Research on Certification of Zero-Carbon Plant Based on Accounting Statement Information

Abstract

Although financial statements are increasingly recognised as containing rich economic and governance information, the existing research on zero-carbon factory certification has primarily focused on external policy instruments, technological pathways, and governance mechanisms, while generally overlooking the role of internal corporate financial information systems in carbon emissions measurement and certification. To address this research gap, this study examines the validity, representational capacity, and underlying mechanisms of financial-statement-based carbon emission intensity as a proxy for actual corporate carbon emissions, using Shanghai and Shenzhen A-share listed companies from 2010 to 2022 as the sample. Findings reveal a significant positive correlation between the two, with this relationship positively moderated by the sustainability of corporate innovation outputs and market competitive position. This effect operates through optimising resource allocation efficiency and strengthening financing constraints. Heterogeneity analysis indicates that explicit environmental signals and engagement levels enhance this representational effect, whereas high green technology levels inhibit it. Furthermore, using the steel industry as a case study, this paper constructs a four-dimensional, nine-indicator zero-carbon factory evaluation system. Employing Entropy-VIKOR model to certify 37 A-share listed steel enterprises in 2022. Carbon emission intensity was identified as the core metric, with C₆ and C₁₅ serving as zero-carbon benchmarks. Evaluation results demonstrated high alignment with official certifications, providing empirical evidence and policy implications for assessing carbon emissions through financial information and establishing zero-carbon factory certification standards.

1. Introduction

China’s ‘dual carbon’ targets require all economic sectors to achieve verifiable emissions reductions, with the industrial sector bearing the greatest burden. As the core operational unit where energy consumption, production processes and carbon emissions converge, zero-carbon factories serve as the most tangible and quantifiable vehicle for translating the dual carbon targets into concrete emissions reduction actions [1]. In July 2025, three ministries jointly issued the ‘Notice on Promoting the Development of Zero-Carbon Industrial Parks’, emphasising that the accounting of industrial carbon emissions should be based on verifiable output-oriented methodologies. As shown in Figure 1, this policy establishes a comprehensive logical framework comprising ‘policy guidance—implementation vehicles—accounting support’, with zero-carbon factories serving as the central hub that closely links macro-level carbon targets with micro-level corporate actions [2]. However, the credibility of zero-carbon factory certification depends on accurate and reliable carbon emissions measurements, whilst direct carbon monitoring is costly and has limited coverage. Data from listed companies’ financial statements, which are subject to regular audits, follow a standardised format and are publicly available, thereby providing a data foundation for systematically calculating corporate carbon emission intensity that has yet to be fully exploited.

From a theoretical perspective, information asymmetry theory suggests that standardised financial disclosure can bridge the information gap between companies and external stakeholders, enabling the use of carbon emission intensity metrics derived from financial data to provide a reliable signal of corporate carbon emission levels. The resource-based view holds that a company’s carbon performance depends on its allocation of resources between high-carbon and low-carbon assets, a process that is systematically reflected through capital structure and operating cost metrics; signal theory further suggests that companies genuinely committed to emissions reduction have an incentive to convey verifiable carbon performance signals through financial reporting channels. In summary, carbon emission intensity calculated from financial statements should be understood as a proxy for corporate carbon emissions rather than a perfect substitute for directly monitored physical emissions. Its value lies in the fact that audited accounting data provide standardised, continuous, and broadly comparable information on firms’ production scale, cost structure, and resource allocation, all of which are closely related to emission-generating activities. At the same time, this proxy may be subject to measurement error because accounting-based estimates cannot fully capture firm-level differences in fuel mix, process technology, and boundary definition. For this reason, the key empirical task of this paper is not to assume perfect accuracy, but to test whether this proxy exhibits sufficient validity, robustness, and practical usefulness relative to actual emissions outcomes, especially in settings where direct emissions disclosure remains incomplete, inconsistent, or vulnerable to selective reporting.

Existing research has systematically examined the impact of policy instruments, environmental regulation, and technological pathways on zero-carbon transition and factory certification [3]. A related stream of literature has explored the connection between ESG disclosure, sustainability reporting, and environmental performance, showing that financial and non-financial disclosures can convey useful information about firms’ carbon-related risks and green transition outcomes [4]. However, these studies mainly focus either on external disclosure quality or on the policy consequences of carbon governance, and pay relatively limited attention to whether routine financial statement information itself can serve as a practical accounting-based proxy for actual corporate emissions. As a result, the carbon accounting literature still lacks a clear framework linking audited financial statement indicators, actual emissions outcomes, and zero-carbon factory certification in a unified analytical setting.

This gap is theoretically important for two reasons. First, the existing ESG and carbon disclosure studies mainly evaluate the informational role of voluntarily or mandatorily reported sustainability data, whereas this paper shifts attention to the carbon-accounting value embedded in conventional financial statements. Second, prior research rarely connects the proxy validity of accounting-based carbon intensity to its governance conditions, transmission mechanisms, and certification application. By doing so, this study moves beyond asking whether disclosure matters and instead examines whether standard financial statement information can function as a scalable, empirically informative foundation for carbon accounting and zero-carbon factory evaluation.

Based on this, this paper examines whether accounting statement information can effectively reflect a company’s carbon emissions, particularly in the context of China’s progressively standardised and regulated carbon accounting system. In addition to testing the baseline association, the empirical design explicitly addresses potential endogeneity arising from reverse causality (see [5,6] for related discussion), omitted variables, and simultaneity between carbon emission intensity and actual emissions. To strengthen causal identification, the study combines firm and year fixed effects with a set of complementary econometric strategies, including lag-based instrumental-variable regressions (see [7,8] for applications), system GMM (see [9] for applications) estimation, placebo tests (see [10,11] for applications), additional governance controls, and pre-pandemic subsample analyses. This design enables the paper not only to assess whether financial-statement-based carbon emission intensity is statistically associated with actual emissions, but also to provide more credible evidence on the direction and robustness of that relationship.

Against this background, this study makes the following contributions:

(1)

This study reveals an intrinsic link between financial statement information and carbon emissions. Analysis of data from Shanghai and Shenzhen A-share listed companies between 2010 and 2022 indicates a significant positive correlation between corporate carbon emission intensity and total carbon emissions. Specifically, companies exhibiting higher carbon emission intensity in their financial statements tend to report greater overall carbon emissions.

(2)

This study introduces moderating factors: the sustainability of corporate innovation outputs and the competitive position of enterprises strengthen the positive relationship between corporate carbon emission intensity and corporate carbon emissions. That is, the higher the levels of these two factors, the more effectively carbon emission intensity reflects the actual corporate carbon emissions of enterprises.

(3)

This study validates two mechanisms—enterprise resource allocation efficiency and financing constraints—as partially mediating the relationship between carbon emission intensity and corporate carbon emissions. It explains how carbon emission intensity reflects actual corporate carbon emissions through resource allocation efficiency and financing constraints.

(4)

This study reveals the heterogeneous relationship between carbon emission intensity and corporate carbon emissions. When enterprises send clear environmental signals and demonstrate strong environmental engagement, carbon emission intensity effectively reflects their actual corporate carbon emissions. Conversely, high levels of green technology within enterprises suppress this positive relationship.

(5)

This study translates empirical findings into practical applications by establishing a zero-carbon factory evaluation framework. Employing the Entropy-VIKOR model, it certifies steel enterprises as zero-carbon factories, identifies industry benchmarks for carbon neutrality, and validates the practical applicability of conducting such certification based on financial statement data. This provides a quantifiable and replicable reference framework for developing zero-carbon factory certification standards.

This study refines the analytical framework linking financial statement information to corporate carbon emissions, providing empirical evidence for assessing corporate carbon emissions using publicly available financial data.

The structure of this paper is as follows: Section 2 reviews relevant literature and proposes research hypotheses; Section 3 outlines sample selection, variable definitions, and empirical methodologies; Section 4 presents benchmark regression results alongside a series of robustness tests; Section 5 further explores policy implications through heterogeneity analysis of corporate environmental signals, competitive positions, and green technology levels; Section 6 constructs a zero-carbon factory evaluation framework based on prior empirical findings, conducts an empirical assessment of zero-carbon factory certification using the steel industry as a case study, and analyses the evaluation outcomes; Section 7 summarises the core findings and offers recommendations for corporate practice and policy formulation.

2. Literature Review and Research Hypotheses

The theoretical framework of this study is designed to extend the existing carbon accounting literature from a static disclosure perspective to a broader accounting–governance–certification perspective. Rather than treating carbon-related information merely as an ESG disclosure outcome, this paper conceptualises financial-statement-based carbon emission intensity as an accounting proxy whose validity depends on governance conditions, transmission mechanisms, and application context. Accordingly, the framework is organised as a four-layer progressive structure. The first layer establishes the baseline hypothesis that carbon emission intensity positively maps onto corporate carbon emissions. The second layer identifies two governance conditions, innovation output sustainability and competitive position, that moderate the strength of this mapping. The third layer unpacks the internal transmission channels through which carbon emission intensity translates into actual emissions via resource allocation efficiency and financing constraints. The fourth layer examines heterogeneity by analysing how environmental signals and engagement strengthen the baseline relationship while high green technology levels attenuate it. These four layers collectively form a unified analytical chain from “whether” to “when,” “how,” and “for whom,” thereby clarifying the theoretical contribution of the paper beyond a simple correlation test.

2.1. Baseline Relationship Between Carbon Emission Intensity and Corporate Carbon Emissions

Theoretically, a company’s carbon emissions are not isolated physical processes but rather the byproduct of converting natural resources into economic value. This “input–output” transformation process is captured by accounting systems through monetary measurement. Based on the logic of accounting–physical isomorphism, financial statements are not merely static records of financial conditions but also digital representations of a company’s underlying production functions and resource allocation efficiency.

It is worth noting that the informational role of carbon emission intensity in financial statements is shaped by the broader financial reporting environment. Under the International Financial Reporting Standards framework, there has long been a lack of uniform guidance on the accounting treatment of carbon emissions; the withdrawal of the 2005 International Financial Reporting Interpretations Committee Interpretation No. 3, ‘Emission Rights’, is a prime example of this, and this has directly led to significant differences in the recognition and disclosure of carbon-related costs across different jurisdictions. The issuance of IFRS S2 by the International Sustainability Standards Board in 2023 marks an important step towards harmonisation, as it explicitly requires the disclosure of both absolute greenhouse gas emissions and emission intensity metrics within a framework connected to general-purpose financial reporting [12]. This evolving convergence between sustainability disclosure and financial reporting standards suggests that carbon emission intensity figures derived from accounting systems are becoming increasingly reliable proxies for actual corporate emission levels, lending institutional support to the analytical approach adopted in this paper.

First, from the perspective of information mapping and disclosure effects, financial statements establish an information infrastructure connecting a company’s physical carbon footprint with external observers. Standardized accounting disclosures can significantly reduce information asymmetry. Mandatory or voluntary non-financial disclosures generate a substantial “veridical effect,” meaning changes in reported data are often accompanied by adjustments in actual business operations. Accounting information systems transform ambiguous climate risks into quantifiable financial metrics through regular reporting mechanisms. In the realm of carbon emissions, greenhouse gas disclosure mechanisms compel companies to internalize environmental externalities as accounting costs. The carbon emission intensity figures reported in financial statements fundamentally reflect both a company’s carbon management capabilities and its exposure to environmental risks. Despite potential measurement error in environmental data, the market is increasingly able to distinguish genuine signals from accounting disclosures [13]. When financial statements indicate high carbon emission intensity, this suggests that the carbon cost per unit of output is relatively high. Based on these publicly available intensity signals, institutional investors assess whether a company is exposed to risks arising from insufficient emission-reduction technologies [14]. Accordingly, such a high-intensity signal is likely to correspond to higher actual corporate carbon emissions.

Second, from the perspective of resource allocation and production efficiency, carbon emission intensity serves as a key proxy variable for measuring a company’s asset utilization efficiency and technological capabilities. Based on the resource-based view, a company’s carbon performance hinges on whether it possesses advanced low-carbon technological assets and efficient operational management capabilities. Long-term value creation depends on companies allocating resources to efficient, sustainable assets. Capital markets can effectively identify substantive operational improvements in enterprises through publicly available ESG news and reporting information [15]; highly efficient corporate capital allocation often manifests in financial statements as superior “green” financial characteristics [16]. Conversely, high carbon emission intensity in financial statements essentially reveals a company’s “high energy consumption, low output” characteristics in resource allocation. Such enterprises are often deeply entrenched in the path dependency of high-carbon assets. Companies with high carbon emission intensity face immense financing pressures due to higher “carbon premiums” [17], while green assets and brown assets exhibit significant divergence in expected returns [18]; the uncertainty and tail risks associated with carbon emission intensity constrain companies’ financial capacity to undertake large-scale technological upgrades, forcing them to maintain high-emission production models [19]. Therefore, carbon emission intensity in financial statements is not merely a ratio metric; it fundamentally anchors a company’s position within the distribution of total carbon emissions. This reasoning is jointly supported by three theoretical pillars: information asymmetry theory predicts that standardised financial disclosure reduces the gap between reported and actual performance; the resource-based view holds that observable capital allocation patterns determine physical emission outcomes; and signal theory implies that carbon emission intensity constitutes a credible, costly-to-fake signal of a firm’s true emission profile [20]. Higher intensity indicates that the company consumes more physical resources that generate carbon emissions relative to its scale, thereby driving up the absolute value of its actual total carbon emissions.

Based on the above evidence, this paper proposes Hypothesis 1.

H1. 

Carbon emission intensity is positively correlated with corporate carbon emissions, meaning that the higher the carbon emission intensity in financial statements, the greater the carbon emissions of the enterprise.

2.2. Moderating Conditions of Innovation Output Sustainability and Competitive Position

As a core indicator of innovation activity stability and continuity, the sustainability of corporate innovation outputs not only reflects long-term strategic orientation in technological R&D and achievement commercialization, but also has the potential to profoundly regulate the relationship between carbon emission intensity and actual corporate emissions by influencing the effectiveness of carbon emission intensity accounting. On the one hand, innovation serves as the core driver for enterprises to optimize their carbon emission performance. Whether through exploratory major technological breakthroughs or incremental process improvements, sustained innovation output can drive enhancements in energy efficiency, the development of low-carbon technologies, or the optimization of production processes. This, in turn, enables changes in carbon emission intensity to better align with fluctuations in the enterprise’s actual emission scale. Especially during times of crisis, sustained investment in specific types of innovation has proven to be key to an organization’s resilience and performance [21]. On the other hand, the sustainability level of corporate innovation outputs determines the representativeness of carbon emission intensity for actual corporate carbon emissions. High-level innovation output sustainability typically implies that enterprises possess stronger green absorption capacity, clearer long-term environmental strategies, and higher levels of trust established with stakeholders. This stability helps drive innovation outcomes to continuously impact carbon emission control processes, enabling carbon emission intensity to more accurately reflect actual corporate emissions.

Conversely, if an enterprise’s innovation output lacks sustainability, even with phased innovation achievements, deviations may occur between carbon emission intensity and actual corporate carbon emissions due to fragmented innovation activities, strategic myopia, or temporary adjustments in resource allocation. For instance, without sustained innovation to underpin it, incremental efficiency gains may serve only short-term cost control—or even trigger a “green paradox.” This occurs when localized efficiency improvements stimulate overall resource consumption expansion, leading to reduced carbon emission intensity but increased actual emissions [22]. In such cases, carbon emission intensity fails to accurately reflect true emission levels. Moreover, the sustainability of high-level innovation outputs helps alleviate financing constraints and uncertainties in the innovation process, providing stable resource support for the long-term R&D and scaled application of low-carbon technologies. From the perspective of dynamic capability theory, sustained innovation output reflects a firm’s capacity to continuously reconfigure its resource base in response to environmental pressures [23], ensuring that efficiency gains are structurally embedded rather than episodic. This further ensures that carbon emission intensity can consistently and accurately reflect a company’s actual corporate carbon emissions.

Based on this reasoning, the following hypothesis is proposed:

H2a. 

The sustainability of corporate innovation outputs significantly and positively moderates the relationship between corporate carbon emission intensity and corporate carbon emissions. That is, the higher the level of sustainability in corporate innovation outputs, the more likely carbon emission intensity is to effectively reflect the company’s actual corporate carbon emissions.

A company’s competitive position is its relative advantage within the industry, based on its unique combination of resources, market positioning, and dynamic capabilities [24]. This position not only determines a company’s profitability and room for survival, but also profoundly influences its willingness and ability to comply with environmental regulations and make long-term strategic investments. The resource-based view and dynamic capability theory indicate that enterprises with highly competitive positions typically possess more abundant financial resources, stronger technological absorption and transformation capabilities, and more solid market influence [25]. This provides them with the critical material and organizational foundations for establishing standardized carbon emission monitoring systems and optimizing production process controls, thereby enabling carbon emission intensity to more accurately reflect actual emission scales. For instance, enterprises with strong competitive positions can more effectively translate external technologies into standardized upgrades of internal processes. Through robust management and control mechanisms, they make it more likely that changes in carbon emission intensity align with corporate carbon emission levels. Conversely, companies with weaker competitive positions often face more severe resource constraints and survival pressures. Their strategic decisions tend to prioritize short-term survival, potentially simplifying carbon emission monitoring and control processes. This can result in carbon emission intensity failing to accurately reflect corporate carbon emission levels.

Specifically, the competitive position of enterprises modulates the effectiveness of the relationship between carbon emission intensity and actual corporate carbon emissions through the following two mechanisms: first, from the perspective of resource buffers, a superior competitive position provides enterprises with a “resource buffer” for optimizing carbon emission management. Such enterprises can allocate a portion of their excess profits or redundant resources toward enhancing carbon emission monitoring equipment, optimizing energy consumption statistical systems, and advancing production process standardization. Through refined management and control, they can ensure carbon emission intensity calculations more accurately reflect corporate carbon emission levels. Second, from a strategic response perspective, a highly competitive position is often associated with stronger market influence and brand value. Maintaining or enhancing a green image is crucial to the long-term competitive advantage of such enterprises. Therefore, faced with compliance and oversight pressures related to carbon emissions, they have greater motivation to establish systematic carbon emission management systems. This ensures that carbon emission intensity effectively reflects corporate carbon emission levels, thereby mitigating reputational risks and solidifying their market leadership position. In contrast, companies with low competitive position may lack the motivation and capacity to allocate resources toward optimizing carbon emission management due to their small market share and low brand premium. Their carbon emission statistics often aim only to meet basic compliance requirements, leading to a disconnect between intensity data and actual emissions. This makes it difficult for intensity metrics to accurately reflect their real-world emission profiles. Consequently, a firm’s competitive advantage significantly enhances the accuracy of carbon emission intensity data in financial statements through a dual pathway; as quantitative indicators of a firm’s carbon emissions, the accuracy of these data stems from resource buffering mechanisms and strategic response measures. This dual-pathway logic is consistent with both the resource-based view and legitimacy theory.

Based on the above theoretical mechanism, the following hypothesis is proposed:

H2b. 

A firm’s competitive position significantly and positively moderates the relationship between its carbon emission intensity and total carbon emissions. That is, the higher a firm’s competitive position, the more likely its carbon emission intensity is to effectively reflect its actual corporate carbon emissions.

2.3. Mediating Mechanisms of Resource Allocation Efficiency and Financing Constraints

The capital structure in financial statements serves as a core proxy variable for resource allocation efficiency. At its core, it represents the allocation decision between high-carbon production capacity and low-carbon transition. Enterprises with high carbon emission intensity are predominantly energy-intensive types, exhibiting strong dependence on traditional high-carbon production pathways [26]. Capital tends to become locked in maintaining high-carbon assets, while the long investment cycles and high uncertainty of low-carbon initiatives lead them to avoid related investments, resulting in inefficient resource allocation. The high-carbon bias in capital structure directly drives increases in energy consumption and total emissions, while resource misallocation further exacerbates pollution through excessive investment [27]. Conversely, directing capital toward low-carbon sectors can reduce emissions through technological upgrades [28]. It is evident that carbon emission intensity regulates resource allocation efficiency through capital structure, thereby influencing carbon emissions. This mechanism is theoretically grounded in the resource-based view’s proposition that heterogeneous asset allocation determines productive outcomes, and is empirically supported by evidence that capital misallocation between high-carbon and low-carbon sectors significantly amplifies industrial pollution [29].

Based on this reasoning, the following hypothesis is proposed:

H3a. 

Carbon emission intensity effectively reflects a company’s actual corporate carbon emissions by optimizing resource allocation efficiency.

The financing constraints reflected in financial statements serve as a key proxy variable for financing constraints, aligning with the capital-dependent nature of the low-carbon transition. At their core, these constraints represent the debt market’s pricing of corporate carbon risk. High carbon emission intensity implies greater carbon risk, leading financial institutions to tighten credit and shorten debt maturities. Notably, debt financing costs for companies in high-carbon industries rise alongside increasing carbon risk, exacerbating financing constraints. Strengthening financing constraints may limit corporate investment in low-carbon technologies, forcing them to maintain high-carbon production or even expand high-carbon capacity [30]. Conversely, easing such constraints can support low-carbon transitions through long-term debt financing, consistent with findings that alleviating financing constraints aids emissions reduction. It is evident that carbon emission intensity indirectly influences carbon emissions by adjusting financing constraints. This channel aligns with carbon risk pricing theory, which demonstrates that debt markets systematically penalise high-carbon firms through elevated borrowing costs [31], and with the financial constraint literature showing that restricted access to external capital limits firms’ ability to undertake low-carbon technological transitions [32].

Based on this reasoning, the following hypothesis is proposed:

H3b. 

Carbon emission intensity effectively reflects a company’s actual corporate carbon emissions by strengthening financing constraints.

2.4. Heterogeneity Analysis of Environmental Signals and Green Technology

If companies can send clearer environmental signals to stakeholders, it will help standardize their carbon emissions accounting processes and ensure that carbon data accurately reflects actual production and operational activities. Therefore, clear environmental signals from enterprises generate a positive signal verification effect. By reducing greenwashing-style data manipulation and enhancing the accuracy of carbon emission intensity accounting, such signals reinforce the positive impact of carbon emission intensity on actual emissions. Companies with strong environmental engagement are more likely to standardize carbon emissions data accounting and management to align with their environmental strategies or meet stakeholder demands, ensuring carbon emission intensity accurately reflects real-world emissions. Therefore, when a company demonstrates strong environmental commitment, carbon emissions intensity is more closely linked to the volume of carbon emissions, making improvements in intensity have a more pronounced effect on emission levels. This prediction is rooted in signal theory: voluntary environmental certifications and executive green backgrounds constitute costly, hard-to-imitate signals that reduce greenwashing incentives and enhance the veridical quality of reported carbon data [33].

Based on this reasoning, the following hypothesis is proposed:

H4a. 

When enterprises can send clear environmental signals and demonstrate strong environmental engagement, this promotes the effective reflection of corporate carbon emissions intensity in their actual corporate carbon emissions.

All other factors being equal, the higher a company’s level of green technology, the stronger its capabilities in monitoring and controlling carbon emissions, and the more advanced its energy-saving and carbon-reduction technologies and processes. This technological advantage does not reinforce the correspondence between carbon emission intensity and actual corporate carbon emissions; rather, it disrupts the former’s accuracy in representing the latter. High-level green technologies disrupt the inherent logical correspondence between carbon emission intensity and actual emissions, creating a “distortion mechanism.” This process may also give rise to the green paradox: after enterprises enhance energy efficiency and reduce carbon emission intensity per unit of output through green technologies, the resulting cost advantages may stimulate production scale expansion, ultimately leading to increased rather than decreased actual corporate carbon emissions. Under such circumstances, the decline in carbon emission intensity fails to accurately reflect changes in a company’s actual emissions, potentially leading to a disconnect where “intensity optimization coincides with an increase in total emissions.” This implies that the higher the level of green technology, the less effectively carbon emission intensity reflects a company’s true carbon footprint, significantly diminishing its representational value. This distortion mechanism is theoretically consistent with the Jevons paradox and the rebound effect literature [34], which demonstrates that technological efficiency gains can stimulate output expansion sufficient to offset or reverse the intended emissions reductions, thereby decoupling intensity metrics from aggregate emission trajectories.

Based on this reasoning, the following hypothesis is proposed:

H4b. 

The higher a company’s level of green technology, the less effectively its carbon emission intensity reflects its actual corporate carbon emissions.

2.5. Map of Impact Mechanisms

Based on the aforementioned theoretical analysis and research hypotheses, we have derived the following diagram illustrating the influencing mechanisms, as shown in Figure 2.

3. Research Design

3.1. Sample Selection and Data Sources

This study selected A-share listed companies on the Shanghai and Shenzhen stock exchanges from 2010 to 2022 as research samples. To ensure data reliability, the following exclusion criteria were applied: (1) exclusion of listed companies in the financial sector; (2) exclusion of listed companies with missing key data; (3) exclusion of ST and *ST companies whose abnormal financial conditions could distort carbon-related estimations; (4) exclusion of companies listed for fewer than one full year, as incomplete data preclude the construction of lagged variables. All remaining continuous variables were winsorised at the 1st and 99th percentiles. The final sample comprised 2503 listed companies and 25,383 firm-year observations. It should be noted that restricting the sample to listed companies subject to mandatory auditing may introduce survivorship bias (see [35,36], for instance) and may represent an upper bound for the reliability of financial carbon metrics, thereby limiting direct generalisation to non-listed enterprises. During the preparation of this manuscript, ChatGPT (GPT-5.4, OpenAI) was used for language polishing and translation. The authors reviewed and edited all AI-assisted output and take full responsibility for the content of the publication.

3.2. Selection of Variable

3.2.1. Dependent Variable: Corporate Carbon Emissions (CEE)

Because relatively few listed firms disclose consistent and verifiable firm-level carbon emissions, and voluntary disclosure may be affected by selection bias (refer to [37,38] for similar discussions) or greenwashing incentives, this paper follows Jiang et al. to construct an estimated measure of corporate carbon emissions [39]. Specifically, a firm’s share of annual operating costs within its industry is used as the weighting factor and multiplied by the industry’s total annual carbon emissions to obtain an approximate firm-level emissions value, which is then converted into carbon dioxide equivalents and log-transformed to form CEE. According to the carbon dioxide conversion standard of the Xiamen Energy Conservation Center, the conversion factor for one metric ton of standard coal is 2.493.

This measure is not intended to replace verified physical emissions data. Rather, it provides a consistent and scalable approximation under conditions of limited disclosure. Its main advantage is broad sample coverage and comparability across firms and years, whereas its main limitation is that it may smooth over firm-level heterogeneity in production technology and energy structure within the same industry. We therefore interpret CEE as an estimated benchmark for empirical validation rather than as a perfect physical measure.

3.2.2. Explanatory Variable: Carbon Emission Intensity (CEI)

The continuous variable measuring firm-level carbon emission intensity follows Chapple et al. [40] and is defined as the ratio of a firm’s carbon dioxide emissions to its main business revenue. The specific formula is as follows:

$$\mathrm{C}\mathrm{E}\mathrm{I}={\displaystyle \frac{\mathrm{C}\mathrm{a}\mathrm{r}\mathrm{b}\mathrm{o}\mathrm{n} \mathrm{d}\mathrm{i}\mathrm{o}\mathrm{x}\mathrm{i}\mathrm{d}\mathrm{e} \mathrm{e}\mathrm{m}\mathrm{i}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{s}}{\mathrm{P}\mathrm{r}\mathrm{i}\mathrm{m}\mathrm{a}\mathrm{r}\mathrm{y} \mathrm{R}\mathrm{e}\mathrm{v}\mathrm{e}\mathrm{n}\mathrm{u}\mathrm{e} \mathrm{S}\mathrm{t}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{m}\mathrm{s}\times \mathrm{1,000,000}}}.$$

This indicator is designed to capture the carbon burden associated with a firm’s economic output on the basis of publicly available accounting information. Compared with existing carbon disclosure databases, the main advantage of this approach is that it relies on regularly audited financial statement items with relatively stable reporting standards and much broader firm coverage, especially in the Chinese listed-company context where direct emissions disclosure is still incomplete. However, because CEI is constructed from accounting-based data, it may contain measurement bias relative to actual monitored emissions, particularly when firms differ substantially in production technology, energy mix, or reporting boundaries. For this reason, the empirical analysis below focuses on whether CEI exhibits stable explanatory power and robustness, rather than assuming that it perfectly captures actual emissions.

3.2.3. Control Variables

The following control variables related to accounting information were selected: Cash Flow Ratio (Cashflow), Return on Assets (ROA), Book-to-Market Value Ratio (BMV), Firm Age (AGE), and Firm Size (SIZE). To avoid the impact of differences in energy utilization and carbon emission efficiency resulting from varying production and operational methods across industries on research conclusions, this paper incorporates industry carbon emission intensity (IEI) as a control variable. Enterprises classified under the six high-energy-consuming industries listed in the 2010 National Economic and Social Development Statistical Bulletin are defined as high-carbon-intensity enterprises, with a value of 1; all other enterprises are defined as low-carbon-intensity enterprises, with a value of 0.

3.2.4. Moderating Variables

The selection of moderating variables is guided by two complementary governance dimensions that shape the reliability of financial carbon accounting. Innovation output sustainability captures the firm’s internal governance capacity for sustained R&D commitment—determining whether carbon emission intensity improvements are structurally embedded or merely transient. Competitive position reflects the external market discipline and resource advantages that enable firms to invest in standardised carbon monitoring systems. Together, these two variables represent the internal and external governance conditions under which financial carbon emission intensity most accurately maps onto actual emissions.

Firstly, the organisation’s innovation output sustainability (OIP). In this paper, we measure the organisation’s innovation persistence [41]. The specific formula (1) is as follows:

$${\mathrm{O}\mathrm{I}\mathrm{P}}_{\mathrm{t}}={\displaystyle \frac{{\mathrm{O}\mathrm{I}\mathrm{N}}_{\mathrm{t}}}{{\mathrm{O}\mathrm{I}\mathrm{N}}_{\mathrm{t}-1}}}\times {\mathrm{O}\mathrm{I}\mathrm{N}}_{\mathrm{t}}$$

(1)

Here, ${\mathrm{O}\mathrm{I}\mathrm{P}}_{\mathrm{t}}$ denotes the sustainability of an enterprise’s innovation output in year t, where ${\mathrm{O}\mathrm{I}\mathrm{N}}_{\mathrm{t}}$ represents the sum of the enterprise’s innovation outputs in years $\mathrm{t}$ and $\mathrm{t}-1$, while ${\mathrm{O}\mathrm{I}\mathrm{N}}_{\mathrm{t}-1}$ denotes the sum of its innovation outputs in years $\mathrm{t}-1$ and $\mathrm{t}-2$.

This metric integrates two key dimensions of innovation: it reflects the growth rate of innovation output through the ratio of consecutive cumulative output, while simultaneously demonstrating the absolute scale of innovation activities by multiplying the current output level. This dual approach provides a comprehensive metric for gauging both the momentum and scale of an organisation’s innovation endeavour. The three-year time window enables this metric to withstand annual fluctuations, thereby providing a balanced perspective on short-term sustainability. The application of patent data in innovation research is widely recognised, with its objectivity and broad availability rendering it a reliable surrogate indicator. Therefore, the OIP indicator provides a robust assessment of the consistency in innovation output generated by enterprises.

Secondly, the company’s competitive position (Pcm). A company’s competitive position refers to the relative strengths and weaknesses it exhibits compared to other competitors within the broader market environment. This comparison encompasses not only market share, brand influence, cost control capabilities, and the quality of customer service, but also profoundly reflects a company’s competitive position and strategic positioning within the marketplace. This paper follows the methodology of Peress et al. in employing the Lerner index to measure a firm’s competitive position. The Lerner index is calculated as the ratio of net revenue—defined as operating revenue minus cost of sales, selling expenses, and administrative expenses—to total operating revenue [42]. The Lerner index reflects the strength of monopoly power in a market by measuring the degree of deviation between prices and marginal costs. The higher the index, the stronger the competitive position of the firm.

3.2.5. Mechanism Variables

The two mechanism variables are chosen to capture the principal governance channels through which carbon emission intensity transmits to actual emissions within the corporate financial system. Resource allocation efficiency reflects the capital governance decision between maintaining high-carbon assets and pursuing low-carbon transition—a choice that is directly observable in the firm’s liability structure. Financing constraints capture the external debt governance mechanism through which creditors price corporate carbon risk, thereby disciplining emissions behaviour via capital market signals. These two channels jointly represent how internal capital governance and external debt governance mediate the carbon intensity–emissions relationship.

Resource allocation efficiency (LEV). This study adopts the research design proposed by Chen et al., employing the debt-to-asset ratio (total liabilities/total assets) as a measure of resource allocation efficiency [43]. This indicator characterises the structure of capital utilisation within enterprises. Research by Chen et al. confirms that reducing resource misallocation is pivotal to enhancing total factor productivity. Consequently, the debt-to-asset ratio may serve as a reliable proxy for resource allocation efficiency.

Financing constraints (KZ). Common methods and indicators for measuring corporate financing constraints include the SA index, KZ index, WW index, and key financial ratios. This study employs the KZ index proposed by Wei et al. as the primary measurement indicator, balancing the principles of comparability and comprehensiveness [44]. This index integrates multi-dimensional corporate financial information including cash flow, dividend payout ratio, leverage ratio, and Tobin’s Q ratio, thereby providing a more comprehensive reflection of the financing constraints faced by energy enterprises. To enhance the robustness of the research findings. For detailed information on the indicators, please refer to Appendix A.

The main variables in this paper are defined as shown in

Table 1. Variables.

Variable Type	Variable Name	Variable Symbol	Variable Definition
Dependent Variable	Corporate Carbon Emissions	CEE	Enterprise approximate carbon emissions multiplied by the carbon dioxide conversion factor
Explanatory Variable	Carbon Emissions Intensity	CEI	Enterprise carbon dioxide emissions/Main business revenue × 1,000,000
Moderator Variable	Sustainability of Innovation Outputs	OIP	OIP denotes the sustainability of an enterprise’s innovation output in year t.
	Competitive Position of Enterprises	Pcm	Lerner Index = (Revenue − Cost of Goods Sold − Selling Expenses − Administrative Expenses)/Revenue
Mechanism Variables	Debt-to-Asset Ratio	LEV	Measuring the efficiency of corporate resource allocation: Total liabilities/Total assets
	Financing constraints	KZ	Assessing corporate debt constraints: Financing constraints based on operating cash flow, cash dividends, cash holdings, leverage ratio, and Tobin’s Q ratio
Control Variable	Cash Flow Ratio	Cashflow	Net cash flows from operating activities/total assets
	Return on Assets	ROA	Net profit/Net assets
	Book-to-market ratio	BMV	Enterprise total market capitalisation/Shareholders’ equity
	Company Age	AGE	Vintage Year − Listing Year + 1, then take the logarithm
	Industry Carbon Emission Intensity	IEI	Virtual variable: takes the value 1 if the enterprise belongs to one of the six high-energy-consuming industries listed in the 2010 National Economic and Social Development Statistical Bulletin, otherwise 0.
	Enterprise size	SIZE	The natural logarithm of the enterprise’s total assets at the end of the previous year

.

3.3. Empirical Methodology

The empirical strategy follows a four-stage framework: baseline two-way fixed effects regression tests H1; interaction models test the moderating roles of innovation sustainability and competitive position (H2a, H2b); mechanism regressions examine partial mediation through resource allocation efficiency and financing constraints (H3a, H3b); and heterogeneity interaction models assess boundary conditions (H4a, H4b). Individual and year fixed effects absorb time-invariant firm heterogeneity and common macroeconomic shocks. Endogeneity is addressed through lagged instruments, system GMM, pre-pandemic subsample regressions, additional controls, and placebo tests.

3.3.1. Baseline Regression Model

To examine the relationship between carbon emission intensity and corporate carbon emissions in financial statements, this paper employs Model (2) for analysis:

$${CEE}_{i,t}={\alpha }_0+{\alpha }_1{CEI}_{i,t}+{\alpha }_2{Controls}_{i,t}+{\lambda }_i+{{\rm T}}_t+{ℇ}_{i,t}$$

(2)

Among these, the dependent variable ${\mathrm{C}\mathrm{E}\mathrm{E}}_{\mathrm{i},\mathrm{t}}$ denotes the total carbon emissions of firm $\mathrm{i}$ in year $\mathrm{t}$. ${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$ represents the carbon emission intensity of firm $\mathrm{i}$ in year $\mathrm{t}$. ${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}\mathrm{s}}_{\mathrm{i},\mathrm{t}}$ denotes a set of control variables. ${\mathrm{\lambda }}_{\mathrm{i}}$ and ${{\rm T}}_t$ respectively represent individual and time fixed effects. ${ℇ}_{i,t}$ denotes the random disturbance term. ${\mathrm{\alpha }}_1$ represents the parameter to be estimated in this study. When ${\mathrm{\alpha }}_1$ is significantly positive, it indicates that carbon emission intensity is positively correlated with corporate carbon emissions; that is, the higher the carbon emission intensity reported in a company’s financial statements, the greater its carbon emissions.

3.3.2. Moderation Effect Model

To examine the moderating effect of the sustainability of corporate innovation outputs on the relationship between carbon emission intensity and corporate carbon emissions, Model (3) was constructed:

$${CEE}_{i,t}={\beta }_0+{\beta }_1{CEI}_{i,t}+{\beta }_2{CEI}_{i,t}\times {OIP}_{i,t}+{\beta }_3{OIP}_{i,t}+{\beta }_4{Controls}_{i,t}+{\lambda }_i+{{\rm T}}_t+{ℇ}_{i,t}$$

(3)

OIP denotes the moderating variable for firm i’s innovation output in year t. CEI × OIP represents the interaction term between carbon emission intensity and innovation output. The specification of other variables remains consistent with Model (2).

To examine the moderating effect of a firm’s competitive position on the relationship between carbon emission intensity and corporate carbon emissions, the following model (4) is constructed:

$${CEE}_{i,t}={\gamma }_0+{\gamma }_1{CEI}_{i,t}+{\gamma }_2{CEI}_{i,t}\times {Pcm}_{i,t}+{\gamma }_3{Pcm}_{i,t}+{\gamma }_4{Controls}_{i,t}+{\lambda }_i+{{\rm T}}_t+{ℇ}_{i,t}$$

(4)

Pcm denotes the moderating variable representing firm i’s competitive position in year t.$\mathrm{C}\mathrm{E}\mathrm{I}\times \mathrm{P}\mathrm{c}\mathrm{m}$ represents the interaction term between carbon emission intensity and corporate competitive position. The settings for other variables are consistent with those in Model (2).

3.3.3. Mechanism Test Model

This study analyses the causal relationship between corporate carbon emission intensity and institutional variables based on existing academic findings. The institutional testing model is presented as shown in Model (5):

$$M_{i,t}={\mu }_0+{\mu }_1{CEI}_{i,t}+{\mu }_2{Controls}_{i,t}+{\lambda }_i+{{\rm T}}_t+{ℇ}_{i,t}$$

(5)

In this model, ${\mathrm{M}}_{\mathrm{i},\mathrm{t}}$ denotes the mechanism variable. The settings for other variables are consistent with those in model (2):

3.3.4. Heterogeneity Test Model

To examine heterogeneity in the relationship between carbon emission intensity and corporate carbon emissions, this paper further investigates the moderating effect model through the construction of Model (6):

$${\mathrm{C}\mathrm{E}\mathrm{E}}_{\mathrm{i},\mathrm{t}}={\mathrm{\eta }}_0+{\mathrm{\eta }}_1{\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}\times {\mathrm{F}}_{\mathrm{i},\mathrm{t}}+{\mathrm{\eta }}_2{\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}+{\mathrm{\eta }}_3{\mathrm{F}}_{\mathrm{i},\mathrm{t}}+{\mathrm{\eta }}_4{Controls}_{i,t}+{\mathrm{\lambda }}_{\mathrm{i}}+{{\rm T}}_t+{ℇ}_{i,t}$$

(6)

In this context, Model (6) not only controls for the core explanatory variable ${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$ but also incorporates the characteristic F, which may influence how carbon emission intensity in financial statements reflects a firm’s actual corporate carbon emissions. The symbols for other variables retain the same meaning as in Model (2). Model (6) focuses on examining the coefficient ${\mathrm{\eta }}_1$ of the cross-product term (${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}\times {\mathrm{F}}_{\mathrm{i},\mathrm{t}}$). When ${\mathrm{\eta }}_1$ is significantly positive, it indicates that the firm’s characteristic F facilitates the reflection of carbon emission intensity on carbon emissions in the financial statements.

4. Empirical Analysis

4.1. Descriptive Statistics

Table 2. Descriptive statistics.

Variable	Obs	Mean	SD	Min	Max
CEE	25,383	10.865	2.052	1.827	20.043
CEI	25,383	0.495	0.625	0.004	19.193
Cashflow	25,383	0.240	0.662	−49.638	34.757
ROA	25,383	0.032	1.818	−76.764	204.690
BMV	25,383	4.666	24.921	−1269.213	2178.310
AGE	25,383	2.106	0.867	0	3.497
IEI	25,383	0.222	0.416	0	1
SIZE	25,383	21.979	1.330	16.412	28.636
OIP	25,377	3.702	1.875	0	10.326
Pcm	25,383	0.099	0.693	−83.497	0.846
LEV	25,383	0.432	1.234	0.007	178.346
KZ	25,383	1.302	2.394	−12.741	14.774

indicates that the standard deviation of corporate carbon emissions is 2.052, with a maximum value of 20.043 and a minimum value of 1.827. This demonstrates that carbon emissions vary among the sample enterprises, though the overall dispersion remains at a moderate level. The mean carbon emission intensity of enterprises stands at 0.495, with a maximum value of 19.193 and a standard deviation of 0.625. This indicates significant variation in carbon emission intensity among the sampled enterprises, with some exhibiting exceptionally high levels, while others display notably low levels. The mean cash flow ratio stands at 0.240, with a maximum value of 34.757, a minimum value of −49.638, and a standard deviation of 0.662. This indicates significant variation in the cash flow status among the sample enterprises, with a few exhibiting extreme cases of either severe cash constraints or substantial cash abundance. The mean return on equity stands at 0.032, with a maximum value of 204.69, a minimum value of −76.764, and a standard deviation of 1.818. This indicates that the overall profitability of the sample enterprises is moderate, with a pronounced divergence in profit levels between companies. The mean book-to-market ratio stands at 4.666, with a maximum value of 2178.31 and a minimum value of −1269.213. The standard deviation is 24.921, indicating that the book-to-market ratios of the sample enterprises exhibit extremely high dispersion, with the presence of outlier observations. The mean age of enterprises is 2.106 years, with a maximum of 3.497 years and a standard deviation of 0.867. This indicates that the operational tenure of the sample enterprises exhibits moderate variation, with the overall cohort being in a mature operational phase. The average carbon emission intensity across industries stands at 0.222, with a standard deviation of 0.416. This indicates that the carbon emission intensity varies moderately among the industries in which the sample enterprises operate, with some sectors exhibiting notably higher carbon emission intensity. The mean enterprise size was 21.979, with a maximum of 28.636, a minimum of 16.412, and a standard deviation of 1.33. This indicates that the variation in scale among the sample enterprises was moderate, with an overall balanced level of size. The mean value for the sustainability of corporate innovation outputs stands at 3.702, with a maximum of 10.326, a minimum of 0, and a standard deviation of 1.875. Moderate variation exists among the sample enterprises, exhibiting a degree of dispersion without extreme divergence. The mean competitive position score for enterprises is 0.099, with a maximum value of 0.846, a minimum value of −83.497, and a standard deviation of 0.693. The sample enterprises exhibit significant divergence, with the presence of outlier observations indicating considerable variation in competitive landscapes. The mean debt-to-asset ratio stands at 0.432, with a maximum value of 178.346 and a standard deviation of 1.234. This indicates that the average debt-to-asset ratio among the sample enterprises is approximately 43.2%, while simultaneously revealing instances of both extremely high and low leverage operations. Significant disparities exist in the capital structures between enterprises. The standard deviation of financing constraints stands at 2.394, with a maximum value of 14.774 and a minimum value of −12.741. This indicates a pronounced divergence in the degree of financing constraints among the sample enterprises, with some firms facing significant financing pressures.

4.2. Baseline Regression

Table 3. Benchmark Regression.

	(1)	(2)	(3)
	CEEi,t	CEEi,t	CEEi,t
${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$	0.540 *** (0.0146)	0.672 *** (0.0106)	0.530 *** (0.0113)
${\mathrm{IEI}}_{\mathrm{i},\mathrm{t}}$	0.880 *** (0.0378)		0.965 *** (0.0292)
${\mathrm{SIZE}}_{\mathrm{i},\mathrm{t}}$		0.828 *** (0.00687)	0.833 *** (0.00671)
${\mathrm{Cashflow}}_{\mathrm{i},\mathrm{t}}$	0.0141 *** (0.00608)	0.0166 *** (0.00481)	0.0111 ** (0.0047)
${\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i},\mathrm{t}}$	0.00343 (0.00213)	0.00682 *** (0.00168)	0.00679 *** (0.00164)
${\mathrm{B}\mathrm{M}\mathrm{V}}_{\mathrm{i},\mathrm{t}}$	−0.000692 *** (0.0000161)	−0.0000652 (0.000128)	−0.0000174 (0.000125)
${\mathrm{A}\mathrm{G}\mathrm{E}}_{\mathrm{i},\mathrm{t}}$	0.319 *** (0.0115)	−0.0920 *** (0.00967)	−0.0875 *** (0.00945)
${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}}_{\mathrm{i},\mathrm{t}}$	9.729 *** (0.026)	−7.477 *** (0.145)	−7.744 *** (0.142)
Individual FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
N	25,383	25,383	25,383
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.936	0.96	0.962

Note: *** and ** denote significance levels of 1% and 5%, respectively; standard errors are shown in parentheses.

presents the baseline regression results. The empirical specification follows standard approaches commonly used in the literature [45]. The first, second and third columns present the regression results under different combinations of control variables, respectively. The first column shows that, without controlling for firm size, both firm carbon emission intensity and industry carbon emission intensity exert a significant positive effect on the dependent variable (at the 1% significance level). In terms of coefficients, each unit increase in CEI corresponds to an approximate increase of 0.540 in the dependent variable, while each unit increase in IEI corresponds to an approximate increase of 0.880 in the dependent variable, indicating a significant impact effect. Upon incorporating the firm size control variable in the second column, the coefficient for firm carbon emission intensity rose to 0.672 and remained highly significant at the 1% level, indicating that its positive impact became further pronounced. The cash flow ratio and return on equity exhibit significant positive coefficients at the 1% significance level, indicating that improvements in a company’s cash flow sufficiency and profitability levels drive growth in the dependent variable. The corporate age coefficient reversed from positive in column (1) to significantly negative in column (2). This reversal is attributable to confounding with firm size: once SIZE is controlled, the residual age effect reveals that mature firms benefit from accumulated energy management experience, leading to relatively lower emissions conditional on scale. The book-to-market ratio coefficient was not significant, suggesting a weaker influence on the dependent variable. The statistical insignificance of book-to-market ratio in this specification is not unexpected, as its primary role is to absorb potential confounding variation associated with firm value characteristics rather than to serve as a substantive predictor. Importantly, its inclusion does not alter the magnitude or significance of the core explanatory variable CEI, confirming that the main findings are not sensitive to this control. The results in the third column, incorporating industry carbon emission intensity and firm size, reveal that the core variables—firm carbon emission intensity and industry carbon emission intensity—maintain a statistically significant positive correlation at the 1% level, with coefficients of 0.530 and 0.965, respectively. This indicates that the core influence effect remains stable. The cash flow ratio is significantly positive at the 5% significance level, the return on equity is significantly positive at the 1% level, and firm age is significantly negative at the 1% level, consistent with the trend observed in the second column.

Overall, the results of the baseline regression analysis support Hypothesis 1, confirming that carbon emission intensity is positively correlated with corporate carbon emissions. That is, the higher the carbon emission intensity in financial statements, the greater the corporate carbon emissions, and this association demonstrates stable significance within the regression model.

4.3. Robustness Tests

To reinforce the causal interpretation of the main results, this paper conducts a series of complementary tests targeting different sources of endogeneity. Lagged-value specifications are used to alleviate reverse causality, system GMM addresses dynamic endogeneity and simultaneity, additional governance controls reduce concerns over omitted variable bias (refer to [46] for more details), pre-2020 subsample regressions rule out distortion from the COVID-19 shock, and placebo tests assess whether the estimated relationship could arise spuriously. Taken together, these tests provide a more stringent basis for interpreting the positive relationship between carbon emission intensity and corporate carbon emissions. They also indirectly speak to measurement reliability, because a proxy severely affected by measurement error would be unlikely to generate such stable results across alternative specifications, subsamples, and identification strategies.

4.3.1. Variable Lag

If an enterprise possesses a larger scale of operations, this may also contribute to improved environmental performance. This implies that the greater the volume of carbon emissions generated by the enterprise, the more it will enhance the carbon emission intensity recorded on its financial statements. Given the bidirectional causal relationship between corporate carbon emissions and carbon emission intensity on financial statements, this may introduce bias into research findings. Therefore, this paper employs an instrumental-variable approach. For details regarding the application of the instrumental variables method (see [47,48] for more applications), please refer to references. The explanatory variable corporate carbon emission intensity (${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$) is regressed using lagged values from one and two periods prior (${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}-1}$, ${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}-2}$). The rationale lies in the fact that, owing to temporal constraints, the lagged one-period and two-period firm carbon intensities bear no direct correlation with current-period indicators such as the cash flow ratio, return on assets, book-to-market ratio, and firm age. This satisfies the exogeneity requirement for instrumental variables. Concurrently, the lagged-period firm carbon emission intensity can influence the current-period firm carbon emissions, thereby satisfying the instrumental variable correlation condition. Analysis of columns (1) and (2) in

Table 4. Lag test.

	(1)	(2)
	CEEi,t	CEEi,t
${\mathrm{L}.\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$	0.493 *** (0.0165)
${\mathrm{L}.2.\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$		0.129 *** (0.011)
${\mathrm{Cashflow}}_{\mathrm{i},\mathrm{t}}$	0.0149 *** (0.00488)	0.0300 *** (0.00727)
${\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i},\mathrm{t}}$	0.0103 *** (0.00270)	0.00743 *** (0.00168)
${\mathrm{B}\mathrm{M}\mathrm{V}}_{\mathrm{i},\mathrm{t}}$	0.000133 (0.0001414)	−0.00002673 (0.000151)
${\mathrm{A}\mathrm{G}\mathrm{E}}_{\mathrm{i},\mathrm{t}}$	0.129 *** (0.0142)	0.163 *** (0.0204)
${\mathrm{IEI}}_{\mathrm{i},\mathrm{t}}$	1.076 *** (0.0337)	1.365 *** (0.0353)
${\mathrm{SIZE}}_{\mathrm{i},\mathrm{t}}$	0.856 *** (0.00747)	0.820 *** (0.00817)
${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}}_{\mathrm{i},\mathrm{t}}$	−8.745 *** (0.162)	−7.937 *** (0.179)
Individual FE	Yes	Yes
Year FE	Yes	Yes
N	22,303	19,950
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.965	0.966

Note: *** denotes significance level of 1%, respectively. Standard errors are shown in parentheses.

reveals that the coefficients for corporate carbon emission intensity ${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}-1}$ and ${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}-2}$ remain statistically significant at the 1% level, indicating a positive correlation between carbon emission intensity and corporate carbon emissions. This finding is consistent with the benchmark regression results.

4.3.2. Regressions on the Pre-2020 Sample

Given the widespread impact of COVID-19 containment measures on routine economic activities, this study treats the period 2020–2022 as an anomalous window and excludes the corresponding data to mitigate interference in empirical findings. The results are presented in

Table 5. Sample period preceding the 2020 COVID-19 outbreak.

	(1)	(2)	(3)
	CEEi,t	CEEi,t	CEEi,t
${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$	0.465 *** (0.0153)	0.585 *** (0.0116)	0.454 *** (0.0122)
${\mathrm{IEI}}_{\mathrm{i},\mathrm{t}}$	0.928 *** (0.0432)		1.013 *** (0.0343)
${\mathrm{SIZE}}_{\mathrm{i},\mathrm{t}}$		0.795 *** (0.00853)	0.801 *** (0.00831)
${\mathrm{Cashflow}}_{\mathrm{i},\mathrm{t}}$	0.000134 (0.00631)	0.00129 (0.00514)	−0.00271 (0.00501)
${\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i},\mathrm{t}}$	0.00389 * (0.00209)	0.00504 *** (0.0017)	0.00508 *** (0.00166)
${\mathrm{B}\mathrm{M}\mathrm{V}}_{\mathrm{i},\mathrm{t}}$	−0.000531 *** (0.000157)	−0.0000782 (0.000128)	−0.0000455 (0.000125)
${\mathrm{A}\mathrm{G}\mathrm{E}}_{\mathrm{i},\mathrm{t}}$	0.282 *** (0.0136)	−0.115 *** (0.0119)	−0.115 *** (0.0116)
${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}}_{\mathrm{i},\mathrm{t}}$	9.753 *** (0.0291)	−6.678 *** (0.18)	−6.981 *** (0.175)
Individual FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
N	18,353	18,353	18,353
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.947	0.965	0.967

Note: *** and * denote significance levels of 1% and 10%, respectively. Standard errors are shown in parentheses.

. The first column shows that, without controlling for firm size, both firm carbon emission intensity and industry carbon emission intensity exert a significant positive effect on the dependent variable at the 1% level, with coefficients of 0.465 and 0.928, respectively. Upon incorporating control variables in the second column, the CEI coefficient rises to 0.585 and remains highly significant, while firm size and return on equity are significantly positive at the 1% level; firm age turns significantly negative, and the cash flow ratio and book-to-market ratio remain statistically insignificant. The third column, which further incorporates industry carbon emission intensity and firm size, confirms that both carbon emission intensity and industry carbon emission intensity maintain significant positive coefficients at the 1% level (0.454 and 1.013, respectively), with the direction and significance of all control variables consistent with the second column. Overall, corporate carbon emission intensity continues to exhibit a positive correlation with corporate carbon emissions at the 1% significance level—that is, the higher the carbon emission intensity reported in financial statements, the greater the volume of carbon emissions produced by the enterprise. Hypothesis 1 is thus validated once more.

4.3.3. Adding Control Variables

Studies have shown that boards of directors, driven by the pursuit of value maximization, tend to push their portfolio companies toward adopting stringent environmental standards in their production processes. The underlying rationale is that building a favorable environmental reputation can ultimately deliver superior returns to shareholders. Independent directors play a critical role in mitigating potential biases in managerial decision-making, strengthening the rigor of strategic choices, and exercising effective supervision over corporate governance. Conversely, when the roles of board chairman and general manager are held by the same individual, the board’s ability to monitor management is compromised, which may give rise to self-serving behaviors that run counter to the firm’s interests. Accordingly, this study incorporates board independence (INDD) and CEO duality (CONP) as control variables. Given that the effects of board independence and CEO duality on corporate carbon emissions may not materialize immediately, both variables are lagged by one period in the regression specification. Specifically, ${INDD}_{i,t-1}$ and ${CONP}_{i,t-1}$ are introduced into the model, and the updated regression results are reported in

Table 6. Test with additional control variables.

	(1)	(2)	(3)
	CEEi,t	CEEi,t	CEEi,t
${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$	0.8713 *** (0.0176)	0.8712 *** (0.0176)	0.8714 *** (0.0176)
${\mathrm{I}\mathrm{N}\mathrm{D}\mathrm{D}}_{\mathrm{i},\mathrm{t}-1}$	0.0007 (0.000783)		0.000672 (0.000783)
${\mathrm{C}\mathrm{O}\mathrm{N}\mathrm{P}}_{\mathrm{i},\mathrm{t}-1}$		0.0160 * (0.00965)	0.0158 (0.00965)
${\mathrm{Cashflow}}_{\mathrm{i},\mathrm{t}}$	0.0215 *** (0.00471)	0.0216 *** (0.00471)	0.0216 *** (0.00471)
${\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i},\mathrm{t}}$	0.0134 *** (0.00261)	0.0134 *** (0.00261)	0.0134 *** (0.00261)
${\mathrm{B}\mathrm{M}\mathrm{V}}_{\mathrm{i},\mathrm{t}}$	0.000241 * (0.000136)	0.000242 * (0.000136)	0.000243 * (0.000136)
${\mathrm{A}\mathrm{G}\mathrm{E}}_{\mathrm{i},\mathrm{t}}$	0.118 *** (0.0137)	0.120 *** (0.0137)	0.120 *** (0.0137)
${\mathrm{IEI}}_{\mathrm{i},\mathrm{t}}$	0.681 *** (0.0338)	0.681 *** (0.0338)	0.681 *** (0.0338)
${\mathrm{SIZE}}_{\mathrm{i},\mathrm{t}}$	0.851 *** (0.00721)	0.851 *** (0.00721)	0.851 *** (0.00721)
${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}}_{\mathrm{i},\mathrm{t}}$	−8.746 *** (0.16)	−8.721 *** (0.156)	−8.752 *** (0.161)
Individual FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
N	22,303	22,303	22,303
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.968	0.968	0.968

Note: *** and * denote significance levels of 1% and 10%, respectively. Standard errors are shown in parentheses.

.

Column (1) includes lagged board independence alongside industry-level carbon emission intensity, firm size, and other control variables. The estimated coefficient on firm-level carbon emission intensity is 0.8713, which is statistically significant at the 1% level. In Column (2), lagged board independence is replaced by lagged CEO duality. The core coefficient remains virtually unchanged at 0.8712 and continues to be significant at the 1% level, while the newly added CEO duality variable is positively significant at the 10% level. Column (3) incorporates both lagged variables simultaneously, yielding a core coefficient of 0.8714 that is again significant at the 1% level. Across all three specifications, the variation in the estimated coefficients on firm-level carbon emission intensity is negligible. The significance levels and directional consistency of other control variables, including industry carbon emission intensity and firm size, remain stable throughout, and the model’s goodness of fit reaches 0.968 in every case. These findings confirm that carbon emission intensity is positively associated with corporate carbon emissions, thereby reinforcing the robustness of the baseline conclusions.

4.3.4. Placebo Test

This paper constructed spurious explanatory variables (see [49,50] for similar discussions) through 1000 random assignments using Monte Carlo simulation, with the results shown in Figure 3. It can be observed that the randomly generated spurious coefficients largely exhibit an approximate normal distribution centred around zero. Moreover, the vast majority of spurious coefficients correspond to p-values exceeding 0.1. This indicates that in hypothetical scenarios devoid of genuine carbon emission intensity effects, variations in carbon emissions are driven solely by random disturbances, with no discernible systematic correlation. Consequently, the robustness of the regression results is indirectly validated.

4.3.5. Endogeneity Test

To address potential endogeneity issues in the research process, this paper employs the dependent variable lagged by one period as an instrumental variable. It is widely recognised within academia that instrumental variables also exert a significant influence in demonstrating the validity of variable relationships, as demonstrated in the literature [51]. The model was regressed using the system Generalised Method of Moments (GMM), with the results presented in

Table 7. Endogeneity Test.

	CEEi,t
${\mathrm{C}\mathrm{E}\mathrm{E}}_{\mathrm{i},\mathrm{t}-1}$	0.875 *** (0.324)
${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$	1.342 *** (0.453)
${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}\mathrm{s}}_{\mathrm{i},\mathrm{t}}$	Yes
${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}}_{\mathrm{i},\mathrm{t}}$	Yes
Individual FE	Yes
Year FE	Yes
AR (1)	0.0240
AR (2)	0.182
Hansen	0.32
N	19,434

Note: *** denotes significance level of 1%, respectively. Standard errors are shown in parentheses.

. Carbon emission intensity remains positively correlated with corporate carbon emissions. The p-values for AR (1) and AR (2) are 0.0240 and 0.182 respectively, indicating the presence of first-order serial correlation but the absence of second-order serial correlation (refer to [52,53] for similar analysis). Furthermore, the Hansen test (refer to [54,55] for similar approaches) yielded a p-value of 0.32, which is greater than 0.1, indicating that the instrumental variables selected are valid.

4.4. Moderation Effect Test

Table 8. Moderating effects test.

	(1)	(2)
	CEEi,t	CEEi,t
${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$	0.454 *** (0.0118)	0.542 *** (0.0112)
${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}\times {\mathrm{O}\mathrm{I}\mathrm{P}}_{\mathrm{i},\mathrm{t}}$	0.0521 *** (0.00271)
${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}\times {\mathrm{P}\mathrm{c}\mathrm{m}}_{\mathrm{i},\mathrm{t}}$		0.0301 *** (0.00189)
${\mathrm{Cashflow}}_{\mathrm{i},\mathrm{t}}$	0.0117 ** (0.00467)	0.0106 ** (0.00468)
${\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i},\mathrm{t}}$	0.00655 *** (0.00163)	0.00691 *** (0.00163)
${\mathrm{B}\mathrm{M}\mathrm{V}}_{\mathrm{i},\mathrm{t}}$	−0.0000495 (0.000124)	−0.0000316 (0.000124)
${\mathrm{A}\mathrm{G}\mathrm{E}}_{\mathrm{i},\mathrm{t}}$	−0.0730 *** (0.00941)	−0.0841 *** (0.0094)
${\mathrm{IEI}}_{\mathrm{i},\mathrm{t}}$	0.893 *** (0.0293)	0.952 *** (0.0291)
${\mathrm{SIZE}}_{\mathrm{i},\mathrm{t}}$	0.818 *** (0.0067)	0.829 *** (0.00668)
${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}}_{\mathrm{i},\mathrm{t}}$	−7.481 *** (0.142)	−7.656 *** (0.142)
Individual FE	Yes	Yes
Year FE	Yes	Yes
N	25,377	25,383
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.963	0.962

Note: *** and ** denote significance levels of 1% and 5%, respectively. Standard errors are shown in parentheses.

presents the results of the moderation effect tests. It is worth noting that all moderator variables (OIP and Pcm) in the regression model were mean-centred (as per the methodology in [56]) to mitigate potential multicollinearity issues and ensure the reliability of interaction effect results. The interaction between carbon emission intensity and innovation output is positive and significant at the 1% level, with a coefficient of 0.0521. This indicates that an increase in the persistence of corporate innovation output reinforces the positive impact of carbon emission intensity on corporate carbon emissions. That is, the higher the level of innovation output, the greater the increase in actual corporate carbon emissions for each unit increase in carbon emission intensity. From an economic perspective, the interaction coefficient of 0.0521 implies that a marginal increase in innovation output will amplify the driving effect of carbon emission intensity on corporate carbon emissions by 0.05 units, providing empirical support for H2a.

Similarly, the interaction between carbon emission intensity and corporate competitive position exhibits a positive and significant correlation at the 1% level, with a coefficient of 0.0301. This finding indicates that an improvement in a firm’s competitive position intensifies the positive correlation between carbon emission intensity and its actual corporate carbon emissions. The higher the firm’s competitive standing, the more pronounced the driving effect of carbon emission intensity on actual emissions becomes. Economically speaking, this coefficient implies that a marginal improvement in competitive position increases the impact of carbon emission intensity on a firm’s emissions by 0.03 units. As a firm’s competitive position improves, carbon emission intensity is more likely to effectively reflect its true carbon emissions, thereby confirming Hypothesis 2b.

As shown in Figure 4, the left panel reveals the positive regulatory effect of OIP. When innovation exhibits high persistence, carbon emission intensity strongly predicts carbon emissions. However, as the rate of innovation continues to decline, this effect will diminish significantly. The divergent trajectories of the two trend lines clearly demonstrate that sustained innovation output enhances the representativeness of carbon emission intensity for corporate carbon emission levels. The right-hand panel demonstrates the positive regulatory effect of Pcm. Under conditions of high competitive standing, the positive correlation between carbon emission intensity and total emissions is markedly amplified. This differentiated pattern confirms that elevated competitive standing enhances the validity and reliability of carbon emission intensity as a metric.

The consistent impact of control variables across models: cash flow ratios, return on assets, industry energy intensity, and firm size all exert a positive driving effect on corporate carbon emissions, and are statistically significant at the 1% or 5% level. The age of an enterprise exhibits a significant negative correlation with carbon emissions, meaning that the longer an enterprise has been in operation, the lower its relative carbon emission levels tend to be. The impact of the book-to-market ratio is limited. Overall, the findings indicate that both innovation output and corporate competitive position significantly moderate the relationship between carbon emission intensity and corporate carbon emissions, highlighting the context-dependent nature of carbon emission intensity effectiveness.

4.5. Mechanism Tests

4.5.1. Mechanistic Effects on Resource Allocation Efficiency

The above verification demonstrates that carbon emission intensity exhibits a significant positive correlation with corporate carbon emissions. Furthermore, the results in Column (1) of

Table 9. Mechanism tests.

	(1)	(2)
	LEVi,t	KZi,t
${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$	0.155 *** (0.0317)	0.353 *** (0.0365)
${\mathrm{Cashflow}}_{\mathrm{i},\mathrm{t}}$	−0.011 (0.0133)	−0.755 *** (0.0152)
${\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i},\mathrm{t}}$	−0.00089 (0.00463)	−0.00759 (0.00533)
${\mathrm{B}\mathrm{M}\mathrm{V}}_{\mathrm{i},\mathrm{t}}$	−0.000715 ** (0.000352)	0.00334 *** (0.000405)
${\mathrm{A}\mathrm{G}\mathrm{E}}_{\mathrm{i},\mathrm{t}}$	0.126 *** (0.0267)	2.035 *** (0.0306)
${\mathrm{IEI}}_{\mathrm{i},\mathrm{t}}$	−0.162 ** (0.0825)	−0.315 *** (0.0948)
${\mathrm{SIZE}}_{\mathrm{i},\mathrm{t}}$	−0.0548 *** (0.0189)	0.235 *** (0.0218)
${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}}_{\mathrm{i},\mathrm{t}}$	1.336 *** (0.401)	−8.086 *** (0.461)
Individual FE	Yes	Yes
Year FE	Yes	Yes
N	25,383	25,383
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.165	0.707

Note: *** and ** denote significance levels of 1% and 5%, respectively. Standard errors are shown in parentheses.

reveal that this correlation is partially mediated by resource allocation efficiency: The carbon emission intensity coefficient is 0.155, significant at the 1% level, indicating that its variation enhances the representativeness of actual corporate carbon emissions by optimising resource allocation efficiency. From an economic perspective, the coefficient of 0.155 indicates that for every unit increase in carbon emission intensity, the debt-to-asset ratio rises by approximately 15.5%. This serves to guide enterprises towards channelling capital and resources towards key emission-reduction areas such as low-carbon R&D and energy-efficient equipment upgrades, thereby securing investment for low-carbon transition and preventing resource stagnation in high-energy-consumption sectors. This optimisation of resource allocation enables fluctuations in carbon emission intensity to precisely align with actual changes in corporate emissions, thereby significantly reducing the risk of resource misallocation causing a disconnect between reported intensity and genuine emissions. Empirical findings indicate that the efficiency of corporate resource allocation plays a partial mediating role, providing a core intrinsic mechanism for explaining the relationship whereby carbon emission intensity effectively reflects a firm’s true carbon emissions through optimising this efficiency. Therefore, we assume that H3a is supported.

4.5.2. The Mechanistic Effects of Financing Constraints

Column (2) of Table 9 reports the results of the corporate debt constraint mechanism. Combined with the positive effect of carbon emission intensity on carbon emissions confirmed by the main regression, the results in this column validate the partial mediating function in this relationship: The carbon emission intensity coefficient is 0.353, significant at the 1% level, indicating that increased carbon emission intensity enhances the representativeness of actual corporate carbon emissions by intensifying financing constraints. High carbon emission intensity is often accompanied by heightened environmental compliance risks and anticipated policy penalties, which may prompt creditors to tighten financing conditions and increase borrowing costs, thereby restricting enterprises’ access to external financing channels. Analysis of the table reveals that for every unit increase in carbon emission intensity, corporate financing constraints rise by approximately 35.3%. This constraint essentially reflects the debt market’s precise pricing of corporate carbon performance, compelling enterprises to accurately report their emissions. The lack of sufficient funds to conceal actual emissions means that their carbon emissions data more closely aligns with the scale of emissions in real-world operations, thereby reducing the likelihood of data distortion. This transmission mechanism clearly demonstrates the value of financing constraints as a regulatory tool, providing a comprehensive explanation of the intrinsic logic whereby carbon emission intensity effectively reflects actual corporate carbon emissions through this pathway. It complements the direct effects observed in the main regression, thereby fully elucidating the underlying logic through which carbon emission intensity influences carbon emissions. Therefore, assume that H3b is confirmed. Figure 5 illustrates the transmission mechanism through which carbon emission intensity influences total carbon emissions by altering firms’ resource allocation efficiency (LEV) and financing constraints (KZ).

It should be noted that both resource allocation efficiency and financing constraints function as partial rather than full mediators, implying that a substantial portion of the relationship between carbon emission intensity and corporate carbon emissions operates through the direct channel or through other transmission pathways not captured by the current model. The partial mediation results suggest that the financial statement-based carbon emission intensity signal transmits to actual emissions through multiple concurrent mechanisms, of which capital structure adjustment and debt market pricing represent only two observable dimensions. A more detailed discussion of these unmodelled pathways is provided in Section 7.1.

5. Heterogeneity Analysis

5.1. Corporate Environmental Signals

To examine whether enterprises issuing explicit environmental signals can enhance the correlation between carbon emission intensity on financial statements and actual corporate carbon emissions, this study employs ISO 14001 [57] environmental management system certification as the primary indicator of environmental signals (EPS). This indicator is then cross-multiplied with the core explanatory variable and regressed using the moderation effect model outlined in Model (6). Specifically, based on ISO 14001 Environmental Management System certification data disclosed by the Certification and Accreditation Administration of the People’s Republic of China, if enterprises in the sample have obtained this certification, their Environmental Performance Signal (EPS) variable is set to 1; otherwise, it is set to 0. It is worth emphasising that corporate participation in such voluntary environmental regulation clearly signals to the outside world their proactive commitment to environmental responsibility. The regression results are presented in Column (1) of

Table 10. Heterogeneity Test Results for Corporate Environmental Signals.

	CEEi,t
	Corporate Environmental Signals
	(1)	(2)
${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$	0.527 *** (0.0114)	0.529 *** (0.0113)
$\mathrm{E}\mathrm{P}\mathrm{S}\times {\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$	0.0173 * (0.0103)
$\mathrm{G}\mathrm{P}\mathrm{B}\times {\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$		0.0618 ** (0.0269)
${\mathrm{Cashflow}}_{\mathrm{i},\mathrm{t}}$	0.0111 ** (0.0047)	0.0111 ** (0.0047)
${\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i},\mathrm{t}}$	0.00680 *** (0.00164)	0.00681 *** (0.00164)
${\mathrm{B}\mathrm{M}\mathrm{V}}_{\mathrm{i},\mathrm{t}}$	−0.0000172 (0.000125)	−0.0000181 (0.000125)
${\mathrm{A}\mathrm{G}\mathrm{E}}_{\mathrm{i},\mathrm{t}}$	−0.0878 *** (0.00946)	−0.0871 *** (0.00945)
${\mathrm{IEI}}_{\mathrm{i},\mathrm{t}}$	0.965 *** (0.0292)	0.963 *** (0.0293)
${\mathrm{SIZE}}_{\mathrm{i},\mathrm{t}}$	0.833 *** (0.00671)	0.833 *** (0.00671)
${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}}_{\mathrm{i},\mathrm{t}}$	−7.748 *** (0.142)	−7.742 *** (0.142)
Individual FE	Yes	Yes
Year FE	Yes	Yes
N	25,383	25,383
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.958	0.958

Note: ***, **, and * denote significance levels of 1%, 5%, and 10%, respectively. Standard errors are shown in parentheses.

. Here, the coefficient for the interaction term ($\mathrm{E}\mathrm{P}\mathrm{S}\times {\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$) is positive but only significant at the 10% level—weaker than the 5% significance achieved by $\mathrm{G}\mathrm{P}\mathrm{B}\times {\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$ in column (2). This likely reflects that ISO 14001 certification is a binary, one-time signal with limited ongoing informational content, whereas a CEO’s green background represents a persistent governance characteristic that continuously shapes carbon accounting quality. Nevertheless, the positive coefficient. This indicates that firms’ voluntary participation in environmental regulations can standardise their carbon emissions data accounting and disclosure processes, thereby enhancing the credibility and accuracy of carbon emission intensity metrics. Consequently, this strengthens the ability of carbon emission intensity to reflect firms’ actual corporate carbon emissions.

The educational background or professional experience of senior executives shapes their cognitive frameworks and knowledge structures, thereby influencing the strategic direction chosen by the enterprise. Therefore, this paper adopts whether the firm’s CEO possesses a green professional background as the second indicator of corporate environmental signals (GPB). This is then cross-multiplied with the core explanatory variables and regressed according to the moderation effect model in Equation (6). Specifically, when a company’s chief executive officer has received green-related education or engaged in green-related work, the CEO’s green professional background (GPB) is assigned a value of 1; otherwise, it is 0. The regression results are presented in Column (2) of Table 10. Here, the coefficient for the interaction term between the firm’s environmental signal and the core explanatory variable ($\mathrm{G}\mathrm{P}\mathrm{B}\times {\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$) is also significantly positive. This indicates that corporate chief executives with green professional backgrounds may be more inclined to integrate environmental principles into strategic decision-making, thereby driving the establishment of robust carbon emissions monitoring, accounting and management systems. This, in turn, strengthens the positive correlation between carbon emission intensity and corporate carbon emissions.

5.2. Corporate Environmental Engagement

If an enterprise demonstrates a sound environmental attitude towards its development strategy, then the correlation between its carbon emission intensity and total carbon emissions on its financial statements becomes stronger. This study employs the incremental frequency of terms such as “environmental protection and low-carbon” within corporate annual reports as an indicator of corporate environmental protection stance (EPW) to conduct heterogeneity tests. The results are shown in column (1) of

Table 11. Heterogeneity Test Results for Corporate Environmental Engagement and Enterprise Technical Level.

	CEEi,t
	Corporate Environmental Stance	Enterprise Technical Level
	(1)	(2)
${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$	0.54594 *** (0.0114)	0.54604 *** (0.0117)
$\mathrm{E}\mathrm{P}\mathrm{W}\times {\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$	0.00592 *** (0.000728)
$\mathrm{G}\mathrm{T}\mathrm{I}\times {\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$		−0.0696 *** (0.0133)
${\mathrm{Cashflow}}_{\mathrm{i},\mathrm{t}}$	0.0114 ** (0.0047)	0.0113 ** (0.0047)
${\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i},\mathrm{t}}$	0.00677 *** (0.00164)	0.00681 *** (0.00164)
${\mathrm{B}\mathrm{M}\mathrm{V}}_{\mathrm{i},\mathrm{t}}$	−0.000016 (0.000125)	−0.0000142 (0.000125)
${\mathrm{A}\mathrm{G}\mathrm{E}}_{\mathrm{i},\mathrm{t}}$	−0.0881 *** (0.00944)	−0.0886 *** (0.00945)
${\mathrm{IEI}}_{\mathrm{i},\mathrm{t}}$	0.947 *** (0.0293)	0.953 *** (0.0293)
${\mathrm{SIZE}}_{\mathrm{i},\mathrm{t}}$	0.833 *** (0.00671)	0.834 *** (0.00671)
${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{t}}_{\mathrm{i},\mathrm{t}}$	−7.751 *** (0.142)	−7.763 *** (0.142)
Individual FE	Yes	Yes
Year FE	Yes	Yes
N	25,383	25,383
$\mathrm{A}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2$	0.958	0.958

Note: *** and ** denote significance levels of 1% and 5%, respectively. Standard errors are shown in parentheses.

. It can be observed that the coefficient for the interaction term between the explanatory variable and the core explanatory variable ($\mathrm{E}\mathrm{P}\mathrm{W}\times {\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$) is significantly positive. This indicates that the more environmentally responsible a company is, the more accurately its accounting carbon emissions intensity reflects reality, thereby exhibiting a stronger correlation with its actual corporate carbon emissions.

5.3. Enterprise Technical Level

When an enterprise possesses a high level of green technology, it indicates that the enterprise has mastered more technical knowledge conducive to carbon reduction. Therefore, this paper regresses the firm’s Green Technology Intensity (GTI) by cross-multiplying it with the core explanatory variables, as per Model (6). When a company’s annual total of green patent applications ranks within the top 10% among all listed enterprises in China, it indicates that the company possesses a higher level of green technology. In such cases, the GTI is assigned a value of 1; otherwise, it is assigned a value of 0. As shown in column (2) of Table 11, the interaction term coefficient ($\mathrm{G}\mathrm{T}\mathrm{I}\times {\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i},\mathrm{t}}$) is significantly negative, indicating that the higher a firm’s green technology level, the stronger its inhibitory effect on carbon emission intensity—a metric representing the firm’s actual corporate carbon emissions.

6. Application of Zero-Carbon Factory Certification Based on Accounting Information

6.1. Establishing an Evaluation Framework for Zero-Carbon Factories

6.1.1. Design Principles for Evaluation Indicator Systems

(1)

The selection of evaluation indicators is not subjectively devised but strictly grounded in the conclusions drawn from the preceding empirical analysis. Indicators incorporated into the VIKOR evaluation framework should be key variables validated in the prior regression model as possessing significant explanatory power, moderating effects, or mediating effects on carbon emissions. This principle ensures the evaluation constitutes an empirical extension of the logic, thereby safeguarding the internal consistency of the research conclusions.

(2)

The core of the VIKOR method lies in ranking sample enterprises by their relative performance through the divergence of indicator data and handling conflicting criteria. Therefore, the selected indicators must exhibit sufficient dispersion and discriminatory power across the sample. Based on this principle, this study excluded certain variables from the preceding empirical analysis: indicators with low discriminatory power and indicators reflecting industry homogeneity were removed (e.g., green technology innovation level, green professional background, ISO 14001 certification, and industry carbon emission intensity). Furthermore, metrics lacking explicit superiority-inferiority attributes (e.g., book-to-market ratio and corporate age) were excluded.

(3)

Specifically, the rationale for selecting the final nine indicators is as follows: Low-carbon performance (CEI) directly measures the physical outcome of emissions reduction, which is the fundamental prerequisite for a zero-carbon factory. Green Momentum (OIP, EPW) captures the enterprise’s sustainable innovation output and strategic focus on environmental issues, representing the internal driving force for long-term zero-carbon transition. Competitive advantage (Pcm, SIZE) reflects the market pricing power and scale effects, which are essential for absorbing the premium costs associated with green technologies. Economic resilience (KZ, LEV, ROA, Cashflow) ensures that the enterprise possesses the financial health, low financing constraints, and operational cash flow necessary to sustain capital-intensive zero-carbon investments. These conflicting yet complementary dimensions perfectly align with the VIKOR method’s requirement for comprehensive, multi-criteria compromise evaluation.

6.1.2. The Composition of the Evaluation Indicator System

The certification of zero-carbon factories constitutes not merely an assessment of physical emissions outcomes, but rather a complex systemic evaluation process encompassing financial backing, governance commitment, and technological impetus. Building upon the empirical mechanism analysis presented earlier, this paper rigorously adheres to principles of scientific rigour and systematic methodology. Leveraging the availability and objectivity of publicly listed companies’ financial data, it constructs a comprehensive evaluation model comprising one primary indicator and nine secondary indicators. This framework aims to comprehensively quantify enterprises’ overall capability to achieve zero-carbon transformation from both internal and external environmental perspectives, thereby establishing quantifiable and replicable evaluation standards for zero-carbon factory certification.

The VIKOR method requires the prior determination of weight values for each attribute. To avoid bias arising from subjective weighting, this paper employs the entropy weighting method to establish the weights for each indicator. The entropy weighting method measures information content based on the degree of dispersion in indicator data, with greater data variation yielding higher weights. The core composition, attribute characteristics, and economic implications of the zero-carbon factory evaluation indicator system are presented in

Table 12. Zero-Carbon Factory Evaluation Indicator System.

Primary Indicator	Level	Secondary Indicators	Indicator Number	Nature of the Indicator
Zero-carbon factory	Low-carbon performance	CEI	X1	−
	Green momentum	OIP	X2	+
		EPW	X3	+
	Competitive advantage	Pcm	X4	+
		SIZE	X5	+
	Economic resilience	KZ	X6	−
		LEV	X7	−
		ROA	X8	+
		Cashflow	X9	+

Note: “+” denotes benefit-type indicators, meaning a higher value indicates better performance; “−” denotes cost-type indicators, meaning a lower value indicates better performance.

below.

6.2. Selection of Research Subjects and Measurement of Indicator Weighting

6.2.1. Data Sources and Research Subjects

The certification assessment of zero-carbon factories constitutes a systematic undertaking involving multidimensional data tracking and accounting. Given the significant variations in carbon emission characteristics and accounting boundaries across different industries, cross-sector hybrid evaluations often struggle to ensure the precision and comparability of results.

As a pillar and foundational raw materials industry within the national economy, the steel sector serves both as a crucial enabler for manufacturing transformation and as a primary source of carbon emissions within the industrial sphere. The success of this industry’s green transition directly impacts the nation’s progress towards achieving its carbon peak and carbon neutrality objectives. Consequently, selecting the steel industry as the subject for zero-carbon factory certification carries significant exemplary value and pressing practical urgency.

Based on data availability and the disclosure standards of listed companies, this study selected 37 listed steel enterprises from the A-share market in 2022 as evaluation subjects. To adhere to the principle of objectivity in academic assessment and facilitate subsequent chart presentation and analysis, these 37 enterprises are sequentially numbered as ${\mathrm{C}}_1,{\mathrm{C}}_2,\dots ,{\mathrm{C}}_{37}$. These enterprises represent the core production capacity of China’s steel smelting and processing sector, exhibiting significant industry representativeness in technological innovation, environmental governance, and capital operations. The indicator data for this study primarily originates from the CSMAR database and the annual reports and social responsibility reports of the sampled enterprises; information on enterprise green factory certifications is sourced from the national-level green factory directory published by the Ministry of Industry and Information Technology.

6.2.2. Calculation of Indicator Weights Based on the Entropy Weighting Method

When calculating the weights of evaluation indicators, the existence of differences between indicators affects their comparability. Therefore, indicators must first undergo standardisation to render them comparable. Simultaneously, considering the direction of influence each indicator exerts on the zero-carbon factory, distinct processing formulas are applied. Positive indicators are processed using Equation (7), while negative indicators are processed using Equation (8). To satisfy the non-negative data requirement of the entropy weight method, the standardised data undergoes non-negation translation by adding a negligible value of 0.0001 to all data points.

$$z_{ij}={\displaystyle \frac{x_{ij}-x_j^{\min }}{x_j^{\max }-x_j^{\min }}}+0.0001$$

(7)

$$z_{ij}={\displaystyle \frac{x_j^{\max }-x_{ij}}{x_j^{\max }-x_j^{\min }}}+0.0001$$

(8)

In this context, $x_{ij}$ denotes the raw value of indicator $j$ for firm $\mathrm{i}$, $x_j^{max}$ and $x_j^{min}$ represent the maximum and minimum values of indicator $j$ across the entire sample, respectively, while $z_{ij}$ indicates the standardised value of the indicator.

Following the aforementioned processing, the standardised matrix Z is obtained. The entropy weighting method determines indicator weights by measuring indicator uncertainty: greater uncertainty signifies greater information content, resulting in higher weights when reflecting the research subject; conversely, weights are lower. The specific calculation steps are shown in Equations (9)–(11), where n denotes the number of companies and m denotes the number of indicators.

First, compute the probability matrix:

$$P_{ij}={\displaystyle \frac{z_{ij}}{\sum _{i=1}^n{z}_{ij}}}, (j=1,2,\dots ,m)$$

(9)

Secondly, calculate the information entropy of each indicator:

$$E_j=-{\displaystyle \frac{1}{\ln n}}\sum _{i=1}^n{P}_{ij} \mathrm{l}\mathrm{n} P_{ij}, (j=1,2,\dots ,m)$$

(10)

Thirdly, calculate the weightings for each indicator using information entropy:

$$W_j={\displaystyle \frac{G_j}{\sum _{j=1}^m{G}_j}}$$

(11)

In this context, $G_j$ denotes the information utility value of an indicator, $W_j$ represents the final weight of the jth indicator, and the sum of all indicator weights equals 1.

Following the aforementioned entropy weighting method’s computational steps, MATLAB 2018a was employed to calculate the final weighting results for each indicator from the standardised data of the nine standardised indicators, as presented in

Table 13. Weighting Results for Zero-Carbon Factory Evaluation Indicators in the Steel Industry Under the Entropy-Based Methodology.

Primary Indicator	Level	Secondary Indicators	Indicator Number	Nature of the Indicator	Total Weight
Zero-carbon factory	Low-carbon performance	CEI	X1	−	0.3141
	Green momentum	OIP	X2	+	0.1359
		EPW	X3	+	0.0382
	Competitive advantage	Pcm	X4	+	0.1411
		SIZE	X5	+	0.0924
	Economic resilience	KZ	X6	−	0.0610
		LEV	X7	−	0.0522
		ROA	X8	+	0.0324
		Cashflow	X9	+	0.1327

Note: “+” denotes benefit-type indicators, meaning a higher value indicates better performance; “−” denotes cost-type indicators, meaning a lower value indicates better performance.

.

6.3. Empirical Evaluation of Zero-Carbon Factories Based on the VIKOR Method

6.3.1. Principles and Justification of the Entropy-VIKOR Method

To enhance the robustness and objectivity of zero-carbon factory assessments, this study adopted the Entropy-VIKOR method rather than AHP, DEA or TOPSIS. As shown in

Table 14. Comparison of Multi-Criteria Decision-Making Methods.

Method	Core Feature	Key Disadvantage	Inappropriateness for This Study	Advantage of Entropy-VIKOR
AHP	Depends on expert scoring and subjective judgment	High subjectivity; difficult to ensure consistency	Zero-carbon certification requires objective accounting data	Uses entropy weighting for fully objective weighting
DEA	Focuses on input–output efficiency	Cannot handle conflicting indicators	Low-carbon performance, innovation, and financial resilience are conflicting	Capable of synthesizing conflicting criteria
TOPSIS	Measures distance to ideal solutions	Ignores criterion importance and individual regret	Cannot identify weak-indicator risks	Balances group utility and individual regret
Entropy-VIKOR	Objective weighting and compromise ranking	None are obvious	—	Optimal for multidimensional, data-based, robust certification

, each of these methods has clear limitations in addressing the multidimensional, conflicting and data-driven nature of zero-carbon factory assessments. The Entropy-VIKOR method avoids the need for subjective weighting, accommodates conflicting evaluation criteria, and balances collective utility with individual regret; it is therefore particularly suitable for the certification of zero-carbon factories based on accounting information.

The VIKOR (Vlsekriterijumska Optimizacija I Kompromisno Resenje) method is a multi-criteria trade-off decision-making approach proposed by Opricovic, Tzeng and others. Its core principle involves defining positive and negative ideal solutions, calculating the distance between each alternative and these ideal solutions, and ultimately obtaining the feasible compromise solution closest to the ideal solutions.

Among these, the Positive-ideal Solution denotes the optimal alternative across all evaluation criteria, while the Negative-ideal Solution represents the worst-performing alternative across all criteria. Subsequently, the evaluation values of each alternative are compared, and their priority order is determined based on the magnitude of their distance from the ideal solutions [58]. The VIKOR method yields a compromise feasible solution closest to the ideal solution. This compromise inherently involves trade-offs between attributes, stemming from the Lp-metric of the compromise programming approach (see [59] for similar discussions). Its defining characteristic is maximising “collective benefits” while minimising “individual regrets of dissent”. Consequently, this compromise solution ultimately gains acceptance from decision-makers.

6.3.2. Data Preprocessing and Determination of the Ideal Solution

This study employs a standardised matrix consistent with entropy weight calculations to ensure uniformity in data measurement throughout the evaluation process, thereby avoiding data bias arising from redundant standardisation. Based on the standardised indicator data, positive ideal solutions and negative ideal solutions are computed using the formulas shown in (12) and (13):

$${Z_j}^{\ast }=\underset{i}{\max } z_{ij}$$

(12)

$${Z_j}^{-}=\underset{i}{\min } z_{ij}$$

(13)

$z_{ij}$ denotes the standardised assessment value of the jth indicator within the zero-carbon factory evaluation system for the ith candidate enterprise. ${Z_j}^{*}$ represents the maximum value of the jth indicator in the candidate enterprise’s zero-carbon factory evaluation system, while ${Z_j}^{-}$ indicates the minimum value of the jth indicator within the same system. Due to the application of range standardisation and non-negative translation, ${Z_j}^{*}=1.0001$ and ${Z_j}^{-}=0.0001$.

6.3.3. Calculate the Group Effect Value S_i and the Individual Regret Value R_i

Based on the determined positive and negative ideal solutions, to further measure the proximity of each sample enterprise to the optimal scheme, this study calculates the distance between each evaluation subject and the ideal solution to derive the group benefit value $S_i$ and the individual regret value $R_i$. A smaller $S_i$ indicates higher comprehensive group benefit for the enterprise, signifying greater overall proximity to the positive ideal solution; a smaller $R_i$ indicates lower maximum individual regret for the enterprise, reflecting weaker shortfall effects in individual indicators. The specific calculations are presented in Equations (14) and (15):

$$S_i=\sum _{j=1}^m{\displaystyle \frac{{\omega }_j\left({Z_j}^{*}-z_{ij}\right)}{\left({Z_j}^{*}-{Z_j}^{-}\right)}}$$

(14)

$$R_i=\underset{i}{\mathrm{m}\mathrm{a}\mathrm{x}}{\displaystyle \frac{{\omega }_j\left({Z_j}^{*}-z_{ij}\right)}{{Z_j}^{*}-{Z_j}^{-}}}$$

(15)

In the formula, ${\omega }_j$ denotes the weight assigned to the j-th indicator as calculated by the entropy weight method described earlier, with the remaining variables retaining their previously defined meanings.

Using MATLAB 2018a for computation, the group efficiency values $S_i$, individual regret values $R_i$, and their rankings were obtained for 37 listed steel companies. The core results are presented in

Table 15. Zero-carbon factory S_i and R_i values and rankings for 37 listed steel companies.

	C1	C2	C3	C4	C5	C6	C7	C8	C9	C10
Si	0.4051	0.5414	0.6813	0.7334	0.7276	0.3551	0.6703	0.6551	0.5572	0.6471
	(4)	(13)	(21)	(32)	(31)	(1)	(26)	(24)	(15)	(20)
Ri	0.1197	0.2554	0.2750	0.3083	0.3065	0.1064	0.2975	0.3001	0.2702	0.2869
	(4)	(13)	(21)	(32)	(31)	(1)	(26)	(24)	(15)	(20)
	C11	C12	C13	C14	C15	C16	C17	C18	C19	C20
Si	0.5872	0.7186	0.7666	0.5930	0.3469	0.4481	0.6219	0.7334	0.5470	0.6645
	(14)	(25)	(30)	(12)	(2)	(7)	(16)	(27)	(17)	(22)
Ri	0.2402	0.2776	0.2886	0.1803	0.1176	0.1261	0.2546	0.2807	0.2863	0.2885
	(14)	(25)	(30)	(12)	(2)	(7)	(16)	(27)	(17)	(22)
	C21	C22	C23	C24	C25	C26	C27	C28	C29	C30
Si	0.7359	0.8818	0.5963	0.4618	0.7655	0.4968	0.6935	0.5711	0.8358	0.8220
	(29)	(37)	(18)	(10)	(33)	(9)	(23)	(11)	(36)	(35)
Ri	0.2987	0.3141	0.2698	0.2027	0.2975	0.1359	0.2781	0.1878	0.3112	0.3077
	(29)	(37)	(18)	(10)	(33)	(9)	(23)	(11)	(36)	(35)
	C31	C32	C33	C34	C35	C36	C37
Si	0.7309	0.4102	0.7648	0.4269	0.4625	0.7027	0.4601
	(28)	(3)	(34)	(6)	(8)	(19)	(5)
Ri	0.2969	0.1134	0.3119	0.1210	0.1344	0.2570	0.1093
	(28)	(3)	(34)	(6)	(8)	(19)	(5)

Note: The values in parentheses denote the rankings of the corresponding companies for each criterion.

.

6.3.4. Calculating the VIKOR Value Q_i Under the Zero-Carbon Factory Evaluation System

The comprehensive evaluation value Q_i represents a compromise metric integrating collective benefits and individual regrets. To establish the final ranking of zero-carbon factory performance for steel enterprises, the VIKOR value for each evaluation subject must be calculated by incorporating decision mechanism coefficients (as applied in [60,61]), as per Formula (16) below:

$$Q_i=v{\displaystyle \frac{S_i-S^{*}}{S^{-}-S^{*}}}+(1-v){\displaystyle \frac{R_i-R^{*}}{R^{-}-R^{*}}}$$

(16)

${\mathrm{Q}}_{\mathrm{i}}$ denotes the VIKOR value for the i-th zero-carbon factory. Here, $S^{*}=\min S_i, S^{-}=\max S_i$; $R^{*}=\min R_i, R^{-}=\max R_i$. That is, $S^{*}$ represents the maximum collective benefit under the zero-carbon factory evaluation system, while $R^{*}$ indicates the minimum collective regret, with $v$ being the weighting factor for maximum collective utility. To simultaneously pursue the maximum group benefit and minimise individual regret within the zero-carbon factory evaluation framework, $v$ is set to 0.5 herein. From the table, we obtain $S^{*}=0.34694$, $S^{-}=0.88179$; $R^{*}=0.10641$, $R^{-}=0.31414$. The value of ${\mathrm{Q}}_{\mathrm{i}}$ can be calculated using Equation (16), as shown in

Table 16. Q_i Values, Rankings and Certification Outcomes for Zero-Carbon Factories Among 37 Listed Steel Companies.

	C1	C2	C3	C4	C5	C6	C7	C8	C9	C10
Qi	0.0865	0.5405	0.7184	0.8474	0.8374	0.0076	0.7623	0.7543	0.5910	0.7151
	(4)	(13)	(21)	(32)	(31)	(1)	(26)	(24)	(15)	(20)
		Yes				Yes	Yes			Yes
	C11	C12	C13	C14	C15	C16	C17	C18	C19	C20
Qi	0.5466	0.7594	0.8308	0.4078	0.0270	0.1419	0.6136	0.7807	0.6200	0.7352
	(14)	(25)	(30)	(12)	(2)	(7)	(16)	(27)	(17)	(22)
	Yes	Yes	Yes		Yes	Yes	Yes	Yes		Yes
	C21	C22	C23	C24	C25	C26	C27	C28	C29	C30
Qi	0.8263	1.0000	0.6263	0.3392	0.8514	0.2111	0.7372	0.4055	0.9499	0.9285
	(29)	(37)	(18)	(10)	(33)	(9)	(23)	(11)	(36)	(35)
	Yes	Yes		Yes	Yes
	C31	C32	C33	C34	C35	C36	C37
Qi	0.8176	0.0761	0.8853	0.1315	0.1754	0.6950	0.1128
	(28)	(3)	(34)	(6)	(8)	(19)	(5)
		Yes	Yes

Note: Companies marked with ‘Yes’ are included in the Ministry of Industry and Information Technology’s list of certified green factories. The values in parentheses denote the rankings of the corresponding companies for each criterion.

.

6.3.5. Comprehensive Evaluation Ranking of Zero-Carbon Factories

Arrange the sample enterprises in ascending order based on their ${\mathrm{Q}}_{\mathrm{i}}$ values. A higher ranking (smaller ${\mathrm{Q}}_{\mathrm{i}}$ value) indicates superior comprehensive performance of the enterprise’s zero-carbon factory, bringing it closer to the ideal solution. According to the core principles of the VIKOR method, the certification of a zero-carbon factory requires simultaneous fulfilment of the following two conditions:

Condition 1: Acceptable Advantage Degree Criterion

$$Q(C^{\left(2\right)})-Q({\mathrm{C}}^{\left(1\right)})\ge DQ={\displaystyle \frac{1}{n-1}}$$

In the formula, ${\mathrm{C}}^{\left(1\right)}$ denotes the enterprise with the highest ${\mathrm{Q}}_{\mathrm{i}}$ value, $C^{\left(2\right)}$ denotes the enterprise with the second-highest ${\mathrm{Q}}_{\mathrm{i}}$ value, n represents the number of sample enterprises, and $DQ$ denotes the dominance threshold.

Condition 2: Acceptable decision stability criteria

The top-ranked entity ${\mathrm{C}}^{\left(1\right)}$ must also be optimal in the ranking of either the group benefit value $S_i$ or the individual regret value $R_i$, thereby ensuring the decision stability of the optimal solution.

Based on the aforementioned rules, the evaluation results of this study are determined as follows: the sample size n of this study is 37; thus, $\mathrm{D}\mathrm{Q}=1/(37-1)\approx 0.0278$. The top-ranked enterprise ${\mathrm{C}}_6$ occupies the first position in the individual regret value $R_i$ ranking, satisfying the acceptable decision stability criterion. The difference in ${\mathrm{Q}}_{\mathrm{i}}$ values between the second-ranked and first-ranked enterprises is $Q(C^{\left(2\right)})-Q({\mathrm{C}}^{\left(1\right)})=0.02700-0.00763=0.01937$. This difference falls below the threshold $DQ=0.0278$, thus failing to meet the acceptable dominance criterion.

According to the compromise solution determination rules of the VIKOR model, if a scheme satisfies condition 2 but fails to meet condition 1, a compromise solution set is generated: all enterprises satisfying $Q(C^{\left(k\right)})-Q({\mathrm{C}}^{\left(1\right)})<DQ$ collectively form the ideal compromise scheme set, i.e., the benchmark group of zero-carbon factories identified in this study.

Calculations reveal that only enterprises ${\mathrm{C}}_6$ and ${\mathrm{C}}_{15}$ exhibit a ${\mathrm{Q}}_{\mathrm{i}}$ value gap within the threshold DQ relative to the leading enterprise. Consequently, these two enterprises collectively form the ideal compromise solution set for evaluating zero-carbon factories in the steel industry and are designated as the benchmark enterprises for zero-carbon factories in the steel sector within this assessment.

6.4. Analysis of Zero-Carbon Factory Evaluation Results in the Steel Industry

6.4.1. Overall Characteristics of Zero-Carbon Factory Assessment Results

The comprehensive evaluation results indicate significant variations in the zero-carbon factory performance of steel enterprises. Among the 37 sample enterprises, the ${\mathrm{Q}}_{\mathrm{i}}$ value ranged from a minimum of 0.00763 to a maximum of 1.00000, with an overall sample mean of 0.581 and a standard deviation of 0.312. Under multi-indicator constraints, enterprises exhibited distinct hierarchical distributions, reflecting markedly uneven comprehensive capabilities across low-carbon performance, green momentum, competitive advantage, and economic resilience. Overall, zero-carbon factory evaluation is not determined by a single indicator but rather the combined effect of multidimensional factors, with inter-firm differences exhibiting strong structural characteristics.

6.4.2. Comparative Analysis of Benchmark Enterprises and Lagging Enterprises

Based on the decision rules derived from the VIKOR method, enterprises ${\mathrm{C}}_6$ and ${\mathrm{C}}_{15}$ have been identified as the “ideal compromise set” for zero-carbon factory evaluation within the steel industry, constituting the benchmark tier for zero-carbon excellence. From the perspective of core metrics, both benchmark enterprises have achieved synergistic development across low-carbon performance, green momentum, and economic resilience. Their carbon emission intensities stand at 0.3644 and 0.2937 respectively, both falling significantly below the sample average of 1.572 and ranking among the top two in the entire sample, aligning with the core requirements for zero-carbon factory construction. In the green momentum dimension, their OIP indicators (6.97 and 4.22) are positioned in the upper range of the sample, demonstrating sustained green innovation capabilities, while both recorded positive EPW growth, reflecting increasing strategic focus on low-carbon transformation. Regarding economic resilience, the two companies’ debt-to-asset ratios (0.5066 and 0.3487) indicate sound financial structures, with ample operating cash flow and positive ROA providing sufficient financial backing for long-term zero-carbon investments. Notably, both benchmark enterprises are nationally certified green factories by the Ministry of Industry and Information Technology, and the alignment between this study’s evaluation outcomes and official accreditation fully validates the effectiveness of the assessment framework developed herein.

In contrast, ${\mathrm{C}}_{22}$, ranked last in the comprehensive evaluation, demonstrated significant shortcomings across all core dimensions. Its carbon emission intensity reached 2.0745—the highest in the entire sample—indicating severely inadequate emission control. Its OIP indicator stood at merely 3.97, while its EPW indicator was negative, revealing insufficient technological support and declining strategic attention towards zero-carbon transition. Financially, its ROA of −0.1023 indicates persistently deteriorating profitability, compounded by elevated gearing ratios and financing constraints that render it unable to support necessary zero-carbon investments. Such enterprises exhibit dual deficiencies in carbon control and operational performance, and under intensifying carbon constraints there is an urgent need to address these shortcomings through production process upgrades, green technological innovation, and strategic transformation.

6.4.3. Distribution Characteristics of Zero-Carbon Factories at the Industry Level

From an industry-wide perspective, the evaluation results for zero-carbon factories in the steel sector reveal a pronounced divergence. Only a handful of enterprises achieved a relatively low overall assessment score, demonstrating strong comprehensive advantages across multiple indicators and possessing considerable potential as zero-carbon exemplars.

Concurrently, the majority of enterprises’ comprehensive evaluation scores cluster within the higher range, indicating that the steel industry as a whole remains in a transitional phase from high-carbon to low-carbon emissions. For these enterprises, a stable and mature development model for zero-carbon transformation has yet to be established, with considerable variation persisting within the industry regarding pathway selection and implementation pace.

From the sample distribution of ${\mathrm{Q}}_{\mathrm{i}}$ values, only two enterprises had ${\mathrm{Q}}_{\mathrm{i}}$ values below the dominance threshold $\mathrm{D}\mathrm{Q}=0.0278$, qualifying them as benchmark zero-carbon factories. Only eight enterprises had ${\mathrm{Q}}_{\mathrm{i}}$ values below 0.2, placing them in the industry’s first tier for zero-carbon transformation. Over 70% of sampled enterprises exhibit ${\mathrm{Q}}_{\mathrm{i}}$ values concentrated within the 0.4–1.0 range, indicating that the steel industry as a whole remains in a transitional phase from high-carbon emissions towards low-carbon and zero-carbon operations. For the majority of enterprises, a stable and mature development model for zero-carbon transformation has yet to be established, with significant disparities persisting within the industry regarding chosen transformation pathways, resource allocation intensity, and implementation effectiveness. In terms of the alignment between evaluation outcomes and official green factory certification, the sample enterprises included in the Ministry of Industry and Information Technology’s national green factory register exhibited an average ${\mathrm{Q}}_{\mathrm{i}}$ value of 0.42. This figure is significantly lower than the overall sample mean of 0.581. This indicates that the zero-carbon factory evaluation framework developed in this study demonstrates a high degree of intrinsic consistency with the official certification criteria for national green factories. This further validates the practical effectiveness and scientific rigour of the evaluation system.

Overall, the steel industry’s transition to net-zero carbon emissions exhibits distinct tiered characteristics. Leading benchmark enterprises have achieved synergistic development of environmental performance and operational efficiency, while the majority of companies still face challenges of insufficient transformation momentum and inadequate capabilities. This distribution pattern provides a practical basis for subsequently advancing net-zero factory construction through industry stratification and categorisation, as well as formulating differentiated transformation support policies.

6.4.4. Sensitivity and Stability Analysis of the Evaluation Model

In the VIKOR framework, the decision-making mechanism coefficient $\mathrm{v}$ represents the weight assigned to maximising group benefit relative to minimising individual regret. The baseline assessment in this study uses $\mathrm{v}=0.5$, which implies that equal weight is given to both objectives. To test whether the assessment results are sensitive to this specific choice, this study generated nine different decision–preference scenarios by adjusting the value of $\mathrm{v}$ in increments of 0.1 from 0.1 to 0.9, and conducted a systematic sensitivity analysis. These scenarios range from strategies that almost entirely prioritise minimising individual regret ($\mathrm{v}=0.1$) to those that almost entirely prioritise maximising group benefit ($\mathrm{v}=0.9$). For each scenario, the VIKOR composite score ${\mathrm{Q}}_{\mathrm{i}}$ was recalculated for all 37 sample steel enterprises.

The results demonstrate that the rankings of both the top-tier and bottom-tier enterprises remain highly stable across all nine scenarios. In the range from $\mathrm{v}=0.1$ to $\mathrm{v}=0.7$, Firm ${\mathrm{C}}_6$ consistently ranked first due to its lowest individual regret value; meanwhile, ${\mathrm{C}}_{15}$ surpassed ${\mathrm{C}}_6$ at $\mathrm{v}=0.8$ and $\mathrm{v}=0.9$ by achieving the lowest social utility value. These two enterprises consistently ranked within the top five across all nine scenarios, and remained in the top two positions in eight of the scenarios ranging from $\mathrm{v}=0.2$ to $\mathrm{v}=0.9$, confirming the robustness of the zero-carbon benchmark enterprise certification results with respect to the choice of v-values. More broadly, the top five enterprises under $\mathrm{v}=0.5$ (${\mathrm{C}}_6$, ${\mathrm{C}}_{15}$, ${\mathrm{C}}_{32}$, ${\mathrm{C}}_1$, ${\mathrm{C}}_{37}$) consistently ranked within the top seven across all scenarios, with ranking fluctuations for ${\mathrm{C}}_6$, ${\mathrm{C}}_{32}$, and ${\mathrm{C}}_{16}$ not exceeding one position, and those for ${\mathrm{C}}_{15}$ and ${\mathrm{C}}_1$ not exceeding three positions. At the other end, the bottom five enterprises (${\mathrm{C}}_{22}$, ${\mathrm{C}}_{29}$, ${\mathrm{C}}_{30}$, ${\mathrm{C}}_{33}$, ${\mathrm{C}}_{25}$) remained anchored in positions 31–37 across all scenarios, with ${\mathrm{C}}_{22}$ consistently ranking last and ${\mathrm{C}}_{29}$ maintaining the 36th position throughout, both exhibiting zero fluctuation.

The average fluctuation in ranking across the 37 enterprises was approximately 3.2 positions. The greatest fluctuations were concentrated in the middle tier (positions 10 to 25), where the standardised S-values and R-values deviated significantly from the expected curve. For instance, ${\mathrm{C}}_{19}$ and ${\mathrm{C}}_{36}$ each exhibited the largest fluctuation of 9 positions due to opposing movements in their S-values and R-values; however, even these fluctuations remained within a limited range and did not alter the fundamental hierarchical structure of the evaluation system. Meanwhile, certain mid-tier enterprises such as ${\mathrm{C}}_{35}$ displayed complete ranking stability (fluctuation of 0), further indicating that the observed mid-tier variability is localised rather than systemic.

To quantify the overall degree of consistency in ranking results across different v-value scenarios, this paper calculated Spearman’s rank correlation coefficients between pairs of the nine scenarios. The results are shown in

Table 17. Spearman Rank Correlation Coefficients Across Nine Decision-Preference Scenarios.

	v = 0.1	v = 0.2	v = 0.3	v = 0.4	v = 0.5	v = 0.6	v = 0.7	v = 0.8	v = 0.9
v = 0.1	1	0.9967	0.9943	0.9867	0.9789	0.9706	0.9583	0.9462	0.9355
v = 0.2	0.9967	1	0.9988	0.9938	0.9884	0.9822	0.9718	0.9623	0.9535
v = 0.3	0.9943	0.9988	1	0.9972	0.9927	0.9877	0.9791	0.9701	0.9621
v = 0.4	0.9867	0.9938	0.9972	1	0.9979	0.9950	0.9891	0.9820	0.9749
v = 0.5	0.9789	0.9884	0.9927	0.9979	1	0.9988	0.9943	0.9889	0.9829
v = 0.6	0.9706	0.9822	0.9877	0.9950	0.9988	1	0.9974	0.9934	0.9884
v = 0.7	0.9583	0.9718	0.9791	0.9891	0.9943	0.9974	1	0.9972	0.9929
v = 0.8	0.9462	0.9623	0.9701	0.9820	0.9889	0.9934	0.9972	1	0.9979
v = 0.9	0.9355	0.9535	0.9621	0.9749	0.9829	0.9884	0.9929	0.9979	1

Note: All pairwise Spearman rank correlation coefficients exceed 0.93, indicating that the ranking outcomes remain highly consistent across different decision-preference scenarios. Bold values indicate the diagonal elements of the correlation matrix, representing the self-correlation of each scenario, which is always equal to 1.

; the rank correlation coefficients for all pairs of scenarios exceed 0.93, with the lowest coefficient between adjacent v-value scenarios being 0.9967 and the rank correlation coefficient between the two most extreme scenarios (v = 0.1 and v = 0.9) standing at 0.9355. All coefficients are well above the threshold of 0.9 for high correlation, indicating a high degree of consistency in the ranking results across different decision-making preferences.

Based on the above analysis, the sensitivity test results indicate that when the decision mechanism coefficient $\mathrm{v}$ is varied from 0.1 to 0.9, the VIKOR rankings of the 37 sample steel enterprises remain generally stable. The rankings of ${\mathrm{C}}_6$, ${\mathrm{C}}_{15}$ and the enterprises at the bottom of the list are virtually unaffected by changes in the value of v, whilst those of the mid-ranking enterprises show some fluctuation in position, albeit to a limited extent. Spearman’s rank correlation coefficients between the various scenarios are all higher than 0.93, indicating that the evaluation results possess excellent robustness. The evaluation rankings obtained in the main text of this paper using a compromise strategy of $\mathrm{v}=0.5$ are fully representative and reliable; the research conclusions do not depend on the selection of a specific value for $\mathrm{v}$.

7. Conclusions and Recommendations

7.1. Conclusions and Discussion

This study employs panel data from Chinese A-share listed companies on the Shanghai and Shenzhen stock exchanges between 2010 and 2022 to systematically examine whether the carbon emission intensity metric in financial statements effectively represents a firm’s actual carbon footprint. It further investigates the moderating effects of innovation output sustainability and competitive position, alongside the transmission mechanisms linking resource allocation efficiency and financing constraints. Empirical findings consistently indicate that corporate carbon emission intensity exhibits a significant positive association with total carbon emissions, and this result remains robust after addressing endogeneity concerns related to reverse causality, omitted variables, and simultaneity through multiple econometric strategies. This demonstrates that under current accounting standards and disclosure frameworks, carbon emission intensity serves not only as a financial ratio metric but also as an effective representation of the physical scale of a company’s carbon emissions. However, this mapping relationship is influenced by both internal corporate characteristics and external environmental factors. Research indicates that both the sustainability of corporate innovation outputs and competitive standing significantly amplify the positive relationship between carbon emission intensity and corporate carbon emissions, thereby enhancing the representational capacity of accounting information regarding an entity’s carbon emissions. Mechanism testing further reveals that carbon emission intensity indirectly influences and maps onto corporate carbon emission levels through two pathways: enhancing resource allocation efficiency and strengthening financing constraints. However, as both channels operate as partial mediators, additional transmission pathways likely coexist. Potential candidates include supply chain carbon cost pass-through, corporate governance oversight mechanisms, information environment pressures from analysts and media, and managerial behavioural responses to carbon risk. These unmodelled factors may simultaneously shape the mapping relationship between carbon emission intensity and actual emissions, and their identification represents an important direction for future research. Heterogeneity analysis indicates that the explicit release of environmental signals and positive environmental attitudes contribute to standardising carbon accounting practices, thereby enhancing the representativeness of carbon emission intensity for actual emissions; a higher level of green technology has to some extent mitigated the positive correlation between carbon emission intensity and carbon emissions. This constitutes a “green paradox” in carbon accounting: as firms adopt advanced green technologies, their carbon emission intensity decreases substantially, yet the corresponding reduction in actual total emissions may be disproportionately smaller due to concurrent production scale expansion or efficiency-driven output growth. Consequently, the representational capacity of financial statement-derived carbon emission intensity as a proxy for actual emissions is distorted, potentially leading to underestimation of the true carbon footprint of technologically advanced firms.

The aforementioned conclusions provide a novel theoretical perspective for zero-carbon factory development and carbon accounting optimisation, revealing the key transmission mechanisms linking financial information and environmental performance while highlighting the institutional and technical challenges of integrating traditional carbon accounting with financial statement data. This study also provides empirical support for establishing a low-cost, comprehensive carbon emissions monitoring system under China’s dual carbon goals.

Notably, differences in innovation governance, environmental attitudes, and green technology foundations produce marked divergence in the relationship between carbon emission intensity and corporate carbon emissions. For enterprises with robust governance and proactive environmental attitudes, innovation strengthens representational efficacy through standardised accounting and optimised resource allocation; conversely, enterprises with advanced green technologies exhibit innovations that disrupt this representation, suppressing the positive correlation between the two. This reveals the limitations of single-indicator assessments, providing crucial reference for policymakers to comprehensively evaluate both technological efficacy and actual emissions control.

Building upon this foundation, this paper focuses on the steel industry and conducts applied research on zero-carbon factory certification using the Entropy-VIKOR model. Through a multi-level evaluation framework encompassing low-carbon performance, green momentum, competitive advantage, and economic resilience, the study identifies enterprises ${\mathrm{C}}_6$ and ${\mathrm{C}}_{15}$ as the “ideal compromise solution set,” demonstrating industry-leading performance across carbon emission control, green innovation, and financial stability. The evaluation outcomes align closely with the Ministry of Industry and Information Technology’s national-level green factory accreditation, further validating the effectiveness and practical applicability of the accounting-based zero-carbon factory evaluation system.

7.2. Policy Implications and Recommendations

Based on the above conclusions, this paper offers the following recommendations to government regulatory bodies, corporate managers and investors:

In the short term, the early-warning function of carbon emission intensity indicators should be fully leveraged while strengthening market-based constraint mechanisms. The confirmed positive mapping between carbon emission intensity and actual emissions provides a direct empirical basis for regulators to deploy financial statement-derived carbon emission intensity as a low-cost, scalable monitoring metric, while the Entropy-VIKOR framework, validated through its high consistency with official green factory certifications, offers a replicable template for standardised zero-carbon factory certification across sectors. For corporate managers, the mediating role of resource allocation efficiency implies that redirecting capital from high-carbon assets towards low-carbon investments can simultaneously reduce actual emissions and improve the reliability of reported carbon emission intensity, while the debt constraint mechanism underscores the financial imperative to proactively lower carbon emission intensity in order to reduce borrowing costs and preserve operational flexibility. The positive moderating effects of innovation sustainability and competitive position further suggest that sustained green innovation and stronger market positioning enhance the accuracy of carbon accounting, thereby facilitating access to green finance. Conversely, the inhibitory effect of high green technology levels cautions policymakers against relying exclusively on intensity metrics for technologically advanced firms, as the green paradox may mask rising total emissions behind declining intensity figures. For investors, the heterogeneity results indicate that financial carbon metrics are most informative for firms with clear environmental signals and strong innovation output sustainability, enabling more precise carbon risk pricing and capital allocation towards enterprises with genuine zero-carbon transition capabilities.

In the long term, the demonstrated validity of financial statement-based carbon emission intensity as an emissions proxy, together with the identified moderating and mediating conditions, calls for the systematic establishment of a carbon information governance ecosystem to drive institutional transformation. A unified national platform for carbon accounting and financial disclosure should be created, integrating carbon risk into the macroprudential regulatory framework to make carbon information a core parameter influencing credit approval and market access. Accounting systems must be deeply integrated with carbon accounting frameworks, while disclosure rules for corporate environmental accounting information should be refined to provide institutional support for carbon emission monitoring and zero-carbon factory assessments based on accounting data. Enterprises must elevate carbon management to a strategic core position, establishing accounting systems that cover all business processes. Carbon cost–benefit analysis should become a routine decision-making tool, and active participation in industry standard-setting is essential to shape green competitiveness. Investors should develop pricing models that integrate carbon risk, using active shareholder engagement to urge companies to enhance disclosure quality. This will guide long-term capital flows towards enterprises with genuine low-carbon transition capabilities, thereby establishing a market-driven sustainable financial constraint mechanism. Tripartite collaboration will ultimately achieve a virtuous cycle of policy effectiveness, corporate competitiveness and investment returns. To realise these recommendations, a tripartite collaborative empowerment framework is proposed, as illustrated in Figure 6.

7.3. Future Perspectives and Limitations

Future research may build upon the findings and theoretical framework of this study to explore the following practical avenues, thereby enhancing the application value of carbon accounting theory within zero-carbon governance. These research pathways may be pursued in a balanced manner:

Firstly, comparative research across nations and systems is a key avenue for expansion. The carbon emission intensity of financial statements in this study is closely correlated with the actual total emissions of enterprises. This study confirms a statistically significant correlation between the two variables; however, this relationship may be influenced by multiple factors including corporate governance structures, industry characteristics, and macroeconomic policies. In particular, differences in environmental regulation regimes and carbon disclosure systems across jurisdictions may fundamentally shape whether this proxy relationship can be generalised beyond China. Subsequent research may apply this study’s framework to different institutional environments. For example, the EU’s mandatory carbon trading scheme and unified sustainability reporting directive may render carbon emission intensity metrics more closely linked to total emissions, whereas under the US’s predominantly voluntary ESG disclosure framework, this relationship may depend more on corporate governance standards and investor pressure. Comparing correlation strength across these differing contexts would systematically elucidate how regulatory pressures, market incentives, and disclosure cultures collectively shape the representational relationship between financial carbon emission intensity and corporate carbon emissions.

Secondly, research should delve deeper into how the efficiency gains and scale expansion effects arising from green technology application influence this representational relationship. Future research may analyse how green technology efficiency improvements, shifts in production scale, and alterations in cost structures interact with corporate carbon emission intensity to modulate its impact on actual carbon emissions. Furthermore, where firm-level actual emissions data are available, future studies should conduct direct comparisons between the financial statement-derived proxy and verified emissions figures, and perform sensitivity and validity tests under varying levels of green technology adoption to quantify the extent of proxy distortion attributable to the green paradox. Such studies could elucidate the mediating role in the “green paradox” or “rebound effect,” providing evidence for policymakers to identify false low-carbon enterprises and optimise emissions reduction incentive policies. Thirdly, expanding the application scope of the accounting-based zero-carbon factory evaluation system. This study conducted single-industry validation using the steel sector. Subsequent research may extend the framework to other high-energy-consuming industries such as building materials, chemicals, and thermal power generation, refining indicator design by integrating sector-specific carbon emission characteristics to establish differentiated certification standards. Concurrently, panel data could be incorporated to conduct dynamic evaluations, further refining the dynamic certification and exit mechanisms for zero-carbon factories.

These three pathways complement one another: transnational comparisons illuminate institutional influences, technical effect analyses dissect the complexities of transformation, and the evaluation framework expands and refines the implementation of zero-carbon factory certification. Together, they propel carbon accounting beyond corporate emissions towards systemic governance, providing the basis for enhancing disclosure quality and designing leak-proof policies, thereby supporting the global transition to net-zero carbon.

Due to limitations in data availability, the conclusions of this study are based on a sample of A-share listed companies, and the application research on zero-carbon factories has only covered listed companies in the A-share steel industry. While the findings cannot be directly extrapolated to non-listed enterprises with opaque data or inadequate accounting systems, nor to jurisdictions with substantially different environmental regulation and carbon disclosure requirements, they nonetheless provide valuable evidence for understanding the role of accounting information in carbon accounting. Specifically, the industry-level apportionment method used to estimate the dependent variable assumes that within-industry emissions are proportional to firms’ cost shares, which may compress firm-level variation arising from heterogeneous production technologies and fuel mixes. This attenuation of observable variation could lead to underestimation of the moderating effects identified in this study, while overstating the mediating role of financing constraints for firms whose true emissions deviate substantially from the industry-proportional estimate. Additionally, while the proxy variable has been validated through statistical correlation and the steel industry case study, direct sensitivity and validity tests comparing the proxy against verified emissions data for individual firms remain limited by data availability. Moreover, the mediation analysis focuses on resource allocation efficiency and financing constraints as two partial mediating channels, leaving other potentially important pathways such as supply chain effects, governance mechanisms, and information environment pressures empirically untested. Future research may validate and extend these findings to broader corporate populations through field investigations, sector-specific tracking, or collaborative efforts with enterprises to obtain primary emissions data, while also employing structural equation modelling or sequential mediation designs to capture these complementary transmission mechanisms. This would enhance carbon monitoring systems and zero-carbon factory evaluation frameworks grounded in accounting information.