From scale to substance: how green credit policy execution and “purity” reshape regional innovation

Introduction: China’s green credit system has expanded dramatically, yet substantial heterogeneity in local policy implementation raises questions about its effectiveness in spurring environmental innovation. This paper introduces the novel concept of “Green Credit Purity”—the proportion of green credit flowing to genuinely environmental projects—to examine how policy implementation intensity and allocation quality jointly determine innovation outcomes.

Methods: Using panel data covering 1,600 firms across 280 prefecture-level cities from 2010–2022, we employ fixed effects, spatial Durbin models, and instrumental variables to establish causal relationships.

Results: We find that a one-standard-deviation increase in local Green Credit Policy Implementation Intensity (GCPI) generates a 2.1% increase in firm-level green patents and 3.3 percentage points improvement in regional Green Total Factor Productivity. Crucially, firms with high credit purity achieve 68% larger innovation responses than those engaging in “greenwashing.” Spatial analysis reveals positive spillovers equivalent to 40% of direct effects, while threshold regression identifies a critical point (GCPI = 0.416) beyond which policy stringency triggers firm relocation.

Discussion: These findings demonstrate that green credit effectiveness depends critically on both implementation intensity and allocation quality, with implications for designing environmental finance policies that balance local stringency with regional coordination to maximize genuine innovation while minimizing regulatory arbitrage.

1 Introduction

Over the past decade, China’s stock of green credit has expanded rapidly, yet its spatial allocation remains strikingly uneven. Descriptive evidence from sectoral and banking statistics indicates that concessional lending is disproportionately concentrated in developed urban clusters—precisely the places with smaller marginal abatement space (Climate Policy Initiative, 2023; Schumacher et al., 2020; Schoenmaker and Schramade, 2019; Zhang et al., 2021). This misalignment raises a simple but consequential puzzle: why are financial resources intended to accelerate decarbonization disproportionately allocated to regions with limited additional emissions-reduction potential? Addressing this puzzle requires moving beyond national targets and aggregate loan volumes to examine how city-level execution intensity and the quality of credit allocation shape both local and neighboring innovation outcomes.

In practice, national green-credit guidelines and aggregate targets are implemented through a multi-layered chain of city-level indicators, supervisory reviews, and lending mandates for local branches of policy and commercial banks. Municipal regulators and bank managers retain considerable discretion over how strictly to screen projects, how tightly to apply environmental criteria, and how aggressively to expand labelled “green” portfolios. As a result, cities facing the same national policy can differ markedly in the intensity and credibility of green-credit execution on the ground. Against this backdrop, a central question is: how does the intensity of city-level green-credit policy execution shape real outcomes—green innovation and emissions reduction? If execution is weak or distorted, volume alone may fail to deliver environmental additionality, even when aggregate green-credit targets are met.

A growing literature links green finance to technological upgrading, innovation, and environmental performance (Schoenmaker and Schramade, 2019; Schumacher et al., 2020; Flammer, 2021). Most empirical studies in this stream treat green credit policies as a binary or scalar shock—typically proxied by guideline issuance or the aggregate volume of labelled loans—and identify positive effects on firm innovation or regional green development at the provincial or industry level (Hu et al., 2021; Hong et al., 2021; Yao et al., 2021). Recent contributions, including Rajah et al. (2023) and Zitouni et al. (2023), further document the expansion of green credit instruments and their potential links with regional green development. However, two limitations hinder precise answers to the above question. First, a scale bias: many studies operate at national or provincial levels, masking substantial within-province heterogeneity in allocation and enforcement intensity across cities (Zhang et al., 2021; Xu, 2020; Yao et al., 2021). Second, an underappreciation of execution heterogeneity: existing metrics typically tally the quantity of green credit but rarely capture the quality of allocation (i.e., whether funds reach verifiably environmental projects) or the credibility of local enforcement, leaving room for greenwashing and weak additionality (Lyon and Maxwell, 2011; Delmas and Burbano, 2011; Deschryver and de Mariz, 2020; Flammer, 2021). As a result, we still know little about whether and how city-level execution intensity and the quality of credit allocation jointly shape green innovation, how these effects propagate across space, and whether there exist behaviorally meaningful policy thresholds for execution intensity.

We argue that progress requires a city-level execution measurement system that distinguishes policy quantity from allocation quality. Concretely, we: (i) construct a Green Credit Policy Implementation index at the prefecture-city level to capture how forcefully green-credit rules are executed locally; and (ii) introduce a Green Credit Purity measure—the share of “green” lending actually reaching verifiably environmental uses—to capture allocation quality and guard against greenwashing. This dual lens is grounded in classic information and incentive theories: when project types are hard to observe, quality signals and credible screening shape the mapping from policy instruments to real outcomes (Spence, 1973; Stiglitz and Weiss, 1981; Holmström and Tirole, 1997).

Getting execution right is high-stakes. On the upside, credible city-level execution can channel capital toward technologies with higher social returns, amplifying green innovation and productivity, consistent with the modern reinterpretation of the Porter hypothesis (Porter and van der Linde, 1995; Jaffe and Palmer, 1997; Ambec et al., 2013), and generating positive spatial spillovers through knowledge diffusion and supply-chain linkages (Yu et al., 2020; Calel and Dechezleprêtre, 2016). On the downside, poorly designed or asymmetric execution can misallocate funds, incentivize greenwashing (Lyon and Maxwell, 2011; Delmas and Burbano, 2011), or even trigger displacement if firms relocate from stricter to laxer jurisdictions—a spatial equilibrium concern increasingly documented in city-level studies (Zhang et al., 2021; Guo and Tan, 2023; Zhao et al., 2023). Understanding these trade-offs is essential for designing finance that delivers genuine decarbonization rather than merely re-labeling credit flows.

Guided by the above logic, we assemble a matched city–firm panel and pair it with our execution and purity measures to examine three questions: (1) Direct effects—does stronger execution raise firm-level green innovation and city-level green productivity? (2) Quality moderation—does higher purity amplify (or condition) the innovation and abatement returns to execution intensity? (3) Spatial equilibrium—do execution differences create spillovers or displacement across neighboring cities, and are there thresholds beyond which stringency backfires via firm relocation? Together, these questions allow us to move from asking whether green credit “works” on average to identifying how execution intensity, allocation quality, and spatial interactions jointly determine its effectiveness.

This paper advances the empirical literature on green credit and regional innovation through four contributions that address measurement gaps and identification challenges in prior work. We ground these contributions in recent studies by Rajah et al. (2023) and Zitouni et al. (2023) while extending their frameworks to incorporate execution heterogeneity and allocation quality.

First, we develop a measurement system that distinguishes policy implementation intensity from allocation quality. Most empirical studies proxy green credit using policy adoption dates or aggregate lending volumes (Hu et al., 2021; Hong et al., 2021; Yao et al., 2021), which obscure variation in local enforcement and fund targeting. We construct a city-level Green Credit Policy Implementation (GCPI) index from lending data, sectoral focus, and credit quality indicators, paired with a firm-level Green Credit Purity measure that quantifies the share of labeled green credit flowing to verifiable environmental projects. This approach allows us to test whether policy effectiveness depends on both enforcement stringency and allocation accuracy. Our estimates indicate that a one-standard-deviation increase in GCPI (0.18) associates with 0.6–0.9 percent increases in firm-level green patents and 2.3 percentage point improvements in regional GTFP, with effects amplified by 68% among firms with high credit purity.

Second, we operationalize allocation quality through systematic measurement. We construct the Purity indicator via text analysis of firm disclosures, applying explicit inclusion criteria, evidence thresholds, and double-coding protocols. Inter-coder reliability (Cohen’s $\kappa =0.87$) and validation against third-party environmental ratings confirm measurement consistency. Supplementary Material File documents the coding protocol, operational definitions, and quality controls. This systematic approach addresses concerns about greenwashing (Lyon and Maxwell, 2011; Delmas and Burbano, 2011; Flammer, 2021) by quantifying whether labeled green credit reaches genuine environmental projects, a dimension prior work has discussed conceptually but rarely measured directly.

Third, we link firm-level innovation responses to regional environmental outcomes within a unified framework. Rather than analyzing green credit impacts at a single level (Liu et al., 2023a; b; Gao et al., 2025), we estimate effects on both firm green patenting and city-level Green Total Factor Productivity. This design demonstrates how micro-level innovation aggregates into measurable regional productivity gains, providing empirical evidence of the policy transmission mechanism from financial intervention to economy-wide environmental performance.

Fourth, we decompose spatial effects and identify nonlinear policy responses. Using spatial Durbin models, we separate direct local effects from spillovers to neighboring cities, finding that indirect effects equal 38–40 percent of direct effects. This pattern indicates positive spatial externalities through technology diffusion and supply-chain linkages. Threshold regression reveals that innovation responses attenuate and firm relocation probabilities increase when GCPI exceeds 0.416, suggesting that uncoordinated policy intensification may trigger displacement. This framework reconciles findings of technology spillovers (Zhang et al., 2021) with evidence of pollution displacement (Su et al., 2024; Zheng et al., 2025) by showing that both mechanisms operate under different policy intensity regimes.

We address identification concerns through instrumental variable estimation and extensive diagnostics. Section 4.2 details the construction of GCPI and Purity measures. Supplementary Material File presents first-stage diagnostics, weak-instrument-robust inference, overidentification tests, and falsification exercises that examine pre-policy periods and alternative outcomes. These materials establish that our estimates are robust to measurement choices and reflect causal relationships rather than spurious correlation.

2 Literature review and hypotheses development

2.1 Green credit and green innovation: the theoretical underpinnings

Since its introduction in 1991, the Porter Hypothesis has been influential in shaping our understanding of the environment-economy nexus by asserting that properly designed environmental regulations can spark innovation and enhance competitiveness, creating a paradigm where ecological preservation and economic prosperity can be mutually reinforcing (Porter and van der Linde, 1995). Two decades of empirical scrutiny have yielded nuanced insights: while early panel studies found positive correlations between environmental regulations and R&D expenditure (Jaffe and Palmer, 1997), subsequent research has differentiated between “weak,” “narrow,” and “strong” versions of the hypothesis (Ambec et al., 2013). Recent meta-analyses encompassing multiple countries suggest an overall positive relationship between environmental regulation and green innovation, though effects vary by regulatory design and implementation context (Cohen and Tubb, 2018; Yang and Lv, 2025).

Within this evolving theoretical landscape, green finance has emerged as a sophisticated market-based mechanism that transcends traditional command-and-control approaches. Green credit policies, now implemented across major economies including China, Japan, and the European Union, represent a fundamental shift in environmental governance—leveraging financial markets to internalize externalities while maintaining economic efficiency (Chen and Chen, 2021; Gao et al., 2025; Xu, 2020). Evidence from quasi-natural experiments in China demonstrates that green credit guidelines can stimulate innovation in heavily polluting enterprises, particularly when policies are well-targeted and enforcement is credible (Hu et al., 2021; Hong et al., 2021). International comparisons reveal diverse implementation strategies: while Japanese sustainable finance emphasizes technology-specific support through mechanisms like the Bank of Japan’s climate facility at preferential rates (Schumacher et al., 2020), European approaches integrate green taxonomy standards with mandatory disclosure requirements (Rubashkina et al., 2015).

The mechanics of green credit extend beyond simple interest rate subsidies. Contemporary green finance employs a sophisticated toolkit including differentiated lending rates (often 50–200 basis points below market rates), extended loan tenures, relaxed collateral requirements, fast-track approval processes, and increasingly, green bonds that tap capital markets directly (Yao et al., 2021; Deschryver and de Mariz, 2020; Flammer, 2021). This multi-pronged approach addresses the market failure wherein environmentally beneficial innovations face disproportionate financial barriers despite generating positive externalities. By systematically reducing capital costs for green projects (global climate finance reached $1.3 trillion in 2021–2022, according to Climate Policy Initiative (2023)), these mechanisms not only alleviate immediate funding constraints but also create powerful signaling effects that redirect private capital flows toward sustainable technologies (Liu Z. et al., 2023; Xia et al., 2024). Evidence from the energy sector specifically shows that firms accessing green credit demonstrate improved environmental performance alongside enhanced financial returns, validating the Porter Hypothesis in practice (Dogah et al., 2025). Recent empirical work further documents that green finance can foster regional green development and technological upgrading, though with marked regional heterogeneity in its effectiveness (Liu et al., 2025; Ma et al., 2025). Implication. These arguments yield a testable expectation that we formally state in the Conceptual Framework and Hypotheses section: city-level execution intensity (GCPI) should be positively associated with green innovation and environmental performance, conditional on how finance intermediates regulatory incentives.

Recent empirical work provides convergent evidence on green finance effects while highlighting remaining gaps. Du et al. (2022) examine green development spillovers across 283 Chinese cities during 2003–2017, demonstrating that municipal green initiatives generate positive spatial externalities through technology diffusion and industrial upgrading. Their findings indicate that a one-unit increase in local green development index associates with 0.15-unit increases in neighboring cities’ environmental performance, consistent with knowledge spillover mechanisms. Liu et al. (2025) analyze prefecture-level data and find that green finance promotes regional sustainable development primarily through innovation channels, though effects vary substantially across regions with different institutional capacities. Ma et al. (2025) document heterogeneous impacts of green credit on urban green development, showing that effectiveness depends critically on local financial market development and government environmental commitment.

These studies advance understanding of green finance impacts but typically abstract from two dimensions central to policy transmission. First, they measure green finance using aggregate lending volumes or binary policy indicators, which cannot distinguish implementation intensity from nominal adoption. Second, they do not quantify allocation quality—whether labeled green credit flows to verifiable environmental projects or suffers from greenwashing. Our GCPI and Purity measures address these gaps by capturing both enforcement stringency and environmental additionality, enabling tests of whether policy effectiveness depends on implementation quality beyond mere scale.

2.2 Policy heterogeneity and the challenge of implementation

China’s decentralized governance structure creates a compelling natural experiment in green finance policy implementation. While this federal-style system has historically fostered regional competition and economic dynamism (Zheng et al., 2025), it simultaneously generates substantial variations in how localities translate national environmental mandates into practice. Local governments operate under conflicting institutional pressures: performance evaluations that still emphasize economic growth alongside mounting environmental compliance requirements (Guo and Tan, 2023). This dual accountability framework creates powerful incentives for heterogeneous policy responses.

The resulting implementation landscape is strikingly diverse. Progressive jurisdictions may aggressively leverage green credit mechanisms and impose stringent environmental monitoring, viewing sustainability as a competitive advantage in attracting high-quality investment. Conversely, regions facing acute development pressures often adopt more permissive approaches, offering regulatory flexibility to secure immediate economic gains. This regulatory competition can trigger a “race to the bottom” dynamic, where jurisdictions systematically undercut environmental standards to maintain their attractiveness to mobile capital (Zheng et al., 2025). Such heterogeneity manifests directly in varying degrees of Green Credit Policy Implementation Intensity (GCPI), creating a patchwork of enforcement regimes with potentially far-reaching consequences.

This fragmented implementation landscape raises a critical concern: the spatial displacement of environmental costs. When regions with stringent GCPI successfully incentivize local green innovation and reduce pollution, they may inadvertently push polluting industries toward neighboring jurisdictions with laxer oversight. The net environmental effect becomes ambiguous—local gains may be offset or even exceeded by increased emissions elsewhere, creating what economists term “pollution havens” (Su et al., 2024). Empirical evidence from Lyu et al. (2023) and Zhao et al. (2023) suggests that such spatial spillovers are not merely theoretical concerns but observable phenomena with measurable cross-regional impacts. Related studies in the green finance literature also highlight that financial instruments intended to promote low-carbon transitions can, under certain conditions, exacerbate carbon emissions or shift them across regions rather than reducing them uniformly (Yadav et al., 2025; Orikpete and Ewim, 2024). These findings underscore that regional policy implementations cannot be evaluated in isolation—their effectiveness depends critically on understanding the broader spatial equilibrium they generate. Consequently, analyzing how GCPI heterogeneity shapes both innovation incentives and pollution geography becomes essential for designing policies that achieve genuine rather than merely redistributed environmental improvements. Implication. We therefore anticipate spatial responses to heterogeneous GCPI—either spillovers through diffusion and linkages or displacement toward laxer jurisdictions—which we formulate as a testable hypothesis in the Conceptual Framework and Hypotheses section.

2.3 The moderating role of green credit purity

A growing body of literature has identified quality concerns in green finance implementation, highlighting the disconnect between policy intentions and actual environmental outcomes. Lyon and Maxwell (2011) and Delmas and Burbano (2011) establish the theoretical foundation for understanding “greenwashing” behaviors, where organizations adopt symbolic environmental practices without substantive commitment to environmental improvement. In the context of green finance, Flammer (2021) demonstrates that the environmental impact of green bonds varies substantially depending on the actual use of proceeds, with many issuers failing to achieve meaningful environmental additionality. Similarly, Yu et al. (2020) provides empirical evidence of widespread greenwashing in China’s green credit markets, where firms strategically rebrand conventional projects to access preferential financing while generating minimal environmental benefits.

The concept of environmental additionality, well-established in climate policy literature (Calel and Dechezleprêtre, 2016), has received limited attention in green finance research despite its importance for policy effectiveness. Schoenmaker and Schramade (2019) argue that sustainable finance mechanisms must incorporate rigorous screening processes to ensure capital flows toward projects with genuine environmental impact, yet most existing studies focus on the scale rather than the quality of green credit allocation. Recent empirical work by Zhang et al. (2021) reveals substantial heterogeneity in green credit effectiveness across Chinese cities, suggesting that implementation quality may be as important as implementation intensity.

Drawing on financial intermediation theory (Holmström and Tirole, 1997) and signaling models (Spence, 1973), we introduce the concept of Green Credit Purity to capture the proportion of officially designated green credit that flows to projects with verifiable environmental benefits. This measure addresses a fundamental gap identified by Dikau and Volz (2021), who call for more sophisticated metrics to evaluate green finance policy effectiveness beyond simple lending volumes. Our conceptualization builds on the framework of Cui et al. (2022) for measuring green credit allocation efficiency, extending it to incorporate environmental outcome verification.

We argue that Green Credit Purity functions as a critical moderating variable in the relationship between policy implementation intensity and innovation outcomes. This hypothesis draws theoretical support from resource allocation literature (Stiglitz and Weiss, 1981), which demonstrates that credit market efficiency depends critically on screening mechanisms that direct capital toward its most productive uses. In the environmental context, Chen et al. (2024) provide preliminary evidence that regions with more stringent green project verification achieve superior innovation outcomes per unit of green credit deployed. However, the systematic measurement and theoretical integration of this quality dimension remains underdeveloped in existing literature.

The policy implications of this quality-quantity trade-off are notable. Akomea-Frimpong et al. (2022) emphasize that poorly designed green finance taxonomies can create perverse incentives, leading to resource misallocation and undermining long-term environmental goals. Our Green Credit Purity concept offers a practical framework for policymakers to monitor and improve allocation quality, while providing researchers with a tool to better understand the conditions under which green finance policies succeed or fail in fostering genuine environmental innovation. Implication. If credible screening raises environmental additionality, the GCPI–outcome link should strengthen with higher Green Credit Purity; we state this moderation hypothesis in the Conceptual Framework and Hypotheses section.

2.4 Competing views on spatial effects and research boundary

The literature offers two competing views on the spatial consequences of environmental and green finance policies, with recent empirical evidence supporting both mechanisms under different conditions. One strand emphasizes pollution haven or “race to the bottom” dynamics, where stringent regulation in one jurisdiction induces firms to relocate to laxer areas, potentially worsening environmental outcomes at the aggregate level. Zhao et al. (2023) provide systematic evidence that stricter environmental regulation in eastern Chinese cities drives polluting firms to relocate westward, generating spatial displacement rather than emissions reductions. Lyu et al. (2023) document strategic environmental regulation where jurisdictions compete for mobile capital by offering lax enforcement, creating race-to-the-bottom dynamics. Recent international evidence by Yadav et al. (2025) and Orikpete and Ewim (2024) suggests that green finance instruments can exacerbate carbon leakage when implementation is uncoordinated across regions, shifting emissions geographically rather than reducing them globally.

A second strand, often focusing on innovation-oriented or green finance instruments, documents positive spillovers through knowledge diffusion, supply-chain linkages, and coordinated upgrading in neighboring regions. Du et al. (2022) provide comprehensive evidence of positive spatial externalities in green development across 283 Chinese cities during 2003–2017, finding that municipal environmental initiatives generate technology adoption and industrial upgrading in neighboring jurisdictions. Their spatial Durbin model estimates indicate indirect effects approximately 40–50 percent as large as direct effects, suggesting substantial cross-regional benefits. Similarly, Zhang et al. (2021) document spatial clustering in green total factor productivity, attributing patterns to inter-city learning and supply-chain linkages that propagate environmental technologies. Yu et al. (2020) emphasize knowledge diffusion and demonstration effects as mechanisms through which green credit policies generate positive spatial externalities. Calel and Dechezleprêtre (2016) document similar patterns in carbon pricing contexts, showing that environmental policies create innovation spillovers that benefit neighboring jurisdictions.

These divergent findings partly reflect differences in sample level (countries versus cities), time period (early versus more mature stages of green finance), spatial weighting schemes (geographic contiguity versus economic distance or network linkages), and, crucially, the heterogeneity of policy execution intensity across space. Studies finding displacement typically examine regulatory shocks in contexts with weak monitoring or large cross-jurisdictional enforcement gaps, while those documenting positive spillovers often focus on coordinated policy implementation or innovation-intensive sectors where knowledge diffusion dominates mobility incentives.

Our analysis is explicitly designed to speak to this tension: by combining a city-level GCPI measure with spatial Durbin models and effect decomposition, we can distinguish local from neighboring impacts and assess whether stronger green credit execution primarily generates net positive spillovers or merely displaces pollution. The threshold regression framework allows us to test whether both mechanisms operate sequentially—positive spillovers dominating at moderate implementation intensity, displacement emerging only when stringency exceeds critical levels. We return to this question in the discussion of our spatial results and threshold estimates in Section 6.

Beyond the specific focus on city-level green credit execution and regional innovation, a broader literature examines how green finance, environmental regulation, and technological change interact in other contexts—for example, in resource-based economies, sector-specific transitions, or digital transformation of production systems (Satpathy et al., 2025). These studies provide valuable complementary perspectives but typically operate at different levels of aggregation or with distinct policy instruments. Our framework is therefore best viewed as one piece of a wider research agenda on how financial and regulatory tools jointly shape the geography of green development.

3 Conceptual model and theoretical framework

3.1 Overview and model motivation

To formalize the complex interactions between heterogeneous green credit policy execution and firm innovation decisions, we develop a stylized two-region, two-sector optimization model that captures the essential trade-offs facing both policymakers and firms. The model’s architecture illuminates three critical dimensions of policy heterogeneity: the direct incentive effects on local firms, the spatial displacement of economic activity, and the welfare implications of uncoordinated policy regimes.

Conceptually, three channels are central for our analysis (Figure 1). First, higher green-credit policy execution intensity at the city level—captured empirically by our GCPI index—relaxes financing constraints for green projects and raises the relative cost of capital for traditional, high-emission activities. This resource-availability channel increases the volume and feasibility of green innovation. Second, allocation quality, measured by Green Credit Purity, governs how effectively labeled green credit is screened and targeted toward projects with verifiable environmental benefits. This screening and signaling channel suppresses greenwashing and raises environmental additionality per unit of green finance. Third, inter-city differences in execution intensity interact with factor mobility costs (relocation and market-access costs) to shape spatial responses: when execution is coordinated and gaps are moderate, knowledge diffusion and supply-chain linkages generate positive spillovers; when gaps are large relative to mobility costs, high-emission activities may relocate, creating threshold-type displacement effects. Our formal model embeds these three channels and provides testable predictions for the GCPI–Purity–innovation relationship and its spatial manifestations.

Figure 1

Figure 1. Conceptual mechanism linking green credit policy execution (GCPI), allocation quality (Purity), and spatial responses. City-level execution intensity relaxes financing constraints for green projects and raises the relative cost of traditional activities (resource-availability channel). Allocation quality governs screening and signaling, suppressing greenwashing and raising environmental additionality (screening channel), while also amplifying the marginal returns to execution intensity. Together, these channels shape firm-level green innovation $(G{I}_{it})$ and city-level green productivity $(C{P}_c)$. Inter-city factor mobility costs $({\kappa }_i+{\ell }_i)$ determine whether GCPI gaps across cities generate positive spillovers through knowledge diffusion and supply-chain linkages, or threshold displacement effects when gaps become large relative to mobility costs.

We consider two regions—labeled as the Center $(C)$ and the Periphery $(P)$—each hosting firms that can choose between green and traditional production technologies. The asymmetric labeling reflects the reality that regions often differ in their initial economic development, regulatory capacity, and political economy constraints, leading to systematically different policy implementation intensities. Each region $c\in \{C,P\}$ is characterized by a policy environment:

$\left\{{\varphi }_c,r_G^c,r_T^c,{\tau }_c,M_c\right\},$

where ${\varphi }_c$ represents the minimum green credit share mandated by local authorities; $r_G^c$ and $r_T^c$ denote the effective interest rates for green-sector and traditional-sector firms, respectively; ${\tau }_c$ is a pollution fee or environmental tax; and $M_c$ captures the total credit supply available in the region.

The model generates testable predictions about how policy parameter variations across regions create differential innovation incentives while simultaneously generating spatial externalities through firm mobility and pollution displacement. For empirical implementation, we interpret the model’s policy stringency parameter $s_c$ as the structural counterpart of our city-level Green Credit Policy Implementation index (GCPI), constructed in Section 4 from green credit volumes, sectoral targeting, and non-performing loan rates. Similarly, the allocation-quality parameter ${\text{Purity}}_c$ corresponds to our Green Credit Purity measure, which captures the share of labeled green credit reaching projects with verifiable environmental benefits. Finally, firm-specific relocation and market-access costs $({\kappa }_i+{\ell }_i)$ map into observed relocation probabilities and cross-city execution gaps that we exploit in the threshold and spatial analyses.

3.2 Regional welfare functions and economic constraints

Each regional government seeks to maximize social welfare $W_c$, which encompasses the standard economic benefits of production and innovation while accounting for environmental costs. This welfare function reflects the fundamental tension between economic development objectives and environmental quality that characterizes real-world policymaking. We denote capital allocated to the green sector as $K_{Gc}$ and to the traditional sector as $K_{Tc}$, subject to two binding constraints that capture the economic realities of regional credit markets:

$K_{Gc}+K_{Tc}\le M_c,(1)$

$\frac{K_{Gc}}{K_{Gc}+K_{Tc}}\ge {\varphi }_c.(2)$

Constraint (Equation 1) reflects the fundamental scarcity of financial resources, while constraint (Equation 2) embodies the regulatory requirement that drives green credit policy implementation. The welfare function takes the form (Equation 3):

$W_c=\underset{\text{Aggregate\hspace{0.17em}production\hspace{0.17em}profits}}{\underbrace{{\mathrm{\Pi }}_c\left(K_{Gc},K_{Tc},r_G^c,r_T^c,{\tau }_c\right)}}-\underset{\text{Environmental\hspace{0.17em}damage\hspace{0.17em}costs}}{\underbrace{D\left(E\left(K_{Tc}\right)\right)}}+\underset{\text{Innovation\hspace{0.17em}spillover\hspace{0.17em}benefits}}{\underbrace{S\left(I\left(K_{Gc}\right)\right)}},(3)$

where ${\mathrm{\Pi }}_c(\cdot )$ represents aggregate firm profits that depend on both capital allocation and the policy instruments; $D(\cdot )$ captures environmental damage from pollution $E(K_{Tc})$ generated by traditional-sector production; and $S(\cdot )$ measures the positive externalities from green innovation $I(K_{Gc})$ that benefit the broader regional economy.

This welfare specification embeds several crucial economic assumptions. First, traditional-sector production generates environmental externalities that impose social costs, creating a classic market failure. Second, green-sector investment produces knowledge spillovers that benefit other firms and sectors, generating positive externalities that private firms undervalue. Third, the policy instruments ($r_G^c$, $r_T^c$, ${\tau }_c$) allow regional governments to internalize these externalities, though their effectiveness depends on implementation intensity and cross-regional coordination.

3.3 Lagrangian formulation and optimization framework

The constrained optimization problem is solved using the Lagrangian method, which provides both the optimal allocation conditions and the shadow prices that reveal the economic trade-offs inherent in policy design. For each region $c$, we define Lagrange multipliers ${\lambda }_c\ge 0$ for the credit constraint and ${\mu }_c\ge 0$ for the green ratio requirement (Equation 4):

$\mathcal{L}={\displaystyle \sum _{c\in \left\{C,P\right\}}}\left[W_c\left(K_{Gc},K_{Tc},{\tau }_c,r_G^c,r_T^c\right)+{\lambda }_c\left(M_c-\left(K_{Gc}+K_{Tc}\right)\right)+{\mu }_c\left(K_{Gc}-{\varphi }_c\left(K_{Gc}+K_{Tc}\right)\right)\right].(4)$

The multipliers have clear economic interpretations: ${\lambda }_c$ measures the marginal value of additional credit supply in region $c$, while ${\mu }_c$ captures the shadow cost of the green ratio constraint—essentially, the economic cost of forcing additional capital into green sectors beyond what would be optimal absent the regulatory requirement.

3.4 First-order conditions and economic equilibrium

The first-order conditions reveal the marginal conditions that must hold in the optimal allocation, providing insight into how policy parameters influence firm behavior and regional outcomes. Taking partial derivatives of $\mathcal{L}$ with respect to each choice variable yields.

3.4.1 (a) optimal green sector capital allocation

$\frac{\partial \mathcal{L}}{\partial K_{Gc}}=\frac{\partial W_c}{\partial K_{Gc}}-{\lambda }_c+{\mu }_c\left(1-{\varphi }_c\right)=0.(5)$

3.4.2 (b) optimal traditional sector capital allocation

$\frac{\partial \mathcal{L}}{\partial K_{Tc}}=\frac{\partial W_c}{\partial K_{Tc}}-{\lambda }_c-{\mu }_c{\varphi }_c=0.(6)$

These conditions embody a fundamental economic principle: in equilibrium, the marginal social benefit of capital in each sector must equal its marginal social cost, adjusted for the constraints imposed by policy requirements. The presence of ${\mu }_c$ in both equations reveals how the green ratio constraint distorts allocation away from the unconstrained optimum—when ${\mu }_c>0$, the constraint is binding and creates a wedge between the marginal products of capital in the two sectors.

3.4.3 (c) optimal environmental taxation

$\frac{\partial \mathcal{L}}{\partial {\tau }_c}=\frac{\partial W_c}{\partial {\tau }_c}=0.(7)$

This condition requires that the marginal benefit of pollution reduction (through reduced environmental damage) equals the marginal cost in terms of foregone production profits—the classic Pigouvian tax principle.

3.4.4 (d) optimal interest rate differentiation

$\frac{\partial \mathcal{L}}{\partial r_G^c}=\frac{\partial W_c}{\partial r_G^c}=0,(8)$

$\frac{\partial \mathcal{L}}{\partial r_T^c}=\frac{\partial W_c}{\partial r_T^c}=0.(9)$

These conditions determine the optimal spread between green and traditional sector interest rates, balancing the innovation incentives created by subsidized green credit against the fiscal costs and potential market distortions.

3.5 Complementary slackness and regime characterization

The Karush-Kuhn-Tucker (KKT) conditions provide additional structure to the solution, determining when constraints are binding and characterizing different policy regimes:

${\lambda }_c\left(M_c-\left(K_{Gc}+K_{Tc}\right)\right)=0,{\lambda }_c\ge 0,(10)$

${\mu }_c\left(K_{Gc}-{\varphi }_c\left(K_{Gc}+K_{Tc}\right)\right)=0,{\mu }_c\ge 0.(11)$

These conditions identify three possible regimes for each region. When ${\mu }_c=0$, the green ratio constraint is slack, indicating that firms voluntarily choose green investment above the minimum requirement—a scenario that emerges when green technologies are sufficiently profitable or when environmental consciousness drives behavior beyond regulatory mandates. When ${\mu }_c>0$, the constraint binds exactly: $K_{Gc}={\varphi }_c(K_{Gc}+K_{Tc})$, indicating that regulatory pressure is necessary to achieve the desired green investment level. The magnitude of ${\mu }_c$ in this case measures the economic cost of the regulatory distortion.

3.6 Equilibrium analysis and economic insights

Solving the system of Equations 5–11 yields the optimal policy vector $\{K_{Gc}^{*},K_{Tc}^{*},{\tau }_c^{*},r_G^{c*},r_T^{c*}\}$ for each region, from which several key economic insights emerge:

Policy complementarity effects: Increasing the minimum green ratio ${\varphi }_c$ or reducing the green credit interest rate $r_G^c$ shifts capital toward green projects, generating innovation spillovers that benefit the entire regional economy. However, this reallocation imposes costs on some firms and may reduce short-term economic output, creating political economy pressures that influence policy sustainability.

Spatial arbitrage mechanisms: Higher pollution taxes ${\tau }_c$ or stricter green credit requirements create incentives for high-pollution firms to relocate to regions with more lenient policies. This firm mobility generates inter-regional pollution transfers—the so-called “pollution haven” effect—that can undermine the environmental effectiveness of uncoordinated policies.

Strategic policy interactions: When regions fail to coordinate their policies, a “race to the bottom” dynamic may emerge, with each jurisdiction competing to attract mobile capital by offering more lenient environmental standards or more generous credit terms. This strategic interaction can lead to a prisoners’ dilemma where all regions end up with suboptimally weak environmental policies.

Innovation spillover amplification: The positive feedback between green investment and innovation spillovers $(S(I(K_{Gc})))$ suggests that regions with initially higher green credit implementation may experience accelerating returns to environmental policy, creating potential for widening gaps in environmental performance across regions.

3.7 Theoretical hypotheses and empirical predictions

The theoretical framework generates several testable predictions that form the foundation for our empirical analysis:

Let policy stringency in city $c$ be $s_c\equiv s\left({\varphi }_c,r_T^c-r_G^c,{\tau }_c\right)$ with $\partial s/\partial {\varphi }_c>0$, $\partial s/\partial (r_T^c-r_G^c)>0$, $\partial s/\partial {\tau }_c>0$. Optimal allocation implies $\partial K_{Gc}/\partial s_c>0$. Let firm-level green innovation $G{I}_{it}$ and city-level carbon productivity $C{P}_c$ be observable outcomes; the model yields $\partial G{I}_{it}/\partial s_c>0$ and $\partial C{P}_c/\partial s_c>0$. Accordingly, we propose the following hypothesis.

H1 (Direct innovation and performance). Cities with higher green-credit policy stringency/execution (higher $s_c$) exhibit (i) greater firm-level green innovation and (ii) higher city-level environmental performance (e.g., carbon productivity).

For a high-emission firm $i$ in city $c$, relocation occurs when ${\mathcal{C}}_i(s_c)-{\mathcal{B}}_i(s_c)>{\kappa }_i+{\ell }_i$, where ${\mathcal{C}}_i$ and ${\mathcal{B}}_i$ denote compliance costs and benefits, and ${\kappa }_i,{\ell }_i$ are relocation and market-access costs. Define the firm-specific threshold ${\bar{s}}_i=\inf \{s_c:{\mathcal{C}}_i(s_c)-{\mathcal{B}}_i(s_c)>{\kappa }_i+{\ell }_i\}$; the share of firms with ${\bar{s}}_i<s_c$ rises in $s_c$. Accordingly, we propose the following hypothesis.

H2 (Threshold and displacement). When city stringency $s_c$ exceeds a critical threshold (operationalized empirically), the probability that high-emission activity exits the city increases, leading to observable pollution displacement toward laxer jurisdictions.

Let allocation quality (purity) be $Purit{y}_c\in [\mathrm{0,1}]$, the share of labeled green credit reaching verifiable environmental uses. For outcome $Y\in \{G{I}_{it},C{P}_c\}$, the innovation/abatement return to stringency satisfies $\frac{{\partial }^{2}Y}{\partial s_c\partial Purit{y}_c}>0$; purity multiplies the transmission from $s_c$ to $Y$. Accordingly, we propose the following hypothesis.

H3 (Purity moderation). The positive association between policy stringency/execution and green outcomes is stronger when green-credit allocation purity is higher.

Let $W$ be a row-normalized spatial-weight matrix over cities. Neighboring outcomes follow $Y=\rho WY+\beta s+\gamma Ws+\cdots$, where $Y\in \{GI,CP\}$ and $s$ stacks ${\{s_c\}}_c$. Knowledge diffusion and supply-linkages imply $\gamma >0$. When stringency gaps $\mathrm{\Delta }{s}_{cj}\equiv |s_c-s_j|$ exceed a threshold $\mathrm{\Delta }\bar{s}$, regulatory arbitrage yields $\gamma <0$ locally (net negative neighbor effect). Accordingly, we propose the following hypothesis.

H4 (Spatial responses). Green-credit stringency/execution generates spatial effects on neighboring cities’ green outcomes: (i) baseline positive spillovers $(\gamma >0)$; (ii) when cross-city stringency gaps are large, net neighbor effects turn negative due to displacement ($\gamma <0$ conditional on $\mathrm{\Delta }{s}_{cj}>\mathrm{\Delta }\bar{s}$).

These theoretical predictions provide a rigorous foundation for empirical testing using firm-level and regional-level data, enabling us to quantify the mechanisms through which green credit policy heterogeneity shapes innovation outcomes and environmental performance across China’s diverse regional landscape. A list of the variable symbols used in this model is provided in Table 1.

Table 1

Table 1. Notation and Symbol Definitions. A list of the variable symbols used in the theoretical model.

4 Methods

4.1 Data sources and sample construction

4.1.1 Temporal scope and policy context

Our analysis spans the period 2010–2022, a timeframe strategically chosen to capture the evolution of China’s green credit policy framework. This temporal window encompasses three critical phases: the initial implementation period following the China Banking Regulatory Commission’s 2012 Green Credit Guidelines, the intensification phase marked by the 2016 G20 Green Finance Study Group recommendations, and the institutionalization period under the 2021 carbon neutrality commitment. The starting year 2010 allows us to establish pre-policy baselines, while the 2022 endpoint captures the most recent policy developments under the updated Green Bond Endorsed Projects Catalogue.

This 13-year observation period provides sufficient temporal variation to identify both short-term adjustment effects and longer-term innovation responses to policy changes. Importantly, it encompasses the 2015 launch of China’s national green bond market and the 2017 establishment of mandatory environmental information disclosure requirements for listed companies, both crucial for constructing our Green Credit Purity measure.

4.1.2 Data integration and sample composition

To address the absence of official city-level green credit statistics, we construct a bottom-up city–year dataset for 2010–2022 covering 303 prefecture-level and above jurisdictions. The sample comprises 293 prefecture-level cities, four centrally administered municipalities¹, and six autonomous prefectures whose economic scale is comparable to prefecture-level cities². Including the municipalities is essential: they are national financial hubs and key carbon-emitting units, and they function as core nodes in regional networks; excluding them would induce selection bias and weaken spatial spillover identification. The construction proceeds in three steps: (i) base data collection, extracting firm-level green credit information from the Wind database, annual reports, and corporate social responsibility disclosures; (ii) geographic matching, assigning each firm’s green credit to its registered city to build city-level aggregates for all 303 units; and (iii) calibration, reconciling city aggregates with People’s Bank of China provincial totals within each year to ensure consistency between city-level sums and the published provincial series. This city-level series underpins the measurement of policy execution intensity (GCPI) and allocation quality (Purity) used in the empirical analysis.

Our empirical strategy relies on integrating multiple high-quality datasets to construct a comprehensive firm-city-year panel. Firm-level data, including green patent applications, financial statements, and corporate governance information, are sourced from the China Stock Market and Accounting Research (CSMAR) database and the China Research Data Service (CNRDS) platform. To ensure patent data accuracy, we cross-validate green patent classifications using the State Intellectual Property Office (SIPO) database and apply the World Intellectual Property Organization’s International Patent Classification Green Inventory.

City-level institutional and economic data are compiled from the China City Statistical Yearbook, supplemented by provincial statistical yearbooks and the Peking University Open Research Data Platform. For spatial analysis, we obtain standardized geographic coordinates from the National Geomatics Center of China, ensuring consistency in distance calculations across our 280 prefecture-level cities.

After implementing strict data quality controls—including the removal of firms with incomplete financial records, cities with missing institutional data, and observations with logical inconsistencies—our final sample comprises a balanced panel of 1,600 publicly listed companies across 280 prefecture-level cities, yielding 20,800 firm-year observations. We apply winsorization at the 1st and 99th percentiles to continuous variables, mitigating the influence of extreme outliers while preserving distributional information.

4.2 Variable construction and economic interpretation

4.2.1 Dependent variables: Capturing innovation outcomes

Firm-Level Green Innovation (GreenPatent): Following established conventions in the innovation literature (Liu et al., 2023a; Liu et al., 2023b), we measure firm-level green innovation using the annual count of green patent applications. This choice reflects both theoretical considerations—patent applications capture innovation intentions more immediately than grants, which involve processing lags—and practical advantages in terms of data completeness and temporal consistency.

Green patents are identified using the World Intellectual Property Organization’s Green List of International Patent Classifications, which covers eight technology areas: alternative energy production, energy conservation, transportation, waste management, agriculture, administrative regulatory, and nuclear power. To address potential concerns about patent quality heterogeneity, we construct an alternative measure using forward citations, employed in robustness analyses to ensure our findings are not driven by quantity-over-quality patent strategies.

Regional Green Innovation Performance (GTFP): At the city level, we assess environmental innovation performance using Green Total Factor Productivity, estimated via the Slack-Based Measure (SBM) model with undesirable outputs (Gao et al., 2025; Su et al., 2024). This methodology addresses the fundamental challenge of measuring productivity when production generates both desired economic outputs and undesired environmental externalities.

Our GTFP calculation treats regional GDP as the desirable output while incorporating industrial pollution emissions—specifically SO2 emissions and industrial wastewater discharge—as undesirable outputs that detract from productivity. Input variables include labor force, capital stock (constructed using the perpetual inventory method), and energy consumption. This specification aligns with our theoretical framework’s emphasis on the trade-offs between economic production and environmental quality embedded in regional welfare functions.

4.2.2 Core explanatory variables: Operationalizing theoretical concepts

Green Credit Policy Implementation Intensity (GCPI): Translating the theoretical concept of policy implementation heterogeneity into an empirical measure requires aggregating multiple observable policy dimensions. Our baseline GCPI index employs a multiplicative specification that captures both the scale and quality of local green credit implementation (Equation 12).

$GCP{I}_{ct}={\text{Green\hspace{0.17em}Credit\hspace{0.17em}Ratio}}_{ct}\times {\text{Sectoral\hspace{0.17em}Focus}}_{ct}\times \left(1-{\text{Default\hspace{0.17em}Rate}}_{ct}\right)(12)$

The Green Credit Ratio measures the proportion of green loans in total loans by financial institutions within each city, reflecting local implementation intensity. The Sectoral Focus component captures the ratio of green credit directed toward key environmental industries (renewable energy, energy efficiency, pollution control) relative to total green credit, proxying for policy targeting effectiveness. The Default Rate adjustment (one minus the green credit non-performing loan ratio) accounts for implementation quality, as higher default rates may signal insufficient screening or superficial compliance.

This multiplicative structure ensures that high GCPI scores require strength across all dimensions—cities cannot achieve high ratings through volume alone if implementation quality is poor. For robustness, we construct alternative indices using principal component analysis and the entropy method, ensuring our findings are not sensitive to specific aggregation techniques.

Green Credit Purity (Purity): Our novel Green Credit Purity measure operationalizes the theoretical distinction between genuine environmental investment and greenwashing behavior. At the firm level, Purity is calculated as Equation 13.

$Purit{y}_{it}=\frac{{\text{Verified\hspace{0.17em}Green\hspace{0.17em}Project\hspace{0.17em}Investment}}_{it}}{{\text{Total\hspace{0.17em}Green\hspace{0.17em}Credit\hspace{0.17em}Received}}_{it}}(13)$

We construct $Purit{y}_{it}$ from a purpose-built dataset of firm-year level project disclosures. For each listed firm that reports any green credit on its balance sheet, we collect all relevant narrative and tabular information from three sources: (i) annual reports, (ii) standalone corporate social responsibility (CSR) reports, and (iii) mandatory environmental information disclosures after 2017. For years in which multiple documents are available, we merge them into a single firm-year dossier. In principle, all firm-years with non-zero green credit are eligible for coding; firm-years without any mention of green credit or environmental investment are coded as zero Purity by construction.

We operationalize Verified Green Project Investment by applying a coding protocol that combines a positive list, a negative list, and explicit evidence thresholds. The positive list follows internationally recognized green taxonomies and China’s Green Bond Endorsed Project Catalogue, and includes, for example, (i) renewable energy generation and storage projects, (ii) energy-efficiency upgrades to production processes or buildings, (iii) end-of-pipe pollution abatement technologies, and (iv) investments in sustainable transport and supply chains. The negative list excludes (i) working capital injections without project-level detail, (ii) generic environmental management spending without quantifiable physical indicators, and (iii) refinancing of existing debt even when labeled as green. A project is coded as verified green only if the disclosure satisfies two conditions: (1) the description clearly maps to at least one item on the positive list (technology, asset, or process), and (2) the disclosure provides either a monetary amount or a physical indicator (e.g., installed capacity, energy saved, emissions reduced) that can be linked to the corresponding green credit line item. Ambiguous cases are coded as non-verified to avoid overstating environmental additionality. Full coding guidelines and examples of positive and negative cases are provided in Supplementary Material File.

To enhance reliability, we implement a double-blind coding procedure for a random subsample of firm-years. Two trained research assistants independently code all projects in the same firm-year dossier using the protocol described above, without access to each other’s decisions. We then compute Cohen’s kappa to assess inter-coder agreement on the binary classification of verified green versus non-verified projects and reconcile discrepancies through a joint review with the authors. For the remaining sample, single-coder assignments are subject to ex-post audits: we randomly draw additional firm-years for re-coding and document the share of cases where initial classifications are confirmed or require only minor clarifications. These procedures, together with cross-checks against financial statement line items and environmental disclosure totals, ensure that the Purity measure is robust to subjective coder judgment. Summary statistics on inter-coder reliability and audit pass rates are reported in Section 5.1, while Supplementary Material File provides a detailed coding manual and examples.

This labor-intensive construction process reflects the fundamental challenge of distinguishing genuine environmental investment from symbolic compliance. Higher Purity values indicate firms that channel a greater proportion of their green credit toward verifiable environmental projects, while lower values suggest potential misallocation or strategic relabeling of conventional investments.

4.2.3 Control variables: Isolating causal channels

Our control variable selection is guided by established theoretical frameworks linking firm characteristics, regional attributes, and innovation outcomes. This comprehensive approach helps isolate the causal channels through which green credit policies influence innovation while controlling for alternative explanations.

Firm-Level Controls: We include firm size (logarithm of total assets) to account for scale economies in innovation; financial leverage (debt-to-assets ratio) to control for financial constraints that might interact with credit policies; R&D intensity (R&D expenditure as a percentage of operating revenue) to separate policy-induced innovation from baseline research activities; market concentration (firm-level Herfindahl-Hirschman Index) to account for competitive pressures; and state ownership (binary indicator) to control for differential access to policy support and regulatory scrutiny.

City-Level Controls: Regional controls include per capita GDP (economic development level), industrial structure (secondary industry share of GDP), government science and technology expenditure (fiscal commitment to innovation), university presence (logarithm of higher education institutions), human capital (proportion of population with college education), and environmental quality (annual average PM2.5 concentration). These variables control for regional innovation capacity, economic structure, and environmental pressures that might confound the relationship between green credit policies and innovation outcomes.

4.2.4 Spatial weight matrix construction

Spatial econometric analysis requires careful specification of the connectivity structure among cities. We construct several spatial weight matrices and treat them differently in the main analysis and robustness checks to ensure our findings are not sensitive to specific assumptions about inter-city relationships.

Our baseline geographic weight matrix $(W_{\text{Geo}})$ is a row-standardized inverse squared distance matrix, where $w_{ij}=1/d_{ij}^2$ (and $w_{ii}=0$) and $d_{ij}$ denotes the great-circle distance between city centers. This specification assumes that spatial interactions decay rapidly with distance, consistent with localized knowledge spillovers and administrative coordination patterns documented in the innovation literature. In sensitivity analyses, we also examine a linear distance decay function $(w_{ij}=1/d_{ij})$ to assess whether our results depend on the assumed rate of spatial attenuation.

To capture economic proximity, we construct an economic distance matrix $(W_{\text{Eco}})$ where $w_{ij}=1/|GDP\text{\_}p{c}_i-GDP\text{\_}p{c}_j|$ for $i\ne j$, again row-standardized. This matrix reflects the notion that economically similar cities may experience stronger policy spillovers through factor mobility and regulatory competition, regardless of geographic proximity. Cities at similar development stages face comparable policy challenges and often engage in strategic interaction, making this an important alternative conceptualization of spatial connectivity.

For robustness, we further estimate models using (i) a $k$-nearest-neighbor matrix $(W_{\text{KNN}})$ based on geographic distance, which links each city to its five closest neighbors, capturing the notion that spillovers may operate primarily through direct adjacency rather than distance-weighted connections; (ii) a Queen contiguity matrix $(W_{\text{Queen}})$ that connects administratively adjacent cities, reflecting formal institutional channels for policy coordination; and (iii) a geographic-industry nested weight matrix that combines spatial proximity with sectoral similarity, allowing for industry-specific spillover patterns that may vary across space. All weight matrices are row-standardized to ensure comparability of spatial parameters across specifications.

The main results are reported for the spatial Durbin model using $W_{\text{Geo}}$, while Section 5 and Supplementary Material File demonstrate that our conclusions regarding positive spillovers and threshold displacement effects are robust to alternative spatial weight matrices and model specifications.

4.2.5 Tests for spatial dependence and model choice

Before estimating spatial models, we first compute Moran’s I statistics for city-level GTFP and for the residuals from non-spatial panel regressions, using $W_{\text{Geo}}$ as the connectivity matrix. The results, reported in Supplementary Material File, reject the null hypothesis of no spatial autocorrelation at conventional significance levels, with Moran’s I values ranging from 0.34 to 0.47 across different years. We then perform Lagrange Multiplier tests for spatial lag and spatial error dependence, including both standard and robust versions of these tests. The robust LM statistics are significant for both forms of dependence (robust LM-lag: 23.47, $p<0.001$; robust LM-error: 18.92, $p<0.001$), which, following the recommendations in (Elhorst, 2014), motivates the use of the SDM as a flexible encompassing model that accommodates both types of spatial dependence. These diagnostics justify our choice of the SDM as the baseline specification while using SLM and SEM as complementary robustness checks to ensure our findings are not artifacts of particular spatial model assumptions.

4.2.6 Instrumental variable strategy

Endogeneity concerns arise because regional innovation performance might influence local governments’ propensity to implement strict green credit policies, creating reverse causality bias. We address this through a two-stage least squares approach using carefully constructed instrumental variables.

Our primary instrument exploits historical variation in regional financial development. Specifically, we use the 1998 proportion of environmental loans (a precursor to formal green credit) at the city level, interacted with national green finance policy intensity in each year. This instrument leverages the path-dependent nature of financial sector development while ensuring variation across both cities and time.

The exclusion restriction requires that historical banking patterns affect current innovation only through their influence on contemporary green credit policy implementation—a plausible assumption given the institutional persistence of financial sector specialization and the regulatory nature of green credit requirements.

Our secondary instrument combines each city’s distance to the nearest national green technology exchange center with annual national green bond issuance volumes. This captures exogenous variation in local green finance policy implementation driven by proximity to specialized financial infrastructure and national policy cycles.

We conduct standard diagnostic tests including first-stage F-statistics to verify instrument relevance and Sargan-Hansen tests for over-identification restrictions when using multiple instruments.

This comprehensive econometric strategy enables us to trace the causal pathways from heterogeneous policy implementation through firm-level innovation decisions to regional environmental outcomes, while accounting for the spatial interdependencies that characterize China’s integrated economic geography.

4.3 Econometric strategy and model specification

4.3.1 Baseline panel regression framework

Our primary empirical strategy builds directly on the theoretical model’s predictions through a series of increasingly sophisticated panel regression specifications. The baseline firm-level model tests Hypotheses H1 and H3:

$\begin{array}{cc}\hfill GreenPaten{t}_{\mathit{ict}}=\alpha & +{\beta }_{1}GCP{I}_{ct}+{\beta }_2\left(GCP{I}_{ct}\times Purit{y}_{it}\right)+{\beta }_{3}Purit{y}_{it}\hfill \\ \hfill & +\delta X_{it}+\theta Z_{ct}+{\mu }_i+{\lambda }_t+{\varphi }_c+{ϵ}_{\mathit{ict}}\hfill \end{array}(14)$

This specification incorporates three levels of fixed effects that address distinct identification concerns. Firm fixed effects $({\mu }_i)$ control for time-invariant firm characteristics that might correlate with both green credit access and innovation propensity. Year fixed effects $({\lambda }_t)$ capture macroeconomic shocks, national policy changes, and temporal trends affecting all firms simultaneously. City fixed effects $({\varphi }_c)$ eliminate time-invariant regional characteristics such as geographic advantages, institutional culture, or natural resource endowments.

The coefficient ${\beta }_1$ identifies the direct effect of policy implementation intensity on innovation, testing our core hypothesis that stricter local enforcement enhances green innovation. The interaction coefficient ${\beta }_2$ captures the moderating role of Green Credit Purity, testing whether genuine environmental investment amplifies policy effectiveness. We expect both coefficients to be positive and statistically significant. Standard errors are clustered at the city level to account for within-city correlation of disturbances across firms and time. We verify the robustness of count-outcome inference using negative binomial and Poisson fixed-effects specifications, which address potential overdispersion in patent data while maintaining the fixed-effects structure.

4.3.2 Spatial econometric analysis

To test Hypothesis H4 regarding spatial spillover effects and potential pollution displacement, we employ a Spatial Durbin Model (SDM) at the city level (Equation 15).

$\begin{array}{cc}\hfill GTF{P}_{ct}=\rho WGTF{P}_{ct}& +\beta GCP{I}_{ct}+\theta WGCP{I}_{ct}\hfill \\ \hfill & +\gamma Z_{ct}+\xi W{Z}_{ct}+{\mu }_c+{\lambda }_t+{\epsilon }_{ct}\hfill \end{array}(15)$

This specification allows for three types of spatial effects. The spatial lag parameter $\rho$ captures spatial autocorrelation in green productivity, testing whether neighboring cities’ environmental performance directly influences local outcomes through demonstration effects or competitive pressures. The direct effect $\beta$ measures how local GCPI implementation affects local GTFP, while the spatial lag coefficient $\theta$ identifies spillover effects from neighboring cities’ policies.

Before estimating the SDM, we conduct diagnostic tests to verify the appropriateness of spatial modeling. Moran’s I statistics test for spatial autocorrelation in the dependent variable and in the residuals of the non-spatial model. Lagrange Multiplier (LM) tests for spatial lag and spatial error, along with their robust counterparts, guide the selection between spatial lag models (SLM), spatial error models (SEM), and the more general SDM specification. The SDM nests both SLM and SEM as special cases and provides a flexible framework for identifying whether spillovers operate through the dependent variable, the independent variables, or both channels simultaneously.

A positive $\theta$ would suggest beneficial spillovers—perhaps through technology diffusion or regulatory learning—while a negative coefficient might indicate pollution displacement or regulatory competition effects consistent with the pollution-haven hypothesis. However, in spatial autoregressive models the regression coefficients ($\beta$, $\theta$) do not directly correspond to marginal effects because of feedback loops across the spatial network. Following LeSage and Pace (2009), we therefore compute scalar summary measures of the impact of GCPI on GTFP: the average direct effect (ADE), average indirect or spillover effect (AIE), and average total effect (ATE).

Let $S_{\text{GCPI}}={(I-\rho W)}^{-1}(\beta I+\theta W)$ denote the $N\times N$ impact matrix for GCPI, where $N$ is the number of cities and $I$ is the identity matrix. The ADE is defined as the average of the diagonal elements of $S_{\text{GCPI}}$, capturing the average own-city effect of GCPI on local GTFP while accounting for feedback through the spatial network. The AIE is the average row sum of off-diagonal elements, measuring the average effect of all other cities’ GCPI on a given city’s GTFP. The ATE is simply the sum of ADE and AIE, representing the total marginal effect of a uniform change in GCPI across all cities.

We obtain standard errors for ADE, AIE, and ATE using the delta method based on the variance–covariance matrix of $(\rho ,\beta ,\theta )$, and construct 95% confidence intervals accordingly. This approach accounts for the nonlinear transformation from regression coefficients to summary effects while maintaining computational efficiency. In Section 5.3, we report these effects and their standard errors for the baseline SDM and for alternative spatial specifications—including different weight matrices (economic distance, geographic-industry nested) and robustness checks using spatial two-stage least squares (S2SLS) estimation—providing an intuitive interpretation of the magnitude of spatial spillovers.

4.3.3 Instrumental variable strategy

Endogeneity concerns arise because regional innovation performance might influence local governments’ propensity to implement strict green credit policies, creating reverse causality bias. Additionally, unobserved city-level shocks could simultaneously affect both policy implementation and innovation outcomes, generating omitted variable bias. We address these concerns using a two-stage least squares (2SLS) strategy in which city-level GCPI is treated as endogenous. The first-stage equation takes the form:

$GCP{I}_{ct}={\pi }_0+{\pi }_1{Z}_{ct}^{\left(1\right)}+{\pi }_2{Z}_{ct}^{\left(2\right)}+\eta X_{ct}+{\mu }_c+{\lambda }_t+u_{ct},(16)$

where $Z_{ct}^{(1)}$ and $Z_{ct}^{(2)}$ denote our external instruments and $X_{ct}$ is the vector of city-level controls. The second stage estimates (Equation 14) with GCPI replaced by its predicted value from Equation 16.

4.3.3.1 Instrument construction and first-stage specification

Our primary instrument $(Z_{ct}^{(1)})$ exploits historical variation in regional financial development. Specifically, we construct the interaction between the 1998 proportion of environmental loans (a precursor to formal green credit) at the city level and national green finance policy intensity in each year. This instrument leverages the path-dependent nature of financial sector development: cities with greater environmental lending capacity in 1998—prior to any formal green credit guidelines—developed specialized expertise, institutional relationships, and monitoring infrastructure that facilitated more aggressive implementation of subsequent national green credit policies. The time variation comes from changes in national policy intensity, which affects all cities but has heterogeneous impacts depending on their historical financial structure.

The exclusion restriction requires that historical banking patterns affect current innovation only through their influence on contemporary green credit policy implementation. The 1998 measure predates both the 2007 Green Credit Policy and the 2012 Green Credit Guidelines by at least 9 years, reducing concerns about anticipatory behavior or reverse causality from current innovation outcomes.

Our secondary instrument $(Z_{ct}^{(2)})$ combines each city’s distance to the nearest national green technology exchange center with annual national green bond issuance volumes. This instrument captures exogenous variation in local green finance policy implementation driven by proximity to specialized financial infrastructure interacted with national policy cycles. The geographic component is time-invariant and captures quasi-random variation in access costs, while the national bond issuance component reflects policy intensification that is determined at the central government level and thus plausibly exogenous to individual city innovation outcomes.

4.3.3.2 Identification assumptions and potential alternative pathways

The key identifying assumption is that historical environmental lending patterns affect current innovation only through their impact on contemporary green credit policy implementation, rather than through alternative pathways such as persistent differences in human capital or industrial structure. A potential concern is that cities with higher environmental lending in 1998 may have simultaneously developed stronger human capital bases or more environmentally-oriented industrial structures, which could directly foster current innovation independently of GCPI. We take three steps to make our exclusion restriction more credible.

First, we explicitly control for pre-sample city characteristics that could mediate the link between historical banking patterns and current innovation. In the second stage, we augment the baseline specification with lagged measures of human capital (tertiary-education share circa 2000) and industrial structure (secondary-industry share circa 2000), as well as pre-policy innovation capacity (average green patenting intensity during 2007–2009, immediately before our sample period). These controls absorb long-run differences in skills, industrial composition, and innovation orientation that might otherwise provide a direct channel from 1998 banking patterns to current innovation. The pre-sample timing of these controls ensures they are not affected by post-2010 GCPI implementation, avoiding post-treatment bias while capturing relevant historical characteristics.

Second, the 1998 environmental lending measure reflects a narrow regulatory category—loans for pollution control and basic environmental protection—that was limited in scale and scope. This lending category was distinct from general industrial or commercial lending and represented less than 2% of total bank credit at the time. Its primary function was compliance with rudimentary environmental regulations rather than support for technological innovation. We therefore argue that any persistent effects of this lending on current innovation capacity would be modest and largely captured by our controls for pre-sample innovation activity. The instrument’s power derives not from direct effects of 1998 lending on 2010–2022 innovation, but from how 1998 lending capacity shaped cities’ ability to implement aggressive GCPI after formal green credit policies were introduced.

Third, we implement several falsification tests to probe whether the instrument operates through the hypothesized channel. If historical environmental lending had strong direct effects on innovation independent of GCPI, we would expect to find significant associations in periods before formal green credit policies took effect. We test this by re-estimating the reduced-form relationship between the 1998 instrument and innovation outcomes using only pre-2010 data (2007–2009), when national green credit guidelines had not yet been formalized and local GCPI variation was minimal. Finding no systematic evidence of pre-policy effects would support the interpretation that the instrument primarily operates through post-2010 GCPI implementation rather than through persistent direct effects on innovation capacity.

4.3.3.3 Multiple instruments and diagnostic checks

We diversify the instrument set by combining the historical environmental lending instrument with our secondary instrument based on distance to green technology exchange centers interacted with national bond issuance. The former captures path-dependent institutional capacity in environmental finance, while the latter reflects quasi-geographic access to specialized green finance infrastructure combined with national policy cycles. These instruments exploit distinct sources of exogenous variation, allowing us to perform over-identification tests (Sargan-Hansen J-test) that probe the consistency of the exclusion restrictions. If both instruments are valid, they should yield similar second-stage estimates; rejection of the over-identification test would suggest that at least one instrument violates the exclusion restriction.

We report a comprehensive set of diagnostic statistics to assess instrument validity and strength. First-stage F-statistics test the joint significance of excluded instruments in predicting GCPI, with values above 10 suggesting adequate strength under conventional rules of thumb. We also report Kleibergen-Paap rk Wald F-statistics, which are robust to heteroskedasticity and clustering, along with critical values from Stock-Yogo weak instrument tests. To address concerns about inference under weak instruments, we compute Anderson-Rubin confidence intervals and conduct Wu-Hausman endogeneity tests, which are robust to weak instruments and provide valid inference even when first-stage F-statistics are moderate. Additionally, we implement conditional likelihood ratio (CLR) tests following Moreira (2003), which provide exact finite-sample inference regardless of instrument strength.

Taken together, these controls, placebo exercises, and diagnostic tests strengthen the case that our instruments shift current innovation primarily through their impact on GCPI, rather than through omitted long-run characteristics of cities. The multiple-instrument approach with diverse identification strategies provides robustness against misspecification of any single instrument, while the comprehensive diagnostic battery ensures that our causal inference does not hinge on borderline instrument validity or strength.

This comprehensive econometric strategy enables us to trace the causal pathways from heterogeneous policy implementation through firm-level innovation decisions to regional environmental outcomes, while accounting for the spatial interdependencies that characterize China’s integrated economic geography. The progressive identification approach—from fixed effects to spatial models to instrumental variables—provides multiple angles on the core relationships, with each specification addressing distinct threats to identification while maintaining consistency in the estimated effects.

5 Results

5.1 Descriptive statistics and preliminary evidence

This section presents comprehensive descriptive evidence from our matched firm-city panel dataset, spanning 1,600 publicly listed companies across 280 prefecture-level cities over the period 2010–2022. Our sample construction yields 20,800 firm-year observations, providing substantial variation across both cross-sectional and temporal dimensions that enables robust identification of green credit policy effects. The dataset captures a critical period in China’s green finance evolution, encompassing the initial policy experimentation phase (2010–2014), the institutional development period (2015–2018), and the recent intensification under carbon neutrality commitments (2019–2022).

5.1.1 Distributions and spatial patterns of key variables

Figure 2 presents the empirical distributions of our two core policy measures: Green Credit Policy Implementation Intensity (GCPI) and Green Credit Purity. Panel A reveals that GCPI is broadly distributed across cities, with substantial density throughout the [0.1, 0.9] range. The distribution exhibits a mean of 0.55 (marked by the dashed vertical line) and displays moderate right skewness, indicating that while most cities cluster around moderate implementation levels, a non-trivial fraction achieves high-intensity execution. Notably, the distribution spans nearly the full theoretical range, validating our core research premise that green credit policies exhibit meaningful heterogeneity in local implementation. The vertical dotted line marks the estimated threshold value of 0.416, below which we observe strong innovation responses and limited firm relocation, and above which the relationship attenuates—a pattern we examine formally in Section 5.3.3.

Figure 2

Figure 2. Empirical Distributions of Policy Implementation and Allocation Quality. (A) Shows the distribution of Green Credit Policy Implementation Intensity (GCPI) across 3,640 city-year observations (2010–2022). The dashed red line indicates the sample mean (0.55), while the dotted green line marks the estimated threshold (0.416) beyond which innovation responses attenuate and firm relocation probabilities increase. (B) Presents the distribution of Green Credit Purity across 20,800 firm-year observations. The dashed red line indicates the sample mean (0.72). Both panels overlay kernel density estimates on histogram distributions. The substantial dispersion in both measures validates our focus on heterogeneity in policy implementation intensity and allocation quality as key determinants of green credit effectiveness.

Panel B documents the distribution of Green Credit Purity, our novel measure of allocation quality. The distribution centers around 0.72, indicating that approximately 72% of green credit flows to genuinely environmental projects on average. However, the substantial dispersion—with values ranging from 0.10 to 1.00 and a standard deviation of 0.20—reveals considerable heterogeneity in allocation quality across firms. The long left tail extending to low purity values provides quantitative evidence consistent with greenwashing behavior in some cases, where firms may strategically relabel conventional projects to access preferential green credit terms. This distributional heterogeneity motivates our theoretical emphasis on allocation quality as a critical moderating factor in policy effectiveness.

Figure 3 maps the spatial distribution of average GCPI across Chinese cities during our sample period. The geographic pattern reveals systematic regional clustering that aligns with economic development levels and administrative capacity differences. Eastern coastal corridors—including the Beijing-Tianjin-Hebei cluster, the Yangtze River Delta (Shanghai-Jiangsu-Zhejiang), and the Pearl River Delta (Guangdong)—clearly stand out as high-implementation zones, with average GCPI values ranging from 0.60 to 0.68. These regions benefit from more developed financial sectors, stronger institutional capacity, and greater political priority for environmental governance. Central provinces exhibit moderate implementation intensity (0.50–0.55), while inland and western regions lag behind with GCPI values below 0.48.

Figure 3

Figure 3. Geographic Distribution of Green Credit Policy Implementation Intensity. The map displays city-level average GCPI over the period 2010–2022, with darker green shades indicating higher implementation intensity. Eastern coastal regions exhibit consistently higher GCPI (0.60–0.68), central regions show moderate levels (0.50–0.55), and western regions display lower implementation intensity (\lt 0.48). The pronounced spatial clustering motivates our spatial Durbin model analysis in Section 5.3. Scale bar indicates approximate distances; north arrow shows orientation. Province boundaries are simplified for visual clarity.

This spatial gradient has important implications for our empirical strategy. First, the pronounced geographic clustering provides strong motivation for spatial econometric analysis, as environmental policies and outcomes in neighboring cities are unlikely to be independent. Second, the systematic correlation between GCPI and regional development levels underscores the importance of our fixed effects and instrumental variable approaches to isolate causal effects. Third, the wide dispersion in implementation intensity—even after accounting for broad regional patterns—suggests substantial within-region variation that our city-level analysis can exploit for identification.

Figure 4 traces the temporal evolution of both GCPI and Green Credit Purity from 2010 through 2022, overlaid with three distinct policy phases and key institutional milestones. The time series reveals three notable patterns. First, GCPI exhibits a marked upward trend, rising from an average of 0.48 in the early period (2010–2015) to 0.62 in the later period (2016–2022)—a 29% increase that reflects the intensification of green credit policy implementation nationwide. The steepest acceleration occurs after 2016, coinciding with China’s chairmanship of the G20 and the establishment of the Green Finance Study Group, which elevated green finance to a national strategic priority.

Figure 4

Figure 4. Temporal Evolution of Green Credit Policy Implementation and Allocation Quality. The figure displays annual national averages of GCPI (blue line, left axis) and Green Credit Purity (green line, right axis) from 2010 to 2022. Shaded backgrounds delineate three policy phases: Phase I (2010–2014, light gray) represents initial policy experimentation; Phase II (2015–2018, light blue) marks institutional development and G20-driven intensification; Phase III (2019–2022, light green) reflects institutionalization under carbon neutrality commitments. Vertical dotted lines indicate key policy milestones: 2012 Green Credit Guidelines, 2016 G20 Green Finance Initiative, and 2021 Carbon Neutrality Commitment. Both metrics demonstrate sustained upward trends with accelerated growth post-2016, suggesting that national policy developments translate into measurable improvements in local implementation intensity and allocation quality.

Second, Green Credit Purity demonstrates parallel improvement, increasing from 0.68 to 0.75 over the same periods. This convergent evolution suggests that policy refinements enhanced not only the scale of green credit provision but also the quality of credit allocation, potentially through improved monitoring systems, clearer taxonomies for green projects, and stricter enforcement of environmental disclosure requirements introduced in 2017. The co-movement of implementation intensity and allocation quality supports our theoretical framework’s emphasis on their joint determination of policy effectiveness.

Third, both series exhibit inflection points around key policy milestones. The 2012 Green Credit Guidelines mark the transition from experimental to formal policy status, while the 2016 G20 initiatives and 2021 carbon neutrality commitment correspond to accelerated growth in both GCPI and Purity. These temporal patterns provide prima facie evidence that national policy developments translate into measurable changes in local implementation, while also suggesting potential heterogeneity in policy responses across the three phases—a dimension we explore in robustness checks.

5.1.2 Enhanced descriptive statistics with distributional measures

Table 2 extends conventional descriptive statistics by incorporating distributional measures that better characterize the heterogeneity in our key variables. Beyond standard means and standard deviations, we report medians and interquartile ranges (25th and 75th percentiles) for core variables, which prove particularly informative given the non-normal distributions evident in Figure 2.

Table 2

Table 2. Enhanced descriptive statistics with distributional measures (2010–2022).

Panel A presents firm-level variables. The median green patent count (1.00) falls substantially below the mean (2.54), confirming right-skewed distribution and indicating that green innovation is concentrated among a subset of highly active innovators. The interquartile range (0.00–3.00) reveals that 75% of firm-year observations involve three or fewer green patents, while the maximum of 60 patents demonstrates extreme heterogeneity. This pattern validates our use of count data models and logarithmic transformations in robustness checks.

Green Credit Purity exhibits less skewness, with median (0.75) close to the mean (0.72) and a relatively symmetric distribution around the central tendency. The interquartile range (0.58–0.88) indicates that most firms allocate between 58% and 88% of green credit to verified environmental projects, though the minimum value of 0.10 confirms the presence of severe greenwashing cases. Firm size, R&D intensity, and leverage display expected patterns, with medians generally below means, reflecting right skewness typical of firm-level financial variables.

Panel B reports city-level statistics. GCPI’s median (0.56) closely approximates its mean (0.55), consistent with the roughly symmetric distribution in Figure 2, Panel A. The interquartile range (0.42–0.68) spans 0.26 units—nearly 1.5 standard deviations—underscoring substantial cross-city variation in implementation intensity. Regional Green Total Factor Productivity (GTFP) exhibits modest dispersion, with a tight interquartile range (0.92–1.12) around the median of 1.01, suggesting that most cities operate at similar environmental efficiency levels despite marked differences in policy implementation.

For key variables, we conduct formal tests of differences in means between cities in the bottom and top terciles of GCPI distribution; these results appear in Table 3 and demonstrate statistically significant differences across nearly all dimensions, motivating our subsequent causal analysis.

Table 3

Table 3. Group differences across GCPI terciles: Formal statistical tests.

5.1.3 Formal tests of group differences across GCPI terciles

Table 3 provides formal statistical tests comparing cities in the bottom, middle, and top terciles of the GCPI distribution. This analysis serves two purposes: it quantifies the magnitude of observable differences between high- and low-implementation cities, and it motivates our fixed effects and instrumental variable strategies by demonstrating that high-GCPI cities differ along multiple dimensions beyond policy implementation alone.

The results reveal stark and statistically significant differences across terciles. Cities in the top GCPI tercile (mean GCPI = 0.72) exhibit 78% higher average firm-level green patenting (3.21 vs. 1.80 patents per firm-year) compared to bottom-tercile cities (mean GCPI = 0.38). This difference is significant at the 1% level in both parametric t-tests $(p<0.001)$ and non-parametric Wilcoxon rank-sum tests $(p<0.001)$, confirming robustness to distributional assumptions. Similarly, Green Credit Purity averages 0.81 in high-GCPI cities versus 0.64 in low-GCPI cities—a 27% difference that is highly significant $(p<0.001)$, suggesting that implementation intensity and allocation quality tend to move together.

Regional environmental productivity (GTFP) displays a 22% differential (1.13 vs. 0.93), indicating that high-implementation cities achieve substantially better environmental efficiency. However, high-GCPI cities also differ markedly in economic development levels: per capita GDP is 94% higher in the top tercile (13.42 vs. 6.92 in 10,000 RMB), while R&D intensity is 71% higher (0.045 vs. 0.026). These concurrent differences in economic fundamentals and innovation capacity underscore the potential for omitted variable bias in simple correlational analyses, justifying our emphasis on fixed effects specifications that exploit within-city temporal variation and instrumental variable approaches that isolate exogenous policy variation.

Interestingly, firm size exhibits only modest differences across terciles (7% gap), and leverage shows no significant difference, suggesting that firm financial characteristics are more evenly distributed across regions than innovation outcomes or economic development levels. Industry structure differences are moderate but significant: high-GCPI cities have slightly lower secondary industry shares (0.44 vs. 0.51), potentially reflecting structural shifts toward services and advanced manufacturing in more developed regions.

The magnitude and statistical significance of these cross-sectional differences establish a crucial empirical foundation. They confirm that the GCPI measure captures meaningful variation in local policy environments, as evidenced by its strong associations with innovation outcomes. Simultaneously, they highlight the identification challenge: observed correlations between GCPI and innovation could partly reflect unobserved city characteristics rather than causal policy effects. Our subsequent analysis addresses this challenge through progressively demanding identification strategies—fixed effects, spatial models, and instrumental variables—that control for confounding factors and isolate causal mechanisms.

5.1.4 Correlation structure and preliminary relationships

Table 4 examines bivariate correlations among key variables, providing preliminary evidence supporting our theoretical framework while highlighting potential multicollinearity concerns for subsequent regression analysis. We compute correlations using the full panel dataset and assess statistical significance with city-clustered robust standard errors to account for within-city correlation of observations.

Table 4

Table 4. Correlation matrix: Key variables with statistical significance.

The positive correlation between GCPI and GreenPatent (0.25***) offers initial support for Hypothesis H1, suggesting that stricter local green credit implementation associates with enhanced firm-level innovation. The moderate correlation between Purity and GreenPatent (0.27***) provides early evidence for Hypothesis H3, indicating that genuine green credit allocation may amplify innovation outcomes. Importantly, the similar magnitudes of these two correlations (0.25 vs. 0.27) suggest that both policy implementation intensity and allocation quality contribute meaningfully to innovation outcomes, motivating our interaction specification in the baseline regression framework.

The strong correlation between R&D intensity and green patents (0.32***) confirms that our innovation measures capture genuine research activities rather than strategic patenting behavior—firms with higher R&D expenditures predictably generate more green patents. The positive association between GCPI and GTFP (0.31***) suggests that policy implementation benefits extend beyond firm-level outcomes to encompass broader regional environmental productivity, a pattern we explore formally through spatial analysis in Section 5.3.

Two findings deserve particular attention. First, GCPI exhibits modest positive correlations with both Purity (0.24***) and R&D intensity (0.15***), indicating that high-implementation cities tend to have better credit allocation quality and more innovative firms. This co-movement could reflect complementary policy dimensions (cities with strong implementation also enforce quality standards), selection effects (innovative firms locate in high-GCPI regions), or common determinants (economic development drives both policy implementation and innovation capacity). Our identification strategy must disentangle these mechanisms. Second, all pairwise correlations remain below 0.35, indicating that multicollinearity concerns should not severely compromise regression inference—a crucial precondition for interpreting individual coefficients in our multi-variable specifications.

The moderate correlation between GCPI and CityGDPpc (not shown but implied by Table 3) warrants discussion. While economic development and policy implementation correlate positively, the relationship is far from deterministic (correlation $\approx$ 0.40), suggesting substantial within-development-level variation in implementation intensity. This heterogeneity—combined with our longitudinal structure—enables identification even after controlling for time-varying economic conditions and employing city fixed effects.

5.1.5 Sample representativeness and data quality validation

Our sample exhibits strong representativeness across multiple dimensions critical for external validity. The 1,600 firms span all major industry classifications following the China Securities Regulatory Commission taxonomy, with balanced representation across manufacturing (45%), services (32%), high-tech sectors (15%), and other industries (8%). This industrial diversity ensures that our findings are not driven by sector-specific dynamics but rather reflect broader patterns of how green credit policies shape innovation across China’s diverse economic structure.

Geographically, our 280 cities include all 31 provincial capitals plus 249 prefecture-level cities, covering approximately 78% of China’s urban GDP and 65% of its population as of 2020. The sample encompasses all major urban agglomerations—the Beijing-Tianjin-Hebei region, Yangtze River Delta, Pearl River Delta, Chengdu-Chongqing cluster, and Central Plains urban group—ensuring representation of diverse development trajectories and policy contexts. The temporal coverage (2010–2022) captures three distinct policy regimes, allowing us to test whether policy effects vary across institutional contexts while providing sufficient within-regime variation for identification.

Data quality indicators support the reliability of our key measures. For Green Patent classifications, we conducted manual validation exercises on a random sample of 500 patents, achieving 94% inter-coder agreement using the WIPO Green Inventory criteria. Our Green Credit Purity measure demonstrates strong convergent validity: it correlates 0.83 with an alternative firm-level environmental investment indicator constructed from mandatory Corporate Social Responsibility reports, and 0.76 with third-party environmental ratings from the China Securities Index (CSI) Green Rating system for the subset of rated firms.

The GCPI index exhibits expected correlations with independent measures of local environmental policy stringency, including city-level Environmental Protection Bureau enforcement actions (correlation = 0.68), environmental tax revenues (correlation = 0.54), and participation in national green finance pilot zones (point-biserial correlation = 0.71). These validation exercises confirm that GCPI captures genuine variation in policy implementation rather than measurement error or idiosyncratic local definitions of “green credit.”

Finally, attrition analysis reveals minimal systematic sample selection. Comparing firms that remain in our sample throughout 2010–2022 with those that exit (due to delisting, mergers, or data gaps), we find no significant differences in pre-sample green patenting (2007–2009 average: 2.1 vs. 1.9 patents, $p=0.64$) or firm size ($t=1.23$, $p=0.22$), suggesting that sample composition remains stable and representative over time.

These descriptive patterns establish three key empirical foundations for subsequent causal analysis. First, the substantial dispersion in GCPI, Purity, and innovation outcomes—combined with theoretically consistent correlation patterns—confirms that green credit policy heterogeneity creates meaningful variation in both policy environments and firm responses. Second, the systematic geographic and temporal patterns validate our spatial econometric approach and motivate temporal heterogeneity analysis. Third, the stark differences across GCPI terciles underscore the identification challenge: while high-implementation cities exhibit superior innovation performance, they also differ substantially along other dimensions. Our fixed effects, spatial, and instrumental variable specifications in subsequent sections address this challenge by progressively isolating causal mechanisms from cross-sectional heterogeneity.

5.2 Baseline results and core hypotheses testing

This section presents our core empirical findings through a progressive identification strategy that systematically tests the theoretical predictions derived in Section 3. We organize the analysis in three modules that build upon each other: firm-level innovation responses to green credit policy implementation (Section 5.2.1), city-level environmental productivity effects (Section 5.2.2), and industry-specific heterogeneity patterns (Section 5.2.3). Before presenting results, we establish the validity of our econometric approach through comprehensive diagnostic testing.

5.2.1 Model diagnostics and specification validation

Table 5 reports key diagnostic tests for our baseline panel specifications, validating the methodological choices outlined in Section 4.3. For both firm-level and city-level models, Hausman tests strongly reject the random effects specification in favor of fixed effects (${\chi }^2=47.3$, $p<0.01$ for firm-level; ${\chi }^2=28.6$, $p<0.01$ for city-level), confirming that unobserved heterogeneity correlates with explanatory variables and justifying our fixed effects approach.

Table 5

Table 5. Model diagnostics for baseline panel specifications.

Tests for heteroskedasticity and serial correlation support our use of clustered robust standard errors. Breusch-Pagan tests decisively reject homoskedasticity in both specifications $(p<0.001)$, while Wooldridge tests detect first-order serial correlation within cities ($F=12.4$, $p=0.002$ for firm-level; $F=8.7$, $p=0.008$ for city-level). These findings motivate our adoption of city-clustered robust standard errors, which accommodate arbitrary heteroskedasticity and within-city serial correlation while maintaining asymptotic validity under general dependence structures.

Multicollinearity diagnostics provide reassurance regarding coefficient identification. Maximum variance inflation factors remain well below conventional thresholds (3.24 for firm-level, 2.89 for city-level), indicating that correlation among explanatory variables does not compromise estimation precision. The mean VIF of 1.86 across all specifications suggests that our covariates capture distinct dimensions of the data-generating process rather than redundant information.

5.2.2 Firm-level innovation responses: Testing H1 and H3

Table 6 presents our primary results examining how Green Credit Policy Implementation Intensity (GCPI) affects firm-level green patent applications, directly testing Hypothesis H1 (direct innovation effects) and Hypothesis H3 (purity moderation). The progressive specification strategy allows us to assess robustness while controlling for increasingly demanding fixed effect structures.

Table 6

Table 6. Baseline results: Green credit policy and firm innovation.

The baseline OLS specification in Column (1) reveals a positive and statistically significant relationship between GCPI and green patenting (coefficient = 0.126, $p<0.01$). Using the most demanding specification with firm, year, and city fixed effects (Column 4), a one-standard-deviation increase in GCPI (0.18) associates with an increase of 0.015 green patents per firm-year—approximately 0.6% of the sample mean (2.54). While modest in percentage terms, this effect is economically meaningful given the baseline low patenting rates and accumulates substantially over time and across the 1,600-firm sample.

The positive coefficient on Green Credit Purity (0.095, $p<0.05$ in Column 4) provides initial evidence that genuine green credit allocation enhances innovation outcomes independently of implementation intensity. More importantly, the significant interaction term (0.053, $p<0.05$) indicates complementarity between policy strictness and allocation quality, directly supporting Hypothesis H3. To interpret the interaction: at low purity levels (25th percentile = 0.58), the marginal effect of GCPI on innovation is 0.082 + 0.053 (0.58) = 0.113; at high purity (75th percentile = 0.88), it rises to 0.082 + 0.053 (0.88) = 0.129—a 14% amplification attributable to improved allocation quality.

The inclusion of firm fixed effects in Column (2) addresses concerns about time-invariant firm heterogeneity. The GCPI coefficient moderately attenuates from 0.126 to 0.103, suggesting that approximately 18% of the OLS effect reflects firm selection rather than pure policy impact. This pattern is consistent with the possibility that inherently more innovative firms locate in high-GCPI cities, though the persistence of substantial coefficients after firm fixed effects confirms genuine policy effects beyond selection.

Year fixed effects in Column (3) control for macroeconomic trends and national policy changes. The GCPI coefficient (0.094) remains economically and statistically significant, indicating that local implementation variations drive innovation beyond national-level policy trends. The full specification in Column (4) includes city fixed effects, exploiting only within-city temporal variation in policy implementation. The GCPI coefficient (0.082) remains significant at the 1% level, demonstrating that policy intensification within cities drives measurable innovation responses even after controlling for time-invariant city characteristics and common time shocks.

Across all specifications, R&D intensity exhibits the expected strong positive relationship with green patenting (coefficients 0.95–1.28), validating our innovation measures and confirming that green patents reflect genuine research activities. The positive firm size effects (0.025–0.048) align with literature on scale economies in innovation (Cohen and Tubb, 2018), while modest coefficient magnitudes suggest policy effects operate beyond simple scale advantages.

5.2.3 City-level environmental productivity: bridging micro to macro

Our firm-level analysis demonstrates that green credit policies enhance micro-level innovation, but environmental challenges require coordinated responses across entire regional economies. Table 7 examines whether policy implementation affects Green Total Factor Productivity (GTFP) at the city level, testing whether firm-level innovation translates into broader environmental efficiency gains and providing a bridge to our subsequent spatial analysis.

Table 7

Table 7. City-level environmental productivity effects.

The OLS specification in Column (1) reveals a substantial positive relationship between GCPI and regional GTFP (coefficient = 0.186). This city-level effect substantially exceeds the firm-level patent response in percentage terms, suggesting that green credit policies generate broader equilibrium effects beyond direct innovation incentives—potentially including supply chain adjustments, factor reallocation, and demonstration effects that propagate through local business networks.

Regional fixed effects in Column (2) exploit within-region variation while controlling for time-invariant geographic and institutional characteristics. The GTFP response (0.154) remains highly significant, indicating that policy implementation improvements within regions drive measurable environmental productivity gains. This within-region identification addresses concerns that cross-regional correlations might reflect unobserved regional characteristics rather than policy effects.

The preferred specification with both region and year fixed effects (Column 3) yields a coefficient of 0.127 $(p<0.01)$. Substantively, a one-standard-deviation increase in GCPI (0.18) associates with a 2.3 percentage point improvement in GTFP. Given that GTFP has a mean of 1.02 and standard deviation of 0.15 in our sample, this represents a meaningful shift—approximately 15% of one standard deviation in the GTFP distribution. The persistence of lagged GTFP (coefficient = 0.385) confirms partial adjustment dynamics in regional environmental efficiency, consistent with Gao et al. (2025) on environmental productivity persistence in Chinese cities.

The magnitude of regional effects relative to firm-level responses merits discussion. While individual firm patent counts increase modestly (0.6% of mean), city-level productivity improvements are more substantial (2.3 percentage points). This divergence likely reflects aggregation: GTFP incorporates not only innovation outputs but also production efficiency gains, emission reductions, and resource reallocation effects that manifest at the regional level. Additionally, spillovers from treated to untreated firms within cities—which we examine formally through spatial models in Section 5.3—may amplify aggregate impacts beyond direct firm responses.

These city-level results complement our firm-level findings by demonstrating economy-wide environmental benefits from green credit policy implementation. The consistency of effects across both micro and macro specifications strengthens confidence that observed relationships reflect genuine policy impacts rather than specification-specific artifacts. The regional results also establish a non-spatial baseline against which we can evaluate the additional explanatory power of spatial econometric models introduced in Section 5.3.

5.2.4 Industry heterogeneity and policy targeting effectiveness

Our theoretical framework predicts that green credit policies should most strongly affect firms in pollution-intensive industries, where environmental regulations create the strongest innovation incentives and compliance pressures. Table 8 tests this prediction by comparing policy effects across restricted (high-pollution) and non-restricted industries, while simultaneously examining how Green Credit Purity moderates these relationships—extending our test of Hypothesis H3 to incorporate industry-specific mechanisms.

Table 8

Table 8. Industry heterogeneity in policy effects.

For firms in restricted industries (Column 1), the GCPI coefficient (0.147) substantially exceeds the baseline sample average (0.082 from Table 6, Column 4), confirming that pollution-intensive firms respond more strongly to green credit policy implementation. The 79% larger coefficient (0.147 vs. 0.082) aligns with theoretical predictions that environmental policies create stronger innovation incentives for firms facing higher compliance costs (Ambec et al., 2013). This amplified response likely reflects multiple mechanisms: restricted firms face greater regulatory scrutiny, have more opportunities for emission-reducing innovation, and benefit more from the reputational advantages of demonstrable environmental improvements.

The interaction between GCPI and Purity in restricted industries (0.089, $p<0.05$) demonstrates that genuine green credit allocation particularly benefits high-pollution firms. At the 75th percentile of purity (0.88), the marginal effect of GCPI reaches 0.147 + 0.089 (0.88) = 0.225—more than double the average effect in non-restricted industries. This pattern suggests that allocation quality becomes critically important precisely when firms have both strong incentives to game the system (to access preferential credit) and genuine technological opportunities for green innovation.

Non-restricted firms (Column 2) exhibit positive but more modest policy responses (GCPI coefficient = 0.082), indicating that green credit policies generate innovation benefits even among cleaner industries. However, the smaller interaction effect (0.041, $p<0.10$) suggests that credit allocation quality matters less for firms already operating with relatively clean technologies. This differential pattern supports our theoretical emphasis on Green Credit Purity as a key moderating factor: allocation quality matters most when information asymmetries and greenwashing incentives are strongest.

The R&D intensity coefficients across both subsamples (1.42 for restricted, 0.87 for non-restricted) merit attention. The larger coefficient in restricted industries likely reflects that pollution-intensive firms face stronger pressures to translate research expenditures into patentable innovations, given regulatory requirements for documented environmental improvements. This pattern further validates our innovation measures as capturing genuine technological responses rather than strategic patent filing.

These industry-specific findings carry important policy implications. They suggest that green credit policies can be particularly effective tools for addressing pollution in heavy industries—precisely the sectors where command-and-control regulations often face the strongest political resistance. However, effectiveness depends critically on robust monitoring to ensure genuine green credit allocation. The large Purity interaction coefficients indicate that without quality controls, increased credit availability may simply fund opportunistic relabeling rather than genuine environmental investment.

5.2.5 Summary and links to spatial and threshold effects

Our baseline results provide robust evidence supporting Hypotheses H1 and H3 across multiple levels of analysis and identification strategies. At the firm level, green credit policy implementation intensity significantly enhances green patenting, with effects that survive progressively demanding fixed effects specifications exploiting only within-city-time variation. A one-standard-deviation increase in GCPI associates with approximately 0.6%–0.9% increases in firm-level green patents—modest in percentage terms but economically meaningful given baseline low patenting rates and substantial when aggregated across our 1,600-firm sample over 13 years.

At the regional level, these micro-level innovation gains translate into broader environmental productivity improvements, with GCPI increases of one standard deviation corresponding to 2.3 percentage point gains in city-level GTFP. The larger magnitude of regional effects relative to firm-level patent responses suggests that policy impacts extend beyond direct innovation incentives to encompass spillovers, factor reallocation, and demonstration effects that manifest at the city level.

Hypothesis H3 receives compelling support through our Green Credit Purity analysis. The positive and significant interaction terms across all specifications confirm that credit allocation quality amplifies policy effectiveness. This moderation effect is particularly pronounced in pollution-intensive industries, where GCPI effects are 79% larger than in cleaner sectors and where purity interactions are twice as strong. These patterns validate our theoretical emphasis on implementation quality: green credit effectiveness depends critically on both the intensity of policy implementation and the genuineness of credit allocation.

The estimated threshold at GCPI = 0.416 (bootstrap $p=0.008$) provides preliminary evidence of nonlinear effects that we examine formally in Section 5.3.3. Below this threshold, innovation responses remain strong and firm relocation probabilities stay low; above it, local innovation gains attenuate while relocation probabilities rise—consistent with regulatory arbitrage mechanisms predicted by our theoretical model. This threshold pattern suggests that unilateral policy escalation without regional coordination may generate diminishing returns and trigger displacement effects.

These findings establish strong average treatment effects while revealing important heterogeneities across industries and policy intensity levels. Section 5.3 extends this analysis to the spatial dimension, examining whether GCPI implementation in one city generates spillovers or displacement effects in neighboring jurisdictions (Hypothesis H4). Section 5.3.3 then formalizes the threshold analysis, decomposing regional responses into innovation enhancement versus firm relocation channels and identifying the conditions under which policy stringency transitions from promoting adaptation to triggering exit.

Results are robust to alternative estimation approaches and inference procedures. Poisson and negative binomial fixed effects models—which address count data properties and potential overdispersion in patent outcomes—yield similar marginal effects with semi-elasticity interpretations consistent with our OLS results. Driscoll-Kraay standard errors robust to spatial correlation produce slightly wider confidence intervals but preserve statistical significance for all core coefficients. These robustness checks, detailed in Supplementary Material File (Supplementary Tables C1–C3), confirm that our findings are not sensitive to functional form assumptions or standard error specifications.

5.2.6 Comparison with recent empirical evidence

Our estimated policy effects align with recent literature on green finance and innovation while revealing larger elasticities attributable to our focus on implementation quality. Liu et al. (2025) find that green finance development associates with 1.8%–2.4% increases in regional green innovation using provincial data, comparable to our city-level estimates of 2.3 percentage point GTFP gains per standard deviation of GCPI. However, their aggregate measure cannot distinguish implementation intensity from nominal policy adoption, potentially attenuating estimated effects through measurement error. Ma et al. (2025) report heterogeneous green credit impacts across cities with varying institutional capacity, consistent with our finding that effects are 68% larger among high-purity firms. Our explicit modeling of allocation quality through the Purity measure provides a mechanism for this heterogeneity: green credit effectiveness depends not only on lending volumes but critically on whether funds reach genuine environmental projects.

The firm-level patent elasticities we estimate (0.6%–0.9% per standard deviation GCPI increase) exceed those reported by Hu et al. (2021) for the 2012 Green Credit Guidelines (approximately 0.4% increase in green patents among targeted firms). This difference likely reflects two factors. First, our sample period (2010–2022) encompasses multiple policy intensifications including the 2016 G20 initiatives and 2021 carbon neutrality commitment, during which green credit implementation became more systematic. Second, our GCPI index captures continuous variation in enforcement stringency rather than a binary treatment, allowing us to identify effects of marginal implementation improvements that discrete policy indicators miss. The consistency of effects across OLS, fixed effects, and IV specifications strengthens confidence that estimates reflect causal relationships rather than selection or measurement artifacts.

5.3 Spatial effects and robustness analysis

Understanding the spatial dimensions of green credit policy implementation is crucial for effective policy design, as environmental challenges transcend administrative boundaries and policy effects can spill over across regions. This section examines four critical aspects of our findings’ robustness through a systematic identification strategy: spatial spillover effects that validate our theoretical predictions about cross-regional externalities (Section 5.3.1), instrumental variable estimation that establishes causal identification (Section 5.3.2), threshold effects and firm relocation responses that illuminate non-linear policy dynamics (Section 5.3.3), and comprehensive robustness checks including alternative innovation measures (Section 5.3.4).

5.3.1 Spatial spillovers and effect decomposition

We next examine the spatial dimension of green credit implementation. Building on the SDM framework introduced in Section 4, Table 9 reports the decomposed direct, indirect (spillover), and total effects of GCPI on city-level GTFP, following the approach of LeSage and Pace (2009). Rather than focusing on raw coefficients, this decomposition translates the spatial feedback loops into economically interpretable marginal effects.

Table 9

Table 9. Spatial effect decomposition: Direct, indirect, and total effects.

The results show three robust patterns. First, the direct effect of GCPI on local GTFP is positive and precisely estimated: a one-standard-deviation increase in GCPI (0.18) raises local green productivity by approximately 2.28–2.52 percentage points in our preferred specifications.³ Second, the indirect effect—capturing the impact of neighboring cities’ GCPI on local GTFP—is also positive and statistically significant. On average, the spillover effect amounts to about 38%–40% of the direct effect, implying that a non-trivial share of the gains from stricter implementation diffuses to surrounding jurisdictions. Third, the total effect (direct + indirect) is substantially larger than the direct effect alone, confirming that treating cities as isolated units would materially understate the productivity benefits of green credit implementation.

These findings speak directly to the competing views summarized in Section 2. While the “pollution haven” literature emphasizes relocation from strict to lax regions, our SDM results indicate that, on balance, green credit implementation generates net positive spatial externalities. Neighboring cities experience higher GTFP when their peers implement green credit more intensively, consistent with mechanisms of technology diffusion, supply-chain upgrading, and regulatory learning. We return to this tension in Section 6, where we contrast our positive-spillover evidence with studies documenting pollution-haven patterns under weaker implementation or different policy regimes.

These positive spillover estimates align with diffusion-oriented findings in recent literature while providing finer-grained decomposition. Du et al. (2022) document that green development initiatives in Chinese cities generate positive spatial externalities to neighbors, with indirect effects approximately 40%–50% of direct effects—closely matching our spillover-to-direct ratio of 38%–40%. Their interpretation emphasizes technology diffusion and demonstration effects, consistent with our finding that neighboring cities benefit from local green credit implementation through knowledge propagation and supply-chain upgrading. The concordance across studies using different green development measures (comprehensive indices versus financial policy implementation) suggests robust spatial complementarities in environmental policy.

However, our threshold analysis in Section 5.3.3 reveals important nonlinearities absent from aggregate spillover estimates. While Du et al. (2022) report uniformly positive spatial effects, we find that spillover patterns depend on policy intensity levels. Below the estimated threshold (GCPI = 0.416), implementation generates strong positive spillovers consistent with their findings. Above this threshold, local innovation gains attenuate and firm relocation probabilities increase, indicating that extremely strict implementation may trigger displacement rather than diffusion. This pattern reconciles positive spillover evidence with pollution haven findings by Zhao et al. (2023) and Lyu et al. (2023), showing that both mechanisms operate but activate under different policy stringency regimes. The threshold provides a quantitative benchmark distinguishing innovation-inducing from displacement-generating implementation intensity.

5.3.2 Instrumental variable evidence: Causal interpretation

While our fixed effects specifications control for many sources of potential bias, endogeneity concerns remain if unobserved city-level shocks simultaneously affect both green credit policy implementation and innovation outcomes. A remaining concern is that our historical banking instrument might affect current innovation through channels other than GCPI—for instance, by shaping long-run human capital accumulation or industrial structure. We address this in three ways. First, we augment the second-stage specifications with controls for historical education levels and industry composition (measured around 2000), and the GCPI coefficient remains stable. Second, we introduce an additional, conceptually distinct instrument based on pre-policy connectivity to national green technology exchanges interacted with annual national green bond issuance, and conduct over-identification tests. The Hansen $J$ statistics do not reject the joint validity of the instruments. Third, Supplementary Material File presents a simple path diagram clarifying that the most plausible channel through which the 1998 environmental loan share influences current innovation is via the evolution of green credit implementation capacity, rather than through direct effects on contemporary R&D decisions.

Table 10 addresses these concerns using an instrumental variable strategy. We use the provincial share of historical environmental loans in 1998 as an external instrument for contemporary GCPI. This measure predates national green credit policies and proxies path-dependent capacity for environmental finance.

Table 10

Table 10. Instrumental variable estimation with diagnostic tests.

The first-stage results demonstrate strong instrument relevance: the historical banking measure significantly predicts contemporary GCPI with a coefficient of 0.248 and an F-statistic of 25.94, well above conventional weak instrument thresholds. The Kleibergen-Paap rk Wald F-statistic confirms instrument strength under heteroskedasticity and clustering. Anderson-Rubin weak instrument tests yield $p<0.01$, rejecting the null of no relationship even allowing for weak instruments. These diagnostics, summarized in Panel A of Table 10, provide confidence in instrument relevance.

The second-stage results provide reassuring evidence of causal identification. The instrumented GCPI coefficient (0.139) closely matches our baseline fixed effects estimates (0.082 in Table 6), suggesting that endogeneity bias does not substantially affect our conclusions. This consistency strengthens confidence in our causal interpretation: green credit policy implementation genuinely drives innovation outcomes rather than merely reflecting reverse causality or omitted variable bias.

The instrumental variable approach also validates our Green Credit Purity findings. The interaction between instrumented GCPI and Purity remains positive and significant (0.067), confirming that credit allocation quality causally moderates policy effectiveness. This result provides evidence on the importance of green credit allocation quality, contributing novel insights to the growing literature on green finance effectiveness (Flammer, 2021).

5.3.3 Thresholds and relocation: Non-linear policy responses

Our theoretical model suggests that extremely strict policy implementation might trigger pollution displacement as firms relocate to jurisdictions with more lenient regulations. To make the threshold estimates more transparent for policy interpretation, we translate the coefficients into predicted relocation probabilities for a representative firm. Using a logit specification for $Relocatio{n}_{i,t+1}$ with the same set of controls and fixed effects, we compute predicted probabilities at different levels of GCPI while holding other covariates at their sample medians.

Figure 5 plots these predicted probabilities together with 95% confidence bands. In the low-to-moderate regime $(GCPI\le \widehat{\gamma }=0.416)$, relocation risks remain low and fairly flat: increasing GCPI from the 25th to the 50th percentile (approximately 0.30–0.39) raises the relocation probability only from about 3.2% to 4.1%. In contrast, once GCPI exceeds the estimated threshold, relocation risks increase more steeply. Moving from $GCPI=0.42$ to $GCPI=0.55$ raises the predicted relocation probability from roughly 4.8%–9.3%, an increase of about 4.5 percentage points. This pattern is consistent with the threshold regression in Table 11: implementation is innovation-enhancing at low and moderate intensity, but extremely strict policies can trigger spatial displacement.

Figure 5

Figure 5. Predicted Firm Relocation Probability Across GCPI Intensity Levels. The figure plots predicted relocation probabilities for a representative firm (median firm size, R&D intensity, and purity) as a function of local GCPI intensity. Probabilities computed from logit estimation of $Relocatio{n}_{i,t+1}$ with firm, year, and city fixed effects. Shaded area represents 95% confidence intervals computed via delta method. Vertical dashed line marks the estimated threshold $\widehat{\gamma }=0.416$ (bootstrap $p=0.008$). Horizontal dotted lines indicate quartiles of sample GCPI distribution (P25 = 0.42, P50 = 0.56, P75 = 0.68). Below the threshold, relocation risks increase slowly; above it, the probability gradient steepens markedly, consistent with regulatory arbitrage mechanisms.

Table 11

Table 11. Threshold regression: Non-linear effects and firm relocation.

Substantively, these magnitudes suggest that most cities in our sample—which cluster below the threshold—can tighten green credit implementation without inducing large relocation responses, especially if accompanied by improvements in Purity. However, cities already operating in the upper tail of the GCPI distribution face a sharper trade-off between local stringency and the risk of pollution displacement, reinforcing the need for regional coordination.

Table 11 presents the underlying threshold regression results that support Figure 5. The estimated threshold at $GCPI=0.416$ (bootstrap $p=0.008$) separates regimes with strong local innovation and low relocation $(\le 0.416)$ from regimes with muted innovation and higher relocation probability $(>0.416)$, consistent with displacement beyond a critical stringency. This provides evidence for Hypothesis H2: beyond a critical threshold, policy stringency can trigger pollution displacement as firms choose exit over adaptation (Zheng et al., 2025).

The threshold analysis yields important policy insights about optimal green credit policy design. The identified threshold (GCPI = 0.416) falls within the upper tercile of our sample distribution, suggesting that most cities operate below this critical level. However, the existence of threshold effects warns against unlimited policy intensification without regional coordination. Cities implementing extremely strict policies may inadvertently reduce aggregate environmental benefits if they push pollution-intensive activities to neighboring jurisdictions with weaker implementation (Su et al., 2024).

The relocation pattern we document aligns with recent evidence on regulatory sensitivity in firm location decisions. Zhao et al. (2025) show that environmental tax stringency drives polluting firms to relocate from eastern to western Chinese cities, with relocation probabilities increasing sharply when tax intensity exceeds specific thresholds. Our finding that relocation risks accelerate above GCPI = 0.416 provides a parallel threshold for financial policy instruments, suggesting that credit-based and regulation-based environmental policies trigger similar behavioral responses among pollution-intensive firms. Similarly, Wu et al. (2017) document that firms facing strict environmental mandates are more likely to shift production to jurisdictions with weaker oversight, particularly when relocation costs are low relative to compliance burdens. This mechanism likely underlies our threshold effect: below 0.416, most firms find adaptation cheaper than exit; above this level, compliance costs for marginal firms exceed relocation costs plus foregone local market access.

The heterogeneity in relocation responses across firm types provides additional insight into threshold mechanisms. Our supplementary analysis (available upon request) shows that relocation probability gradients are steepest among capital-intensive, pollution-heavy manufacturing firms with established production networks in multiple cities—precisely the firms with lower relocation frictions and stronger incentives to exploit regulatory arbitrage opportunities. Small firms and service-sector enterprises exhibit flatter relocation curves, consistent with higher mobility costs and lower exposure to green credit requirements. These patterns indicate that threshold effects primarily reflect strategic responses by mobile, pollution-intensive firms rather than broad-based exit from high-implementation cities, explaining why aggregate regional GTFP continues improving even above the threshold despite rising firm exit rates.

5.3.4 Additional robustness checks

5.3.4.1 Alternative innovation measures: Granted patents

To ensure that our findings reflect genuine innovation responses rather than strategic patent filing behavior, Table 12 replicates our analysis using granted green patents as the dependent variable. Patent grants represent a more stringent innovation measure, as they require passing examination for novelty, non-obviousness, and utility—filtering out applications that may reflect strategic behavior rather than genuine innovation.

Table 12

Table 12. Robustness: Granted green patents as alternative outcome.

The results using granted patents closely parallel our baseline findings, though with appropriately smaller magnitudes reflecting the more restrictive dependent variable. The GCPI coefficients (0.074–0.089) remain highly significant across specifications, indicating that policy implementation drives patents that successfully pass examination. The reduction in magnitude (approximately 25% smaller than application-based results) aligns with typical grant rates and examination lags in patent systems (Hu et al., 2021). Crucially, the interaction between GCPI and Purity (coefficients 0.041–0.052) remains significant across all specifications, confirming that credit allocation quality affects patents that represent genuine innovation rather than mere application volume.

As an additional robustness check, we construct a citation-weighted green patent index that accounts for the technological importance of each patent. For each granted green patent, we compute its forward citation count within a 3-year window and standardize this measure within grant-year–IPC (4-digit) cells. The firm-year index, $CiteWeighted\text{\_}GreenPaten{t}_{it}$, is defined as the sum of standardized citation scores across all green patents owned by firm $i$ in year $t$.

Table 13 shows that our main conclusions are unchanged when using citation-weighted patents as the outcome. The GCPI coefficient remains positive and statistically significant, and the GCPI–Purity interaction continues to amplify innovation responses. Figure 6 compares the coefficients across outcome measures and shows that effect sizes are similar in magnitude, suggesting that green credit policies raise not only the quantity but also the quality of green innovation.

Table 13

Table 13. Robustness: Citation-weighted green patents as quality measure.

Figure 6

Figure 6. Coefficient Comparison Across Alternative Innovation Measures. This figure compares estimated GCPI coefficients (blue bars) and GCPI$\times$ Purity interaction coefficients (green bars) across three dependent variables: (1) green patent applications (baseline), (2) granted green patents, and (3) citation-weighted green patents. Error bars represent 95% confidence intervals computed with city-clustered robust standard errors. All estimates from full fixed effects specifications (firm, year, and city FE). The similar magnitudes across measures confirm that green credit policies enhance both the quantity and quality of green innovation, addressing concerns about strategic patent filing behavior.

Our spatial spillover findings are robust to alternative spatial econometric specifications and weight matrix definitions. Supplementary Table C6 in Supplementary Material File reports results using: (1) spatial lag models (SLM) that restrict spillovers to operate through the dependent variable only; (2) spatial error models (SEM) that attribute spatial correlation to unobserved shocks; (3) k-nearest neighbors weight matrices (k = 5) that emphasize immediate adjacency; (4) Queen contiguity matrices based on administrative boundaries; and (5) economic distance matrices weighted by GDP per capita similarity.

Across all specifications, the core pattern persists: positive and statistically significant spillover effects ranging from 35% to 42% of direct effects. The consistency across different spatial connectivity assumptions—geographic distance, administrative proximity, and economic similarity—strengthens confidence that our spillover estimates capture genuine policy externalities rather than artifacts of particular weight matrix choices. The robustness to SEM specifications further indicates that positive spillovers reflect substantive cross-city transmission mechanisms (technology diffusion, supply-chain linkages) rather than merely correlated unobservables.

The threshold analysis introduces an important caveat: while green credit policies generally enhance innovation, extremely strict implementation may trigger counterproductive firm relocation. This finding highlights the importance of regional policy coordination to prevent pollution displacement while maintaining innovation incentives. These results collectively suggest that optimal green credit policy design requires balancing local implementation intensity with regional coordination mechanisms. The positive spillover effects indicate substantial benefits from policy harmonization across neighboring jurisdictions, while the threshold effects warn against unilateral policy escalation that might undermine broader environmental objectives.

6 Discussion

6.1 Contributions to existing literature

Our findings extend three research streams in environmental finance and innovation. First, we advance measurement of green finance policy heterogeneity. Prior studies typically treat green credit as a binary policy indicator or aggregate loan volume (Hu et al., 2021; Hong et al., 2021; Yao et al., 2021; Rajah et al., 2023). We demonstrate that local implementation intensity and allocation quality jointly determine policy effectiveness. The GCPI index captures variation in enforcement stringency across cities implementing the same national guidelines, while the Purity measure quantifies environmental additionality at the firm level. This measurement approach addresses calls by Schoenmaker and Schramade (2019) and Flammer (2021) for finer-grained indicators of environmental policy execution. The systematic coding protocol documented in Supplementary Material File establishes operational definitions and reliability benchmarks that subsequent research can adapt to other contexts.

Second, we establish the transmission mechanism linking financial policy to environmental outcomes across analytical levels. Most empirical work examines either firm responses (Liu et al., 2023a; Liu et al., 2023b) or regional outcomes (Gao et al., 2025; Ma et al., 2025) separately. By estimating effects on both firm green patenting and city-level GTFP within the same sample, we demonstrate how micro-level innovation aggregates into regional productivity gains. The 2.3 percentage point GTFP improvement associated with a one-standard-deviation GCPI increase indicates that firm innovation responses generate measurable environmental efficiency gains at the regional level. This evidence addresses the “micro–macro disconnect” noted by Cohen and Tubb (2018) and Ambec et al. (2013), showing that financial interventions propagate through the economy to affect aggregate environmental performance.

Third, we reconcile competing spatial findings through threshold modeling. The literature reports both positive spillovers (Zhang et al., 2021) and pollution displacement (Su et al., 2024; Zheng et al., 2025; Yadav et al., 2025) from environmental policies. Our spatial Durbin decomposition shows that neighboring cities benefit from local green credit implementation, with spillovers equal to 38–40 percent of direct effects. Simultaneously, threshold regression identifies a critical GCPI level (0.416) beyond which innovation gains diminish and relocation probabilities increase. This pattern indicates that both technology diffusion and regulatory arbitrage operate, but activate under different policy intensity conditions. The threshold estimate provides a quantitative benchmark for evaluating when implementation stringency transitions from innovation-inducing to displacement-generating.

Supplementary Material File addresses methodological concerns about measurement validity and identification. We document first-stage instrument strength (F-statistics exceeding 20), weak-instrument-robust inference (Anderson-Rubin and conditional likelihood ratio confidence intervals), and falsification tests showing null effects in pre-policy periods. The Purity measure demonstrates 0.83 correlation with alternative environmental investment indicators from CSR reports and 0.76 correlation with third-party green ratings. These diagnostics establish that our estimates reflect causal relationships and are not artifacts of measurement error or specification choices.

This study examines how city-level heterogeneity in green credit policy implementation (GCPI) and the quality of credit allocation (Purity) shape firm innovation and regional environmental performance. Using a matched firm–city panel of 1,600 listed firms across 280 cities from 2010 to 2022, we show that implementation intensity and allocation quality jointly determine the effectiveness of green credit. Baseline panel regressions, spatial Durbin models, and IV–2SLS estimates all point to three robust patterns: (i) stronger local implementation is associated with higher firm-level green innovation and city-level green total factor productivity; (ii) these effects are amplified where a larger share of labelled green credit is channelled to verifiable environmental projects; and (iii) green credit policies generate positive spatial spillovers up to a threshold beyond which relocation risks rise.

6.2 Positioning in the literature and reconciling spatial findings

Our results speak directly to two strands of the environmental economics and green finance literatures. The first is the “innovation-inducing regulation” perspective, in which well-designed environmental policies can stimulate technological change and competitiveness rather than merely imposing static costs (Porter and van der Linde, 1995; Ambec et al., 2013). We complement this view by shifting the focus from policy scale alone to the execution margin: how intensely green credit rules are enforced locally and how credibly green projects are screened. The strong GCPI–Purity interaction documented in the baseline and IV specifications indicates that allocation quality conditions the translation of formal rules into real innovation outcomes, consistent with concerns about greenwashing and symbolic compliance (Lyon and Maxwell, 2011; Flammer, 2021; Schoenmaker and Schramade, 2019).

The second strand concerns the spatial consequences of environmental and green finance policies. A large body of work emphasizes “pollution haven” or “race-to-the-bottom” dynamics, whereby stricter regulation in one jurisdiction induces firms to relocate to laxer areas, with ambiguous or even negative net environmental effects (Zheng et al., 2025; Su et al., 2024; Yadav et al., 2025; Orikpete and Ewim, 2024). By contrast, our spatial Durbin results in Section 5.3 show positive and economically meaningful spillovers: Table 7 indicates that indirect effects account for roughly 38%–40% of the direct impact of GCPI on neighboring cities’ GTFP, and the total effect is substantially larger than the own-city effect alone. This pattern is more in line with diffusion and linkage narratives, in which green policies propagate through technology spillovers, input–output linkages, and demonstration effects (Zhang et al., 2021; Liu et al., 2025; Ma et al., 2025).

Reconciling these apparently conflicting findings requires attention to research design. First, our spatial weights emphasize geographic and economic proximity—inverse-distance and contiguity matrices, with robustness checks using k-nearest neighbours and alternative specifications—whereas many pollution-haven studies rely on trade or FDI linkages across countries or sectors. Our positive indirect effects therefore capture localised diffusion and learning among neighbouring cities, rather than long-distance relocation along global value chains. Second, the time period matters. Our 2010–2022 window covers the gradual institutionalisation of green credit and the post-2016 strengthening of green finance governance; earlier periods with weaker monitoring and looser taxonomies are more likely to generate relocation without genuine greening. Third, the unit of analysis differs. We work at the city level within a single national regulatory regime, where capital and labour mobility are constrained by hukou and land institutions. Cross-country analyses, by contrast, often study relocation across vastly different regulatory and enforcement systems, where pollution-haven mechanisms are stronger. Finally, we explicitly model execution heterogeneity through GCPI and Purity. Sub-sample results show that positive spillovers are strongest where both implementation intensity and allocation quality are high, and weaken in low-purity regimes. This suggests that green finance can reduce, rather than exacerbate, pollution-haven risks when accompanied by credible screening and monitoring.

The threshold and relocation analysis adds an important nuance. While most cities operate below the estimated GCPI threshold and enjoy net positive local and spatial gains from tightening green credit, very stringent implementation in the upper tail is associated with higher relocation probabilities. This pattern helps reconcile our positive-spillover evidence with the pollution-haven literature: relocation pressures appear to be a marginal phenomenon that emerges only beyond a critical stringency level and in the absence of sufficient inter-jurisdictional coordination. In this sense, our results align with recent work arguing that green financial instruments can support low-carbon transitions when embedded in coherent policy mixes, but may trigger displacement where enforcement is uneven or overly unilateral (Satpathy et al., 2025).

6.3 Limitations and directions for future research

Several limitations qualify our findings and point to avenues for future work. A first concern is sample selection. Our firm-level analysis is restricted to listed companies, which are typically larger, more regulated, and more transparent than the universe of firms. These firms are more likely to be direct beneficiaries of formal green credit programmes and to respond to capital-market signals. As a result, the innovation responses we document may overstate impacts relative to small and medium-sized enterprises (SMEs), or may miss informal adaptation margins available to smaller private firms. Extending the analysis to loan-level or credit registry data that cover SMEs would allow researchers to test whether green credit penetrates deeper into the firm size distribution and whether the GCPI–Purity mechanism generalises.

Second, there are measurement and disclosure issues surrounding our Purity indicator. Although we implement a transparent coding protocol with positive and negative lists, double-coding, and high inter-coder agreement, the measure ultimately relies on firms’ narrative disclosures and third-party documents. Firms that under-report environmentally relevant investments may be classified as low-purity despite genuine greening, whereas sophisticated greenwashing may inflate Purity scores if language is carefully crafted. In addition, changes in disclosure regulation over 2010–2022 can mechanically affect measured Purity. These factors would likely attenuate estimated effects, but they nonetheless caution against interpreting Purity as a noise-free measure of allocation quality. Future work could triangulate firm disclosures with bank-level green loan registries, project-level verification data, or remote-sensing indicators of environmental performance.

Third, our policy implementation and outcome measures are subject to non-trivial uncertainty. GCPI is a bottom-up construct that reconciles firm-level lending data with provincial aggregates; coverage gaps in financial databases or misclassification of loan purposes may introduce classical measurement error, biasing coefficients toward zero. GTFP is computed from city-level accounts and pollution statistics that may not fully capture service-sector activities or informal emissions. Although we conduct extensive robustness checks and obtain consistent patterns, residual measurement error remains a potential source of bias.

Fourth, our setting captures a transitional institutional regime. The 2010–2022 horizon coincides with the rollout, scaling-up, and partial consolidation of China’s green credit framework. Dynamic effects may evolve as taxonomies are refined, disclosure requirements are harmonised, and green bonds and other instruments mature. The innovation and relocation responses we identify therefore reflect a particular phase in the policy learning curve. Longer panels and post-2022 data would allow future research to study whether the GCPI–Purity mechanism strengthens, weakens, or changes sign as institutions stabilise.

Fifth, our spatial perspective is primarily geographic. While alternative weight matrices based on economic similarity and contiguity yield similar qualitative conclusions, we do not explicitly model propagation along supply chains or ownership networks. Yet the mechanisms through which green credit travels—upstream and downstream financing, cross-regional corporate groups, or sectoral clusters—may differ from simple geographic proximity. Combining our city-level approach with firm- or plant-level production network data would help to distinguish “neighbourhood” spillovers from “network” spillovers and to map the channels through which green finance reshapes the geography of emissions.

Finally, although our IV strategy and extensive diagnostic tests mitigate concerns about reverse causality and omitted variables, the analysis remains observational. Historical environmental loan shares and connectivity to green finance infrastructure are plausibly exogenous, but they may still correlate with unobserved, slowly evolving institutional traits. Identification based on sharper quasi-experiments—such as staggered expansions of green finance pilot zones, sudden changes in eligibility criteria, or exogenous shocks to green bond demand—would provide additional leverage to isolate causal effects.

Addressing these limitations will require new data and methods. Future research can (i) link bank-level green lending records to a broader firm universe, including SMEs; (ii) integrate third-party verification and remote-sensing information to refine allocation-quality metrics; (iii) extend spatial models to production and ownership networks; and (iv) exploit policy discontinuities to sharpen causal identification. Such extensions would deepen our understanding of how green credit interacts with other regulatory and fiscal instruments in shaping not only the level but also the spatial distribution and quality of low-carbon innovation.

7 Conclusion

This paper has examined how city-level heterogeneity in green credit policy implementation intensity and the quality of credit allocation jointly shape firm innovation and regional environmental performance. Using a matched firm–city panel for 2010–2022, four main findings emerge. First, stronger implementation intensity is consistently associated with higher firm-level green patenting and improved city-level green total factor productivity; a one-standard-deviation increase in implementation corresponds to modest but robust gains in both innovation and productivity. Second, allocation quality acts as a multiplier: the innovation and productivity effects of higher implementation are substantially larger when a greater share of labelled green credit actually finances verifiable environmental projects. Third, spatial decomposition shows that neighbouring cities benefit from each other’s implementation efforts, with indirect spillovers amounting to roughly 38%–40% of the local effect. Fourth, threshold analysis reveals a non-linear pattern: below a Green Credit Policy Implementation index of 0.416, policy tightening is associated with strong innovation responses and limited relocation, whereas above this threshold local innovation gains attenuate and firm relocation probabilities rise.

These findings translate into three concrete policy implications. First, regional coordination is essential to unlock positive spillovers and prevent free-riding. Mechanisms such as regional green finance platforms, joint target-setting, and cross-city performance evaluation can help internalise the documented spillover gains. Second, policy intensity should be calibrated with explicit consideration of thresholds. For most cities, moderate tightening from current levels is likely to yield additional innovation and productivity benefits, but cities approaching the upper range of implementation should prioritise coordination and monitoring of relocation risks rather than unilateral escalation. Third, making allocation quality a core management objective can shift outcomes from labelled green credit volumes to genuine decarbonisation. Standardised verification procedures, periodic audits, and linking concessional terms and supervisory assessments to validated purity scores—especially in high-emission sectors—can reduce greenwashing, enhance innovation returns, and ensure that green credit delivers real environmental improvements rather than merely re-labelling existing financial flows.

Data availability statement

The data analyzed in this study is subject to the following licenses/restrictions: Data are available from the providers upon application; licensing restrictions apply. Requests to access these datasets should be directed to Shuangquan Yang—shuangquanyang97@163.com.

Author contributions

SY: Writing – original draft, Writing – review and editing.

Funding

The author declares that no financial support was received for the research and/or publication of this article.

Conflict of interest

The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The authors declare that no Generative AI was used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenvs.2025.1715753/full#supplementary-material

References

Akomea-Frimpong, I., Adeabah, D., Ofosu, D., and Tenakwah, E. J. (2022). A review of studies on green finance of banks, research gaps and future directions. J. Sustain. Finance and Invest. 12 (4), 1241–1264. doi:10.1080/20430795.2020.1870202

CrossRef Full Text | Google Scholar

Ambec, S., Cohen, M. A., Elgie, S., and Lanoie, P. (2013). The porter hypothesis at 20: can environmental regulation enhance innovation and competitiveness? Rev. Environ. Econ. Policy 7 (1), 2–22. doi:10.1093/reep/res016

CrossRef Full Text | Google Scholar

Calel, R., and Dechezleprêtre, A. (2016). Environmental policy and directed technological change: evidence from the european carbon market. Rev. Econ. Statistics 98 (1), 173–191. doi:10.1162/rest_a_00470

CrossRef Full Text | Google Scholar

Chen, X., and Chen, Z. (2021). Can green finance development reduce carbon emissions? Empirical evidence from 30 Chinese provinces. Sustainability 13 (21), 12137. doi:10.3390/su132112137

CrossRef Full Text | Google Scholar

Chen, Y., Zhang, Y., and Wang, M. (2024). Green credit, financial regulation and corporate green innovation: evidence from China. Fin. Res. Lett. 59, 104768. doi:10.1016/j.frl.2024.104768

CrossRef Full Text | Google Scholar

Climate Policy Initiative (2023). Global landscape of climate finance 2023. Report. Clim. Policy Initiat. Available online at: https://www.climatepolicyinitiative.org/publication/global-landscape-of-climate-finance-2023/.

Google Scholar

Cohen, M. A., and Tubb, A. (2018). The impact of environmental regulation on firm and country competitiveness: a meta-analysis of the porter hypothesis. J. Assoc. Environ. Resour. Econ. 5 (2), 371–399. doi:10.1086/695613

CrossRef Full Text | Google Scholar

Cui, X., Wang, P., Sensoy, A., Nguyen, D. K., and Pan, Y. (2022). Green credit policy and corporate productivity: evidence from a quasi-natural experiment in China. Technol. Forecast. Soc. Change 177, 121516. doi:10.1016/j.techfore.2022.121516

CrossRef Full Text | Google Scholar

Delmas, M. A., and Burbano, V. C. (2011). The drivers of greenwashing. Calif. Manag. Rev. 54 (1), 64–87. doi:10.1525/cmr.2011.54.1.64

CrossRef Full Text | Google Scholar

Deschryver, P., and de Mariz, F. (2020). What future for the green bond market? How can policymakers, companies, and investors unlock the potential of the green bond market? J. Risk Financial Manag. 13 (3), 61. doi:10.3390/jrfm13030061

CrossRef Full Text | Google Scholar

Dikau, S., and Volz, U. (2021). Central bank mandates, sustainability objectives and the promotion of green finance. Ecol. Econ. 184, 107022. doi:10.1016/j.ecolecon.2021.107022

CrossRef Full Text | Google Scholar

Dogah, K. E., Wesseh, P. K. J., and Adomako, S. (2025). Green credit policy, technological innovation, and corporate financial performance: evidence from the energy industry. Bus. Strategy Environ. 34 (1), 1171–1188. doi:10.1002/bse.4041

CrossRef Full Text | Google Scholar

Du, M., Antunes, J., Wanke, P., and Chen, Z. (2022). Ecological efficiency assessment under the construction of low-carbon city: a perspective of green technology innovation. J. Environ. Plan. Manag. 65 (9), 1727–1752. doi:10.1080/09640568.2021.1945552

CrossRef Full Text | Google Scholar

Elhorst, J. P. (2014). Spatial econometrics: from cross-sectional data to spatial panels. Springer.

Google Scholar

Flammer, C. (2021). Corporate green bonds. J. Financial Econ. 142 (2), 499–516. doi:10.1016/j.jfineco.2021.01.010

CrossRef Full Text | Google Scholar

Gao, J., Hua, G., and Huo, B. (2025). Turning “green” into “gold”: a study on the impact of green finance pilot zone policy on energy carbon emission efficiency. Sustain. Dev. 33 (1), 129–143. doi:10.1002/sd.3114

CrossRef Full Text | Google Scholar

Guo, L., and Tan, W. (2023). Relationship between green credit and high-quality development of green and green economy-“pull effect” or “capricorn effect”. Environ. Dev. Sustain. 26, 26449–26473. doi:10.1007/s10668-023-03736-6

CrossRef Full Text | Google Scholar

Holmström, B., and Tirole, J. (1997). Financial intermediation, loanable funds, and the real sector. Q. J. Econ. 112 (3), 663–691. doi:10.1162/003355397555316

CrossRef Full Text | Google Scholar

Hong, M., Li, Z., and Drakeford, B. (2021). Do the green credit guidelines affect corporate green technology innovation? Empirical research from China. Int. J. Environ. Res. Public Health 18 (4), 1682. doi:10.3390/ijerph18041682

PubMed Abstract | CrossRef Full Text | Google Scholar

Hu, G., Wang, X., and Wang, Y. (2021). Can the green credit policy stimulate green innovation in heavily polluting enterprises? Evidence from a quasi-natural experiment in China. Energy Econ. 98, 105134. doi:10.1016/j.eneco.2021.105134

CrossRef Full Text | Google Scholar

Jaffe, A. B., and Palmer, K. (1997). Environmental regulation and innovation: a panel data study. Rev. Econ. Statistics 79 (4), 610–619. doi:10.1162/003465397557196

CrossRef Full Text | Google Scholar

LeSage, J., and Pace, R. K. (2009). Introduction to spatial econometrics. Boca Raton, FL: CRC Press.

Google Scholar

Liu, L., Luo, D., Xing, K., and Chang, K. (2023a). The impact of green technology innovation on carbon dioxide emissions: the role of local environmental regulations. J. Environ. Manag. 340, 117990. doi:10.1016/j.jenvman.2023.117990

PubMed Abstract | CrossRef Full Text | Google Scholar

Liu, Z., He, S., Li, W., and Sun, H. (2023b). Does green credit reduce carbon emissions? Evidence from China. Environ. Sci. Pollut. Res. 30 (10), 26735–26751. doi:10.1007/s11356-022-24011-7

PubMed Abstract | CrossRef Full Text | Google Scholar

Liu, L., Dong, H., and Qi, L. (2025). How directors with green backgrounds drive corporate green innovation: evidence from China. Sustainability 17 (15), 6944. doi:10.3390/su17156944

CrossRef Full Text | Google Scholar

Lyon, T. P., and Maxwell, J. W. (2011). Greenwash: corporate environmental disclosure under threat of audit. J. Econ. and Manag. Strategy 20 (1), 3–41. doi:10.1111/j.1530-9134.2010.00282.x

CrossRef Full Text | Google Scholar

Lyu, Y., Zhao, Y., and Zhang, J. (2023). Green credit and low-carbon development in China: fresh evidence on spatial spillover insights. Environ. Sci. Pollut. Res. 30 (56), 118601–118615. doi:10.1007/s11356-023-30514-8

PubMed Abstract | CrossRef Full Text | Google Scholar

Ma, H., Zhou, Y., Pan, Y., Dong, R., and Tang, D. (2025). Spatial spillover of fintech innovation on green economic growth based on 30 provinces in China. Front. Environ. Sci. 13, 1514403. doi:10.3389/fenvs.2025.1514403

CrossRef Full Text | Google Scholar

Moreira, M. J. (2003). A conditional likelihood ratio test for structural models. Econometrica 71 (4), 1027–1048. doi:10.1111/1468-0262.00438

CrossRef Full Text | Google Scholar

Orikpete, O. F., and Ewim, D. R. E. (2024). “Harmonising efficiency and sustainability: a techno-economic analysis of green hydrogen production methods,” in Challenges and opportunities in green hydrogen production (Springer Nature Singapore), 537–567.

CrossRef Full Text | Google Scholar

Porter, M. E., and van der Linde, C. (1995). Green and competitive: ending the stalemate. Harv. Bus. Rev. 73 (5), 120–134. Available online at: https://hbr.org/1995/09/green-and-competitive-ending-the-stalemate.

Google Scholar

Rajah, K., Ghriss, O., Fguiri, A., and Jeday, M.-R. (2023). “Validating the techno-economic and environmental sustainability of solar pv technology in agriculture using retscreen experts to assess its viability,” in International conference on green energy conversion System (Springer), 795–800.

Google Scholar

Rubashkina, Y., Galeotti, M., and Verdolini, E. (2015). Environmental regulation and competitiveness: empirical evidence on the porter hypothesis from european manufacturing sectors. Energy Policy 83, 288–300. doi:10.1016/j.enpol.2015.02.014

CrossRef Full Text | Google Scholar

Satpathy, I., Chakravarty, D., Patnaik, B. C. M., and Poddar, S. (2025). “Thriving sustainably: redefining business ecosystems through the UN sustainable development goals,” in Sustainable business ecosystems and social perspectives (Hershey, PA: IGI Global), 1–32.

CrossRef Full Text | Google Scholar

Schoenmaker, D., and Schramade, W. (2019). Principles of sustainable finance. Oxford University Press.

Google Scholar

Schumacher, K., Chenet, H., and Volz, U. (2020). Sustainable finance in Japan. J. Sustain. Finance and Invest. 10 (2), 213–246. doi:10.1080/20430795.2020.1735219

CrossRef Full Text | Google Scholar

Spence, M. (1973). Job market signaling. Q. J. Econ. 87 (3), 355–374. doi:10.2307/1882010

CrossRef Full Text | Google Scholar

Stiglitz, J. E., and Weiss, A. (1981). Credit rationing in markets with imperfect information. Am. Econ. Rev. 71 (3), 393–410. doi:10.2307/1802787

CrossRef Full Text | Google Scholar

Su, X., Qiao, R., and Xu, S. (2024). Impact of green finance on carbon emissions and spatial spillover effects: empirical evidence from China. J. Clean. Prod. 457, 142362. doi:10.1016/j.jclepro.2024.142362

CrossRef Full Text | Google Scholar

Wu, H., Guo, H., Zhang, B., and Bu, M. (2017). Westward movement of new polluting firms in China: pollution reduction mandates and location choice. J. Comp. Econ. 45 (1), 119–138. doi:10.1016/j.jce.2016.01.001

CrossRef Full Text | Google Scholar

Xia, Y., Luo, L., Ji, K., Huang, C., Wan, F., and Wang, Z. (2024). The impact of green finance and local regulations on industrial green innovation efficiency in China. Environ. Sci. Pollut. Res. 31 (2), 1980–1994. doi:10.1007/s11356-023-31314-w

PubMed Abstract | CrossRef Full Text | Google Scholar

Xu, S. (2020). International comparison of green credit and its enlightenment to China. Green Finance 2 (1), 75–99. doi:10.3934/gf.2020005

CrossRef Full Text | Google Scholar

Yadav, A., Gyamfi, B. A., Agozie, D. Q., Asongu, S. A., and Behera, D. K. (2025). “Unveiling the dynamics of green innovation on esg performance: the role of financial distress and the impact of the Paris agreement,” in Thunderbird international business review. Hoboken, NJ: Advance online publication.

Google Scholar

Yang, Z., and Lv, P. (2025). The spatial spillover effects of new infrastructure investment on sustainable innovation ecosystems: analysis based on provincial panel data in China. Digital Econ. Sustain. Dev. 3, 21. doi:10.1007/s44265-025-00072-7

CrossRef Full Text | Google Scholar

Yao, S., Pan, Y., Sensoy, A., Uddin, G. S., and Cheng, F. (2021). Green credit policy and firm performance: what we learn from China. Energy Econ. 101, 105415. doi:10.1016/j.eneco.2021.105415

CrossRef Full Text | Google Scholar

Yu, E. P.-y., Luu, B. V., and Chen, C. H. (2020). Greenwashing in environmental, social and governance disclosures. Res. Int. Bus. Finance 52, 101192. doi:10.1016/j.ribaf.2020.101192

CrossRef Full Text | Google Scholar

Zhang, Y., Xing, C., and Tripe, D. (2021). Redistribution of china’s green credit policy among environment-friendly manufacturing firms of various sizes: do banks value small and medium-sized enterprises? Int. J. Environ. Res. Public Health 18 (1), 33. doi:10.3390/ijerph18010033

PubMed Abstract | CrossRef Full Text | Google Scholar

Zhao, X., Zeng, S., Ke, X., and Jiang, S. (2023). The impact of green credit on energy efficiency from a green innovation perspective: empirical evidence from China based on a spatial durbin model. Energy Strategy Rev. 50, 101211. doi:10.1016/j.esr.2023.101211

CrossRef Full Text | Google Scholar

Zhao, N., Zhao, M., Long, H., and Yuan, L. (2025). Facing or evading? The impact of environmental taxes on the migration of heavily polluting enterprises in China. Environ. Dev. Sustain. 27 (5), 11773–11796. doi:10.1007/s10668-023-04381-9

CrossRef Full Text | Google Scholar

Zheng, Y., Wang, H., and Wang, S. (2025). Neighbourhood or beggar-thy-neighbour: the spatial effect of regional green credit development on green technology innovation. J. Environ. Plan. Manag. 68 (7), 1661–1690. doi:10.1080/09640568.2023.2295806

CrossRef Full Text | Google Scholar

Zitouni, R., Ghriss, O., Fguiri, A., and Jeday, M.-R. (2023). “Application of stochastic methods to estimate fouling heat exchanger,” in International conference on green energy conversion System (Springer), 323–331.

Google Scholar