How Digital Technology Shapes the Spatial Evolution of Global Value Chains in Financial Services

Abstract

Rapid advances in digital technologies are reshaping value creation and the trade landscape of global financial services, yet the channels through which they influence the spatial evolution of financial services global value chains (GVCs) remain insufficiently identified. Using a global panel of 52 countries over 2013–2021, we estimate a dynamic Spatial Durbin Model (SDM) to identify overall effects and quantify spatial spillovers and temporal dynamics. We then combine Geographically and Temporally Weighted Regression (GTWR) with spatial mediation models to examine heterogeneity and underlying mechanisms. Our findings show that digital technology significantly drives the spatial evolution of financial services GVCs. Its influence is dominated by spatial diffusion, exhibiting a dynamic pattern of a strong short-run boost followed by long-run reallocation. This dynamic effect is not homogeneous; rather, it reflects a pronounced dual-driver structure: the momentum is more robust when human capital and R&D output reinforce each other, whereas increases in innovation level alone are unlikely to translate into sustained impetus for spatial restructuring. Crucially, digital technologies reshape GVC geography through three core channels: attenuating distance decay, strengthening spatial proximity, and amplifying spatial heterogeneity. These forces deepen the domestic diffusion of knowledge, capital, and technology and extend their spillovers to neighboring and connected economies. The results provide robust empirical evidence on financial geography in the digital era and have clear implications for policies that facilitate cross-border financial services and strengthen regional coordination in support of the 2030 Agenda for Sustainable Development, particularly SDG 8 (financial inclusion) and SDG 10 (global financial governance).

1. Introduction

In recent years, rapid advances in information and communication technologies (ICT) have profoundly reshaped the global division of production, especially in manufacturing, where production processes have been unbundled into multiple discrete stages and outsourced worldwide. This trend gave rise to the concept of global value chain [1]. In a similar vein, the swift development of digital technologies—supported by sustained R&D investment—has been transforming the production and delivery of services through deeper digitalization, standardization, and remote provision. By loosening the geographic constraints that traditionally bounded services trade, these changes have made the spatial organization of service production increasingly dispersed and modular, thereby fostering the emergence of the global service value chains. The degree of digital reshaping varies across industries. Financial services not only occupy pivotal positions in global service networks but also provide critical support for the efficient functioning of the world economy. Yet recent value-added trade statistics show that, relative to other service sectors, trade flows in financial services have declined significantly in recent years. This pattern suggests that the organization of financial services GVCs is undergoing substantial change, and that countries’ R&D capacity and innovation foundations in digital technology may be an important force behind this spatial reconfiguration. Against this backdrop, systematically identifying how digital technology influences the spatial evolution of financial services GVCs is of both theoretical and practical significance.

The literature related to this study can be broadly grouped into three strands. The first examines how digital technologies shape global value chain. Existing research has focused on changes in participation patterns, functional upgrading, organizational restructuring, and system resilience [2,3,4,5,6]. Overall, digital technologies have injected new dynamism into GVC by reshaping modes of production organization and the mechanisms of value creation and distribution, while improving supply–demand matching and information coordination [7,8]. Technologies such as artificial intelligence also appear to complement or substitute for traditional labor across sectors, thereby enhancing service accuracy and automation [9]. Further studies suggest that rising productivity in digitally intensive services has opened new opportunities for the expansion of the global service value chains [10,11,12].

The second strand of literature focuses on how digital technologies reshape the spatial evolution of global value chain. Against the backdrop of the new wave of technological revolution and industrial transformation, digitalization weakens geographic constraints and reduces cross-border coordination and search costs, thereby altering location choices for production and services as well as decisions regarding offshoring and reshoring. In doing so, it reconfigures the spatial layout and division of labor within global value chain [13,14]. At the same time, by improving the efficiency of cross-regional factor mobility and amplifying the comparative advantages associated with institutional differences, digital technologies tend to concentrate higher-end activities in a limited number of hub regions, while standardized and modular tasks are more likely to relocate to locations with lower costs or more permissive regulatory environments. This dynamic further accelerates the uneven spatial evolution of global value chain [15,16,17].

The third strand of literature examines the impacts of digital technologies on financial services. As a concentrated manifestation of digitalization in the financial domain, fintech has become a key engine reshaping financial systems. On the one hand, fintech is expected to expand access to financial services through digital platforms and new business models, thereby enhancing financial inclusion for vulnerable groups and small and medium-sized enterprises. On the other hand, by transcending the geographic limits of traditional branch-based networks, it offers households and firms more diverse and competitive sources of financing, fundamentally altering established financial landscapes. In practice, a growing number of financial products and institutions are rapidly adopting innovations such as big data, artificial intelligence, and cloud computing to improve and automate a wide range of financial services [18,19,20]. The literature further shows that big data and AI enhance credit assessment and risk identification capabilities [21,22], robotic process automation improves back-office processing and compliance efficiency [23], the Internet of Things and edge computing facilitate the cross-regional deployment of front-end service nodes [24,25], and blockchain offers new pathways for auditing, risk management, and financial market transformation [26,27]. Taken together, these technological shifts help loosen the constraints that geographic distance imposes on the efficiency of financial transactions, contributing to more complex patterns of reorganization in both the geographic distribution and operational processes of financial services GVCs.

Building on the existing literature, a key unresolved issue is how digitalization reshapes the cross-border structure of financial services GVCs through spatial mechanisms. Financial geography emphasizes the spatial embeddedness of financial activities, network structures, and the dynamic evolution of core–periphery configurations, offering an important lens for understanding regional differentiation and hierarchical features of financial linkages [28]. The global production network/GVC literature further suggests that digitalization is transforming modes of cross-border organization and coordination, and may trigger a reallocation of functions across value-chain nodes as well as a reconfiguration of spatial coupling relationships [29,30]. Given that much of the extant research focuses on a single technology or a single-country setting, it remains difficult to systematically identify cross-economy spatial restructuring and cross-border transmission mechanisms. Against this backdrop, drawing on a geographical and spatial-interaction perspective, this study addresses three interrelated research questions: (1) How does digital technology influence the spatial evolution of the financial services GVCs in the short run versus the long run? (2) In the presence of substantial cross-country differences in productive capacity and trade conditions, does the impact of digital technology on the spatial evolution of the financial services GVCs exhibit significant spatial heterogeneity? (3) Does digital technology affect the spatial evolution of the financial services GVCs through spatial mechanisms such as distance decay, spatial proximity, and spatial heterogeneity, and if so, how? Correspondingly, Section 2.1 derives and proposes the theoretical hypotheses (H1–H3), which are then examined systematically in Section 3 and Section 4.

The remainder of the paper is organized as follows. Section 2 presents the research design, develops the theoretical analysis and corresponding hypotheses, and describes the sample selection, data sources, and model specifications. Section 3 reports the empirical results and examines both the linear and nonlinear effects of digital technology on the spatial evolution of financial services GVCs. Section 4 conducts mechanism analyses to identify and test the mediating roles of distance decay, spatial proximity, and spatial heterogeneity. Section 5 summarizes the main findings and discusses the policy implications.

2. Research Design

2.1. Theoretical Analyses

Agglomeration economies and industrial clusters are highly sensitive to spatial distance: the shorter the distance, the stronger the agglomeration effects and externalities [31]. In finance, agglomeration not only generates persistent gains for local economic development but also diffuses to surrounding areas through spatial interdependence [32]. Viewing financial firms as cross-border, multi-location actors helps disentangle different types of distance effects and enables a more fine-grained identification of location choices. Multinational enterprises do not choose locations randomly within a country; rather, they tend to embed themselves in financial agglomerations with specific advantages [33] and integrate upstream and downstream activities around bottleneck assets, thereby forming “optimal but asymmetric” geographic footprints [34]. Within this framework, the optimal site for a given production or service stage depends not only on local marginal costs but also on spatial proximity to adjacent stages and the associated coordination and linkage costs [35]. Against the backdrop of concurrent globalization and informatization, the architecture of financial systems has been continuously reshaped. The spatial distribution and concentration of financial service institutions (firms, branches, ATMs) vary markedly across places, jointly shaped by location, market demand, agglomeration externalities, transport accessibility, and the scale and structure of population [36]. Meanwhile, digital platforms have increasingly become key infrastructures linking value-chain activities with end customers. As firms deepen their adoption of internet platforms, the average geographic distance to their customers expands significantly, and value creation becomes more reliant on customer-oriented information acquisition and processing [37]. Importantly, the impact of digital technology on financial network structures appears to be stage-dependent. In the early phase, financial institutions often reallocate resources and costs toward online channels and platform-based services by closing inefficient branches, compressing ATM networks, and consolidating back-office functions. This adjustment may lead to a short-term contraction of physical networks and a reduction in spatial nodes, which is reflected in a decline in the average weighted degree. As digital infrastructures mature and online and platform-based businesses expand, new cross-border connections and multilateral matches may emerge, with network density likely to rebound. Taken together, by reducing information search, matching, and coordination costs and by strengthening remote delivery and cross-jurisdictional collaboration, digital technology substantively weakens the geography-centered “distance-decay” constraint. It extends the reach of financial service linkages and reorganizes the functions and hierarchical structure of nodes, thereby steering the spatial configuration of financial services GVCs toward a more dispersed, cross-regional, and networked pattern.

Accordingly, we propose:

H1. 

Advances in digital technology influence the spatial evolution of financial services GVCs by weakening the distance-decay effect.

Against the backdrop of the pervasive penetration of digital technologies, the organization of value chains and their spatial configurations are being reshaped [38]. Traditional offline finance faces clear structural constraints in spatial diffusion, whereas digitalization opens new channels for financial development to spread across space. Empirical evidence shows that digital finance exerts a significant spatial stimulus, particularly through inter-city diffusion and penetration [39]. Technological innovation and digital finance also display a layered diffusion pattern accompanied by pronounced spillover and proximity effects [40]. From the supply side, supported by technological advances and efficient information flows, digital finance relaxes the time and geographic constraints of conventional financial services. It can foster the sustainable development of firms in both local and neighboring areas [41] and, through new business models, ease financing constraints for innovative actors while improving financing modes and resource allocation [42]. From the demand side, the diffusion of internet retail and digital consumption evolves over time and spreads along spatial proximity: proximity effects dominate in the early stage, while similarity effects become more salient later [43]. Recent studies further suggest that incorporating digital proximity into a multidimensional proximity framework is essential. Digital proximity can substitute for geographic and institutional proximity, while complementing social proximity. Compared with low levels of digital proximity, higher digital proximity significantly promotes inter-city innovation collaboration and generates stronger interaction effects with traditional dimensions of proximity. For peripheral cities, strengthening digital proximity to core cities helps enhance their network positions within regional innovation systems [44]. Taken together, these insights imply that digital technology may reinforce the spatial proximity effect—that is, by amplifying the frequency of local–nearby interactions, improving data-matching efficiency, and deepening collaborative intensity within urban agglomerations and regional networks. This, in turn, can extend and densify financial service linkages, strengthen complementary specialization and functional coupling among nodes, and steer the spatial evolution of financial services GVCs toward more intensive regional coordination, tighter urban-network structures, and more efficient local–neighboring connectivity.

Accordingly, we propose:

H2. 

Advancements in digital technology promote the spatial evolution of financial services GVCs by strengthening the spatial proximity effect.

Digital technology not only reshapes the organization and spatial configuration of value chain; its impacts also exhibit pronounced spatial heterogeneity across countries and regions [45]. A direct real-world basis for this heterogeneity is that factors such as the size of the digital user base, the level of digital infrastructure, the intensity of government support for science and technology, venture capital availability, the depth of the knowledge base, and market size differ in their strength and relative importance across regions. Moreover, different types of digital-technology activities rely on these determinants to varying degrees [46]. From a global value chain perspective, advanced economies are increasingly specializing in high-skill-intensive segments, whereas emerging economies tend to allocate more resources to capital-intensive activities [47]. Although automation can raise productivity and reduce relative labor costs—thereby facilitating firms’ integration into GVC—its effects vary substantially across regions, technology levels, and GVC types [48]. Whether a country possesses “absolute advantage” largely determines its ability to enter GVC, while participation modes and value capture are also highly differentiated across subnational regions within the same country [49]. Meanwhile, firms’ digitalization strategies and internationalization strategies are mutually reinforcing and jointly constrained by cross-country differences in institutional environments and resource endowments [50]. Viewing financial-service actors as cross-border, multi-location entities helps refine our understanding of “border effects”: although national borders manifest as discontinuities in spatial structures, the location choices and functional configurations of these actors are more likely driven by heterogeneous regional conditions such as institutional quality, digital infrastructure, and knowledge networks [33]. Taken together, digital technology is unlikely to homogenize spatial disparities uniformly. Instead, it may amplify and select “advantageous” spatial heterogeneity, steering the financial services GVCs along an asymmetric “advantage–constraint” trajectory. This can strengthen agglomeration and functional upgrading in nodes with abundant digital factors and superior institutions, while generating functional spillovers and cross-border complementarities in surrounding networks, thereby advancing the spatial reorganization and hierarchical restructuring of the value chain.

Accordingly, we propose:

H3. 

The application of digital technology promotes the spatial evolution of financial services GVCs through the spatial heterogeneity effect.

2.2. Model Construction

Traditional OLS panel regressions typically assume cross-sectional independence among observational units, making them ill-suited to capture spatial correlation and spillover effects in countries’ positioning within financial services GVCs [51]. Accordingly, we adopt a spatial econometric framework to systematically examine how digital technology shapes the spatial evolution of financial services GVCs. Specifically, we use a dynamic Spatial Durbin Model as the baseline specification to identify the overall effect and its spatial diffusion. We then incorporate geographically and temporally weighted regression (GTWR) to map spatio-temporal heterogeneity in effect magnitudes, and estimate a spatial mediation model that explicitly tests three channels: the distance-decay effect, the spatial proximity effect, and the spatial heterogeneity effect.

2.2.1. Dynamic Spatial Durbin Model

The financial services GVCs network exhibits pronounced path dependence and spatial interdependence: a country’s current network position depends not only on its contemporaneous level of digital technology but also on the previous-period network structure and on changes in the networks of neighboring economies. Meanwhile, unobserved common shocks—such as cross-border financial cycles, regulatory adjustments, or shifts in global risk appetite—may co-move across geographically or economically proximate countries, implying spatial dependence in the unobserved components captured by the error term and thus a spatial error structure. Taken together, these two sources of spatial dependence motivate the use of the dynamic Spatial Durbin Model, which, within a unified framework, captures the spatio-temporal dynamics of the dependent variable and the spatial spillovers of the explanatory variables, while allowing a more general spatial specification that helps absorb error dependence arising from shared shocks among neighboring countries. Accordingly, we adopt the dynamic SDM as our benchmark empirical model [52].

$$\begin{array}{l}\log {\mathrm{deg}}_{it}=\mathrm{\Psi }+\phi \log {\mathrm{deg}}_{it-1}+\delta {\displaystyle \sum _{j\ne i}^n{w}_{ij}}\log {\mathrm{deg}}_{jt}+\theta {\displaystyle \sum _{j\ne i}^n{w}_{ij}}\log {\mathrm{deg}}_{jt-1}+{\lambda }_1\log dt+{\lambda }_{2}peo+{\lambda }_3\log m\\ +{\lambda }_{4}urban+{\lambda }_{5}fd+{\lambda }_6\log fdi+{\lambda }_{7}dof+{\vartheta }_1{\displaystyle \sum _{j\ne i}^n{w}_{ij}}\log dt+{\vartheta }_2{\displaystyle \sum _{j\ne i}^n{w}_{ij}}peo+{\vartheta }_3{\displaystyle \sum _{j\ne i}^n{w}_{ij}}\log m\\ +{\vartheta }_4{\displaystyle \sum _{j\ne i}^n{w}_{ij}}urban+{\vartheta }_5{\displaystyle \sum _{j\ne i}^n{w}_{ij}}fd+{\vartheta }_6{\displaystyle \sum _{j\ne i}^n{w}_{ij}}\log fdi+{\vartheta }_7{\displaystyle \sum _{j\ne i}^n{w}_{ij}}dof+{\epsilon }_{it}\end{array}$$

(1)

where $i$ and $j$ denote countries $(i,j=1,\dots ,N)$ and $t$ denotes year. The dependent variable $\log {\mathrm{deg}}_{it}$ measures the average weighted degree of country $i$ in year $t$ in the financial services GVCs network. $\log {\mathrm{deg}}_{it-1}$ is its first-order time lag, and ${\displaystyle \sum _{j\ne i}^n{w}_{ij}}\log {\mathrm{deg}}_{jt-1}$ is the corresponding spatial lag term. The key explanatory variable $\log dt$ denotes the digital technology index. $peo$ captures the labor structure, $\log m$ the manufacturing structure, $urban$ the level of urbanization, $fd$ the level of financial development, $\log fdi$ foreign direct investment, and $dof$ the institutional environment (see Section 2.3.3 for detailed definitions and data sources). ${\lambda }_2-{\lambda }_7$ are the coefficients on these control variables, and ${\vartheta }_2-{\vartheta }_7$ are the coefficients on their spatial lags. $\phi$ and $\theta$ denote the elasticities associated with the temporal lag effect and the spatio-temporal lag effect, respectively. $\mathrm{\Psi }$ represents fixed effects, $\delta$ is the corresponding coefficient for spatial spillovers, $w$ is the spatial weight matrix, and $\epsilon$ is the error term.

In estimation, following Elhorst [53], we decompose the impacts of the dynamic SDM into short- and long-run effects, as well as direct and indirect effects.

2.2.2. Geographical and Temporal Weighted Regression (GTWR)

Digital technology may exert markedly different effects across countries and over time. To further capture such spatio-temporal heterogeneity, we employ a geographically and temporally weighted regression (GTWR) model. Compared with the conventional geographically weighted regression (GWR), GTWR allows model coefficients to vary simultaneously across space and time, making it better suited to identifying the spatio-temporal unevenness inherent in the diffusion of digital technologies [54].

$$Y_i={\beta }_0\left(u_i,v_i,t_i\right)+{\displaystyle \sum _{k=1}^N{\beta }_k}\left(u_i,v_i,t_i\right)X_{ik}+{\epsilon }_i$$

(2)

$\left(u_i,v_i\right)$ is the latitude and longitude coordinates of each country, $\left(u_i,v_i,t_i\right)$ is the spatial and temporal coordinates of the country i, and $t_i$ is the year of observation; $Y_i$ denotes the dependent variable, $X_{ik}$ denotes the Kth explanatory variable of the country $i$, and $N$ is the number of explanatory variables. ${\beta }_0\left(u_i,v_i,t_i\right)$ is the intercept term; ${\beta }_k\left(u_i,v_i,t_i\right)$ is the estimated coefficient for the Kth country; and ${\epsilon }_i$ is the error term. We adopt an adaptive kernel function and select the optimal bandwidth using the AICc criterion to obtain local parameter estimates that vary across countries and years. Based on these estimates, we map the spatiotemporal distribution of the effect of digital technology on the average weighted degree of the financial services GVCs network, thereby identifying regional differences in digitalization effects and their evolution over time.

2.2.3. Spatial Mediating Effect Model

Theoretical analysis indicates that digital technology may influence the spatial evolution of financial services GVCs through distance-decay, spatial proximity, and spatial heterogeneity effects. To test these hypotheses, we adopt a standard mediation model and estimate it using a spatial panel econometric approach. Specifically, the average weighted degree (logdeg) is taken as the dependent variable Y; the distance-decay effect (dishar), the spatial proximity effect (inter), and the spatial heterogeneity effect are treated as mediators R; digital technology (logdt) is regarded as the explanatory variable; and other variables are taken as the control variables Z. ${\mu }_i$ denotes area fixed effects; the random error term by ${\epsilon }_{it}$; and the remaining parameters are coefficients to be estimated. Following Cheng and Wang [55], the model is specified as:

$$\log {\mathrm{deg}}_{it}={\beta }_0+{\rho }_1{\displaystyle \sum _{i=1}^n{w}_{ij}}\log {\mathrm{deg}}_{jt}+{\beta }_1\log d{t}_{it}+{\rho }_2{\displaystyle \sum _{i=1}^n{w}_{ij}}\log d{t}_{jt}+{\beta }_x{Z}_{it}+{\rho }_x{\displaystyle \sum _{i=1}^n{w}_{ij}}{Z}_{jt}+{\mu }_i+{\epsilon }_{it}$$

(3)

$$R_{it}={\chi }_0+{\pi }_1{\displaystyle \sum _{i=1}^n{w}_{ij}}{R}_{jt}+{\chi }_1\log d{t}_{it}+{\pi }_2{\displaystyle \sum _{i=1}^n{w}_{ij}}\log d{t}_{jt}+{\chi }_x{Z}_{it}+{\pi }_x{\displaystyle \sum _{i=1}^n{w}_{ij}}{Z}_{jt}+{\mu }_i+{\epsilon }_{it}$$

(4)

$$\begin{array}{l}\log {\mathrm{deg}}_{it}={\gamma }_0+{\theta }_1{\displaystyle \sum _{i=1}^n{w}_{ij}}\log {\mathrm{deg}}_{it}+{\gamma }_1\log d{t}_{it}+{\gamma }_2{R}_{it}+{\theta }_2{\displaystyle \sum _{i=1}^n{w}_{ij}}\log {\mathrm{deg}}_{jt}\\ +{\theta }_3{\displaystyle \sum _{i=1}^n{w}_{ij}}{R}_{jt}+{\gamma }_x{Z}_{it}+{\theta }_x{\displaystyle \sum _{i=1}^n{w}_{ij}}{Z}_{jt}+{\mu }_i+{\epsilon }_{it}\end{array}$$

(5)

According to the stepwise method commonly used in the mediation effect test [56], the specific test steps are as follows: the first step, test the coefficient of the explanatory variable β₁, if β₁ is significant, it is considered that there is a mediation effect; but regardless of whether β₁ is significant or not, we need to carry out the subsequent steps; The second step is to test the explanatory variable coefficient χ₁ and the regression coefficient γ₂ in turn. If both are significant, it indicates that the mediating effect exists. The third step, test whether the coefficient γ₁ of the explanatory variable is significant, if it is significant, indicating that the direct effect exists, the mediation effect is partially mediated; if it is not significant, indicating that the direct effect does not exist, the mediation effect is fully mediated.

2.3. Variables and Data

To measure the impact of digital technology on the spatial evolution of financial services GVCs and its mechanisms, we assemble data for 52 countries over 2013–2021; the sample is determined by data availability. The dependent variable uses financial services GVCs data from the ADB Multi-Region Input–Output (ADB-MRIO) database as provided by the University of International Business and Economics (UIBE) GVC Index database. The core explanatory variable is taken from the digital-technology component of the Global Digital Economy Development Index (TIMG 2023) compiled by the Institute of Finance, Chinese Academy of Social Sciences (CASS). Other country-level control variables are primarily drawn from the World Bank database and the United Nations Conference on Trade and Development (UNCTAD) database.

2.3.1. Network Metrics of Financial Services Global Value Chains

In the course of fine-grained division of labor and modularization in services production, modes of value creation often differ from those in manufacturing value chains. Value in services does not always unfold along a linear chain; more often it is realized through networked activities. Baldwin and Venables [57] refer to this as a “spider” value chain. Accordingly, we employ social network analysis (SNA) to characterize the linkage structure of financial services GVCs from the perspective of overall network features. SNA quantifies “relations” among actors (nodes) to reveal their interactions and network positions [58]. To provide a visual overview of the spatial structure of the financial services GVC network, Figure 1 presents the network in 2013, whereas Figure 2 presents the network in 2021. We use the average weighted degree (logdeg) to capture the average intensity of network linkages, defined as the sum of weighted degrees across all nodes divided by the total number of nodes. A higher weighted degree for a given economy indicates denser cross-border financial services ties with other countries, suggesting that it is more likely to function as a hub-type financial center or a highly interconnected node within the financial services GVCs network. The specific indicators are defined as follows:

$$S={\displaystyle \frac{1}{N}}{\displaystyle {\sum }_i^N{W}_{ij}}$$

(6)

where S is the average weighted degree of the network, N is the number of nodes, and $W_{ij}$ is the weight of node i and node j, that is, the connection strength.

2.3.2. Digital Technology Index (TIMG2023)

Digital technology (logdt) is primarily composed of three dimensions—R&D output, human capital, and innovation level [59]. Its annual distribution from 2013 to 2021 is shown in Figure 3. Firstly, R&D output is the scientific research results after a country has made material and human capital investment, which is a direct reflection of the level of digital technology, which is measured by the indicators of patent application and paper publication related to digital technology. Secondly, the accumulation of human capital is an important basis for the progress of digital technology, which is evaluated through the use of indicators reflecting the degree of access to higher education and the quality of education in related fields. Finally, innovation level serves as a key driver and enabling condition for digital technology development. This component draws on World Economic Forum (WEF) survey data to provide a qualitative assessment of a country’s innovation capacity and the extent of university–industry collaboration.

2.3.3. Control Variables

To more accurately identify the effect of digital technology on the spatial evolution of financial services GVCs, we introduce a set of country-level control variables covering institutions, economic structure, resource allocation, and development [60,61,62,63].

Table 1. Descriptive statistics of the main variables.

Variable	Obs	Mean	Std. Dev.	Min	Max
logdeg	468	3.8528	0.7053	1.79	5.46
logdt	468	1.7418	0.1403	1.18	1.97
peo	468	0.6433	0.1485	0.24	0.89
logm	468	5.0738	0.6798	3.16	6.52
urban	468	1.0947	0.3016	0.29	1.61
fd	468	0.8571	0.3565	0.17	1.52
logfdi	468	5.2404	0.6951	3.52	7.12
dof	468	7.4507	0.7119	4.82	8.83

reports descriptive statistics for the main variables.Specifically: foreign direct investment (logfdi)—log of FDI stock, proxying technology spillovers and capital mobility that may upgrade financial services capacity; institutional environment (dof)—the Heritage Foundation’s Index of Economic Freedom, capturing the role of institutional quality in market openness and GVC participation; labor structure (peo)—service employment as a share of total employment, estimated following the ILO approach, reflecting human-resource support for the sector; manufacturing structure (logm)—industry value added as a share of GDP, gauging the coupling between the manufacturing base and service embedding/outsourcing; urbanization (urban)—the urbanization rate, capturing agglomeration and population-mobility effects on financial services supply; and financial development (fd)—the IMF Financial Development Index, proxying financial resource allocation capacity and the diffusion of innovation.

2.3.4. Setting of Spatial Weight Matrix

This paper uses a combination of geographic distance matrices, economic distance matrices, and nested matrices of geographic and economic distances for spatial econometric analysis. The elements in the geographic distance weight matrix can be set to $W_{d_{ij}}=1/d_{ij}$, where $d_{ij}(i\ne j)$ is the distance between the two countries, calculated using the latitude and longitude coordinates of the capital cities of the two countries. Economic distance spatial weight matrix $W_{pergdp}=1/\mid {\overline{Q}}_i-{\overline{Q}}_j\mid$, where ${\overline{Q}}_i-{\overline{Q}}_j$ is the difference between the mean GDP per capita of the two countries over the period 2013 to 2021. In the specification of spatial weights, the use of only geographic or only economic distance may bias estimates of spatial dependence across regions. Following Parent and LeSage [64], a nested spatial weight matrix is constructed by combining geographic and economic distances $W_{de}=\psi W_d+\left(1-\psi \right)W_e$, where $\psi$ is set to 0.5. All matrices are row-standardized prior to estimation, so that each row sums to one.

3. Empirical Analyses

To assess the appropriateness of adopting a spatial econometric framework, we first examine the spatial dependence of logdt and logdeg using the global Moran’s I, and then conduct model selection based on LM, Hausman, LR, and Wald diagnostics. The results indicate that both logdt and logdeg exhibit significant positive spatial autocorrelation over the sample period, and that a fixed-effects, time-lagged dynamic SDM provides the most suitable specification. The corresponding test statistics are reported in Appendix A

Table A1. Annual Global Moran’s I for Digital Technology and Financial Services GVCs, 2013–2021.

Year	logdeg			logdt
	I	Z	p-Value	I	Z	p-Value
2013	0.225	3.680	0.000	0.306	4.974	0.000
2014	0.221	3.620	0.000	0.327	5.252	0.000
2015	0.229	3.727	0.000	0.327	5.249	0.000
2016	0.226	3.679	0.000	0.309	4.942	0.000
2017	0.207	3.401	0.001	0.309	4.959	0.000
2018	0.207	3.409	0.001	0.306	4.917	0.000
2019	0.211	3.472	0.001	0.306	4.913	0.000
2020	0.212	3.485	0.001	0.312	4.995	0.000
2021	0.196	3.246	0.001	0.301	4.836	0.000

and

Table A2. Spatial Model Diagnostics and Selection.

	${\mathit{\chi }}^2$ Statistical Quantities	p-Value
LMError(Burrideg) test	38.071	0.000
LMError(Robust) test	35.291	0.000
LMLag(Anselin) test	57.475	0.000
LMLag(Robust) test	54.695	0.000
LR(SLM) test	30.54	0.0001
LR(SEM) test	29.56	0.0001
Wald(SLM) test	31.05	0.0001
Wald(SEM) test	29.50	0.0001
Hausman test	16.36	0.0220

.

3.1. Analysis of the Dynamic Spatial Durbin Model Results

As shown in

Table 2. Estimation results of DSDM model under different spatial weight settings.

Variable	Geographic Distance Matrix	Economic Distance Matrix	Nested Matrix
L.logdeg	0.960 ***	0.218 ***	0.234 ***
	(0.0496)	(0.0602)	(0.0602)
logdt	−0.525 ***	−0.575 ***	−0.559 ***
	(0.177)	(0.177)	(0.177)
peo	0.610	−0.135	−0.131
	(0.381)	(0.314)	(0.314)
logm	0.0582	0.213 **	0.206 *
	(0.137)	(0.106)	(0.106)
urban	−0.149	−0.0668	−0.0589
	(0.215)	(0.189)	(0.189)
fd	0.142 **	0.175 ***	0.167 ***
	(0.0631)	(0.0610)	(0.0610)
logfdi	−0.0862 ***	0.0675 **	0.0676 **
	(0.0326)	(0.0269)	(0.0269)
dof	0.115 ***	−0.0398 **	−0.0378 *
	(0.0229)	(0.0202)	(0.0202)
N	416	416	416
R2	0.434	0.292	0.092

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

, the estimates from the dynamic spatial Durbin model indicate that the lagged term of average weighted degree is significantly positive, suggesting pronounced path dependence in the spatial evolution of the financial services GVCs. The persistence of existing cross-border business ties, the continuity of platform and infrastructure configurations, and the inertia of regulatory coordination imply that current network link intensity largely evolves along prior structures, reflecting a logic of gradual adjustment rather than abrupt shifts. The coefficient on the digital technology index is significantly negative, indicating that, within the sample period, improvements in digitalization are accompanied by a contemporaneous decline in average weighted degree. A plausible explanation is that, during digital transformation, traditional financial institutions progressively streamline low-efficiency offline coverage and shift more activities to online channels and platform-based settings, leading to a transitional drop in the measured intensity of cross-border linkages. Importantly, this negative association does not necessarily imply that digital technology weakens the momentum of spatial upgrading in the financial services GVCs. Instead, it may capture a restructuring process in which financial linkages move from conventional physical and organizational forms toward more digital, platform-based, and multi-nodal configurations. Micro-level evidence suggests that firm digitalization reduces customer and supplier concentration, dispersing transactions across a broader set of partners [65], while government open data further weakens dependence on a small number of key counterparties [66]. At the macro level, further evidence points to a stage-dependent or threshold-type relationship between digital technology and GVC position [67], offering an additional rationale for this “transitional negative” pattern.

Regarding the control variables, financial development remains robustly positive. This suggests that more mature financial systems provide a stable institutional and market foundation for cross-border specialization and networked coordination within financial services GVCs. They do so by improving capital allocation efficiency and strengthening cross-border payment, settlement, and risk-sharing mechanisms. By contrast, the overall insignificance of the labor-structure and urbanization variables may indicate that, as digital channels and remote delivery expand, financial service linkages no longer primarily hinge on conventional urban agglomeration or the physical mobility of personnel. Instead, they increasingly depend on “non-geographic” enabling conditions such as digital infrastructure, the organization of data factors, and compliance capacity. Manufacturing structure, foreign direct investment, and the institutional environment display heterogeneous estimates across alternative spatial weight matrices, pointing to clear spatial-context dependence. When economic ties are tighter, cross-border manufacturing and trade may generate more concentrated demand for supply-chain finance, trade credit, and risk-hedging services, thereby more readily strengthening the link intensity of the financial services GVCs. The positive contribution of FDI likewise appears to rely on deep embedding in local industrial chains, markets, and regulatory systems. Institutional differences, in turn, may translate into friction costs in key areas such as cross-border data flows, KYC/AML, and cybersecurity, undermining the efficiency of cross-border coordination among financial service segments. Taken together, the impact of digital technology on the spatial evolution of the financial services GVCs should be interpreted within the joint framework of financial-system foundations, industry-driven demand, and cross-border rules and regulatory coordination.

Parent and LeSage [64] caution that inferring the effects of regressors—and their spatial spillovers—directly from the raw coefficient estimates of the spatial Durbin model (SDM) or its dynamic variant (DSDM) can be misleading. Elhorst [53] likewise emphasizes that one should derive average direct and indirect (spillover) effects from the estimated model. Accordingly, using the coefficient estimates of the DSDM reported in Table 2, we further compute the direct and indirect effects of digital technology on the spatial evolution of financial services GVCs. Because the dynamic SDM includes a lagged dependent variable, these effects are decomposed into short-run and long-run components (

Table 3. Short- and long-run estimates of the direct and indirect effects of digital technology on the spatial evolution of financial services GVCs.

Matrix Type	Variable	Short-Term			Long-Term
		Direct Effects	Indirect Effects	Total Effect	Direct Effects	Indirect Effects	Total Effect
Geographic Distance Matrix	logdt	−0.673 (0.427)	10.14 *** (2.875)	9.467 *** (2.969)	−0.0868 (0.605)	2.018 *** (0.593)	1.931 *** (0.499)
	control variables	YES	YES	YES	YES	YES	YES
Economic Distance Matrix	logdt	0.878 (7.642)	3.219 (7.625)	4.098 *** (0.464)	−0.832 *** (0.251)	4.383 *** (0.529)	3.551 *** (0.404)
	control variables	YES	YES	YES	YES	YES	YES
Nested Matrix	logdt	1.357 (29.19)	2.731 (29.17)	4.088 *** (0.493)	−0.775 *** (0.248)	4.255 *** (0.540)	3.480 *** (0.422)
	control variables	YES	YES	YES	YES	YES	YES

Standard errors in parentheses *** p < 0.01.

).

Overall, across alternative spatial weight specifications, the directions of effects are broadly consistent; in most cases, short-run total effects exceed their long-run counterparts. The indirect (spillover) effect of digital technology is positive and significant at the 1% level, whereas the direct effect is generally insignificant, indicating pronounced cross-border spatial spillovers—digitalization reshapes financial services GVCs primarily by influencing neighboring or economically connected countries. These results suggest that digitalization expands the availability and geographic reach of financial services, strengthens financial inclusion, and creates conditions for inclusive and sustainable development. At the same time, persistent cross-country differences in institutions and regulation raise regulatory complexity and compliance costs in global markets. Policy should therefore prioritize cross-border supervisory cooperation and mutual recognition of rules and develop unified or interoperable frameworks for key functions—know-your-customer/anti-money-laundering (KYC/AML), data governance, and cross-border payments. Subject to prudent risk management, harnessing the benefits of digital finance to promote complementarities between inclusive finance and green finance can help advance the Sustainable Development Goals (SDGs).

3.2. Analysis of the Results of the Geographical and Temporal Weighted Regression Model

Prior to estimating the GTWR, we standardize all variables to ensure comparability across space and time and to mitigate scale effects. To assess potential multicollinearity among the regressors, we compute variance inflation factors (VIFs). All VIFs are below the commonly used threshold of 10 [68], suggesting that multicollinearity is not a material concern.

Table 4. Relevant parameters of Geographical and Temporal Weighted Regression.

Bandwidth	Sigma	Residual Squares	AICc	R2	R2 Adjusted	Spatio-Temporal Distance Ratio
0.1127	0.1482	10.0775	−306.463	0.9562	0.9555	0.2688

reports the GTWR estimates. In terms of goodness of fit, both R² and adjusted R² exceed 0.95, indicating that the GTWR captures the relationship between the explanatory variables and the dependent variable well.

There is pronounced spatial heterogeneity in how the explanatory variables relate to the spatial evolution of financial services GVCs. Specifically, regions with high positive local coefficients for digital technology are concentrated in parts of Asia and Africa, whereas regions with large negative local coefficients are concentrated in the United Kingdom, Turkey, Thailand, Switzerland, Bulgaria, Belgium, Sweden, Slovenia, Austria, Slovakia, Spain, Australia, Singapore, and Denmark—mostly high-income economies. Overall, digital technology is positively associated with the spatial evolution of financial services GVCs, but both the sign and magnitude of the local coefficients vary markedly across countries. Foreign direct investment (FDI) exhibits predominantly positive local GTWR coefficients, with marked spatial differentiation: coefficients are negative in Australia and parts of Europe, whereas positive coefficients prevail elsewhere, especially in North America. Other covariates likewise show clear spatial variation in their local coefficients with respect to the spatial evolution of financial services GVCs. See Figure 4 for details.

3.3. Endogeneity Tests

To alleviate potential endogeneity concerns, we estimate a spatial simultaneous equations model [69] under alternative spatial orders K = 0, 1, 3, and 4. The results are reported in

Table 5. Endogeneity test results.

Variable	Zero-Order		First-Order		Third-Order		Fourth-Order
	Model 1	Model 2	Model 1	Model 2	Model 1	Model 2	Model 1	Model 2
	logdeg	logdt	logdeg	logdt	logdeg	logdt	logdeg	logdt
logdeg		0.172 *** (0.007)		0.178 *** (0.007)		0.171 *** (0.007)		0.172 *** (0.007)
W * logdeg	1.977 *** (0.312)	−0.710 *** (0.075)	2.318 *** (0.258)	−0.848 *** (0.082)	1.729 *** (0.226)	−0.757 *** (0.076)	1.775 *** (0.218)	−0.733 *** (0.075)
logdt	1.775 *** (0.675)		2.459 *** (0.298)		1.657 *** (0.246)		1.883 *** (0.221)
W * logdt	−4.778 *** (0.772)	1.746 *** (0.183)	−5.600 *** (0.636)	2.084 *** (0.201)	−4.119 *** (0.558)	1.856 *** (0.187)	−4.238 *** (0.539)	1.798 *** (0.184)
control variables	control	control	control	control	control	control	control	control
cons	−2.152 *** (0.501)	0.855 *** (0.075)	−2.468 *** (0.255)	0.878 *** (0.070)	−1.834 *** (0.228)	0.866 *** (0.076)	−2.000 *** (0.214)	0.874 *** (0.075)
N	468	468	468	468	468	468	468	468
R2	0.8652	0.6707	0.8597	0.6656	0.8865	0.6752	0.8815	0.6746

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

The estimates remain qualitatively unchanged relative to the baseline—in both sign and significance—suggesting that endogeneity is unlikely to drive our findings.

Following Jin and Sun [69], we assess potential endogeneity with a two-step comparison in a single workflow: we first estimate a benchmark specification with zero spatial order (k = 0) under the maintained assumption of no endogeneity, and then re-estimate the model at higher spatial orders (k = 1, 3, 4) within the spatial simultaneous equations framework, comparing the zero-order estimates with those from the higher-order specifications. If the signs and statistical significance are consistent across orders, endogeneity is unlikely to be driving the results, whereas material discrepancies would call the benchmark into question. In our application, the zero-order estimates are consistent with those from k = 1, 3, 4, which alleviates concerns about endogeneity.

To mitigate potential endogeneity bias, we also estimate difference GMM and system GMM specifications (

Table 6. GMM estimation results.

Variables	SYSGMM	DIFFGMM
L.logdeg	0.721 *** (15.48)	0.287 *** (4.81)
logdt	−0.267 ** (−2.01)	−0.173 * (−1.83)
peo	−0.418 ** (−2.35)	−0.128 (−0.73)
logm	0.099 *** (3.35)	0.144 *** (2.82)
urban	0.376 *** (3.92)	0.286 * (1.76)
fd	0.007 (0.11)	0.014 (0.35)
logfdi	0.023 (1.02)	0.058 *** (3.44)
dof	0.019 (1.07)	−0.025 * (−1.95)
sargan	31.205	26.544
sarganp	0.182	0.543
arm1	−4.257	−3.510
arm2	−0.397	−0.526
ar1p	0.000	0.000
ar2p	0.691	0.599

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

). The diagnostics indicate that the Arellano–Bond tests for serial correlation are satisfied, and the Sargan overidentification test is not rejected, suggesting that the instrument set is valid. The signs and statistical significance of the key coefficients are consistent with those from the dynamic spatial Durbin model (Table 2).

3.4. Robustness Tests

As a further robustness check, we re-estimate non-spatial fixed-effects panel models—alternately imposing unit fixed effects and time fixed effects. The signs and significance of the key coefficients remain broadly unchanged. Applying Moran’s I and Pesaran’s CD tests to the residuals of these non-spatial models reveals persistent, statistically significant cross-sectional/spatial dependence, which further justifies the use of spatial econometric specifications. In addition, to purge possible confounding from the post-2019 COVID-19 shock, we restrict the sample to 2013–2018; the estimates are reported in

Table 7. Results of robustness checks.

	Unit Fixed Effects	Time Fixed Effects	Total Effect (2013–2018)
logdt	0.722 *** (0.198)	0.256 * (0.140)	0.658 * (0.368)
peo	1.013 *** (0.345)	−0.251 (0.249)	2.124 *** (0.731)
logm	0.621 *** (0.0617)	0.280 *** (0.0573)	0.910 *** (0.0916)
urban	1.235 *** (0.308)	0.117 (0.224)	0.667 (0.778)
fd	−0.113 (0.102)	0.0130 (0.0699)	−0.0196 (0.258)
logfdi	0.327 *** (0.0432)	0.144 *** (0.0319)	0.0157 (0.113)
dof	−0.138 *** (0.0243)	−0.0109 (0.0210)	−0.207 *** (0.0491)

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

. The conclusions from both robustness exercises are consistent with the results in Table 3.

3.5. Heterogeneity Tests

Following the World Bank income classification, we divide the 52 economies into a high-income group and a non-high-income group (including upper-middle-, lower-middle-, and low-income economies). As reported in

Table 8. Heterogeneity test results by income group.

Variable	Effect	High-Income Economies	Non–High-Income Economies
	Short-term direct	0.503 * (0.295)	−0.158 (0.273)
	Short-term indirect	3.143 *** (0.888)	−0.0519 (0.479)
logdt	Short-term total effect	3.646 *** (1.066)	−0.210 (0.616)
	Long-term direct	−0.682 (2.174)	−0.340 (0.607)
	Long-term indirect	−4.051 * (2.087)	−0.189 (1.344)
	Long-term total effect	−4.733 *** (1.495)	−0.529 (1.726)

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

, for high-income economies, the short-run total effect of digital technology on the average weighted degree is significantly positive and is driven mainly by spatial indirect effects. In the long run, however, the total effect turns significantly negative, with the adverse impact again primarily attributable to long-run indirect effects. For the non-high-income group, the short- and long-run coefficients of digital technology are negative but statistically insignificant.

These findings suggest that digitalization reshapes cross-border financial linkages in markedly different ways across income groups. For high-income economies, digital technology generates an “efficiency dividend” in the early stage: process digitalization, multi-channel distribution, and platform-based transactions substantially reduce search and coordination costs, enabling banks and non-bank financial institutions to serve a broader range of overseas clients. This strengthens cross-border linkages in the short run and raises the average weighted degree. As digitalization deepens, a “substitution and reallocation” effect increasingly dominates. Advanced automation and cloud architectures facilitate the outsourcing of routine processing and back-office functions to lower-cost locations, while higher value-added activities become further concentrated in a few hubs. At the same time, tighter regulation in data governance, KYC/AML, and cybersecurity raises the fixed costs of maintaining extensive cross-border business relationships. Taken together, these forces compress both the number and intensity of cross-border linkages within the financial services GVCs, consistent with the significantly negative long-run total effect of logdt on the average weighted degree. By contrast, digitalization in non-high-income economies is currently more oriented toward domestic financial inclusion, such as mobile payments, digital credit, and government-to-person transfers. Structural constraints in digital infrastructure (broadband, data centers, and cloud resources), institutional quality (contract enforcement and regulatory capacity), and digital maturity (adoption of advanced analytics, cloud, and platform technologies) limit their ability to embed in cross-border digital financial networks. Institutional gaps in KYC/AML, data protection, and cybersecurity also heighten perceived risk, leaving many cross-border transactions dependent on a small number of traditional correspondent-banking channels. Consequently, digital investment is more clearly reflected in improved domestic access to and use of financial services, but has not yet materially altered these economies’ positions in the financial services GVCs—consistent with the smaller and statistically insignificant coefficients for the non-high-income group. From the perspectives of financial development and inclusiveness, these results imply that high-income economies can reap substantial short-run gains in cross-border connectivity. Yet with further digital deepening, automation and offshore (or cross-border) back-office reconfiguration may channel some linkages and value capture toward a limited set of hubs. Meanwhile, for non-high-income economies to translate domestic digital-finance expansion into more inclusive participation in the financial services GVCs, parallel advances in digital infrastructure and institutional capacity remain essential to meet the basic preconditions for integration into cross-border digital financial networks.

Disaggregating digital technology for the heterogeneity analysis (

Table 9. Heterogeneity test results by digital-technology components.

	R&D Output	Human Capital	Innovation Capacity
logyfch	0.690 *** (0.182)
logrlzb		1.094 *** (0.338)
logcxsp			0.177 (0.173)
peo	−0.299 (0.477)	−0.264 (0.468)	0.0311 (0.724)
logm	0.460 *** (0.0800)	0.600 *** (0.0690)	0.624 *** (0.162)
urban	1.015 *** (0.192)	1.028 *** (0.204)	−1.165 * (0.707)
fd	−0.305 ** (0.143)	−0.341 ** (0.156)	−0.0394 (0.239)
logfdi	0.525 *** (0.0654)	0.556 *** (0.0757)	0.392 *** (0.0876)
dof	−0.358 *** (0.0738)	−0.449 *** (0.0775)	−0.0787 (0.0671)

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

) shows that the total effects of all three dimensions—R&D output, human capital, and innovation level—are positive for the spatial evolution of financial services GVCs, although the coefficient on innovation level is not statistically significant. A plausible explanation is that innovation level captures potential rather than realized digital outputs, leading to measurement error and timing lags. Moreover, innovation tends to operate indirectly—by improving R&D efficiency, facilitating talent attraction and matching, and strengthening institutions and collaboration networks—so its impact is partly absorbed through mediation and spatial spillover channels, attenuating its statistical significance.

4. Mechanisms Testing

To test Hypothesis 1 (distance-decay effect), we adopt the population-weighted bilateral distance indicator (dishar) proposed by Conte et al. [70]. A larger dishar indicates a greater effective distance between two countries and thus a stronger conventional distance-decay constraint. To test Hypothesis 2 (spatial proximity effect), we use the share of internet users in total population (inter) as a proxy for a country’s digital access capacity and “digital proximity”: higher inter implies a stronger ability to form more frequent interactions and tighter network linkages with neighboring and economically similar countries via digital platforms. To test Hypothesis 3 (spatial heterogeneity effect), we construct the interaction term dishar × inter to examine whether the strength of distance decay varies systematically with the level of digital proximity—that is, whether the extent to which digital technology weakens distance constraints differs across spatial contexts.

The econometric results show that (

Table 10. Intermediate mechanism test results.

Variable	Distance-Decay Effect			Spatial Proximity Effect			Spatial Heterogeneity Effect
	Direct Effects	Indirect Effects	Long-Term Total Effect	Direct Effects	Indirect Effects	Long-Term Total Effect	Direct Effects	Indirect Effects	Long-Term Total Effect
R	6.385 *** (1.982)	−19.40 ** (8.187)	−13.01 (8.502)	0.230 *** (0.0576)	0.0502 (0.374)	0.280 (0.383)	0.0494 *** (0.0145)	0.0501 (0.0980)	0.0995 (0.100)
logdt	0.531 *** (0.176)	0.319 (0.990)	0.850 (1.034)	0.434 ** (0.179)	−0.581 (0.867)	−0.147 (0.914)	0.457 ** (0.180)	−0.669 (0.854)	−0.211 (0.901)
peo	0.712 ** (0.282)	3.548 * (2.092)	4.260 ** (2.171)	0.524 * (0.290)	4.930 ** (2.298)	5.454 ** (2.377)	0.559 * (0.293)	4.466 * (2.478)	5.026 ** (2.558)
logm	0.558 *** (0.0552)	−0.0638 (0.0903)	0.494 *** (0.101)	0.482 *** (0.0584)	0.0228 (0.102)	0.505 *** (0.116)	0.491 *** (0.0590)	0.00235 (0.103)	0.493 *** (0.118)
urban	1.332 *** (0.240)	0.297 (1.415)	1.628 (1.475)	0.948 *** (0.257)	−2.542 ** (1.120)	−1.594 (1.194)	0.998 *** (0.259)	−2.719 ** (1.104)	−1.721 (1.179)
fd	−0.0536 (0.0844)	−0.00628 (0.274)	−0.0599 (0.294)	−0.133 (0.0858)	−0.163 (0.289)	−0.296 (0.312)	−0.128 (0.0863)	−0.146 (0.291)	−0.274 (0.314)
logfdi	0.248 *** (0.0489)	−0.0521 (0.112)	0.196 (0.124)	0.247 *** (0.0480)	−0.00322 (0.119)	0.244 * (0.133)	0.247 *** (0.0482)	−0.00142 (0.120)	0.246 * (0.134)
dof	−0.135 *** (0.0212)	−0.0698 (0.0604)	−0.205 *** (0.0640)	−0.140 *** (0.0211)	0.0176 (0.0552)	−0.122 ** (0.0612)	−0.141 *** (0.0212)	0.0287 (0.0575)	−0.112 * (0.0637)

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

) the long-run direct effect of digital technology is positive and significant at the 5% level, whereas the long-run indirect effect is not statistically significant. This pattern suggests that digitalization promotes the spatial evolution of financial services GVCs primarily by attenuating domestic distance decay and strengthening local spatial proximity linkages. For the distance-decay channel specifically, the long-run indirect effect is significantly negative while the long-run direct effect is positive, indicating that cross-border transmission remains constrained by geographic frictions, although domestic digital and network connectivity can partially offset these frictions. Taken together, the evidence corroborates that, as digitalization advances, distance-decay effects weaken while spatial proximity and spatial heterogeneity effects intensify, with more pronounced marginal impacts on the home economy. From a policy perspective, regulatory oversight should focus on cross-border interfaces: improve cross-border information sharing and coordinated enforcement; strengthen dynamic monitoring of cross-border capital and data flows; and guard against cross-border financial risks to protect global financial sustainability. Finally, comparing the estimates from Equation (5) with those from Equation (3) shows smaller coefficient magnitudes in the former, consistent with partial mediation—that is, distance decay, spatial proximity, and spatial heterogeneity operate as mediating channels through which digitalization shapes the spatial evolution of financial services GVCs [56,71].

In the spatial mediation model, all control variables exhibit positive direct effects, except that financial development is not statistically significant and the institutional environment shows a negative direct effect. A plausible interpretation is that the financial sector is jointly shaped by localized and virtualized service delivery; in such a setting, greater institutional freedom facilitates localized learning and interaction in financial services. Consistent with this, labor structure, manufacturing structure, urbanization, and FDI all show positive direct effects. In the digital era, the agglomeration of service-oriented human capital strengthens the regional diffusion capacity of financial functions; the servitization of manufacturing heightens neighboring industry-coordination needs; urbanization-driven connectivity in information and infrastructure promotes service spillovers; and FDI extends financial service linkages through the regional mobility of capital and technology. Taken together, these forces steer the spatial evolution of financial services GVCs toward greater long-distance reach, stronger local–neighbor coupling, and advantage complementarity.

5. Conclusions and Policy Implications

Grounded in global value chain and spatial evolution theories, this study develops a cross-country spatial panel framework covering 52 economies over 2013–2021. By integrating a dynamic spatial Durbin model, geographically and temporally weighted regression (GTWR), and a spatial mediation model, we systematically assess the impact of digital technology on the spatial evolution of financial services GVCs at both the aggregate and component levels. We further characterize spatially uneven effects across countries and development stages. We also unpack three key channels through which digital technology operates: the attenuation of distance-decay effects, the strengthening of spatial proximity effects, and the amplification of spatial heterogeneity effects.

Building on these analyses, we draw three core conclusions. First, digital technology is a central driver of the spatial evolution of financial services GVCs, operating through a dynamic process characterized by spatial diffusion and featuring a pattern of “strong short-run stimulus–long-run redistribution.” Across geographic, economic, and geo-economic nested weight matrices, the short-run direct effects of digital technology on a country’s weighted average degree in the financial services GVCs network are mostly negative or insignificant, and the long-run direct effects become significantly negative under the economic and nested matrices. By contrast, both the short-run and long-run indirect effects transmitted through neighboring and economically similar economies remain significantly positive and dominate in magnitude. As a result, the total effects are positive in both the short and long run, with the short-run total effect substantially larger than the long-run total effect. Second, the impact of digital technology exhibits pronounced “dual heterogeneity,” reflecting differences across both internal components and development stages. Disaggregated evidence shows that the effect does not diffuse uniformly across digital-technology dimensions; instead, it displays a “dual-engine” pattern centered on human capital and R&D outputs. When these two components improve in tandem, the promoting effect of digital technology on the spatial upgrading of financial services GVCs is significantly amplified. In contrast, increases in innovation level alone yield limited marginal gains if technological conversion and application capacity remain weak. Across development stages, high-income economies benefit significantly in the short run from digital efficiency gains, while also facing longer-term structural challenges associated with the outward dispersion of high-value-added functions and the potential relocation of certain key segments. Non-high-income economies, constrained by bottlenecks in digital infrastructure, institutional environments, and human capital, have yet to fully realize the aggregate benefits of digital technology, suggesting a clear “below-threshold” pattern. Third, digital technology reshapes the geography of financial services GVCs primarily by transforming spatial interaction mechanisms. It weakens the traditional distance-decay effect rooted in physical distance, allowing cross-border financial service linkages to become less dependent on strict geographic proximity. At the same time, by improving cross-border data flows and the connectivity of digital infrastructure, it significantly strengthens spatial proximity and spatial heterogeneity effects, tightening intra-regional financial service networks while further accentuating cross-country and cross-regional disparities in digital adoption and network embeddedness. Consequently, the spatial configuration of financial services GVCs increasingly reflects the coexistence of “global specialization and regional agglomeration”: a more polycentric and networked pattern of cross-border linkages at the global level, alongside stronger local focus and more complex regional interdependencies at the regional level.

Based on these findings, we derive three policy implications. First, given the spatially diffusive dynamic of digital technology—characterized by “strong short-run stimulus–long-run redistribution”—policymakers should build a forward-looking regulatory framework that balances domestic resilience with cross-border spillovers. On the one hand, the spatial dimension of digital-finance expansion should be explicitly incorporated into macroprudential and countercyclical toolkits. This includes strengthened, end-to-end oversight of platform-based financial institutions and critical third-party service providers to prevent the amplification of systemic risks through cross-border business. On the other hand, complementary measures in financial inclusion, employment, and consumer protection are needed to provide targeted support for groups and regions that may be crowded out by the contraction or reconfiguration of traditional financial service capacity. Such policies can help translate short-term digital efficiency gains into sustainable financial service provision and greater network resilience.

Second, in light of the “dual-engine” mechanism within digital technology—driven by the complementarity between human capital and R&D outputs—and the heterogeneity across development stages, countries should adopt differentiated national strategies that jointly strengthen foundational capabilities and innovation-to-application conversion. For high-income economies, the priority is to leverage their existing advantages in digital infrastructure and talent. Policy instruments such as tax incentives, regulatory sandboxes, and robust intellectual property protection can support integrated programs that combine digital-skills development with fintech R&D, accelerate the translation of research into deployable digital-finance products and cross-border services, and mitigate the risk of excessive long-run relocation of high value-added functions. For non-high-income economies, policy should focus first on building the necessary “threshold conditions.” This entails accelerating investment in digital infrastructure and foundational platforms for payments and digital identity, improving the business environment and the rule of law, and strengthening data governance and cybersecurity capacity. At the same time, attracting high-quality FDI and deepening international cooperation can facilitate technology spillovers and capability co-building tailored to the financial services sector. With these foundations in place, governments can progressively expand support for digital-skills training and domestic fintech innovation, enabling a sustainable transition from peripheral integration toward deeper participation in financial services GVCs.

Third, because digital technology simultaneously attenuates distance-decay effects and strengthens spatial proximity and spatial heterogeneity effects, regulatory coordination and rule alignment at regional and international levels should be advanced. Specifically, greater harmonization of technical standards and compliance requirements for cross-border payments and settlement, digital identity, and cross-border data flows would lower institutional barriers for emerging markets and smaller economies to connect to global financial services networks. Shared compliance public infrastructure—such as regional KYC/AML information-sharing platforms and joint supervisory databases—can also improve the efficiency of regulatory resources. Finally, mechanisms for mutual recognition and equivalence assessments can enhance the transparency and stability of cross-border rules. This would allow emerging markets to align with global standards while preserving appropriate policy space, enabling a path of “predictable openness” toward more inclusive and resilient integration into financial services GVCs.

This study systematically identifies the pathways and mechanisms through which digital technologies shape the spatial evolution of financial services GVCs. Nevertheless, we note three limitations, each suggesting directions for future research. First, data: our analysis relies on global value-added trade databases with delayed public releases, so the sample ends in 2021. As updated releases become available, future work can reassess the robustness of our findings and, where feasible, draw on higher-frequency sources to capture finer-grained dynamics. Second, sectoral scope: treating financial services as an aggregate masks within-sector heterogeneity; subsequent research could extend the analysis to subsectors such as banking, insurance, and securities to compare how digital technologies differentially shape the spatial configuration of financial services GVCs. Third, level of analysis: our analysis is conducted at the national (macro) level, whereas cross-border financial services are largely intermediated by multinational financial institutions. With access to firm-level cross-border transaction data, future work could integrate heterogeneous-firm frameworks with network-analytic methods to trace the micro-mechanisms by which digital technologies, operating through firm decision-making, reconfigure the spatial structure of financial services GVCs.