Digital Transformation and Supply Chain Resilience in Resource-Constrained Regions: Evidence from Central and Western China

Abstract

In recent years, global supply chains have become increasingly vulnerable to geopolitical tensions, pandemics, and energy crises, particularly in resource-constrained regions characterized by weak infrastructure and high transaction costs. Using panel data on A-share listed firms in China’s central and western regions from 2010 to 2022, this study examines the effect of firm-level digital transformation on supply chain resilience. We construct a digital transformation index and employ an instrumental-variable approach based on the interaction between terrain ruggedness and lagged digital transformation to address endogeneity concerns. Empirical results show that the digital transformation of enterprises has significantly enhanced the resistance and recovery capabilities of the supply chain, verifying its effectiveness in resource-constrained environments. Mechanism analyses reveal that this effect operates through increased supply chain diversification—especially customer diversification—and improved supply–demand matching enabled by more accurate demand forecasting and inventory management. Heterogeneity tests indicate that the resilience-enhancing effects are more pronounced among non-state-owned firms, manufacturing enterprises, and firms in less technology-intensive industries. Overall, our findings provide empirical support for transaction cost economics, dynamic capability theory, and the resource-based view, highlighting the strategic role of digital investment in strengthening supply chain resilience in infrastructure-constrained settings and contributing to the aims of Sustainable Development Goal 9.

1. Introduction

Global supply chains have been repeatedly buffeted by geopolitical disputes, pandemics, and energy crises, revealing that they are vulnerable to prolonged, severe shocks. The experience has highlighted ‘supply chain resilience’, or the capacity of a firm’s supply chain to rapidly restore critical operations after a shock. Numerous studies have examined what makes supply chains resilient; however, most have focused on firms operating in environments with relatively good infrastructure, diverse supplier networks, and efficient logistics systems. It is not clear, then, whether commonly advocated strategies to enhance supply chain resilience in resource-constrained regions will remain effective [1].

Digital transformation has increasingly become a prominent strategy-oriented response to uncertainty. By enabling real-time information sharing, improving forecasting accuracy, and facilitating cross-organizational coordination, digital technologies are widely seen as enhancing firms’ adaptive capacity and operational flexibility. Existing studies show that digital transformation improves firm performance, innovation outcomes, and, more recently, supply chain resilience [2,3]. However, the empirical evidence is predominantly derived from developed economies or regions characterized by more favorable institutional frameworks and infrastructure conditions. In contrast, regions characterized by limited infrastructure, high transaction costs, and concentrated supply networks have received far less systematic attention.

In resource-constrained regions, it is theoretically unclear whether digital transformation is effective. On one hand, inadequate transport and communication infrastructure, fragmented markets, and weak logistics systems may limit returns on digital investment. On the other hand, precisely because such regions suffer from severe information frictions and coordination failures, digital technologies may generate disproportionately large benefits by lowering transaction costs, expanding network reach, and improving supply–demand matching. Whether digital transformation can meaningfully enhance supply chain resilience under such constraints, therefore, remains an open empirical question.

The central and western regions offer a conducive context for examining the impact of digital transformation on supply chain resilience under such conditions. The road density, railway double-track rate, and port throughput in these regions are all far lower than the average level in the eastern region. The distribution of supplier parks is sparse. Freight prices can fluctuate significantly due to weather and road closures. These unfavorable factors are structural rather than cyclical, yet these regions have carried out the majority of the country’s industrial transfer projects. Therefore, by studying enterprises headquartered there, one can strip away conventional buffer factors such as redundant routes, deep supplier resources, and precautionary inventories, thereby isolating the marginal effects of digital adoption [4]. Similar constraints-such as weak backbone infrastructure, high logistics costs, and fragmented supplier networks-are also observed in other infrastructure-constrained regions. While this similarity suggests that the analysis may offer broader analytical insights, it does not imply direct transferability of the findings beyond the studied context.

Against this backdrop, we raise three research questions. (1) When both physical infrastructure and information infrastructure are severely restricted, can digital transformation at the enterprise level strengthen supply chain resilience? (2) Through what mechanisms might this influence enhance supply chain resilience? (3) Will these impacts systematically change due to differences in ownership forms, industry classifications, and technological intensities?

We constructed our sample by selecting A-share listed firms headquartered in China’s central and western regions over the period 2010–2022. The comprehensive supply chain resilience index simultaneously captures both resistance capacity and recovery capability. Digital transformation is quantified based on how often technology-related terms appear in firms’ annual reports. To address potential endogeneity arising from simultaneous causality, we employ the interaction between terrain ruggedness and the lagged digital transformation index employed as an instrument for the current digital transformation level. In addition, we conduct mechanism analyses and heterogeneity analyses for further investigation.

This research makes contributions to the literature in three main ways. First, it provides corporate-level empirical evidence regarding how digital transformation relates to supply chain resilience in regions with limited resources, where the returns to digital strategies are theoretically ambiguous. Second, the analysis offers evidence consistent with supply chain diversification and supply–demand matching as important mechanisms linking digital transformation to resilience. Third, by examining heterogeneity across ownership structures, industries, and technological intensities, this study elucidates the circumstances in which digital transformation is more strongly associated with resilience outcomes.

2. Literature Review, Theoretical Analysis, and Research Hypotheses

2.1. Literature Review

2.1.1. Digital Transformation and Enterprise Performance

As the digital economy accelerates, the deployment of digital technologies has become a standard instrument for firms seeking innovation-led growth and durable competitive advantage. The convergence of artificial intelligence, blockchain, cloud computing, and big-data analytics reconfigures managerial routines and value-creation logics, while simultaneously enhancing organisational agility under uncertainty [5,6]. Empirical studies indicate that digital transformation improves capital-market performance by attenuating information asymmetry, stimulating innovation output, and buttressing financial stability [7,8]; it also raises capital productivity and investment efficiency [9]. Cost reductions and productivity gains are documented across industries, with pronounced effects among frontier firms [10]. A spillover literature further shows that the performance benefits of digital transformation propagate upstream and downstream, elevating the efficiency of supply chain partners [11,12]. Emerging evidence further suggests that digital infrastructure and networked computation can function as system-level coordination mechanisms, enabling firms to reallocate resources, smooth inter-regional frictions, and buffer localized disruptions even under infrastructural constraints [13]. Despite these studies, the primary focus has been on the general context, with no dedicated research conducted specifically on resource-constrained regions.

2.1.2. Supply Chain Resilience and Its Antecedents

Supply chain resilience is conventionally defined as the capacity of a supply chain system to preserve core functions, recover rapidly, and improve through adaptive reconfiguration after disruption [14]. Its architectural dimensions comprise resistance capacity and recovery capacity [15]. Three drivers dominate the extant literature: diversification, collaborative cooperation, and digital integration. Wu and Yao demonstrate that digital transformation fosters supply chain flexibility by encouraging dispersed rather than concentrated resource allocation [16]. Hong and Wang emphasise the safeguarding role of trust-based value-co-creation networks [17]. In a cross-country setting, Alquraish establishes that digital trust and openness facilitate cooperative behaviour in circular supply chains [18]. Collectively, these studies elevate supply chain resilience to the status of a strategic asset that sustains competitive advantage under exogenous volatility.

2.1.3. Digital Transformation and Supply Chain Resilience

Over the past few years, scholarly interest has progressively tied the spread of digital technologies to the robustness of supply chains. Prior research indicates that digital transformation substantially improves firm resilience when supported by favorable institutional and managerial conditions [19] and improves supply chain stability by increasing transparency and balancing power dynamics [20]. However, most of this research focuses on developed economies or globally integrated supply chains, with limited attention given to resource-constrained regions [21,22]. In fact, in such contexts, inadequate infrastructure, high transaction costs, and excessive industrial concentration may hinder the effectiveness of digital transformation. Therefore, there is a pressing need for systematic research on whether and how digital transformation can contribute to supply chain resilience in these environments. This study focuses on the central and western regions of China underexplored yet representative context that not only offers unique insights but also offers analytical insights for regions facing similar infrastructural constraints, without suggesting direct policy transferability.

2.2. Theoretical Analysis and Research Hypotheses

Firstly, transaction cost economics regards information asymmetry and contract execution costs as the main frictional factors affecting organizational efficiency. Digital transformation reduces the accumulated search, negotiation, and supervision costs in inter-enterprise connections through real-time data exchange, smart contracts, and blockchain-based traceability. This cost reduction is most significant in underdeveloped regions where infrastructure is insufficient and law enforcement is weak, resulting in high transaction costs. By narrowing the information gap, digital transformation enhances collaboration efficiency and strengthens the stability of the supply chain [23,24].

Secondly, the dynamic capability theory holds that to achieve competitive survival under uncertain conditions, one needs to possess the capacity to perceive changes in the environment, the ability to capitalize on new opportunities, and flexibly reallocate resources. Digital technology has expanded management cognition: The Internet of Things and cloud-based dashboards can detect fluctuations in demand or supply earlier, while big data algorithms and digital platforms can accelerate the reallocation of resources after a new resource allocation plan is selected. Therefore, digital transformation not only reduces static transaction costs but also lowers dynamic costs resulting from slow adaptation.

Thirdly, the resource-based perspective holds that sustainable advantages stem from those rare, hard-to-imitate, and irreplaceable assets. Digital capabilities meet the conditions of strategic resources: they reconfigure the knowledge reserves and coordination practices within the supply chain. As an illustration, the TADHGCN model—a temporal attention hypergraph convolutional network—enhances the accuracy and interpretability of gasoline price prediction by capturing the high-order interactions among explanatory variables. In resource-constrained environments, such predictive assets, to a certain extent, make up for the weak physical infrastructure and serve as a lever to enhance environmental adaptability [25].

Based on these viewpoints, we have identified two mechanisms by which digital transformation is expected to strengthen supply chain resilience:

(1)

Supply chain diversification: Digital networks have expanded the reach of suppliers and customers, reducing reliance on any single node and thereby alleviating the risk of concentration.

(2)

Supply–Demand Matching: Advanced analysis enhances the accuracy of predictions and the timeliness of coordination, thereby curbing inventory fluctuations and operational uncertainties.

Therefore, we put forward the following hypotheses:

H1 (Overall Effect):

Enterprise digital transformation exerts a notable, favorable effect on supply chain resilience, even in underdeveloped areas characterized by scarce resources and information.

H2 (Diversification Mechanism):

By expanding the customer base, digital transformation has reduced buyer concentration, therefore improving supply chain resilience.

H3 (Supply–Demand Matching Mechanism):

In an environment with large demand fluctuations and fragile logistics, enterprise digital transformation by increasing the efficiency of matching supply with demand.

Figure 1 illustrates the theoretical mechanisms and hypothesized relationships proposed in this study, through which digital transformation (DIG) facilitates supply chain resilience (Resil). Two core pathways are identified:

Figure 1. Mechanism and Hypothesized Relationship Framework of Digital Transformation on Supply Chain Resilience.

Supply Chain Diversification Mechanism. Digital transformation expands firms’ customer and supplier networks, reducing dependence on single nodes. This dispersion of supply chain concentration risk directly strengthens resilience, corresponding to Hypothesis H2.

Supply–Demand Matching Optimization Mechanism. Digital transformation improves demand forecasting accuracy, optimizes inventory management, and enhances resource allocation efficiency. These effects boost supply chain flexibility and responsiveness in high-volatility environments, corresponding to Hypothesis H3.

Arrows in the figure denote hypothesized positive relationships: DIG influences Resil both directly (via the overall effect, Hypothesis H1) and indirectly through the two mechanisms above. Integrating perspectives from transaction cost economics, the dynamic capabilities framework, and the resource-based perspective, this structure explains how digital transformation mitigates information frictions, enhances adaptive capacity, and unlocks strategic resource value in resource-constrained contexts—ultimately driving improvements in supply chain resilience.

3. Research Design

3.1. Sample Selection and Data Sources

The present study uses A-share listed firms located in China’s central and western regions over the period 2010–2022 as the research sample, with corporate-level financial and operational data obtained primarily from the CSMAR database provided by Guotai An and the Juchao Information Network. To ensure the reliability and robustness of the empirical analysis, several data construction and preprocessing procedures are implemented. (1) Firms are classified according to their registered locations following the regional definition of the National Bureau of Statistics, with the central region covering Shanxi, Anhui, Jiangxi, Henan, Hubei, and Hunan, and the western region comprising Inner Mongolia, Guangxi, Chongqing, Sichuan, Guizhou, Yunnan, Xizang, Shanxi, Gansu, Qinghai, Ningxia, and Xinjiang; companies registered in these provinces are included in the sample and used in subsequent heterogeneity analyses. (2) Companies designated for special treatment (e.g., “ST”, “*ST”, or “PT”) over the sample period, as well as companies that were delisted, are excluded to mitigate potential biases arising from abnormal financial conditions and to preserve longitudinal consistency. (3) For control variables and alternative indicators with missing observations, median imputation is employed to retain sample size while limiting the impact of extreme values. (4) All variables of continuous type are winsorized at the 1% level in both tails to reduce the influence of outliers and alleviate potential heteroscedasticity. (5) The baseline regressions are conducted using the full sample defined above, whereas instrumental variable estimations (IV/2SLS) rely on a slightly reduced sample due to the unavailability of instrumental variable data for certain observations. (6) All data cleaning, variable construction, and econometric analyses are performed using Python 3.9 and Stata 18.0.

3.2. Model Specification

To examine how enterprise digital transformation affects supply chain resilience, we employ the following two-way fixed-effects panel regression model for estimation:

$${Resil}_{it}=\alpha +\beta {DIG}_{it}+\gamma X_{it}+{\mathsf{\mu }}_{\left\{\mathrm{i}\mathrm{n}\mathrm{d}\mathrm{u}\mathrm{s}\mathrm{t}\mathrm{r}\mathrm{y}\right(\mathrm{i}\left)\right\}}+{\lambda }_t+{\epsilon }_{it}$$

(1)

Among them, Resil_it denotes the supply chain resilience level of enterprise i in year t; DIG_it indicates the degree of enterprise digital transformation; X_it includes control variables concerning firm characteristics, corporate governance, and financial conditions. μ_{{industry(i)}} and λ_t denote industry and year fixed impacts, respectively, accounting for industry-specific heterogeneity that is time-invariant and for common macroeconomic shocks. ε_it represents the stochastic error term. The core parameter β captures the marginal effect of digital transformation on supply chain resilience. Based on the theoretical inference of H1, we expect β > 0, indicating a positive link connecting digital transformation to supply chain resilience. Robust standard errors, clustered by firm, are used throughout to correct for possible heteroskedasticity and within-firm serial correlation.

3.3. Key Variables

3.3.1. Explained Variable: Supply Chain Resilience (Resil)

Following Wang et al. and Liu et al. [14,19], we gauge supply chain resilience along two core dimensions: the capacity to absorb shocks and the swiftness with which operations rebound.

Resistance (Resis): From a working-capital perspective, stability erodes when downstream customers lock up upstream liquidity. Rising receivables relative to sales indicate intensified financial pressure on suppliers and a more fragile chain. We therefore adopt ln(receivables/sales) as an inverse proxy for resistance; lower values denote slimmer customer-induced capital occupation and hence a robust upstream-downstream interface [26]. Relational stability: Persistent ties facilitate tacit knowledge exchange and coordinated adaptation. We measure this by identifying the number of customers common to a firm’s current-year and prior-year top-five lists, dividing the count by five. A higher fraction signals a denser, repeat-exchange network that is better positioned to withstand and recover from external shocks [27].

Recovery capacity (Recov) is evaluated through the inventory turnover rate and the relative change sensitivity index (SI), which captures the speed at which a firm’s main business income recovers relative to the industry average (Details are provided in Supplementary Materials S1–S3).

To synthesize these dimensions into a single measure, an entropy weighting method is applied to integrate the aforementioned indicators, resulting in a comprehensive supply chain resilience index (Table 1).

Table 1. Supply Chain Resilience Measurement Index.

Primary Dimension	Secondary Dimension	Specific Indicators	Directionality (Positive/Negative)
Resis	Capital Relationship	Natural Logarithm of Accounts Receivable-to-Revenue Ratio	Negative
	Supply–Demand Stability	Count of Non-New Customers Among Top 5 Clients Divided by 5	Positive
Recov	Product Flow Dynamics	Inventory Turnover Ratio	Positive
	Performance Volatility	Economic Performance Sensitivity Index	Positive

3.3.2. Core Explanatory Variable: Digital Transformation (DIG)

Following Liu et al. [19], Qi et al. [28], and a considerable amount of recent studies on digital transformation, we quantify firms’ digital transformation by extracting the occurrence rate of digital-related keywords covering AI, big data analytics, and cloud-based computing, blockchain technologies, and digital innovations applications from their yearly reports. This text-based approach has become a standard proxy in empirical accounting, finance, and management studies, as annual reports are mandatory, audited, and relatively standardized disclosures that reflect firms’ strategic priorities and technology adoption. Prior studies have also shown that digital-related textual disclosures are closely associated with actual IT investment, system upgrading, and digital process reengineering.

To capture the gradual, path-dependent nature of digital transformation, we construct a cumulative digital transformation index. Digital upgrading is inherently a multi-year, irreversible process, and mature digital adopters may mention these terms less frequently in later years despite maintaining a high digital capability. The cumulative measure, therefore, better reflects firms’ long-term digital trajectory and reduces noise induced by short-term disclosure fluctuations. This approach follows Sun et al. [29], who argue that cumulative indicators provide a more stable representation of digital development.

To further ensure content validity, we examine the association between the digital transformation index and firms’ digital-related invention patents (IPC G06, H04, etc.). The two measures are both positively and markedly correlated at the 1% level, confirming that the textual indicator captures substantive digital investment and innovation efforts rather than mere rhetorical disclosure.

3.3.3. Instrumental Variable: Terrain Ruggedness (Ruggedness × DIG_t−1)

To mitigate potential endogeneity stemming from reverse causation and unobserved confounding in firms’ digital transformation decisions, this research utilizes an instrumental variable methodology based on the interaction between terrain ruggedness and lagged digital transformation. Rugged terrain captures exogenous geographic variation that raises infrastructure costs and constrains the deployment and effectiveness of digital technologies, whereas the lagged digital transformation component reflects firms’ pre-existing digital capabilities. The interaction between these two elements generates plausibly exogenous variation in current digital adoption, thereby satisfying the relevance condition. Given that terrain ruggedness is predetermined and time-invariant, it is expected not to be associated with contemporaneous firm-level shocks, supporting the exclusion restriction. Moreover, conditional on industry and year fixed effects as well as a comprehensive collection of firm-level controls, the instrument is justified to influence supply chain resilience only through its effect on current digital transformation, fulfilling the required assumptions for valid instrumentation.

Compared with conventional instruments based on policy shocks or pilot programs, this geography-based interaction instrument offers a complementary and more structural source of exogenous variation. It allows us to identify the causal influence of digital transformation on supply chain resilience in resource-constrained regions, while mitigating concerns related to reverse causality and omitted-variable bias.

3.3.4. Control Variables (Xit)

To alleviate the omitted variable bias, this paper controls for firm characteristics, corporate governance, and financial conditions.

Enterprise characteristics: scale (logarithm of the number of employees), asset-liability ratio, loss dummy variable (takes 1 when net profit < 0), operating cash flow ratio (operating cash flow/current liabilities).

Corporate governance: Audit by the Big Four accounting firms (Big4), board size (log of total board members), concentration of equity (the shareholding ratio of the largest shareholder), and a dummy variable for the concurrent position of chairman and general manager (Dual).

Fixed effects: Fixed effects for year and sector are applied to account for macro shocks and sector-specific fluctuations.

Table 2 below includes variable definitions and sources.

Table 2. Variable Definitions and Sources.

	Variable	Definition	Source
Dependent Variable	Resil	Composite index of supply chain resilience, constructed from resistance (ability to withstand disruptions) and recovery (ability to restore operations) dimensions following Wang et al. (2019) [14] and Liu et al. (2023) [19]	CSMAR/Author’s calculation
Core Explanatory Variable	DIG	Cumulative index of enterprise digital transformation, constructed through keyword frequency in annual reports, capturing firms’ digital strategy, technology adoption, and business process reengineering	CNRDS/Author’s calculation
Instrumental Variable	Ruggedness × DIGt−1	Herein, Ruggedness denotes the topographic ruggedness index of the region, while DIGt−1 stands for the digital transformation level of the enterprise in the prior period.	Geographic data/Author’s calculation
Firm Characteristics (Controls)	Size	Natural logarithm of the number of employees	CSMAR
	Lev	Ratio of total liabilities to total assets	CSMAR
	Loss	Dummy variable = 1 if net profit < 0, else 0	CSMAR
	Cashflow	Net cash flow from operating activities/current liabilities	CSMAR
	TobinQ	It is calculated as (firm’s total market capitalization + book value of liabilities)/book value of total assets, and serves to measure a firm’s investment opportunities and market value.	CSMAR
Corporate Governance (Controls)	Big4	Dummy = 1 if audited by PwC, Deloitte, KPMG, or EY; else 0	CSMAR
	Board	Natural logarithm of (number of board members + 1)	CSMAR
	Top1	Shareholding ratio of the largest shareholder	CSMAR
	Dual	Dummy = 1 if the chairman and CEO are the same person; else 0	CSMAR
	Fixed Effects	Year and industry fixed effects to control for macroeconomic and sector-specific shocks	Author’s setting

4. Empirical Results Analysis

4.1. Descriptive Statistics

Table 3 presents the descriptive statistics of the main variables. Supply chain resilience (Resil) shows considerable variation across firms (mean = 0.239, SD = 0.25), reflecting considerable diversity among firms’ ability to respond to disruptions. The digital transformation index (DIG) also displays significant dispersion (mean = 2.023, SD = 1.75), with the maximum value (7.987) well above the mean. This suggests that while some enterprises in central and western China have made notable progress in digital transformation, most firms are still at an early adoption stage. This disparity reflects the uneven pace of digital upgrading in less-developed regions and provides a solid empirical basis for testing Hypotheses H1–H3.

Table 3. Statistical Summary of Key Variables.

Variables	Obs	Mean	Std. Dev.	Min	Max
Resil	10,269	0.239	0.25	0.003	0.992
Recov	10,269	0.425	0.122	0.015	0.996
Resis	10,269	0.231	0.258	0	0.998
DIG	10,269	2.023	1.75	0	7.987
Lev	10,269	0.44	0.208	0.032	0.908
Size	10,269	7.752	1.228	4.078	11.181
Dual	10,269	0.228	0.42	0	1
Cashflow	10,269	0.047	0.069	−0.222	0.267
Big4	10,269	0.043	0.203	0	1
TobinQ	10,269	2.021	1.431	0.802	15.607
Loss	10,269	0.13	0.336	0	1
Top1	10,269	34.13	14.889	8.02	75.843
Board	10,269	2.146	0.2	1.609	2.708

4.2. Benchmark Regression

Consistent with H1 and prior studies highlighting the positive impact of digital transformation on resilience building [30]. As reported in Table 4, the estimated coefficients suggest a favorable link between digital transformation and supply chain resilience (p < 0.01). In Column (2), the coefficient of 0.0101 (p < 0.01) offers strong evidence that digital transformation meaningfully strengthens firm resilience. Drawing on the sample’s descriptive statistics—where the mean value of DIG is 2.023 (SD = 1.75) and the mean resilience index is 0.239—we further quantify the economic significance of this relationship: a one-standard-deviation increase in DIG (1.75 units) is linked to an approximate 0.018 rise in the resilience index, equivalent to 7.4% of the average resilience level. Moreover, moving from the minimum to the maximum observed value of DIG would lead to an estimated 0.081 increase in the resilience index, accounting for 33.7% of the mean level. These findings underscore that digital transformation can substantially strengthen firms’ capacity to withstand disruptions and restore operations efficiently, even in resource-constrained regions.

Table 4. Benchmark Regression Results.

	(1)	(2)	(3)
Dep. Var.	Recov	Resil	Resis
DIG	0.0060 ***	0.0101 ***	0.0100 ***
	(0.0006)	(0.0019)	(0.0019)
Controls	YES	YES	YES
Industry/year fixed	YES	YES	YES
N	10269	10269	10269
Adj. R2	0.6464	0.1311	0.1296

Note: Standard errors in parentheses *** p < 0.01. Estimated coefficients reflect conditional associations.

Among the control variables, the debt-to-asset ratio (Lev) exhibits a positive correlation with resilience metrics. This suggests that firms with higher leverage—typically those in the expansion phase—may benefit from improved credit access and reduced financing constraints, thereby maintaining greater operational flexibility amid disruptions. Collectively, these results offer strong evidence from empirical analysis for Hypothesis 1 (H1).

4.3. Addressing Endogeneity Issues

To strengthen causal inference, we utilize a two-stage least squares (2SLS) methodology, using terrain ruggedness interacted with lagged digital transformation as an instrumental variable (IV).

$$\mathit{IV}=TerrainRuggedness\times {DIG}_{t-1}$$

(2)

Terrain ruggedness is a time-invariant natural geographical feature that captures regional disparities in digital transformation costs. The one-period lagged digital transformation variable DIG_t−1 represents the exogenous lagged value of a firm’s pre-existing digital infrastructure.

To assess the sensitivity of the 2SLS estimates to potential bias due to weak instruments, we re-estimate the specification using Limited Information Maximum Likelihood (LIML). Because LIML is median-unbiased even when the first-stage F-statistic is only marginally above conventional thresholds, its coefficient vector is typically attenuated relative to 2SLS. Supplementary Material S4 indicates that the sign, magnitude, and statistical importance of the LIML estimates coincide with those produced by 2SLS, while the Kleibergen–Paap rk LM statistic (p < 0.01) and the corresponding Wald F-statistic remain well above the Stock-Yogo 10% critical value, ensuring that the instrumental variable is not weak.

We begin by verifying the reduced-form relationship between the instrument and the endogenous regressor. Column (1) of Table 5 reports the first-stage regression of current digital transformation on the interaction term Terrain ruggedness × Lagged digital transformation. The coefficient is positively and statistically significant at the 1% level (β = 0.1592, s.e. = 0.0036), and the Kleibergen–Paap rk Wald F-statistic equals 1946.22-an order of magnitude above the Stock-Yogo 10% maximal bias threshold, indicating that the instrument exerts strong predictive power over firm-level digital transformation.

Table 5. First-stage results: Dependent variable = DIG.

	(1)
	DIG
IV ($\mathrm{TerrainRuggednes} \times \mathrm{DIG}$t−1)	0.921 ***
	(0.020)
N	8777

Note: Standard errors in parentheses *** p < 0.01.

Table 6 presents the outcomes from the second stage of the instrumental variable analysis (IV) regression employed to examine the influence of digital transformation on supply chain resilience. The coefficient estimate for DIG is positively significant at the 1% level or higher, indicating that the main findings of this study remain robust after mitigating endogeneity stemming from reverse causality.

Table 6. Results of the Instrumental Variable Regression Analysis.

Variable	Instrumental Variable
	Resil	Resis	Recov
	(1)	(2)	(3)
DIG	0.0203 *** (0.0044)	0.0209 *** (0.0045)	0.0038 *** (0.0012)
Control variable	YES	YES	YES
Industry/year fixed	YES	YES	YES
observation	8777	8777	8777
Kleibergen–Paap rk LM statistic	1054.05 ***
Kleibergen–Paap rk Wald F statistic	1946.22 (16.38)

Note: Standard errors in parentheses *** p < 0.01. Estimated coefficients reflect conditional associations. IV estimates should be interpreted as suggestive rather than definitive causal evidence.

4.4. Robustness Test

4.4.1. Propensity Score Matching Method (PSM)

To mitigate sample selection bias, this study conducts robustness checks using propensity score matching (PSM).

First, at the industry-year level, firms with a digital transformation level exceeding the median value are classified as the treatment group (DIG_group = 1). Second, we include the following variables as matching covariates: Size, Lev, Board, Dual, Cashflow, and Big4. Subsequently, a 1-to-4 ratio nearest-neighbor matching approach is employed, using a caliper equal to 0.05 and a common support restriction imposed to eliminate bias arising from distant matches. Post-matching balance is assessed using the standardized mean difference (SMD). Results indicate that the SMD for all covariates falls below 3% after matching, marking a substantial improvement in matching quality (Table 7 and Table 8).

Table 7. Covariate Balance Before and After Matching (Standardized Mean Difference).

Variable	Pre-Matching SMD	Matched SMD
Lev	5.1%	−1.9%
Employee	28.3%	−1.1%
Board	0.6%	−2.1%
Dual	1.8%	3.6%
Cashflow	−0.2%	0.7%
Big4	−3.7%	−0.2%

Table 8. Outcomes of Propensity Score Matching Regression.

	(1)	(2)	(3)
Dep. Var.	Recov	Resil	Resis
DIG	0.0047 ***	0.0087 ***	0.0089 ***
	(0.0009)	(0.0028)	(0.0028)
Controls	YES	YES	YES
Industry/year fixed	YES	YES	YES
N	10,218	10,218	10,218
Adj. R2	0.6601	0.1796	0.1798

Note: Standard errors in parentheses *** p < 0.01. Estimated coefficients reflect conditional associations.

Prior to matching, the maximum standardized mean difference (SMD) across covariates exceeded 28% (e.g., for the Employee variable), indicating pronounced discrepancies in multiple features comparing across the treatment and control groups.

Following matching, the SMD values for all covariates declined to approximately 3%, substantially enhancing group comparability. Notably, this outcome significantly outperforms the widely adopted 10% threshold for balance assessment.

These results suggest that the 1-to-4 ratio nearest-neighbor matching strategy—implemented using a caliper of 0.05 and a common support restriction—effectively achieves covariate balance between the treatment and control groups, laying a robust foundation for causal identification in subsequent regression analyses (Figure 2).

Figure 2. Love plot (Standardized Difference Plot Before and After Matching).

These findings support the stability of our primary findings. Table 8 reports robustness checks showing that the estimated associations remain stable across alternative model specifications.

4.4.2. Additional Controls

Including additional control variables—such as growth rate, fixed assets, office expenses, and state-owned enterprise (SOE) status—does not change the main findings (Table 9). The effect of DIG remains positive and significant, demonstrating robustness to potential omitted variable bias.

Table 9. Regression Results Incorporating Control Variables.

	(1)	(2)	(3)
Dep. Var.	Recov	Resis	Resil
DIG	0.0016 ***	0.0091 ***	0.0088 ***
	(0.0006)	(0.0021)	(0.0020)
Controls	YES	YES	YES
Industry/year fixed	YES	YES	YES
N	9353	9353	9353
Adj. R2	0.7186	0.1291	0.1268

Note: Standard errors in parentheses *** p < 0.01.

4.4.3. Excluding Municipalities

Excluding centrally administered municipalities produces consistent results, indicating that our conclusions are not influenced by regional outliers with distinctive policy environments (Table 10).

Table 10. Regression Results Excluding Municipalities Directly Under the Central Government.

	(1)	(2)	(3)
Dep. Var.	Recov	Resis	Resil
DIG	0.0057 ***	0.0108 ***	0.0106 ***
	(0.0006)	(0.0020)	(0.0019)
Controls	YES	YES	YES
Industry/year fixed	YES	YES	YES
N	9757	9757	9757
Adj. R2	0.6537	0.1374	0.1363

Note: Standard errors in parentheses *** p < 0.01.

4.5. Mechanism Analysis

4.5.1. Supply Chain Diversification Mechanism (H2)

The following mechanism analyses examine whether the empirical patterns are consistent with the proposed channels, rather than establishing definitive causal mechanisms.

This paper adopts the research methodology proposed by Wu Qiang and Yao Yuxiu [16]: customer concentration is measured as the ratio of sales to the top five customers to total sales, while supplier concentration is measured as the ratio of purchases from the top five suppliers to total purchases. Subsequently, the supply chain concentration index is formed by averaging these two ratios.

Table 11 suggests that digital transformation substantially decreases both customer and supplier concentration ratios, evidencing an expansion of the firm’s trading network [31]. Consistent with transaction–cost economics, lower information asymmetry afforded by digital platforms enlarges the feasible set of exchange partners, thereby diluting dependence on any single node. The coefficient for customer concentration is markedly larger than that for supplier concentration, underscoring that demand-side diversification proceeds more rapidly in emerging-market contexts where buyer fragmentation is pronounced. These results correspond to H2.

Table 11. Regression Analysis Results of the Supply Chain Diversification Mechanism.

	(1)	(2)	(3)
Dep. Var.	Refinement of Customer Concentration	Supplier Concentration	Refinement of Supply Chain Concentration
DIG	−2.0624 ***	−1.3936 ***	−1.6906 ***
	(0.1626)	(0.1472)	(0.1235)
Controls	YES	YES	YES
Industry/year fixed	YES	YES	YES
N	9560	8741	9714
Adj. R2	0.3825	0.3189	0.3956

Note: Standard errors in parentheses *** p < 0.01.

4.5.2. Supply–Demand Matching Mechanism (H3)

Digital transformation significantly decreases inventory fluctuation (Table 12), reflecting improved alignment between production and market demand. This result lends supportive evidence to H3 and underscores how real-time data analytics enhance coordination across the supply chain, particularly in markets characterized by high demand volatility and limited logistics infrastructure.

Table 12. Regression Analysis Results of the Supply–Demand Matching Mechanism.

	(1)
Dep. Var.	Inventory changes
DIG	−0.0329 ***
	(0.0110)
Controls	YES
Industry/year fixed	YES
N	9349
Adj. R2	0.0619

Note: Standard errors in parentheses *** p < 0.01.

4.6. Heterogeneity Analysis

4.6.1. Ownership Structure

The beneficial impact of digital transformation is more evident in non-SOEs (Table 13). Non-SOEs utilize their agility and market-driven strategies to fully capitalize on digital capabilities, whereas SOEs may be hindered by bureaucratic inertia and social responsibilities.

Table 13. Regression Analysis Results for State-Owned Investment Enterprises and Non-State-Owned Enterprises.

	State-Owned Enterprises	Non-State-Owned Enterprises
Dep. Var.	Resil	Resil
DIG	0.0085 ***	0.0114 ***
	(0.0032)	(0.0024)
Controls	YES	YES
Industry/year fixed	YES	YES
N	4670	5380
Adj. R2	0.1608	0.1405

Note: Standard errors in parentheses *** p < 0.01.

4.6.2. Industry Type

Manufacturing companies gain more advantages from digital transformation compared to non-manufacturing firms (Table 14), as digital transformation facilitates process reengineering and resource optimization-capabilities that are particularly critical in production-intensive sectors.

Table 14. Regression Results of Manufacturing and Non-manufacturing Industries.

	Manufacturing Enterprises	Non-Manufacturing Type Enterprises
Dep. Var.	Resil	Resil
DIG	0.0108 ***	0.0083 **
	(0.0022)	(0.0034)
Controls	YES	YES
Industry/year fixed	YES	YES
N	6872	3397
Adj. R2	0.0899	0.2167

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

4.6.3. Technological Intensity

To investigate the differential impact of digital transformation on supply chain resilience, firms are categorized as technology-intensive or non-technology-intensive based on official industry classifications. Industry codes follow the CSRC classification and the national standard GB/T 4754, with technology-intensive industries identified according to lists of high-tech and technology-intensive manufacturing, as well as knowledge-intensive service sectors, as published by the National Bureau of Statistics and the Ministry of Industry and Information Technology.

The resilience gains are significant in non-tech-intensive industries but insignificant in tech-intensive sectors (Table 15). This suggests that in the latter, rigid technological architectures and global supply chain dependencies may dampen the marginal benefits of digital upgrading.

Table 15. Regression Results of Technology-intensive and Non-technology-intensive Industries.

	Technology-Intensive	Non-Technology-Intensive
Dep. Var.	Resil	Resil
DIG	0.0044	0.0142 ***
	(0.0027)	(0.0026)
Controls	YES	YES
Industry/year fixed	YES	YES
N	4028	6241
Adj. R2	0.0962	0.1392

Note: Standard errors in parentheses *** p < 0.01.

5. Conclusions and Implications

Based on a panel dataset of firms listed on the A-share market in central and western China over the period 2010–2022, this present study explores the link between firm-level digital transformation and supply chain resilience in environments characterized by relative resource constraints. The empirical results indicate that even in settings marked by underdeveloped transportation infrastructure, concentrated supply chain networks, pronounced information asymmetry, and high coordination costs, digital investments can partially alleviate supply–demand matching frictions. By improving firms’ ability to absorb shocks and restore operational continuity, digital transformation contributes meaningfully to resilience, despite structural disadvantages.

The analysis identifies two complementary mechanisms through which digital transformation enhances resilience. First, digital transformation facilitates supply chain diversification, particularly customer diversification, thereby mitigating risks associated with excessive dependence on concentrated markets. Second, digital capabilities improve supply–demand matching by enhancing data acquisition, forecasting accuracy, and coordination efficiency, which in turn reduces inventory volatility and dampens operational disruptions. Importantly, these mechanisms function as compensatory tools rather than perfect substitutes for physical infrastructure, highlighting that digital investments yield resilience gains primarily by improving coordination and information flows rather than by replacing tangible assets. This finding is also consistent with emerging evidence at broader system levels, suggesting that digital infrastructure and AI-enabled services can function as coordination-enhancing mechanisms, generating positive externalities that facilitate resource reallocation and buffer localized disruptions even under infrastructural constraints [32].

Heterogeneity analyses further demonstrate that the impacts of digital transformation in boosting resilience are unevenly distributed across enterprises. Privately owned manufacturing firms and organizations with greater structural flexibility benefit disproportionately, whereas state-owned enterprises, firms in high-technology-intensive sectors, and organizations characterized by hierarchical rigidity exhibit more muted effects. These findings underscore organizational flexibility as a critical moderator in translating digital capabilities into tangible resilience outcomes, suggesting that internal governance structures shape the effectiveness of digital investments.

From a theoretical standpoint, the results are broadly consistent with the core propositions of transaction cost economics, the theory of dynamic capabilities, and the resource-based perspective in resource-constrained contexts. Digital transformation appears to reduce transaction frictions, strengthen environmental sensing, and enhance firms’ capacity for resource reconfiguration. However, the effectiveness of these mechanisms is contingent on both internal organizational characteristics and external infrastructural conditions. Rather than testing the boundary conditions of these theories, this study provides context-specific empirical evidence illustrating how they operate under infrastructural and resource limitations.

Several implications follow. First, the results suggest that digital transformation can serve as a pragmatic resilience-building strategy for firms operating in constrained environments, particularly when investments are oriented toward coordination, data integration, and demand forecasting rather than purely technological sophistication. Second, the heterogeneous effects highlight that digital investments are most effective when complemented by organizational flexibility, implying that managerial and governance reforms may be necessary to fully realize the resilience benefits of digital transformation.

Finally, this research is based on listed firms in central and western China, and its conclusions should not be automatically generalized to all economies. Nevertheless, because many regions share similar structural features, such as underdeveloped infrastructure, high logistics costs, and fragmented markets, the findings offer analytical insights into how firms facing resource constraints can leverage digital tools to enhance supply chain resilience. While this present study does not directly assess environmental, social, or governance (ESG) outcomes, the evidence on resilience-enhancing digital transformation is relevant to broader sustainability discussions, particularly those related to Sustainable Development Goal 9, by illuminating pathways through which digital capabilities can strengthen industrial adaptability in resource-limited contexts.

6. Future Research Directions

Even though this research offers relatively comprehensive evidence from empirical analysis, several important limitations should be acknowledged.

First, digital transformation and supply chain resilience alike are measured using composite indices constructed from secondary data and textual disclosures. While these proxies are widely used in the literature, they may be subject to measurement error and may not fully capture all dimensions of firms’ digital capabilities or network-level resilience. Such measurement imprecision could attenuate estimated effects and calls for caution in interpreting the magnitude of the results.

Second, although an instrumental-variable strategy is employed to mitigate endogeneity concerns, the identification approach relies on an interaction between terrain ruggedness and lagged digital transformation, which may not fully eliminate all alternative channels through which geographic conditions affect supply chain resilience. Therefore, the instrumental-variable estimates should be interpreted as suggestive rather than definitive causal evidence.

Third, the analysis relies on a non-random sample of firms listed on the A-share market in China’s central and western regions. Since publicly listed companies are typically larger and more formalized, the findings may not readily generalize to smaller or unlisted enterprises, nor to other institutional contexts without comparable structural conditions.

This paper only focuses on the paths at the enterprise level. The marginal contributions of different policy levers, such as industrial upgrading plans, fiscal subsidies, or investment in public digital infrastructure, have not been estimated. Future research can place management decisions within the framework of regional policy portfolios and industry-specific institutional frameworks to determine whether the digital initiatives of the government and the private sector complement or squeeze each other out. Making up for this deficiency will help to more comprehensively and contextually explain how digital transformation can sustain supply chain resilience under different institutional and market conditions.