research landscape, which this study seeks to tackle. First, Further refinement is needed regarding the integrated analysis of the local U-shaped threshold acceleration effect and the spatial inverted U-shaped dual nonlinearity, as well as the identification of critical values. Second, digital payments, digital credit, and digital insurance differ in information flow attributes, yet the heterogeneous spatial spillover patterns of these three services remain under-explored. Additionally, empirical evidence is lacking regarding whether entrepreneurial dynamism and technological innovation information input function as information carriers to exert mediating effects, particularly the existence of spatial mediating effects. Third, How the basic conditions such as the penetration rate of mobile Internet affect the release of the information dividend of digital finance has not been fully discussed. These gaps constrain the advancement of theoretical frameworks and the precision of policy design, necessitating systematic investigation within the context of the Yangtze River Delta integration initiative.

Against this backdrop, this study adopts the perspectives of information geography and spatial analysis, drawing on multi-source panel data of 27 cities in the Yangtze River Delta spanning 2011 to 2023. This paper constructs a composite spatial weight matrix that integrates economic and geographical attributes, and employs spatial econometric models to systematically investigate the spatial spillover effects and mechanism of digital inclusive finance on industrial structure upgrading. The potential contributions of this study are primarily as follows: (1) Redefining digital inclusive finance as a form of regional digital information infrastructure, this study leverages the dual perspectives of information geography and spatial data analysis to characterize the diffusion patterns and spatial transmission mechanisms of digital financial information flows within the Yangtze River Delta urban agglomeration. This effort extends the application scope of information economics in the field of regional development research. (2) It distinguishes the heterogeneous spatial spillover patterns of three types of information flows—digital payment, digital credit, and digital insurance—and empirically validates the mediating effects of entrepreneurial dynamism and technological innovation information input, as well as the spatial transmission characteristics of these effects. This provides quantitative evidence to inform the formulation of differentiated coordinated development policies. (3) By constructing a nonlinear spatial panel econometric model integrated with a threshold model, this study systematically identifies the local U-shaped effect of digital inclusive finance on industrial structure upgrading and the spatial inverted U-shaped spillover effect. This provides empirical evidence for understanding the stage-specific characteristics of digital information benefit release and the attenuation pattern of cross-regional spillover effects.

The subsequent sections of this paper are organized as follows: Section 2 reviews the existing literature on digital inclusive finance and industrial structure upgrading; Section 3 elaborates on the spatial econometric model employed; Section 4 outlines variable selection and data sources;and presents and analyzes the empirical results; Section 5 draws conclusions and discusses key findings, while also summarizing the study’s limitations and proposing future research directions.

2. Literature Review and Research Hypotheses

2.1. Literature Review

In academic research on finance and industrial structure, early studies primarily focused on the direct impacts of traditional finance on industrial structure upgrading and macro-level correlation analyses, with two dominant perspectives: the “promotion thesis” and the “inhibition thesis”. Financial development is argued to facilitate industrial structure upgrading through functions such as resource allocation [12,13,14,15]. Conversely, excessive expansion of the financial sector—when it deviates from the real economy—exerts an inhibitory effect on industrial structure upgrading [16,17].

In the process of the development of the real economy, the traditional financial system, due to prominent issues of information asymmetry and low efficiency in the transmission of credit information, is unable to meet the demands of industrial structure upgrading for high-quality financial information services [18]. Against this backdrop, inclusive finance, which utilizes digital technology to break through the limitations of traditional finance, is regarded as an important path to alleviate the financial information gap and optimize resource allocation [19]. Xu Min and Zhang [20] used the Vector Autoregressive model (VAR) to study and found that there is a long-term equilibrium relationship between the development of inclusive finance and industrial structure upgrading. This conclusion reveals the deep connection between financial information elements and industrial transformation from a time series perspective. However, most of the above studies treat regions as independent entities, ignoring the cross-regional flow and diffusion effect of financial information in the spatial dimension, making it difficult to fully depict the spatial interaction characteristics of financial information services in the digital age.

In recent years, with the release and improvement of the Digital Inclusive Finance Index of Peking University, scholars have begun to systematically examine the relationship between digital inclusive finance and the development of the real economy based on this digital financial information carrier and by drawing on the analytical framework of information network diffusion theory. At the direct effect level, some studies have found a nonlinear relationship between digital inclusive finance and industrial structure upgrading [8,21,22], revealing the threshold characteristics of the release of digital financial information benefits [7]. At the indirect transmission mechanism level, digital inclusive finance indirectly drives the optimization and upgrading of the industrial structure by influencing income disparity, capital accumulation, consumer demand [21,23,24], as well as innovation and entrepreneurship vitality, human capital accumulation [25,26], and other information intermediary factors. In addition, a few studies have begun to introduce spatial econometric methods to examine the spatial information spillover effect of digital inclusive finance on industrial structure upgrading at the provincial level [27,28,29], providing an important reference for this study to explore the spatial diffusion law of digital financial information at the urban agglomeration scale from the perspective of information geography.

To summarize, the advancement of financial infrastructure, informatization, and digitalization has fostered a rich body of academic research on the relationship between financial development and industrial structure upgrading. However, existing studies still exhibit notable limitations in the following dimensions: First, the oversight of spatial information spillover mechanisms. Second, the coarse granularity of information measurement. Third, research on intermediary mechanisms—particularly spatial transmission mechanisms—remains insufficient. From the perspective of information diffusion and spatial interaction, this study takes the Yangtze River Delta urban agglomeration as the research scene, uses spatial econometric models and builds a composite spatial weight matrix integrating economic and geographical features to systematically examine the spatial spillover effect of digital inclusive finance on industrial structure upgrading and the specific local and spatial mechanisms. This study aims to reveal the diffusion rules and transmission heterogeneity of digital financial information among regions, providing empirical evidence and methodological references for understanding the regional synergy effect of digital inclusive finance and promoting the spatially balanced development of industrial structure upgrading.

2.2. Theoretical Hypothesis

2.2.1. The Impact of Digital Inclusive Finance on Industrial Structure Upgrading

Digital inclusive finance, characterized by broad information coverage, low access costs, and high transmission efficiency, is profoundly reshaping the patterns of financial information generation and diffusion. First, leveraging digital information technologies, digital inclusive finance efficiently matches financial service information with the financing needs of diverse entities. By optimizing the allocation efficiency of financial information factors [30], it provides stable capital support for the continuous innovation of enterprises, incentivizing enterprises to increase R&D investment and patent applications, thereby promoting the upgrading of industrial structure toward high-end and diversification [31,32]. Song and Jiang demonstrate that the digital economy’s role in driving industrial structure upgrading and rationalization is primarily mediated by technological innovation [33]. Second, through aggregating financial information, digital inclusive finance lowers the entrepreneurial and information thresholds for enterprises, enabling potential entrepreneurs to quickly access market opportunity information [34,35], which in turn improves the survival rate of newly registered enterprises. The entry of new enterprises enhances industrial diversity, fosters competition and division of labor, and compels existing enterprises to improve production efficiency and technological levels, thus driving the overall evolution of industries toward higher quality. Third, as an important component of the tertiary industry, the development of digital inclusive finance directly contributes to the scale expansion and structural upgrading of the service industry.

On the other hand, digital inclusive finance transcends the reliance of traditional finance on physical branches. By leveraging pre-existing inter-regional economic linkages and information networks, it facilitates the cross-regional flow and diffusion of financial resources, technological information, and digital services in the spatial dimension [36]. This, in turn, exerts a radiating and driving effect on industrial structure upgrading in neighboring regions, giving rise to a significant spatial spillover effect. Building on this, this paper proposes the following research hypotheses:

H1a. 

Digital inclusive finance exerts a significant impact on industrial structure upgrading, and this impact generates a positive information spillover effect through spatial information transmission channels.

H1b. 

Enterprise technological innovation information input plays a mediating role in the impact of digital inclusive finance on industrial structure upgrading.

H1c. 

Enterprise entrepreneurial activity functions as an information mediator in the relationship between digital inclusive finance and industrial structure upgrading.

2.2.2. Nonlinear Impact of Digital Inclusive Finance on Industrial Structure Upgrading

In the research on the relationship between digital inclusive finance and industrial structure upgrading, scholars have gradually focused on the possible nonlinear characteristics between the two. Lin Binghua found that there is a nonlinear relationship between the development of digital inclusive finance and industrial structure upgrading [37]. Yang et al. [7] and Duan et al. [38] further pointed out that the development of digital inclusive finance has bottlenecks and threshold effects. The cause of this nonlinear feature lies in the phased differences in the diffusion of financial information: in the early stage of development, due to the incomplete financial information infrastructure and limited application scenarios within the region, the information transmission efficiency of digital inclusive finance shows a marginal diminishing feature; when development crosses a specific threshold, the network effect and scale effect of information gradually emerge, and its promoting effect on industrial upgrading accelerates, forming a nonlinear influence trajectory [39,40].

In terms of spatial dimensions, the financial information flow carried by digital inclusive finance also presents a complex spatial diffusion pattern. On the one hand, relying on digital platforms, digital inclusive finance optimizes the allocation efficiency of financial information elements among regions, strengthens the financial information connection among regions, alleviates excessive competition for financial resources among regions, and positively drives the industrial structure adjustment of surrounding regions through promoting the spillover of investment information and the diffusion of technological innovation information [41,42]. On the other hand, due to the “siphon effect” of information agglomeration, digital inclusive finance may attract high-quality financial information and resources to high-level regions during its development, thereby exerting an inhibitory effect on the industrial structure upgrading of surrounding regions [11]. In addition, the information absorption capacity of surrounding regions themselves, such as the level of digitalization and the stock of human capital, will also significantly affect the intensity and direction of the spillover effect [43].

In summary, the diffusion of financial information flows carried by digital inclusive finance between local and regional areas is not a simple linear process, but rather exhibits complex nonlinear characteristics and spatial heterogeneity. Based on this, this paper proposes the following research hypothesis:

H2. 

The direct effect and spatial spillover effect of digital inclusive finance on industrial structure upgrading both present nonlinear characteristics, and this nonlinearity stems from the local accumulation threshold of the release of financial information benefits and the degree of improvement of financial infrastructure.

2.2.3. The Differential Effects of Financial Information Flows on Industrial Structure Upgrading

At the business level of digital inclusive finance, different types of financial services influence industrial structure upgrading through distinct information transmission channels:

First, digital payment drives industrial upgrading by enhancing transaction information efficiency. Relying on the rapidly developing third-party payment platforms, digital payment not only facilitates residents’ consumption but also guides the supply side to optimize its structure through the real-time recording and feedback of consumption information flow, thereby promoting industrial upgrading [23]. Meanwhile, digital payment has become the infrastructure for enterprises’ digital operation. By reducing information asymmetry in transactions among enterprises, minimizing friction in factor allocation, and accelerating the coordination of information flow and capital flow among agriculture, industry, and services, it significantly impacts industrial structure upgrading [44].

Second, digital credit alleviates financing constraints by optimizing the allocation of credit information. The convenience and low cost of digital inclusive finance enable more market entities to obtain credit support, effectively easing the financing constraints caused by credit information asymmetry [23,45]. By precisely identifying and evaluating credit information through digital platforms, digital credit lowers the threshold for credit granting, providing capital and information element support for technological upgrading in manufacturing, the development of modern services, and agricultural modernization [46], thereby promoting the evolution of industries towards higher levels.

Thirdly, digital insurance mitigates the risks of factor mobility by providing risk information guarantees. The diverse insurance tools offered by digital inclusive finance provide risk information guarantees for the mobility of labor factors between industries, reducing the uncertainty of factor allocation and facilitating efficient resource allocation guided by information [23], thereby creating a stable factor mobility environment for industrial structure upgrading.

To synthesize the above, the distinct business dimensions of digital inclusive finance—digital payment, digital credit, and digital insurance—carry transaction information flows, credit information flows, and risk information flows, respectively, and act on industrial structure upgrading through heterogeneous information transmission mechanisms. Concurrently, these financial information flows may generate heterogeneous spatial spillover effects in the spatial dimension by leveraging inter-regional economic linkages and information networks. Building on this, this paper proposes the following research hypothesis:

H3. 

The distinct business dimensions of digital inclusive finance exert heterogeneous spatial spillover effects on industrial structure upgrading, with significant heterogeneity across these dimensions.

3. Model Design

3.1. Benchmark Model

Drawing on the theoretical underpinnings established earlier, to conduct an in-depth analysis of the impact of digital inclusive finance development on regional industrial structure upgrading, this paper constructs the following benchmark panel model:

$${\mathrm{isu}}_{it}={\alpha }_0+{\beta }_{1}df{i}_{it}+{\beta }_{2}df{i}_{it}^2+\gamma X_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(1)

where i and t are city and year, respectively; isu_it represents the dependent variable, namely industrial structure upgrading; dfi_it is the primary independent variable—the Digital Inclusive Finance Index; while $df{i}_{it}^2$ captures its squared term to test for potential nonlinear effects; X_it is a vector of control variables; β and γ represent the coefficients to be estimated, regional fixed effects and time fixed effects, respectively; α₀ is the intercept term; ε_it denotes the random error term.

Given that conventional panel models assume spatial independence across regions, omitting spatial dependence may lead to biased and inconsistent parameter estimates. Therefore, building upon the baseline panel model, this study develops a more comprehensive spatial econometric framework to account for potential spatial dependencies.

3.2. Spatial Autocorrelation Analysis

The prerequisite for applying spatial econometric models is that the variables exhibit significant spatial dependence. Therefore, this paper uses the global Moran’s I index and the local Moran’s I index to conduct spatial autocorrelation tests. The global Moran’s I can determine whether the Yangtze River Delta urban agglomeration as a whole shows a spatial agglomeration or dispersion trend, while the local Moran’s I can identify the specific spatial association patterns of each city, thereby providing direct evidence for the rationality of the subsequent spatial weight matrix and the setting of the spatial model [47]. If the test results reject the null hypothesis of spatial random distribution, it indicates that it is necessary to adopt spatial econometric methods.

Moran’s I serves as a statistical measure to assess the presence of spatial correlation [48], and its computational formula is expressed as follows:

$$\mathrm{Moran}’\mathrm{s}I={\displaystyle \frac{{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{W}_{ij}(Y_i-\overline{Y})(Y_j-\overline{Y})}}}{S^2{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{W}_{ij}}}}}$$

(2)

In the above formula, S² represents the sample variance, Y_i is the observed value for the i-th region, n is the total number of regions, and W_ij refers to the element at position (i,j) in the spatial weight matrix, which quantifies the spatial relationship between region i and region j.

The Moran’s I ranges from −1 to 1. A positive value (greater than 0) indicates a positive spatial correlation among regions, meaning that regions with similar characteristics tend to cluster spatially. A negative value (less than 0) denotes a negative spatial correlation, implying that regions with dissimilar characteristics are spatially adjacent. When the value is close to 0, there is no significant spatial correlation, indicating a random spatial distribution pattern.

The critical value of the standard normal can be used to test the spatial autocorrelation coefficient, and the test formula is:

$$Z(d)={\displaystyle \frac{\mathrm{Moran}’\mathrm{s}I-E(I)}{\sqrt{Var(I)}}}$$

(3)

The expected value of Moran’s I is denoted as $E(I)=-{\displaystyle \frac{1}{n-1}}$, and Var(I) represents the variance of Moran’s I.

The global Moran’s I evaluates the spatial agglomeration pattern across an entire spatial dataset. To assess localized spatial clustering around a specific region i, the Local Moran’s I can be employed, which is computed as follows:

$$I_{\mathrm{i}}={\displaystyle \frac{(Y_i-\overline{Y})}{S^2}}{\displaystyle \sum _{j=1}^n{W}_{ij}(}{Y}_j-\overline{Y})$$

(4)

The local Moran’s I and global Moran’s I share conceptual similarities and employ analogous methodological frameworks for spatial autocorrelation assessment.

3.3. Specification of the Spatial Econometric Model

To simultaneously capture the local direct effects of digital inclusive finance information flows on industrial structure upgrading and the cross-regional spatial spillover and diffusion effects, this paper further incorporates spatial information correlation factors to develop a spatial econometric model. Commonly used spatial econometric models primarily include the Spatial Autoregressive Model (SAR), Spatial Error Model (SEM), and Spatial Durbin Model (SDM). The SAR focuses on capturing the spatial information transmission of the explained variable across regions; the SEM emphasizes the spatial diffusion of unobserved information shocks; while the SDM incorporates spatial lag terms of both explanatory and explained variables simultaneously, enabling a more comprehensive characterization of the spatial spillover paths of multi-dimensional financial information elements across regions [49]. To avoid subjectivity in model selection, the optimal-fitting spatial econometric model is selected based on the empirical results of the Lagrange Multiplier (LM) test, Likelihood Ratio (LR) test, and Hausman test. The primary model specifications are set as follows:

SAR:

$${\mathrm{isu}}_{it}=\alpha +\rho w_i^{\prime }is{u}_t+{\beta }_{1}di{f}_{it}+{\beta }_{2}di{f}_{it}^2+{\gamma }_1{X}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(5)

SEM:

$$\begin{array}{l}{\mathrm{isu}}_{it}=\alpha +{\beta }_{1}di{f}_{it}+{\beta }_{2}di{f}_{it}^2+{\gamma }_1{X}_{it}+{\mu }_i+{\lambda }_t+{\nu }_{it}\\ {\nu }_{it}=\tau w_i^{\prime }{\nu }_{it}+{\epsilon }_t\end{array}$$

(6)

SDM:

$${\mathrm{isu}}_{it}=\alpha +\rho w_i^{\prime }is{u}_t+{\beta }_{1}df{i}_{it}+{\theta }_1{w}_i^{\prime }di{f}_t+{\beta }_{2}di{f}_{it}^2+{\theta }_2{w}_i^{\prime }di{f}_t^2+{\gamma }_1{X}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(7)

$w_i^{\prime }$ is the i_th row of the spatial weight matrix W, where θ and ρ represent the spatial correlation coefficients, τ denotes the spatial error coefficient, and ${\nu }_{it}$ is the random disturbance term, the meanings of other variables are the same as above.

3.4. Specification of the Spatial Weight Matrix

The spatial weight matrix serves as the cornerstone of spatial econometric analysis, and its specification directly determines the precision of spatial effect estimation. The diffusion of digital financial information flows is constrained not only by geographical distance but also contingent on the intensity of inter-regional economic linkages. Following the spatial weight matrix construction approach proposed by Shao, Li et al. [50], this paper constructs an economic-geographic distance spatial weight matrix (W). Its specific form is presented as follows:

$$W={\displaystyle \frac{1}{\left|gd{p}_i-gd{p}_j\right|}}\times {\displaystyle \frac{1}{d_{ij}}}$$

(8)

Here, gdp_i and gdp_j stand for the average annual per capita GDP of region i and region j, respectively; d_ij represents the geographical friction in information interaction between region i and region j, measured by the reciprocal of the shortest highway mileage between the provincial capitals of the two regions. This spatial weight matrix not only accounts for the baseline cost of information diffusion determined by inter-regional geographical proximity but also reflects the intensity of information interaction driven by inter-regional economic linkages. As such, W enables a more comprehensive and objective characterization of the spatial correlation strength of inter-regional financial information flows.

3.5. Direct Impact and Spillover Effects

When estimating parameters in SDM, due to the introduction of spatial information transmission terms in the model, if point estimation is directly used to test spatial effects, it may lead to bias because the feedback loop of information between regions is ignored. Therefore, LeSage [49] proposed using partial differentiation methods to decompose the spatial effect of explanatory variables into direct effects, indirect effects (spatial spillover effects), and total effects. The direct effect reflects the marginal impact of changes in explanatory variables in the region on the explained variables in the same region; the indirect effect (spatial spillover effect) describes the cross-regional information radiation impact on the explained variables in neighboring regions caused by changes in explanatory variables in the region through spatial information transmission channels; the total effect is the sum of the two, representing the average comprehensive impact of changes in explanatory variables on all regions within the system. In the SDM specified in this paper (Equation (7)), the direct and indirect effects of digital inclusive finance and its squared term on industrial structure upgrading can be decomposed and estimated using this partial differentiation approach. This decomposition can accurately quantify the local release intensity and cross-regional radiation patterns of digital financial information flows, avoiding the potential inference bias that may result from point estimation.

For simplicity, let isu in Formula (7) be denoted as y and dfi as z, and ignore the control variables and error terms. Formula (7) can then be rewritten in the following form:

$$\mathrm{y}=\rho W_T\mathrm{y}+{\beta }_1\mathrm{z}+{\theta }_1{W}_T\mathrm{z}+{\beta }_2{\mathrm{z}}^{\left(2\right)}+{\theta }_2{W}_T{\mathrm{z}}^{(2)}$$

(9)

Wherein, $\mathrm{y}={\left(y_1^{\prime },\dots ,y_T^{\prime }\right)}^{\prime },\mathrm{z}={\left(z_1^{\prime },\dots ,z_T^{\prime }\right)}^{\prime },{\mathrm{z}}^{(2)}={\left(z_1^{{(2)}^{\prime }},\dots ,z_T^{{(2)}^{\prime }}\right)}^{\prime },W_T=I_T\otimes W,{\mathrm{z}}^{(2)}$ is the element-wise square of z, and rearranging the above equation yields:

$$(I_{NT}-\rho W_T)\mathrm{y}={\beta }_1\mathrm{z}+{\theta }_1{W}_T\mathrm{z}+{\beta }_2{\mathrm{z}}^{\left(2\right)}+{\theta }_2{W}_T{\mathrm{z}}^{(2)}$$

(10)

Suppose $I_{NT}-\rho W_T$ is invertible, then

$$\mathrm{y}=(I_{NT}-\rho W_T{)}^{-1}({\beta }_1\mathrm{z}+{\theta }_1{W}_T\mathrm{z}+{\beta }_2{\mathrm{z}}^{\left(2\right)}+{\theta }_2{W}_T{\mathrm{z}}^{(2)})$$

(11)

To derive the marginal effect of the explanatory variable z on y, taking the differential yields the partial effect matrix:

$$\begin{array}{l}{\displaystyle \frac{\partial \mathrm{y}}{\partial z^{\prime }}}=(I_{NT}-\rho W_T{)}^{-1}\cdot {\displaystyle \frac{\partial }{\partial z^{\prime }}}({\beta }_1\mathrm{z}+{\theta }_1{W}_T\mathrm{z}+{\beta }_2{\mathrm{z}}^{\left(2\right)}+{\theta }_2{W}_T{\mathrm{z}}^{(2)})\\ =(I_{NT}-\rho W_T{)}^{-1}({\beta }_1{I}_{NT}+{\theta }_1{W}_T+2{\beta }_{2}diag(z)+2{\theta }_2{W}_{T}diag(z))\end{array}$$

(12)

For each period t, the partial derivative matrix can be written as:

$${\displaystyle \frac{\partial {\mathrm{y}}_t}{\partial z^{\prime }}}=\left[\begin{array}{ccc}{\displaystyle \frac{\partial {\mathrm{y}}_{1t}}{\partial z_{1t}}}& \cdots & {\displaystyle \frac{\partial {\mathrm{y}}_{1t}}{\partial z_{Nt}}}\\ ⋮& \ddots & ⋮\\ {\displaystyle \frac{\partial {\mathrm{y}}_{Nt}}{\partial z_{1t}}}& \cdots & {\displaystyle \frac{\partial {\mathrm{y}}_{Nt}}{\partial z_{Nt}}}\end{array}\right]=(I_N-\rho W{)}^{-1}({\beta }_1{I}_N+{\theta }_{1}W+2{\beta }_{2}diag(z_t)+2{\theta }_{2}Wdiag(z_t))$$

(13)

diag(z) is a diagonal matrix with each element of z as the diagonal elements.

The mean of the diagonal elements in the aforementioned partial effects matrix represents the direct effect of digital inclusive finance and its quadratic term on regional industrial structure upgrading; the mean of the off-diagonal elements captures the spatial spillover effect of a region’s digital inclusive finance and its quadratic term on industrial structure upgrading in other regions; the sum of these two components constitutes the total effect.

4. Empirical Analysis

4.1. Variable Selection and Data Sources

4.1.1. Dependent Variable: Industrial Structure Upgrading (isu)

Industrial structure upgrading is inherently a process of reallocating information elements across industries. Information carriers—including capital, labor, and technology—flow from low-efficiency sectors to high-efficiency ones, driving the evolution of industries toward service orientation and high-end specialization. Drawing on the approach proposed by Ge Heping et al. [22], this paper constructs an industrial structure upgrading index to quantify the efficiency of such information allocation. The specific calculation method is presented as follows:

$$\mathrm{isu}={\displaystyle \sum _{i=1}^{3}i\times g_i}=1\times g_1+2\times g_2+3\times g_3$$

(14)

g_i is the share of value-added of the i-th industry in regional GDP. A larger value of this index indicates a higher proportion of high information density industries (e.g., modern services, high-tech manufacturing) in the regional economic structure, as well as a more pronounced trend of the industrial structure evolving toward information-intensive sectors.

4.1.2. Core Explanatory Variable: Digital Financial Information Flow

This paper employs the Peking University Digital Inclusive Finance Index as a proxy variable to measure the coverage breadth, usage depth, and digitalization level of financial information services across cities. From an information science perspective, this index can be decomposed into three dimensions:

Digital payment information flow (payment): Reflects the digitalization degree of transaction information and the activity of cross-regional flow.

Digital credit information flow (credit): Captures the efficiency of credit information identification, evaluation, and allocation.

Digital insurance information flow (insurance): Indicates the guarantee capacity of risk information and its supporting role in factor mobility.

4.1.3. Mediating Variables and Threshold Variables

Technological innovation information (info_tech_patent): Measured by the number of patent grants to above-scale industrial enterprises in each city for the current year. This indicator reflects the generation density and accumulation level of technological information within a region, and is subjected to logarithmic transformation in the empirical analysis.

Entrepreneurial activity (new_firm): Measured by the number of newly registered enterprises in each city. This indicator essentially represents the economic decision-making behavior of entrepreneurs, derived from a comprehensive assessment by entrepreneurs of internal and external market information, technological information, and policy information in the regional market.

Information infrastructure level (info_internet_mobile): Measured by the number of mobile phone users per million people in each city. It reflects the real-time accessibility of mobile information flows and serves as the core channel for real-time market information response.

4.1.4. Control Variables

Drawing on findings from existing literature, control variables are selected to mitigate omitted variable bias, focusing on key factors influencing industrial structure upgrading:

Government behavior (gov): Measured by the ratio of total local government fiscal expenditure to regional GDP. This indicator reflects the intensity and direction of public information resource allocation, as government expenditure guides industrial layout via policy information signals. Openness (open): Proxied by total regional import and export volume. It measures the vitality of cross-border information flows and knowledge spillovers—higher openness corresponds to more frequent exchanges of technical and market information across regions. Fixed asset investment (fa_ratio): Captured by the share of fixed asset investment in regional GDP. Economic development level (growth_gdp): Measured by the GDP growth rate, which characterizes the activity and development potential of regional information-based economic activities. The economic growth rate reflects the marginal output efficiency of information factors.The measurement methods of all variables are summarized in

Table 1. Measurement of variables.

Variable Attribute	Variable Name	Measurement Method	Selection Criteria and Importance
Dependent variable	Industrial structure upgrading (isu)	$\mathrm{isu}={\displaystyle \sum _{i=1}^{3}i\times g_i}$	The classical measurement method, which directly captures the allocation efficiency of information factors across industries, serves as the core outcome variable for validating the information transmission effect of digital finance.
Explanatory variables	Digital Financial Information Flow (dfi)	Peking University Digital Inclusive Finance Index	The authoritative and widely adopted index, which systematically captures the multi-dimensional characteristics of digital financial information flows, serves as the core driving force of this study.
	Digital payment information flow (payment)	The “Payment Sub-Index” of the Peking University Digital Inclusive Finance Index
	Digital credit information flow (credit)	The “credit Sub-Index” of the Peking University Digital Inclusive Finance Index
	Digital insurance information flow (insurance)	The “insurance Sub-Index” of the Peking University Digital Inclusive Finance Index
Mediating variables and threshold variables	Technological innovation information (info_tech_patent)	The number of patents authorized to large-scale industrial enterprises in the current year	It captures the generation density and accumulation level of technological information within a region, with high data accessibility. This paper seeks to verify that it exerts a partial mediating effect in the pathway from digital finance to industrial upgrading.
	Entrepreneurial activity (new_firm)	The number of newly registered enterprises in each city during the current year	It directly captures the response efficiency of entrepreneurial information, and this paper seeks to verify that it exerts both local mediating effects and spatial mediating effects.
	Information infrastructure level (info_internet_mobile)	Number of mobile phone users per million people	It captures the regional information absorption capacity, and this paper employs it to test the threshold effect of digital finance’s spatial spillover.
Control variables	Government behavior (gov)	The ratio of government fiscal expenditure to GDP	This variable controls for the interference of the public sector on the industrial structure, while uncovering the nonlinear impact of policy information.
	Openness (open)	Total value of imports and exports of the region	This variable controls for external information shocks, and a higher degree of openness is associated with more frequent exchanges of technological and market information.
	Fixed asset investment (fa_ratio)	The ratio of fixed asset investment to GDP	This variable controls for the influence of traditional capital factors on industrial upgrading, thereby mitigating omitted variable bias.
	Economic development level (growth_gdp)	GDP growth rate	This variable controls for the phased characteristics of macroeconomic growth.

.

This paper integrates multi-source data from the Digital Finance Research Center of Peking University, the China Urban Statistical Yearbook, statistical bulletins of various provinces and cities, and the official website of the National Bureau of Statistics. For partially missing data, linear interpolation is employed for imputation. This paper selects 27 prefecture-level cities in the Yangtze River Delta region over the period 2011–2023 as the research sample, yielding a total of 351 observations. The sample period is determined based on three considerations: First, the Peking University Digital Inclusive Finance Index has been officially released since 2011, and 2023 represents the most recent year with fully available annual data. Second, statistical standards for relevant control variables have been unified since 2011, ensuring high data integrity. Third, the 2011–2023 period fully encompasses the critical phase of digital inclusive finance in the Yangtze River Delta, spanning from initial adoption to deep integration.

4.2. Results of Spatial Autocorrelation Test

Prior to conducting spatial econometric analysis, a preliminary test must be performed to examine whether the variables exhibit spatial information dependence. This paper employs the global Moran’s I and local Moran’s I scatter plots to examine the spatial autocorrelation on the industrial structure upgrading and the digital inclusive finance index of the Yangtze River Delta urban agglomeration, thereby verifying the spatial interaction characteristics of financial information flows across regions.

The results of the global Moran’s I are reported in

Table 2. Global Moran’s I of Industrial Upgrading and Digital Inclusive Finance.

Year	Industrial Structure Upgrading			Digital Inclusive Finance
	Moran’s I	Z Value	p Value	Moran’s I	Z Value	p Value
2011	0.5313	4.7037	0.0000	0.4992	4.4792	0.0000
2012	0.5354	4.7567	0.0000	0.5171	4.5789	0.0000
2013	0.54	4.7983	0.0000	0.5148	4.5253	0.0000
2014	0.5128	4.5798	0.0000	0.5168	4.5796	0.0000
2015	0.4769	4.3233	0.0000	0.5539	4.8526	0.0000
2016	0.4365	3.9789	0.0001	0.6277	5.4529	0.0000
2017	0.198	1.9642	0.0495	0.5821	5.1013	0.0000
2018	0.3705	3.426	0.0006	0.58	5.0704	0.0000
2019	0.3103	2.9521	0.0032	0.573	5.0277	0.0000
2020	0.3584	3.3387	0.0008	0.5727	5.0346	0.0000
2021	0.3352	3.1593	0.0016	0.5055	4.4749	0.0000
2022	0.331	3.1334	0.0017	0.5566	4.8955	0.0000
2023	0.2985	2.8625	0.0042	0.5833	5.1213	0.0000

. Both the Moran’s I values for the industrial structure upgrading and the digital inclusive finance index of the Yangtze River Delta urban agglomeration are positive (greater than 0), with most passing the 1% significance test. The Moran’s I for digital inclusive finance consistently remains above 0.5 and is highly significant, indicating stronger spatial dependence; in contrast, the Moran’s I for industrial structure upgrading experienced a decline in 2017 but remains significantly positive overall. This finding indicates that the distribution pattern of industrial structure information and the spatial allocation of digital financial characterized by “high-high” and “low-low” clustering. Accordingly, spatial correlation should be incorporated into the analysis of the relationship between digital inclusive finance and industrial structure upgrading. This finding provides a necessary prerequisite for employing spatial econometric models to examine the spatial spillover effects between the two variables.

The local Moran’s I scatter plot (Figure 1 and Figure 2) further reveals the information agglomeration patterns and information diffusion hotspots of each city in space. From the distribution of Moran’s I scatter plots for the two variables, cities with a higher level of digital inclusive finance development tend to be situated within regional clusters characterized by a higher level of industrial structure upgrading. To rigorously examine the spatial association between the two, subsequent analysis employs a spatial econometric model and decomposes the direct and indirect effects of digital inclusive finance on industrial structure upgrading via the partial differentiation method.

4.3. Selection of Spatial Econometric Models

For the construction of a scientifically rigorous spatial econometric model, this paper first leverages the Robust LM test to discern the spatial dependence characteristics inherent in the dataset. Subsequently, the LR likelihood ratio test is employed to validate the nested relationships among candidate models. By integrating the findings of these two complementary tests, the final optimal spatial econometric model is determined.

As shown in

Table 3. Results of LM and LR Tests.

Test Method	Test Statistic	p Value
LM-Error Test	2.793	0.095
Robust LM-Error Test	10.929	0.001
LM-Lag Test	16.662	0.000
Robust LM-Lag Test	24.798	0.000
Likelihood-Ratio Test of SDM Against SAR	50.82	0.000
Likelihood-Ratio Test of SDM Against SEM	44.64	0.000

, the LM test rejects the null hypothesis, indicating that the selected samples exhibit both spatial lag dependence and spatial error autocorrelation. Furthermore, the LR test rejects the null hypothesis at the 1% significance level, suggesting that the SDM outperforms both the SAR and the SEM, and does not degenerate into either. Therefore, this study adopts the SDM for empirical analysis. Additionally, based on the Hausman test results (χ² = 114.74, p < 0.001), the p-value is statistically significant at the 1% level, leading to the rejection of the null hypothesis and supporting the use of a two-way fixed effects specification.

4.4. Analysis of Spatial Econometric Model Results

Table 4. Model Regression Results.

Variables	OLS	SAR	SEM	SDM
ln_dfi	−0.332 **	0.087	−0.015	−0.622 **
	(−2.19)	(0.45)	(−0.06)	(−2.41)
ln_dfi2	0.048 ***	0.008	0.022	0.081 ***
	(3.21)	(0.32)	(0.77)	(2.75)
gov	−1.362 **	−0.749 *	−0.763 *	−1.393 ***
	(−2.33)	(−1.95)	(−1.94)	(−3.65)
gov2	3.313 *	1.425	1.739	3.868 ***
	(1.88)	(1.38)	(1.57)	(3.59)
open	0.064 ***	−0.033 **	−0.043 ***	−0.027 *
	(4.97)	(−2.45)	(−3.18)	(−1.94)
fa_ratio	−0.107 ***	−0.019 **	−0.019 **	−0.024 ***
	(−8.20)	(−2.46)	(−2.51)	(−3.15)
growth_gdp	0.317 ***	0.002	−0.002	0.006
	(5.41)	(0.05)	(−0.08)	(0.22)
Constant	2.957 ***
	(7.59)
W*ln_dfi				1.617 ***
				(3.59)
W*ln_dfi2				−0.196 ***
				(−3.64)
ρ		0.483 ***		0.403 ***
		(7.43)		(5.82)
λ			0.491 ***
			(6.95)
Observations	351	351	351	351
R-squared	0.652	0.701	0.710	0.67
Log-likelihood		778.3774	775.2885	800.6972

Note: *** p < 0.01, ** p < 0.05, * p < 0.1. In the OLS, the values in parentheses denote t-statistics; for all other models, represent z-statistics.

presents the regression results of the Ordinary Least Squares (OLS), SAR, SEM and SDM, with the aim of uncovering the nonlinear impacts and spatial effects of digital inclusive finance on industrial structure upgrading. A comparison of the four models reveals that the R² values of the spatial models are all higher than that of the OLS, while the coefficients of ρ and λ are highly significant—indicating the presence of significant spatial autocorrelation in the explained variable. The SDM yields the highest log-likelihood value, thus justifying its adoption. Additionally, the U-shaped effect of the core variable maintains consistent direction across different models, confirming the robustness of the conclusion.

The coefficient of the digital inclusive finance is −0.622, while the coefficient of its quadratic term is 0.081, both statistically significant at conventional levels. Additionally, the spatial interaction coefficient (W*ln_dfi) and the spatial quadratic term coefficient are 1.617 and −0.196, respectively, with both significant at the 1% level. These results suggest that digital inclusive finance exerts a significant impact on industrial structure upgrading and exhibits spatial spillover effects, thereby providing empirical support for Hypothesis 1a.

After excluding the influence of digital inclusive finance in neighboring regions, the digital inclusive finance within the focal region exhibits a U-shaped effect on industrial structure upgrading—indicating the existence of a critical threshold for the release of financial information flows. When the development level of digital inclusive finance is below this value, the diffusion and allocation efficiency of financial information is insufficient to effectively support industrial structure optimization, and may even exert a negative impact due to the initial investment costs of information infrastructure construction. After digital inclusive finance development crosses this specific threshold do information network effects and economies of scale gradually materialize, allowing digital financial information to effectively guide resource allocation and promote factor mobility, thereby generating a significant positive driving effect on local industrial structure upgrading. This finding uncovers the phased regularities governing the release of digital financial information flows. Lin also identified a local U-shaped relationship [37]; however, this study further quantifies the critical threshold and identifies the spatial nonlinear relationship in subsequent sections, thereby verifying the local U-shaped nonlinear relationship proposed in Hypothesis 2.

The financial information flow embedded in digital inclusive finance of neighboring regions exhibits a nonlinear “inverted U-shaped” impact on local industrial structure upgrading. This indicates that the impact of digital inclusive finance in neighboring regions on the focal region is initially positive and then turns negative: during moderate development, cross-regional diffusion of information flows generates positive spillover effects on technology, capital, and markets; with excessive development, the “siphon effect” of information agglomeration gradually becomes pronounced, thereby exerting an inhibitory impact on industrial structure upgrading within the focal region. The spatial inverted U-shaped nonlinear relationship proposed in Hypothesis 2 is hereby verified. This finding aligns with the work of Guo [11]; however, the present study further examines the heterogeneity across payment, credit, and insurance dimensions in subsequent sections. This not only provides empirical evidence for understanding the regional evolution pattern of “initial coordination followed by competition” but also offers a threshold reference for mitigating polarization risks.

In the SDM, the spatial autoregressive coefficient ρ = 0.403 (p < 0.01) suggests that industrial structure upgrading behaviors across adjacent cities exhibit a mutually reinforcing effect, which reflects the synergistic effect of industrial integration in the Yangtze River Delta.

4.5. Decomposition of Spatial Spillover Effects

Since the regression coefficients of explanatory variables in the SDM cannot fully capture marginal information contribution to industrial structure upgrading, this paper employs the partial differentiation approach proposed by LeSage and Pace [49] to decompose the impact of digital inclusive finance on industrial structure upgrading into direct effects, indirect effects, and total effects.

Table 5. Effect Decomposition of the SDM.

Variables	Direct Effects	Indirect Effects	Total Effects
ln_dfi	−0.463 *	2.120 ***	1.657 ***
	(−1.90)	(3.34)	(2.80)
ln_dfi2	0.062 **	−0.252 ***	−0.190 **
	(2.18)	(−3.22)	(−2.46)
gov	−1.528 ***	−2.364 *	−3.892 ***
	(−4.04)	(−1.76)	(−2.60)
gov2	3.974 ***	2.863	6.837 *
	(3.88)	(0.86)	(1.87)
open	−0.019	0.108 **	0.088 *
	(−1.37)	(2.37)	(1.67)
fa_ratio	−0.028 ***	−0.062 **	−0.090 ***
	(−3.30)	(−2.12)	(−2.62)
growth_gdp	0.010	0.056	0.066
	(0.31)	(0.56)	(0.55)

Note: *** p < 0.01, ** p < 0.05, * p < 0.1.

reports the direct, indirect, and total effects of each explanatory variable.

Based on the effect decomposition results in Table 5, first, the coefficient of the direct effect of digital inclusive finance is negative, while its quadratic term coefficient is positive, with a critical threshold of 3.734. This qualitative finding reveals the phased pattern of information release in digital finance. Second, the first-order term coefficient of the indirect effect is 2.120, the second-order term coefficient is −0.252, and the critical threshold is 4.206—all of which are statistically significant at the 1% level. This suggests that the impact of digital inclusive finance in neighboring regions on local industrial structure upgrading exhibits an “inverted U-shaped” relationship, characterized by an initial promotional effect followed by an inhibitory effect. The reason is that the development of digital inclusive finance in the surrounding areas exceeding the critical value does indeed have a “siphon effect” [51]. Hypothesis 1a and Hypothesis 2 are once again verified. This provides empirical evidence for understanding the dynamic transformation of digital financial information flow from “information diffusion” to “information aggregation”.

Among the control variables, the direct effect of government behavior and its quadratic term is significant, and there is a “U-shaped” influence. That is, when the scale of government fiscal expenditure is below the critical value, the policy information signal may have an inhibitory effect on the upgrading of the local industrial structure due to distorted resource allocation or insufficient information transmission efficiency. After the government fiscal expenditure crosses the critical value, the public information guidance effect gradually emerges. Through the effective transmission of industrial policy information and the optimization of public resource allocation information, it has a positive promoting effect on the upgrading of the local industrial structure. At the spatial spillover level, the indirect effect of government behavior is significantly negative, indicating that the fiscal expenditure of the surrounding regions has a policy information suction effect on the upgrading of the local industrial structure—neighboring areas attract high-quality production factors to gather around by releasing more attractive policy signals and public resource allocation information, thereby having a negative spillover effect on the upgrading of the local industrial structure. The direct and indirect effects of fixed asset investment are both negative and pass the significance test. This may be because fixed asset investment is overly concentrated in traditional industries, while the information capital invested in high-tech industries and modern service industries is relatively insufficient, resulting in a low information allocation efficiency of capital factors. At the same time, there are an “crowding-out effect” of investment information. Excessive fixed asset investment crowds out the information capital input of enterprises for R&D innovation and digital transformation, inhibiting the development of high-information-density industries, etc.

4.6. Robustness Testing

To ensure the robustness and reliability of the empirical results, this study employs the approach of replacing the explained variable. Following Yang H et al. [40] Model 1 adopts the ratio of the added value of the tertiary industry to the sum of the added values of the primary and secondary industries in the region as the indicator for regional industrial structure upgrading. Additionally, in line with Li et al. [52], Model 2 uses the ratio of the added value of the tertiary industry to that of the secondary industry as an alternative measure for industrial structure upgrading in the robustness analysis. Furthermore, drawing on the methodology of Ming et al. [53], this paper conducts robustness checks by reducing the sample scope. First, Model 3 excludes data from municipalities directly under the central government. Given that Shanghai, as a central municipality and core city of the Yangtze River Delta, features a service-oriented industrial structure dominated by finance, trade, and logistics, and possesses significant advantages in technological innovation and other domains, it exerts a strong radiating influence on neighboring cities. Therefore, it is necessary to examine whether the inclusion of such megacities affects the validity of the findings. Second, Model 4 shortens the time span by excluding data from 2011 and 2023 to assess the stability of the estimates. The results of the robustness tests are presented in

Table 6. Robustness Test Results.

Variables	Model 1	Model 2	Model 3	Model 4
ln_dfi	−5.287 ***	−5.061 ***	−0.614 **	−1.294 **
	(−4.06)	(−4.30)	(−2.17)	(−2.29)
ln_dfi2	0.582 ***	0.558 ***	0.080 **	0.137 **
	(3.90)	(4.14)	(2.45)	(2.38)
gov	−7.463 ***	−5.990 ***	−1.490 **	−1.860 ***
	(−3.87)	(−3.44)	(−2.05)	(−4.51)
gov2	23.123 ***	17.974 ***	4.206 **	5.057 ***
	(4.26)	(3.66)	(2.35)	(4.42)
open	−0.227 ***	−0.204 ***	−0.023	−0.033 *
	(−3.28)	(−3.26)	(−0.51)	(−1.86)
fa_ratio	−0.137 ***	−0.153 ***	−0.024 *	−0.023 **
	(−3.58)	(−4.41)	(−1.74)	(−2.24)
zgdp	0.155	0.129	0.004	0.015
	(1.15)	(1.05)	(0.22)	(0.53)
W*ln_dfi	8.919 ***	8.485 ***	1.715 **	4.702 ***
	(3.89)	(4.09)	(2.53)	(4.92)
W*ln_dfi2	−1.134 ***	−1.087 ***	−0.207 ***	−0.516 ***
	(−4.15)	(−4.40)	(−2.58)	(−5.21)
rho	0.167 **	0.157 *	0.390 ***	0.331 ***
	(2.08)	(1.95)	(4.68)	(4.21)

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1.

.

According to the results, both the coefficients of digital inclusive finance and its quadratic term are statistically significant at the 1% level, and the results are consistent with the previous results. This suggests that the development of digital inclusive finance exerts a nonlinear influence on regional industrial structure upgrading and exhibits a spatial spillover effect, indicating that the practical implications derived in this paper are robust and reliable.

4.7. The Impact of Different Sub-Sectors of Digital Inclusive Finance on Industrial Upgrading

While previous analysis has confirmed the U-shaped impact of digital inclusive finance and its quadratic term on industrial structure upgrading, the specific sources of the overall financial information benefits remain undetermined. As a comprehensive financial information service system, digital inclusive finance encompasses distinct business dimensions that embed heterogeneous information flows: digital payments primarily transmit transaction information flows, digital credit mainly carries credit information flows, and digital insurance predominantly provides risk information flows. These variations in information functions may give rise to heterogeneous impacts on industrial upgrading. Accordingly, this paper further incorporates digital payment information flow, digital credit information flow and digital insurance information flow into the spatial econometric model for estimation, in order to systematically examine the local direct effects and spatial spillover effects of different financial information flows on industrial structure upgrading.

According to

Table 7. Effect Decomposition of Digital Inclusive Finance Sub-sectors.

Variables	Direct Effects	Indirect Effects	Total Effects
ln_payment	0.182 ***	0.144 *	0.326 ***
	(4.59)	(1.89)	(4.97)
ln_credit	0.090 **	0.463 ***	0.554 ***
	(2.54)	(4.64)	(5.15)
ln_insurance	0.081 ***	−0.039	0.042
	(4.17)	(−0.72)	(0.71)

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1.

, digital payment and digital credit exhibit statistically significant positive direct, indirect, and total effects. Specifically, digital payment services demonstrate a stronger local driving effect, whereas digital credit services feature a more prominent cross-regional transmission effect. This indicates that transaction information flows and credit information flows not only promote local industrial structure upgrading but also generate positive spillover effects on neighboring regions via spatial information transmission channels. In contrast, for the risk information flow embedded in digital insurance services, the direct effect is statistically significant and positive, whereas neither the indirect effect nor the total effect passes the significance test. This suggests that the financial information flows of the three types of services play heterogeneous roles in industrial structure upgrading, thereby validating Hypothesis 3.

First, digital payments leverage information and communication technologies (ICTs) to mitigate information asymmetry and reduce financing constraints [54], thereby effectively lowering transaction costs and information frictions while expanding the coverage of financial services [55]. Concurrently, digital payments can generate spatial spillovers through cross-city consumption and industrial chains.

Second, enabled by technological underpinnings, the digital credit information flow expands access to formal financial services for excluded and underserved entities [6]—specifically, delivering productive capital support to information-constrained small and medium-sized enterprises, start-ups, and individual business owners. Spatially, it facilitates cross-regional capital allocation via internet platforms and supply chains.

Finally, risk information flows effectively facilitate industrial structure upgrading within the local region, yet no significant spatial spillover effect has been identified. Insurance products are characterized by localized information attributes and exhibit distinct regional heterogeneity [56]. Furthermore, the insurance sector still has substantial room for development, rendering direct cross-regional replication challenging and thus constraining the spatial diffusion of risk information flows.

4.8. Analysis of the Impact Path of Financial Information Flow

Theoretical analysis suggests that digital inclusive finance can influence industrial structure upgrading through micro-level information transmission pathways. To test the mediating effects of enterprise entrepreneurial activity and enterprise innovation information input, this paper employs the stepwise regression method [57] to incorporate mediating variables into the SDM.

$$M_{it}={\alpha }_4+z{w}_i^{\prime }\mathrm{i}\mathrm{s}{\mathrm{u}}_t+c_{1}df{i}_{it}+{\omega }_1{w}_i^{\prime }di{f}_t+c_{2}di{f}_{it}^2+{\omega }_2{w}_i^{\prime }di{f}_t^2+\gamma X_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(15)

$${\mathrm{isu}}_{it}={\alpha }_5+\rho w_i^{\prime }\mathrm{i}\mathrm{s}{\mathrm{u}}_t+{\beta }_{1}df{i}_{it}+{\theta }_1{w}_i^{\prime }di{f}_t+{\beta }_{2}di{f}_{it}^2+{\theta }_2{w}_i^{\prime }di{f}_t^2+{\beta }_3{M}_{it}+{\theta }_3{w}_i^{\prime }{M}_t+\gamma X_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(16)

M_it represents the mediating variables, namely the entrepreneurial activity of enterprises and the information input for innovation of enterprises. z, θ and ρ represents the spatial correlation coefficient. The meanings of other variables are consistent with those in the previous text. The calculation results of the model are shown in the following

Table 8. Test of Mediating Effect of Entrepreneurial Activity.

	(1) isu	(2) ln_new_firm	(3) isu
Direct effect
ln_dfi	−0.463 *	−4.548 **	−0.388 *
	(0.244)	(1.840)	(0.232)
ln_dfi2	0.062 **	0.474 **	0.055 **
	(0.028)	(0.211)	(0.027)
ln_new_firm			0.029 ***
			(0.008)
Indirect effect
ln_dfi	2.120 ***	6.447 **	2.013 ***
	(0.635)	(3.255)	(0.616)
ln_dfi2	−0.252 ***	−0.622	−0.246 ***
	(0.078)	(0.387)	(0.077)
ln_new_firm			0.060 *
			(0.032)
Total effect
ln_dfi	1.657 ***	1.899	1.625 ***
	(0.592)	(2.485)	(0.599)
ln_dfi2	−0.190 **	−0.148	−0.191 **
	(0.077)	(0.322)	(0.079)
ln_new_firm			0.089 **
			(0.036)
Control variable	yes	yes	yes
R2	0.66	0.54	0.68
N	351	351	351

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1.

:

Model (1) in Table 8 corresponds to the empirical results of the SDM presented earlier. Models (2) and (3) denote the model estimating the impact of digital inclusive finance on entrepreneurial activity, and the full model incorporating the mediating variable, respectively.

Digital inclusive finance exhibits a direct U-shaped impact on enterprise entrepreneurial activity. Meanwhile, digital inclusive finance from neighboring regions exerts a positive effect on local entrepreneurial activity via spatial diffusion (6.447, p < 0.01). The direct, indirect, and total effects of entrepreneurial dynamism on industrial structure upgrading are all statistically significant (Model 3). Moreover, when entrepreneurial dynamism is introduced as a covariate, the corresponding direct, indirect, and total effects of digital inclusive finance on industrial structure upgrading attenuate—suggesting a partial mediation or competitive absorption of explanatory power. The results indicate that entrepreneurial dynamism exerts a partial U-shaped mediating effect in the relationship between digital inclusive finance and industrial structure upgrading, with a spatial mediating effect also present. Hypothesis 1c is thus validated.

At low levels of financial agglomeration, insufficient financial support suppresses the enhancement of innovation and entrepreneurship dynamism, thereby undermining urban economic resilience. As financial agglomeration deepens, the credit, transaction, and risk information flows unlocked by digital inclusive finance are captured, evaluated, and translated into tangible entrepreneurial decision-making behaviors by potential entrepreneurs [21,58]. This leads to a significant increase in innovation and entrepreneurship dynamism, ultimately giving rise to a U-shaped mediating effect [59]. The emergence of new startups introduces competitive forces at the boundaries of industries, transforming traditional business models with new technological means, and thereby compelling existing enterprises to enhance efficiency and adjust structures [60]. The entrepreneurial information enabled by digital inclusive finance is not confined to signal feedback and resource reallocation within local markets. It also facilitates the growth of enterprise counts in neighboring regions through the synergy between cross-regional market linkages and transportation network [61], thereby exerting an impact on the industrial structure of adjacent areas.

Table 9. Mediating Effect Test of Information Input for Technological Innovation.

	isu	ln_info_tech_patent	isu
Direct effect
ln_dfi	−0.463 *	1.179 ***	−0.393 *
	(0.244)	(0.363)	(0.229)
ln_dfi2	0.062 **		0.051 *
	(0.028)		(0.027)
ln_info_tech_patent			0.029 ***
			(0.007)
Indirect effect
ln_dfi	2.120 ***	−0.134	1.925 ***
	(0.635)	(0.760)	(0.643)
ln_dfi2	−0.252 ***		−0.230 ***
	(0.078)		(0.080)
ln_info_tech_patent			−0.008
			(0.026)
Total effect
ln_dfi	1.657 ***	1.045	1.532 **
	(0.592)	(0.688)	(0.641)
ln_dfi2	−0.190 **		−0.179 **
	(0.077)		(0.084)
ln_info_tech_patent			0.020
			(0.029)
Control variable	yes	yes	yes
R2	0.66	0.66	0.65
N	351	351	351

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

results show that digital inclusive finance has a positive direct effect on technological innovation information input, but the spatial effect is not significant. After adding technological innovation information input, the absolute value of the direct effect of digital inclusive finance decreases (0.463 decreases to 0.393), and the square term significantly decreases, indicating that technological innovation information input has played a partial mediating effect in the impact of digital inclusive finance on industrial structure upgrading. However, this mediating effect is mainly limited to the local transmission path and has not yet formed a significant spatial spillover. Hypothesis 1b is partially verified.

Specifically, digital inclusive finance can mitigate financing constraints, incentivize firms’ technological innovation, and drive high-quality economic development as well as industrial structure upgrading [32,62]. Financial innovation integrated with technological innovation provides a robust foundation for local industrial structure transformation via diverse innovative instruments [63]. The stock of technological innovation information, proxied by the number of patent grants, possesses the attribute of tacit knowledge [64]. The effective absorption and reutilization of technological information require recipients to have a corresponding knowledge base and learning capacity, thereby constraining cross-regional knowledge spillovers [65]—hence, no significant spatial mediating effect is observed.

4.9. Threshold Effect Analysis

Considering the nonlinear relationship between digital inclusive finance and industrial structure, this paper further constructs a threshold effect model to analyze the nonlinear relationship between digital inclusive finance and industrial structure upgrading under different levels of digital inclusive finance and information infrastructure.

Based on the theoretical analysis in the previous section, a nonlinear panel threshold model is constructed with the levels of digital inclusive finance and information infrastructure as threshold variables. The specific model is as follows:

$${\mathrm{isu}}_{it}={\alpha }_0+{\beta }_{1}df{i}_{it}\times I(q_{it}\le {\delta }_1)+{\beta }_{2}dfi\times I({\delta }_1\le q_{it}\le {\delta }_2)+{\beta }_{3}dfi\times I(q_{it}\ge {\delta }_2)+\gamma X_{it}+{\epsilon }_{it}$$

(17)

Here, q_it is the threshold variable, δ represents a specific threshold value, and I(.) stands for the indicator function of the threshold model. The definitions of other variables are consistent with those specified in the preceding text.

The relevant test and regression results with digital inclusive finance and information infrastructure level as threshold variables are presented in

Table 10. Tests for Threshold Effects.

Threshold Variable	Number of Thresholds	F- Statistic	p- Value	Critical Value (10%)	Critical Value (5%)	Critical Value (1%)	Threshold Value	95% Confidence Interval
ln_dfi	Single threshold	37.50 *	0.080	34.72	42.77	56.80	5.15	(5.4561, 5.5200)
	Double threshold	19.32 *	0.068	16.33	21.81	32.10	5.51	(5.1279, 5.1538)
	Triple threshold	7.21	0.622	24.78	30.32	35.98	-	-
info_internet_mobile	Single threshold	65.09 ***	0.000	22.70	32.36	41.99	0.79	(0.7700, 0.8000)
	Double threshold	16.67	0.202	20.92	26.12	37.35	-	-

Notes: * p < 0.1, *** p < 0.01.

and

Table 11. Threshold Effect Regression Results.

Variables	isu	isu
ln_dfi(info_internet_mobile ≤ 0.79)	0.0881 ***
	(0.0136)
ln_dfi(info_internet_mobile > 0.79)	0.1007 ***
	(0.0120)
ln_dfi(ln_dfi ≤ 5.15)		0.0608 ***
		(0.0087)
ln_dfi(5.15 < ln_dfi ≤ 5.51)		0.0680 ***
		(0.0090)
ln_dfi(ln_dfi > 5.51)		0.0726 ***
		(0.0086)
Constant	1.9347 ***	2.1418 ***
	(0.0651)	(0.0857)
Observations	351	351
Number of id	27	27
R-squared	0.756	0.750

Notes: Robust standard errors in parentheses: *** p < 0.01.

. This paper employs Hansen’s [66] bootstrap method (with 500 replications) to test the number of thresholds in the model, and the test results are reported in Table 10. For digital inclusive finance, there exist two thresholds with values of 5.15 and 5.51, both of which exceed the critical point where the “U-shaped” relationship transitions from negative to positive. For the level of information infrastructure, there is a single threshold with a value of 0.79.

Table 11 presents the impact of digital inclusive finance on industrial structure upgrading across different threshold intervals. The results reveal two key findings: First, across all threshold intervals of digital inclusive finance, the coefficient is positive and gradually increases and statistically significant at the 1% level, indicating that digital inclusive finance in the Yangtze River Delta region has generally crossed the U-shaped inflection point and entered a stage of positive promotion. As a financial infrastructure, digital inclusive finance utilizes IT platforms to process and transmit information. Its marginal cost decreases as the business scale expands, while the density and value of the financial information flows it carries continue to increase. Small and micro enterprises (SMEs) can better align the development with inclusive goals. access more and higher-quality financial services, thereby endowing digital inclusive finance with the characteristic of increasing marginal returns [67].

Second, across all threshold intervals of the information infrastructure level, the coefficient of digital inclusive finance is positive and statistically significant at the 1% level. This suggests that as information infrastructure improves and user scale expands, the facilitating effect of digital inclusive finance on industrial structure upgrading exhibits a pattern of marginal enhancement. The proliferation of mobile network infrastructure has significantly enhanced information mobility, reduced information search costs, improved market efficiency, expanded financial service accessibility, and fundamentally reshaped the coverage density and spatial equilibrium of information flows—thereby providing foundational support for digital inclusive finance to enable industrial structure upgrading.

5. Conclusions and Discussion

5.1. Conclusions

This paper primarily investigates the spatial information transmission effect of digital inclusive finance on industrial structure upgrading in the Yangtze River Delta urban agglomeration from the perspective of spatial information dependence, employing a spatial panel econometric model. The main conclusions are as follows:

(1)

The local impact of digital inclusive finance on industrial structure upgrading exhibits a U-shaped pattern, while spatial spillover effect follows an inverted U-shaped pattern. Following a suite of robustness tests—including the exclusion of extreme samples and replacement of the explained variable—this nonlinear relationship remains statistically significant.

(2)

Digital payment and digital credit constitute the core information flows driving industrial structure upgrading, while the spatial spillover effect of digital insurance lacks statistical significance. All three local business segments significantly facilitate industrial upgrading; however, only payment and credit services generate positive spatial spillovers (coefficients = 0.144 and 0.463, respectively; p < 0.01).

(3)

Further mechanism tests reveal that entrepreneurial activity exerts a partial mediating effect with spatial transmission, whereas the mediating effect of technological innovation information input is only significant locally. Entrepreneurial activity reduces the direct effect of digital inclusive finance by 16.2%, with the mediating effect transmitting across regions. Due to the tacit knowledge attribute of technological innovation information and heterogeneity in absorptive capacity, no significant spatial spillover is observed. The mechanism tests validate the transmission chain: “information release → response of information carriers → optimization of industrial information allocation”.

In conclusion, this study’s findings underscore the pivotal role of digital inclusive finance in advancing industrial structure upgrading, identifying a nonlinear evolutionary path: local information accumulation (U-shaped upward trend) → cross-regional diffusion (rising segment of the inverted U-shape) → over-agglomeration inhibition (declining segment of the inverted U-shape). The results indicate that moderate regional cooperation facilitates resource allocation optimization, yet caution is warranted regarding the “siphon effect” induced by excessive digital finance agglomeration. These empirical analyses complement existing literature’s understanding of the nonlinear spatial transmission mechanism of digital financial information and enrich spatial spillover theory. Future policies should prioritize enhancing the information absorptive capacity of peripheral regions and promoting cross-regional flow of technological innovation information, thereby enabling the balanced distribution of digital finance dividends across regions.

5.2. Policy Recommendations

Based on the empirical findings of this study, actionable policy recommendations are proposed for governments at all levels, financial institutions, and relevant industrial stakeholders in the Yangtze River Delta:

(1)

This study finds that digital inclusive finance exerts a significant spatial spillover effect on industrial structure upgrading, indicating that financial information flow serves as the core driver of regional coordination. Regions should continuously advance the development of digital inclusive finance and co-construct a digital financial information sharing network to enhance the transmission efficiency of transaction information flows, credit information flows, and risk information flows, reduce information friction costs, accelerate digital inclusive finance’s crossing of the critical value, and facilitate the early arrival of the inflection point for industrial structure upgrading.

(2)

The local U-shaped threshold acceleration effect and spatial inverted U-shaped characteristic provide a quantitative basis for tiered policy formulation by governments at all levels. For regions below the critical threshold, financial institutions are encouraged to develop small-denomination, high-frequency simplified digital financial products targeting groups with underdeveloped digital infrastructure. For core cities above the critical threshold, a siphon effect monitoring mechanism and compensation fund should be established to prevent excessive resource agglomeration and mitigate the “inverted U-shaped” siphon effect.

(3)

Payment and credit information flows exert significant positive spatial spillover effects, whereas insurance information flows do not exhibit such spatial spillovers. Governments at all levels and financial institutions should optimize the structure of financial information flows in a classified manner to enhance the synergistic support efficacy for industrial upgrading. Sustained efforts should be directed toward channeling digital payment information flows into industrial extension, thereby providing data support for industrial digital transformation; optimizing the structure of digital credit information flows, drawing on the experience of “three-tier service” models (ultra-simplified application, ultra-fast approval, and ultra-high-quality service), simplify approval processes, and improving the information response speed for small and micro enterprises.

(4)

The “information decoding” function that activates entrepreneurial dynamism should be leveraged to cultivate the peripheral technological absorption capacity. This study finds that entrepreneurial information can spill over naturally: the “market opportunity information” embedded within it is highly accessible for capture and imitation, whereas technological patents require other regions to first enhance absorption capacity. Thus, peripheral regions should be encouraged to collaborate with universities and research institutions in core cities, and local technology transfer talents should be cultivated by drawing on the G60 Sci-Tech Innovation Manager System currently being advanced [68].

5.3. Limitations and Directions for Future Research

While this study employs fixed-effects models and spatial econometric models to mitigate spatial endogeneity, unobserved time-varying and policy confounding factors may still introduce biases into the estimates. Thus, future research could adopt quasi-experimental designs—such as policy shocks or instrumental variable (IV) methods rooted in digital finance or inclusive finance—to further elucidate the impact of digital inclusive finance on industrial structure upgrading.

The Yangtze River Delta region exhibits distinct regional integration characteristics, which may not be fully generalizable to other regions in China or even globally. Thus, additional empirical data and evidence are required to further analyze the relationship between digital inclusive finance and industrial structure upgrading. Future research could conduct comparative analyses across provinces with diverse geographical and institutional attributes, or even in other countries worldwide, to enhance the generalizability and transferability of the research findings. Despite the aforementioned limitations, this study offers valuable empirical insights into the nexus between digital financial information and industrial structure, and deepens the understanding of spatial information transmission.