How Business Environments Affect Enterprise Vitality: A Complex Adaptive Systems Theory Perspective

Abstract

As a complex ecosystem, a business environment plays a structural role in shaping enterprise vitality, yet its multidimensional mechanisms remain underexplored, particularly within transitioning economies. This study employs a time-series Global Principal Component Analysis (GPCA) model to measure the vitality levels of 1475 A-share listed enterprises and the quality of business environments across 284 cities between 2008 and 2022 in China. Based on Complex Adaptive Systems (CAS) theory, we propose a three-dimensional “institution–resource–capability” theoretical framework to analyze the impact of a business environment on enterprise vitality and its underlying complex mechanisms. Our results reveal that, (1) a business environment and its constituent subsystems significantly enhance enterprise vitality, with the market environment and innovation ecosystem exhibiting the strongest effects; (2) the revitalizing impact of a business environment is more pronounced for enterprises in the tertiary industry, manufacturing, regulated sectors, and foreign-invested enterprises (FIEs), as well as those operating in Eastern China; (3) mechanistically, the positive association between a business environment and enterprise vitality is consistent with the following three core pathways: mitigating enterprise risks, restructuring resource provision, and cultivating enterprise capability. This research enriches theoretical frameworks for enterprise sustainable development within complex economic systems, while providing valuable policy implications for optimizing business environments in global transitioning economies.

1. Introduction

Amid the complex environment shaped by the technological revolution, the trend of anti-globalization, major-power competition, and drastic market shifts, enterprises are facing unprecedented challenges to both their survival and sustainable development. The fundamental pathway to address these multidimensional pressures lies in the capacity of enterprises to systematically cultivate and continuously strengthen their enterprise vitality. As a comprehensive metric measuring production, operational, and managerial capabilities, enterprise vitality emphasizes long-term growth over short-term prosperity and has become a core strategic objective for enterprise development [1,2]. Enhancing enterprise vitality requires both internal capabilities and supportive external conditions, compelling governments to optimize business environment for institutional dividends [3]. Consequently, the Chinese government has elevated business environment optimization to national strategic status. Data released by the National Bureau of Statistics (NBS) indicate that 27.37 million new business entities were established in China in 2024, among which innovative enterprises exceeded 2.39 million and foreign-invested enterprises totaled 59,080. This demonstrates the effective transmission of an optimized business environment to the release of enterprise vitality. However, enterprises still face institutional bottlenecks throughout their lifecycle, including financing constraints, slow innovation conversion, and market access barriers, which severely hamper the unleashing of enterprise vitality. Therefore, investigating how a business environment influences enterprise vitality has become vital for deepening factor marketization reforms and establishing sustainable growth paradigms.

Current research on business environment and enterprise behavior exhibits a fragmented nature. Scholars have amassed substantial evidence across multiple dimensions, including enterprise investment, financing, trade activities, innovation, and performance. However, disjointed theoretical frameworks and oversimplified variables have prevented comprehensive understanding. This limitation obscures the complex mechanisms through which business environments shape enterprise vitality from an evolving complex systems perspective. Specifically, current research exhibits the following three limitations. First, research dimensions remain fragmented, failing to analyze interdependencies among systemic elements. Scholars have predominantly focused on the impact of a business environment on isolated enterprise activities. For instance, some have examined the influence of the institutional or legal environment on enterprise investment or financing [4,5]; others have explored the relationship between specific business environment subsystems and enterprise imports or exports, such as the market environment [6], political environment [7], legal and institutional environment [8], and infrastructure environment [9], while a further group has investigated the links between the legal environment, market environment, and enterprise innovation or performance [10,11,12,13]. While these studies have deepened domain-specific understanding, their isolated examination of singular enterprise behaviors fails to deconstruct the synergistic transmission network of “institution–resource–capability”. Second, theoretical integration remains deficient, lacking a dynamic evolutionary framework. Current research compartmentalizes Institution-Based View (IBV), Resource-Based View (RBV), and Dynamic Capabilities Theory (DCT), neglecting their synthesis in modeling the co-evolution of agent-adaptive learning and systemic rules. Crucially, studies remain unanchored in the Complex Adaptive Systems (CAS) paradigm of “rulesets–resource flows–adaptive learning”, resulting in superficial systemic mechanism explanations. For instance, an improved legal environment facilitating enterprise financing fundamentally reflects transaction cost reduction [14], while export boundary expansion stems from enhanced resource mobility [8], both representing differentiated transmissions of the same institutional dividend yet mechanically separated through theoretical fragmentation. Third, research directly linking the business environment to enterprise vitality remains surprisingly limited. One possible explanation lies in the inherent complexity of enterprise vitality that complicates measurement, leading extant literature to oversimplify it as enterprise performance [15]. This reductionism ignores interactive dimensions of vitality and cannot explain how business environments synergistically activate enterprise vitality through tripartite pathways: risk mitigation, resource restructuring, and capability cultivation.

Investigations on enterprise vitality have evolved from conceptual foundations to quantitative methodologies, tracing a research trajectory marked by increasing methodological sophistication and multidimensional analysis. In the beginning, Li and Yuan (2002) [16] defined it as the survival capacity, self-development capability, and regenerative power manifested during enterprise development, emphasizing its comprehensive nature. Building upon this, early studies attempted to assess enterprise vitality using integrated subjective–objective approaches: fuzzy clustering analysis, analytic hierarchy process, and fuzzy mathematics evaluation methods [16]. Due to the limitations of methodological maturity, this early-stage research was largely confined to theoretical frameworks, thus lacking dynamic adaptability. In recent years, scholars have shifted their focus to quantifiable models of enterprise vitality, with measurement becoming increasingly refined and multidimensional. Zhu et al. (2019) [17] constructed an enterprise vitality index model spanning business operations, innovation, and information-sharing dimensions. Chen et al. (2020) [18] employed novel TFP-FN-CHK metrics to evaluate vitality. Yang and Deng (2023) [2] adopted the entropy weight method to measure enterprise vitality. Meanwhile, scholars have begun exploring mechanisms through which exogenous factors affect vitality. Bripi (2016), Yu and Liang (2019), and Wang et al. (2024) [19,20,21] examined impacts of policy uncertainty, tax reforms, and similar factors on private enterprise vitality. Nevertheless, such research predominantly isolates the effects of discrete policy variables, lacking systematic deconstruction of the “institution–resource–capability” transmission chain.

In summary, while the existing literature has collectively highlighted the facilitative role of the business environment on enterprise innovation and other activities from perspectives such as IBV, RBV, and DCT, most of these studies are grounded in singular theoretical paradigms. They lack an integrated analytical framework capable of systematically capturing the complex interactive effects and underlying mechanisms among various subsystems of a business environment. This theoretical limitation further leads to the research gap: methodologically, existing studies struggle to effectively examine multiple concurrent causal mechanisms and their context-dependent nature. To address the limitation, this paper develops a three-dimensional theoretical framework of “institution–resource–capability” based on CAS theory. Extending from this framework, we further integrate the IBV, RBV, and DCT to systematically deconstruct the impact of a business environment on enterprise vitality and unravel its underlying complex mechanisms. Specifically, this study addresses the following fundamental questions:

Q1: Through what specific mechanisms does the business environment influence enterprise vitality?

Q2: Does the impact of the business environment on enterprise vitality exhibit systematic heterogeneity?

Q3: How can enterprise vitality and the business environment be empirically measured to capture dynamic changes in vitality across an evolving business environment?

This paper aims to achieve breakthroughs in theoretical framework, conceptual deconstruction, and research methodology, thereby forming marginal contributions in three aspects.

First, regarding the theoretical framework, CAS theory is innovatively introduced, moving beyond the singular theoretical paradigms prevalent in the existing literature. A three-dimensional “institution–resource–capability” theoretical framework is constructed. This framework not only reveals how business environments may influence enterprise vitality through three possible pathways: mitigating enterprise risks, restructuring resource provision, and cultivating enterprise capability, but also emphasizes the synergistic effects among its subsystems. Consequently, a novel theoretical perspective is provided for understanding the dynamic and complex relationship between a business environment and enterprise vitality.

Second, in terms of conceptual deconstruction, the conventional one-dimensional definition that simply equates enterprise vitality with financial performance or innovation output is overcome. Grounded in CAS theory, enterprise vitality is conceptualized as an emergent outcome of non-linear interactions among institutions, resources, and capabilities. It is thus comprehensively measured across three dimensions: viability, growth, and regeneration. This conceptual operationalization not only captures the true developmental state of enterprises in dynamic environments more accurately but also offers a referential multidimensional assessment framework for subsequent research.

Third, in terms of research methodology, the GPCA model is employed to measure both business environment and enterprise vitality. This approach effectively overcomes the limitations inherent in traditional static cross-sectional models, such as the entropy weight method and principal component analysis (PCA), thereby enabling long-term dynamic tracking of their evolutionary patterns within panel data dimensions. Furthermore, this study systematically integrates a suite of advanced econometric tools, including Bartik-type instrumental variables (IVs), dynamic factor models, multi-way clustered standard errors, and temporal mediation models, to construct a multi-level, dynamic, and robust testing framework. This systematic methodological innovation not only enhances the reliability of causal inference and the robustness of conclusions, but also provides a referential analytical paradigm for future empirical research dedicated to complex economic systems.

Additionally, we conduct multidimensional heterogeneity tests across institutional subsystems, enterprise ownership types, industries, and regions. These analyses examine differential impacts of institutional environments on enterprise vitality, thereby furnishing empirical evidence for refining policy systems.

The remainder of the paper is structured as follows. Section 2 establishes the theoretical foundations and develops hypotheses. Section 3 outlines the research design and methodology, with Section 4 presenting empirical findings. Section 5 examines heterogeneous effects across key dimensions, followed by Section 6 testing the hypothesized mechanisms. Section 7 concludes with policy implications, and Section 8 discusses theoretical and practical contributions, limitations, and suggestions for future research.

2. Theoretical Framework

To transcend the limitations of single-theory analyses in extant literature, we ground our “institution–resource–capability” three-dimensional theoretical framework in CAS theory’s core triad: rulesets, resource flows, and adaptive learning. Through integrating IBV, RBV, and DCT to further explicate these theoretical dimensions, we unravel the complex mechanisms through which a business environment influences enterprise vitality more comprehensively.

CAS theory represents a core achievement in complexity science, seeking to explain formation mechanisms of macroscopic order within biological, ecological, social, and economic systems [22]. Its central tenet, “adaptability creates complexity”, posits that systemic complexity emerges from adaptive behaviors and interactions among adaptive agents. CAS theory conceptualizes economic ecosystems as dynamic networks formed through non-linear interplay among adaptive agents, environmental rulesets, resource flows, and collective learning feedback mechanisms [23,24].

Based on CAS theory, this study conceptualizes the business environment as a prototypical CAS: a dynamic system formed through non-linear interactions among multiple adaptive agents (governments, enterprises, financial institutions, etc.), wherein enterprises constitute core adaptive agents with learning capabilities [25]. Enterprise vitality emerges not as a direct determinant of the business environment, but rather through continuous interaction, learning, and adaptation between enterprises and the complex business environment system, including other agents within it (governments, financial institutions, other enterprises, intermediaries, labor forces, etc.). To deconstruct the intrinsic mechanisms of this emergent process, we develop a three-dimensional theoretical framework of “institution–resource–capability” based on CAS theory’s ternary structure of “rulesets–resource flows–adaptive learning”. Specifically, the institutional dimension embodies the dynamic adjustment of environmental rulesets; the resource dimension drives the reorganization feedback of factor flows; and the capability dimension activates agents’ adaptive learning. In addition, this framework further integrates IBV, RBV, and DCT to elucidate micro-level mechanisms across these three dimensions. The theoretical framework is illustrated in Figure 1.

3. Theoretical Mechanism and Research Hypothesis

3.1. Institutional Dimension: Enterprise Risk Mitigation Mechanism

Within the three-dimensional theoretical framework, the institutional dimension serves as the environmental rulesets of CAS, which constrains the boundaries of enterprise risks through adaptive adjustments. To elucidate the regulatory mechanism of the rulesets on enterprise risks, we introduce IBV to reveal its micro-level pathways of the influence. IBV reveals that formal institutions and informal constraints jointly shape enterprises’ risk perception and behavioral boundaries [26]. Consequently, business environment systematically reduces enterprise risks throughout the whole lifecycle via institutional designs characterized by rule of law, transparency, and stability, thereby laying the foundations for unleashing enterprise vitality. At the market entry stage, standardized registration procedures eliminate regulatory gaps between licensing and operational approval, thereby reducing experimentation costs for startups and incentivizing investor engagement in emerging sectors [27]. During the market competition phase, impartial review mechanisms prohibit regional protectionism to mitigate compression risks for small and medium-sized enterprises (SMEs), thereby changing competition toward productive efficiency [28]. In the investment and financing phase, on one front, credit information- sharing platforms demolish bank-enterprise information barriers, attenuate precautionary cash hoarding by enterprises [29], and catalyze the conversion of idle capital into R&D investment. On another front, confronting uncertainties in long-term investments, commitments to policy continuity curb regulatory arbitrariness, enabling enterprises to precisely gauge investment returns and avoid disruptions caused by institutional jumps during transitional periods [30].

3.2. Resource Dimension: Resource Provision Reconfiguration Mechanism

In the theoretical framework, the resource dimension functions as the factor flow in CAS, driving resource allocation optimization through reorganization feedback. To deconstruct how factor flows reshape enterprise resource scarcity and value, we employ RBV to explicate micro-level mechanisms. RBV posits enterprises as bundles of tangible and intangible resources, where sustaining enterprise vitality and competitive advantage hinges on possessing idiosyncratic and immobile resources [31,32]. Continuous improvements in business environment propel enterprise vitality by reconfiguring enterprises’ resource access boundaries, intensifying the competitive leverage effect of resource scarcity, and enhancing resource valuation. This manifests through three pathways: First, agglomeration and retention of explicit regional resources. A conducive business environment facilitates the clustering of critical resources, including capital, technology, and talent, at the enterprise level by reducing institutional barriers in factor mobility [33,34]. Second, gravitational effects for transregionally scarce resources. Driven by the profit-maximization imperative of resource allocation, business environment optimization establishes cross-regional resource gradients, thereby attracting inflows of high-value mobile resources from extra-regional sources, such as capital, commodity flows, and information flows [35]. Such agglomeration not only yields scale economies but also confers inimitable locational embeddedness advantages through resource complementarity. Third, intangible resource empowerment derived from government–enterprise synergy. Optimized environments facilitate access to policy dividends and relational capital, thereby reducing investment arbitrariness, preventing resource misallocation, and enhancing resource circulation resilience and vitality levels [20].

3.3. Capability Dimension: Enterprise Capability Cultivation Mechanism

Within this framework, the capability dimension plays the role of the adaptive learning feedback in CAS, enhancing enterprise capability through activating agents’ adaptive learning. To elucidate how learning feedback catalyzes core enterprise competencies, this study employs DCT to deconstruct micro-level operational paradigms. While RBV reveals unique resources as foundational sources of competitive advantage, DCT further posits that enterprises’ sustained acquisition and maintenance of this advantage depends critically on their core capacities to dynamically integrate and configure resources [36]. These core capacities constitute both the source and manifestation of enterprise vitality. Reforms of the business environment systematically enhance vitality by strengthening enterprises’ strategic focus capabilities, innovation iteration capacities, and competitive adaptation competencies. First, business environment optimization reduces institutional transaction costs, redirecting entrepreneurial energy from non-productive activities towards productive pursuits, thereby augmenting strategic execution capabilities [37,38]. This energy reallocation essentially economizes capability-building costs. Robust property rights protection, judicial impartiality, and contract enforcement diminish defensive resource expenditures, enabling focus on core capability cultivation while preventing capacity fragmentation through unrelated diversification [39,40]. Second, the business environment strengthens innovation transformation capabilities by stabilizing innovation expectations and mitigating investment risks [41,42]. On one hand, optimization measures, such as targeted R&D subsidies and knowledge-sharing platforms, effectively stimulate entrepreneurship, propelling technological leaps through organizational learning and knowledge management [13,43]; on the other hand, a rule-of-law environment safeguards the appropriability of innovation returns, transforming innovation from uncertainty-driven experimentation into sustainable capacity-building [44]. Finally, equitable competitive environments enhance dynamic adaptation capabilities through dual mechanisms: competitive pressures compel resource allocation efficiency, while calibrated market access regulations prevent resource dissipation from cutthroat competition [21], which jointly ensure continuous core competency upgrading within compliant frameworks.

3.4. Heterogeneity Analysis

From the evolutionary perspective of CAS, interactions between business environment and enterprise vitality exhibit significant heterogeneity, fundamentally reflecting asymmetric responses of the “institution–resource–capability” triad to heterogeneous agents within the three-dimensional framework. Firstly, multidimensional complexity intrinsic to the business environment induces functional divergence among subsystems. As an institutional system deeply embedded in economic, political, and social structures [39], its subsystems, such as market environment and innovation ecosystem, exhibit differential contributions to corporate production functions, operational costs, and developmental trajectories. This variation consequently gives rise to heterogeneous effects on enterprise vitality across subsystems. Secondly, enterprises exhibit significant heterogeneity in responsiveness to the business environment due to variations in ownership types, sectoral attributes, and other micro-level characteristics. Enterprises with distinct ownership types and industry attributes demonstrate structural disparities in risk tolerance, resource accessibility, and capability development [2], thus the need to optimize business environment aims precisely to dismantle ownership discrimination and sectoral entry barriers, and reconstruct equitable competitive foundations to synergistically release vitality across heterogeneous agents. Finally, regional disparities in the business environment will induce significant divergence in key dimensions such as marketization levels and institutional enforcement within economic zones, which further leads to systematic variation in the effectiveness of the three mechanisms mentioned above [15].

Based on the above discussion, the hypotheses below are formulated, and the research model is illustrated in Figure 2:

H1. 

Business environment optimization effectively improves enterprise vitality.

H1a. 

Optimizing the business environment facilitates the enhancement of enterprise vitality by mitigating enterprise risks.

H1b. 

Optimizing the business environment facilitates the enhancement of enterprise vitality by reconfiguring resource provision.

H1c. 

Optimizing the business environment facilitates the enhancement of enterprise vitality by cultivating enterprise capability.

H2. 

Business environment subsystems exert heterogeneous effects on enterprise vitality.

H3a. 

The impact on enterprise vitality exhibits heterogeneity across industry sectors.

H3b. 

The impact on enterprise vitality exhibits heterogeneity across ownership types.

H3c. 

The impact on enterprise vitality exhibits heterogeneity across regional disparities.

In summary, the three-dimensional theoretical framework of “institution–resource–capability” constructed in this study does not represent a simple linear addition of the IBV, RBV, and DCT, but could constitute an integrative transcendence from the perspective of CAS. As shown in

Table 1. Comparison of the theoretical frameworks.

Existing Theories	CAS Theoretical Framework	Falsifiable Hypotheses
IBV • Core proposition: Institutions are viewed as exogenous rule sets, with enterprises adapting to institutional constraints through strategic choices to mitigate risks. • Limitation: Treats institutions as static, thereby overlooking mutual interaction and co-evolution between enterprises and institutions.	Institutional dimension • Institutions are endogenously generated rules within the system, constantly debugged and reshaped through interactions among various actors. Their optimization aims to mitigate systemic risks arising from rule uncertainty and friction, thereby creating stability expectations. • Corresponding subsystems: Government and Public Services Market environment	• H1a and H3b Improvements in the business environment, particularly in the government and public services, are hypothesized to exert a stronger enhancing effect on the vitality of private enterprises. This is because private enterprises are more disadvantaged in interactions with governmental authorities and are consequently more sensitive to refinements in “rulesets”. Should empirical findings demonstrate a stronger effect on state-owned enterprises (SOEs), this hypothesis would be falsified.
RBV • Core proposition: Focuses on internal resource endowments of enterprises, which are viewed as the source of competitive advantage. • Limitation: Exhibits a static analytical orientation, failing to adequately explain how external environments dynamically influence the reconfiguration of resource flows and access channels.	Resource Dimension • Resources are conceptualized as “resource flows” circulating within the system. System optimization can reconfigure the pathways, velocity, and direction of these flows, thereby directly reducing factor allocation costs. • Corresponding subsystems: Operational costs Infrastructure Quality	• H1b and H3a Improvements in the business environment, particularly through reductions in operational costs, are hypothesized to exert a stronger enhancing effect on the vitality of manufacturing industries within the secondary industry. This is because such enterprises are most sensitive to cost fluctuations. Should empirical findings demonstrate that business environment enhancements produce stronger effects on the broader secondary industry than specifically on manufacturing industries, this hypothesis would be falsified.
DCT • Core proposition: Enterprises adapt to environmental changes through higher-order capabilities (integrating, building…). • Limitation: Often treats capabilities as a “black-box” attribute of enterprises, underestimating the role of external systems in triggering and cultivating such adaptive learning processes.	Capability Dimension • Dynamic capabilities are conceptualized as adaptive learning behaviors manifested through enterprises’ interactions with complex environments. The external environment serves as the bedrock for cultivating and activating such capabilities. • Corresponding subsystems: Innovation Ecosystem Natural Ecology and Environment	• H1c and H3c Optimizing the business environment, particularly the innovation ecosystem, is hypothesized to exert a stronger enhancing effect on enterprise vitality in the eastern region. This is because enterprises in this region exhibit greater dependence on knowledge spillovers and innovation ecosystems for their adaptive learning processes. Should empirical evidence demonstrate stronger effects in the central or western regions, this hypothesis would be falsified.
Gaps in Theoretical Synthesis • Limited theoretical perspectives with divergent thematic focus • Emphasis on static analysis to the neglect of dynamic dimensions	Theoretical Breakthrough • Systemic emergence: Enterprise vitality is conceptualized as a macro-level emergent outcome resulting from interactions and couplings among the six subsystems. • Non-linear interactions: The influences of the business environment and its subsystems are not fixed but dynamically evolve across contexts.	• H1 and H2 The theoretical framework predicts that neither business environment nor all its subsystems are universally equally important across all contexts. Their dominant effects can be a priori and directionally predicted based on enterprise characteristics (sector, ownership, region), as specified in H3a, b, and c above. These predictions will be rigorously tested in subsequent empirical analyses.

, this framework conceptualizes institutions as endogenous “rulesets”, resources as dynamic “resource flows”, and capabilities as environmentally triggered “adaptive learning”, thereby providing a more systematic and dynamic explanation of the multiple pathways through which the business environment influences enterprise vitality.

More significantly, this framework deduces a series of falsifiable and context-dependent hypotheses (see Table 1), pre-registering their differential directional effects across various industries, ownership structures, and geographical regions. This approach not only establishes a theoretical contribution to providing a testable integrative framework, but also creates a foundation for subsequent empirical interaction tests.

4. Research Design and Methodology

4.1. Econometric Model Specification

To examine the impact of a business environment on enterprise vitality, we construct the following econometric model:

$${\mathrm{v}\mathrm{i}\mathrm{t}}_{\mathrm{i}\mathrm{t}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\mathrm{e}\mathrm{n}\mathrm{v}\mathrm{i}\mathrm{r}}_{\mathrm{i}\mathrm{j}\mathrm{t}}+\mathsf{\beta }{\mathrm{x}}_{\mathrm{i}\mathrm{t}}+\mathsf{\phi }{\mathrm{m}}_{\mathrm{j}\mathrm{t}}+{\mathrm{r}}_{\mathrm{t}}+{\mathrm{r}}_{\mathrm{i}}+{\mathrm{r}}_{\mathrm{g}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{j}\mathrm{t}}$$

(1)

where ${\mathrm{v}\mathrm{i}\mathrm{t}}_{\mathrm{i}\mathrm{t}}$ represents the vitality level of enterprise $\mathrm{i}$ in year $\mathrm{t}$, and ${\mathrm{e}\mathrm{n}\mathrm{v}\mathrm{i}\mathrm{r}}_{\mathrm{i}\mathrm{j}\mathrm{t}}$ indicates the business environment quality of city $\mathrm{j}$ where enterprise $\mathrm{i}$ is located in year $\mathrm{t}$. The variables ${\mathrm{x}}_{\mathrm{i}\mathrm{t}}$ and ${\mathrm{m}}_{\mathrm{j}\mathrm{t}}$ denote enterprise-level and city-level control variables, respectively. ${\mathsf{\alpha }}_0$ is the constant term, while ${\mathsf{\alpha }}_1$, $\mathsf{\beta },$ and $\mathsf{\phi }$ are regression coefficients. The terms ${\mathrm{r}}_{\mathrm{t}}$, ${\mathrm{r}}_{\mathrm{i}},$ and ${\mathrm{r}}_{\mathrm{g}}$ represent the time fixed effects, individual fixed effects, and industry fixed effects, respectively. ${\mathsf{\epsilon }}_{\mathrm{i}\mathrm{j}\mathrm{t}}$ is the error term.

4.2. Variable Selection and Measurement

4.2.1. Dependent Variable

The dependent variable is enterprise vitality (vit). Aligned with CAS theoretical definition, enterprise vitality represents a dynamic composite state manifested through tripartite “institution–resource–capability” interactions. On the basis of this conceptualization, we draw inspiration from the approach of Li and Yuan (2002) [16] and deconstruct enterprise vitality across three life-cycle characteristics: viability, growth, and regeneration. Consequently, following the principles of quantifiability, comprehensiveness, and practicality, a comprehensive indicator evaluation system is constructed, as detailed in

Table 2. Indicator system for enterprise vitality evaluation ¹.

Objective Level	Tier-1 Indicator	Tier-2 Indicator	Tier-3 Indicator	Calculation Formula	Unit	Data Source
Enterprise Vitality	Viability	Solvency Capacity	Debt-to- Asset Ratio	Total Liabilities divided by Total Assets	/	CSMAR Database-China Listed Firms Research Series- Financial Indicators- Solvency
			Current Ratio	Current Assets divided by Current Liabilities	/
		Operational Capacity	Total Asset Turnover	Net Operating Revenue divided by Average Total Assets	/	CSMAR Database-China Listed Firms Research Series- Financial Indicators-Operating Capacity
			Current Asset Turnover	Net Operating Revenue divided by Average Current Assets	/
	Growth	Profitability	Return on Equity	Net Profit divided by Average Net Assets	/	CSMAR Database-China Listed Firms Research Series- Financial Indicators- Growth Capability
			Return on Assets	(Total Profit plus Financial Expenses) divided by Average Total Assets	/	CSMAR Database-China Listed Firms Research Series- Financial Indicators-Earning Capacity
		Development Capacity	Operating Profit Growth Rate	Current Year Operating Profit Increase divided by Prior Year Operating Profit	/	CSMAR Database-China Listed Firms Research Series- Financial Indicators-Growth Capability
			Capital Accumulation Rate	Current Year Equity Increase divided by Beginning Equity	/
		Competitive Capacity	Market Share	Enterprise Operating Revenue divided by Industry Total Revenue	/	CSMAR Database-China Listed Firms Research Series- Financial Indicators-Earning Capacity
	Regeneration	Innovation Capacity	R&D Intensity	Total R&D Expenditure divided by Operating Revenue	/	CSMAR Database-China Listed Firms Research Series-Listed Firm’s R&D and Innovation-R&D
			Intangible Assets Ratio	Year-End Net Intangible Assets divided by Year-End Total Assets	/	CSMAR Database-China Listed Firms Research Series- Financial Indicators-Ratio Structure
		Transformation Capacity	1 divided by Fixed Assets Allocation Ratio	Total Assets divided by Net Fixed Assets	/	CSMAR Database-China Listed Firms Research Series- Financial Indicators-Ratio Structure
			Non-Core Business Profit Margin	Non-Core Business Profit divided by Operating Revenue	/	CSMAR Database-China Listed Firms Research Series- Financial Indicators-Growth Capability

¹ Since all variables are expressed as ratios or proportional indicators, which are dimensionless and bounded within specific numerical ranges, no data transformation is performed during processing.

.

Based on this indicator system, we employ the GPCA model for measurement. This methodological selection is justified by the limitation of conventional PCA or factor analysis, which are only applicable to two-dimensional cross-sectional data comprising samples and indicators, yet unsuitable for panel data incorporating dimensions of samples, indicators, and temporal. Conversely, the core advantage of the GPCA model lies in its use of fixed weights to establish a unified metric, thus ensuring complete temporal comparability of the resultant composite indices [45]. Consequently, this approach not only resolves intertemporal incomparability issues of enterprise vitality values inherent in traditional methods but also eliminates indicator compatibility interference through orthogonal factor loadings separation, thereby enabling accurate measurement of enterprise vitality. With reference to the approach of Bonzo and Hermosilla (2002) [46], the GPCA model for evaluating enterprise vitality is constructed, entailing the following specific steps:

We first construct a time-indexed three-dimensional data table $\mathrm{X}$ for GPCA modeling. Given $\mathrm{m}$ variables used, for each year $\mathrm{t}$ there exists a data table ${\mathrm{X}}_{\mathrm{t}}={\left({\mathrm{X}}_{\mathrm{i}}\right)}_{\mathrm{n}\times \mathrm{m}}$, where $\mathrm{n}$ represents the number of sample points and $\mathrm{i}$ takes values 1, 2, 3,.... These annual tables are then arranged chronologically to form a panel dataset, constituting a large matrix ${\mathrm{T}}_{\mathrm{n}\times \mathrm{m}}$ defined as the global data table, denoted as follows: $\mathrm{X}={\left({\mathrm{X}}^1,{\mathrm{X}}^2,\cdots ,{\mathrm{X}}^{\mathrm{t}}\right)}_{{\mathrm{T}}_{\mathrm{n}\times \mathrm{m}}}^{\prime }={\left({\mathrm{X}}_{\mathrm{i}}\right)}_{{\mathrm{T}}_{\mathrm{n}\times \mathrm{m}}}$. This study compiled data for 1475 listed enterprises, with a measurement system encompassing 13 variables and spanning 15 annual periods, thus establishing a $1475\times 13\times 15$ time-series panel data table. Let ${\mathrm{p}}_{\mathrm{i}}^{\mathrm{t}}$ denote the weight of sample unit ${\mathrm{s}}_{\mathrm{i}}$ in year $\mathrm{t}$, satisfying ${\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{p}}_{\mathrm{i}}^{\mathrm{t}}=1/\mathrm{T}$ and ${\sum }_{\mathrm{t}=1}^{\mathrm{T}}{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{p}}_{\mathrm{i}}^{\mathrm{t}}=1$. The global variable for enterprise vitality can then be defined as ${\mathrm{X}}_{\mathrm{j}}=\left({\mathrm{x}}_{1\mathrm{j}}^1,\cdots ,{\mathrm{x}}_{\mathrm{n}\mathrm{j}}^1;\cdots ;{\mathrm{x}}_{1\mathrm{j}}^{\mathrm{T}},\cdots ,{\mathrm{x}}_{\mathrm{n}\mathrm{j}}^{\mathrm{T}}\right)\in \mathrm{X}$. The corresponding global variance function for this variable is ${\mathrm{e}}_{\mathrm{j}}^2=\mathrm{V}\mathrm{a}\mathrm{r}\left({\mathrm{x}}_{\mathrm{j}}\right)={\sum }_{\mathrm{t}=1}^{\mathrm{T}}{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{p}}_{\mathrm{i}}^{\mathrm{t}}{\left({\mathrm{x}}_{\mathrm{i}\mathrm{j}}^{\mathrm{t}}-\overline{{\mathrm{x}}_{\mathrm{j}}}\right)}^2$, and the global covariance function is ${\mathrm{e}}_{\mathrm{j}\mathrm{k}}=\mathrm{c}\mathrm{o}\mathrm{v}\left({\mathrm{x}}_{\mathrm{j}},{\mathrm{x}}_{\mathrm{k}}\right)={\sum }_{\mathrm{t}=1}^{\mathrm{T}}{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{p}}_{\mathrm{i}}^{\mathrm{t}}\left({\mathrm{x}}_{\mathrm{i}\mathrm{j}}^{\mathrm{t}}-\overline{{\mathrm{x}}_{\mathrm{j}}}\right)\left({\mathrm{x}}_{\mathrm{i}\mathrm{k}}^{\mathrm{t}}-\overline{{\mathrm{x}}_{\mathrm{k}}}\right)$. Since the centroid of the three-dimensional data table $\mathrm{X}$ is $\mathrm{x}={\left(\overline{{\mathrm{x}}_1},\overline{{\mathrm{x}}_2},\cdots ,\overline{{\mathrm{x}}_{\mathrm{p}}}\right)}^{\prime }={\sum }_{\mathrm{t}=1}^{\mathrm{T}}{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{p}}_{\mathrm{i}}^{\mathrm{t}}{\mathrm{s}}_{\mathrm{i}}^{\mathrm{t}}$, the global covariance matrix can be expressed as $\mathrm{V}={\sum }_{\mathrm{t}=1}^{\mathrm{T}}{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{p}}_{\mathrm{i}}^{\mathrm{t}}\left({\mathrm{s}}_{\mathrm{i}}^{\mathrm{t}}-\mathrm{x}\right){\left({\mathrm{s}}_{\mathrm{i}}^{\mathrm{t}}-\mathrm{x}\right)}^{\prime }$.

Moreover, given the data metric matrix $\mathrm{Z}$, denote the first $\mathrm{m}$ eigenvalues of the matrix $\mathrm{Z}\mathrm{V}$ as ${\mathsf{\lambda }}_1\ge {\mathsf{\lambda }}_2\ge \cdots {\mathsf{\lambda }}_{\mathrm{m}}$, with corresponding orthogonal eigenvectors ${\mathrm{\alpha }}_1,{\mathrm{\alpha }}_2,\cdots {\mathrm{\alpha }}_{\mathrm{m}}$, also referred to as the global principal axes $\mathsf{\omega }$ of the panel data. Let ${\mathrm{C}}_{\mathrm{h}}$ represent the $\mathrm{h}-\mathrm{t}\mathrm{h}$ global principal component, which constitutes the collection of projection variables of the set of group points onto the global principal axis ${\mathsf{\omega }}_{\mathrm{h}}$. This corresponds to ${\mathrm{c}}_{\mathrm{h}}\left(\mathrm{t},\mathrm{i}\right)={\left({\mathrm{s}}_{\mathrm{i}}^{\mathrm{t}}-\mathrm{x}\right)}^{\prime }{\mathrm{Z}}_{\mathrm{a}\mathrm{h}}$, where ${\mathrm{c}}_{\mathrm{h}}\left(\mathrm{t},\mathrm{i}\right)={\left({\mathrm{C}}_{\mathrm{h}}\left(\mathrm{1,1}\right),\cdots ,{\mathrm{C}}_{\mathrm{h}}\left(1,\mathrm{n}\right),\cdots ,{\mathrm{C}}_{\mathrm{h}}\left(\mathrm{T},\mathrm{n}\right)\right)}^{\prime }\in {\mathrm{R}}^{\mathrm{T}\mathrm{h}}$. As established in classical principal component theory, the global principal components provide the optimal linear summaries of the original panel data variable system $\mathrm{X}$, with the first global principal component capturing the direction of maximum variance in the data, and so forth.

Finally, in panel data cluster analysis, the similarity measures between enterprise vitality evaluation indicators can be represented as follows: The supremum difference is defined as ${\mathsf{\delta }}_{\mathrm{i}\mathrm{j}}^{\left(1\right)}=\mathrm{s}\mathrm{u}\mathrm{p}\left\{\left|{\mathrm{x}}_{\mathrm{i}}\left(\mathrm{t}\right)-{\mathrm{x}}_{\mathrm{j}}\left(\mathrm{t}\right)\right|,0\le \mathrm{t}\le \mathrm{T}\right\}$; the uniform discrepancy is given by ${\mathsf{\delta }}_{\mathrm{i}\mathrm{j}}^{\left(2\right)}={\int }_0^{\mathrm{T}}\left|{\mathrm{x}}_{\mathrm{i}}\left(\mathrm{t}\right)-{\mathrm{x}}_{\mathrm{j}}\left(\mathrm{t}\right)\right|{\mathrm{d}}_{\mathrm{t}}$; the absolute sum of discrepancy is expressed as ${\mathsf{\delta }}_{\mathrm{i}\mathrm{j}}^{\left(3\right)}={\sum }_{\mathrm{k}=1}^{\mathrm{m}}\left|{\mathrm{x}}_{\mathrm{i}}\left(\mathrm{t}\right)-{\mathrm{x}}_{\mathrm{j}}\left(\mathrm{t}\right)\right|$,where $0\le {\mathrm{t}}_1<\cdots {\mathrm{t}}_{\mathrm{m}}\le \mathrm{T}$; and the Euclidean distance is ${\mathsf{\delta }}_{\mathrm{i}\mathrm{j}}^{\left(3\right)}={\left({\mathrm{x}}_{\mathrm{i}}\left({\mathrm{t}}_{\mathrm{k}}\right)-{\mathrm{x}}_{\mathrm{j}}\left({\mathrm{t}}_{\mathrm{k}}\right)\right)}^2$.

Subsequently, we implement the temporal GPCA using Stata 18.0 in the following six steps, on the basis of the GPCA evaluation model for enterprise vitality and the corresponding indicator data established above. The base-period weights are fixed and applied to all subsequent years to ensure strict comparability of the composite index across different time periods.

Step 1: Direction harmonization and global standardization are performed on the raw data to resolve inconsistencies in measurement units and scales. Logarithmic transformation is applied to inverse indicators for directional alignment. Standardization follows the formula: $\widetilde{{\mathrm{X}}_{\mathrm{j}}}=\left({\mathrm{X}}_{\mathrm{j}}{-\mathsf{\mu }}_{\mathrm{j}}\right)/{\mathsf{\sigma }}_{\mathrm{j}}$, where ${\mathrm{X}}_{\mathrm{j}}$ denotes the original indicator value of the $\mathrm{j}$-th indicator, $\widetilde{{\mathrm{X}}_{\mathrm{j}}}$ represents the standardized value, ${\mathrm{and} \mathsf{\mu }}_{\mathrm{j}}$ and ${\mathsf{\sigma }}_{\mathrm{j}}$ are the mean and standard deviation of the $\mathrm{j}$-th indicator respectively. Subsequently, Kaiser–Meyer–Olkin (KMO) sampling adequacy and Bartlett’s sphericity tests are conducted on the processed global data to assess suitability for GPCA. The results show a KMO statistic of 0.736, indicating sufficient common variance for PCA. Bartlett’s test achieves statistical significance (p < 0.01), rejecting the null hypothesis and confirming the validity of GPCA application.

Step 2: The variance-covariance matrix $\mathrm{V}$ is computed for the standardized data matrix $\widetilde{{\mathrm{X}}_{\mathrm{j}}}$, generating the global data covariance matrix.

Step 3: Eigenvalues ${\mathsf{\lambda }}_{\mathrm{i}}$ and corresponding eigenvectors ${\mathsf{\alpha }}_{\mathrm{i}}$ of $\mathrm{V}$ are calculated, where $\mathrm{i}=1,2,3\cdots$.

Step 4: Variance explained ratios ${\mathsf{\beta }}_{\mathrm{i}}$ and cumulative variance explained ratios ${\mathsf{\theta }}_{\mathrm{k}}$ are determined. The variance explained ratio for the $\mathrm{i}$-th global principal component ${\mathrm{C}}_{\mathrm{i}}\left(1\le \mathrm{i}\le \mathrm{m}\right)$ is given by ${\mathsf{\beta }}_{\mathrm{i}}={\mathsf{\lambda }}_{\mathrm{i}}/{\sum }_{\mathrm{i}=1}^{\mathrm{m}}{\mathsf{\lambda }}_{\mathrm{i}}$, where $\mathrm{m}$ denotes the number of variables. The cumulative variance explained ratio up to the $\mathrm{k}$-th component $\left(1\le \mathrm{k}\le \mathrm{l}\right)$ is: ${\mathsf{\theta }}_{\mathrm{k}}={\sum }_{\mathrm{i}=1}^{\mathrm{l}}{\mathsf{\lambda }}_{\mathrm{i}}/{\sum }_{\mathrm{i}=1}^{\mathrm{m}}{\mathsf{\lambda }}_{\mathrm{i}}$. In time-series GPCA, components with eigenvalues exceeding unity (Kaiser criterion) and cumulative variance explained below 80% are typically retained. Nine principal components with eigenvalues exceeding unity (λ > 1) were identified, explaining 78.69% of the total variance. The consistency of cumulative variance explained before and after rotation supports their use in constructing the composite enterprise vitality index as proxies for original variables.

Step 5: Scores for each global principal component are computed. The score for the $\mathrm{h}$-th component is derived as: ${\mathrm{C}}_{\mathrm{h}}={\mathsf{\alpha }}_1\widetilde{{\mathrm{X}}_1}+{\mathsf{\alpha }}_2\widetilde{{\mathrm{X}}_2}+\cdots +{\mathsf{\alpha }}_{\mathrm{m}}\widetilde{{\mathrm{X}}_{\mathrm{m}}}$, where $\widetilde{{\mathrm{X}}_{\mathrm{i}}}$ denotes the standardized value of the $\mathrm{i}$-th variable.

Step 6: Composite vitality scores are derived. The composite enterprise vitality index for an enterprise in year $\mathrm{t}$ is constructed as: ${\mathrm{C}}_{\mathrm{t}}={\mathsf{\beta }}_1{\mathrm{C}}_1+{\mathsf{\beta }}_2{\mathrm{C}}_2+···+{\mathsf{\beta }}_{\mathrm{i}}{\mathrm{C}}_{\mathrm{i}}$, where ${\mathrm{C}}_{\mathrm{i}}$ is the score of the $\mathrm{i}$-th principal component in year $\mathrm{t}$, $\mathrm{a}\mathrm{n}\mathrm{d} {\mathsf{\beta }}_{\mathrm{i}}$ denotes the variance explained ratio of the $\mathrm{i}$-th component (as defined in Step 4).

To verify the temporal comparability of the composite index, the following robustness checks are conducted. First, to examine the temporal robustness of the factor structure, the full sample is divided into earlier (2008–2015) and later (2016–2022) sub-periods, with GPCA performed separately to construct the indices. The sub-period indices are found to be highly correlated with the full-sample index, with correlation coefficients of 0.93 and 0.91, respectively, demonstrating consistent assessment outcomes across different time frames. Furthermore, the Procrustes similarity statistics for the factor loading matrices of the earlier and later sub-periods are calculated, yielding values of 0.9865 and 0.9814 respectively, both approaching unity, thereby substantiating the high stability of the factor structure over time.

Secondly, to examine the rationality of the weighting assumption, a comparison is conducted between fixed weights and implicit time-varying weights. By calculating both the correlation coefficients and cosine similarity between the first principal component loadings from annual PCA and the global loadings, it is found that the average correlation coefficient across all years is 0.965 (minimum value: 0.932), while the average cosine similarity reached 0.991 (minimum value: 0.974). These results indicate that the global weights maintain strong representativeness across all time periods.

Furthermore, robustness evidence is supplemented from multiple dimensions: First, annual PCA diagnostics demonstrate that all yearly KMO values exceeded 0.7, Bartlett’s tests are consistently significant, and cumulative variance explanation rates fluctuate reasonably around the global value of 78.7%, indicating the suitability of annual data for PCA and the reliability of the global framework. Second, indices generated by different rotation methods (Varimax and Promax) exhibited correlation coefficients exceeding 0.95, confirming the insensitivity of the index to rotation methods. Third, comparisons are made between the GPCA index and indices constructed using both the Bai–Ng dynamic factor model and rolling-window dynamic factor approaches. The results show that correlation coefficients between GPCA and both alternative indices exceeded 0.90 for both level values and first-differenced values, while city rankings remain highly stable, thereby reaffirming the robustness of core findings to factor construction methodology selection.

In summary, multiple tests consistently demonstrate that the composite index developed in this study possesses the temporal robustness and methodological robustness, providing substantial assurance for the reliability of the core conclusions.

4.2.2. Independent Variable

The independent variable is the business environment (envir). Conceptualized through CAS theory, the business environment constitutes a comprehensive external system encountered throughout the enterprise lifecycle, manifesting core subsystems, including market environment, innovation ecosystem, and other critical rulesets. Based on the conceptual foundations, determinants, and evaluation principles of the business environment, and following the methodology established by Zhang et al. (2020) [47], we construct a comprehensive assessment index system as presented in

Table 3. Indicator system for urban business environment evaluation ¹.

Objective Level	Tier-1 Indicator	Tier-2 Indicator	Tier-3 Indicator	Transformations	Unit	Data Source
Business Environment	Operational Costs (cost)	Labor Cost Intensity	Average Annual Salary	natural logarithm	Chinese Yuan (CNY)	EPS Data Platform- China City Database
		Tax Burden	Local Fiscal Revenue divided by GDP	/	/
	Market Environment (market)	Economic Development	GDP per Capita	natural logarithm	CNY
		FDI Absorption	FDI Inflows	natural logarithm	USD 10,000
		Household Consumption Expenditure	Retail Sales Volume	natural logarithm	CNY 10,000
		Fixed Capital Formation	Gross Fixed Capital Formation	natural logarithm	CNY 10,000
	Government and Public Services (govr)	Government Expenditure-to-GDP	Per Capita Fiscal Expenditure	natural logarithm	/
		Education Service Accessibility	Education Expenditure divided by Total Fiscal Expenditure	/	%
		Healthcare Service Coverage	Hospital Beds per 10,000 Population	natural logarithm	unit
		Financial Market Depth	Outstanding Bank Loans	natural logarithm	CNY 10,000
	Infrastructure Quality (infra)	Road Infrastructure Density	Road Area per Capita	natural logarithm	m2	EPS Data Platform- China Urban and Rural Construction Database
		Digital Infrastructure Penetration	Broadband Subscriptions	natural logarithm	10,000 household	EPS Data Platform- China City Database
		Freight Logistics Infrastructure	Road Freight Traffic	natural logarithm	10,000 metric tons
	Innovation Ecosystem (innov)	Innovation Performance	Patents Granted per 10,000 Population	natural logarithm	per 10,000 persons	China Urban Statistical Yearbook https://www.stats.gov.cn/sj/ndsj/ (accessed on 10 January 2025)
		R&D Expenditure Intensity	R&D Expenditure divided by Fiscal Expenditure	/	%	EPS Data Platform- China City Database
		Human Capital Stock	Tertiary Students per 10,000 Population	natural logarithm	person
	Natural Ecology and Environment (natur)	Air Quality Index	PM2.5 Concentration	natural logarithm	μg/m3	Macro Datas https://www.macrodatas.cn/article/1147473457 (accessed on 25 January 2025)
		Urban Green Space Ratio	Green Space per Capita	natural logarithm	m2	EPS Data Platform-China Urban and Rural Construction Database

¹ Patents Granted per 10,000 Population is defined as the annual count of invention patents granted within a city (excluding utility models and design patents) divided by the year-end resident population; Broadband Subscriptions refers to the number of fixed broadband internet access subscriptions per 10,000 households; PM2.5 concentration for each city is derived using global PM2.5 raster data calibrated through Geographically Weighted Regression (GWR) from Washington University in St. Louis, integrated with observed data from air quality monitoring stations across China. The zonal statistics tool within ArcGIS software is employed to extract the annual mean values of all raster cells within each prefecture-level city boundary, representing the PM2.5 concentration of the city.

. Similarly, we employ the GPCA model that is suitable for panel data, to measure business environment quality. The computational procedures are detailed in Section 4.2.1. The initial sample comprised 312 cities. Due to substantial missing data, some cities are excluded from the sample, resulting in a total of 284 cities being retained for analysis.

In addition, we further calculate metric values for the six subsystems of the business environment. Standardization of the original variables is performed based on the means and standard deviations of the internal variables within each subsystem. Subsequently, PCA was conducted on the standardized variables. Based on the total variance explained tables and factor loading matrices, we computed the variance proportion of each common factor relative to total communality. Subsequently, we derived corresponding indicator weights, and then calculated subsystem scores by aggregating constituent indicator values.

4.2.3. Control Variables

The control variables encompass a range of factors influencing enterprise vitality, which are structured hierarchically at both municipal and enterprise dimensions, drawing on approaches from Yu and Liang (2019) and Gao et al. (2024) [13,20]. Specifically, the control variables are as follows: (1) economic development level (lnpgdp), measured by the logarithm of GDP per capita; (2) population density (lnden), represented by the logarithm of the population density; (3) industrial structure (ins), defined as the ratio of secondary and tertiary industry value-added to GDP; (4) enterprise scale (lnsize), measured by the logarithm of the total assets of the enterprise; (5) ownership concentration (conce), quantified using the shareholding percentage of the largest shareholder; (6) enterprise age (age) is measured by the years since establishment, calculated as: $\mathrm{a}\mathrm{g}\mathrm{e}=\ln \left({\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}}_{\mathrm{c}}-{\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}}_{\mathrm{f}}+1\right)$, where ${\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}}_{\mathrm{c}}$ denotes the current year, ${\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}}_{\mathrm{f}}$ represents the enterprise’s founding year, and an extra 1 is added to avoid logarithm of zero in cases of newly established enterprises; and (7) the quadratic term of enterprise age (sage), which is introduced to capture non-linear dynamics between organizational maturity and enterprise vitality. The definitions of all control variables are presented in

Table 4. Control variables definition.

Variable Name	Variable Symbol	Variable Definition/Transformations	Unit	Data Source
Economic development level	lnpgdp	the logarithm of GDP per capita	CNY per person	EPS Data Platform-China City Database
Population density	lnden	the logarithm of population density	Persons per square kilometer
Industrial structure	ins	the ratio of secondary and tertiary industry value-added to GDP	/
Enterprise scale	lnsize	the logarithm of the total assets of the enterprise	CNY	CSMAR Database-China Listed Firms Research Series-Financial Indicators-Earning Capacity
Ownership concentration	conce	the shareholding percentage of the largest shareholder	%	CSMAR Database-China Listed Firms Research Series-Equity Nature
Enterprise age	age	ln(yearc − yearf + 1)	/	CSMAR Database-China Listed Firms Research Series-China Listed Firm’s Basic Information
The quadratic term of enterprise age	sage	the quadratic term of enterprise age	/

.

4.3. Data Sources and Processing Procedures

The initial sample was sourced from the CSMAR database, comprising 80,460 observations from 5364 A-share listed enterprises spanning the period 2008 to 2022. To ensure data quality and alignment with the research objectives, the following procedures were applied to the raw sample:

(1) Financial enterprises, enterprises with ST or *ST trading status, and entities with significant missing financial data were excluded, resulting in the removal of 3482, 88, and 319 enterprises, respectively.

(2) An additional 28 enterprises located in cities with incomplete business environment information were omitted.

(3) With regard to missing values, the methodology adopted by Li (2019) [48] was applied. The specific procedures implemented were as follows: For isolated missing observations occurring in intermediate years, the arithmetic mean values of the preceding and subsequent years were utilized for imputation (e.g., the 2015 and 2017 values were averaged to impute missing 2016 data). For consecutive missing values at the terminal period of the sample, the available historical growth rate of the variable was first calculated, and extrapolation based on this growth rate was subsequently applied. For more complex missing data patterns where these methods proved inadequate, regression imputation was performed by specifying the variable with missing values as the dependent variable and including per capita GDP, population size, and fiscal expenditure as independent variables, with the predicted values of the model used for imputation. It is recognized that these deterministic imputation methods may lead to underestimated standard errors and fail to account for uncertainty attributable to missing data. Consequently, complete case analysis results are provided in the robustness check section to evaluate potential biases arising from the imputation approach.

(4) To mitigate the influence of outliers on the reliability of empirical estimates, all continuous variables were Winsorized at the 1st and 99th percentiles. The city- and enterprise-level datasets were subsequently merged by matching each enterprise to its city of registration (238 cities in total), yielding a panel of 1447 Chinese A-share listed enterprises spanning 2008–2022, with a total of 21,705 observations. The detailed sample screening procedure is illustrated in Figure 3.

Additionally, the Winsorised descriptive statistics for all variables were reported in

Table 5. Descriptive statistics of variables.

Variables	Mean	Median	Standard Deviation	Minimum	Maximum	Observations
vit	−0.005	−0.023	0.202	−0.547	0.787	21,705
envir	1.392	0.920	1.655	−0.952	5.983	21,705
cost	0.011	0.001	0.031	−0.023	0.127	21,705
market	0.153	0.008	0.429	−0.324	1.767	21,705
govr	0.089	0.005	0.248	−0.188	1.023	21,705
infra	0.091	0.005	0.254	−0.192	1.047	21,705
innov	0.103	0.005	0.290	−0.219	1.192	21,705
natur	0.015	0.001	0.043	−0.032	0.176	21,705
lnpgdp	10.97	11.07	0.668	9.184	12.15	21,705
lnden	7.971	7.977	0.669	6.317	9.392	21,705
ins	0.945	0.962	0.056	0.750	1	21,705
lnsize	22.00	21.88	1.359	19.03	25.81	21,705
conce	0.359	0.338	0.153	0.0870	0.750	21,705
age	2.792	2.833	0.356	1.792	3.466	21,705
sage	7.915	8.027	1.914	3.210	12.01	21,705

.

To eliminate dimensional influences and ensure comparability of effect sizes, all variables in this study were standardized based on the means and standard deviations of the full 2008–2022 sample prior to econometric analysis.

5. Empirical Results and Analysis

5.1. Impact of Business Environment on Enterprise Vitality

Prior to estimating Model (1) with panel data, we conducted F-test, Breusch–Pagan (BP) and Hausman tests, which collectively indicated the superiority of the fixed-effects specification. The baseline regression results are presented in

Table 6. Results of the baseline regression ¹.

Variables	Enterprise-Clustered SE				City-Clustered SE	Enterprise-Year Clustered SE	City-Year Clustered SE	Conley Spatial SE
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
envir	0.027 ***	0.025 **	0.016 **	0.017 ***	0.017 ***	0.017 **	0.017 **	0.017 **
	(8.12)	(2.33)	(2.33)	(3.77)	(2.82)	(2.50)	(2.46)	(2.42)
age	0.124 ***	0.179 ***	0.262 ***	0.086 ***	0.086 ***	0.086 ***	0.086 **	0.086 **
	(3.02)	(3.51)	(2.64)	(2.62)	(2.75)	(2.61)	(2.07)	(2.03)
sage	−0.024 ***	−0.040 ***	−0.059 **	−0.005 **	−0.005 **	−0.005 **	−0.005 **	−0.005 **
	(−3.97)	(−3.74)	(−2.27)	(−2.13)	(−2.15)	(−2.11)	(−2.10)	(−2.02)
lnsize	−0.023 ***	−0.011 ***	−0.012 ***	−0.022 ***	−0.022 ***	−0.022 ***	−0.022 ***	−0.022 ***
	(−4.32)	(−3.42)	(−3.65)	(−4.34)	(−3.62)	(−4.30)	(−3.61)	(−3.49)
conce	−0.154 ***	−0.020 **	−0.026 ***	−0.095 ***	−0.095 ***	−0.095 ***	−0.095 ***	−0.095 ***
	(−4.83)	(−2.09)	(−3.90)	(−3.29)	(−4.07)	(−3.21)	(−2.82)	(−3.65)
lnpgdp	0.027 ***	0.012 ***	0.012 **	0.014 ***	0.014 **	0.014 **	0.014 **	0.014 **
	(5.80)	(3.61)	(2.36)	(2.83)	(2.29)	(2.05)	(2.01)	(2.02)
lnden	0.014 **	0.002 **	0.003 *	0.004 *	0.004 *	0.004 *	0.004 *	0.004 *
	(2.02)	(2.49)	(1.88)	(1.85)	(1.84)	(1.80)	(1.79)	(1.74)
ins	0.182 ***	0.016 **	0.017 **	0.091 ***	0.091 ***	0.091 **	0.091 **	0.091 **
	(3.19)	(2.01)	(2.34)	(2.78)	(2.69)	(2.46)	(2.07)	(1.99)
Constant	0.723 ***	0.061 **	0.166 **	0.494 **	0.494 **	0.494 **	0.494 **	0.494 **
	(4.82)	(2.39)	(1.99)	(2.14)	(2.45)	(2.08)	(2.40)	(2.06)
Individual FE	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time FE	No	No	Yes	Yes	Yes	Yes	Yes	Yes
Industry FE	No	No	No	Yes	Yes	Yes	Yes	Yes
Adjusted R2	0.036	0.018	0.024	0.606	0.606	0.606	0.606	0.606
[95% CI]	[0.02, 0.03]	[0.00, 0.05]	[0.00, 0.03]	[0.01, 0.03]	[0.01, 0.03]	[0.00, 0.03]	[0.00, 0.03]	[0.00, 0.03]
Pesaran CD Test								14.73 *** (p = 0.000)
Observations	21,705	21,705	21,705	21,705	21,705	21,705	21,705	21,705

¹ the values in parentheses are the t-test statistics; *, **, *** indicate significance levels at the 10%, 5%, and 1% levels; regarding the clustering levels of standard errors: enterprise-clustered standard errors were employed in columns (1) to (4); city-clustered standard errors in column (5); two-way clustered standard errors along enterprise-year dimensions in column (6); two-way clustered standard errors along city-year dimensions in column (7); and Conley spatial standard errors in column (8); since each enterprise belongs to only one city, the two-way clustered standard errors along enterprise-city dimensions effectively reduce to city-clustered standard errors.

. Column (1) of Table 6 reports the estimation results without controlling for fixed effects. Column (2) incorporates individual fixed effects, while column (3) further introduces time fixed effects. Column (4) simultaneously controls for individual, time, and industry fixed effects.

Furthermore, we note that the core explanatory variable is measured at the city-year level, while the explained variable is measured at the enterprise level. This may introduce correlation in the model residuals among enterprises within the same city. If clustering is applied only at the enterprise level, this intra-city correlation may not be fully captured. Therefore, the clustering level is raised to the city level, and two-way clustering is further employed across three dimensions: enterprise-city, enterprise-year, and city-year. It should be specifically noted that, since each enterprise belongs to only one city, and enterprises are nested within cities, clustering at the city level inherently accounts for enterprise-level correlation to some extent. As a result, the two-way clustered standard errors along enterprise and city dimensions effectively reduce to one-way clustering at the city level, while the estimated coefficients remain unchanged.

To control for potential spatial spillover effects and cross-sectional dependence, Conley’s spatial heteroskedasticity-autocorrelation robust standard errors are employed in this study. The spatial cutoff distance is set at 500 km, with a triangular kernel function used to weight neighboring observations, thereby capturing the decay of spatial correlation with increasing distance. To verify the robustness of this specification, multiple distance thresholds (300 km and 800 km) and alternative kernel functions are tested, with key estimation results remaining consistent. Further support for this approach is provided by Pesaran’s cross-sectional dependence test: the CD statistic of 14.73 (p < 0.01) strongly rejects the null hypothesis of no cross-sectional dependence, confirming that spatial correlation constitutes a genuine econometric concern and justifying the use of Conley standard errors as both necessary and appropriate.

The results in Table 6 indicate that the magnitude of the coefficient for the core explanatory variable remains consistent across different clustering levels, with only the standard errors changing. When more stringent multi-way clustered standard errors are applied, the standard errors of the estimates increase. Following the adoption of more stringent standard errors, the standard errors of the estimates increased, yet the coefficient signs, magnitudes, and statistical significance of the core variables remain highly stable, maintaining significance at least at the 5% level. This suggests that the main findings are reasonably robust to alternative adjustments of standard errors.

As shown in Table 6, the business environment coefficient (envir) is positive and statistically significant at either the 5% or 1% level, thereby validating Hypothesis H1. To evaluate the economic significance of business environment improvements, the baseline regression results from Column 5 of Table 6 are interpreted as follows. First, the standardized effect indicates that a one-standard-deviation increase in the business environment index leads to a significant rise of 0.017 standard deviations in enterprise vitality during the sample period. Second, for a representative enterprise located in a city where the business environment improves from the sample mean (1.392) to the top 10th percentile level (3.943), equivalent to an improvement of 1.54 standard deviations, enterprise vitality is expected to increase by 0.026 standard deviations. Finally, relative to the overall distribution, this effect size (0.026 standard deviations) accounts for approximately 0.4% of the full range of enterprise vitality (6.6 standard deviations), indicating that the improvement in the business environment possesses substantive economic significance.

In Figure 4 we visualize the estimation results in Table 6 using a forest plot, where the impact coefficients of the business environment along with their corresponding 95% confidence intervals (CIs) are displayed. The point estimates of the coefficients are represented by dots, while the horizontal lines indicate the associated CIs.

5.2. Effects of Business Environment Subsystems on Enterprise Vitality

Table 7. Estimation results for business environment subsystems ¹.

Variables	cost	market	govr	infra	innov	natur
factor	0.138 **	0.317 ***	0.162 **	0.194 ***	0.235 ***	0.095 *
	(2.03)	(6.36)	(2.28)	(4.08)	(4.64)	(1.86)
Constant	−0.455 ***	−0.498 ***	−0.421 ***	−0.490 ***	−0.463 ***	−0.402 ***
	(−3.56)	(−3.87)	(−4.35)	(−3.47)	(−4.01)	(−2.63)
Control Variables	Yes	Yes	Yes	Yes	Yes	Yes
Individual FE	Yes	Yes	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes	Yes	Yes
Industry FE	Yes	Yes	Yes	Yes	Yes	Yes
Adjusted R2	0.152	0.157	0.209	0.123	0.223	0.164
Partial R2	0.008	0.038	0.011	0.015	0.022	0.003
Quantile Shift Effect	0.182	0.418	0.214	0.256	0.310	0.125
[95% CI]	[0.00, 0.27]	[0.22, 0.41]	[0.02. 0.30]	[0.10, 0.29]	[0.14, 0.33]	[−0.01, 0.20]
Observations	21,705	21,705	21,705	21,705	21,705	21,705

¹ *, **, *** denote statistical significance at 10%, 5%, and 1% levels, respectively; t-statistics appear in parentheses; full control variable estimates omitted for brevity but available upon request; the standard errors are robust and clustered at the city level.

presents regression results examining the impact of business environment subsystems on enterprise vitality. All six business environment subsystems exert statistically significant positive effects on enterprise vitality, albeit with substantial heterogeneity in effect magnitudes, providing empirical support for Hypothesis H2. Specifically, market environment and innovation ecosystem exhibit the most pronounced vitality-enhancing effects. One standard deviation improvement in their respective indices is associated with an increase of 0.317 and 0.235 standard deviations in enterprise vitality, respectively, which are substantially larger than those of other subsystems. From the perspective of quantile shift effects, when the market environment and innovation ecosystem of a region improve from relatively poor (25th percentile) to relatively good (75th percentile), this is projected to bring about an increase in enterprise vitality of 0.418 and 0.310 standard deviations, respectively. Such a result confirms the pivotal status of fair competition mechanisms and innovation safeguards within business environment reforms. In contrast, while statistically significant, infrastructure quality and government and public services demonstrate standardized coefficients below 0.2 standard deviations, indicating relatively limited practical contributions. This reflects that the scale effects of infrastructure investment have not yet been fully realized, necessitating future governments’ focus on smart infrastructure aligned with industrial needs. The weak statistical significance of operational costs and natural ecology and environment indicates diminishing marginal returns from cost-reduction policies, alongside an urgent government imperative to amplify the positive externalities of ecological improvements.

In Figure 5 we visualize the estimation results in Table 7 using a forest plot, where the impact coefficients of business environment subsystems along with their corresponding 95% CIs are displayed.

5.3. Robustness Check

Robustness checks are conducted through five complementary approaches.

Firstly, sample adjustments to control for specialized cohorts. (1) Exclusion of capital-intensive real estate sector enterprises subject to heightened policy intervention, resulting in the removal of 100 enterprises and a final sample size of 20,205. (2) Removal of enterprises in the four direct-administered municipalities, which exhibit distinct administrative hierarchies and socio-economic environments, leading to the exclusion of 295 firms and a final sample size of 17,280.

Secondly, alternative estimation methods for the baseline model. (1) Application of maximum likelihood estimation (MLE) leveraging its large-sample efficiency. (2) Complete case analysis is employed, whereby all samples with missing values (538 enterprises) are excluded, and estimation is conducted using only complete observations, resulting in a total sample size of 13,635. (3) Re-estimation using raw data without Winsorisation. (4) Continuous variables are Winsorised at the 5th and 95th percentiles.

Thirdly, incorporation of entrepreneur-level controls. (1) Entrepreneur age (bossage), measured by the natural logarithm of actual age. (2) Educational attainment (edu), ordinally scaled (secondary education or below = 1, college = 2, university = 3, postgraduate = 4).

Fourthly, a dynamic factor model is employed to measure the core variables. Given that both business environment and enterprise vitality are likely to exhibit dynamic persistence, the Bai–Ng method is employed to develop a Dynamic Factor Model for the measurement of these two variables.

Table 8. Results of robustness check ¹.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Excluding Real Estate Companies	Excluding Municipalities	Maximum Likelihood Estimation	Complete-Case Analysis	Without Winsorizing the Sample	Winsorized at the 5% Level	Entrepreneur Characteristic	Measured by a Dynamic Factor Model
envir	0.030 ***	0.034 ***	0.016 ***	0.019 ***	0.037 **	0.017 ***	0.027 ***	0.042 ***
	(2.81)	(3.25)	(3.69)	(3.34)	(2.24)	(4.48)	(3.29)	(4.75)
bossage							0.153
							(0.58)
edu							0.338 **
							(2.31)
Constant	−0.679 ***	−0.509 **	−0.256 ***	−0.312 ***	−0.309 **	−0.558***	−0.763 ***	−0.461 ***
	(−2.81)	(−2.13)	(−2.79)	(−3.02)	(−2.06)	(−2.75)	(−4.82)	(−3.34)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Individual FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Industry FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Adjusted R2	0.169	0.158		0.153	0.061	0.165	0.036	0.308
F-statistic	69.767	60.103		42.965	39.378	89.03	73.161	54.823
Observations	20,205	17,280	21,705	13,635	21,705	21,705	21,705	21,705

¹ **, *** denote statistical significance at 5%, and 1% levels, respectively; t-statistics appear in parentheses; full control variable estimates omitted for brevity but available upon request; the standard errors are robust and clustered at the city level.

reports outcomes of the robustness check for the aforementioned four methods, which confirm result stability through consistent coefficient signs across all specifications despite marginal variations in significance levels and magnitude, thus validating the robustness of the baseline regression.

Fifthly, estimation is conducted using a dynamic panel model. This approach is adopted to account for the potential path-dependent nature of enterprise vitality, wherein current vitality levels may be influenced by their prior values. Omission of the lagged dependent variable could introduce bias in the estimates derived from the baseline fixed-effects model due to the neglect of this dynamic adjustment mechanism. To address this, the model is extended to a dynamic specification by incorporating the first-order lag term of enterprise vitality (L.vit) into the set of explanatory variables. To mitigate endogeneity concerns, the System Generalized Method of Moments (SGMM) estimator is employed. This method treats business environment variables and the lagged vitality term as endogenous variables, utilizing all lags from the second period and earlier while adopting a collapsed matrix approach to construct GMM IVs. This ensures instrumental validity while strictly controlling the total number of instruments to avoid overfitting.

It should be noted that the estimated sample size for SGMM (N = 20,145) is smaller than that of the baseline regression (N = 21,705) for two reasons: first, the introduction of the first-order lag term of enterprise vitality (L.vit) leads to the loss of observations from the initial period (2008); and second, the use of higher-order lags to meet instrumental validity requirements inevitably further reduces the time-series sample. Such sample reduction due to dynamic model specification and instrument construction is a common phenomenon in econometric estimation.

Furthermore, to demonstrate that the estimation results are not influenced by the choice of specific sample periods or the order of lag of IVs, the following robustness tests are conducted: (1) Adjustment of the sample time range (Trimming T). SGMM estimations are reperformed using samples from 2009–2022 and 2008–2021. (2) Limiting the maximum lag order of IVs (Limiting Lag Depth). Estimations are conducted by successively restricting instruments to periods $\mathrm{t}-2$, $\mathrm{t}-2$ to $\mathrm{t}-3$, and $\mathrm{t}-2$ to $\mathrm{t}-4$.

The results are presented in

Table 9. Results of SGMM estimation ¹.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	SGMM (2008–2022)	SGMM (2009–2022)	SGMM (2008–2021)	SGMM Lag (2,2)	SGMM Lag (2,3)	SGMM Lag (2,4)
envir	0.050 **	0.015 ***	0.017 ***	0.003 **	0.004 ***	0.003 ***
	(2.19)	(2.77)	(3.39)	(2.44)	(2.85)	(2.62)
L.vit	0.202 **	0.487 ***	0.447 ***	0.823 ***	0.826 ***	0.822 ***
	(2.43)	(3.33)	(3.35)	(33.70)	(36.20)	(39.19)
Constant	0.191 **	0.107 ***	0.116 ***	0.129 **	0.133 **	0.136 **
	(2.28)	(3.08)	(3.14)	(2.06)	(2.17)	(2.24)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Individual FE	Yes	Yes	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes	Yes	Yes
Industry FE	No	No	No	No	No	No
AR(1) [P]	0.001	0.000	0.000	0.000	0.000	0.000
AR(2) [P]	0.528	0.139	0.179	0.149	0.136	0.140
Hansen test [P]	0.307	0.295	0.312	0.322	0.311	0.309
IV(N)	28	45	42	66	87	106
IV-to-panels ratio	0.019	0.031	0.029	0.045	0.060	0.073
Diff-in-Hansen tests [P]	0.114	0.157	0.128	0.172	0.123	0.118
Observations	20,145	18,600	18,600	20,145	20,145	20,145

¹ **, *** denote statistical significance at 5%, and 1% levels, respectively; t-statistics appear in parentheses; full control variable estimates omitted for brevity but available upon request; to mitigate the risk of instrument proliferation, the collapsed instrument matrix is employed for all GMM-style instruments; the standard errors are robust and clustered at the city level.

. All model specifications pass the key diagnostic tests for SGMM: the differenced error terms exhibit significant first-order serial correlation (AR(1) p < 0.01) but no second-order serial correlation (AR(2) p > 0.1); the instruments-to-panels ratio ranges from 0.019 to 0.073, well below 1, indicating no serious instrument proliferation; and the Hansen overidentification tests fail to reject the null hypothesis (p > 0.29), suggesting overall validity of the instruments. On this basis, as shown in Table 9, the coefficients for a business environment (envir) are found to be significantly positive across all specifications (coefficient range: 0.003–0.050, p < 0.05), while the first-lagged enterprise vitality (L.vit) also demonstrate significant positive persistence (coefficient range: 0.202–0.826, p < 0.05). These results collectively indicate that, even after controlling for the dynamic persistence of enterprise vitality, improvements in a business environment continue to exert a statistically significant positive effect on enterprise vitality, and this conclusion remains robust across different sample periods and choices of IV lag orders.

5.4. Endogeneity Test

This study may face endogeneity issues stemming from two sources: bidirectional causality between business environment and enterprise vitality, as well as omitted variables in the model. To address this, endogeneity diagnostics were performed for the business environment variable. The Hausman test result (p = 0.005) strongly rejects the null hypothesis of exogeneity, confirming endogeneity issues. Consequently, two IVs were constructed to mitigate estimation bias, as detailed below.

(1) The first IV is designated as L.envir, which leverages the dynamic persistence of an urban business environment. This selection is grounded in the dynamic persistence characteristic of an urban business environment, where significant autocorrelation is observed between period $\mathrm{t}$ and $\mathrm{t}-1$ conditions. Crucially, the one-period lagged value influences enterprise vitality exclusively through its impact on the contemporaneous business environment, thereby satisfying the exclusion restriction requirement. In accordance with Sun and Wang (2022) [12], L.envir is therefore specified as the first IV. (2) The second IV is the mean value of business environments in other cities within the same province (envir_iv). Its construction is predicated on two fundamental attributes: an intra-provincial business environment exhibits significant spatial interdependence, while conditions in neighboring cities exert no direct causal effect on local enterprise vitality, influencing outcomes exclusively through local environmental transmission. Building upon Yu et al. (2020) [49], envir_iv for city $\mathrm{a}$ is consequently formulated as the arithmetic mean of business environment indices across all other sample cities within the same province, formally specified as: ${\mathrm{e}\mathrm{n}\mathrm{v}\mathrm{i}\mathrm{r}\_\mathrm{i}\mathrm{v}}_{\mathrm{a}\mathrm{t}}=1/{\mathrm{N}}_{\mathrm{a}}{\sum }_{\mathrm{j}\ne \mathrm{a}}^{{\mathrm{N}}_{\mathrm{a}}}{\mathrm{e}\mathrm{n}\mathrm{v}\mathrm{i}\mathrm{r}}_{\mathrm{j}\mathrm{t}}$, where $\mathrm{t}$ denotes the year, $\mathrm{j}$ represents cities within the same province of the target city $\mathrm{a}$, and ${\mathrm{N}}_{\mathrm{a}}$ is the count of such cities. To ensure metric reliability, target cities with ${\mathrm{N}}_{\mathrm{a}}\le 2$ were excluded.

Table 10. Estimation results of endogeneity test ¹.

Variables	L.envir	envir_iv	L.envir and envir_iv	Bartik	All
envir	0.061 ***	0.108 ***	0.074 ***	0.856 ***	0.410 ***
	(4.81)	(4.77)	(3.98)	(4.63)	(2.89)
Conley spatial SE	0.059	0.068	0.061	0.203	0.312
Chi-sq(1) p-value	0.000	0.000	0.000	0.000	0.000
Kleibergen–Paap rk Wald F statistic	24.371	20.034	20.027	22.368	33.421
Sanderson–Windmeijer F			35.221		21.654
Anderson–Rubin 95% CI	[0.036, 0.086]	[0.063, 0.153]	[0.037, 0.111]	[0.492, 1.220]	[0.132, 0.688]
Hansen J statistic (p-value)			0.342		0.255
Control variables	Yes	Yes	Yes	Yes	Yes
Individual FE	Yes	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes	Yes
Industry FE	Yes	Yes	Yes	Yes	Yes
Adjusted R2	0.163	0.085	0.164	0.174	0.796
F-statistic	188.147	249.966	188.606	105.27	75.49
Observations	15,495	16,695	15,495	21,705	15,495

¹ *** indicates 1% significance level; t-statistics in parentheses; LM statistic p-value (Chi-sq(1)) tests instrument relevance; Kleibergen–Paap rk Wald F statistic evaluates weak instruments; Hansen J statistic (p-value) examines overidentification restrictions; complete control variable results archived for researcher correspondence; the standard errors are robust and clustered at the city level.

reports the validity test results for L.envir and envir_iv. The underidentification test yields a p-value < 0.05, rejecting the null hypothesis of weak instrument relevance. The Kleibergen–Paap rk Wald F statistic substantially exceeds the critical threshold of 10, indicating no weak instruments concern. Sargan’s overidentification test produces a p-value > 0.1, failing to reject instrument exogeneity. Collectively, these diagnostics confirm instrument validity. Furthermore, given the Generalised Method of Moments (GMM) estimator’s efficacy in addressing endogeneity from bidirectional causality and omitted variables [50], we re-estimated the model using GMM.

Although both IVs discussed above have passed the weak instrument test and the underidentification test, a discussion regarding potential challenges to the exclusion restriction remains necessary. Firstly, considering the lagged business environment variable (L.envir), while it mitigates reverse causality to some extent, its validity strictly depends on satisfying the exclusion restriction. Specifically, it must affect enterprise vitality solely through its influence on the current business environment. If the business environment exhibits strong persistence, the lagged term might capture information related to earlier historical factors. If these unobserved factors continue to have a direct impact on current enterprise vitality, the estimation results would be subject to bias. Secondly, the validity of the province-level mean IV (envir_iv) is predicated on the assumption that a business environment in a given city is not directly affected by the enterprise vitality of other cities within the same province.

However, in reality, common provincial policy interventions or province-level economic spillover effects may simultaneously influence a business environment and enterprise vitality across cities within the province, potentially violating the exclusion restriction. Despite these theoretical concerns, it is noted that the IVs demonstrate statistically strong performance in the first-stage regression, with the Kleibergen–Paap rk Wald F statistic substantially exceeding conventional critical values, indicating a strong correlation between the IVs and the endogenous variables. While this statistical strength is acknowledged, further examination of these potential issues will be conducted through alternative identification strategies with stronger exogeneity, such as Bartik-type IVs.

To address the aforementioned challenges, an IV with greater exogeneity was constructed following the approach of Bartik (1991) [51]. The variable is defined as ${\mathrm{B}\mathrm{a}\mathrm{r}\mathrm{t}\mathrm{i}\mathrm{k}}_{\mathrm{j}\mathrm{t}}={\mathrm{N}\mathrm{I}}_{\mathrm{t}}\times {\mathrm{I}\mathrm{C}\mathrm{T}\mathrm{S}}_{\mathrm{j},2008}$, where ${\mathrm{N}\mathrm{I}}_{\mathrm{t}}$ denotes the national internet penetration rate in year $\mathrm{t}$, and ${\mathrm{I}\mathrm{C}\mathrm{T}\mathrm{S}}_{\mathrm{j},2008}$ represents the share of establishments in information transmission, computer services, and software (ICT) industries relative to the total number of establishments in city $\mathrm{j}$ as of 2008. The exclusion restriction of this IV is derived from its temporal variation originating from the exogenous national internet penetration shock, while its cross-sectional variation across cities stems from historical industrial shares. The effects of these historical shares have already been absorbed by city fixed effects, and the national shock is unrelated to city-specific disturbances likewise. The first-stage regression results demonstrate a strong correlation between the IV and business environment, as evidenced by a Kleibergen–Paap rk Wald F-statistic of 22.368, which substantially exceeds the critical value for weak identification.

Despite the preliminary satisfaction of IV requirements, we acknowledge the challenges to identification (such as common trends and spatial correlation). Consequently, a series of rigorous robustness analyses are conducted: (1) Weak instrument robust tests are reported, including the Kleibergen–Paap rk Wald F statistic and Anderson–Rubin confidence sets. (2) Leave-one-province-out tests are performed to demonstrate that results do not depend on any single province. (3) A pseudo-Bartik instrument is constructed for placebo testing by interacting unrelated national shocks (growth rate of national grain output) with city-level ICT shares in 2008. (4) City-specific linear time trends are controlled to eliminate confounding from local development trajectories. (5) Spatial correlation is accounted for through reporting Conley spatial standard errors. The results from all these tests consistently support the robustness of the core findings.

As evidenced by the estimation results in Table 10, the coefficients for a business environment remain statistically significant and positive at the 1% level across different IV specifications, indicating robust promotional effects of business environment improvement on enterprise vitality. Specifically, when L.envir is employed as the IV, the estimated coefficient is 0.061; when envir_iv is adopted, the coefficient increases to 1.465, while using the more exogenous Bartik IV yields an estimate of 0.856, which falls between the two aforementioned values. It is notable that IV estimates are generally higher than the OLS results (0.017), consistent with theoretical expectations regarding OLS attenuation bias caused by measurement error and the identification of the local average treatment effect by IVs.

6. Heterogeneity Test

6.1. Industry Heterogeneity

To examine industry heterogeneity, we categorize sample enterprises into primary, secondary, and tertiary industries following China’s Industrial Classification for National Economic Activities (GB/T 4754). Given the central role of manufacturing within the secondary industry, we conduct separate estimations for this subsector to analyze within- industry heterogeneity in greater depth. Furthermore, following Yu and Liang (2019) [20] and considering differentiated business policies for regulated sectors (involving national security, natural monopoly, public services, and high technology) versus non-regulated sectors, we partition the full sample into these two categories to examine heterogeneous effects of policy sensitivity on enterprise vitality.

Table 11. Industry heterogeneity analysis results ¹.

Variables	Primary Industry (1)	Secondary Industry (2)	Tertiary Industry (3)	Manufacturing (4)	Regulated Sectors (5)	Non-Regulated Sectors (6)
envir	0.038	0.014 *	0.057 ***	0.037 ***	0.043 **	0.024 **
	(0.51)	(1.83)	(2.62)	(2.70)	(2.28)	(2.05)
Constant	0.294 ***	−0.637 **	−0.340 ***	−0.454 *	−0.744 **	−0.690 **
	(3.02)	(−2.34)	(−3.62)	(−1.85)	(−2.08)	(−2.19)
Control Variables	Yes	Yes	Yes	Yes	Yes	Yes
Individual FE	Yes	Yes	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes	Yes	Yes
Industry FE	Yes	Yes	Yes	Yes	Yes	Yes
[95%CI]	[−0.11, 0.18]	[0.00, 0.03]	[0.01, 0.10]	[0.01, 0.06]	[0.01, 0.06]	[0.01, 0.08]
Adjusted R2	0.201	0.275	0.130	0.302	0.153	0.249
F-statistic	7.260	57.140	7.941	56.726	8.144	53.278
Observations	330	15,060	6315	13,020	5640	16,065
Interaction terms	envir × Primary −0.021 (−1.6) envir × Tertiary 0.033 ** (2.5)				envir × Regulated 0.019 ** (2.00)
Wald(P)	(1) vs. (2) 0.329 (3) vs. (2) 0.018 (1) vs. (3) 0.048				(5) vs. (6) 0.025
BH Adj. Significant	envir × Primary No envir × Tertiary Yes				envir × Regulated Yes

¹ *, **, *** denote statistical significance at 10%, 5%, and 1% levels, respectively; t-statistics appear in parentheses; full control variable estimates omitted for brevity but available upon request; the standard errors are robust and clustered at the city level; the BH Adj. Significant row indicates whether the interaction term coefficients remain statistically significant after Benjamini–Hochberg adjustment.

reports the estimation results of industry heterogeneity. The facilitative effect of a business environment on enterprise vitality exhibits a significant hierarchical gradient across industries, providing empirical support for Hypothesis H3a. Explicitly, the efficacy of business environment optimization follows the hierarchy: Tertiary Industry > Regulated Sectors > Manufacturing > Non-regulated Sectors> Secondary Industry > Primary Industry. Firstly, the tertiary industry demonstrates the strongest impact, with a standardized coefficient of 0.057, substantially exceeding that of the secondary industry. This highlights that the service industry is more responsive to business environment reforms. The primary industrial statistical insignificance stems fundamentally from its resource-constrained nature. Secondly, the manufacturing demonstrates a highly significant coefficient at the 1% level, along with a larger effect size compared to the secondary industry. Such a phenomenon is attributable to deliberate industrial policy interventions aligned with national manufacturing power strategies. Finally, the promotional effect of business environment optimization on enterprise vitality is significantly higher in regulated sectors (0.043) than in non-regulated sectors (0.024). This divergence stems primarily from differential allocation of policy resources.

To rigorously examine the industry heterogeneity in the impact of a business environment, a model with an interaction term is introduced in addition to the split-sample regressions. In this model, industry types are incorporated as categorical variables, along with interaction terms between the business environment and these industry classifications. For the analysis of heterogeneity across the three major sectors, the secondary industry is designated as the baseline category. Similarly, when examining heterogeneity based on regulatory attributes, non-regulated sectors are used as the baseline group.

Based on the interaction model, Wald tests are employed to conduct statistical inferences regarding coefficient differences between groups, with Benjamini–Hochberg false discovery rate adjustment applied to all interaction term p-values to control for biases arising from multiple hypothesis testing. The results presented in Table 10 indicate significant heterogeneous effects among the three major industries. Specifically, statistically significant differences at the 5% level are identified between the tertiary and secondary industries, as well as between the tertiary and primary industries. A marginally significant difference (p = 0.082) is also observed between the primary and secondary industries. Simultaneously, a significant difference at the 5% level (p = 0.025) is detected between regulated and non-regulated sectors, indicating that regulatory attributes represent another important factor influencing the effectiveness of a business environment.

6.2. Ownership Heterogeneity

Classified by ownership structure, the full sample is partitioned into three subsamples: SOEs (12,465), privately-owned enterprises (POEs) (8730), and FIEs (510). To further examine differential effects of a business environment on enterprise vitality across ownership types, in addition to estimating each subsample separately, interaction terms are incorporated into the full-sample model, with POEs designated as the baseline category. The coefficients of the interaction term reflect differences relative to this baseline group, with their statistical significance determined through Wald tests. To control for bias arising from multiple testing, Benjamini–Hochberg false discovery rate adjustment is applied to all interaction term p-values.

The estimation results in

Table 12. Heterogeneity analysis by ownership structure and region ¹.

Variables	Ownership Heterogeneity	Regional Disparities
envir (POEs/Central)	0.051 *** (3.54)	0.109 *** (2.65)
envir × SOEs	−0.019 ** (−2.13)
envir × FIE	0.012 ** (2.01)
envir × East		0.331 *** (7.36)
envir × West		−0.038 *** (−4.89)
envir × Northeast		−0.032 *** (−3.28)
Constant	0.243 *** (2.76)	0.316 *** (4.56)
BH Adj. Significant	Yes	Yes
[95% CI]	[0.02, 0.08]	[0.03, 0.19]
Control Variables	Yes	Yes
Individual FE	Yes	Yes
Time FE	Yes	Yes
Industry FE	Yes	Yes
Adjusted R2	0.216	0.218
Observations	21,705	21,705
Wald(P)	SOEs vs. POEs 0.038 FIEs vs. POEs 0.061 SOEs vs. FIEs 0.002	Eastern vs. Central 0.000 Western vs. Central 0.040 Northeast vs. Central 0.062 Eastern vs. Western 0.000 Eastern vs. Northeast 0.000 Western vs. Northeast 0.125
Subsample Results	SOEs 0.029 ** (2.50) POEs 0.047 ** (2.43) FIEs 0.068 *** (3.80)	Eastern 0.066 *** (3.87) Central 0.049 ** (2.31) Western 0.018(1.35) Northeast 0.021(0.88)

¹ **, *** denote statistical significance at 5%, and 1% levels, respectively; t-statistics appear in parentheses; full control variable estimates omitted for brevity but available upon request; the standard errors are robust and clustered at the city level; the BH Adj. Significant row indicates whether the interaction term coefficients remain statistically significant after Benjamini–Hochberg adjustment.

support Hypothesis H3b, indicating significant ownership heterogeneity in the impact of a business environment on enterprise vitality, demonstrating a descending pattern of “FIEs > POEs > SOEs”. Using POEs as the baseline (0.051), the estimated effect for SOEs is significantly lower by 0.019, while FIEs are significantly higher by 0.012, resulting in the highest total effect for FIEs (0.063) and the lowest for SOEs (0.032). Wald tests reveal that the difference between FIEs and SOEs is significant at the 1% level (p = 0.002), while the difference between FIEs and POEs is marginally significant at the 10% level (p = 0.061). These findings suggest that FIEs respond most sensitively to business environment improvements, reflecting their superior ability to leverage institutional advantages; POEs still face certain transitional barriers such as financing constraints, whereas SOEs exhibit relatively lower sensitivity to market-oriented business environment changes due to their inherent resource advantages and institutional attributes.

6.3. Regional Disparities in Business Environment

The full sample is stratified into four region-based subsamples: Eastern, Central, Western, and Northeast China, with subsample sizes of 13,485, 3525, 3420, and 1275, respectively. To examine regional heterogeneity in the impact of a business environment on enterprise vitality, in addition to estimating each subsample separately, interaction terms between business environment and regional dummy variables are incorporated into the full-sample model, with the Central region designated as the baseline category. The coefficients of the interaction terms reflect differences between each group and the baseline group, with their statistical significance determined by Wald tests.

As shown in Table 12, the estimation results support Hypothesis H3c, indicating significant regional divergence in the promotional effect of a business environment on enterprise vitality. Using the Central region as the baseline (0.109), the estimated effect for the Eastern region is significantly higher by 0.331, while the Western region is significantly lower by 0.038 and the Northeast region significantly lower by 0.032. Consequently, the total effect is highest in the Eastern region (0.440) and lowest in the Western region (0.071). Wald tests demonstrate that the effect in the Eastern region is significantly greater than all other regions at the 1% level, while the Central region also shows significantly higher effects than the Western region. Overall, a descending pattern is observed: Eastern > Central > Northeast > Western. This suggests that Eastern and Central regions leverage institutional innovation advantages to reap reform dividends, while Western and Northeast regions, constrained by underdeveloped business environments and institutional inertia, have not yet fully activated policy transmission effectiveness. Therefore, these regions urgently require institutional reforms to eliminate regulatory blockages and boost enterprise vitality.

Finally, the results of the heterogeneity analysis are visualized in Figure 6 as a forest plot, which displays the impact coefficients of business environments across different subgroups along with their corresponding 95% CIs.

7. Mechanism Tests

Drawing on the approach of Voelkle et al. (2012) [52], a temporal mediation model is constructed to examine the pathways through which a business environment influences enterprise vitality. Through the inclusion of lagged terms, this model ensures a clear temporal sequence among business environment, mediating mechanisms, and enterprise vitality, aligning with the chronological requirements of causal inference. Specifically, the models are as follows:

$${\mathrm{M}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{e}\mathrm{n}\mathrm{v}\mathrm{i}\mathrm{r}}_{\mathrm{i}\mathrm{j},\mathrm{t}-1}+\mathsf{\beta }{\mathrm{x}}_{\mathrm{i}\mathrm{t}}+\mathsf{\phi }{\mathrm{m}}_{\mathrm{j}\mathrm{t}}+{\mathrm{r}}_{\mathrm{t}}+{\mathrm{r}}_{\mathrm{i}}+{\mathrm{r}}_{\mathrm{g}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{j}\mathrm{t}}$$

(2)

$${\mathrm{v}\mathrm{i}\mathrm{t}}_{\mathrm{i},\mathrm{t}+1}={\mathsf{\gamma }}_0+{\mathsf{\gamma }}_1{\mathrm{e}\mathrm{n}\mathrm{v}\mathrm{i}\mathrm{r}}_{\mathrm{i}\mathrm{j},\mathrm{t}-1}+{\mathsf{\gamma }}_2{\mathrm{M}}_{\mathrm{i}\mathrm{t}}+\mathsf{\beta }{\mathrm{x}}_{\mathrm{i}\mathrm{t}}+\mathsf{\phi }{\mathrm{m}}_{\mathrm{j}\mathrm{t}}+{\mathrm{r}}_{\mathrm{t}}+{\mathrm{r}}_{\mathrm{i}}+{\mathrm{r}}_{\mathrm{g}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{j}\mathrm{t}}$$

(3)

where ${\mathrm{M}}_{\mathrm{i}\mathrm{t}}$ represents the mediating variable in period $\mathrm{t}$, ${\mathrm{e}\mathrm{n}\mathrm{v}\mathrm{i}\mathrm{r}}_{\mathrm{i}\mathrm{j},\mathrm{t}-1}$ represents the business environment in period $\mathrm{t}-1$, and ${\mathrm{v}\mathrm{i}\mathrm{t}}_{\mathrm{i},\mathrm{t}+1}$ indicates enterprise vitality in period $\mathrm{t}+1$. The terms ${\mathsf{\beta }}_0$ and ${\mathsf{\gamma }}_0$ denote constant terms (intercepts), while ${\mathsf{\beta }}_1$, ${\mathsf{\gamma }}_1$, and ${\mathsf{\gamma }}_2$ represent the estimated coefficients for the respective variables. The other variables remain consistent with the definitions provided previously.

7.1. Test of Enterprise Risk Mitigation Mechanism

Given data availability constraints, we focus on business risk exposure (lever) as the mediating variable. Drawing on the approach of Du and Yue (2025) [53], we measure business risk using earnings volatility, specifically defined as the industry-adjusted return on total assets for each enterprise in each observation period. The calculation formula is as follows: ${\mathrm{l}\mathrm{e}\mathrm{v}\mathrm{e}\mathrm{r}}_{\mathrm{i}\mathrm{t}}={\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i}\mathrm{t}}-1/\mathrm{N}{\sum }_{\mathrm{n}=1}^{\mathrm{N}}{\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i}\mathrm{t}}$, where ${\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i}\mathrm{t}}$ represents the return on total assets for enterprise $\mathrm{i}$ in year $\mathrm{t}$, and $\mathrm{N}$ is the total number of enterprises in the industry. Higher values indicate greater earnings volatility and consequently higher business risk for the enterprise. Columns (1) and (2) of

Table 13. Results of the mechanism test ¹.

Variables	Testing the Mechanism of Enterprise Risk Mitigation		Testing the Mechanism of Resource Provision Reconfiguration		Testing the Mechanism of Enterprise Capability Cultivation
	(1)	(2)	(3)	(4)	(5)	(6)
L_envir	−0.025 ** (−2.23)	0.015 ** (2.02)	0.018 ** (2.41)	0.071 ** (2.11)	0.204 *** (5.36)	0.198 *** (3.88)
lever		−0.001 *** (−4.40)
resour				0.183 *** (9.14)
innov						0.213 *** (4.73)
Constant	2.044 *** (3.49)	0.865 *** (4.08)	−0.529 *** (−3.78)	−0.760 *** (−3.60)	0.523 *** (4.03)	−0.37 *** (−3.44)
Control Variables	Yes	Yes	Yes	Yes	Yes	Yes
Individual FE	Yes	Yes	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes	Yes	Yes
Industry FE	Yes	Yes	Yes	Yes	Yes	Yes
Adjusted R2	0.248	0.625	0.610	0.632	0.421	0.072
Observations	20,145	18,600	20,145	18,600	20,145	18,600

¹ **, *** denote statistical significance at 5%, and 1% levels, respectively; t-statistics appear in parentheses; full control variable estimates omitted for brevity but available upon request; the standard errors are robust and clustered at the city level.

indicate that business risk represents a potentially significant transmission channel. The identified mechanism is consistent with optimizing business environment enhancing enterprise vitality by mitigating enterprise risks, thereby validating Hypothesis H1a.

7.2. Test of Resource Provision Reconfiguration Mechanism

Considering that mitigating financing constraints is a pivotal policy orientation for enhancing a business environment, this study selects enterprise credit resources (resour) as the mediating variable to elucidate the resource provision mechanism. Following Jin (2024) [29], we measure it as the ratio of the sum of long-term and short-term borrowings to total assets. The estimation results in Columns (3) and (4) of Table 13 indicate the presence of a mediating effect. This demonstrates that optimizing a business environment aligns with the pathway of enhancing enterprise vitality by reconfiguring resource provision mechanism, which validates Hypothesis H1b.

7.3. Test of Regional Disparities in a Business Environment

Due to the constraints of data availability, we primarily examine innovation capability (innov) as the mediating variable and measure it as the number of invention patents applied for by the enterprise in the given year, following the methodology established by Sun et al. (2025) [44]. Patents are the most commonly used indicator for measuring enterprise innovation, and invention patents represent the highest level of originality among patent types, which establish them a robust proxy for the quality of enterprise innovation. The estimation results presented in Columns (5) and (6) of Table 13 support the presence of a mediating effect. This mechanism is consistent with optimizing business environment enhancing enterprise vitality by cultivating enterprise capability, thereby validating Hypothesis H1c.

7.4. Placebo Testing

To enhance the robustness of the findings and further identify causal direction, a placebo test is conducted. Specifically, the future mediating variable ${\mathrm{M}}_{\mathrm{i},\mathrm{t}+1}$ is regressed on the current business environment ${\mathrm{e}\mathrm{n}\mathrm{v}\mathrm{i}\mathrm{r}}_{\mathrm{i}\mathrm{j}\mathrm{t}}$. If the business environment exhibits no significant predictive power over future mediating variables, the possibility of estimation bias due to reverse causality can be partially excluded. As shown in

Table 14. Results of the placebo test ¹.

Variables	Testing the Mechanism of Enterprise Risk Mitigation (1)	Testing the Mechanism of Resource Provision Reconfiguration (2)	Testing the Mechanism of Enterprise Capability Cultivation (3)
envir	0.001 (0.08)	0.002 (0.43)	0.016 (1.17)
Constant	2.945 *** (6.25)	−0.618 *** (−4.23)	0.589 *** (5.01)
Control Variables	Yes	Yes	Yes
Individual FE	Yes	Yes	Yes
Time FE	Yes	Yes	Yes
Industry FE	Yes	Yes	Yes
Adjusted R2	0.248	0.611	0.403
Observations	20,145	20,145	20,145

¹ *** denote statistical significance at 1% levels, respectively; t-statistics appear in parentheses; full control variable estimates omitted for brevity but available upon request; the standard errors are robust and clustered at the city level.

, the coefficient for the business environment is statistically insignificant, indicating a low likelihood of endogeneity caused by reverse causality, thereby providing additional robustness support for the previously discussed mechanistic analysis.

8. Conclusions and Policy Implications

8.1. Conclusions

Based on theoretical analysis and empirical tests, the principal findings of this study are summarized as follows.

First, while the optimization of the overall business environment and its individual subsystems all significantly enhance enterprise vitality, the effects of the subsystems exhibit heterogeneity. Notably, the market environment and innovation ecosystem substantially outperform other subsystems in boosting enterprise vitality, emerging as the core driving forces.

Second, the impact of a business environment on enterprise vitality manifests gradient variations across industries, ownership types, and geographic regions. Specifically: (1) the industrial gradient follows: Tertiary Industry > Regulated Industries > Manufacturing > Non-regulated Sectors > Secondary Industry > Primary Industry. (2) Ownership-type variations exhibit: FIEs > POEs > SOEs. (3) Regional disparities display: Eastern China > Central China > Northeast China > Western China.

Third, mechanism tests demonstrate that the process through which business environment enhances enterprise vitality aligns with three potential pathways: mitigating enterprise risks, restructuring resource provision, and cultivating enterprise capability. These findings collectively support the tripartite theoretical framework of “institution–resource–capability” proposed in this study.

8.2. Policy Implications

Based on the research conclusions, the following policy recommendations are proposed.

Firstly, reforms of a business environment should prioritize breakthroughs in the development of the market environment and innovation ecosystem. Specifically, governments must enhance market access facilitation, strengthen intellectual property protection mechanisms, and refine innovation incentive policy design, thereby unlocking the catalytic effect of the market environment and innovation ecosystem on enterprise vitality.

Secondly, the reforms should also establish a gradient-based policy system for industries and market entities, employing differentiated approaches to energize all sectors and achieve full-scale efficiency enhancement. Explicitly, sectoral policy should do the following: (a) advance service-sector marketization with incentive-balanced regulation, (b) accelerate manufacturing’s digital transformation, and (c) enhance factor mobility in foundational industries through infrastructure upgrades. Moreover, governments should simplify negative lists and cross-border capital flow management for FIEs, expand financing channels and ensure competitive neutrality for private enterprises, and advance mixed-ownership reforms with market-based evaluations for SOEs. Regionally, governments should designate Eastern China as a pilot zone, and then transfer proven reforms to the Central, Western, and Northeast regions through inter-regional collaboration mechanisms with calibrated phasing.

9. Discussion

We compare the findings of this study with results from the existing literature employing similar methodologies. The convergence lies in the confirmation of the positive impact of a business environment on enterprise behavior and the existence of heterogeneity in this impact, aligning with the research of Zhang (2022), Li et al. (2024), Dabboussi (2023), and Asadzade (2024) [5,15,28,35]. However, divergences emerge in the three aspects below. Firstly, we innovatively utilize the “rulesets–resource flows–adaptive learning” tripartite structure from complex adaptive systems theory to develop a novel three-dimensional “institution–resource–capability” theoretical framework. Integrating IBV, RBV, and DCT, this framework provides a deeper analysis of the micro-level mechanisms operating within each dimension. Consequently, theoretical hypotheses were proposed suggesting that a business environment may influence enterprise vitality through three pathways: risk mitigation, resource restructuring, and capability cultivation. This theoretical framework transcends the static limitations in the existing literature that separately examine institutional factors, resources, and capabilities [8,13,14]. It restructures the transmission mechanism through which business environment influences enterprise vitality, thereby systematically revealing complex effects that cannot be explained by any single theory. Secondly, in contrast to the static cross-sectional metrics (e.g., entropy weight method, PCA) in the extant literature [2,10,11,12], this research pioneers a modeling via GPCA to measure business environment and enterprise vitality. Notably, our approach captures dynamic evolutionary patterns of enterprise vitality during business environment optimization, achieving dual breakthroughs in measurement accuracy and methodological suitability. Thirdly, compared to the studies by Bripi (2016), Yu and Liang (2019), and Wang et al. (2024) [19,20,21], our heterogeneity tests are more granular and comprehensive, yielding new insights. For instance, we have identified the pivotal enhancing role of the market environment and innovation ecosystem among the six subsystems of a business environment, and the differential effects observed between regulated and non-regulated sectors.

Specifically, the findings of this study indicate that the policy effectiveness in the Western and Northeast regions has not been fully activated. This phenomenon is rooted in deep-seated institutional bottlenecks and structural challenges prevalent in these areas. Firstly, market fragmentation and barriers to factor mobility are identified as core obstacles. Compared to the Eastern region, these areas exhibit higher internal market barriers, where the flow of production factors such as labor, capital, and technology continues to face restrictions both within and across regions. As a result, effects of business environment optimization policies are severely constrained. Secondly, strong institutional inertia plays a critical role. The historical dominance of traditional industries and SOEs in the economic structure, along with entrenched governance mentalities and established government–business relations, has led to an institutional lock-in effect. Consequently, new policies aimed at streamlining approvals, promoting fair competition, and incentivizing innovation struggle to overcome existing path dependency, resulting in policy ineffectiveness. Finally, inadequate policy implementation capacity and a lack of tolerance for trials and errors further undermine policy efficacy. Local governments often face dual pressures of deepening reforms while ensuring stability. Without effective incentives or an environment that allows experimentation, policy implementation often becomes conservative and ritualistic. Consequently, many well-designed policies fail to deliver tangible benefits to enterprises during execution.

Therefore, this study holds significant theoretical and practical implications. Theoretically, we integrate multiple theories including CAS to examine the dynamic mechanisms by which business environments shape enterprise vitality, thereby advancing institutional theory from a paradigm of unidirectional environmental constraints toward a multidimensional co-evolutionary theoretical framework. Moreover, this study systematically constructs a conceptual framework of enterprise vitality, which overcomes the limitation in traditional literature that focuses on performance of enterprises while neglecting the dynamic behavior in their development. Consequently, it provides a new perspective for understanding the sustainable development capabilities of enterprises within complex economic ecosystems. In practice, this study establishes pathways for precisely implementing targeted policies to optimize a business environment, based on its proposed multidimensional gradient response patterns and differentiated policy recommendations.

In spite of the insights gained, this study is subject to several limitations which also indicate directions for future research.

Firstly, a limitation pertains to the sample coverage. Although the use of data from A-share listed companies in China ensures standardized data quality, the vast number of unlisted enterprises, especially small and SMEs were not included. Given the significant sectoral heterogeneity identified in this study, it is speculated that the sensitivity to business environments and the underlying mechanisms for unlisted SMEs may differ systematically from those of large, listed enterprises. Therefore, caution should be exercised when generalizing these findings to unlisted enterprises. Future research could focus on collecting micro-level data from unlisted enterprises to directly test whether the heterogeneous patterns hold in broader samples, thereby providing a more comprehensive understanding of the impact spectrum of a business environment.

Secondly, a limitation pertains to the contemporaneity of variable measurement. During the construction of indicators, constraints of data availability resulted in insufficient incorporation of key digital transformation dimensions in the measurement of core variables. For instance, business environment indicators lacked assessments of digital infrastructure (such as 5G coverage) and the digitalization level of policy implementation; meanwhile, enterprise vitality measurements failed to effectively capture dynamic capabilities in leveraging data element and implementing digital transformation. Consequently, the positive effects of a business environment may have been underestimated in regions and industries at the forefront of digital economic development. Future research should prioritize the development of an integrated indicator system that combines both digital and physical dimensions. This would not only complement the current measurement methodology but also represent a significant expansion of the connotations of “institution–resource–capability” theoretical framework in the digital era.

Thirdly, a limitation exists regarding the methodological alignment with the theoretical framework. Although the theoretical framework was constructed based on CAS theory, empirical analysis primarily employed regression-based econometric models. These methods possess inherent limitations in capturing core CAS characteristics such as multi-agent non-linear interactions and systemic emergence. For instance, while interactive effects among subsystems were identified, econometric models cannot simulate the non-linear threshold effects or path dependence that these interactions may generate over the long term. Therefore, a promising direction for future research would be to utilize the empirical findings of this study (such as the impact coefficients of various subsystems) as parameter inputs for constructing Agent-Based Modeling simulations. This approach will enable the simulation of adaptive behaviors across different types of enterprises, thereby achieving genuine integration of theory and methodology within the CAS paradigm. Furthermore, more sophisticated methodologies such as panel SEM and g-computation could be adopted in future research as potential directions for further mechanistic investigations.